diff --git "a/checkpoint-12588/trainer_state.json" "b/checkpoint-12588/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/checkpoint-12588/trainer_state.json"
@@ -0,0 +1,93923 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.0,
+  "global_step": 12588,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.968253968253968e-06,
+      "loss": 1.3695,
+      "theoretical_loss": 3.515365452996946,
+      "tokens_seen": 1650130944
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 7.936507936507936e-06,
+      "loss": 1.3931,
+      "theoretical_loss": 3.515341478637045,
+      "tokens_seen": 1650262016
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 1.1904761904761905e-05,
+      "loss": 1.3746,
+      "theoretical_loss": 3.5153175067143483,
+      "tokens_seen": 1650393088
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 1.5873015873015872e-05,
+      "loss": 1.4017,
+      "theoretical_loss": 3.5152935372284144,
+      "tokens_seen": 1650524160
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 1.984126984126984e-05,
+      "loss": 1.3655,
+      "theoretical_loss": 3.515269570178803,
+      "tokens_seen": 1650655232
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 2.380952380952381e-05,
+      "loss": 1.3475,
+      "theoretical_loss": 3.515245605565072,
+      "tokens_seen": 1650786304
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 2.7777777777777776e-05,
+      "loss": 1.3167,
+      "theoretical_loss": 3.5152216433867807,
+      "tokens_seen": 1650917376
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.1746031746031745e-05,
+      "loss": 1.3606,
+      "theoretical_loss": 3.515197683643489,
+      "tokens_seen": 1651048448
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.571428571428571e-05,
+      "loss": 1.4394,
+      "theoretical_loss": 3.515173726334755,
+      "tokens_seen": 1651179520
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.968253968253968e-05,
+      "loss": 1.3852,
+      "theoretical_loss": 3.5151497714601385,
+      "tokens_seen": 1651310592
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 4.365079365079365e-05,
+      "loss": 1.3411,
+      "theoretical_loss": 3.5151258190191994,
+      "tokens_seen": 1651441664
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 4.761904761904762e-05,
+      "loss": 1.3545,
+      "theoretical_loss": 3.5151018690114975,
+      "tokens_seen": 1651572736
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 5.1587301587301586e-05,
+      "loss": 1.3834,
+      "theoretical_loss": 3.5150779214365917,
+      "tokens_seen": 1651703808
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 5.555555555555555e-05,
+      "loss": 1.4067,
+      "theoretical_loss": 3.515053976294043,
+      "tokens_seen": 1651834880
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 5.9523809523809524e-05,
+      "loss": 1.3485,
+      "theoretical_loss": 3.5150300335834097,
+      "tokens_seen": 1651965952
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 6.349206349206349e-05,
+      "loss": 1.3379,
+      "theoretical_loss": 3.5150060933042533,
+      "tokens_seen": 1652097024
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 6.746031746031745e-05,
+      "loss": 1.4202,
+      "theoretical_loss": 3.5149821554561336,
+      "tokens_seen": 1652228096
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 7.142857142857142e-05,
+      "loss": 1.4509,
+      "theoretical_loss": 3.5149582200386105,
+      "tokens_seen": 1652359168
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 7.53968253968254e-05,
+      "loss": 1.3737,
+      "theoretical_loss": 3.514934287051245,
+      "tokens_seen": 1652490240
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 7.936507936507937e-05,
+      "loss": 1.3169,
+      "theoretical_loss": 3.5149103564935977,
+      "tokens_seen": 1652621312
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 8.333333333333333e-05,
+      "loss": 1.319,
+      "theoretical_loss": 3.5148864283652292,
+      "tokens_seen": 1652752384
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 8.73015873015873e-05,
+      "loss": 1.4098,
+      "theoretical_loss": 3.514862502665699,
+      "tokens_seen": 1652883456
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.126984126984126e-05,
+      "loss": 1.3878,
+      "theoretical_loss": 3.51483857939457,
+      "tokens_seen": 1653014528
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.523809523809524e-05,
+      "loss": 1.3756,
+      "theoretical_loss": 3.514814658551402,
+      "tokens_seen": 1653145600
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.92063492063492e-05,
+      "loss": 1.3952,
+      "theoretical_loss": 3.5147907401357563,
+      "tokens_seen": 1653276672
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00010317460317460317,
+      "loss": 1.4438,
+      "theoretical_loss": 3.514766824147194,
+      "tokens_seen": 1653407744
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00010714285714285714,
+      "loss": 1.3289,
+      "theoretical_loss": 3.514742910585277,
+      "tokens_seen": 1653538816
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0001111111111111111,
+      "loss": 1.3928,
+      "theoretical_loss": 3.514718999449566,
+      "tokens_seen": 1653669888
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00011507936507936508,
+      "loss": 1.3424,
+      "theoretical_loss": 3.5146950907396235,
+      "tokens_seen": 1653800960
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00011904761904761905,
+      "loss": 1.3681,
+      "theoretical_loss": 3.5146711844550103,
+      "tokens_seen": 1653932032
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.000123015873015873,
+      "loss": 1.3379,
+      "theoretical_loss": 3.5146472805952884,
+      "tokens_seen": 1654063104
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00012698412698412698,
+      "loss": 1.3265,
+      "theoretical_loss": 3.5146233791600197,
+      "tokens_seen": 1654194176
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00013095238095238096,
+      "loss": 1.4431,
+      "theoretical_loss": 3.5145994801487666,
+      "tokens_seen": 1654325248
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0001349206349206349,
+      "loss": 1.3846,
+      "theoretical_loss": 3.514575583561091,
+      "tokens_seen": 1654456320
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0001388888888888889,
+      "loss": 1.3978,
+      "theoretical_loss": 3.514551689396555,
+      "tokens_seen": 1654587392
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00014285714285714284,
+      "loss": 1.3472,
+      "theoretical_loss": 3.514527797654721,
+      "tokens_seen": 1654718464
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00014682539682539685,
+      "loss": 1.4142,
+      "theoretical_loss": 3.514503908335152,
+      "tokens_seen": 1654849536
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0001507936507936508,
+      "loss": 1.3236,
+      "theoretical_loss": 3.51448002143741,
+      "tokens_seen": 1654980608
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00015476190476190478,
+      "loss": 1.375,
+      "theoretical_loss": 3.5144561369610576,
+      "tokens_seen": 1655111680
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00015873015873015873,
+      "loss": 1.3313,
+      "theoretical_loss": 3.5144322549056586,
+      "tokens_seen": 1655242752
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0001626984126984127,
+      "loss": 1.3801,
+      "theoretical_loss": 3.5144083752707744,
+      "tokens_seen": 1655373824
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00016666666666666666,
+      "loss": 1.4213,
+      "theoretical_loss": 3.5143844980559695,
+      "tokens_seen": 1655504896
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00017063492063492064,
+      "loss": 1.3847,
+      "theoretical_loss": 3.514360623260806,
+      "tokens_seen": 1655635968
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0001746031746031746,
+      "loss": 1.3418,
+      "theoretical_loss": 3.5143367508848478,
+      "tokens_seen": 1655767040
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00017857142857142857,
+      "loss": 1.3739,
+      "theoretical_loss": 3.5143128809276583,
+      "tokens_seen": 1655898112
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00018253968253968252,
+      "loss": 1.3561,
+      "theoretical_loss": 3.514289013388801,
+      "tokens_seen": 1656029184
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0001865079365079365,
+      "loss": 1.3653,
+      "theoretical_loss": 3.514265148267839,
+      "tokens_seen": 1656160256
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00019047619047619048,
+      "loss": 1.3505,
+      "theoretical_loss": 3.5142412855643363,
+      "tokens_seen": 1656291328
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00019444444444444446,
+      "loss": 1.3245,
+      "theoretical_loss": 3.514217425277857,
+      "tokens_seen": 1656422400
+    },
+    {
+      "epoch": 0.0,
+      "objective/train/advantage_avg": 0.49312451481819153,
+      "objective/train/docs_used": 604805,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3617349863052368,
+      "objective/train/original_loss": 1.3617348670959473,
+      "objective/train/theoretical_loss": 3.5141935674079647,
+      "objective/train/tokens_used": 27012576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24361729621887207,
+      "objective/train/value_max": -0.5,
+      "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": 1.4305692911148071,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.5141935674079647,
+      "tokens_seen": 1656553472
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0001984126984126984,
+      "loss": 1.4058,
+      "theoretical_loss": 3.5141935674079647,
+      "tokens_seen": 1656553472
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0002023809523809524,
+      "loss": 1.3224,
+      "theoretical_loss": 3.5141697119542235,
+      "tokens_seen": 1656684544
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00020634920634920634,
+      "loss": 1.358,
+      "theoretical_loss": 3.514145858916198,
+      "tokens_seen": 1656815616
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00021031746031746032,
+      "loss": 1.3574,
+      "theoretical_loss": 3.5141220082934517,
+      "tokens_seen": 1656946688
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00021428571428571427,
+      "loss": 1.4469,
+      "theoretical_loss": 3.51409816008555,
+      "tokens_seen": 1657077760
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00021825396825396825,
+      "loss": 1.3044,
+      "theoretical_loss": 3.5140743142920567,
+      "tokens_seen": 1657208832
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0002222222222222222,
+      "loss": 1.3209,
+      "theoretical_loss": 3.5140504709125366,
+      "tokens_seen": 1657339904
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00022619047619047618,
+      "loss": 1.3432,
+      "theoretical_loss": 3.514026629946555,
+      "tokens_seen": 1657470976
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00023015873015873016,
+      "loss": 1.2955,
+      "theoretical_loss": 3.5140027913936756,
+      "tokens_seen": 1657602048
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00023412698412698414,
+      "loss": 1.403,
+      "theoretical_loss": 3.5139789552534646,
+      "tokens_seen": 1657733120
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0002380952380952381,
+      "loss": 1.3106,
+      "theoretical_loss": 3.513955121525486,
+      "tokens_seen": 1657864192
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00024206349206349207,
+      "loss": 1.3722,
+      "theoretical_loss": 3.513931290209306,
+      "tokens_seen": 1657995264
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.000246031746031746,
+      "loss": 1.3989,
+      "theoretical_loss": 3.5139074613044885,
+      "tokens_seen": 1658126336
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00025,
+      "loss": 1.288,
+      "theoretical_loss": 3.5138836348106004,
+      "tokens_seen": 1658257408
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00025396825396825396,
+      "loss": 1.3981,
+      "theoretical_loss": 3.5138598107272068,
+      "tokens_seen": 1658388480
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00025793650793650796,
+      "loss": 1.432,
+      "theoretical_loss": 3.513835989053873,
+      "tokens_seen": 1658519552
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0002619047619047619,
+      "loss": 1.4049,
+      "theoretical_loss": 3.513812169790165,
+      "tokens_seen": 1658650624
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00026587301587301587,
+      "loss": 1.3807,
+      "theoretical_loss": 3.5137883529356486,
+      "tokens_seen": 1658781696
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0002698412698412698,
+      "loss": 1.3385,
+      "theoretical_loss": 3.51376453848989,
+      "tokens_seen": 1658912768
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0002738095238095238,
+      "loss": 1.4366,
+      "theoretical_loss": 3.5137407264524554,
+      "tokens_seen": 1659043840
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0002777777777777778,
+      "loss": 1.4143,
+      "theoretical_loss": 3.5137169168229105,
+      "tokens_seen": 1659174912
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00028174603174603173,
+      "loss": 1.4107,
+      "theoretical_loss": 3.5136931096008217,
+      "tokens_seen": 1659305984
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0002857142857142857,
+      "loss": 1.3443,
+      "theoretical_loss": 3.5136693047857555,
+      "tokens_seen": 1659437056
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0002896825396825397,
+      "loss": 1.2858,
+      "theoretical_loss": 3.513645502377279,
+      "tokens_seen": 1659568128
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0002936507936507937,
+      "loss": 1.5038,
+      "theoretical_loss": 3.513621702374958,
+      "tokens_seen": 1659699200
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00029761904761904765,
+      "loss": 1.3992,
+      "theoretical_loss": 3.5135979047783596,
+      "tokens_seen": 1659830272
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0003015873015873016,
+      "loss": 1.3845,
+      "theoretical_loss": 3.5135741095870507,
+      "tokens_seen": 1659961344
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0003055555555555556,
+      "loss": 1.2885,
+      "theoretical_loss": 3.5135503168005986,
+      "tokens_seen": 1660092416
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00030952380952380956,
+      "loss": 1.4511,
+      "theoretical_loss": 3.5135265264185707,
+      "tokens_seen": 1660223488
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0003134920634920635,
+      "loss": 1.3512,
+      "theoretical_loss": 3.5135027384405326,
+      "tokens_seen": 1660354560
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00031746031746031746,
+      "loss": 1.3552,
+      "theoretical_loss": 3.5134789528660537,
+      "tokens_seen": 1660485632
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00032142857142857147,
+      "loss": 1.3153,
+      "theoretical_loss": 3.5134551696946996,
+      "tokens_seen": 1660616704
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0003253968253968254,
+      "loss": 1.2872,
+      "theoretical_loss": 3.5134313889260396,
+      "tokens_seen": 1660747776
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00032936507936507937,
+      "loss": 1.3429,
+      "theoretical_loss": 3.5134076105596397,
+      "tokens_seen": 1660878848
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0003333333333333333,
+      "loss": 1.4111,
+      "theoretical_loss": 3.5133838345950688,
+      "tokens_seen": 1661009920
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00033730158730158733,
+      "loss": 1.4397,
+      "theoretical_loss": 3.5133600610318942,
+      "tokens_seen": 1661140992
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0003412698412698413,
+      "loss": 1.4037,
+      "theoretical_loss": 3.5133362898696845,
+      "tokens_seen": 1661272064
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00034523809523809523,
+      "loss": 1.3321,
+      "theoretical_loss": 3.5133125211080074,
+      "tokens_seen": 1661403136
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0003492063492063492,
+      "loss": 1.3974,
+      "theoretical_loss": 3.513288754746431,
+      "tokens_seen": 1661534208
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0003531746031746032,
+      "loss": 1.3761,
+      "theoretical_loss": 3.513264990784524,
+      "tokens_seen": 1661665280
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00035714285714285714,
+      "loss": 1.3624,
+      "theoretical_loss": 3.5132412292218547,
+      "tokens_seen": 1661796352
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0003611111111111111,
+      "loss": 1.3429,
+      "theoretical_loss": 3.5132174700579917,
+      "tokens_seen": 1661927424
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00036507936507936505,
+      "loss": 1.402,
+      "theoretical_loss": 3.5131937132925035,
+      "tokens_seen": 1662058496
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00036904761904761905,
+      "loss": 1.4001,
+      "theoretical_loss": 3.5131699589249585,
+      "tokens_seen": 1662189568
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000373015873015873,
+      "loss": 1.3363,
+      "theoretical_loss": 3.513146206954927,
+      "tokens_seen": 1662320640
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000376984126984127,
+      "loss": 1.3404,
+      "theoretical_loss": 3.513122457381977,
+      "tokens_seen": 1662451712
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00038095238095238096,
+      "loss": 1.4184,
+      "theoretical_loss": 3.5130987102056768,
+      "tokens_seen": 1662582784
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00038492063492063497,
+      "loss": 1.3575,
+      "theoretical_loss": 3.5130749654255977,
+      "tokens_seen": 1662713856
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0003888888888888889,
+      "loss": 1.3939,
+      "theoretical_loss": 3.513051223041307,
+      "tokens_seen": 1662844928
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0003928571428571429,
+      "loss": 1.4304,
+      "theoretical_loss": 3.513027483052376,
+      "tokens_seen": 1662976000
+    },
+    {
+      "epoch": 0.01,
+      "objective/train/advantage_avg": 0.4920444190502167,
+      "objective/train/docs_used": 607267,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3075224161148071,
+      "objective/train/original_loss": 1.3075224161148071,
+      "objective/train/theoretical_loss": 3.5130037454583727,
+      "objective/train/tokens_used": 33566176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24258464574813843,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504376888275146,
+      "objective/train/weighted_lm_loss": 1.3734228610992432,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.5130037454583727,
+      "tokens_seen": 1663107072
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0003968253968253968,
+      "loss": 1.3401,
+      "theoretical_loss": 3.5130037454583727,
+      "tokens_seen": 1663107072
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00040079365079365083,
+      "loss": 1.4049,
+      "theoretical_loss": 3.5129800102588673,
+      "tokens_seen": 1663238144
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004047619047619048,
+      "loss": 1.4203,
+      "theoretical_loss": 3.51295627745343,
+      "tokens_seen": 1663369216
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00040873015873015874,
+      "loss": 1.3671,
+      "theoretical_loss": 3.51293254704163,
+      "tokens_seen": 1663500288
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004126984126984127,
+      "loss": 1.3695,
+      "theoretical_loss": 3.512908819023038,
+      "tokens_seen": 1663631360
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004166666666666667,
+      "loss": 1.3762,
+      "theoretical_loss": 3.512885093397224,
+      "tokens_seen": 1663762432
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00042063492063492065,
+      "loss": 1.3731,
+      "theoretical_loss": 3.5128613701637574,
+      "tokens_seen": 1663893504
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004246031746031746,
+      "loss": 1.4363,
+      "theoretical_loss": 3.5128376493222095,
+      "tokens_seen": 1664024576
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00042857142857142855,
+      "loss": 1.3627,
+      "theoretical_loss": 3.512813930872151,
+      "tokens_seen": 1664155648
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00043253968253968256,
+      "loss": 1.364,
+      "theoretical_loss": 3.512790214813151,
+      "tokens_seen": 1664286720
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004365079365079365,
+      "loss": 1.3679,
+      "theoretical_loss": 3.5127665011447817,
+      "tokens_seen": 1664417792
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00044047619047619046,
+      "loss": 1.3338,
+      "theoretical_loss": 3.5127427898666133,
+      "tokens_seen": 1664548864
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004444444444444444,
+      "loss": 1.3595,
+      "theoretical_loss": 3.5127190809782167,
+      "tokens_seen": 1664679936
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004484126984126984,
+      "loss": 1.3742,
+      "theoretical_loss": 3.5126953744791627,
+      "tokens_seen": 1664811008
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00045238095238095237,
+      "loss": 1.3755,
+      "theoretical_loss": 3.5126716703690226,
+      "tokens_seen": 1664942080
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004563492063492063,
+      "loss": 1.3729,
+      "theoretical_loss": 3.5126479686473675,
+      "tokens_seen": 1665073152
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00046031746031746033,
+      "loss": 1.3381,
+      "theoretical_loss": 3.5126242693137693,
+      "tokens_seen": 1665204224
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00046428571428571433,
+      "loss": 1.4221,
+      "theoretical_loss": 3.5126005723677984,
+      "tokens_seen": 1665335296
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004682539682539683,
+      "loss": 1.3567,
+      "theoretical_loss": 3.5125768778090274,
+      "tokens_seen": 1665466368
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00047222222222222224,
+      "loss": 1.4678,
+      "theoretical_loss": 3.5125531856370276,
+      "tokens_seen": 1665597440
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004761904761904762,
+      "loss": 1.3589,
+      "theoretical_loss": 3.5125294958513704,
+      "tokens_seen": 1665728512
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004801587301587302,
+      "loss": 1.4317,
+      "theoretical_loss": 3.5125058084516283,
+      "tokens_seen": 1665859584
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00048412698412698415,
+      "loss": 1.4362,
+      "theoretical_loss": 3.512482123437373,
+      "tokens_seen": 1665990656
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004880952380952381,
+      "loss": 1.3499,
+      "theoretical_loss": 3.5124584408081767,
+      "tokens_seen": 1666121728
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000492063492063492,
+      "loss": 1.4308,
+      "theoretical_loss": 3.5124347605636115,
+      "tokens_seen": 1666252800
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000496031746031746,
+      "loss": 1.3653,
+      "theoretical_loss": 3.5124110827032498,
+      "tokens_seen": 1666383872
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005,
+      "loss": 1.4815,
+      "theoretical_loss": 3.5123874072266643,
+      "tokens_seen": 1666514944
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004999598780292088,
+      "loss": 1.4123,
+      "theoretical_loss": 3.512363734133427,
+      "tokens_seen": 1666646016
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004999197560584176,
+      "loss": 1.45,
+      "theoretical_loss": 3.5123400634231112,
+      "tokens_seen": 1666777088
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004998796340876264,
+      "loss": 1.2764,
+      "theoretical_loss": 3.512316395095289,
+      "tokens_seen": 1666908160
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004998395121168352,
+      "loss": 1.3985,
+      "theoretical_loss": 3.5122927291495336,
+      "tokens_seen": 1667039232
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004997993901460439,
+      "loss": 1.2939,
+      "theoretical_loss": 3.5122690655854187,
+      "tokens_seen": 1667170304
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004997592681752528,
+      "loss": 1.3953,
+      "theoretical_loss": 3.512245404402516,
+      "tokens_seen": 1667301376
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004997191462044615,
+      "loss": 1.3859,
+      "theoretical_loss": 3.5122217456004,
+      "tokens_seen": 1667432448
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004996790242336704,
+      "loss": 1.3152,
+      "theoretical_loss": 3.512198089178643,
+      "tokens_seen": 1667563520
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004996389022628792,
+      "loss": 1.3683,
+      "theoretical_loss": 3.5121744351368194,
+      "tokens_seen": 1667694592
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000499598780292088,
+      "loss": 1.3496,
+      "theoretical_loss": 3.512150783474502,
+      "tokens_seen": 1667825664
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004995586583212968,
+      "loss": 1.4414,
+      "theoretical_loss": 3.5121271341912648,
+      "tokens_seen": 1667956736
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004995185363505056,
+      "loss": 1.3534,
+      "theoretical_loss": 3.5121034872866814,
+      "tokens_seen": 1668087808
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004994784143797144,
+      "loss": 1.3453,
+      "theoretical_loss": 3.512079842760326,
+      "tokens_seen": 1668218880
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004994382924089231,
+      "loss": 1.3539,
+      "theoretical_loss": 3.512056200611772,
+      "tokens_seen": 1668349952
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004993981704381319,
+      "loss": 1.474,
+      "theoretical_loss": 3.512032560840594,
+      "tokens_seen": 1668481024
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004993580484673407,
+      "loss": 1.3211,
+      "theoretical_loss": 3.5120089234463663,
+      "tokens_seen": 1668612096
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004993179264965495,
+      "loss": 1.4576,
+      "theoretical_loss": 3.511985288428663,
+      "tokens_seen": 1668743168
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004992778045257583,
+      "loss": 1.3862,
+      "theoretical_loss": 3.5119616557870583,
+      "tokens_seen": 1668874240
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004992376825549671,
+      "loss": 1.3536,
+      "theoretical_loss": 3.511938025521127,
+      "tokens_seen": 1669005312
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004991975605841759,
+      "loss": 1.4104,
+      "theoretical_loss": 3.5119143976304437,
+      "tokens_seen": 1669136384
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004991574386133847,
+      "loss": 1.4271,
+      "theoretical_loss": 3.511890772114583,
+      "tokens_seen": 1669267456
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004991173166425935,
+      "loss": 1.4434,
+      "theoretical_loss": 3.51186714897312,
+      "tokens_seen": 1669398528
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004990771946718023,
+      "loss": 1.4699,
+      "theoretical_loss": 3.5118435282056297,
+      "tokens_seen": 1669529600
+    },
+    {
+      "epoch": 0.01,
+      "objective/train/advantage_avg": 0.49601492285728455,
+      "objective/train/docs_used": 609762,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.4597387313842773,
+      "objective/train/original_loss": 1.4597387313842773,
+      "objective/train/theoretical_loss": 3.5118199098116873,
+      "objective/train/tokens_used": 40119776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24622058868408203,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0508533716201782,
+      "objective/train/weighted_lm_loss": 1.5338997840881348,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.5118199098116873,
+      "tokens_seen": 1669660672
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000499037072701011,
+      "loss": 1.4602,
+      "theoretical_loss": 3.5118199098116873,
+      "tokens_seen": 1669660672
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004989969507302198,
+      "loss": 1.4326,
+      "theoretical_loss": 3.5117962937908676,
+      "tokens_seen": 1669791744
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004989568287594286,
+      "loss": 1.3369,
+      "theoretical_loss": 3.5117726801427454,
+      "tokens_seen": 1669922816
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004989167067886375,
+      "loss": 1.3943,
+      "theoretical_loss": 3.5117490688668975,
+      "tokens_seen": 1670053888
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004988765848178463,
+      "loss": 1.4945,
+      "theoretical_loss": 3.511725459962898,
+      "tokens_seen": 1670184960
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004988364628470551,
+      "loss": 1.416,
+      "theoretical_loss": 3.511701853430324,
+      "tokens_seen": 1670316032
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004987963408762639,
+      "loss": 1.3914,
+      "theoretical_loss": 3.5116782492687504,
+      "tokens_seen": 1670447104
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004987562189054726,
+      "loss": 1.3483,
+      "theoretical_loss": 3.5116546474777524,
+      "tokens_seen": 1670578176
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004987160969346815,
+      "loss": 1.4315,
+      "theoretical_loss": 3.5116310480569073,
+      "tokens_seen": 1670709248
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004986759749638902,
+      "loss": 1.3635,
+      "theoretical_loss": 3.51160745100579,
+      "tokens_seen": 1670840320
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004986358529930991,
+      "loss": 1.3853,
+      "theoretical_loss": 3.511583856323978,
+      "tokens_seen": 1670971392
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004985957310223078,
+      "loss": 1.3968,
+      "theoretical_loss": 3.511560264011046,
+      "tokens_seen": 1671102464
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004985556090515166,
+      "loss": 1.4094,
+      "theoretical_loss": 3.5115366740665714,
+      "tokens_seen": 1671233536
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004985154870807254,
+      "loss": 1.4298,
+      "theoretical_loss": 3.5115130864901305,
+      "tokens_seen": 1671364608
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004984753651099342,
+      "loss": 1.4488,
+      "theoretical_loss": 3.5114895012813,
+      "tokens_seen": 1671495680
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000498435243139143,
+      "loss": 1.4073,
+      "theoretical_loss": 3.5114659184396566,
+      "tokens_seen": 1671626752
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004983951211683518,
+      "loss": 1.4102,
+      "theoretical_loss": 3.511442337964777,
+      "tokens_seen": 1671757824
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004983549991975606,
+      "loss": 1.4759,
+      "theoretical_loss": 3.5114187598562374,
+      "tokens_seen": 1671888896
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004983148772267693,
+      "loss": 1.3719,
+      "theoretical_loss": 3.5113951841136166,
+      "tokens_seen": 1672019968
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004982747552559782,
+      "loss": 1.429,
+      "theoretical_loss": 3.5113716107364903,
+      "tokens_seen": 1672151040
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004982346332851869,
+      "loss": 1.4434,
+      "theoretical_loss": 3.5113480397244365,
+      "tokens_seen": 1672282112
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004981945113143958,
+      "loss": 1.3819,
+      "theoretical_loss": 3.5113244710770326,
+      "tokens_seen": 1672413184
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004981543893436046,
+      "loss": 1.414,
+      "theoretical_loss": 3.5113009047938553,
+      "tokens_seen": 1672544256
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004981142673728134,
+      "loss": 1.3956,
+      "theoretical_loss": 3.5112773408744826,
+      "tokens_seen": 1672675328
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004980741454020222,
+      "loss": 1.3643,
+      "theoretical_loss": 3.5112537793184924,
+      "tokens_seen": 1672806400
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000498034023431231,
+      "loss": 1.4305,
+      "theoretical_loss": 3.511230220125462,
+      "tokens_seen": 1672937472
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004979939014604398,
+      "loss": 1.4303,
+      "theoretical_loss": 3.5112066632949706,
+      "tokens_seen": 1673068544
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004979537794896486,
+      "loss": 1.4142,
+      "theoretical_loss": 3.511183108826595,
+      "tokens_seen": 1673199616
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004979136575188573,
+      "loss": 1.4683,
+      "theoretical_loss": 3.5111595567199134,
+      "tokens_seen": 1673330688
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004978735355480661,
+      "loss": 1.3987,
+      "theoretical_loss": 3.5111360069745046,
+      "tokens_seen": 1673461760
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004978334135772749,
+      "loss": 1.4134,
+      "theoretical_loss": 3.5111124595899463,
+      "tokens_seen": 1673592832
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004977932916064837,
+      "loss": 1.3569,
+      "theoretical_loss": 3.5110889145658177,
+      "tokens_seen": 1673723904
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004977531696356925,
+      "loss": 1.3864,
+      "theoretical_loss": 3.5110653719016964,
+      "tokens_seen": 1673854976
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004977130476649013,
+      "loss": 1.4096,
+      "theoretical_loss": 3.5110418315971623,
+      "tokens_seen": 1673986048
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004976729256941101,
+      "loss": 1.4337,
+      "theoretical_loss": 3.511018293651793,
+      "tokens_seen": 1674117120
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004976328037233188,
+      "loss": 1.3006,
+      "theoretical_loss": 3.510994758065168,
+      "tokens_seen": 1674248192
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004975926817525277,
+      "loss": 1.4556,
+      "theoretical_loss": 3.5109712248368665,
+      "tokens_seen": 1674379264
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004975525597817364,
+      "loss": 1.392,
+      "theoretical_loss": 3.510947693966467,
+      "tokens_seen": 1674510336
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004975124378109452,
+      "loss": 1.4857,
+      "theoretical_loss": 3.5109241654535497,
+      "tokens_seen": 1674641408
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000497472315840154,
+      "loss": 1.437,
+      "theoretical_loss": 3.5109006392976925,
+      "tokens_seen": 1674772480
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004974321938693629,
+      "loss": 1.3713,
+      "theoretical_loss": 3.510877115498476,
+      "tokens_seen": 1674903552
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004973920718985717,
+      "loss": 1.3887,
+      "theoretical_loss": 3.510853594055479,
+      "tokens_seen": 1675034624
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004973519499277805,
+      "loss": 1.4352,
+      "theoretical_loss": 3.5108300749682817,
+      "tokens_seen": 1675165696
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004973118279569893,
+      "loss": 1.3809,
+      "theoretical_loss": 3.510806558236464,
+      "tokens_seen": 1675296768
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004972717059861981,
+      "loss": 1.2798,
+      "theoretical_loss": 3.510783043859605,
+      "tokens_seen": 1675427840
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004972315840154069,
+      "loss": 1.3397,
+      "theoretical_loss": 3.5107595318372846,
+      "tokens_seen": 1675558912
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004971914620446156,
+      "loss": 1.3052,
+      "theoretical_loss": 3.510736022169084,
+      "tokens_seen": 1675689984
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004971513400738245,
+      "loss": 1.3646,
+      "theoretical_loss": 3.5107125148545824,
+      "tokens_seen": 1675821056
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004971112181030332,
+      "loss": 1.4565,
+      "theoretical_loss": 3.51068900989336,
+      "tokens_seen": 1675952128
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000497071096132242,
+      "loss": 1.373,
+      "theoretical_loss": 3.5106655072849984,
+      "tokens_seen": 1676083200
+    },
+    {
+      "epoch": 0.02,
+      "objective/train/advantage_avg": 0.49048250913619995,
+      "objective/train/docs_used": 612232,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.310646653175354,
+      "objective/train/original_loss": 1.3106465339660645,
+      "objective/train/theoretical_loss": 3.5106420070290767,
+      "objective/train/tokens_used": 46673376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2414161115884781,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0502755641937256,
+      "objective/train/weighted_lm_loss": 1.3763915300369263,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.5106420070290767,
+      "tokens_seen": 1676214272
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004970309741614508,
+      "loss": 1.3443,
+      "theoretical_loss": 3.5106420070290767,
+      "tokens_seen": 1676214272
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004969908521906596,
+      "loss": 1.3382,
+      "theoretical_loss": 3.5106185091251767,
+      "tokens_seen": 1676345344
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004969507302198684,
+      "loss": 1.3725,
+      "theoretical_loss": 3.5105950135728783,
+      "tokens_seen": 1676476416
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004969106082490772,
+      "loss": 1.331,
+      "theoretical_loss": 3.5105715203717627,
+      "tokens_seen": 1676607488
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000496870486278286,
+      "loss": 1.3484,
+      "theoretical_loss": 3.5105480295214107,
+      "tokens_seen": 1676738560
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004968303643074947,
+      "loss": 1.3638,
+      "theoretical_loss": 3.510524541021403,
+      "tokens_seen": 1676869632
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004967902423367036,
+      "loss": 1.3657,
+      "theoretical_loss": 3.5105010548713214,
+      "tokens_seen": 1677000704
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004967501203659123,
+      "loss": 1.3866,
+      "theoretical_loss": 3.5104775710707474,
+      "tokens_seen": 1677131776
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004967099983951212,
+      "loss": 1.3396,
+      "theoretical_loss": 3.510454089619261,
+      "tokens_seen": 1677262848
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00049666987642433,
+      "loss": 1.4666,
+      "theoretical_loss": 3.510430610516445,
+      "tokens_seen": 1677393920
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004966297544535388,
+      "loss": 1.2801,
+      "theoretical_loss": 3.5104071337618805,
+      "tokens_seen": 1677524992
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004965896324827476,
+      "loss": 1.3018,
+      "theoretical_loss": 3.510383659355149,
+      "tokens_seen": 1677656064
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004965495105119564,
+      "loss": 1.3302,
+      "theoretical_loss": 3.5103601872958325,
+      "tokens_seen": 1677787136
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004965093885411652,
+      "loss": 1.4484,
+      "theoretical_loss": 3.510336717583513,
+      "tokens_seen": 1677918208
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000496469266570374,
+      "loss": 1.3864,
+      "theoretical_loss": 3.5103132502177727,
+      "tokens_seen": 1678049280
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004964291445995827,
+      "loss": 1.4445,
+      "theoretical_loss": 3.5102897851981933,
+      "tokens_seen": 1678180352
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004963890226287915,
+      "loss": 1.4149,
+      "theoretical_loss": 3.510266322524357,
+      "tokens_seen": 1678311424
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004963489006580003,
+      "loss": 1.4412,
+      "theoretical_loss": 3.510242862195846,
+      "tokens_seen": 1678442496
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004963087786872091,
+      "loss": 1.4124,
+      "theoretical_loss": 3.510219404212244,
+      "tokens_seen": 1678573568
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004962686567164179,
+      "loss": 1.4134,
+      "theoretical_loss": 3.5101959485731316,
+      "tokens_seen": 1678704640
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004962285347456267,
+      "loss": 1.3986,
+      "theoretical_loss": 3.510172495278093,
+      "tokens_seen": 1678835712
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004961884127748355,
+      "loss": 1.3678,
+      "theoretical_loss": 3.51014904432671,
+      "tokens_seen": 1678966784
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004961482908040443,
+      "loss": 1.4253,
+      "theoretical_loss": 3.5101255957185664,
+      "tokens_seen": 1679097856
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004961081688332531,
+      "loss": 1.421,
+      "theoretical_loss": 3.5101021494532443,
+      "tokens_seen": 1679228928
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004960680468624618,
+      "loss": 1.408,
+      "theoretical_loss": 3.5100787055303266,
+      "tokens_seen": 1679360000
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004960279248916708,
+      "loss": 1.3072,
+      "theoretical_loss": 3.510055263949398,
+      "tokens_seen": 1679491072
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004959878029208794,
+      "loss": 1.3419,
+      "theoretical_loss": 3.51003182471004,
+      "tokens_seen": 1679622144
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004959476809500883,
+      "loss": 1.4195,
+      "theoretical_loss": 3.510008387811837,
+      "tokens_seen": 1679753216
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004959075589792971,
+      "loss": 1.3175,
+      "theoretical_loss": 3.5099849532543725,
+      "tokens_seen": 1679884288
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004958674370085059,
+      "loss": 1.3609,
+      "theoretical_loss": 3.5099615210372295,
+      "tokens_seen": 1680015360
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004958273150377147,
+      "loss": 1.3781,
+      "theoretical_loss": 3.5099380911599924,
+      "tokens_seen": 1680146432
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004957871930669235,
+      "loss": 1.3961,
+      "theoretical_loss": 3.5099146636222445,
+      "tokens_seen": 1680277504
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004957470710961323,
+      "loss": 1.2856,
+      "theoretical_loss": 3.5098912384235703,
+      "tokens_seen": 1680408576
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000495706949125341,
+      "loss": 1.3224,
+      "theoretical_loss": 3.5098678155635534,
+      "tokens_seen": 1680539648
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004956668271545499,
+      "loss": 1.2885,
+      "theoretical_loss": 3.509844395041778,
+      "tokens_seen": 1680670720
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004956267051837586,
+      "loss": 1.3963,
+      "theoretical_loss": 3.509820976857829,
+      "tokens_seen": 1680801792
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004955865832129674,
+      "loss": 1.4677,
+      "theoretical_loss": 3.5097975610112897,
+      "tokens_seen": 1680932864
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004955464612421762,
+      "loss": 1.4489,
+      "theoretical_loss": 3.509774147501745,
+      "tokens_seen": 1681063936
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000495506339271385,
+      "loss": 1.3752,
+      "theoretical_loss": 3.5097507363287797,
+      "tokens_seen": 1681195008
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004954662173005938,
+      "loss": 1.384,
+      "theoretical_loss": 3.5097273274919782,
+      "tokens_seen": 1681326080
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004954260953298026,
+      "loss": 1.3558,
+      "theoretical_loss": 3.5097039209909253,
+      "tokens_seen": 1681457152
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004953859733590114,
+      "loss": 1.45,
+      "theoretical_loss": 3.509680516825206,
+      "tokens_seen": 1681588224
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004953458513882202,
+      "loss": 1.386,
+      "theoretical_loss": 3.509657114994406,
+      "tokens_seen": 1681719296
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000495305729417429,
+      "loss": 1.3814,
+      "theoretical_loss": 3.5096337154981088,
+      "tokens_seen": 1681850368
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004952656074466377,
+      "loss": 1.4282,
+      "theoretical_loss": 3.5096103183359,
+      "tokens_seen": 1681981440
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004952254854758466,
+      "loss": 1.3896,
+      "theoretical_loss": 3.5095869235073667,
+      "tokens_seen": 1682112512
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004951853635050554,
+      "loss": 1.3856,
+      "theoretical_loss": 3.509563531012092,
+      "tokens_seen": 1682243584
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004951452415342642,
+      "loss": 1.4611,
+      "theoretical_loss": 3.509540140849663,
+      "tokens_seen": 1682374656
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000495105119563473,
+      "loss": 1.3615,
+      "theoretical_loss": 3.509516753019665,
+      "tokens_seen": 1682505728
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004950649975926818,
+      "loss": 1.3666,
+      "theoretical_loss": 3.5094933675216833,
+      "tokens_seen": 1682636800
+    },
+    {
+      "epoch": 0.02,
+      "objective/train/advantage_avg": 0.4924430251121521,
+      "objective/train/docs_used": 614786,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2686177492141724,
+      "objective/train/original_loss": 1.2686177492141724,
+      "objective/train/theoretical_loss": 3.509469984355303,
+      "objective/train/tokens_used": 53226976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24293343722820282,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504794120788574,
+      "objective/train/weighted_lm_loss": 1.3326752185821533,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.509469984355303,
+      "tokens_seen": 1682767872
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004950248756218906,
+      "loss": 1.3521,
+      "theoretical_loss": 3.509469984355303,
+      "tokens_seen": 1682767872
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004949847536510994,
+      "loss": 1.3523,
+      "theoretical_loss": 3.5094466035201126,
+      "tokens_seen": 1682898944
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004949446316803081,
+      "loss": 1.4183,
+      "theoretical_loss": 3.5094232250156954,
+      "tokens_seen": 1683030016
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000494904509709517,
+      "loss": 1.4028,
+      "theoretical_loss": 3.509399848841639,
+      "tokens_seen": 1683161088
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004948643877387257,
+      "loss": 1.4222,
+      "theoretical_loss": 3.50937647499753,
+      "tokens_seen": 1683292160
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004948242657679345,
+      "loss": 1.3506,
+      "theoretical_loss": 3.509353103482953,
+      "tokens_seen": 1683423232
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004947841437971433,
+      "loss": 1.3954,
+      "theoretical_loss": 3.5093297342974967,
+      "tokens_seen": 1683554304
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004947440218263521,
+      "loss": 1.4068,
+      "theoretical_loss": 3.509306367440746,
+      "tokens_seen": 1683685376
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004947038998555609,
+      "loss": 1.4483,
+      "theoretical_loss": 3.509283002912288,
+      "tokens_seen": 1683816448
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004946637778847697,
+      "loss": 1.4106,
+      "theoretical_loss": 3.5092596407117105,
+      "tokens_seen": 1683947520
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004946236559139785,
+      "loss": 1.4925,
+      "theoretical_loss": 3.509236280838599,
+      "tokens_seen": 1684078592
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004945835339431872,
+      "loss": 1.4076,
+      "theoretical_loss": 3.5092129232925413,
+      "tokens_seen": 1684209664
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004945434119723962,
+      "loss": 1.4439,
+      "theoretical_loss": 3.5091895680731247,
+      "tokens_seen": 1684340736
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004945032900016048,
+      "loss": 1.3127,
+      "theoretical_loss": 3.5091662151799357,
+      "tokens_seen": 1684471808
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004944631680308137,
+      "loss": 1.4491,
+      "theoretical_loss": 3.5091428646125618,
+      "tokens_seen": 1684602880
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004944230460600225,
+      "loss": 1.4211,
+      "theoretical_loss": 3.5091195163705904,
+      "tokens_seen": 1684733952
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004943829240892313,
+      "loss": 1.4627,
+      "theoretical_loss": 3.5090961704536094,
+      "tokens_seen": 1684865024
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004943428021184401,
+      "loss": 1.5669,
+      "theoretical_loss": 3.509072826861207,
+      "tokens_seen": 1684996096
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004943026801476489,
+      "loss": 1.3618,
+      "theoretical_loss": 3.509049485592969,
+      "tokens_seen": 1685127168
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004942625581768577,
+      "loss": 1.3727,
+      "theoretical_loss": 3.509026146648485,
+      "tokens_seen": 1685258240
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004942224362060665,
+      "loss": 1.3035,
+      "theoretical_loss": 3.5090028100273427,
+      "tokens_seen": 1685389312
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004941823142352753,
+      "loss": 1.4248,
+      "theoretical_loss": 3.508979475729129,
+      "tokens_seen": 1685520384
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000494142192264484,
+      "loss": 1.4647,
+      "theoretical_loss": 3.5089561437534336,
+      "tokens_seen": 1685651456
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004941020702936929,
+      "loss": 1.3657,
+      "theoretical_loss": 3.508932814099844,
+      "tokens_seen": 1685782528
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004940619483229016,
+      "loss": 1.415,
+      "theoretical_loss": 3.5089094867679487,
+      "tokens_seen": 1685913600
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004940218263521104,
+      "loss": 1.4542,
+      "theoretical_loss": 3.5088861617573364,
+      "tokens_seen": 1686044672
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004939817043813192,
+      "loss": 1.4074,
+      "theoretical_loss": 3.5088628390675947,
+      "tokens_seen": 1686175744
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000493941582410528,
+      "loss": 1.4187,
+      "theoretical_loss": 3.5088395186983137,
+      "tokens_seen": 1686306816
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004939014604397368,
+      "loss": 1.397,
+      "theoretical_loss": 3.508816200649081,
+      "tokens_seen": 1686437888
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004938613384689456,
+      "loss": 1.4359,
+      "theoretical_loss": 3.5087928849194867,
+      "tokens_seen": 1686568960
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004938212164981544,
+      "loss": 1.4589,
+      "theoretical_loss": 3.5087695715091183,
+      "tokens_seen": 1686700032
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004937810945273631,
+      "loss": 1.3124,
+      "theoretical_loss": 3.508746260417566,
+      "tokens_seen": 1686831104
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000493740972556572,
+      "loss": 1.4518,
+      "theoretical_loss": 3.508722951644419,
+      "tokens_seen": 1686962176
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004937008505857808,
+      "loss": 1.3451,
+      "theoretical_loss": 3.5086996451892656,
+      "tokens_seen": 1687093248
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004936607286149896,
+      "loss": 1.3101,
+      "theoretical_loss": 3.5086763410516966,
+      "tokens_seen": 1687224320
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004936206066441984,
+      "loss": 1.3028,
+      "theoretical_loss": 3.5086530392313007,
+      "tokens_seen": 1687355392
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004935804846734072,
+      "loss": 1.428,
+      "theoretical_loss": 3.5086297397276676,
+      "tokens_seen": 1687486464
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000493540362702616,
+      "loss": 1.3774,
+      "theoretical_loss": 3.508606442540388,
+      "tokens_seen": 1687617536
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004935002407318248,
+      "loss": 1.4242,
+      "theoretical_loss": 3.5085831476690497,
+      "tokens_seen": 1687748608
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004934601187610335,
+      "loss": 1.364,
+      "theoretical_loss": 3.508559855113244,
+      "tokens_seen": 1687879680
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004934199967902424,
+      "loss": 1.4848,
+      "theoretical_loss": 3.5085365648725615,
+      "tokens_seen": 1688010752
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004933798748194511,
+      "loss": 1.4678,
+      "theoretical_loss": 3.508513276946591,
+      "tokens_seen": 1688141824
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004933397528486599,
+      "loss": 1.4497,
+      "theoretical_loss": 3.5084899913349235,
+      "tokens_seen": 1688272896
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004932996308778687,
+      "loss": 1.3989,
+      "theoretical_loss": 3.5084667080371497,
+      "tokens_seen": 1688403968
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004932595089070775,
+      "loss": 1.34,
+      "theoretical_loss": 3.5084434270528595,
+      "tokens_seen": 1688535040
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004932193869362863,
+      "loss": 1.3089,
+      "theoretical_loss": 3.508420148381643,
+      "tokens_seen": 1688666112
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004931792649654951,
+      "loss": 1.3543,
+      "theoretical_loss": 3.5083968720230923,
+      "tokens_seen": 1688797184
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004931391429947039,
+      "loss": 1.3645,
+      "theoretical_loss": 3.5083735979767967,
+      "tokens_seen": 1688928256
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004930990210239126,
+      "loss": 1.3935,
+      "theoretical_loss": 3.508350326242348,
+      "tokens_seen": 1689059328
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004930588990531215,
+      "loss": 1.4298,
+      "theoretical_loss": 3.5083270568193368,
+      "tokens_seen": 1689190400
+    },
+    {
+      "epoch": 0.02,
+      "objective/train/advantage_avg": 0.4927401542663574,
+      "objective/train/docs_used": 617011,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.4579259157180786,
+      "objective/train/original_loss": 1.457925796508789,
+      "objective/train/theoretical_loss": 3.5083037897073543,
+      "objective/train/tokens_used": 59780576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2431456297636032,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505101680755615,
+      "objective/train/weighted_lm_loss": 1.5314990282058716,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.5083037897073543,
+      "tokens_seen": 1689321472
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004930187770823302,
+      "loss": 1.4624,
+      "theoretical_loss": 3.5083037897073543,
+      "tokens_seen": 1689321472
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004929786551115392,
+      "loss": 1.2762,
+      "theoretical_loss": 3.5082805249059916,
+      "tokens_seen": 1689452544
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004929385331407479,
+      "loss": 1.3146,
+      "theoretical_loss": 3.5082572624148405,
+      "tokens_seen": 1689583616
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004928984111699567,
+      "loss": 1.4033,
+      "theoretical_loss": 3.5082340022334915,
+      "tokens_seen": 1689714688
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004928582891991655,
+      "loss": 1.3834,
+      "theoretical_loss": 3.508210744361537,
+      "tokens_seen": 1689845760
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004928181672283743,
+      "loss": 1.3702,
+      "theoretical_loss": 3.5081874887985682,
+      "tokens_seen": 1689976832
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004927780452575831,
+      "loss": 1.3917,
+      "theoretical_loss": 3.5081642355441764,
+      "tokens_seen": 1690107904
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004927379232867919,
+      "loss": 1.3198,
+      "theoretical_loss": 3.5081409845979543,
+      "tokens_seen": 1690238976
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004926978013160007,
+      "loss": 1.3811,
+      "theoretical_loss": 3.508117735959493,
+      "tokens_seen": 1690370048
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004926576793452094,
+      "loss": 1.4454,
+      "theoretical_loss": 3.508094489628385,
+      "tokens_seen": 1690501120
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004926175573744183,
+      "loss": 1.5402,
+      "theoretical_loss": 3.508071245604223,
+      "tokens_seen": 1690632192
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000492577435403627,
+      "loss": 1.368,
+      "theoretical_loss": 3.5080480038865978,
+      "tokens_seen": 1690763264
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004925373134328358,
+      "loss": 1.4413,
+      "theoretical_loss": 3.5080247644751026,
+      "tokens_seen": 1690894336
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004924971914620446,
+      "loss": 1.3646,
+      "theoretical_loss": 3.50800152736933,
+      "tokens_seen": 1691025408
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004924570694912534,
+      "loss": 1.3989,
+      "theoretical_loss": 3.507978292568872,
+      "tokens_seen": 1691156480
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004924169475204622,
+      "loss": 1.3613,
+      "theoretical_loss": 3.507955060073322,
+      "tokens_seen": 1691287552
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000492376825549671,
+      "loss": 1.424,
+      "theoretical_loss": 3.507931829882272,
+      "tokens_seen": 1691418624
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004923367035788798,
+      "loss": 1.4875,
+      "theoretical_loss": 3.507908601995315,
+      "tokens_seen": 1691549696
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004922965816080887,
+      "loss": 1.3386,
+      "theoretical_loss": 3.507885376412044,
+      "tokens_seen": 1691680768
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004922564596372974,
+      "loss": 1.4318,
+      "theoretical_loss": 3.5078621531320526,
+      "tokens_seen": 1691811840
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004922163376665062,
+      "loss": 1.344,
+      "theoretical_loss": 3.5078389321549333,
+      "tokens_seen": 1691942912
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000492176215695715,
+      "loss": 1.4108,
+      "theoretical_loss": 3.5078157134802797,
+      "tokens_seen": 1692073984
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004921360937249238,
+      "loss": 1.3601,
+      "theoretical_loss": 3.507792497107685,
+      "tokens_seen": 1692205056
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004920959717541326,
+      "loss": 1.4081,
+      "theoretical_loss": 3.5077692830367426,
+      "tokens_seen": 1692336128
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004920558497833414,
+      "loss": 1.4221,
+      "theoretical_loss": 3.5077460712670465,
+      "tokens_seen": 1692467200
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004920157278125502,
+      "loss": 1.3689,
+      "theoretical_loss": 3.5077228617981895,
+      "tokens_seen": 1692598272
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004919756058417589,
+      "loss": 1.4327,
+      "theoretical_loss": 3.507699654629767,
+      "tokens_seen": 1692729344
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004919354838709678,
+      "loss": 1.44,
+      "theoretical_loss": 3.5076764497613713,
+      "tokens_seen": 1692860416
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004918953619001765,
+      "loss": 1.3976,
+      "theoretical_loss": 3.5076532471925965,
+      "tokens_seen": 1692991488
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004918552399293853,
+      "loss": 1.4101,
+      "theoretical_loss": 3.5076300469230377,
+      "tokens_seen": 1693122560
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004918151179585941,
+      "loss": 1.3839,
+      "theoretical_loss": 3.507606848952289,
+      "tokens_seen": 1693253632
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004917749959878029,
+      "loss": 1.4341,
+      "theoretical_loss": 3.5075836532799434,
+      "tokens_seen": 1693384704
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004917348740170117,
+      "loss": 1.374,
+      "theoretical_loss": 3.507560459905597,
+      "tokens_seen": 1693515776
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004916947520462205,
+      "loss": 1.3621,
+      "theoretical_loss": 3.507537268828843,
+      "tokens_seen": 1693646848
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004916546300754293,
+      "loss": 1.4272,
+      "theoretical_loss": 3.507514080049276,
+      "tokens_seen": 1693777920
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004916145081046381,
+      "loss": 1.469,
+      "theoretical_loss": 3.5074908935664917,
+      "tokens_seen": 1693908992
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000491574386133847,
+      "loss": 1.3858,
+      "theoretical_loss": 3.5074677093800846,
+      "tokens_seen": 1694040064
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004915342641630556,
+      "loss": 1.4938,
+      "theoretical_loss": 3.5074445274896493,
+      "tokens_seen": 1694171136
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004914941421922646,
+      "loss": 1.3329,
+      "theoretical_loss": 3.507421347894781,
+      "tokens_seen": 1694302208
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004914540202214733,
+      "loss": 1.4584,
+      "theoretical_loss": 3.5073981705950743,
+      "tokens_seen": 1694433280
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004914138982506821,
+      "loss": 1.3845,
+      "theoretical_loss": 3.5073749955901254,
+      "tokens_seen": 1694564352
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004913737762798909,
+      "loss": 1.4548,
+      "theoretical_loss": 3.507351822879529,
+      "tokens_seen": 1694695424
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004913336543090997,
+      "loss": 1.3957,
+      "theoretical_loss": 3.507328652462881,
+      "tokens_seen": 1694826496
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004912935323383085,
+      "loss": 1.4543,
+      "theoretical_loss": 3.507305484339776,
+      "tokens_seen": 1694957568
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004912534103675173,
+      "loss": 1.4045,
+      "theoretical_loss": 3.5072823185098105,
+      "tokens_seen": 1695088640
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004912132883967261,
+      "loss": 1.3963,
+      "theoretical_loss": 3.50725915497258,
+      "tokens_seen": 1695219712
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004911731664259348,
+      "loss": 1.3517,
+      "theoretical_loss": 3.5072359937276802,
+      "tokens_seen": 1695350784
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004911330444551437,
+      "loss": 1.393,
+      "theoretical_loss": 3.507212834774707,
+      "tokens_seen": 1695481856
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004910929224843524,
+      "loss": 1.3629,
+      "theoretical_loss": 3.507189678113257,
+      "tokens_seen": 1695612928
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004910528005135612,
+      "loss": 1.4062,
+      "theoretical_loss": 3.507166523742926,
+      "tokens_seen": 1695744000
+    },
+    {
+      "epoch": 0.03,
+      "objective/train/advantage_avg": 0.49190375208854675,
+      "objective/train/docs_used": 619583,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.27505362033844,
+      "objective/train/original_loss": 1.27505362033844,
+      "objective/train/theoretical_loss": 3.50714337166331,
+      "objective/train/tokens_used": 66334176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2425486445426941,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050423502922058,
+      "objective/train/weighted_lm_loss": 1.3394197225570679,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.50714337166331,
+      "tokens_seen": 1695875072
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00049101267854277,
+      "loss": 1.3463,
+      "theoretical_loss": 3.50714337166331,
+      "tokens_seen": 1695875072
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004909725565719788,
+      "loss": 1.4812,
+      "theoretical_loss": 3.507120221874006,
+      "tokens_seen": 1696006144
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004909324346011876,
+      "loss": 1.3517,
+      "theoretical_loss": 3.507097074374609,
+      "tokens_seen": 1696137216
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004908923126303964,
+      "loss": 1.4115,
+      "theoretical_loss": 3.5070739291647173,
+      "tokens_seen": 1696268288
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004908521906596052,
+      "loss": 1.4356,
+      "theoretical_loss": 3.5070507862439264,
+      "tokens_seen": 1696399360
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000490812068688814,
+      "loss": 1.4689,
+      "theoretical_loss": 3.507027645611834,
+      "tokens_seen": 1696530432
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004907719467180228,
+      "loss": 1.3751,
+      "theoretical_loss": 3.5070045072680367,
+      "tokens_seen": 1696661504
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004907318247472316,
+      "loss": 1.2868,
+      "theoretical_loss": 3.5069813712121305,
+      "tokens_seen": 1696792576
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004906917027764404,
+      "loss": 1.4022,
+      "theoretical_loss": 3.5069582374437136,
+      "tokens_seen": 1696923648
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004906515808056492,
+      "loss": 1.2913,
+      "theoretical_loss": 3.5069351059623823,
+      "tokens_seen": 1697054720
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000490611458834858,
+      "loss": 1.4676,
+      "theoretical_loss": 3.506911976767735,
+      "tokens_seen": 1697185792
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004905713368640668,
+      "loss": 1.4048,
+      "theoretical_loss": 3.5068888498593678,
+      "tokens_seen": 1697316864
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004905312148932756,
+      "loss": 1.2725,
+      "theoretical_loss": 3.5068657252368793,
+      "tokens_seen": 1697447936
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004904910929224843,
+      "loss": 1.4036,
+      "theoretical_loss": 3.5068426028998667,
+      "tokens_seen": 1697579008
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004904509709516932,
+      "loss": 1.3666,
+      "theoretical_loss": 3.506819482847927,
+      "tokens_seen": 1697710080
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004904108489809019,
+      "loss": 1.4334,
+      "theoretical_loss": 3.5067963650806586,
+      "tokens_seen": 1697841152
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004903707270101108,
+      "loss": 1.3676,
+      "theoretical_loss": 3.50677324959766,
+      "tokens_seen": 1697972224
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004903306050393195,
+      "loss": 1.3502,
+      "theoretical_loss": 3.5067501363985283,
+      "tokens_seen": 1698103296
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004902904830685283,
+      "loss": 1.4625,
+      "theoretical_loss": 3.5067270254828613,
+      "tokens_seen": 1698234368
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004902503610977371,
+      "loss": 1.4125,
+      "theoretical_loss": 3.506703916850258,
+      "tokens_seen": 1698365440
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004902102391269459,
+      "loss": 1.4983,
+      "theoretical_loss": 3.5066808105003164,
+      "tokens_seen": 1698496512
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004901701171561547,
+      "loss": 1.3733,
+      "theoretical_loss": 3.5066577064326347,
+      "tokens_seen": 1698627584
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004901299951853635,
+      "loss": 1.4763,
+      "theoretical_loss": 3.5066346046468118,
+      "tokens_seen": 1698758656
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004900898732145723,
+      "loss": 1.313,
+      "theoretical_loss": 3.5066115051424465,
+      "tokens_seen": 1698889728
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000490049751243781,
+      "loss": 1.3631,
+      "theoretical_loss": 3.5065884079191365,
+      "tokens_seen": 1699020800
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00049000962927299,
+      "loss": 1.426,
+      "theoretical_loss": 3.5065653129764813,
+      "tokens_seen": 1699151872
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004899695073021987,
+      "loss": 1.3082,
+      "theoretical_loss": 3.5065422203140795,
+      "tokens_seen": 1699282944
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004899293853314075,
+      "loss": 1.3317,
+      "theoretical_loss": 3.5065191299315304,
+      "tokens_seen": 1699414016
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004898892633606163,
+      "loss": 1.3585,
+      "theoretical_loss": 3.506496041828433,
+      "tokens_seen": 1699545088
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004898491413898251,
+      "loss": 1.3166,
+      "theoretical_loss": 3.5064729560043864,
+      "tokens_seen": 1699676160
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004898090194190339,
+      "loss": 1.3516,
+      "theoretical_loss": 3.5064498724589903,
+      "tokens_seen": 1699807232
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004897688974482427,
+      "loss": 1.3948,
+      "theoretical_loss": 3.5064267911918434,
+      "tokens_seen": 1699938304
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004897287754774515,
+      "loss": 1.3142,
+      "theoretical_loss": 3.506403712202546,
+      "tokens_seen": 1700069376
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004896886535066603,
+      "loss": 1.343,
+      "theoretical_loss": 3.5063806354906975,
+      "tokens_seen": 1700200448
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004896485315358691,
+      "loss": 1.346,
+      "theoretical_loss": 3.5063575610558964,
+      "tokens_seen": 1700331520
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004896084095650778,
+      "loss": 1.3024,
+      "theoretical_loss": 3.5063344888977444,
+      "tokens_seen": 1700462592
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004895682875942866,
+      "loss": 1.3599,
+      "theoretical_loss": 3.506311419015841,
+      "tokens_seen": 1700593664
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004895281656234954,
+      "loss": 1.4807,
+      "theoretical_loss": 3.506288351409785,
+      "tokens_seen": 1700724736
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004894880436527042,
+      "loss": 1.335,
+      "theoretical_loss": 3.506265286079177,
+      "tokens_seen": 1700855808
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000489447921681913,
+      "loss": 1.3381,
+      "theoretical_loss": 3.5062422230236185,
+      "tokens_seen": 1700986880
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004894077997111218,
+      "loss": 1.5128,
+      "theoretical_loss": 3.5062191622427084,
+      "tokens_seen": 1701117952
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004893676777403306,
+      "loss": 1.4283,
+      "theoretical_loss": 3.5061961037360474,
+      "tokens_seen": 1701249024
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004893275557695395,
+      "loss": 1.4178,
+      "theoretical_loss": 3.5061730475032364,
+      "tokens_seen": 1701380096
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004892874337987481,
+      "loss": 1.353,
+      "theoretical_loss": 3.506149993543876,
+      "tokens_seen": 1701511168
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000489247311827957,
+      "loss": 1.4015,
+      "theoretical_loss": 3.5061269418575662,
+      "tokens_seen": 1701642240
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004892071898571658,
+      "loss": 1.3166,
+      "theoretical_loss": 3.506103892443909,
+      "tokens_seen": 1701773312
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004891670678863746,
+      "loss": 1.3793,
+      "theoretical_loss": 3.5060808453025043,
+      "tokens_seen": 1701904384
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004891269459155834,
+      "loss": 1.4553,
+      "theoretical_loss": 3.506057800432953,
+      "tokens_seen": 1702035456
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004890868239447922,
+      "loss": 1.4088,
+      "theoretical_loss": 3.5060347578348576,
+      "tokens_seen": 1702166528
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000489046701974001,
+      "loss": 1.3492,
+      "theoretical_loss": 3.506011717507818,
+      "tokens_seen": 1702297600
+    },
+    {
+      "epoch": 0.03,
+      "objective/train/advantage_avg": 0.49052882194519043,
+      "objective/train/docs_used": 621885,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2854690551757812,
+      "objective/train/original_loss": 1.2854689359664917,
+      "objective/train/theoretical_loss": 3.5059886794514363,
+      "objective/train/tokens_used": 72887776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2413872331380844,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050279974937439,
+      "objective/train/weighted_lm_loss": 1.3500052690505981,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.7788007855415344,
+      "theoretical_loss": 3.5059886794514363,
+      "tokens_seen": 1702428672
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004890065800032098,
+      "loss": 1.3614,
+      "theoretical_loss": 3.5059886794514363,
+      "tokens_seen": 1702428672
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004889664580324186,
+      "loss": 1.3057,
+      "theoretical_loss": 3.5059656436653137,
+      "tokens_seen": 1702559744
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004889263360616273,
+      "loss": 1.3432,
+      "theoretical_loss": 3.5059426101490514,
+      "tokens_seen": 1702690816
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004888862140908362,
+      "loss": 1.4112,
+      "theoretical_loss": 3.505919578902251,
+      "tokens_seen": 1702821888
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004888460921200449,
+      "loss": 1.4196,
+      "theoretical_loss": 3.5058965499245147,
+      "tokens_seen": 1702952960
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004888059701492537,
+      "loss": 1.4086,
+      "theoretical_loss": 3.505873523215444,
+      "tokens_seen": 1703084032
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004887658481784625,
+      "loss": 1.3893,
+      "theoretical_loss": 3.5058504987746413,
+      "tokens_seen": 1703215104
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004887257262076713,
+      "loss": 1.3835,
+      "theoretical_loss": 3.5058274766017083,
+      "tokens_seen": 1703346176
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004886856042368801,
+      "loss": 1.3415,
+      "theoretical_loss": 3.505804456696247,
+      "tokens_seen": 1703477248
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004886454822660889,
+      "loss": 1.4897,
+      "theoretical_loss": 3.5057814390578597,
+      "tokens_seen": 1703608320
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004886053602952977,
+      "loss": 1.4454,
+      "theoretical_loss": 3.5057584236861485,
+      "tokens_seen": 1703739392
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004885652383245064,
+      "loss": 1.3234,
+      "theoretical_loss": 3.5057354105807166,
+      "tokens_seen": 1703870464
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004885251163537154,
+      "loss": 1.3479,
+      "theoretical_loss": 3.5057123997411663,
+      "tokens_seen": 1704001536
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004884849943829241,
+      "loss": 1.3406,
+      "theoretical_loss": 3.5056893911670994,
+      "tokens_seen": 1704132608
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004884448724121329,
+      "loss": 1.4138,
+      "theoretical_loss": 3.5056663848581193,
+      "tokens_seen": 1704263680
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004884047504413417,
+      "loss": 1.4841,
+      "theoretical_loss": 3.5056433808138294,
+      "tokens_seen": 1704394752
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004883646284705505,
+      "loss": 1.355,
+      "theoretical_loss": 3.5056203790338314,
+      "tokens_seen": 1704525824
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004883245064997593,
+      "loss": 1.3701,
+      "theoretical_loss": 3.505597379517729,
+      "tokens_seen": 1704656896
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004882843845289681,
+      "loss": 1.4738,
+      "theoretical_loss": 3.5055743822651255,
+      "tokens_seen": 1704787968
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004882442625581769,
+      "loss": 1.3874,
+      "theoretical_loss": 3.5055513872756237,
+      "tokens_seen": 1704919040
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00048820414058738565,
+      "loss": 1.4192,
+      "theoretical_loss": 3.5055283945488274,
+      "tokens_seen": 1705050112
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00048816401861659446,
+      "loss": 1.3632,
+      "theoretical_loss": 3.5055054040843396,
+      "tokens_seen": 1705181184
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00048812389664580326,
+      "loss": 1.4627,
+      "theoretical_loss": 3.505482415881764,
+      "tokens_seen": 1705312256
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000488083774675012,
+      "loss": 1.3819,
+      "theoretical_loss": 3.505459429940704,
+      "tokens_seen": 1705443328
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004880436527042209,
+      "loss": 1.3774,
+      "theoretical_loss": 3.505436446260764,
+      "tokens_seen": 1705574400
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004880035307334296,
+      "loss": 1.3677,
+      "theoretical_loss": 3.5054134648415474,
+      "tokens_seen": 1705705472
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00048796340876263843,
+      "loss": 1.3999,
+      "theoretical_loss": 3.505390485682658,
+      "tokens_seen": 1705836544
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00048792328679184724,
+      "loss": 1.4424,
+      "theoretical_loss": 3.5053675087836997,
+      "tokens_seen": 1705967616
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00048788316482105605,
+      "loss": 1.3888,
+      "theoretical_loss": 3.5053445341442773,
+      "tokens_seen": 1706098688
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004878430428502648,
+      "loss": 1.3541,
+      "theoretical_loss": 3.505321561763995,
+      "tokens_seen": 1706229760
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004878029208794736,
+      "loss": 1.3888,
+      "theoretical_loss": 3.505298591642456,
+      "tokens_seen": 1706360832
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004877627989086824,
+      "loss": 1.2804,
+      "theoretical_loss": 3.5052756237792657,
+      "tokens_seen": 1706491904
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00048772267693789116,
+      "loss": 1.4138,
+      "theoretical_loss": 3.5052526581740286,
+      "tokens_seen": 1706622976
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004876825549671,
+      "loss": 1.4446,
+      "theoretical_loss": 3.5052296948263493,
+      "tokens_seen": 1706754048
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004876424329963088,
+      "loss": 1.4425,
+      "theoretical_loss": 3.5052067337358324,
+      "tokens_seen": 1706885120
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00048760231102551764,
+      "loss": 1.4045,
+      "theoretical_loss": 3.5051837749020827,
+      "tokens_seen": 1707016192
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0004875621890547264,
+      "loss": 1.3042,
+      "theoretical_loss": 3.5051608183247054,
+      "tokens_seen": 1707147264
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00048752206708393514,
+      "loss": 1.3233,
+      "theoretical_loss": 3.505137864003305,
+      "tokens_seen": 1707278336
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000487481945113144,
+      "loss": 1.3754,
+      "theoretical_loss": 3.505114911937487,
+      "tokens_seen": 1707409408
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00048744182314235275,
+      "loss": 1.3874,
+      "theoretical_loss": 3.5050919621268566,
+      "tokens_seen": 1707540480
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00048740170117156156,
+      "loss": 1.433,
+      "theoretical_loss": 3.5050690145710193,
+      "tokens_seen": 1707671552
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048736157920077036,
+      "loss": 1.3168,
+      "theoretical_loss": 3.5050460692695804,
+      "tokens_seen": 1707802624
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048732145722997917,
+      "loss": 1.3977,
+      "theoretical_loss": 3.505023126222145,
+      "tokens_seen": 1707933696
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004872813352591879,
+      "loss": 1.3055,
+      "theoretical_loss": 3.505000185428319,
+      "tokens_seen": 1708064768
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048724121328839673,
+      "loss": 1.3294,
+      "theoretical_loss": 3.5049772468877087,
+      "tokens_seen": 1708195840
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048720109131760554,
+      "loss": 1.4196,
+      "theoretical_loss": 3.504954310599919,
+      "tokens_seen": 1708326912
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004871609693468143,
+      "loss": 1.4162,
+      "theoretical_loss": 3.5049313765645564,
+      "tokens_seen": 1708457984
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048712084737602315,
+      "loss": 1.3132,
+      "theoretical_loss": 3.504908444781227,
+      "tokens_seen": 1708589056
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004870807254052319,
+      "loss": 1.3592,
+      "theoretical_loss": 3.5048855152495366,
+      "tokens_seen": 1708720128
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004870406034344407,
+      "loss": 1.3633,
+      "theoretical_loss": 3.5048625879690913,
+      "tokens_seen": 1708851200
+    },
+    {
+      "epoch": 0.04,
+      "objective/train/advantage_avg": 0.4949110746383667,
+      "objective/train/docs_used": 624223,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3441896438598633,
+      "objective/train/original_loss": 1.3441895246505737,
+      "objective/train/theoretical_loss": 3.5048396629394976,
+      "objective/train/tokens_used": 79441376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24522891640663147,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0507378578186035,
+      "objective/train/weighted_lm_loss": 1.4123423099517822,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.5048396629394976,
+      "tokens_seen": 1708982272
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004870004814636495,
+      "loss": 1.3507,
+      "theoretical_loss": 3.5048396629394976,
+      "tokens_seen": 1708982272
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048696035949285827,
+      "loss": 1.3123,
+      "theoretical_loss": 3.5048167401603623,
+      "tokens_seen": 1709113344
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004869202375220671,
+      "loss": 1.3463,
+      "theoretical_loss": 3.504793819631292,
+      "tokens_seen": 1709244416
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004868801155512759,
+      "loss": 1.354,
+      "theoretical_loss": 3.5047709013518924,
+      "tokens_seen": 1709375488
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004868399935804847,
+      "loss": 1.3742,
+      "theoretical_loss": 3.5047479853217705,
+      "tokens_seen": 1709506560
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004867998716096935,
+      "loss": 1.32,
+      "theoretical_loss": 3.5047250715405336,
+      "tokens_seen": 1709637632
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004867597496389023,
+      "loss": 1.403,
+      "theoretical_loss": 3.5047021600077883,
+      "tokens_seen": 1709768704
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048671962766811105,
+      "loss": 1.3924,
+      "theoretical_loss": 3.5046792507231417,
+      "tokens_seen": 1709899776
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048667950569731986,
+      "loss": 1.4011,
+      "theoretical_loss": 3.504656343686201,
+      "tokens_seen": 1710030848
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048663938372652866,
+      "loss": 1.4582,
+      "theoretical_loss": 3.504633438896573,
+      "tokens_seen": 1710161920
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004865992617557374,
+      "loss": 1.3964,
+      "theoretical_loss": 3.5046105363538658,
+      "tokens_seen": 1710292992
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004865591397849463,
+      "loss": 1.4266,
+      "theoretical_loss": 3.5045876360576855,
+      "tokens_seen": 1710424064
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000486519017814155,
+      "loss": 1.4188,
+      "theoretical_loss": 3.504564738007641,
+      "tokens_seen": 1710555136
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048647889584336383,
+      "loss": 1.4181,
+      "theoretical_loss": 3.5045418422033396,
+      "tokens_seen": 1710686208
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048643877387257264,
+      "loss": 1.4143,
+      "theoretical_loss": 3.5045189486443884,
+      "tokens_seen": 1710817280
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048639865190178145,
+      "loss": 1.3471,
+      "theoretical_loss": 3.5044960573303956,
+      "tokens_seen": 1710948352
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004863585299309902,
+      "loss": 1.416,
+      "theoretical_loss": 3.5044731682609687,
+      "tokens_seen": 1711079424
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000486318407960199,
+      "loss": 1.3342,
+      "theoretical_loss": 3.504450281435717,
+      "tokens_seen": 1711210496
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004862782859894078,
+      "loss": 1.3177,
+      "theoretical_loss": 3.5044273968542465,
+      "tokens_seen": 1711341568
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004862381640186166,
+      "loss": 1.3781,
+      "theoretical_loss": 3.504404514516167,
+      "tokens_seen": 1711472640
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004861980420478254,
+      "loss": 1.3511,
+      "theoretical_loss": 3.5043816344210863,
+      "tokens_seen": 1711603712
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004861579200770342,
+      "loss": 1.4424,
+      "theoretical_loss": 3.504358756568613,
+      "tokens_seen": 1711734784
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048611779810624304,
+      "loss": 1.4104,
+      "theoretical_loss": 3.5043358809583554,
+      "tokens_seen": 1711865856
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004860776761354518,
+      "loss": 1.4438,
+      "theoretical_loss": 3.504313007589922,
+      "tokens_seen": 1711996928
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048603755416466054,
+      "loss": 1.3842,
+      "theoretical_loss": 3.5042901364629215,
+      "tokens_seen": 1712128000
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004859974321938694,
+      "loss": 1.3769,
+      "theoretical_loss": 3.504267267576963,
+      "tokens_seen": 1712259072
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048595731022307815,
+      "loss": 1.3582,
+      "theoretical_loss": 3.504244400931655,
+      "tokens_seen": 1712390144
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048591718825228696,
+      "loss": 1.3545,
+      "theoretical_loss": 3.5042215365266065,
+      "tokens_seen": 1712521216
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048587706628149576,
+      "loss": 1.3112,
+      "theoretical_loss": 3.504198674361427,
+      "tokens_seen": 1712652288
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048583694431070457,
+      "loss": 1.3648,
+      "theoretical_loss": 3.504175814435725,
+      "tokens_seen": 1712783360
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004857968223399133,
+      "loss": 1.3526,
+      "theoretical_loss": 3.504152956749111,
+      "tokens_seen": 1712914432
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048575670036912213,
+      "loss": 1.3206,
+      "theoretical_loss": 3.504130101301193,
+      "tokens_seen": 1713045504
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048571657839833094,
+      "loss": 1.3754,
+      "theoretical_loss": 3.5041072480915814,
+      "tokens_seen": 1713176576
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004856764564275397,
+      "loss": 1.4286,
+      "theoretical_loss": 3.504084397119885,
+      "tokens_seen": 1713307648
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048563633445674855,
+      "loss": 1.4068,
+      "theoretical_loss": 3.504061548385714,
+      "tokens_seen": 1713438720
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004855962124859573,
+      "loss": 1.3297,
+      "theoretical_loss": 3.5040387018886783,
+      "tokens_seen": 1713569792
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048555609051516616,
+      "loss": 1.3759,
+      "theoretical_loss": 3.504015857628387,
+      "tokens_seen": 1713700864
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004855159685443749,
+      "loss": 1.4079,
+      "theoretical_loss": 3.5039930156044514,
+      "tokens_seen": 1713831936
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048547584657358366,
+      "loss": 1.4472,
+      "theoretical_loss": 3.5039701758164803,
+      "tokens_seen": 1713963008
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004854357246027925,
+      "loss": 1.4433,
+      "theoretical_loss": 3.503947338264084,
+      "tokens_seen": 1714094080
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004853956026320013,
+      "loss": 1.3467,
+      "theoretical_loss": 3.503924502946873,
+      "tokens_seen": 1714225152
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004853554806612101,
+      "loss": 1.3861,
+      "theoretical_loss": 3.503901669864458,
+      "tokens_seen": 1714356224
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004853153586904189,
+      "loss": 1.3619,
+      "theoretical_loss": 3.5038788390164495,
+      "tokens_seen": 1714487296
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004852752367196277,
+      "loss": 1.403,
+      "theoretical_loss": 3.5038560104024574,
+      "tokens_seen": 1714618368
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048523511474883645,
+      "loss": 1.4212,
+      "theoretical_loss": 3.503833184022093,
+      "tokens_seen": 1714749440
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048519499277804525,
+      "loss": 1.4686,
+      "theoretical_loss": 3.5038103598749664,
+      "tokens_seen": 1714880512
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048515487080725406,
+      "loss": 1.3193,
+      "theoretical_loss": 3.5037875379606884,
+      "tokens_seen": 1715011584
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004851147488364628,
+      "loss": 1.3895,
+      "theoretical_loss": 3.503764718278871,
+      "tokens_seen": 1715142656
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004850746268656717,
+      "loss": 1.4159,
+      "theoretical_loss": 3.503741900829124,
+      "tokens_seen": 1715273728
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004850345048948804,
+      "loss": 1.4651,
+      "theoretical_loss": 3.5037190856110594,
+      "tokens_seen": 1715404800
+    },
+    {
+      "epoch": 0.04,
+      "objective/train/advantage_avg": 0.49151313304901123,
+      "objective/train/docs_used": 626732,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2593648433685303,
+      "objective/train/original_loss": 1.2593647241592407,
+      "objective/train/theoretical_loss": 3.503696272624288,
+      "objective/train/tokens_used": 85994976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24228765070438385,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503829717636108,
+      "objective/train/weighted_lm_loss": 1.3227307796478271,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9048373699188232,
+      "theoretical_loss": 3.503696272624288,
+      "tokens_seen": 1715535872
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004849943829240893,
+      "loss": 1.3737,
+      "theoretical_loss": 3.503696272624288,
+      "tokens_seen": 1715535872
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048495426095329804,
+      "loss": 1.3679,
+      "theoretical_loss": 3.503673461868421,
+      "tokens_seen": 1715666944
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048491413898250684,
+      "loss": 1.4244,
+      "theoretical_loss": 3.5036506533430707,
+      "tokens_seen": 1715798016
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048487401701171565,
+      "loss": 1.4526,
+      "theoretical_loss": 3.5036278470478477,
+      "tokens_seen": 1715929088
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004848338950409244,
+      "loss": 1.4118,
+      "theoretical_loss": 3.5036050429823637,
+      "tokens_seen": 1716060160
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004847937730701332,
+      "loss": 1.4924,
+      "theoretical_loss": 3.5035822411462307,
+      "tokens_seen": 1716191232
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000484753651099342,
+      "loss": 1.375,
+      "theoretical_loss": 3.5035594415390605,
+      "tokens_seen": 1716322304
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004847135291285508,
+      "loss": 1.3806,
+      "theoretical_loss": 3.503536644160465,
+      "tokens_seen": 1716453376
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004846734071577596,
+      "loss": 1.3953,
+      "theoretical_loss": 3.503513849010056,
+      "tokens_seen": 1716584448
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048463328518696843,
+      "loss": 1.4334,
+      "theoretical_loss": 3.5034910560874457,
+      "tokens_seen": 1716715520
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004845931632161772,
+      "loss": 1.3839,
+      "theoretical_loss": 3.5034682653922466,
+      "tokens_seen": 1716846592
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048455304124538594,
+      "loss": 1.4543,
+      "theoretical_loss": 3.5034454769240706,
+      "tokens_seen": 1716977664
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004845129192745948,
+      "loss": 1.3923,
+      "theoretical_loss": 3.5034226906825303,
+      "tokens_seen": 1717108736
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048447279730380355,
+      "loss": 1.4074,
+      "theoretical_loss": 3.5033999066672386,
+      "tokens_seen": 1717239808
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048443267533301236,
+      "loss": 1.4582,
+      "theoretical_loss": 3.503377124877807,
+      "tokens_seen": 1717370880
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048439255336222116,
+      "loss": 1.3336,
+      "theoretical_loss": 3.5033543453138494,
+      "tokens_seen": 1717501952
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048435243139142997,
+      "loss": 1.3966,
+      "theoretical_loss": 3.5033315679749784,
+      "tokens_seen": 1717633024
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004843123094206388,
+      "loss": 1.3846,
+      "theoretical_loss": 3.503308792860806,
+      "tokens_seen": 1717764096
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048427218744984753,
+      "loss": 1.4581,
+      "theoretical_loss": 3.503286019970946,
+      "tokens_seen": 1717895168
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048423206547905634,
+      "loss": 1.3864,
+      "theoretical_loss": 3.503263249305011,
+      "tokens_seen": 1718026240
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048419194350826514,
+      "loss": 1.3432,
+      "theoretical_loss": 3.5032404808626145,
+      "tokens_seen": 1718157312
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048415182153747395,
+      "loss": 1.4182,
+      "theoretical_loss": 3.5032177146433696,
+      "tokens_seen": 1718288384
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004841116995666827,
+      "loss": 1.3283,
+      "theoretical_loss": 3.5031949506468893,
+      "tokens_seen": 1718419456
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048407157759589156,
+      "loss": 1.4123,
+      "theoretical_loss": 3.503172188872788,
+      "tokens_seen": 1718550528
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004840314556251003,
+      "loss": 1.3581,
+      "theoretical_loss": 3.5031494293206786,
+      "tokens_seen": 1718681600
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048399133365430906,
+      "loss": 1.426,
+      "theoretical_loss": 3.503126671990175,
+      "tokens_seen": 1718812672
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004839512116835179,
+      "loss": 1.4592,
+      "theoretical_loss": 3.5031039168808906,
+      "tokens_seen": 1718943744
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004839110897127267,
+      "loss": 1.358,
+      "theoretical_loss": 3.50308116399244,
+      "tokens_seen": 1719074816
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004838709677419355,
+      "loss": 1.3032,
+      "theoretical_loss": 3.5030584133244362,
+      "tokens_seen": 1719205888
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004838308457711443,
+      "loss": 1.39,
+      "theoretical_loss": 3.503035664876494,
+      "tokens_seen": 1719336960
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004837907238003531,
+      "loss": 1.4449,
+      "theoretical_loss": 3.5030129186482277,
+      "tokens_seen": 1719468032
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004837506018295619,
+      "loss": 1.4471,
+      "theoretical_loss": 3.50299017463925,
+      "tokens_seen": 1719599104
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048371047985877065,
+      "loss": 1.3771,
+      "theoretical_loss": 3.502967432849177,
+      "tokens_seen": 1719730176
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048367035788797946,
+      "loss": 1.4727,
+      "theoretical_loss": 3.5029446932776223,
+      "tokens_seen": 1719861248
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048363023591718827,
+      "loss": 1.3861,
+      "theoretical_loss": 3.502921955924201,
+      "tokens_seen": 1719992320
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004835901139463971,
+      "loss": 1.4263,
+      "theoretical_loss": 3.502899220788527,
+      "tokens_seen": 1720123392
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004835499919756058,
+      "loss": 1.3646,
+      "theoretical_loss": 3.502876487870215,
+      "tokens_seen": 1720254464
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004835098700048147,
+      "loss": 1.401,
+      "theoretical_loss": 3.5028537571688805,
+      "tokens_seen": 1720385536
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048346974803402344,
+      "loss": 1.3677,
+      "theoretical_loss": 3.502831028684138,
+      "tokens_seen": 1720516608
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004834296260632322,
+      "loss": 1.3551,
+      "theoretical_loss": 3.5028083024156027,
+      "tokens_seen": 1720647680
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048338950409244105,
+      "loss": 1.4438,
+      "theoretical_loss": 3.5027855783628894,
+      "tokens_seen": 1720778752
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004833493821216498,
+      "loss": 1.3952,
+      "theoretical_loss": 3.5027628565256137,
+      "tokens_seen": 1720909824
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004833092601508586,
+      "loss": 1.4922,
+      "theoretical_loss": 3.50274013690339,
+      "tokens_seen": 1721040896
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004832691381800674,
+      "loss": 1.4844,
+      "theoretical_loss": 3.502717419495835,
+      "tokens_seen": 1721171968
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004832290162092762,
+      "loss": 1.518,
+      "theoretical_loss": 3.502694704302564,
+      "tokens_seen": 1721303040
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000483188894238485,
+      "loss": 1.4,
+      "theoretical_loss": 3.5026719913231914,
+      "tokens_seen": 1721434112
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048314877226769383,
+      "loss": 1.3203,
+      "theoretical_loss": 3.5026492805573337,
+      "tokens_seen": 1721565184
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004831086502969026,
+      "loss": 1.4647,
+      "theoretical_loss": 3.5026265720046066,
+      "tokens_seen": 1721696256
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004830685283261114,
+      "loss": 1.3258,
+      "theoretical_loss": 3.5026038656646263,
+      "tokens_seen": 1721827328
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004830284063553202,
+      "loss": 1.4015,
+      "theoretical_loss": 3.502581161537008,
+      "tokens_seen": 1721958400
+    },
+    {
+      "epoch": 0.04,
+      "objective/train/advantage_avg": 0.4925839602947235,
+      "objective/train/docs_used": 629016,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3220460414886475,
+      "objective/train/original_loss": 1.3220460414886475,
+      "objective/train/theoretical_loss": 3.502558459621369,
+      "objective/train/tokens_used": 92548576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2432505190372467,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504951477050781,
+      "objective/train/weighted_lm_loss": 1.38889479637146,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.502558459621369,
+      "tokens_seen": 1722089472
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048298828438452895,
+      "loss": 1.3917,
+      "theoretical_loss": 3.502558459621369,
+      "tokens_seen": 1722089472
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004829481624137378,
+      "loss": 1.3244,
+      "theoretical_loss": 3.5025357599173237,
+      "tokens_seen": 1722220544
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048290804044294656,
+      "loss": 1.4472,
+      "theoretical_loss": 3.50251306242449,
+      "tokens_seen": 1722351616
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048286791847215537,
+      "loss": 1.3493,
+      "theoretical_loss": 3.5024903671424834,
+      "tokens_seen": 1722482688
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004828277965013642,
+      "loss": 1.5627,
+      "theoretical_loss": 3.50246767407092,
+      "tokens_seen": 1722613760
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048278767453057293,
+      "loss": 1.4022,
+      "theoretical_loss": 3.5024449832094176,
+      "tokens_seen": 1722744832
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048274755255978173,
+      "loss": 1.501,
+      "theoretical_loss": 3.502422294557592,
+      "tokens_seen": 1722875904
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048270743058899054,
+      "loss": 1.3874,
+      "theoretical_loss": 3.50239960811506,
+      "tokens_seen": 1723006976
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048266730861819935,
+      "loss": 1.3936,
+      "theoretical_loss": 3.502376923881438,
+      "tokens_seen": 1723138048
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004826271866474081,
+      "loss": 1.4146,
+      "theoretical_loss": 3.5023542418563443,
+      "tokens_seen": 1723269120
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048258706467661696,
+      "loss": 1.4053,
+      "theoretical_loss": 3.5023315620393944,
+      "tokens_seen": 1723400192
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004825469427058257,
+      "loss": 1.3813,
+      "theoretical_loss": 3.502308884430206,
+      "tokens_seen": 1723531264
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048250682073503446,
+      "loss": 1.407,
+      "theoretical_loss": 3.5022862090283966,
+      "tokens_seen": 1723662336
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004824666987642433,
+      "loss": 1.4086,
+      "theoretical_loss": 3.502263535833583,
+      "tokens_seen": 1723793408
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004824265767934521,
+      "loss": 1.4098,
+      "theoretical_loss": 3.5022408648453833,
+      "tokens_seen": 1723924480
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00048238645482266094,
+      "loss": 1.3123,
+      "theoretical_loss": 3.5022181960634144,
+      "tokens_seen": 1724055552
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0004823463328518697,
+      "loss": 1.3539,
+      "theoretical_loss": 3.5021955294872944,
+      "tokens_seen": 1724186624
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004823062108810785,
+      "loss": 1.4709,
+      "theoretical_loss": 3.5021728651166404,
+      "tokens_seen": 1724317696
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004822660889102873,
+      "loss": 1.4136,
+      "theoretical_loss": 3.50215020295107,
+      "tokens_seen": 1724448768
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00048222596693949605,
+      "loss": 1.3738,
+      "theoretical_loss": 3.502127542990202,
+      "tokens_seen": 1724579840
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00048218584496870486,
+      "loss": 1.4433,
+      "theoretical_loss": 3.5021048852336536,
+      "tokens_seen": 1724710912
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00048214572299791367,
+      "loss": 1.4102,
+      "theoretical_loss": 3.5020822296810437,
+      "tokens_seen": 1724841984
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004821056010271225,
+      "loss": 1.4627,
+      "theoretical_loss": 3.5020595763319897,
+      "tokens_seen": 1724973056
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004820654790563312,
+      "loss": 1.4272,
+      "theoretical_loss": 3.50203692518611,
+      "tokens_seen": 1725104128
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004820253570855401,
+      "loss": 1.4197,
+      "theoretical_loss": 3.502014276243023,
+      "tokens_seen": 1725235200
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00048198523511474884,
+      "loss": 1.4336,
+      "theoretical_loss": 3.5019916295023474,
+      "tokens_seen": 1725366272
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004819451131439576,
+      "loss": 1.4075,
+      "theoretical_loss": 3.5019689849637015,
+      "tokens_seen": 1725497344
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00048190499117316645,
+      "loss": 1.3836,
+      "theoretical_loss": 3.501946342626704,
+      "tokens_seen": 1725628416
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004818648692023752,
+      "loss": 1.3642,
+      "theoretical_loss": 3.5019237024909735,
+      "tokens_seen": 1725759488
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00048182474723158406,
+      "loss": 1.4915,
+      "theoretical_loss": 3.5019010645561295,
+      "tokens_seen": 1725890560
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004817846252607928,
+      "loss": 1.4807,
+      "theoretical_loss": 3.5018784288217897,
+      "tokens_seen": 1726021632
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004817445032900016,
+      "loss": 1.4511,
+      "theoretical_loss": 3.501855795287574,
+      "tokens_seen": 1726152704
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00048170438131921043,
+      "loss": 1.3969,
+      "theoretical_loss": 3.5018331639531017,
+      "tokens_seen": 1726283776
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00048166425934841923,
+      "loss": 1.4158,
+      "theoretical_loss": 3.5018105348179915,
+      "tokens_seen": 1726414848
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000481624137377628,
+      "loss": 1.3511,
+      "theoretical_loss": 3.5017879078818623,
+      "tokens_seen": 1726545920
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004815840154068368,
+      "loss": 1.5111,
+      "theoretical_loss": 3.501765283144335,
+      "tokens_seen": 1726676992
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004815438934360456,
+      "loss": 1.3697,
+      "theoretical_loss": 3.5017426606050277,
+      "tokens_seen": 1726808064
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00048150377146525435,
+      "loss": 1.3282,
+      "theoretical_loss": 3.50172004026356,
+      "tokens_seen": 1726939136
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004814636494944632,
+      "loss": 1.42,
+      "theoretical_loss": 3.5016974221195527,
+      "tokens_seen": 1727070208
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00048142352752367196,
+      "loss": 1.3163,
+      "theoretical_loss": 3.501674806172625,
+      "tokens_seen": 1727201280
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00048138340555288077,
+      "loss": 1.4558,
+      "theoretical_loss": 3.5016521924223962,
+      "tokens_seen": 1727332352
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004813432835820896,
+      "loss": 1.4546,
+      "theoretical_loss": 3.501629580868487,
+      "tokens_seen": 1727463424
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00048130316161129833,
+      "loss": 1.3761,
+      "theoretical_loss": 3.501606971510517,
+      "tokens_seen": 1727594496
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00048126303964050713,
+      "loss": 1.3958,
+      "theoretical_loss": 3.501584364348107,
+      "tokens_seen": 1727725568
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00048122291766971594,
+      "loss": 1.3872,
+      "theoretical_loss": 3.5015617593808765,
+      "tokens_seen": 1727856640
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00048118279569892475,
+      "loss": 1.4534,
+      "theoretical_loss": 3.5015391566084464,
+      "tokens_seen": 1727987712
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00048114267372813355,
+      "loss": 1.4246,
+      "theoretical_loss": 3.5015165560304364,
+      "tokens_seen": 1728118784
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00048110255175734236,
+      "loss": 1.3759,
+      "theoretical_loss": 3.5014939576464683,
+      "tokens_seen": 1728249856
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004810624297865511,
+      "loss": 1.3462,
+      "theoretical_loss": 3.5014713614561614,
+      "tokens_seen": 1728380928
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004810223078157599,
+      "loss": 1.3743,
+      "theoretical_loss": 3.5014487674591375,
+      "tokens_seen": 1728512000
+    },
+    {
+      "epoch": 0.05,
+      "objective/train/advantage_avg": 0.49412211775779724,
+      "objective/train/docs_used": 631392,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3023947477340698,
+      "objective/train/original_loss": 1.3023947477340698,
+      "objective/train/theoretical_loss": 3.5014261756550167,
+      "objective/train/tokens_used": 99102176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24453096091747284,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506553649902344,
+      "objective/train/weighted_lm_loss": 1.3682907819747925,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.5014261756550167,
+      "tokens_seen": 1728643072
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004809821858449687,
+      "loss": 1.387,
+      "theoretical_loss": 3.5014261756550167,
+      "tokens_seen": 1728643072
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004809420638741775,
+      "loss": 1.433,
+      "theoretical_loss": 3.50140358604342,
+      "tokens_seen": 1728774144
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00048090194190338634,
+      "loss": 1.412,
+      "theoretical_loss": 3.5013809986239686,
+      "tokens_seen": 1728905216
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004808618199325951,
+      "loss": 1.3311,
+      "theoretical_loss": 3.501358413396284,
+      "tokens_seen": 1729036288
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004808216979618039,
+      "loss": 1.3396,
+      "theoretical_loss": 3.5013358303599866,
+      "tokens_seen": 1729167360
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004807815759910127,
+      "loss": 1.4216,
+      "theoretical_loss": 3.5013132495146984,
+      "tokens_seen": 1729298432
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00048074145402022145,
+      "loss": 1.4348,
+      "theoretical_loss": 3.5012906708600395,
+      "tokens_seen": 1729429504
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00048070133204943026,
+      "loss": 1.3673,
+      "theoretical_loss": 3.501268094395633,
+      "tokens_seen": 1729560576
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00048066121007863907,
+      "loss": 1.461,
+      "theoretical_loss": 3.5012455201211,
+      "tokens_seen": 1729691648
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00048062108810784787,
+      "loss": 1.421,
+      "theoretical_loss": 3.5012229480360615,
+      "tokens_seen": 1729822720
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004805809661370567,
+      "loss": 1.3532,
+      "theoretical_loss": 3.50120037814014,
+      "tokens_seen": 1729953792
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004805408441662655,
+      "loss": 1.3785,
+      "theoretical_loss": 3.501177810432957,
+      "tokens_seen": 1730084864
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00048050072219547424,
+      "loss": 1.5111,
+      "theoretical_loss": 3.501155244914134,
+      "tokens_seen": 1730215936
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00048046060022468304,
+      "loss": 1.3553,
+      "theoretical_loss": 3.5011326815832944,
+      "tokens_seen": 1730347008
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00048042047825389185,
+      "loss": 1.3683,
+      "theoretical_loss": 3.5011101204400594,
+      "tokens_seen": 1730478080
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004803803562831006,
+      "loss": 1.3687,
+      "theoretical_loss": 3.501087561484051,
+      "tokens_seen": 1730609152
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00048034023431230946,
+      "loss": 1.4292,
+      "theoretical_loss": 3.501065004714892,
+      "tokens_seen": 1730740224
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004803001123415182,
+      "loss": 1.322,
+      "theoretical_loss": 3.5010424501322044,
+      "tokens_seen": 1730871296
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000480259990370727,
+      "loss": 1.3532,
+      "theoretical_loss": 3.501019897735611,
+      "tokens_seen": 1731002368
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00048021986839993583,
+      "loss": 1.4111,
+      "theoretical_loss": 3.5009973475247347,
+      "tokens_seen": 1731133440
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00048017974642914463,
+      "loss": 1.4742,
+      "theoretical_loss": 3.5009747994991978,
+      "tokens_seen": 1731264512
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004801396244583534,
+      "loss": 1.4505,
+      "theoretical_loss": 3.5009522536586233,
+      "tokens_seen": 1731395584
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004800995024875622,
+      "loss": 1.3271,
+      "theoretical_loss": 3.5009297100026338,
+      "tokens_seen": 1731526656
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000480059380516771,
+      "loss": 1.32,
+      "theoretical_loss": 3.500907168530852,
+      "tokens_seen": 1731657728
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00048001925854597975,
+      "loss": 1.3814,
+      "theoretical_loss": 3.5008846292429023,
+      "tokens_seen": 1731788800
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004799791365751886,
+      "loss": 1.4206,
+      "theoretical_loss": 3.500862092138407,
+      "tokens_seen": 1731919872
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047993901460439736,
+      "loss": 1.4495,
+      "theoretical_loss": 3.5008395572169886,
+      "tokens_seen": 1732050944
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004798988926336062,
+      "loss": 1.3806,
+      "theoretical_loss": 3.5008170244782715,
+      "tokens_seen": 1732182016
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000479858770662815,
+      "loss": 1.4287,
+      "theoretical_loss": 3.500794493921879,
+      "tokens_seen": 1732313088
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047981864869202373,
+      "loss": 1.4619,
+      "theoretical_loss": 3.5007719655474343,
+      "tokens_seen": 1732444160
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004797785267212326,
+      "loss": 1.311,
+      "theoretical_loss": 3.5007494393545615,
+      "tokens_seen": 1732575232
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047973840475044134,
+      "loss": 1.3706,
+      "theoretical_loss": 3.5007269153428835,
+      "tokens_seen": 1732706304
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047969828277965015,
+      "loss": 1.4103,
+      "theoretical_loss": 3.5007043935120254,
+      "tokens_seen": 1732837376
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047965816080885895,
+      "loss": 1.4389,
+      "theoretical_loss": 3.50068187386161,
+      "tokens_seen": 1732968448
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047961803883806776,
+      "loss": 1.4495,
+      "theoretical_loss": 3.500659356391262,
+      "tokens_seen": 1733099520
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004795779168672765,
+      "loss": 1.4116,
+      "theoretical_loss": 3.5006368411006052,
+      "tokens_seen": 1733230592
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004795377948964853,
+      "loss": 1.4194,
+      "theoretical_loss": 3.5006143279892634,
+      "tokens_seen": 1733361664
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004794976729256941,
+      "loss": 1.4449,
+      "theoretical_loss": 3.5005918170568613,
+      "tokens_seen": 1733492736
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004794575509549029,
+      "loss": 1.4123,
+      "theoretical_loss": 3.5005693083030236,
+      "tokens_seen": 1733623808
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047941742898411174,
+      "loss": 1.4532,
+      "theoretical_loss": 3.500546801727374,
+      "tokens_seen": 1733754880
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004793773070133205,
+      "loss": 1.377,
+      "theoretical_loss": 3.5005242973295383,
+      "tokens_seen": 1733885952
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047933718504252935,
+      "loss": 1.4296,
+      "theoretical_loss": 3.50050179510914,
+      "tokens_seen": 1734017024
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004792970630717381,
+      "loss": 1.444,
+      "theoretical_loss": 3.500479295065804,
+      "tokens_seen": 1734148096
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047925694110094685,
+      "loss": 1.373,
+      "theoretical_loss": 3.500456797199156,
+      "tokens_seen": 1734279168
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004792168191301557,
+      "loss": 1.3896,
+      "theoretical_loss": 3.50043430150882,
+      "tokens_seen": 1734410240
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047917669715936447,
+      "loss": 1.3758,
+      "theoretical_loss": 3.5004118079944213,
+      "tokens_seen": 1734541312
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047913657518857327,
+      "loss": 1.4271,
+      "theoretical_loss": 3.5003893166555846,
+      "tokens_seen": 1734672384
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004790964532177821,
+      "loss": 1.4959,
+      "theoretical_loss": 3.500366827491936,
+      "tokens_seen": 1734803456
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004790563312469909,
+      "loss": 1.4317,
+      "theoretical_loss": 3.5003443405031005,
+      "tokens_seen": 1734934528
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047901620927619964,
+      "loss": 1.3889,
+      "theoretical_loss": 3.5003218556887035,
+      "tokens_seen": 1735065600
+    },
+    {
+      "epoch": 0.05,
+      "objective/train/advantage_avg": 0.49361297488212585,
+      "objective/train/docs_used": 633815,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.361792802810669,
+      "objective/train/original_loss": 1.3617925643920898,
+      "objective/train/theoretical_loss": 3.50029937304837,
+      "objective/train/tokens_used": 105655776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24423593282699585,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506030321121216,
+      "objective/train/weighted_lm_loss": 1.4306267499923706,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9048373699188232,
+      "theoretical_loss": 3.50029937304837,
+      "tokens_seen": 1735196672
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047897608730540844,
+      "loss": 1.4218,
+      "theoretical_loss": 3.50029937304837,
+      "tokens_seen": 1735196672
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047893596533461725,
+      "loss": 1.406,
+      "theoretical_loss": 3.5002768925817263,
+      "tokens_seen": 1735327744
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000478895843363826,
+      "loss": 1.3574,
+      "theoretical_loss": 3.5002544142883982,
+      "tokens_seen": 1735458816
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047885572139303486,
+      "loss": 1.3608,
+      "theoretical_loss": 3.5002319381680103,
+      "tokens_seen": 1735589888
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004788155994222436,
+      "loss": 1.4567,
+      "theoretical_loss": 3.5002094642201897,
+      "tokens_seen": 1735720960
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004787754774514524,
+      "loss": 1.3214,
+      "theoretical_loss": 3.500186992444562,
+      "tokens_seen": 1735852032
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004787353554806612,
+      "loss": 1.3948,
+      "theoretical_loss": 3.5001645228407527,
+      "tokens_seen": 1735983104
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047869523350987,
+      "loss": 1.4158,
+      "theoretical_loss": 3.5001420554083884,
+      "tokens_seen": 1736114176
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047865511153907884,
+      "loss": 1.3588,
+      "theoretical_loss": 3.500119590147096,
+      "tokens_seen": 1736245248
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004786149895682876,
+      "loss": 1.4953,
+      "theoretical_loss": 3.5000971270565007,
+      "tokens_seen": 1736376320
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004785748675974964,
+      "loss": 1.4121,
+      "theoretical_loss": 3.5000746661362294,
+      "tokens_seen": 1736507392
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004785347456267052,
+      "loss": 1.4259,
+      "theoretical_loss": 3.5000522073859086,
+      "tokens_seen": 1736638464
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000478494623655914,
+      "loss": 1.4549,
+      "theoretical_loss": 3.5000297508051643,
+      "tokens_seen": 1736769536
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047845450168512276,
+      "loss": 1.3305,
+      "theoretical_loss": 3.5000072963936244,
+      "tokens_seen": 1736900608
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004784143797143316,
+      "loss": 1.4002,
+      "theoretical_loss": 3.4999848441509154,
+      "tokens_seen": 1737031680
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004783742577435404,
+      "loss": 1.4674,
+      "theoretical_loss": 3.4999623940766638,
+      "tokens_seen": 1737162752
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047833413577274913,
+      "loss": 1.4144,
+      "theoretical_loss": 3.499939946170496,
+      "tokens_seen": 1737293824
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000478294013801958,
+      "loss": 1.4431,
+      "theoretical_loss": 3.49991750043204,
+      "tokens_seen": 1737424896
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047825389183116674,
+      "loss": 1.361,
+      "theoretical_loss": 3.4998950568609235,
+      "tokens_seen": 1737555968
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047821376986037555,
+      "loss": 1.4464,
+      "theoretical_loss": 3.499872615456772,
+      "tokens_seen": 1737687040
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047817364788958435,
+      "loss": 1.3413,
+      "theoretical_loss": 3.4998501762192142,
+      "tokens_seen": 1737818112
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047813352591879316,
+      "loss": 1.4199,
+      "theoretical_loss": 3.499827739147877,
+      "tokens_seen": 1737949184
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004780934039480019,
+      "loss": 1.304,
+      "theoretical_loss": 3.499805304242388,
+      "tokens_seen": 1738080256
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004780532819772107,
+      "loss": 1.5515,
+      "theoretical_loss": 3.4997828715023753,
+      "tokens_seen": 1738211328
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004780131600064195,
+      "loss": 1.3821,
+      "theoretical_loss": 3.4997604409274663,
+      "tokens_seen": 1738342400
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047797303803562833,
+      "loss": 1.3868,
+      "theoretical_loss": 3.4997380125172883,
+      "tokens_seen": 1738473472
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047793291606483714,
+      "loss": 1.418,
+      "theoretical_loss": 3.4997155862714697,
+      "tokens_seen": 1738604544
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004778927940940459,
+      "loss": 1.3233,
+      "theoretical_loss": 3.4996931621896383,
+      "tokens_seen": 1738735616
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047785267212325475,
+      "loss": 1.393,
+      "theoretical_loss": 3.4996707402714224,
+      "tokens_seen": 1738866688
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004778125501524635,
+      "loss": 1.4033,
+      "theoretical_loss": 3.4996483205164504,
+      "tokens_seen": 1738997760
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047777242818167225,
+      "loss": 1.465,
+      "theoretical_loss": 3.4996259029243495,
+      "tokens_seen": 1739128832
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004777323062108811,
+      "loss": 1.3677,
+      "theoretical_loss": 3.4996034874947495,
+      "tokens_seen": 1739259904
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047769218424008987,
+      "loss": 1.3856,
+      "theoretical_loss": 3.4995810742272777,
+      "tokens_seen": 1739390976
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047765206226929867,
+      "loss": 1.4257,
+      "theoretical_loss": 3.4995586631215634,
+      "tokens_seen": 1739522048
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004776119402985075,
+      "loss": 1.3375,
+      "theoretical_loss": 3.499536254177235,
+      "tokens_seen": 1739653120
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004775718183277163,
+      "loss": 1.4334,
+      "theoretical_loss": 3.4995138473939207,
+      "tokens_seen": 1739784192
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047753169635692504,
+      "loss": 1.3819,
+      "theoretical_loss": 3.49949144277125,
+      "tokens_seen": 1739915264
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047749157438613384,
+      "loss": 1.4338,
+      "theoretical_loss": 3.4994690403088518,
+      "tokens_seen": 1740046336
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047745145241534265,
+      "loss": 1.478,
+      "theoretical_loss": 3.4994466400063544,
+      "tokens_seen": 1740177408
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004774113304445514,
+      "loss": 1.3336,
+      "theoretical_loss": 3.499424241863388,
+      "tokens_seen": 1740308480
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00047737120847376026,
+      "loss": 1.4062,
+      "theoretical_loss": 3.499401845879581,
+      "tokens_seen": 1740439552
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000477331086502969,
+      "loss": 1.3962,
+      "theoretical_loss": 3.4993794520545625,
+      "tokens_seen": 1740570624
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0004772909645321779,
+      "loss": 1.3908,
+      "theoretical_loss": 3.4993570603879625,
+      "tokens_seen": 1740701696
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004772508425613866,
+      "loss": 1.4763,
+      "theoretical_loss": 3.49933467087941,
+      "tokens_seen": 1740832768
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004772107205905954,
+      "loss": 1.4323,
+      "theoretical_loss": 3.4993122835285346,
+      "tokens_seen": 1740963840
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047717059861980424,
+      "loss": 1.401,
+      "theoretical_loss": 3.4992898983349665,
+      "tokens_seen": 1741094912
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000477130476649013,
+      "loss": 1.3849,
+      "theoretical_loss": 3.499267515298335,
+      "tokens_seen": 1741225984
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004770903546782218,
+      "loss": 1.3533,
+      "theoretical_loss": 3.4992451344182696,
+      "tokens_seen": 1741357056
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004770502327074306,
+      "loss": 1.3774,
+      "theoretical_loss": 3.499222755694401,
+      "tokens_seen": 1741488128
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004770101107366394,
+      "loss": 1.3959,
+      "theoretical_loss": 3.499200379126358,
+      "tokens_seen": 1741619200
+    },
+    {
+      "epoch": 0.06,
+      "objective/train/advantage_avg": 0.49168017506599426,
+      "objective/train/docs_used": 636395,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2947736978530884,
+      "objective/train/original_loss": 1.2947735786437988,
+      "objective/train/theoretical_loss": 3.499178004713772,
+      "objective/train/tokens_used": 112209376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2425571233034134,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504010915756226,
+      "objective/train/weighted_lm_loss": 1.360120415687561,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.499178004713772,
+      "tokens_seen": 1741750272
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047696998876584816,
+      "loss": 1.3963,
+      "theoretical_loss": 3.499178004713772,
+      "tokens_seen": 1741750272
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000476929866795057,
+      "loss": 1.3416,
+      "theoretical_loss": 3.4991556324562723,
+      "tokens_seen": 1741881344
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004768897448242658,
+      "loss": 1.4068,
+      "theoretical_loss": 3.4991332623534896,
+      "tokens_seen": 1742012416
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004768496228534745,
+      "loss": 1.4136,
+      "theoretical_loss": 3.499110894405055,
+      "tokens_seen": 1742143488
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004768095008826834,
+      "loss": 1.349,
+      "theoretical_loss": 3.4990885286105975,
+      "tokens_seen": 1742274560
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047676937891189214,
+      "loss": 1.368,
+      "theoretical_loss": 3.499066164969748,
+      "tokens_seen": 1742405632
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000476729256941101,
+      "loss": 1.4452,
+      "theoretical_loss": 3.499043803482138,
+      "tokens_seen": 1742536704
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047668913497030975,
+      "loss": 1.3665,
+      "theoretical_loss": 3.4990214441473975,
+      "tokens_seen": 1742667776
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047664901299951856,
+      "loss": 1.4362,
+      "theoretical_loss": 3.4989990869651573,
+      "tokens_seen": 1742798848
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047660889102872736,
+      "loss": 1.3993,
+      "theoretical_loss": 3.498976731935049,
+      "tokens_seen": 1742929920
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004765687690579361,
+      "loss": 1.4327,
+      "theoretical_loss": 3.498954379056703,
+      "tokens_seen": 1743060992
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004765286470871449,
+      "loss": 1.46,
+      "theoretical_loss": 3.4989320283297505,
+      "tokens_seen": 1743192064
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047648852511635373,
+      "loss": 1.447,
+      "theoretical_loss": 3.498909679753823,
+      "tokens_seen": 1743323136
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047644840314556254,
+      "loss": 1.4799,
+      "theoretical_loss": 3.4988873333285513,
+      "tokens_seen": 1743454208
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004764082811747713,
+      "loss": 1.4211,
+      "theoretical_loss": 3.4988649890535672,
+      "tokens_seen": 1743585280
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047636815920398015,
+      "loss": 1.4373,
+      "theoretical_loss": 3.4988426469285017,
+      "tokens_seen": 1743716352
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004763280372331889,
+      "loss": 1.3581,
+      "theoretical_loss": 3.498820306952987,
+      "tokens_seen": 1743847424
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047628791526239765,
+      "loss": 1.4384,
+      "theoretical_loss": 3.498797969126654,
+      "tokens_seen": 1743978496
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004762477932916065,
+      "loss": 1.4286,
+      "theoretical_loss": 3.498775633449135,
+      "tokens_seen": 1744109568
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047620767132081527,
+      "loss": 1.4041,
+      "theoretical_loss": 3.498753299920062,
+      "tokens_seen": 1744240640
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047616754935002407,
+      "loss": 1.4122,
+      "theoretical_loss": 3.498730968539066,
+      "tokens_seen": 1744371712
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004761274273792329,
+      "loss": 1.3555,
+      "theoretical_loss": 3.49870863930578,
+      "tokens_seen": 1744502784
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004760873054084417,
+      "loss": 1.4106,
+      "theoretical_loss": 3.498686312219836,
+      "tokens_seen": 1744633856
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004760471834376505,
+      "loss": 1.4688,
+      "theoretical_loss": 3.498663987280865,
+      "tokens_seen": 1744764928
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047600706146685924,
+      "loss": 1.4378,
+      "theoretical_loss": 3.498641664488501,
+      "tokens_seen": 1744896000
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047596693949606805,
+      "loss": 1.2829,
+      "theoretical_loss": 3.4986193438423747,
+      "tokens_seen": 1745027072
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047592681752527686,
+      "loss": 1.3428,
+      "theoretical_loss": 3.4985970253421206,
+      "tokens_seen": 1745158144
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047588669555448566,
+      "loss": 1.4081,
+      "theoretical_loss": 3.498574708987369,
+      "tokens_seen": 1745289216
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004758465735836944,
+      "loss": 1.3981,
+      "theoretical_loss": 3.498552394777754,
+      "tokens_seen": 1745420288
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004758064516129033,
+      "loss": 1.4249,
+      "theoretical_loss": 3.498530082712908,
+      "tokens_seen": 1745551360
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000475766329642112,
+      "loss": 1.4097,
+      "theoretical_loss": 3.498507772792464,
+      "tokens_seen": 1745682432
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004757262076713208,
+      "loss": 1.4305,
+      "theoretical_loss": 3.498485465016054,
+      "tokens_seen": 1745813504
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047568608570052964,
+      "loss": 1.3859,
+      "theoretical_loss": 3.498463159383312,
+      "tokens_seen": 1745944576
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004756459637297384,
+      "loss": 1.4817,
+      "theoretical_loss": 3.498440855893871,
+      "tokens_seen": 1746075648
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004756058417589472,
+      "loss": 1.4339,
+      "theoretical_loss": 3.498418554547364,
+      "tokens_seen": 1746206720
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000475565719788156,
+      "loss": 1.3576,
+      "theoretical_loss": 3.4983962553434242,
+      "tokens_seen": 1746337792
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004755255978173648,
+      "loss": 1.355,
+      "theoretical_loss": 3.498373958281685,
+      "tokens_seen": 1746468864
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004754854758465736,
+      "loss": 1.3822,
+      "theoretical_loss": 3.4983516633617797,
+      "tokens_seen": 1746599936
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004754453538757824,
+      "loss": 1.4294,
+      "theoretical_loss": 3.4983293705833423,
+      "tokens_seen": 1746731008
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004754052319049912,
+      "loss": 1.3347,
+      "theoretical_loss": 3.498307079946006,
+      "tokens_seen": 1746862080
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004753651099342,
+      "loss": 1.4022,
+      "theoretical_loss": 3.4982847914494046,
+      "tokens_seen": 1746993152
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004753249879634088,
+      "loss": 1.3744,
+      "theoretical_loss": 3.498262505093172,
+      "tokens_seen": 1747124224
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047528486599261754,
+      "loss": 1.4711,
+      "theoretical_loss": 3.4982402208769425,
+      "tokens_seen": 1747255296
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004752447440218264,
+      "loss": 1.4915,
+      "theoretical_loss": 3.49821793880035,
+      "tokens_seen": 1747386368
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047520462205103515,
+      "loss": 1.384,
+      "theoretical_loss": 3.4981956588630276,
+      "tokens_seen": 1747517440
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047516450008024396,
+      "loss": 1.3903,
+      "theoretical_loss": 3.4981733810646105,
+      "tokens_seen": 1747648512
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047512437810945276,
+      "loss": 1.4074,
+      "theoretical_loss": 3.498151105404733,
+      "tokens_seen": 1747779584
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004750842561386615,
+      "loss": 1.3836,
+      "theoretical_loss": 3.4981288318830286,
+      "tokens_seen": 1747910656
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004750441341678703,
+      "loss": 1.4662,
+      "theoretical_loss": 3.4981065604991324,
+      "tokens_seen": 1748041728
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047500401219707913,
+      "loss": 1.4044,
+      "theoretical_loss": 3.498084291252679,
+      "tokens_seen": 1748172800
+    },
+    {
+      "epoch": 0.06,
+      "objective/train/advantage_avg": 0.492466539144516,
+      "objective/train/docs_used": 638722,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.236612319946289,
+      "objective/train/original_loss": 1.23661208152771,
+      "objective/train/theoretical_loss": 3.4980620241433025,
+      "objective/train/tokens_used": 118762976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2430907040834427,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504825115203857,
+      "objective/train/weighted_lm_loss": 1.2990974187850952,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4980620241433025,
+      "tokens_seen": 1748303872
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047496389022628794,
+      "loss": 1.3456,
+      "theoretical_loss": 3.4980620241433025,
+      "tokens_seen": 1748303872
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004749237682554967,
+      "loss": 1.3803,
+      "theoretical_loss": 3.498039759170638,
+      "tokens_seen": 1748434944
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047488364628470555,
+      "loss": 1.362,
+      "theoretical_loss": 3.4980174963343202,
+      "tokens_seen": 1748566016
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004748435243139143,
+      "loss": 1.336,
+      "theoretical_loss": 3.4979952356339847,
+      "tokens_seen": 1748697088
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004748034023431231,
+      "loss": 1.4101,
+      "theoretical_loss": 3.4979729770692654,
+      "tokens_seen": 1748828160
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004747632803723319,
+      "loss": 1.4015,
+      "theoretical_loss": 3.497950720639798,
+      "tokens_seen": 1748959232
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047472315840154066,
+      "loss": 1.4387,
+      "theoretical_loss": 3.4979284663452175,
+      "tokens_seen": 1749090304
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004746830364307495,
+      "loss": 1.4141,
+      "theoretical_loss": 3.497906214185159,
+      "tokens_seen": 1749221376
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004746429144599583,
+      "loss": 1.4591,
+      "theoretical_loss": 3.497883964159258,
+      "tokens_seen": 1749352448
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004746027924891671,
+      "loss": 1.3502,
+      "theoretical_loss": 3.49786171626715,
+      "tokens_seen": 1749483520
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004745626705183759,
+      "loss": 1.4107,
+      "theoretical_loss": 3.4978394705084703,
+      "tokens_seen": 1749614592
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047452254854758464,
+      "loss": 1.4137,
+      "theoretical_loss": 3.497817226882855,
+      "tokens_seen": 1749745664
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047448242657679345,
+      "loss": 1.457,
+      "theoretical_loss": 3.4977949853899393,
+      "tokens_seen": 1749876736
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047444230460600225,
+      "loss": 1.4541,
+      "theoretical_loss": 3.4977727460293595,
+      "tokens_seen": 1750007808
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047440218263521106,
+      "loss": 1.3769,
+      "theoretical_loss": 3.497750508800751,
+      "tokens_seen": 1750138880
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004743620606644198,
+      "loss": 1.3937,
+      "theoretical_loss": 3.49772827370375,
+      "tokens_seen": 1750269952
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004743219386936287,
+      "loss": 1.384,
+      "theoretical_loss": 3.497706040737992,
+      "tokens_seen": 1750401024
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004742818167228374,
+      "loss": 1.3599,
+      "theoretical_loss": 3.4976838099031142,
+      "tokens_seen": 1750532096
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004742416947520462,
+      "loss": 1.4512,
+      "theoretical_loss": 3.497661581198752,
+      "tokens_seen": 1750663168
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047420157278125504,
+      "loss": 1.4062,
+      "theoretical_loss": 3.4976393546245426,
+      "tokens_seen": 1750794240
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004741614508104638,
+      "loss": 1.4069,
+      "theoretical_loss": 3.497617130180121,
+      "tokens_seen": 1750925312
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047412132883967265,
+      "loss": 1.3906,
+      "theoretical_loss": 3.497594907865125,
+      "tokens_seen": 1751056384
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004740812068688814,
+      "loss": 1.4606,
+      "theoretical_loss": 3.497572687679191,
+      "tokens_seen": 1751187456
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004740410848980902,
+      "loss": 1.4379,
+      "theoretical_loss": 3.497550469621955,
+      "tokens_seen": 1751318528
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000474000962927299,
+      "loss": 1.4368,
+      "theoretical_loss": 3.4975282536930545,
+      "tokens_seen": 1751449600
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047396084095650777,
+      "loss": 1.3707,
+      "theoretical_loss": 3.4975060398921265,
+      "tokens_seen": 1751580672
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004739207189857166,
+      "loss": 1.4131,
+      "theoretical_loss": 3.4974838282188068,
+      "tokens_seen": 1751711744
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004738805970149254,
+      "loss": 1.5052,
+      "theoretical_loss": 3.4974616186727334,
+      "tokens_seen": 1751842816
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004738404750441342,
+      "loss": 1.3824,
+      "theoretical_loss": 3.497439411253543,
+      "tokens_seen": 1751973888
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047380035307334294,
+      "loss": 1.4208,
+      "theoretical_loss": 3.4974172059608737,
+      "tokens_seen": 1752104960
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004737602311025518,
+      "loss": 1.4053,
+      "theoretical_loss": 3.497395002794362,
+      "tokens_seen": 1752236032
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047372010913176055,
+      "loss": 1.3878,
+      "theoretical_loss": 3.497372801753645,
+      "tokens_seen": 1752367104
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047367998716096936,
+      "loss": 1.4734,
+      "theoretical_loss": 3.497350602838361,
+      "tokens_seen": 1752498176
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047363986519017816,
+      "loss": 1.471,
+      "theoretical_loss": 3.4973284060481475,
+      "tokens_seen": 1752629248
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004735997432193869,
+      "loss": 1.3662,
+      "theoretical_loss": 3.497306211382641,
+      "tokens_seen": 1752760320
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004735596212485958,
+      "loss": 1.369,
+      "theoretical_loss": 3.497284018841481,
+      "tokens_seen": 1752891392
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047351949927780453,
+      "loss": 1.5142,
+      "theoretical_loss": 3.497261828424304,
+      "tokens_seen": 1753022464
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047347937730701334,
+      "loss": 1.3607,
+      "theoretical_loss": 3.497239640130749,
+      "tokens_seen": 1753153536
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047343925533622214,
+      "loss": 1.429,
+      "theoretical_loss": 3.4972174539604532,
+      "tokens_seen": 1753284608
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047339913336543095,
+      "loss": 1.3766,
+      "theoretical_loss": 3.4971952699130546,
+      "tokens_seen": 1753415680
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004733590113946397,
+      "loss": 1.4197,
+      "theoretical_loss": 3.497173087988192,
+      "tokens_seen": 1753546752
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004733188894238485,
+      "loss": 1.4737,
+      "theoretical_loss": 3.4971509081855037,
+      "tokens_seen": 1753677824
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004732787674530573,
+      "loss": 1.408,
+      "theoretical_loss": 3.4971287305046275,
+      "tokens_seen": 1753808896
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047323864548226606,
+      "loss": 1.4443,
+      "theoretical_loss": 3.4971065549452023,
+      "tokens_seen": 1753939968
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004731985235114749,
+      "loss": 1.4109,
+      "theoretical_loss": 3.497084381506867,
+      "tokens_seen": 1754071040
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004731584015406837,
+      "loss": 1.4536,
+      "theoretical_loss": 3.4970622101892594,
+      "tokens_seen": 1754202112
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004731182795698925,
+      "loss": 1.4405,
+      "theoretical_loss": 3.497040040992019,
+      "tokens_seen": 1754333184
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004730781575991013,
+      "loss": 1.4848,
+      "theoretical_loss": 3.4970178739147837,
+      "tokens_seen": 1754464256
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047303803562831004,
+      "loss": 1.4891,
+      "theoretical_loss": 3.496995708957193,
+      "tokens_seen": 1754595328
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047299791365751885,
+      "loss": 1.3721,
+      "theoretical_loss": 3.496973546118886,
+      "tokens_seen": 1754726400
+    },
+    {
+      "epoch": 0.06,
+      "objective/train/advantage_avg": 0.49056705832481384,
+      "objective/train/docs_used": 641116,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3271706104278564,
+      "objective/train/original_loss": 1.327170491218567,
+      "objective/train/theoretical_loss": 3.496951385399502,
+      "objective/train/tokens_used": 125316576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2415105402469635,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050284504890442,
+      "objective/train/weighted_lm_loss": 1.393988847732544,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.496951385399502,
+      "tokens_seen": 1754857472
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047295779168672765,
+      "loss": 1.4117,
+      "theoretical_loss": 3.496951385399502,
+      "tokens_seen": 1754857472
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047291766971593646,
+      "loss": 1.3989,
+      "theoretical_loss": 3.496929226798679,
+      "tokens_seen": 1754988544
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047287754774514527,
+      "loss": 1.4092,
+      "theoretical_loss": 3.496907070316058,
+      "tokens_seen": 1755119616
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004728374257743541,
+      "loss": 1.3963,
+      "theoretical_loss": 3.496884915951277,
+      "tokens_seen": 1755250688
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004727973038035628,
+      "loss": 1.4248,
+      "theoretical_loss": 3.4968627637039758,
+      "tokens_seen": 1755381760
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047275718183277163,
+      "loss": 1.4215,
+      "theoretical_loss": 3.496840613573794,
+      "tokens_seen": 1755512832
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047271705986198044,
+      "loss": 1.3768,
+      "theoretical_loss": 3.4968184655603713,
+      "tokens_seen": 1755643904
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004726769378911892,
+      "loss": 1.4028,
+      "theoretical_loss": 3.4967963196633476,
+      "tokens_seen": 1755774976
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047263681592039805,
+      "loss": 1.3764,
+      "theoretical_loss": 3.4967741758823623,
+      "tokens_seen": 1755906048
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004725966939496068,
+      "loss": 1.3686,
+      "theoretical_loss": 3.496752034217055,
+      "tokens_seen": 1756037120
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004725565719788156,
+      "loss": 1.3359,
+      "theoretical_loss": 3.4967298946670664,
+      "tokens_seen": 1756168192
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004725164500080244,
+      "loss": 1.3983,
+      "theoretical_loss": 3.496707757232036,
+      "tokens_seen": 1756299264
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047247632803723317,
+      "loss": 1.4416,
+      "theoretical_loss": 3.4966856219116043,
+      "tokens_seen": 1756430336
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000472436206066442,
+      "loss": 1.4342,
+      "theoretical_loss": 3.4966634887054115,
+      "tokens_seen": 1756561408
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004723960840956508,
+      "loss": 1.4403,
+      "theoretical_loss": 3.4966413576130977,
+      "tokens_seen": 1756692480
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004723559621248596,
+      "loss": 1.3869,
+      "theoretical_loss": 3.4966192286343034,
+      "tokens_seen": 1756823552
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004723158401540684,
+      "loss": 1.4642,
+      "theoretical_loss": 3.4965971017686694,
+      "tokens_seen": 1756954624
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004722757181832772,
+      "loss": 1.484,
+      "theoretical_loss": 3.496574977015836,
+      "tokens_seen": 1757085696
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00047223559621248595,
+      "loss": 1.4613,
+      "theoretical_loss": 3.4965528543754436,
+      "tokens_seen": 1757216768
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004721954742416948,
+      "loss": 1.3182,
+      "theoretical_loss": 3.496530733847133,
+      "tokens_seen": 1757347840
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00047215535227090356,
+      "loss": 1.4038,
+      "theoretical_loss": 3.496508615430546,
+      "tokens_seen": 1757478912
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004721152303001123,
+      "loss": 1.4181,
+      "theoretical_loss": 3.4964864991253224,
+      "tokens_seen": 1757609984
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004720751083293212,
+      "loss": 1.3865,
+      "theoretical_loss": 3.496464384931104,
+      "tokens_seen": 1757741056
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00047203498635852993,
+      "loss": 1.4539,
+      "theoretical_loss": 3.4964422728475313,
+      "tokens_seen": 1757872128
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00047199486438773873,
+      "loss": 1.4211,
+      "theoretical_loss": 3.4964201628742457,
+      "tokens_seen": 1758003200
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00047195474241694754,
+      "loss": 1.3867,
+      "theoretical_loss": 3.4963980550108884,
+      "tokens_seen": 1758134272
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00047191462044615635,
+      "loss": 1.3554,
+      "theoretical_loss": 3.4963759492571014,
+      "tokens_seen": 1758265344
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004718744984753651,
+      "loss": 1.4188,
+      "theoretical_loss": 3.496353845612526,
+      "tokens_seen": 1758396416
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004718343765045739,
+      "loss": 1.3313,
+      "theoretical_loss": 3.4963317440768025,
+      "tokens_seen": 1758527488
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004717942545337827,
+      "loss": 1.4627,
+      "theoretical_loss": 3.496309644649574,
+      "tokens_seen": 1758658560
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00047175413256299146,
+      "loss": 1.4871,
+      "theoretical_loss": 3.4962875473304815,
+      "tokens_seen": 1758789632
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004717140105922003,
+      "loss": 1.5018,
+      "theoretical_loss": 3.496265452119167,
+      "tokens_seen": 1758920704
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004716738886214091,
+      "loss": 1.4509,
+      "theoretical_loss": 3.496243359015273,
+      "tokens_seen": 1759051776
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00047163376665061794,
+      "loss": 1.3809,
+      "theoretical_loss": 3.4962212680184406,
+      "tokens_seen": 1759182848
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004715936446798267,
+      "loss": 1.3618,
+      "theoretical_loss": 3.496199179128312,
+      "tokens_seen": 1759313920
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00047155352270903544,
+      "loss": 1.4103,
+      "theoretical_loss": 3.4961770923445297,
+      "tokens_seen": 1759444992
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004715134007382443,
+      "loss": 1.4053,
+      "theoretical_loss": 3.4961550076667356,
+      "tokens_seen": 1759576064
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00047147327876745305,
+      "loss": 1.3475,
+      "theoretical_loss": 3.4961329250945727,
+      "tokens_seen": 1759707136
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00047143315679666186,
+      "loss": 1.3359,
+      "theoretical_loss": 3.4961108446276823,
+      "tokens_seen": 1759838208
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00047139303482587067,
+      "loss": 1.4874,
+      "theoretical_loss": 3.496088766265708,
+      "tokens_seen": 1759969280
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004713529128550795,
+      "loss": 1.504,
+      "theoretical_loss": 3.496066690008292,
+      "tokens_seen": 1760100352
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004713127908842882,
+      "loss": 1.4465,
+      "theoretical_loss": 3.4960446158550775,
+      "tokens_seen": 1760231424
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00047127266891349703,
+      "loss": 1.3419,
+      "theoretical_loss": 3.496022543805706,
+      "tokens_seen": 1760362496
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00047123254694270584,
+      "loss": 1.464,
+      "theoretical_loss": 3.496000473859821,
+      "tokens_seen": 1760493568
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004711924249719146,
+      "loss": 1.3898,
+      "theoretical_loss": 3.495978406017066,
+      "tokens_seen": 1760624640
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00047115230300112345,
+      "loss": 1.3775,
+      "theoretical_loss": 3.4959563402770835,
+      "tokens_seen": 1760755712
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004711121810303322,
+      "loss": 1.4458,
+      "theoretical_loss": 3.495934276639517,
+      "tokens_seen": 1760886784
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00047107205905954106,
+      "loss": 1.4233,
+      "theoretical_loss": 3.495912215104009,
+      "tokens_seen": 1761017856
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004710319370887498,
+      "loss": 1.3738,
+      "theoretical_loss": 3.4958901556702036,
+      "tokens_seen": 1761148928
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00047099181511795857,
+      "loss": 1.4149,
+      "theoretical_loss": 3.495868098337744,
+      "tokens_seen": 1761280000
+    },
+    {
+      "epoch": 0.07,
+      "objective/train/advantage_avg": 0.49273422360420227,
+      "objective/train/docs_used": 643454,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.286047101020813,
+      "objective/train/original_loss": 1.2860469818115234,
+      "objective/train/theoretical_loss": 3.4958460431062726,
+      "objective/train/tokens_used": 131870176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24346250295639038,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505112409591675,
+      "objective/train/weighted_lm_loss": 1.3509286642074585,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4958460431062726,
+      "tokens_seen": 1761411072
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00047095169314716743,
+      "loss": 1.343,
+      "theoretical_loss": 3.4958460431062726,
+      "tokens_seen": 1761411072
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004709115711763762,
+      "loss": 1.4811,
+      "theoretical_loss": 3.4958239899754346,
+      "tokens_seen": 1761542144
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000470871449205585,
+      "loss": 1.4304,
+      "theoretical_loss": 3.4958019389448727,
+      "tokens_seen": 1761673216
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004708313272347938,
+      "loss": 1.4259,
+      "theoretical_loss": 3.495779890014231,
+      "tokens_seen": 1761804288
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004707912052640026,
+      "loss": 1.3619,
+      "theoretical_loss": 3.495757843183153,
+      "tokens_seen": 1761935360
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00047075108329321135,
+      "loss": 1.5369,
+      "theoretical_loss": 3.4957357984512827,
+      "tokens_seen": 1762066432
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004707109613224202,
+      "loss": 1.3894,
+      "theoretical_loss": 3.495713755818264,
+      "tokens_seen": 1762197504
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00047067083935162896,
+      "loss": 1.4179,
+      "theoretical_loss": 3.4956917152837415,
+      "tokens_seen": 1762328576
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004706307173808377,
+      "loss": 1.5179,
+      "theoretical_loss": 3.4956696768473585,
+      "tokens_seen": 1762459648
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004705905954100466,
+      "loss": 1.4123,
+      "theoretical_loss": 3.4956476405087598,
+      "tokens_seen": 1762590720
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00047055047343925533,
+      "loss": 1.3942,
+      "theoretical_loss": 3.4956256062675903,
+      "tokens_seen": 1762721792
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00047051035146846413,
+      "loss": 1.4852,
+      "theoretical_loss": 3.4956035741234937,
+      "tokens_seen": 1762852864
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00047047022949767294,
+      "loss": 1.4001,
+      "theoretical_loss": 3.4955815440761144,
+      "tokens_seen": 1762983936
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00047043010752688175,
+      "loss": 1.4817,
+      "theoretical_loss": 3.495559516125097,
+      "tokens_seen": 1763115008
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00047038998555609055,
+      "loss": 1.3962,
+      "theoretical_loss": 3.4955374902700864,
+      "tokens_seen": 1763246080
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004703498635852993,
+      "loss": 1.4471,
+      "theoretical_loss": 3.495515466510728,
+      "tokens_seen": 1763377152
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004703097416145081,
+      "loss": 1.4398,
+      "theoretical_loss": 3.495493444846666,
+      "tokens_seen": 1763508224
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004702696196437169,
+      "loss": 1.6096,
+      "theoretical_loss": 3.495471425277545,
+      "tokens_seen": 1763639296
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004702294976729257,
+      "loss": 1.4173,
+      "theoretical_loss": 3.4954494078030107,
+      "tokens_seen": 1763770368
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004701893757021345,
+      "loss": 1.3199,
+      "theoretical_loss": 3.495427392422708,
+      "tokens_seen": 1763901440
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00047014925373134334,
+      "loss": 1.4743,
+      "theoretical_loss": 3.495405379136282,
+      "tokens_seen": 1764032512
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004701091317605521,
+      "loss": 1.3642,
+      "theoretical_loss": 3.495383367943378,
+      "tokens_seen": 1764163584
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00047006900978976084,
+      "loss": 1.4358,
+      "theoretical_loss": 3.4953613588436414,
+      "tokens_seen": 1764294656
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004700288878189697,
+      "loss": 1.417,
+      "theoretical_loss": 3.4953393518367175,
+      "tokens_seen": 1764425728
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046998876584817845,
+      "loss": 1.4093,
+      "theoretical_loss": 3.4953173469222527,
+      "tokens_seen": 1764556800
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046994864387738726,
+      "loss": 1.4064,
+      "theoretical_loss": 3.4952953440998917,
+      "tokens_seen": 1764687872
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046990852190659607,
+      "loss": 1.4594,
+      "theoretical_loss": 3.4952733433692806,
+      "tokens_seen": 1764818944
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046986839993580487,
+      "loss": 1.4714,
+      "theoretical_loss": 3.495251344730065,
+      "tokens_seen": 1764950016
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004698282779650136,
+      "loss": 1.4884,
+      "theoretical_loss": 3.495229348181891,
+      "tokens_seen": 1765081088
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046978815599422243,
+      "loss": 1.4131,
+      "theoretical_loss": 3.4952073537244046,
+      "tokens_seen": 1765212160
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046974803402343124,
+      "loss": 1.351,
+      "theoretical_loss": 3.495185361357252,
+      "tokens_seen": 1765343232
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046970791205264004,
+      "loss": 1.4988,
+      "theoretical_loss": 3.4951633710800785,
+      "tokens_seen": 1765474304
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046966779008184885,
+      "loss": 1.4082,
+      "theoretical_loss": 3.4951413828925317,
+      "tokens_seen": 1765605376
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004696276681110576,
+      "loss": 1.392,
+      "theoretical_loss": 3.495119396794257,
+      "tokens_seen": 1765736448
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046958754614026646,
+      "loss": 1.37,
+      "theoretical_loss": 3.495097412784901,
+      "tokens_seen": 1765867520
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004695474241694752,
+      "loss": 1.4852,
+      "theoretical_loss": 3.4950754308641105,
+      "tokens_seen": 1765998592
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046950730219868397,
+      "loss": 1.3748,
+      "theoretical_loss": 3.495053451031531,
+      "tokens_seen": 1766129664
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046946718022789283,
+      "loss": 1.3577,
+      "theoretical_loss": 3.4950314732868106,
+      "tokens_seen": 1766260736
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004694270582571016,
+      "loss": 1.4256,
+      "theoretical_loss": 3.495009497629596,
+      "tokens_seen": 1766391808
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004693869362863104,
+      "loss": 1.3964,
+      "theoretical_loss": 3.4949875240595327,
+      "tokens_seen": 1766522880
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004693468143155192,
+      "loss": 1.4036,
+      "theoretical_loss": 3.494965552576269,
+      "tokens_seen": 1766653952
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000469306692344728,
+      "loss": 1.529,
+      "theoretical_loss": 3.4949435831794506,
+      "tokens_seen": 1766785024
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046926657037393675,
+      "loss": 1.3521,
+      "theoretical_loss": 3.4949216158687264,
+      "tokens_seen": 1766916096
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046922644840314556,
+      "loss": 1.4109,
+      "theoretical_loss": 3.4948996506437418,
+      "tokens_seen": 1767047168
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046918632643235436,
+      "loss": 1.4182,
+      "theoretical_loss": 3.494877687504145,
+      "tokens_seen": 1767178240
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004691462044615631,
+      "loss": 1.407,
+      "theoretical_loss": 3.4948557264495834,
+      "tokens_seen": 1767309312
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000469106082490772,
+      "loss": 1.4009,
+      "theoretical_loss": 3.494833767479704,
+      "tokens_seen": 1767440384
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046906596051998073,
+      "loss": 1.39,
+      "theoretical_loss": 3.494811810594155,
+      "tokens_seen": 1767571456
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004690258385491896,
+      "loss": 1.3686,
+      "theoretical_loss": 3.494789855792583,
+      "tokens_seen": 1767702528
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046898571657839834,
+      "loss": 1.4794,
+      "theoretical_loss": 3.494767903074637,
+      "tokens_seen": 1767833600
+    },
+    {
+      "epoch": 0.07,
+      "objective/train/advantage_avg": 0.49285978078842163,
+      "objective/train/docs_used": 645687,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.344320297241211,
+      "objective/train/original_loss": 1.344320297241211,
+      "objective/train/theoretical_loss": 3.4947459524399633,
+      "objective/train/tokens_used": 138423776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24335119128227234,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505231618881226,
+      "objective/train/weighted_lm_loss": 1.4121487140655518,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4947459524399633,
+      "tokens_seen": 1767964672
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046894559460760715,
+      "loss": 1.3848,
+      "theoretical_loss": 3.4947459524399633,
+      "tokens_seen": 1767964672
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046890547263681595,
+      "loss": 1.4119,
+      "theoretical_loss": 3.494724003888211,
+      "tokens_seen": 1768095744
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004688653506660247,
+      "loss": 1.3518,
+      "theoretical_loss": 3.494702057419027,
+      "tokens_seen": 1768226816
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004688252286952335,
+      "loss": 1.3641,
+      "theoretical_loss": 3.494680113032061,
+      "tokens_seen": 1768357888
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004687851067244423,
+      "loss": 1.4583,
+      "theoretical_loss": 3.494658170726959,
+      "tokens_seen": 1768488960
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004687449847536511,
+      "loss": 1.4447,
+      "theoretical_loss": 3.494636230503371,
+      "tokens_seen": 1768620032
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004687048627828599,
+      "loss": 1.5009,
+      "theoretical_loss": 3.4946142923609447,
+      "tokens_seen": 1768751104
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046866474081206874,
+      "loss": 1.3929,
+      "theoretical_loss": 3.4945923562993277,
+      "tokens_seen": 1768882176
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004686246188412775,
+      "loss": 1.3733,
+      "theoretical_loss": 3.49457042231817,
+      "tokens_seen": 1769013248
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046858449687048624,
+      "loss": 1.3414,
+      "theoretical_loss": 3.4945484904171193,
+      "tokens_seen": 1769144320
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004685443748996951,
+      "loss": 1.4149,
+      "theoretical_loss": 3.4945265605958236,
+      "tokens_seen": 1769275392
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046850425292890385,
+      "loss": 1.4036,
+      "theoretical_loss": 3.494504632853933,
+      "tokens_seen": 1769406464
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004684641309581127,
+      "loss": 1.4623,
+      "theoretical_loss": 3.4944827071910955,
+      "tokens_seen": 1769537536
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046842400898732147,
+      "loss": 1.4257,
+      "theoretical_loss": 3.49446078360696,
+      "tokens_seen": 1769668608
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046838388701653027,
+      "loss": 1.4052,
+      "theoretical_loss": 3.494438862101175,
+      "tokens_seen": 1769799680
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004683437650457391,
+      "loss": 1.3456,
+      "theoretical_loss": 3.494416942673391,
+      "tokens_seen": 1769930752
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046830364307494783,
+      "loss": 1.3818,
+      "theoretical_loss": 3.4943950253232563,
+      "tokens_seen": 1770061824
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046826352110415664,
+      "loss": 1.398,
+      "theoretical_loss": 3.49437311005042,
+      "tokens_seen": 1770192896
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046822339913336544,
+      "loss": 1.3084,
+      "theoretical_loss": 3.4943511968545313,
+      "tokens_seen": 1770323968
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046818327716257425,
+      "loss": 1.4041,
+      "theoretical_loss": 3.4943292857352404,
+      "tokens_seen": 1770455040
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000468143155191783,
+      "loss": 1.4086,
+      "theoretical_loss": 3.4943073766921957,
+      "tokens_seen": 1770586112
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046810303322099186,
+      "loss": 1.4229,
+      "theoretical_loss": 3.4942854697250483,
+      "tokens_seen": 1770717184
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004680629112502006,
+      "loss": 1.4034,
+      "theoretical_loss": 3.494263564833447,
+      "tokens_seen": 1770848256
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046802278927940937,
+      "loss": 1.3942,
+      "theoretical_loss": 3.4942416620170405,
+      "tokens_seen": 1770979328
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004679826673086182,
+      "loss": 1.3677,
+      "theoretical_loss": 3.49421976127548,
+      "tokens_seen": 1771110400
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000467942545337827,
+      "loss": 1.4067,
+      "theoretical_loss": 3.4941978626084156,
+      "tokens_seen": 1771241472
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004679024233670358,
+      "loss": 1.3997,
+      "theoretical_loss": 3.4941759660154963,
+      "tokens_seen": 1771372544
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004678623013962446,
+      "loss": 1.3412,
+      "theoretical_loss": 3.494154071496373,
+      "tokens_seen": 1771503616
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004678221794254534,
+      "loss": 1.3885,
+      "theoretical_loss": 3.4941321790506956,
+      "tokens_seen": 1771634688
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004677820574546622,
+      "loss": 1.3853,
+      "theoretical_loss": 3.4941102886781135,
+      "tokens_seen": 1771765760
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046774193548387096,
+      "loss": 1.4281,
+      "theoretical_loss": 3.4940884003782786,
+      "tokens_seen": 1771896832
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046770181351307976,
+      "loss": 1.386,
+      "theoretical_loss": 3.4940665141508407,
+      "tokens_seen": 1772027904
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046766169154228857,
+      "loss": 1.3218,
+      "theoretical_loss": 3.49404462999545,
+      "tokens_seen": 1772158976
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004676215695714974,
+      "loss": 1.5046,
+      "theoretical_loss": 3.494022747911757,
+      "tokens_seen": 1772290048
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046758144760070613,
+      "loss": 1.4087,
+      "theoretical_loss": 3.4940008678994134,
+      "tokens_seen": 1772421120
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000467541325629915,
+      "loss": 1.4306,
+      "theoretical_loss": 3.4939789899580687,
+      "tokens_seen": 1772552192
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046750120365912374,
+      "loss": 1.3863,
+      "theoretical_loss": 3.4939571140873746,
+      "tokens_seen": 1772683264
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046746108168833255,
+      "loss": 1.4147,
+      "theoretical_loss": 3.493935240286982,
+      "tokens_seen": 1772814336
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046742095971754135,
+      "loss": 1.3725,
+      "theoretical_loss": 3.493913368556541,
+      "tokens_seen": 1772945408
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004673808377467501,
+      "loss": 1.3994,
+      "theoretical_loss": 3.493891498895704,
+      "tokens_seen": 1773076480
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004673407157759589,
+      "loss": 1.3856,
+      "theoretical_loss": 3.493869631304121,
+      "tokens_seen": 1773207552
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004673005938051677,
+      "loss": 1.3801,
+      "theoretical_loss": 3.4938477657814446,
+      "tokens_seen": 1773338624
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0004672604718343765,
+      "loss": 1.4392,
+      "theoretical_loss": 3.493825902327325,
+      "tokens_seen": 1773469696
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046722034986358533,
+      "loss": 1.4037,
+      "theoretical_loss": 3.493804040941414,
+      "tokens_seen": 1773600768
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00046718022789279414,
+      "loss": 1.417,
+      "theoretical_loss": 3.493782181623364,
+      "tokens_seen": 1773731840
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004671401059220029,
+      "loss": 1.545,
+      "theoretical_loss": 3.493760324372825,
+      "tokens_seen": 1773862912
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004670999839512117,
+      "loss": 1.486,
+      "theoretical_loss": 3.4937384691894495,
+      "tokens_seen": 1773993984
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004670598619804205,
+      "loss": 1.3595,
+      "theoretical_loss": 3.4937166160728896,
+      "tokens_seen": 1774125056
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046701974000962925,
+      "loss": 1.4164,
+      "theoretical_loss": 3.4936947650227967,
+      "tokens_seen": 1774256128
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004669796180388381,
+      "loss": 1.3797,
+      "theoretical_loss": 3.4936729160388236,
+      "tokens_seen": 1774387200
+    },
+    {
+      "epoch": 0.08,
+      "objective/train/advantage_avg": 0.4899574816226959,
+      "objective/train/docs_used": 647978,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3552584648132324,
+      "objective/train/original_loss": 1.3552584648132324,
+      "objective/train/theoretical_loss": 3.493651069120621,
+      "objective/train/tokens_used": 144977376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24081814289093018,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0502198934555054,
+      "objective/train/weighted_lm_loss": 1.4234589338302612,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.493651069120621,
+      "tokens_seen": 1774518272
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046693949606804687,
+      "loss": 1.4321,
+      "theoretical_loss": 3.493651069120621,
+      "tokens_seen": 1774518272
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046689937409725567,
+      "loss": 1.3909,
+      "theoretical_loss": 3.493629224267842,
+      "tokens_seen": 1774649344
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004668592521264645,
+      "loss": 1.2969,
+      "theoretical_loss": 3.4936073814801385,
+      "tokens_seen": 1774780416
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046681913015567323,
+      "loss": 1.4725,
+      "theoretical_loss": 3.493585540757163,
+      "tokens_seen": 1774911488
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046677900818488204,
+      "loss": 1.389,
+      "theoretical_loss": 3.4935637020985673,
+      "tokens_seen": 1775042560
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046673888621409084,
+      "loss": 1.3922,
+      "theoretical_loss": 3.4935418655040045,
+      "tokens_seen": 1775173632
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046669876424329965,
+      "loss": 1.365,
+      "theoretical_loss": 3.4935200309731265,
+      "tokens_seen": 1775304704
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004666586422725084,
+      "loss": 1.3742,
+      "theoretical_loss": 3.493498198505587,
+      "tokens_seen": 1775435776
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046661852030171726,
+      "loss": 1.407,
+      "theoretical_loss": 3.493476368101037,
+      "tokens_seen": 1775566848
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000466578398330926,
+      "loss": 1.4304,
+      "theoretical_loss": 3.4934545397591314,
+      "tokens_seen": 1775697920
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004665382763601348,
+      "loss": 1.4123,
+      "theoretical_loss": 3.493432713479521,
+      "tokens_seen": 1775828992
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004664981543893436,
+      "loss": 1.4686,
+      "theoretical_loss": 3.4934108892618605,
+      "tokens_seen": 1775960064
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004664580324185524,
+      "loss": 1.4135,
+      "theoretical_loss": 3.4933890671058023,
+      "tokens_seen": 1776091136
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046641791044776124,
+      "loss": 1.4224,
+      "theoretical_loss": 3.493367247010999,
+      "tokens_seen": 1776222208
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046637778847697,
+      "loss": 1.3993,
+      "theoretical_loss": 3.493345428977104,
+      "tokens_seen": 1776353280
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004663376665061788,
+      "loss": 1.3776,
+      "theoretical_loss": 3.4933236130037715,
+      "tokens_seen": 1776484352
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004662975445353876,
+      "loss": 1.3902,
+      "theoretical_loss": 3.493301799090654,
+      "tokens_seen": 1776615424
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046625742256459636,
+      "loss": 1.4527,
+      "theoretical_loss": 3.4932799872374054,
+      "tokens_seen": 1776746496
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046621730059380516,
+      "loss": 1.3498,
+      "theoretical_loss": 3.4932581774436793,
+      "tokens_seen": 1776877568
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046617717862301397,
+      "loss": 1.3575,
+      "theoretical_loss": 3.4932363697091287,
+      "tokens_seen": 1777008640
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004661370566522228,
+      "loss": 1.3981,
+      "theoretical_loss": 3.4932145640334076,
+      "tokens_seen": 1777139712
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046609693468143153,
+      "loss": 1.3868,
+      "theoretical_loss": 3.4931927604161697,
+      "tokens_seen": 1777270784
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004660568127106404,
+      "loss": 1.4095,
+      "theoretical_loss": 3.4931709588570694,
+      "tokens_seen": 1777401856
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046601669073984914,
+      "loss": 1.4184,
+      "theoretical_loss": 3.4931491593557604,
+      "tokens_seen": 1777532928
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000465976568769058,
+      "loss": 1.449,
+      "theoretical_loss": 3.493127361911897,
+      "tokens_seen": 1777664000
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046593644679826675,
+      "loss": 1.3256,
+      "theoretical_loss": 3.493105566525133,
+      "tokens_seen": 1777795072
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004658963248274755,
+      "loss": 1.3769,
+      "theoretical_loss": 3.4930837731951225,
+      "tokens_seen": 1777926144
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046585620285668436,
+      "loss": 1.3877,
+      "theoretical_loss": 3.49306198192152,
+      "tokens_seen": 1778057216
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004658160808858931,
+      "loss": 1.3658,
+      "theoretical_loss": 3.49304019270398,
+      "tokens_seen": 1778188288
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004657759589151019,
+      "loss": 1.3658,
+      "theoretical_loss": 3.4930184055421565,
+      "tokens_seen": 1778319360
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046573583694431073,
+      "loss": 1.4172,
+      "theoretical_loss": 3.4929966204357052,
+      "tokens_seen": 1778450432
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046569571497351954,
+      "loss": 1.3046,
+      "theoretical_loss": 3.4929748373842795,
+      "tokens_seen": 1778581504
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004656555930027283,
+      "loss": 1.3592,
+      "theoretical_loss": 3.4929530563875346,
+      "tokens_seen": 1778712576
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004656154710319371,
+      "loss": 1.3173,
+      "theoretical_loss": 3.492931277445125,
+      "tokens_seen": 1778843648
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004655753490611459,
+      "loss": 1.3754,
+      "theoretical_loss": 3.4929095005567063,
+      "tokens_seen": 1778974720
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046553522709035465,
+      "loss": 1.3335,
+      "theoretical_loss": 3.492887725721933,
+      "tokens_seen": 1779105792
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004654951051195635,
+      "loss": 1.3702,
+      "theoretical_loss": 3.49286595294046,
+      "tokens_seen": 1779236864
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046545498314877227,
+      "loss": 1.3171,
+      "theoretical_loss": 3.4928441822119423,
+      "tokens_seen": 1779367936
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046541486117798107,
+      "loss": 1.3645,
+      "theoretical_loss": 3.492822413536036,
+      "tokens_seen": 1779499008
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004653747392071899,
+      "loss": 1.3836,
+      "theoretical_loss": 3.4928006469123956,
+      "tokens_seen": 1779630080
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046533461723639863,
+      "loss": 1.3687,
+      "theoretical_loss": 3.4927788823406773,
+      "tokens_seen": 1779761152
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004652944952656075,
+      "loss": 1.3278,
+      "theoretical_loss": 3.492757119820536,
+      "tokens_seen": 1779892224
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046525437329481624,
+      "loss": 1.4656,
+      "theoretical_loss": 3.4927353593516264,
+      "tokens_seen": 1780023296
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046521425132402505,
+      "loss": 1.411,
+      "theoretical_loss": 3.4927136009336057,
+      "tokens_seen": 1780154368
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046517412935323386,
+      "loss": 1.3895,
+      "theoretical_loss": 3.492691844566129,
+      "tokens_seen": 1780285440
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046513400738244266,
+      "loss": 1.3516,
+      "theoretical_loss": 3.492670090248852,
+      "tokens_seen": 1780416512
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004650938854116514,
+      "loss": 1.315,
+      "theoretical_loss": 3.4926483379814304,
+      "tokens_seen": 1780547584
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004650537634408602,
+      "loss": 1.4452,
+      "theoretical_loss": 3.4926265877635205,
+      "tokens_seen": 1780678656
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000465013641470069,
+      "loss": 1.4551,
+      "theoretical_loss": 3.4926048395947786,
+      "tokens_seen": 1780809728
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004649735194992778,
+      "loss": 1.3336,
+      "theoretical_loss": 3.4925830934748605,
+      "tokens_seen": 1780940800
+    },
+    {
+      "epoch": 0.08,
+      "objective/train/advantage_avg": 0.4912663996219635,
+      "objective/train/docs_used": 650314,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3729509115219116,
+      "objective/train/original_loss": 1.372950792312622,
+      "objective/train/theoretical_loss": 3.492561349403422,
+      "objective/train/tokens_used": 151530976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24169018864631653,
+      "objective/train/value_max": -0.5,
+      "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": 1.442193865776062,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.492561349403422,
+      "tokens_seen": 1781071872
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046493339752848664,
+      "loss": 1.3877,
+      "theoretical_loss": 3.492561349403422,
+      "tokens_seen": 1781071872
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004648932755576954,
+      "loss": 1.3924,
+      "theoretical_loss": 3.4925396073801203,
+      "tokens_seen": 1781202944
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004648531535869042,
+      "loss": 1.4307,
+      "theoretical_loss": 3.4925178674046107,
+      "tokens_seen": 1781334016
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000464813031616113,
+      "loss": 1.484,
+      "theoretical_loss": 3.4924961294765504,
+      "tokens_seen": 1781465088
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046477290964532176,
+      "loss": 1.3833,
+      "theoretical_loss": 3.4924743935955966,
+      "tokens_seen": 1781596160
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046473278767453056,
+      "loss": 1.4096,
+      "theoretical_loss": 3.4924526597614047,
+      "tokens_seen": 1781727232
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046469266570373937,
+      "loss": 1.328,
+      "theoretical_loss": 3.492430927973632,
+      "tokens_seen": 1781858304
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004646525437329482,
+      "loss": 1.5021,
+      "theoretical_loss": 3.492409198231935,
+      "tokens_seen": 1781989376
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000464612421762157,
+      "loss": 1.4351,
+      "theoretical_loss": 3.492387470535971,
+      "tokens_seen": 1782120448
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004645722997913658,
+      "loss": 1.3862,
+      "theoretical_loss": 3.4923657448853964,
+      "tokens_seen": 1782251520
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046453217782057454,
+      "loss": 1.3937,
+      "theoretical_loss": 3.492344021279869,
+      "tokens_seen": 1782382592
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046449205584978335,
+      "loss": 1.4512,
+      "theoretical_loss": 3.4923222997190457,
+      "tokens_seen": 1782513664
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046445193387899215,
+      "loss": 1.3934,
+      "theoretical_loss": 3.492300580202583,
+      "tokens_seen": 1782644736
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004644118119082009,
+      "loss": 1.3862,
+      "theoretical_loss": 3.4922788627301387,
+      "tokens_seen": 1782775808
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046437168993740976,
+      "loss": 1.3396,
+      "theoretical_loss": 3.4922571473013706,
+      "tokens_seen": 1782906880
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004643315679666185,
+      "loss": 1.3623,
+      "theoretical_loss": 3.492235433915936,
+      "tokens_seen": 1783037952
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004642914459958273,
+      "loss": 1.322,
+      "theoretical_loss": 3.492213722573492,
+      "tokens_seen": 1783169024
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046425132402503613,
+      "loss": 1.3992,
+      "theoretical_loss": 3.4921920132736965,
+      "tokens_seen": 1783300096
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046421120205424494,
+      "loss": 1.365,
+      "theoretical_loss": 3.492170306016207,
+      "tokens_seen": 1783431168
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004641710800834537,
+      "loss": 1.3772,
+      "theoretical_loss": 3.4921486008006815,
+      "tokens_seen": 1783562240
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004641309581126625,
+      "loss": 1.3885,
+      "theoretical_loss": 3.492126897626778,
+      "tokens_seen": 1783693312
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004640908361418713,
+      "loss": 1.4758,
+      "theoretical_loss": 3.4921051964941543,
+      "tokens_seen": 1783824384
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046405071417108005,
+      "loss": 1.391,
+      "theoretical_loss": 3.4920834974024686,
+      "tokens_seen": 1783955456
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004640105922002889,
+      "loss": 1.5249,
+      "theoretical_loss": 3.4920618003513786,
+      "tokens_seen": 1784086528
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046397047022949767,
+      "loss": 1.389,
+      "theoretical_loss": 3.4920401053405428,
+      "tokens_seen": 1784217600
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004639303482587065,
+      "loss": 1.4134,
+      "theoretical_loss": 3.492018412369619,
+      "tokens_seen": 1784348672
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004638902262879153,
+      "loss": 1.4559,
+      "theoretical_loss": 3.491996721438267,
+      "tokens_seen": 1784479744
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046385010431712403,
+      "loss": 1.4851,
+      "theoretical_loss": 3.4919750325461436,
+      "tokens_seen": 1784610816
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004638099823463329,
+      "loss": 1.4535,
+      "theoretical_loss": 3.491953345692908,
+      "tokens_seen": 1784741888
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046376986037554164,
+      "loss": 1.5221,
+      "theoretical_loss": 3.491931660878219,
+      "tokens_seen": 1784872960
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046372973840475045,
+      "loss": 1.4102,
+      "theoretical_loss": 3.4919099781017353,
+      "tokens_seen": 1785004032
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046368961643395926,
+      "loss": 1.5223,
+      "theoretical_loss": 3.491888297363116,
+      "tokens_seen": 1785135104
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046364949446316806,
+      "loss": 1.4652,
+      "theoretical_loss": 3.4918666186620184,
+      "tokens_seen": 1785266176
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004636093724923768,
+      "loss": 1.3631,
+      "theoretical_loss": 3.491844941998103,
+      "tokens_seen": 1785397248
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004635692505215856,
+      "loss": 1.469,
+      "theoretical_loss": 3.491823267371028,
+      "tokens_seen": 1785528320
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004635291285507944,
+      "loss": 1.4515,
+      "theoretical_loss": 3.4918015947804535,
+      "tokens_seen": 1785659392
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004634890065800032,
+      "loss": 1.3234,
+      "theoretical_loss": 3.4917799242260372,
+      "tokens_seen": 1785790464
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046344888460921204,
+      "loss": 1.3705,
+      "theoretical_loss": 3.49175825570744,
+      "tokens_seen": 1785921536
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004634087626384208,
+      "loss": 1.4018,
+      "theoretical_loss": 3.4917365892243204,
+      "tokens_seen": 1786052608
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046336864066762965,
+      "loss": 1.3161,
+      "theoretical_loss": 3.4917149247763377,
+      "tokens_seen": 1786183680
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004633285186968384,
+      "loss": 1.4149,
+      "theoretical_loss": 3.4916932623631514,
+      "tokens_seen": 1786314752
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046328839672604716,
+      "loss": 1.3928,
+      "theoretical_loss": 3.4916716019844216,
+      "tokens_seen": 1786445824
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000463248274755256,
+      "loss": 1.3512,
+      "theoretical_loss": 3.491649943639808,
+      "tokens_seen": 1786576896
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046320815278446477,
+      "loss": 1.4882,
+      "theoretical_loss": 3.49162828732897,
+      "tokens_seen": 1786707968
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004631680308136736,
+      "loss": 1.522,
+      "theoretical_loss": 3.491606633051567,
+      "tokens_seen": 1786839040
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004631279088428824,
+      "loss": 1.3754,
+      "theoretical_loss": 3.49158498080726,
+      "tokens_seen": 1786970112
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004630877868720912,
+      "loss": 1.3437,
+      "theoretical_loss": 3.4915633305957083,
+      "tokens_seen": 1787101184
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046304766490129994,
+      "loss": 1.3892,
+      "theoretical_loss": 3.4915416824165724,
+      "tokens_seen": 1787232256
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046300754293050875,
+      "loss": 1.4086,
+      "theoretical_loss": 3.491520036269512,
+      "tokens_seen": 1787363328
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046296742095971755,
+      "loss": 1.3771,
+      "theoretical_loss": 3.4914983921541873,
+      "tokens_seen": 1787494400
+    },
+    {
+      "epoch": 0.08,
+      "objective/train/advantage_avg": 0.48783355951309204,
+      "objective/train/docs_used": 652772,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3736180067062378,
+      "objective/train/original_loss": 1.3736178874969482,
+      "objective/train/theoretical_loss": 3.4914767500702597,
+      "objective/train/tokens_used": 158084576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.23931410908699036,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0499999523162842,
+      "objective/train/weighted_lm_loss": 1.4424537420272827,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4914767500702597,
+      "tokens_seen": 1787625472
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004629272989889263,
+      "loss": 1.4772,
+      "theoretical_loss": 3.4914767500702597,
+      "tokens_seen": 1787625472
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046288717701813516,
+      "loss": 1.395,
+      "theoretical_loss": 3.4914551100173883,
+      "tokens_seen": 1787756544
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004628470550473439,
+      "loss": 1.4367,
+      "theoretical_loss": 3.491433471995234,
+      "tokens_seen": 1787887616
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004628069330765527,
+      "loss": 1.4672,
+      "theoretical_loss": 3.491411836003458,
+      "tokens_seen": 1788018688
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046276681110576153,
+      "loss": 1.4195,
+      "theoretical_loss": 3.4913902020417202,
+      "tokens_seen": 1788149760
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046272668913497034,
+      "loss": 1.3587,
+      "theoretical_loss": 3.4913685701096817,
+      "tokens_seen": 1788280832
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046268656716417914,
+      "loss": 1.3721,
+      "theoretical_loss": 3.4913469402070034,
+      "tokens_seen": 1788411904
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004626464451933879,
+      "loss": 1.3191,
+      "theoretical_loss": 3.491325312333346,
+      "tokens_seen": 1788542976
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004626063232225967,
+      "loss": 1.399,
+      "theoretical_loss": 3.4913036864883704,
+      "tokens_seen": 1788674048
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004625662012518055,
+      "loss": 1.3488,
+      "theoretical_loss": 3.491282062671738,
+      "tokens_seen": 1788805120
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004625260792810143,
+      "loss": 1.4584,
+      "theoretical_loss": 3.49126044088311,
+      "tokens_seen": 1788936192
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046248595731022306,
+      "loss": 1.375,
+      "theoretical_loss": 3.4912388211221472,
+      "tokens_seen": 1789067264
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004624458353394319,
+      "loss": 1.3887,
+      "theoretical_loss": 3.4912172033885116,
+      "tokens_seen": 1789198336
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004624057133686407,
+      "loss": 1.3422,
+      "theoretical_loss": 3.491195587681864,
+      "tokens_seen": 1789329408
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046236559139784943,
+      "loss": 1.3447,
+      "theoretical_loss": 3.4911739740018657,
+      "tokens_seen": 1789460480
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0004623254694270583,
+      "loss": 1.3679,
+      "theoretical_loss": 3.4911523623481786,
+      "tokens_seen": 1789591552
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046228534745626704,
+      "loss": 1.3332,
+      "theoretical_loss": 3.4911307527204647,
+      "tokens_seen": 1789722624
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046224522548547585,
+      "loss": 1.4201,
+      "theoretical_loss": 3.4911091451183847,
+      "tokens_seen": 1789853696
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046220510351468465,
+      "loss": 1.3483,
+      "theoretical_loss": 3.491087539541602,
+      "tokens_seen": 1789984768
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00046216498154389346,
+      "loss": 1.3046,
+      "theoretical_loss": 3.491065935989777,
+      "tokens_seen": 1790115840
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00046212485957310227,
+      "loss": 1.4567,
+      "theoretical_loss": 3.491044334462572,
+      "tokens_seen": 1790246912
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000462084737602311,
+      "loss": 1.4978,
+      "theoretical_loss": 3.4910227349596497,
+      "tokens_seen": 1790377984
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004620446156315198,
+      "loss": 1.3076,
+      "theoretical_loss": 3.4910011374806715,
+      "tokens_seen": 1790509056
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00046200449366072863,
+      "loss": 1.4552,
+      "theoretical_loss": 3.4909795420253,
+      "tokens_seen": 1790640128
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00046196437168993744,
+      "loss": 1.333,
+      "theoretical_loss": 3.4909579485931976,
+      "tokens_seen": 1790771200
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004619242497191462,
+      "loss": 1.4726,
+      "theoretical_loss": 3.490936357184026,
+      "tokens_seen": 1790902272
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00046188412774835505,
+      "loss": 1.3395,
+      "theoretical_loss": 3.490914767797449,
+      "tokens_seen": 1791033344
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004618440057775638,
+      "loss": 1.4167,
+      "theoretical_loss": 3.4908931804331274,
+      "tokens_seen": 1791164416
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00046180388380677256,
+      "loss": 1.3863,
+      "theoretical_loss": 3.4908715950907245,
+      "tokens_seen": 1791295488
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004617637618359814,
+      "loss": 1.3023,
+      "theoretical_loss": 3.490850011769904,
+      "tokens_seen": 1791426560
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00046172363986519017,
+      "loss": 1.2854,
+      "theoretical_loss": 3.4908284304703274,
+      "tokens_seen": 1791557632
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000461683517894399,
+      "loss": 1.3906,
+      "theoretical_loss": 3.490806851191658,
+      "tokens_seen": 1791688704
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004616433959236078,
+      "loss": 1.4203,
+      "theoretical_loss": 3.4907852739335588,
+      "tokens_seen": 1791819776
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004616032739528166,
+      "loss": 1.372,
+      "theoretical_loss": 3.4907636986956927,
+      "tokens_seen": 1791950848
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00046156315198202534,
+      "loss": 1.4511,
+      "theoretical_loss": 3.4907421254777233,
+      "tokens_seen": 1792081920
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00046152303001123415,
+      "loss": 1.298,
+      "theoretical_loss": 3.4907205542793127,
+      "tokens_seen": 1792212992
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00046148290804044295,
+      "loss": 1.457,
+      "theoretical_loss": 3.490698985100125,
+      "tokens_seen": 1792344064
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00046144278606965176,
+      "loss": 1.3956,
+      "theoretical_loss": 3.4906774179398234,
+      "tokens_seen": 1792475136
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00046140266409886056,
+      "loss": 1.2974,
+      "theoretical_loss": 3.4906558527980716,
+      "tokens_seen": 1792606208
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004613625421280693,
+      "loss": 1.381,
+      "theoretical_loss": 3.4906342896745324,
+      "tokens_seen": 1792737280
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004613224201572782,
+      "loss": 1.3598,
+      "theoretical_loss": 3.4906127285688697,
+      "tokens_seen": 1792868352
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00046128229818648693,
+      "loss": 1.3677,
+      "theoretical_loss": 3.490591169480747,
+      "tokens_seen": 1792999424
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004612421762156957,
+      "loss": 1.3893,
+      "theoretical_loss": 3.490569612409829,
+      "tokens_seen": 1793130496
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00046120205424490454,
+      "loss": 1.3018,
+      "theoretical_loss": 3.4905480573557783,
+      "tokens_seen": 1793261568
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004611619322741133,
+      "loss": 1.3812,
+      "theoretical_loss": 3.490526504318259,
+      "tokens_seen": 1793392640
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004611218103033221,
+      "loss": 1.3555,
+      "theoretical_loss": 3.490504953296936,
+      "tokens_seen": 1793523712
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004610816883325309,
+      "loss": 1.4616,
+      "theoretical_loss": 3.490483404291473,
+      "tokens_seen": 1793654784
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004610415663617397,
+      "loss": 1.4194,
+      "theoretical_loss": 3.4904618573015336,
+      "tokens_seen": 1793785856
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00046100144439094846,
+      "loss": 1.3661,
+      "theoretical_loss": 3.490440312326782,
+      "tokens_seen": 1793916928
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004609613224201573,
+      "loss": 1.3207,
+      "theoretical_loss": 3.490418769366883,
+      "tokens_seen": 1794048000
+    },
+    {
+      "epoch": 0.09,
+      "objective/train/advantage_avg": 0.4925491511821747,
+      "objective/train/docs_used": 655001,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3235520124435425,
+      "objective/train/original_loss": 1.323551893234253,
+      "objective/train/theoretical_loss": 3.4903972284215015,
+      "objective/train/tokens_used": 164638176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24312761425971985,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504910945892334,
+      "objective/train/weighted_lm_loss": 1.3904740810394287,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4903972284215015,
+      "tokens_seen": 1794179072
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004609212004493661,
+      "loss": 1.4141,
+      "theoretical_loss": 3.4903972284215015,
+      "tokens_seen": 1794179072
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00046088107847857483,
+      "loss": 1.3348,
+      "theoretical_loss": 3.4903756894903006,
+      "tokens_seen": 1794310144
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004608409565077837,
+      "loss": 1.2953,
+      "theoretical_loss": 3.4903541525729462,
+      "tokens_seen": 1794441216
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00046080083453699244,
+      "loss": 1.3655,
+      "theoretical_loss": 3.4903326176691016,
+      "tokens_seen": 1794572288
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004607607125662013,
+      "loss": 1.4641,
+      "theoretical_loss": 3.4903110847784324,
+      "tokens_seen": 1794703360
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00046072059059541005,
+      "loss": 1.3793,
+      "theoretical_loss": 3.4902895539006034,
+      "tokens_seen": 1794834432
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00046068046862461886,
+      "loss": 1.3882,
+      "theoretical_loss": 3.4902680250352796,
+      "tokens_seen": 1794965504
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00046064034665382767,
+      "loss": 1.3405,
+      "theoretical_loss": 3.490246498182125,
+      "tokens_seen": 1795096576
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004606002246830364,
+      "loss": 1.3946,
+      "theoretical_loss": 3.4902249733408057,
+      "tokens_seen": 1795227648
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004605601027122452,
+      "loss": 1.2954,
+      "theoretical_loss": 3.490203450510986,
+      "tokens_seen": 1795358720
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00046051998074145403,
+      "loss": 1.3489,
+      "theoretical_loss": 3.4901819296923318,
+      "tokens_seen": 1795489792
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00046047985877066284,
+      "loss": 1.2574,
+      "theoretical_loss": 3.490160410884508,
+      "tokens_seen": 1795620864
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004604397367998716,
+      "loss": 1.3662,
+      "theoretical_loss": 3.49013889408718,
+      "tokens_seen": 1795751936
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00046039961482908045,
+      "loss": 1.4807,
+      "theoretical_loss": 3.4901173793000133,
+      "tokens_seen": 1795883008
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004603594928582892,
+      "loss": 1.444,
+      "theoretical_loss": 3.4900958665226733,
+      "tokens_seen": 1796014080
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00046031937088749795,
+      "loss": 1.4149,
+      "theoretical_loss": 3.4900743557548255,
+      "tokens_seen": 1796145152
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004602792489167068,
+      "loss": 1.3275,
+      "theoretical_loss": 3.490052846996136,
+      "tokens_seen": 1796276224
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00046023912694591557,
+      "loss": 1.3151,
+      "theoretical_loss": 3.49003134024627,
+      "tokens_seen": 1796407296
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00046019900497512443,
+      "loss": 1.3856,
+      "theoretical_loss": 3.4900098355048934,
+      "tokens_seen": 1796538368
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004601588830043332,
+      "loss": 1.4538,
+      "theoretical_loss": 3.4899883327716728,
+      "tokens_seen": 1796669440
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000460118761033542,
+      "loss": 1.3708,
+      "theoretical_loss": 3.4899668320462736,
+      "tokens_seen": 1796800512
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004600786390627508,
+      "loss": 1.3659,
+      "theoretical_loss": 3.489945333328362,
+      "tokens_seen": 1796931584
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00046003851709195954,
+      "loss": 1.3644,
+      "theoretical_loss": 3.489923836617604,
+      "tokens_seen": 1797062656
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045999839512116835,
+      "loss": 1.4878,
+      "theoretical_loss": 3.489902341913666,
+      "tokens_seen": 1797193728
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045995827315037716,
+      "loss": 1.4066,
+      "theoretical_loss": 3.489880849216215,
+      "tokens_seen": 1797324800
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045991815117958596,
+      "loss": 1.4032,
+      "theoretical_loss": 3.4898593585249156,
+      "tokens_seen": 1797455872
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004598780292087947,
+      "loss": 1.433,
+      "theoretical_loss": 3.489837869839436,
+      "tokens_seen": 1797586944
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004598379072380036,
+      "loss": 1.3226,
+      "theoretical_loss": 3.489816383159442,
+      "tokens_seen": 1797718016
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045979778526721233,
+      "loss": 1.4067,
+      "theoretical_loss": 3.4897948984846003,
+      "tokens_seen": 1797849088
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004597576632964211,
+      "loss": 1.4423,
+      "theoretical_loss": 3.489773415814578,
+      "tokens_seen": 1797980160
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045971754132562994,
+      "loss": 1.3848,
+      "theoretical_loss": 3.489751935149041,
+      "tokens_seen": 1798111232
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004596774193548387,
+      "loss": 1.3452,
+      "theoretical_loss": 3.4897304564876572,
+      "tokens_seen": 1798242304
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004596372973840475,
+      "loss": 1.4638,
+      "theoretical_loss": 3.489708979830093,
+      "tokens_seen": 1798373376
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004595971754132563,
+      "loss": 1.3964,
+      "theoretical_loss": 3.4896875051760152,
+      "tokens_seen": 1798504448
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004595570534424651,
+      "loss": 1.4449,
+      "theoretical_loss": 3.4896660325250917,
+      "tokens_seen": 1798635520
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004595169314716739,
+      "loss": 1.3501,
+      "theoretical_loss": 3.4896445618769887,
+      "tokens_seen": 1798766592
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004594768095008827,
+      "loss": 1.415,
+      "theoretical_loss": 3.4896230932313745,
+      "tokens_seen": 1798897664
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004594366875300915,
+      "loss": 1.3526,
+      "theoretical_loss": 3.4896016265879153,
+      "tokens_seen": 1799028736
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004593965655593003,
+      "loss": 1.3284,
+      "theoretical_loss": 3.4895801619462796,
+      "tokens_seen": 1799159808
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004593564435885091,
+      "loss": 1.3149,
+      "theoretical_loss": 3.4895586993061345,
+      "tokens_seen": 1799290880
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045931632161771784,
+      "loss": 1.3745,
+      "theoretical_loss": 3.4895372386671477,
+      "tokens_seen": 1799421952
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004592761996469267,
+      "loss": 1.3746,
+      "theoretical_loss": 3.489515780028986,
+      "tokens_seen": 1799553024
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045923607767613545,
+      "loss": 1.2985,
+      "theoretical_loss": 3.4894943233913187,
+      "tokens_seen": 1799684096
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045919595570534426,
+      "loss": 1.394,
+      "theoretical_loss": 3.489472868753813,
+      "tokens_seen": 1799815168
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045915583373455307,
+      "loss": 1.4108,
+      "theoretical_loss": 3.4894514161161356,
+      "tokens_seen": 1799946240
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004591157117637618,
+      "loss": 1.3693,
+      "theoretical_loss": 3.489429965477956,
+      "tokens_seen": 1800077312
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004590755897929706,
+      "loss": 1.3306,
+      "theoretical_loss": 3.489408516838942,
+      "tokens_seen": 1800208384
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045903546782217943,
+      "loss": 1.4662,
+      "theoretical_loss": 3.4893870701987613,
+      "tokens_seen": 1800339456
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045899534585138824,
+      "loss": 1.462,
+      "theoretical_loss": 3.489365625557082,
+      "tokens_seen": 1800470528
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045895522388059704,
+      "loss": 1.3938,
+      "theoretical_loss": 3.489344182913573,
+      "tokens_seen": 1800601600
+    },
+    {
+      "epoch": 0.09,
+      "objective/train/advantage_avg": 0.49205124378204346,
+      "objective/train/docs_used": 657521,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3897008895874023,
+      "objective/train/original_loss": 1.3897008895874023,
+      "objective/train/theoretical_loss": 3.4893227422679027,
+      "objective/train/tokens_used": 171191776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24267229437828064,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504387617111206,
+      "objective/train/weighted_lm_loss": 1.4597222805023193,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4893227422679027,
+      "tokens_seen": 1800732672
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045891510190980585,
+      "loss": 1.4666,
+      "theoretical_loss": 3.4893227422679027,
+      "tokens_seen": 1800732672
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004588749799390146,
+      "loss": 1.3576,
+      "theoretical_loss": 3.4893013036197384,
+      "tokens_seen": 1800863744
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004588348579682234,
+      "loss": 1.3903,
+      "theoretical_loss": 3.4892798669687504,
+      "tokens_seen": 1800994816
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004587947359974322,
+      "loss": 1.3719,
+      "theoretical_loss": 3.489258432314606,
+      "tokens_seen": 1801125888
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045875461402664097,
+      "loss": 1.4091,
+      "theoretical_loss": 3.4892369996569745,
+      "tokens_seen": 1801256960
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045871449205584983,
+      "loss": 1.3348,
+      "theoretical_loss": 3.489215568995524,
+      "tokens_seen": 1801388032
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004586743700850586,
+      "loss": 1.4173,
+      "theoretical_loss": 3.489194140329925,
+      "tokens_seen": 1801519104
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004586342481142674,
+      "loss": 1.418,
+      "theoretical_loss": 3.489172713659844,
+      "tokens_seen": 1801650176
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004585941261434762,
+      "loss": 1.3744,
+      "theoretical_loss": 3.489151288984952,
+      "tokens_seen": 1801781248
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045855400417268494,
+      "loss": 1.4436,
+      "theoretical_loss": 3.4891298663049177,
+      "tokens_seen": 1801912320
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045851388220189375,
+      "loss": 1.4099,
+      "theoretical_loss": 3.4891084456194092,
+      "tokens_seen": 1802043392
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045847376023110256,
+      "loss": 1.3591,
+      "theoretical_loss": 3.4890870269280967,
+      "tokens_seen": 1802174464
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045843363826031136,
+      "loss": 1.3032,
+      "theoretical_loss": 3.4890656102306496,
+      "tokens_seen": 1802305536
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004583935162895201,
+      "loss": 1.2776,
+      "theoretical_loss": 3.489044195526737,
+      "tokens_seen": 1802436608
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000458353394318729,
+      "loss": 1.4757,
+      "theoretical_loss": 3.4890227828160283,
+      "tokens_seen": 1802567680
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045831327234793773,
+      "loss": 1.4042,
+      "theoretical_loss": 3.489001372098193,
+      "tokens_seen": 1802698752
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045827315037714653,
+      "loss": 1.4169,
+      "theoretical_loss": 3.4889799633729015,
+      "tokens_seen": 1802829824
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045823302840635534,
+      "loss": 1.3211,
+      "theoretical_loss": 3.4889585566398225,
+      "tokens_seen": 1802960896
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004581929064355641,
+      "loss": 1.445,
+      "theoretical_loss": 3.488937151898627,
+      "tokens_seen": 1803091968
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045815278446477295,
+      "loss": 1.4012,
+      "theoretical_loss": 3.4889157491489833,
+      "tokens_seen": 1803223040
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004581126624939817,
+      "loss": 1.4272,
+      "theoretical_loss": 3.488894348390563,
+      "tokens_seen": 1803354112
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004580725405231905,
+      "loss": 1.364,
+      "theoretical_loss": 3.4888729496230346,
+      "tokens_seen": 1803485184
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004580324185523993,
+      "loss": 1.3407,
+      "theoretical_loss": 3.4888515528460693,
+      "tokens_seen": 1803616256
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004579922965816081,
+      "loss": 1.3262,
+      "theoretical_loss": 3.4888301580593373,
+      "tokens_seen": 1803747328
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004579521746108169,
+      "loss": 1.4088,
+      "theoretical_loss": 3.488808765262508,
+      "tokens_seen": 1803878400
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004579120526400257,
+      "loss": 1.448,
+      "theoretical_loss": 3.4887873744552524,
+      "tokens_seen": 1804009472
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004578719306692345,
+      "loss": 1.3589,
+      "theoretical_loss": 3.488765985637241,
+      "tokens_seen": 1804140544
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045783180869844324,
+      "loss": 1.3807,
+      "theoretical_loss": 3.488744598808144,
+      "tokens_seen": 1804271616
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004577916867276521,
+      "loss": 1.3253,
+      "theoretical_loss": 3.4887232139676323,
+      "tokens_seen": 1804402688
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045775156475686085,
+      "loss": 1.3669,
+      "theoretical_loss": 3.4887018311153764,
+      "tokens_seen": 1804533760
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004577114427860697,
+      "loss": 1.3644,
+      "theoretical_loss": 3.4886804502510467,
+      "tokens_seen": 1804664832
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045767132081527847,
+      "loss": 1.4264,
+      "theoretical_loss": 3.4886590713743146,
+      "tokens_seen": 1804795904
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004576311988444872,
+      "loss": 1.2801,
+      "theoretical_loss": 3.4886376944848507,
+      "tokens_seen": 1804926976
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004575910768736961,
+      "loss": 1.3641,
+      "theoretical_loss": 3.4886163195823263,
+      "tokens_seen": 1805058048
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045755095490290483,
+      "loss": 1.3585,
+      "theoretical_loss": 3.488594946666412,
+      "tokens_seen": 1805189120
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045751083293211364,
+      "loss": 1.3523,
+      "theoretical_loss": 3.4885735757367793,
+      "tokens_seen": 1805320192
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045747071096132244,
+      "loss": 1.4357,
+      "theoretical_loss": 3.488552206793099,
+      "tokens_seen": 1805451264
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045743058899053125,
+      "loss": 1.3468,
+      "theoretical_loss": 3.4885308398350428,
+      "tokens_seen": 1805582336
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045739046701974,
+      "loss": 1.4144,
+      "theoretical_loss": 3.4885094748622816,
+      "tokens_seen": 1805713408
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004573503450489488,
+      "loss": 1.4181,
+      "theoretical_loss": 3.4884881118744877,
+      "tokens_seen": 1805844480
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004573102230781576,
+      "loss": 1.421,
+      "theoretical_loss": 3.488466750871332,
+      "tokens_seen": 1805975552
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045727010110736637,
+      "loss": 1.3479,
+      "theoretical_loss": 3.488445391852486,
+      "tokens_seen": 1806106624
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00045722997913657523,
+      "loss": 1.4099,
+      "theoretical_loss": 3.488424034817622,
+      "tokens_seen": 1806237696
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000457189857165784,
+      "loss": 1.3085,
+      "theoretical_loss": 3.488402679766411,
+      "tokens_seen": 1806368768
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004571497351949928,
+      "loss": 1.4751,
+      "theoretical_loss": 3.4883813266985246,
+      "tokens_seen": 1806499840
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0004571096132242016,
+      "loss": 1.4085,
+      "theoretical_loss": 3.488359975613636,
+      "tokens_seen": 1806630912
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045706949125341034,
+      "loss": 1.4334,
+      "theoretical_loss": 3.4883386265114167,
+      "tokens_seen": 1806761984
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004570293692826192,
+      "loss": 1.3661,
+      "theoretical_loss": 3.4883172793915387,
+      "tokens_seen": 1806893056
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045698924731182796,
+      "loss": 1.3703,
+      "theoretical_loss": 3.4882959342536735,
+      "tokens_seen": 1807024128
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045694912534103676,
+      "loss": 1.3639,
+      "theoretical_loss": 3.488274591097494,
+      "tokens_seen": 1807155200
+    },
+    {
+      "epoch": 0.1,
+      "objective/train/advantage_avg": 0.49131128191947937,
+      "objective/train/docs_used": 659683,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.283807635307312,
+      "objective/train/original_loss": 1.283807396888733,
+      "objective/train/theoretical_loss": 3.4882532499226726,
+      "objective/train/tokens_used": 177745376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2418506145477295,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503605604171753,
+      "objective/train/weighted_lm_loss": 1.3484594821929932,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4882532499226726,
+      "tokens_seen": 1807286272
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045690900337024557,
+      "loss": 1.3755,
+      "theoretical_loss": 3.4882532499226726,
+      "tokens_seen": 1807286272
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004568688813994544,
+      "loss": 1.4385,
+      "theoretical_loss": 3.4882319107288815,
+      "tokens_seen": 1807417344
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045682875942866313,
+      "loss": 1.4172,
+      "theoretical_loss": 3.4882105735157927,
+      "tokens_seen": 1807548416
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045678863745787193,
+      "loss": 1.4269,
+      "theoretical_loss": 3.4881892382830797,
+      "tokens_seen": 1807679488
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045674851548708074,
+      "loss": 1.3757,
+      "theoretical_loss": 3.488167905030414,
+      "tokens_seen": 1807810560
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004567083935162895,
+      "loss": 1.3513,
+      "theoretical_loss": 3.4881465737574695,
+      "tokens_seen": 1807941632
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045666827154549835,
+      "loss": 1.3722,
+      "theoretical_loss": 3.488125244463918,
+      "tokens_seen": 1808072704
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004566281495747071,
+      "loss": 1.4493,
+      "theoretical_loss": 3.488103917149433,
+      "tokens_seen": 1808203776
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004565880276039159,
+      "loss": 1.3919,
+      "theoretical_loss": 3.4880825918136873,
+      "tokens_seen": 1808334848
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004565479056331247,
+      "loss": 1.3776,
+      "theoretical_loss": 3.4880612684563532,
+      "tokens_seen": 1808465920
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045650778366233347,
+      "loss": 1.4485,
+      "theoretical_loss": 3.488039947077105,
+      "tokens_seen": 1808596992
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004564676616915423,
+      "loss": 1.3966,
+      "theoretical_loss": 3.488018627675615,
+      "tokens_seen": 1808728064
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004564275397207511,
+      "loss": 1.3783,
+      "theoretical_loss": 3.4879973102515565,
+      "tokens_seen": 1808859136
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004563874177499599,
+      "loss": 1.3835,
+      "theoretical_loss": 3.4879759948046027,
+      "tokens_seen": 1808990208
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004563472957791687,
+      "loss": 1.3948,
+      "theoretical_loss": 3.487954681334428,
+      "tokens_seen": 1809121280
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004563071738083775,
+      "loss": 1.4036,
+      "theoretical_loss": 3.4879333698407047,
+      "tokens_seen": 1809252352
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045626705183758625,
+      "loss": 1.4434,
+      "theoretical_loss": 3.487912060323107,
+      "tokens_seen": 1809383424
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004562269298667951,
+      "loss": 1.3572,
+      "theoretical_loss": 3.4878907527813086,
+      "tokens_seen": 1809514496
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045618680789600387,
+      "loss": 1.4478,
+      "theoretical_loss": 3.4878694472149823,
+      "tokens_seen": 1809645568
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004561466859252126,
+      "loss": 1.3794,
+      "theoretical_loss": 3.487848143623803,
+      "tokens_seen": 1809776640
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004561065639544215,
+      "loss": 1.3571,
+      "theoretical_loss": 3.4878268420074443,
+      "tokens_seen": 1809907712
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045606644198363023,
+      "loss": 1.3647,
+      "theoretical_loss": 3.48780554236558,
+      "tokens_seen": 1810038784
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045602632001283904,
+      "loss": 1.3836,
+      "theoretical_loss": 3.4877842446978833,
+      "tokens_seen": 1810169856
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045598619804204784,
+      "loss": 1.3899,
+      "theoretical_loss": 3.48776294900403,
+      "tokens_seen": 1810300928
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045594607607125665,
+      "loss": 1.3787,
+      "theoretical_loss": 3.487741655283693,
+      "tokens_seen": 1810432000
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004559059541004654,
+      "loss": 1.3635,
+      "theoretical_loss": 3.487720363536546,
+      "tokens_seen": 1810563072
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004558658321296742,
+      "loss": 1.4199,
+      "theoretical_loss": 3.4876990737622657,
+      "tokens_seen": 1810694144
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.000455825710158883,
+      "loss": 1.4297,
+      "theoretical_loss": 3.487677785960524,
+      "tokens_seen": 1810825216
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045578558818809177,
+      "loss": 1.3671,
+      "theoretical_loss": 3.4876565001309965,
+      "tokens_seen": 1810956288
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004557454662173006,
+      "loss": 1.4268,
+      "theoretical_loss": 3.4876352162733584,
+      "tokens_seen": 1811087360
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004557053442465094,
+      "loss": 1.4166,
+      "theoretical_loss": 3.4876139343872827,
+      "tokens_seen": 1811218432
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045566522227571824,
+      "loss": 1.417,
+      "theoretical_loss": 3.4875926544724454,
+      "tokens_seen": 1811349504
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.000455625100304927,
+      "loss": 1.4273,
+      "theoretical_loss": 3.4875713765285212,
+      "tokens_seen": 1811480576
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045558497833413574,
+      "loss": 1.4137,
+      "theoretical_loss": 3.4875501005551843,
+      "tokens_seen": 1811611648
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004555448563633446,
+      "loss": 1.405,
+      "theoretical_loss": 3.4875288265521105,
+      "tokens_seen": 1811742720
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045550473439255336,
+      "loss": 1.376,
+      "theoretical_loss": 3.4875075545189738,
+      "tokens_seen": 1811873792
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045546461242176216,
+      "loss": 1.4102,
+      "theoretical_loss": 3.48748628445545,
+      "tokens_seen": 1812004864
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045542449045097097,
+      "loss": 1.388,
+      "theoretical_loss": 3.487465016361214,
+      "tokens_seen": 1812135936
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004553843684801798,
+      "loss": 1.3927,
+      "theoretical_loss": 3.4874437502359417,
+      "tokens_seen": 1812267008
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045534424650938853,
+      "loss": 1.3421,
+      "theoretical_loss": 3.4874224860793075,
+      "tokens_seen": 1812398080
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045530412453859733,
+      "loss": 1.373,
+      "theoretical_loss": 3.4874012238909877,
+      "tokens_seen": 1812529152
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045526400256780614,
+      "loss": 1.4442,
+      "theoretical_loss": 3.487379963670657,
+      "tokens_seen": 1812660224
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004552238805970149,
+      "loss": 1.4027,
+      "theoretical_loss": 3.487358705417991,
+      "tokens_seen": 1812791296
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045518375862622375,
+      "loss": 1.371,
+      "theoretical_loss": 3.4873374491326654,
+      "tokens_seen": 1812922368
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004551436366554325,
+      "loss": 1.3411,
+      "theoretical_loss": 3.4873161948143565,
+      "tokens_seen": 1813053440
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045510351468464137,
+      "loss": 1.3511,
+      "theoretical_loss": 3.48729494246274,
+      "tokens_seen": 1813184512
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004550633927138501,
+      "loss": 1.4231,
+      "theoretical_loss": 3.4872736920774914,
+      "tokens_seen": 1813315584
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045502327074305887,
+      "loss": 1.3615,
+      "theoretical_loss": 3.487252443658287,
+      "tokens_seen": 1813446656
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045498314877226773,
+      "loss": 1.3598,
+      "theoretical_loss": 3.487231197204802,
+      "tokens_seen": 1813577728
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004549430268014765,
+      "loss": 1.4255,
+      "theoretical_loss": 3.487209952716713,
+      "tokens_seen": 1813708800
+    },
+    {
+      "epoch": 0.1,
+      "objective/train/advantage_avg": 0.4918607771396637,
+      "objective/train/docs_used": 661946,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.244071125984192,
+      "objective/train/original_loss": 1.2440710067749023,
+      "objective/train/theoretical_loss": 3.4871887101936965,
+      "objective/train/tokens_used": 184298976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24258999526500702,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504192113876343,
+      "objective/train/weighted_lm_loss": 1.3068296909332275,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4871887101936965,
+      "tokens_seen": 1813839872
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004549029048306853,
+      "loss": 1.3029,
+      "theoretical_loss": 3.4871887101936965,
+      "tokens_seen": 1813839872
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004548627828598941,
+      "loss": 1.4936,
+      "theoretical_loss": 3.4871674696354287,
+      "tokens_seen": 1813970944
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004548226608891029,
+      "loss": 1.4304,
+      "theoretical_loss": 3.487146231041585,
+      "tokens_seen": 1814102016
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045478253891831165,
+      "loss": 1.3829,
+      "theoretical_loss": 3.487124994411843,
+      "tokens_seen": 1814233088
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004547424169475205,
+      "loss": 1.3425,
+      "theoretical_loss": 3.487103759745879,
+      "tokens_seen": 1814364160
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045470229497672927,
+      "loss": 1.4601,
+      "theoretical_loss": 3.487082527043369,
+      "tokens_seen": 1814495232
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.000454662173005938,
+      "loss": 1.4824,
+      "theoretical_loss": 3.48706129630399,
+      "tokens_seen": 1814626304
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004546220510351469,
+      "loss": 1.37,
+      "theoretical_loss": 3.4870400675274182,
+      "tokens_seen": 1814757376
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045458192906435563,
+      "loss": 1.4805,
+      "theoretical_loss": 3.4870188407133313,
+      "tokens_seen": 1814888448
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045454180709356444,
+      "loss": 1.3789,
+      "theoretical_loss": 3.486997615861406,
+      "tokens_seen": 1815019520
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045450168512277324,
+      "loss": 1.4339,
+      "theoretical_loss": 3.486976392971318,
+      "tokens_seen": 1815150592
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045446156315198205,
+      "loss": 1.42,
+      "theoretical_loss": 3.486955172042746,
+      "tokens_seen": 1815281664
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045442144118119086,
+      "loss": 1.3904,
+      "theoretical_loss": 3.486933953075366,
+      "tokens_seen": 1815412736
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004543813192103996,
+      "loss": 1.4356,
+      "theoretical_loss": 3.486912736068856,
+      "tokens_seen": 1815543808
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004543411972396084,
+      "loss": 1.3613,
+      "theoretical_loss": 3.4868915210228923,
+      "tokens_seen": 1815674880
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004543010752688172,
+      "loss": 1.3808,
+      "theoretical_loss": 3.486870307937153,
+      "tokens_seen": 1815805952
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.000454260953298026,
+      "loss": 1.3701,
+      "theoretical_loss": 3.4868490968113144,
+      "tokens_seen": 1815937024
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004542208313272348,
+      "loss": 1.3322,
+      "theoretical_loss": 3.4868278876450556,
+      "tokens_seen": 1816068096
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045418070935644364,
+      "loss": 1.3169,
+      "theoretical_loss": 3.486806680438053,
+      "tokens_seen": 1816199168
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004541405873856524,
+      "loss": 1.4063,
+      "theoretical_loss": 3.4867854751899845,
+      "tokens_seen": 1816330240
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045410046541486114,
+      "loss": 1.4537,
+      "theoretical_loss": 3.486764271900528,
+      "tokens_seen": 1816461312
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045406034344407,
+      "loss": 1.3102,
+      "theoretical_loss": 3.4867430705693616,
+      "tokens_seen": 1816592384
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045402022147327876,
+      "loss": 1.3982,
+      "theoretical_loss": 3.486721871196162,
+      "tokens_seen": 1816723456
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045398009950248756,
+      "loss": 1.3972,
+      "theoretical_loss": 3.486700673780608,
+      "tokens_seen": 1816854528
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045393997753169637,
+      "loss": 1.3506,
+      "theoretical_loss": 3.486679478322378,
+      "tokens_seen": 1816985600
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004538998555609052,
+      "loss": 1.478,
+      "theoretical_loss": 3.486658284821149,
+      "tokens_seen": 1817116672
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.000453859733590114,
+      "loss": 1.3784,
+      "theoretical_loss": 3.4866370932765998,
+      "tokens_seen": 1817247744
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045381961161932273,
+      "loss": 1.4172,
+      "theoretical_loss": 3.486615903688409,
+      "tokens_seen": 1817378816
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045377948964853154,
+      "loss": 1.3719,
+      "theoretical_loss": 3.486594716056254,
+      "tokens_seen": 1817509888
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045373936767774035,
+      "loss": 1.3264,
+      "theoretical_loss": 3.4865735303798138,
+      "tokens_seen": 1817640960
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045369924570694915,
+      "loss": 1.3521,
+      "theoretical_loss": 3.486552346658767,
+      "tokens_seen": 1817772032
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004536591237361579,
+      "loss": 1.3172,
+      "theoretical_loss": 3.4865311648927912,
+      "tokens_seen": 1817903104
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045361900176536676,
+      "loss": 1.3662,
+      "theoretical_loss": 3.4865099850815664,
+      "tokens_seen": 1818034176
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004535788797945755,
+      "loss": 1.3052,
+      "theoretical_loss": 3.48648880722477,
+      "tokens_seen": 1818165248
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045353875782378427,
+      "loss": 1.4569,
+      "theoretical_loss": 3.486467631322082,
+      "tokens_seen": 1818296320
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045349863585299313,
+      "loss": 1.4315,
+      "theoretical_loss": 3.4864464573731806,
+      "tokens_seen": 1818427392
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004534585138822019,
+      "loss": 1.3464,
+      "theoretical_loss": 3.486425285377744,
+      "tokens_seen": 1818558464
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004534183919114107,
+      "loss": 1.4535,
+      "theoretical_loss": 3.486404115335453,
+      "tokens_seen": 1818689536
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004533782699406195,
+      "loss": 1.4063,
+      "theoretical_loss": 3.486382947245985,
+      "tokens_seen": 1818820608
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004533381479698283,
+      "loss": 1.4327,
+      "theoretical_loss": 3.48636178110902,
+      "tokens_seen": 1818951680
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045329802599903705,
+      "loss": 1.4588,
+      "theoretical_loss": 3.4863406169242372,
+      "tokens_seen": 1819082752
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004532579040282459,
+      "loss": 1.3944,
+      "theoretical_loss": 3.486319454691315,
+      "tokens_seen": 1819213824
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045321778205745467,
+      "loss": 1.3594,
+      "theoretical_loss": 3.4862982944099343,
+      "tokens_seen": 1819344896
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045317766008666347,
+      "loss": 1.3899,
+      "theoretical_loss": 3.4862771360797735,
+      "tokens_seen": 1819475968
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004531375381158723,
+      "loss": 1.4815,
+      "theoretical_loss": 3.486255979700512,
+      "tokens_seen": 1819607040
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045309741614508103,
+      "loss": 1.3797,
+      "theoretical_loss": 3.48623482527183,
+      "tokens_seen": 1819738112
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004530572941742899,
+      "loss": 1.4142,
+      "theoretical_loss": 3.486213672793408,
+      "tokens_seen": 1819869184
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045301717220349864,
+      "loss": 1.3612,
+      "theoretical_loss": 3.4861925222649233,
+      "tokens_seen": 1820000256
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045297705023270745,
+      "loss": 1.3693,
+      "theoretical_loss": 3.486171373686058,
+      "tokens_seen": 1820131328
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045293692826191626,
+      "loss": 1.4478,
+      "theoretical_loss": 3.4861502270564904,
+      "tokens_seen": 1820262400
+    },
+    {
+      "epoch": 0.1,
+      "objective/train/advantage_avg": 0.4930785000324249,
+      "objective/train/docs_used": 664244,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.374138355255127,
+      "objective/train/original_loss": 1.374138355255127,
+      "objective/train/theoretical_loss": 3.486129082375902,
+      "objective/train/tokens_used": 190852576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24354740977287292,
+      "objective/train/value_max": -0.5,
+      "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": 1.4435352087020874,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.486129082375902,
+      "tokens_seen": 1820393472
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.000452896806291125,
+      "loss": 1.3988,
+      "theoretical_loss": 3.486129082375902,
+      "tokens_seen": 1820393472
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004528566843203338,
+      "loss": 1.4661,
+      "theoretical_loss": 3.486107939643972,
+      "tokens_seen": 1820524544
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004528165623495426,
+      "loss": 1.4424,
+      "theoretical_loss": 3.4860867988603803,
+      "tokens_seen": 1820655616
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004527764403787514,
+      "loss": 1.284,
+      "theoretical_loss": 3.486065660024808,
+      "tokens_seen": 1820786688
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004527363184079602,
+      "loss": 1.5066,
+      "theoretical_loss": 3.4860445231369344,
+      "tokens_seen": 1820917760
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045269619643716904,
+      "loss": 1.43,
+      "theoretical_loss": 3.4860233881964406,
+      "tokens_seen": 1821048832
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004526560744663778,
+      "loss": 1.3621,
+      "theoretical_loss": 3.4860022552030068,
+      "tokens_seen": 1821179904
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045261595249558654,
+      "loss": 1.4524,
+      "theoretical_loss": 3.4859811241563134,
+      "tokens_seen": 1821310976
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004525758305247954,
+      "loss": 1.4234,
+      "theoretical_loss": 3.4859599950560414,
+      "tokens_seen": 1821442048
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045253570855400416,
+      "loss": 1.3911,
+      "theoretical_loss": 3.4859388679018704,
+      "tokens_seen": 1821573120
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.000452495586583213,
+      "loss": 1.3788,
+      "theoretical_loss": 3.4859177426934824,
+      "tokens_seen": 1821704192
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045245546461242177,
+      "loss": 1.374,
+      "theoretical_loss": 3.485896619430558,
+      "tokens_seen": 1821835264
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004524153426416306,
+      "loss": 1.4438,
+      "theoretical_loss": 3.485875498112777,
+      "tokens_seen": 1821966336
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004523752206708394,
+      "loss": 1.3483,
+      "theoretical_loss": 3.4858543787398215,
+      "tokens_seen": 1822097408
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045233509870004813,
+      "loss": 1.472,
+      "theoretical_loss": 3.4858332613113725,
+      "tokens_seen": 1822228480
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045229497672925694,
+      "loss": 1.4128,
+      "theoretical_loss": 3.4858121458271105,
+      "tokens_seen": 1822359552
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045225485475846575,
+      "loss": 1.4733,
+      "theoretical_loss": 3.485791032286717,
+      "tokens_seen": 1822490624
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045221473278767455,
+      "loss": 1.4855,
+      "theoretical_loss": 3.4857699206898736,
+      "tokens_seen": 1822621696
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004521746108168833,
+      "loss": 1.4685,
+      "theoretical_loss": 3.4857488110362613,
+      "tokens_seen": 1822752768
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045213448884609216,
+      "loss": 1.3589,
+      "theoretical_loss": 3.485727703325561,
+      "tokens_seen": 1822883840
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0004520943668753009,
+      "loss": 1.4224,
+      "theoretical_loss": 3.485706597557455,
+      "tokens_seen": 1823014912
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00045205424490450967,
+      "loss": 1.3727,
+      "theoretical_loss": 3.4856854937316246,
+      "tokens_seen": 1823145984
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00045201412293371853,
+      "loss": 1.3986,
+      "theoretical_loss": 3.4856643918477515,
+      "tokens_seen": 1823277056
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004519740009629273,
+      "loss": 1.3619,
+      "theoretical_loss": 3.485643291905517,
+      "tokens_seen": 1823408128
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00045193387899213614,
+      "loss": 1.3918,
+      "theoretical_loss": 3.4856221939046037,
+      "tokens_seen": 1823539200
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004518937570213449,
+      "loss": 1.4431,
+      "theoretical_loss": 3.485601097844693,
+      "tokens_seen": 1823670272
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004518536350505537,
+      "loss": 1.4079,
+      "theoretical_loss": 3.4855800037254667,
+      "tokens_seen": 1823801344
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004518135130797625,
+      "loss": 1.4358,
+      "theoretical_loss": 3.4855589115466072,
+      "tokens_seen": 1823932416
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00045177339110897126,
+      "loss": 1.421,
+      "theoretical_loss": 3.4855378213077963,
+      "tokens_seen": 1824063488
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00045173326913818006,
+      "loss": 1.4204,
+      "theoretical_loss": 3.4855167330087165,
+      "tokens_seen": 1824194560
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00045169314716738887,
+      "loss": 1.3462,
+      "theoretical_loss": 3.485495646649049,
+      "tokens_seen": 1824325632
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004516530251965977,
+      "loss": 1.4199,
+      "theoretical_loss": 3.4854745622284775,
+      "tokens_seen": 1824456704
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00045161290322580643,
+      "loss": 1.3629,
+      "theoretical_loss": 3.485453479746684,
+      "tokens_seen": 1824587776
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004515727812550153,
+      "loss": 1.363,
+      "theoretical_loss": 3.48543239920335,
+      "tokens_seen": 1824718848
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00045153265928422404,
+      "loss": 1.2871,
+      "theoretical_loss": 3.4854113205981596,
+      "tokens_seen": 1824849920
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00045149253731343285,
+      "loss": 1.4158,
+      "theoretical_loss": 3.485390243930794,
+      "tokens_seen": 1824980992
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00045145241534264165,
+      "loss": 1.3413,
+      "theoretical_loss": 3.485369169200937,
+      "tokens_seen": 1825112064
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004514122933718504,
+      "loss": 1.4291,
+      "theoretical_loss": 3.485348096408271,
+      "tokens_seen": 1825243136
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004513721714010592,
+      "loss": 1.3758,
+      "theoretical_loss": 3.4853270255524778,
+      "tokens_seen": 1825374208
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.000451332049430268,
+      "loss": 1.4089,
+      "theoretical_loss": 3.485305956633242,
+      "tokens_seen": 1825505280
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004512919274594768,
+      "loss": 1.384,
+      "theoretical_loss": 3.4852848896502455,
+      "tokens_seen": 1825636352
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00045125180548868563,
+      "loss": 1.4175,
+      "theoretical_loss": 3.4852638246031717,
+      "tokens_seen": 1825767424
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00045121168351789444,
+      "loss": 1.3301,
+      "theoretical_loss": 3.485242761491704,
+      "tokens_seen": 1825898496
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004511715615471032,
+      "loss": 1.4141,
+      "theoretical_loss": 3.4852217003155253,
+      "tokens_seen": 1826029568
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.000451131439576312,
+      "loss": 1.4812,
+      "theoretical_loss": 3.485200641074319,
+      "tokens_seen": 1826160640
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004510913176055208,
+      "loss": 1.3939,
+      "theoretical_loss": 3.485179583767768,
+      "tokens_seen": 1826291712
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00045105119563472956,
+      "loss": 1.3542,
+      "theoretical_loss": 3.4851585283955564,
+      "tokens_seen": 1826422784
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004510110736639384,
+      "loss": 1.3448,
+      "theoretical_loss": 3.4851374749573676,
+      "tokens_seen": 1826553856
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00045097095169314717,
+      "loss": 1.4108,
+      "theoretical_loss": 3.485116423452885,
+      "tokens_seen": 1826684928
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.000450930829722356,
+      "loss": 1.3578,
+      "theoretical_loss": 3.485095373881792,
+      "tokens_seen": 1826816000
+    },
+    {
+      "epoch": 0.11,
+      "objective/train/advantage_avg": 0.49358558654785156,
+      "objective/train/docs_used": 666029,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3123897314071655,
+      "objective/train/original_loss": 1.312389612197876,
+      "objective/train/theoretical_loss": 3.485074326243773,
+      "objective/train/tokens_used": 197406176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24399851262569427,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505989789962769,
+      "objective/train/weighted_lm_loss": 1.3788729906082153,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.485074326243773,
+      "tokens_seen": 1826947072
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004508907077515648,
+      "loss": 1.3088,
+      "theoretical_loss": 3.485074326243773,
+      "tokens_seen": 1826947072
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00045085058578077353,
+      "loss": 1.3708,
+      "theoretical_loss": 3.4850532805385113,
+      "tokens_seen": 1827078144
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00045081046380998234,
+      "loss": 1.4899,
+      "theoretical_loss": 3.485032236765691,
+      "tokens_seen": 1827209216
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00045077034183919115,
+      "loss": 1.3715,
+      "theoretical_loss": 3.4850111949249953,
+      "tokens_seen": 1827340288
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00045073021986839995,
+      "loss": 1.3949,
+      "theoretical_loss": 3.48499015501611,
+      "tokens_seen": 1827471360
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00045069009789760876,
+      "loss": 1.395,
+      "theoretical_loss": 3.484969117038718,
+      "tokens_seen": 1827602432
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00045064997592681756,
+      "loss": 1.4265,
+      "theoretical_loss": 3.4849480809925026,
+      "tokens_seen": 1827733504
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004506098539560263,
+      "loss": 1.3781,
+      "theoretical_loss": 3.48492704687715,
+      "tokens_seen": 1827864576
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004505697319852351,
+      "loss": 1.4163,
+      "theoretical_loss": 3.484906014692344,
+      "tokens_seen": 1827995648
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00045052961001444393,
+      "loss": 1.3599,
+      "theoretical_loss": 3.484884984437768,
+      "tokens_seen": 1828126720
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004504894880436527,
+      "loss": 1.4277,
+      "theoretical_loss": 3.4848639561131067,
+      "tokens_seen": 1828257792
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00045044936607286154,
+      "loss": 1.4105,
+      "theoretical_loss": 3.484842929718045,
+      "tokens_seen": 1828388864
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004504092441020703,
+      "loss": 1.3958,
+      "theoretical_loss": 3.4848219052522684,
+      "tokens_seen": 1828519936
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004503691221312791,
+      "loss": 1.456,
+      "theoretical_loss": 3.48480088271546,
+      "tokens_seen": 1828651008
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004503290001604879,
+      "loss": 1.3832,
+      "theoretical_loss": 3.484779862107306,
+      "tokens_seen": 1828782080
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00045028887818969666,
+      "loss": 1.3423,
+      "theoretical_loss": 3.4847588434274908,
+      "tokens_seen": 1828913152
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00045024875621890546,
+      "loss": 1.3357,
+      "theoretical_loss": 3.4847378266756985,
+      "tokens_seen": 1829044224
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00045020863424811427,
+      "loss": 1.3006,
+      "theoretical_loss": 3.484716811851615,
+      "tokens_seen": 1829175296
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004501685122773231,
+      "loss": 1.3509,
+      "theoretical_loss": 3.4846957989549248,
+      "tokens_seen": 1829306368
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00045012839030653183,
+      "loss": 1.3655,
+      "theoretical_loss": 3.4846747879853135,
+      "tokens_seen": 1829437440
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004500882683357407,
+      "loss": 1.3743,
+      "theoretical_loss": 3.484653778942466,
+      "tokens_seen": 1829568512
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00045004814636494944,
+      "loss": 1.3687,
+      "theoretical_loss": 3.484632771826068,
+      "tokens_seen": 1829699584
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004500080243941583,
+      "loss": 1.3914,
+      "theoretical_loss": 3.4846117666358047,
+      "tokens_seen": 1829830656
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044996790242336705,
+      "loss": 1.435,
+      "theoretical_loss": 3.484590763371361,
+      "tokens_seen": 1829961728
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004499277804525758,
+      "loss": 1.3915,
+      "theoretical_loss": 3.4845697620324234,
+      "tokens_seen": 1830092800
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044988765848178467,
+      "loss": 1.3406,
+      "theoretical_loss": 3.484548762618677,
+      "tokens_seen": 1830223872
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004498475365109934,
+      "loss": 1.3957,
+      "theoretical_loss": 3.484527765129807,
+      "tokens_seen": 1830354944
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004498074145402022,
+      "loss": 1.3166,
+      "theoretical_loss": 3.4845067695655,
+      "tokens_seen": 1830486016
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044976729256941103,
+      "loss": 1.3055,
+      "theoretical_loss": 3.4844857759254406,
+      "tokens_seen": 1830617088
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044972717059861984,
+      "loss": 1.4126,
+      "theoretical_loss": 3.484464784209316,
+      "tokens_seen": 1830748160
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004496870486278286,
+      "loss": 1.3462,
+      "theoretical_loss": 3.484443794416812,
+      "tokens_seen": 1830879232
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004496469266570374,
+      "loss": 1.3854,
+      "theoretical_loss": 3.484422806547614,
+      "tokens_seen": 1831010304
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004496068046862462,
+      "loss": 1.445,
+      "theoretical_loss": 3.4844018206014082,
+      "tokens_seen": 1831141376
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044956668271545495,
+      "loss": 1.3598,
+      "theoretical_loss": 3.4843808365778814,
+      "tokens_seen": 1831272448
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004495265607446638,
+      "loss": 1.3956,
+      "theoretical_loss": 3.4843598544767187,
+      "tokens_seen": 1831403520
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044948643877387257,
+      "loss": 1.2941,
+      "theoretical_loss": 3.484338874297608,
+      "tokens_seen": 1831534592
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044944631680308143,
+      "loss": 1.3541,
+      "theoretical_loss": 3.484317896040234,
+      "tokens_seen": 1831665664
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004494061948322902,
+      "loss": 1.3672,
+      "theoretical_loss": 3.484296919704284,
+      "tokens_seen": 1831796736
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044936607286149893,
+      "loss": 1.3801,
+      "theoretical_loss": 3.484275945289445,
+      "tokens_seen": 1831927808
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004493259508907078,
+      "loss": 1.4064,
+      "theoretical_loss": 3.4842549727954033,
+      "tokens_seen": 1832058880
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044928582891991654,
+      "loss": 1.3867,
+      "theoretical_loss": 3.4842340022218448,
+      "tokens_seen": 1832189952
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044924570694912535,
+      "loss": 1.4482,
+      "theoretical_loss": 3.4842130335684574,
+      "tokens_seen": 1832321024
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044920558497833416,
+      "loss": 1.3892,
+      "theoretical_loss": 3.4841920668349275,
+      "tokens_seen": 1832452096
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044916546300754296,
+      "loss": 1.4388,
+      "theoretical_loss": 3.484171102020942,
+      "tokens_seen": 1832583168
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004491253410367517,
+      "loss": 1.4,
+      "theoretical_loss": 3.4841501391261875,
+      "tokens_seen": 1832714240
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004490852190659605,
+      "loss": 1.3346,
+      "theoretical_loss": 3.484129178150352,
+      "tokens_seen": 1832845312
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044904509709516933,
+      "loss": 1.381,
+      "theoretical_loss": 3.484108219093122,
+      "tokens_seen": 1832976384
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004490049751243781,
+      "loss": 1.3395,
+      "theoretical_loss": 3.4840872619541843,
+      "tokens_seen": 1833107456
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044896485315358694,
+      "loss": 1.403,
+      "theoretical_loss": 3.484066306733227,
+      "tokens_seen": 1833238528
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004489247311827957,
+      "loss": 1.4662,
+      "theoretical_loss": 3.484045353429937,
+      "tokens_seen": 1833369600
+    },
+    {
+      "epoch": 0.11,
+      "objective/train/advantage_avg": 0.4920766353607178,
+      "objective/train/docs_used": 668449,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.340402364730835,
+      "objective/train/original_loss": 1.3404021263122559,
+      "objective/train/theoretical_loss": 3.4840244020440023,
+      "objective/train/tokens_used": 203959776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24279150366783142,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050441861152649,
+      "objective/train/weighted_lm_loss": 1.4080448150634766,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4840244020440023,
+      "tokens_seen": 1833500672
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004488846092120045,
+      "loss": 1.4252,
+      "theoretical_loss": 3.4840244020440023,
+      "tokens_seen": 1833500672
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004488444872412133,
+      "loss": 1.4468,
+      "theoretical_loss": 3.4840034525751093,
+      "tokens_seen": 1833631744
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044880436527042206,
+      "loss": 1.3924,
+      "theoretical_loss": 3.483982505022947,
+      "tokens_seen": 1833762816
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004487642432996309,
+      "loss": 1.3521,
+      "theoretical_loss": 3.4839615593872018,
+      "tokens_seen": 1833893888
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044872412132883967,
+      "loss": 1.3656,
+      "theoretical_loss": 3.483940615667562,
+      "tokens_seen": 1834024960
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004486839993580485,
+      "loss": 1.3854,
+      "theoretical_loss": 3.4839196738637153,
+      "tokens_seen": 1834156032
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004486438773872573,
+      "loss": 1.4361,
+      "theoretical_loss": 3.4838987339753498,
+      "tokens_seen": 1834287104
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004486037554164661,
+      "loss": 1.3817,
+      "theoretical_loss": 3.4838777960021527,
+      "tokens_seen": 1834418176
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044856363344567484,
+      "loss": 1.4701,
+      "theoretical_loss": 3.483856859943813,
+      "tokens_seen": 1834549248
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004485235114748837,
+      "loss": 1.4236,
+      "theoretical_loss": 3.483835925800018,
+      "tokens_seen": 1834680320
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044848338950409245,
+      "loss": 1.2915,
+      "theoretical_loss": 3.4838149935704563,
+      "tokens_seen": 1834811392
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004484432675333012,
+      "loss": 1.2873,
+      "theoretical_loss": 3.483794063254816,
+      "tokens_seen": 1834942464
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044840314556251007,
+      "loss": 1.3644,
+      "theoretical_loss": 3.4837731348527856,
+      "tokens_seen": 1835073536
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004483630235917188,
+      "loss": 1.3861,
+      "theoretical_loss": 3.483752208364053,
+      "tokens_seen": 1835204608
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004483229016209276,
+      "loss": 1.3625,
+      "theoretical_loss": 3.483731283788307,
+      "tokens_seen": 1835335680
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044828277965013643,
+      "loss": 1.3073,
+      "theoretical_loss": 3.4837103611252367,
+      "tokens_seen": 1835466752
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044824265767934524,
+      "loss": 1.4238,
+      "theoretical_loss": 3.4836894403745298,
+      "tokens_seen": 1835597824
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.000448202535708554,
+      "loss": 1.3519,
+      "theoretical_loss": 3.4836685215358747,
+      "tokens_seen": 1835728896
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004481624137377628,
+      "loss": 1.3903,
+      "theoretical_loss": 3.483647604608961,
+      "tokens_seen": 1835859968
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004481222917669716,
+      "loss": 1.4488,
+      "theoretical_loss": 3.4836266895934775,
+      "tokens_seen": 1835991040
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004480821697961804,
+      "loss": 1.3926,
+      "theoretical_loss": 3.4836057764891124,
+      "tokens_seen": 1836122112
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004480420478253892,
+      "loss": 1.3249,
+      "theoretical_loss": 3.4835848652955557,
+      "tokens_seen": 1836253184
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044800192585459797,
+      "loss": 1.3516,
+      "theoretical_loss": 3.483563956012495,
+      "tokens_seen": 1836384256
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044796180388380683,
+      "loss": 1.3699,
+      "theoretical_loss": 3.4835430486396204,
+      "tokens_seen": 1836515328
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004479216819130156,
+      "loss": 1.369,
+      "theoretical_loss": 3.4835221431766215,
+      "tokens_seen": 1836646400
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044788155994222433,
+      "loss": 1.4159,
+      "theoretical_loss": 3.4835012396231866,
+      "tokens_seen": 1836777472
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004478414379714332,
+      "loss": 1.3468,
+      "theoretical_loss": 3.483480337979005,
+      "tokens_seen": 1836908544
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044780131600064194,
+      "loss": 1.4129,
+      "theoretical_loss": 3.483459438243767,
+      "tokens_seen": 1837039616
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044776119402985075,
+      "loss": 1.4486,
+      "theoretical_loss": 3.4834385404171613,
+      "tokens_seen": 1837170688
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044772107205905956,
+      "loss": 1.388,
+      "theoretical_loss": 3.4834176444988776,
+      "tokens_seen": 1837301760
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044768095008826836,
+      "loss": 1.3008,
+      "theoretical_loss": 3.483396750488606,
+      "tokens_seen": 1837432832
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004476408281174771,
+      "loss": 1.4473,
+      "theoretical_loss": 3.4833758583860357,
+      "tokens_seen": 1837563904
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004476007061466859,
+      "loss": 1.5034,
+      "theoretical_loss": 3.4833549681908567,
+      "tokens_seen": 1837694976
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044756058417589473,
+      "loss": 1.3968,
+      "theoretical_loss": 3.4833340799027583,
+      "tokens_seen": 1837826048
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004475204622051035,
+      "loss": 1.4175,
+      "theoretical_loss": 3.4833131935214308,
+      "tokens_seen": 1837957120
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044748034023431234,
+      "loss": 1.4282,
+      "theoretical_loss": 3.483292309046565,
+      "tokens_seen": 1838088192
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004474402182635211,
+      "loss": 1.3508,
+      "theoretical_loss": 3.483271426477849,
+      "tokens_seen": 1838219264
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044740009629272995,
+      "loss": 1.3715,
+      "theoretical_loss": 3.4832505458149745,
+      "tokens_seen": 1838350336
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004473599743219387,
+      "loss": 1.3094,
+      "theoretical_loss": 3.483229667057632,
+      "tokens_seen": 1838481408
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044731985235114746,
+      "loss": 1.3688,
+      "theoretical_loss": 3.48320879020551,
+      "tokens_seen": 1838612480
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004472797303803563,
+      "loss": 1.3873,
+      "theoretical_loss": 3.4831879152583003,
+      "tokens_seen": 1838743552
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044723960840956507,
+      "loss": 1.3297,
+      "theoretical_loss": 3.483167042215693,
+      "tokens_seen": 1838874624
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004471994864387739,
+      "loss": 1.3694,
+      "theoretical_loss": 3.4831461710773786,
+      "tokens_seen": 1839005696
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004471593644679827,
+      "loss": 1.4097,
+      "theoretical_loss": 3.4831253018430477,
+      "tokens_seen": 1839136768
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0004471192424971915,
+      "loss": 1.4248,
+      "theoretical_loss": 3.4831044345123905,
+      "tokens_seen": 1839267840
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044707912052640024,
+      "loss": 1.3345,
+      "theoretical_loss": 3.483083569085098,
+      "tokens_seen": 1839398912
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044703899855560905,
+      "loss": 1.409,
+      "theoretical_loss": 3.4830627055608607,
+      "tokens_seen": 1839529984
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00044699887658481785,
+      "loss": 1.3696,
+      "theoretical_loss": 3.48304184393937,
+      "tokens_seen": 1839661056
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004469587546140266,
+      "loss": 1.3715,
+      "theoretical_loss": 3.4830209842203166,
+      "tokens_seen": 1839792128
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044691863264323547,
+      "loss": 1.3593,
+      "theoretical_loss": 3.4830001264033914,
+      "tokens_seen": 1839923200
+    },
+    {
+      "epoch": 0.12,
+      "objective/train/advantage_avg": 0.4910498857498169,
+      "objective/train/docs_used": 670853,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.246853232383728,
+      "objective/train/original_loss": 1.2468531131744385,
+      "objective/train/theoretical_loss": 3.4829792704882854,
+      "objective/train/tokens_used": 210513376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24191436171531677,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503346920013428,
+      "objective/train/weighted_lm_loss": 1.3095362186431885,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4829792704882854,
+      "tokens_seen": 1840054272
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004468785106724442,
+      "loss": 1.3686,
+      "theoretical_loss": 3.4829792704882854,
+      "tokens_seen": 1840054272
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004468383887016531,
+      "loss": 1.3804,
+      "theoretical_loss": 3.4829584164746903,
+      "tokens_seen": 1840185344
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044679826673086183,
+      "loss": 1.4111,
+      "theoretical_loss": 3.482937564362296,
+      "tokens_seen": 1840316416
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044675814476007064,
+      "loss": 1.3384,
+      "theoretical_loss": 3.4829167141507957,
+      "tokens_seen": 1840447488
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044671802278927944,
+      "loss": 1.3616,
+      "theoretical_loss": 3.482895865839879,
+      "tokens_seen": 1840578560
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004466779008184882,
+      "loss": 1.3774,
+      "theoretical_loss": 3.482875019429238,
+      "tokens_seen": 1840709632
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.000446637778847697,
+      "loss": 1.3396,
+      "theoretical_loss": 3.4828541749185646,
+      "tokens_seen": 1840840704
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004465976568769058,
+      "loss": 1.4376,
+      "theoretical_loss": 3.4828333323075498,
+      "tokens_seen": 1840971776
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004465575349061146,
+      "loss": 1.3454,
+      "theoretical_loss": 3.4828124915958854,
+      "tokens_seen": 1841102848
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044651741293532337,
+      "loss": 1.3726,
+      "theoretical_loss": 3.482791652783263,
+      "tokens_seen": 1841233920
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044647729096453223,
+      "loss": 1.3949,
+      "theoretical_loss": 3.482770815869375,
+      "tokens_seen": 1841364992
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.000446437168993741,
+      "loss": 1.4421,
+      "theoretical_loss": 3.4827499808539124,
+      "tokens_seen": 1841496064
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044639704702294973,
+      "loss": 1.4141,
+      "theoretical_loss": 3.482729147736568,
+      "tokens_seen": 1841627136
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004463569250521586,
+      "loss": 1.3103,
+      "theoretical_loss": 3.4827083165170327,
+      "tokens_seen": 1841758208
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044631680308136734,
+      "loss": 1.3858,
+      "theoretical_loss": 3.4826874871949993,
+      "tokens_seen": 1841889280
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044627668111057615,
+      "loss": 1.3257,
+      "theoretical_loss": 3.4826666597701603,
+      "tokens_seen": 1842020352
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044623655913978496,
+      "loss": 1.4092,
+      "theoretical_loss": 3.4826458342422066,
+      "tokens_seen": 1842151424
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044619643716899376,
+      "loss": 1.4249,
+      "theoretical_loss": 3.4826250106108323,
+      "tokens_seen": 1842282496
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044615631519820257,
+      "loss": 1.3837,
+      "theoretical_loss": 3.4826041888757286,
+      "tokens_seen": 1842413568
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004461161932274113,
+      "loss": 1.4136,
+      "theoretical_loss": 3.482583369036588,
+      "tokens_seen": 1842544640
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044607607125662013,
+      "loss": 1.4548,
+      "theoretical_loss": 3.4825625510931033,
+      "tokens_seen": 1842675712
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044603594928582893,
+      "loss": 1.3707,
+      "theoretical_loss": 3.4825417350449666,
+      "tokens_seen": 1842806784
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044599582731503774,
+      "loss": 1.3876,
+      "theoretical_loss": 3.4825209208918713,
+      "tokens_seen": 1842937856
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004459557053442465,
+      "loss": 1.4564,
+      "theoretical_loss": 3.4825001086335092,
+      "tokens_seen": 1843068928
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044591558337345535,
+      "loss": 1.3374,
+      "theoretical_loss": 3.482479298269574,
+      "tokens_seen": 1843200000
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004458754614026641,
+      "loss": 1.3979,
+      "theoretical_loss": 3.4824584897997584,
+      "tokens_seen": 1843331072
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044583533943187286,
+      "loss": 1.386,
+      "theoretical_loss": 3.482437683223755,
+      "tokens_seen": 1843462144
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004457952174610817,
+      "loss": 1.3341,
+      "theoretical_loss": 3.4824168785412564,
+      "tokens_seen": 1843593216
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044575509549029047,
+      "loss": 1.4137,
+      "theoretical_loss": 3.4823960757519563,
+      "tokens_seen": 1843724288
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004457149735194993,
+      "loss": 1.3667,
+      "theoretical_loss": 3.4823752748555483,
+      "tokens_seen": 1843855360
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004456748515487081,
+      "loss": 1.429,
+      "theoretical_loss": 3.4823544758517246,
+      "tokens_seen": 1843986432
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004456347295779169,
+      "loss": 1.4354,
+      "theoretical_loss": 3.4823336787401793,
+      "tokens_seen": 1844117504
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004455946076071257,
+      "loss": 1.4392,
+      "theoretical_loss": 3.482312883520605,
+      "tokens_seen": 1844248576
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044555448563633445,
+      "loss": 1.3632,
+      "theoretical_loss": 3.4822920901926957,
+      "tokens_seen": 1844379648
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044551436366554325,
+      "loss": 1.3642,
+      "theoretical_loss": 3.482271298756145,
+      "tokens_seen": 1844510720
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044547424169475206,
+      "loss": 1.3613,
+      "theoretical_loss": 3.4822505092106457,
+      "tokens_seen": 1844641792
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044543411972396087,
+      "loss": 1.4459,
+      "theoretical_loss": 3.4822297215558926,
+      "tokens_seen": 1844772864
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004453939977531696,
+      "loss": 1.3117,
+      "theoretical_loss": 3.4822089357915784,
+      "tokens_seen": 1844903936
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004453538757823785,
+      "loss": 1.4002,
+      "theoretical_loss": 3.4821881519173976,
+      "tokens_seen": 1845035008
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044531375381158723,
+      "loss": 1.4043,
+      "theoretical_loss": 3.4821673699330438,
+      "tokens_seen": 1845166080
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044527363184079604,
+      "loss": 1.4272,
+      "theoretical_loss": 3.482146589838211,
+      "tokens_seen": 1845297152
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044523350987000484,
+      "loss": 1.3415,
+      "theoretical_loss": 3.482125811632593,
+      "tokens_seen": 1845428224
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004451933878992136,
+      "loss": 1.3553,
+      "theoretical_loss": 3.482105035315884,
+      "tokens_seen": 1845559296
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004451532659284224,
+      "loss": 1.4534,
+      "theoretical_loss": 3.482084260887778,
+      "tokens_seen": 1845690368
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004451131439576312,
+      "loss": 1.3353,
+      "theoretical_loss": 3.4820634883479693,
+      "tokens_seen": 1845821440
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044507302198684,
+      "loss": 1.4535,
+      "theoretical_loss": 3.4820427176961526,
+      "tokens_seen": 1845952512
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044503290001604877,
+      "loss": 1.4547,
+      "theoretical_loss": 3.482021948932022,
+      "tokens_seen": 1846083584
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004449927780452576,
+      "loss": 1.4088,
+      "theoretical_loss": 3.4820011820552716,
+      "tokens_seen": 1846214656
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004449526560744664,
+      "loss": 1.3235,
+      "theoretical_loss": 3.481980417065596,
+      "tokens_seen": 1846345728
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004449125341036752,
+      "loss": 1.4615,
+      "theoretical_loss": 3.48195965396269,
+      "tokens_seen": 1846476800
+    },
+    {
+      "epoch": 0.12,
+      "objective/train/advantage_avg": 0.4902796447277069,
+      "objective/train/docs_used": 673286,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.4052106142044067,
+      "objective/train/original_loss": 1.4052107334136963,
+      "objective/train/theoretical_loss": 3.4819388927462485,
+      "objective/train/tokens_used": 217066976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24099278450012207,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0502530336380005,
+      "objective/train/weighted_lm_loss": 1.4756853580474854,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4819388927462485,
+      "tokens_seen": 1846607872
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.000444872412132884,
+      "loss": 1.4399,
+      "theoretical_loss": 3.4819388927462485,
+      "tokens_seen": 1846607872
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044483229016209274,
+      "loss": 1.3801,
+      "theoretical_loss": 3.4819181334159657,
+      "tokens_seen": 1846738944
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004447921681913016,
+      "loss": 1.3909,
+      "theoretical_loss": 3.4818973759715366,
+      "tokens_seen": 1846870016
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044475204622051036,
+      "loss": 1.4092,
+      "theoretical_loss": 3.481876620412656,
+      "tokens_seen": 1847001088
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044471192424971916,
+      "loss": 1.4178,
+      "theoretical_loss": 3.4818558667390196,
+      "tokens_seen": 1847132160
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044467180227892797,
+      "loss": 1.4,
+      "theoretical_loss": 3.4818351149503206,
+      "tokens_seen": 1847263232
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004446316803081367,
+      "loss": 1.3854,
+      "theoretical_loss": 3.4818143650462563,
+      "tokens_seen": 1847394304
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044459155833734553,
+      "loss": 1.3702,
+      "theoretical_loss": 3.4817936170265202,
+      "tokens_seen": 1847525376
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044455143636655433,
+      "loss": 1.3358,
+      "theoretical_loss": 3.481772870890808,
+      "tokens_seen": 1847656448
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044451131439576314,
+      "loss": 1.4719,
+      "theoretical_loss": 3.4817521266388147,
+      "tokens_seen": 1847787520
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004444711924249719,
+      "loss": 1.4299,
+      "theoretical_loss": 3.4817313842702364,
+      "tokens_seen": 1847918592
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044443107045418075,
+      "loss": 1.4327,
+      "theoretical_loss": 3.4817106437847682,
+      "tokens_seen": 1848049664
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004443909484833895,
+      "loss": 1.385,
+      "theoretical_loss": 3.4816899051821055,
+      "tokens_seen": 1848180736
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044435082651259826,
+      "loss": 1.3943,
+      "theoretical_loss": 3.4816691684619436,
+      "tokens_seen": 1848311808
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004443107045418071,
+      "loss": 1.3888,
+      "theoretical_loss": 3.4816484336239784,
+      "tokens_seen": 1848442880
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044427058257101587,
+      "loss": 1.4187,
+      "theoretical_loss": 3.481627700667906,
+      "tokens_seen": 1848573952
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044423046060022473,
+      "loss": 1.3762,
+      "theoretical_loss": 3.4816069695934218,
+      "tokens_seen": 1848705024
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004441903386294335,
+      "loss": 1.4806,
+      "theoretical_loss": 3.4815862404002216,
+      "tokens_seen": 1848836096
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004441502166586423,
+      "loss": 1.3654,
+      "theoretical_loss": 3.481565513088001,
+      "tokens_seen": 1848967168
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004441100946878511,
+      "loss": 1.381,
+      "theoretical_loss": 3.481544787656457,
+      "tokens_seen": 1849098240
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044406997271705985,
+      "loss": 1.3566,
+      "theoretical_loss": 3.481524064105285,
+      "tokens_seen": 1849229312
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044402985074626865,
+      "loss": 1.4256,
+      "theoretical_loss": 3.4815033424341806,
+      "tokens_seen": 1849360384
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044398972877547746,
+      "loss": 1.3588,
+      "theoretical_loss": 3.481482622642841,
+      "tokens_seen": 1849491456
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044394960680468627,
+      "loss": 1.4388,
+      "theoretical_loss": 3.4814619047309625,
+      "tokens_seen": 1849622528
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.000443909484833895,
+      "loss": 1.4007,
+      "theoretical_loss": 3.481441188698241,
+      "tokens_seen": 1849753600
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004438693628631039,
+      "loss": 1.3876,
+      "theoretical_loss": 3.481420474544372,
+      "tokens_seen": 1849884672
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044382924089231263,
+      "loss": 1.4046,
+      "theoretical_loss": 3.481399762269054,
+      "tokens_seen": 1850015744
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044378911892152144,
+      "loss": 1.4403,
+      "theoretical_loss": 3.481379051871982,
+      "tokens_seen": 1850146816
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044374899695073024,
+      "loss": 1.5157,
+      "theoretical_loss": 3.481358343352853,
+      "tokens_seen": 1850277888
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.000443708874979939,
+      "loss": 1.3661,
+      "theoretical_loss": 3.4813376367113644,
+      "tokens_seen": 1850408960
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044366875300914786,
+      "loss": 1.3688,
+      "theoretical_loss": 3.481316931947212,
+      "tokens_seen": 1850540032
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004436286310383566,
+      "loss": 1.2976,
+      "theoretical_loss": 3.481296229060093,
+      "tokens_seen": 1850671104
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004435885090675654,
+      "loss": 1.3647,
+      "theoretical_loss": 3.4812755280497045,
+      "tokens_seen": 1850802176
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004435483870967742,
+      "loss": 1.4287,
+      "theoretical_loss": 3.4812548289157434,
+      "tokens_seen": 1850933248
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.000443508265125983,
+      "loss": 1.3768,
+      "theoretical_loss": 3.481234131657907,
+      "tokens_seen": 1851064320
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004434681431551918,
+      "loss": 1.4105,
+      "theoretical_loss": 3.4812134362758913,
+      "tokens_seen": 1851195392
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004434280211844006,
+      "loss": 1.344,
+      "theoretical_loss": 3.4811927427693954,
+      "tokens_seen": 1851326464
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004433878992136094,
+      "loss": 1.3587,
+      "theoretical_loss": 3.4811720511381146,
+      "tokens_seen": 1851457536
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044334777724281814,
+      "loss": 1.3404,
+      "theoretical_loss": 3.4811513613817473,
+      "tokens_seen": 1851588608
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.000443307655272027,
+      "loss": 1.3612,
+      "theoretical_loss": 3.481130673499991,
+      "tokens_seen": 1851719680
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044326753330123576,
+      "loss": 1.386,
+      "theoretical_loss": 3.481109987492543,
+      "tokens_seen": 1851850752
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044322741133044456,
+      "loss": 1.3515,
+      "theoretical_loss": 3.4810893033591004,
+      "tokens_seen": 1851981824
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044318728935965337,
+      "loss": 1.4022,
+      "theoretical_loss": 3.481068621099361,
+      "tokens_seen": 1852112896
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004431471673888621,
+      "loss": 1.4085,
+      "theoretical_loss": 3.4810479407130233,
+      "tokens_seen": 1852243968
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004431070454180709,
+      "loss": 1.3562,
+      "theoretical_loss": 3.4810272621997838,
+      "tokens_seen": 1852375040
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044306692344727973,
+      "loss": 1.4337,
+      "theoretical_loss": 3.4810065855593413,
+      "tokens_seen": 1852506112
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044302680147648854,
+      "loss": 1.4347,
+      "theoretical_loss": 3.480985910791393,
+      "tokens_seen": 1852637184
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044298667950569735,
+      "loss": 1.3398,
+      "theoretical_loss": 3.4809652378956373,
+      "tokens_seen": 1852768256
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044294655753490615,
+      "loss": 1.3249,
+      "theoretical_loss": 3.4809445668717722,
+      "tokens_seen": 1852899328
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004429064355641149,
+      "loss": 1.4167,
+      "theoretical_loss": 3.480923897719496,
+      "tokens_seen": 1853030400
+    },
+    {
+      "epoch": 0.12,
+      "objective/train/advantage_avg": 0.489789217710495,
+      "objective/train/docs_used": 675642,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3568042516708374,
+      "objective/train/original_loss": 1.356804370880127,
+      "objective/train/theoretical_loss": 3.480903230438506,
+      "objective/train/tokens_used": 223620576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2405891865491867,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0502018928527832,
+      "objective/train/weighted_lm_loss": 1.4249345064163208,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.480903230438506,
+      "tokens_seen": 1853161472
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004428663135933237,
+      "loss": 1.4345,
+      "theoretical_loss": 3.480903230438506,
+      "tokens_seen": 1853161472
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004428261916225325,
+      "loss": 1.4082,
+      "theoretical_loss": 3.4808825650285016,
+      "tokens_seen": 1853292544
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044278606965174127,
+      "loss": 1.3963,
+      "theoretical_loss": 3.48086190148918,
+      "tokens_seen": 1853423616
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044274594768095013,
+      "loss": 1.3701,
+      "theoretical_loss": 3.480841239820241,
+      "tokens_seen": 1853554688
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004427058257101589,
+      "loss": 1.3768,
+      "theoretical_loss": 3.4808205800213816,
+      "tokens_seen": 1853685760
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004426657037393677,
+      "loss": 1.3318,
+      "theoretical_loss": 3.4807999220923014,
+      "tokens_seen": 1853816832
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004426255817685765,
+      "loss": 1.4265,
+      "theoretical_loss": 3.480779266032698,
+      "tokens_seen": 1853947904
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044258545979778525,
+      "loss": 1.3891,
+      "theoretical_loss": 3.480758611842271,
+      "tokens_seen": 1854078976
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044254533782699405,
+      "loss": 1.3698,
+      "theoretical_loss": 3.4807379595207184,
+      "tokens_seen": 1854210048
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044250521585620286,
+      "loss": 1.405,
+      "theoretical_loss": 3.4807173090677397,
+      "tokens_seen": 1854341120
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044246509388541167,
+      "loss": 1.3618,
+      "theoretical_loss": 3.480696660483033,
+      "tokens_seen": 1854472192
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004424249719146204,
+      "loss": 1.3265,
+      "theoretical_loss": 3.480676013766298,
+      "tokens_seen": 1854603264
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004423848499438293,
+      "loss": 1.3407,
+      "theoretical_loss": 3.4806553689172333,
+      "tokens_seen": 1854734336
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044234472797303803,
+      "loss": 1.3631,
+      "theoretical_loss": 3.480634725935538,
+      "tokens_seen": 1854865408
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044230460600224684,
+      "loss": 1.4185,
+      "theoretical_loss": 3.4806140848209113,
+      "tokens_seen": 1854996480
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044226448403145564,
+      "loss": 1.3539,
+      "theoretical_loss": 3.480593445573053,
+      "tokens_seen": 1855127552
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004422243620606644,
+      "loss": 1.4121,
+      "theoretical_loss": 3.480572808191661,
+      "tokens_seen": 1855258624
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00044218424008987326,
+      "loss": 1.3445,
+      "theoretical_loss": 3.480552172676436,
+      "tokens_seen": 1855389696
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.000442144118119082,
+      "loss": 1.4243,
+      "theoretical_loss": 3.4805315390270772,
+      "tokens_seen": 1855520768
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004421039961482908,
+      "loss": 1.3811,
+      "theoretical_loss": 3.4805109072432834,
+      "tokens_seen": 1855651840
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004420638741774996,
+      "loss": 1.406,
+      "theoretical_loss": 3.480490277324755,
+      "tokens_seen": 1855782912
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004420237522067084,
+      "loss": 1.2858,
+      "theoretical_loss": 3.480469649271191,
+      "tokens_seen": 1855913984
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0004419836302359172,
+      "loss": 1.3786,
+      "theoretical_loss": 3.480449023082291,
+      "tokens_seen": 1856045056
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.000441943508265126,
+      "loss": 1.3217,
+      "theoretical_loss": 3.4804283987577556,
+      "tokens_seen": 1856176128
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004419033862943348,
+      "loss": 1.3446,
+      "theoretical_loss": 3.480407776297284,
+      "tokens_seen": 1856307200
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00044186326432354354,
+      "loss": 1.3359,
+      "theoretical_loss": 3.4803871557005763,
+      "tokens_seen": 1856438272
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004418231423527524,
+      "loss": 1.4432,
+      "theoretical_loss": 3.480366536967333,
+      "tokens_seen": 1856569344
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00044178302038196116,
+      "loss": 1.3658,
+      "theoretical_loss": 3.4803459200972533,
+      "tokens_seen": 1856700416
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00044174289841117,
+      "loss": 1.3702,
+      "theoretical_loss": 3.4803253050900373,
+      "tokens_seen": 1856831488
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00044170277644037877,
+      "loss": 1.3661,
+      "theoretical_loss": 3.4803046919453866,
+      "tokens_seen": 1856962560
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004416626544695875,
+      "loss": 1.4024,
+      "theoretical_loss": 3.480284080663,
+      "tokens_seen": 1857093632
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004416225324987964,
+      "loss": 1.4183,
+      "theoretical_loss": 3.480263471242578,
+      "tokens_seen": 1857224704
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00044158241052800513,
+      "loss": 1.3457,
+      "theoretical_loss": 3.4802428636838214,
+      "tokens_seen": 1857355776
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00044154228855721394,
+      "loss": 1.4095,
+      "theoretical_loss": 3.480222257986431,
+      "tokens_seen": 1857486848
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00044150216658642275,
+      "loss": 1.4217,
+      "theoretical_loss": 3.4802016541501066,
+      "tokens_seen": 1857617920
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00044146204461563155,
+      "loss": 1.4579,
+      "theoretical_loss": 3.480181052174549,
+      "tokens_seen": 1857748992
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004414219226448403,
+      "loss": 1.3172,
+      "theoretical_loss": 3.48016045205946,
+      "tokens_seen": 1857880064
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004413818006740491,
+      "loss": 1.3783,
+      "theoretical_loss": 3.480139853804538,
+      "tokens_seen": 1858011136
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004413416787032579,
+      "loss": 1.388,
+      "theoretical_loss": 3.480119257409486,
+      "tokens_seen": 1858142208
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00044130155673246667,
+      "loss": 1.4427,
+      "theoretical_loss": 3.480098662874004,
+      "tokens_seen": 1858273280
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00044126143476167553,
+      "loss": 1.4132,
+      "theoretical_loss": 3.480078070197793,
+      "tokens_seen": 1858404352
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004412213127908843,
+      "loss": 1.419,
+      "theoretical_loss": 3.4800574793805543,
+      "tokens_seen": 1858535424
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004411811908200931,
+      "loss": 1.3341,
+      "theoretical_loss": 3.4800368904219887,
+      "tokens_seen": 1858666496
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004411410688493019,
+      "loss": 1.3286,
+      "theoretical_loss": 3.480016303321797,
+      "tokens_seen": 1858797568
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00044110094687851065,
+      "loss": 1.4095,
+      "theoretical_loss": 3.4799957180796817,
+      "tokens_seen": 1858928640
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004410608249077195,
+      "loss": 1.2971,
+      "theoretical_loss": 3.4799751346953425,
+      "tokens_seen": 1859059712
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00044102070293692826,
+      "loss": 1.4639,
+      "theoretical_loss": 3.479954553168482,
+      "tokens_seen": 1859190784
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00044098058096613706,
+      "loss": 1.2664,
+      "theoretical_loss": 3.4799339734988015,
+      "tokens_seen": 1859321856
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00044094045899534587,
+      "loss": 1.3877,
+      "theoretical_loss": 3.479913395686002,
+      "tokens_seen": 1859452928
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004409003370245547,
+      "loss": 1.3243,
+      "theoretical_loss": 3.479892819729785,
+      "tokens_seen": 1859584000
+    },
+    {
+      "epoch": 0.13,
+      "objective/train/advantage_avg": 0.4955404996871948,
+      "objective/train/docs_used": 678160,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2684444189071655,
+      "objective/train/original_loss": 1.2684444189071655,
+      "objective/train/theoretical_loss": 3.4798722456298528,
+      "objective/train/tokens_used": 230174176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24573126435279846,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0508034229278564,
+      "objective/train/weighted_lm_loss": 1.3327796459197998,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4798722456298528,
+      "tokens_seen": 1859715072
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00044086021505376343,
+      "loss": 1.3856,
+      "theoretical_loss": 3.4798722456298528,
+      "tokens_seen": 1859715072
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00044082009308297224,
+      "loss": 1.4043,
+      "theoretical_loss": 3.4798516733859066,
+      "tokens_seen": 1859846144
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00044077997111218104,
+      "loss": 1.3693,
+      "theoretical_loss": 3.4798311029976485,
+      "tokens_seen": 1859977216
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004407398491413898,
+      "loss": 1.3736,
+      "theoretical_loss": 3.4798105344647805,
+      "tokens_seen": 1860108288
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00044069972717059865,
+      "loss": 1.4098,
+      "theoretical_loss": 3.4797899677870037,
+      "tokens_seen": 1860239360
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004406596051998074,
+      "loss": 1.379,
+      "theoretical_loss": 3.479769402964021,
+      "tokens_seen": 1860370432
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004406194832290162,
+      "loss": 1.3413,
+      "theoretical_loss": 3.4797488399955343,
+      "tokens_seen": 1860501504
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.000440579361258225,
+      "loss": 1.4251,
+      "theoretical_loss": 3.479728278881245,
+      "tokens_seen": 1860632576
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004405392392874338,
+      "loss": 1.3594,
+      "theoretical_loss": 3.4797077196208566,
+      "tokens_seen": 1860763648
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00044049911731664263,
+      "loss": 1.3598,
+      "theoretical_loss": 3.47968716221407,
+      "tokens_seen": 1860894720
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004404589953458514,
+      "loss": 1.3356,
+      "theoretical_loss": 3.4796666066605884,
+      "tokens_seen": 1861025792
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004404188733750602,
+      "loss": 1.3471,
+      "theoretical_loss": 3.4796460529601143,
+      "tokens_seen": 1861156864
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.000440378751404269,
+      "loss": 1.4103,
+      "theoretical_loss": 3.4796255011123494,
+      "tokens_seen": 1861287936
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004403386294334778,
+      "loss": 1.3619,
+      "theoretical_loss": 3.479604951116997,
+      "tokens_seen": 1861419008
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00044029850746268656,
+      "loss": 1.4358,
+      "theoretical_loss": 3.4795844029737593,
+      "tokens_seen": 1861550080
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004402583854918954,
+      "loss": 1.4009,
+      "theoretical_loss": 3.479563856682339,
+      "tokens_seen": 1861681152
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00044021826352110417,
+      "loss": 1.4379,
+      "theoretical_loss": 3.4795433122424395,
+      "tokens_seen": 1861812224
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004401781415503129,
+      "loss": 1.3569,
+      "theoretical_loss": 3.4795227696537623,
+      "tokens_seen": 1861943296
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004401380195795218,
+      "loss": 1.5066,
+      "theoretical_loss": 3.4795022289160116,
+      "tokens_seen": 1862074368
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00044009789760873053,
+      "loss": 1.3574,
+      "theoretical_loss": 3.47948169002889,
+      "tokens_seen": 1862205440
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00044005777563793934,
+      "loss": 1.4047,
+      "theoretical_loss": 3.4794611529920996,
+      "tokens_seen": 1862336512
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00044001765366714815,
+      "loss": 1.3777,
+      "theoretical_loss": 3.479440617805345,
+      "tokens_seen": 1862467584
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043997753169635695,
+      "loss": 1.348,
+      "theoretical_loss": 3.4794200844683285,
+      "tokens_seen": 1862598656
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004399374097255657,
+      "loss": 1.3758,
+      "theoretical_loss": 3.479399552980753,
+      "tokens_seen": 1862729728
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004398972877547745,
+      "loss": 1.3707,
+      "theoretical_loss": 3.4793790233423225,
+      "tokens_seen": 1862860800
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004398571657839833,
+      "loss": 1.2904,
+      "theoretical_loss": 3.4793584955527406,
+      "tokens_seen": 1862991872
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004398170438131921,
+      "loss": 1.3634,
+      "theoretical_loss": 3.4793379696117093,
+      "tokens_seen": 1863122944
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043977692184240093,
+      "loss": 1.339,
+      "theoretical_loss": 3.4793174455189337,
+      "tokens_seen": 1863254016
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004397367998716097,
+      "loss": 1.3619,
+      "theoretical_loss": 3.4792969232741164,
+      "tokens_seen": 1863385088
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043969667790081854,
+      "loss": 1.3437,
+      "theoretical_loss": 3.479276402876962,
+      "tokens_seen": 1863516160
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004396565559300273,
+      "loss": 1.4257,
+      "theoretical_loss": 3.4792558843271735,
+      "tokens_seen": 1863647232
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043961643395923605,
+      "loss": 1.4027,
+      "theoretical_loss": 3.4792353676244545,
+      "tokens_seen": 1863778304
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004395763119884449,
+      "loss": 1.4093,
+      "theoretical_loss": 3.479214852768509,
+      "tokens_seen": 1863909376
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043953619001765366,
+      "loss": 1.385,
+      "theoretical_loss": 3.479194339759041,
+      "tokens_seen": 1864040448
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043949606804686246,
+      "loss": 1.4099,
+      "theoretical_loss": 3.479173828595755,
+      "tokens_seen": 1864171520
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043945594607607127,
+      "loss": 1.4216,
+      "theoretical_loss": 3.479153319278354,
+      "tokens_seen": 1864302592
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004394158241052801,
+      "loss": 1.4015,
+      "theoretical_loss": 3.4791328118065428,
+      "tokens_seen": 1864433664
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043937570213448883,
+      "loss": 1.3062,
+      "theoretical_loss": 3.479112306180026,
+      "tokens_seen": 1864564736
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043933558016369764,
+      "loss": 1.3544,
+      "theoretical_loss": 3.4790918023985067,
+      "tokens_seen": 1864695808
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043929545819290644,
+      "loss": 1.4341,
+      "theoretical_loss": 3.4790713004616904,
+      "tokens_seen": 1864826880
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004392553362221152,
+      "loss": 1.3147,
+      "theoretical_loss": 3.479050800369281,
+      "tokens_seen": 1864957952
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043921521425132405,
+      "loss": 1.3996,
+      "theoretical_loss": 3.4790303021209823,
+      "tokens_seen": 1865089024
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004391750922805328,
+      "loss": 1.337,
+      "theoretical_loss": 3.4790098057164998,
+      "tokens_seen": 1865220096
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043913497030974167,
+      "loss": 1.3397,
+      "theoretical_loss": 3.478989311155538,
+      "tokens_seen": 1865351168
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004390948483389504,
+      "loss": 1.4063,
+      "theoretical_loss": 3.478968818437801,
+      "tokens_seen": 1865482240
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004390547263681592,
+      "loss": 1.4111,
+      "theoretical_loss": 3.478948327562994,
+      "tokens_seen": 1865613312
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043901460439736803,
+      "loss": 1.3802,
+      "theoretical_loss": 3.4789278385308218,
+      "tokens_seen": 1865744384
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004389744824265768,
+      "loss": 1.3687,
+      "theoretical_loss": 3.4789073513409887,
+      "tokens_seen": 1865875456
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004389343604557856,
+      "loss": 1.3641,
+      "theoretical_loss": 3.478886865993201,
+      "tokens_seen": 1866006528
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004388942384849944,
+      "loss": 1.4456,
+      "theoretical_loss": 3.478866382487162,
+      "tokens_seen": 1866137600
+    },
+    {
+      "epoch": 0.13,
+      "objective/train/advantage_avg": 0.4927670657634735,
+      "objective/train/docs_used": 681054,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3475145101547241,
+      "objective/train/original_loss": 1.3475145101547241,
+      "objective/train/theoretical_loss": 3.478845900822577,
+      "objective/train/tokens_used": 236727776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24354243278503418,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505149364471436,
+      "objective/train/weighted_lm_loss": 1.415496826171875,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.478845900822577,
+      "tokens_seen": 1866268672
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004388541165142032,
+      "loss": 1.4121,
+      "theoretical_loss": 3.478845900822577,
+      "tokens_seen": 1866268672
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043881399454341195,
+      "loss": 1.3009,
+      "theoretical_loss": 3.4788254209991525,
+      "tokens_seen": 1866399744
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004387738725726208,
+      "loss": 1.4071,
+      "theoretical_loss": 3.478804943016593,
+      "tokens_seen": 1866530816
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043873375060182957,
+      "loss": 1.3187,
+      "theoretical_loss": 3.4787844668746035,
+      "tokens_seen": 1866661888
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004386936286310383,
+      "loss": 1.3979,
+      "theoretical_loss": 3.4787639925728895,
+      "tokens_seen": 1866792960
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004386535066602472,
+      "loss": 1.4372,
+      "theoretical_loss": 3.4787435201111565,
+      "tokens_seen": 1866924032
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043861338468945593,
+      "loss": 1.3772,
+      "theoretical_loss": 3.47872304948911,
+      "tokens_seen": 1867055104
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004385732627186648,
+      "loss": 1.3795,
+      "theoretical_loss": 3.4787025807064555,
+      "tokens_seen": 1867186176
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043853314074787354,
+      "loss": 1.489,
+      "theoretical_loss": 3.4786821137628983,
+      "tokens_seen": 1867317248
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043849301877708235,
+      "loss": 1.3975,
+      "theoretical_loss": 3.4786616486581448,
+      "tokens_seen": 1867448320
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043845289680629116,
+      "loss": 1.3942,
+      "theoretical_loss": 3.4786411853918997,
+      "tokens_seen": 1867579392
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004384127748354999,
+      "loss": 1.4172,
+      "theoretical_loss": 3.4786207239638696,
+      "tokens_seen": 1867710464
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004383726528647087,
+      "loss": 1.3948,
+      "theoretical_loss": 3.478600264373761,
+      "tokens_seen": 1867841536
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004383325308939175,
+      "loss": 1.4285,
+      "theoretical_loss": 3.478579806621278,
+      "tokens_seen": 1867972608
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043829240892312633,
+      "loss": 1.4439,
+      "theoretical_loss": 3.4785593507061288,
+      "tokens_seen": 1868103680
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004382522869523351,
+      "loss": 1.4411,
+      "theoretical_loss": 3.478538896628018,
+      "tokens_seen": 1868234752
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043821216498154394,
+      "loss": 1.3388,
+      "theoretical_loss": 3.478518444386652,
+      "tokens_seen": 1868365824
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004381720430107527,
+      "loss": 1.4278,
+      "theoretical_loss": 3.478497993981737,
+      "tokens_seen": 1868496896
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043813192103996145,
+      "loss": 1.4567,
+      "theoretical_loss": 3.47847754541298,
+      "tokens_seen": 1868627968
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004380917990691703,
+      "loss": 1.4203,
+      "theoretical_loss": 3.478457098680087,
+      "tokens_seen": 1868759040
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043805167709837906,
+      "loss": 1.5401,
+      "theoretical_loss": 3.4784366537827633,
+      "tokens_seen": 1868890112
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043801155512758786,
+      "loss": 1.4186,
+      "theoretical_loss": 3.4784162107207175,
+      "tokens_seen": 1869021184
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043797143315679667,
+      "loss": 1.3772,
+      "theoretical_loss": 3.4783957694936545,
+      "tokens_seen": 1869152256
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004379313111860055,
+      "loss": 1.3656,
+      "theoretical_loss": 3.478375330101281,
+      "tokens_seen": 1869283328
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004378911892152143,
+      "loss": 1.3855,
+      "theoretical_loss": 3.478354892543305,
+      "tokens_seen": 1869414400
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043785106724442304,
+      "loss": 1.447,
+      "theoretical_loss": 3.4783344568194323,
+      "tokens_seen": 1869545472
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043781094527363184,
+      "loss": 1.4198,
+      "theoretical_loss": 3.4783140229293696,
+      "tokens_seen": 1869676544
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043777082330284065,
+      "loss": 1.5236,
+      "theoretical_loss": 3.4782935908728243,
+      "tokens_seen": 1869807616
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043773070133204945,
+      "loss": 1.4248,
+      "theoretical_loss": 3.478273160649503,
+      "tokens_seen": 1869938688
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004376905793612582,
+      "loss": 1.2784,
+      "theoretical_loss": 3.4782527322591132,
+      "tokens_seen": 1870069760
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043765045739046707,
+      "loss": 1.4033,
+      "theoretical_loss": 3.4782323057013613,
+      "tokens_seen": 1870200832
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004376103354196758,
+      "loss": 1.3903,
+      "theoretical_loss": 3.4782118809759552,
+      "tokens_seen": 1870331904
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043757021344888457,
+      "loss": 1.3575,
+      "theoretical_loss": 3.4781914580826014,
+      "tokens_seen": 1870462976
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043753009147809343,
+      "loss": 1.3145,
+      "theoretical_loss": 3.478171037021008,
+      "tokens_seen": 1870594048
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004374899695073022,
+      "loss": 1.3254,
+      "theoretical_loss": 3.478150617790882,
+      "tokens_seen": 1870725120
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.000437449847536511,
+      "loss": 1.4512,
+      "theoretical_loss": 3.4781302003919303,
+      "tokens_seen": 1870856192
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004374097255657198,
+      "loss": 1.3569,
+      "theoretical_loss": 3.478109784823861,
+      "tokens_seen": 1870987264
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004373696035949286,
+      "loss": 1.4818,
+      "theoretical_loss": 3.4780893710863823,
+      "tokens_seen": 1871118336
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004373294816241374,
+      "loss": 1.4608,
+      "theoretical_loss": 3.4780689591792004,
+      "tokens_seen": 1871249408
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004372893596533462,
+      "loss": 1.4489,
+      "theoretical_loss": 3.478048549102024,
+      "tokens_seen": 1871380480
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043724923768255497,
+      "loss": 1.4421,
+      "theoretical_loss": 3.478028140854561,
+      "tokens_seen": 1871511552
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004372091157117638,
+      "loss": 1.3975,
+      "theoretical_loss": 3.4780077344365186,
+      "tokens_seen": 1871642624
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004371689937409726,
+      "loss": 1.4638,
+      "theoretical_loss": 3.477987329847605,
+      "tokens_seen": 1871773696
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043712887177018133,
+      "loss": 1.3577,
+      "theoretical_loss": 3.477966927087528,
+      "tokens_seen": 1871904768
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004370887497993902,
+      "loss": 1.4332,
+      "theoretical_loss": 3.477946526155996,
+      "tokens_seen": 1872035840
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043704862782859894,
+      "loss": 1.4207,
+      "theoretical_loss": 3.477926127052717,
+      "tokens_seen": 1872166912
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043700850585780775,
+      "loss": 1.3433,
+      "theoretical_loss": 3.4779057297773983,
+      "tokens_seen": 1872297984
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00043696838388701656,
+      "loss": 1.3529,
+      "theoretical_loss": 3.47788533432975,
+      "tokens_seen": 1872429056
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004369282619162253,
+      "loss": 1.3738,
+      "theoretical_loss": 3.4778649407094786,
+      "tokens_seen": 1872560128
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0004368881399454341,
+      "loss": 1.4353,
+      "theoretical_loss": 3.477844548916294,
+      "tokens_seen": 1872691200
+    },
+    {
+      "epoch": 0.13,
+      "objective/train/advantage_avg": 0.4942508935928345,
+      "objective/train/docs_used": 683408,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.4058138132095337,
+      "objective/train/original_loss": 1.4058139324188232,
+      "objective/train/theoretical_loss": 3.477824158949903,
+      "objective/train/tokens_used": 243281376,
+      "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": 1.050668716430664,
+      "objective/train/weighted_lm_loss": 1.4771074056625366,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.477824158949903,
+      "tokens_seen": 1872822272
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004368480179746429,
+      "loss": 1.4073,
+      "theoretical_loss": 3.477824158949903,
+      "tokens_seen": 1872822272
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043680789600385173,
+      "loss": 1.4796,
+      "theoretical_loss": 3.4778037708100156,
+      "tokens_seen": 1872953344
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004367677740330605,
+      "loss": 1.4078,
+      "theoretical_loss": 3.4777833844963397,
+      "tokens_seen": 1873084416
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043672765206226934,
+      "loss": 1.4631,
+      "theoretical_loss": 3.4777630000085837,
+      "tokens_seen": 1873215488
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004366875300914781,
+      "loss": 1.416,
+      "theoretical_loss": 3.4777426173464567,
+      "tokens_seen": 1873346560
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004366474081206869,
+      "loss": 1.417,
+      "theoretical_loss": 3.477722236509668,
+      "tokens_seen": 1873477632
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004366072861498957,
+      "loss": 1.4917,
+      "theoretical_loss": 3.4777018574979253,
+      "tokens_seen": 1873608704
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043656716417910446,
+      "loss": 1.4188,
+      "theoretical_loss": 3.477681480310938,
+      "tokens_seen": 1873739776
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004365270422083133,
+      "loss": 1.3478,
+      "theoretical_loss": 3.4776611049484156,
+      "tokens_seen": 1873870848
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043648692023752207,
+      "loss": 1.2863,
+      "theoretical_loss": 3.477640731410067,
+      "tokens_seen": 1874001920
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004364467982667309,
+      "loss": 1.4165,
+      "theoretical_loss": 3.4776203596956003,
+      "tokens_seen": 1874132992
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004364066762959397,
+      "loss": 1.4497,
+      "theoretical_loss": 3.4775999898047254,
+      "tokens_seen": 1874264064
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043636655432514843,
+      "loss": 1.3819,
+      "theoretical_loss": 3.477579621737152,
+      "tokens_seen": 1874395136
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043632643235435724,
+      "loss": 1.4412,
+      "theoretical_loss": 3.4775592554925887,
+      "tokens_seen": 1874526208
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043628631038356605,
+      "loss": 1.3951,
+      "theoretical_loss": 3.477538891070745,
+      "tokens_seen": 1874657280
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043624618841277485,
+      "loss": 1.4328,
+      "theoretical_loss": 3.4775185284713306,
+      "tokens_seen": 1874788352
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004362060664419836,
+      "loss": 1.4566,
+      "theoretical_loss": 3.4774981676940553,
+      "tokens_seen": 1874919424
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043616594447119247,
+      "loss": 1.3691,
+      "theoretical_loss": 3.477477808738628,
+      "tokens_seen": 1875050496
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004361258225004012,
+      "loss": 1.4266,
+      "theoretical_loss": 3.4774574516047587,
+      "tokens_seen": 1875181568
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043608570052960997,
+      "loss": 1.3665,
+      "theoretical_loss": 3.477437096292157,
+      "tokens_seen": 1875312640
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043604557855881883,
+      "loss": 1.4057,
+      "theoretical_loss": 3.477416742800533,
+      "tokens_seen": 1875443712
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004360054565880276,
+      "loss": 1.4107,
+      "theoretical_loss": 3.477396391129596,
+      "tokens_seen": 1875574784
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043596533461723644,
+      "loss": 1.4038,
+      "theoretical_loss": 3.477376041279056,
+      "tokens_seen": 1875705856
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004359252126464452,
+      "loss": 1.3286,
+      "theoretical_loss": 3.477355693248623,
+      "tokens_seen": 1875836928
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.000435885090675654,
+      "loss": 1.4333,
+      "theoretical_loss": 3.477335347038008,
+      "tokens_seen": 1875968000
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004358449687048628,
+      "loss": 1.3803,
+      "theoretical_loss": 3.47731500264692,
+      "tokens_seen": 1876099072
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004358048467340716,
+      "loss": 1.4167,
+      "theoretical_loss": 3.47729466007507,
+      "tokens_seen": 1876230144
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043576472476328037,
+      "loss": 1.3923,
+      "theoretical_loss": 3.477274319322167,
+      "tokens_seen": 1876361216
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004357246027924892,
+      "loss": 1.4138,
+      "theoretical_loss": 3.477253980387922,
+      "tokens_seen": 1876492288
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.000435684480821698,
+      "loss": 1.3583,
+      "theoretical_loss": 3.4772336432720463,
+      "tokens_seen": 1876623360
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043564435885090673,
+      "loss": 1.4215,
+      "theoretical_loss": 3.4772133079742487,
+      "tokens_seen": 1876754432
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004356042368801156,
+      "loss": 1.4234,
+      "theoretical_loss": 3.4771929744942414,
+      "tokens_seen": 1876885504
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043556411490932434,
+      "loss": 1.4184,
+      "theoretical_loss": 3.477172642831733,
+      "tokens_seen": 1877016576
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043552399293853315,
+      "loss": 1.4336,
+      "theoretical_loss": 3.477152312986436,
+      "tokens_seen": 1877147648
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043548387096774196,
+      "loss": 1.4165,
+      "theoretical_loss": 3.47713198495806,
+      "tokens_seen": 1877278720
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004354437489969507,
+      "loss": 1.3999,
+      "theoretical_loss": 3.4771116587463164,
+      "tokens_seen": 1877409792
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043540362702615957,
+      "loss": 1.4136,
+      "theoretical_loss": 3.4770913343509156,
+      "tokens_seen": 1877540864
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004353635050553683,
+      "loss": 1.4947,
+      "theoretical_loss": 3.477071011771569,
+      "tokens_seen": 1877671936
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043532338308457713,
+      "loss": 1.3959,
+      "theoretical_loss": 3.477050691007987,
+      "tokens_seen": 1877803008
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043528326111378593,
+      "loss": 1.4201,
+      "theoretical_loss": 3.4770303720598807,
+      "tokens_seen": 1877934080
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043524313914299474,
+      "loss": 1.4421,
+      "theoretical_loss": 3.4770100549269616,
+      "tokens_seen": 1878065152
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004352030171722035,
+      "loss": 1.4367,
+      "theoretical_loss": 3.476989739608941,
+      "tokens_seen": 1878196224
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004351628952014123,
+      "loss": 1.4143,
+      "theoretical_loss": 3.476969426105529,
+      "tokens_seen": 1878327296
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004351227732306211,
+      "loss": 1.3732,
+      "theoretical_loss": 3.476949114416439,
+      "tokens_seen": 1878458368
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043508265125982986,
+      "loss": 1.4283,
+      "theoretical_loss": 3.47692880454138,
+      "tokens_seen": 1878589440
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004350425292890387,
+      "loss": 1.3632,
+      "theoretical_loss": 3.476908496480065,
+      "tokens_seen": 1878720512
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043500240731824747,
+      "loss": 1.3135,
+      "theoretical_loss": 3.476888190232205,
+      "tokens_seen": 1878851584
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004349622853474563,
+      "loss": 1.3752,
+      "theoretical_loss": 3.4768678857975113,
+      "tokens_seen": 1878982656
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004349221633766651,
+      "loss": 1.4447,
+      "theoretical_loss": 3.4768475831756964,
+      "tokens_seen": 1879113728
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043488204140587383,
+      "loss": 1.3687,
+      "theoretical_loss": 3.4768272823664708,
+      "tokens_seen": 1879244800
+    },
+    {
+      "epoch": 0.14,
+      "objective/train/advantage_avg": 0.4950573444366455,
+      "objective/train/docs_used": 685672,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2780570983886719,
+      "objective/train/original_loss": 1.2780569791793823,
+      "objective/train/theoretical_loss": 3.4768069833695474,
+      "objective/train/tokens_used": 249834976,
+      "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": 1.0507532358169556,
+      "objective/train/weighted_lm_loss": 1.3429361581802368,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4768069833695474,
+      "tokens_seen": 1879375872
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043484191943508264,
+      "loss": 1.3337,
+      "theoretical_loss": 3.4768069833695474,
+      "tokens_seen": 1879375872
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043480179746429145,
+      "loss": 1.443,
+      "theoretical_loss": 3.4767866861846377,
+      "tokens_seen": 1879506944
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043476167549350025,
+      "loss": 1.3249,
+      "theoretical_loss": 3.476766390811453,
+      "tokens_seen": 1879638016
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043472155352270906,
+      "loss": 1.4682,
+      "theoretical_loss": 3.4767460972497064,
+      "tokens_seen": 1879769088
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043468143155191787,
+      "loss": 1.3702,
+      "theoretical_loss": 3.4767258054991084,
+      "tokens_seen": 1879900160
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004346413095811266,
+      "loss": 1.3874,
+      "theoretical_loss": 3.476705515559372,
+      "tokens_seen": 1880031232
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004346011876103354,
+      "loss": 1.395,
+      "theoretical_loss": 3.47668522743021,
+      "tokens_seen": 1880162304
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043456106563954423,
+      "loss": 1.374,
+      "theoretical_loss": 3.4766649411113337,
+      "tokens_seen": 1880293376
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.000434520943668753,
+      "loss": 1.3788,
+      "theoretical_loss": 3.4766446566024554,
+      "tokens_seen": 1880424448
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043448082169796184,
+      "loss": 1.4146,
+      "theoretical_loss": 3.476624373903288,
+      "tokens_seen": 1880555520
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004344406997271706,
+      "loss": 1.3459,
+      "theoretical_loss": 3.4766040930135436,
+      "tokens_seen": 1880686592
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004344005777563794,
+      "loss": 1.3922,
+      "theoretical_loss": 3.476583813932934,
+      "tokens_seen": 1880817664
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004343604557855882,
+      "loss": 1.387,
+      "theoretical_loss": 3.476563536661174,
+      "tokens_seen": 1880948736
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.000434320333814797,
+      "loss": 1.4434,
+      "theoretical_loss": 3.476543261197973,
+      "tokens_seen": 1881079808
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043428021184400577,
+      "loss": 1.4037,
+      "theoretical_loss": 3.4765229875430466,
+      "tokens_seen": 1881210880
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043424008987321457,
+      "loss": 1.3566,
+      "theoretical_loss": 3.476502715696106,
+      "tokens_seen": 1881341952
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004341999679024234,
+      "loss": 1.3786,
+      "theoretical_loss": 3.476482445656864,
+      "tokens_seen": 1881473024
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043415984593163213,
+      "loss": 1.3883,
+      "theoretical_loss": 3.4764621774250344,
+      "tokens_seen": 1881604096
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.000434119723960841,
+      "loss": 1.3172,
+      "theoretical_loss": 3.4764419110003293,
+      "tokens_seen": 1881735168
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043407960199004974,
+      "loss": 1.4395,
+      "theoretical_loss": 3.476421646382462,
+      "tokens_seen": 1881866240
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004340394800192586,
+      "loss": 1.3826,
+      "theoretical_loss": 3.4764013835711456,
+      "tokens_seen": 1881997312
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043399935804846736,
+      "loss": 1.3931,
+      "theoretical_loss": 3.4763811225660937,
+      "tokens_seen": 1882128384
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004339592360776761,
+      "loss": 1.5057,
+      "theoretical_loss": 3.4763608633670184,
+      "tokens_seen": 1882259456
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043391911410688497,
+      "loss": 1.3474,
+      "theoretical_loss": 3.4763406059736344,
+      "tokens_seen": 1882390528
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004338789921360937,
+      "loss": 1.2992,
+      "theoretical_loss": 3.4763203503856537,
+      "tokens_seen": 1882521600
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043383887016530253,
+      "loss": 1.5014,
+      "theoretical_loss": 3.4763000966027904,
+      "tokens_seen": 1882652672
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043379874819451133,
+      "loss": 1.4079,
+      "theoretical_loss": 3.476279844624758,
+      "tokens_seen": 1882783744
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043375862622372014,
+      "loss": 1.4581,
+      "theoretical_loss": 3.47625959445127,
+      "tokens_seen": 1882914816
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004337185042529289,
+      "loss": 1.3862,
+      "theoretical_loss": 3.47623934608204,
+      "tokens_seen": 1883045888
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004336783822821377,
+      "loss": 1.4365,
+      "theoretical_loss": 3.4762190995167814,
+      "tokens_seen": 1883176960
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004336382603113465,
+      "loss": 1.4771,
+      "theoretical_loss": 3.476198854755208,
+      "tokens_seen": 1883308032
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043359813834055526,
+      "loss": 1.4052,
+      "theoretical_loss": 3.476178611797034,
+      "tokens_seen": 1883439104
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004335580163697641,
+      "loss": 1.4581,
+      "theoretical_loss": 3.476158370641973,
+      "tokens_seen": 1883570176
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043351789439897287,
+      "loss": 1.4026,
+      "theoretical_loss": 3.476138131289739,
+      "tokens_seen": 1883701248
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043347777242818173,
+      "loss": 1.3298,
+      "theoretical_loss": 3.476117893740046,
+      "tokens_seen": 1883832320
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004334376504573905,
+      "loss": 1.4138,
+      "theoretical_loss": 3.4760976579926077,
+      "tokens_seen": 1883963392
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043339752848659923,
+      "loss": 1.3026,
+      "theoretical_loss": 3.476077424047139,
+      "tokens_seen": 1884094464
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004333574065158081,
+      "loss": 1.353,
+      "theoretical_loss": 3.476057191903353,
+      "tokens_seen": 1884225536
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043331728454501685,
+      "loss": 1.3938,
+      "theoretical_loss": 3.4760369615609648,
+      "tokens_seen": 1884356608
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043327716257422565,
+      "loss": 1.3729,
+      "theoretical_loss": 3.476016733019689,
+      "tokens_seen": 1884487680
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043323704060343446,
+      "loss": 1.3604,
+      "theoretical_loss": 3.4759965062792393,
+      "tokens_seen": 1884618752
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043319691863264327,
+      "loss": 1.3243,
+      "theoretical_loss": 3.47597628133933,
+      "tokens_seen": 1884749824
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.000433156796661852,
+      "loss": 1.4426,
+      "theoretical_loss": 3.4759560581996762,
+      "tokens_seen": 1884880896
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004331166746910608,
+      "loss": 1.382,
+      "theoretical_loss": 3.4759358368599926,
+      "tokens_seen": 1885011968
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043307655272026963,
+      "loss": 1.3413,
+      "theoretical_loss": 3.475915617319993,
+      "tokens_seen": 1885143040
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004330364307494784,
+      "loss": 1.3706,
+      "theoretical_loss": 3.475895399579393,
+      "tokens_seen": 1885274112
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043299630877868724,
+      "loss": 1.3684,
+      "theoretical_loss": 3.475875183637907,
+      "tokens_seen": 1885405184
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.000432956186807896,
+      "loss": 1.4297,
+      "theoretical_loss": 3.4758549694952494,
+      "tokens_seen": 1885536256
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004329160648371048,
+      "loss": 1.448,
+      "theoretical_loss": 3.4758347571511363,
+      "tokens_seen": 1885667328
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004328759428663136,
+      "loss": 1.3981,
+      "theoretical_loss": 3.4758145466052817,
+      "tokens_seen": 1885798400
+    },
+    {
+      "epoch": 0.14,
+      "objective/train/advantage_avg": 0.49301230907440186,
+      "objective/train/docs_used": 687569,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.322345495223999,
+      "objective/train/original_loss": 1.322345495223999,
+      "objective/train/theoretical_loss": 3.4757943378574008,
+      "objective/train/tokens_used": 256388576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24363450706005096,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505398511886597,
+      "objective/train/weighted_lm_loss": 1.3890397548675537,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9048373699188232,
+      "theoretical_loss": 3.4757943378574008,
+      "tokens_seen": 1885929472
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043283582089552236,
+      "loss": 1.4512,
+      "theoretical_loss": 3.4757943378574008,
+      "tokens_seen": 1885929472
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004327956989247312,
+      "loss": 1.4693,
+      "theoretical_loss": 3.475774130907209,
+      "tokens_seen": 1886060544
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043275557695393997,
+      "loss": 1.4121,
+      "theoretical_loss": 3.4757539257544208,
+      "tokens_seen": 1886191616
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004327154549831488,
+      "loss": 1.3561,
+      "theoretical_loss": 3.4757337223987523,
+      "tokens_seen": 1886322688
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004326753330123576,
+      "loss": 1.4957,
+      "theoretical_loss": 3.475713520839918,
+      "tokens_seen": 1886453760
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004326352110415664,
+      "loss": 1.5185,
+      "theoretical_loss": 3.4756933210776344,
+      "tokens_seen": 1886584832
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043259508907077514,
+      "loss": 1.3536,
+      "theoretical_loss": 3.4756731231116156,
+      "tokens_seen": 1886715904
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.000432554967099984,
+      "loss": 1.3712,
+      "theoretical_loss": 3.475652926941578,
+      "tokens_seen": 1886846976
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043251484512919276,
+      "loss": 1.4145,
+      "theoretical_loss": 3.4756327325672367,
+      "tokens_seen": 1886978048
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004324747231584015,
+      "loss": 1.459,
+      "theoretical_loss": 3.4756125399883073,
+      "tokens_seen": 1887109120
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043243460118761037,
+      "loss": 1.4386,
+      "theoretical_loss": 3.475592349204506,
+      "tokens_seen": 1887240192
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004323944792168191,
+      "loss": 1.4241,
+      "theoretical_loss": 3.475572160215548,
+      "tokens_seen": 1887371264
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043235435724602793,
+      "loss": 1.347,
+      "theoretical_loss": 3.47555197302115,
+      "tokens_seen": 1887502336
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043231423527523673,
+      "loss": 1.4383,
+      "theoretical_loss": 3.4755317876210263,
+      "tokens_seen": 1887633408
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043227411330444554,
+      "loss": 1.3464,
+      "theoretical_loss": 3.4755116040148946,
+      "tokens_seen": 1887764480
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043223399133365435,
+      "loss": 1.4394,
+      "theoretical_loss": 3.4754914222024698,
+      "tokens_seen": 1887895552
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004321938693628631,
+      "loss": 1.3487,
+      "theoretical_loss": 3.4754712421834677,
+      "tokens_seen": 1888026624
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004321537473920719,
+      "loss": 1.4395,
+      "theoretical_loss": 3.475451063957606,
+      "tokens_seen": 1888157696
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004321136254212807,
+      "loss": 1.3109,
+      "theoretical_loss": 3.475430887524599,
+      "tokens_seen": 1888288768
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004320735034504895,
+      "loss": 1.3865,
+      "theoretical_loss": 3.4754107128841647,
+      "tokens_seen": 1888419840
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043203338147969827,
+      "loss": 1.3555,
+      "theoretical_loss": 3.4753905400360185,
+      "tokens_seen": 1888550912
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043199325950890713,
+      "loss": 1.4695,
+      "theoretical_loss": 3.4753703689798763,
+      "tokens_seen": 1888681984
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004319531375381159,
+      "loss": 1.3193,
+      "theoretical_loss": 3.4753501997154554,
+      "tokens_seen": 1888813056
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043191301556732463,
+      "loss": 1.4067,
+      "theoretical_loss": 3.4753300322424727,
+      "tokens_seen": 1888944128
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0004318728935965335,
+      "loss": 1.4239,
+      "theoretical_loss": 3.475309866560644,
+      "tokens_seen": 1889075200
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00043183277162574225,
+      "loss": 1.3948,
+      "theoretical_loss": 3.475289702669686,
+      "tokens_seen": 1889206272
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00043179264965495105,
+      "loss": 1.3735,
+      "theoretical_loss": 3.4752695405693155,
+      "tokens_seen": 1889337344
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00043175252768415986,
+      "loss": 1.3502,
+      "theoretical_loss": 3.4752493802592497,
+      "tokens_seen": 1889468416
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00043171240571336867,
+      "loss": 1.3376,
+      "theoretical_loss": 3.475229221739205,
+      "tokens_seen": 1889599488
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004316722837425774,
+      "loss": 1.3868,
+      "theoretical_loss": 3.4752090650088983,
+      "tokens_seen": 1889730560
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004316321617717862,
+      "loss": 1.3794,
+      "theoretical_loss": 3.475188910068047,
+      "tokens_seen": 1889861632
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00043159203980099503,
+      "loss": 1.3589,
+      "theoretical_loss": 3.475168756916368,
+      "tokens_seen": 1889992704
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00043155191783020384,
+      "loss": 1.4133,
+      "theoretical_loss": 3.475148605553578,
+      "tokens_seen": 1890123776
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00043151179585941264,
+      "loss": 1.3566,
+      "theoretical_loss": 3.4751284559793945,
+      "tokens_seen": 1890254848
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004314716738886214,
+      "loss": 1.4119,
+      "theoretical_loss": 3.4751083081935343,
+      "tokens_seen": 1890385920
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00043143155191783026,
+      "loss": 1.4843,
+      "theoretical_loss": 3.475088162195716,
+      "tokens_seen": 1890516992
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.000431391429947039,
+      "loss": 1.3532,
+      "theoretical_loss": 3.4750680179856555,
+      "tokens_seen": 1890648064
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00043135130797624776,
+      "loss": 1.3463,
+      "theoretical_loss": 3.475047875563071,
+      "tokens_seen": 1890779136
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004313111860054566,
+      "loss": 1.3708,
+      "theoretical_loss": 3.4750277349276795,
+      "tokens_seen": 1890910208
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00043127106403466537,
+      "loss": 1.4184,
+      "theoretical_loss": 3.4750075960791995,
+      "tokens_seen": 1891041280
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004312309420638742,
+      "loss": 1.3405,
+      "theoretical_loss": 3.4749874590173473,
+      "tokens_seen": 1891172352
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.000431190820093083,
+      "loss": 1.3852,
+      "theoretical_loss": 3.474967323741842,
+      "tokens_seen": 1891303424
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004311506981222918,
+      "loss": 1.4182,
+      "theoretical_loss": 3.4749471902524003,
+      "tokens_seen": 1891434496
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00043111057615150054,
+      "loss": 1.3952,
+      "theoretical_loss": 3.47492705854874,
+      "tokens_seen": 1891565568
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004310704541807094,
+      "loss": 1.2897,
+      "theoretical_loss": 3.4749069286305803,
+      "tokens_seen": 1891696640
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00043103033220991816,
+      "loss": 1.3543,
+      "theoretical_loss": 3.4748868004976377,
+      "tokens_seen": 1891827712
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004309902102391269,
+      "loss": 1.3838,
+      "theoretical_loss": 3.47486667414963,
+      "tokens_seen": 1891958784
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00043095008826833577,
+      "loss": 1.317,
+      "theoretical_loss": 3.474846549586277,
+      "tokens_seen": 1892089856
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004309099662975445,
+      "loss": 1.3757,
+      "theoretical_loss": 3.474826426807295,
+      "tokens_seen": 1892220928
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004308698443267534,
+      "loss": 1.4128,
+      "theoretical_loss": 3.474806305812403,
+      "tokens_seen": 1892352000
+    },
+    {
+      "epoch": 0.15,
+      "objective/train/advantage_avg": 0.49397557973861694,
+      "objective/train/docs_used": 690244,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.4016882181167603,
+      "objective/train/original_loss": 1.4016883373260498,
+      "objective/train/theoretical_loss": 3.47478618660132,
+      "objective/train/tokens_used": 262942176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2446504682302475,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506412982940674,
+      "objective/train/weighted_lm_loss": 1.4726468324661255,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.47478618660132,
+      "tokens_seen": 1892483072
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00043082972235596213,
+      "loss": 1.4139,
+      "theoretical_loss": 3.47478618660132,
+      "tokens_seen": 1892483072
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00043078960038517094,
+      "loss": 1.4717,
+      "theoretical_loss": 3.4747660691737634,
+      "tokens_seen": 1892614144
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00043074947841437975,
+      "loss": 1.3704,
+      "theoretical_loss": 3.4747459535294514,
+      "tokens_seen": 1892745216
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004307093564435885,
+      "loss": 1.3772,
+      "theoretical_loss": 3.474725839668103,
+      "tokens_seen": 1892876288
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004306692344727973,
+      "loss": 1.3592,
+      "theoretical_loss": 3.4747057275894364,
+      "tokens_seen": 1893007360
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004306291125020061,
+      "loss": 1.4048,
+      "theoretical_loss": 3.4746856172931704,
+      "tokens_seen": 1893138432
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004305889905312149,
+      "loss": 1.4004,
+      "theoretical_loss": 3.474665508779024,
+      "tokens_seen": 1893269504
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00043054886856042367,
+      "loss": 1.3591,
+      "theoretical_loss": 3.4746454020467152,
+      "tokens_seen": 1893400576
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00043050874658963253,
+      "loss": 1.3436,
+      "theoretical_loss": 3.474625297095963,
+      "tokens_seen": 1893531648
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004304686246188413,
+      "loss": 1.4484,
+      "theoretical_loss": 3.474605193926487,
+      "tokens_seen": 1893662720
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00043042850264805003,
+      "loss": 1.4606,
+      "theoretical_loss": 3.4745850925380046,
+      "tokens_seen": 1893793792
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004303883806772589,
+      "loss": 1.3664,
+      "theoretical_loss": 3.474564992930236,
+      "tokens_seen": 1893924864
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00043034825870646765,
+      "loss": 1.3931,
+      "theoretical_loss": 3.4745448951028997,
+      "tokens_seen": 1894055936
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004303081367356765,
+      "loss": 1.4276,
+      "theoretical_loss": 3.4745247990557147,
+      "tokens_seen": 1894187008
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00043026801476488526,
+      "loss": 1.3855,
+      "theoretical_loss": 3.4745047047884006,
+      "tokens_seen": 1894318080
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00043022789279409407,
+      "loss": 1.388,
+      "theoretical_loss": 3.4744846123006767,
+      "tokens_seen": 1894449152
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00043018777082330287,
+      "loss": 1.4184,
+      "theoretical_loss": 3.474464521592262,
+      "tokens_seen": 1894580224
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004301476488525116,
+      "loss": 1.3614,
+      "theoretical_loss": 3.4744444326628754,
+      "tokens_seen": 1894711296
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00043010752688172043,
+      "loss": 1.3534,
+      "theoretical_loss": 3.474424345512237,
+      "tokens_seen": 1894842368
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00043006740491092924,
+      "loss": 1.4113,
+      "theoretical_loss": 3.4744042601400658,
+      "tokens_seen": 1894973440
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00043002728294013804,
+      "loss": 1.4264,
+      "theoretical_loss": 3.474384176546082,
+      "tokens_seen": 1895104512
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004299871609693468,
+      "loss": 1.433,
+      "theoretical_loss": 3.4743640947300047,
+      "tokens_seen": 1895235584
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042994703899855565,
+      "loss": 1.3993,
+      "theoretical_loss": 3.4743440146915536,
+      "tokens_seen": 1895366656
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004299069170277644,
+      "loss": 1.4461,
+      "theoretical_loss": 3.4743239364304483,
+      "tokens_seen": 1895497728
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042986679505697316,
+      "loss": 1.4458,
+      "theoretical_loss": 3.474303859946409,
+      "tokens_seen": 1895628800
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.000429826673086182,
+      "loss": 1.3885,
+      "theoretical_loss": 3.4742837852391553,
+      "tokens_seen": 1895759872
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042978655111539077,
+      "loss": 1.4188,
+      "theoretical_loss": 3.4742637123084075,
+      "tokens_seen": 1895890944
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004297464291445996,
+      "loss": 1.3917,
+      "theoretical_loss": 3.474243641153885,
+      "tokens_seen": 1896022016
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004297063071738084,
+      "loss": 1.4183,
+      "theoretical_loss": 3.474223571775308,
+      "tokens_seen": 1896153088
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004296661852030172,
+      "loss": 1.3482,
+      "theoretical_loss": 3.4742035041723964,
+      "tokens_seen": 1896284160
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.000429626063232226,
+      "loss": 1.3199,
+      "theoretical_loss": 3.4741834383448715,
+      "tokens_seen": 1896415232
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004295859412614348,
+      "loss": 1.3903,
+      "theoretical_loss": 3.474163374292452,
+      "tokens_seen": 1896546304
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042954581929064356,
+      "loss": 1.4078,
+      "theoretical_loss": 3.474143312014859,
+      "tokens_seen": 1896677376
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042950569731985236,
+      "loss": 1.412,
+      "theoretical_loss": 3.4741232515118137,
+      "tokens_seen": 1896808448
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042946557534906117,
+      "loss": 1.3268,
+      "theoretical_loss": 3.474103192783035,
+      "tokens_seen": 1896939520
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004294254533782699,
+      "loss": 1.3857,
+      "theoretical_loss": 3.474083135828244,
+      "tokens_seen": 1897070592
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004293853314074788,
+      "loss": 1.3985,
+      "theoretical_loss": 3.474063080647161,
+      "tokens_seen": 1897201664
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042934520943668753,
+      "loss": 1.4297,
+      "theoretical_loss": 3.4740430272395075,
+      "tokens_seen": 1897332736
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042930508746589634,
+      "loss": 1.3985,
+      "theoretical_loss": 3.4740229756050036,
+      "tokens_seen": 1897463808
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042926496549510515,
+      "loss": 1.3523,
+      "theoretical_loss": 3.4740029257433696,
+      "tokens_seen": 1897594880
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004292248435243139,
+      "loss": 1.3716,
+      "theoretical_loss": 3.4739828776543265,
+      "tokens_seen": 1897725952
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004291847215535227,
+      "loss": 1.3374,
+      "theoretical_loss": 3.4739628313375963,
+      "tokens_seen": 1897857024
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004291445995827315,
+      "loss": 1.3625,
+      "theoretical_loss": 3.4739427867928985,
+      "tokens_seen": 1897988096
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004291044776119403,
+      "loss": 1.3591,
+      "theoretical_loss": 3.473922744019955,
+      "tokens_seen": 1898119168
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004290643556411491,
+      "loss": 1.425,
+      "theoretical_loss": 3.473902703018486,
+      "tokens_seen": 1898250240
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042902423367035793,
+      "loss": 1.3815,
+      "theoretical_loss": 3.4738826637882134,
+      "tokens_seen": 1898381312
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004289841116995667,
+      "loss": 1.3888,
+      "theoretical_loss": 3.4738626263288586,
+      "tokens_seen": 1898512384
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004289439897287755,
+      "loss": 1.3815,
+      "theoretical_loss": 3.473842590640142,
+      "tokens_seen": 1898643456
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004289038677579843,
+      "loss": 1.3957,
+      "theoretical_loss": 3.4738225567217853,
+      "tokens_seen": 1898774528
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042886374578719305,
+      "loss": 1.3619,
+      "theoretical_loss": 3.47380252457351,
+      "tokens_seen": 1898905600
+    },
+    {
+      "epoch": 0.15,
+      "objective/train/advantage_avg": 0.491893470287323,
+      "objective/train/docs_used": 692658,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3227930068969727,
+      "objective/train/original_loss": 1.322792887687683,
+      "objective/train/theoretical_loss": 3.4737824941950373,
+      "objective/train/tokens_used": 269495776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24248704314231873,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504220724105835,
+      "objective/train/weighted_lm_loss": 1.3894788026809692,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4737824941950373,
+      "tokens_seen": 1899036672
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004288236238164019,
+      "loss": 1.4005,
+      "theoretical_loss": 3.4737824941950373,
+      "tokens_seen": 1899036672
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042878350184561066,
+      "loss": 1.4405,
+      "theoretical_loss": 3.473762465586089,
+      "tokens_seen": 1899167744
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042874337987481946,
+      "loss": 1.3826,
+      "theoretical_loss": 3.4737424387463864,
+      "tokens_seen": 1899298816
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042870325790402827,
+      "loss": 1.403,
+      "theoretical_loss": 3.4737224136756515,
+      "tokens_seen": 1899429888
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.000428663135933237,
+      "loss": 1.3951,
+      "theoretical_loss": 3.4737023903736053,
+      "tokens_seen": 1899560960
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042862301396244583,
+      "loss": 1.3059,
+      "theoretical_loss": 3.4736823688399703,
+      "tokens_seen": 1899692032
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042858289199165464,
+      "loss": 1.3787,
+      "theoretical_loss": 3.473662349074468,
+      "tokens_seen": 1899823104
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042854277002086344,
+      "loss": 1.3454,
+      "theoretical_loss": 3.47364233107682,
+      "tokens_seen": 1899954176
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004285026480500722,
+      "loss": 1.3301,
+      "theoretical_loss": 3.473622314846749,
+      "tokens_seen": 1900085248
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042846252607928105,
+      "loss": 1.4114,
+      "theoretical_loss": 3.473602300383976,
+      "tokens_seen": 1900216320
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004284224041084898,
+      "loss": 1.3565,
+      "theoretical_loss": 3.473582287688224,
+      "tokens_seen": 1900347392
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004283822821376986,
+      "loss": 1.3889,
+      "theoretical_loss": 3.473562276759215,
+      "tokens_seen": 1900478464
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004283421601669074,
+      "loss": 1.3495,
+      "theoretical_loss": 3.4735422675966703,
+      "tokens_seen": 1900609536
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042830203819611617,
+      "loss": 1.442,
+      "theoretical_loss": 3.4735222602003133,
+      "tokens_seen": 1900740608
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042826191622532503,
+      "loss": 1.361,
+      "theoretical_loss": 3.4735022545698655,
+      "tokens_seen": 1900871680
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004282217942545338,
+      "loss": 1.3445,
+      "theoretical_loss": 3.47348225070505,
+      "tokens_seen": 1901002752
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004281816722837426,
+      "loss": 1.3673,
+      "theoretical_loss": 3.4734622486055886,
+      "tokens_seen": 1901133824
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004281415503129514,
+      "loss": 1.4427,
+      "theoretical_loss": 3.473442248271204,
+      "tokens_seen": 1901264896
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042810142834216015,
+      "loss": 1.4224,
+      "theoretical_loss": 3.473422249701619,
+      "tokens_seen": 1901395968
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042806130637136896,
+      "loss": 1.4414,
+      "theoretical_loss": 3.4734022528965562,
+      "tokens_seen": 1901527040
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042802118440057776,
+      "loss": 1.4201,
+      "theoretical_loss": 3.4733822578557385,
+      "tokens_seen": 1901658112
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042798106242978657,
+      "loss": 1.3934,
+      "theoretical_loss": 3.4733622645788875,
+      "tokens_seen": 1901789184
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004279409404589953,
+      "loss": 1.4225,
+      "theoretical_loss": 3.473342273065728,
+      "tokens_seen": 1901920256
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004279008184882042,
+      "loss": 1.4675,
+      "theoretical_loss": 3.473322283315981,
+      "tokens_seen": 1902051328
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042786069651741293,
+      "loss": 1.4256,
+      "theoretical_loss": 3.47330229532937,
+      "tokens_seen": 1902182400
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004278205745466218,
+      "loss": 1.3714,
+      "theoretical_loss": 3.4732823091056186,
+      "tokens_seen": 1902313472
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042778045257583054,
+      "loss": 1.3847,
+      "theoretical_loss": 3.4732623246444496,
+      "tokens_seen": 1902444544
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004277403306050393,
+      "loss": 1.4093,
+      "theoretical_loss": 3.473242341945586,
+      "tokens_seen": 1902575616
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042770020863424816,
+      "loss": 1.2655,
+      "theoretical_loss": 3.4732223610087507,
+      "tokens_seen": 1902706688
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004276600866634569,
+      "loss": 1.4533,
+      "theoretical_loss": 3.473202381833668,
+      "tokens_seen": 1902837760
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004276199646926657,
+      "loss": 1.367,
+      "theoretical_loss": 3.47318240442006,
+      "tokens_seen": 1902968832
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004275798427218745,
+      "loss": 1.3048,
+      "theoretical_loss": 3.473162428767651,
+      "tokens_seen": 1903099904
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042753972075108333,
+      "loss": 1.3906,
+      "theoretical_loss": 3.4731424548761636,
+      "tokens_seen": 1903230976
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004274995987802921,
+      "loss": 1.3822,
+      "theoretical_loss": 3.473122482745322,
+      "tokens_seen": 1903362048
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004274594768095009,
+      "loss": 1.4679,
+      "theoretical_loss": 3.47310251237485,
+      "tokens_seen": 1903493120
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004274193548387097,
+      "loss": 1.382,
+      "theoretical_loss": 3.47308254376447,
+      "tokens_seen": 1903624192
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042737923286791845,
+      "loss": 1.3254,
+      "theoretical_loss": 3.4730625769139074,
+      "tokens_seen": 1903755264
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004273391108971273,
+      "loss": 1.3249,
+      "theoretical_loss": 3.4730426118228843,
+      "tokens_seen": 1903886336
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042729898892633606,
+      "loss": 1.3686,
+      "theoretical_loss": 3.473022648491126,
+      "tokens_seen": 1904017408
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042725886695554486,
+      "loss": 1.4147,
+      "theoretical_loss": 3.4730026869183552,
+      "tokens_seen": 1904148480
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042721874498475367,
+      "loss": 1.3265,
+      "theoretical_loss": 3.472982727104296,
+      "tokens_seen": 1904279552
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004271786230139624,
+      "loss": 1.3482,
+      "theoretical_loss": 3.4729627690486735,
+      "tokens_seen": 1904410624
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004271385010431713,
+      "loss": 1.3694,
+      "theoretical_loss": 3.47294281275121,
+      "tokens_seen": 1904541696
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042709837907238004,
+      "loss": 1.2925,
+      "theoretical_loss": 3.472922858211632,
+      "tokens_seen": 1904672768
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042705825710158884,
+      "loss": 1.3002,
+      "theoretical_loss": 3.4729029054296614,
+      "tokens_seen": 1904803840
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042701813513079765,
+      "loss": 1.3453,
+      "theoretical_loss": 3.4728829544050237,
+      "tokens_seen": 1904934912
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042697801316000645,
+      "loss": 1.3794,
+      "theoretical_loss": 3.4728630051374427,
+      "tokens_seen": 1905065984
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004269378911892152,
+      "loss": 1.457,
+      "theoretical_loss": 3.4728430576266436,
+      "tokens_seen": 1905197056
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.000426897769218424,
+      "loss": 1.4102,
+      "theoretical_loss": 3.4728231118723496,
+      "tokens_seen": 1905328128
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0004268576472476328,
+      "loss": 1.3641,
+      "theoretical_loss": 3.472803167874287,
+      "tokens_seen": 1905459200
+    },
+    {
+      "epoch": 0.15,
+      "objective/train/advantage_avg": 0.4944094121456146,
+      "objective/train/docs_used": 694687,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2877788543701172,
+      "objective/train/original_loss": 1.2877788543701172,
+      "objective/train/theoretical_loss": 3.472783225632178,
+      "objective/train/tokens_used": 276049376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24509765207767487,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506869554519653,
+      "objective/train/weighted_lm_loss": 1.3530761003494263,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.472783225632178,
+      "tokens_seen": 1905590272
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042681752527684157,
+      "loss": 1.335,
+      "theoretical_loss": 3.472783225632178,
+      "tokens_seen": 1905590272
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00042677740330605043,
+      "loss": 1.3958,
+      "theoretical_loss": 3.472763285145749,
+      "tokens_seen": 1905721344
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004267372813352592,
+      "loss": 1.423,
+      "theoretical_loss": 3.472743346414724,
+      "tokens_seen": 1905852416
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.000426697159364468,
+      "loss": 1.3349,
+      "theoretical_loss": 3.472723409438828,
+      "tokens_seen": 1905983488
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004266570373936768,
+      "loss": 1.4693,
+      "theoretical_loss": 3.472703474217786,
+      "tokens_seen": 1906114560
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042661691542288555,
+      "loss": 1.3642,
+      "theoretical_loss": 3.472683540751323,
+      "tokens_seen": 1906245632
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042657679345209435,
+      "loss": 1.351,
+      "theoretical_loss": 3.472663609039163,
+      "tokens_seen": 1906376704
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042653667148130316,
+      "loss": 1.3603,
+      "theoretical_loss": 3.4726436790810316,
+      "tokens_seen": 1906507776
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042649654951051197,
+      "loss": 1.3873,
+      "theoretical_loss": 3.4726237508766538,
+      "tokens_seen": 1906638848
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004264564275397208,
+      "loss": 1.4874,
+      "theoretical_loss": 3.472603824425755,
+      "tokens_seen": 1906769920
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004264163055689296,
+      "loss": 1.402,
+      "theoretical_loss": 3.4725838997280603,
+      "tokens_seen": 1906900992
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042637618359813833,
+      "loss": 1.4832,
+      "theoretical_loss": 3.4725639767832948,
+      "tokens_seen": 1907032064
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004263360616273472,
+      "loss": 1.4207,
+      "theoretical_loss": 3.4725440555911837,
+      "tokens_seen": 1907163136
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042629593965655594,
+      "loss": 1.3468,
+      "theoretical_loss": 3.4725241361514527,
+      "tokens_seen": 1907294208
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004262558176857647,
+      "loss": 1.3761,
+      "theoretical_loss": 3.472504218463827,
+      "tokens_seen": 1907425280
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042621569571497356,
+      "loss": 1.3741,
+      "theoretical_loss": 3.4724843025280325,
+      "tokens_seen": 1907556352
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004261755737441823,
+      "loss": 1.3534,
+      "theoretical_loss": 3.472464388343794,
+      "tokens_seen": 1907687424
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004261354517733911,
+      "loss": 1.431,
+      "theoretical_loss": 3.4724444759108373,
+      "tokens_seen": 1907818496
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004260953298025999,
+      "loss": 1.4445,
+      "theoretical_loss": 3.472424565228889,
+      "tokens_seen": 1907949568
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042605520783180873,
+      "loss": 1.375,
+      "theoretical_loss": 3.472404656297674,
+      "tokens_seen": 1908080640
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004260150858610175,
+      "loss": 1.3749,
+      "theoretical_loss": 3.4723847491169186,
+      "tokens_seen": 1908211712
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004259749638902263,
+      "loss": 1.3385,
+      "theoretical_loss": 3.472364843686348,
+      "tokens_seen": 1908342784
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004259348419194351,
+      "loss": 1.3252,
+      "theoretical_loss": 3.472344940005689,
+      "tokens_seen": 1908473856
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042589471994864385,
+      "loss": 1.4862,
+      "theoretical_loss": 3.4723250380746666,
+      "tokens_seen": 1908604928
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004258545979778527,
+      "loss": 1.3673,
+      "theoretical_loss": 3.4723051378930077,
+      "tokens_seen": 1908736000
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042581447600706146,
+      "loss": 1.3717,
+      "theoretical_loss": 3.472285239460438,
+      "tokens_seen": 1908867072
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004257743540362703,
+      "loss": 1.3388,
+      "theoretical_loss": 3.4722653427766836,
+      "tokens_seen": 1908998144
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042573423206547907,
+      "loss": 1.4272,
+      "theoretical_loss": 3.4722454478414715,
+      "tokens_seen": 1909129216
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004256941100946878,
+      "loss": 1.4218,
+      "theoretical_loss": 3.472225554654527,
+      "tokens_seen": 1909260288
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004256539881238967,
+      "loss": 1.3339,
+      "theoretical_loss": 3.4722056632155773,
+      "tokens_seen": 1909391360
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042561386615310543,
+      "loss": 1.38,
+      "theoretical_loss": 3.472185773524348,
+      "tokens_seen": 1909522432
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042557374418231424,
+      "loss": 1.3789,
+      "theoretical_loss": 3.472165885580566,
+      "tokens_seen": 1909653504
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042553362221152305,
+      "loss": 1.3811,
+      "theoretical_loss": 3.472145999383958,
+      "tokens_seen": 1909784576
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042549350024073185,
+      "loss": 1.2678,
+      "theoretical_loss": 3.472126114934251,
+      "tokens_seen": 1909915648
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004254533782699406,
+      "loss": 1.4248,
+      "theoretical_loss": 3.4721062322311704,
+      "tokens_seen": 1910046720
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004254132562991494,
+      "loss": 1.4677,
+      "theoretical_loss": 3.472086351274444,
+      "tokens_seen": 1910177792
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004253731343283582,
+      "loss": 1.3775,
+      "theoretical_loss": 3.4720664720637986,
+      "tokens_seen": 1910308864
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042533301235756697,
+      "loss": 1.3154,
+      "theoretical_loss": 3.4720465945989605,
+      "tokens_seen": 1910439936
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042529289038677583,
+      "loss": 1.3959,
+      "theoretical_loss": 3.4720267188796567,
+      "tokens_seen": 1910571008
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004252527684159846,
+      "loss": 1.3532,
+      "theoretical_loss": 3.472006844905615,
+      "tokens_seen": 1910702080
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042521264644519344,
+      "loss": 1.3667,
+      "theoretical_loss": 3.471986972676561,
+      "tokens_seen": 1910833152
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004251725244744022,
+      "loss": 1.3692,
+      "theoretical_loss": 3.471967102192223,
+      "tokens_seen": 1910964224
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042513240250361095,
+      "loss": 1.4086,
+      "theoretical_loss": 3.471947233452328,
+      "tokens_seen": 1911095296
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004250922805328198,
+      "loss": 1.3572,
+      "theoretical_loss": 3.471927366456603,
+      "tokens_seen": 1911226368
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042505215856202856,
+      "loss": 1.2964,
+      "theoretical_loss": 3.471907501204775,
+      "tokens_seen": 1911357440
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042501203659123737,
+      "loss": 1.3867,
+      "theoretical_loss": 3.4718876376965717,
+      "tokens_seen": 1911488512
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004249719146204462,
+      "loss": 1.4191,
+      "theoretical_loss": 3.4718677759317202,
+      "tokens_seen": 1911619584
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.000424931792649655,
+      "loss": 1.3268,
+      "theoretical_loss": 3.4718479159099487,
+      "tokens_seen": 1911750656
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042489167067886373,
+      "loss": 1.4197,
+      "theoretical_loss": 3.471828057630984,
+      "tokens_seen": 1911881728
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042485154870807254,
+      "loss": 1.4548,
+      "theoretical_loss": 3.471808201094554,
+      "tokens_seen": 1912012800
+    },
+    {
+      "epoch": 0.16,
+      "objective/train/advantage_avg": 0.49468478560447693,
+      "objective/train/docs_used": 697086,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3361613750457764,
+      "objective/train/original_loss": 1.3361610174179077,
+      "objective/train/theoretical_loss": 3.4717883463003867,
+      "objective/train/tokens_used": 282602976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24504251778125763,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050714135169983,
+      "objective/train/weighted_lm_loss": 1.4039626121520996,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4717883463003867,
+      "tokens_seen": 1912143872
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042481142673728134,
+      "loss": 1.4055,
+      "theoretical_loss": 3.4717883463003867,
+      "tokens_seen": 1912143872
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004247713047664901,
+      "loss": 1.4412,
+      "theoretical_loss": 3.471768493248209,
+      "tokens_seen": 1912274944
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042473118279569896,
+      "loss": 1.3733,
+      "theoretical_loss": 3.471748641937749,
+      "tokens_seen": 1912406016
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004246910608249077,
+      "loss": 1.364,
+      "theoretical_loss": 3.4717287923687348,
+      "tokens_seen": 1912537088
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004246509388541165,
+      "loss": 1.3218,
+      "theoretical_loss": 3.471708944540895,
+      "tokens_seen": 1912668160
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004246108168833253,
+      "loss": 1.4132,
+      "theoretical_loss": 3.4716890984539557,
+      "tokens_seen": 1912799232
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042457069491253413,
+      "loss": 1.3795,
+      "theoretical_loss": 3.4716692541076464,
+      "tokens_seen": 1912930304
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042453057294174293,
+      "loss": 1.3713,
+      "theoretical_loss": 3.4716494115016943,
+      "tokens_seen": 1913061376
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004244904509709517,
+      "loss": 1.3496,
+      "theoretical_loss": 3.4716295706358284,
+      "tokens_seen": 1913192448
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004244503290001605,
+      "loss": 1.4255,
+      "theoretical_loss": 3.4716097315097763,
+      "tokens_seen": 1913323520
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004244102070293693,
+      "loss": 1.3506,
+      "theoretical_loss": 3.471589894123266,
+      "tokens_seen": 1913454592
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004243700850585781,
+      "loss": 1.3789,
+      "theoretical_loss": 3.471570058476027,
+      "tokens_seen": 1913585664
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042432996308778686,
+      "loss": 1.3561,
+      "theoretical_loss": 3.4715502245677867,
+      "tokens_seen": 1913716736
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004242898411169957,
+      "loss": 1.3276,
+      "theoretical_loss": 3.471530392398274,
+      "tokens_seen": 1913847808
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042424971914620447,
+      "loss": 1.4229,
+      "theoretical_loss": 3.471510561967217,
+      "tokens_seen": 1913978880
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004242095971754132,
+      "loss": 1.3093,
+      "theoretical_loss": 3.4714907332743445,
+      "tokens_seen": 1914109952
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004241694752046221,
+      "loss": 1.2875,
+      "theoretical_loss": 3.4714709063193854,
+      "tokens_seen": 1914241024
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042412935323383083,
+      "loss": 1.3576,
+      "theoretical_loss": 3.4714510811020682,
+      "tokens_seen": 1914372096
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042408923126303964,
+      "loss": 1.4611,
+      "theoretical_loss": 3.471431257622121,
+      "tokens_seen": 1914503168
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042404910929224845,
+      "loss": 1.3934,
+      "theoretical_loss": 3.471411435879274,
+      "tokens_seen": 1914634240
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042400898732145725,
+      "loss": 1.283,
+      "theoretical_loss": 3.4713916158732547,
+      "tokens_seen": 1914765312
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042396886535066606,
+      "loss": 1.4162,
+      "theoretical_loss": 3.4713717976037928,
+      "tokens_seen": 1914896384
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004239287433798748,
+      "loss": 1.425,
+      "theoretical_loss": 3.471351981070617,
+      "tokens_seen": 1915027456
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004238886214090836,
+      "loss": 1.4585,
+      "theoretical_loss": 3.4713321662734566,
+      "tokens_seen": 1915158528
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004238484994382924,
+      "loss": 1.422,
+      "theoretical_loss": 3.471312353212041,
+      "tokens_seen": 1915289600
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042380837746750123,
+      "loss": 1.3958,
+      "theoretical_loss": 3.4712925418860987,
+      "tokens_seen": 1915420672
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042376825549671,
+      "loss": 1.4051,
+      "theoretical_loss": 3.471272732295359,
+      "tokens_seen": 1915551744
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042372813352591884,
+      "loss": 1.376,
+      "theoretical_loss": 3.4712529244395514,
+      "tokens_seen": 1915682816
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004236880115551276,
+      "loss": 1.4327,
+      "theoretical_loss": 3.4712331183184055,
+      "tokens_seen": 1915813888
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042364788958433635,
+      "loss": 1.4543,
+      "theoretical_loss": 3.471213313931651,
+      "tokens_seen": 1915944960
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004236077676135452,
+      "loss": 1.351,
+      "theoretical_loss": 3.4711935112790164,
+      "tokens_seen": 1916076032
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042356764564275396,
+      "loss": 1.4428,
+      "theoretical_loss": 3.4711737103602323,
+      "tokens_seen": 1916207104
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042352752367196277,
+      "loss": 1.426,
+      "theoretical_loss": 3.471153911175027,
+      "tokens_seen": 1916338176
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004234874017011716,
+      "loss": 1.3139,
+      "theoretical_loss": 3.4711341137231315,
+      "tokens_seen": 1916469248
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004234472797303804,
+      "loss": 1.3871,
+      "theoretical_loss": 3.4711143180042745,
+      "tokens_seen": 1916600320
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042340715775958913,
+      "loss": 1.3833,
+      "theoretical_loss": 3.4710945240181865,
+      "tokens_seen": 1916731392
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042336703578879794,
+      "loss": 1.3298,
+      "theoretical_loss": 3.471074731764597,
+      "tokens_seen": 1916862464
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042332691381800674,
+      "loss": 1.4206,
+      "theoretical_loss": 3.4710549412432363,
+      "tokens_seen": 1916993536
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042328679184721555,
+      "loss": 1.336,
+      "theoretical_loss": 3.471035152453834,
+      "tokens_seen": 1917124608
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042324666987642436,
+      "loss": 1.3905,
+      "theoretical_loss": 3.47101536539612,
+      "tokens_seen": 1917255680
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004232065479056331,
+      "loss": 1.3492,
+      "theoretical_loss": 3.470995580069825,
+      "tokens_seen": 1917386752
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042316642593484197,
+      "loss": 1.4274,
+      "theoretical_loss": 3.4709757964746784,
+      "tokens_seen": 1917517824
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004231263039640507,
+      "loss": 1.3863,
+      "theoretical_loss": 3.470956014610411,
+      "tokens_seen": 1917648896
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042308618199325953,
+      "loss": 1.4353,
+      "theoretical_loss": 3.4709362344767527,
+      "tokens_seen": 1917779968
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042304606002246833,
+      "loss": 1.3505,
+      "theoretical_loss": 3.470916456073434,
+      "tokens_seen": 1917911040
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004230059380516771,
+      "loss": 1.3456,
+      "theoretical_loss": 3.4708966794001856,
+      "tokens_seen": 1918042112
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004229658160808859,
+      "loss": 1.4093,
+      "theoretical_loss": 3.4708769044567376,
+      "tokens_seen": 1918173184
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004229256941100947,
+      "loss": 1.3892,
+      "theoretical_loss": 3.47085713124282,
+      "tokens_seen": 1918304256
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004228855721393035,
+      "loss": 1.4033,
+      "theoretical_loss": 3.470837359758164,
+      "tokens_seen": 1918435328
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042284545016851226,
+      "loss": 1.3696,
+      "theoretical_loss": 3.4708175900025005,
+      "tokens_seen": 1918566400
+    },
+    {
+      "epoch": 0.16,
+      "objective/train/advantage_avg": 0.49230635166168213,
+      "objective/train/docs_used": 699453,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3334330320358276,
+      "objective/train/original_loss": 1.3334331512451172,
+      "objective/train/theoretical_loss": 3.4707978219755597,
+      "objective/train/tokens_used": 289156576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24277792870998383,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504647493362427,
+      "objective/train/weighted_lm_loss": 1.4007240533828735,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4707978219755597,
+      "tokens_seen": 1918697472
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004228053281977211,
+      "loss": 1.4165,
+      "theoretical_loss": 3.4707978219755597,
+      "tokens_seen": 1918697472
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042276520622692987,
+      "loss": 1.3598,
+      "theoretical_loss": 3.470778055677073,
+      "tokens_seen": 1918828544
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004227250842561386,
+      "loss": 1.3803,
+      "theoretical_loss": 3.4707582911067703,
+      "tokens_seen": 1918959616
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004226849622853475,
+      "loss": 1.3981,
+      "theoretical_loss": 3.4707385282643832,
+      "tokens_seen": 1919090688
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042264484031455623,
+      "loss": 1.3997,
+      "theoretical_loss": 3.470718767149642,
+      "tokens_seen": 1919221760
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004226047183437651,
+      "loss": 1.3908,
+      "theoretical_loss": 3.4706990077622777,
+      "tokens_seen": 1919352832
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042256459637297385,
+      "loss": 1.3774,
+      "theoretical_loss": 3.4706792501020223,
+      "tokens_seen": 1919483904
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042252447440218265,
+      "loss": 1.3189,
+      "theoretical_loss": 3.4706594941686064,
+      "tokens_seen": 1919614976
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042248435243139146,
+      "loss": 1.3345,
+      "theoretical_loss": 3.4706397399617606,
+      "tokens_seen": 1919746048
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004224442304606002,
+      "loss": 1.3753,
+      "theoretical_loss": 3.470619987481217,
+      "tokens_seen": 1919877120
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.000422404108489809,
+      "loss": 1.365,
+      "theoretical_loss": 3.470600236726707,
+      "tokens_seen": 1920008192
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004223639865190178,
+      "loss": 1.4322,
+      "theoretical_loss": 3.4705804876979602,
+      "tokens_seen": 1920139264
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042232386454822663,
+      "loss": 1.3413,
+      "theoretical_loss": 3.4705607403947107,
+      "tokens_seen": 1920270336
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004222837425774354,
+      "loss": 1.3218,
+      "theoretical_loss": 3.470540994816688,
+      "tokens_seen": 1920401408
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042224362060664424,
+      "loss": 1.2498,
+      "theoretical_loss": 3.470521250963624,
+      "tokens_seen": 1920532480
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.000422203498635853,
+      "loss": 1.4037,
+      "theoretical_loss": 3.470501508835251,
+      "tokens_seen": 1920663552
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042216337666506175,
+      "loss": 1.4492,
+      "theoretical_loss": 3.4704817684313,
+      "tokens_seen": 1920794624
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004221232546942706,
+      "loss": 1.3379,
+      "theoretical_loss": 3.470462029751503,
+      "tokens_seen": 1920925696
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042208313272347936,
+      "loss": 1.347,
+      "theoretical_loss": 3.4704422927955916,
+      "tokens_seen": 1921056768
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004220430107526882,
+      "loss": 1.3912,
+      "theoretical_loss": 3.4704225575632974,
+      "tokens_seen": 1921187840
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042200288878189697,
+      "loss": 1.3817,
+      "theoretical_loss": 3.4704028240543527,
+      "tokens_seen": 1921318912
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004219627668111058,
+      "loss": 1.407,
+      "theoretical_loss": 3.4703830922684897,
+      "tokens_seen": 1921449984
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004219226448403146,
+      "loss": 1.4145,
+      "theoretical_loss": 3.4703633622054406,
+      "tokens_seen": 1921581056
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042188252286952334,
+      "loss": 1.3363,
+      "theoretical_loss": 3.4703436338649363,
+      "tokens_seen": 1921712128
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042184240089873214,
+      "loss": 1.3229,
+      "theoretical_loss": 3.47032390724671,
+      "tokens_seen": 1921843200
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042180227892794095,
+      "loss": 1.3749,
+      "theoretical_loss": 3.4703041823504934,
+      "tokens_seen": 1921974272
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00042176215695714976,
+      "loss": 1.4074,
+      "theoretical_loss": 3.4702844591760185,
+      "tokens_seen": 1922105344
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0004217220349863585,
+      "loss": 1.3848,
+      "theoretical_loss": 3.470264737723019,
+      "tokens_seen": 1922236416
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00042168191301556737,
+      "loss": 1.3804,
+      "theoretical_loss": 3.470245017991225,
+      "tokens_seen": 1922367488
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004216417910447761,
+      "loss": 1.3626,
+      "theoretical_loss": 3.4702252999803713,
+      "tokens_seen": 1922498560
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00042160166907398493,
+      "loss": 1.3865,
+      "theoretical_loss": 3.4702055836901886,
+      "tokens_seen": 1922629632
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00042156154710319373,
+      "loss": 1.4315,
+      "theoretical_loss": 3.470185869120411,
+      "tokens_seen": 1922760704
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004215214251324025,
+      "loss": 1.3757,
+      "theoretical_loss": 3.4701661562707695,
+      "tokens_seen": 1922891776
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004214813031616113,
+      "loss": 1.2424,
+      "theoretical_loss": 3.4701464451409985,
+      "tokens_seen": 1923022848
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004214411811908201,
+      "loss": 1.4942,
+      "theoretical_loss": 3.4701267357308287,
+      "tokens_seen": 1923153920
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004214010592200289,
+      "loss": 1.399,
+      "theoretical_loss": 3.470107028039995,
+      "tokens_seen": 1923284992
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004213609372492377,
+      "loss": 1.362,
+      "theoretical_loss": 3.4700873220682285,
+      "tokens_seen": 1923416064
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004213208152784465,
+      "loss": 1.2706,
+      "theoretical_loss": 3.4700676178152636,
+      "tokens_seen": 1923547136
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00042128069330765527,
+      "loss": 1.3644,
+      "theoretical_loss": 3.4700479152808326,
+      "tokens_seen": 1923678208
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004212405713368641,
+      "loss": 1.3017,
+      "theoretical_loss": 3.470028214464668,
+      "tokens_seen": 1923809280
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004212004493660729,
+      "loss": 1.3238,
+      "theoretical_loss": 3.4700085153665032,
+      "tokens_seen": 1923940352
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00042116032739528163,
+      "loss": 1.3651,
+      "theoretical_loss": 3.4699888179860725,
+      "tokens_seen": 1924071424
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004211202054244905,
+      "loss": 1.3832,
+      "theoretical_loss": 3.4699691223231075,
+      "tokens_seen": 1924202496
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00042108008345369925,
+      "loss": 1.3622,
+      "theoretical_loss": 3.469949428377342,
+      "tokens_seen": 1924333568
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00042103996148290805,
+      "loss": 1.4016,
+      "theoretical_loss": 3.4699297361485097,
+      "tokens_seen": 1924464640
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00042099983951211686,
+      "loss": 1.3676,
+      "theoretical_loss": 3.469910045636344,
+      "tokens_seen": 1924595712
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004209597175413256,
+      "loss": 1.3882,
+      "theoretical_loss": 3.4698903568405783,
+      "tokens_seen": 1924726784
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004209195955705344,
+      "loss": 1.4111,
+      "theoretical_loss": 3.4698706697609456,
+      "tokens_seen": 1924857856
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004208794735997432,
+      "loss": 1.3507,
+      "theoretical_loss": 3.46985098439718,
+      "tokens_seen": 1924988928
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00042083935162895203,
+      "loss": 1.3829,
+      "theoretical_loss": 3.469831300749015,
+      "tokens_seen": 1925120000
+    },
+    {
+      "epoch": 0.17,
+      "objective/train/advantage_avg": 0.493378609418869,
+      "objective/train/docs_used": 701973,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.390122890472412,
+      "objective/train/original_loss": 1.390122890472412,
+      "objective/train/theoretical_loss": 3.469811618816184,
+      "objective/train/tokens_used": 295710176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24403424561023712,
+      "objective/train/value_max": -0.5,
+      "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": 1.4604884386062622,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.469811618816184,
+      "tokens_seen": 1925251072
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004207992296581608,
+      "loss": 1.4556,
+      "theoretical_loss": 3.469811618816184,
+      "tokens_seen": 1925251072
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00042075910768736964,
+      "loss": 1.3051,
+      "theoretical_loss": 3.469791938598421,
+      "tokens_seen": 1925382144
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004207189857165784,
+      "loss": 1.3854,
+      "theoretical_loss": 3.46977226009546,
+      "tokens_seen": 1925513216
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004206788637457872,
+      "loss": 1.4188,
+      "theoretical_loss": 3.469752583307035,
+      "tokens_seen": 1925644288
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.000420638741774996,
+      "loss": 1.4736,
+      "theoretical_loss": 3.469732908232879,
+      "tokens_seen": 1925775360
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00042059861980420476,
+      "loss": 1.4637,
+      "theoretical_loss": 3.469713234872727,
+      "tokens_seen": 1925906432
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004205584978334136,
+      "loss": 1.3448,
+      "theoretical_loss": 3.4696935632263126,
+      "tokens_seen": 1926037504
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00042051837586262237,
+      "loss": 1.3902,
+      "theoretical_loss": 3.46967389329337,
+      "tokens_seen": 1926168576
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004204782538918312,
+      "loss": 1.3212,
+      "theoretical_loss": 3.469654225073633,
+      "tokens_seen": 1926299648
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00042043813192104,
+      "loss": 1.4292,
+      "theoretical_loss": 3.4696345585668364,
+      "tokens_seen": 1926430720
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00042039800995024874,
+      "loss": 1.3173,
+      "theoretical_loss": 3.4696148937727145,
+      "tokens_seen": 1926561792
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00042035788797945754,
+      "loss": 1.4611,
+      "theoretical_loss": 3.469595230691001,
+      "tokens_seen": 1926692864
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00042031776600866635,
+      "loss": 1.3731,
+      "theoretical_loss": 3.4695755693214307,
+      "tokens_seen": 1926823936
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00042027764403787516,
+      "loss": 1.4031,
+      "theoretical_loss": 3.469555909663738,
+      "tokens_seen": 1926955008
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004202375220670839,
+      "loss": 1.4299,
+      "theoretical_loss": 3.469536251717658,
+      "tokens_seen": 1927086080
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00042019740009629277,
+      "loss": 1.405,
+      "theoretical_loss": 3.4695165954829243,
+      "tokens_seen": 1927217152
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004201572781255015,
+      "loss": 1.3547,
+      "theoretical_loss": 3.4694969409592717,
+      "tokens_seen": 1927348224
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004201171561547103,
+      "loss": 1.3651,
+      "theoretical_loss": 3.469477288146436,
+      "tokens_seen": 1927479296
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00042007703418391913,
+      "loss": 1.4535,
+      "theoretical_loss": 3.4694576370441506,
+      "tokens_seen": 1927610368
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004200369122131279,
+      "loss": 1.4342,
+      "theoretical_loss": 3.4694379876521513,
+      "tokens_seen": 1927741440
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041999679024233675,
+      "loss": 1.4675,
+      "theoretical_loss": 3.469418339970172,
+      "tokens_seen": 1927872512
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004199566682715455,
+      "loss": 1.2992,
+      "theoretical_loss": 3.469398693997949,
+      "tokens_seen": 1928003584
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004199165463007543,
+      "loss": 1.4159,
+      "theoretical_loss": 3.469379049735216,
+      "tokens_seen": 1928134656
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004198764243299631,
+      "loss": 1.4027,
+      "theoretical_loss": 3.469359407181708,
+      "tokens_seen": 1928265728
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004198363023591719,
+      "loss": 1.3815,
+      "theoretical_loss": 3.4693397663371615,
+      "tokens_seen": 1928396800
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041979618038838067,
+      "loss": 1.3904,
+      "theoretical_loss": 3.4693201272013106,
+      "tokens_seen": 1928527872
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004197560584175895,
+      "loss": 1.3416,
+      "theoretical_loss": 3.4693004897738904,
+      "tokens_seen": 1928658944
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004197159364467983,
+      "loss": 1.4238,
+      "theoretical_loss": 3.469280854054637,
+      "tokens_seen": 1928790016
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041967581447600703,
+      "loss": 1.3563,
+      "theoretical_loss": 3.4692612200432853,
+      "tokens_seen": 1928921088
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004196356925052159,
+      "loss": 1.4545,
+      "theoretical_loss": 3.46924158773957,
+      "tokens_seen": 1929052160
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041959557053442465,
+      "loss": 1.3334,
+      "theoretical_loss": 3.4692219571432283,
+      "tokens_seen": 1929183232
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041955544856363345,
+      "loss": 1.3386,
+      "theoretical_loss": 3.469202328253994,
+      "tokens_seen": 1929314304
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041951532659284226,
+      "loss": 1.3868,
+      "theoretical_loss": 3.469182701071604,
+      "tokens_seen": 1929445376
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.000419475204622051,
+      "loss": 1.5001,
+      "theoretical_loss": 3.4691630755957927,
+      "tokens_seen": 1929576448
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041943508265125987,
+      "loss": 1.4281,
+      "theoretical_loss": 3.469143451826297,
+      "tokens_seen": 1929707520
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004193949606804686,
+      "loss": 1.4083,
+      "theoretical_loss": 3.469123829762852,
+      "tokens_seen": 1929838592
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041935483870967743,
+      "loss": 1.4504,
+      "theoretical_loss": 3.4691042094051934,
+      "tokens_seen": 1929969664
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041931471673888624,
+      "loss": 1.4134,
+      "theoretical_loss": 3.4690845907530576,
+      "tokens_seen": 1930100736
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041927459476809504,
+      "loss": 1.4323,
+      "theoretical_loss": 3.46906497380618,
+      "tokens_seen": 1930231808
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004192344727973038,
+      "loss": 1.367,
+      "theoretical_loss": 3.469045358564297,
+      "tokens_seen": 1930362880
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004191943508265126,
+      "loss": 1.4139,
+      "theoretical_loss": 3.4690257450271442,
+      "tokens_seen": 1930493952
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004191542288557214,
+      "loss": 1.4492,
+      "theoretical_loss": 3.4690061331944584,
+      "tokens_seen": 1930625024
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041911410688493016,
+      "loss": 1.4073,
+      "theoretical_loss": 3.4689865230659755,
+      "tokens_seen": 1930756096
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.000419073984914139,
+      "loss": 1.4443,
+      "theoretical_loss": 3.4689669146414315,
+      "tokens_seen": 1930887168
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041903386294334777,
+      "loss": 1.4216,
+      "theoretical_loss": 3.468947307920563,
+      "tokens_seen": 1931018240
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004189937409725566,
+      "loss": 1.3718,
+      "theoretical_loss": 3.4689277029031054,
+      "tokens_seen": 1931149312
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004189536190017654,
+      "loss": 1.3309,
+      "theoretical_loss": 3.4689080995887966,
+      "tokens_seen": 1931280384
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041891349703097414,
+      "loss": 1.4011,
+      "theoretical_loss": 3.468888497977372,
+      "tokens_seen": 1931411456
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.000418873375060183,
+      "loss": 1.3932,
+      "theoretical_loss": 3.468868898068569,
+      "tokens_seen": 1931542528
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041883325308939175,
+      "loss": 1.4107,
+      "theoretical_loss": 3.468849299862123,
+      "tokens_seen": 1931673600
+    },
+    {
+      "epoch": 0.17,
+      "objective/train/advantage_avg": 0.49324527382850647,
+      "objective/train/docs_used": 704289,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.4504508972167969,
+      "objective/train/original_loss": 1.4504506587982178,
+      "objective/train/theoretical_loss": 3.4688297033577715,
+      "objective/train/tokens_used": 302263776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2437809705734253,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050563931465149,
+      "objective/train/weighted_lm_loss": 1.5236562490463257,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4688297033577715,
+      "tokens_seen": 1931804672
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041879313111860056,
+      "loss": 1.5063,
+      "theoretical_loss": 3.4688297033577715,
+      "tokens_seen": 1931804672
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041875300914780936,
+      "loss": 1.4441,
+      "theoretical_loss": 3.4688101085552514,
+      "tokens_seen": 1931935744
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041871288717701817,
+      "loss": 1.3574,
+      "theoretical_loss": 3.468790515454299,
+      "tokens_seen": 1932066816
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004186727652062269,
+      "loss": 1.3895,
+      "theoretical_loss": 3.4687709240546507,
+      "tokens_seen": 1932197888
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004186326432354357,
+      "loss": 1.2798,
+      "theoretical_loss": 3.4687513343560443,
+      "tokens_seen": 1932328960
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041859252126464453,
+      "loss": 1.4396,
+      "theoretical_loss": 3.468731746358216,
+      "tokens_seen": 1932460032
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004185523992938533,
+      "loss": 1.375,
+      "theoretical_loss": 3.4687121600609037,
+      "tokens_seen": 1932591104
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041851227732306215,
+      "loss": 1.3269,
+      "theoretical_loss": 3.4686925754638436,
+      "tokens_seen": 1932722176
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004184721553522709,
+      "loss": 1.4177,
+      "theoretical_loss": 3.4686729925667725,
+      "tokens_seen": 1932853248
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004184320333814797,
+      "loss": 1.3876,
+      "theoretical_loss": 3.468653411369429,
+      "tokens_seen": 1932984320
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004183919114106885,
+      "loss": 1.3826,
+      "theoretical_loss": 3.468633831871549,
+      "tokens_seen": 1933115392
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004183517894398973,
+      "loss": 1.417,
+      "theoretical_loss": 3.4686142540728704,
+      "tokens_seen": 1933246464
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041831166746910607,
+      "loss": 1.4329,
+      "theoretical_loss": 3.46859467797313,
+      "tokens_seen": 1933377536
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004182715454983149,
+      "loss": 1.3314,
+      "theoretical_loss": 3.468575103572066,
+      "tokens_seen": 1933508608
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004182314235275237,
+      "loss": 1.3395,
+      "theoretical_loss": 3.4685555308694154,
+      "tokens_seen": 1933639680
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004181913015567325,
+      "loss": 1.4001,
+      "theoretical_loss": 3.468535959864916,
+      "tokens_seen": 1933770752
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004181511795859413,
+      "loss": 1.3977,
+      "theoretical_loss": 3.468516390558305,
+      "tokens_seen": 1933901824
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041811105761515005,
+      "loss": 1.4582,
+      "theoretical_loss": 3.4684968229493203,
+      "tokens_seen": 1934032896
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004180709356443589,
+      "loss": 1.4667,
+      "theoretical_loss": 3.4684772570376996,
+      "tokens_seen": 1934163968
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041803081367356766,
+      "loss": 1.3583,
+      "theoretical_loss": 3.46845769282318,
+      "tokens_seen": 1934295040
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004179906917027764,
+      "loss": 1.3869,
+      "theoretical_loss": 3.4684381303055005,
+      "tokens_seen": 1934426112
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041795056973198527,
+      "loss": 1.3437,
+      "theoretical_loss": 3.468418569484398,
+      "tokens_seen": 1934557184
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.000417910447761194,
+      "loss": 1.4422,
+      "theoretical_loss": 3.4683990103596107,
+      "tokens_seen": 1934688256
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041787032579040283,
+      "loss": 1.3946,
+      "theoretical_loss": 3.4683794529308773,
+      "tokens_seen": 1934819328
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041783020381961164,
+      "loss": 1.3171,
+      "theoretical_loss": 3.4683598971979346,
+      "tokens_seen": 1934950400
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041779008184882044,
+      "loss": 1.4193,
+      "theoretical_loss": 3.4683403431605218,
+      "tokens_seen": 1935081472
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004177499598780292,
+      "loss": 1.3431,
+      "theoretical_loss": 3.4683207908183764,
+      "tokens_seen": 1935212544
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.000417709837907238,
+      "loss": 1.3611,
+      "theoretical_loss": 3.4683012401712365,
+      "tokens_seen": 1935343616
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004176697159364468,
+      "loss": 1.3698,
+      "theoretical_loss": 3.4682816912188406,
+      "tokens_seen": 1935474688
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041762959396565556,
+      "loss": 1.3092,
+      "theoretical_loss": 3.468262143960928,
+      "tokens_seen": 1935605760
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004175894719948644,
+      "loss": 1.4153,
+      "theoretical_loss": 3.468242598397235,
+      "tokens_seen": 1935736832
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041754935002407317,
+      "loss": 1.3293,
+      "theoretical_loss": 3.468223054527502,
+      "tokens_seen": 1935867904
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041750922805328203,
+      "loss": 1.3277,
+      "theoretical_loss": 3.468203512351467,
+      "tokens_seen": 1935998976
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004174691060824908,
+      "loss": 1.3674,
+      "theoretical_loss": 3.468183971868868,
+      "tokens_seen": 1936130048
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041742898411169954,
+      "loss": 1.4189,
+      "theoretical_loss": 3.4681644330794437,
+      "tokens_seen": 1936261120
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004173888621409084,
+      "loss": 1.3617,
+      "theoretical_loss": 3.4681448959829337,
+      "tokens_seen": 1936392192
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041734874017011715,
+      "loss": 1.3357,
+      "theoretical_loss": 3.4681253605790756,
+      "tokens_seen": 1936523264
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041730861819932596,
+      "loss": 1.4204,
+      "theoretical_loss": 3.4681058268676086,
+      "tokens_seen": 1936654336
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041726849622853476,
+      "loss": 1.3866,
+      "theoretical_loss": 3.4680862948482716,
+      "tokens_seen": 1936785408
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041722837425774357,
+      "loss": 1.3861,
+      "theoretical_loss": 3.4680667645208043,
+      "tokens_seen": 1936916480
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004171882522869523,
+      "loss": 1.3865,
+      "theoretical_loss": 3.468047235884945,
+      "tokens_seen": 1937047552
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004171481303161611,
+      "loss": 1.4226,
+      "theoretical_loss": 3.468027708940432,
+      "tokens_seen": 1937178624
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041710800834536993,
+      "loss": 1.3361,
+      "theoretical_loss": 3.468008183687005,
+      "tokens_seen": 1937309696
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004170678863745787,
+      "loss": 1.4089,
+      "theoretical_loss": 3.467988660124403,
+      "tokens_seen": 1937440768
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041702776440378755,
+      "loss": 1.4065,
+      "theoretical_loss": 3.467969138252366,
+      "tokens_seen": 1937571840
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004169876424329963,
+      "loss": 1.403,
+      "theoretical_loss": 3.467949618070633,
+      "tokens_seen": 1937702912
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00041694752046220516,
+      "loss": 1.4263,
+      "theoretical_loss": 3.467930099578943,
+      "tokens_seen": 1937833984
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004169073984914139,
+      "loss": 1.364,
+      "theoretical_loss": 3.4679105827770345,
+      "tokens_seen": 1937965056
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004168672765206227,
+      "loss": 1.4006,
+      "theoretical_loss": 3.467891067664649,
+      "tokens_seen": 1938096128
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004168271545498315,
+      "loss": 1.3779,
+      "theoretical_loss": 3.467871554241524,
+      "tokens_seen": 1938227200
+    },
+    {
+      "epoch": 0.17,
+      "objective/train/advantage_avg": 0.491678923368454,
+      "objective/train/docs_used": 706849,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2877845764160156,
+      "objective/train/original_loss": 1.2877845764160156,
+      "objective/train/theoretical_loss": 3.4678520425074,
+      "objective/train/tokens_used": 308817376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24220307171344757,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503991842269897,
+      "objective/train/weighted_lm_loss": 1.3524404764175415,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4678520425074,
+      "tokens_seen": 1938358272
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004167870325790403,
+      "loss": 1.312,
+      "theoretical_loss": 3.4678520425074,
+      "tokens_seen": 1938358272
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004167469106082491,
+      "loss": 1.4107,
+      "theoretical_loss": 3.4678325324620167,
+      "tokens_seen": 1938489344
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0004167067886374579,
+      "loss": 1.3137,
+      "theoretical_loss": 3.4678130241051135,
+      "tokens_seen": 1938620416
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004166666666666667,
+      "loss": 1.4301,
+      "theoretical_loss": 3.46779351743643,
+      "tokens_seen": 1938751488
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041662654469587545,
+      "loss": 1.3962,
+      "theoretical_loss": 3.4677740124557066,
+      "tokens_seen": 1938882560
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004165864227250843,
+      "loss": 1.3495,
+      "theoretical_loss": 3.467754509162683,
+      "tokens_seen": 1939013632
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041654630075429306,
+      "loss": 1.4134,
+      "theoretical_loss": 3.467735007557098,
+      "tokens_seen": 1939144704
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004165061787835018,
+      "loss": 1.3493,
+      "theoretical_loss": 3.467715507638693,
+      "tokens_seen": 1939275776
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041646605681271067,
+      "loss": 1.2851,
+      "theoretical_loss": 3.4676960094072067,
+      "tokens_seen": 1939406848
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004164259348419194,
+      "loss": 1.3812,
+      "theoretical_loss": 3.46767651286238,
+      "tokens_seen": 1939537920
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041638581287112823,
+      "loss": 1.4248,
+      "theoretical_loss": 3.467657018003954,
+      "tokens_seen": 1939668992
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041634569090033704,
+      "loss": 1.393,
+      "theoretical_loss": 3.4676375248316664,
+      "tokens_seen": 1939800064
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041630556892954584,
+      "loss": 1.4027,
+      "theoretical_loss": 3.4676180333452598,
+      "tokens_seen": 1939931136
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041626544695875465,
+      "loss": 1.3713,
+      "theoretical_loss": 3.467598543544473,
+      "tokens_seen": 1940062208
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004162253249879634,
+      "loss": 1.3786,
+      "theoretical_loss": 3.467579055429047,
+      "tokens_seen": 1940193280
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004161852030171722,
+      "loss": 1.3993,
+      "theoretical_loss": 3.4675595689987224,
+      "tokens_seen": 1940324352
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.000416145081046381,
+      "loss": 1.4287,
+      "theoretical_loss": 3.4675400842532387,
+      "tokens_seen": 1940455424
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004161049590755898,
+      "loss": 1.4604,
+      "theoretical_loss": 3.4675206011923376,
+      "tokens_seen": 1940586496
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041606483710479857,
+      "loss": 1.3848,
+      "theoretical_loss": 3.4675011198157595,
+      "tokens_seen": 1940717568
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041602471513400743,
+      "loss": 1.4354,
+      "theoretical_loss": 3.4674816401232444,
+      "tokens_seen": 1940848640
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004159845931632162,
+      "loss": 1.4068,
+      "theoretical_loss": 3.4674621621145336,
+      "tokens_seen": 1940979712
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041594447119242494,
+      "loss": 1.3412,
+      "theoretical_loss": 3.467442685789367,
+      "tokens_seen": 1941110784
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004159043492216338,
+      "loss": 1.3645,
+      "theoretical_loss": 3.467423211147487,
+      "tokens_seen": 1941241856
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041586422725084255,
+      "loss": 1.401,
+      "theoretical_loss": 3.467403738188633,
+      "tokens_seen": 1941372928
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041582410528005135,
+      "loss": 1.3324,
+      "theoretical_loss": 3.4673842669125463,
+      "tokens_seen": 1941504000
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041578398330926016,
+      "loss": 1.3613,
+      "theoretical_loss": 3.4673647973189685,
+      "tokens_seen": 1941635072
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041574386133846897,
+      "loss": 1.3655,
+      "theoretical_loss": 3.4673453294076397,
+      "tokens_seen": 1941766144
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004157037393676778,
+      "loss": 1.3492,
+      "theoretical_loss": 3.467325863178302,
+      "tokens_seen": 1941897216
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004156636173968865,
+      "loss": 1.3377,
+      "theoretical_loss": 3.4673063986306962,
+      "tokens_seen": 1942028288
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041562349542609533,
+      "loss": 1.3696,
+      "theoretical_loss": 3.4672869357645633,
+      "tokens_seen": 1942159360
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041558337345530414,
+      "loss": 1.3262,
+      "theoretical_loss": 3.4672674745796446,
+      "tokens_seen": 1942290432
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041554325148451294,
+      "loss": 1.2663,
+      "theoretical_loss": 3.467248015075681,
+      "tokens_seen": 1942421504
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004155031295137217,
+      "loss": 1.3631,
+      "theoretical_loss": 3.4672285572524153,
+      "tokens_seen": 1942552576
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041546300754293056,
+      "loss": 1.4497,
+      "theoretical_loss": 3.467209101109587,
+      "tokens_seen": 1942683648
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004154228855721393,
+      "loss": 1.4101,
+      "theoretical_loss": 3.46718964664694,
+      "tokens_seen": 1942814720
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041538276360134806,
+      "loss": 1.3893,
+      "theoretical_loss": 3.4671701938642134,
+      "tokens_seen": 1942945792
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004153426416305569,
+      "loss": 1.3247,
+      "theoretical_loss": 3.4671507427611505,
+      "tokens_seen": 1943076864
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004153025196597657,
+      "loss": 1.4034,
+      "theoretical_loss": 3.4671312933374923,
+      "tokens_seen": 1943207936
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004152623976889745,
+      "loss": 1.3501,
+      "theoretical_loss": 3.4671118455929806,
+      "tokens_seen": 1943339008
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004152222757181833,
+      "loss": 1.3803,
+      "theoretical_loss": 3.4670923995273575,
+      "tokens_seen": 1943470080
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004151821537473921,
+      "loss": 1.3506,
+      "theoretical_loss": 3.4670729551403645,
+      "tokens_seen": 1943601152
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041514203177660085,
+      "loss": 1.4444,
+      "theoretical_loss": 3.4670535124317436,
+      "tokens_seen": 1943732224
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004151019098058097,
+      "loss": 1.343,
+      "theoretical_loss": 3.467034071401237,
+      "tokens_seen": 1943863296
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041506178783501846,
+      "loss": 1.4313,
+      "theoretical_loss": 3.4670146320485866,
+      "tokens_seen": 1943994368
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041502166586422726,
+      "loss": 1.4152,
+      "theoretical_loss": 3.466995194373534,
+      "tokens_seen": 1944125440
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041498154389343607,
+      "loss": 1.3492,
+      "theoretical_loss": 3.4669757583758223,
+      "tokens_seen": 1944256512
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004149414219226448,
+      "loss": 1.3312,
+      "theoretical_loss": 3.466956324055192,
+      "tokens_seen": 1944387584
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004149012999518537,
+      "loss": 1.3649,
+      "theoretical_loss": 3.466936891411388,
+      "tokens_seen": 1944518656
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041486117798106244,
+      "loss": 1.3794,
+      "theoretical_loss": 3.46691746044415,
+      "tokens_seen": 1944649728
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041482105601027124,
+      "loss": 1.3919,
+      "theoretical_loss": 3.466898031153222,
+      "tokens_seen": 1944780800
+    },
+    {
+      "epoch": 0.18,
+      "objective/train/advantage_avg": 0.4927396774291992,
+      "objective/train/docs_used": 709206,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.4127498865127563,
+      "objective/train/original_loss": 1.412750005722046,
+      "objective/train/theoretical_loss": 3.466878603538346,
+      "objective/train/tokens_used": 315370976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2433270364999771,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505108833312988,
+      "objective/train/weighted_lm_loss": 1.4839540719985962,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9048373699188232,
+      "theoretical_loss": 3.466878603538346,
+      "tokens_seen": 1944911872
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041478093403948005,
+      "loss": 1.4648,
+      "theoretical_loss": 3.466878603538346,
+      "tokens_seen": 1944911872
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004147408120686888,
+      "loss": 1.4949,
+      "theoretical_loss": 3.4668591775992637,
+      "tokens_seen": 1945042944
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004147006900978976,
+      "loss": 1.4098,
+      "theoretical_loss": 3.4668397533357185,
+      "tokens_seen": 1945174016
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004146605681271064,
+      "loss": 1.4517,
+      "theoretical_loss": 3.4668203307474528,
+      "tokens_seen": 1945305088
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004146204461563152,
+      "loss": 1.3819,
+      "theoretical_loss": 3.4668009098342094,
+      "tokens_seen": 1945436160
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041458032418552397,
+      "loss": 1.3421,
+      "theoretical_loss": 3.466781490595731,
+      "tokens_seen": 1945567232
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041454020221473283,
+      "loss": 1.4468,
+      "theoretical_loss": 3.4667620730317594,
+      "tokens_seen": 1945698304
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004145000802439416,
+      "loss": 1.3056,
+      "theoretical_loss": 3.4667426571420394,
+      "tokens_seen": 1945829376
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041445995827315034,
+      "loss": 1.404,
+      "theoretical_loss": 3.466723242926312,
+      "tokens_seen": 1945960448
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004144198363023592,
+      "loss": 1.3421,
+      "theoretical_loss": 3.466703830384321,
+      "tokens_seen": 1946091520
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041437971433156795,
+      "loss": 1.3975,
+      "theoretical_loss": 3.466684419515809,
+      "tokens_seen": 1946222592
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004143395923607768,
+      "loss": 1.4116,
+      "theoretical_loss": 3.4666650103205194,
+      "tokens_seen": 1946353664
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041429947038998556,
+      "loss": 1.4359,
+      "theoretical_loss": 3.466645602798195,
+      "tokens_seen": 1946484736
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041425934841919437,
+      "loss": 1.3927,
+      "theoretical_loss": 3.4666261969485794,
+      "tokens_seen": 1946615808
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004142192264484032,
+      "loss": 1.3996,
+      "theoretical_loss": 3.4666067927714157,
+      "tokens_seen": 1946746880
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004141791044776119,
+      "loss": 1.4289,
+      "theoretical_loss": 3.466587390266447,
+      "tokens_seen": 1946877952
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041413898250682073,
+      "loss": 1.3924,
+      "theoretical_loss": 3.4665679894334165,
+      "tokens_seen": 1947009024
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041409886053602954,
+      "loss": 1.356,
+      "theoretical_loss": 3.466548590272067,
+      "tokens_seen": 1947140096
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041405873856523834,
+      "loss": 1.3912,
+      "theoretical_loss": 3.4665291927821436,
+      "tokens_seen": 1947271168
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004140186165944471,
+      "loss": 1.3929,
+      "theoretical_loss": 3.466509796963389,
+      "tokens_seen": 1947402240
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041397849462365596,
+      "loss": 1.4836,
+      "theoretical_loss": 3.466490402815546,
+      "tokens_seen": 1947533312
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004139383726528647,
+      "loss": 1.411,
+      "theoretical_loss": 3.4664710103383594,
+      "tokens_seen": 1947664384
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041389825068207346,
+      "loss": 1.4063,
+      "theoretical_loss": 3.466451619531572,
+      "tokens_seen": 1947795456
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004138581287112823,
+      "loss": 1.4742,
+      "theoretical_loss": 3.466432230394928,
+      "tokens_seen": 1947926528
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004138180067404911,
+      "loss": 1.4426,
+      "theoretical_loss": 3.466412842928171,
+      "tokens_seen": 1948057600
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041377788476969993,
+      "loss": 1.3458,
+      "theoretical_loss": 3.466393457131045,
+      "tokens_seen": 1948188672
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004137377627989087,
+      "loss": 1.3925,
+      "theoretical_loss": 3.4663740730032933,
+      "tokens_seen": 1948319744
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004136976408281175,
+      "loss": 1.4027,
+      "theoretical_loss": 3.4663546905446605,
+      "tokens_seen": 1948450816
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004136575188573263,
+      "loss": 1.4396,
+      "theoretical_loss": 3.466335309754891,
+      "tokens_seen": 1948581888
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004136173968865351,
+      "loss": 1.4736,
+      "theoretical_loss": 3.4663159306337277,
+      "tokens_seen": 1948712960
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041357727491574386,
+      "loss": 1.3762,
+      "theoretical_loss": 3.4662965531809156,
+      "tokens_seen": 1948844032
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041353715294495266,
+      "loss": 1.3189,
+      "theoretical_loss": 3.4662771773961985,
+      "tokens_seen": 1948975104
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041349703097416147,
+      "loss": 1.4248,
+      "theoretical_loss": 3.466257803279321,
+      "tokens_seen": 1949106176
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004134569090033702,
+      "loss": 1.3602,
+      "theoretical_loss": 3.4662384308300265,
+      "tokens_seen": 1949237248
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004134167870325791,
+      "loss": 1.4117,
+      "theoretical_loss": 3.4662190600480605,
+      "tokens_seen": 1949368320
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041337666506178783,
+      "loss": 1.4072,
+      "theoretical_loss": 3.4661996909331667,
+      "tokens_seen": 1949499392
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041333654309099664,
+      "loss": 1.3638,
+      "theoretical_loss": 3.46618032348509,
+      "tokens_seen": 1949630464
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041329642112020545,
+      "loss": 1.3552,
+      "theoretical_loss": 3.4661609577035746,
+      "tokens_seen": 1949761536
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004132562991494142,
+      "loss": 1.3216,
+      "theoretical_loss": 3.4661415935883646,
+      "tokens_seen": 1949892608
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.000413216177178623,
+      "loss": 1.3456,
+      "theoretical_loss": 3.4661222311392055,
+      "tokens_seen": 1950023680
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004131760552078318,
+      "loss": 1.3881,
+      "theoretical_loss": 3.466102870355842,
+      "tokens_seen": 1950154752
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004131359332370406,
+      "loss": 1.4723,
+      "theoretical_loss": 3.466083511238018,
+      "tokens_seen": 1950285824
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004130958112662494,
+      "loss": 1.4312,
+      "theoretical_loss": 3.466064153785479,
+      "tokens_seen": 1950416896
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041305568929545823,
+      "loss": 1.3683,
+      "theoretical_loss": 3.4660447979979696,
+      "tokens_seen": 1950547968
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.000413015567324667,
+      "loss": 1.4233,
+      "theoretical_loss": 3.4660254438752345,
+      "tokens_seen": 1950679040
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004129754453538758,
+      "loss": 1.4722,
+      "theoretical_loss": 3.4660060914170194,
+      "tokens_seen": 1950810112
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004129353233830846,
+      "loss": 1.3615,
+      "theoretical_loss": 3.4659867406230687,
+      "tokens_seen": 1950941184
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041289520141229335,
+      "loss": 1.4508,
+      "theoretical_loss": 3.4659673914931277,
+      "tokens_seen": 1951072256
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004128550794415022,
+      "loss": 1.3186,
+      "theoretical_loss": 3.4659480440269412,
+      "tokens_seen": 1951203328
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041281495747071096,
+      "loss": 1.4441,
+      "theoretical_loss": 3.4659286982242548,
+      "tokens_seen": 1951334400
+    },
+    {
+      "epoch": 0.18,
+      "objective/train/advantage_avg": 0.49048033356666565,
+      "objective/train/docs_used": 711575,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3126404285430908,
+      "objective/train/original_loss": 1.3126404285430908,
+      "objective/train/theoretical_loss": 3.465909354084814,
+      "objective/train/tokens_used": 321924576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2413894087076187,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050275206565857,
+      "objective/train/weighted_lm_loss": 1.3783910274505615,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.465909354084814,
+      "tokens_seen": 1951465472
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041277483549991977,
+      "loss": 1.3892,
+      "theoretical_loss": 3.465909354084814,
+      "tokens_seen": 1951465472
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004127347135291286,
+      "loss": 1.3927,
+      "theoretical_loss": 3.465890011608363,
+      "tokens_seen": 1951596544
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004126945915583373,
+      "loss": 1.4082,
+      "theoretical_loss": 3.465870670794648,
+      "tokens_seen": 1951727616
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041265446958754613,
+      "loss": 1.3522,
+      "theoretical_loss": 3.4658513316434147,
+      "tokens_seen": 1951858688
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041261434761675494,
+      "loss": 1.3247,
+      "theoretical_loss": 3.465831994154408,
+      "tokens_seen": 1951989760
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041257422564596374,
+      "loss": 1.4672,
+      "theoretical_loss": 3.465812658327373,
+      "tokens_seen": 1952120832
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004125341036751725,
+      "loss": 1.4226,
+      "theoretical_loss": 3.465793324162057,
+      "tokens_seen": 1952251904
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041249398170438136,
+      "loss": 1.4221,
+      "theoretical_loss": 3.4657739916582035,
+      "tokens_seen": 1952382976
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004124538597335901,
+      "loss": 1.4196,
+      "theoretical_loss": 3.4657546608155596,
+      "tokens_seen": 1952514048
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004124137377627989,
+      "loss": 1.3835,
+      "theoretical_loss": 3.465735331633871,
+      "tokens_seen": 1952645120
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004123736157920077,
+      "loss": 1.4176,
+      "theoretical_loss": 3.465716004112883,
+      "tokens_seen": 1952776192
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004123334938212165,
+      "loss": 1.4325,
+      "theoretical_loss": 3.465696678252341,
+      "tokens_seen": 1952907264
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041229337185042533,
+      "loss": 1.392,
+      "theoretical_loss": 3.465677354051992,
+      "tokens_seen": 1953038336
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004122532498796341,
+      "loss": 1.4183,
+      "theoretical_loss": 3.4656580315115817,
+      "tokens_seen": 1953169408
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004122131279088429,
+      "loss": 1.3936,
+      "theoretical_loss": 3.4656387106308557,
+      "tokens_seen": 1953300480
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004121730059380517,
+      "loss": 1.4345,
+      "theoretical_loss": 3.4656193914095605,
+      "tokens_seen": 1953431552
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004121328839672605,
+      "loss": 1.4327,
+      "theoretical_loss": 3.465600073847442,
+      "tokens_seen": 1953562624
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041209276199646926,
+      "loss": 1.417,
+      "theoretical_loss": 3.465580757944247,
+      "tokens_seen": 1953693696
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041205264002567806,
+      "loss": 1.3663,
+      "theoretical_loss": 3.4655614436997206,
+      "tokens_seen": 1953824768
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041201251805488687,
+      "loss": 1.3429,
+      "theoretical_loss": 3.46554213111361,
+      "tokens_seen": 1953955840
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004119723960840956,
+      "loss": 1.4372,
+      "theoretical_loss": 3.4655228201856616,
+      "tokens_seen": 1954086912
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004119322741133045,
+      "loss": 1.3769,
+      "theoretical_loss": 3.465503510915621,
+      "tokens_seen": 1954217984
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041189215214251323,
+      "loss": 1.4131,
+      "theoretical_loss": 3.4654842033032356,
+      "tokens_seen": 1954349056
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004118520301717221,
+      "loss": 1.4387,
+      "theoretical_loss": 3.4654648973482516,
+      "tokens_seen": 1954480128
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041181190820093085,
+      "loss": 1.4199,
+      "theoretical_loss": 3.465445593050415,
+      "tokens_seen": 1954611200
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004117717862301396,
+      "loss": 1.3541,
+      "theoretical_loss": 3.4654262904094737,
+      "tokens_seen": 1954742272
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00041173166425934846,
+      "loss": 1.5357,
+      "theoretical_loss": 3.4654069894251736,
+      "tokens_seen": 1954873344
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0004116915422885572,
+      "loss": 1.4016,
+      "theoretical_loss": 3.4653876900972613,
+      "tokens_seen": 1955004416
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.000411651420317766,
+      "loss": 1.4824,
+      "theoretical_loss": 3.465368392425484,
+      "tokens_seen": 1955135488
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004116112983469748,
+      "loss": 1.3995,
+      "theoretical_loss": 3.4653490964095885,
+      "tokens_seen": 1955266560
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00041157117637618363,
+      "loss": 1.3759,
+      "theoretical_loss": 3.4653298020493217,
+      "tokens_seen": 1955397632
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004115310544053924,
+      "loss": 1.3503,
+      "theoretical_loss": 3.4653105093444303,
+      "tokens_seen": 1955528704
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004114909324346012,
+      "loss": 1.4092,
+      "theoretical_loss": 3.465291218294662,
+      "tokens_seen": 1955659776
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00041145081046381,
+      "loss": 1.4058,
+      "theoretical_loss": 3.465271928899763,
+      "tokens_seen": 1955790848
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00041141068849301875,
+      "loss": 1.3829,
+      "theoretical_loss": 3.4652526411594815,
+      "tokens_seen": 1955921920
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004113705665222276,
+      "loss": 1.356,
+      "theoretical_loss": 3.465233355073564,
+      "tokens_seen": 1956052992
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00041133044455143636,
+      "loss": 1.3545,
+      "theoretical_loss": 3.4652140706417573,
+      "tokens_seen": 1956184064
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00041129032258064517,
+      "loss": 1.3444,
+      "theoretical_loss": 3.46519478786381,
+      "tokens_seen": 1956315136
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00041125020060985397,
+      "loss": 1.4034,
+      "theoretical_loss": 3.4651755067394685,
+      "tokens_seen": 1956446208
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004112100786390627,
+      "loss": 1.3507,
+      "theoretical_loss": 3.4651562272684804,
+      "tokens_seen": 1956577280
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004111699566682716,
+      "loss": 1.4038,
+      "theoretical_loss": 3.4651369494505935,
+      "tokens_seen": 1956708352
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00041112983469748034,
+      "loss": 1.3812,
+      "theoretical_loss": 3.4651176732855546,
+      "tokens_seen": 1956839424
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00041108971272668914,
+      "loss": 1.44,
+      "theoretical_loss": 3.4650983987731125,
+      "tokens_seen": 1956970496
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00041104959075589795,
+      "loss": 1.403,
+      "theoretical_loss": 3.465079125913014,
+      "tokens_seen": 1957101568
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00041100946878510676,
+      "loss": 1.4229,
+      "theoretical_loss": 3.4650598547050064,
+      "tokens_seen": 1957232640
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004109693468143155,
+      "loss": 1.441,
+      "theoretical_loss": 3.465040585148839,
+      "tokens_seen": 1957363712
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004109292248435243,
+      "loss": 1.4511,
+      "theoretical_loss": 3.4650213172442577,
+      "tokens_seen": 1957494784
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004108891028727331,
+      "loss": 1.3787,
+      "theoretical_loss": 3.4650020509910116,
+      "tokens_seen": 1957625856
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004108489809019419,
+      "loss": 1.3685,
+      "theoretical_loss": 3.4649827863888483,
+      "tokens_seen": 1957756928
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00041080885893115073,
+      "loss": 1.4116,
+      "theoretical_loss": 3.4649635234375165,
+      "tokens_seen": 1957888000
+    },
+    {
+      "epoch": 0.19,
+      "objective/train/advantage_avg": 0.49267223477363586,
+      "objective/train/docs_used": 713999,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3758057355880737,
+      "objective/train/original_loss": 1.3758057355880737,
+      "objective/train/theoretical_loss": 3.4649442621367625,
+      "objective/train/tokens_used": 328478176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24330894649028778,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505040884017944,
+      "objective/train/weighted_lm_loss": 1.4452353715896606,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4649442621367625,
+      "tokens_seen": 1958019072
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004107687369603595,
+      "loss": 1.4272,
+      "theoretical_loss": 3.4649442621367625,
+      "tokens_seen": 1958019072
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004107286149895683,
+      "loss": 1.3807,
+      "theoretical_loss": 3.4649250024863356,
+      "tokens_seen": 1958150144
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004106884930187771,
+      "loss": 1.4441,
+      "theoretical_loss": 3.4649057444859843,
+      "tokens_seen": 1958281216
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00041064837104798585,
+      "loss": 1.453,
+      "theoretical_loss": 3.4648864881354564,
+      "tokens_seen": 1958412288
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004106082490771947,
+      "loss": 1.391,
+      "theoretical_loss": 3.464867233434499,
+      "tokens_seen": 1958543360
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00041056812710640346,
+      "loss": 1.3476,
+      "theoretical_loss": 3.4648479803828627,
+      "tokens_seen": 1958674432
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00041052800513561227,
+      "loss": 1.4654,
+      "theoretical_loss": 3.4648287289802937,
+      "tokens_seen": 1958805504
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004104878831648211,
+      "loss": 1.5173,
+      "theoretical_loss": 3.464809479226542,
+      "tokens_seen": 1958936576
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004104477611940299,
+      "loss": 1.3946,
+      "theoretical_loss": 3.4647902311213548,
+      "tokens_seen": 1959067648
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00041040763922323863,
+      "loss": 1.4361,
+      "theoretical_loss": 3.464770984664481,
+      "tokens_seen": 1959198720
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004103675172524475,
+      "loss": 1.4581,
+      "theoretical_loss": 3.4647517398556698,
+      "tokens_seen": 1959329792
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00041032739528165625,
+      "loss": 1.4334,
+      "theoretical_loss": 3.46473249669467,
+      "tokens_seen": 1959460864
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.000410287273310865,
+      "loss": 1.3679,
+      "theoretical_loss": 3.4647132551812287,
+      "tokens_seen": 1959591936
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00041024715134007386,
+      "loss": 1.3688,
+      "theoretical_loss": 3.4646940153150965,
+      "tokens_seen": 1959723008
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004102070293692826,
+      "loss": 1.4718,
+      "theoretical_loss": 3.4646747770960213,
+      "tokens_seen": 1959854080
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004101669073984914,
+      "loss": 1.3819,
+      "theoretical_loss": 3.4646555405237516,
+      "tokens_seen": 1959985152
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004101267854277002,
+      "loss": 1.3357,
+      "theoretical_loss": 3.464636305598037,
+      "tokens_seen": 1960116224
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00041008666345690903,
+      "loss": 1.3282,
+      "theoretical_loss": 3.464617072318627,
+      "tokens_seen": 1960247296
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004100465414861178,
+      "loss": 1.4071,
+      "theoretical_loss": 3.4645978406852693,
+      "tokens_seen": 1960378368
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004100064195153266,
+      "loss": 1.4146,
+      "theoretical_loss": 3.4645786106977137,
+      "tokens_seen": 1960509440
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004099662975445354,
+      "loss": 1.4067,
+      "theoretical_loss": 3.4645593823557093,
+      "tokens_seen": 1960640512
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004099261755737442,
+      "loss": 1.3901,
+      "theoretical_loss": 3.464540155659005,
+      "tokens_seen": 1960771584
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.000409886053602953,
+      "loss": 1.3414,
+      "theoretical_loss": 3.4645209306073506,
+      "tokens_seen": 1960902656
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040984593163216176,
+      "loss": 1.4561,
+      "theoretical_loss": 3.464501707200495,
+      "tokens_seen": 1961033728
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004098058096613706,
+      "loss": 1.4105,
+      "theoretical_loss": 3.464482485438187,
+      "tokens_seen": 1961164800
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040976568769057937,
+      "loss": 1.3652,
+      "theoretical_loss": 3.464463265320177,
+      "tokens_seen": 1961295872
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004097255657197881,
+      "loss": 1.4078,
+      "theoretical_loss": 3.4644440468462143,
+      "tokens_seen": 1961426944
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.000409685443748997,
+      "loss": 1.3631,
+      "theoretical_loss": 3.464424830016048,
+      "tokens_seen": 1961558016
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040964532177820574,
+      "loss": 1.4006,
+      "theoretical_loss": 3.4644056148294284,
+      "tokens_seen": 1961689088
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040960519980741454,
+      "loss": 1.3674,
+      "theoretical_loss": 3.464386401286104,
+      "tokens_seen": 1961820160
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040956507783662335,
+      "loss": 1.4403,
+      "theoretical_loss": 3.4643671893858254,
+      "tokens_seen": 1961951232
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040952495586583216,
+      "loss": 1.4313,
+      "theoretical_loss": 3.464347979128342,
+      "tokens_seen": 1962082304
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004094848338950409,
+      "loss": 1.4354,
+      "theoretical_loss": 3.464328770513404,
+      "tokens_seen": 1962213376
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004094447119242497,
+      "loss": 1.3837,
+      "theoretical_loss": 3.464309563540761,
+      "tokens_seen": 1962344448
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004094045899534585,
+      "loss": 1.3579,
+      "theoretical_loss": 3.464290358210162,
+      "tokens_seen": 1962475520
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040936446798266727,
+      "loss": 1.3832,
+      "theoretical_loss": 3.4642711545213585,
+      "tokens_seen": 1962606592
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040932434601187613,
+      "loss": 1.3184,
+      "theoretical_loss": 3.4642519524740996,
+      "tokens_seen": 1962737664
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004092842240410849,
+      "loss": 1.38,
+      "theoretical_loss": 3.4642327520681353,
+      "tokens_seen": 1962868736
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040924410207029375,
+      "loss": 1.3752,
+      "theoretical_loss": 3.464213553303216,
+      "tokens_seen": 1962999808
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004092039800995025,
+      "loss": 1.4363,
+      "theoretical_loss": 3.4641943561790924,
+      "tokens_seen": 1963130880
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040916385812871125,
+      "loss": 1.4114,
+      "theoretical_loss": 3.464175160695514,
+      "tokens_seen": 1963261952
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004091237361579201,
+      "loss": 1.4301,
+      "theoretical_loss": 3.464155966852231,
+      "tokens_seen": 1963393024
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040908361418712886,
+      "loss": 1.4095,
+      "theoretical_loss": 3.4641367746489946,
+      "tokens_seen": 1963524096
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040904349221633767,
+      "loss": 1.4287,
+      "theoretical_loss": 3.464117584085554,
+      "tokens_seen": 1963655168
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004090033702455465,
+      "loss": 1.3209,
+      "theoretical_loss": 3.4640983951616606,
+      "tokens_seen": 1963786240
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004089632482747553,
+      "loss": 1.3533,
+      "theoretical_loss": 3.464079207877065,
+      "tokens_seen": 1963917312
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040892312630396403,
+      "loss": 1.4224,
+      "theoretical_loss": 3.464060022231517,
+      "tokens_seen": 1964048384
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004088830043331729,
+      "loss": 1.4148,
+      "theoretical_loss": 3.4640408382247676,
+      "tokens_seen": 1964179456
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040884288236238165,
+      "loss": 1.4012,
+      "theoretical_loss": 3.464021655856567,
+      "tokens_seen": 1964310528
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004088027603915904,
+      "loss": 1.4123,
+      "theoretical_loss": 3.464002475126667,
+      "tokens_seen": 1964441600
+    },
+    {
+      "epoch": 0.19,
+      "objective/train/advantage_avg": 0.49150922894477844,
+      "objective/train/docs_used": 716326,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3085780143737793,
+      "objective/train/original_loss": 1.3085778951644897,
+      "objective/train/theoretical_loss": 3.463983296034818,
+      "objective/train/tokens_used": 335031776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24220247566699982,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503822565078735,
+      "objective/train/weighted_lm_loss": 1.3745098114013672,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.463983296034818,
+      "tokens_seen": 1964572672
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040876263842079926,
+      "loss": 1.4086,
+      "theoretical_loss": 3.463983296034818,
+      "tokens_seen": 1964572672
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.000408722516450008,
+      "loss": 1.3957,
+      "theoretical_loss": 3.46396411858077,
+      "tokens_seen": 1964703744
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040868239447921687,
+      "loss": 1.3723,
+      "theoretical_loss": 3.463944942764275,
+      "tokens_seen": 1964834816
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004086422725084256,
+      "loss": 1.3871,
+      "theoretical_loss": 3.463925768585083,
+      "tokens_seen": 1964965888
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040860215053763443,
+      "loss": 1.4799,
+      "theoretical_loss": 3.4639065960429454,
+      "tokens_seen": 1965096960
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040856202856684324,
+      "loss": 1.4055,
+      "theoretical_loss": 3.4638874251376137,
+      "tokens_seen": 1965228032
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.000408521906596052,
+      "loss": 1.3816,
+      "theoretical_loss": 3.4638682558688383,
+      "tokens_seen": 1965359104
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004084817846252608,
+      "loss": 1.4347,
+      "theoretical_loss": 3.463849088236371,
+      "tokens_seen": 1965490176
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004084416626544696,
+      "loss": 1.304,
+      "theoretical_loss": 3.4638299222399622,
+      "tokens_seen": 1965621248
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004084015406836784,
+      "loss": 1.3421,
+      "theoretical_loss": 3.463810757879364,
+      "tokens_seen": 1965752320
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040836141871288716,
+      "loss": 1.4068,
+      "theoretical_loss": 3.463791595154327,
+      "tokens_seen": 1965883392
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.000408321296742096,
+      "loss": 1.4032,
+      "theoretical_loss": 3.4637724340646034,
+      "tokens_seen": 1966014464
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040828117477130477,
+      "loss": 1.3231,
+      "theoretical_loss": 3.463753274609944,
+      "tokens_seen": 1966145536
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004082410528005135,
+      "loss": 1.4284,
+      "theoretical_loss": 3.463734116790101,
+      "tokens_seen": 1966276608
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004082009308297224,
+      "loss": 1.412,
+      "theoretical_loss": 3.4637149606048245,
+      "tokens_seen": 1966407680
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040816080885893114,
+      "loss": 1.4069,
+      "theoretical_loss": 3.4636958060538676,
+      "tokens_seen": 1966538752
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040812068688813994,
+      "loss": 1.3414,
+      "theoretical_loss": 3.4636766531369814,
+      "tokens_seen": 1966669824
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040808056491734875,
+      "loss": 1.3656,
+      "theoretical_loss": 3.463657501853918,
+      "tokens_seen": 1966800896
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040804044294655756,
+      "loss": 1.4139,
+      "theoretical_loss": 3.463638352204428,
+      "tokens_seen": 1966931968
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040800032097576636,
+      "loss": 1.4176,
+      "theoretical_loss": 3.463619204188264,
+      "tokens_seen": 1967063040
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004079601990049751,
+      "loss": 1.4153,
+      "theoretical_loss": 3.463600057805178,
+      "tokens_seen": 1967194112
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004079200770341839,
+      "loss": 1.379,
+      "theoretical_loss": 3.4635809130549218,
+      "tokens_seen": 1967325184
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004078799550633927,
+      "loss": 1.3339,
+      "theoretical_loss": 3.463561769937247,
+      "tokens_seen": 1967456256
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040783983309260153,
+      "loss": 1.5114,
+      "theoretical_loss": 3.463542628451907,
+      "tokens_seen": 1967587328
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004077997111218103,
+      "loss": 1.4,
+      "theoretical_loss": 3.4635234885986517,
+      "tokens_seen": 1967718400
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040775958915101915,
+      "loss": 1.4345,
+      "theoretical_loss": 3.4635043503772347,
+      "tokens_seen": 1967849472
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004077194671802279,
+      "loss": 1.3503,
+      "theoretical_loss": 3.463485213787408,
+      "tokens_seen": 1967980544
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040767934520943665,
+      "loss": 1.4607,
+      "theoretical_loss": 3.4634660788289238,
+      "tokens_seen": 1968111616
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004076392232386455,
+      "loss": 1.3669,
+      "theoretical_loss": 3.463446945501534,
+      "tokens_seen": 1968242688
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040759910126785426,
+      "loss": 1.3867,
+      "theoretical_loss": 3.4634278138049917,
+      "tokens_seen": 1968373760
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040755897929706307,
+      "loss": 1.403,
+      "theoretical_loss": 3.4634086837390488,
+      "tokens_seen": 1968504832
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004075188573262719,
+      "loss": 1.4065,
+      "theoretical_loss": 3.4633895553034577,
+      "tokens_seen": 1968635904
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004074787353554807,
+      "loss": 1.3585,
+      "theoretical_loss": 3.463370428497971,
+      "tokens_seen": 1968766976
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004074386133846895,
+      "loss": 1.3608,
+      "theoretical_loss": 3.4633513033223418,
+      "tokens_seen": 1968898048
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004073984914138983,
+      "loss": 1.3373,
+      "theoretical_loss": 3.463332179776322,
+      "tokens_seen": 1969029120
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040735836944310705,
+      "loss": 1.3869,
+      "theoretical_loss": 3.463313057859665,
+      "tokens_seen": 1969160192
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040731824747231585,
+      "loss": 1.4328,
+      "theoretical_loss": 3.4632939375721223,
+      "tokens_seen": 1969291264
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040727812550152466,
+      "loss": 1.4017,
+      "theoretical_loss": 3.463274818913448,
+      "tokens_seen": 1969422336
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004072380035307334,
+      "loss": 1.4477,
+      "theoretical_loss": 3.4632557018833943,
+      "tokens_seen": 1969553408
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040719788155994227,
+      "loss": 1.3887,
+      "theoretical_loss": 3.463236586481714,
+      "tokens_seen": 1969684480
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.000407157759589151,
+      "loss": 1.4096,
+      "theoretical_loss": 3.463217472708161,
+      "tokens_seen": 1969815552
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040711763761835983,
+      "loss": 1.338,
+      "theoretical_loss": 3.4631983605624868,
+      "tokens_seen": 1969946624
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040707751564756864,
+      "loss": 1.4284,
+      "theoretical_loss": 3.4631792500444454,
+      "tokens_seen": 1970077696
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004070373936767774,
+      "loss": 1.3543,
+      "theoretical_loss": 3.4631601411537902,
+      "tokens_seen": 1970208768
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004069972717059862,
+      "loss": 1.3734,
+      "theoretical_loss": 3.463141033890273,
+      "tokens_seen": 1970339840
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.000406957149735195,
+      "loss": 1.3059,
+      "theoretical_loss": 3.463121928253649,
+      "tokens_seen": 1970470912
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004069170277644038,
+      "loss": 1.3921,
+      "theoretical_loss": 3.46310282424367,
+      "tokens_seen": 1970601984
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040687690579361256,
+      "loss": 1.3309,
+      "theoretical_loss": 3.4630837218600896,
+      "tokens_seen": 1970733056
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004068367838228214,
+      "loss": 1.3685,
+      "theoretical_loss": 3.4630646211026614,
+      "tokens_seen": 1970864128
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040679666185203017,
+      "loss": 1.4224,
+      "theoretical_loss": 3.463045521971139,
+      "tokens_seen": 1970995200
+    },
+    {
+      "epoch": 0.19,
+      "objective/train/advantage_avg": 0.49450939893722534,
+      "objective/train/docs_used": 718798,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.4068423509597778,
+      "objective/train/original_loss": 1.4068424701690674,
+      "objective/train/theoretical_loss": 3.4630264244652755,
+      "objective/train/tokens_used": 341585376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24477232992649078,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506953001022339,
+      "objective/train/weighted_lm_loss": 1.4780831336975098,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4630264244652755,
+      "tokens_seen": 1971126272
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.000406756539881239,
+      "loss": 1.4177,
+      "theoretical_loss": 3.4630264244652755,
+      "tokens_seen": 1971126272
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004067164179104478,
+      "loss": 1.4782,
+      "theoretical_loss": 3.4630073285848244,
+      "tokens_seen": 1971257344
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040667629593965654,
+      "loss": 1.4358,
+      "theoretical_loss": 3.46298823432954,
+      "tokens_seen": 1971388416
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0004066361739688654,
+      "loss": 1.3605,
+      "theoretical_loss": 3.462969141699175,
+      "tokens_seen": 1971519488
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00040659605199807415,
+      "loss": 1.4567,
+      "theoretical_loss": 3.4629500506934834,
+      "tokens_seen": 1971650560
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040655593002728296,
+      "loss": 1.3378,
+      "theoretical_loss": 3.4629309613122192,
+      "tokens_seen": 1971781632
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040651580805649176,
+      "loss": 1.3771,
+      "theoretical_loss": 3.4629118735551363,
+      "tokens_seen": 1971912704
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004064756860857005,
+      "loss": 1.3936,
+      "theoretical_loss": 3.462892787421988,
+      "tokens_seen": 1972043776
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004064355641149093,
+      "loss": 1.369,
+      "theoretical_loss": 3.462873702912529,
+      "tokens_seen": 1972174848
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004063954421441181,
+      "loss": 1.3644,
+      "theoretical_loss": 3.462854620026513,
+      "tokens_seen": 1972305920
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040635532017332693,
+      "loss": 1.4199,
+      "theoretical_loss": 3.4628355387636933,
+      "tokens_seen": 1972436992
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004063151982025357,
+      "loss": 1.4357,
+      "theoretical_loss": 3.4628164591238253,
+      "tokens_seen": 1972568064
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040627507623174455,
+      "loss": 1.3384,
+      "theoretical_loss": 3.462797381106662,
+      "tokens_seen": 1972699136
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004062349542609533,
+      "loss": 1.4473,
+      "theoretical_loss": 3.462778304711958,
+      "tokens_seen": 1972830208
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040619483229016205,
+      "loss": 1.5017,
+      "theoretical_loss": 3.4627592299394676,
+      "tokens_seen": 1972961280
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004061547103193709,
+      "loss": 1.3776,
+      "theoretical_loss": 3.4627401567889446,
+      "tokens_seen": 1973092352
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040611458834857966,
+      "loss": 1.455,
+      "theoretical_loss": 3.4627210852601444,
+      "tokens_seen": 1973223424
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004060744663777885,
+      "loss": 1.4689,
+      "theoretical_loss": 3.4627020153528205,
+      "tokens_seen": 1973354496
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004060343444069973,
+      "loss": 1.3476,
+      "theoretical_loss": 3.462682947066728,
+      "tokens_seen": 1973485568
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004059942224362061,
+      "loss": 1.3917,
+      "theoretical_loss": 3.4626638804016205,
+      "tokens_seen": 1973616640
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004059541004654149,
+      "loss": 1.3614,
+      "theoretical_loss": 3.4626448153572533,
+      "tokens_seen": 1973747712
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040591397849462364,
+      "loss": 1.3588,
+      "theoretical_loss": 3.462625751933381,
+      "tokens_seen": 1973878784
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040587385652383245,
+      "loss": 1.4709,
+      "theoretical_loss": 3.462606690129758,
+      "tokens_seen": 1974009856
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040583373455304125,
+      "loss": 1.4445,
+      "theoretical_loss": 3.462587629946139,
+      "tokens_seen": 1974140928
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040579361258225006,
+      "loss": 1.3953,
+      "theoretical_loss": 3.4625685713822785,
+      "tokens_seen": 1974272000
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004057534906114588,
+      "loss": 1.3125,
+      "theoretical_loss": 3.4625495144379324,
+      "tokens_seen": 1974403072
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040571336864066767,
+      "loss": 1.4379,
+      "theoretical_loss": 3.4625304591128545,
+      "tokens_seen": 1974534144
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004056732466698764,
+      "loss": 1.5163,
+      "theoretical_loss": 3.4625114054068007,
+      "tokens_seen": 1974665216
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040563312469908523,
+      "loss": 1.3984,
+      "theoretical_loss": 3.462492353319525,
+      "tokens_seen": 1974796288
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040559300272829404,
+      "loss": 1.4159,
+      "theoretical_loss": 3.4624733028507824,
+      "tokens_seen": 1974927360
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004055528807575028,
+      "loss": 1.3903,
+      "theoretical_loss": 3.4624542540003294,
+      "tokens_seen": 1975058432
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040551275878671165,
+      "loss": 1.4169,
+      "theoretical_loss": 3.4624352067679194,
+      "tokens_seen": 1975189504
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004054726368159204,
+      "loss": 1.4519,
+      "theoretical_loss": 3.4624161611533086,
+      "tokens_seen": 1975320576
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004054325148451292,
+      "loss": 1.3813,
+      "theoretical_loss": 3.462397117156252,
+      "tokens_seen": 1975451648
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.000405392392874338,
+      "loss": 1.4333,
+      "theoretical_loss": 3.462378074776505,
+      "tokens_seen": 1975582720
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004053522709035468,
+      "loss": 1.4178,
+      "theoretical_loss": 3.462359034013823,
+      "tokens_seen": 1975713792
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040531214893275557,
+      "loss": 1.4204,
+      "theoretical_loss": 3.4623399948679614,
+      "tokens_seen": 1975844864
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004052720269619644,
+      "loss": 1.4593,
+      "theoretical_loss": 3.462320957338676,
+      "tokens_seen": 1975975936
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004052319049911732,
+      "loss": 1.3732,
+      "theoretical_loss": 3.4623019214257216,
+      "tokens_seen": 1976107008
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040519178302038194,
+      "loss": 1.3693,
+      "theoretical_loss": 3.462282887128854,
+      "tokens_seen": 1976238080
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004051516610495908,
+      "loss": 1.3499,
+      "theoretical_loss": 3.462263854447829,
+      "tokens_seen": 1976369152
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040511153907879955,
+      "loss": 1.396,
+      "theoretical_loss": 3.462244823382402,
+      "tokens_seen": 1976500224
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040507141710800835,
+      "loss": 1.4005,
+      "theoretical_loss": 3.462225793932329,
+      "tokens_seen": 1976631296
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040503129513721716,
+      "loss": 1.4407,
+      "theoretical_loss": 3.462206766097366,
+      "tokens_seen": 1976762368
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004049911731664259,
+      "loss": 1.3789,
+      "theoretical_loss": 3.462187739877269,
+      "tokens_seen": 1976893440
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004049510511956347,
+      "loss": 1.3257,
+      "theoretical_loss": 3.4621687152717926,
+      "tokens_seen": 1977024512
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004049109292248435,
+      "loss": 1.3849,
+      "theoretical_loss": 3.4621496922806942,
+      "tokens_seen": 1977155584
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040487080725405233,
+      "loss": 1.3579,
+      "theoretical_loss": 3.4621306709037287,
+      "tokens_seen": 1977286656
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040483068528326114,
+      "loss": 1.2953,
+      "theoretical_loss": 3.4621116511406527,
+      "tokens_seen": 1977417728
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040479056331246994,
+      "loss": 1.3487,
+      "theoretical_loss": 3.4620926329912227,
+      "tokens_seen": 1977548800
+    },
+    {
+      "epoch": 0.2,
+      "objective/train/advantage_avg": 0.4914051294326782,
+      "objective/train/docs_used": 721216,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3864001035690308,
+      "objective/train/original_loss": 1.3864001035690308,
+      "objective/train/theoretical_loss": 3.4620736164551946,
+      "objective/train/tokens_used": 348138976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24286571145057678,
+      "objective/train/value_max": -0.5,
+      "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": 1.4562426805496216,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.7788007855415344,
+      "theoretical_loss": 3.4620736164551946,
+      "tokens_seen": 1977679872
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004047504413416787,
+      "loss": 1.4015,
+      "theoretical_loss": 3.4620736164551946,
+      "tokens_seen": 1977679872
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004047103193708875,
+      "loss": 1.3424,
+      "theoretical_loss": 3.4620546015323237,
+      "tokens_seen": 1977810944
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004046701974000963,
+      "loss": 1.3545,
+      "theoretical_loss": 3.462035588222368,
+      "tokens_seen": 1977942016
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040463007542930506,
+      "loss": 1.3052,
+      "theoretical_loss": 3.462016576525082,
+      "tokens_seen": 1978073088
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004045899534585139,
+      "loss": 1.3823,
+      "theoretical_loss": 3.461997566440223,
+      "tokens_seen": 1978204160
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004045498314877227,
+      "loss": 1.3904,
+      "theoretical_loss": 3.461978557967548,
+      "tokens_seen": 1978335232
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004045097095169315,
+      "loss": 1.3868,
+      "theoretical_loss": 3.461959551106812,
+      "tokens_seen": 1978466304
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004044695875461403,
+      "loss": 1.4984,
+      "theoretical_loss": 3.4619405458577726,
+      "tokens_seen": 1978597376
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040442946557534904,
+      "loss": 1.3859,
+      "theoretical_loss": 3.4619215422201863,
+      "tokens_seen": 1978728448
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040438934360455785,
+      "loss": 1.3891,
+      "theoretical_loss": 3.461902540193809,
+      "tokens_seen": 1978859520
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040434922163376665,
+      "loss": 1.337,
+      "theoretical_loss": 3.4618835397783982,
+      "tokens_seen": 1978990592
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040430909966297546,
+      "loss": 1.3315,
+      "theoretical_loss": 3.461864540973711,
+      "tokens_seen": 1979121664
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004042689776921842,
+      "loss": 1.4529,
+      "theoretical_loss": 3.461845543779503,
+      "tokens_seen": 1979252736
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040422885572139307,
+      "loss": 1.338,
+      "theoretical_loss": 3.461826548195532,
+      "tokens_seen": 1979383808
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004041887337506018,
+      "loss": 1.3846,
+      "theoretical_loss": 3.461807554221554,
+      "tokens_seen": 1979514880
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004041486117798107,
+      "loss": 1.341,
+      "theoretical_loss": 3.4617885618573268,
+      "tokens_seen": 1979645952
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040410848980901944,
+      "loss": 1.3135,
+      "theoretical_loss": 3.461769571102607,
+      "tokens_seen": 1979777024
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004040683678382282,
+      "loss": 1.4306,
+      "theoretical_loss": 3.4617505819571512,
+      "tokens_seen": 1979908096
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040402824586743705,
+      "loss": 1.3472,
+      "theoretical_loss": 3.4617315944207174,
+      "tokens_seen": 1980039168
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004039881238966458,
+      "loss": 1.3783,
+      "theoretical_loss": 3.461712608493063,
+      "tokens_seen": 1980170240
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004039480019258546,
+      "loss": 1.3699,
+      "theoretical_loss": 3.461693624173944,
+      "tokens_seen": 1980301312
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004039078799550634,
+      "loss": 1.3639,
+      "theoretical_loss": 3.4616746414631185,
+      "tokens_seen": 1980432384
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004038677579842722,
+      "loss": 1.3726,
+      "theoretical_loss": 3.4616556603603437,
+      "tokens_seen": 1980563456
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040382763601348097,
+      "loss": 1.3451,
+      "theoretical_loss": 3.461636680865377,
+      "tokens_seen": 1980694528
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004037875140426898,
+      "loss": 1.3352,
+      "theoretical_loss": 3.461617702977975,
+      "tokens_seen": 1980825600
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004037473920718986,
+      "loss": 1.4402,
+      "theoretical_loss": 3.4615987266978965,
+      "tokens_seen": 1980956672
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040370727010110734,
+      "loss": 1.4155,
+      "theoretical_loss": 3.4615797520248983,
+      "tokens_seen": 1981087744
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004036671481303162,
+      "loss": 1.4265,
+      "theoretical_loss": 3.461560778958738,
+      "tokens_seen": 1981218816
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040362702615952495,
+      "loss": 1.4067,
+      "theoretical_loss": 3.4615418074991737,
+      "tokens_seen": 1981349888
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004035869041887338,
+      "loss": 1.337,
+      "theoretical_loss": 3.4615228376459624,
+      "tokens_seen": 1981480960
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040354678221794256,
+      "loss": 1.3683,
+      "theoretical_loss": 3.4615038693988622,
+      "tokens_seen": 1981612032
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004035066602471513,
+      "loss": 1.3439,
+      "theoretical_loss": 3.461484902757631,
+      "tokens_seen": 1981743104
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004034665382763602,
+      "loss": 1.3827,
+      "theoretical_loss": 3.4614659377220263,
+      "tokens_seen": 1981874176
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004034264163055689,
+      "loss": 1.4105,
+      "theoretical_loss": 3.4614469742918064,
+      "tokens_seen": 1982005248
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040338629433477773,
+      "loss": 1.3318,
+      "theoretical_loss": 3.461428012466729,
+      "tokens_seen": 1982136320
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040334617236398654,
+      "loss": 1.3253,
+      "theoretical_loss": 3.461409052246552,
+      "tokens_seen": 1982267392
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040330605039319534,
+      "loss": 1.3924,
+      "theoretical_loss": 3.4613900936310342,
+      "tokens_seen": 1982398464
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004032659284224041,
+      "loss": 1.4029,
+      "theoretical_loss": 3.461371136619933,
+      "tokens_seen": 1982529536
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004032258064516129,
+      "loss": 1.3549,
+      "theoretical_loss": 3.461352181213006,
+      "tokens_seen": 1982660608
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004031856844808217,
+      "loss": 1.4193,
+      "theoretical_loss": 3.461333227410013,
+      "tokens_seen": 1982791680
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040314556251003046,
+      "loss": 1.3819,
+      "theoretical_loss": 3.461314275210711,
+      "tokens_seen": 1982922752
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004031054405392393,
+      "loss": 1.3861,
+      "theoretical_loss": 3.4612953246148583,
+      "tokens_seen": 1983053824
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004030653185684481,
+      "loss": 1.337,
+      "theoretical_loss": 3.461276375622214,
+      "tokens_seen": 1983184896
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004030251965976569,
+      "loss": 1.3586,
+      "theoretical_loss": 3.4612574282325363,
+      "tokens_seen": 1983315968
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004029850746268657,
+      "loss": 1.3626,
+      "theoretical_loss": 3.4612384824455837,
+      "tokens_seen": 1983447040
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040294495265607444,
+      "loss": 1.4421,
+      "theoretical_loss": 3.4612195382611146,
+      "tokens_seen": 1983578112
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004029048306852833,
+      "loss": 1.4,
+      "theoretical_loss": 3.4612005956788874,
+      "tokens_seen": 1983709184
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040286470871449205,
+      "loss": 1.3591,
+      "theoretical_loss": 3.4611816546986613,
+      "tokens_seen": 1983840256
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040282458674370086,
+      "loss": 1.307,
+      "theoretical_loss": 3.461162715320194,
+      "tokens_seen": 1983971328
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040278446477290966,
+      "loss": 1.3866,
+      "theoretical_loss": 3.4611437775432456,
+      "tokens_seen": 1984102400
+    },
+    {
+      "epoch": 0.2,
+      "objective/train/advantage_avg": 0.4923151135444641,
+      "objective/train/docs_used": 723589,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.219787359237671,
+      "objective/train/original_loss": 1.2197872400283813,
+      "objective/train/theoretical_loss": 3.4611248413675737,
+      "objective/train/tokens_used": 354692576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24292775988578796,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504664182662964,
+      "objective/train/weighted_lm_loss": 1.2810142040252686,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4611248413675737,
+      "tokens_seen": 1984233472
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040274434280211847,
+      "loss": 1.3036,
+      "theoretical_loss": 3.4611248413675737,
+      "tokens_seen": 1984233472
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004027042208313272,
+      "loss": 1.4381,
+      "theoretical_loss": 3.4611059067929375,
+      "tokens_seen": 1984364544
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004026640988605361,
+      "loss": 1.4214,
+      "theoretical_loss": 3.461086973819096,
+      "tokens_seen": 1984495616
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040262397688974483,
+      "loss": 1.3865,
+      "theoretical_loss": 3.4610680424458082,
+      "tokens_seen": 1984626688
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004025838549189536,
+      "loss": 1.4281,
+      "theoretical_loss": 3.461049112672833,
+      "tokens_seen": 1984757760
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040254373294816245,
+      "loss": 1.4016,
+      "theoretical_loss": 3.461030184499929,
+      "tokens_seen": 1984888832
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004025036109773712,
+      "loss": 1.3815,
+      "theoretical_loss": 3.461011257926857,
+      "tokens_seen": 1985019904
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040246348900658,
+      "loss": 1.4778,
+      "theoretical_loss": 3.4609923329533743,
+      "tokens_seen": 1985150976
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004024233670357888,
+      "loss": 1.4037,
+      "theoretical_loss": 3.4609734095792404,
+      "tokens_seen": 1985282048
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004023832450649976,
+      "loss": 1.3614,
+      "theoretical_loss": 3.460954487804215,
+      "tokens_seen": 1985413120
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004023431230942064,
+      "loss": 1.3553,
+      "theoretical_loss": 3.4609355676280575,
+      "tokens_seen": 1985544192
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004023030011234152,
+      "loss": 1.3495,
+      "theoretical_loss": 3.460916649050527,
+      "tokens_seen": 1985675264
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.000402262879152624,
+      "loss": 1.432,
+      "theoretical_loss": 3.460897732071383,
+      "tokens_seen": 1985806336
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004022227571818328,
+      "loss": 1.4002,
+      "theoretical_loss": 3.4608788166903857,
+      "tokens_seen": 1985937408
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004021826352110416,
+      "loss": 1.4628,
+      "theoretical_loss": 3.460859902907293,
+      "tokens_seen": 1986068480
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040214251324025035,
+      "loss": 1.4139,
+      "theoretical_loss": 3.4608409907218656,
+      "tokens_seen": 1986199552
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004021023912694592,
+      "loss": 1.3596,
+      "theoretical_loss": 3.4608220801338625,
+      "tokens_seen": 1986330624
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040206226929866796,
+      "loss": 1.4078,
+      "theoretical_loss": 3.4608031711430445,
+      "tokens_seen": 1986461696
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004020221473278767,
+      "loss": 1.4309,
+      "theoretical_loss": 3.4607842637491704,
+      "tokens_seen": 1986592768
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004019820253570856,
+      "loss": 1.3025,
+      "theoretical_loss": 3.460765357952,
+      "tokens_seen": 1986723840
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004019419033862943,
+      "loss": 1.3743,
+      "theoretical_loss": 3.460746453751293,
+      "tokens_seen": 1986854912
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040190178141550313,
+      "loss": 1.3654,
+      "theoretical_loss": 3.4607275511468103,
+      "tokens_seen": 1986985984
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040186165944471194,
+      "loss": 1.4017,
+      "theoretical_loss": 3.4607086501383106,
+      "tokens_seen": 1987117056
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040182153747392074,
+      "loss": 1.4169,
+      "theoretical_loss": 3.4606897507255545,
+      "tokens_seen": 1987248128
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004017814155031295,
+      "loss": 1.4462,
+      "theoretical_loss": 3.4606708529083017,
+      "tokens_seen": 1987379200
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004017412935323383,
+      "loss": 1.3544,
+      "theoretical_loss": 3.460651956686313,
+      "tokens_seen": 1987510272
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004017011715615471,
+      "loss": 1.3125,
+      "theoretical_loss": 3.4606330620593484,
+      "tokens_seen": 1987641344
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004016610495907559,
+      "loss": 1.3766,
+      "theoretical_loss": 3.460614169027167,
+      "tokens_seen": 1987772416
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004016209276199647,
+      "loss": 1.3233,
+      "theoretical_loss": 3.4605952775895306,
+      "tokens_seen": 1987903488
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0004015808056491735,
+      "loss": 1.443,
+      "theoretical_loss": 3.4605763877461984,
+      "tokens_seen": 1988034560
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00040154068367838233,
+      "loss": 1.3527,
+      "theoretical_loss": 3.4605574994969306,
+      "tokens_seen": 1988165632
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0004015005617075911,
+      "loss": 1.4544,
+      "theoretical_loss": 3.460538612841489,
+      "tokens_seen": 1988296704
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00040146043973679984,
+      "loss": 1.3755,
+      "theoretical_loss": 3.4605197277796327,
+      "tokens_seen": 1988427776
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0004014203177660087,
+      "loss": 1.3971,
+      "theoretical_loss": 3.4605008443111225,
+      "tokens_seen": 1988558848
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00040138019579521745,
+      "loss": 1.3689,
+      "theoretical_loss": 3.4604819624357193,
+      "tokens_seen": 1988689920
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00040134007382442626,
+      "loss": 1.3652,
+      "theoretical_loss": 3.4604630821531837,
+      "tokens_seen": 1988820992
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00040129995185363506,
+      "loss": 1.3594,
+      "theoretical_loss": 3.460444203463276,
+      "tokens_seen": 1988952064
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00040125982988284387,
+      "loss": 1.3248,
+      "theoretical_loss": 3.460425326365757,
+      "tokens_seen": 1989083136
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0004012197079120526,
+      "loss": 1.42,
+      "theoretical_loss": 3.4604064508603876,
+      "tokens_seen": 1989214208
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00040117958594126143,
+      "loss": 1.3658,
+      "theoretical_loss": 3.4603875769469283,
+      "tokens_seen": 1989345280
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00040113946397047023,
+      "loss": 1.4075,
+      "theoretical_loss": 3.4603687046251412,
+      "tokens_seen": 1989476352
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.000401099341999679,
+      "loss": 1.3833,
+      "theoretical_loss": 3.4603498338947856,
+      "tokens_seen": 1989607424
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00040105922002888785,
+      "loss": 1.3619,
+      "theoretical_loss": 3.4603309647556233,
+      "tokens_seen": 1989738496
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0004010190980580966,
+      "loss": 1.3305,
+      "theoretical_loss": 3.460312097207415,
+      "tokens_seen": 1989869568
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00040097897608730546,
+      "loss": 1.3684,
+      "theoretical_loss": 3.4602932312499224,
+      "tokens_seen": 1990000640
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0004009388541165142,
+      "loss": 1.3946,
+      "theoretical_loss": 3.4602743668829055,
+      "tokens_seen": 1990131712
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.000400898732145723,
+      "loss": 1.4252,
+      "theoretical_loss": 3.4602555041061267,
+      "tokens_seen": 1990262784
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0004008586101749318,
+      "loss": 1.4067,
+      "theoretical_loss": 3.4602366429193463,
+      "tokens_seen": 1990393856
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0004008184882041406,
+      "loss": 1.4033,
+      "theoretical_loss": 3.4602177833223267,
+      "tokens_seen": 1990524928
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0004007783662333494,
+      "loss": 1.4631,
+      "theoretical_loss": 3.4601989253148275,
+      "tokens_seen": 1990656000
+    },
+    {
+      "epoch": 0.21,
+      "objective/train/advantage_avg": 0.49091672897338867,
+      "objective/train/docs_used": 726036,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3494302034378052,
+      "objective/train/original_loss": 1.3494300842285156,
+      "objective/train/theoretical_loss": 3.4601800688966122,
+      "objective/train/tokens_used": 361246176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24158227443695068,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503196716308594,
+      "objective/train/weighted_lm_loss": 1.4173425436019897,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4601800688966122,
+      "tokens_seen": 1990787072
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0004007382442625582,
+      "loss": 1.4325,
+      "theoretical_loss": 3.4601800688966122,
+      "tokens_seen": 1990787072
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.000400698122291767,
+      "loss": 1.3702,
+      "theoretical_loss": 3.4601612140674405,
+      "tokens_seen": 1990918144
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00040065800032097575,
+      "loss": 1.4781,
+      "theoretical_loss": 3.4601423608270743,
+      "tokens_seen": 1991049216
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0004006178783501846,
+      "loss": 1.4255,
+      "theoretical_loss": 3.4601235091752756,
+      "tokens_seen": 1991180288
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00040057775637939336,
+      "loss": 1.3043,
+      "theoretical_loss": 3.4601046591118063,
+      "tokens_seen": 1991311360
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0004005376344086021,
+      "loss": 1.4977,
+      "theoretical_loss": 3.4600858106364267,
+      "tokens_seen": 1991442432
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00040049751243781097,
+      "loss": 1.4082,
+      "theoretical_loss": 3.4600669637488997,
+      "tokens_seen": 1991573504
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0004004573904670197,
+      "loss": 1.4263,
+      "theoretical_loss": 3.4600481184489866,
+      "tokens_seen": 1991704576
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0004004172684962286,
+      "loss": 1.3695,
+      "theoretical_loss": 3.460029274736449,
+      "tokens_seen": 1991835648
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00040037714652543734,
+      "loss": 1.4313,
+      "theoretical_loss": 3.4600104326110492,
+      "tokens_seen": 1991966720
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00040033702455464614,
+      "loss": 1.3836,
+      "theoretical_loss": 3.459991592072549,
+      "tokens_seen": 1992097792
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00040029690258385495,
+      "loss": 1.393,
+      "theoretical_loss": 3.4599727531207103,
+      "tokens_seen": 1992228864
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0004002567806130637,
+      "loss": 1.3876,
+      "theoretical_loss": 3.4599539157552943,
+      "tokens_seen": 1992359936
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0004002166586422725,
+      "loss": 1.3912,
+      "theoretical_loss": 3.4599350799760646,
+      "tokens_seen": 1992491008
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0004001765366714813,
+      "loss": 1.4201,
+      "theoretical_loss": 3.4599162457827823,
+      "tokens_seen": 1992622080
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0004001364147006901,
+      "loss": 1.4263,
+      "theoretical_loss": 3.4598974131752094,
+      "tokens_seen": 1992753152
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0004000962927298989,
+      "loss": 1.4258,
+      "theoretical_loss": 3.459878582153109,
+      "tokens_seen": 1992884224
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00040005617075910773,
+      "loss": 1.3892,
+      "theoretical_loss": 3.4598597527162425,
+      "tokens_seen": 1993015296
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0004000160487883165,
+      "loss": 1.4084,
+      "theoretical_loss": 3.4598409248643724,
+      "tokens_seen": 1993146368
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039997592681752524,
+      "loss": 1.395,
+      "theoretical_loss": 3.459822098597262,
+      "tokens_seen": 1993277440
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003999358048467341,
+      "loss": 1.3852,
+      "theoretical_loss": 3.459803273914672,
+      "tokens_seen": 1993408512
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039989568287594285,
+      "loss": 1.3896,
+      "theoretical_loss": 3.4597844508163655,
+      "tokens_seen": 1993539584
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039985556090515166,
+      "loss": 1.3527,
+      "theoretical_loss": 3.4597656293021064,
+      "tokens_seen": 1993670656
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039981543893436046,
+      "loss": 1.4139,
+      "theoretical_loss": 3.4597468093716555,
+      "tokens_seen": 1993801728
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039977531696356927,
+      "loss": 1.3975,
+      "theoretical_loss": 3.459727991024776,
+      "tokens_seen": 1993932800
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003997351949927781,
+      "loss": 1.3048,
+      "theoretical_loss": 3.459709174261231,
+      "tokens_seen": 1994063872
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039969507302198683,
+      "loss": 1.3233,
+      "theoretical_loss": 3.459690359080783,
+      "tokens_seen": 1994194944
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039965495105119563,
+      "loss": 1.4332,
+      "theoretical_loss": 3.459671545483194,
+      "tokens_seen": 1994326016
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039961482908040444,
+      "loss": 1.3923,
+      "theoretical_loss": 3.4596527334682277,
+      "tokens_seen": 1994457088
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039957470710961325,
+      "loss": 1.4239,
+      "theoretical_loss": 3.459633923035647,
+      "tokens_seen": 1994588160
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.000399534585138822,
+      "loss": 1.4184,
+      "theoretical_loss": 3.4596151141852145,
+      "tokens_seen": 1994719232
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039949446316803086,
+      "loss": 1.3193,
+      "theoretical_loss": 3.4595963069166933,
+      "tokens_seen": 1994850304
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003994543411972396,
+      "loss": 1.3958,
+      "theoretical_loss": 3.459577501229846,
+      "tokens_seen": 1994981376
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003994142192264484,
+      "loss": 1.3845,
+      "theoretical_loss": 3.4595586971244363,
+      "tokens_seen": 1995112448
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003993740972556572,
+      "loss": 1.3394,
+      "theoretical_loss": 3.459539894600227,
+      "tokens_seen": 1995243520
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.000399333975284866,
+      "loss": 1.4946,
+      "theoretical_loss": 3.4595210936569813,
+      "tokens_seen": 1995374592
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003992938533140748,
+      "loss": 1.3986,
+      "theoretical_loss": 3.459502294294462,
+      "tokens_seen": 1995505664
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003992537313432836,
+      "loss": 1.347,
+      "theoretical_loss": 3.459483496512434,
+      "tokens_seen": 1995636736
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003992136093724924,
+      "loss": 1.3463,
+      "theoretical_loss": 3.459464700310659,
+      "tokens_seen": 1995767808
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039917348740170115,
+      "loss": 1.3175,
+      "theoretical_loss": 3.4594459056889004,
+      "tokens_seen": 1995898880
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039913336543091,
+      "loss": 1.3229,
+      "theoretical_loss": 3.4594271126469227,
+      "tokens_seen": 1996029952
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039909324346011876,
+      "loss": 1.2427,
+      "theoretical_loss": 3.4594083211844886,
+      "tokens_seen": 1996161024
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039905312148932757,
+      "loss": 1.4293,
+      "theoretical_loss": 3.459389531301362,
+      "tokens_seen": 1996292096
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039901299951853637,
+      "loss": 1.387,
+      "theoretical_loss": 3.459370742997306,
+      "tokens_seen": 1996423168
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003989728775477451,
+      "loss": 1.3908,
+      "theoretical_loss": 3.459351956272085,
+      "tokens_seen": 1996554240
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.000398932755576954,
+      "loss": 1.4189,
+      "theoretical_loss": 3.4593331711254622,
+      "tokens_seen": 1996685312
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039889263360616274,
+      "loss": 1.3756,
+      "theoretical_loss": 3.459314387557201,
+      "tokens_seen": 1996816384
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039885251163537154,
+      "loss": 1.4502,
+      "theoretical_loss": 3.459295605567066,
+      "tokens_seen": 1996947456
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039881238966458035,
+      "loss": 1.3996,
+      "theoretical_loss": 3.45927682515482,
+      "tokens_seen": 1997078528
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003987722676937891,
+      "loss": 1.4367,
+      "theoretical_loss": 3.4592580463202283,
+      "tokens_seen": 1997209600
+    },
+    {
+      "epoch": 0.21,
+      "objective/train/advantage_avg": 0.49159374833106995,
+      "objective/train/docs_used": 728467,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2781449556350708,
+      "objective/train/original_loss": 1.2781447172164917,
+      "objective/train/theoretical_loss": 3.459239269063054,
+      "objective/train/tokens_used": 367799776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24257053434848785,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050392508506775,
+      "objective/train/weighted_lm_loss": 1.342505693435669,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.459239269063054,
+      "tokens_seen": 1997340672
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003987321457229979,
+      "loss": 1.3671,
+      "theoretical_loss": 3.459239269063054,
+      "tokens_seen": 1997340672
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003986920237522067,
+      "loss": 1.3627,
+      "theoretical_loss": 3.459220493383061,
+      "tokens_seen": 1997471744
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003986519017814155,
+      "loss": 1.4409,
+      "theoretical_loss": 3.459201719280013,
+      "tokens_seen": 1997602816
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039861177981062427,
+      "loss": 1.4338,
+      "theoretical_loss": 3.459182946753675,
+      "tokens_seen": 1997733888
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039857165783983313,
+      "loss": 1.3179,
+      "theoretical_loss": 3.459164175803811,
+      "tokens_seen": 1997864960
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003985315358690419,
+      "loss": 1.3414,
+      "theoretical_loss": 3.4591454064301845,
+      "tokens_seen": 1997996032
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003984914138982507,
+      "loss": 1.3112,
+      "theoretical_loss": 3.4591266386325605,
+      "tokens_seen": 1998127104
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003984512919274595,
+      "loss": 1.3586,
+      "theoretical_loss": 3.459107872410703,
+      "tokens_seen": 1998258176
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039841116995666825,
+      "loss": 1.3902,
+      "theoretical_loss": 3.4590891077643766,
+      "tokens_seen": 1998389248
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003983710479858771,
+      "loss": 1.4321,
+      "theoretical_loss": 3.459070344693345,
+      "tokens_seen": 1998520320
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039833092601508586,
+      "loss": 1.4316,
+      "theoretical_loss": 3.4590515831973736,
+      "tokens_seen": 1998651392
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039829080404429467,
+      "loss": 1.3714,
+      "theoretical_loss": 3.4590328232762255,
+      "tokens_seen": 1998782464
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003982506820735035,
+      "loss": 1.3298,
+      "theoretical_loss": 3.459014064929667,
+      "tokens_seen": 1998913536
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039821056010271223,
+      "loss": 1.4036,
+      "theoretical_loss": 3.4589953081574616,
+      "tokens_seen": 1999044608
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039817043813192103,
+      "loss": 1.3792,
+      "theoretical_loss": 3.458976552959374,
+      "tokens_seen": 1999175680
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039813031616112984,
+      "loss": 1.3719,
+      "theoretical_loss": 3.4589577993351694,
+      "tokens_seen": 1999306752
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039809019419033865,
+      "loss": 1.4216,
+      "theoretical_loss": 3.4589390472846127,
+      "tokens_seen": 1999437824
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003980500722195474,
+      "loss": 1.2957,
+      "theoretical_loss": 3.4589202968074675,
+      "tokens_seen": 1999568896
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039800995024875626,
+      "loss": 1.3797,
+      "theoretical_loss": 3.4589015479035,
+      "tokens_seen": 1999699968
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.000397969828277965,
+      "loss": 1.5021,
+      "theoretical_loss": 3.458882800572474,
+      "tokens_seen": 1999831040
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003979297063071738,
+      "loss": 1.4604,
+      "theoretical_loss": 3.4588640548141556,
+      "tokens_seen": 1999962112
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003978895843363826,
+      "loss": 1.3084,
+      "theoretical_loss": 3.458845310628309,
+      "tokens_seen": 2000093184
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003978494623655914,
+      "loss": 1.353,
+      "theoretical_loss": 3.4588265680146995,
+      "tokens_seen": 2000224256
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039780934039480024,
+      "loss": 1.3736,
+      "theoretical_loss": 3.458807826973092,
+      "tokens_seen": 2000355328
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.000397769218424009,
+      "loss": 1.4307,
+      "theoretical_loss": 3.458789087503252,
+      "tokens_seen": 2000486400
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003977290964532178,
+      "loss": 1.3606,
+      "theoretical_loss": 3.4587703496049444,
+      "tokens_seen": 2000617472
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003976889744824266,
+      "loss": 1.4031,
+      "theoretical_loss": 3.458751613277935,
+      "tokens_seen": 2000748544
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003976488525116354,
+      "loss": 1.3763,
+      "theoretical_loss": 3.4587328785219884,
+      "tokens_seen": 2000879616
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039760873054084416,
+      "loss": 1.372,
+      "theoretical_loss": 3.45871414533687,
+      "tokens_seen": 2001010688
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039756860857005297,
+      "loss": 1.3173,
+      "theoretical_loss": 3.4586954137223467,
+      "tokens_seen": 2001141760
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039752848659926177,
+      "loss": 1.3891,
+      "theoretical_loss": 3.4586766836781813,
+      "tokens_seen": 2001272832
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003974883646284705,
+      "loss": 1.4406,
+      "theoretical_loss": 3.4586579552041417,
+      "tokens_seen": 2001403904
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003974482426576794,
+      "loss": 1.4337,
+      "theoretical_loss": 3.458639228299992,
+      "tokens_seen": 2001534976
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039740812068688814,
+      "loss": 1.4202,
+      "theoretical_loss": 3.458620502965499,
+      "tokens_seen": 2001666048
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039736799871609694,
+      "loss": 1.3705,
+      "theoretical_loss": 3.458601779200427,
+      "tokens_seen": 2001797120
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039732787674530575,
+      "loss": 1.4037,
+      "theoretical_loss": 3.458583057004543,
+      "tokens_seen": 2001928192
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003972877547745145,
+      "loss": 1.3401,
+      "theoretical_loss": 3.458564336377612,
+      "tokens_seen": 2002059264
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039724763280372336,
+      "loss": 1.3432,
+      "theoretical_loss": 3.4585456173194,
+      "tokens_seen": 2002190336
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003972075108329321,
+      "loss": 1.378,
+      "theoretical_loss": 3.4585268998296725,
+      "tokens_seen": 2002321408
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003971673888621409,
+      "loss": 1.4046,
+      "theoretical_loss": 3.4585081839081964,
+      "tokens_seen": 2002452480
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003971272668913497,
+      "loss": 1.3831,
+      "theoretical_loss": 3.4584894695547366,
+      "tokens_seen": 2002583552
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039708714492055853,
+      "loss": 1.4288,
+      "theoretical_loss": 3.4584707567690596,
+      "tokens_seen": 2002714624
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003970470229497673,
+      "loss": 1.4003,
+      "theoretical_loss": 3.4584520455509313,
+      "tokens_seen": 2002845696
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003970069009789761,
+      "loss": 1.3171,
+      "theoretical_loss": 3.4584333359001187,
+      "tokens_seen": 2002976768
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003969667790081849,
+      "loss": 1.4225,
+      "theoretical_loss": 3.4584146278163868,
+      "tokens_seen": 2003107840
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039692665703739365,
+      "loss": 1.3698,
+      "theoretical_loss": 3.458395921299502,
+      "tokens_seen": 2003238912
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003968865350666025,
+      "loss": 1.4187,
+      "theoretical_loss": 3.458377216349231,
+      "tokens_seen": 2003369984
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039684641309581126,
+      "loss": 1.4337,
+      "theoretical_loss": 3.45835851296534,
+      "tokens_seen": 2003501056
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039680629112502007,
+      "loss": 1.3812,
+      "theoretical_loss": 3.458339811147595,
+      "tokens_seen": 2003632128
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003967661691542289,
+      "loss": 1.3721,
+      "theoretical_loss": 3.458321110895763,
+      "tokens_seen": 2003763200
+    },
+    {
+      "epoch": 0.21,
+      "objective/train/advantage_avg": 0.4897269308567047,
+      "objective/train/docs_used": 730811,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.387560248374939,
+      "objective/train/original_loss": 1.387560248374939,
+      "objective/train/theoretical_loss": 3.45830241220961,
+      "objective/train/tokens_used": 374353376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24060115218162537,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0501958131790161,
+      "objective/train/weighted_lm_loss": 1.4570908546447754,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9048373699188232,
+      "theoretical_loss": 3.45830241220961,
+      "tokens_seen": 2003894272
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039672604718343763,
+      "loss": 1.4127,
+      "theoretical_loss": 3.45830241220961,
+      "tokens_seen": 2003894272
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039668592521264643,
+      "loss": 1.472,
+      "theoretical_loss": 3.4582837150889025,
+      "tokens_seen": 2004025344
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039664580324185524,
+      "loss": 1.3861,
+      "theoretical_loss": 3.4582650195334077,
+      "tokens_seen": 2004156416
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039660568127106405,
+      "loss": 1.3713,
+      "theoretical_loss": 3.4582463255428917,
+      "tokens_seen": 2004287488
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039656555930027285,
+      "loss": 1.4203,
+      "theoretical_loss": 3.4582276331171213,
+      "tokens_seen": 2004418560
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00039652543732948166,
+      "loss": 1.3749,
+      "theoretical_loss": 3.4582089422558626,
+      "tokens_seen": 2004549632
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0003964853153586904,
+      "loss": 1.4042,
+      "theoretical_loss": 3.4581902529588833,
+      "tokens_seen": 2004680704
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003964451933878992,
+      "loss": 1.3534,
+      "theoretical_loss": 3.4581715652259497,
+      "tokens_seen": 2004811776
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.000396405071417108,
+      "loss": 1.4267,
+      "theoretical_loss": 3.458152879056829,
+      "tokens_seen": 2004942848
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003963649494463168,
+      "loss": 1.3509,
+      "theoretical_loss": 3.4581341944512882,
+      "tokens_seen": 2005073920
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039632482747552564,
+      "loss": 1.4872,
+      "theoretical_loss": 3.458115511409094,
+      "tokens_seen": 2005204992
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003962847055047344,
+      "loss": 1.3483,
+      "theoretical_loss": 3.458096829930013,
+      "tokens_seen": 2005336064
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003962445835339432,
+      "loss": 1.3852,
+      "theoretical_loss": 3.4580781500138125,
+      "tokens_seen": 2005467136
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.000396204461563152,
+      "loss": 1.365,
+      "theoretical_loss": 3.4580594716602597,
+      "tokens_seen": 2005598208
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003961643395923608,
+      "loss": 1.4106,
+      "theoretical_loss": 3.458040794869123,
+      "tokens_seen": 2005729280
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039612421762156956,
+      "loss": 1.3605,
+      "theoretical_loss": 3.4580221196401673,
+      "tokens_seen": 2005860352
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039608409565077837,
+      "loss": 1.3914,
+      "theoretical_loss": 3.4580034459731612,
+      "tokens_seen": 2005991424
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039604397367998717,
+      "loss": 1.4189,
+      "theoretical_loss": 3.4579847738678726,
+      "tokens_seen": 2006122496
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003960038517091959,
+      "loss": 1.3462,
+      "theoretical_loss": 3.4579661033240674,
+      "tokens_seen": 2006253568
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003959637297384048,
+      "loss": 1.3433,
+      "theoretical_loss": 3.457947434341514,
+      "tokens_seen": 2006384640
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039592360776761354,
+      "loss": 1.4095,
+      "theoretical_loss": 3.45792876691998,
+      "tokens_seen": 2006515712
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003958834857968224,
+      "loss": 1.4096,
+      "theoretical_loss": 3.457910101059232,
+      "tokens_seen": 2006646784
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039584336382603115,
+      "loss": 1.3958,
+      "theoretical_loss": 3.4578914367590388,
+      "tokens_seen": 2006777856
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003958032418552399,
+      "loss": 1.303,
+      "theoretical_loss": 3.457872774019166,
+      "tokens_seen": 2006908928
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039576311988444876,
+      "loss": 1.3258,
+      "theoretical_loss": 3.457854112839384,
+      "tokens_seen": 2007040000
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003957229979136575,
+      "loss": 1.4108,
+      "theoretical_loss": 3.4578354532194586,
+      "tokens_seen": 2007171072
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003956828759428663,
+      "loss": 1.4434,
+      "theoretical_loss": 3.4578167951591574,
+      "tokens_seen": 2007302144
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003956427539720751,
+      "loss": 1.3982,
+      "theoretical_loss": 3.4577981386582497,
+      "tokens_seen": 2007433216
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039560263200128393,
+      "loss": 1.401,
+      "theoretical_loss": 3.4577794837165023,
+      "tokens_seen": 2007564288
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003955625100304927,
+      "loss": 1.3946,
+      "theoretical_loss": 3.457760830333683,
+      "tokens_seen": 2007695360
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003955223880597015,
+      "loss": 1.375,
+      "theoretical_loss": 3.4577421785095606,
+      "tokens_seen": 2007826432
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003954822660889103,
+      "loss": 1.3212,
+      "theoretical_loss": 3.4577235282439025,
+      "tokens_seen": 2007957504
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039544214411811905,
+      "loss": 1.2658,
+      "theoretical_loss": 3.457704879536477,
+      "tokens_seen": 2008088576
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003954020221473279,
+      "loss": 1.4363,
+      "theoretical_loss": 3.457686232387052,
+      "tokens_seen": 2008219648
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039536190017653666,
+      "loss": 1.3249,
+      "theoretical_loss": 3.457667586795396,
+      "tokens_seen": 2008350720
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003953217782057455,
+      "loss": 1.4248,
+      "theoretical_loss": 3.457648942761277,
+      "tokens_seen": 2008481792
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003952816562349543,
+      "loss": 1.384,
+      "theoretical_loss": 3.457630300284463,
+      "tokens_seen": 2008612864
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.000395241534264163,
+      "loss": 1.3887,
+      "theoretical_loss": 3.4576116593647224,
+      "tokens_seen": 2008743936
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003952014122933719,
+      "loss": 1.3544,
+      "theoretical_loss": 3.457593020001824,
+      "tokens_seen": 2008875008
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039516129032258064,
+      "loss": 1.3904,
+      "theoretical_loss": 3.457574382195536,
+      "tokens_seen": 2009006080
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039512116835178945,
+      "loss": 1.3882,
+      "theoretical_loss": 3.4575557459456263,
+      "tokens_seen": 2009137152
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039508104638099825,
+      "loss": 1.3803,
+      "theoretical_loss": 3.457537111251864,
+      "tokens_seen": 2009268224
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039504092441020706,
+      "loss": 1.3911,
+      "theoretical_loss": 3.457518478114018,
+      "tokens_seen": 2009399296
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003950008024394158,
+      "loss": 1.4589,
+      "theoretical_loss": 3.4574998465318565,
+      "tokens_seen": 2009530368
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003949606804686246,
+      "loss": 1.4201,
+      "theoretical_loss": 3.4574812165051476,
+      "tokens_seen": 2009661440
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003949205584978334,
+      "loss": 1.4518,
+      "theoretical_loss": 3.457462588033661,
+      "tokens_seen": 2009792512
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003948804365270422,
+      "loss": 1.4245,
+      "theoretical_loss": 3.457443961117165,
+      "tokens_seen": 2009923584
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039484031455625104,
+      "loss": 1.3548,
+      "theoretical_loss": 3.457425335755428,
+      "tokens_seen": 2010054656
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003948001925854598,
+      "loss": 1.3581,
+      "theoretical_loss": 3.4574067119482197,
+      "tokens_seen": 2010185728
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003947600706146686,
+      "loss": 1.4259,
+      "theoretical_loss": 3.4573880896953084,
+      "tokens_seen": 2010316800
+    },
+    {
+      "epoch": 0.22,
+      "objective/train/advantage_avg": 0.4927845001220703,
+      "objective/train/docs_used": 733267,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2820780277252197,
+      "objective/train/original_loss": 1.2820780277252197,
+      "objective/train/theoretical_loss": 3.457369468996463,
+      "objective/train/tokens_used": 380906976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2431802898645401,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050514817237854,
+      "objective/train/weighted_lm_loss": 1.3468701839447021,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.457369468996463,
+      "tokens_seen": 2010447872
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003947199486438774,
+      "loss": 1.3571,
+      "theoretical_loss": 3.457369468996463,
+      "tokens_seen": 2010447872
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003946798266730862,
+      "loss": 1.3682,
+      "theoretical_loss": 3.457350849851453,
+      "tokens_seen": 2010578944
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.000394639704702295,
+      "loss": 1.348,
+      "theoretical_loss": 3.4573322322600473,
+      "tokens_seen": 2010710016
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039459958273150377,
+      "loss": 1.3265,
+      "theoretical_loss": 3.457313616222015,
+      "tokens_seen": 2010841088
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039455946076071257,
+      "loss": 1.4141,
+      "theoretical_loss": 3.4572950017371245,
+      "tokens_seen": 2010972160
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003945193387899214,
+      "loss": 1.4268,
+      "theoretical_loss": 3.4572763888051457,
+      "tokens_seen": 2011103232
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003944792168191302,
+      "loss": 1.3737,
+      "theoretical_loss": 3.4572577774258484,
+      "tokens_seen": 2011234304
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039443909484833894,
+      "loss": 1.3792,
+      "theoretical_loss": 3.457239167599001,
+      "tokens_seen": 2011365376
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003943989728775478,
+      "loss": 1.3551,
+      "theoretical_loss": 3.457220559324373,
+      "tokens_seen": 2011496448
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039435885090675655,
+      "loss": 1.4019,
+      "theoretical_loss": 3.457201952601734,
+      "tokens_seen": 2011627520
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003943187289359653,
+      "loss": 1.2829,
+      "theoretical_loss": 3.4571833474308535,
+      "tokens_seen": 2011758592
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039427860696517416,
+      "loss": 1.3878,
+      "theoretical_loss": 3.457164743811501,
+      "tokens_seen": 2011889664
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003942384849943829,
+      "loss": 1.3734,
+      "theoretical_loss": 3.4571461417434453,
+      "tokens_seen": 2012020736
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003941983630235917,
+      "loss": 1.4137,
+      "theoretical_loss": 3.4571275412264573,
+      "tokens_seen": 2012151808
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003941582410528005,
+      "loss": 1.3243,
+      "theoretical_loss": 3.4571089422603056,
+      "tokens_seen": 2012282880
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039411811908200933,
+      "loss": 1.4043,
+      "theoretical_loss": 3.4570903448447607,
+      "tokens_seen": 2012413952
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039407799711121814,
+      "loss": 1.3202,
+      "theoretical_loss": 3.4570717489795912,
+      "tokens_seen": 2012545024
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003940378751404269,
+      "loss": 1.4129,
+      "theoretical_loss": 3.4570531546645684,
+      "tokens_seen": 2012676096
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003939977531696357,
+      "loss": 1.3955,
+      "theoretical_loss": 3.457034561899461,
+      "tokens_seen": 2012807168
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003939576311988445,
+      "loss": 1.366,
+      "theoretical_loss": 3.457015970684039,
+      "tokens_seen": 2012938240
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003939175092280533,
+      "loss": 1.334,
+      "theoretical_loss": 3.4569973810180725,
+      "tokens_seen": 2013069312
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039387738725726206,
+      "loss": 1.3551,
+      "theoretical_loss": 3.4569787929013325,
+      "tokens_seen": 2013200384
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003938372652864709,
+      "loss": 1.3734,
+      "theoretical_loss": 3.456960206333587,
+      "tokens_seen": 2013331456
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003937971433156797,
+      "loss": 1.4296,
+      "theoretical_loss": 3.4569416213146082,
+      "tokens_seen": 2013462528
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003937570213448884,
+      "loss": 1.2998,
+      "theoretical_loss": 3.4569230378441644,
+      "tokens_seen": 2013593600
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003937168993740973,
+      "loss": 1.4236,
+      "theoretical_loss": 3.456904455922027,
+      "tokens_seen": 2013724672
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039367677740330604,
+      "loss": 1.3995,
+      "theoretical_loss": 3.456885875547966,
+      "tokens_seen": 2013855744
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039363665543251485,
+      "loss": 1.3561,
+      "theoretical_loss": 3.456867296721751,
+      "tokens_seen": 2013986816
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039359653346172365,
+      "loss": 1.3559,
+      "theoretical_loss": 3.4568487194431534,
+      "tokens_seen": 2014117888
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039355641149093246,
+      "loss": 1.4389,
+      "theoretical_loss": 3.456830143711943,
+      "tokens_seen": 2014248960
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003935162895201412,
+      "loss": 1.3045,
+      "theoretical_loss": 3.4568115695278903,
+      "tokens_seen": 2014380032
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039347616754935,
+      "loss": 1.3797,
+      "theoretical_loss": 3.4567929968907656,
+      "tokens_seen": 2014511104
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003934360455785588,
+      "loss": 1.4324,
+      "theoretical_loss": 3.4567744258003392,
+      "tokens_seen": 2014642176
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039339592360776763,
+      "loss": 1.3521,
+      "theoretical_loss": 3.456755856256383,
+      "tokens_seen": 2014773248
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039335580163697644,
+      "loss": 1.3576,
+      "theoretical_loss": 3.456737288258666,
+      "tokens_seen": 2014904320
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003933156796661852,
+      "loss": 1.2716,
+      "theoretical_loss": 3.45671872180696,
+      "tokens_seen": 2015035392
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039327555769539405,
+      "loss": 1.3969,
+      "theoretical_loss": 3.456700156901035,
+      "tokens_seen": 2015166464
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003932354357246028,
+      "loss": 1.34,
+      "theoretical_loss": 3.4566815935406625,
+      "tokens_seen": 2015297536
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003931953137538116,
+      "loss": 1.3664,
+      "theoretical_loss": 3.456663031725612,
+      "tokens_seen": 2015428608
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003931551917830204,
+      "loss": 1.3628,
+      "theoretical_loss": 3.456644471455656,
+      "tokens_seen": 2015559680
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039311506981222916,
+      "loss": 1.3578,
+      "theoretical_loss": 3.4566259127305647,
+      "tokens_seen": 2015690752
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039307494784143797,
+      "loss": 1.3499,
+      "theoretical_loss": 3.456607355550109,
+      "tokens_seen": 2015821824
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003930348258706468,
+      "loss": 1.3911,
+      "theoretical_loss": 3.45658879991406,
+      "tokens_seen": 2015952896
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003929947038998556,
+      "loss": 1.403,
+      "theoretical_loss": 3.4565702458221885,
+      "tokens_seen": 2016083968
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039295458192906434,
+      "loss": 1.3541,
+      "theoretical_loss": 3.4565516932742657,
+      "tokens_seen": 2016215040
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003929144599582732,
+      "loss": 1.3771,
+      "theoretical_loss": 3.4565331422700636,
+      "tokens_seen": 2016346112
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039287433798748195,
+      "loss": 1.35,
+      "theoretical_loss": 3.456514592809352,
+      "tokens_seen": 2016477184
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003928342160166907,
+      "loss": 1.38,
+      "theoretical_loss": 3.4564960448919035,
+      "tokens_seen": 2016608256
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039279409404589956,
+      "loss": 1.4053,
+      "theoretical_loss": 3.4564774985174886,
+      "tokens_seen": 2016739328
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003927539720751083,
+      "loss": 1.4133,
+      "theoretical_loss": 3.456458953685879,
+      "tokens_seen": 2016870400
+    },
+    {
+      "epoch": 0.22,
+      "objective/train/advantage_avg": 0.4912193715572357,
+      "objective/train/docs_used": 735776,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.4399480819702148,
+      "objective/train/original_loss": 1.4399480819702148,
+      "objective/train/theoretical_loss": 3.456440410396846,
+      "objective/train/tokens_used": 387460576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24175754189491272,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503509044647217,
+      "objective/train/weighted_lm_loss": 1.5122746229171753,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.456440410396846,
+      "tokens_seen": 2017001472
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003927138501043172,
+      "loss": 1.5043,
+      "theoretical_loss": 3.456440410396846,
+      "tokens_seen": 2017001472
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003926737281335259,
+      "loss": 1.3345,
+      "theoretical_loss": 3.456421868650161,
+      "tokens_seen": 2017132544
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039263360616273473,
+      "loss": 1.4017,
+      "theoretical_loss": 3.456403328445595,
+      "tokens_seen": 2017263616
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039259348419194354,
+      "loss": 1.3727,
+      "theoretical_loss": 3.4563847897829207,
+      "tokens_seen": 2017394688
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003925533622211523,
+      "loss": 1.4396,
+      "theoretical_loss": 3.456366252661909,
+      "tokens_seen": 2017525760
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003925132402503611,
+      "loss": 1.3063,
+      "theoretical_loss": 3.456347717082332,
+      "tokens_seen": 2017656832
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003924731182795699,
+      "loss": 1.3722,
+      "theoretical_loss": 3.4563291830439606,
+      "tokens_seen": 2017787904
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003924329963087787,
+      "loss": 1.3379,
+      "theoretical_loss": 3.456310650546567,
+      "tokens_seen": 2017918976
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039239287433798746,
+      "loss": 1.3224,
+      "theoretical_loss": 3.456292119589923,
+      "tokens_seen": 2018050048
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003923527523671963,
+      "loss": 1.3444,
+      "theoretical_loss": 3.456273590173801,
+      "tokens_seen": 2018181120
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003923126303964051,
+      "loss": 1.3811,
+      "theoretical_loss": 3.4562550622979717,
+      "tokens_seen": 2018312192
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003922725084256138,
+      "loss": 1.4358,
+      "theoretical_loss": 3.456236535962208,
+      "tokens_seen": 2018443264
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003922323864548227,
+      "loss": 1.4015,
+      "theoretical_loss": 3.4562180111662815,
+      "tokens_seen": 2018574336
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039219226448403144,
+      "loss": 1.4098,
+      "theoretical_loss": 3.4561994879099647,
+      "tokens_seen": 2018705408
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003921521425132403,
+      "loss": 1.3418,
+      "theoretical_loss": 3.4561809661930285,
+      "tokens_seen": 2018836480
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039211202054244905,
+      "loss": 1.4184,
+      "theoretical_loss": 3.4561624460152465,
+      "tokens_seen": 2018967552
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039207189857165786,
+      "loss": 1.3621,
+      "theoretical_loss": 3.45614392737639,
+      "tokens_seen": 2019098624
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039203177660086666,
+      "loss": 1.328,
+      "theoretical_loss": 3.4561254102762318,
+      "tokens_seen": 2019229696
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003919916546300754,
+      "loss": 1.401,
+      "theoretical_loss": 3.4561068947145435,
+      "tokens_seen": 2019360768
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003919515326592842,
+      "loss": 1.3869,
+      "theoretical_loss": 3.4560883806910976,
+      "tokens_seen": 2019491840
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039191141068849303,
+      "loss": 1.3604,
+      "theoretical_loss": 3.456069868205667,
+      "tokens_seen": 2019622912
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039187128871770183,
+      "loss": 1.4177,
+      "theoretical_loss": 3.4560513572580236,
+      "tokens_seen": 2019753984
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003918311667469106,
+      "loss": 1.3612,
+      "theoretical_loss": 3.4560328478479407,
+      "tokens_seen": 2019885056
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039179104477611945,
+      "loss": 1.3945,
+      "theoretical_loss": 3.4560143399751895,
+      "tokens_seen": 2020016128
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003917509228053282,
+      "loss": 1.364,
+      "theoretical_loss": 3.4559958336395438,
+      "tokens_seen": 2020147200
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039171080083453695,
+      "loss": 1.3409,
+      "theoretical_loss": 3.455977328840775,
+      "tokens_seen": 2020278272
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003916706788637458,
+      "loss": 1.31,
+      "theoretical_loss": 3.455958825578657,
+      "tokens_seen": 2020409344
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039163055689295456,
+      "loss": 1.3543,
+      "theoretical_loss": 3.4559403238529622,
+      "tokens_seen": 2020540416
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00039159043492216337,
+      "loss": 1.4006,
+      "theoretical_loss": 3.455921823663463,
+      "tokens_seen": 2020671488
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003915503129513722,
+      "loss": 1.4132,
+      "theoretical_loss": 3.455903325009932,
+      "tokens_seen": 2020802560
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.000391510190980581,
+      "loss": 1.4315,
+      "theoretical_loss": 3.4558848278921426,
+      "tokens_seen": 2020933632
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003914700690097898,
+      "loss": 1.4074,
+      "theoretical_loss": 3.4558663323098675,
+      "tokens_seen": 2021064704
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003914299470389986,
+      "loss": 1.4034,
+      "theoretical_loss": 3.45584783826288,
+      "tokens_seen": 2021195776
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00039138982506820735,
+      "loss": 1.3315,
+      "theoretical_loss": 3.455829345750953,
+      "tokens_seen": 2021326848
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00039134970309741615,
+      "loss": 1.3436,
+      "theoretical_loss": 3.4558108547738584,
+      "tokens_seen": 2021457920
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00039130958112662496,
+      "loss": 1.3867,
+      "theoretical_loss": 3.455792365331371,
+      "tokens_seen": 2021588992
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003912694591558337,
+      "loss": 1.3334,
+      "theoretical_loss": 3.4557738774232636,
+      "tokens_seen": 2021720064
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003912293371850426,
+      "loss": 1.4077,
+      "theoretical_loss": 3.4557553910493084,
+      "tokens_seen": 2021851136
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003911892152142513,
+      "loss": 1.4062,
+      "theoretical_loss": 3.455736906209279,
+      "tokens_seen": 2021982208
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00039114909324346013,
+      "loss": 1.3664,
+      "theoretical_loss": 3.45571842290295,
+      "tokens_seen": 2022113280
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00039110897127266894,
+      "loss": 1.4843,
+      "theoretical_loss": 3.455699941130093,
+      "tokens_seen": 2022244352
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003910688493018777,
+      "loss": 1.3891,
+      "theoretical_loss": 3.4556814608904824,
+      "tokens_seen": 2022375424
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003910287273310865,
+      "loss": 1.4079,
+      "theoretical_loss": 3.455662982183892,
+      "tokens_seen": 2022506496
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003909886053602953,
+      "loss": 1.3961,
+      "theoretical_loss": 3.4556445050100937,
+      "tokens_seen": 2022637568
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003909484833895041,
+      "loss": 1.3485,
+      "theoretical_loss": 3.455626029368862,
+      "tokens_seen": 2022768640
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00039090836141871286,
+      "loss": 1.4065,
+      "theoretical_loss": 3.455607555259971,
+      "tokens_seen": 2022899712
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003908682394479217,
+      "loss": 1.4052,
+      "theoretical_loss": 3.4555890826831934,
+      "tokens_seen": 2023030784
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003908281174771305,
+      "loss": 1.3306,
+      "theoretical_loss": 3.455570611638304,
+      "tokens_seen": 2023161856
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003907879955063393,
+      "loss": 1.413,
+      "theoretical_loss": 3.455552142125075,
+      "tokens_seen": 2023292928
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003907478735355481,
+      "loss": 1.3005,
+      "theoretical_loss": 3.4555336741432816,
+      "tokens_seen": 2023424000
+    },
+    {
+      "epoch": 0.23,
+      "objective/train/advantage_avg": 0.49416378140449524,
+      "objective/train/docs_used": 738179,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.317606806755066,
+      "objective/train/original_loss": 1.3176066875457764,
+      "objective/train/theoretical_loss": 3.455515207692697,
+      "objective/train/tokens_used": 394014176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24452967941761017,
+      "objective/train/value_max": -0.5,
+      "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": 1.3843556642532349,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.455515207692697,
+      "tokens_seen": 2023555072
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00039070775156475684,
+      "loss": 1.287,
+      "theoretical_loss": 3.455515207692697,
+      "tokens_seen": 2023555072
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003906676295939657,
+      "loss": 1.3576,
+      "theoretical_loss": 3.455496742773095,
+      "tokens_seen": 2023686144
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00039062750762317445,
+      "loss": 1.3764,
+      "theoretical_loss": 3.45547827938425,
+      "tokens_seen": 2023817216
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00039058738565238326,
+      "loss": 1.3515,
+      "theoretical_loss": 3.4554598175259352,
+      "tokens_seen": 2023948288
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00039054726368159206,
+      "loss": 1.4206,
+      "theoretical_loss": 3.455441357197925,
+      "tokens_seen": 2024079360
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003905071417108008,
+      "loss": 1.3664,
+      "theoretical_loss": 3.455422898399994,
+      "tokens_seen": 2024210432
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003904670197400096,
+      "loss": 1.3471,
+      "theoretical_loss": 3.4554044411319156,
+      "tokens_seen": 2024341504
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00039042689776921843,
+      "loss": 1.3008,
+      "theoretical_loss": 3.4553859853934643,
+      "tokens_seen": 2024472576
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00039038677579842723,
+      "loss": 1.3301,
+      "theoretical_loss": 3.4553675311844136,
+      "tokens_seen": 2024603648
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.000390346653827636,
+      "loss": 1.3662,
+      "theoretical_loss": 3.455349078504539,
+      "tokens_seen": 2024734720
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00039030653185684485,
+      "loss": 1.4866,
+      "theoretical_loss": 3.455330627353614,
+      "tokens_seen": 2024865792
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003902664098860536,
+      "loss": 1.4071,
+      "theoretical_loss": 3.455312177731413,
+      "tokens_seen": 2024996864
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003902262879152624,
+      "loss": 1.3984,
+      "theoretical_loss": 3.455293729637711,
+      "tokens_seen": 2025127936
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003901861659444712,
+      "loss": 1.3923,
+      "theoretical_loss": 3.4552752830722815,
+      "tokens_seen": 2025259008
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00039014604397367996,
+      "loss": 1.4041,
+      "theoretical_loss": 3.4552568380348996,
+      "tokens_seen": 2025390080
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003901059220028888,
+      "loss": 1.378,
+      "theoretical_loss": 3.45523839452534,
+      "tokens_seen": 2025521152
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003900658000320976,
+      "loss": 1.3783,
+      "theoretical_loss": 3.4552199525433767,
+      "tokens_seen": 2025652224
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003900256780613064,
+      "loss": 1.3744,
+      "theoretical_loss": 3.4552015120887845,
+      "tokens_seen": 2025783296
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003899855560905152,
+      "loss": 1.3358,
+      "theoretical_loss": 3.455183073161339,
+      "tokens_seen": 2025914368
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.000389945434119724,
+      "loss": 1.4461,
+      "theoretical_loss": 3.4551646357608137,
+      "tokens_seen": 2026045440
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038990531214893275,
+      "loss": 1.2929,
+      "theoretical_loss": 3.455146199886984,
+      "tokens_seen": 2026176512
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038986519017814155,
+      "loss": 1.4144,
+      "theoretical_loss": 3.4551277655396246,
+      "tokens_seen": 2026307584
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038982506820735036,
+      "loss": 1.4203,
+      "theoretical_loss": 3.4551093327185103,
+      "tokens_seen": 2026438656
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003897849462365591,
+      "loss": 1.393,
+      "theoretical_loss": 3.455090901423416,
+      "tokens_seen": 2026569728
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038974482426576797,
+      "loss": 1.3776,
+      "theoretical_loss": 3.4550724716541175,
+      "tokens_seen": 2026700800
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003897047022949767,
+      "loss": 1.3147,
+      "theoretical_loss": 3.4550540434103887,
+      "tokens_seen": 2026831872
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038966458032418553,
+      "loss": 1.4311,
+      "theoretical_loss": 3.455035616692005,
+      "tokens_seen": 2026962944
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038962445835339434,
+      "loss": 1.3945,
+      "theoretical_loss": 3.455017191498742,
+      "tokens_seen": 2027094016
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003895843363826031,
+      "loss": 1.2911,
+      "theoretical_loss": 3.454998767830374,
+      "tokens_seen": 2027225088
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038954421441181195,
+      "loss": 1.3344,
+      "theoretical_loss": 3.4549803456866774,
+      "tokens_seen": 2027356160
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003895040924410207,
+      "loss": 1.3709,
+      "theoretical_loss": 3.454961925067426,
+      "tokens_seen": 2027487232
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003894639704702295,
+      "loss": 1.3773,
+      "theoretical_loss": 3.4549435059723965,
+      "tokens_seen": 2027618304
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003894238484994383,
+      "loss": 1.3533,
+      "theoretical_loss": 3.4549250884013634,
+      "tokens_seen": 2027749376
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003893837265286471,
+      "loss": 1.404,
+      "theoretical_loss": 3.4549066723541024,
+      "tokens_seen": 2027880448
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003893436045578559,
+      "loss": 1.4163,
+      "theoretical_loss": 3.454888257830389,
+      "tokens_seen": 2028011520
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003893034825870647,
+      "loss": 1.4831,
+      "theoretical_loss": 3.4548698448299984,
+      "tokens_seen": 2028142592
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003892633606162735,
+      "loss": 1.4216,
+      "theoretical_loss": 3.4548514333527067,
+      "tokens_seen": 2028273664
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038922323864548224,
+      "loss": 1.3847,
+      "theoretical_loss": 3.454833023398289,
+      "tokens_seen": 2028404736
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003891831166746911,
+      "loss": 1.3806,
+      "theoretical_loss": 3.454814614966521,
+      "tokens_seen": 2028535808
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038914299470389985,
+      "loss": 1.3707,
+      "theoretical_loss": 3.4547962080571786,
+      "tokens_seen": 2028666880
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038910287273310866,
+      "loss": 1.4284,
+      "theoretical_loss": 3.4547778026700375,
+      "tokens_seen": 2028797952
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038906275076231746,
+      "loss": 1.3751,
+      "theoretical_loss": 3.4547593988048737,
+      "tokens_seen": 2028929024
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003890226287915262,
+      "loss": 1.3565,
+      "theoretical_loss": 3.454740996461462,
+      "tokens_seen": 2029060096
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003889825068207351,
+      "loss": 1.4578,
+      "theoretical_loss": 3.4547225956395797,
+      "tokens_seen": 2029191168
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038894238484994383,
+      "loss": 1.4077,
+      "theoretical_loss": 3.454704196339002,
+      "tokens_seen": 2029322240
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038890226287915263,
+      "loss": 1.396,
+      "theoretical_loss": 3.4546857985595048,
+      "tokens_seen": 2029453312
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038886214090836144,
+      "loss": 1.4124,
+      "theoretical_loss": 3.4546674023008643,
+      "tokens_seen": 2029584384
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038882201893757025,
+      "loss": 1.3258,
+      "theoretical_loss": 3.454649007562857,
+      "tokens_seen": 2029715456
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.000388781896966779,
+      "loss": 1.3516,
+      "theoretical_loss": 3.4546306143452576,
+      "tokens_seen": 2029846528
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003887417749959878,
+      "loss": 1.3912,
+      "theoretical_loss": 3.4546122226478437,
+      "tokens_seen": 2029977600
+    },
+    {
+      "epoch": 0.23,
+      "objective/train/advantage_avg": 0.4942469894886017,
+      "objective/train/docs_used": 740596,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2641459703445435,
+      "objective/train/original_loss": 1.2641457319259644,
+      "objective/train/theoretical_loss": 3.4545938324703913,
+      "objective/train/tokens_used": 400567776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2446666657924652,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506685972213745,
+      "objective/train/weighted_lm_loss": 1.3280733823776245,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4545938324703913,
+      "tokens_seen": 2030108672
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003887016530251966,
+      "loss": 1.3789,
+      "theoretical_loss": 3.4545938324703913,
+      "tokens_seen": 2030108672
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038866153105440536,
+      "loss": 1.3138,
+      "theoretical_loss": 3.4545754438126766,
+      "tokens_seen": 2030239744
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003886214090836142,
+      "loss": 1.404,
+      "theoretical_loss": 3.454557056674475,
+      "tokens_seen": 2030370816
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.000388581287112823,
+      "loss": 1.413,
+      "theoretical_loss": 3.454538671055564,
+      "tokens_seen": 2030501888
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003885411651420318,
+      "loss": 1.2894,
+      "theoretical_loss": 3.4545202869557197,
+      "tokens_seen": 2030632960
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003885010431712406,
+      "loss": 1.3559,
+      "theoretical_loss": 3.4545019043747183,
+      "tokens_seen": 2030764032
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038846092120044934,
+      "loss": 1.3636,
+      "theoretical_loss": 3.4544835233123363,
+      "tokens_seen": 2030895104
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038842079922965815,
+      "loss": 1.3617,
+      "theoretical_loss": 3.4544651437683505,
+      "tokens_seen": 2031026176
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038838067725886695,
+      "loss": 1.3754,
+      "theoretical_loss": 3.4544467657425377,
+      "tokens_seen": 2031157248
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038834055528807576,
+      "loss": 1.3279,
+      "theoretical_loss": 3.454428389234674,
+      "tokens_seen": 2031288320
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038830043331728457,
+      "loss": 1.4065,
+      "theoretical_loss": 3.4544100142445355,
+      "tokens_seen": 2031419392
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038826031134649337,
+      "loss": 1.4007,
+      "theoretical_loss": 3.4543916407719006,
+      "tokens_seen": 2031550464
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003882201893757021,
+      "loss": 1.4061,
+      "theoretical_loss": 3.4543732688165454,
+      "tokens_seen": 2031681536
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.000388180067404911,
+      "loss": 1.3779,
+      "theoretical_loss": 3.454354898378246,
+      "tokens_seen": 2031812608
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038813994543411974,
+      "loss": 1.3916,
+      "theoretical_loss": 3.4543365294567803,
+      "tokens_seen": 2031943680
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003880998234633285,
+      "loss": 1.3562,
+      "theoretical_loss": 3.454318162051924,
+      "tokens_seen": 2032074752
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038805970149253735,
+      "loss": 1.4809,
+      "theoretical_loss": 3.4542997961634554,
+      "tokens_seen": 2032205824
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003880195795217461,
+      "loss": 1.3901,
+      "theoretical_loss": 3.454281431791151,
+      "tokens_seen": 2032336896
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003879794575509549,
+      "loss": 1.3791,
+      "theoretical_loss": 3.4542630689347877,
+      "tokens_seen": 2032467968
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003879393355801637,
+      "loss": 1.3791,
+      "theoretical_loss": 3.4542447075941425,
+      "tokens_seen": 2032599040
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003878992136093725,
+      "loss": 1.4119,
+      "theoretical_loss": 3.454226347768993,
+      "tokens_seen": 2032730112
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003878590916385813,
+      "loss": 1.3098,
+      "theoretical_loss": 3.4542079894591162,
+      "tokens_seen": 2032861184
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003878189696677901,
+      "loss": 1.3985,
+      "theoretical_loss": 3.4541896326642894,
+      "tokens_seen": 2032992256
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003877788476969989,
+      "loss": 1.3817,
+      "theoretical_loss": 3.4541712773842894,
+      "tokens_seen": 2033123328
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038773872572620764,
+      "loss": 1.413,
+      "theoretical_loss": 3.454152923618895,
+      "tokens_seen": 2033254400
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003876986037554165,
+      "loss": 1.3788,
+      "theoretical_loss": 3.4541345713678817,
+      "tokens_seen": 2033385472
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038765848178462525,
+      "loss": 1.4175,
+      "theoretical_loss": 3.4541162206310276,
+      "tokens_seen": 2033516544
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003876183598138341,
+      "loss": 1.3806,
+      "theoretical_loss": 3.454097871408111,
+      "tokens_seen": 2033647616
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038757823784304286,
+      "loss": 1.3312,
+      "theoretical_loss": 3.4540795236989084,
+      "tokens_seen": 2033778688
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003875381158722516,
+      "loss": 1.4104,
+      "theoretical_loss": 3.454061177503198,
+      "tokens_seen": 2033909760
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003874979939014605,
+      "loss": 1.3548,
+      "theoretical_loss": 3.4540428328207575,
+      "tokens_seen": 2034040832
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038745787193066923,
+      "loss": 1.33,
+      "theoretical_loss": 3.4540244896513643,
+      "tokens_seen": 2034171904
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038741774995987803,
+      "loss": 1.4061,
+      "theoretical_loss": 3.4540061479947965,
+      "tokens_seen": 2034302976
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038737762798908684,
+      "loss": 1.35,
+      "theoretical_loss": 3.453987807850831,
+      "tokens_seen": 2034434048
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038733750601829565,
+      "loss": 1.3661,
+      "theoretical_loss": 3.4539694692192464,
+      "tokens_seen": 2034565120
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003872973840475044,
+      "loss": 1.3553,
+      "theoretical_loss": 3.45395113209982,
+      "tokens_seen": 2034696192
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003872572620767132,
+      "loss": 1.3338,
+      "theoretical_loss": 3.45393279649233,
+      "tokens_seen": 2034827264
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.000387217140105922,
+      "loss": 1.3228,
+      "theoretical_loss": 3.453914462396555,
+      "tokens_seen": 2034958336
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038717701813513076,
+      "loss": 1.435,
+      "theoretical_loss": 3.4538961298122715,
+      "tokens_seen": 2035089408
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003871368961643396,
+      "loss": 1.3159,
+      "theoretical_loss": 3.453877798739259,
+      "tokens_seen": 2035220480
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003870967741935484,
+      "loss": 1.3567,
+      "theoretical_loss": 3.4538594691772952,
+      "tokens_seen": 2035351552
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038705665222275724,
+      "loss": 1.3506,
+      "theoretical_loss": 3.453841141126158,
+      "tokens_seen": 2035482624
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.000387016530251966,
+      "loss": 1.3335,
+      "theoretical_loss": 3.453822814585626,
+      "tokens_seen": 2035613696
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038697640828117474,
+      "loss": 1.4086,
+      "theoretical_loss": 3.4538044895554765,
+      "tokens_seen": 2035744768
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003869362863103836,
+      "loss": 1.4138,
+      "theoretical_loss": 3.4537861660354885,
+      "tokens_seen": 2035875840
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038689616433959235,
+      "loss": 1.4291,
+      "theoretical_loss": 3.4537678440254407,
+      "tokens_seen": 2036006912
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038685604236880116,
+      "loss": 1.3207,
+      "theoretical_loss": 3.4537495235251106,
+      "tokens_seen": 2036137984
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038681592039800997,
+      "loss": 1.3734,
+      "theoretical_loss": 3.4537312045342774,
+      "tokens_seen": 2036269056
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038677579842721877,
+      "loss": 1.3884,
+      "theoretical_loss": 3.4537128870527187,
+      "tokens_seen": 2036400128
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003867356764564275,
+      "loss": 1.4184,
+      "theoretical_loss": 3.4536945710802143,
+      "tokens_seen": 2036531200
+    },
+    {
+      "epoch": 0.23,
+      "objective/train/advantage_avg": 0.49468788504600525,
+      "objective/train/docs_used": 742794,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.430019497871399,
+      "objective/train/original_loss": 1.4300193786621094,
+      "objective/train/theoretical_loss": 3.4536762566165415,
+      "objective/train/tokens_used": 407121376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2449086457490921,
+      "objective/train/value_max": -0.5,
+      "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": 1.5024975538253784,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4536762566165415,
+      "tokens_seen": 2036662272
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003866955544856364,
+      "loss": 1.4441,
+      "theoretical_loss": 3.4536762566165415,
+      "tokens_seen": 2036662272
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038665543251484514,
+      "loss": 1.4891,
+      "theoretical_loss": 3.4536579436614794,
+      "tokens_seen": 2036793344
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003866153105440539,
+      "loss": 1.361,
+      "theoretical_loss": 3.453639632214807,
+      "tokens_seen": 2036924416
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00038657518857326275,
+      "loss": 1.3706,
+      "theoretical_loss": 3.4536213222763026,
+      "tokens_seen": 2037055488
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003865350666024715,
+      "loss": 1.3928,
+      "theoretical_loss": 3.453603013845745,
+      "tokens_seen": 2037186560
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003864949446316803,
+      "loss": 1.3065,
+      "theoretical_loss": 3.4535847069229133,
+      "tokens_seen": 2037317632
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003864548226608891,
+      "loss": 1.3997,
+      "theoretical_loss": 3.4535664015075858,
+      "tokens_seen": 2037448704
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003864147006900979,
+      "loss": 1.4123,
+      "theoretical_loss": 3.4535480975995423,
+      "tokens_seen": 2037579776
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003863745787193067,
+      "loss": 1.447,
+      "theoretical_loss": 3.453529795198561,
+      "tokens_seen": 2037710848
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003863344567485155,
+      "loss": 1.2913,
+      "theoretical_loss": 3.4535114943044203,
+      "tokens_seen": 2037841920
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003862943347777243,
+      "loss": 1.3997,
+      "theoretical_loss": 3.4534931949169003,
+      "tokens_seen": 2037972992
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003862542128069331,
+      "loss": 1.3067,
+      "theoretical_loss": 3.45347489703578,
+      "tokens_seen": 2038104064
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003862140908361419,
+      "loss": 1.3846,
+      "theoretical_loss": 3.4534566006608385,
+      "tokens_seen": 2038235136
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038617396886535065,
+      "loss": 1.3305,
+      "theoretical_loss": 3.4534383057918543,
+      "tokens_seen": 2038366208
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003861338468945595,
+      "loss": 1.3465,
+      "theoretical_loss": 3.453420012428607,
+      "tokens_seen": 2038497280
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038609372492376826,
+      "loss": 1.4042,
+      "theoretical_loss": 3.4534017205708762,
+      "tokens_seen": 2038628352
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.000386053602952977,
+      "loss": 1.3341,
+      "theoretical_loss": 3.453383430218441,
+      "tokens_seen": 2038759424
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003860134809821859,
+      "loss": 1.447,
+      "theoretical_loss": 3.453365141371081,
+      "tokens_seen": 2038890496
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038597335901139463,
+      "loss": 1.3729,
+      "theoretical_loss": 3.4533468540285748,
+      "tokens_seen": 2039021568
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038593323704060343,
+      "loss": 1.3828,
+      "theoretical_loss": 3.4533285681907024,
+      "tokens_seen": 2039152640
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038589311506981224,
+      "loss": 1.4021,
+      "theoretical_loss": 3.4533102838572436,
+      "tokens_seen": 2039283712
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038585299309902105,
+      "loss": 1.4337,
+      "theoretical_loss": 3.453292001027977,
+      "tokens_seen": 2039414784
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038581287112822985,
+      "loss": 1.4145,
+      "theoretical_loss": 3.4532737197026835,
+      "tokens_seen": 2039545856
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003857727491574386,
+      "loss": 1.4081,
+      "theoretical_loss": 3.4532554398811417,
+      "tokens_seen": 2039676928
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003857326271866474,
+      "loss": 1.3878,
+      "theoretical_loss": 3.4532371615631323,
+      "tokens_seen": 2039808000
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003856925052158562,
+      "loss": 1.4292,
+      "theoretical_loss": 3.4532188847484337,
+      "tokens_seen": 2039939072
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.000385652383245065,
+      "loss": 1.3745,
+      "theoretical_loss": 3.4532006094368266,
+      "tokens_seen": 2040070144
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003856122612742738,
+      "loss": 1.3767,
+      "theoretical_loss": 3.4531823356280906,
+      "tokens_seen": 2040201216
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038557213930348264,
+      "loss": 1.4051,
+      "theoretical_loss": 3.4531640633220055,
+      "tokens_seen": 2040332288
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003855320173326914,
+      "loss": 1.2914,
+      "theoretical_loss": 3.4531457925183515,
+      "tokens_seen": 2040463360
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038549189536190014,
+      "loss": 1.4105,
+      "theoretical_loss": 3.453127523216908,
+      "tokens_seen": 2040594432
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.000385451773391109,
+      "loss": 1.4147,
+      "theoretical_loss": 3.4531092554174556,
+      "tokens_seen": 2040725504
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038541165142031775,
+      "loss": 1.4066,
+      "theoretical_loss": 3.4530909891197736,
+      "tokens_seen": 2040856576
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038537152944952656,
+      "loss": 1.4112,
+      "theoretical_loss": 3.453072724323643,
+      "tokens_seen": 2040987648
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038533140747873537,
+      "loss": 1.3117,
+      "theoretical_loss": 3.4530544610288434,
+      "tokens_seen": 2041118720
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038529128550794417,
+      "loss": 1.3289,
+      "theoretical_loss": 3.4530361992351555,
+      "tokens_seen": 2041249792
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003852511635371529,
+      "loss": 1.3752,
+      "theoretical_loss": 3.453017938942359,
+      "tokens_seen": 2041380864
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003852110415663618,
+      "loss": 1.3822,
+      "theoretical_loss": 3.4529996801502345,
+      "tokens_seen": 2041511936
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038517091959557054,
+      "loss": 1.4631,
+      "theoretical_loss": 3.452981422858562,
+      "tokens_seen": 2041643008
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038513079762477934,
+      "loss": 1.4639,
+      "theoretical_loss": 3.4529631670671224,
+      "tokens_seen": 2041774080
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038509067565398815,
+      "loss": 1.3889,
+      "theoretical_loss": 3.4529449127756955,
+      "tokens_seen": 2041905152
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003850505536831969,
+      "loss": 1.3933,
+      "theoretical_loss": 3.4529266599840627,
+      "tokens_seen": 2042036224
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038501043171240576,
+      "loss": 1.3792,
+      "theoretical_loss": 3.452908408692003,
+      "tokens_seen": 2042167296
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003849703097416145,
+      "loss": 1.3408,
+      "theoretical_loss": 3.4528901588992986,
+      "tokens_seen": 2042298368
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003849301877708233,
+      "loss": 1.4156,
+      "theoretical_loss": 3.4528719106057295,
+      "tokens_seen": 2042429440
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003848900658000321,
+      "loss": 1.4368,
+      "theoretical_loss": 3.4528536638110756,
+      "tokens_seen": 2042560512
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003848499438292409,
+      "loss": 1.3797,
+      "theoretical_loss": 3.452835418515119,
+      "tokens_seen": 2042691584
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003848098218584497,
+      "loss": 1.4053,
+      "theoretical_loss": 3.452817174717639,
+      "tokens_seen": 2042822656
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003847696998876585,
+      "loss": 1.4011,
+      "theoretical_loss": 3.4527989324184176,
+      "tokens_seen": 2042953728
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003847295779168673,
+      "loss": 1.4189,
+      "theoretical_loss": 3.452780691617235,
+      "tokens_seen": 2043084800
+    },
+    {
+      "epoch": 0.24,
+      "objective/train/advantage_avg": 0.4943661093711853,
+      "objective/train/docs_used": 745253,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.326207160949707,
+      "objective/train/original_loss": 1.326207160949707,
+      "objective/train/theoretical_loss": 3.4527624523138725,
+      "objective/train/tokens_used": 413674976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2446451038122177,
+      "objective/train/value_max": -0.5,
+      "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": 1.3933334350585938,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4527624523138725,
+      "tokens_seen": 2043215872
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038468945594607605,
+      "loss": 1.348,
+      "theoretical_loss": 3.4527624523138725,
+      "tokens_seen": 2043215872
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003846493339752849,
+      "loss": 1.3162,
+      "theoretical_loss": 3.452744214508111,
+      "tokens_seen": 2043346944
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038460921200449366,
+      "loss": 1.319,
+      "theoretical_loss": 3.45272597819973,
+      "tokens_seen": 2043478016
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003845690900337024,
+      "loss": 1.3674,
+      "theoretical_loss": 3.4527077433885127,
+      "tokens_seen": 2043609088
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003845289680629113,
+      "loss": 1.3144,
+      "theoretical_loss": 3.4526895100742396,
+      "tokens_seen": 2043740160
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038448884609212,
+      "loss": 1.3089,
+      "theoretical_loss": 3.452671278256691,
+      "tokens_seen": 2043871232
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003844487241213289,
+      "loss": 1.3472,
+      "theoretical_loss": 3.452653047935649,
+      "tokens_seen": 2044002304
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038440860215053764,
+      "loss": 1.3379,
+      "theoretical_loss": 3.4526348191108935,
+      "tokens_seen": 2044133376
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038436848017974645,
+      "loss": 1.3783,
+      "theoretical_loss": 3.452616591782207,
+      "tokens_seen": 2044264448
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038432835820895525,
+      "loss": 1.3407,
+      "theoretical_loss": 3.4525983659493713,
+      "tokens_seen": 2044395520
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.000384288236238164,
+      "loss": 1.2965,
+      "theoretical_loss": 3.4525801416121658,
+      "tokens_seen": 2044526592
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003842481142673728,
+      "loss": 1.4553,
+      "theoretical_loss": 3.4525619187703738,
+      "tokens_seen": 2044657664
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003842079922965816,
+      "loss": 1.4088,
+      "theoretical_loss": 3.452543697423775,
+      "tokens_seen": 2044788736
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003841678703257904,
+      "loss": 1.3538,
+      "theoretical_loss": 3.452525477572152,
+      "tokens_seen": 2044919808
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003841277483549992,
+      "loss": 1.3077,
+      "theoretical_loss": 3.452507259215287,
+      "tokens_seen": 2045050880
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038408762638420804,
+      "loss": 1.381,
+      "theoretical_loss": 3.4524890423529597,
+      "tokens_seen": 2045181952
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003840475044134168,
+      "loss": 1.3502,
+      "theoretical_loss": 3.4524708269849533,
+      "tokens_seen": 2045313024
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038400738244262554,
+      "loss": 1.3522,
+      "theoretical_loss": 3.4524526131110487,
+      "tokens_seen": 2045444096
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003839672604718344,
+      "loss": 1.3362,
+      "theoretical_loss": 3.452434400731028,
+      "tokens_seen": 2045575168
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038392713850104315,
+      "loss": 1.318,
+      "theoretical_loss": 3.452416189844673,
+      "tokens_seen": 2045706240
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.000383887016530252,
+      "loss": 1.325,
+      "theoretical_loss": 3.4523979804517646,
+      "tokens_seen": 2045837312
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038384689455946077,
+      "loss": 1.3612,
+      "theoretical_loss": 3.4523797725520855,
+      "tokens_seen": 2045968384
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038380677258866957,
+      "loss": 1.4422,
+      "theoretical_loss": 3.452361566145418,
+      "tokens_seen": 2046099456
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003837666506178784,
+      "loss": 1.4353,
+      "theoretical_loss": 3.452343361231543,
+      "tokens_seen": 2046230528
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038372652864708713,
+      "loss": 1.4041,
+      "theoretical_loss": 3.452325157810243,
+      "tokens_seen": 2046361600
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038368640667629594,
+      "loss": 1.3133,
+      "theoretical_loss": 3.4523069558813004,
+      "tokens_seen": 2046492672
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038364628470550474,
+      "loss": 1.3579,
+      "theoretical_loss": 3.4522887554444965,
+      "tokens_seen": 2046623744
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038360616273471355,
+      "loss": 1.3514,
+      "theoretical_loss": 3.4522705564996143,
+      "tokens_seen": 2046754816
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003835660407639223,
+      "loss": 1.4604,
+      "theoretical_loss": 3.4522523590464353,
+      "tokens_seen": 2046885888
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038352591879313116,
+      "loss": 1.3682,
+      "theoretical_loss": 3.452234163084742,
+      "tokens_seen": 2047016960
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003834857968223399,
+      "loss": 1.3398,
+      "theoretical_loss": 3.4522159686143166,
+      "tokens_seen": 2047148032
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003834456748515487,
+      "loss": 1.4151,
+      "theoretical_loss": 3.452197775634941,
+      "tokens_seen": 2047279104
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003834055528807575,
+      "loss": 1.3737,
+      "theoretical_loss": 3.4521795841463985,
+      "tokens_seen": 2047410176
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003833654309099663,
+      "loss": 1.3357,
+      "theoretical_loss": 3.4521613941484706,
+      "tokens_seen": 2047541248
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003833253089391751,
+      "loss": 1.3451,
+      "theoretical_loss": 3.4521432056409402,
+      "tokens_seen": 2047672320
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003832851869683839,
+      "loss": 1.3646,
+      "theoretical_loss": 3.4521250186235903,
+      "tokens_seen": 2047803392
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003832450649975927,
+      "loss": 1.4082,
+      "theoretical_loss": 3.4521068330962024,
+      "tokens_seen": 2047934464
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003832049430268015,
+      "loss": 1.3381,
+      "theoretical_loss": 3.4520886490585596,
+      "tokens_seen": 2048065536
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003831648210560103,
+      "loss": 1.4026,
+      "theoretical_loss": 3.4520704665104445,
+      "tokens_seen": 2048196608
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038312469908521906,
+      "loss": 1.4122,
+      "theoretical_loss": 3.4520522854516402,
+      "tokens_seen": 2048327680
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038308457711442787,
+      "loss": 1.3628,
+      "theoretical_loss": 3.4520341058819284,
+      "tokens_seen": 2048458752
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003830444551436367,
+      "loss": 1.3314,
+      "theoretical_loss": 3.4520159278010922,
+      "tokens_seen": 2048589824
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003830043331728454,
+      "loss": 1.375,
+      "theoretical_loss": 3.451997751208915,
+      "tokens_seen": 2048720896
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003829642112020543,
+      "loss": 1.3862,
+      "theoretical_loss": 3.4519795761051797,
+      "tokens_seen": 2048851968
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038292408923126304,
+      "loss": 1.4113,
+      "theoretical_loss": 3.4519614024896685,
+      "tokens_seen": 2048983040
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038288396726047185,
+      "loss": 1.3556,
+      "theoretical_loss": 3.451943230362165,
+      "tokens_seen": 2049114112
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038284384528968065,
+      "loss": 1.3412,
+      "theoretical_loss": 3.4519250597224516,
+      "tokens_seen": 2049245184
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003828037233188894,
+      "loss": 1.3857,
+      "theoretical_loss": 3.451906890570312,
+      "tokens_seen": 2049376256
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003827636013480982,
+      "loss": 1.4199,
+      "theoretical_loss": 3.451888722905528,
+      "tokens_seen": 2049507328
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.000382723479377307,
+      "loss": 1.3777,
+      "theoretical_loss": 3.4518705567278847,
+      "tokens_seen": 2049638400
+    },
+    {
+      "epoch": 0.24,
+      "objective/train/advantage_avg": 0.49186035990715027,
+      "objective/train/docs_used": 747702,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3505499362945557,
+      "objective/train/original_loss": 1.3505499362945557,
+      "objective/train/theoretical_loss": 3.4518523920371633,
+      "objective/train/tokens_used": 420228576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2424211949110031,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504183769226074,
+      "objective/train/weighted_lm_loss": 1.4186336994171143,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4518523920371633,
+      "tokens_seen": 2049769472
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003826833574065158,
+      "loss": 1.3756,
+      "theoretical_loss": 3.4518523920371633,
+      "tokens_seen": 2049769472
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003826432354357246,
+      "loss": 1.3473,
+      "theoretical_loss": 3.4518342288331487,
+      "tokens_seen": 2049900544
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038260311346493344,
+      "loss": 1.3437,
+      "theoretical_loss": 3.4518160671156233,
+      "tokens_seen": 2050031616
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003825629914941422,
+      "loss": 1.4386,
+      "theoretical_loss": 3.4517979068843703,
+      "tokens_seen": 2050162688
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.000382522869523351,
+      "loss": 1.486,
+      "theoretical_loss": 3.4517797481391734,
+      "tokens_seen": 2050293760
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003824827475525598,
+      "loss": 1.3603,
+      "theoretical_loss": 3.4517615908798156,
+      "tokens_seen": 2050424832
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038244262558176855,
+      "loss": 1.3367,
+      "theoretical_loss": 3.4517434351060814,
+      "tokens_seen": 2050555904
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003824025036109774,
+      "loss": 1.4174,
+      "theoretical_loss": 3.4517252808177528,
+      "tokens_seen": 2050686976
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038236238164018616,
+      "loss": 1.3629,
+      "theoretical_loss": 3.4517071280146148,
+      "tokens_seen": 2050818048
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038232225966939497,
+      "loss": 1.3238,
+      "theoretical_loss": 3.45168897669645,
+      "tokens_seen": 2050949120
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003822821376986038,
+      "loss": 1.436,
+      "theoretical_loss": 3.451670826863042,
+      "tokens_seen": 2051080192
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038224201572781253,
+      "loss": 1.3843,
+      "theoretical_loss": 3.451652678514175,
+      "tokens_seen": 2051211264
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038220189375702134,
+      "loss": 1.4162,
+      "theoretical_loss": 3.451634531649632,
+      "tokens_seen": 2051342336
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038216177178623014,
+      "loss": 1.403,
+      "theoretical_loss": 3.4516163862691984,
+      "tokens_seen": 2051473408
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038212164981543895,
+      "loss": 1.3609,
+      "theoretical_loss": 3.451598242372656,
+      "tokens_seen": 2051604480
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003820815278446477,
+      "loss": 1.3502,
+      "theoretical_loss": 3.45158009995979,
+      "tokens_seen": 2051735552
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038204140587385656,
+      "loss": 1.3531,
+      "theoretical_loss": 3.451561959030384,
+      "tokens_seen": 2051866624
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003820012839030653,
+      "loss": 1.3782,
+      "theoretical_loss": 3.4515438195842214,
+      "tokens_seen": 2051997696
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003819611619322742,
+      "loss": 1.4253,
+      "theoretical_loss": 3.451525681621087,
+      "tokens_seen": 2052128768
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003819210399614829,
+      "loss": 1.3497,
+      "theoretical_loss": 3.451507545140764,
+      "tokens_seen": 2052259840
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003818809179906917,
+      "loss": 1.3533,
+      "theoretical_loss": 3.4514894101430373,
+      "tokens_seen": 2052390912
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038184079601990054,
+      "loss": 1.3953,
+      "theoretical_loss": 3.4514712766276907,
+      "tokens_seen": 2052521984
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003818006740491093,
+      "loss": 1.391,
+      "theoretical_loss": 3.4514531445945082,
+      "tokens_seen": 2052653056
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003817605520783181,
+      "loss": 1.4366,
+      "theoretical_loss": 3.451435014043274,
+      "tokens_seen": 2052784128
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003817204301075269,
+      "loss": 1.4117,
+      "theoretical_loss": 3.451416884973773,
+      "tokens_seen": 2052915200
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003816803081367357,
+      "loss": 1.3924,
+      "theoretical_loss": 3.451398757385789,
+      "tokens_seen": 2053046272
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038164018616594446,
+      "loss": 1.4246,
+      "theoretical_loss": 3.4513806312791058,
+      "tokens_seen": 2053177344
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038160006419515327,
+      "loss": 1.4046,
+      "theoretical_loss": 3.451362506653509,
+      "tokens_seen": 2053308416
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003815599422243621,
+      "loss": 1.3986,
+      "theoretical_loss": 3.4513443835087823,
+      "tokens_seen": 2053439488
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003815198202535708,
+      "loss": 1.4288,
+      "theoretical_loss": 3.4513262618447103,
+      "tokens_seen": 2053570560
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003814796982827797,
+      "loss": 1.4119,
+      "theoretical_loss": 3.4513081416610776,
+      "tokens_seen": 2053701632
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038143957631198844,
+      "loss": 1.366,
+      "theoretical_loss": 3.451290022957669,
+      "tokens_seen": 2053832704
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038139945434119725,
+      "loss": 1.3799,
+      "theoretical_loss": 3.4512719057342682,
+      "tokens_seen": 2053963776
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00038135933237040605,
+      "loss": 1.3184,
+      "theoretical_loss": 3.4512537899906612,
+      "tokens_seen": 2054094848
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003813192103996148,
+      "loss": 1.3835,
+      "theoretical_loss": 3.451235675726632,
+      "tokens_seen": 2054225920
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00038127908842882366,
+      "loss": 1.4332,
+      "theoretical_loss": 3.4512175629419657,
+      "tokens_seen": 2054356992
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003812389664580324,
+      "loss": 1.3855,
+      "theoretical_loss": 3.4511994516364464,
+      "tokens_seen": 2054488064
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003811988444872412,
+      "loss": 1.3719,
+      "theoretical_loss": 3.4511813418098596,
+      "tokens_seen": 2054619136
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00038115872251645003,
+      "loss": 1.3735,
+      "theoretical_loss": 3.4511632334619895,
+      "tokens_seen": 2054750208
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00038111860054565884,
+      "loss": 1.5031,
+      "theoretical_loss": 3.451145126592622,
+      "tokens_seen": 2054881280
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003810784785748676,
+      "loss": 1.4197,
+      "theoretical_loss": 3.451127021201542,
+      "tokens_seen": 2055012352
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003810383566040764,
+      "loss": 1.43,
+      "theoretical_loss": 3.4511089172885336,
+      "tokens_seen": 2055143424
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003809982346332852,
+      "loss": 1.3393,
+      "theoretical_loss": 3.4510908148533828,
+      "tokens_seen": 2055274496
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00038095811266249395,
+      "loss": 1.3861,
+      "theoretical_loss": 3.4510727138958748,
+      "tokens_seen": 2055405568
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003809179906917028,
+      "loss": 1.3933,
+      "theoretical_loss": 3.4510546144157934,
+      "tokens_seen": 2055536640
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00038087786872091156,
+      "loss": 1.4031,
+      "theoretical_loss": 3.451036516412925,
+      "tokens_seen": 2055667712
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00038083774675012037,
+      "loss": 1.3964,
+      "theoretical_loss": 3.4510184198870553,
+      "tokens_seen": 2055798784
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003807976247793292,
+      "loss": 1.4143,
+      "theoretical_loss": 3.451000324837968,
+      "tokens_seen": 2055929856
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00038075750280853793,
+      "loss": 1.4366,
+      "theoretical_loss": 3.45098223126545,
+      "tokens_seen": 2056060928
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003807173808377468,
+      "loss": 1.3943,
+      "theoretical_loss": 3.450964139169286,
+      "tokens_seen": 2056192000
+    },
+    {
+      "epoch": 0.25,
+      "objective/train/advantage_avg": 0.4949285089969635,
+      "objective/train/docs_used": 750166,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.270849347114563,
+      "objective/train/original_loss": 1.270849347114563,
+      "objective/train/theoretical_loss": 3.450946048549261,
+      "objective/train/tokens_used": 426782176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24527603387832642,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0507398843765259,
+      "objective/train/weighted_lm_loss": 1.3353068828582764,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.450946048549261,
+      "tokens_seen": 2056323072
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00038067725886695554,
+      "loss": 1.3519,
+      "theoretical_loss": 3.450946048549261,
+      "tokens_seen": 2056323072
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00038063713689616435,
+      "loss": 1.4489,
+      "theoretical_loss": 3.4509279594051616,
+      "tokens_seen": 2056454144
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00038059701492537315,
+      "loss": 1.3482,
+      "theoretical_loss": 3.4509098717367723,
+      "tokens_seen": 2056585216
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00038055689295458196,
+      "loss": 1.3693,
+      "theoretical_loss": 3.450891785543879,
+      "tokens_seen": 2056716288
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003805167709837907,
+      "loss": 1.3312,
+      "theoretical_loss": 3.4508737008262678,
+      "tokens_seen": 2056847360
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003804766490129996,
+      "loss": 1.3994,
+      "theoretical_loss": 3.450855617583724,
+      "tokens_seen": 2056978432
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003804365270422083,
+      "loss": 1.3131,
+      "theoretical_loss": 3.4508375358160333,
+      "tokens_seen": 2057109504
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003803964050714171,
+      "loss": 1.4188,
+      "theoretical_loss": 3.4508194555229807,
+      "tokens_seen": 2057240576
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00038035628310062594,
+      "loss": 1.3829,
+      "theoretical_loss": 3.4508013767043533,
+      "tokens_seen": 2057371648
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003803161611298347,
+      "loss": 1.3612,
+      "theoretical_loss": 3.4507832993599363,
+      "tokens_seen": 2057502720
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003802760391590435,
+      "loss": 1.2719,
+      "theoretical_loss": 3.4507652234895154,
+      "tokens_seen": 2057633792
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003802359171882523,
+      "loss": 1.4181,
+      "theoretical_loss": 3.450747149092877,
+      "tokens_seen": 2057764864
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003801957952174611,
+      "loss": 1.2949,
+      "theoretical_loss": 3.4507290761698064,
+      "tokens_seen": 2057895936
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00038015567324666986,
+      "loss": 1.413,
+      "theoretical_loss": 3.450711004720091,
+      "tokens_seen": 2058027008
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00038011555127587867,
+      "loss": 1.3543,
+      "theoretical_loss": 3.450692934743515,
+      "tokens_seen": 2058158080
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003800754293050875,
+      "loss": 1.4405,
+      "theoretical_loss": 3.450674866239866,
+      "tokens_seen": 2058289152
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003800353073342963,
+      "loss": 1.3349,
+      "theoretical_loss": 3.450656799208929,
+      "tokens_seen": 2058420224
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003799951853635051,
+      "loss": 1.3508,
+      "theoretical_loss": 3.4506387336504916,
+      "tokens_seen": 2058551296
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037995506339271384,
+      "loss": 1.3681,
+      "theoretical_loss": 3.450620669564339,
+      "tokens_seen": 2058682368
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003799149414219227,
+      "loss": 1.3449,
+      "theoretical_loss": 3.450602606950257,
+      "tokens_seen": 2058813440
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037987481945113145,
+      "loss": 1.4607,
+      "theoretical_loss": 3.4505845458080335,
+      "tokens_seen": 2058944512
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003798346974803402,
+      "loss": 1.2986,
+      "theoretical_loss": 3.4505664861374536,
+      "tokens_seen": 2059075584
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037979457550954906,
+      "loss": 1.2589,
+      "theoretical_loss": 3.4505484279383043,
+      "tokens_seen": 2059206656
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003797544535387578,
+      "loss": 1.3798,
+      "theoretical_loss": 3.450530371210372,
+      "tokens_seen": 2059337728
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003797143315679666,
+      "loss": 1.3831,
+      "theoretical_loss": 3.4505123159534428,
+      "tokens_seen": 2059468800
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037967420959717543,
+      "loss": 1.4145,
+      "theoretical_loss": 3.450494262167304,
+      "tokens_seen": 2059599872
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037963408762638423,
+      "loss": 1.3193,
+      "theoretical_loss": 3.4504762098517414,
+      "tokens_seen": 2059730944
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.000379593965655593,
+      "loss": 1.4326,
+      "theoretical_loss": 3.4504581590065424,
+      "tokens_seen": 2059862016
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003795538436848018,
+      "loss": 1.4036,
+      "theoretical_loss": 3.4504401096314927,
+      "tokens_seen": 2059993088
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003795137217140106,
+      "loss": 1.3292,
+      "theoretical_loss": 3.45042206172638,
+      "tokens_seen": 2060124160
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037947359974321935,
+      "loss": 1.3461,
+      "theoretical_loss": 3.450404015290991,
+      "tokens_seen": 2060255232
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003794334777724282,
+      "loss": 1.3984,
+      "theoretical_loss": 3.4503859703251116,
+      "tokens_seen": 2060386304
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037939335580163696,
+      "loss": 1.4144,
+      "theoretical_loss": 3.45036792682853,
+      "tokens_seen": 2060517376
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003793532338308458,
+      "loss": 1.4025,
+      "theoretical_loss": 3.450349884801032,
+      "tokens_seen": 2060648448
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003793131118600546,
+      "loss": 1.3577,
+      "theoretical_loss": 3.450331844242405,
+      "tokens_seen": 2060779520
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037927298988926333,
+      "loss": 1.3726,
+      "theoretical_loss": 3.4503138051524367,
+      "tokens_seen": 2060910592
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003792328679184722,
+      "loss": 1.3413,
+      "theoretical_loss": 3.4502957675309123,
+      "tokens_seen": 2061041664
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037919274594768094,
+      "loss": 1.4223,
+      "theoretical_loss": 3.4502777313776205,
+      "tokens_seen": 2061172736
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037915262397688975,
+      "loss": 1.3076,
+      "theoretical_loss": 3.450259696692348,
+      "tokens_seen": 2061303808
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037911250200609855,
+      "loss": 1.3979,
+      "theoretical_loss": 3.4502416634748823,
+      "tokens_seen": 2061434880
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037907238003530736,
+      "loss": 1.3865,
+      "theoretical_loss": 3.4502236317250095,
+      "tokens_seen": 2061565952
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003790322580645161,
+      "loss": 1.2984,
+      "theoretical_loss": 3.4502056014425184,
+      "tokens_seen": 2061697024
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003789921360937249,
+      "loss": 1.3847,
+      "theoretical_loss": 3.450187572627195,
+      "tokens_seen": 2061828096
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003789520141229337,
+      "loss": 1.3993,
+      "theoretical_loss": 3.4501695452788272,
+      "tokens_seen": 2061959168
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003789118921521425,
+      "loss": 1.4208,
+      "theoretical_loss": 3.4501515193972025,
+      "tokens_seen": 2062090240
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037887177018135134,
+      "loss": 1.362,
+      "theoretical_loss": 3.4501334949821083,
+      "tokens_seen": 2062221312
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003788316482105601,
+      "loss": 1.3857,
+      "theoretical_loss": 3.450115472033332,
+      "tokens_seen": 2062352384
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037879152623976895,
+      "loss": 1.3833,
+      "theoretical_loss": 3.450097450550661,
+      "tokens_seen": 2062483456
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003787514042689777,
+      "loss": 1.3159,
+      "theoretical_loss": 3.450079430533883,
+      "tokens_seen": 2062614528
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003787112822981865,
+      "loss": 1.2923,
+      "theoretical_loss": 3.4500614119827855,
+      "tokens_seen": 2062745600
+    },
+    {
+      "epoch": 0.25,
+      "objective/train/advantage_avg": 0.4908670485019684,
+      "objective/train/docs_used": 752534,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.307652473449707,
+      "objective/train/original_loss": 1.307652473449707,
+      "objective/train/theoretical_loss": 3.4500433948971567,
+      "objective/train/tokens_used": 433335776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2418198585510254,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503159761428833,
+      "objective/train/weighted_lm_loss": 1.373425841331482,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4500433948971567,
+      "tokens_seen": 2062876672
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003786711603273953,
+      "loss": 1.3995,
+      "theoretical_loss": 3.4500433948971567,
+      "tokens_seen": 2062876672
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037863103835660407,
+      "loss": 1.344,
+      "theoretical_loss": 3.450025379276784,
+      "tokens_seen": 2063007744
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003785909163858129,
+      "loss": 1.4235,
+      "theoretical_loss": 3.4500073651214547,
+      "tokens_seen": 2063138816
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003785507944150217,
+      "loss": 1.3794,
+      "theoretical_loss": 3.449989352430957,
+      "tokens_seen": 2063269888
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003785106724442305,
+      "loss": 1.3863,
+      "theoretical_loss": 3.449971341205079,
+      "tokens_seen": 2063400960
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037847055047343924,
+      "loss": 1.4023,
+      "theoretical_loss": 3.4499533314436084,
+      "tokens_seen": 2063532032
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003784304285026481,
+      "loss": 1.3251,
+      "theoretical_loss": 3.4499353231463328,
+      "tokens_seen": 2063663104
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037839030653185685,
+      "loss": 1.3709,
+      "theoretical_loss": 3.44991731631304,
+      "tokens_seen": 2063794176
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003783501845610656,
+      "loss": 1.3317,
+      "theoretical_loss": 3.4498993109435188,
+      "tokens_seen": 2063925248
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037831006259027446,
+      "loss": 1.3473,
+      "theoretical_loss": 3.4498813070375567,
+      "tokens_seen": 2064056320
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003782699406194832,
+      "loss": 1.3516,
+      "theoretical_loss": 3.4498633045949423,
+      "tokens_seen": 2064187392
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.000378229818648692,
+      "loss": 1.4391,
+      "theoretical_loss": 3.4498453036154633,
+      "tokens_seen": 2064318464
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037818969667790083,
+      "loss": 1.3336,
+      "theoretical_loss": 3.449827304098908,
+      "tokens_seen": 2064449536
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037814957470710963,
+      "loss": 1.3674,
+      "theoretical_loss": 3.449809306045065,
+      "tokens_seen": 2064580608
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037810945273631844,
+      "loss": 1.4556,
+      "theoretical_loss": 3.4497913094537216,
+      "tokens_seen": 2064711680
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003780693307655272,
+      "loss": 1.4208,
+      "theoretical_loss": 3.449773314324667,
+      "tokens_seen": 2064842752
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.000378029208794736,
+      "loss": 1.4132,
+      "theoretical_loss": 3.44975532065769,
+      "tokens_seen": 2064973824
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003779890868239448,
+      "loss": 1.4003,
+      "theoretical_loss": 3.449737328452578,
+      "tokens_seen": 2065104896
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003779489648531536,
+      "loss": 1.4316,
+      "theoretical_loss": 3.4497193377091193,
+      "tokens_seen": 2065235968
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037790884288236236,
+      "loss": 1.3765,
+      "theoretical_loss": 3.4497013484271037,
+      "tokens_seen": 2065367040
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003778687209115712,
+      "loss": 1.3518,
+      "theoretical_loss": 3.449683360606319,
+      "tokens_seen": 2065498112
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037782859894078,
+      "loss": 1.3459,
+      "theoretical_loss": 3.4496653742465533,
+      "tokens_seen": 2065629184
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037778847696998873,
+      "loss": 1.3364,
+      "theoretical_loss": 3.4496473893475956,
+      "tokens_seen": 2065760256
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003777483549991976,
+      "loss": 1.4724,
+      "theoretical_loss": 3.449629405909235,
+      "tokens_seen": 2065891328
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037770823302840634,
+      "loss": 1.3406,
+      "theoretical_loss": 3.4496114239312603,
+      "tokens_seen": 2066022400
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037766811105761515,
+      "loss": 1.3254,
+      "theoretical_loss": 3.4495934434134594,
+      "tokens_seen": 2066153472
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037762798908682395,
+      "loss": 1.3056,
+      "theoretical_loss": 3.449575464355622,
+      "tokens_seen": 2066284544
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037758786711603276,
+      "loss": 1.3684,
+      "theoretical_loss": 3.449557486757536,
+      "tokens_seen": 2066415616
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003775477451452415,
+      "loss": 1.4452,
+      "theoretical_loss": 3.4495395106189912,
+      "tokens_seen": 2066546688
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003775076231744503,
+      "loss": 1.359,
+      "theoretical_loss": 3.449521535939776,
+      "tokens_seen": 2066677760
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003774675012036591,
+      "loss": 1.3093,
+      "theoretical_loss": 3.44950356271968,
+      "tokens_seen": 2066808832
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037742737923286793,
+      "loss": 1.3693,
+      "theoretical_loss": 3.449485590958492,
+      "tokens_seen": 2066939904
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037738725726207674,
+      "loss": 1.4192,
+      "theoretical_loss": 3.449467620656,
+      "tokens_seen": 2067070976
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003773471352912855,
+      "loss": 1.3579,
+      "theoretical_loss": 3.449449651811995,
+      "tokens_seen": 2067202048
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037730701332049435,
+      "loss": 1.3077,
+      "theoretical_loss": 3.4494316844262647,
+      "tokens_seen": 2067333120
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003772668913497031,
+      "loss": 1.3561,
+      "theoretical_loss": 3.4494137184985982,
+      "tokens_seen": 2067464192
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003772267693789119,
+      "loss": 1.399,
+      "theoretical_loss": 3.449395754028786,
+      "tokens_seen": 2067595264
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003771866474081207,
+      "loss": 1.3793,
+      "theoretical_loss": 3.449377791016617,
+      "tokens_seen": 2067726336
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037714652543732947,
+      "loss": 1.349,
+      "theoretical_loss": 3.44935982946188,
+      "tokens_seen": 2067857408
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003771064034665383,
+      "loss": 1.3748,
+      "theoretical_loss": 3.4493418693643645,
+      "tokens_seen": 2067988480
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003770662814957471,
+      "loss": 1.3702,
+      "theoretical_loss": 3.4493239107238605,
+      "tokens_seen": 2068119552
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003770261595249559,
+      "loss": 1.3778,
+      "theoretical_loss": 3.4493059535401565,
+      "tokens_seen": 2068250624
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037698603755416464,
+      "loss": 1.378,
+      "theoretical_loss": 3.449287997813043,
+      "tokens_seen": 2068381696
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003769459155833735,
+      "loss": 1.3698,
+      "theoretical_loss": 3.4492700435423087,
+      "tokens_seen": 2068512768
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037690579361258225,
+      "loss": 1.4543,
+      "theoretical_loss": 3.449252090727744,
+      "tokens_seen": 2068643840
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037686567164179106,
+      "loss": 1.3565,
+      "theoretical_loss": 3.4492341393691377,
+      "tokens_seen": 2068774912
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037682554967099986,
+      "loss": 1.3878,
+      "theoretical_loss": 3.4492161894662803,
+      "tokens_seen": 2068905984
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003767854277002086,
+      "loss": 1.4222,
+      "theoretical_loss": 3.449198241018961,
+      "tokens_seen": 2069037056
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003767453057294175,
+      "loss": 1.3917,
+      "theoretical_loss": 3.44918029402697,
+      "tokens_seen": 2069168128
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037670518375862623,
+      "loss": 1.3434,
+      "theoretical_loss": 3.4491623484900966,
+      "tokens_seen": 2069299200
+    },
+    {
+      "epoch": 0.25,
+      "objective/train/advantage_avg": 0.48946264386177063,
+      "objective/train/docs_used": 754612,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.323708176612854,
+      "objective/train/original_loss": 1.3237080574035645,
+      "objective/train/theoretical_loss": 3.4491444044081314,
+      "objective/train/tokens_used": 439889376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24038410186767578,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0501681566238403,
+      "objective/train/weighted_lm_loss": 1.3900035619735718,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.7788007855415344,
+      "theoretical_loss": 3.4491444044081314,
+      "tokens_seen": 2069430272
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037666506178783503,
+      "loss": 1.4361,
+      "theoretical_loss": 3.4491444044081314,
+      "tokens_seen": 2069430272
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037662493981704384,
+      "loss": 1.3439,
+      "theoretical_loss": 3.449126461780864,
+      "tokens_seen": 2069561344
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003765848178462526,
+      "loss": 1.4334,
+      "theoretical_loss": 3.4491085206080836,
+      "tokens_seen": 2069692416
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003765446958754614,
+      "loss": 1.3725,
+      "theoretical_loss": 3.449090580889581,
+      "tokens_seen": 2069823488
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003765045739046702,
+      "loss": 1.4147,
+      "theoretical_loss": 3.449072642625146,
+      "tokens_seen": 2069954560
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.000376464451933879,
+      "loss": 1.3668,
+      "theoretical_loss": 3.449054705814569,
+      "tokens_seen": 2070085632
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037642432996308776,
+      "loss": 1.4287,
+      "theoretical_loss": 3.4490367704576395,
+      "tokens_seen": 2070216704
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003763842079922966,
+      "loss": 1.3805,
+      "theoretical_loss": 3.449018836554148,
+      "tokens_seen": 2070347776
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003763440860215054,
+      "loss": 1.4046,
+      "theoretical_loss": 3.449000904103885,
+      "tokens_seen": 2070478848
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00037630396405071413,
+      "loss": 1.3651,
+      "theoretical_loss": 3.44898297310664,
+      "tokens_seen": 2070609920
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.000376263842079923,
+      "loss": 1.4082,
+      "theoretical_loss": 3.448965043562204,
+      "tokens_seen": 2070740992
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037622372010913174,
+      "loss": 1.3577,
+      "theoretical_loss": 3.448947115470367,
+      "tokens_seen": 2070872064
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003761835981383406,
+      "loss": 1.3273,
+      "theoretical_loss": 3.4489291888309204,
+      "tokens_seen": 2071003136
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037614347616754935,
+      "loss": 1.3974,
+      "theoretical_loss": 3.4489112636436525,
+      "tokens_seen": 2071134208
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037610335419675816,
+      "loss": 1.438,
+      "theoretical_loss": 3.4488933399083557,
+      "tokens_seen": 2071265280
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037606323222596697,
+      "loss": 1.4361,
+      "theoretical_loss": 3.44887541762482,
+      "tokens_seen": 2071396352
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003760231102551757,
+      "loss": 1.3103,
+      "theoretical_loss": 3.4488574967928356,
+      "tokens_seen": 2071527424
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003759829882843845,
+      "loss": 1.4086,
+      "theoretical_loss": 3.448839577412193,
+      "tokens_seen": 2071658496
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037594286631359333,
+      "loss": 1.3168,
+      "theoretical_loss": 3.4488216594826833,
+      "tokens_seen": 2071789568
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037590274434280214,
+      "loss": 1.3339,
+      "theoretical_loss": 3.448803743004097,
+      "tokens_seen": 2071920640
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003758626223720109,
+      "loss": 1.3863,
+      "theoretical_loss": 3.4487858279762253,
+      "tokens_seen": 2072051712
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037582250040121975,
+      "loss": 1.3357,
+      "theoretical_loss": 3.448767914398858,
+      "tokens_seen": 2072182784
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003757823784304285,
+      "loss": 1.2986,
+      "theoretical_loss": 3.4487500022717867,
+      "tokens_seen": 2072313856
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003757422564596373,
+      "loss": 1.4384,
+      "theoretical_loss": 3.4487320915948017,
+      "tokens_seen": 2072444928
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003757021344888461,
+      "loss": 1.4179,
+      "theoretical_loss": 3.4487141823676946,
+      "tokens_seen": 2072576000
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037566201251805487,
+      "loss": 1.3232,
+      "theoretical_loss": 3.448696274590256,
+      "tokens_seen": 2072707072
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003756218905472637,
+      "loss": 1.365,
+      "theoretical_loss": 3.448678368262277,
+      "tokens_seen": 2072838144
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003755817685764725,
+      "loss": 1.4693,
+      "theoretical_loss": 3.448660463383548,
+      "tokens_seen": 2072969216
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003755416466056813,
+      "loss": 1.3891,
+      "theoretical_loss": 3.4486425599538606,
+      "tokens_seen": 2073100288
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003755015246348901,
+      "loss": 1.4035,
+      "theoretical_loss": 3.4486246579730064,
+      "tokens_seen": 2073231360
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003754614026640989,
+      "loss": 1.4246,
+      "theoretical_loss": 3.4486067574407757,
+      "tokens_seen": 2073362432
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037542128069330765,
+      "loss": 1.3226,
+      "theoretical_loss": 3.4485888583569606,
+      "tokens_seen": 2073493504
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037538115872251646,
+      "loss": 1.3154,
+      "theoretical_loss": 3.4485709607213515,
+      "tokens_seen": 2073624576
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037534103675172526,
+      "loss": 1.303,
+      "theoretical_loss": 3.4485530645337397,
+      "tokens_seen": 2073755648
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.000375300914780934,
+      "loss": 1.3634,
+      "theoretical_loss": 3.4485351697939173,
+      "tokens_seen": 2073886720
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003752607928101429,
+      "loss": 1.3311,
+      "theoretical_loss": 3.4485172765016756,
+      "tokens_seen": 2074017792
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037522067083935163,
+      "loss": 1.2882,
+      "theoretical_loss": 3.448499384656805,
+      "tokens_seen": 2074148864
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037518054886856043,
+      "loss": 1.3745,
+      "theoretical_loss": 3.4484814942590987,
+      "tokens_seen": 2074279936
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037514042689776924,
+      "loss": 1.4163,
+      "theoretical_loss": 3.448463605308346,
+      "tokens_seen": 2074411008
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.000375100304926978,
+      "loss": 1.4346,
+      "theoretical_loss": 3.4484457178043404,
+      "tokens_seen": 2074542080
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003750601829561868,
+      "loss": 1.4193,
+      "theoretical_loss": 3.4484278317468724,
+      "tokens_seen": 2074673152
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003750200609853956,
+      "loss": 1.2721,
+      "theoretical_loss": 3.448409947135734,
+      "tokens_seen": 2074804224
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003749799390146044,
+      "loss": 1.4195,
+      "theoretical_loss": 3.4483920639707173,
+      "tokens_seen": 2074935296
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003749398170438132,
+      "loss": 1.3096,
+      "theoretical_loss": 3.448374182251613,
+      "tokens_seen": 2075066368
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.000374899695073022,
+      "loss": 1.3203,
+      "theoretical_loss": 3.448356301978214,
+      "tokens_seen": 2075197440
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003748595731022308,
+      "loss": 1.3872,
+      "theoretical_loss": 3.4483384231503114,
+      "tokens_seen": 2075328512
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003748194511314396,
+      "loss": 1.3445,
+      "theoretical_loss": 3.4483205457676975,
+      "tokens_seen": 2075459584
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003747793291606484,
+      "loss": 1.3573,
+      "theoretical_loss": 3.4483026698301638,
+      "tokens_seen": 2075590656
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037473920718985714,
+      "loss": 1.3565,
+      "theoretical_loss": 3.4482847953375018,
+      "tokens_seen": 2075721728
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.000374699085219066,
+      "loss": 1.3366,
+      "theoretical_loss": 3.4482669222895046,
+      "tokens_seen": 2075852800
+    },
+    {
+      "epoch": 0.26,
+      "objective/train/advantage_avg": 0.4904538094997406,
+      "objective/train/docs_used": 756913,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.1641877889633179,
+      "objective/train/original_loss": 1.1641876697540283,
+      "objective/train/theoretical_loss": 3.4482490506859635,
+      "objective/train/tokens_used": 446442976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24124419689178467,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.05027174949646,
+      "objective/train/weighted_lm_loss": 1.222683072090149,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9048373699188232,
+      "theoretical_loss": 3.4482490506859635,
+      "tokens_seen": 2075983872
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037465896324827475,
+      "loss": 1.2813,
+      "theoretical_loss": 3.4482490506859635,
+      "tokens_seen": 2075983872
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037461884127748356,
+      "loss": 1.4046,
+      "theoretical_loss": 3.4482311805266717,
+      "tokens_seen": 2076114944
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037457871930669237,
+      "loss": 1.3595,
+      "theoretical_loss": 3.4482133118114193,
+      "tokens_seen": 2076246016
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003745385973359011,
+      "loss": 1.4271,
+      "theoretical_loss": 3.44819544454,
+      "tokens_seen": 2076377088
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003744984753651099,
+      "loss": 1.3263,
+      "theoretical_loss": 3.4481775787122055,
+      "tokens_seen": 2076508160
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037445835339431873,
+      "loss": 1.3348,
+      "theoretical_loss": 3.4481597143278284,
+      "tokens_seen": 2076639232
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037441823142352754,
+      "loss": 1.3952,
+      "theoretical_loss": 3.4481418513866604,
+      "tokens_seen": 2076770304
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003743781094527363,
+      "loss": 1.3516,
+      "theoretical_loss": 3.4481239898884946,
+      "tokens_seen": 2076901376
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037433798748194515,
+      "loss": 1.3765,
+      "theoretical_loss": 3.448106129833122,
+      "tokens_seen": 2077032448
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003742978655111539,
+      "loss": 1.339,
+      "theoretical_loss": 3.4480882712203367,
+      "tokens_seen": 2077163520
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003742577435403627,
+      "loss": 1.3035,
+      "theoretical_loss": 3.4480704140499303,
+      "tokens_seen": 2077294592
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003742176215695715,
+      "loss": 1.3545,
+      "theoretical_loss": 3.448052558321695,
+      "tokens_seen": 2077425664
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037417749959878027,
+      "loss": 1.3283,
+      "theoretical_loss": 3.4480347040354244,
+      "tokens_seen": 2077556736
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003741373776279891,
+      "loss": 1.3784,
+      "theoretical_loss": 3.44801685119091,
+      "tokens_seen": 2077687808
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003740972556571979,
+      "loss": 1.333,
+      "theoretical_loss": 3.447998999787945,
+      "tokens_seen": 2077818880
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003740571336864067,
+      "loss": 1.343,
+      "theoretical_loss": 3.447981149826322,
+      "tokens_seen": 2077949952
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003740170117156155,
+      "loss": 1.3717,
+      "theoretical_loss": 3.4479633013058337,
+      "tokens_seen": 2078081024
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003739768897448243,
+      "loss": 1.4078,
+      "theoretical_loss": 3.447945454226273,
+      "tokens_seen": 2078212096
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037393676777403305,
+      "loss": 1.364,
+      "theoretical_loss": 3.447927608587432,
+      "tokens_seen": 2078343168
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037389664580324186,
+      "loss": 1.3688,
+      "theoretical_loss": 3.4479097643891046,
+      "tokens_seen": 2078474240
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037385652383245066,
+      "loss": 1.3266,
+      "theoretical_loss": 3.447891921631083,
+      "tokens_seen": 2078605312
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003738164018616594,
+      "loss": 1.3356,
+      "theoretical_loss": 3.4478740803131602,
+      "tokens_seen": 2078736384
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003737762798908683,
+      "loss": 1.3168,
+      "theoretical_loss": 3.4478562404351294,
+      "tokens_seen": 2078867456
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.000373736157920077,
+      "loss": 1.3808,
+      "theoretical_loss": 3.4478384019967834,
+      "tokens_seen": 2078998528
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003736960359492859,
+      "loss": 1.3382,
+      "theoretical_loss": 3.4478205649979157,
+      "tokens_seen": 2079129600
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037365591397849464,
+      "loss": 1.4276,
+      "theoretical_loss": 3.447802729438319,
+      "tokens_seen": 2079260672
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003736157920077034,
+      "loss": 1.3743,
+      "theoretical_loss": 3.447784895317786,
+      "tokens_seen": 2079391744
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037357567003691225,
+      "loss": 1.3894,
+      "theoretical_loss": 3.447767062636111,
+      "tokens_seen": 2079522816
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.000373535548066121,
+      "loss": 1.3967,
+      "theoretical_loss": 3.4477492313930864,
+      "tokens_seen": 2079653888
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003734954260953298,
+      "loss": 1.4028,
+      "theoretical_loss": 3.4477314015885057,
+      "tokens_seen": 2079784960
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003734553041245386,
+      "loss": 1.4157,
+      "theoretical_loss": 3.4477135732221624,
+      "tokens_seen": 2079916032
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003734151821537474,
+      "loss": 1.3805,
+      "theoretical_loss": 3.4476957462938493,
+      "tokens_seen": 2080047104
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003733750601829562,
+      "loss": 1.3644,
+      "theoretical_loss": 3.4476779208033608,
+      "tokens_seen": 2080178176
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.000373334938212165,
+      "loss": 1.3788,
+      "theoretical_loss": 3.4476600967504893,
+      "tokens_seen": 2080309248
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003732948162413738,
+      "loss": 1.3956,
+      "theoretical_loss": 3.447642274135029,
+      "tokens_seen": 2080440320
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037325469427058254,
+      "loss": 1.3116,
+      "theoretical_loss": 3.447624452956773,
+      "tokens_seen": 2080571392
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003732145722997914,
+      "loss": 1.4106,
+      "theoretical_loss": 3.447606633215515,
+      "tokens_seen": 2080702464
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037317445032900015,
+      "loss": 1.3269,
+      "theoretical_loss": 3.447588814911049,
+      "tokens_seen": 2080833536
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037313432835820896,
+      "loss": 1.3279,
+      "theoretical_loss": 3.447570998043168,
+      "tokens_seen": 2080964608
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037309420638741777,
+      "loss": 1.3258,
+      "theoretical_loss": 3.4475531826116663,
+      "tokens_seen": 2081095680
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003730540844166265,
+      "loss": 1.4452,
+      "theoretical_loss": 3.447535368616337,
+      "tokens_seen": 2081226752
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003730139624458354,
+      "loss": 1.3837,
+      "theoretical_loss": 3.447517556056974,
+      "tokens_seen": 2081357824
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037297384047504413,
+      "loss": 1.4194,
+      "theoretical_loss": 3.4474997449333724,
+      "tokens_seen": 2081488896
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037293371850425294,
+      "loss": 1.3379,
+      "theoretical_loss": 3.4474819352453245,
+      "tokens_seen": 2081619968
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037289359653346174,
+      "loss": 1.3133,
+      "theoretical_loss": 3.4474641269926245,
+      "tokens_seen": 2081751040
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037285347456267055,
+      "loss": 1.3584,
+      "theoretical_loss": 3.447446320175067,
+      "tokens_seen": 2081882112
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003728133525918793,
+      "loss": 1.3551,
+      "theoretical_loss": 3.4474285147924455,
+      "tokens_seen": 2082013184
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003727732306210881,
+      "loss": 1.4128,
+      "theoretical_loss": 3.4474107108445544,
+      "tokens_seen": 2082144256
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003727331086502969,
+      "loss": 1.373,
+      "theoretical_loss": 3.4473929083311874,
+      "tokens_seen": 2082275328
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037269298667950567,
+      "loss": 1.4211,
+      "theoretical_loss": 3.447375107252139,
+      "tokens_seen": 2082406400
+    },
+    {
+      "epoch": 0.26,
+      "objective/train/advantage_avg": 0.48736584186553955,
+      "objective/train/docs_used": 759184,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2452672719955444,
+      "objective/train/original_loss": 1.245267391204834,
+      "objective/train/theoretical_loss": 3.447357307607203,
+      "objective/train/tokens_used": 452996576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2386443316936493,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0499496459960938,
+      "objective/train/weighted_lm_loss": 1.3076441287994385,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.447357307607203,
+      "tokens_seen": 2082537472
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003726528647087145,
+      "loss": 1.3253,
+      "theoretical_loss": 3.447357307607203,
+      "tokens_seen": 2082537472
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003726127427379233,
+      "loss": 1.3722,
+      "theoretical_loss": 3.4473395093961736,
+      "tokens_seen": 2082668544
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003725726207671321,
+      "loss": 1.4096,
+      "theoretical_loss": 3.4473217126188453,
+      "tokens_seen": 2082799616
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003725324987963409,
+      "loss": 1.355,
+      "theoretical_loss": 3.447303917275012,
+      "tokens_seen": 2082930688
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003724923768255497,
+      "loss": 1.3073,
+      "theoretical_loss": 3.447286123364469,
+      "tokens_seen": 2083061760
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003724522548547585,
+      "loss": 1.4073,
+      "theoretical_loss": 3.44726833088701,
+      "tokens_seen": 2083192832
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037241213288396726,
+      "loss": 1.4508,
+      "theoretical_loss": 3.4472505398424294,
+      "tokens_seen": 2083323904
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037237201091317606,
+      "loss": 1.3687,
+      "theoretical_loss": 3.447232750230522,
+      "tokens_seen": 2083454976
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037233188894238487,
+      "loss": 1.3777,
+      "theoretical_loss": 3.4472149620510812,
+      "tokens_seen": 2083586048
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003722917669715937,
+      "loss": 1.3726,
+      "theoretical_loss": 3.447197175303903,
+      "tokens_seen": 2083717120
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003722516450008024,
+      "loss": 1.3765,
+      "theoretical_loss": 3.447179389988781,
+      "tokens_seen": 2083848192
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003722115230300113,
+      "loss": 1.4252,
+      "theoretical_loss": 3.447161606105511,
+      "tokens_seen": 2083979264
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037217140105922004,
+      "loss": 1.3253,
+      "theoretical_loss": 3.4471438236538865,
+      "tokens_seen": 2084110336
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003721312790884288,
+      "loss": 1.3729,
+      "theoretical_loss": 3.4471260426337027,
+      "tokens_seen": 2084241408
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037209115711763765,
+      "loss": 1.3584,
+      "theoretical_loss": 3.447108263044754,
+      "tokens_seen": 2084372480
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003720510351468464,
+      "loss": 1.3656,
+      "theoretical_loss": 3.447090484886836,
+      "tokens_seen": 2084503552
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003720109131760552,
+      "loss": 1.4463,
+      "theoretical_loss": 3.4470727081597428,
+      "tokens_seen": 2084634624
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.000371970791205264,
+      "loss": 1.3142,
+      "theoretical_loss": 3.44705493286327,
+      "tokens_seen": 2084765696
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003719306692344728,
+      "loss": 1.4269,
+      "theoretical_loss": 3.447037158997211,
+      "tokens_seen": 2084896768
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003718905472636816,
+      "loss": 1.3641,
+      "theoretical_loss": 3.4470193865613625,
+      "tokens_seen": 2085027840
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003718504252928904,
+      "loss": 1.3366,
+      "theoretical_loss": 3.4470016155555188,
+      "tokens_seen": 2085158912
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003718103033220992,
+      "loss": 1.3693,
+      "theoretical_loss": 3.4469838459794753,
+      "tokens_seen": 2085289984
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.000371770181351308,
+      "loss": 1.3091,
+      "theoretical_loss": 3.446966077833027,
+      "tokens_seen": 2085421056
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003717300593805168,
+      "loss": 1.3522,
+      "theoretical_loss": 3.4469483111159684,
+      "tokens_seen": 2085552128
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037168993740972555,
+      "loss": 1.4513,
+      "theoretical_loss": 3.4469305458280957,
+      "tokens_seen": 2085683200
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003716498154389344,
+      "loss": 1.4282,
+      "theoretical_loss": 3.4469127819692034,
+      "tokens_seen": 2085814272
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037160969346814316,
+      "loss": 1.3218,
+      "theoretical_loss": 3.4468950195390864,
+      "tokens_seen": 2085945344
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003715695714973519,
+      "loss": 1.4172,
+      "theoretical_loss": 3.4468772585375413,
+      "tokens_seen": 2086076416
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003715294495265608,
+      "loss": 1.3988,
+      "theoretical_loss": 3.4468594989643626,
+      "tokens_seen": 2086207488
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037148932755576953,
+      "loss": 1.4579,
+      "theoretical_loss": 3.4468417408193455,
+      "tokens_seen": 2086338560
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037144920558497834,
+      "loss": 1.4484,
+      "theoretical_loss": 3.446823984102286,
+      "tokens_seen": 2086469632
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037140908361418714,
+      "loss": 1.3238,
+      "theoretical_loss": 3.4468062288129797,
+      "tokens_seen": 2086600704
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00037136896164339595,
+      "loss": 1.3069,
+      "theoretical_loss": 3.446788474951221,
+      "tokens_seen": 2086731776
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003713288396726047,
+      "loss": 1.416,
+      "theoretical_loss": 3.446770722516807,
+      "tokens_seen": 2086862848
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003712887177018135,
+      "loss": 1.4173,
+      "theoretical_loss": 3.4467529715095324,
+      "tokens_seen": 2086993920
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003712485957310223,
+      "loss": 1.3349,
+      "theoretical_loss": 3.446735221929193,
+      "tokens_seen": 2087124992
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00037120847376023107,
+      "loss": 1.3374,
+      "theoretical_loss": 3.4467174737755846,
+      "tokens_seen": 2087256064
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003711683517894399,
+      "loss": 1.3334,
+      "theoretical_loss": 3.446699727048503,
+      "tokens_seen": 2087387136
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003711282298186487,
+      "loss": 1.356,
+      "theoretical_loss": 3.4466819817477434,
+      "tokens_seen": 2087518208
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00037108810784785754,
+      "loss": 1.4179,
+      "theoretical_loss": 3.4466642378731023,
+      "tokens_seen": 2087649280
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003710479858770663,
+      "loss": 1.3957,
+      "theoretical_loss": 3.446646495424375,
+      "tokens_seen": 2087780352
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003710078639062751,
+      "loss": 1.3509,
+      "theoretical_loss": 3.4466287544013583,
+      "tokens_seen": 2087911424
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003709677419354839,
+      "loss": 1.3498,
+      "theoretical_loss": 3.446611014803847,
+      "tokens_seen": 2088042496
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00037092761996469266,
+      "loss": 1.3667,
+      "theoretical_loss": 3.4465932766316376,
+      "tokens_seen": 2088173568
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00037088749799390146,
+      "loss": 1.3619,
+      "theoretical_loss": 3.4465755398845266,
+      "tokens_seen": 2088304640
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00037084737602311027,
+      "loss": 1.3372,
+      "theoretical_loss": 3.4465578045623095,
+      "tokens_seen": 2088435712
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003708072540523191,
+      "loss": 1.331,
+      "theoretical_loss": 3.4465400706647826,
+      "tokens_seen": 2088566784
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003707671320815278,
+      "loss": 1.3775,
+      "theoretical_loss": 3.446522338191742,
+      "tokens_seen": 2088697856
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003707270101107367,
+      "loss": 1.3511,
+      "theoretical_loss": 3.446504607142984,
+      "tokens_seen": 2088828928
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00037068688813994544,
+      "loss": 1.4151,
+      "theoretical_loss": 3.4464868775183044,
+      "tokens_seen": 2088960000
+    },
+    {
+      "epoch": 0.27,
+      "objective/train/advantage_avg": 0.4929499924182892,
+      "objective/train/docs_used": 761616,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2782905101776123,
+      "objective/train/original_loss": 1.2782905101776123,
+      "objective/train/theoretical_loss": 3.4464691493175,
+      "objective/train/tokens_used": 459550176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24340718984603882,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505324602127075,
+      "objective/train/weighted_lm_loss": 1.342782974243164,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4464691493175,
+      "tokens_seen": 2089091072
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003706467661691542,
+      "loss": 1.3635,
+      "theoretical_loss": 3.4464691493175,
+      "tokens_seen": 2089091072
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00037060664419836305,
+      "loss": 1.3394,
+      "theoretical_loss": 3.446451422540367,
+      "tokens_seen": 2089222144
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003705665222275718,
+      "loss": 1.4074,
+      "theoretical_loss": 3.446433697186702,
+      "tokens_seen": 2089353216
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00037052640025678066,
+      "loss": 1.3889,
+      "theoretical_loss": 3.4464159732563004,
+      "tokens_seen": 2089484288
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003704862782859894,
+      "loss": 1.365,
+      "theoretical_loss": 3.44639825074896,
+      "tokens_seen": 2089615360
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003704461563151982,
+      "loss": 1.3081,
+      "theoretical_loss": 3.4463805296644767,
+      "tokens_seen": 2089746432
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00037040603434440703,
+      "loss": 1.3492,
+      "theoretical_loss": 3.446362810002647,
+      "tokens_seen": 2089877504
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003703659123736158,
+      "loss": 1.3775,
+      "theoretical_loss": 3.446345091763267,
+      "tokens_seen": 2090008576
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003703257904028246,
+      "loss": 1.3254,
+      "theoretical_loss": 3.446327374946134,
+      "tokens_seen": 2090139648
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003702856684320334,
+      "loss": 1.3578,
+      "theoretical_loss": 3.4463096595510443,
+      "tokens_seen": 2090270720
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003702455464612422,
+      "loss": 1.3596,
+      "theoretical_loss": 3.446291945577795,
+      "tokens_seen": 2090401792
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00037020542449045095,
+      "loss": 1.3916,
+      "theoretical_loss": 3.4462742330261826,
+      "tokens_seen": 2090532864
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003701653025196598,
+      "loss": 1.3077,
+      "theoretical_loss": 3.4462565218960037,
+      "tokens_seen": 2090663936
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00037012518054886856,
+      "loss": 1.384,
+      "theoretical_loss": 3.4462388121870555,
+      "tokens_seen": 2090795008
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003700850585780773,
+      "loss": 1.3948,
+      "theoretical_loss": 3.4462211038991346,
+      "tokens_seen": 2090926080
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003700449366072862,
+      "loss": 1.3809,
+      "theoretical_loss": 3.446203397032038,
+      "tokens_seen": 2091057152
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00037000481463649493,
+      "loss": 1.4145,
+      "theoretical_loss": 3.4461856915855624,
+      "tokens_seen": 2091188224
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036996469266570374,
+      "loss": 1.3987,
+      "theoretical_loss": 3.4461679875595053,
+      "tokens_seen": 2091319296
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036992457069491254,
+      "loss": 1.3957,
+      "theoretical_loss": 3.446150284953663,
+      "tokens_seen": 2091450368
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036988444872412135,
+      "loss": 1.4788,
+      "theoretical_loss": 3.4461325837678336,
+      "tokens_seen": 2091581440
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036984432675333015,
+      "loss": 1.3023,
+      "theoretical_loss": 3.4461148840018136,
+      "tokens_seen": 2091712512
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003698042047825389,
+      "loss": 1.3952,
+      "theoretical_loss": 3.4460971856554,
+      "tokens_seen": 2091843584
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003697640828117477,
+      "loss": 1.3973,
+      "theoretical_loss": 3.4460794887283903,
+      "tokens_seen": 2091974656
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003697239608409565,
+      "loss": 1.3646,
+      "theoretical_loss": 3.446061793220581,
+      "tokens_seen": 2092105728
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003696838388701653,
+      "loss": 1.3865,
+      "theoretical_loss": 3.446044099131771,
+      "tokens_seen": 2092236800
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003696437168993741,
+      "loss": 1.399,
+      "theoretical_loss": 3.446026406461756,
+      "tokens_seen": 2092367872
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036960359492858294,
+      "loss": 1.3712,
+      "theoretical_loss": 3.4460087152103345,
+      "tokens_seen": 2092498944
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003695634729577917,
+      "loss": 1.3498,
+      "theoretical_loss": 3.445991025377303,
+      "tokens_seen": 2092630016
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036952335098700044,
+      "loss": 1.4004,
+      "theoretical_loss": 3.4459733369624597,
+      "tokens_seen": 2092761088
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003694832290162093,
+      "loss": 1.3995,
+      "theoretical_loss": 3.445955649965602,
+      "tokens_seen": 2092892160
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036944310704541805,
+      "loss": 1.3863,
+      "theoretical_loss": 3.4459379643865264,
+      "tokens_seen": 2093023232
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036940298507462686,
+      "loss": 1.3771,
+      "theoretical_loss": 3.4459202802250317,
+      "tokens_seen": 2093154304
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036936286310383567,
+      "loss": 1.3355,
+      "theoretical_loss": 3.445902597480915,
+      "tokens_seen": 2093285376
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003693227411330445,
+      "loss": 1.3789,
+      "theoretical_loss": 3.4458849161539744,
+      "tokens_seen": 2093416448
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003692826191622532,
+      "loss": 1.3268,
+      "theoretical_loss": 3.4458672362440073,
+      "tokens_seen": 2093547520
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003692424971914621,
+      "loss": 1.386,
+      "theoretical_loss": 3.445849557750811,
+      "tokens_seen": 2093678592
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036920237522067084,
+      "loss": 1.3363,
+      "theoretical_loss": 3.445831880674184,
+      "tokens_seen": 2093809664
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036916225324987964,
+      "loss": 1.3619,
+      "theoretical_loss": 3.4458142050139235,
+      "tokens_seen": 2093940736
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036912213127908845,
+      "loss": 1.3135,
+      "theoretical_loss": 3.4457965307698277,
+      "tokens_seen": 2094071808
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003690820093082972,
+      "loss": 1.4447,
+      "theoretical_loss": 3.445778857941695,
+      "tokens_seen": 2094202880
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036904188733750606,
+      "loss": 1.2819,
+      "theoretical_loss": 3.4457611865293223,
+      "tokens_seen": 2094333952
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003690017653667148,
+      "loss": 1.3812,
+      "theoretical_loss": 3.4457435165325085,
+      "tokens_seen": 2094465024
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003689616433959236,
+      "loss": 1.3994,
+      "theoretical_loss": 3.445725847951051,
+      "tokens_seen": 2094596096
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036892152142513243,
+      "loss": 1.339,
+      "theoretical_loss": 3.445708180784748,
+      "tokens_seen": 2094727168
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003688813994543412,
+      "loss": 1.3712,
+      "theoretical_loss": 3.445690515033398,
+      "tokens_seen": 2094858240
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036884127748355,
+      "loss": 1.4052,
+      "theoretical_loss": 3.445672850696799,
+      "tokens_seen": 2094989312
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003688011555127588,
+      "loss": 1.3864,
+      "theoretical_loss": 3.445655187774749,
+      "tokens_seen": 2095120384
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003687610335419676,
+      "loss": 1.3563,
+      "theoretical_loss": 3.445637526267046,
+      "tokens_seen": 2095251456
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036872091157117635,
+      "loss": 1.3992,
+      "theoretical_loss": 3.445619866173489,
+      "tokens_seen": 2095382528
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003686807896003852,
+      "loss": 1.3061,
+      "theoretical_loss": 3.4456022074938755,
+      "tokens_seen": 2095513600
+    },
+    {
+      "epoch": 0.27,
+      "objective/train/advantage_avg": 0.49521300196647644,
+      "objective/train/docs_used": 764034,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2210099697113037,
+      "objective/train/original_loss": 1.2210098505020142,
+      "objective/train/theoretical_loss": 3.4455845502280047,
+      "objective/train/tokens_used": 466103776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2458733767271042,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0507712364196777,
+      "objective/train/weighted_lm_loss": 1.2829445600509644,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.7788007855415344,
+      "theoretical_loss": 3.4455845502280047,
+      "tokens_seen": 2095644672
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036864066762959396,
+      "loss": 1.3365,
+      "theoretical_loss": 3.4455845502280047,
+      "tokens_seen": 2095644672
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036860054565880277,
+      "loss": 1.4325,
+      "theoretical_loss": 3.4455668943756748,
+      "tokens_seen": 2095775744
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003685604236880116,
+      "loss": 1.3479,
+      "theoretical_loss": 3.4455492399366836,
+      "tokens_seen": 2095906816
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036852030171722033,
+      "loss": 1.3717,
+      "theoretical_loss": 3.44553158691083,
+      "tokens_seen": 2096037888
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003684801797464292,
+      "loss": 1.4245,
+      "theoretical_loss": 3.445513935297913,
+      "tokens_seen": 2096168960
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036844005777563794,
+      "loss": 1.4274,
+      "theoretical_loss": 3.4454962850977306,
+      "tokens_seen": 2096300032
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036839993580484675,
+      "loss": 1.4044,
+      "theoretical_loss": 3.4454786363100816,
+      "tokens_seen": 2096431104
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036835981383405555,
+      "loss": 1.4642,
+      "theoretical_loss": 3.445460988934765,
+      "tokens_seen": 2096562176
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003683196918632643,
+      "loss": 1.3505,
+      "theoretical_loss": 3.4454433429715783,
+      "tokens_seen": 2096693248
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003682795698924731,
+      "loss": 1.4607,
+      "theoretical_loss": 3.4454256984203213,
+      "tokens_seen": 2096824320
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003682394479216819,
+      "loss": 1.376,
+      "theoretical_loss": 3.445408055280793,
+      "tokens_seen": 2096955392
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003681993259508907,
+      "loss": 1.4299,
+      "theoretical_loss": 3.445390413552791,
+      "tokens_seen": 2097086464
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003681592039800995,
+      "loss": 1.3974,
+      "theoretical_loss": 3.445372773236115,
+      "tokens_seen": 2097217536
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036811908200930834,
+      "loss": 1.4533,
+      "theoretical_loss": 3.445355134330564,
+      "tokens_seen": 2097348608
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003680789600385171,
+      "loss": 1.3528,
+      "theoretical_loss": 3.445337496835937,
+      "tokens_seen": 2097479680
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036803883806772584,
+      "loss": 1.2719,
+      "theoretical_loss": 3.4453198607520323,
+      "tokens_seen": 2097610752
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003679987160969347,
+      "loss": 1.3836,
+      "theoretical_loss": 3.44530222607865,
+      "tokens_seen": 2097741824
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036795859412614345,
+      "loss": 1.3115,
+      "theoretical_loss": 3.4452845928155877,
+      "tokens_seen": 2097872896
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003679184721553523,
+      "loss": 1.3817,
+      "theoretical_loss": 3.4452669609626456,
+      "tokens_seen": 2098003968
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036787835018456107,
+      "loss": 1.3899,
+      "theoretical_loss": 3.445249330519623,
+      "tokens_seen": 2098135040
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003678382282137699,
+      "loss": 1.3408,
+      "theoretical_loss": 3.4452317014863176,
+      "tokens_seen": 2098266112
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003677981062429787,
+      "loss": 1.4767,
+      "theoretical_loss": 3.4452140738625303,
+      "tokens_seen": 2098397184
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003677579842721875,
+      "loss": 1.2995,
+      "theoretical_loss": 3.44519644764806,
+      "tokens_seen": 2098528256
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036771786230139624,
+      "loss": 1.3655,
+      "theoretical_loss": 3.4451788228427054,
+      "tokens_seen": 2098659328
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036767774033060504,
+      "loss": 1.387,
+      "theoretical_loss": 3.4451611994462663,
+      "tokens_seen": 2098790400
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036763761835981385,
+      "loss": 1.4023,
+      "theoretical_loss": 3.445143577458542,
+      "tokens_seen": 2098921472
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003675974963890226,
+      "loss": 1.3275,
+      "theoretical_loss": 3.445125956879332,
+      "tokens_seen": 2099052544
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036755737441823146,
+      "loss": 1.3761,
+      "theoretical_loss": 3.4451083377084357,
+      "tokens_seen": 2099183616
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003675172524474402,
+      "loss": 1.3789,
+      "theoretical_loss": 3.445090719945653,
+      "tokens_seen": 2099314688
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.000367477130476649,
+      "loss": 1.4014,
+      "theoretical_loss": 3.4450731035907824,
+      "tokens_seen": 2099445760
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036743700850585783,
+      "loss": 1.3363,
+      "theoretical_loss": 3.445055488643624,
+      "tokens_seen": 2099576832
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003673968865350666,
+      "loss": 1.3314,
+      "theoretical_loss": 3.4450378751039787,
+      "tokens_seen": 2099707904
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036735676456427544,
+      "loss": 1.4053,
+      "theoretical_loss": 3.4450202629716444,
+      "tokens_seen": 2099838976
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003673166425934842,
+      "loss": 1.3196,
+      "theoretical_loss": 3.445002652246422,
+      "tokens_seen": 2099970048
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.000367276520622693,
+      "loss": 1.3617,
+      "theoretical_loss": 3.4449850429281104,
+      "tokens_seen": 2100101120
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003672363986519018,
+      "loss": 1.3232,
+      "theoretical_loss": 3.44496743501651,
+      "tokens_seen": 2100232192
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003671962766811106,
+      "loss": 1.3438,
+      "theoretical_loss": 3.4449498285114206,
+      "tokens_seen": 2100363264
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036715615471031936,
+      "loss": 1.4289,
+      "theoretical_loss": 3.4449322234126414,
+      "tokens_seen": 2100494336
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036711603273952817,
+      "loss": 1.3903,
+      "theoretical_loss": 3.4449146197199734,
+      "tokens_seen": 2100625408
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.000367075910768737,
+      "loss": 1.3425,
+      "theoretical_loss": 3.444897017433216,
+      "tokens_seen": 2100756480
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036703578879794573,
+      "loss": 1.3487,
+      "theoretical_loss": 3.444879416552169,
+      "tokens_seen": 2100887552
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003669956668271546,
+      "loss": 1.3531,
+      "theoretical_loss": 3.4448618170766325,
+      "tokens_seen": 2101018624
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036695554485636334,
+      "loss": 1.3304,
+      "theoretical_loss": 3.4448442190064075,
+      "tokens_seen": 2101149696
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036691542288557215,
+      "loss": 1.3226,
+      "theoretical_loss": 3.4448266223412927,
+      "tokens_seen": 2101280768
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036687530091478095,
+      "loss": 1.3571,
+      "theoretical_loss": 3.4448090270810896,
+      "tokens_seen": 2101411840
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003668351789439897,
+      "loss": 1.3395,
+      "theoretical_loss": 3.444791433225597,
+      "tokens_seen": 2101542912
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003667950569731985,
+      "loss": 1.3863,
+      "theoretical_loss": 3.4447738407746167,
+      "tokens_seen": 2101673984
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003667549350024073,
+      "loss": 1.4341,
+      "theoretical_loss": 3.4447562497279485,
+      "tokens_seen": 2101805056
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003667148130316161,
+      "loss": 1.3587,
+      "theoretical_loss": 3.4447386600853918,
+      "tokens_seen": 2101936128
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036667469106082493,
+      "loss": 1.2891,
+      "theoretical_loss": 3.4447210718467476,
+      "tokens_seen": 2102067200
+    },
+    {
+      "epoch": 0.27,
+      "objective/train/advantage_avg": 0.4941035211086273,
+      "objective/train/docs_used": 765966,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2915301322937012,
+      "objective/train/original_loss": 1.2915300130844116,
+      "objective/train/theoretical_loss": 3.444703485011817,
+      "objective/train/tokens_used": 472657376,
+      "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": 1.0506526231765747,
+      "objective/train/weighted_lm_loss": 1.3568816184997559,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.444703485011817,
+      "tokens_seen": 2102198272
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036663456909003374,
+      "loss": 1.3244,
+      "theoretical_loss": 3.444703485011817,
+      "tokens_seen": 2102198272
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003665944471192425,
+      "loss": 1.3169,
+      "theoretical_loss": 3.4446858995803993,
+      "tokens_seen": 2102329344
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003665543251484513,
+      "loss": 1.334,
+      "theoretical_loss": 3.444668315552296,
+      "tokens_seen": 2102460416
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003665142031776601,
+      "loss": 1.3352,
+      "theoretical_loss": 3.4446507329273066,
+      "tokens_seen": 2102591488
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036647408120686885,
+      "loss": 1.2969,
+      "theoretical_loss": 3.444633151705233,
+      "tokens_seen": 2102722560
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003664339592360777,
+      "loss": 1.3545,
+      "theoretical_loss": 3.4446155718858744,
+      "tokens_seen": 2102853632
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036639383726528647,
+      "loss": 1.3794,
+      "theoretical_loss": 3.4445979934690327,
+      "tokens_seen": 2102984704
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003663537152944953,
+      "loss": 1.3894,
+      "theoretical_loss": 3.444580416454508,
+      "tokens_seen": 2103115776
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003663135933237041,
+      "loss": 1.3137,
+      "theoretical_loss": 3.444562840842101,
+      "tokens_seen": 2103246848
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003662734713529129,
+      "loss": 1.4274,
+      "theoretical_loss": 3.4445452666316125,
+      "tokens_seen": 2103377920
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036623334938212164,
+      "loss": 1.307,
+      "theoretical_loss": 3.4445276938228435,
+      "tokens_seen": 2103508992
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00036619322741133044,
+      "loss": 1.3437,
+      "theoretical_loss": 3.4445101224155947,
+      "tokens_seen": 2103640064
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036615310544053925,
+      "loss": 1.4089,
+      "theoretical_loss": 3.4444925524096672,
+      "tokens_seen": 2103771136
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.000366112983469748,
+      "loss": 1.3465,
+      "theoretical_loss": 3.444474983804862,
+      "tokens_seen": 2103902208
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036607286149895686,
+      "loss": 1.3512,
+      "theoretical_loss": 3.4444574166009794,
+      "tokens_seen": 2104033280
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003660327395281656,
+      "loss": 1.3479,
+      "theoretical_loss": 3.4444398507978216,
+      "tokens_seen": 2104164352
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003659926175573745,
+      "loss": 1.3996,
+      "theoretical_loss": 3.4444222863951888,
+      "tokens_seen": 2104295424
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036595249558658323,
+      "loss": 1.3274,
+      "theoretical_loss": 3.444404723392882,
+      "tokens_seen": 2104426496
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.000365912373615792,
+      "loss": 1.2559,
+      "theoretical_loss": 3.444387161790703,
+      "tokens_seen": 2104557568
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036587225164500084,
+      "loss": 1.3602,
+      "theoretical_loss": 3.4443696015884524,
+      "tokens_seen": 2104688640
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003658321296742096,
+      "loss": 1.3384,
+      "theoretical_loss": 3.444352042785932,
+      "tokens_seen": 2104819712
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003657920077034184,
+      "loss": 1.3712,
+      "theoretical_loss": 3.4443344853829427,
+      "tokens_seen": 2104950784
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003657518857326272,
+      "loss": 1.3311,
+      "theoretical_loss": 3.444316929379286,
+      "tokens_seen": 2105081856
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.000365711763761836,
+      "loss": 1.3237,
+      "theoretical_loss": 3.4442993747747632,
+      "tokens_seen": 2105212928
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036567164179104476,
+      "loss": 1.3469,
+      "theoretical_loss": 3.4442818215691755,
+      "tokens_seen": 2105344000
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036563151982025357,
+      "loss": 1.3291,
+      "theoretical_loss": 3.444264269762324,
+      "tokens_seen": 2105475072
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003655913978494624,
+      "loss": 1.379,
+      "theoretical_loss": 3.4442467193540116,
+      "tokens_seen": 2105606144
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036555127587867113,
+      "loss": 1.3465,
+      "theoretical_loss": 3.4442291703440384,
+      "tokens_seen": 2105737216
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036551115390788,
+      "loss": 1.3289,
+      "theoretical_loss": 3.4442116227322064,
+      "tokens_seen": 2105868288
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036547103193708874,
+      "loss": 1.373,
+      "theoretical_loss": 3.4441940765183174,
+      "tokens_seen": 2105999360
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003654309099662976,
+      "loss": 1.3451,
+      "theoretical_loss": 3.4441765317021726,
+      "tokens_seen": 2106130432
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036539078799550635,
+      "loss": 1.3774,
+      "theoretical_loss": 3.444158988283574,
+      "tokens_seen": 2106261504
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003653506660247151,
+      "loss": 1.3217,
+      "theoretical_loss": 3.444141446262323,
+      "tokens_seen": 2106392576
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036531054405392397,
+      "loss": 1.3783,
+      "theoretical_loss": 3.4441239056382225,
+      "tokens_seen": 2106523648
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003652704220831327,
+      "loss": 1.3783,
+      "theoretical_loss": 3.4441063664110727,
+      "tokens_seen": 2106654720
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003652303001123415,
+      "loss": 1.306,
+      "theoretical_loss": 3.4440888285806768,
+      "tokens_seen": 2106785792
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036519017814155033,
+      "loss": 1.3355,
+      "theoretical_loss": 3.4440712921468357,
+      "tokens_seen": 2106916864
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036515005617075914,
+      "loss": 1.2918,
+      "theoretical_loss": 3.4440537571093515,
+      "tokens_seen": 2107047936
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003651099341999679,
+      "loss": 1.2878,
+      "theoretical_loss": 3.444036223468026,
+      "tokens_seen": 2107179008
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003650698122291767,
+      "loss": 1.2591,
+      "theoretical_loss": 3.444018691222662,
+      "tokens_seen": 2107310080
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003650296902583855,
+      "loss": 1.2939,
+      "theoretical_loss": 3.444001160373061,
+      "tokens_seen": 2107441152
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036498956828759425,
+      "loss": 1.3914,
+      "theoretical_loss": 3.4439836309190253,
+      "tokens_seen": 2107572224
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003649494463168031,
+      "loss": 1.3646,
+      "theoretical_loss": 3.4439661028603568,
+      "tokens_seen": 2107703296
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036490932434601187,
+      "loss": 1.3478,
+      "theoretical_loss": 3.4439485761968576,
+      "tokens_seen": 2107834368
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036486920237522067,
+      "loss": 1.3252,
+      "theoretical_loss": 3.44393105092833,
+      "tokens_seen": 2107965440
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003648290804044295,
+      "loss": 1.3304,
+      "theoretical_loss": 3.4439135270545767,
+      "tokens_seen": 2108096512
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036478895843363823,
+      "loss": 1.3487,
+      "theoretical_loss": 3.4438960045753992,
+      "tokens_seen": 2108227584
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003647488364628471,
+      "loss": 1.3679,
+      "theoretical_loss": 3.4438784834906,
+      "tokens_seen": 2108358656
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036470871449205584,
+      "loss": 1.3077,
+      "theoretical_loss": 3.4438609637999824,
+      "tokens_seen": 2108489728
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036466859252126465,
+      "loss": 1.4136,
+      "theoretical_loss": 3.4438434455033473,
+      "tokens_seen": 2108620800
+    },
+    {
+      "epoch": 0.28,
+      "objective/train/advantage_avg": 0.4933674931526184,
+      "objective/train/docs_used": 768317,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.157778024673462,
+      "objective/train/original_loss": 1.1577777862548828,
+      "objective/train/theoretical_loss": 3.443825928600498,
+      "objective/train/tokens_used": 479210976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24379408359527588,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050576090812683,
+      "objective/train/weighted_lm_loss": 1.2162023782730103,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.443825928600498,
+      "tokens_seen": 2108751872
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036462847055047346,
+      "loss": 1.3087,
+      "theoretical_loss": 3.443825928600498,
+      "tokens_seen": 2108751872
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036458834857968226,
+      "loss": 1.3431,
+      "theoretical_loss": 3.4438084130912374,
+      "tokens_seen": 2108882944
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.000364548226608891,
+      "loss": 1.3757,
+      "theoretical_loss": 3.4437908989753674,
+      "tokens_seen": 2109014016
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003645081046380999,
+      "loss": 1.3863,
+      "theoretical_loss": 3.4437733862526905,
+      "tokens_seen": 2109145088
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036446798266730863,
+      "loss": 1.3175,
+      "theoretical_loss": 3.44375587492301,
+      "tokens_seen": 2109276160
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003644278606965174,
+      "loss": 1.4058,
+      "theoretical_loss": 3.4437383649861273,
+      "tokens_seen": 2109407232
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036438773872572624,
+      "loss": 1.3848,
+      "theoretical_loss": 3.4437208564418462,
+      "tokens_seen": 2109538304
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.000364347616754935,
+      "loss": 1.324,
+      "theoretical_loss": 3.4437033492899687,
+      "tokens_seen": 2109669376
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003643074947841438,
+      "loss": 1.3007,
+      "theoretical_loss": 3.4436858435302984,
+      "tokens_seen": 2109800448
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003642673728133526,
+      "loss": 1.3296,
+      "theoretical_loss": 3.4436683391626373,
+      "tokens_seen": 2109931520
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003642272508425614,
+      "loss": 1.2758,
+      "theoretical_loss": 3.443650836186789,
+      "tokens_seen": 2110062592
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036418712887177016,
+      "loss": 1.3532,
+      "theoretical_loss": 3.4436333346025556,
+      "tokens_seen": 2110193664
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036414700690097897,
+      "loss": 1.3667,
+      "theoretical_loss": 3.4436158344097407,
+      "tokens_seen": 2110324736
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003641068849301878,
+      "loss": 1.3231,
+      "theoretical_loss": 3.4435983356081468,
+      "tokens_seen": 2110455808
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003640667629593966,
+      "loss": 1.3596,
+      "theoretical_loss": 3.443580838197577,
+      "tokens_seen": 2110586880
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003640266409886054,
+      "loss": 1.2862,
+      "theoretical_loss": 3.4435633421778347,
+      "tokens_seen": 2110717952
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036398651901781414,
+      "loss": 1.3275,
+      "theoretical_loss": 3.443545847548722,
+      "tokens_seen": 2110849024
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.000363946397047023,
+      "loss": 1.2846,
+      "theoretical_loss": 3.4435283543100437,
+      "tokens_seen": 2110980096
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036390627507623175,
+      "loss": 1.2051,
+      "theoretical_loss": 3.4435108624616015,
+      "tokens_seen": 2111111168
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003638661531054405,
+      "loss": 1.3826,
+      "theoretical_loss": 3.443493372003199,
+      "tokens_seen": 2111242240
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036382603113464937,
+      "loss": 1.3175,
+      "theoretical_loss": 3.44347588293464,
+      "tokens_seen": 2111373312
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003637859091638581,
+      "loss": 1.3155,
+      "theoretical_loss": 3.4434583952557274,
+      "tokens_seen": 2111504384
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003637457871930669,
+      "loss": 1.2539,
+      "theoretical_loss": 3.443440908966264,
+      "tokens_seen": 2111635456
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036370566522227573,
+      "loss": 1.3464,
+      "theoretical_loss": 3.4434234240660535,
+      "tokens_seen": 2111766528
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036366554325148454,
+      "loss": 1.2823,
+      "theoretical_loss": 3.4434059405549,
+      "tokens_seen": 2111897600
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003636254212806933,
+      "loss": 1.3884,
+      "theoretical_loss": 3.443388458432606,
+      "tokens_seen": 2112028672
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003635852993099021,
+      "loss": 1.3485,
+      "theoretical_loss": 3.443370977698976,
+      "tokens_seen": 2112159744
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003635451773391109,
+      "loss": 1.3349,
+      "theoretical_loss": 3.4433534983538125,
+      "tokens_seen": 2112290816
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003635050553683197,
+      "loss": 1.3091,
+      "theoretical_loss": 3.44333602039692,
+      "tokens_seen": 2112421888
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003634649333975285,
+      "loss": 1.3186,
+      "theoretical_loss": 3.4433185438281013,
+      "tokens_seen": 2112552960
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036342481142673727,
+      "loss": 1.3411,
+      "theoretical_loss": 3.44330106864716,
+      "tokens_seen": 2112684032
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003633846894559461,
+      "loss": 1.3785,
+      "theoretical_loss": 3.4432835948539005,
+      "tokens_seen": 2112815104
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003633445674851549,
+      "loss": 1.4155,
+      "theoretical_loss": 3.443266122448126,
+      "tokens_seen": 2112946176
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036330444551436363,
+      "loss": 1.3748,
+      "theoretical_loss": 3.4432486514296405,
+      "tokens_seen": 2113077248
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003632643235435725,
+      "loss": 1.2587,
+      "theoretical_loss": 3.443231181798248,
+      "tokens_seen": 2113208320
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036322420157278124,
+      "loss": 1.2897,
+      "theoretical_loss": 3.443213713553752,
+      "tokens_seen": 2113339392
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036318407960199005,
+      "loss": 1.3793,
+      "theoretical_loss": 3.4431962466959565,
+      "tokens_seen": 2113470464
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036314395763119886,
+      "loss": 1.3441,
+      "theoretical_loss": 3.443178781224665,
+      "tokens_seen": 2113601536
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036310383566040766,
+      "loss": 1.2914,
+      "theoretical_loss": 3.4431613171396824,
+      "tokens_seen": 2113732608
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003630637136896164,
+      "loss": 1.2782,
+      "theoretical_loss": 3.4431438544408124,
+      "tokens_seen": 2113863680
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003630235917188253,
+      "loss": 1.4518,
+      "theoretical_loss": 3.443126393127858,
+      "tokens_seen": 2113994752
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.000362983469748034,
+      "loss": 1.3728,
+      "theoretical_loss": 3.443108933200625,
+      "tokens_seen": 2114125824
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003629433477772428,
+      "loss": 1.3568,
+      "theoretical_loss": 3.4430914746589165,
+      "tokens_seen": 2114256896
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036290322580645164,
+      "loss": 1.371,
+      "theoretical_loss": 3.4430740175025365,
+      "tokens_seen": 2114387968
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003628631038356604,
+      "loss": 1.3456,
+      "theoretical_loss": 3.4430565617312894,
+      "tokens_seen": 2114519040
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036282298186486925,
+      "loss": 1.3815,
+      "theoretical_loss": 3.44303910734498,
+      "tokens_seen": 2114650112
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.000362782859894078,
+      "loss": 1.2921,
+      "theoretical_loss": 3.4430216543434122,
+      "tokens_seen": 2114781184
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003627427379232868,
+      "loss": 1.3407,
+      "theoretical_loss": 3.4430042027263905,
+      "tokens_seen": 2114912256
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003627026159524956,
+      "loss": 1.3062,
+      "theoretical_loss": 3.4429867524937183,
+      "tokens_seen": 2115043328
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036266249398170437,
+      "loss": 1.2742,
+      "theoretical_loss": 3.442969303645201,
+      "tokens_seen": 2115174400
+    },
+    {
+      "epoch": 0.28,
+      "objective/train/advantage_avg": 0.49164819717407227,
+      "objective/train/docs_used": 770743,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2092077732086182,
+      "objective/train/original_loss": 1.209207534790039,
+      "objective/train/theoretical_loss": 3.4429518561806436,
+      "objective/train/tokens_used": 485764576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2424975335597992,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503976345062256,
+      "objective/train/weighted_lm_loss": 1.2701698541641235,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4429518561806436,
+      "tokens_seen": 2115305472
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003626223720109132,
+      "loss": 1.3454,
+      "theoretical_loss": 3.4429518561806436,
+      "tokens_seen": 2115305472
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.000362582250040122,
+      "loss": 1.3654,
+      "theoretical_loss": 3.4429344100998494,
+      "tokens_seen": 2115436544
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003625421280693308,
+      "loss": 1.334,
+      "theoretical_loss": 3.442916965402623,
+      "tokens_seen": 2115567616
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036250200609853954,
+      "loss": 1.2874,
+      "theoretical_loss": 3.4428995220887693,
+      "tokens_seen": 2115698688
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003624618841277484,
+      "loss": 1.3531,
+      "theoretical_loss": 3.4428820801580935,
+      "tokens_seen": 2115829760
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036242176215695715,
+      "loss": 1.3225,
+      "theoretical_loss": 3.4428646396103986,
+      "tokens_seen": 2115960832
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003623816401861659,
+      "loss": 1.3566,
+      "theoretical_loss": 3.4428472004454913,
+      "tokens_seen": 2116091904
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036234151821537477,
+      "loss": 1.3516,
+      "theoretical_loss": 3.442829762663175,
+      "tokens_seen": 2116222976
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003623013962445835,
+      "loss": 1.3279,
+      "theoretical_loss": 3.442812326263255,
+      "tokens_seen": 2116354048
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003622612742737924,
+      "loss": 1.4017,
+      "theoretical_loss": 3.4427948912455357,
+      "tokens_seen": 2116485120
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036222115230300113,
+      "loss": 1.2853,
+      "theoretical_loss": 3.4427774576098225,
+      "tokens_seen": 2116616192
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036218103033220994,
+      "loss": 1.3905,
+      "theoretical_loss": 3.44276002535592,
+      "tokens_seen": 2116747264
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036214090836141874,
+      "loss": 1.3899,
+      "theoretical_loss": 3.4427425944836325,
+      "tokens_seen": 2116878336
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003621007863906275,
+      "loss": 1.3417,
+      "theoretical_loss": 3.442725164992766,
+      "tokens_seen": 2117009408
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003620606644198363,
+      "loss": 1.3054,
+      "theoretical_loss": 3.4427077368831247,
+      "tokens_seen": 2117140480
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003620205424490451,
+      "loss": 1.3515,
+      "theoretical_loss": 3.4426903101545143,
+      "tokens_seen": 2117271552
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003619804204782539,
+      "loss": 1.2964,
+      "theoretical_loss": 3.4426728848067394,
+      "tokens_seen": 2117402624
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036194029850746267,
+      "loss": 1.3521,
+      "theoretical_loss": 3.4426554608396054,
+      "tokens_seen": 2117533696
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003619001765366715,
+      "loss": 1.2932,
+      "theoretical_loss": 3.442638038252917,
+      "tokens_seen": 2117664768
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003618600545658803,
+      "loss": 1.3243,
+      "theoretical_loss": 3.4426206170464804,
+      "tokens_seen": 2117795840
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036181993259508903,
+      "loss": 1.3044,
+      "theoretical_loss": 3.4426031972201,
+      "tokens_seen": 2117926912
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003617798106242979,
+      "loss": 1.3053,
+      "theoretical_loss": 3.442585778773581,
+      "tokens_seen": 2118057984
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036173968865350664,
+      "loss": 1.3808,
+      "theoretical_loss": 3.442568361706729,
+      "tokens_seen": 2118189056
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036169956668271545,
+      "loss": 1.3569,
+      "theoretical_loss": 3.442550946019349,
+      "tokens_seen": 2118320128
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036165944471192426,
+      "loss": 1.3521,
+      "theoretical_loss": 3.4425335317112475,
+      "tokens_seen": 2118451200
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036161932274113306,
+      "loss": 1.3946,
+      "theoretical_loss": 3.442516118782229,
+      "tokens_seen": 2118582272
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036157920077034187,
+      "loss": 1.378,
+      "theoretical_loss": 3.4424987072320983,
+      "tokens_seen": 2118713344
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003615390787995507,
+      "loss": 1.341,
+      "theoretical_loss": 3.4424812970606626,
+      "tokens_seen": 2118844416
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003614989568287594,
+      "loss": 1.3097,
+      "theoretical_loss": 3.4424638882677265,
+      "tokens_seen": 2118975488
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036145883485796823,
+      "loss": 1.3793,
+      "theoretical_loss": 3.4424464808530955,
+      "tokens_seen": 2119106560
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036141871288717704,
+      "loss": 1.3563,
+      "theoretical_loss": 3.442429074816576,
+      "tokens_seen": 2119237632
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003613785909163858,
+      "loss": 1.4202,
+      "theoretical_loss": 3.4424116701579726,
+      "tokens_seen": 2119368704
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036133846894559465,
+      "loss": 1.3522,
+      "theoretical_loss": 3.4423942668770913,
+      "tokens_seen": 2119499776
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003612983469748034,
+      "loss": 1.4047,
+      "theoretical_loss": 3.4423768649737387,
+      "tokens_seen": 2119630848
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003612582250040122,
+      "loss": 1.3481,
+      "theoretical_loss": 3.44235946444772,
+      "tokens_seen": 2119761920
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.000361218103033221,
+      "loss": 1.3544,
+      "theoretical_loss": 3.44234206529884,
+      "tokens_seen": 2119892992
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00036117798106242977,
+      "loss": 1.3469,
+      "theoretical_loss": 3.442324667526907,
+      "tokens_seen": 2120024064
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003611378590916386,
+      "loss": 1.4069,
+      "theoretical_loss": 3.4423072711317246,
+      "tokens_seen": 2120155136
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003610977371208474,
+      "loss": 1.3929,
+      "theoretical_loss": 3.4422898761131,
+      "tokens_seen": 2120286208
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003610576151500562,
+      "loss": 1.3397,
+      "theoretical_loss": 3.4422724824708384,
+      "tokens_seen": 2120417280
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00036101749317926494,
+      "loss": 1.3068,
+      "theoretical_loss": 3.442255090204746,
+      "tokens_seen": 2120548352
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003609773712084738,
+      "loss": 1.2793,
+      "theoretical_loss": 3.4422376993146298,
+      "tokens_seen": 2120679424
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00036093724923768255,
+      "loss": 1.3372,
+      "theoretical_loss": 3.4422203098002955,
+      "tokens_seen": 2120810496
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00036089712726689136,
+      "loss": 1.467,
+      "theoretical_loss": 3.4422029216615484,
+      "tokens_seen": 2120941568
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00036085700529610017,
+      "loss": 1.2791,
+      "theoretical_loss": 3.4421855348981953,
+      "tokens_seen": 2121072640
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003608168833253089,
+      "loss": 1.2599,
+      "theoretical_loss": 3.4421681495100422,
+      "tokens_seen": 2121203712
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003607767613545178,
+      "loss": 1.312,
+      "theoretical_loss": 3.442150765496896,
+      "tokens_seen": 2121334784
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00036073663938372653,
+      "loss": 1.3473,
+      "theoretical_loss": 3.4421333828585627,
+      "tokens_seen": 2121465856
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00036069651741293534,
+      "loss": 1.3409,
+      "theoretical_loss": 3.442116001594848,
+      "tokens_seen": 2121596928
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00036065639544214414,
+      "loss": 1.316,
+      "theoretical_loss": 3.4420986217055587,
+      "tokens_seen": 2121728000
+    },
+    {
+      "epoch": 0.29,
+      "objective/train/advantage_avg": 0.49318167567253113,
+      "objective/train/docs_used": 773131,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2524040937423706,
+      "objective/train/original_loss": 1.252403974533081,
+      "objective/train/theoretical_loss": 3.442081243190502,
+      "objective/train/tokens_used": 492318176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24356888234615326,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505564212799072,
+      "objective/train/weighted_lm_loss": 1.3156495094299316,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9834713935852051,
+      "theoretical_loss": 3.442081243190502,
+      "tokens_seen": 2121859072
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003606162734713529,
+      "loss": 1.2837,
+      "theoretical_loss": 3.442081243190502,
+      "tokens_seen": 2121859072
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003605761515005617,
+      "loss": 1.3832,
+      "theoretical_loss": 3.442063866049483,
+      "tokens_seen": 2121990144
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003605360295297705,
+      "loss": 1.3969,
+      "theoretical_loss": 3.4420464902823094,
+      "tokens_seen": 2122121216
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003604959075589793,
+      "loss": 1.3339,
+      "theoretical_loss": 3.4420291158887864,
+      "tokens_seen": 2122252288
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00036045578558818807,
+      "loss": 1.3558,
+      "theoretical_loss": 3.442011742868722,
+      "tokens_seen": 2122383360
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003604156636173969,
+      "loss": 1.4009,
+      "theoretical_loss": 3.441994371221922,
+      "tokens_seen": 2122514432
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003603755416466057,
+      "loss": 1.3917,
+      "theoretical_loss": 3.441977000948193,
+      "tokens_seen": 2122645504
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003603354196758145,
+      "loss": 1.3846,
+      "theoretical_loss": 3.4419596320473422,
+      "tokens_seen": 2122776576
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003602952977050233,
+      "loss": 1.4195,
+      "theoretical_loss": 3.4419422645191764,
+      "tokens_seen": 2122907648
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00036025517573423204,
+      "loss": 1.4082,
+      "theoretical_loss": 3.441924898363501,
+      "tokens_seen": 2123038720
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003602150537634409,
+      "loss": 1.3072,
+      "theoretical_loss": 3.441907533580125,
+      "tokens_seen": 2123169792
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00036017493179264966,
+      "loss": 1.3699,
+      "theoretical_loss": 3.441890170168854,
+      "tokens_seen": 2123300864
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00036013480982185846,
+      "loss": 1.3772,
+      "theoretical_loss": 3.441872808129494,
+      "tokens_seen": 2123431936
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00036009468785106727,
+      "loss": 1.3738,
+      "theoretical_loss": 3.4418554474618537,
+      "tokens_seen": 2123563008
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.000360054565880276,
+      "loss": 1.3562,
+      "theoretical_loss": 3.4418380881657393,
+      "tokens_seen": 2123694080
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003600144439094848,
+      "loss": 1.3618,
+      "theoretical_loss": 3.441820730240958,
+      "tokens_seen": 2123825152
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035997432193869363,
+      "loss": 1.3323,
+      "theoretical_loss": 3.441803373687316,
+      "tokens_seen": 2123956224
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035993419996790244,
+      "loss": 1.3627,
+      "theoretical_loss": 3.4417860185046214,
+      "tokens_seen": 2124087296
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003598940779971112,
+      "loss": 1.3419,
+      "theoretical_loss": 3.441768664692681,
+      "tokens_seen": 2124218368
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035985395602632005,
+      "loss": 1.363,
+      "theoretical_loss": 3.4417513122513017,
+      "tokens_seen": 2124349440
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003598138340555288,
+      "loss": 1.4755,
+      "theoretical_loss": 3.441733961180291,
+      "tokens_seen": 2124480512
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003597737120847376,
+      "loss": 1.2959,
+      "theoretical_loss": 3.441716611479456,
+      "tokens_seen": 2124611584
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003597335901139464,
+      "loss": 1.3326,
+      "theoretical_loss": 3.4416992631486045,
+      "tokens_seen": 2124742656
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035969346814315517,
+      "loss": 1.3298,
+      "theoretical_loss": 3.441681916187543,
+      "tokens_seen": 2124873728
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035965334617236403,
+      "loss": 1.3609,
+      "theoretical_loss": 3.441664570596079,
+      "tokens_seen": 2125004800
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003596132242015728,
+      "loss": 1.3981,
+      "theoretical_loss": 3.4416472263740205,
+      "tokens_seen": 2125135872
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003595731022307816,
+      "loss": 1.3766,
+      "theoretical_loss": 3.441629883521175,
+      "tokens_seen": 2125266944
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003595329802599904,
+      "loss": 1.4468,
+      "theoretical_loss": 3.441612542037349,
+      "tokens_seen": 2125398016
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003594928582891992,
+      "loss": 1.2892,
+      "theoretical_loss": 3.4415952019223504,
+      "tokens_seen": 2125529088
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035945273631840795,
+      "loss": 1.3447,
+      "theoretical_loss": 3.4415778631759872,
+      "tokens_seen": 2125660160
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035941261434761676,
+      "loss": 1.3488,
+      "theoretical_loss": 3.441560525798067,
+      "tokens_seen": 2125791232
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035937249237682556,
+      "loss": 1.4316,
+      "theoretical_loss": 3.4415431897883964,
+      "tokens_seen": 2125922304
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003593323704060343,
+      "loss": 1.3569,
+      "theoretical_loss": 3.441525855146784,
+      "tokens_seen": 2126053376
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003592922484352432,
+      "loss": 1.34,
+      "theoretical_loss": 3.441508521873038,
+      "tokens_seen": 2126184448
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035925212646445193,
+      "loss": 1.3642,
+      "theoretical_loss": 3.441491189966965,
+      "tokens_seen": 2126315520
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035921200449366074,
+      "loss": 1.3093,
+      "theoretical_loss": 3.4414738594283727,
+      "tokens_seen": 2126446592
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035917188252286954,
+      "loss": 1.3911,
+      "theoretical_loss": 3.44145653025707,
+      "tokens_seen": 2126577664
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003591317605520783,
+      "loss": 1.3332,
+      "theoretical_loss": 3.441439202452864,
+      "tokens_seen": 2126708736
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035909163858128715,
+      "loss": 1.3908,
+      "theoretical_loss": 3.4414218760155633,
+      "tokens_seen": 2126839808
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003590515166104959,
+      "loss": 1.3488,
+      "theoretical_loss": 3.4414045509449753,
+      "tokens_seen": 2126970880
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003590113946397047,
+      "loss": 1.282,
+      "theoretical_loss": 3.4413872272409076,
+      "tokens_seen": 2127101952
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003589712726689135,
+      "loss": 1.4341,
+      "theoretical_loss": 3.441369904903169,
+      "tokens_seen": 2127233024
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003589311506981223,
+      "loss": 1.406,
+      "theoretical_loss": 3.4413525839315673,
+      "tokens_seen": 2127364096
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003588910287273311,
+      "loss": 1.3378,
+      "theoretical_loss": 3.441335264325911,
+      "tokens_seen": 2127495168
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003588509067565399,
+      "loss": 1.3441,
+      "theoretical_loss": 3.441317946086007,
+      "tokens_seen": 2127626240
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003588107847857487,
+      "loss": 1.4181,
+      "theoretical_loss": 3.4413006292116646,
+      "tokens_seen": 2127757312
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035877066281495744,
+      "loss": 1.2563,
+      "theoretical_loss": 3.4412833137026917,
+      "tokens_seen": 2127888384
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003587305408441663,
+      "loss": 1.299,
+      "theoretical_loss": 3.4412659995588966,
+      "tokens_seen": 2128019456
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035869041887337506,
+      "loss": 1.3199,
+      "theoretical_loss": 3.441248686780087,
+      "tokens_seen": 2128150528
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035865029690258386,
+      "loss": 1.3454,
+      "theoretical_loss": 3.441231375366072,
+      "tokens_seen": 2128281600
+    },
+    {
+      "epoch": 0.29,
+      "objective/train/advantage_avg": 0.49311745166778564,
+      "objective/train/docs_used": 775219,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2643911838531494,
+      "objective/train/original_loss": 1.2643911838531494,
+      "objective/train/theoretical_loss": 3.44121406531666,
+      "objective/train/tokens_used": 498871776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24350152909755707,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505497455596924,
+      "objective/train/weighted_lm_loss": 1.3284060955047607,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.44121406531666,
+      "tokens_seen": 2128412672
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035861017493179267,
+      "loss": 1.3089,
+      "theoretical_loss": 3.44121406531666,
+      "tokens_seen": 2128412672
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003585700529610014,
+      "loss": 1.4147,
+      "theoretical_loss": 3.4411967566316592,
+      "tokens_seen": 2128543744
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003585299309902102,
+      "loss": 1.3449,
+      "theoretical_loss": 3.4411794493108774,
+      "tokens_seen": 2128674816
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035848980901941903,
+      "loss": 1.3395,
+      "theoretical_loss": 3.441162143354124,
+      "tokens_seen": 2128805888
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035844968704862784,
+      "loss": 1.3186,
+      "theoretical_loss": 3.441144838761207,
+      "tokens_seen": 2128936960
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035840956507783664,
+      "loss": 1.256,
+      "theoretical_loss": 3.4411275355319355,
+      "tokens_seen": 2129068032
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035836944310704545,
+      "loss": 1.2933,
+      "theoretical_loss": 3.4411102336661177,
+      "tokens_seen": 2129199104
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003583293211362542,
+      "loss": 1.4246,
+      "theoretical_loss": 3.441092933163562,
+      "tokens_seen": 2129330176
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035828919916546306,
+      "loss": 1.3989,
+      "theoretical_loss": 3.4410756340240773,
+      "tokens_seen": 2129461248
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003582490771946718,
+      "loss": 1.3756,
+      "theoretical_loss": 3.4410583362474725,
+      "tokens_seen": 2129592320
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035820895522388057,
+      "loss": 1.3258,
+      "theoretical_loss": 3.4410410398335562,
+      "tokens_seen": 2129723392
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035816883325308943,
+      "loss": 1.3578,
+      "theoretical_loss": 3.4410237447821377,
+      "tokens_seen": 2129854464
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003581287112822982,
+      "loss": 1.4464,
+      "theoretical_loss": 3.441006451093025,
+      "tokens_seen": 2129985536
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.000358088589311507,
+      "loss": 1.4805,
+      "theoretical_loss": 3.440989158766027,
+      "tokens_seen": 2130116608
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003580484673407158,
+      "loss": 1.334,
+      "theoretical_loss": 3.440971867800953,
+      "tokens_seen": 2130247680
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003580083453699246,
+      "loss": 1.3122,
+      "theoretical_loss": 3.440954578197613,
+      "tokens_seen": 2130378752
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035796822339913335,
+      "loss": 1.4246,
+      "theoretical_loss": 3.440937289955814,
+      "tokens_seen": 2130509824
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035792810142834216,
+      "loss": 1.3898,
+      "theoretical_loss": 3.4409200030753655,
+      "tokens_seen": 2130640896
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035788797945755096,
+      "loss": 1.391,
+      "theoretical_loss": 3.4409027175560776,
+      "tokens_seen": 2130771968
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003578478574867597,
+      "loss": 1.3208,
+      "theoretical_loss": 3.4408854333977583,
+      "tokens_seen": 2130903040
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003578077355159686,
+      "loss": 1.294,
+      "theoretical_loss": 3.4408681506002177,
+      "tokens_seen": 2131034112
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035776761354517733,
+      "loss": 1.3409,
+      "theoretical_loss": 3.440850869163264,
+      "tokens_seen": 2131165184
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003577274915743862,
+      "loss": 1.3008,
+      "theoretical_loss": 3.440833589086707,
+      "tokens_seen": 2131296256
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035768736960359494,
+      "loss": 1.2605,
+      "theoretical_loss": 3.4408163103703564,
+      "tokens_seen": 2131427328
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003576472476328037,
+      "loss": 1.3333,
+      "theoretical_loss": 3.4407990330140206,
+      "tokens_seen": 2131558400
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035760712566201255,
+      "loss": 1.4377,
+      "theoretical_loss": 3.4407817570175094,
+      "tokens_seen": 2131689472
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003575670036912213,
+      "loss": 1.3646,
+      "theoretical_loss": 3.440764482380632,
+      "tokens_seen": 2131820544
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003575268817204301,
+      "loss": 1.288,
+      "theoretical_loss": 3.440747209103198,
+      "tokens_seen": 2131951616
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003574867597496389,
+      "loss": 1.3493,
+      "theoretical_loss": 3.440729937185016,
+      "tokens_seen": 2132082688
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003574466377788477,
+      "loss": 1.3274,
+      "theoretical_loss": 3.440712666625897,
+      "tokens_seen": 2132213760
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003574065158080565,
+      "loss": 1.4044,
+      "theoretical_loss": 3.4406953974256496,
+      "tokens_seen": 2132344832
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003573663938372653,
+      "loss": 1.4783,
+      "theoretical_loss": 3.440678129584083,
+      "tokens_seen": 2132475904
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003573262718664741,
+      "loss": 1.4123,
+      "theoretical_loss": 3.440660863101008,
+      "tokens_seen": 2132606976
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035728614989568284,
+      "loss": 1.3809,
+      "theoretical_loss": 3.440643597976233,
+      "tokens_seen": 2132738048
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003572460279248917,
+      "loss": 1.3876,
+      "theoretical_loss": 3.440626334209569,
+      "tokens_seen": 2132869120
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035720590595410045,
+      "loss": 1.3212,
+      "theoretical_loss": 3.440609071800824,
+      "tokens_seen": 2133000192
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003571657839833093,
+      "loss": 1.2932,
+      "theoretical_loss": 3.440591810749809,
+      "tokens_seen": 2133131264
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035712566201251807,
+      "loss": 1.3907,
+      "theoretical_loss": 3.4405745510563337,
+      "tokens_seen": 2133262336
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003570855400417268,
+      "loss": 1.353,
+      "theoretical_loss": 3.440557292720208,
+      "tokens_seen": 2133393408
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003570454180709357,
+      "loss": 1.3741,
+      "theoretical_loss": 3.440540035741241,
+      "tokens_seen": 2133524480
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035700529610014443,
+      "loss": 1.3276,
+      "theoretical_loss": 3.4405227801192435,
+      "tokens_seen": 2133655552
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035696517412935324,
+      "loss": 1.3634,
+      "theoretical_loss": 3.4405055258540242,
+      "tokens_seen": 2133786624
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035692505215856204,
+      "loss": 1.4125,
+      "theoretical_loss": 3.440488272945395,
+      "tokens_seen": 2133917696
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035688493018777085,
+      "loss": 1.3378,
+      "theoretical_loss": 3.440471021393164,
+      "tokens_seen": 2134048768
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003568448082169796,
+      "loss": 1.3849,
+      "theoretical_loss": 3.440453771197143,
+      "tokens_seen": 2134179840
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003568046862461884,
+      "loss": 1.3033,
+      "theoretical_loss": 3.4404365223571407,
+      "tokens_seen": 2134310912
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003567645642753972,
+      "loss": 1.2894,
+      "theoretical_loss": 3.440419274872968,
+      "tokens_seen": 2134441984
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035672444230460597,
+      "loss": 1.4617,
+      "theoretical_loss": 3.440402028744434,
+      "tokens_seen": 2134573056
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035668432033381483,
+      "loss": 1.3344,
+      "theoretical_loss": 3.440384783971351,
+      "tokens_seen": 2134704128
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003566441983630236,
+      "loss": 1.3519,
+      "theoretical_loss": 3.4403675405535274,
+      "tokens_seen": 2134835200
+    },
+    {
+      "epoch": 0.29,
+      "objective/train/advantage_avg": 0.49208778142929077,
+      "objective/train/docs_used": 777595,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3036603927612305,
+      "objective/train/original_loss": 1.303660273551941,
+      "objective/train/theoretical_loss": 3.440350298490774,
+      "objective/train/tokens_used": 505425376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2426285296678543,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504422187805176,
+      "objective/train/weighted_lm_loss": 1.3694040775299072,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.440350298490774,
+      "tokens_seen": 2134966272
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003566040763922324,
+      "loss": 1.3744,
+      "theoretical_loss": 3.440350298490774,
+      "tokens_seen": 2134966272
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003565639544214412,
+      "loss": 1.3144,
+      "theoretical_loss": 3.4403330577829014,
+      "tokens_seen": 2135097344
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035652383245065,
+      "loss": 1.4051,
+      "theoretical_loss": 3.4403158184297196,
+      "tokens_seen": 2135228416
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003564837104798588,
+      "loss": 1.3177,
+      "theoretical_loss": 3.4402985804310395,
+      "tokens_seen": 2135359488
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035644358850906756,
+      "loss": 1.2738,
+      "theoretical_loss": 3.4402813437866713,
+      "tokens_seen": 2135490560
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035640346653827636,
+      "loss": 1.3602,
+      "theoretical_loss": 3.440264108496425,
+      "tokens_seen": 2135621632
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035636334456748517,
+      "loss": 1.338,
+      "theoretical_loss": 3.440246874560112,
+      "tokens_seen": 2135752704
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.000356323222596694,
+      "loss": 1.2851,
+      "theoretical_loss": 3.4402296419775427,
+      "tokens_seen": 2135883776
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035628310062590273,
+      "loss": 1.306,
+      "theoretical_loss": 3.4402124107485266,
+      "tokens_seen": 2136014848
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003562429786551116,
+      "loss": 1.4534,
+      "theoretical_loss": 3.4401951808728755,
+      "tokens_seen": 2136145920
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035620285668432034,
+      "loss": 1.3078,
+      "theoretical_loss": 3.4401779523503997,
+      "tokens_seen": 2136276992
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003561627347135291,
+      "loss": 1.3139,
+      "theoretical_loss": 3.44016072518091,
+      "tokens_seen": 2136408064
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035612261274273795,
+      "loss": 1.3453,
+      "theoretical_loss": 3.4401434993642175,
+      "tokens_seen": 2136539136
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003560824907719467,
+      "loss": 1.3434,
+      "theoretical_loss": 3.440126274900132,
+      "tokens_seen": 2136670208
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003560423688011555,
+      "loss": 1.3236,
+      "theoretical_loss": 3.4401090517884647,
+      "tokens_seen": 2136801280
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003560022468303643,
+      "loss": 1.3287,
+      "theoretical_loss": 3.440091830029027,
+      "tokens_seen": 2136932352
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003559621248595731,
+      "loss": 1.3709,
+      "theoretical_loss": 3.4400746096216297,
+      "tokens_seen": 2137063424
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003559220028887819,
+      "loss": 1.3041,
+      "theoretical_loss": 3.440057390566083,
+      "tokens_seen": 2137194496
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003558818809179907,
+      "loss": 1.3249,
+      "theoretical_loss": 3.440040172862199,
+      "tokens_seen": 2137325568
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003558417589471995,
+      "loss": 1.467,
+      "theoretical_loss": 3.4400229565097877,
+      "tokens_seen": 2137456640
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003558016369764083,
+      "loss": 1.3552,
+      "theoretical_loss": 3.44000574150866,
+      "tokens_seen": 2137587712
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003557615150056171,
+      "loss": 1.3358,
+      "theoretical_loss": 3.4399885278586284,
+      "tokens_seen": 2137718784
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035572139303482585,
+      "loss": 1.3692,
+      "theoretical_loss": 3.439971315559503,
+      "tokens_seen": 2137849856
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003556812710640347,
+      "loss": 1.3815,
+      "theoretical_loss": 3.439954104611095,
+      "tokens_seen": 2137980928
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035564114909324347,
+      "loss": 1.3485,
+      "theoretical_loss": 3.439936895013216,
+      "tokens_seen": 2138112000
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003556010271224522,
+      "loss": 1.4182,
+      "theoretical_loss": 3.4399196867656765,
+      "tokens_seen": 2138243072
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003555609051516611,
+      "loss": 1.318,
+      "theoretical_loss": 3.4399024798682887,
+      "tokens_seen": 2138374144
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035552078318086983,
+      "loss": 1.3467,
+      "theoretical_loss": 3.439885274320863,
+      "tokens_seen": 2138505216
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035548066121007864,
+      "loss": 1.2704,
+      "theoretical_loss": 3.439868070123212,
+      "tokens_seen": 2138636288
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035544053923928744,
+      "loss": 1.3688,
+      "theoretical_loss": 3.4398508672751458,
+      "tokens_seen": 2138767360
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035540041726849625,
+      "loss": 1.2033,
+      "theoretical_loss": 3.4398336657764768,
+      "tokens_seen": 2138898432
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.000355360295297705,
+      "loss": 1.3016,
+      "theoretical_loss": 3.4398164656270156,
+      "tokens_seen": 2139029504
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003553201733269138,
+      "loss": 1.4354,
+      "theoretical_loss": 3.439799266826574,
+      "tokens_seen": 2139160576
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003552800513561226,
+      "loss": 1.3379,
+      "theoretical_loss": 3.439782069374964,
+      "tokens_seen": 2139291648
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003552399293853314,
+      "loss": 1.3617,
+      "theoretical_loss": 3.439764873271997,
+      "tokens_seen": 2139422720
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035519980741454023,
+      "loss": 1.3025,
+      "theoretical_loss": 3.4397476785174845,
+      "tokens_seen": 2139553792
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.000355159685443749,
+      "loss": 1.3479,
+      "theoretical_loss": 3.4397304851112382,
+      "tokens_seen": 2139684864
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035511956347295784,
+      "loss": 1.3463,
+      "theoretical_loss": 3.439713293053069,
+      "tokens_seen": 2139815936
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003550794415021666,
+      "loss": 1.3066,
+      "theoretical_loss": 3.4396961023427908,
+      "tokens_seen": 2139947008
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003550393195313754,
+      "loss": 1.3531,
+      "theoretical_loss": 3.439678912980213,
+      "tokens_seen": 2140078080
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003549991975605842,
+      "loss": 1.3401,
+      "theoretical_loss": 3.4396617249651484,
+      "tokens_seen": 2140209152
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035495907558979296,
+      "loss": 1.299,
+      "theoretical_loss": 3.4396445382974097,
+      "tokens_seen": 2140340224
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035491895361900176,
+      "loss": 1.3598,
+      "theoretical_loss": 3.4396273529768067,
+      "tokens_seen": 2140471296
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035487883164821057,
+      "loss": 1.3198,
+      "theoretical_loss": 3.4396101690031533,
+      "tokens_seen": 2140602368
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003548387096774194,
+      "loss": 1.3216,
+      "theoretical_loss": 3.439592986376261,
+      "tokens_seen": 2140733440
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035479858770662813,
+      "loss": 1.333,
+      "theoretical_loss": 3.4395758050959406,
+      "tokens_seen": 2140864512
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.000354758465735837,
+      "loss": 1.3133,
+      "theoretical_loss": 3.439558625162006,
+      "tokens_seen": 2140995584
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035471834376504574,
+      "loss": 1.3116,
+      "theoretical_loss": 3.439541446574268,
+      "tokens_seen": 2141126656
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003546782217942545,
+      "loss": 1.3424,
+      "theoretical_loss": 3.439524269332539,
+      "tokens_seen": 2141257728
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035463809982346335,
+      "loss": 1.3786,
+      "theoretical_loss": 3.439507093436631,
+      "tokens_seen": 2141388800
+    },
+    {
+      "epoch": 0.3,
+      "objective/train/advantage_avg": 0.4912111163139343,
+      "objective/train/docs_used": 779818,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.272933840751648,
+      "objective/train/original_loss": 1.2729336023330688,
+      "objective/train/theoretical_loss": 3.4394899188863564,
+      "objective/train/tokens_used": 511978976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2420736402273178,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503517389297485,
+      "objective/train/weighted_lm_loss": 1.3371341228485107,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4394899188863564,
+      "tokens_seen": 2141519872
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003545979778526721,
+      "loss": 1.35,
+      "theoretical_loss": 3.4394899188863564,
+      "tokens_seen": 2141519872
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035455785588188097,
+      "loss": 1.3525,
+      "theoretical_loss": 3.4394727456815275,
+      "tokens_seen": 2141650944
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003545177339110897,
+      "loss": 1.3336,
+      "theoretical_loss": 3.4394555738219568,
+      "tokens_seen": 2141782016
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003544776119402985,
+      "loss": 1.354,
+      "theoretical_loss": 3.4394384033074563,
+      "tokens_seen": 2141913088
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035443748996950733,
+      "loss": 1.3306,
+      "theoretical_loss": 3.4394212341378383,
+      "tokens_seen": 2142044160
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003543973679987161,
+      "loss": 1.3816,
+      "theoretical_loss": 3.439404066312915,
+      "tokens_seen": 2142175232
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003543572460279249,
+      "loss": 1.3775,
+      "theoretical_loss": 3.439386899832499,
+      "tokens_seen": 2142306304
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003543171240571337,
+      "loss": 1.4204,
+      "theoretical_loss": 3.4393697346964025,
+      "tokens_seen": 2142437376
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003542770020863425,
+      "loss": 1.2977,
+      "theoretical_loss": 3.439352570904439,
+      "tokens_seen": 2142568448
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035423688011555125,
+      "loss": 1.3403,
+      "theoretical_loss": 3.4393354084564196,
+      "tokens_seen": 2142699520
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003541967581447601,
+      "loss": 1.3482,
+      "theoretical_loss": 3.4393182473521575,
+      "tokens_seen": 2142830592
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035415663617396887,
+      "loss": 1.3254,
+      "theoretical_loss": 3.439301087591466,
+      "tokens_seen": 2142961664
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003541165142031776,
+      "loss": 1.3445,
+      "theoretical_loss": 3.439283929174157,
+      "tokens_seen": 2143092736
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003540763922323865,
+      "loss": 1.3283,
+      "theoretical_loss": 3.439266772100043,
+      "tokens_seen": 2143223808
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035403627026159523,
+      "loss": 1.4237,
+      "theoretical_loss": 3.439249616368937,
+      "tokens_seen": 2143354880
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003539961482908041,
+      "loss": 1.2654,
+      "theoretical_loss": 3.4392324619806516,
+      "tokens_seen": 2143485952
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035395602632001284,
+      "loss": 1.3151,
+      "theoretical_loss": 3.439215308935,
+      "tokens_seen": 2143617024
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035391590434922165,
+      "loss": 1.4002,
+      "theoretical_loss": 3.439198157231795,
+      "tokens_seen": 2143748096
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035387578237843046,
+      "loss": 1.3056,
+      "theoretical_loss": 3.4391810068708493,
+      "tokens_seen": 2143879168
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003538356604076392,
+      "loss": 1.3022,
+      "theoretical_loss": 3.439163857851975,
+      "tokens_seen": 2144010240
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.000353795538436848,
+      "loss": 1.3979,
+      "theoretical_loss": 3.4391467101749864,
+      "tokens_seen": 2144141312
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003537554164660568,
+      "loss": 1.3561,
+      "theoretical_loss": 3.439129563839696,
+      "tokens_seen": 2144272384
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035371529449526563,
+      "loss": 1.2641,
+      "theoretical_loss": 3.4391124188459163,
+      "tokens_seen": 2144403456
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003536751725244744,
+      "loss": 1.3572,
+      "theoretical_loss": 3.439095275193461,
+      "tokens_seen": 2144534528
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035363505055368324,
+      "loss": 1.3046,
+      "theoretical_loss": 3.4390781328821425,
+      "tokens_seen": 2144665600
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.000353594928582892,
+      "loss": 1.3905,
+      "theoretical_loss": 3.4390609919117745,
+      "tokens_seen": 2144796672
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003535548066121008,
+      "loss": 1.3435,
+      "theoretical_loss": 3.4390438522821705,
+      "tokens_seen": 2144927744
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003535146846413096,
+      "loss": 1.3569,
+      "theoretical_loss": 3.439026713993143,
+      "tokens_seen": 2145058816
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035347456267051836,
+      "loss": 1.3842,
+      "theoretical_loss": 3.439009577044505,
+      "tokens_seen": 2145189888
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035343444069972716,
+      "loss": 1.3014,
+      "theoretical_loss": 3.4389924414360706,
+      "tokens_seen": 2145320960
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035339431872893597,
+      "loss": 1.3094,
+      "theoretical_loss": 3.438975307167653,
+      "tokens_seen": 2145452032
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003533541967581448,
+      "loss": 1.3533,
+      "theoretical_loss": 3.4389581742390654,
+      "tokens_seen": 2145583104
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003533140747873536,
+      "loss": 1.3545,
+      "theoretical_loss": 3.4389410426501206,
+      "tokens_seen": 2145714176
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003532739528165624,
+      "loss": 1.4075,
+      "theoretical_loss": 3.438923912400633,
+      "tokens_seen": 2145845248
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035323383084577114,
+      "loss": 1.4296,
+      "theoretical_loss": 3.438906783490415,
+      "tokens_seen": 2145976320
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035319370887497995,
+      "loss": 1.2793,
+      "theoretical_loss": 3.438889655919281,
+      "tokens_seen": 2146107392
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035315358690418875,
+      "loss": 1.3695,
+      "theoretical_loss": 3.438872529687045,
+      "tokens_seen": 2146238464
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003531134649333975,
+      "loss": 1.3153,
+      "theoretical_loss": 3.438855404793519,
+      "tokens_seen": 2146369536
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035307334296260637,
+      "loss": 1.3704,
+      "theoretical_loss": 3.4388382812385174,
+      "tokens_seen": 2146500608
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003530332209918151,
+      "loss": 1.3245,
+      "theoretical_loss": 3.4388211590218543,
+      "tokens_seen": 2146631680
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003529930990210239,
+      "loss": 1.3739,
+      "theoretical_loss": 3.438804038143343,
+      "tokens_seen": 2146762752
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035295297705023273,
+      "loss": 1.3275,
+      "theoretical_loss": 3.438786918602797,
+      "tokens_seen": 2146893824
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003529128550794415,
+      "loss": 1.3575,
+      "theoretical_loss": 3.4387698004000304,
+      "tokens_seen": 2147024896
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003528727331086503,
+      "loss": 1.3527,
+      "theoretical_loss": 3.4387526835348567,
+      "tokens_seen": 2147155968
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003528326111378591,
+      "loss": 1.2825,
+      "theoretical_loss": 3.4387355680070897,
+      "tokens_seen": 2147287040
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003527924891670679,
+      "loss": 1.3574,
+      "theoretical_loss": 3.438718453816544,
+      "tokens_seen": 2147418112
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035275236719627665,
+      "loss": 1.3272,
+      "theoretical_loss": 3.438701340963033,
+      "tokens_seen": 2147549184
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003527122452254855,
+      "loss": 1.3884,
+      "theoretical_loss": 3.4386842294463706,
+      "tokens_seen": 2147680256
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035267212325469427,
+      "loss": 1.3174,
+      "theoretical_loss": 3.4386671192663707,
+      "tokens_seen": 2147811328
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035263200128390307,
+      "loss": 1.3679,
+      "theoretical_loss": 3.4386500104228475,
+      "tokens_seen": 2147942400
+    },
+    {
+      "epoch": 0.3,
+      "objective/train/advantage_avg": 0.494305282831192,
+      "objective/train/docs_used": 782164,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3467038869857788,
+      "objective/train/original_loss": 1.3467040061950684,
+      "objective/train/theoretical_loss": 3.4386329029156153,
+      "objective/train/tokens_used": 518532576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2445875108242035,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506740808486938,
+      "objective/train/weighted_lm_loss": 1.4148141145706177,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 1.0,
+      "theoretical_loss": 3.4386329029156153,
+      "tokens_seen": 2148073472
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003525918793131119,
+      "loss": 1.3948,
+      "theoretical_loss": 3.4386329029156153,
+      "tokens_seen": 2148073472
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035255175734232063,
+      "loss": 1.3827,
+      "theoretical_loss": 3.4386157967444877,
+      "tokens_seen": 2148204544
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003525116353715295,
+      "loss": 1.3449,
+      "theoretical_loss": 3.438598691909279,
+      "tokens_seen": 2148335616
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035247151340073824,
+      "loss": 1.3641,
+      "theoretical_loss": 3.4385815884098037,
+      "tokens_seen": 2148466688
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035243139142994705,
+      "loss": 1.345,
+      "theoretical_loss": 3.438564486245875,
+      "tokens_seen": 2148597760
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035239126945915586,
+      "loss": 1.3522,
+      "theoretical_loss": 3.4385473854173085,
+      "tokens_seen": 2148728832
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003523511474883646,
+      "loss": 1.3393,
+      "theoretical_loss": 3.4385302859239184,
+      "tokens_seen": 2148859904
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003523110255175734,
+      "loss": 1.3082,
+      "theoretical_loss": 3.438513187765518,
+      "tokens_seen": 2148990976
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003522709035467822,
+      "loss": 1.3373,
+      "theoretical_loss": 3.438496090941922,
+      "tokens_seen": 2149122048
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035223078157599103,
+      "loss": 1.3526,
+      "theoretical_loss": 3.4384789954529453,
+      "tokens_seen": 2149253120
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003521906596051998,
+      "loss": 1.3467,
+      "theoretical_loss": 3.438461901298402,
+      "tokens_seen": 2149384192
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035215053763440864,
+      "loss": 1.3559,
+      "theoretical_loss": 3.4384448084781063,
+      "tokens_seen": 2149515264
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003521104156636174,
+      "loss": 1.3569,
+      "theoretical_loss": 3.4384277169918733,
+      "tokens_seen": 2149646336
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003520702936928262,
+      "loss": 1.31,
+      "theoretical_loss": 3.4384106268395174,
+      "tokens_seen": 2149777408
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.000352030171722035,
+      "loss": 1.3635,
+      "theoretical_loss": 3.4383935380208523,
+      "tokens_seen": 2149908480
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035199004975124376,
+      "loss": 1.3978,
+      "theoretical_loss": 3.438376450535694,
+      "tokens_seen": 2150039552
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003519499277804526,
+      "loss": 1.3141,
+      "theoretical_loss": 3.438359364383856,
+      "tokens_seen": 2150170624
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035190980580966137,
+      "loss": 1.3075,
+      "theoretical_loss": 3.4383422795651537,
+      "tokens_seen": 2150301696
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003518696838388702,
+      "loss": 1.3342,
+      "theoretical_loss": 3.4383251960794015,
+      "tokens_seen": 2150432768
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.000351829561868079,
+      "loss": 1.3897,
+      "theoretical_loss": 3.4383081139264147,
+      "tokens_seen": 2150563840
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003517894398972878,
+      "loss": 1.3608,
+      "theoretical_loss": 3.4382910331060073,
+      "tokens_seen": 2150694912
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035174931792649654,
+      "loss": 1.3725,
+      "theoretical_loss": 3.438273953617995,
+      "tokens_seen": 2150825984
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035170919595570535,
+      "loss": 1.3945,
+      "theoretical_loss": 3.4382568754621916,
+      "tokens_seen": 2150957056
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035166907398491415,
+      "loss": 1.375,
+      "theoretical_loss": 3.4382397986384134,
+      "tokens_seen": 2151088128
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003516289520141229,
+      "loss": 1.4657,
+      "theoretical_loss": 3.438222723146474,
+      "tokens_seen": 2151219200
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035158883004333177,
+      "loss": 1.313,
+      "theoretical_loss": 3.438205648986189,
+      "tokens_seen": 2151350272
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003515487080725405,
+      "loss": 1.3097,
+      "theoretical_loss": 3.4381885761573736,
+      "tokens_seen": 2151481344
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003515085861017493,
+      "loss": 1.271,
+      "theoretical_loss": 3.4381715046598424,
+      "tokens_seen": 2151612416
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035146846413095813,
+      "loss": 1.3624,
+      "theoretical_loss": 3.438154434493411,
+      "tokens_seen": 2151743488
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003514283421601669,
+      "loss": 1.4337,
+      "theoretical_loss": 3.4381373656578944,
+      "tokens_seen": 2151874560
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035138822018937574,
+      "loss": 1.3013,
+      "theoretical_loss": 3.4381202981531076,
+      "tokens_seen": 2152005632
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003513480982185845,
+      "loss": 1.4015,
+      "theoretical_loss": 3.438103231978866,
+      "tokens_seen": 2152136704
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003513079762477933,
+      "loss": 1.3397,
+      "theoretical_loss": 3.4380861671349843,
+      "tokens_seen": 2152267776
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003512678542770021,
+      "loss": 1.3247,
+      "theoretical_loss": 3.4380691036212783,
+      "tokens_seen": 2152398848
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003512277323062109,
+      "loss": 1.2973,
+      "theoretical_loss": 3.4380520414375635,
+      "tokens_seen": 2152529920
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035118761033541967,
+      "loss": 1.3385,
+      "theoretical_loss": 3.438034980583655,
+      "tokens_seen": 2152660992
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035114748836462847,
+      "loss": 1.358,
+      "theoretical_loss": 3.438017921059368,
+      "tokens_seen": 2152792064
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003511073663938373,
+      "loss": 1.386,
+      "theoretical_loss": 3.4380008628645182,
+      "tokens_seen": 2152923136
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035106724442304603,
+      "loss": 1.3555,
+      "theoretical_loss": 3.4379838059989214,
+      "tokens_seen": 2153054208
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003510271224522549,
+      "loss": 1.2968,
+      "theoretical_loss": 3.437966750462392,
+      "tokens_seen": 2153185280
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00035098700048146364,
+      "loss": 1.3582,
+      "theoretical_loss": 3.437949696254747,
+      "tokens_seen": 2153316352
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00035094687851067245,
+      "loss": 1.354,
+      "theoretical_loss": 3.4379326433758006,
+      "tokens_seen": 2153447424
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00035090675653988126,
+      "loss": 1.3503,
+      "theoretical_loss": 3.4379155918253694,
+      "tokens_seen": 2153578496
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00035086663456909,
+      "loss": 1.3194,
+      "theoretical_loss": 3.437898541603269,
+      "tokens_seen": 2153709568
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00035082651259829887,
+      "loss": 1.3626,
+      "theoretical_loss": 3.437881492709314,
+      "tokens_seen": 2153840640
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003507863906275076,
+      "loss": 1.3979,
+      "theoretical_loss": 3.4378644451433216,
+      "tokens_seen": 2153971712
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003507462686567164,
+      "loss": 1.3382,
+      "theoretical_loss": 3.437847398905107,
+      "tokens_seen": 2154102784
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00035070614668592523,
+      "loss": 1.3337,
+      "theoretical_loss": 3.437830353994485,
+      "tokens_seen": 2154233856
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00035066602471513404,
+      "loss": 1.407,
+      "theoretical_loss": 3.4378133104112734,
+      "tokens_seen": 2154364928
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003506259027443428,
+      "loss": 1.3425,
+      "theoretical_loss": 3.4377962681552865,
+      "tokens_seen": 2154496000
+    },
+    {
+      "epoch": 0.31,
+      "objective/train/advantage_avg": 0.4932624399662018,
+      "objective/train/docs_used": 784476,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3405694961547852,
+      "objective/train/original_loss": 1.3405693769454956,
+      "objective/train/theoretical_loss": 3.4377792272263408,
+      "objective/train/tokens_used": 525086176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24378855526447296,
+      "objective/train/value_max": -0.5,
+      "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": 1.4083142280578613,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4377792272263408,
+      "tokens_seen": 2154627072
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003505857807735516,
+      "loss": 1.3315,
+      "theoretical_loss": 3.4377792272263408,
+      "tokens_seen": 2154627072
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003505456588027604,
+      "loss": 1.325,
+      "theoretical_loss": 3.4377621876242523,
+      "tokens_seen": 2154758144
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00035050553683196916,
+      "loss": 1.3954,
+      "theoretical_loss": 3.4377451493488365,
+      "tokens_seen": 2154889216
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.000350465414861178,
+      "loss": 1.3835,
+      "theoretical_loss": 3.4377281123999097,
+      "tokens_seen": 2155020288
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00035042529289038677,
+      "loss": 1.2876,
+      "theoretical_loss": 3.4377110767772887,
+      "tokens_seen": 2155151360
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003503851709195956,
+      "loss": 1.3386,
+      "theoretical_loss": 3.4376940424807882,
+      "tokens_seen": 2155282432
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003503450489488044,
+      "loss": 1.3175,
+      "theoretical_loss": 3.4376770095102254,
+      "tokens_seen": 2155413504
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003503049269780132,
+      "loss": 1.4363,
+      "theoretical_loss": 3.437659977865416,
+      "tokens_seen": 2155544576
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00035026480500722194,
+      "loss": 1.3553,
+      "theoretical_loss": 3.437642947546176,
+      "tokens_seen": 2155675648
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00035022468303643075,
+      "loss": 1.3236,
+      "theoretical_loss": 3.437625918552323,
+      "tokens_seen": 2155806720
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00035018456106563955,
+      "loss": 1.356,
+      "theoretical_loss": 3.4376088908836717,
+      "tokens_seen": 2155937792
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00035014443909484836,
+      "loss": 1.3687,
+      "theoretical_loss": 3.4375918645400385,
+      "tokens_seen": 2156068864
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00035010431712405717,
+      "loss": 1.3058,
+      "theoretical_loss": 3.4375748395212407,
+      "tokens_seen": 2156199936
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003500641951532659,
+      "loss": 1.2525,
+      "theoretical_loss": 3.437557815827094,
+      "tokens_seen": 2156331008
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003500240731824748,
+      "loss": 1.3054,
+      "theoretical_loss": 3.437540793457415,
+      "tokens_seen": 2156462080
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034998395121168353,
+      "loss": 1.3529,
+      "theoretical_loss": 3.4375237724120202,
+      "tokens_seen": 2156593152
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003499438292408923,
+      "loss": 1.3369,
+      "theoretical_loss": 3.4375067526907266,
+      "tokens_seen": 2156724224
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034990370727010114,
+      "loss": 1.3922,
+      "theoretical_loss": 3.43748973429335,
+      "tokens_seen": 2156855296
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003498635852993099,
+      "loss": 1.2878,
+      "theoretical_loss": 3.437472717219707,
+      "tokens_seen": 2156986368
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003498234633285187,
+      "loss": 1.3126,
+      "theoretical_loss": 3.4374557014696143,
+      "tokens_seen": 2157117440
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003497833413577275,
+      "loss": 1.4097,
+      "theoretical_loss": 3.4374386870428886,
+      "tokens_seen": 2157248512
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003497432193869363,
+      "loss": 1.3917,
+      "theoretical_loss": 3.4374216739393466,
+      "tokens_seen": 2157379584
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034970309741614507,
+      "loss": 1.4311,
+      "theoretical_loss": 3.437404662158806,
+      "tokens_seen": 2157510656
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034966297544535387,
+      "loss": 1.3259,
+      "theoretical_loss": 3.4373876517010813,
+      "tokens_seen": 2157641728
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003496228534745627,
+      "loss": 1.3653,
+      "theoretical_loss": 3.437370642565991,
+      "tokens_seen": 2157772800
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034958273150377143,
+      "loss": 1.3832,
+      "theoretical_loss": 3.437353634753352,
+      "tokens_seen": 2157903872
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003495426095329803,
+      "loss": 1.343,
+      "theoretical_loss": 3.4373366282629805,
+      "tokens_seen": 2158034944
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034950248756218904,
+      "loss": 1.3356,
+      "theoretical_loss": 3.437319623094693,
+      "tokens_seen": 2158166016
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003494623655913979,
+      "loss": 1.3427,
+      "theoretical_loss": 3.4373026192483076,
+      "tokens_seen": 2158297088
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034942224362060666,
+      "loss": 1.393,
+      "theoretical_loss": 3.4372856167236403,
+      "tokens_seen": 2158428160
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003493821216498154,
+      "loss": 1.3821,
+      "theoretical_loss": 3.4372686155205088,
+      "tokens_seen": 2158559232
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034934199967902427,
+      "loss": 1.3578,
+      "theoretical_loss": 3.43725161563873,
+      "tokens_seen": 2158690304
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.000349301877708233,
+      "loss": 1.2903,
+      "theoretical_loss": 3.4372346170781203,
+      "tokens_seen": 2158821376
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003492617557374418,
+      "loss": 1.3351,
+      "theoretical_loss": 3.437217619838498,
+      "tokens_seen": 2158952448
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034922163376665063,
+      "loss": 1.3762,
+      "theoretical_loss": 3.437200623919679,
+      "tokens_seen": 2159083520
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034918151179585944,
+      "loss": 1.341,
+      "theoretical_loss": 3.4371836293214812,
+      "tokens_seen": 2159214592
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003491413898250682,
+      "loss": 1.364,
+      "theoretical_loss": 3.437166636043722,
+      "tokens_seen": 2159345664
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.000349101267854277,
+      "loss": 1.4009,
+      "theoretical_loss": 3.4371496440862184,
+      "tokens_seen": 2159476736
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003490611458834858,
+      "loss": 1.3426,
+      "theoretical_loss": 3.437132653448787,
+      "tokens_seen": 2159607808
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034902102391269456,
+      "loss": 1.3346,
+      "theoretical_loss": 3.4371156641312464,
+      "tokens_seen": 2159738880
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003489809019419034,
+      "loss": 1.3792,
+      "theoretical_loss": 3.4370986761334135,
+      "tokens_seen": 2159869952
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034894077997111217,
+      "loss": 1.3447,
+      "theoretical_loss": 3.4370816894551055,
+      "tokens_seen": 2160001024
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034890065800032103,
+      "loss": 1.3987,
+      "theoretical_loss": 3.43706470409614,
+      "tokens_seen": 2160132096
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003488605360295298,
+      "loss": 1.4152,
+      "theoretical_loss": 3.4370477200563343,
+      "tokens_seen": 2160263168
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003488204140587386,
+      "loss": 1.3714,
+      "theoretical_loss": 3.437030737335506,
+      "tokens_seen": 2160394240
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003487802920879474,
+      "loss": 1.3416,
+      "theoretical_loss": 3.4370137559334726,
+      "tokens_seen": 2160525312
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034874017011715615,
+      "loss": 1.4786,
+      "theoretical_loss": 3.436996775850052,
+      "tokens_seen": 2160656384
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034870004814636495,
+      "loss": 1.3692,
+      "theoretical_loss": 3.4369797970850615,
+      "tokens_seen": 2160787456
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034865992617557376,
+      "loss": 1.3761,
+      "theoretical_loss": 3.436962819638319,
+      "tokens_seen": 2160918528
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034861980420478256,
+      "loss": 1.2693,
+      "theoretical_loss": 3.436945843509642,
+      "tokens_seen": 2161049600
+    },
+    {
+      "epoch": 0.31,
+      "objective/train/advantage_avg": 0.4920182228088379,
+      "objective/train/docs_used": 786923,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.153581142425537,
+      "objective/train/original_loss": 1.153581142425537,
+      "objective/train/theoretical_loss": 3.436928868698848,
+      "objective/train/tokens_used": 531639776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24248407781124115,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504344701766968,
+      "objective/train/weighted_lm_loss": 1.2118184566497803,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9834713935852051,
+      "theoretical_loss": 3.436928868698848,
+      "tokens_seen": 2161180672
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003485796822339913,
+      "loss": 1.3085,
+      "theoretical_loss": 3.436928868698848,
+      "tokens_seen": 2161180672
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003485395602632002,
+      "loss": 1.3817,
+      "theoretical_loss": 3.436911895205756,
+      "tokens_seen": 2161311744
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034849943829240893,
+      "loss": 1.2576,
+      "theoretical_loss": 3.436894923030182,
+      "tokens_seen": 2161442816
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003484593163216177,
+      "loss": 1.2803,
+      "theoretical_loss": 3.4368779521719452,
+      "tokens_seen": 2161573888
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034841919435082654,
+      "loss": 1.3665,
+      "theoretical_loss": 3.4368609826308627,
+      "tokens_seen": 2161704960
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003483790723800353,
+      "loss": 1.2684,
+      "theoretical_loss": 3.436844014406753,
+      "tokens_seen": 2161836032
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003483389504092441,
+      "loss": 1.4346,
+      "theoretical_loss": 3.436827047499434,
+      "tokens_seen": 2161967104
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003482988284384529,
+      "loss": 1.3732,
+      "theoretical_loss": 3.4368100819087233,
+      "tokens_seen": 2162098176
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003482587064676617,
+      "loss": 1.366,
+      "theoretical_loss": 3.436793117634439,
+      "tokens_seen": 2162229248
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003482185844968705,
+      "loss": 1.3223,
+      "theoretical_loss": 3.4367761546763993,
+      "tokens_seen": 2162360320
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034817846252607927,
+      "loss": 1.3666,
+      "theoretical_loss": 3.4367591930344226,
+      "tokens_seen": 2162491392
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003481383405552881,
+      "loss": 1.314,
+      "theoretical_loss": 3.436742232708326,
+      "tokens_seen": 2162622464
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003480982185844969,
+      "loss": 1.3496,
+      "theoretical_loss": 3.4367252736979292,
+      "tokens_seen": 2162753536
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003480580966137057,
+      "loss": 1.408,
+      "theoretical_loss": 3.436708316003049,
+      "tokens_seen": 2162884608
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034801797464291444,
+      "loss": 1.3626,
+      "theoretical_loss": 3.4366913596235045,
+      "tokens_seen": 2163015680
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003479778526721233,
+      "loss": 1.3073,
+      "theoretical_loss": 3.4366744045591138,
+      "tokens_seen": 2163146752
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034793773070133206,
+      "loss": 1.357,
+      "theoretical_loss": 3.4366574508096948,
+      "tokens_seen": 2163277824
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003478976087305408,
+      "loss": 1.347,
+      "theoretical_loss": 3.4366404983750667,
+      "tokens_seen": 2163408896
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034785748675974967,
+      "loss": 1.2942,
+      "theoretical_loss": 3.436623547255047,
+      "tokens_seen": 2163539968
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003478173647889584,
+      "loss": 1.3428,
+      "theoretical_loss": 3.4366065974494546,
+      "tokens_seen": 2163671040
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003477772428181672,
+      "loss": 1.3426,
+      "theoretical_loss": 3.4365896489581074,
+      "tokens_seen": 2163802112
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034773712084737603,
+      "loss": 1.3772,
+      "theoretical_loss": 3.436572701780825,
+      "tokens_seen": 2163933184
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034769699887658484,
+      "loss": 1.4832,
+      "theoretical_loss": 3.436555755917425,
+      "tokens_seen": 2164064256
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003476568769057936,
+      "loss": 1.3652,
+      "theoretical_loss": 3.436538811367726,
+      "tokens_seen": 2164195328
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003476167549350024,
+      "loss": 1.3777,
+      "theoretical_loss": 3.436521868131547,
+      "tokens_seen": 2164326400
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003475766329642112,
+      "loss": 1.4403,
+      "theoretical_loss": 3.436504926208707,
+      "tokens_seen": 2164457472
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034753651099342,
+      "loss": 1.3308,
+      "theoretical_loss": 3.4364879855990234,
+      "tokens_seen": 2164588544
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003474963890226288,
+      "loss": 1.3318,
+      "theoretical_loss": 3.436471046302316,
+      "tokens_seen": 2164719616
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034745626705183757,
+      "loss": 1.3354,
+      "theoretical_loss": 3.4364541083184035,
+      "tokens_seen": 2164850688
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034741614508104643,
+      "loss": 1.3065,
+      "theoretical_loss": 3.4364371716471043,
+      "tokens_seen": 2164981760
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003473760231102552,
+      "loss": 1.3895,
+      "theoretical_loss": 3.436420236288237,
+      "tokens_seen": 2165112832
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034733590113946393,
+      "loss": 1.3963,
+      "theoretical_loss": 3.436403302241621,
+      "tokens_seen": 2165243904
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003472957791686728,
+      "loss": 1.4312,
+      "theoretical_loss": 3.436386369507075,
+      "tokens_seen": 2165374976
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034725565719788155,
+      "loss": 1.3451,
+      "theoretical_loss": 3.4363694380844176,
+      "tokens_seen": 2165506048
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034721553522709035,
+      "loss": 1.2882,
+      "theoretical_loss": 3.4363525079734685,
+      "tokens_seen": 2165637120
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034717541325629916,
+      "loss": 1.4359,
+      "theoretical_loss": 3.4363355791740458,
+      "tokens_seen": 2165768192
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034713529128550796,
+      "loss": 1.3219,
+      "theoretical_loss": 3.436318651685969,
+      "tokens_seen": 2165899264
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003470951693147167,
+      "loss": 1.3773,
+      "theoretical_loss": 3.4363017255090575,
+      "tokens_seen": 2166030336
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003470550473439256,
+      "loss": 1.3158,
+      "theoretical_loss": 3.43628480064313,
+      "tokens_seen": 2166161408
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034701492537313433,
+      "loss": 1.3796,
+      "theoretical_loss": 3.436267877088006,
+      "tokens_seen": 2166292480
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034697480340234314,
+      "loss": 1.3454,
+      "theoretical_loss": 3.436250954843504,
+      "tokens_seen": 2166423552
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034693468143155194,
+      "loss": 1.2861,
+      "theoretical_loss": 3.436234033909444,
+      "tokens_seen": 2166554624
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003468945594607607,
+      "loss": 1.3388,
+      "theoretical_loss": 3.4362171142856446,
+      "tokens_seen": 2166685696
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034685443748996955,
+      "loss": 1.3969,
+      "theoretical_loss": 3.4362001959719253,
+      "tokens_seen": 2166816768
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003468143155191783,
+      "loss": 1.3662,
+      "theoretical_loss": 3.4361832789681057,
+      "tokens_seen": 2166947840
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003467741935483871,
+      "loss": 1.3941,
+      "theoretical_loss": 3.4361663632740047,
+      "tokens_seen": 2167078912
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003467340715775959,
+      "loss": 1.3316,
+      "theoretical_loss": 3.4361494488894424,
+      "tokens_seen": 2167209984
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034669394960680467,
+      "loss": 1.2969,
+      "theoretical_loss": 3.436132535814237,
+      "tokens_seen": 2167341056
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003466538276360135,
+      "loss": 1.3625,
+      "theoretical_loss": 3.4361156240482096,
+      "tokens_seen": 2167472128
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003466137056652223,
+      "loss": 1.429,
+      "theoretical_loss": 3.4360987135911785,
+      "tokens_seen": 2167603200
+    },
+    {
+      "epoch": 0.31,
+      "objective/train/advantage_avg": 0.49082377552986145,
+      "objective/train/docs_used": 789244,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2739579677581787,
+      "objective/train/original_loss": 1.2739579677581787,
+      "objective/train/theoretical_loss": 3.436081804442963,
+      "objective/train/tokens_used": 538193376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2417551577091217,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503113269805908,
+      "objective/train/weighted_lm_loss": 1.3380993604660034,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.436081804442963,
+      "tokens_seen": 2167734272
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003465735836944311,
+      "loss": 1.3457,
+      "theoretical_loss": 3.436081804442963,
+      "tokens_seen": 2167734272
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034653346172363984,
+      "loss": 1.2903,
+      "theoretical_loss": 3.4360648966033835,
+      "tokens_seen": 2167865344
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003464933397528487,
+      "loss": 1.4347,
+      "theoretical_loss": 3.4360479900722596,
+      "tokens_seen": 2167996416
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034645321778205745,
+      "loss": 1.4319,
+      "theoretical_loss": 3.436031084849411,
+      "tokens_seen": 2168127488
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003464130958112662,
+      "loss": 1.3685,
+      "theoretical_loss": 3.4360141809346567,
+      "tokens_seen": 2168258560
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034637297384047507,
+      "loss": 1.3392,
+      "theoretical_loss": 3.4359972783278163,
+      "tokens_seen": 2168389632
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003463328518696838,
+      "loss": 1.3125,
+      "theoretical_loss": 3.4359803770287107,
+      "tokens_seen": 2168520704
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003462927298988927,
+      "loss": 1.3936,
+      "theoretical_loss": 3.435963477037159,
+      "tokens_seen": 2168651776
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034625260792810143,
+      "loss": 1.3199,
+      "theoretical_loss": 3.435946578352981,
+      "tokens_seen": 2168782848
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034621248595731024,
+      "loss": 1.3501,
+      "theoretical_loss": 3.4359296809759963,
+      "tokens_seen": 2168913920
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034617236398651904,
+      "loss": 1.4054,
+      "theoretical_loss": 3.4359127849060256,
+      "tokens_seen": 2169044992
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003461322420157278,
+      "loss": 1.2944,
+      "theoretical_loss": 3.4358958901428878,
+      "tokens_seen": 2169176064
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003460921200449366,
+      "loss": 1.3698,
+      "theoretical_loss": 3.435878996686404,
+      "tokens_seen": 2169307136
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003460519980741454,
+      "loss": 1.3417,
+      "theoretical_loss": 3.435862104536393,
+      "tokens_seen": 2169438208
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003460118761033542,
+      "loss": 1.4059,
+      "theoretical_loss": 3.435845213692676,
+      "tokens_seen": 2169569280
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034597175413256297,
+      "loss": 1.2733,
+      "theoretical_loss": 3.4358283241550724,
+      "tokens_seen": 2169700352
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034593163216177183,
+      "loss": 1.3038,
+      "theoretical_loss": 3.435811435923402,
+      "tokens_seen": 2169831424
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003458915101909806,
+      "loss": 1.3281,
+      "theoretical_loss": 3.4357945489974857,
+      "tokens_seen": 2169962496
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034585138822018933,
+      "loss": 1.3437,
+      "theoretical_loss": 3.4357776633771433,
+      "tokens_seen": 2170093568
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003458112662493982,
+      "loss": 1.3905,
+      "theoretical_loss": 3.4357607790621953,
+      "tokens_seen": 2170224640
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034577114427860695,
+      "loss": 1.402,
+      "theoretical_loss": 3.4357438960524616,
+      "tokens_seen": 2170355712
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003457310223078158,
+      "loss": 1.378,
+      "theoretical_loss": 3.4357270143477625,
+      "tokens_seen": 2170486784
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034569090033702456,
+      "loss": 1.3997,
+      "theoretical_loss": 3.4357101339479184,
+      "tokens_seen": 2170617856
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034565077836623336,
+      "loss": 1.4222,
+      "theoretical_loss": 3.4356932548527497,
+      "tokens_seen": 2170748928
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034561065639544217,
+      "loss": 1.421,
+      "theoretical_loss": 3.435676377062077,
+      "tokens_seen": 2170880000
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.000345570534424651,
+      "loss": 1.3733,
+      "theoretical_loss": 3.4356595005757202,
+      "tokens_seen": 2171011072
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034553041245385973,
+      "loss": 1.3801,
+      "theoretical_loss": 3.4356426253935,
+      "tokens_seen": 2171142144
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034549029048306854,
+      "loss": 1.3632,
+      "theoretical_loss": 3.4356257515152375,
+      "tokens_seen": 2171273216
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034545016851227734,
+      "loss": 1.3607,
+      "theoretical_loss": 3.4356088789407524,
+      "tokens_seen": 2171404288
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003454100465414861,
+      "loss": 1.4386,
+      "theoretical_loss": 3.4355920076698654,
+      "tokens_seen": 2171535360
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034536992457069495,
+      "loss": 1.3747,
+      "theoretical_loss": 3.435575137702398,
+      "tokens_seen": 2171666432
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003453298025999037,
+      "loss": 1.4039,
+      "theoretical_loss": 3.4355582690381694,
+      "tokens_seen": 2171797504
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003452896806291125,
+      "loss": 1.4278,
+      "theoretical_loss": 3.4355414016770016,
+      "tokens_seen": 2171928576
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003452495586583213,
+      "loss": 1.3782,
+      "theoretical_loss": 3.4355245356187143,
+      "tokens_seen": 2172059648
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034520943668753007,
+      "loss": 1.3473,
+      "theoretical_loss": 3.435507670863129,
+      "tokens_seen": 2172190720
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003451693147167389,
+      "loss": 1.3534,
+      "theoretical_loss": 3.4354908074100656,
+      "tokens_seen": 2172321792
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003451291927459477,
+      "loss": 1.3076,
+      "theoretical_loss": 3.435473945259346,
+      "tokens_seen": 2172452864
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003450890707751565,
+      "loss": 1.3805,
+      "theoretical_loss": 3.4354570844107903,
+      "tokens_seen": 2172583936
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003450489488043653,
+      "loss": 1.2917,
+      "theoretical_loss": 3.4354402248642195,
+      "tokens_seen": 2172715008
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003450088268335741,
+      "loss": 1.3891,
+      "theoretical_loss": 3.435423366619455,
+      "tokens_seen": 2172846080
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034496870486278285,
+      "loss": 1.3262,
+      "theoretical_loss": 3.4354065096763176,
+      "tokens_seen": 2172977152
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034492858289199166,
+      "loss": 1.203,
+      "theoretical_loss": 3.4353896540346276,
+      "tokens_seen": 2173108224
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034488846092120047,
+      "loss": 1.2733,
+      "theoretical_loss": 3.4353727996942065,
+      "tokens_seen": 2173239296
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003448483389504092,
+      "loss": 1.4116,
+      "theoretical_loss": 3.4353559466548758,
+      "tokens_seen": 2173370368
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003448082169796181,
+      "loss": 1.3272,
+      "theoretical_loss": 3.435339094916456,
+      "tokens_seen": 2173501440
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034476809500882683,
+      "loss": 1.354,
+      "theoretical_loss": 3.435322244478769,
+      "tokens_seen": 2173632512
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034472797303803564,
+      "loss": 1.3525,
+      "theoretical_loss": 3.4353053953416346,
+      "tokens_seen": 2173763584
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034468785106724444,
+      "loss": 1.3149,
+      "theoretical_loss": 3.435288547504875,
+      "tokens_seen": 2173894656
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003446477290964532,
+      "loss": 1.3166,
+      "theoretical_loss": 3.435271700968312,
+      "tokens_seen": 2174025728
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.000344607607125662,
+      "loss": 1.3359,
+      "theoretical_loss": 3.4352548557317655,
+      "tokens_seen": 2174156800
+    },
+    {
+      "epoch": 0.32,
+      "objective/train/advantage_avg": 0.4926750063896179,
+      "objective/train/docs_used": 791682,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3083956241607666,
+      "objective/train/original_loss": 1.3083957433700562,
+      "objective/train/theoretical_loss": 3.435238011795058,
+      "objective/train/tokens_used": 544746976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24315457046031952,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505037307739258,
+      "objective/train/weighted_lm_loss": 1.37453031539917,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.435238011795058,
+      "tokens_seen": 2174287872
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003445674851548708,
+      "loss": 1.3981,
+      "theoretical_loss": 3.435238011795058,
+      "tokens_seen": 2174287872
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003445273631840796,
+      "loss": 1.3494,
+      "theoretical_loss": 3.4352211691580097,
+      "tokens_seen": 2174418944
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034448724121328837,
+      "loss": 1.3818,
+      "theoretical_loss": 3.435204327820443,
+      "tokens_seen": 2174550016
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034444711924249723,
+      "loss": 1.4043,
+      "theoretical_loss": 3.435187487782179,
+      "tokens_seen": 2174681088
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.000344406997271706,
+      "loss": 1.3072,
+      "theoretical_loss": 3.435170649043039,
+      "tokens_seen": 2174812160
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003443668753009148,
+      "loss": 1.387,
+      "theoretical_loss": 3.4351538116028446,
+      "tokens_seen": 2174943232
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003443267533301236,
+      "loss": 1.3685,
+      "theoretical_loss": 3.4351369754614174,
+      "tokens_seen": 2175074304
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034428663135933234,
+      "loss": 1.3537,
+      "theoretical_loss": 3.4351201406185785,
+      "tokens_seen": 2175205376
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003442465093885412,
+      "loss": 1.2844,
+      "theoretical_loss": 3.4351033070741503,
+      "tokens_seen": 2175336448
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034420638741774996,
+      "loss": 1.3052,
+      "theoretical_loss": 3.435086474827954,
+      "tokens_seen": 2175467520
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034416626544695876,
+      "loss": 1.2418,
+      "theoretical_loss": 3.435069643879811,
+      "tokens_seen": 2175598592
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034412614347616757,
+      "loss": 1.3648,
+      "theoretical_loss": 3.4350528142295436,
+      "tokens_seen": 2175729664
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003440860215053764,
+      "loss": 1.3099,
+      "theoretical_loss": 3.435035985876973,
+      "tokens_seen": 2175860736
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034404589953458513,
+      "loss": 1.3628,
+      "theoretical_loss": 3.4350191588219214,
+      "tokens_seen": 2175991808
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034400577756379393,
+      "loss": 1.2747,
+      "theoretical_loss": 3.4350023330642103,
+      "tokens_seen": 2176122880
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034396565559300274,
+      "loss": 1.2591,
+      "theoretical_loss": 3.4349855086036616,
+      "tokens_seen": 2176253952
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003439255336222115,
+      "loss": 1.3164,
+      "theoretical_loss": 3.4349686854400976,
+      "tokens_seen": 2176385024
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034388541165142035,
+      "loss": 1.408,
+      "theoretical_loss": 3.4349518635733394,
+      "tokens_seen": 2176516096
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003438452896806291,
+      "loss": 1.3347,
+      "theoretical_loss": 3.4349350430032097,
+      "tokens_seen": 2176647168
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034380516770983797,
+      "loss": 1.2746,
+      "theoretical_loss": 3.4349182237295297,
+      "tokens_seen": 2176778240
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003437650457390467,
+      "loss": 1.3478,
+      "theoretical_loss": 3.4349014057521225,
+      "tokens_seen": 2176909312
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034372492376825547,
+      "loss": 1.4248,
+      "theoretical_loss": 3.4348845890708093,
+      "tokens_seen": 2177040384
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034368480179746433,
+      "loss": 1.3722,
+      "theoretical_loss": 3.4348677736854123,
+      "tokens_seen": 2177171456
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003436446798266731,
+      "loss": 1.4103,
+      "theoretical_loss": 3.4348509595957544,
+      "tokens_seen": 2177302528
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003436045578558819,
+      "loss": 1.4181,
+      "theoretical_loss": 3.4348341468016566,
+      "tokens_seen": 2177433600
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003435644358850907,
+      "loss": 1.3551,
+      "theoretical_loss": 3.4348173353029416,
+      "tokens_seen": 2177564672
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003435243139142995,
+      "loss": 1.4055,
+      "theoretical_loss": 3.434800525099432,
+      "tokens_seen": 2177695744
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034348419194350825,
+      "loss": 1.3066,
+      "theoretical_loss": 3.4347837161909496,
+      "tokens_seen": 2177826816
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034344406997271706,
+      "loss": 1.3712,
+      "theoretical_loss": 3.4347669085773167,
+      "tokens_seen": 2177957888
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034340394800192587,
+      "loss": 1.3354,
+      "theoretical_loss": 3.4347501022583558,
+      "tokens_seen": 2178088960
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003433638260311346,
+      "loss": 1.3411,
+      "theoretical_loss": 3.4347332972338895,
+      "tokens_seen": 2178220032
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003433237040603435,
+      "loss": 1.3293,
+      "theoretical_loss": 3.4347164935037395,
+      "tokens_seen": 2178351104
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034328358208955223,
+      "loss": 1.2624,
+      "theoretical_loss": 3.434699691067729,
+      "tokens_seen": 2178482176
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034324346011876104,
+      "loss": 1.3883,
+      "theoretical_loss": 3.4346828899256803,
+      "tokens_seen": 2178613248
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034320333814796984,
+      "loss": 1.3494,
+      "theoretical_loss": 3.4346660900774157,
+      "tokens_seen": 2178744320
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003431632161771786,
+      "loss": 1.2752,
+      "theoretical_loss": 3.4346492915227573,
+      "tokens_seen": 2178875392
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034312309420638746,
+      "loss": 1.3192,
+      "theoretical_loss": 3.4346324942615287,
+      "tokens_seen": 2179006464
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003430829722355962,
+      "loss": 1.3642,
+      "theoretical_loss": 3.4346156982935523,
+      "tokens_seen": 2179137536
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.000343042850264805,
+      "loss": 1.3014,
+      "theoretical_loss": 3.43459890361865,
+      "tokens_seen": 2179268608
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003430027282940138,
+      "loss": 1.3746,
+      "theoretical_loss": 3.4345821102366454,
+      "tokens_seen": 2179399680
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034296260632322263,
+      "loss": 1.319,
+      "theoretical_loss": 3.4345653181473605,
+      "tokens_seen": 2179530752
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003429224843524314,
+      "loss": 1.3585,
+      "theoretical_loss": 3.434548527350618,
+      "tokens_seen": 2179661824
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003428823623816402,
+      "loss": 1.3415,
+      "theoretical_loss": 3.4345317378462417,
+      "tokens_seen": 2179792896
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.000342842240410849,
+      "loss": 1.312,
+      "theoretical_loss": 3.4345149496340532,
+      "tokens_seen": 2179923968
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034280211844005774,
+      "loss": 1.3651,
+      "theoretical_loss": 3.4344981627138766,
+      "tokens_seen": 2180055040
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003427619964692666,
+      "loss": 1.4635,
+      "theoretical_loss": 3.4344813770855334,
+      "tokens_seen": 2180186112
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034272187449847536,
+      "loss": 1.3709,
+      "theoretical_loss": 3.4344645927488475,
+      "tokens_seen": 2180317184
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034268175252768416,
+      "loss": 1.3913,
+      "theoretical_loss": 3.434447809703642,
+      "tokens_seen": 2180448256
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034264163055689297,
+      "loss": 1.3985,
+      "theoretical_loss": 3.4344310279497394,
+      "tokens_seen": 2180579328
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003426015085861017,
+      "loss": 1.3732,
+      "theoretical_loss": 3.4344142474869623,
+      "tokens_seen": 2180710400
+    },
+    {
+      "epoch": 0.32,
+      "objective/train/advantage_avg": 0.49382683634757996,
+      "objective/train/docs_used": 794081,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2644366025924683,
+      "objective/train/original_loss": 1.2644366025924683,
+      "objective/train/theoretical_loss": 3.434397468315135,
+      "objective/train/tokens_used": 551300576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2445559948682785,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506260395050049,
+      "objective/train/weighted_lm_loss": 1.3283647298812866,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.434397468315135,
+      "tokens_seen": 2180841472
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034256138661531053,
+      "loss": 1.3475,
+      "theoretical_loss": 3.434397468315135,
+      "tokens_seen": 2180841472
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034252126464451933,
+      "loss": 1.3321,
+      "theoretical_loss": 3.4343806904340797,
+      "tokens_seen": 2180972544
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034248114267372814,
+      "loss": 1.3681,
+      "theoretical_loss": 3.4343639138436197,
+      "tokens_seen": 2181103616
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034244102070293695,
+      "loss": 1.3197,
+      "theoretical_loss": 3.434347138543578,
+      "tokens_seen": 2181234688
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034240089873214575,
+      "loss": 1.2787,
+      "theoretical_loss": 3.4343303645337784,
+      "tokens_seen": 2181365760
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003423607767613545,
+      "loss": 1.3967,
+      "theoretical_loss": 3.434313591814044,
+      "tokens_seen": 2181496832
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034232065479056337,
+      "loss": 1.3051,
+      "theoretical_loss": 3.4342968203841977,
+      "tokens_seen": 2181627904
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003422805328197721,
+      "loss": 1.4292,
+      "theoretical_loss": 3.434280050244063,
+      "tokens_seen": 2181758976
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034224041084898087,
+      "loss": 1.3019,
+      "theoretical_loss": 3.434263281393463,
+      "tokens_seen": 2181890048
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034220028887818973,
+      "loss": 1.3877,
+      "theoretical_loss": 3.4342465138322216,
+      "tokens_seen": 2182021120
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003421601669073985,
+      "loss": 1.286,
+      "theoretical_loss": 3.434229747560162,
+      "tokens_seen": 2182152192
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003421200449366073,
+      "loss": 1.31,
+      "theoretical_loss": 3.4342129825771073,
+      "tokens_seen": 2182283264
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003420799229658161,
+      "loss": 1.3544,
+      "theoretical_loss": 3.4341962188828816,
+      "tokens_seen": 2182414336
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003420398009950249,
+      "loss": 1.3781,
+      "theoretical_loss": 3.434179456477308,
+      "tokens_seen": 2182545408
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034199967902423365,
+      "loss": 1.3904,
+      "theoretical_loss": 3.4341626953602105,
+      "tokens_seen": 2182676480
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034195955705344246,
+      "loss": 1.3221,
+      "theoretical_loss": 3.434145935531412,
+      "tokens_seen": 2182807552
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034191943508265127,
+      "loss": 1.4036,
+      "theoretical_loss": 3.4341291769907367,
+      "tokens_seen": 2182938624
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034187931311186007,
+      "loss": 1.3379,
+      "theoretical_loss": 3.4341124197380077,
+      "tokens_seen": 2183069696
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003418391911410689,
+      "loss": 1.297,
+      "theoretical_loss": 3.4340956637730495,
+      "tokens_seen": 2183200768
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034179906917027763,
+      "loss": 1.3324,
+      "theoretical_loss": 3.4340789090956854,
+      "tokens_seen": 2183331840
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003417589471994865,
+      "loss": 1.4004,
+      "theoretical_loss": 3.434062155705739,
+      "tokens_seen": 2183462912
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034171882522869524,
+      "loss": 1.3392,
+      "theoretical_loss": 3.4340454036030343,
+      "tokens_seen": 2183593984
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.000341678703257904,
+      "loss": 1.3585,
+      "theoretical_loss": 3.434028652787395,
+      "tokens_seen": 2183725056
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034163858128711286,
+      "loss": 1.346,
+      "theoretical_loss": 3.434011903258645,
+      "tokens_seen": 2183856128
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003415984593163216,
+      "loss": 1.3332,
+      "theoretical_loss": 3.433995155016609,
+      "tokens_seen": 2183987200
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003415583373455304,
+      "loss": 1.4337,
+      "theoretical_loss": 3.433978408061109,
+      "tokens_seen": 2184118272
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003415182153747392,
+      "loss": 1.3684,
+      "theoretical_loss": 3.433961662391971,
+      "tokens_seen": 2184249344
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034147809340394803,
+      "loss": 1.4139,
+      "theoretical_loss": 3.433944918009018,
+      "tokens_seen": 2184380416
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003414379714331568,
+      "loss": 1.3795,
+      "theoretical_loss": 3.4339281749120745,
+      "tokens_seen": 2184511488
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003413978494623656,
+      "loss": 1.403,
+      "theoretical_loss": 3.433911433100964,
+      "tokens_seen": 2184642560
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003413577274915744,
+      "loss": 1.3372,
+      "theoretical_loss": 3.433894692575511,
+      "tokens_seen": 2184773632
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034131760552078314,
+      "loss": 1.3492,
+      "theoretical_loss": 3.4338779533355397,
+      "tokens_seen": 2184904704
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.000341277483549992,
+      "loss": 1.326,
+      "theoretical_loss": 3.4338612153808743,
+      "tokens_seen": 2185035776
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034123736157920076,
+      "loss": 1.3173,
+      "theoretical_loss": 3.4338444787113387,
+      "tokens_seen": 2185166848
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003411972396084096,
+      "loss": 1.4115,
+      "theoretical_loss": 3.4338277433267574,
+      "tokens_seen": 2185297920
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034115711763761837,
+      "loss": 1.3737,
+      "theoretical_loss": 3.4338110092269547,
+      "tokens_seen": 2185428992
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003411169956668271,
+      "loss": 1.3348,
+      "theoretical_loss": 3.4337942764117546,
+      "tokens_seen": 2185560064
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.000341076873696036,
+      "loss": 1.3975,
+      "theoretical_loss": 3.4337775448809817,
+      "tokens_seen": 2185691136
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034103675172524473,
+      "loss": 1.385,
+      "theoretical_loss": 3.4337608146344607,
+      "tokens_seen": 2185822208
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034099662975445354,
+      "loss": 1.404,
+      "theoretical_loss": 3.4337440856720156,
+      "tokens_seen": 2185953280
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034095650778366235,
+      "loss": 1.372,
+      "theoretical_loss": 3.433727357993471,
+      "tokens_seen": 2186084352
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034091638581287115,
+      "loss": 1.4376,
+      "theoretical_loss": 3.433710631598651,
+      "tokens_seen": 2186215424
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003408762638420799,
+      "loss": 1.3801,
+      "theoretical_loss": 3.433693906487381,
+      "tokens_seen": 2186346496
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00034083614187128877,
+      "loss": 1.4143,
+      "theoretical_loss": 3.433677182659485,
+      "tokens_seen": 2186477568
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003407960199004975,
+      "loss": 1.3159,
+      "theoretical_loss": 3.4336604601147878,
+      "tokens_seen": 2186608640
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00034075589792970627,
+      "loss": 1.328,
+      "theoretical_loss": 3.4336437388531134,
+      "tokens_seen": 2186739712
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00034071577595891513,
+      "loss": 1.415,
+      "theoretical_loss": 3.4336270188742875,
+      "tokens_seen": 2186870784
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003406756539881239,
+      "loss": 1.3363,
+      "theoretical_loss": 3.4336103001781337,
+      "tokens_seen": 2187001856
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00034063553201733274,
+      "loss": 1.4398,
+      "theoretical_loss": 3.4335935827644777,
+      "tokens_seen": 2187132928
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003405954100465415,
+      "loss": 1.3444,
+      "theoretical_loss": 3.4335768666331434,
+      "tokens_seen": 2187264000
+    },
+    {
+      "epoch": 0.33,
+      "objective/train/advantage_avg": 0.49426645040512085,
+      "objective/train/docs_used": 796380,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.399160623550415,
+      "objective/train/original_loss": 1.399160385131836,
+      "objective/train/theoretical_loss": 3.4335601517839565,
+      "objective/train/tokens_used": 557854176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24459707736968994,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506702661514282,
+      "objective/train/weighted_lm_loss": 1.4699373245239258,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4335601517839565,
+      "tokens_seen": 2187395072
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003405552880757503,
+      "loss": 1.4075,
+      "theoretical_loss": 3.4335601517839565,
+      "tokens_seen": 2187395072
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003405151661049591,
+      "loss": 1.3474,
+      "theoretical_loss": 3.4335434382167414,
+      "tokens_seen": 2187526144
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00034047504413416786,
+      "loss": 1.3996,
+      "theoretical_loss": 3.4335267259313227,
+      "tokens_seen": 2187657216
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00034043492216337667,
+      "loss": 1.3399,
+      "theoretical_loss": 3.433510014927526,
+      "tokens_seen": 2187788288
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00034039480019258547,
+      "loss": 1.3612,
+      "theoretical_loss": 3.4334933052051753,
+      "tokens_seen": 2187919360
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003403546782217943,
+      "loss": 1.3794,
+      "theoretical_loss": 3.4334765967640966,
+      "tokens_seen": 2188050432
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00034031455625100303,
+      "loss": 1.3616,
+      "theoretical_loss": 3.4334598896041144,
+      "tokens_seen": 2188181504
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003402744342802119,
+      "loss": 1.3361,
+      "theoretical_loss": 3.4334431837250534,
+      "tokens_seen": 2188312576
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00034023431230942064,
+      "loss": 1.4225,
+      "theoretical_loss": 3.433426479126739,
+      "tokens_seen": 2188443648
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003401941903386294,
+      "loss": 1.3913,
+      "theoretical_loss": 3.4334097758089968,
+      "tokens_seen": 2188574720
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00034015406836783826,
+      "loss": 1.3902,
+      "theoretical_loss": 3.433393073771651,
+      "tokens_seen": 2188705792
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.000340113946397047,
+      "loss": 1.3449,
+      "theoretical_loss": 3.4333763730145277,
+      "tokens_seen": 2188836864
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003400738244262558,
+      "loss": 1.3968,
+      "theoretical_loss": 3.433359673537452,
+      "tokens_seen": 2188967936
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003400337024554646,
+      "loss": 1.3887,
+      "theoretical_loss": 3.4333429753402482,
+      "tokens_seen": 2189099008
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003399935804846734,
+      "loss": 1.3243,
+      "theoretical_loss": 3.4333262784227427,
+      "tokens_seen": 2189230080
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033995345851388223,
+      "loss": 1.2798,
+      "theoretical_loss": 3.43330958278476,
+      "tokens_seen": 2189361152
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.000339913336543091,
+      "loss": 1.2972,
+      "theoretical_loss": 3.433292888426126,
+      "tokens_seen": 2189492224
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003398732145722998,
+      "loss": 1.3574,
+      "theoretical_loss": 3.4332761953466653,
+      "tokens_seen": 2189623296
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003398330926015086,
+      "loss": 1.3862,
+      "theoretical_loss": 3.433259503546205,
+      "tokens_seen": 2189754368
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003397929706307174,
+      "loss": 1.3797,
+      "theoretical_loss": 3.4332428130245685,
+      "tokens_seen": 2189885440
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033975284865992616,
+      "loss": 1.3655,
+      "theoretical_loss": 3.4332261237815827,
+      "tokens_seen": 2190016512
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.000339712726689135,
+      "loss": 1.4168,
+      "theoretical_loss": 3.433209435817073,
+      "tokens_seen": 2190147584
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033967260471834377,
+      "loss": 1.3439,
+      "theoretical_loss": 3.433192749130864,
+      "tokens_seen": 2190278656
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003396324827475525,
+      "loss": 1.3782,
+      "theoretical_loss": 3.433176063722782,
+      "tokens_seen": 2190409728
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003395923607767614,
+      "loss": 1.3278,
+      "theoretical_loss": 3.433159379592653,
+      "tokens_seen": 2190540800
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033955223880597013,
+      "loss": 1.342,
+      "theoretical_loss": 3.4331426967403025,
+      "tokens_seen": 2190671872
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033951211683517894,
+      "loss": 1.3223,
+      "theoretical_loss": 3.433126015165555,
+      "tokens_seen": 2190802944
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033947199486438775,
+      "loss": 1.3604,
+      "theoretical_loss": 3.433109334868238,
+      "tokens_seen": 2190934016
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033943187289359655,
+      "loss": 1.3884,
+      "theoretical_loss": 3.433092655848176,
+      "tokens_seen": 2191065088
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003393917509228053,
+      "loss": 1.3248,
+      "theoretical_loss": 3.4330759781051956,
+      "tokens_seen": 2191196160
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033935162895201417,
+      "loss": 1.3377,
+      "theoretical_loss": 3.4330593016391218,
+      "tokens_seen": 2191327232
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003393115069812229,
+      "loss": 1.3387,
+      "theoretical_loss": 3.4330426264497813,
+      "tokens_seen": 2191458304
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003392713850104317,
+      "loss": 1.3347,
+      "theoretical_loss": 3.4330259525369993,
+      "tokens_seen": 2191589376
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033923126303964053,
+      "loss": 1.3071,
+      "theoretical_loss": 3.4330092799006025,
+      "tokens_seen": 2191720448
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003391911410688493,
+      "loss": 1.4236,
+      "theoretical_loss": 3.432992608540416,
+      "tokens_seen": 2191851520
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033915101909805814,
+      "loss": 1.3669,
+      "theoretical_loss": 3.4329759384562664,
+      "tokens_seen": 2191982592
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003391108971272669,
+      "loss": 1.3174,
+      "theoretical_loss": 3.43295926964798,
+      "tokens_seen": 2192113664
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003390707751564757,
+      "loss": 1.409,
+      "theoretical_loss": 3.4329426021153817,
+      "tokens_seen": 2192244736
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003390306531856845,
+      "loss": 1.2578,
+      "theoretical_loss": 3.432925935858299,
+      "tokens_seen": 2192375808
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033899053121489326,
+      "loss": 1.3348,
+      "theoretical_loss": 3.4329092708765567,
+      "tokens_seen": 2192506880
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033895040924410207,
+      "loss": 1.333,
+      "theoretical_loss": 3.432892607169982,
+      "tokens_seen": 2192637952
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033891028727331087,
+      "loss": 1.3372,
+      "theoretical_loss": 3.432875944738401,
+      "tokens_seen": 2192769024
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003388701653025197,
+      "loss": 1.3847,
+      "theoretical_loss": 3.4328592835816396,
+      "tokens_seen": 2192900096
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033883004333172843,
+      "loss": 1.2944,
+      "theoretical_loss": 3.432842623699524,
+      "tokens_seen": 2193031168
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003387899213609373,
+      "loss": 1.3595,
+      "theoretical_loss": 3.4328259650918813,
+      "tokens_seen": 2193162240
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033874979939014604,
+      "loss": 1.3669,
+      "theoretical_loss": 3.4328093077585367,
+      "tokens_seen": 2193293312
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033870967741935485,
+      "loss": 1.3429,
+      "theoretical_loss": 3.4327926516993172,
+      "tokens_seen": 2193424384
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033866955544856366,
+      "loss": 1.3618,
+      "theoretical_loss": 3.432775996914049,
+      "tokens_seen": 2193555456
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003386294334777724,
+      "loss": 1.3628,
+      "theoretical_loss": 3.432759343402559,
+      "tokens_seen": 2193686528
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033858931150698127,
+      "loss": 1.3943,
+      "theoretical_loss": 3.4327426911646732,
+      "tokens_seen": 2193817600
+    },
+    {
+      "epoch": 0.33,
+      "objective/train/advantage_avg": 0.4935292601585388,
+      "objective/train/docs_used": 798764,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3344637155532837,
+      "objective/train/original_loss": 1.3344637155532837,
+      "objective/train/theoretical_loss": 3.4327260402002184,
+      "objective/train/tokens_used": 564407776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24394921958446503,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505931377410889,
+      "objective/train/weighted_lm_loss": 1.4019125699996948,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4327260402002184,
+      "tokens_seen": 2193948672
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033854918953619,
+      "loss": 1.3893,
+      "theoretical_loss": 3.4327260402002184,
+      "tokens_seen": 2193948672
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003385090675653988,
+      "loss": 1.3924,
+      "theoretical_loss": 3.432709390509021,
+      "tokens_seen": 2194079744
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033846894559460763,
+      "loss": 1.3621,
+      "theoretical_loss": 3.4326927420909077,
+      "tokens_seen": 2194210816
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003384288236238164,
+      "loss": 1.3482,
+      "theoretical_loss": 3.432676094945705,
+      "tokens_seen": 2194341888
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003383887016530252,
+      "loss": 1.3624,
+      "theoretical_loss": 3.4326594490732396,
+      "tokens_seen": 2194472960
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.000338348579682234,
+      "loss": 1.3499,
+      "theoretical_loss": 3.432642804473338,
+      "tokens_seen": 2194604032
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003383084577114428,
+      "loss": 1.313,
+      "theoretical_loss": 3.4326261611458273,
+      "tokens_seen": 2194735104
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033826833574065156,
+      "loss": 1.4182,
+      "theoretical_loss": 3.432609519090534,
+      "tokens_seen": 2194866176
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003382282137698604,
+      "loss": 1.3968,
+      "theoretical_loss": 3.4325928783072848,
+      "tokens_seen": 2194997248
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033818809179906917,
+      "loss": 1.3314,
+      "theoretical_loss": 3.4325762387959067,
+      "tokens_seen": 2195128320
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003381479698282779,
+      "loss": 1.3494,
+      "theoretical_loss": 3.4325596005562264,
+      "tokens_seen": 2195259392
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003381078478574868,
+      "loss": 1.2939,
+      "theoretical_loss": 3.4325429635880713,
+      "tokens_seen": 2195390464
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033806772588669553,
+      "loss": 1.3697,
+      "theoretical_loss": 3.4325263278912677,
+      "tokens_seen": 2195521536
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003380276039159044,
+      "loss": 1.3076,
+      "theoretical_loss": 3.4325096934656427,
+      "tokens_seen": 2195652608
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033798748194511315,
+      "loss": 1.4516,
+      "theoretical_loss": 3.4324930603110233,
+      "tokens_seen": 2195783680
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033794735997432195,
+      "loss": 1.3649,
+      "theoretical_loss": 3.4324764284272367,
+      "tokens_seen": 2195914752
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033790723800353076,
+      "loss": 1.3357,
+      "theoretical_loss": 3.43245979781411,
+      "tokens_seen": 2196045824
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003378671160327395,
+      "loss": 1.3302,
+      "theoretical_loss": 3.43244316847147,
+      "tokens_seen": 2196176896
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003378269940619483,
+      "loss": 1.3068,
+      "theoretical_loss": 3.4324265403991436,
+      "tokens_seen": 2196307968
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003377868720911571,
+      "loss": 1.3205,
+      "theoretical_loss": 3.432409913596959,
+      "tokens_seen": 2196439040
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033774675012036593,
+      "loss": 1.3935,
+      "theoretical_loss": 3.432393288064742,
+      "tokens_seen": 2196570112
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003377066281495747,
+      "loss": 1.4087,
+      "theoretical_loss": 3.432376663802321,
+      "tokens_seen": 2196701184
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033766650617878354,
+      "loss": 1.3461,
+      "theoretical_loss": 3.432360040809523,
+      "tokens_seen": 2196832256
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003376263842079923,
+      "loss": 1.3456,
+      "theoretical_loss": 3.4323434190861746,
+      "tokens_seen": 2196963328
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003375862622372011,
+      "loss": 1.3851,
+      "theoretical_loss": 3.432326798632104,
+      "tokens_seen": 2197094400
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003375461402664099,
+      "loss": 1.352,
+      "theoretical_loss": 3.4323101794471382,
+      "tokens_seen": 2197225472
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033750601829561866,
+      "loss": 1.332,
+      "theoretical_loss": 3.4322935615311043,
+      "tokens_seen": 2197356544
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003374658963248275,
+      "loss": 1.3662,
+      "theoretical_loss": 3.43227694488383,
+      "tokens_seen": 2197487616
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033742577435403627,
+      "loss": 1.3804,
+      "theoretical_loss": 3.432260329505143,
+      "tokens_seen": 2197618688
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003373856523832451,
+      "loss": 1.3365,
+      "theoretical_loss": 3.4322437153948706,
+      "tokens_seen": 2197749760
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003373455304124539,
+      "loss": 1.4049,
+      "theoretical_loss": 3.43222710255284,
+      "tokens_seen": 2197880832
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003373054084416627,
+      "loss": 1.3817,
+      "theoretical_loss": 3.4322104909788793,
+      "tokens_seen": 2198011904
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033726528647087144,
+      "loss": 1.3972,
+      "theoretical_loss": 3.432193880672816,
+      "tokens_seen": 2198142976
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033722516450008025,
+      "loss": 1.3423,
+      "theoretical_loss": 3.432177271634477,
+      "tokens_seen": 2198274048
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033718504252928906,
+      "loss": 1.3804,
+      "theoretical_loss": 3.4321606638636912,
+      "tokens_seen": 2198405120
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003371449205584978,
+      "loss": 1.3191,
+      "theoretical_loss": 3.4321440573602855,
+      "tokens_seen": 2198536192
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033710479858770667,
+      "loss": 1.351,
+      "theoretical_loss": 3.432127452124088,
+      "tokens_seen": 2198667264
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003370646766169154,
+      "loss": 1.3787,
+      "theoretical_loss": 3.4321108481549256,
+      "tokens_seen": 2198798336
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003370245546461242,
+      "loss": 1.32,
+      "theoretical_loss": 3.4320942454526273,
+      "tokens_seen": 2198929408
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033698443267533303,
+      "loss": 1.2746,
+      "theoretical_loss": 3.43207764401702,
+      "tokens_seen": 2199060480
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003369443107045418,
+      "loss": 1.3929,
+      "theoretical_loss": 3.4320610438479324,
+      "tokens_seen": 2199191552
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003369041887337506,
+      "loss": 1.3095,
+      "theoretical_loss": 3.4320444449451917,
+      "tokens_seen": 2199322624
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003368640667629594,
+      "loss": 1.341,
+      "theoretical_loss": 3.432027847308626,
+      "tokens_seen": 2199453696
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003368239447921682,
+      "loss": 1.3532,
+      "theoretical_loss": 3.432011250938064,
+      "tokens_seen": 2199584768
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.000336783822821377,
+      "loss": 1.4017,
+      "theoretical_loss": 3.4319946558333325,
+      "tokens_seen": 2199715840
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003367437008505858,
+      "loss": 1.3581,
+      "theoretical_loss": 3.4319780619942604,
+      "tokens_seen": 2199846912
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033670357887979457,
+      "loss": 1.321,
+      "theoretical_loss": 3.431961469420675,
+      "tokens_seen": 2199977984
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003366634569090034,
+      "loss": 1.3497,
+      "theoretical_loss": 3.4319448781124056,
+      "tokens_seen": 2200109056
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003366233349382122,
+      "loss": 1.4175,
+      "theoretical_loss": 3.431928288069279,
+      "tokens_seen": 2200240128
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033658321296742093,
+      "loss": 1.3744,
+      "theoretical_loss": 3.4319116992911245,
+      "tokens_seen": 2200371200
+    },
+    {
+      "epoch": 0.33,
+      "objective/train/advantage_avg": 0.49178948998451233,
+      "objective/train/docs_used": 801168,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2348886728286743,
+      "objective/train/original_loss": 1.2348886728286743,
+      "objective/train/theoretical_loss": 3.4318951117777696,
+      "objective/train/tokens_used": 570961376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24227096140384674,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504106283187866,
+      "objective/train/weighted_lm_loss": 1.2970620393753052,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4318951117777696,
+      "tokens_seen": 2200502272
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003365430909966298,
+      "loss": 1.3148,
+      "theoretical_loss": 3.4318951117777696,
+      "tokens_seen": 2200502272
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033650296902583855,
+      "loss": 1.3714,
+      "theoretical_loss": 3.4318785255290427,
+      "tokens_seen": 2200633344
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033646284705504735,
+      "loss": 1.2853,
+      "theoretical_loss": 3.431861940544772,
+      "tokens_seen": 2200764416
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033642272508425616,
+      "loss": 1.3309,
+      "theoretical_loss": 3.4318453568247858,
+      "tokens_seen": 2200895488
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003363826031134649,
+      "loss": 1.3507,
+      "theoretical_loss": 3.431828774368913,
+      "tokens_seen": 2201026560
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003363424811426737,
+      "loss": 1.3309,
+      "theoretical_loss": 3.431812193176981,
+      "tokens_seen": 2201157632
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003363023591718825,
+      "loss": 1.3568,
+      "theoretical_loss": 3.431795613248819,
+      "tokens_seen": 2201288704
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033626223720109133,
+      "loss": 1.3066,
+      "theoretical_loss": 3.431779034584255,
+      "tokens_seen": 2201419776
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003362221152303001,
+      "loss": 1.3531,
+      "theoretical_loss": 3.431762457183118,
+      "tokens_seen": 2201550848
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033618199325950894,
+      "loss": 1.3209,
+      "theoretical_loss": 3.431745881045236,
+      "tokens_seen": 2201681920
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003361418712887177,
+      "loss": 1.339,
+      "theoretical_loss": 3.4317293061704377,
+      "tokens_seen": 2201812992
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033610174931792655,
+      "loss": 1.3087,
+      "theoretical_loss": 3.4317127325585517,
+      "tokens_seen": 2201944064
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003360616273471353,
+      "loss": 1.3854,
+      "theoretical_loss": 3.4316961602094063,
+      "tokens_seen": 2202075136
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033602150537634406,
+      "loss": 1.3552,
+      "theoretical_loss": 3.4316795891228304,
+      "tokens_seen": 2202206208
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003359813834055529,
+      "loss": 1.3855,
+      "theoretical_loss": 3.431663019298653,
+      "tokens_seen": 2202337280
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00033594126143476167,
+      "loss": 1.3596,
+      "theoretical_loss": 3.431646450736702,
+      "tokens_seen": 2202468352
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003359011394639705,
+      "loss": 1.3154,
+      "theoretical_loss": 3.431629883436807,
+      "tokens_seen": 2202599424
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003358610174931793,
+      "loss": 1.3408,
+      "theoretical_loss": 3.431613317398796,
+      "tokens_seen": 2202730496
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003358208955223881,
+      "loss": 1.5205,
+      "theoretical_loss": 3.4315967526224984,
+      "tokens_seen": 2202861568
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033578077355159684,
+      "loss": 1.3871,
+      "theoretical_loss": 3.431580189107743,
+      "tokens_seen": 2202992640
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033574065158080565,
+      "loss": 1.3462,
+      "theoretical_loss": 3.431563626854359,
+      "tokens_seen": 2203123712
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033570052961001445,
+      "loss": 1.4022,
+      "theoretical_loss": 3.431547065862174,
+      "tokens_seen": 2203254784
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003356604076392232,
+      "loss": 1.3429,
+      "theoretical_loss": 3.431530506131018,
+      "tokens_seen": 2203385856
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033562028566843207,
+      "loss": 1.3537,
+      "theoretical_loss": 3.4315139476607195,
+      "tokens_seen": 2203516928
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003355801636976408,
+      "loss": 1.342,
+      "theoretical_loss": 3.431497390451108,
+      "tokens_seen": 2203648000
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003355400417268497,
+      "loss": 1.3609,
+      "theoretical_loss": 3.4314808345020125,
+      "tokens_seen": 2203779072
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033549991975605843,
+      "loss": 1.4011,
+      "theoretical_loss": 3.4314642798132615,
+      "tokens_seen": 2203910144
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003354597977852672,
+      "loss": 1.3306,
+      "theoretical_loss": 3.4314477263846848,
+      "tokens_seen": 2204041216
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033541967581447604,
+      "loss": 1.2885,
+      "theoretical_loss": 3.4314311742161108,
+      "tokens_seen": 2204172288
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003353795538436848,
+      "loss": 1.3436,
+      "theoretical_loss": 3.4314146233073695,
+      "tokens_seen": 2204303360
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003353394318728936,
+      "loss": 1.4281,
+      "theoretical_loss": 3.4313980736582894,
+      "tokens_seen": 2204434432
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003352993099021024,
+      "loss": 1.4027,
+      "theoretical_loss": 3.4313815252687,
+      "tokens_seen": 2204565504
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003352591879313112,
+      "loss": 1.3648,
+      "theoretical_loss": 3.4313649781384306,
+      "tokens_seen": 2204696576
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033521906596051997,
+      "loss": 1.3546,
+      "theoretical_loss": 3.4313484322673107,
+      "tokens_seen": 2204827648
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003351789439897288,
+      "loss": 1.3696,
+      "theoretical_loss": 3.431331887655169,
+      "tokens_seen": 2204958720
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003351388220189376,
+      "loss": 1.3307,
+      "theoretical_loss": 3.4313153443018356,
+      "tokens_seen": 2205089792
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033509870004814633,
+      "loss": 1.3461,
+      "theoretical_loss": 3.43129880220714,
+      "tokens_seen": 2205220864
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003350585780773552,
+      "loss": 1.4698,
+      "theoretical_loss": 3.4312822613709106,
+      "tokens_seen": 2205351936
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033501845610656395,
+      "loss": 1.3687,
+      "theoretical_loss": 3.431265721792978,
+      "tokens_seen": 2205483008
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033497833413577275,
+      "loss": 1.3893,
+      "theoretical_loss": 3.431249183473171,
+      "tokens_seen": 2205614080
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033493821216498156,
+      "loss": 1.3993,
+      "theoretical_loss": 3.4312326464113196,
+      "tokens_seen": 2205745152
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003348980901941903,
+      "loss": 1.3727,
+      "theoretical_loss": 3.431216110607253,
+      "tokens_seen": 2205876224
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033485796822339917,
+      "loss": 1.3665,
+      "theoretical_loss": 3.431199576060801,
+      "tokens_seen": 2206007296
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003348178462526079,
+      "loss": 1.3457,
+      "theoretical_loss": 3.4311830427717935,
+      "tokens_seen": 2206138368
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033477772428181673,
+      "loss": 1.3617,
+      "theoretical_loss": 3.4311665107400597,
+      "tokens_seen": 2206269440
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033473760231102554,
+      "loss": 1.3204,
+      "theoretical_loss": 3.4311499799654293,
+      "tokens_seen": 2206400512
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033469748034023434,
+      "loss": 1.3452,
+      "theoretical_loss": 3.4311334504477324,
+      "tokens_seen": 2206531584
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003346573583694431,
+      "loss": 1.3525,
+      "theoretical_loss": 3.4311169221867988,
+      "tokens_seen": 2206662656
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033461723639865195,
+      "loss": 1.4016,
+      "theoretical_loss": 3.4311003951824577,
+      "tokens_seen": 2206793728
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003345771144278607,
+      "loss": 1.3259,
+      "theoretical_loss": 3.4310838694345396,
+      "tokens_seen": 2206924800
+    },
+    {
+      "epoch": 0.34,
+      "objective/train/advantage_avg": 0.4934345781803131,
+      "objective/train/docs_used": 803600,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2919442653656006,
+      "objective/train/original_loss": 1.2919442653656006,
+      "objective/train/theoretical_loss": 3.431067344942874,
+      "objective/train/tokens_used": 577514976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2438177615404129,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050583004951477,
+      "objective/train/weighted_lm_loss": 1.357256293296814,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.431067344942874,
+      "tokens_seen": 2207055872
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033453699245706946,
+      "loss": 1.4178,
+      "theoretical_loss": 3.431067344942874,
+      "tokens_seen": 2207055872
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003344968704862783,
+      "loss": 1.3497,
+      "theoretical_loss": 3.4310508217072906,
+      "tokens_seen": 2207186944
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033445674851548707,
+      "loss": 1.2779,
+      "theoretical_loss": 3.43103429972762,
+      "tokens_seen": 2207318016
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003344166265446959,
+      "loss": 1.396,
+      "theoretical_loss": 3.4310177790036915,
+      "tokens_seen": 2207449088
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003343765045739047,
+      "loss": 1.3573,
+      "theoretical_loss": 3.431001259535336,
+      "tokens_seen": 2207580160
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003343363826031135,
+      "loss": 1.3704,
+      "theoretical_loss": 3.4309847413223826,
+      "tokens_seen": 2207711232
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033429626063232224,
+      "loss": 1.3542,
+      "theoretical_loss": 3.430968224364662,
+      "tokens_seen": 2207842304
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033425613866153105,
+      "loss": 1.3236,
+      "theoretical_loss": 3.4309517086620036,
+      "tokens_seen": 2207973376
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033421601669073985,
+      "loss": 1.4298,
+      "theoretical_loss": 3.4309351942142383,
+      "tokens_seen": 2208104448
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033417589471994866,
+      "loss": 1.3053,
+      "theoretical_loss": 3.430918681021196,
+      "tokens_seen": 2208235520
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033413577274915747,
+      "loss": 1.3677,
+      "theoretical_loss": 3.4309021690827066,
+      "tokens_seen": 2208366592
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003340956507783662,
+      "loss": 1.3056,
+      "theoretical_loss": 3.430885658398601,
+      "tokens_seen": 2208497664
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003340555288075751,
+      "loss": 1.3524,
+      "theoretical_loss": 3.430869148968709,
+      "tokens_seen": 2208628736
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033401540683678383,
+      "loss": 1.3803,
+      "theoretical_loss": 3.4308526407928603,
+      "tokens_seen": 2208759808
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003339752848659926,
+      "loss": 1.3963,
+      "theoretical_loss": 3.430836133870886,
+      "tokens_seen": 2208890880
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033393516289520144,
+      "loss": 1.3886,
+      "theoretical_loss": 3.430819628202617,
+      "tokens_seen": 2209021952
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003338950409244102,
+      "loss": 1.3385,
+      "theoretical_loss": 3.4308031237878827,
+      "tokens_seen": 2209153024
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.000333854918953619,
+      "loss": 1.4559,
+      "theoretical_loss": 3.430786620626514,
+      "tokens_seen": 2209284096
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003338147969828278,
+      "loss": 1.4174,
+      "theoretical_loss": 3.4307701187183413,
+      "tokens_seen": 2209415168
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003337746750120366,
+      "loss": 1.4195,
+      "theoretical_loss": 3.4307536180631946,
+      "tokens_seen": 2209546240
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033373455304124537,
+      "loss": 1.299,
+      "theoretical_loss": 3.4307371186609057,
+      "tokens_seen": 2209677312
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003336944310704542,
+      "loss": 1.3939,
+      "theoretical_loss": 3.4307206205113037,
+      "tokens_seen": 2209808384
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.000333654309099663,
+      "loss": 1.3495,
+      "theoretical_loss": 3.43070412361422,
+      "tokens_seen": 2209939456
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003336141871288718,
+      "loss": 1.3268,
+      "theoretical_loss": 3.430687627969485,
+      "tokens_seen": 2210070528
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003335740651580806,
+      "loss": 1.3515,
+      "theoretical_loss": 3.4306711335769298,
+      "tokens_seen": 2210201600
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033353394318728934,
+      "loss": 1.3167,
+      "theoretical_loss": 3.4306546404363845,
+      "tokens_seen": 2210332672
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003334938212164982,
+      "loss": 1.3597,
+      "theoretical_loss": 3.43063814854768,
+      "tokens_seen": 2210463744
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033345369924570696,
+      "loss": 1.3747,
+      "theoretical_loss": 3.430621657910647,
+      "tokens_seen": 2210594816
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003334135772749157,
+      "loss": 1.345,
+      "theoretical_loss": 3.4306051685251173,
+      "tokens_seen": 2210725888
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033337345530412457,
+      "loss": 1.3063,
+      "theoretical_loss": 3.4305886803909202,
+      "tokens_seen": 2210856960
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003333333333333333,
+      "loss": 1.3464,
+      "theoretical_loss": 3.4305721935078872,
+      "tokens_seen": 2210988032
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033329321136254213,
+      "loss": 1.4277,
+      "theoretical_loss": 3.4305557078758495,
+      "tokens_seen": 2211119104
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033325308939175093,
+      "loss": 1.4133,
+      "theoretical_loss": 3.4305392234946375,
+      "tokens_seen": 2211250176
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033321296742095974,
+      "loss": 1.3211,
+      "theoretical_loss": 3.430522740364083,
+      "tokens_seen": 2211381248
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003331728454501685,
+      "loss": 1.4312,
+      "theoretical_loss": 3.430506258484016,
+      "tokens_seen": 2211512320
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003331327234793773,
+      "loss": 1.3155,
+      "theoretical_loss": 3.430489777854268,
+      "tokens_seen": 2211643392
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003330926015085861,
+      "loss": 1.4038,
+      "theoretical_loss": 3.4304732984746695,
+      "tokens_seen": 2211774464
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033305247953779486,
+      "loss": 1.4728,
+      "theoretical_loss": 3.4304568203450523,
+      "tokens_seen": 2211905536
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003330123575670037,
+      "loss": 1.3639,
+      "theoretical_loss": 3.4304403434652473,
+      "tokens_seen": 2212036608
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033297223559621247,
+      "loss": 1.3366,
+      "theoretical_loss": 3.430423867835086,
+      "tokens_seen": 2212167680
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033293211362542133,
+      "loss": 1.3491,
+      "theoretical_loss": 3.4304073934543995,
+      "tokens_seen": 2212298752
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003328919916546301,
+      "loss": 1.35,
+      "theoretical_loss": 3.430390920323018,
+      "tokens_seen": 2212429824
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003328518696838389,
+      "loss": 1.3802,
+      "theoretical_loss": 3.430374448440774,
+      "tokens_seen": 2212560896
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003328117477130477,
+      "loss": 1.4024,
+      "theoretical_loss": 3.4303579778074984,
+      "tokens_seen": 2212691968
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033277162574225645,
+      "loss": 1.3588,
+      "theoretical_loss": 3.430341508423022,
+      "tokens_seen": 2212823040
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033273150377146525,
+      "loss": 1.4364,
+      "theoretical_loss": 3.4303250402871774,
+      "tokens_seen": 2212954112
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033269138180067406,
+      "loss": 1.3101,
+      "theoretical_loss": 3.4303085733997944,
+      "tokens_seen": 2213085184
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033265125982988287,
+      "loss": 1.3668,
+      "theoretical_loss": 3.430292107760706,
+      "tokens_seen": 2213216256
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003326111378590916,
+      "loss": 1.3365,
+      "theoretical_loss": 3.4302756433697423,
+      "tokens_seen": 2213347328
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003325710158883005,
+      "loss": 1.3119,
+      "theoretical_loss": 3.4302591802267357,
+      "tokens_seen": 2213478400
+    },
+    {
+      "epoch": 0.34,
+      "objective/train/advantage_avg": 0.4922281801700592,
+      "objective/train/docs_used": 805966,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3034181594848633,
+      "objective/train/original_loss": 1.3034181594848633,
+      "objective/train/theoretical_loss": 3.430242718331517,
+      "objective/train/tokens_used": 584068576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24269194900989532,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504565238952637,
+      "objective/train/weighted_lm_loss": 1.3691374063491821,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.430242718331517,
+      "tokens_seen": 2213609472
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033253089391750923,
+      "loss": 1.393,
+      "theoretical_loss": 3.430242718331517,
+      "tokens_seen": 2213609472
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.000332490771946718,
+      "loss": 1.3724,
+      "theoretical_loss": 3.4302262576839184,
+      "tokens_seen": 2213740544
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033245064997592684,
+      "loss": 1.3906,
+      "theoretical_loss": 3.430209798283771,
+      "tokens_seen": 2213871616
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003324105280051356,
+      "loss": 1.3624,
+      "theoretical_loss": 3.4301933401309066,
+      "tokens_seen": 2214002688
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033237040603434446,
+      "loss": 1.3263,
+      "theoretical_loss": 3.4301768832251573,
+      "tokens_seen": 2214133760
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003323302840635532,
+      "loss": 1.3675,
+      "theoretical_loss": 3.4301604275663538,
+      "tokens_seen": 2214264832
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.000332290162092762,
+      "loss": 1.3473,
+      "theoretical_loss": 3.430143973154329,
+      "tokens_seen": 2214395904
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003322500401219708,
+      "loss": 1.2783,
+      "theoretical_loss": 3.4301275199889134,
+      "tokens_seen": 2214526976
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003322099181511796,
+      "loss": 1.3941,
+      "theoretical_loss": 3.4301110680699396,
+      "tokens_seen": 2214658048
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003321697961803884,
+      "loss": 1.4164,
+      "theoretical_loss": 3.4300946173972395,
+      "tokens_seen": 2214789120
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003321296742095972,
+      "loss": 1.2651,
+      "theoretical_loss": 3.4300781679706445,
+      "tokens_seen": 2214920192
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.000332089552238806,
+      "loss": 1.3303,
+      "theoretical_loss": 3.430061719789986,
+      "tokens_seen": 2215051264
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033204943026801474,
+      "loss": 1.328,
+      "theoretical_loss": 3.4300452728550974,
+      "tokens_seen": 2215182336
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003320093082972236,
+      "loss": 1.2824,
+      "theoretical_loss": 3.4300288271658093,
+      "tokens_seen": 2215313408
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033196918632643236,
+      "loss": 1.3273,
+      "theoretical_loss": 3.4300123827219546,
+      "tokens_seen": 2215444480
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003319290643556411,
+      "loss": 1.3714,
+      "theoretical_loss": 3.429995939523365,
+      "tokens_seen": 2215575552
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033188894238484997,
+      "loss": 1.3242,
+      "theoretical_loss": 3.429979497569872,
+      "tokens_seen": 2215706624
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003318488204140587,
+      "loss": 1.4746,
+      "theoretical_loss": 3.4299630568613075,
+      "tokens_seen": 2215837696
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033180869844326753,
+      "loss": 1.4091,
+      "theoretical_loss": 3.429946617397505,
+      "tokens_seen": 2215968768
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033176857647247633,
+      "loss": 1.2747,
+      "theoretical_loss": 3.4299301791782955,
+      "tokens_seen": 2216099840
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033172845450168514,
+      "loss": 1.4046,
+      "theoretical_loss": 3.429913742203512,
+      "tokens_seen": 2216230912
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033168833253089395,
+      "loss": 1.393,
+      "theoretical_loss": 3.4298973064729856,
+      "tokens_seen": 2216361984
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003316482105601027,
+      "loss": 1.3001,
+      "theoretical_loss": 3.4298808719865495,
+      "tokens_seen": 2216493056
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003316080885893115,
+      "loss": 1.3575,
+      "theoretical_loss": 3.429864438744035,
+      "tokens_seen": 2216624128
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003315679666185203,
+      "loss": 1.3219,
+      "theoretical_loss": 3.429848006745276,
+      "tokens_seen": 2216755200
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003315278446477291,
+      "loss": 1.3334,
+      "theoretical_loss": 3.4298315759901032,
+      "tokens_seen": 2216886272
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033148772267693787,
+      "loss": 1.3019,
+      "theoretical_loss": 3.42981514647835,
+      "tokens_seen": 2217017344
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033144760070614673,
+      "loss": 1.3721,
+      "theoretical_loss": 3.429798718209848,
+      "tokens_seen": 2217148416
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003314074787353555,
+      "loss": 1.2649,
+      "theoretical_loss": 3.4297822911844302,
+      "tokens_seen": 2217279488
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003313673567645643,
+      "loss": 1.4124,
+      "theoretical_loss": 3.429765865401929,
+      "tokens_seen": 2217410560
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003313272347937731,
+      "loss": 1.3712,
+      "theoretical_loss": 3.4297494408621767,
+      "tokens_seen": 2217541632
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033128711282298185,
+      "loss": 1.3034,
+      "theoretical_loss": 3.429733017565006,
+      "tokens_seen": 2217672704
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033124699085219065,
+      "loss": 1.3273,
+      "theoretical_loss": 3.429716595510249,
+      "tokens_seen": 2217803776
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033120686888139946,
+      "loss": 1.282,
+      "theoretical_loss": 3.4297001746977394,
+      "tokens_seen": 2217934848
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033116674691060827,
+      "loss": 1.3928,
+      "theoretical_loss": 3.4296837551273085,
+      "tokens_seen": 2218065920
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.000331126624939817,
+      "loss": 1.3282,
+      "theoretical_loss": 3.42966733679879,
+      "tokens_seen": 2218196992
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003310865029690259,
+      "loss": 1.3673,
+      "theoretical_loss": 3.429650919712016,
+      "tokens_seen": 2218328064
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033104638099823463,
+      "loss": 1.3818,
+      "theoretical_loss": 3.4296345038668195,
+      "tokens_seen": 2218459136
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033100625902744344,
+      "loss": 1.3528,
+      "theoretical_loss": 3.429618089263033,
+      "tokens_seen": 2218590208
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033096613705665224,
+      "loss": 1.3551,
+      "theoretical_loss": 3.4296016759004893,
+      "tokens_seen": 2218721280
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.000330926015085861,
+      "loss": 1.2936,
+      "theoretical_loss": 3.4295852637790216,
+      "tokens_seen": 2218852352
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00033088589311506986,
+      "loss": 1.341,
+      "theoretical_loss": 3.4295688528984623,
+      "tokens_seen": 2218983424
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003308457711442786,
+      "loss": 1.331,
+      "theoretical_loss": 3.4295524432586446,
+      "tokens_seen": 2219114496
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003308056491734874,
+      "loss": 1.2502,
+      "theoretical_loss": 3.4295360348594013,
+      "tokens_seen": 2219245568
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003307655272026962,
+      "loss": 1.4124,
+      "theoretical_loss": 3.4295196277005653,
+      "tokens_seen": 2219376640
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.000330725405231905,
+      "loss": 1.317,
+      "theoretical_loss": 3.4295032217819696,
+      "tokens_seen": 2219507712
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003306852832611138,
+      "loss": 1.2863,
+      "theoretical_loss": 3.4294868171034474,
+      "tokens_seen": 2219638784
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003306451612903226,
+      "loss": 1.361,
+      "theoretical_loss": 3.4294704136648315,
+      "tokens_seen": 2219769856
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003306050393195314,
+      "loss": 1.2858,
+      "theoretical_loss": 3.429454011465955,
+      "tokens_seen": 2219900928
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00033056491734874014,
+      "loss": 1.291,
+      "theoretical_loss": 3.4294376105066515,
+      "tokens_seen": 2220032000
+    },
+    {
+      "epoch": 0.35,
+      "objective/train/advantage_avg": 0.4895266592502594,
+      "objective/train/docs_used": 808099,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2698285579681396,
+      "objective/train/original_loss": 1.2698285579681396,
+      "objective/train/theoretical_loss": 3.4294212107867534,
+      "objective/train/tokens_used": 590622176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2401658147573471,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0501734018325806,
+      "objective/train/weighted_lm_loss": 1.3336659669876099,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4294212107867534,
+      "tokens_seen": 2220163072
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.000330524795377949,
+      "loss": 1.2871,
+      "theoretical_loss": 3.4294212107867534,
+      "tokens_seen": 2220163072
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00033048467340715776,
+      "loss": 1.3736,
+      "theoretical_loss": 3.4294048123060943,
+      "tokens_seen": 2220294144
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00033044455143636656,
+      "loss": 1.3216,
+      "theoretical_loss": 3.4293884150645075,
+      "tokens_seen": 2220425216
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00033040442946557537,
+      "loss": 1.3607,
+      "theoretical_loss": 3.4293720190618258,
+      "tokens_seen": 2220556288
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003303643074947841,
+      "loss": 1.3285,
+      "theoretical_loss": 3.429355624297883,
+      "tokens_seen": 2220687360
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.000330324185523993,
+      "loss": 1.4195,
+      "theoretical_loss": 3.429339230772512,
+      "tokens_seen": 2220818432
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00033028406355320173,
+      "loss": 1.3815,
+      "theoretical_loss": 3.429322838485546,
+      "tokens_seen": 2220949504
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00033024394158241054,
+      "loss": 1.4198,
+      "theoretical_loss": 3.4293064474368187,
+      "tokens_seen": 2221080576
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00033020381961161935,
+      "loss": 1.3461,
+      "theoretical_loss": 3.429290057626164,
+      "tokens_seen": 2221211648
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003301636976408281,
+      "loss": 1.3128,
+      "theoretical_loss": 3.4292736690534142,
+      "tokens_seen": 2221342720
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003301235756700369,
+      "loss": 1.3712,
+      "theoretical_loss": 3.4292572817184035,
+      "tokens_seen": 2221473792
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003300834536992457,
+      "loss": 1.3257,
+      "theoretical_loss": 3.4292408956209655,
+      "tokens_seen": 2221604864
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003300433317284545,
+      "loss": 1.3319,
+      "theoretical_loss": 3.429224510760933,
+      "tokens_seen": 2221735936
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00033000320975766327,
+      "loss": 1.3113,
+      "theoretical_loss": 3.429208127138141,
+      "tokens_seen": 2221867008
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032996308778687213,
+      "loss": 1.3072,
+      "theoretical_loss": 3.429191744752421,
+      "tokens_seen": 2221998080
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003299229658160809,
+      "loss": 1.315,
+      "theoretical_loss": 3.4291753636036084,
+      "tokens_seen": 2222129152
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003298828438452897,
+      "loss": 1.4009,
+      "theoretical_loss": 3.4291589836915364,
+      "tokens_seen": 2222260224
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003298427218744985,
+      "loss": 1.3757,
+      "theoretical_loss": 3.4291426050160383,
+      "tokens_seen": 2222391296
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032980259990370725,
+      "loss": 1.4129,
+      "theoretical_loss": 3.4291262275769476,
+      "tokens_seen": 2222522368
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003297624779329161,
+      "loss": 1.3307,
+      "theoretical_loss": 3.429109851374099,
+      "tokens_seen": 2222653440
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032972235596212486,
+      "loss": 1.3444,
+      "theoretical_loss": 3.429093476407326,
+      "tokens_seen": 2222784512
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032968223399133367,
+      "loss": 1.3329,
+      "theoretical_loss": 3.429077102676462,
+      "tokens_seen": 2222915584
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032964211202054247,
+      "loss": 1.3843,
+      "theoretical_loss": 3.429060730181341,
+      "tokens_seen": 2223046656
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003296019900497513,
+      "loss": 1.3076,
+      "theoretical_loss": 3.429044358921797,
+      "tokens_seen": 2223177728
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032956186807896003,
+      "loss": 1.285,
+      "theoretical_loss": 3.429027988897664,
+      "tokens_seen": 2223308800
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032952174610816884,
+      "loss": 1.4034,
+      "theoretical_loss": 3.4290116201087755,
+      "tokens_seen": 2223439872
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032948162413737764,
+      "loss": 1.3865,
+      "theoretical_loss": 3.428995252554966,
+      "tokens_seen": 2223570944
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003294415021665864,
+      "loss": 1.3645,
+      "theoretical_loss": 3.4289788862360693,
+      "tokens_seen": 2223702016
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032940138019579526,
+      "loss": 1.3928,
+      "theoretical_loss": 3.4289625211519192,
+      "tokens_seen": 2223833088
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.000329361258225004,
+      "loss": 1.467,
+      "theoretical_loss": 3.4289461573023505,
+      "tokens_seen": 2223964160
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003293211362542128,
+      "loss": 1.2974,
+      "theoretical_loss": 3.4289297946871966,
+      "tokens_seen": 2224095232
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003292810142834216,
+      "loss": 1.3819,
+      "theoretical_loss": 3.428913433306292,
+      "tokens_seen": 2224226304
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032924089231263037,
+      "loss": 1.4726,
+      "theoretical_loss": 3.428897073159471,
+      "tokens_seen": 2224357376
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032920077034183923,
+      "loss": 1.2681,
+      "theoretical_loss": 3.428880714246567,
+      "tokens_seen": 2224488448
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.000329160648371048,
+      "loss": 1.3434,
+      "theoretical_loss": 3.4288643565674155,
+      "tokens_seen": 2224619520
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003291205264002568,
+      "loss": 1.3192,
+      "theoretical_loss": 3.4288480001218495,
+      "tokens_seen": 2224750592
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003290804044294656,
+      "loss": 1.359,
+      "theoretical_loss": 3.428831644909704,
+      "tokens_seen": 2224881664
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003290402824586744,
+      "loss": 1.4429,
+      "theoretical_loss": 3.4288152909308134,
+      "tokens_seen": 2225012736
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032900016048788316,
+      "loss": 1.4079,
+      "theoretical_loss": 3.428798938185012,
+      "tokens_seen": 2225143808
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032896003851709196,
+      "loss": 1.3606,
+      "theoretical_loss": 3.428782586672134,
+      "tokens_seen": 2225274880
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032891991654630077,
+      "loss": 1.3469,
+      "theoretical_loss": 3.4287662363920135,
+      "tokens_seen": 2225405952
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003288797945755095,
+      "loss": 1.3257,
+      "theoretical_loss": 3.4287498873444857,
+      "tokens_seen": 2225537024
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003288396726047184,
+      "loss": 1.3335,
+      "theoretical_loss": 3.428733539529385,
+      "tokens_seen": 2225668096
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032879955063392713,
+      "loss": 1.3516,
+      "theoretical_loss": 3.4287171929465456,
+      "tokens_seen": 2225799168
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032875942866313594,
+      "loss": 1.3463,
+      "theoretical_loss": 3.4287008475958016,
+      "tokens_seen": 2225930240
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032871930669234475,
+      "loss": 1.2843,
+      "theoretical_loss": 3.4286845034769886,
+      "tokens_seen": 2226061312
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003286791847215535,
+      "loss": 1.3635,
+      "theoretical_loss": 3.428668160589941,
+      "tokens_seen": 2226192384
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003286390627507623,
+      "loss": 1.3433,
+      "theoretical_loss": 3.428651818934493,
+      "tokens_seen": 2226323456
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003285989407799711,
+      "loss": 1.439,
+      "theoretical_loss": 3.428635478510479,
+      "tokens_seen": 2226454528
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003285588188091799,
+      "loss": 1.2454,
+      "theoretical_loss": 3.428619139317735,
+      "tokens_seen": 2226585600
+    },
+    {
+      "epoch": 0.35,
+      "objective/train/advantage_avg": 0.4919842779636383,
+      "objective/train/docs_used": 810670,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2596139907836914,
+      "objective/train/original_loss": 1.259614109992981,
+      "objective/train/theoretical_loss": 3.428602801356095,
+      "objective/train/tokens_used": 597175776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24256180226802826,
+      "objective/train/value_max": -0.5,
+      "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": 1.3230178356170654,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.428602801356095,
+      "tokens_seen": 2226716672
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003285186968383887,
+      "loss": 1.3546,
+      "theoretical_loss": 3.428602801356095,
+      "tokens_seen": 2226716672
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032847857486759753,
+      "loss": 1.3323,
+      "theoretical_loss": 3.4285864646253934,
+      "tokens_seen": 2226847744
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003284384528968063,
+      "loss": 1.3518,
+      "theoretical_loss": 3.428570129125465,
+      "tokens_seen": 2226978816
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003283983309260151,
+      "loss": 1.3978,
+      "theoretical_loss": 3.4285537948561458,
+      "tokens_seen": 2227109888
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003283582089552239,
+      "loss": 1.3175,
+      "theoretical_loss": 3.4285374618172693,
+      "tokens_seen": 2227240960
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032831808698443265,
+      "loss": 1.3432,
+      "theoretical_loss": 3.4285211300086713,
+      "tokens_seen": 2227372032
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003282779650136415,
+      "loss": 1.3619,
+      "theoretical_loss": 3.4285047994301863,
+      "tokens_seen": 2227503104
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032823784304285026,
+      "loss": 1.3248,
+      "theoretical_loss": 3.42848847008165,
+      "tokens_seen": 2227634176
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032819772107205907,
+      "loss": 1.3785,
+      "theoretical_loss": 3.428472141962896,
+      "tokens_seen": 2227765248
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032815759910126787,
+      "loss": 1.2849,
+      "theoretical_loss": 3.4284558150737605,
+      "tokens_seen": 2227896320
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003281174771304767,
+      "loss": 1.3936,
+      "theoretical_loss": 3.428439489414078,
+      "tokens_seen": 2228027392
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032807735515968543,
+      "loss": 1.3422,
+      "theoretical_loss": 3.428423164983684,
+      "tokens_seen": 2228158464
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032803723318889424,
+      "loss": 1.2599,
+      "theoretical_loss": 3.428406841782414,
+      "tokens_seen": 2228289536
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032799711121810304,
+      "loss": 1.3731,
+      "theoretical_loss": 3.428390519810102,
+      "tokens_seen": 2228420608
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003279569892473118,
+      "loss": 1.3665,
+      "theoretical_loss": 3.428374199066584,
+      "tokens_seen": 2228551680
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032791686727652066,
+      "loss": 1.3868,
+      "theoretical_loss": 3.4283578795516947,
+      "tokens_seen": 2228682752
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003278767453057294,
+      "loss": 1.2852,
+      "theoretical_loss": 3.4283415612652703,
+      "tokens_seen": 2228813824
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032783662333493827,
+      "loss": 1.3422,
+      "theoretical_loss": 3.4283252442071452,
+      "tokens_seen": 2228944896
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.000327796501364147,
+      "loss": 1.3451,
+      "theoretical_loss": 3.428308928377155,
+      "tokens_seen": 2229075968
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032775637939335577,
+      "loss": 1.4013,
+      "theoretical_loss": 3.428292613775135,
+      "tokens_seen": 2229207040
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032771625742256463,
+      "loss": 1.359,
+      "theoretical_loss": 3.4282763004009205,
+      "tokens_seen": 2229338112
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003276761354517734,
+      "loss": 1.4091,
+      "theoretical_loss": 3.4282599882543474,
+      "tokens_seen": 2229469184
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003276360134809822,
+      "loss": 1.365,
+      "theoretical_loss": 3.428243677335251,
+      "tokens_seen": 2229600256
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.000327595891510191,
+      "loss": 1.3047,
+      "theoretical_loss": 3.4282273676434665,
+      "tokens_seen": 2229731328
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003275557695393998,
+      "loss": 1.38,
+      "theoretical_loss": 3.428211059178829,
+      "tokens_seen": 2229862400
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032751564756860856,
+      "loss": 1.3661,
+      "theoretical_loss": 3.428194751941175,
+      "tokens_seen": 2229993472
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032747552559781736,
+      "loss": 1.3616,
+      "theoretical_loss": 3.4281784459303397,
+      "tokens_seen": 2230124544
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032743540362702617,
+      "loss": 1.3785,
+      "theoretical_loss": 3.4281621411461587,
+      "tokens_seen": 2230255616
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003273952816562349,
+      "loss": 1.3154,
+      "theoretical_loss": 3.4281458375884677,
+      "tokens_seen": 2230386688
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003273551596854438,
+      "loss": 1.36,
+      "theoretical_loss": 3.428129535257102,
+      "tokens_seen": 2230517760
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032731503771465253,
+      "loss": 1.3512,
+      "theoretical_loss": 3.4281132341518976,
+      "tokens_seen": 2230648832
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003272749157438614,
+      "loss": 1.3605,
+      "theoretical_loss": 3.42809693427269,
+      "tokens_seen": 2230779904
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032723479377307015,
+      "loss": 1.3349,
+      "theoretical_loss": 3.4280806356193154,
+      "tokens_seen": 2230910976
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003271946718022789,
+      "loss": 1.3645,
+      "theoretical_loss": 3.4280643381916094,
+      "tokens_seen": 2231042048
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032715454983148776,
+      "loss": 1.299,
+      "theoretical_loss": 3.428048041989408,
+      "tokens_seen": 2231173120
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003271144278606965,
+      "loss": 1.2867,
+      "theoretical_loss": 3.4280317470125463,
+      "tokens_seen": 2231304192
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003270743058899053,
+      "loss": 1.28,
+      "theoretical_loss": 3.428015453260861,
+      "tokens_seen": 2231435264
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003270341839191141,
+      "loss": 1.3821,
+      "theoretical_loss": 3.427999160734188,
+      "tokens_seen": 2231566336
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032699406194832293,
+      "loss": 1.4206,
+      "theoretical_loss": 3.4279828694323626,
+      "tokens_seen": 2231697408
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003269539399775317,
+      "loss": 1.3976,
+      "theoretical_loss": 3.427966579355222,
+      "tokens_seen": 2231828480
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003269138180067405,
+      "loss": 1.3481,
+      "theoretical_loss": 3.4279502905026007,
+      "tokens_seen": 2231959552
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003268736960359493,
+      "loss": 1.3835,
+      "theoretical_loss": 3.427934002874336,
+      "tokens_seen": 2232090624
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032683357406515805,
+      "loss": 1.389,
+      "theoretical_loss": 3.427917716470263,
+      "tokens_seen": 2232221696
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003267934520943669,
+      "loss": 1.3695,
+      "theoretical_loss": 3.4279014312902185,
+      "tokens_seen": 2232352768
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032675333012357566,
+      "loss": 1.3274,
+      "theoretical_loss": 3.4278851473340386,
+      "tokens_seen": 2232483840
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032671320815278447,
+      "loss": 1.3459,
+      "theoretical_loss": 3.4278688646015594,
+      "tokens_seen": 2232614912
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032667308618199327,
+      "loss": 1.3343,
+      "theoretical_loss": 3.4278525830926165,
+      "tokens_seen": 2232745984
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003266329642112021,
+      "loss": 1.3473,
+      "theoretical_loss": 3.427836302807047,
+      "tokens_seen": 2232877056
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003265928422404109,
+      "loss": 1.3398,
+      "theoretical_loss": 3.4278200237446868,
+      "tokens_seen": 2233008128
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032655272026961964,
+      "loss": 1.3609,
+      "theoretical_loss": 3.4278037459053725,
+      "tokens_seen": 2233139200
+    },
+    {
+      "epoch": 0.35,
+      "objective/train/advantage_avg": 0.4938317835330963,
+      "objective/train/docs_used": 813220,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.234928011894226,
+      "objective/train/original_loss": 1.2349278926849365,
+      "objective/train/theoretical_loss": 3.4277874692889396,
+      "objective/train/tokens_used": 603729376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2441910207271576,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506246089935303,
+      "objective/train/weighted_lm_loss": 1.2974351644515991,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4277874692889396,
+      "tokens_seen": 2233270272
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032651259829882844,
+      "loss": 1.3663,
+      "theoretical_loss": 3.4277874692889396,
+      "tokens_seen": 2233270272
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032647247632803725,
+      "loss": 1.3861,
+      "theoretical_loss": 3.4277711938952256,
+      "tokens_seen": 2233401344
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032643235435724606,
+      "loss": 1.2924,
+      "theoretical_loss": 3.427754919724066,
+      "tokens_seen": 2233532416
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003263922323864548,
+      "loss": 1.3697,
+      "theoretical_loss": 3.427738646775298,
+      "tokens_seen": 2233663488
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032635211041566367,
+      "loss": 1.284,
+      "theoretical_loss": 3.427722375048757,
+      "tokens_seen": 2233794560
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003263119884448724,
+      "loss": 1.3061,
+      "theoretical_loss": 3.4277061045442805,
+      "tokens_seen": 2233925632
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032627186647408117,
+      "loss": 1.3246,
+      "theoretical_loss": 3.4276898352617047,
+      "tokens_seen": 2234056704
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032623174450329003,
+      "loss": 1.4119,
+      "theoretical_loss": 3.4276735672008662,
+      "tokens_seen": 2234187776
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003261916225324988,
+      "loss": 1.4163,
+      "theoretical_loss": 3.4276573003616013,
+      "tokens_seen": 2234318848
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003261515005617076,
+      "loss": 1.4024,
+      "theoretical_loss": 3.4276410347437465,
+      "tokens_seen": 2234449920
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003261113785909164,
+      "loss": 1.3309,
+      "theoretical_loss": 3.427624770347139,
+      "tokens_seen": 2234580992
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003260712566201252,
+      "loss": 1.3456,
+      "theoretical_loss": 3.4276085071716156,
+      "tokens_seen": 2234712064
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032603113464933396,
+      "loss": 1.3527,
+      "theoretical_loss": 3.4275922452170122,
+      "tokens_seen": 2234843136
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032599101267854276,
+      "loss": 1.4484,
+      "theoretical_loss": 3.427575984483166,
+      "tokens_seen": 2234974208
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032595089070775157,
+      "loss": 1.3251,
+      "theoretical_loss": 3.427559724969914,
+      "tokens_seen": 2235105280
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003259107687369604,
+      "loss": 1.3458,
+      "theoretical_loss": 3.427543466677093,
+      "tokens_seen": 2235236352
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003258706467661692,
+      "loss": 1.4158,
+      "theoretical_loss": 3.427527209604539,
+      "tokens_seen": 2235367424
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00032583052479537793,
+      "loss": 1.3036,
+      "theoretical_loss": 3.42751095375209,
+      "tokens_seen": 2235498496
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003257904028245868,
+      "loss": 1.3678,
+      "theoretical_loss": 3.427494699119582,
+      "tokens_seen": 2235629568
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032575028085379555,
+      "loss": 1.4,
+      "theoretical_loss": 3.4274784457068526,
+      "tokens_seen": 2235760640
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003257101588830043,
+      "loss": 1.3507,
+      "theoretical_loss": 3.4274621935137386,
+      "tokens_seen": 2235891712
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032567003691221316,
+      "loss": 1.3257,
+      "theoretical_loss": 3.4274459425400767,
+      "tokens_seen": 2236022784
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003256299149414219,
+      "loss": 1.3114,
+      "theoretical_loss": 3.427429692785704,
+      "tokens_seen": 2236153856
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003255897929706307,
+      "loss": 1.345,
+      "theoretical_loss": 3.427413444250458,
+      "tokens_seen": 2236284928
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003255496709998395,
+      "loss": 1.3756,
+      "theoretical_loss": 3.427397196934175,
+      "tokens_seen": 2236416000
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032550954902904833,
+      "loss": 1.3338,
+      "theoretical_loss": 3.4273809508366933,
+      "tokens_seen": 2236547072
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003254694270582571,
+      "loss": 1.4444,
+      "theoretical_loss": 3.4273647059578485,
+      "tokens_seen": 2236678144
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003254293050874659,
+      "loss": 1.3823,
+      "theoretical_loss": 3.427348462297479,
+      "tokens_seen": 2236809216
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003253891831166747,
+      "loss": 1.382,
+      "theoretical_loss": 3.4273322198554217,
+      "tokens_seen": 2236940288
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003253490611458835,
+      "loss": 1.3718,
+      "theoretical_loss": 3.4273159786315137,
+      "tokens_seen": 2237071360
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003253089391750923,
+      "loss": 1.3712,
+      "theoretical_loss": 3.4272997386255923,
+      "tokens_seen": 2237202432
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032526881720430106,
+      "loss": 1.4332,
+      "theoretical_loss": 3.4272834998374946,
+      "tokens_seen": 2237333504
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003252286952335099,
+      "loss": 1.3829,
+      "theoretical_loss": 3.4272672622670584,
+      "tokens_seen": 2237464576
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032518857326271867,
+      "loss": 1.312,
+      "theoretical_loss": 3.427251025914121,
+      "tokens_seen": 2237595648
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003251484512919275,
+      "loss": 1.3537,
+      "theoretical_loss": 3.4272347907785194,
+      "tokens_seen": 2237726720
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003251083293211363,
+      "loss": 1.3561,
+      "theoretical_loss": 3.4272185568600917,
+      "tokens_seen": 2237857792
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032506820735034504,
+      "loss": 1.4494,
+      "theoretical_loss": 3.427202324158675,
+      "tokens_seen": 2237988864
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032502808537955384,
+      "loss": 1.3465,
+      "theoretical_loss": 3.4271860926741065,
+      "tokens_seen": 2238119936
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032498796340876265,
+      "loss": 1.3949,
+      "theoretical_loss": 3.427169862406224,
+      "tokens_seen": 2238251008
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032494784143797145,
+      "loss": 1.4266,
+      "theoretical_loss": 3.427153633354865,
+      "tokens_seen": 2238382080
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003249077194671802,
+      "loss": 1.349,
+      "theoretical_loss": 3.427137405519867,
+      "tokens_seen": 2238513152
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032486759749638907,
+      "loss": 1.388,
+      "theoretical_loss": 3.427121178901068,
+      "tokens_seen": 2238644224
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003248274755255978,
+      "loss": 1.352,
+      "theoretical_loss": 3.427104953498305,
+      "tokens_seen": 2238775296
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032478735355480657,
+      "loss": 1.4151,
+      "theoretical_loss": 3.427088729311417,
+      "tokens_seen": 2238906368
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032474723158401543,
+      "loss": 1.4149,
+      "theoretical_loss": 3.42707250634024,
+      "tokens_seen": 2239037440
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003247071096132242,
+      "loss": 1.351,
+      "theoretical_loss": 3.4270562845846126,
+      "tokens_seen": 2239168512
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032466698764243304,
+      "loss": 1.3013,
+      "theoretical_loss": 3.427040064044373,
+      "tokens_seen": 2239299584
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003246268656716418,
+      "loss": 1.3736,
+      "theoretical_loss": 3.427023844719358,
+      "tokens_seen": 2239430656
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003245867437008506,
+      "loss": 1.3235,
+      "theoretical_loss": 3.4270076266094063,
+      "tokens_seen": 2239561728
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003245466217300594,
+      "loss": 1.3186,
+      "theoretical_loss": 3.4269914097143555,
+      "tokens_seen": 2239692800
+    },
+    {
+      "epoch": 0.36,
+      "objective/train/advantage_avg": 0.4927559196949005,
+      "objective/train/docs_used": 815683,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3075639009475708,
+      "objective/train/original_loss": 1.3075636625289917,
+      "objective/train/theoretical_loss": 3.426975194034043,
+      "objective/train/tokens_used": 610282976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24313655495643616,
+      "objective/train/value_max": -0.5,
+      "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": 1.3736765384674072,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.426975194034043,
+      "tokens_seen": 2239823872
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032450649975926816,
+      "loss": 1.3479,
+      "theoretical_loss": 3.426975194034043,
+      "tokens_seen": 2239823872
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032446637778847697,
+      "loss": 1.3249,
+      "theoretical_loss": 3.4269589795683073,
+      "tokens_seen": 2239954944
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003244262558176858,
+      "loss": 1.3972,
+      "theoretical_loss": 3.4269427663169867,
+      "tokens_seen": 2240086016
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003243861338468946,
+      "loss": 1.3388,
+      "theoretical_loss": 3.4269265542799183,
+      "tokens_seen": 2240217088
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032434601187610333,
+      "loss": 1.4159,
+      "theoretical_loss": 3.4269103434569406,
+      "tokens_seen": 2240348160
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003243058899053122,
+      "loss": 1.4013,
+      "theoretical_loss": 3.4268941338478918,
+      "tokens_seen": 2240479232
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032426576793452095,
+      "loss": 1.3796,
+      "theoretical_loss": 3.42687792545261,
+      "tokens_seen": 2240610304
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003242256459637297,
+      "loss": 1.3596,
+      "theoretical_loss": 3.4268617182709327,
+      "tokens_seen": 2240741376
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032418552399293856,
+      "loss": 1.3446,
+      "theoretical_loss": 3.426845512302698,
+      "tokens_seen": 2240872448
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003241454020221473,
+      "loss": 1.3777,
+      "theoretical_loss": 3.4268293075477456,
+      "tokens_seen": 2241003520
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032410528005135617,
+      "loss": 1.3452,
+      "theoretical_loss": 3.426813104005912,
+      "tokens_seen": 2241134592
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003240651580805649,
+      "loss": 1.322,
+      "theoretical_loss": 3.4267969016770365,
+      "tokens_seen": 2241265664
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032402503610977373,
+      "loss": 1.331,
+      "theoretical_loss": 3.4267807005609567,
+      "tokens_seen": 2241396736
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032398491413898254,
+      "loss": 1.3061,
+      "theoretical_loss": 3.4267645006575114,
+      "tokens_seen": 2241527808
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003239447921681913,
+      "loss": 1.3419,
+      "theoretical_loss": 3.4267483019665383,
+      "tokens_seen": 2241658880
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003239046701974001,
+      "loss": 1.3473,
+      "theoretical_loss": 3.4267321044878765,
+      "tokens_seen": 2241789952
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003238645482266089,
+      "loss": 1.3748,
+      "theoretical_loss": 3.426715908221364,
+      "tokens_seen": 2241921024
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003238244262558177,
+      "loss": 1.4226,
+      "theoretical_loss": 3.4266997131668395,
+      "tokens_seen": 2242052096
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032378430428502646,
+      "loss": 1.2911,
+      "theoretical_loss": 3.4266835193241407,
+      "tokens_seen": 2242183168
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003237441823142353,
+      "loss": 1.3626,
+      "theoretical_loss": 3.4266673266931074,
+      "tokens_seen": 2242314240
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032370406034344407,
+      "loss": 1.3254,
+      "theoretical_loss": 3.426651135273577,
+      "tokens_seen": 2242445312
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003236639383726528,
+      "loss": 1.3694,
+      "theoretical_loss": 3.4266349450653886,
+      "tokens_seen": 2242576384
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003236238164018617,
+      "loss": 1.3156,
+      "theoretical_loss": 3.4266187560683803,
+      "tokens_seen": 2242707456
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032358369443107044,
+      "loss": 1.414,
+      "theoretical_loss": 3.4266025682823913,
+      "tokens_seen": 2242838528
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032354357246027924,
+      "loss": 1.3191,
+      "theoretical_loss": 3.42658638170726,
+      "tokens_seen": 2242969600
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032350345048948805,
+      "loss": 1.2871,
+      "theoretical_loss": 3.4265701963428246,
+      "tokens_seen": 2243100672
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032346332851869685,
+      "loss": 1.26,
+      "theoretical_loss": 3.426554012188925,
+      "tokens_seen": 2243231744
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032342320654790566,
+      "loss": 1.3726,
+      "theoretical_loss": 3.4265378292453983,
+      "tokens_seen": 2243362816
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032338308457711447,
+      "loss": 1.3785,
+      "theoretical_loss": 3.426521647512085,
+      "tokens_seen": 2243493888
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003233429626063232,
+      "loss": 1.4314,
+      "theoretical_loss": 3.4265054669888224,
+      "tokens_seen": 2243624960
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.000323302840635532,
+      "loss": 1.4221,
+      "theoretical_loss": 3.4264892876754507,
+      "tokens_seen": 2243756032
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032326271866474083,
+      "loss": 1.3157,
+      "theoretical_loss": 3.4264731095718077,
+      "tokens_seen": 2243887104
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003232225966939496,
+      "loss": 1.421,
+      "theoretical_loss": 3.4264569326777323,
+      "tokens_seen": 2244018176
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032318247472315844,
+      "loss": 1.3587,
+      "theoretical_loss": 3.426440756993064,
+      "tokens_seen": 2244149248
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003231423527523672,
+      "loss": 1.4534,
+      "theoretical_loss": 3.426424582517642,
+      "tokens_seen": 2244280320
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.000323102230781576,
+      "loss": 1.3983,
+      "theoretical_loss": 3.4264084092513043,
+      "tokens_seen": 2244411392
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003230621088107848,
+      "loss": 1.3035,
+      "theoretical_loss": 3.4263922371938906,
+      "tokens_seen": 2244542464
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032302198683999356,
+      "loss": 1.3633,
+      "theoretical_loss": 3.42637606634524,
+      "tokens_seen": 2244673536
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032298186486920237,
+      "loss": 1.3581,
+      "theoretical_loss": 3.426359896705191,
+      "tokens_seen": 2244804608
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003229417428984112,
+      "loss": 1.3719,
+      "theoretical_loss": 3.426343728273583,
+      "tokens_seen": 2244935680
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032290162092762,
+      "loss": 1.3894,
+      "theoretical_loss": 3.4263275610502557,
+      "tokens_seen": 2245066752
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032286149895682873,
+      "loss": 1.4105,
+      "theoretical_loss": 3.4263113950350474,
+      "tokens_seen": 2245197824
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003228213769860376,
+      "loss": 1.3477,
+      "theoretical_loss": 3.426295230227798,
+      "tokens_seen": 2245328896
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032278125501524634,
+      "loss": 1.3638,
+      "theoretical_loss": 3.4262790666283456,
+      "tokens_seen": 2245459968
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032274113304445515,
+      "loss": 1.3611,
+      "theoretical_loss": 3.426262904236531,
+      "tokens_seen": 2245591040
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032270101107366396,
+      "loss": 1.3497,
+      "theoretical_loss": 3.4262467430521926,
+      "tokens_seen": 2245722112
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003226608891028727,
+      "loss": 1.428,
+      "theoretical_loss": 3.42623058307517,
+      "tokens_seen": 2245853184
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032262076713208157,
+      "loss": 1.4054,
+      "theoretical_loss": 3.4262144243053023,
+      "tokens_seen": 2245984256
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003225806451612903,
+      "loss": 1.2992,
+      "theoretical_loss": 3.426198266742429,
+      "tokens_seen": 2246115328
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032254052319049913,
+      "loss": 1.3577,
+      "theoretical_loss": 3.4261821103863896,
+      "tokens_seen": 2246246400
+    },
+    {
+      "epoch": 0.36,
+      "objective/train/advantage_avg": 0.49195271730422974,
+      "objective/train/docs_used": 817879,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.291707158088684,
+      "objective/train/original_loss": 1.2917070388793945,
+      "objective/train/theoretical_loss": 3.4261659552370234,
+      "objective/train/tokens_used": 616836576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24257628619670868,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504285097122192,
+      "objective/train/weighted_lm_loss": 1.3568472862243652,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4261659552370234,
+      "tokens_seen": 2246377472
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032250040121970793,
+      "loss": 1.3542,
+      "theoretical_loss": 3.4261659552370234,
+      "tokens_seen": 2246377472
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003224602792489167,
+      "loss": 1.3884,
+      "theoretical_loss": 3.42614980129417,
+      "tokens_seen": 2246508544
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003224201572781255,
+      "loss": 1.2792,
+      "theoretical_loss": 3.426133648557669,
+      "tokens_seen": 2246639616
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003223800353073343,
+      "loss": 1.4629,
+      "theoretical_loss": 3.4261174970273593,
+      "tokens_seen": 2246770688
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003223399133365431,
+      "loss": 1.4325,
+      "theoretical_loss": 3.4261013467030814,
+      "tokens_seen": 2246901760
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032229979136575186,
+      "loss": 1.342,
+      "theoretical_loss": 3.4260851975846744,
+      "tokens_seen": 2247032832
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003222596693949607,
+      "loss": 1.3303,
+      "theoretical_loss": 3.4260690496719786,
+      "tokens_seen": 2247163904
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032221954742416947,
+      "loss": 1.327,
+      "theoretical_loss": 3.426052902964832,
+      "tokens_seen": 2247294976
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003221794254533782,
+      "loss": 1.3928,
+      "theoretical_loss": 3.426036757463076,
+      "tokens_seen": 2247426048
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003221393034825871,
+      "loss": 1.3399,
+      "theoretical_loss": 3.42602061316655,
+      "tokens_seen": 2247557120
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032209918151179584,
+      "loss": 1.2751,
+      "theoretical_loss": 3.426004470075093,
+      "tokens_seen": 2247688192
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003220590595410047,
+      "loss": 1.3344,
+      "theoretical_loss": 3.425988328188545,
+      "tokens_seen": 2247819264
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032201893757021345,
+      "loss": 1.3901,
+      "theoretical_loss": 3.425972187506747,
+      "tokens_seen": 2247950336
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032197881559942225,
+      "loss": 1.4046,
+      "theoretical_loss": 3.425956048029537,
+      "tokens_seen": 2248081408
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032193869362863106,
+      "loss": 1.3536,
+      "theoretical_loss": 3.4259399097567558,
+      "tokens_seen": 2248212480
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032189857165783987,
+      "loss": 1.3664,
+      "theoretical_loss": 3.425923772688243,
+      "tokens_seen": 2248343552
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003218584496870486,
+      "loss": 1.3716,
+      "theoretical_loss": 3.4259076368238395,
+      "tokens_seen": 2248474624
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003218183277162574,
+      "loss": 1.3973,
+      "theoretical_loss": 3.4258915021633847,
+      "tokens_seen": 2248605696
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032177820574546623,
+      "loss": 1.4364,
+      "theoretical_loss": 3.425875368706718,
+      "tokens_seen": 2248736768
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.000321738083774675,
+      "loss": 1.2976,
+      "theoretical_loss": 3.42585923645368,
+      "tokens_seen": 2248867840
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032169796180388384,
+      "loss": 1.4059,
+      "theoretical_loss": 3.4258431054041107,
+      "tokens_seen": 2248998912
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003216578398330926,
+      "loss": 1.3013,
+      "theoretical_loss": 3.4258269755578503,
+      "tokens_seen": 2249129984
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003216177178623014,
+      "loss": 1.3543,
+      "theoretical_loss": 3.4258108469147386,
+      "tokens_seen": 2249261056
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003215775958915102,
+      "loss": 1.2876,
+      "theoretical_loss": 3.425794719474616,
+      "tokens_seen": 2249392128
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032153747392071896,
+      "loss": 1.3544,
+      "theoretical_loss": 3.4257785932373226,
+      "tokens_seen": 2249523200
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003214973519499278,
+      "loss": 1.3436,
+      "theoretical_loss": 3.4257624682026986,
+      "tokens_seen": 2249654272
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003214572299791366,
+      "loss": 1.3105,
+      "theoretical_loss": 3.4257463443705847,
+      "tokens_seen": 2249785344
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003214171080083454,
+      "loss": 1.382,
+      "theoretical_loss": 3.42573022174082,
+      "tokens_seen": 2249916416
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003213769860375542,
+      "loss": 1.349,
+      "theoretical_loss": 3.425714100313246,
+      "tokens_seen": 2250047488
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.000321336864066763,
+      "loss": 1.3699,
+      "theoretical_loss": 3.425697980087702,
+      "tokens_seen": 2250178560
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032129674209597174,
+      "loss": 1.4187,
+      "theoretical_loss": 3.42568186106403,
+      "tokens_seen": 2250309632
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032125662012518055,
+      "loss": 1.2829,
+      "theoretical_loss": 3.4256657432420683,
+      "tokens_seen": 2250440704
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032121649815438936,
+      "loss": 1.2885,
+      "theoretical_loss": 3.425649626621659,
+      "tokens_seen": 2250571776
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003211763761835981,
+      "loss": 1.369,
+      "theoretical_loss": 3.425633511202642,
+      "tokens_seen": 2250702848
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032113625421280697,
+      "loss": 1.2911,
+      "theoretical_loss": 3.4256173969848573,
+      "tokens_seen": 2250833920
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003210961322420157,
+      "loss": 1.4022,
+      "theoretical_loss": 3.4256012839681462,
+      "tokens_seen": 2250964992
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032105601027122453,
+      "loss": 1.4178,
+      "theoretical_loss": 3.4255851721523487,
+      "tokens_seen": 2251096064
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032101588830043333,
+      "loss": 1.3547,
+      "theoretical_loss": 3.4255690615373053,
+      "tokens_seen": 2251227136
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003209757663296421,
+      "loss": 1.382,
+      "theoretical_loss": 3.425552952122857,
+      "tokens_seen": 2251358208
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003209356443588509,
+      "loss": 1.359,
+      "theoretical_loss": 3.4255368439088447,
+      "tokens_seen": 2251489280
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003208955223880597,
+      "loss": 1.3577,
+      "theoretical_loss": 3.425520736895108,
+      "tokens_seen": 2251620352
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003208554004172685,
+      "loss": 1.3541,
+      "theoretical_loss": 3.425504631081489,
+      "tokens_seen": 2251751424
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003208152784464773,
+      "loss": 1.3456,
+      "theoretical_loss": 3.425488526467827,
+      "tokens_seen": 2251882496
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003207751564756861,
+      "loss": 1.3557,
+      "theoretical_loss": 3.4254724230539644,
+      "tokens_seen": 2252013568
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00032073503450489487,
+      "loss": 1.3344,
+      "theoretical_loss": 3.4254563208397406,
+      "tokens_seen": 2252144640
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003206949125341037,
+      "loss": 1.3752,
+      "theoretical_loss": 3.4254402198249965,
+      "tokens_seen": 2252275712
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003206547905633125,
+      "loss": 1.3352,
+      "theoretical_loss": 3.425424120009574,
+      "tokens_seen": 2252406784
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00032061466859252123,
+      "loss": 1.3582,
+      "theoretical_loss": 3.425408021393313,
+      "tokens_seen": 2252537856
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003205745466217301,
+      "loss": 1.3274,
+      "theoretical_loss": 3.4253919239760546,
+      "tokens_seen": 2252668928
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00032053442465093885,
+      "loss": 1.332,
+      "theoretical_loss": 3.42537582775764,
+      "tokens_seen": 2252800000
+    },
+    {
+      "epoch": 0.37,
+      "objective/train/advantage_avg": 0.4918114244937897,
+      "objective/train/docs_used": 820362,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3642363548278809,
+      "objective/train/original_loss": 1.3642362356185913,
+      "objective/train/theoretical_loss": 3.4253597327379097,
+      "objective/train/tokens_used": 623390176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2422666996717453,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504127740859985,
+      "objective/train/weighted_lm_loss": 1.4330031871795654,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4253597327379097,
+      "tokens_seen": 2252931072
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00032049430268014765,
+      "loss": 1.3862,
+      "theoretical_loss": 3.4253597327379097,
+      "tokens_seen": 2252931072
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00032045418070935646,
+      "loss": 1.3327,
+      "theoretical_loss": 3.4253436389167056,
+      "tokens_seen": 2253062144
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003204140587385652,
+      "loss": 1.345,
+      "theoretical_loss": 3.425327546293868,
+      "tokens_seen": 2253193216
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.000320373936767774,
+      "loss": 1.382,
+      "theoretical_loss": 3.425311454869238,
+      "tokens_seen": 2253324288
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003203338147969828,
+      "loss": 1.3236,
+      "theoretical_loss": 3.425295364642657,
+      "tokens_seen": 2253455360
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00032029369282619163,
+      "loss": 1.3315,
+      "theoretical_loss": 3.425279275613966,
+      "tokens_seen": 2253586432
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00032025357085540044,
+      "loss": 1.3972,
+      "theoretical_loss": 3.425263187783006,
+      "tokens_seen": 2253717504
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00032021344888460924,
+      "loss": 1.3618,
+      "theoretical_loss": 3.4252471011496186,
+      "tokens_seen": 2253848576
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.000320173326913818,
+      "loss": 1.4364,
+      "theoretical_loss": 3.4252310157136443,
+      "tokens_seen": 2253979648
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00032013320494302686,
+      "loss": 1.3915,
+      "theoretical_loss": 3.425214931474925,
+      "tokens_seen": 2254110720
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003200930829722356,
+      "loss": 1.3379,
+      "theoretical_loss": 3.425198848433302,
+      "tokens_seen": 2254241792
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00032005296100144436,
+      "loss": 1.335,
+      "theoretical_loss": 3.425182766588616,
+      "tokens_seen": 2254372864
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003200128390306532,
+      "loss": 1.3884,
+      "theoretical_loss": 3.4251666859407086,
+      "tokens_seen": 2254503936
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.000319972717059862,
+      "loss": 1.3411,
+      "theoretical_loss": 3.4251506064894217,
+      "tokens_seen": 2254635008
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003199325950890708,
+      "loss": 1.3039,
+      "theoretical_loss": 3.425134528234596,
+      "tokens_seen": 2254766080
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003198924731182796,
+      "loss": 1.3922,
+      "theoretical_loss": 3.4251184511760733,
+      "tokens_seen": 2254897152
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003198523511474884,
+      "loss": 1.4012,
+      "theoretical_loss": 3.425102375313695,
+      "tokens_seen": 2255028224
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031981222917669714,
+      "loss": 1.4039,
+      "theoretical_loss": 3.4250863006473025,
+      "tokens_seen": 2255159296
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031977210720590595,
+      "loss": 1.3403,
+      "theoretical_loss": 3.4250702271767373,
+      "tokens_seen": 2255290368
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031973198523511476,
+      "loss": 1.3377,
+      "theoretical_loss": 3.425054154901841,
+      "tokens_seen": 2255421440
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003196918632643235,
+      "loss": 1.3851,
+      "theoretical_loss": 3.4250380838224554,
+      "tokens_seen": 2255552512
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031965174129353237,
+      "loss": 1.344,
+      "theoretical_loss": 3.4250220139384213,
+      "tokens_seen": 2255683584
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003196116193227411,
+      "loss": 1.3721,
+      "theoretical_loss": 3.4250059452495814,
+      "tokens_seen": 2255814656
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031957149735195,
+      "loss": 1.3115,
+      "theoretical_loss": 3.424989877755777,
+      "tokens_seen": 2255945728
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031953137538115873,
+      "loss": 1.3806,
+      "theoretical_loss": 3.4249738114568498,
+      "tokens_seen": 2256076800
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003194912534103675,
+      "loss": 1.3211,
+      "theoretical_loss": 3.424957746352641,
+      "tokens_seen": 2256207872
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031945113143957635,
+      "loss": 1.3332,
+      "theoretical_loss": 3.4249416824429932,
+      "tokens_seen": 2256338944
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003194110094687851,
+      "loss": 1.3756,
+      "theoretical_loss": 3.4249256197277473,
+      "tokens_seen": 2256470016
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003193708874979939,
+      "loss": 1.3841,
+      "theoretical_loss": 3.424909558206746,
+      "tokens_seen": 2256601088
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003193307655272027,
+      "loss": 1.3546,
+      "theoretical_loss": 3.4248934978798307,
+      "tokens_seen": 2256732160
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003192906435564115,
+      "loss": 1.4018,
+      "theoretical_loss": 3.4248774387468433,
+      "tokens_seen": 2256863232
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031925052158562027,
+      "loss": 1.3783,
+      "theoretical_loss": 3.424861380807626,
+      "tokens_seen": 2256994304
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003192103996148291,
+      "loss": 1.3253,
+      "theoretical_loss": 3.4248453240620202,
+      "tokens_seen": 2257125376
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003191702776440379,
+      "loss": 1.3895,
+      "theoretical_loss": 3.424829268509868,
+      "tokens_seen": 2257256448
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031913015567324663,
+      "loss": 1.3616,
+      "theoretical_loss": 3.424813214151012,
+      "tokens_seen": 2257387520
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003190900337024555,
+      "loss": 1.3337,
+      "theoretical_loss": 3.4247971609852934,
+      "tokens_seen": 2257518592
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031904991173166425,
+      "loss": 1.3544,
+      "theoretical_loss": 3.424781109012555,
+      "tokens_seen": 2257649664
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003190097897608731,
+      "loss": 1.366,
+      "theoretical_loss": 3.424765058232638,
+      "tokens_seen": 2257780736
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031896966779008186,
+      "loss": 1.3256,
+      "theoretical_loss": 3.424749008645386,
+      "tokens_seen": 2257911808
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003189295458192906,
+      "loss": 1.3351,
+      "theoretical_loss": 3.4247329602506396,
+      "tokens_seen": 2258042880
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031888942384849947,
+      "loss": 1.4015,
+      "theoretical_loss": 3.4247169130482416,
+      "tokens_seen": 2258173952
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003188493018777082,
+      "loss": 1.4236,
+      "theoretical_loss": 3.4247008670380343,
+      "tokens_seen": 2258305024
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031880917990691703,
+      "loss": 1.3765,
+      "theoretical_loss": 3.42468482221986,
+      "tokens_seen": 2258436096
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031876905793612584,
+      "loss": 1.385,
+      "theoretical_loss": 3.424668778593561,
+      "tokens_seen": 2258567168
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031872893596533464,
+      "loss": 1.3512,
+      "theoretical_loss": 3.424652736158979,
+      "tokens_seen": 2258698240
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003186888139945434,
+      "loss": 1.3376,
+      "theoretical_loss": 3.424636694915957,
+      "tokens_seen": 2258829312
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031864869202375226,
+      "loss": 1.2936,
+      "theoretical_loss": 3.424620654864337,
+      "tokens_seen": 2258960384
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.000318608570052961,
+      "loss": 1.378,
+      "theoretical_loss": 3.424604616003961,
+      "tokens_seen": 2259091456
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031856844808216976,
+      "loss": 1.2797,
+      "theoretical_loss": 3.424588578334673,
+      "tokens_seen": 2259222528
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003185283261113786,
+      "loss": 1.378,
+      "theoretical_loss": 3.424572541856314,
+      "tokens_seen": 2259353600
+    },
+    {
+      "epoch": 0.37,
+      "objective/train/advantage_avg": 0.49134349822998047,
+      "objective/train/docs_used": 822867,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.337546467781067,
+      "objective/train/original_loss": 1.3375463485717773,
+      "objective/train/theoretical_loss": 3.4245565065687265,
+      "objective/train/tokens_used": 629943776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24214965105056763,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503653287887573,
+      "objective/train/weighted_lm_loss": 1.4047635793685913,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4245565065687265,
+      "tokens_seen": 2259484672
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003184882041405874,
+      "loss": 1.3729,
+      "theoretical_loss": 3.4245565065687265,
+      "tokens_seen": 2259484672
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003184480821697962,
+      "loss": 1.3008,
+      "theoretical_loss": 3.4245404724717536,
+      "tokens_seen": 2259615744
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.000318407960199005,
+      "loss": 1.3299,
+      "theoretical_loss": 3.424524439565238,
+      "tokens_seen": 2259746816
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003183678382282138,
+      "loss": 1.3784,
+      "theoretical_loss": 3.4245084078490216,
+      "tokens_seen": 2259877888
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003183277162574226,
+      "loss": 1.3042,
+      "theoretical_loss": 3.424492377322947,
+      "tokens_seen": 2260008960
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031828759428663135,
+      "loss": 1.3257,
+      "theoretical_loss": 3.4244763479868574,
+      "tokens_seen": 2260140032
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031824747231584016,
+      "loss": 1.2953,
+      "theoretical_loss": 3.4244603198405956,
+      "tokens_seen": 2260271104
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031820735034504896,
+      "loss": 1.3548,
+      "theoretical_loss": 3.4244442928840035,
+      "tokens_seen": 2260402176
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031816722837425777,
+      "loss": 1.3093,
+      "theoretical_loss": 3.424428267116924,
+      "tokens_seen": 2260533248
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003181271064034665,
+      "loss": 1.3476,
+      "theoretical_loss": 3.4244122425392005,
+      "tokens_seen": 2260664320
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003180869844326754,
+      "loss": 1.4463,
+      "theoretical_loss": 3.424396219150675,
+      "tokens_seen": 2260795392
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031804686246188413,
+      "loss": 1.4129,
+      "theoretical_loss": 3.4243801969511907,
+      "tokens_seen": 2260926464
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003180067404910929,
+      "loss": 1.336,
+      "theoretical_loss": 3.4243641759405907,
+      "tokens_seen": 2261057536
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031796661852030175,
+      "loss": 1.3915,
+      "theoretical_loss": 3.4243481561187172,
+      "tokens_seen": 2261188608
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003179264965495105,
+      "loss": 1.3151,
+      "theoretical_loss": 3.424332137485414,
+      "tokens_seen": 2261319680
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003178863745787193,
+      "loss": 1.3513,
+      "theoretical_loss": 3.4243161200405225,
+      "tokens_seen": 2261450752
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003178462526079281,
+      "loss": 1.3668,
+      "theoretical_loss": 3.424300103783888,
+      "tokens_seen": 2261581824
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003178061306371369,
+      "loss": 1.3843,
+      "theoretical_loss": 3.424284088715351,
+      "tokens_seen": 2261712896
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031776600866634567,
+      "loss": 1.4359,
+      "theoretical_loss": 3.4242680748347563,
+      "tokens_seen": 2261843968
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003177258866955545,
+      "loss": 1.3494,
+      "theoretical_loss": 3.424252062141946,
+      "tokens_seen": 2261975040
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003176857647247633,
+      "loss": 1.2523,
+      "theoretical_loss": 3.4242360506367637,
+      "tokens_seen": 2262106112
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003176456427539721,
+      "loss": 1.3759,
+      "theoretical_loss": 3.424220040319052,
+      "tokens_seen": 2262237184
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003176055207831809,
+      "loss": 1.3701,
+      "theoretical_loss": 3.424204031188655,
+      "tokens_seen": 2262368256
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031756539881238965,
+      "loss": 1.4549,
+      "theoretical_loss": 3.4241880232454145,
+      "tokens_seen": 2262499328
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003175252768415985,
+      "loss": 1.3666,
+      "theoretical_loss": 3.424172016489175,
+      "tokens_seen": 2262630400
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031748515487080726,
+      "loss": 1.3549,
+      "theoretical_loss": 3.4241560109197784,
+      "tokens_seen": 2262761472
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.000317445032900016,
+      "loss": 1.3699,
+      "theoretical_loss": 3.424140006537069,
+      "tokens_seen": 2262892544
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031740491092922487,
+      "loss": 1.3452,
+      "theoretical_loss": 3.42412400334089,
+      "tokens_seen": 2263023616
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003173647889584336,
+      "loss": 1.2858,
+      "theoretical_loss": 3.4241080013310845,
+      "tokens_seen": 2263154688
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031732466698764243,
+      "loss": 1.3566,
+      "theoretical_loss": 3.424092000507496,
+      "tokens_seen": 2263285760
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031728454501685124,
+      "loss": 1.3585,
+      "theoretical_loss": 3.4240760008699676,
+      "tokens_seen": 2263416832
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031724442304606004,
+      "loss": 1.3501,
+      "theoretical_loss": 3.4240600024183427,
+      "tokens_seen": 2263547904
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003172043010752688,
+      "loss": 1.4252,
+      "theoretical_loss": 3.424044005152465,
+      "tokens_seen": 2263678976
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031716417910447766,
+      "loss": 1.3871,
+      "theoretical_loss": 3.4240280090721775,
+      "tokens_seen": 2263810048
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003171240571336864,
+      "loss": 1.404,
+      "theoretical_loss": 3.4240120141773245,
+      "tokens_seen": 2263941120
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003170839351628952,
+      "loss": 1.3761,
+      "theoretical_loss": 3.423996020467749,
+      "tokens_seen": 2264072192
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.000317043813192104,
+      "loss": 1.3585,
+      "theoretical_loss": 3.4239800279432946,
+      "tokens_seen": 2264203264
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031700369122131277,
+      "loss": 1.3801,
+      "theoretical_loss": 3.423964036603805,
+      "tokens_seen": 2264334336
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031696356925052163,
+      "loss": 1.3013,
+      "theoretical_loss": 3.4239480464491234,
+      "tokens_seen": 2264465408
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003169234472797304,
+      "loss": 1.3172,
+      "theoretical_loss": 3.423932057479094,
+      "tokens_seen": 2264596480
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003168833253089392,
+      "loss": 1.3756,
+      "theoretical_loss": 3.4239160696935604,
+      "tokens_seen": 2264727552
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.000316843203338148,
+      "loss": 1.396,
+      "theoretical_loss": 3.4239000830923656,
+      "tokens_seen": 2264858624
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031680308136735675,
+      "loss": 1.4,
+      "theoretical_loss": 3.423884097675354,
+      "tokens_seen": 2264989696
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031676295939656556,
+      "loss": 1.4207,
+      "theoretical_loss": 3.42386811344237,
+      "tokens_seen": 2265120768
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031672283742577436,
+      "loss": 1.3812,
+      "theoretical_loss": 3.4238521303932563,
+      "tokens_seen": 2265251840
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031668271545498317,
+      "loss": 1.3782,
+      "theoretical_loss": 3.423836148527857,
+      "tokens_seen": 2265382912
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003166425934841919,
+      "loss": 1.3049,
+      "theoretical_loss": 3.423820167846016,
+      "tokens_seen": 2265513984
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003166024715134008,
+      "loss": 1.3492,
+      "theoretical_loss": 3.4238041883475767,
+      "tokens_seen": 2265645056
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031656234954260953,
+      "loss": 1.2915,
+      "theoretical_loss": 3.423788210032384,
+      "tokens_seen": 2265776128
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003165222275718183,
+      "loss": 1.4189,
+      "theoretical_loss": 3.4237722329002813,
+      "tokens_seen": 2265907200
+    },
+    {
+      "epoch": 0.37,
+      "objective/train/advantage_avg": 0.49391239881515503,
+      "objective/train/docs_used": 825438,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2457987070083618,
+      "objective/train/original_loss": 1.2457985877990723,
+      "objective/train/theoretical_loss": 3.423756256951113,
+      "objective/train/tokens_used": 636497376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24460281431674957,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050634741783142,
+      "objective/train/weighted_lm_loss": 1.308835744857788,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.423756256951113,
+      "tokens_seen": 2266038272
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031648210560102715,
+      "loss": 1.3173,
+      "theoretical_loss": 3.423756256951113,
+      "tokens_seen": 2266038272
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003164419836302359,
+      "loss": 1.3232,
+      "theoretical_loss": 3.4237402821847223,
+      "tokens_seen": 2266169344
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031640186165944476,
+      "loss": 1.3132,
+      "theoretical_loss": 3.4237243086009537,
+      "tokens_seen": 2266300416
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003163617396886535,
+      "loss": 1.4733,
+      "theoretical_loss": 3.423708336199651,
+      "tokens_seen": 2266431488
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003163216177178623,
+      "loss": 1.3383,
+      "theoretical_loss": 3.423692364980659,
+      "tokens_seen": 2266562560
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003162814957470711,
+      "loss": 1.4201,
+      "theoretical_loss": 3.423676394943821,
+      "tokens_seen": 2266693632
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003162413737762799,
+      "loss": 1.4072,
+      "theoretical_loss": 3.423660426088982,
+      "tokens_seen": 2266824704
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003162012518054887,
+      "loss": 1.3433,
+      "theoretical_loss": 3.4236444584159846,
+      "tokens_seen": 2266955776
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003161611298346975,
+      "loss": 1.3784,
+      "theoretical_loss": 3.423628491924675,
+      "tokens_seen": 2267086848
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003161210078639063,
+      "loss": 1.3843,
+      "theoretical_loss": 3.423612526614896,
+      "tokens_seen": 2267217920
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031608088589311505,
+      "loss": 1.3698,
+      "theoretical_loss": 3.423596562486493,
+      "tokens_seen": 2267348992
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003160407639223239,
+      "loss": 1.3174,
+      "theoretical_loss": 3.423580599539309,
+      "tokens_seen": 2267480064
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031600064195153266,
+      "loss": 1.3145,
+      "theoretical_loss": 3.423564637773189,
+      "tokens_seen": 2267611136
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003159605199807414,
+      "loss": 1.3307,
+      "theoretical_loss": 3.423548677187978,
+      "tokens_seen": 2267742208
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031592039800995027,
+      "loss": 1.3244,
+      "theoretical_loss": 3.423532717783519,
+      "tokens_seen": 2267873280
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.000315880276039159,
+      "loss": 1.3516,
+      "theoretical_loss": 3.4235167595596576,
+      "tokens_seen": 2268004352
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003158401540683679,
+      "loss": 1.4466,
+      "theoretical_loss": 3.4235008025162377,
+      "tokens_seen": 2268135424
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031580003209757664,
+      "loss": 1.3814,
+      "theoretical_loss": 3.423484846653104,
+      "tokens_seen": 2268266496
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031575991012678544,
+      "loss": 1.336,
+      "theoretical_loss": 3.4234688919701,
+      "tokens_seen": 2268397568
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00031571978815599425,
+      "loss": 1.3646,
+      "theoretical_loss": 3.423452938467072,
+      "tokens_seen": 2268528640
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.000315679666185203,
+      "loss": 1.2822,
+      "theoretical_loss": 3.4234369861438636,
+      "tokens_seen": 2268659712
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003156395442144118,
+      "loss": 1.3676,
+      "theoretical_loss": 3.4234210350003194,
+      "tokens_seen": 2268790784
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003155994222436206,
+      "loss": 1.3688,
+      "theoretical_loss": 3.4234050850362836,
+      "tokens_seen": 2268921856
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003155593002728294,
+      "loss": 1.3564,
+      "theoretical_loss": 3.423389136251602,
+      "tokens_seen": 2269052928
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031551917830203817,
+      "loss": 1.264,
+      "theoretical_loss": 3.423373188646118,
+      "tokens_seen": 2269184000
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031547905633124703,
+      "loss": 1.3305,
+      "theoretical_loss": 3.4233572422196774,
+      "tokens_seen": 2269315072
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003154389343604558,
+      "loss": 1.3602,
+      "theoretical_loss": 3.4233412969721244,
+      "tokens_seen": 2269446144
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003153988123896646,
+      "loss": 1.3547,
+      "theoretical_loss": 3.4233253529033036,
+      "tokens_seen": 2269577216
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003153586904188734,
+      "loss": 1.3113,
+      "theoretical_loss": 3.42330941001306,
+      "tokens_seen": 2269708288
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031531856844808215,
+      "loss": 1.3315,
+      "theoretical_loss": 3.423293468301239,
+      "tokens_seen": 2269839360
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031527844647729096,
+      "loss": 1.3998,
+      "theoretical_loss": 3.423277527767684,
+      "tokens_seen": 2269970432
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031523832450649976,
+      "loss": 1.2732,
+      "theoretical_loss": 3.4232615884122413,
+      "tokens_seen": 2270101504
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031519820253570857,
+      "loss": 1.2873,
+      "theoretical_loss": 3.4232456502347555,
+      "tokens_seen": 2270232576
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003151580805649174,
+      "loss": 1.2988,
+      "theoretical_loss": 3.4232297132350706,
+      "tokens_seen": 2270363648
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003151179585941262,
+      "loss": 1.3127,
+      "theoretical_loss": 3.423213777413033,
+      "tokens_seen": 2270494720
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031507783662333493,
+      "loss": 1.3672,
+      "theoretical_loss": 3.423197842768487,
+      "tokens_seen": 2270625792
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031503771465254374,
+      "loss": 1.3556,
+      "theoretical_loss": 3.4231819093012774,
+      "tokens_seen": 2270756864
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031499759268175255,
+      "loss": 1.3487,
+      "theoretical_loss": 3.4231659770112497,
+      "tokens_seen": 2270887936
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003149574707109613,
+      "loss": 1.2191,
+      "theoretical_loss": 3.4231500458982485,
+      "tokens_seen": 2271019008
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031491734874017016,
+      "loss": 1.3754,
+      "theoretical_loss": 3.423134115962119,
+      "tokens_seen": 2271150080
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003148772267693789,
+      "loss": 1.4345,
+      "theoretical_loss": 3.423118187202707,
+      "tokens_seen": 2271281152
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003148371047985877,
+      "loss": 1.2865,
+      "theoretical_loss": 3.4231022596198577,
+      "tokens_seen": 2271412224
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003147969828277965,
+      "loss": 1.3635,
+      "theoretical_loss": 3.4230863332134156,
+      "tokens_seen": 2271543296
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003147568608570053,
+      "loss": 1.2894,
+      "theoretical_loss": 3.423070407983226,
+      "tokens_seen": 2271674368
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003147167388862141,
+      "loss": 1.3549,
+      "theoretical_loss": 3.423054483929134,
+      "tokens_seen": 2271805440
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003146766169154229,
+      "loss": 1.3688,
+      "theoretical_loss": 3.423038561050986,
+      "tokens_seen": 2271936512
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003146364949446317,
+      "loss": 1.307,
+      "theoretical_loss": 3.423022639348626,
+      "tokens_seen": 2272067584
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031459637297384045,
+      "loss": 1.3817,
+      "theoretical_loss": 3.4230067188219,
+      "tokens_seen": 2272198656
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003145562510030493,
+      "loss": 1.3486,
+      "theoretical_loss": 3.4229907994706537,
+      "tokens_seen": 2272329728
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031451612903225806,
+      "loss": 1.3594,
+      "theoretical_loss": 3.4229748812947323,
+      "tokens_seen": 2272460800
+    },
+    {
+      "epoch": 0.38,
+      "objective/train/advantage_avg": 0.49286961555480957,
+      "objective/train/docs_used": 827716,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2670907974243164,
+      "objective/train/original_loss": 1.2670907974243164,
+      "objective/train/theoretical_loss": 3.4229589642939806,
+      "objective/train/tokens_used": 643050976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24323828518390656,
+      "objective/train/value_max": -0.5,
+      "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": 1.331016182899475,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9834713935852051,
+      "theoretical_loss": 3.4229589642939806,
+      "tokens_seen": 2272591872
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031447600706146687,
+      "loss": 1.2894,
+      "theoretical_loss": 3.4229589642939806,
+      "tokens_seen": 2272591872
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031443588509067567,
+      "loss": 1.3753,
+      "theoretical_loss": 3.4229430484682446,
+      "tokens_seen": 2272722944
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003143957631198844,
+      "loss": 1.3638,
+      "theoretical_loss": 3.42292713381737,
+      "tokens_seen": 2272854016
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003143556411490933,
+      "loss": 1.3525,
+      "theoretical_loss": 3.4229112203412018,
+      "tokens_seen": 2272985088
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031431551917830204,
+      "loss": 1.3231,
+      "theoretical_loss": 3.4228953080395863,
+      "tokens_seen": 2273116160
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031427539720751084,
+      "loss": 1.3395,
+      "theoretical_loss": 3.422879396912368,
+      "tokens_seen": 2273247232
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031423527523671965,
+      "loss": 1.3521,
+      "theoretical_loss": 3.4228634869593937,
+      "tokens_seen": 2273378304
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003141951532659284,
+      "loss": 1.3704,
+      "theoretical_loss": 3.4228475781805083,
+      "tokens_seen": 2273509376
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003141550312951372,
+      "loss": 1.41,
+      "theoretical_loss": 3.4228316705755577,
+      "tokens_seen": 2273640448
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.000314114909324346,
+      "loss": 1.4493,
+      "theoretical_loss": 3.4228157641443873,
+      "tokens_seen": 2273771520
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003140747873535548,
+      "loss": 1.3516,
+      "theoretical_loss": 3.4227998588868433,
+      "tokens_seen": 2273902592
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031403466538276357,
+      "loss": 1.3886,
+      "theoretical_loss": 3.4227839548027714,
+      "tokens_seen": 2274033664
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031399454341197243,
+      "loss": 1.3034,
+      "theoretical_loss": 3.4227680518920174,
+      "tokens_seen": 2274164736
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003139544214411812,
+      "loss": 1.413,
+      "theoretical_loss": 3.4227521501544267,
+      "tokens_seen": 2274295808
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031391429947039004,
+      "loss": 1.3905,
+      "theoretical_loss": 3.4227362495898457,
+      "tokens_seen": 2274426880
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003138741774995988,
+      "loss": 1.4471,
+      "theoretical_loss": 3.42272035019812,
+      "tokens_seen": 2274557952
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031383405552880755,
+      "loss": 1.3329,
+      "theoretical_loss": 3.4227044519790955,
+      "tokens_seen": 2274689024
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003137939335580164,
+      "loss": 1.2768,
+      "theoretical_loss": 3.4226885549326176,
+      "tokens_seen": 2274820096
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031375381158722516,
+      "loss": 1.3081,
+      "theoretical_loss": 3.4226726590585335,
+      "tokens_seen": 2274951168
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031371368961643397,
+      "loss": 1.4036,
+      "theoretical_loss": 3.4226567643566885,
+      "tokens_seen": 2275082240
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003136735676456428,
+      "loss": 1.3698,
+      "theoretical_loss": 3.4226408708269282,
+      "tokens_seen": 2275213312
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003136334456748516,
+      "loss": 1.3344,
+      "theoretical_loss": 3.4226249784691,
+      "tokens_seen": 2275344384
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031359332370406033,
+      "loss": 1.3762,
+      "theoretical_loss": 3.4226090872830484,
+      "tokens_seen": 2275475456
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031355320173326914,
+      "loss": 1.3505,
+      "theoretical_loss": 3.42259319726862,
+      "tokens_seen": 2275606528
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031351307976247795,
+      "loss": 1.4014,
+      "theoretical_loss": 3.422577308425662,
+      "tokens_seen": 2275737600
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003134729577916867,
+      "loss": 1.3751,
+      "theoretical_loss": 3.422561420754019,
+      "tokens_seen": 2275868672
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031343283582089556,
+      "loss": 1.3289,
+      "theoretical_loss": 3.4225455342535382,
+      "tokens_seen": 2275999744
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003133927138501043,
+      "loss": 1.3882,
+      "theoretical_loss": 3.4225296489240655,
+      "tokens_seen": 2276130816
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003133525918793131,
+      "loss": 1.3539,
+      "theoretical_loss": 3.4225137647654473,
+      "tokens_seen": 2276261888
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003133124699085219,
+      "loss": 1.3662,
+      "theoretical_loss": 3.42249788177753,
+      "tokens_seen": 2276392960
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003132723479377307,
+      "loss": 1.314,
+      "theoretical_loss": 3.4224819999601594,
+      "tokens_seen": 2276524032
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031323222596693954,
+      "loss": 1.3855,
+      "theoretical_loss": 3.422466119313182,
+      "tokens_seen": 2276655104
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003131921039961483,
+      "loss": 1.3582,
+      "theoretical_loss": 3.4224502398364445,
+      "tokens_seen": 2276786176
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003131519820253571,
+      "loss": 1.3393,
+      "theoretical_loss": 3.4224343615297936,
+      "tokens_seen": 2276917248
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003131118600545659,
+      "loss": 1.3138,
+      "theoretical_loss": 3.422418484393075,
+      "tokens_seen": 2277048320
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003130717380837747,
+      "loss": 1.3559,
+      "theoretical_loss": 3.422402608426135,
+      "tokens_seen": 2277179392
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031303161611298346,
+      "loss": 1.3677,
+      "theoretical_loss": 3.422386733628821,
+      "tokens_seen": 2277310464
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031299149414219226,
+      "loss": 1.3552,
+      "theoretical_loss": 3.422370860000979,
+      "tokens_seen": 2277441536
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031295137217140107,
+      "loss": 1.361,
+      "theoretical_loss": 3.4223549875424557,
+      "tokens_seen": 2277572608
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003129112502006098,
+      "loss": 1.3713,
+      "theoretical_loss": 3.422339116253097,
+      "tokens_seen": 2277703680
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003128711282298187,
+      "loss": 1.4285,
+      "theoretical_loss": 3.4223232461327506,
+      "tokens_seen": 2277834752
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031283100625902744,
+      "loss": 1.3604,
+      "theoretical_loss": 3.4223073771812627,
+      "tokens_seen": 2277965824
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031279088428823624,
+      "loss": 1.3762,
+      "theoretical_loss": 3.4222915093984794,
+      "tokens_seen": 2278096896
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031275076231744505,
+      "loss": 1.3493,
+      "theoretical_loss": 3.4222756427842485,
+      "tokens_seen": 2278227968
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003127106403466538,
+      "loss": 1.279,
+      "theoretical_loss": 3.4222597773384154,
+      "tokens_seen": 2278359040
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003126705183758626,
+      "loss": 1.3681,
+      "theoretical_loss": 3.422243913060828,
+      "tokens_seen": 2278490112
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003126303964050714,
+      "loss": 1.4355,
+      "theoretical_loss": 3.422228049951333,
+      "tokens_seen": 2278621184
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003125902744342802,
+      "loss": 1.3675,
+      "theoretical_loss": 3.422212188009776,
+      "tokens_seen": 2278752256
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.000312550152463489,
+      "loss": 1.3096,
+      "theoretical_loss": 3.422196327236005,
+      "tokens_seen": 2278883328
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031251003049269783,
+      "loss": 1.2897,
+      "theoretical_loss": 3.422180467629867,
+      "tokens_seen": 2279014400
+    },
+    {
+      "epoch": 0.38,
+      "objective/train/advantage_avg": 0.494416743516922,
+      "objective/train/docs_used": 830070,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3059881925582886,
+      "objective/train/original_loss": 1.3059881925582886,
+      "objective/train/theoretical_loss": 3.4221646091912077,
+      "objective/train/tokens_used": 649604576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24475479125976562,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050686001777649,
+      "objective/train/weighted_lm_loss": 1.3720581531524658,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4221646091912077,
+      "tokens_seen": 2279145472
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003124699085219066,
+      "loss": 1.3496,
+      "theoretical_loss": 3.4221646091912077,
+      "tokens_seen": 2279145472
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031242978655111544,
+      "loss": 1.3532,
+      "theoretical_loss": 3.4221487519198748,
+      "tokens_seen": 2279276544
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003123896645803242,
+      "loss": 1.2694,
+      "theoretical_loss": 3.422132895815716,
+      "tokens_seen": 2279407616
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031234954260953295,
+      "loss": 1.419,
+      "theoretical_loss": 3.422117040878577,
+      "tokens_seen": 2279538688
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003123094206387418,
+      "loss": 1.3654,
+      "theoretical_loss": 3.422101187108305,
+      "tokens_seen": 2279669760
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031226929866795056,
+      "loss": 1.359,
+      "theoretical_loss": 3.4220853345047475,
+      "tokens_seen": 2279800832
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031222917669715937,
+      "loss": 1.387,
+      "theoretical_loss": 3.4220694830677516,
+      "tokens_seen": 2279931904
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003121890547263682,
+      "loss": 1.2961,
+      "theoretical_loss": 3.422053632797164,
+      "tokens_seen": 2280062976
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.000312148932755577,
+      "loss": 1.3346,
+      "theoretical_loss": 3.4220377836928324,
+      "tokens_seen": 2280194048
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031210881078478573,
+      "loss": 1.3553,
+      "theoretical_loss": 3.4220219357546036,
+      "tokens_seen": 2280325120
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031206868881399454,
+      "loss": 1.2967,
+      "theoretical_loss": 3.4220060889823243,
+      "tokens_seen": 2280456192
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031202856684320335,
+      "loss": 1.417,
+      "theoretical_loss": 3.4219902433758427,
+      "tokens_seen": 2280587264
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031198844487241215,
+      "loss": 1.3212,
+      "theoretical_loss": 3.421974398935005,
+      "tokens_seen": 2280718336
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031194832290162096,
+      "loss": 1.3987,
+      "theoretical_loss": 3.4219585556596597,
+      "tokens_seen": 2280849408
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003119082009308297,
+      "loss": 1.3073,
+      "theoretical_loss": 3.4219427135496527,
+      "tokens_seen": 2280980480
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031186807896003857,
+      "loss": 1.3646,
+      "theoretical_loss": 3.4219268726048324,
+      "tokens_seen": 2281111552
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003118279569892473,
+      "loss": 1.4089,
+      "theoretical_loss": 3.421911032825046,
+      "tokens_seen": 2281242624
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003117878350184561,
+      "loss": 1.3196,
+      "theoretical_loss": 3.42189519421014,
+      "tokens_seen": 2281373696
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031174771304766494,
+      "loss": 1.2598,
+      "theoretical_loss": 3.421879356759963,
+      "tokens_seen": 2281504768
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003117075910768737,
+      "loss": 1.3085,
+      "theoretical_loss": 3.421863520474362,
+      "tokens_seen": 2281635840
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003116674691060825,
+      "loss": 1.4017,
+      "theoretical_loss": 3.421847685353184,
+      "tokens_seen": 2281766912
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003116273471352913,
+      "loss": 1.3826,
+      "theoretical_loss": 3.421831851396277,
+      "tokens_seen": 2281897984
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003115872251645001,
+      "loss": 1.372,
+      "theoretical_loss": 3.4218160186034887,
+      "tokens_seen": 2282029056
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031154710319370886,
+      "loss": 1.3385,
+      "theoretical_loss": 3.421800186974666,
+      "tokens_seen": 2282160128
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031150698122291766,
+      "loss": 1.252,
+      "theoretical_loss": 3.421784356509657,
+      "tokens_seen": 2282291200
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031146685925212647,
+      "loss": 1.3572,
+      "theoretical_loss": 3.4217685272083096,
+      "tokens_seen": 2282422272
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003114267372813352,
+      "loss": 1.3391,
+      "theoretical_loss": 3.4217526990704705,
+      "tokens_seen": 2282553344
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003113866153105441,
+      "loss": 1.3216,
+      "theoretical_loss": 3.4217368720959884,
+      "tokens_seen": 2282684416
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031134649333975284,
+      "loss": 1.3935,
+      "theoretical_loss": 3.42172104628471,
+      "tokens_seen": 2282815488
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003113063713689617,
+      "loss": 1.3785,
+      "theoretical_loss": 3.421705221636484,
+      "tokens_seen": 2282946560
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031126624939817045,
+      "loss": 1.3246,
+      "theoretical_loss": 3.421689398151157,
+      "tokens_seen": 2283077632
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003112261274273792,
+      "loss": 1.3058,
+      "theoretical_loss": 3.4216735758285783,
+      "tokens_seen": 2283208704
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031118600545658806,
+      "loss": 1.274,
+      "theoretical_loss": 3.4216577546685945,
+      "tokens_seen": 2283339776
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003111458834857968,
+      "loss": 1.3976,
+      "theoretical_loss": 3.421641934671054,
+      "tokens_seen": 2283470848
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003111057615150056,
+      "loss": 1.4235,
+      "theoretical_loss": 3.4216261158358043,
+      "tokens_seen": 2283601920
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003110656395442144,
+      "loss": 1.3454,
+      "theoretical_loss": 3.421610298162694,
+      "tokens_seen": 2283732992
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031102551757342323,
+      "loss": 1.3482,
+      "theoretical_loss": 3.42159448165157,
+      "tokens_seen": 2283864064
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.000310985395602632,
+      "loss": 1.397,
+      "theoretical_loss": 3.421578666302281,
+      "tokens_seen": 2283995136
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003109452736318408,
+      "loss": 1.3209,
+      "theoretical_loss": 3.4215628521146746,
+      "tokens_seen": 2284126208
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003109051516610496,
+      "loss": 1.3022,
+      "theoretical_loss": 3.4215470390885994,
+      "tokens_seen": 2284257280
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031086502969025835,
+      "loss": 1.3345,
+      "theoretical_loss": 3.4215312272239027,
+      "tokens_seen": 2284388352
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003108249077194672,
+      "loss": 1.3314,
+      "theoretical_loss": 3.4215154165204336,
+      "tokens_seen": 2284519424
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00031078478574867596,
+      "loss": 1.3317,
+      "theoretical_loss": 3.421499606978039,
+      "tokens_seen": 2284650496
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003107446637778848,
+      "loss": 1.3269,
+      "theoretical_loss": 3.4214837985965674,
+      "tokens_seen": 2284781568
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003107045418070936,
+      "loss": 1.3922,
+      "theoretical_loss": 3.4214679913758674,
+      "tokens_seen": 2284912640
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003106644198363024,
+      "loss": 1.299,
+      "theoretical_loss": 3.4214521853157867,
+      "tokens_seen": 2285043712
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003106242978655112,
+      "loss": 1.3912,
+      "theoretical_loss": 3.421436380416174,
+      "tokens_seen": 2285174784
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00031058417589471994,
+      "loss": 1.3895,
+      "theoretical_loss": 3.4214205766768773,
+      "tokens_seen": 2285305856
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00031054405392392874,
+      "loss": 1.385,
+      "theoretical_loss": 3.4214047740977445,
+      "tokens_seen": 2285436928
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00031050393195313755,
+      "loss": 1.2709,
+      "theoretical_loss": 3.4213889726786246,
+      "tokens_seen": 2285568000
+    },
+    {
+      "epoch": 0.39,
+      "objective/train/advantage_avg": 0.4909844696521759,
+      "objective/train/docs_used": 832464,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2576274871826172,
+      "objective/train/original_loss": 1.2576276063919067,
+      "objective/train/theoretical_loss": 3.4213731724193654,
+      "objective/train/tokens_used": 656158176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24156850576400757,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503263473510742,
+      "objective/train/weighted_lm_loss": 1.3209099769592285,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4213731724193654,
+      "tokens_seen": 2285699072
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00031046380998234636,
+      "loss": 1.3092,
+      "theoretical_loss": 3.4213731724193654,
+      "tokens_seen": 2285699072
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003104236880115551,
+      "loss": 1.3474,
+      "theoretical_loss": 3.4213573733198155,
+      "tokens_seen": 2285830144
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00031038356604076397,
+      "loss": 1.3083,
+      "theoretical_loss": 3.4213415753798233,
+      "tokens_seen": 2285961216
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003103434440699727,
+      "loss": 1.3725,
+      "theoretical_loss": 3.421325778599237,
+      "tokens_seen": 2286092288
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003103033220991815,
+      "loss": 1.2633,
+      "theoretical_loss": 3.4213099829779052,
+      "tokens_seen": 2286223360
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00031026320012839033,
+      "loss": 1.3989,
+      "theoretical_loss": 3.4212941885156765,
+      "tokens_seen": 2286354432
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003102230781575991,
+      "loss": 1.337,
+      "theoretical_loss": 3.4212783952123993,
+      "tokens_seen": 2286485504
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003101829561868079,
+      "loss": 1.3469,
+      "theoretical_loss": 3.421262603067922,
+      "tokens_seen": 2286616576
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003101428342160167,
+      "loss": 1.3353,
+      "theoretical_loss": 3.421246812082093,
+      "tokens_seen": 2286747648
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003101027122452255,
+      "loss": 1.4001,
+      "theoretical_loss": 3.4212310222547617,
+      "tokens_seen": 2286878720
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003100625902744343,
+      "loss": 1.4178,
+      "theoretical_loss": 3.4212152335857757,
+      "tokens_seen": 2287009792
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00031002246830364306,
+      "loss": 1.3734,
+      "theoretical_loss": 3.4211994460749846,
+      "tokens_seen": 2287140864
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030998234633285187,
+      "loss": 1.3542,
+      "theoretical_loss": 3.421183659722236,
+      "tokens_seen": 2287271936
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003099422243620607,
+      "loss": 1.3981,
+      "theoretical_loss": 3.4211678745273795,
+      "tokens_seen": 2287403008
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003099021023912695,
+      "loss": 1.2883,
+      "theoretical_loss": 3.4211520904902635,
+      "tokens_seen": 2287534080
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030986198042047824,
+      "loss": 1.3513,
+      "theoretical_loss": 3.4211363076107366,
+      "tokens_seen": 2287665152
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003098218584496871,
+      "loss": 1.3913,
+      "theoretical_loss": 3.421120525888648,
+      "tokens_seen": 2287796224
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030978173647889585,
+      "loss": 1.3942,
+      "theoretical_loss": 3.421104745323846,
+      "tokens_seen": 2287927296
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003097416145081046,
+      "loss": 1.4096,
+      "theoretical_loss": 3.42108896591618,
+      "tokens_seen": 2288058368
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030970149253731346,
+      "loss": 1.3998,
+      "theoretical_loss": 3.4210731876654985,
+      "tokens_seen": 2288189440
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003096613705665222,
+      "loss": 1.2819,
+      "theoretical_loss": 3.4210574105716502,
+      "tokens_seen": 2288320512
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.000309621248595731,
+      "loss": 1.4063,
+      "theoretical_loss": 3.4210416346344847,
+      "tokens_seen": 2288451584
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003095811266249398,
+      "loss": 1.3078,
+      "theoretical_loss": 3.4210258598538505,
+      "tokens_seen": 2288582656
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030954100465414863,
+      "loss": 1.3675,
+      "theoretical_loss": 3.4210100862295962,
+      "tokens_seen": 2288713728
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003095008826833574,
+      "loss": 1.2951,
+      "theoretical_loss": 3.4209943137615717,
+      "tokens_seen": 2288844800
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003094607607125662,
+      "loss": 1.3366,
+      "theoretical_loss": 3.4209785424496255,
+      "tokens_seen": 2288975872
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.000309420638741775,
+      "loss": 1.386,
+      "theoretical_loss": 3.420962772293607,
+      "tokens_seen": 2289106944
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003093805167709838,
+      "loss": 1.3014,
+      "theoretical_loss": 3.4209470032933647,
+      "tokens_seen": 2289238016
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003093403948001926,
+      "loss": 1.354,
+      "theoretical_loss": 3.4209312354487484,
+      "tokens_seen": 2289369088
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030930027282940136,
+      "loss": 1.4067,
+      "theoretical_loss": 3.4209154687596066,
+      "tokens_seen": 2289500160
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003092601508586102,
+      "loss": 1.3441,
+      "theoretical_loss": 3.420899703225789,
+      "tokens_seen": 2289631232
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.000309220028887819,
+      "loss": 1.3762,
+      "theoretical_loss": 3.420883938847145,
+      "tokens_seen": 2289762304
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003091799069170278,
+      "loss": 1.2942,
+      "theoretical_loss": 3.4208681756235233,
+      "tokens_seen": 2289893376
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003091397849462366,
+      "loss": 1.411,
+      "theoretical_loss": 3.420852413554773,
+      "tokens_seen": 2290024448
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030909966297544534,
+      "loss": 1.3161,
+      "theoretical_loss": 3.420836652640744,
+      "tokens_seen": 2290155520
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030905954100465414,
+      "loss": 1.3467,
+      "theoretical_loss": 3.4208208928812853,
+      "tokens_seen": 2290286592
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030901941903386295,
+      "loss": 1.2912,
+      "theoretical_loss": 3.4208051342762467,
+      "tokens_seen": 2290417664
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030897929706307176,
+      "loss": 1.3227,
+      "theoretical_loss": 3.4207893768254767,
+      "tokens_seen": 2290548736
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003089391750922805,
+      "loss": 1.3375,
+      "theoretical_loss": 3.4207736205288257,
+      "tokens_seen": 2290679808
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030889905312148937,
+      "loss": 1.3238,
+      "theoretical_loss": 3.4207578653861423,
+      "tokens_seen": 2290810880
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003088589311506981,
+      "loss": 1.3778,
+      "theoretical_loss": 3.420742111397277,
+      "tokens_seen": 2290941952
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030881880917990693,
+      "loss": 1.3507,
+      "theoretical_loss": 3.420726358562078,
+      "tokens_seen": 2291073024
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030877868720911573,
+      "loss": 1.3418,
+      "theoretical_loss": 3.4207106068803954,
+      "tokens_seen": 2291204096
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003087385652383245,
+      "loss": 1.3457,
+      "theoretical_loss": 3.420694856352079,
+      "tokens_seen": 2291335168
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030869844326753335,
+      "loss": 1.3814,
+      "theoretical_loss": 3.420679106976978,
+      "tokens_seen": 2291466240
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003086583212967421,
+      "loss": 1.3154,
+      "theoretical_loss": 3.4206633587549424,
+      "tokens_seen": 2291597312
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003086181993259509,
+      "loss": 1.3597,
+      "theoretical_loss": 3.4206476116858218,
+      "tokens_seen": 2291728384
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003085780773551597,
+      "loss": 1.3718,
+      "theoretical_loss": 3.4206318657694657,
+      "tokens_seen": 2291859456
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030853795538436846,
+      "loss": 1.3637,
+      "theoretical_loss": 3.4206161210057235,
+      "tokens_seen": 2291990528
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030849783341357727,
+      "loss": 1.3332,
+      "theoretical_loss": 3.420600377394445,
+      "tokens_seen": 2292121600
+    },
+    {
+      "epoch": 0.39,
+      "objective/train/advantage_avg": 0.4938550293445587,
+      "objective/train/docs_used": 834793,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2289206981658936,
+      "objective/train/original_loss": 1.2289206981658936,
+      "objective/train/theoretical_loss": 3.4205846349354805,
+      "objective/train/tokens_used": 662711776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24419793486595154,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506269931793213,
+      "objective/train/weighted_lm_loss": 1.2910094261169434,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4205846349354805,
+      "tokens_seen": 2292252672
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003084577114427861,
+      "loss": 1.2655,
+      "theoretical_loss": 3.4205846349354805,
+      "tokens_seen": 2292252672
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003084175894719949,
+      "loss": 1.2741,
+      "theoretical_loss": 3.4205688936286793,
+      "tokens_seen": 2292383744
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030837746750120363,
+      "loss": 1.3179,
+      "theoretical_loss": 3.4205531534738913,
+      "tokens_seen": 2292514816
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003083373455304125,
+      "loss": 1.3607,
+      "theoretical_loss": 3.4205374144709664,
+      "tokens_seen": 2292645888
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030829722355962125,
+      "loss": 1.4118,
+      "theoretical_loss": 3.420521676619755,
+      "tokens_seen": 2292776960
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030825710158883,
+      "loss": 1.3942,
+      "theoretical_loss": 3.4205059399201057,
+      "tokens_seen": 2292908032
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030821697961803886,
+      "loss": 1.2818,
+      "theoretical_loss": 3.420490204371869,
+      "tokens_seen": 2293039104
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003081768576472476,
+      "loss": 1.2528,
+      "theoretical_loss": 3.420474469974895,
+      "tokens_seen": 2293170176
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030813673567645647,
+      "loss": 1.3827,
+      "theoretical_loss": 3.420458736729034,
+      "tokens_seen": 2293301248
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003080966137056652,
+      "loss": 1.3147,
+      "theoretical_loss": 3.420443004634136,
+      "tokens_seen": 2293432320
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030805649173487403,
+      "loss": 1.4262,
+      "theoretical_loss": 3.4204272736900503,
+      "tokens_seen": 2293563392
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030801636976408284,
+      "loss": 1.3028,
+      "theoretical_loss": 3.420411543896627,
+      "tokens_seen": 2293694464
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003079762477932916,
+      "loss": 1.2548,
+      "theoretical_loss": 3.420395815253717,
+      "tokens_seen": 2293825536
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003079361258225004,
+      "loss": 1.4063,
+      "theoretical_loss": 3.4203800877611696,
+      "tokens_seen": 2293956608
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003078960038517092,
+      "loss": 1.396,
+      "theoretical_loss": 3.4203643614188355,
+      "tokens_seen": 2294087680
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.000307855881880918,
+      "loss": 1.3759,
+      "theoretical_loss": 3.420348636226565,
+      "tokens_seen": 2294218752
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030781575991012676,
+      "loss": 1.3182,
+      "theoretical_loss": 3.4203329121842074,
+      "tokens_seen": 2294349824
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003077756379393356,
+      "loss": 1.3377,
+      "theoretical_loss": 3.4203171892916138,
+      "tokens_seen": 2294480896
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003077355159685444,
+      "loss": 1.4037,
+      "theoretical_loss": 3.420301467548634,
+      "tokens_seen": 2294611968
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003076953939977532,
+      "loss": 1.2765,
+      "theoretical_loss": 3.4202857469551184,
+      "tokens_seen": 2294743040
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.000307655272026962,
+      "loss": 1.4145,
+      "theoretical_loss": 3.4202700275109175,
+      "tokens_seen": 2294874112
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030761515005617074,
+      "loss": 1.3763,
+      "theoretical_loss": 3.4202543092158812,
+      "tokens_seen": 2295005184
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003075750280853796,
+      "loss": 1.3309,
+      "theoretical_loss": 3.42023859206986,
+      "tokens_seen": 2295136256
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030753490611458835,
+      "loss": 1.2661,
+      "theoretical_loss": 3.420222876072705,
+      "tokens_seen": 2295267328
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030749478414379716,
+      "loss": 1.3489,
+      "theoretical_loss": 3.420207161224266,
+      "tokens_seen": 2295398400
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030745466217300596,
+      "loss": 1.2562,
+      "theoretical_loss": 3.420191447524393,
+      "tokens_seen": 2295529472
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030741454020221477,
+      "loss": 1.3802,
+      "theoretical_loss": 3.420175734972937,
+      "tokens_seen": 2295660544
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003073744182314235,
+      "loss": 1.2953,
+      "theoretical_loss": 3.420160023569749,
+      "tokens_seen": 2295791616
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030733429626063233,
+      "loss": 1.4242,
+      "theoretical_loss": 3.4201443133146787,
+      "tokens_seen": 2295922688
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030729417428984113,
+      "loss": 1.4096,
+      "theoretical_loss": 3.4201286042075774,
+      "tokens_seen": 2296053760
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003072540523190499,
+      "loss": 1.3454,
+      "theoretical_loss": 3.4201128962482947,
+      "tokens_seen": 2296184832
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030721393034825875,
+      "loss": 1.4033,
+      "theoretical_loss": 3.420097189436682,
+      "tokens_seen": 2296315904
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003071738083774675,
+      "loss": 1.3417,
+      "theoretical_loss": 3.4200814837725897,
+      "tokens_seen": 2296446976
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003071336864066763,
+      "loss": 1.4781,
+      "theoretical_loss": 3.420065779255869,
+      "tokens_seen": 2296578048
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003070935644358851,
+      "loss": 1.3585,
+      "theoretical_loss": 3.4200500758863694,
+      "tokens_seen": 2296709120
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030705344246509386,
+      "loss": 1.3286,
+      "theoretical_loss": 3.4200343736639427,
+      "tokens_seen": 2296840192
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030701332049430267,
+      "loss": 1.3367,
+      "theoretical_loss": 3.4200186725884394,
+      "tokens_seen": 2296971264
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003069731985235115,
+      "loss": 1.3743,
+      "theoretical_loss": 3.42000297265971,
+      "tokens_seen": 2297102336
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003069330765527203,
+      "loss": 1.3145,
+      "theoretical_loss": 3.419987273877606,
+      "tokens_seen": 2297233408
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003068929545819291,
+      "loss": 1.3045,
+      "theoretical_loss": 3.419971576241977,
+      "tokens_seen": 2297364480
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003068528326111379,
+      "loss": 1.3442,
+      "theoretical_loss": 3.419955879752675,
+      "tokens_seen": 2297495552
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030681271064034665,
+      "loss": 1.3612,
+      "theoretical_loss": 3.4199401844095507,
+      "tokens_seen": 2297626624
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030677258866955545,
+      "loss": 1.2995,
+      "theoretical_loss": 3.4199244902124546,
+      "tokens_seen": 2297757696
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030673246669876426,
+      "loss": 1.3122,
+      "theoretical_loss": 3.419908797161238,
+      "tokens_seen": 2297888768
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.000306692344727973,
+      "loss": 1.3697,
+      "theoretical_loss": 3.4198931052557517,
+      "tokens_seen": 2298019840
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030665222275718187,
+      "loss": 1.4036,
+      "theoretical_loss": 3.419877414495847,
+      "tokens_seen": 2298150912
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003066121007863906,
+      "loss": 1.3628,
+      "theoretical_loss": 3.4198617248813745,
+      "tokens_seen": 2298281984
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030657197881559943,
+      "loss": 1.3354,
+      "theoretical_loss": 3.4198460364121854,
+      "tokens_seen": 2298413056
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030653185684480824,
+      "loss": 1.3162,
+      "theoretical_loss": 3.419830349088131,
+      "tokens_seen": 2298544128
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.000306491734874017,
+      "loss": 1.4156,
+      "theoretical_loss": 3.4198146629090624,
+      "tokens_seen": 2298675200
+    },
+    {
+      "epoch": 0.39,
+      "objective/train/advantage_avg": 0.49411168694496155,
+      "objective/train/docs_used": 837138,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.1697930097579956,
+      "objective/train/original_loss": 1.1697931289672852,
+      "objective/train/theoretical_loss": 3.419798977874831,
+      "objective/train/tokens_used": 669265376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24450013041496277,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506541728973389,
+      "objective/train/weighted_lm_loss": 1.2290570735931396,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9048373699188232,
+      "theoretical_loss": 3.419798977874831,
+      "tokens_seen": 2298806272
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003064516129032258,
+      "loss": 1.2413,
+      "theoretical_loss": 3.419798977874831,
+      "tokens_seen": 2298806272
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003064114909324346,
+      "loss": 1.4575,
+      "theoretical_loss": 3.4197832939852875,
+      "tokens_seen": 2298937344
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003063713689616434,
+      "loss": 1.2729,
+      "theoretical_loss": 3.4197676112402826,
+      "tokens_seen": 2299068416
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030633124699085216,
+      "loss": 1.3516,
+      "theoretical_loss": 3.419751929639669,
+      "tokens_seen": 2299199488
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.000306291125020061,
+      "loss": 1.3275,
+      "theoretical_loss": 3.4197362491832966,
+      "tokens_seen": 2299330560
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030625100304926977,
+      "loss": 1.3405,
+      "theoretical_loss": 3.419720569871018,
+      "tokens_seen": 2299461632
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003062108810784786,
+      "loss": 1.344,
+      "theoretical_loss": 3.4197048917026827,
+      "tokens_seen": 2299592704
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003061707591076874,
+      "loss": 1.3025,
+      "theoretical_loss": 3.4196892146781437,
+      "tokens_seen": 2299723776
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030613063713689614,
+      "loss": 1.3204,
+      "theoretical_loss": 3.4196735387972517,
+      "tokens_seen": 2299854848
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.000306090515166105,
+      "loss": 1.3766,
+      "theoretical_loss": 3.419657864059858,
+      "tokens_seen": 2299985920
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030605039319531375,
+      "loss": 1.4295,
+      "theoretical_loss": 3.4196421904658143,
+      "tokens_seen": 2300116992
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030601027122452256,
+      "loss": 1.4074,
+      "theoretical_loss": 3.4196265180149714,
+      "tokens_seen": 2300248064
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030597014925373136,
+      "loss": 1.3568,
+      "theoretical_loss": 3.419610846707182,
+      "tokens_seen": 2300379136
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030593002728294017,
+      "loss": 1.3588,
+      "theoretical_loss": 3.419595176542297,
+      "tokens_seen": 2300510208
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003058899053121489,
+      "loss": 1.3554,
+      "theoretical_loss": 3.4195795075201674,
+      "tokens_seen": 2300641280
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030584978334135773,
+      "loss": 1.3454,
+      "theoretical_loss": 3.4195638396406456,
+      "tokens_seen": 2300772352
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030580966137056653,
+      "loss": 1.4286,
+      "theoretical_loss": 3.419548172903583,
+      "tokens_seen": 2300903424
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003057695393997753,
+      "loss": 1.3621,
+      "theoretical_loss": 3.4195325073088307,
+      "tokens_seen": 2301034496
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030572941742898415,
+      "loss": 1.3912,
+      "theoretical_loss": 3.4195168428562406,
+      "tokens_seen": 2301165568
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003056892954581929,
+      "loss": 1.3455,
+      "theoretical_loss": 3.4195011795456645,
+      "tokens_seen": 2301296640
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030564917348740176,
+      "loss": 1.4542,
+      "theoretical_loss": 3.419485517376954,
+      "tokens_seen": 2301427712
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003056090515166105,
+      "loss": 1.3192,
+      "theoretical_loss": 3.4194698563499615,
+      "tokens_seen": 2301558784
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00030556892954581926,
+      "loss": 1.3238,
+      "theoretical_loss": 3.4194541964645375,
+      "tokens_seen": 2301689856
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003055288075750281,
+      "loss": 1.2737,
+      "theoretical_loss": 3.419438537720535,
+      "tokens_seen": 2301820928
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003054886856042369,
+      "loss": 1.3719,
+      "theoretical_loss": 3.4194228801178053,
+      "tokens_seen": 2301952000
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003054485636334457,
+      "loss": 1.358,
+      "theoretical_loss": 3.4194072236562,
+      "tokens_seen": 2302083072
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003054084416626545,
+      "loss": 1.3712,
+      "theoretical_loss": 3.4193915683355707,
+      "tokens_seen": 2302214144
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003053683196918633,
+      "loss": 1.3742,
+      "theoretical_loss": 3.4193759141557702,
+      "tokens_seen": 2302345216
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030532819772107205,
+      "loss": 1.3604,
+      "theoretical_loss": 3.4193602611166503,
+      "tokens_seen": 2302476288
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030528807575028085,
+      "loss": 1.3166,
+      "theoretical_loss": 3.4193446092180624,
+      "tokens_seen": 2302607360
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030524795377948966,
+      "loss": 1.3121,
+      "theoretical_loss": 3.419328958459859,
+      "tokens_seen": 2302738432
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003052078318086984,
+      "loss": 1.3822,
+      "theoretical_loss": 3.4193133088418914,
+      "tokens_seen": 2302869504
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030516770983790727,
+      "loss": 1.4073,
+      "theoretical_loss": 3.4192976603640126,
+      "tokens_seen": 2303000576
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.000305127587867116,
+      "loss": 1.3612,
+      "theoretical_loss": 3.4192820130260735,
+      "tokens_seen": 2303131648
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030508746589632483,
+      "loss": 1.3918,
+      "theoretical_loss": 3.4192663668279275,
+      "tokens_seen": 2303262720
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030504734392553364,
+      "loss": 1.3033,
+      "theoretical_loss": 3.419250721769426,
+      "tokens_seen": 2303393792
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003050072219547424,
+      "loss": 1.4032,
+      "theoretical_loss": 3.419235077850421,
+      "tokens_seen": 2303524864
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030496709998395125,
+      "loss": 1.269,
+      "theoretical_loss": 3.419219435070765,
+      "tokens_seen": 2303655936
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030492697801316,
+      "loss": 1.3592,
+      "theoretical_loss": 3.4192037934303103,
+      "tokens_seen": 2303787008
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003048868560423688,
+      "loss": 1.3177,
+      "theoretical_loss": 3.4191881529289083,
+      "tokens_seen": 2303918080
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003048467340715776,
+      "loss": 1.4336,
+      "theoretical_loss": 3.419172513566412,
+      "tokens_seen": 2304049152
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003048066121007864,
+      "loss": 1.3377,
+      "theoretical_loss": 3.4191568753426744,
+      "tokens_seen": 2304180224
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030476649012999517,
+      "loss": 1.33,
+      "theoretical_loss": 3.4191412382575463,
+      "tokens_seen": 2304311296
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.000304726368159204,
+      "loss": 1.3631,
+      "theoretical_loss": 3.419125602310881,
+      "tokens_seen": 2304442368
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003046862461884128,
+      "loss": 1.3595,
+      "theoretical_loss": 3.4191099675025303,
+      "tokens_seen": 2304573440
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030464612421762154,
+      "loss": 1.4268,
+      "theoretical_loss": 3.419094333832347,
+      "tokens_seen": 2304704512
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003046060022468304,
+      "loss": 1.3634,
+      "theoretical_loss": 3.4190787013001835,
+      "tokens_seen": 2304835584
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030456588027603915,
+      "loss": 1.2987,
+      "theoretical_loss": 3.419063069905892,
+      "tokens_seen": 2304966656
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030452575830524796,
+      "loss": 1.3601,
+      "theoretical_loss": 3.4190474396493253,
+      "tokens_seen": 2305097728
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030448563633445676,
+      "loss": 1.2903,
+      "theoretical_loss": 3.4190318105303357,
+      "tokens_seen": 2305228800
+    },
+    {
+      "debugging/Compilability": 1.0,
+      "debugging/distinct-1-grams": 0.7431796110913657,
+      "debugging/entropy-1-grams": 5.572979440278162,
+      "debugging/length": 451.2962962962963,
+      "debugging/num_segments": 27,
+      "debugging/raw_token_scores_avg": 0.0062319617718458176,
+      "debugging/raw_token_scores_std": 0.020211543887853622,
+      "debugging/score": 0.010468038290363413,
+      "debugging/score_std": 0.015814533663842854,
+      "epoch": 0.4,
+      "objective/train/advantage_avg": 0.4937704801559448,
+      "objective/train/docs_used": 839437,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.372235894203186,
+      "objective/train/original_loss": 1.372235894203186,
+      "objective/train/theoretical_loss": 3.4190161825487757,
+      "objective/train/tokens_used": 675818976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24421776831150055,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506187677383423,
+      "objective/train/weighted_lm_loss": 1.4415522813796997,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9048373699188232,
+      "theoretical_loss": 3.4190161825487757,
+      "tokens_seen": 2305359872
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030444551436366557,
+      "loss": 1.3832,
+      "theoretical_loss": 3.4190161825487757,
+      "tokens_seen": 2305359872
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003044053923928743,
+      "loss": 1.3909,
+      "theoretical_loss": 3.419000555704498,
+      "tokens_seen": 2305490944
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003043652704220831,
+      "loss": 1.3798,
+      "theoretical_loss": 3.4189849299973547,
+      "tokens_seen": 2305622016
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030432514845129193,
+      "loss": 1.4506,
+      "theoretical_loss": 3.4189693054271997,
+      "tokens_seen": 2305753088
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030428502648050074,
+      "loss": 1.3905,
+      "theoretical_loss": 3.418953681993884,
+      "tokens_seen": 2305884160
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030424490450970955,
+      "loss": 1.2971,
+      "theoretical_loss": 3.418938059697261,
+      "tokens_seen": 2306015232
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003042047825389183,
+      "loss": 1.394,
+      "theoretical_loss": 3.418922438537184,
+      "tokens_seen": 2306146304
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030416466056812716,
+      "loss": 1.2936,
+      "theoretical_loss": 3.418906818513505,
+      "tokens_seen": 2306277376
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003041245385973359,
+      "loss": 1.3839,
+      "theoretical_loss": 3.418891199626077,
+      "tokens_seen": 2306408448
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030408441662654466,
+      "loss": 1.3261,
+      "theoretical_loss": 3.4188755818747527,
+      "tokens_seen": 2306539520
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003040442946557535,
+      "loss": 1.2736,
+      "theoretical_loss": 3.4188599652593847,
+      "tokens_seen": 2306670592
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003040041726849623,
+      "loss": 1.3586,
+      "theoretical_loss": 3.4188443497798264,
+      "tokens_seen": 2306801664
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003039640507141711,
+      "loss": 1.3605,
+      "theoretical_loss": 3.41882873543593,
+      "tokens_seen": 2306932736
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003039239287433799,
+      "loss": 1.4424,
+      "theoretical_loss": 3.4188131222275486,
+      "tokens_seen": 2307063808
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003038838067725887,
+      "loss": 1.4501,
+      "theoretical_loss": 3.418797510154535,
+      "tokens_seen": 2307194880
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030384368480179745,
+      "loss": 1.3773,
+      "theoretical_loss": 3.418781899216743,
+      "tokens_seen": 2307325952
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030380356283100625,
+      "loss": 1.3262,
+      "theoretical_loss": 3.418766289414025,
+      "tokens_seen": 2307457024
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030376344086021506,
+      "loss": 1.3453,
+      "theoretical_loss": 3.4187506807462333,
+      "tokens_seen": 2307588096
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030372331888942387,
+      "loss": 1.4132,
+      "theoretical_loss": 3.418735073213222,
+      "tokens_seen": 2307719168
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030368319691863267,
+      "loss": 1.3347,
+      "theoretical_loss": 3.4187194668148435,
+      "tokens_seen": 2307850240
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003036430749478414,
+      "loss": 1.3585,
+      "theoretical_loss": 3.418703861550951,
+      "tokens_seen": 2307981312
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003036029529770503,
+      "loss": 1.3873,
+      "theoretical_loss": 3.418688257421398,
+      "tokens_seen": 2308112384
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030356283100625904,
+      "loss": 1.3726,
+      "theoretical_loss": 3.418672654426037,
+      "tokens_seen": 2308243456
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003035227090354678,
+      "loss": 1.358,
+      "theoretical_loss": 3.418657052564722,
+      "tokens_seen": 2308374528
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030348258706467665,
+      "loss": 1.4434,
+      "theoretical_loss": 3.418641451837305,
+      "tokens_seen": 2308505600
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003034424650938854,
+      "loss": 1.3791,
+      "theoretical_loss": 3.4186258522436406,
+      "tokens_seen": 2308636672
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003034023431230942,
+      "loss": 1.3252,
+      "theoretical_loss": 3.418610253783581,
+      "tokens_seen": 2308767744
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.000303362221152303,
+      "loss": 1.3489,
+      "theoretical_loss": 3.4185946564569796,
+      "tokens_seen": 2308898816
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003033220991815118,
+      "loss": 1.3697,
+      "theoretical_loss": 3.4185790602636903,
+      "tokens_seen": 2309029888
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030328197721072057,
+      "loss": 1.3104,
+      "theoretical_loss": 3.418563465203566,
+      "tokens_seen": 2309160960
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003032418552399294,
+      "loss": 1.3473,
+      "theoretical_loss": 3.41854787127646,
+      "tokens_seen": 2309292032
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003032017332691382,
+      "loss": 1.3182,
+      "theoretical_loss": 3.4185322784822256,
+      "tokens_seen": 2309423104
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030316161129834694,
+      "loss": 1.3657,
+      "theoretical_loss": 3.418516686820717,
+      "tokens_seen": 2309554176
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003031214893275558,
+      "loss": 1.4383,
+      "theoretical_loss": 3.4185010962917866,
+      "tokens_seen": 2309685248
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030308136735676455,
+      "loss": 1.36,
+      "theoretical_loss": 3.4184855068952884,
+      "tokens_seen": 2309816320
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003030412453859734,
+      "loss": 1.4166,
+      "theoretical_loss": 3.4184699186310756,
+      "tokens_seen": 2309947392
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030300112341518216,
+      "loss": 1.4252,
+      "theoretical_loss": 3.418454331499002,
+      "tokens_seen": 2310078464
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030296100144439097,
+      "loss": 1.3697,
+      "theoretical_loss": 3.4184387454989213,
+      "tokens_seen": 2310209536
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003029208794735998,
+      "loss": 1.3265,
+      "theoretical_loss": 3.4184231606306867,
+      "tokens_seen": 2310340608
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003028807575028085,
+      "loss": 1.3657,
+      "theoretical_loss": 3.4184075768941518,
+      "tokens_seen": 2310471680
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030284063553201733,
+      "loss": 1.3613,
+      "theoretical_loss": 3.4183919942891707,
+      "tokens_seen": 2310602752
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030280051356122614,
+      "loss": 1.3633,
+      "theoretical_loss": 3.4183764128155962,
+      "tokens_seen": 2310733824
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030276039159043495,
+      "loss": 1.3132,
+      "theoretical_loss": 3.418360832473283,
+      "tokens_seen": 2310864896
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003027202696196437,
+      "loss": 1.4044,
+      "theoretical_loss": 3.4183452532620837,
+      "tokens_seen": 2310995968
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030268014764885256,
+      "loss": 1.377,
+      "theoretical_loss": 3.418329675181853,
+      "tokens_seen": 2311127040
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003026400256780613,
+      "loss": 1.3848,
+      "theoretical_loss": 3.4183140982324445,
+      "tokens_seen": 2311258112
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030259990370727006,
+      "loss": 1.3593,
+      "theoretical_loss": 3.4182985224137115,
+      "tokens_seen": 2311389184
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003025597817364789,
+      "loss": 1.3733,
+      "theoretical_loss": 3.4182829477255083,
+      "tokens_seen": 2311520256
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003025196597656877,
+      "loss": 1.3407,
+      "theoretical_loss": 3.4182673741676886,
+      "tokens_seen": 2311651328
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030247953779489654,
+      "loss": 1.3729,
+      "theoretical_loss": 3.418251801740106,
+      "tokens_seen": 2311782400
+    },
+    {
+      "epoch": 0.4,
+      "objective/train/advantage_avg": 0.49308401346206665,
+      "objective/train/docs_used": 841951,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2734869718551636,
+      "objective/train/original_loss": 1.2734869718551636,
+      "objective/train/theoretical_loss": 3.4182362304426146,
+      "objective/train/tokens_used": 682372576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24355031549930573,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050546646118164,
+      "objective/train/weighted_lm_loss": 1.3378567695617676,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4182362304426146,
+      "tokens_seen": 2311913472
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003024394158241053,
+      "loss": 1.2886,
+      "theoretical_loss": 3.4182362304426146,
+      "tokens_seen": 2311913472
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003023992938533141,
+      "loss": 1.328,
+      "theoretical_loss": 3.4182206602750687,
+      "tokens_seen": 2312044544
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003023591718825229,
+      "loss": 1.3819,
+      "theoretical_loss": 3.4182050912373216,
+      "tokens_seen": 2312175616
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030231904991173165,
+      "loss": 1.3782,
+      "theoretical_loss": 3.418189523329228,
+      "tokens_seen": 2312306688
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030227892794094046,
+      "loss": 1.3696,
+      "theoretical_loss": 3.418173956550641,
+      "tokens_seen": 2312437760
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030223880597014926,
+      "loss": 1.2819,
+      "theoretical_loss": 3.4181583909014153,
+      "tokens_seen": 2312568832
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030219868399935807,
+      "loss": 1.3878,
+      "theoretical_loss": 3.4181428263814047,
+      "tokens_seen": 2312699904
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003021585620285668,
+      "loss": 1.2309,
+      "theoretical_loss": 3.4181272629904638,
+      "tokens_seen": 2312830976
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003021184400577757,
+      "loss": 1.3594,
+      "theoretical_loss": 3.4181117007284456,
+      "tokens_seen": 2312962048
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030207831808698444,
+      "loss": 1.3821,
+      "theoretical_loss": 3.418096139595206,
+      "tokens_seen": 2313093120
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003020381961161932,
+      "loss": 1.3867,
+      "theoretical_loss": 3.418080579590597,
+      "tokens_seen": 2313224192
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030199807414540205,
+      "loss": 1.4176,
+      "theoretical_loss": 3.4180650207144745,
+      "tokens_seen": 2313355264
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003019579521746108,
+      "loss": 1.3839,
+      "theoretical_loss": 3.418049462966692,
+      "tokens_seen": 2313486336
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003019178302038196,
+      "loss": 1.3612,
+      "theoretical_loss": 3.418033906347104,
+      "tokens_seen": 2313617408
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003018777082330284,
+      "loss": 1.3695,
+      "theoretical_loss": 3.4180183508555646,
+      "tokens_seen": 2313748480
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003018375862622372,
+      "loss": 1.3589,
+      "theoretical_loss": 3.418002796491928,
+      "tokens_seen": 2313879552
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.000301797464291446,
+      "loss": 1.3609,
+      "theoretical_loss": 3.417987243256049,
+      "tokens_seen": 2314010624
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003017573423206548,
+      "loss": 1.324,
+      "theoretical_loss": 3.4179716911477818,
+      "tokens_seen": 2314141696
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003017172203498636,
+      "loss": 1.3503,
+      "theoretical_loss": 3.4179561401669805,
+      "tokens_seen": 2314272768
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003016770983790724,
+      "loss": 1.4248,
+      "theoretical_loss": 3.417940590313499,
+      "tokens_seen": 2314403840
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003016369764082812,
+      "loss": 1.352,
+      "theoretical_loss": 3.4179250415871936,
+      "tokens_seen": 2314534912
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030159685443748995,
+      "loss": 1.3082,
+      "theoretical_loss": 3.4179094939879167,
+      "tokens_seen": 2314665984
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003015567324666988,
+      "loss": 1.3526,
+      "theoretical_loss": 3.417893947515524,
+      "tokens_seen": 2314797056
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030151661049590756,
+      "loss": 1.4038,
+      "theoretical_loss": 3.41787840216987,
+      "tokens_seen": 2314928128
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003014764885251163,
+      "loss": 1.343,
+      "theoretical_loss": 3.4178628579508086,
+      "tokens_seen": 2315059200
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003014363665543252,
+      "loss": 1.4172,
+      "theoretical_loss": 3.4178473148581947,
+      "tokens_seen": 2315190272
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003013962445835339,
+      "loss": 1.3682,
+      "theoretical_loss": 3.4178317728918826,
+      "tokens_seen": 2315321344
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030135612261274273,
+      "loss": 1.2995,
+      "theoretical_loss": 3.417816232051728,
+      "tokens_seen": 2315452416
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030131600064195154,
+      "loss": 1.3362,
+      "theoretical_loss": 3.4178006923375843,
+      "tokens_seen": 2315583488
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030127587867116035,
+      "loss": 1.4101,
+      "theoretical_loss": 3.417785153749307,
+      "tokens_seen": 2315714560
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003012357567003691,
+      "loss": 1.3793,
+      "theoretical_loss": 3.4177696162867504,
+      "tokens_seen": 2315845632
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030119563472957796,
+      "loss": 1.2859,
+      "theoretical_loss": 3.4177540799497694,
+      "tokens_seen": 2315976704
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003011555127587867,
+      "loss": 1.3496,
+      "theoretical_loss": 3.4177385447382185,
+      "tokens_seen": 2316107776
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003011153907879955,
+      "loss": 1.367,
+      "theoretical_loss": 3.4177230106519527,
+      "tokens_seen": 2316238848
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003010752688172043,
+      "loss": 1.3431,
+      "theoretical_loss": 3.4177074776908274,
+      "tokens_seen": 2316369920
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003010351468464131,
+      "loss": 1.3454,
+      "theoretical_loss": 3.4176919458546964,
+      "tokens_seen": 2316500992
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030099502487562194,
+      "loss": 1.3401,
+      "theoretical_loss": 3.417676415143415,
+      "tokens_seen": 2316632064
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003009549029048307,
+      "loss": 1.3774,
+      "theoretical_loss": 3.4176608855568382,
+      "tokens_seen": 2316763136
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003009147809340395,
+      "loss": 1.3568,
+      "theoretical_loss": 3.4176453570948215,
+      "tokens_seen": 2316894208
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003008746589632483,
+      "loss": 1.3614,
+      "theoretical_loss": 3.4176298297572183,
+      "tokens_seen": 2317025280
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030083453699245705,
+      "loss": 1.3917,
+      "theoretical_loss": 3.417614303543885,
+      "tokens_seen": 2317156352
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030079441502166586,
+      "loss": 1.3514,
+      "theoretical_loss": 3.417598778454676,
+      "tokens_seen": 2317287424
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030075429305087466,
+      "loss": 1.3126,
+      "theoretical_loss": 3.417583254489447,
+      "tokens_seen": 2317418496
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030071417108008347,
+      "loss": 1.3708,
+      "theoretical_loss": 3.4175677316480515,
+      "tokens_seen": 2317549568
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003006740491092922,
+      "loss": 1.3601,
+      "theoretical_loss": 3.4175522099303466,
+      "tokens_seen": 2317680640
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003006339271385011,
+      "loss": 1.3859,
+      "theoretical_loss": 3.4175366893361856,
+      "tokens_seen": 2317811712
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030059380516770984,
+      "loss": 1.4521,
+      "theoretical_loss": 3.417521169865425,
+      "tokens_seen": 2317942784
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003005536831969186,
+      "loss": 1.3069,
+      "theoretical_loss": 3.4175056515179194,
+      "tokens_seen": 2318073856
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00030051356122612745,
+      "loss": 1.3719,
+      "theoretical_loss": 3.4174901342935238,
+      "tokens_seen": 2318204928
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0003004734392553362,
+      "loss": 1.3816,
+      "theoretical_loss": 3.417474618192094,
+      "tokens_seen": 2318336000
+    },
+    {
+      "epoch": 0.41,
+      "objective/train/advantage_avg": 0.4904979169368744,
+      "objective/train/docs_used": 844398,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3182473182678223,
+      "objective/train/original_loss": 1.3182473182678223,
+      "objective/train/theoretical_loss": 3.4174591032134845,
+      "objective/train/tokens_used": 688926176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24121245741844177,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0502760410308838,
+      "objective/train/weighted_lm_loss": 1.3844895362854004,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4174591032134845,
+      "tokens_seen": 2318467072
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00030043331728454506,
+      "loss": 1.3893,
+      "theoretical_loss": 3.4174591032134845,
+      "tokens_seen": 2318467072
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0003003931953137538,
+      "loss": 1.4257,
+      "theoretical_loss": 3.417443589357551,
+      "tokens_seen": 2318598144
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0003003530733429626,
+      "loss": 1.3334,
+      "theoretical_loss": 3.417428076624149,
+      "tokens_seen": 2318729216
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0003003129513721714,
+      "loss": 1.3237,
+      "theoretical_loss": 3.4174125650131337,
+      "tokens_seen": 2318860288
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0003002728294013802,
+      "loss": 1.3653,
+      "theoretical_loss": 3.4173970545243604,
+      "tokens_seen": 2318991360
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.000300232707430589,
+      "loss": 1.3548,
+      "theoretical_loss": 3.417381545157684,
+      "tokens_seen": 2319122432
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0003001925854597978,
+      "loss": 1.3925,
+      "theoretical_loss": 3.417366036912961,
+      "tokens_seen": 2319253504
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0003001524634890066,
+      "loss": 1.297,
+      "theoretical_loss": 3.4173505297900455,
+      "tokens_seen": 2319384576
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00030011234151821535,
+      "loss": 1.3498,
+      "theoretical_loss": 3.417335023788794,
+      "tokens_seen": 2319515648
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0003000722195474242,
+      "loss": 1.3901,
+      "theoretical_loss": 3.417319518909062,
+      "tokens_seen": 2319646720
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00030003209757663296,
+      "loss": 1.3682,
+      "theoretical_loss": 3.4173040151507044,
+      "tokens_seen": 2319777792
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002999919756058417,
+      "loss": 1.4069,
+      "theoretical_loss": 3.417288512513577,
+      "tokens_seen": 2319908864
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002999518536350506,
+      "loss": 1.4254,
+      "theoretical_loss": 3.417273010997535,
+      "tokens_seen": 2320039936
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002999117316642593,
+      "loss": 1.361,
+      "theoretical_loss": 3.417257510602435,
+      "tokens_seen": 2320171008
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002998716096934682,
+      "loss": 1.3946,
+      "theoretical_loss": 3.417242011328132,
+      "tokens_seen": 2320302080
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029983148772267694,
+      "loss": 1.3355,
+      "theoretical_loss": 3.4172265131744815,
+      "tokens_seen": 2320433152
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029979136575188574,
+      "loss": 1.4012,
+      "theoretical_loss": 3.4172110161413394,
+      "tokens_seen": 2320564224
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029975124378109455,
+      "loss": 1.3389,
+      "theoretical_loss": 3.417195520228561,
+      "tokens_seen": 2320695296
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029971112181030336,
+      "loss": 1.3433,
+      "theoretical_loss": 3.4171800254360027,
+      "tokens_seen": 2320826368
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002996709998395121,
+      "loss": 1.2846,
+      "theoretical_loss": 3.41716453176352,
+      "tokens_seen": 2320957440
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002996308778687209,
+      "loss": 1.4277,
+      "theoretical_loss": 3.4171490392109685,
+      "tokens_seen": 2321088512
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002995907558979297,
+      "loss": 1.3522,
+      "theoretical_loss": 3.4171335477782043,
+      "tokens_seen": 2321219584
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002995506339271385,
+      "loss": 1.3787,
+      "theoretical_loss": 3.417118057465083,
+      "tokens_seen": 2321350656
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029951051195634733,
+      "loss": 1.2928,
+      "theoretical_loss": 3.4171025682714604,
+      "tokens_seen": 2321481728
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002994703899855561,
+      "loss": 1.36,
+      "theoretical_loss": 3.417087080197193,
+      "tokens_seen": 2321612800
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002994302680147649,
+      "loss": 1.3479,
+      "theoretical_loss": 3.4170715932421354,
+      "tokens_seen": 2321743872
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002993901460439737,
+      "loss": 1.2757,
+      "theoretical_loss": 3.417056107406145,
+      "tokens_seen": 2321874944
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029935002407318245,
+      "loss": 1.3384,
+      "theoretical_loss": 3.4170406226890773,
+      "tokens_seen": 2322006016
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029930990210239126,
+      "loss": 1.3167,
+      "theoretical_loss": 3.417025139090788,
+      "tokens_seen": 2322137088
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029926978013160006,
+      "loss": 1.3876,
+      "theoretical_loss": 3.417009656611133,
+      "tokens_seen": 2322268160
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029922965816080887,
+      "loss": 1.3704,
+      "theoretical_loss": 3.416994175249969,
+      "tokens_seen": 2322399232
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002991895361900177,
+      "loss": 1.2823,
+      "theoretical_loss": 3.4169786950071517,
+      "tokens_seen": 2322530304
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002991494142192265,
+      "loss": 1.4055,
+      "theoretical_loss": 3.416963215882537,
+      "tokens_seen": 2322661376
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029910929224843524,
+      "loss": 1.3627,
+      "theoretical_loss": 3.4169477378759816,
+      "tokens_seen": 2322792448
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029906917027764404,
+      "loss": 1.5067,
+      "theoretical_loss": 3.416932260987341,
+      "tokens_seen": 2322923520
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029902904830685285,
+      "loss": 1.4324,
+      "theoretical_loss": 3.4169167852164724,
+      "tokens_seen": 2323054592
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002989889263360616,
+      "loss": 1.41,
+      "theoretical_loss": 3.4169013105632304,
+      "tokens_seen": 2323185664
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029894880436527046,
+      "loss": 1.2982,
+      "theoretical_loss": 3.416885837027473,
+      "tokens_seen": 2323316736
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002989086823944792,
+      "loss": 1.3011,
+      "theoretical_loss": 3.4168703646090552,
+      "tokens_seen": 2323447808
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.000298868560423688,
+      "loss": 1.366,
+      "theoretical_loss": 3.4168548933078338,
+      "tokens_seen": 2323578880
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002988284384528968,
+      "loss": 1.314,
+      "theoretical_loss": 3.4168394231236645,
+      "tokens_seen": 2323709952
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002987883164821056,
+      "loss": 1.3518,
+      "theoretical_loss": 3.416823954056405,
+      "tokens_seen": 2323841024
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002987481945113144,
+      "loss": 1.3576,
+      "theoretical_loss": 3.4168084861059107,
+      "tokens_seen": 2323972096
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002987080725405232,
+      "loss": 1.3535,
+      "theoretical_loss": 3.4167930192720375,
+      "tokens_seen": 2324103168
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.000298667950569732,
+      "loss": 1.3811,
+      "theoretical_loss": 3.416777553554643,
+      "tokens_seen": 2324234240
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002986278285989408,
+      "loss": 1.3351,
+      "theoretical_loss": 3.4167620889535835,
+      "tokens_seen": 2324365312
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002985877066281496,
+      "loss": 1.2979,
+      "theoretical_loss": 3.4167466254687144,
+      "tokens_seen": 2324496384
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029854758465735836,
+      "loss": 1.3413,
+      "theoretical_loss": 3.416731163099893,
+      "tokens_seen": 2324627456
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029850746268656717,
+      "loss": 1.3761,
+      "theoretical_loss": 3.416715701846976,
+      "tokens_seen": 2324758528
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.000298467340715776,
+      "loss": 1.3288,
+      "theoretical_loss": 3.41670024170982,
+      "tokens_seen": 2324889600
+    },
+    {
+      "epoch": 0.41,
+      "objective/train/advantage_avg": 0.49318841099739075,
+      "objective/train/docs_used": 846164,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2886375188827515,
+      "objective/train/original_loss": 1.288637399673462,
+      "objective/train/theoretical_loss": 3.4166847826882805,
+      "objective/train/tokens_used": 695479776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24386224150657654,
+      "objective/train/value_max": -0.5,
+      "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": 1.3537626266479492,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4166847826882805,
+      "tokens_seen": 2325020672
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002984272187449847,
+      "loss": 1.3561,
+      "theoretical_loss": 3.4166847826882805,
+      "tokens_seen": 2325020672
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002983870967741936,
+      "loss": 1.3767,
+      "theoretical_loss": 3.4166693247822155,
+      "tokens_seen": 2325151744
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029834697480340234,
+      "loss": 1.32,
+      "theoretical_loss": 3.416653867991481,
+      "tokens_seen": 2325282816
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029830685283261114,
+      "loss": 1.2952,
+      "theoretical_loss": 3.4166384123159332,
+      "tokens_seen": 2325413888
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029826673086181995,
+      "loss": 1.3553,
+      "theoretical_loss": 3.41662295775543,
+      "tokens_seen": 2325544960
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029822660889102876,
+      "loss": 1.3813,
+      "theoretical_loss": 3.416607504309827,
+      "tokens_seen": 2325676032
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002981864869202375,
+      "loss": 1.3135,
+      "theoretical_loss": 3.416592051978981,
+      "tokens_seen": 2325807104
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002981463649494463,
+      "loss": 1.3962,
+      "theoretical_loss": 3.41657660076275,
+      "tokens_seen": 2325938176
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002981062429786551,
+      "loss": 1.3639,
+      "theoretical_loss": 3.416561150660989,
+      "tokens_seen": 2326069248
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002980661210078639,
+      "loss": 1.3766,
+      "theoretical_loss": 3.4165457016735563,
+      "tokens_seen": 2326200320
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029802599903707273,
+      "loss": 1.3133,
+      "theoretical_loss": 3.416530253800308,
+      "tokens_seen": 2326331392
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002979858770662815,
+      "loss": 1.3174,
+      "theoretical_loss": 3.4165148070411013,
+      "tokens_seen": 2326462464
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029794575509549035,
+      "loss": 1.3696,
+      "theoretical_loss": 3.416499361395793,
+      "tokens_seen": 2326593536
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002979056331246991,
+      "loss": 1.3291,
+      "theoretical_loss": 3.4164839168642396,
+      "tokens_seen": 2326724608
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029786551115390785,
+      "loss": 1.2988,
+      "theoretical_loss": 3.4164684734462987,
+      "tokens_seen": 2326855680
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002978253891831167,
+      "loss": 1.3416,
+      "theoretical_loss": 3.4164530311418266,
+      "tokens_seen": 2326986752
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029778526721232546,
+      "loss": 1.3589,
+      "theoretical_loss": 3.4164375899506814,
+      "tokens_seen": 2327117824
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029774514524153427,
+      "loss": 1.3933,
+      "theoretical_loss": 3.4164221498727194,
+      "tokens_seen": 2327248896
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002977050232707431,
+      "loss": 1.3509,
+      "theoretical_loss": 3.4164067109077974,
+      "tokens_seen": 2327379968
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002976649012999519,
+      "loss": 1.2881,
+      "theoretical_loss": 3.4163912730557726,
+      "tokens_seen": 2327511040
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029762477932916063,
+      "loss": 1.3436,
+      "theoretical_loss": 3.4163758363165027,
+      "tokens_seen": 2327642112
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029758465735836944,
+      "loss": 1.3283,
+      "theoretical_loss": 3.4163604006898445,
+      "tokens_seen": 2327773184
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029754453538757825,
+      "loss": 1.3062,
+      "theoretical_loss": 3.416344966175655,
+      "tokens_seen": 2327904256
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.000297504413416787,
+      "loss": 1.3792,
+      "theoretical_loss": 3.4163295327737915,
+      "tokens_seen": 2328035328
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029746429144599586,
+      "loss": 1.3705,
+      "theoretical_loss": 3.416314100484111,
+      "tokens_seen": 2328166400
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002974241694752046,
+      "loss": 1.3075,
+      "theoretical_loss": 3.4162986693064714,
+      "tokens_seen": 2328297472
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029738404750441347,
+      "loss": 1.3116,
+      "theoretical_loss": 3.4162832392407294,
+      "tokens_seen": 2328428544
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002973439255336222,
+      "loss": 1.3597,
+      "theoretical_loss": 3.4162678102867425,
+      "tokens_seen": 2328559616
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.000297303803562831,
+      "loss": 1.3943,
+      "theoretical_loss": 3.416252382444368,
+      "tokens_seen": 2328690688
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029726368159203984,
+      "loss": 1.3449,
+      "theoretical_loss": 3.416236955713463,
+      "tokens_seen": 2328821760
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002972235596212486,
+      "loss": 1.4046,
+      "theoretical_loss": 3.4162215300938854,
+      "tokens_seen": 2328952832
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002971834376504574,
+      "loss": 1.4166,
+      "theoretical_loss": 3.416206105585492,
+      "tokens_seen": 2329083904
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002971433156796662,
+      "loss": 1.4032,
+      "theoretical_loss": 3.416190682188141,
+      "tokens_seen": 2329214976
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.000297103193708875,
+      "loss": 1.3586,
+      "theoretical_loss": 3.416175259901689,
+      "tokens_seen": 2329346048
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029706307173808376,
+      "loss": 1.3219,
+      "theoretical_loss": 3.416159838725994,
+      "tokens_seen": 2329477120
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029702294976729257,
+      "loss": 1.31,
+      "theoretical_loss": 3.416144418660913,
+      "tokens_seen": 2329608192
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002969828277965014,
+      "loss": 1.2763,
+      "theoretical_loss": 3.416128999706305,
+      "tokens_seen": 2329739264
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002969427058257101,
+      "loss": 1.2996,
+      "theoretical_loss": 3.4161135818620254,
+      "tokens_seen": 2329870336
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.000296902583854919,
+      "loss": 1.3287,
+      "theoretical_loss": 3.4160981651279334,
+      "tokens_seen": 2330001408
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029686246188412774,
+      "loss": 1.4377,
+      "theoretical_loss": 3.416082749503886,
+      "tokens_seen": 2330132480
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029682233991333654,
+      "loss": 1.3012,
+      "theoretical_loss": 3.416067334989741,
+      "tokens_seen": 2330263552
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029678221794254535,
+      "loss": 1.3239,
+      "theoretical_loss": 3.416051921585355,
+      "tokens_seen": 2330394624
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002967420959717541,
+      "loss": 1.3866,
+      "theoretical_loss": 3.4160365092905876,
+      "tokens_seen": 2330525696
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029670197400096296,
+      "loss": 1.3155,
+      "theoretical_loss": 3.4160210981052956,
+      "tokens_seen": 2330656768
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002966618520301717,
+      "loss": 1.3836,
+      "theoretical_loss": 3.416005688029337,
+      "tokens_seen": 2330787840
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002966217300593805,
+      "loss": 1.3341,
+      "theoretical_loss": 3.4159902790625685,
+      "tokens_seen": 2330918912
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029658160808858933,
+      "loss": 1.3593,
+      "theoretical_loss": 3.415974871204849,
+      "tokens_seen": 2331049984
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029654148611779813,
+      "loss": 1.2816,
+      "theoretical_loss": 3.4159594644560363,
+      "tokens_seen": 2331181056
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002965013641470069,
+      "loss": 1.3439,
+      "theoretical_loss": 3.4159440588159877,
+      "tokens_seen": 2331312128
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029646124217621575,
+      "loss": 1.3914,
+      "theoretical_loss": 3.4159286542845617,
+      "tokens_seen": 2331443200
+    },
+    {
+      "epoch": 0.41,
+      "objective/train/advantage_avg": 0.4901066720485687,
+      "objective/train/docs_used": 848727,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2700831890106201,
+      "objective/train/original_loss": 1.2700828313827515,
+      "objective/train/theoretical_loss": 3.4159132508616157,
+      "objective/train/tokens_used": 702033376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24101822078227997,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0502359867095947,
+      "objective/train/weighted_lm_loss": 1.3337695598602295,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4159132508616157,
+      "tokens_seen": 2331574272
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002964211202054245,
+      "loss": 1.3134,
+      "theoretical_loss": 3.4159132508616157,
+      "tokens_seen": 2331574272
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029638099823463325,
+      "loss": 1.4194,
+      "theoretical_loss": 3.4158978485470075,
+      "tokens_seen": 2331705344
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002963408762638421,
+      "loss": 1.3341,
+      "theoretical_loss": 3.4158824473405955,
+      "tokens_seen": 2331836416
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029630075429305086,
+      "loss": 1.2887,
+      "theoretical_loss": 3.4158670472422377,
+      "tokens_seen": 2331967488
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029626063232225967,
+      "loss": 1.3447,
+      "theoretical_loss": 3.4158516482517918,
+      "tokens_seen": 2332098560
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002962205103514685,
+      "loss": 1.3455,
+      "theoretical_loss": 3.415836250369116,
+      "tokens_seen": 2332229632
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002961803883806773,
+      "loss": 1.3606,
+      "theoretical_loss": 3.4158208535940684,
+      "tokens_seen": 2332360704
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029614026640988603,
+      "loss": 1.3749,
+      "theoretical_loss": 3.415805457926507,
+      "tokens_seen": 2332491776
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029610014443909484,
+      "loss": 1.2729,
+      "theoretical_loss": 3.41579006336629,
+      "tokens_seen": 2332622848
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029606002246830365,
+      "loss": 1.3863,
+      "theoretical_loss": 3.415774669913275,
+      "tokens_seen": 2332753920
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029601990049751245,
+      "loss": 1.3004,
+      "theoretical_loss": 3.4157592775673207,
+      "tokens_seen": 2332884992
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029597977852672126,
+      "loss": 1.3785,
+      "theoretical_loss": 3.4157438863282854,
+      "tokens_seen": 2333016064
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029593965655593,
+      "loss": 1.3775,
+      "theoretical_loss": 3.4157284961960275,
+      "tokens_seen": 2333147136
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029589953458513887,
+      "loss": 1.3774,
+      "theoretical_loss": 3.4157131071704043,
+      "tokens_seen": 2333278208
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002958594126143476,
+      "loss": 1.3573,
+      "theoretical_loss": 3.4156977192512747,
+      "tokens_seen": 2333409280
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002958192906435564,
+      "loss": 1.4071,
+      "theoretical_loss": 3.4156823324384975,
+      "tokens_seen": 2333540352
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029577916867276524,
+      "loss": 1.379,
+      "theoretical_loss": 3.4156669467319296,
+      "tokens_seen": 2333671424
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.000295739046701974,
+      "loss": 1.3792,
+      "theoretical_loss": 3.4156515621314307,
+      "tokens_seen": 2333802496
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002956989247311828,
+      "loss": 1.3669,
+      "theoretical_loss": 3.4156361786368583,
+      "tokens_seen": 2333933568
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002956588027603916,
+      "loss": 1.3072,
+      "theoretical_loss": 3.4156207962480716,
+      "tokens_seen": 2334064640
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002956186807896004,
+      "loss": 1.3182,
+      "theoretical_loss": 3.4156054149649284,
+      "tokens_seen": 2334195712
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029557855881880916,
+      "loss": 1.2936,
+      "theoretical_loss": 3.4155900347872876,
+      "tokens_seen": 2334326784
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029553843684801797,
+      "loss": 1.4021,
+      "theoretical_loss": 3.415574655715007,
+      "tokens_seen": 2334457856
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029549831487722677,
+      "loss": 1.3638,
+      "theoretical_loss": 3.415559277747946,
+      "tokens_seen": 2334588928
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002954581929064356,
+      "loss": 1.3714,
+      "theoretical_loss": 3.4155439008859623,
+      "tokens_seen": 2334720000
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002954180709356444,
+      "loss": 1.372,
+      "theoretical_loss": 3.4155285251289147,
+      "tokens_seen": 2334851072
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029537794896485314,
+      "loss": 1.385,
+      "theoretical_loss": 3.415513150476662,
+      "tokens_seen": 2334982144
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.000295337826994062,
+      "loss": 1.2338,
+      "theoretical_loss": 3.4154977769290626,
+      "tokens_seen": 2335113216
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029529770502327075,
+      "loss": 1.4547,
+      "theoretical_loss": 3.4154824044859757,
+      "tokens_seen": 2335244288
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002952575830524795,
+      "loss": 1.4275,
+      "theoretical_loss": 3.415467033147259,
+      "tokens_seen": 2335375360
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029521746108168836,
+      "loss": 1.3593,
+      "theoretical_loss": 3.415451662912772,
+      "tokens_seen": 2335506432
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002951773391108971,
+      "loss": 1.3023,
+      "theoretical_loss": 3.4154362937823732,
+      "tokens_seen": 2335637504
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002951372171401059,
+      "loss": 1.3403,
+      "theoretical_loss": 3.415420925755921,
+      "tokens_seen": 2335768576
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029509709516931473,
+      "loss": 1.3905,
+      "theoretical_loss": 3.415405558833274,
+      "tokens_seen": 2335899648
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029505697319852353,
+      "loss": 1.3758,
+      "theoretical_loss": 3.4153901930142916,
+      "tokens_seen": 2336030720
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002950168512277323,
+      "loss": 1.3221,
+      "theoretical_loss": 3.415374828298833,
+      "tokens_seen": 2336161792
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029497672925694115,
+      "loss": 1.3112,
+      "theoretical_loss": 3.4153594646867558,
+      "tokens_seen": 2336292864
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002949366072861499,
+      "loss": 1.2959,
+      "theoretical_loss": 3.4153441021779196,
+      "tokens_seen": 2336423936
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029489648531535865,
+      "loss": 1.4017,
+      "theoretical_loss": 3.415328740772183,
+      "tokens_seen": 2336555008
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002948563633445675,
+      "loss": 1.3685,
+      "theoretical_loss": 3.4153133804694056,
+      "tokens_seen": 2336686080
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029481624137377626,
+      "loss": 1.4217,
+      "theoretical_loss": 3.4152980212694453,
+      "tokens_seen": 2336817152
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002947761194029851,
+      "loss": 1.3188,
+      "theoretical_loss": 3.415282663172162,
+      "tokens_seen": 2336948224
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002947359974321939,
+      "loss": 1.3464,
+      "theoretical_loss": 3.415267306177414,
+      "tokens_seen": 2337079296
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002946958754614027,
+      "loss": 1.3381,
+      "theoretical_loss": 3.415251950285061,
+      "tokens_seen": 2337210368
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002946557534906115,
+      "loss": 1.404,
+      "theoretical_loss": 3.415236595494962,
+      "tokens_seen": 2337341440
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029461563151982024,
+      "loss": 1.4105,
+      "theoretical_loss": 3.415221241806975,
+      "tokens_seen": 2337472512
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029457550954902905,
+      "loss": 1.3462,
+      "theoretical_loss": 3.4152058892209602,
+      "tokens_seen": 2337603584
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029453538757823785,
+      "loss": 1.2863,
+      "theoretical_loss": 3.4151905377367764,
+      "tokens_seen": 2337734656
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029449526560744666,
+      "loss": 1.2933,
+      "theoretical_loss": 3.415175187354283,
+      "tokens_seen": 2337865728
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002944551436366554,
+      "loss": 1.4207,
+      "theoretical_loss": 3.415159838073339,
+      "tokens_seen": 2337996800
+    },
+    {
+      "epoch": 0.42,
+      "objective/train/advantage_avg": 0.49227625131607056,
+      "objective/train/docs_used": 851124,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3456754684448242,
+      "objective/train/original_loss": 1.3456752300262451,
+      "objective/train/theoretical_loss": 3.4151444898938026,
+      "objective/train/tokens_used": 708586976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24319976568222046,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504639148712158,
+      "objective/train/weighted_lm_loss": 1.4136109352111816,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4151444898938026,
+      "tokens_seen": 2338127872
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029441502166586427,
+      "loss": 1.397,
+      "theoretical_loss": 3.4151444898938026,
+      "tokens_seen": 2338127872
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.000294374899695073,
+      "loss": 1.3622,
+      "theoretical_loss": 3.415129142815535,
+      "tokens_seen": 2338258944
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002943347777242818,
+      "loss": 1.335,
+      "theoretical_loss": 3.415113796838394,
+      "tokens_seen": 2338390016
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029429465575349064,
+      "loss": 1.2451,
+      "theoretical_loss": 3.415098451962239,
+      "tokens_seen": 2338521088
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002942545337826994,
+      "loss": 1.4395,
+      "theoretical_loss": 3.41508310818693,
+      "tokens_seen": 2338652160
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029421441181190825,
+      "loss": 1.3812,
+      "theoretical_loss": 3.415067765512326,
+      "tokens_seen": 2338783232
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.000294174289841117,
+      "loss": 1.356,
+      "theoretical_loss": 3.4150524239382865,
+      "tokens_seen": 2338914304
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002941341678703258,
+      "loss": 1.2775,
+      "theoretical_loss": 3.41503708346467,
+      "tokens_seen": 2339045376
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002940940458995346,
+      "loss": 1.2846,
+      "theoretical_loss": 3.4150217440913373,
+      "tokens_seen": 2339176448
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029405392392874337,
+      "loss": 1.3315,
+      "theoretical_loss": 3.4150064058181466,
+      "tokens_seen": 2339307520
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029401380195795217,
+      "loss": 1.3705,
+      "theoretical_loss": 3.414991068644958,
+      "tokens_seen": 2339438592
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.000293973679987161,
+      "loss": 1.4727,
+      "theoretical_loss": 3.414975732571631,
+      "tokens_seen": 2339569664
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002939335580163698,
+      "loss": 1.3473,
+      "theoretical_loss": 3.414960397598025,
+      "tokens_seen": 2339700736
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029389343604557854,
+      "loss": 1.2926,
+      "theoretical_loss": 3.4149450637239998,
+      "tokens_seen": 2339831808
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002938533140747874,
+      "loss": 1.3557,
+      "theoretical_loss": 3.414929730949414,
+      "tokens_seen": 2339962880
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029381319210399615,
+      "loss": 1.3379,
+      "theoretical_loss": 3.4149143992741284,
+      "tokens_seen": 2340093952
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002937730701332049,
+      "loss": 1.3535,
+      "theoretical_loss": 3.414899068698002,
+      "tokens_seen": 2340225024
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029373294816241376,
+      "loss": 1.3879,
+      "theoretical_loss": 3.414883739220895,
+      "tokens_seen": 2340356096
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002936928261916225,
+      "loss": 1.378,
+      "theoretical_loss": 3.414868410842666,
+      "tokens_seen": 2340487168
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002936527042208313,
+      "loss": 1.3694,
+      "theoretical_loss": 3.414853083563176,
+      "tokens_seen": 2340618240
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002936125822500401,
+      "loss": 1.3835,
+      "theoretical_loss": 3.4148377573822835,
+      "tokens_seen": 2340749312
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029357246027924893,
+      "loss": 1.322,
+      "theoretical_loss": 3.414822432299849,
+      "tokens_seen": 2340880384
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029353233830845774,
+      "loss": 1.4032,
+      "theoretical_loss": 3.414807108315732,
+      "tokens_seen": 2341011456
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029349221633766655,
+      "loss": 1.2847,
+      "theoretical_loss": 3.414791785429792,
+      "tokens_seen": 2341142528
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002934520943668753,
+      "loss": 1.3517,
+      "theoretical_loss": 3.41477646364189,
+      "tokens_seen": 2341273600
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002934119723960841,
+      "loss": 1.3532,
+      "theoretical_loss": 3.4147611429518845,
+      "tokens_seen": 2341404672
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002933718504252929,
+      "loss": 1.3391,
+      "theoretical_loss": 3.414745823359636,
+      "tokens_seen": 2341535744
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029333172845450166,
+      "loss": 1.3667,
+      "theoretical_loss": 3.4147305048650045,
+      "tokens_seen": 2341666816
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002932916064837105,
+      "loss": 1.3615,
+      "theoretical_loss": 3.4147151874678494,
+      "tokens_seen": 2341797888
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002932514845129193,
+      "loss": 1.3169,
+      "theoretical_loss": 3.4146998711680308,
+      "tokens_seen": 2341928960
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002932113625421281,
+      "loss": 1.3436,
+      "theoretical_loss": 3.4146845559654095,
+      "tokens_seen": 2342060032
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002931712405713369,
+      "loss": 1.3544,
+      "theoretical_loss": 3.4146692418598446,
+      "tokens_seen": 2342191104
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029313111860054564,
+      "loss": 1.4441,
+      "theoretical_loss": 3.4146539288511963,
+      "tokens_seen": 2342322176
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029309099662975445,
+      "loss": 1.3613,
+      "theoretical_loss": 3.414638616939325,
+      "tokens_seen": 2342453248
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029305087465896325,
+      "loss": 1.3643,
+      "theoretical_loss": 3.4146233061240903,
+      "tokens_seen": 2342584320
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029301075268817206,
+      "loss": 1.3721,
+      "theoretical_loss": 3.414607996405353,
+      "tokens_seen": 2342715392
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002929706307173808,
+      "loss": 1.424,
+      "theoretical_loss": 3.414592687782972,
+      "tokens_seen": 2342846464
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029293050874658967,
+      "loss": 1.3327,
+      "theoretical_loss": 3.414577380256809,
+      "tokens_seen": 2342977536
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002928903867757984,
+      "loss": 1.3258,
+      "theoretical_loss": 3.414562073826723,
+      "tokens_seen": 2343108608
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029285026480500723,
+      "loss": 1.4085,
+      "theoretical_loss": 3.414546768492575,
+      "tokens_seen": 2343239680
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029281014283421604,
+      "loss": 1.3163,
+      "theoretical_loss": 3.4145314642542246,
+      "tokens_seen": 2343370752
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002927700208634248,
+      "loss": 1.3275,
+      "theoretical_loss": 3.4145161611115324,
+      "tokens_seen": 2343501824
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029272989889263365,
+      "loss": 1.444,
+      "theoretical_loss": 3.4145008590643586,
+      "tokens_seen": 2343632896
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002926897769218424,
+      "loss": 1.355,
+      "theoretical_loss": 3.4144855581125633,
+      "tokens_seen": 2343763968
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002926496549510512,
+      "loss": 1.3447,
+      "theoretical_loss": 3.4144702582560074,
+      "tokens_seen": 2343895040
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029260953298026,
+      "loss": 1.3635,
+      "theoretical_loss": 3.414454959494551,
+      "tokens_seen": 2344026112
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029256941100946877,
+      "loss": 1.3627,
+      "theoretical_loss": 3.4144396618280544,
+      "tokens_seen": 2344157184
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029252928903867757,
+      "loss": 1.3601,
+      "theoretical_loss": 3.4144243652563775,
+      "tokens_seen": 2344288256
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002924891670678864,
+      "loss": 1.3259,
+      "theoretical_loss": 3.4144090697793814,
+      "tokens_seen": 2344419328
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002924490450970952,
+      "loss": 1.2973,
+      "theoretical_loss": 3.4143937753969267,
+      "tokens_seen": 2344550400
+    },
+    {
+      "epoch": 0.42,
+      "objective/train/advantage_avg": 0.4915265440940857,
+      "objective/train/docs_used": 853299,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2067394256591797,
+      "objective/train/original_loss": 1.2067394256591797,
+      "objective/train/theoretical_loss": 3.414378482108874,
+      "objective/train/tokens_used": 715140576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24238237738609314,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503848791122437,
+      "objective/train/weighted_lm_loss": 1.2674344778060913,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.414378482108874,
+      "tokens_seen": 2344681472
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029240892312630394,
+      "loss": 1.3381,
+      "theoretical_loss": 3.414378482108874,
+      "tokens_seen": 2344681472
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002923688011555128,
+      "loss": 1.3524,
+      "theoretical_loss": 3.414363189915083,
+      "tokens_seen": 2344812544
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029232867918472155,
+      "loss": 1.3029,
+      "theoretical_loss": 3.4143478988154143,
+      "tokens_seen": 2344943616
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002922885572139303,
+      "loss": 1.4521,
+      "theoretical_loss": 3.4143326088097297,
+      "tokens_seen": 2345074688
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029224843524313916,
+      "loss": 1.3617,
+      "theoretical_loss": 3.4143173198978882,
+      "tokens_seen": 2345205760
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002922083132723479,
+      "loss": 1.2675,
+      "theoretical_loss": 3.4143020320797515,
+      "tokens_seen": 2345336832
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002921681913015568,
+      "loss": 1.4353,
+      "theoretical_loss": 3.41428674535518,
+      "tokens_seen": 2345467904
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002921280693307655,
+      "loss": 1.3591,
+      "theoretical_loss": 3.4142714597240342,
+      "tokens_seen": 2345598976
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029208794735997433,
+      "loss": 1.256,
+      "theoretical_loss": 3.414256175186175,
+      "tokens_seen": 2345730048
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029204782538918314,
+      "loss": 1.3422,
+      "theoretical_loss": 3.414240891741463,
+      "tokens_seen": 2345861120
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002920077034183919,
+      "loss": 1.3439,
+      "theoretical_loss": 3.4142256093897587,
+      "tokens_seen": 2345992192
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002919675814476007,
+      "loss": 1.3612,
+      "theoretical_loss": 3.414210328130923,
+      "tokens_seen": 2346123264
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002919274594768095,
+      "loss": 1.345,
+      "theoretical_loss": 3.4141950479648173,
+      "tokens_seen": 2346254336
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002918873375060183,
+      "loss": 1.3303,
+      "theoretical_loss": 3.4141797688913016,
+      "tokens_seen": 2346385408
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029184721553522706,
+      "loss": 1.3718,
+      "theoretical_loss": 3.4141644909102373,
+      "tokens_seen": 2346516480
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002918070935644359,
+      "loss": 1.3521,
+      "theoretical_loss": 3.414149214021485,
+      "tokens_seen": 2346647552
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002917669715936447,
+      "loss": 1.3821,
+      "theoretical_loss": 3.4141339382249054,
+      "tokens_seen": 2346778624
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002917268496228535,
+      "loss": 1.3292,
+      "theoretical_loss": 3.4141186635203598,
+      "tokens_seen": 2346909696
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002916867276520623,
+      "loss": 1.3549,
+      "theoretical_loss": 3.4141033899077096,
+      "tokens_seen": 2347040768
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029164660568127104,
+      "loss": 1.388,
+      "theoretical_loss": 3.4140881173868145,
+      "tokens_seen": 2347171840
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002916064837104799,
+      "loss": 1.3835,
+      "theoretical_loss": 3.4140728459575365,
+      "tokens_seen": 2347302912
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029156636173968865,
+      "loss": 1.3386,
+      "theoretical_loss": 3.4140575756197364,
+      "tokens_seen": 2347433984
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029152623976889746,
+      "loss": 1.321,
+      "theoretical_loss": 3.414042306373275,
+      "tokens_seen": 2347565056
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029148611779810626,
+      "loss": 1.4088,
+      "theoretical_loss": 3.4140270382180136,
+      "tokens_seen": 2347696128
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029144599582731507,
+      "loss": 1.3558,
+      "theoretical_loss": 3.414011771153813,
+      "tokens_seen": 2347827200
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002914058738565238,
+      "loss": 1.3343,
+      "theoretical_loss": 3.413996505180535,
+      "tokens_seen": 2347958272
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029136575188573263,
+      "loss": 1.3779,
+      "theoretical_loss": 3.4139812402980407,
+      "tokens_seen": 2348089344
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029132562991494144,
+      "loss": 1.2601,
+      "theoretical_loss": 3.4139659765061907,
+      "tokens_seen": 2348220416
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002912855079441502,
+      "loss": 1.2521,
+      "theoretical_loss": 3.413950713804846,
+      "tokens_seen": 2348351488
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029124538597335905,
+      "loss": 1.3443,
+      "theoretical_loss": 3.4139354521938685,
+      "tokens_seen": 2348482560
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002912052640025678,
+      "loss": 1.3592,
+      "theoretical_loss": 3.413920191673119,
+      "tokens_seen": 2348613632
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002911651420317766,
+      "loss": 1.4283,
+      "theoretical_loss": 3.4139049322424597,
+      "tokens_seen": 2348744704
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002911250200609854,
+      "loss": 1.2424,
+      "theoretical_loss": 3.4138896739017506,
+      "tokens_seen": 2348875776
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029108489809019417,
+      "loss": 1.3491,
+      "theoretical_loss": 3.413874416650854,
+      "tokens_seen": 2349006848
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029104477611940297,
+      "loss": 1.3713,
+      "theoretical_loss": 3.413859160489631,
+      "tokens_seen": 2349137920
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002910046541486118,
+      "loss": 1.3813,
+      "theoretical_loss": 3.4138439054179424,
+      "tokens_seen": 2349268992
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002909645321778206,
+      "loss": 1.3468,
+      "theoretical_loss": 3.4138286514356504,
+      "tokens_seen": 2349400064
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002909244102070294,
+      "loss": 1.3704,
+      "theoretical_loss": 3.4138133985426165,
+      "tokens_seen": 2349531136
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002908842882362382,
+      "loss": 1.3605,
+      "theoretical_loss": 3.413798146738701,
+      "tokens_seen": 2349662208
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029084416626544695,
+      "loss": 1.2915,
+      "theoretical_loss": 3.4137828960237666,
+      "tokens_seen": 2349793280
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029080404429465576,
+      "loss": 1.3923,
+      "theoretical_loss": 3.4137676463976745,
+      "tokens_seen": 2349924352
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029076392232386456,
+      "loss": 1.3486,
+      "theoretical_loss": 3.413752397860286,
+      "tokens_seen": 2350055424
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002907238003530733,
+      "loss": 1.3471,
+      "theoretical_loss": 3.4137371504114626,
+      "tokens_seen": 2350186496
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002906836783822822,
+      "loss": 1.3708,
+      "theoretical_loss": 3.4137219040510667,
+      "tokens_seen": 2350317568
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002906435564114909,
+      "loss": 1.3908,
+      "theoretical_loss": 3.4137066587789584,
+      "tokens_seen": 2350448640
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029060343444069973,
+      "loss": 1.3774,
+      "theoretical_loss": 3.4136914145950006,
+      "tokens_seen": 2350579712
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029056331246990854,
+      "loss": 1.4296,
+      "theoretical_loss": 3.4136761714990547,
+      "tokens_seen": 2350710784
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002905231904991173,
+      "loss": 1.3778,
+      "theoretical_loss": 3.413660929490982,
+      "tokens_seen": 2350841856
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002904830685283261,
+      "loss": 1.3876,
+      "theoretical_loss": 3.4136456885706448,
+      "tokens_seen": 2350972928
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002904429465575349,
+      "loss": 1.3405,
+      "theoretical_loss": 3.413630448737904,
+      "tokens_seen": 2351104000
+    },
+    {
+      "epoch": 0.42,
+      "objective/train/advantage_avg": 0.4919123351573944,
+      "objective/train/docs_used": 855824,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3030980825424194,
+      "objective/train/original_loss": 1.3030980825424194,
+      "objective/train/theoretical_loss": 3.4136152099926225,
+      "objective/train/tokens_used": 721694176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2427648901939392,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050425410270691,
+      "objective/train/weighted_lm_loss": 1.3687880039215088,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4136152099926225,
+      "tokens_seen": 2351235072
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002904028245867437,
+      "loss": 1.3582,
+      "theoretical_loss": 3.4136152099926225,
+      "tokens_seen": 2351235072
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002903627026159525,
+      "loss": 1.3869,
+      "theoretical_loss": 3.4135999723346613,
+      "tokens_seen": 2351366144
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002903225806451613,
+      "loss": 1.3871,
+      "theoretical_loss": 3.413584735763882,
+      "tokens_seen": 2351497216
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002902824586743701,
+      "loss": 1.3012,
+      "theoretical_loss": 3.413569500280147,
+      "tokens_seen": 2351628288
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00029024233670357894,
+      "loss": 1.3678,
+      "theoretical_loss": 3.413554265883318,
+      "tokens_seen": 2351759360
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002902022147327877,
+      "loss": 1.2976,
+      "theoretical_loss": 3.413539032573257,
+      "tokens_seen": 2351890432
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00029016209276199644,
+      "loss": 1.4245,
+      "theoretical_loss": 3.4135238003498256,
+      "tokens_seen": 2352021504
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002901219707912053,
+      "loss": 1.3382,
+      "theoretical_loss": 3.413508569212886,
+      "tokens_seen": 2352152576
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00029008184882041405,
+      "loss": 1.3886,
+      "theoretical_loss": 3.4134933391623,
+      "tokens_seen": 2352283648
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00029004172684962286,
+      "loss": 1.4735,
+      "theoretical_loss": 3.41347811019793,
+      "tokens_seen": 2352414720
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00029000160487883166,
+      "loss": 1.3741,
+      "theoretical_loss": 3.4134628823196373,
+      "tokens_seen": 2352545792
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028996148290804047,
+      "loss": 1.3444,
+      "theoretical_loss": 3.4134476555272846,
+      "tokens_seen": 2352676864
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002899213609372492,
+      "loss": 1.3253,
+      "theoretical_loss": 3.4134324298207335,
+      "tokens_seen": 2352807936
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028988123896645803,
+      "loss": 1.335,
+      "theoretical_loss": 3.4134172051998464,
+      "tokens_seen": 2352939008
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028984111699566684,
+      "loss": 1.3539,
+      "theoretical_loss": 3.413401981664485,
+      "tokens_seen": 2353070080
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002898009950248756,
+      "loss": 1.3123,
+      "theoretical_loss": 3.4133867592145117,
+      "tokens_seen": 2353201152
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028976087305408445,
+      "loss": 1.3715,
+      "theoretical_loss": 3.4133715378497893,
+      "tokens_seen": 2353332224
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002897207510832932,
+      "loss": 1.36,
+      "theoretical_loss": 3.4133563175701793,
+      "tokens_seen": 2353463296
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028968062911250206,
+      "loss": 1.3741,
+      "theoretical_loss": 3.4133410983755437,
+      "tokens_seen": 2353594368
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002896405071417108,
+      "loss": 1.3689,
+      "theoretical_loss": 3.413325880265745,
+      "tokens_seen": 2353725440
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028960038517091957,
+      "loss": 1.3773,
+      "theoretical_loss": 3.4133106632406456,
+      "tokens_seen": 2353856512
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002895602632001284,
+      "loss": 1.3227,
+      "theoretical_loss": 3.413295447300108,
+      "tokens_seen": 2353987584
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002895201412293372,
+      "loss": 1.2982,
+      "theoretical_loss": 3.413280232443994,
+      "tokens_seen": 2354118656
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.000289480019258546,
+      "loss": 1.3214,
+      "theoretical_loss": 3.413265018672166,
+      "tokens_seen": 2354249728
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002894398972877548,
+      "loss": 1.3713,
+      "theoretical_loss": 3.413249805984487,
+      "tokens_seen": 2354380800
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002893997753169636,
+      "loss": 1.3554,
+      "theoretical_loss": 3.4132345943808184,
+      "tokens_seen": 2354511872
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028935965334617235,
+      "loss": 1.3683,
+      "theoretical_loss": 3.4132193838610236,
+      "tokens_seen": 2354642944
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028931953137538115,
+      "loss": 1.3461,
+      "theoretical_loss": 3.4132041744249646,
+      "tokens_seen": 2354774016
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028927940940458996,
+      "loss": 1.2857,
+      "theoretical_loss": 3.4131889660725037,
+      "tokens_seen": 2354905088
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002892392874337987,
+      "loss": 1.356,
+      "theoretical_loss": 3.4131737588035036,
+      "tokens_seen": 2355036160
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002891991654630076,
+      "loss": 1.3642,
+      "theoretical_loss": 3.4131585526178267,
+      "tokens_seen": 2355167232
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002891590434922163,
+      "loss": 1.3409,
+      "theoretical_loss": 3.413143347515336,
+      "tokens_seen": 2355298304
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002891189215214252,
+      "loss": 1.3539,
+      "theoretical_loss": 3.4131281434958933,
+      "tokens_seen": 2355429376
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028907879955063394,
+      "loss": 1.3587,
+      "theoretical_loss": 3.4131129405593614,
+      "tokens_seen": 2355560448
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002890386775798427,
+      "loss": 1.3453,
+      "theoretical_loss": 3.4130977387056034,
+      "tokens_seen": 2355691520
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028899855560905155,
+      "loss": 1.3142,
+      "theoretical_loss": 3.4130825379344816,
+      "tokens_seen": 2355822592
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002889584336382603,
+      "loss": 1.371,
+      "theoretical_loss": 3.4130673382458587,
+      "tokens_seen": 2355953664
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002889183116674691,
+      "loss": 1.3411,
+      "theoretical_loss": 3.4130521396395976,
+      "tokens_seen": 2356084736
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002888781896966779,
+      "loss": 1.3568,
+      "theoretical_loss": 3.4130369421155606,
+      "tokens_seen": 2356215808
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002888380677258867,
+      "loss": 1.3256,
+      "theoretical_loss": 3.4130217456736105,
+      "tokens_seen": 2356346880
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002887979457550955,
+      "loss": 1.3524,
+      "theoretical_loss": 3.413006550313611,
+      "tokens_seen": 2356477952
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028875782378430433,
+      "loss": 1.3115,
+      "theoretical_loss": 3.4129913560354233,
+      "tokens_seen": 2356609024
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002887177018135131,
+      "loss": 1.3061,
+      "theoretical_loss": 3.4129761628389113,
+      "tokens_seen": 2356740096
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028867757984272184,
+      "loss": 1.3949,
+      "theoretical_loss": 3.4129609707239377,
+      "tokens_seen": 2356871168
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002886374578719307,
+      "loss": 1.3222,
+      "theoretical_loss": 3.4129457796903653,
+      "tokens_seen": 2357002240
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028859733590113945,
+      "loss": 1.339,
+      "theoretical_loss": 3.412930589738057,
+      "tokens_seen": 2357133312
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028855721393034826,
+      "loss": 1.32,
+      "theoretical_loss": 3.4129154008668756,
+      "tokens_seen": 2357264384
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028851709195955706,
+      "loss": 1.4107,
+      "theoretical_loss": 3.412900213076684,
+      "tokens_seen": 2357395456
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028847696998876587,
+      "loss": 1.3904,
+      "theoretical_loss": 3.4128850263673454,
+      "tokens_seen": 2357526528
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002884368480179747,
+      "loss": 1.3538,
+      "theoretical_loss": 3.412869840738723,
+      "tokens_seen": 2357657600
+    },
+    {
+      "epoch": 0.43,
+      "objective/train/advantage_avg": 0.4918232560157776,
+      "objective/train/docs_used": 858137,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.209593415260315,
+      "objective/train/original_loss": 1.2095931768417358,
+      "objective/train/theoretical_loss": 3.412854656190679,
+      "objective/train/tokens_used": 728247776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24252358078956604,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050415277481079,
+      "objective/train/weighted_lm_loss": 1.2705448865890503,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9048373699188232,
+      "theoretical_loss": 3.412854656190679,
+      "tokens_seen": 2357788672
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028839672604718343,
+      "loss": 1.2867,
+      "theoretical_loss": 3.412854656190679,
+      "tokens_seen": 2357788672
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028835660407639224,
+      "loss": 1.309,
+      "theoretical_loss": 3.4128394727230766,
+      "tokens_seen": 2357919744
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028831648210560104,
+      "loss": 1.3671,
+      "theoretical_loss": 3.41282429033578,
+      "tokens_seen": 2358050816
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028827636013480985,
+      "loss": 1.3528,
+      "theoretical_loss": 3.412809109028651,
+      "tokens_seen": 2358181888
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002882362381640186,
+      "loss": 1.3794,
+      "theoretical_loss": 3.4127939288015536,
+      "tokens_seen": 2358312960
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028819611619322746,
+      "loss": 1.3441,
+      "theoretical_loss": 3.4127787496543505,
+      "tokens_seen": 2358444032
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002881559942224362,
+      "loss": 1.3197,
+      "theoretical_loss": 3.412763571586905,
+      "tokens_seen": 2358575104
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028811587225164496,
+      "loss": 1.3011,
+      "theoretical_loss": 3.41274839459908,
+      "tokens_seen": 2358706176
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002880757502808538,
+      "loss": 1.4215,
+      "theoretical_loss": 3.412733218690739,
+      "tokens_seen": 2358837248
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002880356283100626,
+      "loss": 1.3566,
+      "theoretical_loss": 3.4127180438617457,
+      "tokens_seen": 2358968320
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002879955063392714,
+      "loss": 1.3912,
+      "theoretical_loss": 3.4127028701119624,
+      "tokens_seen": 2359099392
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002879553843684802,
+      "loss": 1.3493,
+      "theoretical_loss": 3.4126876974412528,
+      "tokens_seen": 2359230464
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.000287915262397689,
+      "loss": 1.3086,
+      "theoretical_loss": 3.4126725258494806,
+      "tokens_seen": 2359361536
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028787514042689775,
+      "loss": 1.3224,
+      "theoretical_loss": 3.4126573553365085,
+      "tokens_seen": 2359492608
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028783501845610655,
+      "loss": 1.3627,
+      "theoretical_loss": 3.4126421859022003,
+      "tokens_seen": 2359623680
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028779489648531536,
+      "loss": 1.3242,
+      "theoretical_loss": 3.412627017546419,
+      "tokens_seen": 2359754752
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028775477451452417,
+      "loss": 1.41,
+      "theoretical_loss": 3.412611850269029,
+      "tokens_seen": 2359885824
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.000287714652543733,
+      "loss": 1.333,
+      "theoretical_loss": 3.412596684069893,
+      "tokens_seen": 2360016896
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002876745305729417,
+      "loss": 1.3256,
+      "theoretical_loss": 3.4125815189488735,
+      "tokens_seen": 2360147968
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002876344086021506,
+      "loss": 1.3503,
+      "theoretical_loss": 3.4125663549058363,
+      "tokens_seen": 2360279040
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028759428663135934,
+      "loss": 1.2811,
+      "theoretical_loss": 3.4125511919406426,
+      "tokens_seen": 2360410112
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002875541646605681,
+      "loss": 1.3404,
+      "theoretical_loss": 3.4125360300531575,
+      "tokens_seen": 2360541184
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028751404268977695,
+      "loss": 1.3345,
+      "theoretical_loss": 3.412520869243244,
+      "tokens_seen": 2360672256
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002874739207189857,
+      "loss": 1.3225,
+      "theoretical_loss": 3.4125057095107656,
+      "tokens_seen": 2360803328
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002874337987481945,
+      "loss": 1.3954,
+      "theoretical_loss": 3.412490550855586,
+      "tokens_seen": 2360934400
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002873936767774033,
+      "loss": 1.3447,
+      "theoretical_loss": 3.412475393277569,
+      "tokens_seen": 2361065472
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002873535548066121,
+      "loss": 1.3826,
+      "theoretical_loss": 3.412460236776578,
+      "tokens_seen": 2361196544
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002873134328358209,
+      "loss": 1.326,
+      "theoretical_loss": 3.412445081352477,
+      "tokens_seen": 2361327616
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002872733108650297,
+      "loss": 1.2758,
+      "theoretical_loss": 3.4124299270051295,
+      "tokens_seen": 2361458688
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002872331888942385,
+      "loss": 1.3634,
+      "theoretical_loss": 3.4124147737343993,
+      "tokens_seen": 2361589760
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002871930669234473,
+      "loss": 1.2575,
+      "theoretical_loss": 3.4123996215401506,
+      "tokens_seen": 2361720832
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002871529449526561,
+      "loss": 1.2997,
+      "theoretical_loss": 3.4123844704222464,
+      "tokens_seen": 2361851904
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028711282298186485,
+      "loss": 1.3721,
+      "theoretical_loss": 3.4123693203805505,
+      "tokens_seen": 2361982976
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002870727010110737,
+      "loss": 1.3682,
+      "theoretical_loss": 3.4123541714149273,
+      "tokens_seen": 2362114048
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028703257904028246,
+      "loss": 1.4006,
+      "theoretical_loss": 3.4123390235252407,
+      "tokens_seen": 2362245120
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028699245706949127,
+      "loss": 1.274,
+      "theoretical_loss": 3.4123238767113544,
+      "tokens_seen": 2362376192
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002869523350987001,
+      "loss": 1.4061,
+      "theoretical_loss": 3.412308730973132,
+      "tokens_seen": 2362507264
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028691221312790883,
+      "loss": 1.3434,
+      "theoretical_loss": 3.412293586310438,
+      "tokens_seen": 2362638336
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028687209115711763,
+      "loss": 1.3139,
+      "theoretical_loss": 3.412278442723136,
+      "tokens_seen": 2362769408
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028683196918632644,
+      "loss": 1.4122,
+      "theoretical_loss": 3.41226330021109,
+      "tokens_seen": 2362900480
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028679184721553525,
+      "loss": 1.3975,
+      "theoretical_loss": 3.412248158774164,
+      "tokens_seen": 2363031552
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.000286751725244744,
+      "loss": 1.3332,
+      "theoretical_loss": 3.4122330184122225,
+      "tokens_seen": 2363162624
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028671160327395286,
+      "loss": 1.3349,
+      "theoretical_loss": 3.4122178791251287,
+      "tokens_seen": 2363293696
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002866714813031616,
+      "loss": 1.358,
+      "theoretical_loss": 3.4122027409127473,
+      "tokens_seen": 2363424768
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028663135933237036,
+      "loss": 1.4155,
+      "theoretical_loss": 3.4121876037749423,
+      "tokens_seen": 2363555840
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002865912373615792,
+      "loss": 1.2928,
+      "theoretical_loss": 3.412172467711578,
+      "tokens_seen": 2363686912
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.000286551115390788,
+      "loss": 1.3848,
+      "theoretical_loss": 3.4121573327225185,
+      "tokens_seen": 2363817984
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028651099341999684,
+      "loss": 1.35,
+      "theoretical_loss": 3.4121421988076275,
+      "tokens_seen": 2363949056
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002864708714492056,
+      "loss": 1.326,
+      "theoretical_loss": 3.4121270659667697,
+      "tokens_seen": 2364080128
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002864307494784144,
+      "loss": 1.312,
+      "theoretical_loss": 3.4121119341998094,
+      "tokens_seen": 2364211200
+    },
+    {
+      "epoch": 0.43,
+      "objective/train/advantage_avg": 0.49361056089401245,
+      "objective/train/docs_used": 860660,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2614537477493286,
+      "objective/train/original_loss": 1.2614537477493286,
+      "objective/train/theoretical_loss": 3.4120968035066106,
+      "objective/train/tokens_used": 734801376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24412928521633148,
+      "objective/train/value_max": -0.5,
+      "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": 1.3253023624420166,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4120968035066106,
+      "tokens_seen": 2364342272
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002863906275076232,
+      "loss": 1.3251,
+      "theoretical_loss": 3.4120968035066106,
+      "tokens_seen": 2364342272
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028635050553683195,
+      "loss": 1.4461,
+      "theoretical_loss": 3.4120816738870374,
+      "tokens_seen": 2364473344
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028631038356604076,
+      "loss": 1.3093,
+      "theoretical_loss": 3.412066545340955,
+      "tokens_seen": 2364604416
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028627026159524957,
+      "loss": 1.3505,
+      "theoretical_loss": 3.4120514178682266,
+      "tokens_seen": 2364735488
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002862301396244584,
+      "loss": 1.3795,
+      "theoretical_loss": 3.4120362914687172,
+      "tokens_seen": 2364866560
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002861900176536671,
+      "loss": 1.3826,
+      "theoretical_loss": 3.412021166142291,
+      "tokens_seen": 2364997632
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.000286149895682876,
+      "loss": 1.3584,
+      "theoretical_loss": 3.4120060418888123,
+      "tokens_seen": 2365128704
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028610977371208474,
+      "loss": 1.3519,
+      "theoretical_loss": 3.411990918708146,
+      "tokens_seen": 2365259776
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002860696517412935,
+      "loss": 1.3336,
+      "theoretical_loss": 3.411975796600156,
+      "tokens_seen": 2365390848
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028602952977050235,
+      "loss": 1.4088,
+      "theoretical_loss": 3.4119606755647074,
+      "tokens_seen": 2365521920
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002859894077997111,
+      "loss": 1.4038,
+      "theoretical_loss": 3.411945555601664,
+      "tokens_seen": 2365652992
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028594928582891996,
+      "loss": 1.2392,
+      "theoretical_loss": 3.411930436710891,
+      "tokens_seen": 2365784064
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002859091638581287,
+      "loss": 1.3588,
+      "theoretical_loss": 3.411915318892252,
+      "tokens_seen": 2365915136
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002858690418873375,
+      "loss": 1.3133,
+      "theoretical_loss": 3.411900202145613,
+      "tokens_seen": 2366046208
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028582891991654633,
+      "loss": 1.3392,
+      "theoretical_loss": 3.4118850864708374,
+      "tokens_seen": 2366177280
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002857887979457551,
+      "loss": 1.3857,
+      "theoretical_loss": 3.41186997186779,
+      "tokens_seen": 2366308352
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002857486759749639,
+      "loss": 1.3845,
+      "theoretical_loss": 3.411854858336336,
+      "tokens_seen": 2366439424
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002857085540041727,
+      "loss": 1.3704,
+      "theoretical_loss": 3.41183974587634,
+      "tokens_seen": 2366570496
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002856684320333815,
+      "loss": 1.3328,
+      "theoretical_loss": 3.411824634487666,
+      "tokens_seen": 2366701568
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028562831006259025,
+      "loss": 1.3409,
+      "theoretical_loss": 3.41180952417018,
+      "tokens_seen": 2366832640
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002855881880917991,
+      "loss": 1.3555,
+      "theoretical_loss": 3.411794414923745,
+      "tokens_seen": 2366963712
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028554806612100786,
+      "loss": 1.3289,
+      "theoretical_loss": 3.4117793067482274,
+      "tokens_seen": 2367094784
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028550794415021667,
+      "loss": 1.3638,
+      "theoretical_loss": 3.4117641996434913,
+      "tokens_seen": 2367225856
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002854678221794255,
+      "loss": 1.3587,
+      "theoretical_loss": 3.4117490936094015,
+      "tokens_seen": 2367356928
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028542770020863423,
+      "loss": 1.2434,
+      "theoretical_loss": 3.411733988645823,
+      "tokens_seen": 2367488000
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00028538757823784303,
+      "loss": 1.3823,
+      "theoretical_loss": 3.41171888475262,
+      "tokens_seen": 2367619072
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028534745626705184,
+      "loss": 1.347,
+      "theoretical_loss": 3.411703781929659,
+      "tokens_seen": 2367750144
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028530733429626065,
+      "loss": 1.3181,
+      "theoretical_loss": 3.411688680176803,
+      "tokens_seen": 2367881216
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028526721232546945,
+      "loss": 1.3814,
+      "theoretical_loss": 3.411673579493918,
+      "tokens_seen": 2368012288
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028522709035467826,
+      "loss": 1.243,
+      "theoretical_loss": 3.4116584798808693,
+      "tokens_seen": 2368143360
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.000285186968383887,
+      "loss": 1.4575,
+      "theoretical_loss": 3.4116433813375213,
+      "tokens_seen": 2368274432
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002851468464130958,
+      "loss": 1.3145,
+      "theoretical_loss": 3.411628283863739,
+      "tokens_seen": 2368405504
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002851067244423046,
+      "loss": 1.3578,
+      "theoretical_loss": 3.411613187459388,
+      "tokens_seen": 2368536576
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002850666024715134,
+      "loss": 1.3579,
+      "theoretical_loss": 3.4115980921243327,
+      "tokens_seen": 2368667648
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028502648050072224,
+      "loss": 1.4006,
+      "theoretical_loss": 3.4115829978584387,
+      "tokens_seen": 2368798720
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.000284986358529931,
+      "loss": 1.4145,
+      "theoretical_loss": 3.411567904661571,
+      "tokens_seen": 2368929792
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002849462365591398,
+      "loss": 1.3407,
+      "theoretical_loss": 3.411552812533594,
+      "tokens_seen": 2369060864
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002849061145883486,
+      "loss": 1.2329,
+      "theoretical_loss": 3.411537721474374,
+      "tokens_seen": 2369191936
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028486599261755735,
+      "loss": 1.3914,
+      "theoretical_loss": 3.4115226314837757,
+      "tokens_seen": 2369323008
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028482587064676616,
+      "loss": 1.3269,
+      "theoretical_loss": 3.4115075425616643,
+      "tokens_seen": 2369454080
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028478574867597497,
+      "loss": 1.3719,
+      "theoretical_loss": 3.411492454707905,
+      "tokens_seen": 2369585152
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028474562670518377,
+      "loss": 1.2862,
+      "theoretical_loss": 3.4114773679223633,
+      "tokens_seen": 2369716224
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002847055047343925,
+      "loss": 1.3594,
+      "theoretical_loss": 3.411462282204904,
+      "tokens_seen": 2369847296
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002846653827636014,
+      "loss": 1.3344,
+      "theoretical_loss": 3.4114471975553933,
+      "tokens_seen": 2369978368
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028462526079281014,
+      "loss": 1.3582,
+      "theoretical_loss": 3.411432113973696,
+      "tokens_seen": 2370109440
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028458513882201894,
+      "loss": 1.381,
+      "theoretical_loss": 3.411417031459677,
+      "tokens_seen": 2370240512
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028454501685122775,
+      "loss": 1.3592,
+      "theoretical_loss": 3.4114019500132025,
+      "tokens_seen": 2370371584
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002845048948804365,
+      "loss": 1.2738,
+      "theoretical_loss": 3.4113868696341374,
+      "tokens_seen": 2370502656
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028446477290964536,
+      "loss": 1.3667,
+      "theoretical_loss": 3.411371790322347,
+      "tokens_seen": 2370633728
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002844246509388541,
+      "loss": 1.3268,
+      "theoretical_loss": 3.4113567120776978,
+      "tokens_seen": 2370764800
+    },
+    {
+      "epoch": 0.44,
+      "objective/train/advantage_avg": 0.4924556612968445,
+      "objective/train/docs_used": 863154,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2023653984069824,
+      "objective/train/original_loss": 1.2023653984069824,
+      "objective/train/theoretical_loss": 3.4113416349000536,
+      "objective/train/tokens_used": 741354976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24306902289390564,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504813194274902,
+      "objective/train/weighted_lm_loss": 1.2630406618118286,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4113416349000536,
+      "tokens_seen": 2370895872
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002843845289680629,
+      "loss": 1.2984,
+      "theoretical_loss": 3.4113416349000536,
+      "tokens_seen": 2370895872
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028434440699727173,
+      "loss": 1.3878,
+      "theoretical_loss": 3.411326558789282,
+      "tokens_seen": 2371026944
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002843042850264805,
+      "loss": 1.4265,
+      "theoretical_loss": 3.4113114837452465,
+      "tokens_seen": 2371158016
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002842641630556893,
+      "loss": 1.4033,
+      "theoretical_loss": 3.411296409767814,
+      "tokens_seen": 2371289088
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002842240410848981,
+      "loss": 1.3807,
+      "theoretical_loss": 3.4112813368568498,
+      "tokens_seen": 2371420160
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002841839191141069,
+      "loss": 1.4216,
+      "theoretical_loss": 3.4112662650122187,
+      "tokens_seen": 2371551232
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028414379714331565,
+      "loss": 1.3972,
+      "theoretical_loss": 3.4112511942337873,
+      "tokens_seen": 2371682304
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002841036751725245,
+      "loss": 1.3648,
+      "theoretical_loss": 3.4112361245214213,
+      "tokens_seen": 2371813376
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028406355320173326,
+      "loss": 1.3607,
+      "theoretical_loss": 3.411221055874986,
+      "tokens_seen": 2371944448
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.000284023431230942,
+      "loss": 1.3358,
+      "theoretical_loss": 3.4112059882943466,
+      "tokens_seen": 2372075520
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002839833092601509,
+      "loss": 1.3176,
+      "theoretical_loss": 3.4111909217793697,
+      "tokens_seen": 2372206592
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028394318728935963,
+      "loss": 1.2778,
+      "theoretical_loss": 3.411175856329921,
+      "tokens_seen": 2372337664
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002839030653185685,
+      "loss": 1.273,
+      "theoretical_loss": 3.4111607919458655,
+      "tokens_seen": 2372468736
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028386294334777724,
+      "loss": 1.3997,
+      "theoretical_loss": 3.4111457286270697,
+      "tokens_seen": 2372599808
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028382282137698605,
+      "loss": 1.3627,
+      "theoretical_loss": 3.4111306663733996,
+      "tokens_seen": 2372730880
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028378269940619485,
+      "loss": 1.4142,
+      "theoretical_loss": 3.4111156051847202,
+      "tokens_seen": 2372861952
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028374257743540366,
+      "loss": 1.2902,
+      "theoretical_loss": 3.411100545060898,
+      "tokens_seen": 2372993024
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002837024554646124,
+      "loss": 1.4456,
+      "theoretical_loss": 3.4110854860017987,
+      "tokens_seen": 2373124096
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002836623334938212,
+      "loss": 1.4608,
+      "theoretical_loss": 3.4110704280072888,
+      "tokens_seen": 2373255168
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028362221152303,
+      "loss": 1.3286,
+      "theoretical_loss": 3.411055371077233,
+      "tokens_seen": 2373386240
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002835820895522388,
+      "loss": 1.3488,
+      "theoretical_loss": 3.4110403152114985,
+      "tokens_seen": 2373517312
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028354196758144764,
+      "loss": 1.3725,
+      "theoretical_loss": 3.411025260409951,
+      "tokens_seen": 2373648384
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002835018456106564,
+      "loss": 1.3008,
+      "theoretical_loss": 3.411010206672456,
+      "tokens_seen": 2373779456
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002834617236398652,
+      "loss": 1.3269,
+      "theoretical_loss": 3.41099515399888,
+      "tokens_seen": 2373910528
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.000283421601669074,
+      "loss": 1.3494,
+      "theoretical_loss": 3.410980102389089,
+      "tokens_seen": 2374041600
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028338147969828275,
+      "loss": 1.2852,
+      "theoretical_loss": 3.410965051842949,
+      "tokens_seen": 2374172672
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002833413577274916,
+      "loss": 1.2806,
+      "theoretical_loss": 3.4109500023603263,
+      "tokens_seen": 2374303744
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028330123575670037,
+      "loss": 1.3591,
+      "theoretical_loss": 3.4109349539410867,
+      "tokens_seen": 2374434816
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028326111378590917,
+      "loss": 1.3795,
+      "theoretical_loss": 3.4109199065850966,
+      "tokens_seen": 2374565888
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.000283220991815118,
+      "loss": 1.3267,
+      "theoretical_loss": 3.4109048602922223,
+      "tokens_seen": 2374696960
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002831808698443268,
+      "loss": 1.3208,
+      "theoretical_loss": 3.41088981506233,
+      "tokens_seen": 2374828032
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028314074787353554,
+      "loss": 1.3406,
+      "theoretical_loss": 3.4108747708952856,
+      "tokens_seen": 2374959104
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028310062590274434,
+      "loss": 1.4505,
+      "theoretical_loss": 3.410859727790956,
+      "tokens_seen": 2375090176
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028306050393195315,
+      "loss": 1.337,
+      "theoretical_loss": 3.4108446857492067,
+      "tokens_seen": 2375221248
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002830203819611619,
+      "loss": 1.3314,
+      "theoretical_loss": 3.4108296447699047,
+      "tokens_seen": 2375352320
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028298025999037076,
+      "loss": 1.3469,
+      "theoretical_loss": 3.4108146048529155,
+      "tokens_seen": 2375483392
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002829401380195795,
+      "loss": 1.32,
+      "theoretical_loss": 3.4107995659981065,
+      "tokens_seen": 2375614464
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002829000160487883,
+      "loss": 1.3913,
+      "theoretical_loss": 3.4107845282053435,
+      "tokens_seen": 2375745536
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028285989407799713,
+      "loss": 1.3445,
+      "theoretical_loss": 3.4107694914744933,
+      "tokens_seen": 2375876608
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002828197721072059,
+      "loss": 1.3806,
+      "theoretical_loss": 3.4107544558054217,
+      "tokens_seen": 2376007680
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002827796501364147,
+      "loss": 1.3521,
+      "theoretical_loss": 3.410739421197995,
+      "tokens_seen": 2376138752
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002827395281656235,
+      "loss": 1.3457,
+      "theoretical_loss": 3.410724387652081,
+      "tokens_seen": 2376269824
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002826994061948323,
+      "loss": 1.269,
+      "theoretical_loss": 3.410709355167545,
+      "tokens_seen": 2376400896
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002826592842240411,
+      "loss": 1.3751,
+      "theoretical_loss": 3.410694323744254,
+      "tokens_seen": 2376531968
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002826191622532499,
+      "loss": 1.3655,
+      "theoretical_loss": 3.4106792933820747,
+      "tokens_seen": 2376663040
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028257904028245866,
+      "loss": 1.2911,
+      "theoretical_loss": 3.4106642640808733,
+      "tokens_seen": 2376794112
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028253891831166747,
+      "loss": 1.3293,
+      "theoretical_loss": 3.4106492358405163,
+      "tokens_seen": 2376925184
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002824987963408763,
+      "loss": 1.3172,
+      "theoretical_loss": 3.4106342086608707,
+      "tokens_seen": 2377056256
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028245867437008503,
+      "loss": 1.3774,
+      "theoretical_loss": 3.4106191825418035,
+      "tokens_seen": 2377187328
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002824185523992939,
+      "loss": 1.3701,
+      "theoretical_loss": 3.4106041574831805,
+      "tokens_seen": 2377318400
+    },
+    {
+      "epoch": 0.44,
+      "objective/train/advantage_avg": 0.4902985692024231,
+      "objective/train/docs_used": 865580,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.330042839050293,
+      "objective/train/original_loss": 1.330042839050293,
+      "objective/train/theoretical_loss": 3.410589133484869,
+      "objective/train/tokens_used": 747908576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24141673743724823,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0502572059631348,
+      "objective/train/weighted_lm_loss": 1.3968614339828491,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.410589133484869,
+      "tokens_seen": 2377449472
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028237843042850264,
+      "loss": 1.3792,
+      "theoretical_loss": 3.410589133484869,
+      "tokens_seen": 2377449472
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028233830845771145,
+      "loss": 1.3428,
+      "theoretical_loss": 3.410574110546735,
+      "tokens_seen": 2377580544
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028229818648692025,
+      "loss": 1.2771,
+      "theoretical_loss": 3.4105590886686463,
+      "tokens_seen": 2377711616
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028225806451612906,
+      "loss": 1.3282,
+      "theoretical_loss": 3.410544067850469,
+      "tokens_seen": 2377842688
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002822179425453378,
+      "loss": 1.3689,
+      "theoretical_loss": 3.4105290480920702,
+      "tokens_seen": 2377973760
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002821778205745466,
+      "loss": 1.3634,
+      "theoretical_loss": 3.4105140293933163,
+      "tokens_seen": 2378104832
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002821376986037554,
+      "loss": 1.3589,
+      "theoretical_loss": 3.4104990117540748,
+      "tokens_seen": 2378235904
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028209757663296423,
+      "loss": 1.3526,
+      "theoretical_loss": 3.4104839951742116,
+      "tokens_seen": 2378366976
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028205745466217304,
+      "loss": 1.3631,
+      "theoretical_loss": 3.410468979653595,
+      "tokens_seen": 2378498048
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002820173326913818,
+      "loss": 1.351,
+      "theoretical_loss": 3.41045396519209,
+      "tokens_seen": 2378629120
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028197721072059065,
+      "loss": 1.3046,
+      "theoretical_loss": 3.4104389517895655,
+      "tokens_seen": 2378760192
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002819370887497994,
+      "loss": 1.3357,
+      "theoretical_loss": 3.4104239394458875,
+      "tokens_seen": 2378891264
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028189696677900815,
+      "loss": 1.4125,
+      "theoretical_loss": 3.410408928160923,
+      "tokens_seen": 2379022336
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.000281856844808217,
+      "loss": 1.4335,
+      "theoretical_loss": 3.410393917934539,
+      "tokens_seen": 2379153408
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028181672283742577,
+      "loss": 1.3294,
+      "theoretical_loss": 3.410378908766603,
+      "tokens_seen": 2379284480
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028177660086663457,
+      "loss": 1.3715,
+      "theoretical_loss": 3.4103639006569813,
+      "tokens_seen": 2379415552
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002817364788958434,
+      "loss": 1.3486,
+      "theoretical_loss": 3.4103488936055415,
+      "tokens_seen": 2379546624
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002816963569250522,
+      "loss": 1.3386,
+      "theoretical_loss": 3.4103338876121505,
+      "tokens_seen": 2379677696
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028165623495426094,
+      "loss": 1.3799,
+      "theoretical_loss": 3.410318882676676,
+      "tokens_seen": 2379808768
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028161611298346974,
+      "loss": 1.3605,
+      "theoretical_loss": 3.410303878798984,
+      "tokens_seen": 2379939840
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028157599101267855,
+      "loss": 1.3367,
+      "theoretical_loss": 3.410288875978943,
+      "tokens_seen": 2380070912
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002815358690418873,
+      "loss": 1.2964,
+      "theoretical_loss": 3.4102738742164194,
+      "tokens_seen": 2380201984
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028149574707109616,
+      "loss": 1.3496,
+      "theoretical_loss": 3.4102588735112804,
+      "tokens_seen": 2380333056
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002814556251003049,
+      "loss": 1.4077,
+      "theoretical_loss": 3.4102438738633936,
+      "tokens_seen": 2380464128
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002814155031295138,
+      "loss": 1.3477,
+      "theoretical_loss": 3.410228875272626,
+      "tokens_seen": 2380595200
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002813753811587225,
+      "loss": 1.3777,
+      "theoretical_loss": 3.410213877738845,
+      "tokens_seen": 2380726272
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002813352591879313,
+      "loss": 1.3874,
+      "theoretical_loss": 3.410198881261918,
+      "tokens_seen": 2380857344
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028129513721714014,
+      "loss": 1.3299,
+      "theoretical_loss": 3.4101838858417124,
+      "tokens_seen": 2380988416
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002812550152463489,
+      "loss": 1.3417,
+      "theoretical_loss": 3.4101688914780954,
+      "tokens_seen": 2381119488
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002812148932755577,
+      "loss": 1.4053,
+      "theoretical_loss": 3.4101538981709347,
+      "tokens_seen": 2381250560
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002811747713047665,
+      "loss": 1.4166,
+      "theoretical_loss": 3.410138905920097,
+      "tokens_seen": 2381381632
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002811346493339753,
+      "loss": 1.3539,
+      "theoretical_loss": 3.4101239147254505,
+      "tokens_seen": 2381512704
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028109452736318406,
+      "loss": 1.3115,
+      "theoretical_loss": 3.4101089245868628,
+      "tokens_seen": 2381643776
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028105440539239287,
+      "loss": 1.4215,
+      "theoretical_loss": 3.4100939355042006,
+      "tokens_seen": 2381774848
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002810142834216017,
+      "loss": 1.3696,
+      "theoretical_loss": 3.4100789474773316,
+      "tokens_seen": 2381905920
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028097416145081043,
+      "loss": 1.3283,
+      "theoretical_loss": 3.410063960506124,
+      "tokens_seen": 2382036992
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002809340394800193,
+      "loss": 1.39,
+      "theoretical_loss": 3.410048974590445,
+      "tokens_seen": 2382168064
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028089391750922804,
+      "loss": 1.4094,
+      "theoretical_loss": 3.4100339897301617,
+      "tokens_seen": 2382299136
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002808537955384369,
+      "loss": 1.3245,
+      "theoretical_loss": 3.4100190059251423,
+      "tokens_seen": 2382430208
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028081367356764565,
+      "loss": 1.4041,
+      "theoretical_loss": 3.4100040231752544,
+      "tokens_seen": 2382561280
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028077355159685446,
+      "loss": 1.3794,
+      "theoretical_loss": 3.4099890414803653,
+      "tokens_seen": 2382692352
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028073342962606327,
+      "loss": 1.3673,
+      "theoretical_loss": 3.409974060840343,
+      "tokens_seen": 2382823424
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.000280693307655272,
+      "loss": 1.4009,
+      "theoretical_loss": 3.409959081255055,
+      "tokens_seen": 2382954496
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002806531856844808,
+      "loss": 1.2989,
+      "theoretical_loss": 3.4099441027243693,
+      "tokens_seen": 2383085568
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028061306371368963,
+      "loss": 1.3769,
+      "theoretical_loss": 3.4099291252481536,
+      "tokens_seen": 2383216640
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028057294174289844,
+      "loss": 1.4325,
+      "theoretical_loss": 3.4099141488262754,
+      "tokens_seen": 2383347712
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002805328197721072,
+      "loss": 1.2815,
+      "theoretical_loss": 3.4098991734586024,
+      "tokens_seen": 2383478784
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028049269780131605,
+      "loss": 1.3301,
+      "theoretical_loss": 3.409884199145003,
+      "tokens_seen": 2383609856
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002804525758305248,
+      "loss": 1.4343,
+      "theoretical_loss": 3.4098692258853447,
+      "tokens_seen": 2383740928
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028041245385973355,
+      "loss": 1.3278,
+      "theoretical_loss": 3.409854253679496,
+      "tokens_seen": 2383872000
+    },
+    {
+      "epoch": 0.44,
+      "objective/train/advantage_avg": 0.4913197159767151,
+      "objective/train/docs_used": 868029,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2209957838058472,
+      "objective/train/original_loss": 1.2209957838058472,
+      "objective/train/theoretical_loss": 3.409839282527323,
+      "objective/train/tokens_used": 754462176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24252137541770935,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503648519515991,
+      "objective/train/weighted_lm_loss": 1.282531499862671,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9048373699188232,
+      "theoretical_loss": 3.409839282527323,
+      "tokens_seen": 2384003072
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002803723318889424,
+      "loss": 1.2974,
+      "theoretical_loss": 3.409839282527323,
+      "tokens_seen": 2384003072
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00028033220991815117,
+      "loss": 1.3126,
+      "theoretical_loss": 3.4098243124286958,
+      "tokens_seen": 2384134144
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00028029208794735997,
+      "loss": 1.3097,
+      "theoretical_loss": 3.4098093433834813,
+      "tokens_seen": 2384265216
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002802519659765688,
+      "loss": 1.3866,
+      "theoretical_loss": 3.4097943753915474,
+      "tokens_seen": 2384396288
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002802118440057776,
+      "loss": 1.2987,
+      "theoretical_loss": 3.4097794084527626,
+      "tokens_seen": 2384527360
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002801717220349864,
+      "loss": 1.3897,
+      "theoretical_loss": 3.409764442566994,
+      "tokens_seen": 2384658432
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00028013160006419514,
+      "loss": 1.3494,
+      "theoretical_loss": 3.4097494777341106,
+      "tokens_seen": 2384789504
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00028009147809340395,
+      "loss": 1.344,
+      "theoretical_loss": 3.40973451395398,
+      "tokens_seen": 2384920576
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00028005135612261276,
+      "loss": 1.3507,
+      "theoretical_loss": 3.40971955122647,
+      "tokens_seen": 2385051648
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00028001123415182156,
+      "loss": 1.3668,
+      "theoretical_loss": 3.4097045895514495,
+      "tokens_seen": 2385182720
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002799711121810303,
+      "loss": 1.2666,
+      "theoretical_loss": 3.409689628928786,
+      "tokens_seen": 2385313792
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002799309902102392,
+      "loss": 1.3279,
+      "theoretical_loss": 3.409674669358348,
+      "tokens_seen": 2385444864
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002798908682394479,
+      "loss": 1.3104,
+      "theoretical_loss": 3.4096597108400037,
+      "tokens_seen": 2385575936
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002798507462686567,
+      "loss": 1.3508,
+      "theoretical_loss": 3.4096447533736214,
+      "tokens_seen": 2385707008
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027981062429786554,
+      "loss": 1.3797,
+      "theoretical_loss": 3.4096297969590688,
+      "tokens_seen": 2385838080
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002797705023270743,
+      "loss": 1.2872,
+      "theoretical_loss": 3.4096148415962144,
+      "tokens_seen": 2385969152
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002797303803562831,
+      "loss": 1.3744,
+      "theoretical_loss": 3.4095998872849265,
+      "tokens_seen": 2386100224
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002796902583854919,
+      "loss": 1.3863,
+      "theoretical_loss": 3.4095849340250735,
+      "tokens_seen": 2386231296
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002796501364147007,
+      "loss": 1.3275,
+      "theoretical_loss": 3.4095699818165235,
+      "tokens_seen": 2386362368
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027961001444390946,
+      "loss": 1.4154,
+      "theoretical_loss": 3.409555030659145,
+      "tokens_seen": 2386493440
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027956989247311827,
+      "loss": 1.4164,
+      "theoretical_loss": 3.409540080552807,
+      "tokens_seen": 2386624512
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002795297705023271,
+      "loss": 1.3311,
+      "theoretical_loss": 3.4095251314973765,
+      "tokens_seen": 2386755584
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002794896485315359,
+      "loss": 1.3873,
+      "theoretical_loss": 3.4095101834927233,
+      "tokens_seen": 2386886656
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002794495265607447,
+      "loss": 1.2882,
+      "theoretical_loss": 3.409495236538715,
+      "tokens_seen": 2387017728
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027940940458995344,
+      "loss": 1.3694,
+      "theoretical_loss": 3.40948029063522,
+      "tokens_seen": 2387148800
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002793692826191623,
+      "loss": 1.33,
+      "theoretical_loss": 3.409465345782108,
+      "tokens_seen": 2387279872
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027932916064837105,
+      "loss": 1.3197,
+      "theoretical_loss": 3.409450401979246,
+      "tokens_seen": 2387410944
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002792890386775798,
+      "loss": 1.3783,
+      "theoretical_loss": 3.4094354592265033,
+      "tokens_seen": 2387542016
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027924891670678866,
+      "loss": 1.4494,
+      "theoretical_loss": 3.409420517523748,
+      "tokens_seen": 2387673088
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002792087947359974,
+      "loss": 1.2954,
+      "theoretical_loss": 3.4094055768708493,
+      "tokens_seen": 2387804160
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002791686727652062,
+      "loss": 1.3543,
+      "theoretical_loss": 3.4093906372676757,
+      "tokens_seen": 2387935232
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027912855079441503,
+      "loss": 1.3703,
+      "theoretical_loss": 3.4093756987140953,
+      "tokens_seen": 2388066304
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027908842882362384,
+      "loss": 1.3318,
+      "theoretical_loss": 3.409360761209977,
+      "tokens_seen": 2388197376
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002790483068528326,
+      "loss": 1.4256,
+      "theoretical_loss": 3.4093458247551904,
+      "tokens_seen": 2388328448
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027900818488204145,
+      "loss": 1.376,
+      "theoretical_loss": 3.4093308893496026,
+      "tokens_seen": 2388459520
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002789680629112502,
+      "loss": 1.3516,
+      "theoretical_loss": 3.4093159549930836,
+      "tokens_seen": 2388590592
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027892794094045895,
+      "loss": 1.259,
+      "theoretical_loss": 3.4093010216855015,
+      "tokens_seen": 2388721664
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002788878189696678,
+      "loss": 1.3841,
+      "theoretical_loss": 3.4092860894267254,
+      "tokens_seen": 2388852736
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027884769699887657,
+      "loss": 1.2705,
+      "theoretical_loss": 3.4092711582166237,
+      "tokens_seen": 2388983808
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002788075750280854,
+      "loss": 1.3481,
+      "theoretical_loss": 3.4092562280550656,
+      "tokens_seen": 2389114880
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002787674530572942,
+      "loss": 1.3891,
+      "theoretical_loss": 3.4092412989419203,
+      "tokens_seen": 2389245952
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.000278727331086503,
+      "loss": 1.3297,
+      "theoretical_loss": 3.4092263708770556,
+      "tokens_seen": 2389377024
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002786872091157118,
+      "loss": 1.3199,
+      "theoretical_loss": 3.4092114438603414,
+      "tokens_seen": 2389508096
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027864708714492054,
+      "loss": 1.3727,
+      "theoretical_loss": 3.4091965178916457,
+      "tokens_seen": 2389639168
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027860696517412935,
+      "loss": 1.3894,
+      "theoretical_loss": 3.4091815929708384,
+      "tokens_seen": 2389770240
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027856684320333816,
+      "loss": 1.3817,
+      "theoretical_loss": 3.4091666690977878,
+      "tokens_seen": 2389901312
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027852672123254696,
+      "loss": 1.3202,
+      "theoretical_loss": 3.4091517462723635,
+      "tokens_seen": 2390032384
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002784865992617557,
+      "loss": 1.3523,
+      "theoretical_loss": 3.409136824494434,
+      "tokens_seen": 2390163456
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002784464772909646,
+      "loss": 1.3118,
+      "theoretical_loss": 3.409121903763868,
+      "tokens_seen": 2390294528
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002784063553201733,
+      "loss": 1.3531,
+      "theoretical_loss": 3.4091069840805353,
+      "tokens_seen": 2390425600
+    },
+    {
+      "epoch": 0.45,
+      "objective/train/advantage_avg": 0.4902363419532776,
+      "objective/train/docs_used": 870459,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2631425857543945,
+      "objective/train/original_loss": 1.2631425857543945,
+      "objective/train/theoretical_loss": 3.4090920654443044,
+      "objective/train/tokens_used": 761015776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24124352633953094,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0502500534057617,
+      "objective/train/weighted_lm_loss": 1.3266063928604126,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4090920654443044,
+      "tokens_seen": 2390556672
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002783662333493821,
+      "loss": 1.3386,
+      "theoretical_loss": 3.4090920654443044,
+      "tokens_seen": 2390556672
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027832611137859094,
+      "loss": 1.3165,
+      "theoretical_loss": 3.409077147855045,
+      "tokens_seen": 2390687744
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002782859894077997,
+      "loss": 1.3671,
+      "theoretical_loss": 3.409062231312626,
+      "tokens_seen": 2390818816
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027824586743700855,
+      "loss": 1.3924,
+      "theoretical_loss": 3.409047315816917,
+      "tokens_seen": 2390949888
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002782057454662173,
+      "loss": 1.3757,
+      "theoretical_loss": 3.4090324013677855,
+      "tokens_seen": 2391080960
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002781656234954261,
+      "loss": 1.3839,
+      "theoretical_loss": 3.4090174879651025,
+      "tokens_seen": 2391212032
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002781255015246349,
+      "loss": 1.3119,
+      "theoretical_loss": 3.4090025756087368,
+      "tokens_seen": 2391343104
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027808537955384367,
+      "loss": 1.3623,
+      "theoretical_loss": 3.4089876642985573,
+      "tokens_seen": 2391474176
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002780452575830525,
+      "loss": 1.3833,
+      "theoretical_loss": 3.408972754034433,
+      "tokens_seen": 2391605248
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002780051356122613,
+      "loss": 1.3511,
+      "theoretical_loss": 3.408957844816234,
+      "tokens_seen": 2391736320
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002779650136414701,
+      "loss": 1.4185,
+      "theoretical_loss": 3.408942936643829,
+      "tokens_seen": 2391867392
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027792489167067884,
+      "loss": 1.3894,
+      "theoretical_loss": 3.408928029517088,
+      "tokens_seen": 2391998464
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002778847696998877,
+      "loss": 1.3235,
+      "theoretical_loss": 3.4089131234358794,
+      "tokens_seen": 2392129536
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027784464772909645,
+      "loss": 1.3974,
+      "theoretical_loss": 3.408898218400073,
+      "tokens_seen": 2392260608
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002778045257583052,
+      "loss": 1.3958,
+      "theoretical_loss": 3.4088833144095387,
+      "tokens_seen": 2392391680
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027776440378751406,
+      "loss": 1.3628,
+      "theoretical_loss": 3.4088684114641454,
+      "tokens_seen": 2392522752
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002777242818167228,
+      "loss": 1.3291,
+      "theoretical_loss": 3.408853509563763,
+      "tokens_seen": 2392653824
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002776841598459316,
+      "loss": 1.4298,
+      "theoretical_loss": 3.4088386087082605,
+      "tokens_seen": 2392784896
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027764403787514043,
+      "loss": 1.3034,
+      "theoretical_loss": 3.4088237088975073,
+      "tokens_seen": 2392915968
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027760391590434924,
+      "loss": 1.3262,
+      "theoretical_loss": 3.4088088101313736,
+      "tokens_seen": 2393047040
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027756379393355804,
+      "loss": 1.3187,
+      "theoretical_loss": 3.408793912409729,
+      "tokens_seen": 2393178112
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027752367196276685,
+      "loss": 1.4481,
+      "theoretical_loss": 3.408779015732442,
+      "tokens_seen": 2393309184
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002774835499919756,
+      "loss": 1.3085,
+      "theoretical_loss": 3.408764120099383,
+      "tokens_seen": 2393440256
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002774434280211844,
+      "loss": 1.2981,
+      "theoretical_loss": 3.4087492255104213,
+      "tokens_seen": 2393571328
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002774033060503932,
+      "loss": 1.2822,
+      "theoretical_loss": 3.408734331965427,
+      "tokens_seen": 2393702400
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027736318407960196,
+      "loss": 1.2873,
+      "theoretical_loss": 3.4087194394642695,
+      "tokens_seen": 2393833472
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002773230621088108,
+      "loss": 1.2695,
+      "theoretical_loss": 3.4087045480068183,
+      "tokens_seen": 2393964544
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002772829401380196,
+      "loss": 1.3569,
+      "theoretical_loss": 3.4086896575929435,
+      "tokens_seen": 2394095616
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002772428181672284,
+      "loss": 1.3797,
+      "theoretical_loss": 3.408674768222515,
+      "tokens_seen": 2394226688
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002772026961964372,
+      "loss": 1.373,
+      "theoretical_loss": 3.4086598798954015,
+      "tokens_seen": 2394357760
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027716257422564594,
+      "loss": 1.3687,
+      "theoretical_loss": 3.4086449926114737,
+      "tokens_seen": 2394488832
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027712245225485475,
+      "loss": 1.2827,
+      "theoretical_loss": 3.4086301063706017,
+      "tokens_seen": 2394619904
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027708233028406355,
+      "loss": 1.263,
+      "theoretical_loss": 3.408615221172654,
+      "tokens_seen": 2394750976
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027704220831327236,
+      "loss": 1.3153,
+      "theoretical_loss": 3.408600337017502,
+      "tokens_seen": 2394882048
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027700208634248117,
+      "loss": 1.3775,
+      "theoretical_loss": 3.4085854539050144,
+      "tokens_seen": 2395013120
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027696196437169,
+      "loss": 1.386,
+      "theoretical_loss": 3.408570571835062,
+      "tokens_seen": 2395144192
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002769218424008987,
+      "loss": 1.3654,
+      "theoretical_loss": 3.408555690807514,
+      "tokens_seen": 2395275264
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027688172043010753,
+      "loss": 1.3681,
+      "theoretical_loss": 3.408540810822241,
+      "tokens_seen": 2395406336
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027684159845931634,
+      "loss": 1.4078,
+      "theoretical_loss": 3.4085259318791126,
+      "tokens_seen": 2395537408
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002768014764885251,
+      "loss": 1.2879,
+      "theoretical_loss": 3.4085110539779984,
+      "tokens_seen": 2395668480
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027676135451773395,
+      "loss": 1.3015,
+      "theoretical_loss": 3.408496177118769,
+      "tokens_seen": 2395799552
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002767212325469427,
+      "loss": 1.4214,
+      "theoretical_loss": 3.4084813013012942,
+      "tokens_seen": 2395930624
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002766811105761515,
+      "loss": 1.3504,
+      "theoretical_loss": 3.4084664265254445,
+      "tokens_seen": 2396061696
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002766409886053603,
+      "loss": 1.3183,
+      "theoretical_loss": 3.4084515527910897,
+      "tokens_seen": 2396192768
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027660086663456907,
+      "loss": 1.4023,
+      "theoretical_loss": 3.4084366800980996,
+      "tokens_seen": 2396323840
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002765607446637779,
+      "loss": 1.3502,
+      "theoretical_loss": 3.4084218084463447,
+      "tokens_seen": 2396454912
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002765206226929867,
+      "loss": 1.3786,
+      "theoretical_loss": 3.4084069378356947,
+      "tokens_seen": 2396585984
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002764805007221955,
+      "loss": 1.3252,
+      "theoretical_loss": 3.4083920682660205,
+      "tokens_seen": 2396717056
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027644037875140424,
+      "loss": 1.3045,
+      "theoretical_loss": 3.408377199737192,
+      "tokens_seen": 2396848128
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002764002567806131,
+      "loss": 1.3587,
+      "theoretical_loss": 3.4083623322490793,
+      "tokens_seen": 2396979200
+    },
+    {
+      "epoch": 0.45,
+      "objective/train/advantage_avg": 0.493741899728775,
+      "objective/train/docs_used": 872754,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3581417798995972,
+      "objective/train/original_loss": 1.3581417798995972,
+      "objective/train/theoretical_loss": 3.4083474658015525,
+      "objective/train/tokens_used": 767569376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.244066521525383,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506150722503662,
+      "objective/train/weighted_lm_loss": 1.4266716241836548,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4083474658015525,
+      "tokens_seen": 2397110272
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027636013480982185,
+      "loss": 1.4142,
+      "theoretical_loss": 3.4083474658015525,
+      "tokens_seen": 2397110272
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027632001283903066,
+      "loss": 1.3413,
+      "theoretical_loss": 3.4083326003944823,
+      "tokens_seen": 2397241344
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027627989086823946,
+      "loss": 1.3948,
+      "theoretical_loss": 3.4083177360277386,
+      "tokens_seen": 2397372416
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002762397688974482,
+      "loss": 1.4053,
+      "theoretical_loss": 3.408302872701192,
+      "tokens_seen": 2397503488
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002761996469266571,
+      "loss": 1.311,
+      "theoretical_loss": 3.4082880104147133,
+      "tokens_seen": 2397634560
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027615952495586583,
+      "loss": 1.3595,
+      "theoretical_loss": 3.4082731491681715,
+      "tokens_seen": 2397765632
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027611940298507464,
+      "loss": 1.3182,
+      "theoretical_loss": 3.408258288961439,
+      "tokens_seen": 2397896704
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027607928101428344,
+      "loss": 1.369,
+      "theoretical_loss": 3.4082434297943838,
+      "tokens_seen": 2398027776
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027603915904349225,
+      "loss": 1.3845,
+      "theoretical_loss": 3.408228571666878,
+      "tokens_seen": 2398158848
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.000275999037072701,
+      "loss": 1.3631,
+      "theoretical_loss": 3.4082137145787916,
+      "tokens_seen": 2398289920
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002759589151019098,
+      "loss": 1.3173,
+      "theoretical_loss": 3.4081988585299956,
+      "tokens_seen": 2398420992
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002759187931311186,
+      "loss": 1.4099,
+      "theoretical_loss": 3.4081840035203594,
+      "tokens_seen": 2398552064
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027587867116032736,
+      "loss": 1.4197,
+      "theoretical_loss": 3.408169149549755,
+      "tokens_seen": 2398683136
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002758385491895362,
+      "loss": 1.3891,
+      "theoretical_loss": 3.4081542966180516,
+      "tokens_seen": 2398814208
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.000275798427218745,
+      "loss": 1.3442,
+      "theoretical_loss": 3.40813944472512,
+      "tokens_seen": 2398945280
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027575830524795384,
+      "loss": 1.3491,
+      "theoretical_loss": 3.4081245938708316,
+      "tokens_seen": 2399076352
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002757181832771626,
+      "loss": 1.364,
+      "theoretical_loss": 3.4081097440550563,
+      "tokens_seen": 2399207424
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027567806130637134,
+      "loss": 1.353,
+      "theoretical_loss": 3.408094895277665,
+      "tokens_seen": 2399338496
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002756379393355802,
+      "loss": 1.3763,
+      "theoretical_loss": 3.4080800475385287,
+      "tokens_seen": 2399469568
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027559781736478895,
+      "loss": 1.3084,
+      "theoretical_loss": 3.4080652008375174,
+      "tokens_seen": 2399600640
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027555769539399776,
+      "loss": 1.2837,
+      "theoretical_loss": 3.4080503551745025,
+      "tokens_seen": 2399731712
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027551757342320657,
+      "loss": 1.3341,
+      "theoretical_loss": 3.408035510549354,
+      "tokens_seen": 2399862784
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002754774514524154,
+      "loss": 1.3479,
+      "theoretical_loss": 3.408020666961943,
+      "tokens_seen": 2399993856
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002754373294816241,
+      "loss": 1.336,
+      "theoretical_loss": 3.408005824412141,
+      "tokens_seen": 2400124928
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027539720751083293,
+      "loss": 1.4108,
+      "theoretical_loss": 3.4079909828998174,
+      "tokens_seen": 2400256000
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027535708554004174,
+      "loss": 1.3841,
+      "theoretical_loss": 3.407976142424844,
+      "tokens_seen": 2400387072
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002753169635692505,
+      "loss": 1.3614,
+      "theoretical_loss": 3.4079613029870917,
+      "tokens_seen": 2400518144
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00027527684159845935,
+      "loss": 1.4328,
+      "theoretical_loss": 3.407946464586431,
+      "tokens_seen": 2400649216
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002752367196276681,
+      "loss": 1.3095,
+      "theoretical_loss": 3.4079316272227334,
+      "tokens_seen": 2400780288
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002751965976568769,
+      "loss": 1.3088,
+      "theoretical_loss": 3.407916790895869,
+      "tokens_seen": 2400911360
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002751564756860857,
+      "loss": 1.414,
+      "theoretical_loss": 3.4079019556057086,
+      "tokens_seen": 2401042432
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027511635371529447,
+      "loss": 1.3181,
+      "theoretical_loss": 3.4078871213521245,
+      "tokens_seen": 2401173504
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027507623174450333,
+      "loss": 1.3907,
+      "theoretical_loss": 3.4078722881349863,
+      "tokens_seen": 2401304576
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002750361097737121,
+      "loss": 1.4441,
+      "theoretical_loss": 3.407857455954166,
+      "tokens_seen": 2401435648
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002749959878029209,
+      "loss": 1.2848,
+      "theoretical_loss": 3.407842624809534,
+      "tokens_seen": 2401566720
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002749558658321297,
+      "loss": 1.3987,
+      "theoretical_loss": 3.4078277947009616,
+      "tokens_seen": 2401697792
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002749157438613385,
+      "loss": 1.3556,
+      "theoretical_loss": 3.40781296562832,
+      "tokens_seen": 2401828864
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027487562189054725,
+      "loss": 1.2868,
+      "theoretical_loss": 3.40779813759148,
+      "tokens_seen": 2401959936
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027483549991975606,
+      "loss": 1.3146,
+      "theoretical_loss": 3.4077833105903133,
+      "tokens_seen": 2402091008
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027479537794896486,
+      "loss": 1.3668,
+      "theoretical_loss": 3.4077684846246905,
+      "tokens_seen": 2402222080
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002747552559781736,
+      "loss": 1.4476,
+      "theoretical_loss": 3.4077536596944826,
+      "tokens_seen": 2402353152
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002747151340073825,
+      "loss": 1.2553,
+      "theoretical_loss": 3.4077388357995617,
+      "tokens_seen": 2402484224
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027467501203659123,
+      "loss": 1.4234,
+      "theoretical_loss": 3.407724012939798,
+      "tokens_seen": 2402615296
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027463489006580003,
+      "loss": 1.3487,
+      "theoretical_loss": 3.407709191115064,
+      "tokens_seen": 2402746368
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027459476809500884,
+      "loss": 1.4015,
+      "theoretical_loss": 3.4076943703252294,
+      "tokens_seen": 2402877440
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002745546461242176,
+      "loss": 1.3633,
+      "theoretical_loss": 3.4076795505701667,
+      "tokens_seen": 2403008512
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002745145241534264,
+      "loss": 1.3086,
+      "theoretical_loss": 3.4076647318497466,
+      "tokens_seen": 2403139584
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002744744021826352,
+      "loss": 1.3398,
+      "theoretical_loss": 3.407649914163841,
+      "tokens_seen": 2403270656
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.000274434280211844,
+      "loss": 1.3007,
+      "theoretical_loss": 3.407635097512321,
+      "tokens_seen": 2403401728
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002743941582410528,
+      "loss": 1.3832,
+      "theoretical_loss": 3.4076202818950576,
+      "tokens_seen": 2403532800
+    },
+    {
+      "epoch": 0.46,
+      "objective/train/advantage_avg": 0.4917581379413605,
+      "objective/train/docs_used": 875100,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3093440532684326,
+      "objective/train/original_loss": 1.3093440532684326,
+      "objective/train/theoretical_loss": 3.4076054673119227,
+      "objective/train/tokens_used": 774122976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24291522800922394,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504106283187866,
+      "objective/train/weighted_lm_loss": 1.3753472566604614,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4076054673119227,
+      "tokens_seen": 2403663872
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002743540362702616,
+      "loss": 1.3932,
+      "theoretical_loss": 3.4076054673119227,
+      "tokens_seen": 2403663872
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002743139142994704,
+      "loss": 1.4134,
+      "theoretical_loss": 3.407590653762788,
+      "tokens_seen": 2403794944
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027427379232867924,
+      "loss": 1.3723,
+      "theoretical_loss": 3.4075758412475237,
+      "tokens_seen": 2403926016
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.000274233670357888,
+      "loss": 1.4131,
+      "theoretical_loss": 3.4075610297660024,
+      "tokens_seen": 2404057088
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027419354838709674,
+      "loss": 1.369,
+      "theoretical_loss": 3.407546219318095,
+      "tokens_seen": 2404188160
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002741534264163056,
+      "loss": 1.4605,
+      "theoretical_loss": 3.407531409903674,
+      "tokens_seen": 2404319232
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027411330444551435,
+      "loss": 1.4718,
+      "theoretical_loss": 3.40751660152261,
+      "tokens_seen": 2404450304
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027407318247472316,
+      "loss": 1.3568,
+      "theoretical_loss": 3.4075017941747747,
+      "tokens_seen": 2404581376
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027403306050393197,
+      "loss": 1.2627,
+      "theoretical_loss": 3.40748698786004,
+      "tokens_seen": 2404712448
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002739929385331408,
+      "loss": 1.3575,
+      "theoretical_loss": 3.407472182578277,
+      "tokens_seen": 2404843520
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002739528165623495,
+      "loss": 1.4001,
+      "theoretical_loss": 3.407457378329358,
+      "tokens_seen": 2404974592
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027391269459155833,
+      "loss": 1.3579,
+      "theoretical_loss": 3.4074425751131545,
+      "tokens_seen": 2405105664
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027387257262076714,
+      "loss": 1.4932,
+      "theoretical_loss": 3.407427772929538,
+      "tokens_seen": 2405236736
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027383245064997594,
+      "loss": 1.4724,
+      "theoretical_loss": 3.40741297177838,
+      "tokens_seen": 2405367808
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027379232867918475,
+      "loss": 1.3317,
+      "theoretical_loss": 3.4073981716595525,
+      "tokens_seen": 2405498880
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002737522067083935,
+      "loss": 1.3637,
+      "theoretical_loss": 3.4073833725729274,
+      "tokens_seen": 2405629952
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027371208473760236,
+      "loss": 1.3753,
+      "theoretical_loss": 3.407368574518376,
+      "tokens_seen": 2405761024
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002736719627668111,
+      "loss": 1.3034,
+      "theoretical_loss": 3.407353777495771,
+      "tokens_seen": 2405892096
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027363184079601987,
+      "loss": 1.4699,
+      "theoretical_loss": 3.4073389815049833,
+      "tokens_seen": 2406023168
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027359171882522873,
+      "loss": 1.3934,
+      "theoretical_loss": 3.407324186545885,
+      "tokens_seen": 2406154240
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002735515968544375,
+      "loss": 1.3563,
+      "theoretical_loss": 3.407309392618348,
+      "tokens_seen": 2406285312
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002735114748836463,
+      "loss": 1.4166,
+      "theoretical_loss": 3.4072945997222446,
+      "tokens_seen": 2406416384
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002734713529128551,
+      "loss": 1.4425,
+      "theoretical_loss": 3.4072798078574458,
+      "tokens_seen": 2406547456
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002734312309420639,
+      "loss": 1.372,
+      "theoretical_loss": 3.4072650170238243,
+      "tokens_seen": 2406678528
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027339110897127265,
+      "loss": 1.3927,
+      "theoretical_loss": 3.407250227221252,
+      "tokens_seen": 2406809600
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027335098700048146,
+      "loss": 1.4156,
+      "theoretical_loss": 3.407235438449601,
+      "tokens_seen": 2406940672
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027331086502969026,
+      "loss": 1.3088,
+      "theoretical_loss": 3.4072206507087426,
+      "tokens_seen": 2407071744
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.000273270743058899,
+      "loss": 1.3557,
+      "theoretical_loss": 3.4072058639985494,
+      "tokens_seen": 2407202816
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002732306210881079,
+      "loss": 1.4879,
+      "theoretical_loss": 3.4071910783188932,
+      "tokens_seen": 2407333888
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027319049911731663,
+      "loss": 1.3804,
+      "theoretical_loss": 3.407176293669646,
+      "tokens_seen": 2407464960
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002731503771465255,
+      "loss": 1.3019,
+      "theoretical_loss": 3.4071615100506807,
+      "tokens_seen": 2407596032
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027311025517573424,
+      "loss": 1.3599,
+      "theoretical_loss": 3.4071467274618685,
+      "tokens_seen": 2407727104
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.000273070133204943,
+      "loss": 1.3531,
+      "theoretical_loss": 3.4071319459030818,
+      "tokens_seen": 2407858176
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027303001123415185,
+      "loss": 1.3785,
+      "theoretical_loss": 3.407117165374193,
+      "tokens_seen": 2407989248
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002729898892633606,
+      "loss": 1.3587,
+      "theoretical_loss": 3.407102385875074,
+      "tokens_seen": 2408120320
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002729497672925694,
+      "loss": 1.4191,
+      "theoretical_loss": 3.407087607405597,
+      "tokens_seen": 2408251392
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002729096453217782,
+      "loss": 1.3533,
+      "theoretical_loss": 3.4070728299656348,
+      "tokens_seen": 2408382464
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.000272869523350987,
+      "loss": 1.3897,
+      "theoretical_loss": 3.4070580535550588,
+      "tokens_seen": 2408513536
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002728294013801958,
+      "loss": 1.3066,
+      "theoretical_loss": 3.4070432781737416,
+      "tokens_seen": 2408644608
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027278927940940464,
+      "loss": 1.3554,
+      "theoretical_loss": 3.407028503821556,
+      "tokens_seen": 2408775680
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002727491574386134,
+      "loss": 1.3048,
+      "theoretical_loss": 3.407013730498374,
+      "tokens_seen": 2408906752
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027270903546782214,
+      "loss": 1.3227,
+      "theoretical_loss": 3.406998958204067,
+      "tokens_seen": 2409037824
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.000272668913497031,
+      "loss": 1.2993,
+      "theoretical_loss": 3.406984186938509,
+      "tokens_seen": 2409168896
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027262879152623975,
+      "loss": 1.3856,
+      "theoretical_loss": 3.4069694167015716,
+      "tokens_seen": 2409299968
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002725886695554486,
+      "loss": 1.423,
+      "theoretical_loss": 3.406954647493127,
+      "tokens_seen": 2409431040
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027254854758465737,
+      "loss": 1.3978,
+      "theoretical_loss": 3.406939879313048,
+      "tokens_seen": 2409562112
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027250842561386617,
+      "loss": 1.5155,
+      "theoretical_loss": 3.406925112161207,
+      "tokens_seen": 2409693184
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.000272468303643075,
+      "loss": 1.3992,
+      "theoretical_loss": 3.4069103460374763,
+      "tokens_seen": 2409824256
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027242818167228373,
+      "loss": 1.3351,
+      "theoretical_loss": 3.406895580941729,
+      "tokens_seen": 2409955328
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027238805970149254,
+      "loss": 1.4357,
+      "theoretical_loss": 3.4068808168738367,
+      "tokens_seen": 2410086400
+    },
+    {
+      "epoch": 0.46,
+      "objective/train/advantage_avg": 0.4906492233276367,
+      "objective/train/docs_used": 877499,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.299674391746521,
+      "objective/train/original_loss": 1.2996745109558105,
+      "objective/train/theoretical_loss": 3.4068660538336726,
+      "objective/train/tokens_used": 780676576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24124418199062347,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0502912998199463,
+      "objective/train/weighted_lm_loss": 1.3649916648864746,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4068660538336726,
+      "tokens_seen": 2410217472
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027234793773070134,
+      "loss": 1.346,
+      "theoretical_loss": 3.4068660538336726,
+      "tokens_seen": 2410217472
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027230781575991015,
+      "loss": 1.3836,
+      "theoretical_loss": 3.406851291821109,
+      "tokens_seen": 2410348544
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002722676937891189,
+      "loss": 1.3917,
+      "theoretical_loss": 3.406836530836019,
+      "tokens_seen": 2410479616
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027222757181832776,
+      "loss": 1.3683,
+      "theoretical_loss": 3.406821770878274,
+      "tokens_seen": 2410610688
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002721874498475365,
+      "loss": 1.381,
+      "theoretical_loss": 3.4068070119477483,
+      "tokens_seen": 2410741760
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027214732787674527,
+      "loss": 1.385,
+      "theoretical_loss": 3.406792254044313,
+      "tokens_seen": 2410872832
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027210720590595413,
+      "loss": 1.4034,
+      "theoretical_loss": 3.406777497167842,
+      "tokens_seen": 2411003904
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002720670839351629,
+      "loss": 1.3533,
+      "theoretical_loss": 3.4067627413182078,
+      "tokens_seen": 2411134976
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002720269619643717,
+      "loss": 1.359,
+      "theoretical_loss": 3.4067479864952825,
+      "tokens_seen": 2411266048
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002719868399935805,
+      "loss": 1.3632,
+      "theoretical_loss": 3.4067332326989392,
+      "tokens_seen": 2411397120
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002719467180227893,
+      "loss": 1.3574,
+      "theoretical_loss": 3.406718479929051,
+      "tokens_seen": 2411528192
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002719065960519981,
+      "loss": 1.4744,
+      "theoretical_loss": 3.40670372818549,
+      "tokens_seen": 2411659264
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027186647408120686,
+      "loss": 1.2979,
+      "theoretical_loss": 3.4066889774681295,
+      "tokens_seen": 2411790336
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027182635211041566,
+      "loss": 1.3574,
+      "theoretical_loss": 3.4066742277768425,
+      "tokens_seen": 2411921408
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027178623013962447,
+      "loss": 1.3834,
+      "theoretical_loss": 3.4066594791115015,
+      "tokens_seen": 2412052480
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002717461081688333,
+      "loss": 1.3838,
+      "theoretical_loss": 3.40664473147198,
+      "tokens_seen": 2412183552
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027170598619804203,
+      "loss": 1.339,
+      "theoretical_loss": 3.40662998485815,
+      "tokens_seen": 2412314624
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002716658642272509,
+      "loss": 1.3553,
+      "theoretical_loss": 3.4066152392698847,
+      "tokens_seen": 2412445696
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027162574225645964,
+      "loss": 1.389,
+      "theoretical_loss": 3.4066004947070576,
+      "tokens_seen": 2412576768
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002715856202856684,
+      "loss": 1.4564,
+      "theoretical_loss": 3.4065857511695414,
+      "tokens_seen": 2412707840
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027154549831487725,
+      "loss": 1.3854,
+      "theoretical_loss": 3.406571008657209,
+      "tokens_seen": 2412838912
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.000271505376344086,
+      "loss": 1.3281,
+      "theoretical_loss": 3.4065562671699334,
+      "tokens_seen": 2412969984
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002714652543732948,
+      "loss": 1.4044,
+      "theoretical_loss": 3.406541526707588,
+      "tokens_seen": 2413101056
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002714251324025036,
+      "loss": 1.3347,
+      "theoretical_loss": 3.4065267872700455,
+      "tokens_seen": 2413232128
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002713850104317124,
+      "loss": 1.3892,
+      "theoretical_loss": 3.406512048857179,
+      "tokens_seen": 2413363200
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002713448884609212,
+      "loss": 1.3847,
+      "theoretical_loss": 3.406497311468862,
+      "tokens_seen": 2413494272
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027130476649013004,
+      "loss": 1.2876,
+      "theoretical_loss": 3.406482575104967,
+      "tokens_seen": 2413625344
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002712646445193388,
+      "loss": 1.3629,
+      "theoretical_loss": 3.4064678397653676,
+      "tokens_seen": 2413756416
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002712245225485476,
+      "loss": 1.3655,
+      "theoretical_loss": 3.406453105449937,
+      "tokens_seen": 2413887488
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002711844005777564,
+      "loss": 1.3957,
+      "theoretical_loss": 3.4064383721585485,
+      "tokens_seen": 2414018560
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027114427860696515,
+      "loss": 1.4015,
+      "theoretical_loss": 3.4064236398910754,
+      "tokens_seen": 2414149632
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.000271104156636174,
+      "loss": 1.3993,
+      "theoretical_loss": 3.40640890864739,
+      "tokens_seen": 2414280704
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027106403466538277,
+      "loss": 1.361,
+      "theoretical_loss": 3.406394178427367,
+      "tokens_seen": 2414411776
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027102391269459157,
+      "loss": 1.3745,
+      "theoretical_loss": 3.4063794492308785,
+      "tokens_seen": 2414542848
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002709837907238004,
+      "loss": 1.4156,
+      "theoretical_loss": 3.4063647210577988,
+      "tokens_seen": 2414673920
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027094366875300913,
+      "loss": 1.3635,
+      "theoretical_loss": 3.4063499939080004,
+      "tokens_seen": 2414804992
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027090354678221794,
+      "loss": 1.3095,
+      "theoretical_loss": 3.406335267781357,
+      "tokens_seen": 2414936064
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027086342481142674,
+      "loss": 1.3955,
+      "theoretical_loss": 3.4063205426777423,
+      "tokens_seen": 2415067136
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027082330284063555,
+      "loss": 1.334,
+      "theoretical_loss": 3.4063058185970294,
+      "tokens_seen": 2415198208
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002707831808698443,
+      "loss": 1.3249,
+      "theoretical_loss": 3.4062910955390917,
+      "tokens_seen": 2415329280
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027074305889905316,
+      "loss": 1.4039,
+      "theoretical_loss": 3.4062763735038026,
+      "tokens_seen": 2415460352
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002707029369282619,
+      "loss": 1.3594,
+      "theoretical_loss": 3.406261652491036,
+      "tokens_seen": 2415591424
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027066281495747067,
+      "loss": 1.4225,
+      "theoretical_loss": 3.406246932500665,
+      "tokens_seen": 2415722496
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002706226929866795,
+      "loss": 1.3382,
+      "theoretical_loss": 3.4062322135325633,
+      "tokens_seen": 2415853568
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002705825710158883,
+      "loss": 1.3981,
+      "theoretical_loss": 3.4062174955866045,
+      "tokens_seen": 2415984640
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027054244904509714,
+      "loss": 1.3403,
+      "theoretical_loss": 3.406202778662662,
+      "tokens_seen": 2416115712
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002705023270743059,
+      "loss": 1.4256,
+      "theoretical_loss": 3.406188062760609,
+      "tokens_seen": 2416246784
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002704622051035147,
+      "loss": 1.317,
+      "theoretical_loss": 3.40617334788032,
+      "tokens_seen": 2416377856
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002704220831327235,
+      "loss": 1.432,
+      "theoretical_loss": 3.406158634021668,
+      "tokens_seen": 2416508928
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027038196116193226,
+      "loss": 1.3748,
+      "theoretical_loss": 3.406143921184527,
+      "tokens_seen": 2416640000
+    },
+    {
+      "epoch": 0.46,
+      "objective/train/advantage_avg": 0.4937954843044281,
+      "objective/train/docs_used": 880009,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.312087893486023,
+      "objective/train/original_loss": 1.3120877742767334,
+      "objective/train/theoretical_loss": 3.4061292093687707,
+      "objective/train/tokens_used": 787230176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24416138231754303,
+      "objective/train/value_max": -0.5,
+      "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": 1.3784689903259277,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4061292093687707,
+      "tokens_seen": 2416771072
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027034183919114106,
+      "loss": 1.3754,
+      "theoretical_loss": 3.4061292093687707,
+      "tokens_seen": 2416771072
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027030171722034987,
+      "loss": 1.4558,
+      "theoretical_loss": 3.4061144985742726,
+      "tokens_seen": 2416902144
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002702615952495587,
+      "loss": 1.3601,
+      "theoretical_loss": 3.4060997888009066,
+      "tokens_seen": 2417033216
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00027022147327876743,
+      "loss": 1.3667,
+      "theoretical_loss": 3.406085080048546,
+      "tokens_seen": 2417164288
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002701813513079763,
+      "loss": 1.5393,
+      "theoretical_loss": 3.4060703723170653,
+      "tokens_seen": 2417295360
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00027014122933718504,
+      "loss": 1.3901,
+      "theoretical_loss": 3.406055665606338,
+      "tokens_seen": 2417426432
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002701011073663938,
+      "loss": 1.4136,
+      "theoretical_loss": 3.4060409599162376,
+      "tokens_seen": 2417557504
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00027006098539560265,
+      "loss": 1.3504,
+      "theoretical_loss": 3.406026255246638,
+      "tokens_seen": 2417688576
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002700208634248114,
+      "loss": 1.3714,
+      "theoretical_loss": 3.4060115515974134,
+      "tokens_seen": 2417819648
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026998074145402027,
+      "loss": 1.3095,
+      "theoretical_loss": 3.4059968489684374,
+      "tokens_seen": 2417950720
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.000269940619483229,
+      "loss": 1.3109,
+      "theoretical_loss": 3.4059821473595844,
+      "tokens_seen": 2418081792
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002699004975124378,
+      "loss": 1.3569,
+      "theoretical_loss": 3.4059674467707275,
+      "tokens_seen": 2418212864
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026986037554164663,
+      "loss": 1.3724,
+      "theoretical_loss": 3.4059527472017415,
+      "tokens_seen": 2418343936
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002698202535708554,
+      "loss": 1.3097,
+      "theoretical_loss": 3.4059380486524997,
+      "tokens_seen": 2418475008
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002697801316000642,
+      "loss": 1.3481,
+      "theoretical_loss": 3.4059233511228766,
+      "tokens_seen": 2418606080
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.000269740009629273,
+      "loss": 1.4488,
+      "theoretical_loss": 3.405908654612746,
+      "tokens_seen": 2418737152
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002696998876584818,
+      "loss": 1.4065,
+      "theoretical_loss": 3.4058939591219817,
+      "tokens_seen": 2418868224
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026965976568769055,
+      "loss": 1.3136,
+      "theoretical_loss": 3.405879264650458,
+      "tokens_seen": 2418999296
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002696196437168994,
+      "loss": 1.4062,
+      "theoretical_loss": 3.4058645711980495,
+      "tokens_seen": 2419130368
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026957952174610817,
+      "loss": 1.3615,
+      "theoretical_loss": 3.4058498787646294,
+      "tokens_seen": 2419261440
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026953939977531697,
+      "loss": 1.327,
+      "theoretical_loss": 3.405835187350072,
+      "tokens_seen": 2419392512
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002694992778045258,
+      "loss": 1.3625,
+      "theoretical_loss": 3.4058204969542523,
+      "tokens_seen": 2419523584
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026945915583373453,
+      "loss": 1.3427,
+      "theoretical_loss": 3.4058058075770434,
+      "tokens_seen": 2419654656
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026941903386294334,
+      "loss": 1.406,
+      "theoretical_loss": 3.4057911192183203,
+      "tokens_seen": 2419785728
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026937891189215214,
+      "loss": 1.355,
+      "theoretical_loss": 3.405776431877957,
+      "tokens_seen": 2419916800
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026933878992136095,
+      "loss": 1.3582,
+      "theoretical_loss": 3.405761745555827,
+      "tokens_seen": 2420047872
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026929866795056976,
+      "loss": 1.3331,
+      "theoretical_loss": 3.4057470602518056,
+      "tokens_seen": 2420178944
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026925854597977856,
+      "loss": 1.4517,
+      "theoretical_loss": 3.405732375965767,
+      "tokens_seen": 2420310016
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002692184240089873,
+      "loss": 1.3616,
+      "theoretical_loss": 3.4057176926975847,
+      "tokens_seen": 2420441088
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002691783020381961,
+      "loss": 1.3896,
+      "theoretical_loss": 3.4057030104471337,
+      "tokens_seen": 2420572160
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002691381800674049,
+      "loss": 1.3042,
+      "theoretical_loss": 3.4056883292142883,
+      "tokens_seen": 2420703232
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002690980580966137,
+      "loss": 1.3709,
+      "theoretical_loss": 3.4056736489989223,
+      "tokens_seen": 2420834304
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026905793612582254,
+      "loss": 1.3581,
+      "theoretical_loss": 3.405658969800911,
+      "tokens_seen": 2420965376
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002690178141550313,
+      "loss": 1.3851,
+      "theoretical_loss": 3.4056442916201286,
+      "tokens_seen": 2421096448
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002689776921842401,
+      "loss": 1.3988,
+      "theoretical_loss": 3.4056296144564486,
+      "tokens_seen": 2421227520
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002689375702134489,
+      "loss": 1.3719,
+      "theoretical_loss": 3.4056149383097467,
+      "tokens_seen": 2421358592
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026889744824265766,
+      "loss": 1.4686,
+      "theoretical_loss": 3.4056002631798967,
+      "tokens_seen": 2421489664
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026885732627186646,
+      "loss": 1.3331,
+      "theoretical_loss": 3.4055855890667734,
+      "tokens_seen": 2421620736
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026881720430107527,
+      "loss": 1.4207,
+      "theoretical_loss": 3.4055709159702516,
+      "tokens_seen": 2421751808
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002687770823302841,
+      "loss": 1.3419,
+      "theoretical_loss": 3.4055562438902047,
+      "tokens_seen": 2421882880
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002687369603594929,
+      "loss": 1.4039,
+      "theoretical_loss": 3.405541572826509,
+      "tokens_seen": 2422013952
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002686968383887017,
+      "loss": 1.4331,
+      "theoretical_loss": 3.4055269027790374,
+      "tokens_seen": 2422145024
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026865671641791044,
+      "loss": 1.2591,
+      "theoretical_loss": 3.4055122337476655,
+      "tokens_seen": 2422276096
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026861659444711925,
+      "loss": 1.3648,
+      "theoretical_loss": 3.4054975657322677,
+      "tokens_seen": 2422407168
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026857647247632805,
+      "loss": 1.4183,
+      "theoretical_loss": 3.4054828987327186,
+      "tokens_seen": 2422538240
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002685363505055368,
+      "loss": 1.4113,
+      "theoretical_loss": 3.4054682327488934,
+      "tokens_seen": 2422669312
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026849622853474566,
+      "loss": 1.3405,
+      "theoretical_loss": 3.405453567780666,
+      "tokens_seen": 2422800384
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002684561065639544,
+      "loss": 1.2877,
+      "theoretical_loss": 3.405438903827912,
+      "tokens_seen": 2422931456
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002684159845931632,
+      "loss": 1.4011,
+      "theoretical_loss": 3.4054242408905053,
+      "tokens_seen": 2423062528
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026837586262237203,
+      "loss": 1.3335,
+      "theoretical_loss": 3.405409578968321,
+      "tokens_seen": 2423193600
+    },
+    {
+      "epoch": 0.47,
+      "objective/train/advantage_avg": 0.4934743046760559,
+      "objective/train/docs_used": 882232,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.305745005607605,
+      "objective/train/original_loss": 1.3057448863983154,
+      "objective/train/theoretical_loss": 3.405394918061234,
+      "objective/train/tokens_used": 793783776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24409565329551697,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505883693695068,
+      "objective/train/weighted_lm_loss": 1.3715765476226807,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.405394918061234,
+      "tokens_seen": 2423324672
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002683357406515808,
+      "loss": 1.364,
+      "theoretical_loss": 3.405394918061234,
+      "tokens_seen": 2423324672
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002682956186807896,
+      "loss": 1.3723,
+      "theoretical_loss": 3.405380258169119,
+      "tokens_seen": 2423455744
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002682554967099984,
+      "loss": 1.4662,
+      "theoretical_loss": 3.4053655992918515,
+      "tokens_seen": 2423586816
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002682153747392072,
+      "loss": 1.4801,
+      "theoretical_loss": 3.4053509414293055,
+      "tokens_seen": 2423717888
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026817525276841595,
+      "loss": 1.4956,
+      "theoretical_loss": 3.405336284581356,
+      "tokens_seen": 2423848960
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002681351307976248,
+      "loss": 1.3401,
+      "theoretical_loss": 3.4053216287478785,
+      "tokens_seen": 2423980032
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026809500882683357,
+      "loss": 1.4039,
+      "theoretical_loss": 3.405306973928747,
+      "tokens_seen": 2424111104
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002680548868560424,
+      "loss": 1.3957,
+      "theoretical_loss": 3.4052923201238374,
+      "tokens_seen": 2424242176
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002680147648852512,
+      "loss": 1.3306,
+      "theoretical_loss": 3.405277667333024,
+      "tokens_seen": 2424373248
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026797464291445993,
+      "loss": 1.3226,
+      "theoretical_loss": 3.4052630155561823,
+      "tokens_seen": 2424504320
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002679345209436688,
+      "loss": 1.3482,
+      "theoretical_loss": 3.405248364793187,
+      "tokens_seen": 2424635392
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026789439897287754,
+      "loss": 1.3057,
+      "theoretical_loss": 3.405233715043913,
+      "tokens_seen": 2424766464
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026785427700208635,
+      "loss": 1.3837,
+      "theoretical_loss": 3.405219066308236,
+      "tokens_seen": 2424897536
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026781415503129516,
+      "loss": 1.4108,
+      "theoretical_loss": 3.4052044185860306,
+      "tokens_seen": 2425028608
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026777403306050396,
+      "loss": 1.3365,
+      "theoretical_loss": 3.405189771877172,
+      "tokens_seen": 2425159680
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002677339110897127,
+      "loss": 1.4534,
+      "theoretical_loss": 3.4051751261815353,
+      "tokens_seen": 2425290752
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002676937891189215,
+      "loss": 1.3155,
+      "theoretical_loss": 3.4051604814989953,
+      "tokens_seen": 2425421824
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002676536671481303,
+      "loss": 1.3454,
+      "theoretical_loss": 3.405145837829428,
+      "tokens_seen": 2425552896
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002676135451773391,
+      "loss": 1.3293,
+      "theoretical_loss": 3.405131195172708,
+      "tokens_seen": 2425683968
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026757342320654794,
+      "loss": 1.4441,
+      "theoretical_loss": 3.405116553528711,
+      "tokens_seen": 2425815040
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002675333012357567,
+      "loss": 1.4051,
+      "theoretical_loss": 3.4051019128973117,
+      "tokens_seen": 2425946112
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026749317926496555,
+      "loss": 1.3896,
+      "theoretical_loss": 3.4050872732783857,
+      "tokens_seen": 2426077184
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002674530572941743,
+      "loss": 1.3457,
+      "theoretical_loss": 3.405072634671808,
+      "tokens_seen": 2426208256
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026741293532338306,
+      "loss": 1.442,
+      "theoretical_loss": 3.405057997077454,
+      "tokens_seen": 2426339328
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002673728133525919,
+      "loss": 1.3549,
+      "theoretical_loss": 3.4050433604951995,
+      "tokens_seen": 2426470400
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026733269138180067,
+      "loss": 1.2644,
+      "theoretical_loss": 3.4050287249249194,
+      "tokens_seen": 2426601472
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002672925694110095,
+      "loss": 1.4224,
+      "theoretical_loss": 3.405014090366489,
+      "tokens_seen": 2426732544
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002672524474402183,
+      "loss": 1.3624,
+      "theoretical_loss": 3.404999456819784,
+      "tokens_seen": 2426863616
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002672123254694271,
+      "loss": 1.2523,
+      "theoretical_loss": 3.4049848242846794,
+      "tokens_seen": 2426994688
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026717220349863584,
+      "loss": 1.3235,
+      "theoretical_loss": 3.404970192761051,
+      "tokens_seen": 2427125760
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026713208152784465,
+      "loss": 1.3515,
+      "theoretical_loss": 3.404955562248775,
+      "tokens_seen": 2427256832
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026709195955705345,
+      "loss": 1.2854,
+      "theoretical_loss": 3.404940932747725,
+      "tokens_seen": 2427387904
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002670518375862622,
+      "loss": 1.4444,
+      "theoretical_loss": 3.404926304257778,
+      "tokens_seen": 2427518976
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026701171561547106,
+      "loss": 1.3065,
+      "theoretical_loss": 3.4049116767788092,
+      "tokens_seen": 2427650048
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002669715936446798,
+      "loss": 1.4587,
+      "theoretical_loss": 3.404897050310694,
+      "tokens_seen": 2427781120
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002669314716738886,
+      "loss": 1.4293,
+      "theoretical_loss": 3.404882424853308,
+      "tokens_seen": 2427912192
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026689134970309743,
+      "loss": 1.3271,
+      "theoretical_loss": 3.4048678004065267,
+      "tokens_seen": 2428043264
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002668512277323062,
+      "loss": 1.4726,
+      "theoretical_loss": 3.404853176970226,
+      "tokens_seen": 2428174336
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026681110576151504,
+      "loss": 1.4288,
+      "theoretical_loss": 3.4048385545442814,
+      "tokens_seen": 2428305408
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002667709837907238,
+      "loss": 1.3388,
+      "theoretical_loss": 3.4048239331285686,
+      "tokens_seen": 2428436480
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002667308618199326,
+      "loss": 1.3463,
+      "theoretical_loss": 3.4048093127229633,
+      "tokens_seen": 2428567552
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002666907398491414,
+      "loss": 1.3452,
+      "theoretical_loss": 3.4047946933273407,
+      "tokens_seen": 2428698624
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002666506178783502,
+      "loss": 1.301,
+      "theoretical_loss": 3.4047800749415775,
+      "tokens_seen": 2428829696
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026661049590755896,
+      "loss": 1.3367,
+      "theoretical_loss": 3.404765457565549,
+      "tokens_seen": 2428960768
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002665703739367678,
+      "loss": 1.3249,
+      "theoretical_loss": 3.4047508411991303,
+      "tokens_seen": 2429091840
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002665302519659766,
+      "loss": 1.4187,
+      "theoretical_loss": 3.404736225842198,
+      "tokens_seen": 2429222912
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026649012999518533,
+      "loss": 1.3348,
+      "theoretical_loss": 3.404721611494628,
+      "tokens_seen": 2429353984
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002664500080243942,
+      "loss": 1.2485,
+      "theoretical_loss": 3.4047069981562954,
+      "tokens_seen": 2429485056
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026640988605360294,
+      "loss": 1.351,
+      "theoretical_loss": 3.4046923858270772,
+      "tokens_seen": 2429616128
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026636976408281175,
+      "loss": 1.401,
+      "theoretical_loss": 3.4046777745068484,
+      "tokens_seen": 2429747200
+    },
+    {
+      "epoch": 0.47,
+      "objective/train/advantage_avg": 0.4913833737373352,
+      "objective/train/docs_used": 884501,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3249716758728027,
+      "objective/train/original_loss": 1.3249716758728027,
+      "objective/train/theoretical_loss": 3.4046631641954845,
+      "objective/train/tokens_used": 800337376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2420549988746643,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503689050674438,
+      "objective/train/weighted_lm_loss": 1.3915722370147705,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.7788007855415344,
+      "theoretical_loss": 3.4046631641954845,
+      "tokens_seen": 2429878272
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026632964211202055,
+      "loss": 1.3436,
+      "theoretical_loss": 3.4046631641954845,
+      "tokens_seen": 2429878272
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026628952014122936,
+      "loss": 1.3805,
+      "theoretical_loss": 3.4046485548928627,
+      "tokens_seen": 2430009344
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002662493981704381,
+      "loss": 1.3698,
+      "theoretical_loss": 3.404633946598858,
+      "tokens_seen": 2430140416
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002662092761996469,
+      "loss": 1.4134,
+      "theoretical_loss": 3.4046193393133466,
+      "tokens_seen": 2430271488
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002661691542288557,
+      "loss": 1.4728,
+      "theoretical_loss": 3.4046047330362046,
+      "tokens_seen": 2430402560
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026612903225806453,
+      "loss": 1.2853,
+      "theoretical_loss": 3.4045901277673076,
+      "tokens_seen": 2430533632
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026608891028727334,
+      "loss": 1.4208,
+      "theoretical_loss": 3.4045755235065327,
+      "tokens_seen": 2430664704
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002660487883164821,
+      "loss": 1.2794,
+      "theoretical_loss": 3.404560920253755,
+      "tokens_seen": 2430795776
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026600866634569095,
+      "loss": 1.3169,
+      "theoretical_loss": 3.4045463180088507,
+      "tokens_seen": 2430926848
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002659685443748997,
+      "loss": 1.3436,
+      "theoretical_loss": 3.4045317167716957,
+      "tokens_seen": 2431057920
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026592842240410846,
+      "loss": 1.394,
+      "theoretical_loss": 3.4045171165421673,
+      "tokens_seen": 2431188992
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002658883004333173,
+      "loss": 1.4003,
+      "theoretical_loss": 3.40450251732014,
+      "tokens_seen": 2431320064
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026584817846252607,
+      "loss": 1.2952,
+      "theoretical_loss": 3.404487919105491,
+      "tokens_seen": 2431451136
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002658080564917349,
+      "loss": 1.3668,
+      "theoretical_loss": 3.4044733218980965,
+      "tokens_seen": 2431582208
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002657679345209437,
+      "loss": 1.4166,
+      "theoretical_loss": 3.4044587256978325,
+      "tokens_seen": 2431713280
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002657278125501525,
+      "loss": 1.2796,
+      "theoretical_loss": 3.404444130504575,
+      "tokens_seen": 2431844352
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026568769057936124,
+      "loss": 1.3683,
+      "theoretical_loss": 3.4044295363182,
+      "tokens_seen": 2431975424
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026564756860857005,
+      "loss": 1.377,
+      "theoretical_loss": 3.404414943138585,
+      "tokens_seen": 2432106496
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026560744663777885,
+      "loss": 1.3491,
+      "theoretical_loss": 3.4044003509656053,
+      "tokens_seen": 2432237568
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026556732466698766,
+      "loss": 1.3259,
+      "theoretical_loss": 3.4043857597991374,
+      "tokens_seen": 2432368640
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026552720269619646,
+      "loss": 1.353,
+      "theoretical_loss": 3.4043711696390577,
+      "tokens_seen": 2432499712
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002654870807254052,
+      "loss": 1.3492,
+      "theoretical_loss": 3.4043565804852425,
+      "tokens_seen": 2432630784
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002654469587546141,
+      "loss": 1.415,
+      "theoretical_loss": 3.404341992337568,
+      "tokens_seen": 2432761856
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026540683678382283,
+      "loss": 1.3411,
+      "theoretical_loss": 3.4043274051959114,
+      "tokens_seen": 2432892928
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002653667148130316,
+      "loss": 1.298,
+      "theoretical_loss": 3.4043128190601486,
+      "tokens_seen": 2433024000
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00026532659284224044,
+      "loss": 1.3259,
+      "theoretical_loss": 3.4042982339301555,
+      "tokens_seen": 2433155072
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002652864708714492,
+      "loss": 1.3674,
+      "theoretical_loss": 3.4042836498058096,
+      "tokens_seen": 2433286144
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.000265246348900658,
+      "loss": 1.3328,
+      "theoretical_loss": 3.404269066686987,
+      "tokens_seen": 2433417216
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002652062269298668,
+      "loss": 1.2638,
+      "theoretical_loss": 3.4042544845735634,
+      "tokens_seen": 2433548288
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002651661049590756,
+      "loss": 1.4273,
+      "theoretical_loss": 3.4042399034654167,
+      "tokens_seen": 2433679360
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026512598298828436,
+      "loss": 1.4483,
+      "theoretical_loss": 3.4042253233624225,
+      "tokens_seen": 2433810432
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026508586101749317,
+      "loss": 1.3237,
+      "theoretical_loss": 3.404210744264458,
+      "tokens_seen": 2433941504
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.000265045739046702,
+      "loss": 1.4228,
+      "theoretical_loss": 3.404196166171399,
+      "tokens_seen": 2434072576
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026500561707591073,
+      "loss": 1.3749,
+      "theoretical_loss": 3.404181589083123,
+      "tokens_seen": 2434203648
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002649654951051196,
+      "loss": 1.2866,
+      "theoretical_loss": 3.4041670129995065,
+      "tokens_seen": 2434334720
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026492537313432834,
+      "loss": 1.405,
+      "theoretical_loss": 3.404152437920426,
+      "tokens_seen": 2434465792
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002648852511635372,
+      "loss": 1.3865,
+      "theoretical_loss": 3.404137863845758,
+      "tokens_seen": 2434596864
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026484512919274595,
+      "loss": 1.3023,
+      "theoretical_loss": 3.404123290775379,
+      "tokens_seen": 2434727936
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026480500722195476,
+      "loss": 1.3374,
+      "theoretical_loss": 3.4041087187091663,
+      "tokens_seen": 2434859008
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026476488525116357,
+      "loss": 1.3046,
+      "theoretical_loss": 3.4040941476469966,
+      "tokens_seen": 2434990080
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002647247632803723,
+      "loss": 1.3732,
+      "theoretical_loss": 3.4040795775887465,
+      "tokens_seen": 2435121152
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002646846413095811,
+      "loss": 1.3854,
+      "theoretical_loss": 3.404065008534293,
+      "tokens_seen": 2435252224
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026464451933878993,
+      "loss": 1.3391,
+      "theoretical_loss": 3.4040504404835126,
+      "tokens_seen": 2435383296
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026460439736799874,
+      "loss": 1.2795,
+      "theoretical_loss": 3.404035873436282,
+      "tokens_seen": 2435514368
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002645642753972075,
+      "loss": 1.3991,
+      "theoretical_loss": 3.4040213073924788,
+      "tokens_seen": 2435645440
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026452415342641635,
+      "loss": 1.3149,
+      "theoretical_loss": 3.4040067423519793,
+      "tokens_seen": 2435776512
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002644840314556251,
+      "loss": 1.3566,
+      "theoretical_loss": 3.4039921783146605,
+      "tokens_seen": 2435907584
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026444390948483385,
+      "loss": 1.2886,
+      "theoretical_loss": 3.4039776152803993,
+      "tokens_seen": 2436038656
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002644037875140427,
+      "loss": 1.4235,
+      "theoretical_loss": 3.4039630532490732,
+      "tokens_seen": 2436169728
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026436366554325147,
+      "loss": 1.4225,
+      "theoretical_loss": 3.4039484922205583,
+      "tokens_seen": 2436300800
+    },
+    {
+      "epoch": 0.48,
+      "objective/train/advantage_avg": 0.4900150001049042,
+      "objective/train/docs_used": 886917,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3052598237991333,
+      "objective/train/original_loss": 1.3052597045898438,
+      "objective/train/theoretical_loss": 3.4039339321947324,
+      "objective/train/tokens_used": 806890976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24133111536502838,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050228238105774,
+      "objective/train/weighted_lm_loss": 1.3706624507904053,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9048373699188232,
+      "theoretical_loss": 3.4039339321947324,
+      "tokens_seen": 2436431872
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026432354357246033,
+      "loss": 1.3466,
+      "theoretical_loss": 3.4039339321947324,
+      "tokens_seen": 2436431872
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002642834216016691,
+      "loss": 1.3529,
+      "theoretical_loss": 3.4039193731714716,
+      "tokens_seen": 2436562944
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002642432996308779,
+      "loss": 1.3438,
+      "theoretical_loss": 3.4039048151506544,
+      "tokens_seen": 2436694016
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002642031776600867,
+      "loss": 1.3691,
+      "theoretical_loss": 3.403890258132156,
+      "tokens_seen": 2436825088
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026416305568929544,
+      "loss": 1.2824,
+      "theoretical_loss": 3.403875702115855,
+      "tokens_seen": 2436956160
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026412293371850425,
+      "loss": 1.3295,
+      "theoretical_loss": 3.403861147101628,
+      "tokens_seen": 2437087232
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026408281174771306,
+      "loss": 1.3407,
+      "theoretical_loss": 3.4038465930893516,
+      "tokens_seen": 2437218304
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026404268977692186,
+      "loss": 1.3454,
+      "theoretical_loss": 3.4038320400789033,
+      "tokens_seen": 2437349376
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002640025678061306,
+      "loss": 1.403,
+      "theoretical_loss": 3.403817488070161,
+      "tokens_seen": 2437480448
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002639624458353395,
+      "loss": 1.4051,
+      "theoretical_loss": 3.4038029370630007,
+      "tokens_seen": 2437611520
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026392232386454823,
+      "loss": 1.3696,
+      "theoretical_loss": 3.4037883870573005,
+      "tokens_seen": 2437742592
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.000263882201893757,
+      "loss": 1.3932,
+      "theoretical_loss": 3.403773838052937,
+      "tokens_seen": 2437873664
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026384207992296584,
+      "loss": 1.3585,
+      "theoretical_loss": 3.4037592900497877,
+      "tokens_seen": 2438004736
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002638019579521746,
+      "loss": 1.4633,
+      "theoretical_loss": 3.4037447430477306,
+      "tokens_seen": 2438135808
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002637618359813834,
+      "loss": 1.3125,
+      "theoretical_loss": 3.4037301970466416,
+      "tokens_seen": 2438266880
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002637217140105922,
+      "loss": 1.3153,
+      "theoretical_loss": 3.403715652046399,
+      "tokens_seen": 2438397952
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.000263681592039801,
+      "loss": 1.3485,
+      "theoretical_loss": 3.4037011080468806,
+      "tokens_seen": 2438529024
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002636414700690098,
+      "loss": 1.3566,
+      "theoretical_loss": 3.403686565047962,
+      "tokens_seen": 2438660096
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026360134809821857,
+      "loss": 1.3417,
+      "theoretical_loss": 3.4036720230495225,
+      "tokens_seen": 2438791168
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002635612261274274,
+      "loss": 1.3579,
+      "theoretical_loss": 3.4036574820514387,
+      "tokens_seen": 2438922240
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002635211041566362,
+      "loss": 1.35,
+      "theoretical_loss": 3.403642942053587,
+      "tokens_seen": 2439053312
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.000263480982185845,
+      "loss": 1.3687,
+      "theoretical_loss": 3.403628403055847,
+      "tokens_seen": 2439184384
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026344086021505374,
+      "loss": 1.3496,
+      "theoretical_loss": 3.4036138650580945,
+      "tokens_seen": 2439315456
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002634007382442626,
+      "loss": 1.3401,
+      "theoretical_loss": 3.4035993280602073,
+      "tokens_seen": 2439446528
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026336061627347135,
+      "loss": 1.3332,
+      "theoretical_loss": 3.4035847920620634,
+      "tokens_seen": 2439577600
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026332049430268016,
+      "loss": 1.4092,
+      "theoretical_loss": 3.4035702570635404,
+      "tokens_seen": 2439708672
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026328037233188897,
+      "loss": 1.3657,
+      "theoretical_loss": 3.403555723064515,
+      "tokens_seen": 2439839744
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002632402503610977,
+      "loss": 1.3365,
+      "theoretical_loss": 3.4035411900648653,
+      "tokens_seen": 2439970816
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002632001283903065,
+      "loss": 1.3595,
+      "theoretical_loss": 3.4035266580644694,
+      "tokens_seen": 2440101888
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026316000641951533,
+      "loss": 1.3851,
+      "theoretical_loss": 3.403512127063204,
+      "tokens_seen": 2440232960
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026311988444872414,
+      "loss": 1.283,
+      "theoretical_loss": 3.403497597060947,
+      "tokens_seen": 2440364032
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002630797624779329,
+      "loss": 1.3685,
+      "theoretical_loss": 3.403483068057577,
+      "tokens_seen": 2440495104
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026303964050714175,
+      "loss": 1.37,
+      "theoretical_loss": 3.4034685400529705,
+      "tokens_seen": 2440626176
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002629995185363505,
+      "loss": 1.3716,
+      "theoretical_loss": 3.4034540130470057,
+      "tokens_seen": 2440757248
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002629593965655593,
+      "loss": 1.3514,
+      "theoretical_loss": 3.4034394870395603,
+      "tokens_seen": 2440888320
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002629192745947681,
+      "loss": 1.3565,
+      "theoretical_loss": 3.403424962030512,
+      "tokens_seen": 2441019392
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026287915262397687,
+      "loss": 1.3962,
+      "theoretical_loss": 3.4034104380197383,
+      "tokens_seen": 2441150464
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026283903065318573,
+      "loss": 1.3304,
+      "theoretical_loss": 3.403395915007118,
+      "tokens_seen": 2441281536
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002627989086823945,
+      "loss": 1.3076,
+      "theoretical_loss": 3.4033813929925274,
+      "tokens_seen": 2441412608
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002627587867116033,
+      "loss": 1.4259,
+      "theoretical_loss": 3.4033668719758454,
+      "tokens_seen": 2441543680
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002627186647408121,
+      "loss": 1.4407,
+      "theoretical_loss": 3.4033523519569497,
+      "tokens_seen": 2441674752
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026267854277002084,
+      "loss": 1.3148,
+      "theoretical_loss": 3.4033378329357182,
+      "tokens_seen": 2441805824
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026263842079922965,
+      "loss": 1.3851,
+      "theoretical_loss": 3.4033233149120288,
+      "tokens_seen": 2441936896
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026259829882843846,
+      "loss": 1.4299,
+      "theoretical_loss": 3.4033087978857592,
+      "tokens_seen": 2442067968
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026255817685764726,
+      "loss": 1.2889,
+      "theoretical_loss": 3.4032942818567875,
+      "tokens_seen": 2442199040
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.000262518054886856,
+      "loss": 1.3559,
+      "theoretical_loss": 3.4032797668249914,
+      "tokens_seen": 2442330112
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002624779329160649,
+      "loss": 1.3916,
+      "theoretical_loss": 3.4032652527902494,
+      "tokens_seen": 2442461184
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026243781094527363,
+      "loss": 1.3671,
+      "theoretical_loss": 3.403250739752439,
+      "tokens_seen": 2442592256
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002623976889744824,
+      "loss": 1.2952,
+      "theoretical_loss": 3.403236227711439,
+      "tokens_seen": 2442723328
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026235756700369124,
+      "loss": 1.3865,
+      "theoretical_loss": 3.403221716667127,
+      "tokens_seen": 2442854400
+    },
+    {
+      "epoch": 0.48,
+      "objective/train/advantage_avg": 0.49215129017829895,
+      "objective/train/docs_used": 889343,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2749754190444946,
+      "objective/train/original_loss": 1.274975299835205,
+      "objective/train/theoretical_loss": 3.4032072066193804,
+      "objective/train/tokens_used": 813444576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24250178039073944,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050447940826416,
+      "objective/train/weighted_lm_loss": 1.3394321203231812,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4032072066193804,
+      "tokens_seen": 2442985472
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002623174450329,
+      "loss": 1.2867,
+      "theoretical_loss": 3.4032072066193804,
+      "tokens_seen": 2442985472
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026227732306210885,
+      "loss": 1.4135,
+      "theoretical_loss": 3.4031926975680777,
+      "tokens_seen": 2443116544
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002622372010913176,
+      "loss": 1.3528,
+      "theoretical_loss": 3.403178189513098,
+      "tokens_seen": 2443247616
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002621970791205264,
+      "loss": 1.3179,
+      "theoretical_loss": 3.403163682454318,
+      "tokens_seen": 2443378688
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002621569571497352,
+      "loss": 1.4434,
+      "theoretical_loss": 3.403149176391617,
+      "tokens_seen": 2443509760
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026211683517894397,
+      "loss": 1.4214,
+      "theoretical_loss": 3.4031346713248727,
+      "tokens_seen": 2443640832
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002620767132081528,
+      "loss": 1.3824,
+      "theoretical_loss": 3.403120167253963,
+      "tokens_seen": 2443771904
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002620365912373616,
+      "loss": 1.3671,
+      "theoretical_loss": 3.4031056641787663,
+      "tokens_seen": 2443902976
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002619964692665704,
+      "loss": 1.3547,
+      "theoretical_loss": 3.4030911620991615,
+      "tokens_seen": 2444034048
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026195634729577914,
+      "loss": 1.3015,
+      "theoretical_loss": 3.4030766610150263,
+      "tokens_seen": 2444165120
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.000261916225324988,
+      "loss": 1.3656,
+      "theoretical_loss": 3.4030621609262384,
+      "tokens_seen": 2444296192
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026187610335419675,
+      "loss": 1.3673,
+      "theoretical_loss": 3.4030476618326775,
+      "tokens_seen": 2444427264
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026183598138340556,
+      "loss": 1.3288,
+      "theoretical_loss": 3.4030331637342206,
+      "tokens_seen": 2444558336
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026179585941261437,
+      "loss": 1.4097,
+      "theoretical_loss": 3.403018666630747,
+      "tokens_seen": 2444689408
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002617557374418231,
+      "loss": 1.3422,
+      "theoretical_loss": 3.4030041705221343,
+      "tokens_seen": 2444820480
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.000261715615471032,
+      "loss": 1.3162,
+      "theoretical_loss": 3.4029896754082616,
+      "tokens_seen": 2444951552
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026167549350024073,
+      "loss": 1.348,
+      "theoretical_loss": 3.4029751812890074,
+      "tokens_seen": 2445082624
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026163537152944954,
+      "loss": 1.4252,
+      "theoretical_loss": 3.4029606881642493,
+      "tokens_seen": 2445213696
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026159524955865834,
+      "loss": 1.3873,
+      "theoretical_loss": 3.4029461960338665,
+      "tokens_seen": 2445344768
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026155512758786715,
+      "loss": 1.4288,
+      "theoretical_loss": 3.4029317048977368,
+      "tokens_seen": 2445475840
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002615150056170759,
+      "loss": 1.3504,
+      "theoretical_loss": 3.4029172147557394,
+      "tokens_seen": 2445606912
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002614748836462847,
+      "loss": 1.3061,
+      "theoretical_loss": 3.4029027256077526,
+      "tokens_seen": 2445737984
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002614347616754935,
+      "loss": 1.3827,
+      "theoretical_loss": 3.4028882374536544,
+      "tokens_seen": 2445869056
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026139463970470227,
+      "loss": 1.3524,
+      "theoretical_loss": 3.4028737502933244,
+      "tokens_seen": 2446000128
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026135451773391113,
+      "loss": 1.3082,
+      "theoretical_loss": 3.4028592641266404,
+      "tokens_seen": 2446131200
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002613143957631199,
+      "loss": 1.3383,
+      "theoretical_loss": 3.4028447789534813,
+      "tokens_seen": 2446262272
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002612742737923287,
+      "loss": 1.3855,
+      "theoretical_loss": 3.4028302947737257,
+      "tokens_seen": 2446393344
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002612341518215375,
+      "loss": 1.3901,
+      "theoretical_loss": 3.402815811587252,
+      "tokens_seen": 2446524416
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026119402985074624,
+      "loss": 1.2735,
+      "theoretical_loss": 3.4028013293939394,
+      "tokens_seen": 2446655488
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026115390787995505,
+      "loss": 1.3941,
+      "theoretical_loss": 3.4027868481936663,
+      "tokens_seen": 2446786560
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026111378590916386,
+      "loss": 1.3278,
+      "theoretical_loss": 3.4027723679863113,
+      "tokens_seen": 2446917632
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026107366393837266,
+      "loss": 1.3576,
+      "theoretical_loss": 3.4027578887717533,
+      "tokens_seen": 2447048704
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026103354196758147,
+      "loss": 1.3078,
+      "theoretical_loss": 3.402743410549871,
+      "tokens_seen": 2447179776
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002609934199967903,
+      "loss": 1.3247,
+      "theoretical_loss": 3.4027289333205433,
+      "tokens_seen": 2447310848
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026095329802599903,
+      "loss": 1.2909,
+      "theoretical_loss": 3.4027144570836487,
+      "tokens_seen": 2447441920
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026091317605520783,
+      "loss": 1.3118,
+      "theoretical_loss": 3.4026999818390666,
+      "tokens_seen": 2447572992
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026087305408441664,
+      "loss": 1.3915,
+      "theoretical_loss": 3.4026855075866753,
+      "tokens_seen": 2447704064
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002608329321136254,
+      "loss": 1.3737,
+      "theoretical_loss": 3.402671034326354,
+      "tokens_seen": 2447835136
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026079281014283425,
+      "loss": 1.3587,
+      "theoretical_loss": 3.402656562057981,
+      "tokens_seen": 2447966208
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.000260752688172043,
+      "loss": 1.318,
+      "theoretical_loss": 3.402642090781436,
+      "tokens_seen": 2448097280
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002607125662012518,
+      "loss": 1.3343,
+      "theoretical_loss": 3.402627620496597,
+      "tokens_seen": 2448228352
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002606724442304606,
+      "loss": 1.4047,
+      "theoretical_loss": 3.4026131512033437,
+      "tokens_seen": 2448359424
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026063232225966937,
+      "loss": 1.3685,
+      "theoretical_loss": 3.402598682901555,
+      "tokens_seen": 2448490496
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002605922002888782,
+      "loss": 1.4233,
+      "theoretical_loss": 3.4025842155911104,
+      "tokens_seen": 2448621568
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.000260552078318087,
+      "loss": 1.3529,
+      "theoretical_loss": 3.4025697492718874,
+      "tokens_seen": 2448752640
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002605119563472958,
+      "loss": 1.2902,
+      "theoretical_loss": 3.4025552839437663,
+      "tokens_seen": 2448883712
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002604718343765046,
+      "loss": 1.3761,
+      "theoretical_loss": 3.402540819606626,
+      "tokens_seen": 2449014784
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002604317124057134,
+      "loss": 1.3372,
+      "theoretical_loss": 3.4025263562603447,
+      "tokens_seen": 2449145856
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026039159043492215,
+      "loss": 1.4012,
+      "theoretical_loss": 3.4025118939048027,
+      "tokens_seen": 2449276928
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026035146846413096,
+      "loss": 1.2874,
+      "theoretical_loss": 3.402497432539879,
+      "tokens_seen": 2449408000
+    },
+    {
+      "epoch": 0.48,
+      "objective/train/advantage_avg": 0.49413785338401794,
+      "objective/train/docs_used": 891543,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2185678482055664,
+      "objective/train/original_loss": 1.2185678482055664,
+      "objective/train/theoretical_loss": 3.4024829721654513,
+      "objective/train/tokens_used": 819998176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24446628987789154,
+      "objective/train/value_max": -0.5,
+      "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": 1.2802948951721191,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4024829721654513,
+      "tokens_seen": 2449539072
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026031134649333977,
+      "loss": 1.2402,
+      "theoretical_loss": 3.4024829721654513,
+      "tokens_seen": 2449539072
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002602712245225485,
+      "loss": 1.2682,
+      "theoretical_loss": 3.4024685127814003,
+      "tokens_seen": 2449670144
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002602311025517574,
+      "loss": 1.3047,
+      "theoretical_loss": 3.4024540543876047,
+      "tokens_seen": 2449801216
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026019098058096613,
+      "loss": 1.2681,
+      "theoretical_loss": 3.4024395969839434,
+      "tokens_seen": 2449932288
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026015085861017494,
+      "loss": 1.2501,
+      "theoretical_loss": 3.4024251405702963,
+      "tokens_seen": 2450063360
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00026011073663938374,
+      "loss": 1.299,
+      "theoretical_loss": 3.4024106851465423,
+      "tokens_seen": 2450194432
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00026007061466859255,
+      "loss": 1.235,
+      "theoretical_loss": 3.4023962307125606,
+      "tokens_seen": 2450325504
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002600304926978013,
+      "loss": 1.2269,
+      "theoretical_loss": 3.40238177726823,
+      "tokens_seen": 2450456576
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002599903707270101,
+      "loss": 1.3725,
+      "theoretical_loss": 3.4023673248134307,
+      "tokens_seen": 2450587648
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002599502487562189,
+      "loss": 1.3414,
+      "theoretical_loss": 3.4023528733480415,
+      "tokens_seen": 2450718720
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025991012678542767,
+      "loss": 1.2975,
+      "theoretical_loss": 3.402338422871942,
+      "tokens_seen": 2450849792
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002598700048146365,
+      "loss": 1.3277,
+      "theoretical_loss": 3.4023239733850117,
+      "tokens_seen": 2450980864
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002598298828438453,
+      "loss": 1.3015,
+      "theoretical_loss": 3.40230952488713,
+      "tokens_seen": 2451111936
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025978976087305414,
+      "loss": 1.3207,
+      "theoretical_loss": 3.4022950773781755,
+      "tokens_seen": 2451243008
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002597496389022629,
+      "loss": 1.2419,
+      "theoretical_loss": 3.4022806308580282,
+      "tokens_seen": 2451374080
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025970951693147164,
+      "loss": 1.3199,
+      "theoretical_loss": 3.4022661853265683,
+      "tokens_seen": 2451505152
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002596693949606805,
+      "loss": 1.2878,
+      "theoretical_loss": 3.402251740783674,
+      "tokens_seen": 2451636224
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025962927298988926,
+      "loss": 1.3697,
+      "theoretical_loss": 3.4022372972292256,
+      "tokens_seen": 2451767296
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025958915101909806,
+      "loss": 1.233,
+      "theoretical_loss": 3.402222854663102,
+      "tokens_seen": 2451898368
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025954902904830687,
+      "loss": 1.29,
+      "theoretical_loss": 3.4022084130851837,
+      "tokens_seen": 2452029440
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002595089070775157,
+      "loss": 1.2582,
+      "theoretical_loss": 3.4021939724953496,
+      "tokens_seen": 2452160512
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025946878510672443,
+      "loss": 1.2699,
+      "theoretical_loss": 3.4021795328934794,
+      "tokens_seen": 2452291584
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025942866313593323,
+      "loss": 1.3177,
+      "theoretical_loss": 3.4021650942794524,
+      "tokens_seen": 2452422656
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025938854116514204,
+      "loss": 1.3563,
+      "theoretical_loss": 3.402150656653149,
+      "tokens_seen": 2452553728
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002593484191943508,
+      "loss": 1.2605,
+      "theoretical_loss": 3.402136220014448,
+      "tokens_seen": 2452684800
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025930829722355965,
+      "loss": 1.3362,
+      "theoretical_loss": 3.4021217843632297,
+      "tokens_seen": 2452815872
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002592681752527684,
+      "loss": 1.3089,
+      "theoretical_loss": 3.4021073496993735,
+      "tokens_seen": 2452946944
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025922805328197727,
+      "loss": 1.3253,
+      "theoretical_loss": 3.402092916022759,
+      "tokens_seen": 2453078016
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.000259187931311186,
+      "loss": 1.3905,
+      "theoretical_loss": 3.4020784833332662,
+      "tokens_seen": 2453209088
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025914780934039477,
+      "loss": 1.3015,
+      "theoretical_loss": 3.402064051630775,
+      "tokens_seen": 2453340160
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025910768736960363,
+      "loss": 1.2753,
+      "theoretical_loss": 3.4020496209151645,
+      "tokens_seen": 2453471232
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002590675653988124,
+      "loss": 1.2764,
+      "theoretical_loss": 3.4020351911863154,
+      "tokens_seen": 2453602304
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002590274434280212,
+      "loss": 1.3802,
+      "theoretical_loss": 3.4020207624441063,
+      "tokens_seen": 2453733376
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025898732145723,
+      "loss": 1.2453,
+      "theoretical_loss": 3.4020063346884184,
+      "tokens_seen": 2453864448
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002589471994864388,
+      "loss": 1.2928,
+      "theoretical_loss": 3.401991907919131,
+      "tokens_seen": 2453995520
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025890707751564755,
+      "loss": 1.3055,
+      "theoretical_loss": 3.401977482136124,
+      "tokens_seen": 2454126592
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025886695554485636,
+      "loss": 1.3996,
+      "theoretical_loss": 3.4019630573392767,
+      "tokens_seen": 2454257664
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025882683357406517,
+      "loss": 1.2656,
+      "theoretical_loss": 3.40194863352847,
+      "tokens_seen": 2454388736
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002587867116032739,
+      "loss": 1.2871,
+      "theoretical_loss": 3.4019342107035833,
+      "tokens_seen": 2454519808
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002587465896324828,
+      "loss": 1.3561,
+      "theoretical_loss": 3.4019197888644968,
+      "tokens_seen": 2454650880
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025870646766169153,
+      "loss": 1.3522,
+      "theoretical_loss": 3.40190536801109,
+      "tokens_seen": 2454781952
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025866634569090034,
+      "loss": 1.3143,
+      "theoretical_loss": 3.4018909481432433,
+      "tokens_seen": 2454913024
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025862622372010914,
+      "loss": 1.2815,
+      "theoretical_loss": 3.401876529260837,
+      "tokens_seen": 2455044096
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025858610174931795,
+      "loss": 1.3811,
+      "theoretical_loss": 3.4018621113637506,
+      "tokens_seen": 2455175168
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025854597977852676,
+      "loss": 1.345,
+      "theoretical_loss": 3.4018476944518645,
+      "tokens_seen": 2455306240
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002585058578077355,
+      "loss": 1.3369,
+      "theoretical_loss": 3.401833278525059,
+      "tokens_seen": 2455437312
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002584657358369443,
+      "loss": 1.3454,
+      "theoretical_loss": 3.4018188635832134,
+      "tokens_seen": 2455568384
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002584256138661531,
+      "loss": 1.3096,
+      "theoretical_loss": 3.4018044496262085,
+      "tokens_seen": 2455699456
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002583854918953619,
+      "loss": 1.2863,
+      "theoretical_loss": 3.4017900366539244,
+      "tokens_seen": 2455830528
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002583453699245707,
+      "loss": 1.2827,
+      "theoretical_loss": 3.401775624666241,
+      "tokens_seen": 2455961600
+    },
+    {
+      "epoch": 0.49,
+      "objective/train/advantage_avg": 0.49538013339042664,
+      "objective/train/docs_used": 893908,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.1229428052902222,
+      "objective/train/original_loss": 1.1229428052902222,
+      "objective/train/theoretical_loss": 3.401761213663039,
+      "objective/train/tokens_used": 826551776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24557486176490784,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0507864952087402,
+      "objective/train/weighted_lm_loss": 1.179932951927185,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.401761213663039,
+      "tokens_seen": 2456092672
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025830524795377954,
+      "loss": 1.2334,
+      "theoretical_loss": 3.401761213663039,
+      "tokens_seen": 2456092672
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002582651259829883,
+      "loss": 1.2823,
+      "theoretical_loss": 3.401746803644198,
+      "tokens_seen": 2456223744
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025822500401219704,
+      "loss": 1.2695,
+      "theoretical_loss": 3.4017323946095988,
+      "tokens_seen": 2456354816
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002581848820414059,
+      "loss": 1.3382,
+      "theoretical_loss": 3.401717986559121,
+      "tokens_seen": 2456485888
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025814476007061466,
+      "loss": 1.3693,
+      "theoretical_loss": 3.4017035794926453,
+      "tokens_seen": 2456616960
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025810463809982346,
+      "loss": 1.2785,
+      "theoretical_loss": 3.4016891734100523,
+      "tokens_seen": 2456748032
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025806451612903227,
+      "loss": 1.2945,
+      "theoretical_loss": 3.401674768311222,
+      "tokens_seen": 2456879104
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002580243941582411,
+      "loss": 1.3303,
+      "theoretical_loss": 3.401660364196035,
+      "tokens_seen": 2457010176
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002579842721874498,
+      "loss": 1.2673,
+      "theoretical_loss": 3.4016459610643706,
+      "tokens_seen": 2457141248
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025794415021665863,
+      "loss": 1.3501,
+      "theoretical_loss": 3.4016315589161104,
+      "tokens_seen": 2457272320
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025790402824586744,
+      "loss": 1.3376,
+      "theoretical_loss": 3.4016171577511343,
+      "tokens_seen": 2457403392
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025786390627507625,
+      "loss": 1.4024,
+      "theoretical_loss": 3.4016027575693233,
+      "tokens_seen": 2457534464
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025782378430428505,
+      "loss": 1.3338,
+      "theoretical_loss": 3.4015883583705566,
+      "tokens_seen": 2457665536
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002577836623334938,
+      "loss": 1.2815,
+      "theoretical_loss": 3.401573960154716,
+      "tokens_seen": 2457796608
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025774354036270266,
+      "loss": 1.2962,
+      "theoretical_loss": 3.4015595629216815,
+      "tokens_seen": 2457927680
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002577034183919114,
+      "loss": 1.2798,
+      "theoretical_loss": 3.401545166671333,
+      "tokens_seen": 2458058752
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025766329642112017,
+      "loss": 1.2948,
+      "theoretical_loss": 3.401530771403552,
+      "tokens_seen": 2458189824
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025762317445032903,
+      "loss": 1.3121,
+      "theoretical_loss": 3.4015163771182184,
+      "tokens_seen": 2458320896
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002575830524795378,
+      "loss": 1.2919,
+      "theoretical_loss": 3.4015019838152134,
+      "tokens_seen": 2458451968
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002575429305087466,
+      "loss": 1.315,
+      "theoretical_loss": 3.401487591494417,
+      "tokens_seen": 2458583040
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002575028085379554,
+      "loss": 1.288,
+      "theoretical_loss": 3.40147320015571,
+      "tokens_seen": 2458714112
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002574626865671642,
+      "loss": 1.2947,
+      "theoretical_loss": 3.4014588097989726,
+      "tokens_seen": 2458845184
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025742256459637295,
+      "loss": 1.3208,
+      "theoretical_loss": 3.4014444204240863,
+      "tokens_seen": 2458976256
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025738244262558176,
+      "loss": 1.3455,
+      "theoretical_loss": 3.401430032030931,
+      "tokens_seen": 2459107328
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025734232065479057,
+      "loss": 1.3511,
+      "theoretical_loss": 3.4014156446193886,
+      "tokens_seen": 2459238400
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002573021986839993,
+      "loss": 1.2066,
+      "theoretical_loss": 3.4014012581893383,
+      "tokens_seen": 2459369472
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002572620767132082,
+      "loss": 1.3113,
+      "theoretical_loss": 3.401386872740662,
+      "tokens_seen": 2459500544
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025722195474241693,
+      "loss": 1.29,
+      "theoretical_loss": 3.40137248827324,
+      "tokens_seen": 2459631616
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002571818327716258,
+      "loss": 1.2456,
+      "theoretical_loss": 3.4013581047869526,
+      "tokens_seen": 2459762688
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025714171080083454,
+      "loss": 1.2869,
+      "theoretical_loss": 3.4013437222816814,
+      "tokens_seen": 2459893760
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025710158883004335,
+      "loss": 1.3024,
+      "theoretical_loss": 3.401329340757307,
+      "tokens_seen": 2460024832
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025706146685925216,
+      "loss": 1.3753,
+      "theoretical_loss": 3.4013149602137096,
+      "tokens_seen": 2460155904
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002570213448884609,
+      "loss": 1.2239,
+      "theoretical_loss": 3.401300580650771,
+      "tokens_seen": 2460286976
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002569812229176697,
+      "loss": 1.2395,
+      "theoretical_loss": 3.4012862020683716,
+      "tokens_seen": 2460418048
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002569411009468785,
+      "loss": 1.3492,
+      "theoretical_loss": 3.4012718244663924,
+      "tokens_seen": 2460549120
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002569009789760873,
+      "loss": 1.2598,
+      "theoretical_loss": 3.401257447844715,
+      "tokens_seen": 2460680192
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002568608570052961,
+      "loss": 1.3143,
+      "theoretical_loss": 3.401243072203219,
+      "tokens_seen": 2460811264
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025682073503450494,
+      "loss": 1.2981,
+      "theoretical_loss": 3.4012286975417863,
+      "tokens_seen": 2460942336
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002567806130637137,
+      "loss": 1.3593,
+      "theoretical_loss": 3.4012143238602977,
+      "tokens_seen": 2461073408
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025674049109292244,
+      "loss": 1.3816,
+      "theoretical_loss": 3.4011999511586337,
+      "tokens_seen": 2461204480
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002567003691221313,
+      "loss": 1.2819,
+      "theoretical_loss": 3.4011855794366763,
+      "tokens_seen": 2461335552
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025666024715134006,
+      "loss": 1.2592,
+      "theoretical_loss": 3.4011712086943056,
+      "tokens_seen": 2461466624
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002566201251805489,
+      "loss": 1.3335,
+      "theoretical_loss": 3.4011568389314037,
+      "tokens_seen": 2461597696
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025658000320975767,
+      "loss": 1.3146,
+      "theoretical_loss": 3.4011424701478505,
+      "tokens_seen": 2461728768
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002565398812389665,
+      "loss": 1.2698,
+      "theoretical_loss": 3.4011281023435282,
+      "tokens_seen": 2461859840
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002564997592681753,
+      "loss": 1.3021,
+      "theoretical_loss": 3.401113735518317,
+      "tokens_seen": 2461990912
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025645963729738403,
+      "loss": 1.3387,
+      "theoretical_loss": 3.401099369672099,
+      "tokens_seen": 2462121984
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025641951532659284,
+      "loss": 1.2834,
+      "theoretical_loss": 3.4010850048047545,
+      "tokens_seen": 2462253056
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025637939335580165,
+      "loss": 1.3256,
+      "theoretical_loss": 3.4010706409161653,
+      "tokens_seen": 2462384128
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025633927138501045,
+      "loss": 1.3506,
+      "theoretical_loss": 3.4010562780062124,
+      "tokens_seen": 2462515200
+    },
+    {
+      "epoch": 0.49,
+      "objective/train/advantage_avg": 0.4915258288383484,
+      "objective/train/docs_used": 896263,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3069236278533936,
+      "objective/train/original_loss": 1.3069236278533936,
+      "objective/train/theoretical_loss": 3.401041916074777,
+      "objective/train/tokens_used": 833105376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2423502653837204,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503846406936646,
+      "objective/train/weighted_lm_loss": 1.3727760314941406,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.401041916074777,
+      "tokens_seen": 2462646272
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002562991494142192,
+      "loss": 1.3004,
+      "theoretical_loss": 3.401041916074777,
+      "tokens_seen": 2462646272
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025625902744342806,
+      "loss": 1.2332,
+      "theoretical_loss": 3.4010275551217406,
+      "tokens_seen": 2462777344
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002562189054726368,
+      "loss": 1.3507,
+      "theoretical_loss": 3.401013195146984,
+      "tokens_seen": 2462908416
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025617878350184557,
+      "loss": 1.1911,
+      "theoretical_loss": 3.4009988361503893,
+      "tokens_seen": 2463039488
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025613866153105443,
+      "loss": 1.3973,
+      "theoretical_loss": 3.4009844781318366,
+      "tokens_seen": 2463170560
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002560985395602632,
+      "loss": 1.32,
+      "theoretical_loss": 3.400970121091208,
+      "tokens_seen": 2463301632
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.000256058417589472,
+      "loss": 1.2613,
+      "theoretical_loss": 3.4009557650283857,
+      "tokens_seen": 2463432704
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002560182956186808,
+      "loss": 1.2653,
+      "theoretical_loss": 3.4009414099432496,
+      "tokens_seen": 2463563776
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002559781736478896,
+      "loss": 1.3545,
+      "theoretical_loss": 3.4009270558356817,
+      "tokens_seen": 2463694848
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002559380516770984,
+      "loss": 1.2606,
+      "theoretical_loss": 3.4009127027055635,
+      "tokens_seen": 2463825920
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025589792970630716,
+      "loss": 1.2452,
+      "theoretical_loss": 3.4008983505527763,
+      "tokens_seen": 2463956992
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025585780773551597,
+      "loss": 1.3662,
+      "theoretical_loss": 3.4008839993772018,
+      "tokens_seen": 2464088064
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025581768576472477,
+      "loss": 1.2886,
+      "theoretical_loss": 3.400869649178721,
+      "tokens_seen": 2464219136
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002557775637939336,
+      "loss": 1.2842,
+      "theoretical_loss": 3.4008552999572164,
+      "tokens_seen": 2464350208
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025573744182314233,
+      "loss": 1.3059,
+      "theoretical_loss": 3.4008409517125684,
+      "tokens_seen": 2464481280
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002556973198523512,
+      "loss": 1.2997,
+      "theoretical_loss": 3.400826604444659,
+      "tokens_seen": 2464612352
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025565719788155994,
+      "loss": 1.28,
+      "theoretical_loss": 3.40081225815337,
+      "tokens_seen": 2464743424
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002556170759107687,
+      "loss": 1.2612,
+      "theoretical_loss": 3.400797912838583,
+      "tokens_seen": 2464874496
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025557695393997755,
+      "loss": 1.3404,
+      "theoretical_loss": 3.400783568500179,
+      "tokens_seen": 2465005568
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002555368319691863,
+      "loss": 1.2712,
+      "theoretical_loss": 3.40076922513804,
+      "tokens_seen": 2465136640
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002554967099983951,
+      "loss": 1.3545,
+      "theoretical_loss": 3.400754882752048,
+      "tokens_seen": 2465267712
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002554565880276039,
+      "loss": 1.3497,
+      "theoretical_loss": 3.400740541342084,
+      "tokens_seen": 2465398784
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002554164660568127,
+      "loss": 1.233,
+      "theoretical_loss": 3.4007262009080303,
+      "tokens_seen": 2465529856
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025537634408602153,
+      "loss": 1.3466,
+      "theoretical_loss": 3.4007118614497682,
+      "tokens_seen": 2465660928
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025533622211523034,
+      "loss": 1.3358,
+      "theoretical_loss": 3.40069752296718,
+      "tokens_seen": 2465792000
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002552961001444391,
+      "loss": 1.3139,
+      "theoretical_loss": 3.4006831854601467,
+      "tokens_seen": 2465923072
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002552559781736479,
+      "loss": 1.3312,
+      "theoretical_loss": 3.4006688489285506,
+      "tokens_seen": 2466054144
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002552158562028567,
+      "loss": 1.3178,
+      "theoretical_loss": 3.400654513372273,
+      "tokens_seen": 2466185216
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025517573423206546,
+      "loss": 1.4073,
+      "theoretical_loss": 3.4006401787911966,
+      "tokens_seen": 2466316288
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002551356122612743,
+      "loss": 1.243,
+      "theoretical_loss": 3.4006258451852025,
+      "tokens_seen": 2466447360
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00025509549029048307,
+      "loss": 1.3079,
+      "theoretical_loss": 3.4006115125541725,
+      "tokens_seen": 2466578432
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002550553683196919,
+      "loss": 1.2491,
+      "theoretical_loss": 3.400597180897989,
+      "tokens_seen": 2466709504
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002550152463489007,
+      "loss": 1.2755,
+      "theoretical_loss": 3.4005828502165336,
+      "tokens_seen": 2466840576
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025497512437810943,
+      "loss": 1.4093,
+      "theoretical_loss": 3.400568520509688,
+      "tokens_seen": 2466971648
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025493500240731824,
+      "loss": 1.3266,
+      "theoretical_loss": 3.4005541917773345,
+      "tokens_seen": 2467102720
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025489488043652705,
+      "loss": 1.3326,
+      "theoretical_loss": 3.400539864019355,
+      "tokens_seen": 2467233792
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025485475846573585,
+      "loss": 1.2961,
+      "theoretical_loss": 3.4005255372356316,
+      "tokens_seen": 2467364864
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002548146364949446,
+      "loss": 1.2215,
+      "theoretical_loss": 3.4005112114260463,
+      "tokens_seen": 2467495936
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025477451452415346,
+      "loss": 1.3591,
+      "theoretical_loss": 3.4004968865904805,
+      "tokens_seen": 2467627008
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002547343925533622,
+      "loss": 1.2904,
+      "theoretical_loss": 3.4004825627288167,
+      "tokens_seen": 2467758080
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.000254694270582571,
+      "loss": 1.3193,
+      "theoretical_loss": 3.4004682398409374,
+      "tokens_seen": 2467889152
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025465414861177983,
+      "loss": 1.4058,
+      "theoretical_loss": 3.400453917926724,
+      "tokens_seen": 2468020224
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002546140266409886,
+      "loss": 1.3014,
+      "theoretical_loss": 3.400439596986059,
+      "tokens_seen": 2468151296
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025457390467019744,
+      "loss": 1.308,
+      "theoretical_loss": 3.400425277018824,
+      "tokens_seen": 2468282368
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002545337826994062,
+      "loss": 1.3659,
+      "theoretical_loss": 3.400410958024902,
+      "tokens_seen": 2468413440
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.000254493660728615,
+      "loss": 1.3145,
+      "theoretical_loss": 3.4003966400041747,
+      "tokens_seen": 2468544512
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002544535387578238,
+      "loss": 1.3335,
+      "theoretical_loss": 3.400382322956524,
+      "tokens_seen": 2468675584
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025441341678703256,
+      "loss": 1.3633,
+      "theoretical_loss": 3.4003680068818327,
+      "tokens_seen": 2468806656
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025437329481624136,
+      "loss": 1.3352,
+      "theoretical_loss": 3.4003536917799826,
+      "tokens_seen": 2468937728
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025433317284545017,
+      "loss": 1.2485,
+      "theoretical_loss": 3.400339377650856,
+      "tokens_seen": 2469068800
+    },
+    {
+      "epoch": 0.5,
+      "objective/train/advantage_avg": 0.49276605248451233,
+      "objective/train/docs_used": 898257,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.237341284751892,
+      "objective/train/original_loss": 1.237341284751892,
+      "objective/train/theoretical_loss": 3.4003250644943352,
+      "objective/train/tokens_used": 839658976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24350085854530334,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050514578819275,
+      "objective/train/weighted_lm_loss": 1.2997381687164307,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.4003250644943352,
+      "tokens_seen": 2469199872
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.000254293050874659,
+      "loss": 1.3595,
+      "theoretical_loss": 3.4003250644943352,
+      "tokens_seen": 2469199872
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025425292890386773,
+      "loss": 1.2671,
+      "theoretical_loss": 3.4003107523103027,
+      "tokens_seen": 2469330944
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002542128069330766,
+      "loss": 1.3045,
+      "theoretical_loss": 3.4002964410986407,
+      "tokens_seen": 2469462016
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025417268496228534,
+      "loss": 1.3526,
+      "theoretical_loss": 3.4002821308592317,
+      "tokens_seen": 2469593088
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002541325629914941,
+      "loss": 1.3696,
+      "theoretical_loss": 3.4002678215919575,
+      "tokens_seen": 2469724160
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025409244102070295,
+      "loss": 1.2619,
+      "theoretical_loss": 3.4002535132967004,
+      "tokens_seen": 2469855232
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002540523190499117,
+      "loss": 1.3361,
+      "theoretical_loss": 3.400239205973344,
+      "tokens_seen": 2469986304
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025401219707912057,
+      "loss": 1.4321,
+      "theoretical_loss": 3.4002248996217697,
+      "tokens_seen": 2470117376
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002539720751083293,
+      "loss": 1.3095,
+      "theoretical_loss": 3.40021059424186,
+      "tokens_seen": 2470248448
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002539319531375381,
+      "loss": 1.4529,
+      "theoretical_loss": 3.400196289833498,
+      "tokens_seen": 2470379520
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025389183116674693,
+      "loss": 1.3687,
+      "theoretical_loss": 3.400181986396565,
+      "tokens_seen": 2470510592
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025385170919595574,
+      "loss": 1.3453,
+      "theoretical_loss": 3.4001676839309445,
+      "tokens_seen": 2470641664
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002538115872251645,
+      "loss": 1.4057,
+      "theoretical_loss": 3.4001533824365184,
+      "tokens_seen": 2470772736
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002537714652543733,
+      "loss": 1.3222,
+      "theoretical_loss": 3.40013908191317,
+      "tokens_seen": 2470903808
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002537313432835821,
+      "loss": 1.3901,
+      "theoretical_loss": 3.4001247823607814,
+      "tokens_seen": 2471034880
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025369122131279086,
+      "loss": 1.2735,
+      "theoretical_loss": 3.4001104837792346,
+      "tokens_seen": 2471165952
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002536510993419997,
+      "loss": 1.4012,
+      "theoretical_loss": 3.400096186168413,
+      "tokens_seen": 2471297024
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025361097737120847,
+      "loss": 1.2811,
+      "theoretical_loss": 3.4000818895281992,
+      "tokens_seen": 2471428096
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002535708554004173,
+      "loss": 1.3188,
+      "theoretical_loss": 3.4000675938584757,
+      "tokens_seen": 2471559168
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002535307334296261,
+      "loss": 1.3332,
+      "theoretical_loss": 3.400053299159125,
+      "tokens_seen": 2471690240
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025349061145883483,
+      "loss": 1.3665,
+      "theoretical_loss": 3.4000390054300293,
+      "tokens_seen": 2471821312
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002534504894880437,
+      "loss": 1.2756,
+      "theoretical_loss": 3.4000247126710725,
+      "tokens_seen": 2471952384
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025341036751725244,
+      "loss": 1.4017,
+      "theoretical_loss": 3.4000104208821367,
+      "tokens_seen": 2472083456
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025337024554646125,
+      "loss": 1.247,
+      "theoretical_loss": 3.399996130063104,
+      "tokens_seen": 2472214528
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025333012357567006,
+      "loss": 1.431,
+      "theoretical_loss": 3.3999818402138584,
+      "tokens_seen": 2472345600
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025329000160487886,
+      "loss": 1.3509,
+      "theoretical_loss": 3.399967551334282,
+      "tokens_seen": 2472476672
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002532498796340876,
+      "loss": 1.1837,
+      "theoretical_loss": 3.3999532634242575,
+      "tokens_seen": 2472607744
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002532097576632964,
+      "loss": 1.365,
+      "theoretical_loss": 3.399938976483668,
+      "tokens_seen": 2472738816
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025316963569250523,
+      "loss": 1.4541,
+      "theoretical_loss": 3.399924690512396,
+      "tokens_seen": 2472869888
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.000253129513721714,
+      "loss": 1.319,
+      "theoretical_loss": 3.3999104055103246,
+      "tokens_seen": 2473000960
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025308939175092284,
+      "loss": 1.3566,
+      "theoretical_loss": 3.3998961214773367,
+      "tokens_seen": 2473132032
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002530492697801316,
+      "loss": 1.3587,
+      "theoretical_loss": 3.3998818384133154,
+      "tokens_seen": 2473263104
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002530091478093404,
+      "loss": 1.3155,
+      "theoretical_loss": 3.399867556318143,
+      "tokens_seen": 2473394176
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002529690258385492,
+      "loss": 1.3373,
+      "theoretical_loss": 3.3998532751917034,
+      "tokens_seen": 2473525248
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025292890386775796,
+      "loss": 1.3092,
+      "theoretical_loss": 3.3998389950338788,
+      "tokens_seen": 2473656320
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025288878189696676,
+      "loss": 1.3559,
+      "theoretical_loss": 3.3998247158445523,
+      "tokens_seen": 2473787392
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025284865992617557,
+      "loss": 1.3144,
+      "theoretical_loss": 3.399810437623607,
+      "tokens_seen": 2473918464
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002528085379553844,
+      "loss": 1.3107,
+      "theoretical_loss": 3.399796160370926,
+      "tokens_seen": 2474049536
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002527684159845932,
+      "loss": 1.3355,
+      "theoretical_loss": 3.399781884086392,
+      "tokens_seen": 2474180608
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.000252728294013802,
+      "loss": 1.3191,
+      "theoretical_loss": 3.399767608769889,
+      "tokens_seen": 2474311680
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025268817204301074,
+      "loss": 1.3056,
+      "theoretical_loss": 3.3997533344212987,
+      "tokens_seen": 2474442752
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025264805007221955,
+      "loss": 1.3832,
+      "theoretical_loss": 3.3997390610405054,
+      "tokens_seen": 2474573824
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025260792810142835,
+      "loss": 1.3504,
+      "theoretical_loss": 3.3997247886273914,
+      "tokens_seen": 2474704896
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002525678061306371,
+      "loss": 1.2759,
+      "theoretical_loss": 3.3997105171818403,
+      "tokens_seen": 2474835968
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025252768415984597,
+      "loss": 1.2872,
+      "theoretical_loss": 3.3996962467037353,
+      "tokens_seen": 2474967040
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002524875621890547,
+      "loss": 1.3002,
+      "theoretical_loss": 3.3996819771929596,
+      "tokens_seen": 2475098112
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002524474402182635,
+      "loss": 1.3805,
+      "theoretical_loss": 3.399667708649396,
+      "tokens_seen": 2475229184
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025240731824747233,
+      "loss": 1.3799,
+      "theoretical_loss": 3.399653441072928,
+      "tokens_seen": 2475360256
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025236719627668114,
+      "loss": 1.373,
+      "theoretical_loss": 3.399639174463439,
+      "tokens_seen": 2475491328
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002523270743058899,
+      "loss": 1.3169,
+      "theoretical_loss": 3.399624908820812,
+      "tokens_seen": 2475622400
+    },
+    {
+      "epoch": 0.5,
+      "objective/train/advantage_avg": 0.4961351156234741,
+      "objective/train/docs_used": 901124,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.354691982269287,
+      "objective/train/original_loss": 1.354691982269287,
+      "objective/train/theoretical_loss": 3.3996106441449307,
+      "objective/train/tokens_used": 846212576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24634544551372528,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0508660078048706,
+      "objective/train/weighted_lm_loss": 1.4235517978668213,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3996106441449307,
+      "tokens_seen": 2475753472
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002522869523350987,
+      "loss": 1.3975,
+      "theoretical_loss": 3.3996106441449307,
+      "tokens_seen": 2475753472
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002522468303643075,
+      "loss": 1.3221,
+      "theoretical_loss": 3.399596380435678,
+      "tokens_seen": 2475884544
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002522067083935163,
+      "loss": 1.4109,
+      "theoretical_loss": 3.399582117692937,
+      "tokens_seen": 2476015616
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002521665864227251,
+      "loss": 1.343,
+      "theoretical_loss": 3.3995678559165916,
+      "tokens_seen": 2476146688
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025212646445193387,
+      "loss": 1.3338,
+      "theoretical_loss": 3.399553595106525,
+      "tokens_seen": 2476277760
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025208634248114273,
+      "loss": 1.3866,
+      "theoretical_loss": 3.39953933526262,
+      "tokens_seen": 2476408832
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002520462205103515,
+      "loss": 1.345,
+      "theoretical_loss": 3.3995250763847613,
+      "tokens_seen": 2476539904
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025200609853956023,
+      "loss": 1.2897,
+      "theoretical_loss": 3.3995108184728315,
+      "tokens_seen": 2476670976
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002519659765687691,
+      "loss": 1.3702,
+      "theoretical_loss": 3.399496561526714,
+      "tokens_seen": 2476802048
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025192585459797784,
+      "loss": 1.2872,
+      "theoretical_loss": 3.3994823055462926,
+      "tokens_seen": 2476933120
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025188573262718665,
+      "loss": 1.3179,
+      "theoretical_loss": 3.3994680505314507,
+      "tokens_seen": 2477064192
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025184561065639546,
+      "loss": 1.3471,
+      "theoretical_loss": 3.3994537964820712,
+      "tokens_seen": 2477195264
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025180548868560426,
+      "loss": 1.3255,
+      "theoretical_loss": 3.3994395433980387,
+      "tokens_seen": 2477326336
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.000251765366714813,
+      "loss": 1.4162,
+      "theoretical_loss": 3.399425291279236,
+      "tokens_seen": 2477457408
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002517252447440218,
+      "loss": 1.4006,
+      "theoretical_loss": 3.399411040125547,
+      "tokens_seen": 2477588480
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025168512277323063,
+      "loss": 1.331,
+      "theoretical_loss": 3.399396789936855,
+      "tokens_seen": 2477719552
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002516450008024394,
+      "loss": 1.2797,
+      "theoretical_loss": 3.399382540713044,
+      "tokens_seen": 2477850624
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025160487883164824,
+      "loss": 1.4296,
+      "theoretical_loss": 3.399368292453997,
+      "tokens_seen": 2477981696
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.000251564756860857,
+      "loss": 1.3732,
+      "theoretical_loss": 3.3993540451595985,
+      "tokens_seen": 2478112768
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025152463489006585,
+      "loss": 1.3689,
+      "theoretical_loss": 3.3993397988297316,
+      "tokens_seen": 2478243840
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002514845129192746,
+      "loss": 1.3489,
+      "theoretical_loss": 3.39932555346428,
+      "tokens_seen": 2478374912
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025144439094848336,
+      "loss": 1.3525,
+      "theoretical_loss": 3.3993113090631275,
+      "tokens_seen": 2478505984
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002514042689776922,
+      "loss": 1.4181,
+      "theoretical_loss": 3.399297065626158,
+      "tokens_seen": 2478637056
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025136414700690097,
+      "loss": 1.3668,
+      "theoretical_loss": 3.3992828231532553,
+      "tokens_seen": 2478768128
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002513240250361098,
+      "loss": 1.4409,
+      "theoretical_loss": 3.399268581644303,
+      "tokens_seen": 2478899200
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002512839030653186,
+      "loss": 1.336,
+      "theoretical_loss": 3.399254341099185,
+      "tokens_seen": 2479030272
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002512437810945274,
+      "loss": 1.2932,
+      "theoretical_loss": 3.3992401015177847,
+      "tokens_seen": 2479161344
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025120365912373614,
+      "loss": 1.3159,
+      "theoretical_loss": 3.3992258628999865,
+      "tokens_seen": 2479292416
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025116353715294495,
+      "loss": 1.309,
+      "theoretical_loss": 3.399211625245674,
+      "tokens_seen": 2479423488
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025112341518215375,
+      "loss": 1.3489,
+      "theoretical_loss": 3.399197388554731,
+      "tokens_seen": 2479554560
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002510832932113625,
+      "loss": 1.2793,
+      "theoretical_loss": 3.3991831528270415,
+      "tokens_seen": 2479685632
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025104317124057137,
+      "loss": 1.3948,
+      "theoretical_loss": 3.3991689180624896,
+      "tokens_seen": 2479816704
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002510030492697801,
+      "loss": 1.314,
+      "theoretical_loss": 3.3991546842609592,
+      "tokens_seen": 2479947776
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.000250962927298989,
+      "loss": 1.4094,
+      "theoretical_loss": 3.3991404514223333,
+      "tokens_seen": 2480078848
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025092280532819773,
+      "loss": 1.3279,
+      "theoretical_loss": 3.3991262195464973,
+      "tokens_seen": 2480209920
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002508826833574065,
+      "loss": 1.3638,
+      "theoretical_loss": 3.3991119886333347,
+      "tokens_seen": 2480340992
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025084256138661534,
+      "loss": 1.2928,
+      "theoretical_loss": 3.3990977586827293,
+      "tokens_seen": 2480472064
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002508024394158241,
+      "loss": 1.3108,
+      "theoretical_loss": 3.3990835296945647,
+      "tokens_seen": 2480603136
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002507623174450329,
+      "loss": 1.302,
+      "theoretical_loss": 3.399069301668726,
+      "tokens_seen": 2480734208
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002507221954742417,
+      "loss": 1.3371,
+      "theoretical_loss": 3.3990550746050965,
+      "tokens_seen": 2480865280
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002506820735034505,
+      "loss": 1.4533,
+      "theoretical_loss": 3.399040848503561,
+      "tokens_seen": 2480996352
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025064195153265927,
+      "loss": 1.3699,
+      "theoretical_loss": 3.3990266233640023,
+      "tokens_seen": 2481127424
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025060182956186813,
+      "loss": 1.3514,
+      "theoretical_loss": 3.399012399186306,
+      "tokens_seen": 2481258496
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002505617075910769,
+      "loss": 1.2695,
+      "theoretical_loss": 3.398998175970356,
+      "tokens_seen": 2481389568
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025052158562028563,
+      "loss": 1.3485,
+      "theoretical_loss": 3.3989839537160353,
+      "tokens_seen": 2481520640
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002504814636494945,
+      "loss": 1.3874,
+      "theoretical_loss": 3.3989697324232298,
+      "tokens_seen": 2481651712
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025044134167870324,
+      "loss": 1.2937,
+      "theoretical_loss": 3.3989555120918222,
+      "tokens_seen": 2481782784
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025040121970791205,
+      "loss": 1.3621,
+      "theoretical_loss": 3.3989412927216978,
+      "tokens_seen": 2481913856
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025036109773712086,
+      "loss": 1.3496,
+      "theoretical_loss": 3.39892707431274,
+      "tokens_seen": 2482044928
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025032097576632966,
+      "loss": 1.4085,
+      "theoretical_loss": 3.398912856864834,
+      "tokens_seen": 2482176000
+    },
+    {
+      "epoch": 0.5,
+      "objective/train/advantage_avg": 0.49090903997421265,
+      "objective/train/docs_used": 903113,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3238605260849,
+      "objective/train/original_loss": 1.3238604068756104,
+      "objective/train/theoretical_loss": 3.3988986403778636,
+      "objective/train/tokens_used": 852766176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24179239571094513,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050320029258728,
+      "objective/train/weighted_lm_loss": 1.390445590019226,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.8607079386711121,
+      "theoretical_loss": 3.3988986403778636,
+      "tokens_seen": 2482307072
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025028085379553847,
+      "loss": 1.3799,
+      "theoretical_loss": 3.3988986403778636,
+      "tokens_seen": 2482307072
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002502407318247472,
+      "loss": 1.2814,
+      "theoretical_loss": 3.3988844248517127,
+      "tokens_seen": 2482438144
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025020060985395603,
+      "loss": 1.4092,
+      "theoretical_loss": 3.3988702102862662,
+      "tokens_seen": 2482569216
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025016048788316483,
+      "loss": 1.3663,
+      "theoretical_loss": 3.3988559966814087,
+      "tokens_seen": 2482700288
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025012036591237364,
+      "loss": 1.2903,
+      "theoretical_loss": 3.3988417840370237,
+      "tokens_seen": 2482831360
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002500802439415824,
+      "loss": 1.2904,
+      "theoretical_loss": 3.3988275723529964,
+      "tokens_seen": 2482962432
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00025004012197079125,
+      "loss": 1.3234,
+      "theoretical_loss": 3.398813361629211,
+      "tokens_seen": 2483093504
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00025,
+      "loss": 1.4352,
+      "theoretical_loss": 3.3987991518655516,
+      "tokens_seen": 2483224576
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002499598780292088,
+      "loss": 1.3733,
+      "theoretical_loss": 3.3987849430619033,
+      "tokens_seen": 2483355648
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002499197560584176,
+      "loss": 1.3163,
+      "theoretical_loss": 3.3987707352181498,
+      "tokens_seen": 2483486720
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002498796340876264,
+      "loss": 1.4243,
+      "theoretical_loss": 3.398756528334176,
+      "tokens_seen": 2483617792
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002498395121168352,
+      "loss": 1.3313,
+      "theoretical_loss": 3.3987423224098667,
+      "tokens_seen": 2483748864
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.000249799390146044,
+      "loss": 1.2802,
+      "theoretical_loss": 3.3987281174451063,
+      "tokens_seen": 2483879936
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002497592681752528,
+      "loss": 1.4419,
+      "theoretical_loss": 3.398713913439779,
+      "tokens_seen": 2484011008
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024971914620446154,
+      "loss": 1.3179,
+      "theoretical_loss": 3.3986997103937693,
+      "tokens_seen": 2484142080
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024967902423367035,
+      "loss": 1.2831,
+      "theoretical_loss": 3.3986855083069623,
+      "tokens_seen": 2484273152
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024963890226287915,
+      "loss": 1.4508,
+      "theoretical_loss": 3.3986713071792423,
+      "tokens_seen": 2484404224
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024959878029208796,
+      "loss": 1.3463,
+      "theoretical_loss": 3.3986571070104943,
+      "tokens_seen": 2484535296
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024955865832129677,
+      "loss": 1.3633,
+      "theoretical_loss": 3.3986429078006024,
+      "tokens_seen": 2484666368
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002495185363505055,
+      "loss": 1.3936,
+      "theoretical_loss": 3.3986287095494516,
+      "tokens_seen": 2484797440
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002494784143797143,
+      "loss": 1.3156,
+      "theoretical_loss": 3.398614512256927,
+      "tokens_seen": 2484928512
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024943829240892313,
+      "loss": 1.2825,
+      "theoretical_loss": 3.3986003159229123,
+      "tokens_seen": 2485059584
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024939817043813194,
+      "loss": 1.3133,
+      "theoretical_loss": 3.398586120547293,
+      "tokens_seen": 2485190656
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024935804846734074,
+      "loss": 1.3008,
+      "theoretical_loss": 3.3985719261299536,
+      "tokens_seen": 2485321728
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024931792649654955,
+      "loss": 1.3115,
+      "theoretical_loss": 3.3985577326707794,
+      "tokens_seen": 2485452800
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002492778045257583,
+      "loss": 1.2932,
+      "theoretical_loss": 3.3985435401696544,
+      "tokens_seen": 2485583872
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002492376825549671,
+      "loss": 1.4269,
+      "theoretical_loss": 3.3985293486264636,
+      "tokens_seen": 2485714944
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002491975605841759,
+      "loss": 1.3342,
+      "theoretical_loss": 3.3985151580410924,
+      "tokens_seen": 2485846016
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024915743861338467,
+      "loss": 1.2451,
+      "theoretical_loss": 3.3985009684134253,
+      "tokens_seen": 2485977088
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024911731664259347,
+      "loss": 1.3556,
+      "theoretical_loss": 3.3984867797433465,
+      "tokens_seen": 2486108160
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002490771946718023,
+      "loss": 1.3406,
+      "theoretical_loss": 3.398472592030742,
+      "tokens_seen": 2486239232
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002490370727010111,
+      "loss": 1.3148,
+      "theoretical_loss": 3.3984584052754965,
+      "tokens_seen": 2486370304
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002489969507302199,
+      "loss": 1.2978,
+      "theoretical_loss": 3.3984442194774944,
+      "tokens_seen": 2486501376
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024895682875942864,
+      "loss": 1.3129,
+      "theoretical_loss": 3.398430034636621,
+      "tokens_seen": 2486632448
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024891670678863745,
+      "loss": 1.4148,
+      "theoretical_loss": 3.3984158507527606,
+      "tokens_seen": 2486763520
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024887658481784626,
+      "loss": 1.296,
+      "theoretical_loss": 3.3984016678257998,
+      "tokens_seen": 2486894592
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024883646284705506,
+      "loss": 1.3761,
+      "theoretical_loss": 3.398387485855622,
+      "tokens_seen": 2487025664
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024879634087626387,
+      "loss": 1.3644,
+      "theoretical_loss": 3.398373304842113,
+      "tokens_seen": 2487156736
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002487562189054726,
+      "loss": 1.336,
+      "theoretical_loss": 3.398359124785158,
+      "tokens_seen": 2487287808
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024871609693468143,
+      "loss": 1.2582,
+      "theoretical_loss": 3.3983449456846415,
+      "tokens_seen": 2487418880
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024867597496389023,
+      "loss": 1.3417,
+      "theoretical_loss": 3.3983307675404486,
+      "tokens_seen": 2487549952
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024863585299309904,
+      "loss": 1.3049,
+      "theoretical_loss": 3.398316590352465,
+      "tokens_seen": 2487681024
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002485957310223078,
+      "loss": 1.3794,
+      "theoretical_loss": 3.3983024141205758,
+      "tokens_seen": 2487812096
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002485556090515166,
+      "loss": 1.378,
+      "theoretical_loss": 3.398288238844666,
+      "tokens_seen": 2487943168
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002485154870807254,
+      "loss": 1.3541,
+      "theoretical_loss": 3.39827406452462,
+      "tokens_seen": 2488074240
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002484753651099342,
+      "loss": 1.3203,
+      "theoretical_loss": 3.398259891160324,
+      "tokens_seen": 2488205312
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.000248435243139143,
+      "loss": 1.335,
+      "theoretical_loss": 3.398245718751663,
+      "tokens_seen": 2488336384
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002483951211683518,
+      "loss": 1.3879,
+      "theoretical_loss": 3.398231547298522,
+      "tokens_seen": 2488467456
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002483549991975606,
+      "loss": 1.3729,
+      "theoretical_loss": 3.3982173768007864,
+      "tokens_seen": 2488598528
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002483148772267694,
+      "loss": 1.3965,
+      "theoretical_loss": 3.3982032072583412,
+      "tokens_seen": 2488729600
+    },
+    {
+      "epoch": 0.51,
+      "objective/train/advantage_avg": 0.4946686923503876,
+      "objective/train/docs_used": 905609,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.175655484199524,
+      "objective/train/original_loss": 1.1756553649902344,
+      "objective/train/theoretical_loss": 3.398189038671072,
+      "objective/train/tokens_used": 859319776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24499139189720154,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0507123470306396,
+      "objective/train/weighted_lm_loss": 1.2351675033569336,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.398189038671072,
+      "tokens_seen": 2488860672
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002482747552559782,
+      "loss": 1.3114,
+      "theoretical_loss": 3.398189038671072,
+      "tokens_seen": 2488860672
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.000248234633285187,
+      "loss": 1.3072,
+      "theoretical_loss": 3.3981748710388646,
+      "tokens_seen": 2488991744
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024819451131439575,
+      "loss": 1.3794,
+      "theoretical_loss": 3.398160704361603,
+      "tokens_seen": 2489122816
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024815438934360455,
+      "loss": 1.4007,
+      "theoretical_loss": 3.398146538639174,
+      "tokens_seen": 2489253888
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024811426737281336,
+      "loss": 1.3418,
+      "theoretical_loss": 3.398132373871462,
+      "tokens_seen": 2489384960
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024807414540202217,
+      "loss": 1.3935,
+      "theoretical_loss": 3.3981182100583527,
+      "tokens_seen": 2489516032
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002480340234312309,
+      "loss": 1.321,
+      "theoretical_loss": 3.3981040471997312,
+      "tokens_seen": 2489647104
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002479939014604397,
+      "loss": 1.2964,
+      "theoretical_loss": 3.3980898852954837,
+      "tokens_seen": 2489778176
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024795377948964853,
+      "loss": 1.3899,
+      "theoretical_loss": 3.398075724345495,
+      "tokens_seen": 2489909248
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024791365751885734,
+      "loss": 1.4199,
+      "theoretical_loss": 3.398061564349651,
+      "tokens_seen": 2490040320
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024787353554806614,
+      "loss": 1.3314,
+      "theoretical_loss": 3.3980474053078367,
+      "tokens_seen": 2490171392
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024783341357727495,
+      "loss": 1.3469,
+      "theoretical_loss": 3.398033247219938,
+      "tokens_seen": 2490302464
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002477932916064837,
+      "loss": 1.3285,
+      "theoretical_loss": 3.3980190900858402,
+      "tokens_seen": 2490433536
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002477531696356925,
+      "loss": 1.3637,
+      "theoretical_loss": 3.398004933905429,
+      "tokens_seen": 2490564608
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002477130476649013,
+      "loss": 1.3512,
+      "theoretical_loss": 3.39799077867859,
+      "tokens_seen": 2490695680
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002476729256941101,
+      "loss": 1.3434,
+      "theoretical_loss": 3.3979766244052083,
+      "tokens_seen": 2490826752
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024763280372331887,
+      "loss": 1.3234,
+      "theoretical_loss": 3.3979624710851706,
+      "tokens_seen": 2490957824
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002475926817525277,
+      "loss": 1.3964,
+      "theoretical_loss": 3.3979483187183614,
+      "tokens_seen": 2491088896
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002475525597817365,
+      "loss": 1.3488,
+      "theoretical_loss": 3.397934167304667,
+      "tokens_seen": 2491219968
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002475124378109453,
+      "loss": 1.3467,
+      "theoretical_loss": 3.3979200168439725,
+      "tokens_seen": 2491351040
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024747231584015404,
+      "loss": 1.3306,
+      "theoretical_loss": 3.3979058673361644,
+      "tokens_seen": 2491482112
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024743219386936285,
+      "loss": 1.4324,
+      "theoretical_loss": 3.3978917187811275,
+      "tokens_seen": 2491613184
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024739207189857166,
+      "loss": 1.3408,
+      "theoretical_loss": 3.3978775711787486,
+      "tokens_seen": 2491744256
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024735194992778046,
+      "loss": 1.3484,
+      "theoretical_loss": 3.3978634245289125,
+      "tokens_seen": 2491875328
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024731182795698927,
+      "loss": 1.3278,
+      "theoretical_loss": 3.3978492788315053,
+      "tokens_seen": 2492006400
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002472717059861981,
+      "loss": 1.2668,
+      "theoretical_loss": 3.397835134086413,
+      "tokens_seen": 2492137472
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024723158401540683,
+      "loss": 1.4509,
+      "theoretical_loss": 3.397820990293521,
+      "tokens_seen": 2492268544
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024719146204461563,
+      "loss": 1.3283,
+      "theoretical_loss": 3.397806847452715,
+      "tokens_seen": 2492399616
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024715134007382444,
+      "loss": 1.2949,
+      "theoretical_loss": 3.397792705563882,
+      "tokens_seen": 2492530688
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024711121810303325,
+      "loss": 1.3008,
+      "theoretical_loss": 3.3977785646269063,
+      "tokens_seen": 2492661760
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.000247071096132242,
+      "loss": 1.2968,
+      "theoretical_loss": 3.397764424641675,
+      "tokens_seen": 2492792832
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002470309741614508,
+      "loss": 1.4094,
+      "theoretical_loss": 3.3977502856080735,
+      "tokens_seen": 2492923904
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002469908521906596,
+      "loss": 1.3467,
+      "theoretical_loss": 3.3977361475259875,
+      "tokens_seen": 2493054976
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002469507302198684,
+      "loss": 1.3744,
+      "theoretical_loss": 3.397722010395303,
+      "tokens_seen": 2493186048
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002469106082490772,
+      "loss": 1.3409,
+      "theoretical_loss": 3.397707874215907,
+      "tokens_seen": 2493317120
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.000246870486278286,
+      "loss": 1.294,
+      "theoretical_loss": 3.397693738987684,
+      "tokens_seen": 2493448192
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002468303643074948,
+      "loss": 1.3068,
+      "theoretical_loss": 3.397679604710521,
+      "tokens_seen": 2493579264
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002467902423367036,
+      "loss": 1.2887,
+      "theoretical_loss": 3.397665471384303,
+      "tokens_seen": 2493710336
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002467501203659124,
+      "loss": 1.2308,
+      "theoretical_loss": 3.3976513390089176,
+      "tokens_seen": 2493841408
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002467099983951212,
+      "loss": 1.3392,
+      "theoretical_loss": 3.3976372075842494,
+      "tokens_seen": 2493972480
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024666987642432995,
+      "loss": 1.3245,
+      "theoretical_loss": 3.397623077110185,
+      "tokens_seen": 2494103552
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024662975445353876,
+      "loss": 1.3415,
+      "theoretical_loss": 3.3976089475866114,
+      "tokens_seen": 2494234624
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024658963248274757,
+      "loss": 1.3558,
+      "theoretical_loss": 3.3975948190134133,
+      "tokens_seen": 2494365696
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002465495105119563,
+      "loss": 1.3353,
+      "theoretical_loss": 3.3975806913904774,
+      "tokens_seen": 2494496768
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002465093885411651,
+      "loss": 1.3689,
+      "theoretical_loss": 3.39756656471769,
+      "tokens_seen": 2494627840
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024646926657037393,
+      "loss": 1.3081,
+      "theoretical_loss": 3.3975524389949374,
+      "tokens_seen": 2494758912
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024642914459958274,
+      "loss": 1.2774,
+      "theoretical_loss": 3.397538314222105,
+      "tokens_seen": 2494889984
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024638902262879154,
+      "loss": 1.4141,
+      "theoretical_loss": 3.3975241903990803,
+      "tokens_seen": 2495021056
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024634890065800035,
+      "loss": 1.3274,
+      "theoretical_loss": 3.397510067525748,
+      "tokens_seen": 2495152128
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024630877868720916,
+      "loss": 1.265,
+      "theoretical_loss": 3.397495945601996,
+      "tokens_seen": 2495283200
+    },
+    {
+      "epoch": 0.51,
+      "objective/train/advantage_avg": 0.49343815445899963,
+      "objective/train/docs_used": 908214,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2351884841918945,
+      "objective/train/original_loss": 1.235188364982605,
+      "objective/train/theoretical_loss": 3.3974818246277096,
+      "objective/train/tokens_used": 865873376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24385757744312286,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505836009979248,
+      "objective/train/weighted_lm_loss": 1.2976628541946411,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3974818246277096,
+      "tokens_seen": 2495414272
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002462686567164179,
+      "loss": 1.3551,
+      "theoretical_loss": 3.3974818246277096,
+      "tokens_seen": 2495414272
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002462285347456267,
+      "loss": 1.2964,
+      "theoretical_loss": 3.397467704602775,
+      "tokens_seen": 2495545344
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002461884127748355,
+      "loss": 1.2818,
+      "theoretical_loss": 3.397453585527079,
+      "tokens_seen": 2495676416
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002461482908040443,
+      "loss": 1.3418,
+      "theoretical_loss": 3.397439467400508,
+      "tokens_seen": 2495807488
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002461081688332531,
+      "loss": 1.3875,
+      "theoretical_loss": 3.3974253502229477,
+      "tokens_seen": 2495938560
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002460680468624619,
+      "loss": 1.347,
+      "theoretical_loss": 3.397411233994285,
+      "tokens_seen": 2496069632
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002460279248916707,
+      "loss": 1.4089,
+      "theoretical_loss": 3.397397118714406,
+      "tokens_seen": 2496200704
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024598780292087944,
+      "loss": 1.4229,
+      "theoretical_loss": 3.3973830043831974,
+      "tokens_seen": 2496331776
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024594768095008825,
+      "loss": 1.3225,
+      "theoretical_loss": 3.3973688910005455,
+      "tokens_seen": 2496462848
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024590755897929706,
+      "loss": 1.3287,
+      "theoretical_loss": 3.3973547785663367,
+      "tokens_seen": 2496593920
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024586743700850586,
+      "loss": 1.3858,
+      "theoretical_loss": 3.3973406670804573,
+      "tokens_seen": 2496724992
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024582731503771467,
+      "loss": 1.3121,
+      "theoretical_loss": 3.3973265565427946,
+      "tokens_seen": 2496856064
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002457871930669235,
+      "loss": 1.3925,
+      "theoretical_loss": 3.3973124469532343,
+      "tokens_seen": 2496987136
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002457470710961323,
+      "loss": 1.3546,
+      "theoretical_loss": 3.3972983383116633,
+      "tokens_seen": 2497118208
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024570694912534103,
+      "loss": 1.2684,
+      "theoretical_loss": 3.397284230617968,
+      "tokens_seen": 2497249280
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024566682715454984,
+      "loss": 1.417,
+      "theoretical_loss": 3.397270123872035,
+      "tokens_seen": 2497380352
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024562670518375865,
+      "loss": 1.427,
+      "theoretical_loss": 3.3972560180737505,
+      "tokens_seen": 2497511424
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002455865832129674,
+      "loss": 1.3894,
+      "theoretical_loss": 3.397241913223002,
+      "tokens_seen": 2497642496
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002455464612421762,
+      "loss": 1.3124,
+      "theoretical_loss": 3.3972278093196753,
+      "tokens_seen": 2497773568
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.000245506339271385,
+      "loss": 1.3714,
+      "theoretical_loss": 3.3972137063636576,
+      "tokens_seen": 2497904640
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002454662173005938,
+      "loss": 1.3041,
+      "theoretical_loss": 3.397199604354835,
+      "tokens_seen": 2498035712
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002454260953298026,
+      "loss": 1.3526,
+      "theoretical_loss": 3.3971855032930955,
+      "tokens_seen": 2498166784
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002453859733590114,
+      "loss": 1.3564,
+      "theoretical_loss": 3.397171403178324,
+      "tokens_seen": 2498297856
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002453458513882202,
+      "loss": 1.301,
+      "theoretical_loss": 3.3971573040104084,
+      "tokens_seen": 2498428928
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.000245305729417429,
+      "loss": 1.3032,
+      "theoretical_loss": 3.3971432057892352,
+      "tokens_seen": 2498560000
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002452656074466378,
+      "loss": 1.3262,
+      "theoretical_loss": 3.397129108514691,
+      "tokens_seen": 2498691072
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002452254854758466,
+      "loss": 1.3631,
+      "theoretical_loss": 3.3971150121866627,
+      "tokens_seen": 2498822144
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002451853635050554,
+      "loss": 1.2829,
+      "theoretical_loss": 3.397100916805037,
+      "tokens_seen": 2498953216
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024514524153426416,
+      "loss": 1.3978,
+      "theoretical_loss": 3.3970868223697006,
+      "tokens_seen": 2499084288
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024510511956347297,
+      "loss": 1.3305,
+      "theoretical_loss": 3.397072728880541,
+      "tokens_seen": 2499215360
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024506499759268177,
+      "loss": 1.4171,
+      "theoretical_loss": 3.3970586363374444,
+      "tokens_seen": 2499346432
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002450248756218905,
+      "loss": 1.332,
+      "theoretical_loss": 3.397044544740298,
+      "tokens_seen": 2499477504
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00024498475365109933,
+      "loss": 1.2845,
+      "theoretical_loss": 3.3970304540889886,
+      "tokens_seen": 2499608576
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024494463168030814,
+      "loss": 1.3764,
+      "theoretical_loss": 3.3970163643834033,
+      "tokens_seen": 2499739648
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024490450970951694,
+      "loss": 1.3381,
+      "theoretical_loss": 3.3970022756234286,
+      "tokens_seen": 2499870720
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024486438773872575,
+      "loss": 1.322,
+      "theoretical_loss": 3.396988187808952,
+      "tokens_seen": 2500001792
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024482426576793456,
+      "loss": 1.2593,
+      "theoretical_loss": 3.3969741009398597,
+      "tokens_seen": 2500132864
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002447841437971433,
+      "loss": 1.2989,
+      "theoretical_loss": 3.39696001501604,
+      "tokens_seen": 2500263936
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002447440218263521,
+      "loss": 1.3218,
+      "theoretical_loss": 3.3969459300373783,
+      "tokens_seen": 2500395008
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002447038998555609,
+      "loss": 1.328,
+      "theoretical_loss": 3.3969318460037634,
+      "tokens_seen": 2500526080
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002446637778847697,
+      "loss": 1.2325,
+      "theoretical_loss": 3.3969177629150806,
+      "tokens_seen": 2500657152
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002446236559139785,
+      "loss": 1.4212,
+      "theoretical_loss": 3.3969036807712185,
+      "tokens_seen": 2500788224
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002445835339431873,
+      "loss": 1.2429,
+      "theoretical_loss": 3.396889599572063,
+      "tokens_seen": 2500919296
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002445434119723961,
+      "loss": 1.3719,
+      "theoretical_loss": 3.3968755193175024,
+      "tokens_seen": 2501050368
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002445032900016049,
+      "loss": 1.3559,
+      "theoretical_loss": 3.396861440007423,
+      "tokens_seen": 2501181440
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024446316803081365,
+      "loss": 1.3054,
+      "theoretical_loss": 3.3968473616417114,
+      "tokens_seen": 2501312512
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024442304606002246,
+      "loss": 1.3367,
+      "theoretical_loss": 3.396833284220256,
+      "tokens_seen": 2501443584
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024438292408923126,
+      "loss": 1.3307,
+      "theoretical_loss": 3.3968192077429435,
+      "tokens_seen": 2501574656
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024434280211844007,
+      "loss": 1.3216,
+      "theoretical_loss": 3.396805132209661,
+      "tokens_seen": 2501705728
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002443026801476489,
+      "loss": 1.3655,
+      "theoretical_loss": 3.396791057620296,
+      "tokens_seen": 2501836800
+    },
+    {
+      "epoch": 0.52,
+      "objective/train/advantage_avg": 0.49200841784477234,
+      "objective/train/docs_used": 910572,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.254350185394287,
+      "objective/train/original_loss": 1.2543503046035767,
+      "objective/train/theoretical_loss": 3.3967769839747355,
+      "objective/train/tokens_used": 872426976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24272920191287994,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504347085952759,
+      "objective/train/weighted_lm_loss": 1.3174842596054077,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3967769839747355,
+      "tokens_seen": 2501967872
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002442625581768577,
+      "loss": 1.2982,
+      "theoretical_loss": 3.3967769839747355,
+      "tokens_seen": 2501967872
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024422243620606643,
+      "loss": 1.2855,
+      "theoretical_loss": 3.3967629112728672,
+      "tokens_seen": 2502098944
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024418231423527524,
+      "loss": 1.2363,
+      "theoretical_loss": 3.3967488395145775,
+      "tokens_seen": 2502230016
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024414219226448405,
+      "loss": 1.3538,
+      "theoretical_loss": 3.396734768699755,
+      "tokens_seen": 2502361088
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024410207029369282,
+      "loss": 1.3007,
+      "theoretical_loss": 3.396720698828286,
+      "tokens_seen": 2502492160
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024406194832290163,
+      "loss": 1.2889,
+      "theoretical_loss": 3.396706629900058,
+      "tokens_seen": 2502623232
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024402182635211044,
+      "loss": 1.2578,
+      "theoretical_loss": 3.3966925619149584,
+      "tokens_seen": 2502754304
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024398170438131922,
+      "loss": 1.3007,
+      "theoretical_loss": 3.3966784948728748,
+      "tokens_seen": 2502885376
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024394158241052802,
+      "loss": 1.396,
+      "theoretical_loss": 3.396664428773695,
+      "tokens_seen": 2503016448
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002439014604397368,
+      "loss": 1.2756,
+      "theoretical_loss": 3.3966503636173053,
+      "tokens_seen": 2503147520
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024386133846894558,
+      "loss": 1.3864,
+      "theoretical_loss": 3.3966362994035944,
+      "tokens_seen": 2503278592
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002438212164981544,
+      "loss": 1.3423,
+      "theoretical_loss": 3.3966222361324485,
+      "tokens_seen": 2503409664
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002437810945273632,
+      "loss": 1.2357,
+      "theoretical_loss": 3.3966081738037563,
+      "tokens_seen": 2503540736
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.000243740972556572,
+      "loss": 1.4236,
+      "theoretical_loss": 3.3965941124174046,
+      "tokens_seen": 2503671808
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024370085058578078,
+      "loss": 1.3667,
+      "theoretical_loss": 3.396580051973281,
+      "tokens_seen": 2503802880
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024366072861498959,
+      "loss": 1.183,
+      "theoretical_loss": 3.396565992471273,
+      "tokens_seen": 2503933952
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024362060664419836,
+      "loss": 1.3477,
+      "theoretical_loss": 3.3965519339112684,
+      "tokens_seen": 2504065024
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024358048467340714,
+      "loss": 1.291,
+      "theoretical_loss": 3.396537876293155,
+      "tokens_seen": 2504196096
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024354036270261595,
+      "loss": 1.2918,
+      "theoretical_loss": 3.3965238196168195,
+      "tokens_seen": 2504327168
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024350024073182476,
+      "loss": 1.3217,
+      "theoretical_loss": 3.3965097638821504,
+      "tokens_seen": 2504458240
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024346011876103356,
+      "loss": 1.3282,
+      "theoretical_loss": 3.396495709089035,
+      "tokens_seen": 2504589312
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024341999679024234,
+      "loss": 1.3045,
+      "theoretical_loss": 3.396481655237361,
+      "tokens_seen": 2504720384
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024337987481945115,
+      "loss": 1.3138,
+      "theoretical_loss": 3.3964676023270157,
+      "tokens_seen": 2504851456
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024333975284865993,
+      "loss": 1.2801,
+      "theoretical_loss": 3.3964535503578874,
+      "tokens_seen": 2504982528
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002432996308778687,
+      "loss": 1.2795,
+      "theoretical_loss": 3.3964394993298637,
+      "tokens_seen": 2505113600
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002432595089070775,
+      "loss": 1.2529,
+      "theoretical_loss": 3.396425449242832,
+      "tokens_seen": 2505244672
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024321938693628632,
+      "loss": 1.3496,
+      "theoretical_loss": 3.3964114000966807,
+      "tokens_seen": 2505375744
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002431792649654951,
+      "loss": 1.2831,
+      "theoretical_loss": 3.396397351891297,
+      "tokens_seen": 2505506816
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002431391429947039,
+      "loss": 1.277,
+      "theoretical_loss": 3.3963833046265686,
+      "tokens_seen": 2505637888
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002430990210239127,
+      "loss": 1.4114,
+      "theoretical_loss": 3.3963692583023835,
+      "tokens_seen": 2505768960
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024305889905312152,
+      "loss": 1.3715,
+      "theoretical_loss": 3.3963552129186296,
+      "tokens_seen": 2505900032
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024301877708233027,
+      "loss": 1.3705,
+      "theoretical_loss": 3.396341168475195,
+      "tokens_seen": 2506031104
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024297865511153908,
+      "loss": 1.2315,
+      "theoretical_loss": 3.3963271249719673,
+      "tokens_seen": 2506162176
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024293853314074788,
+      "loss": 1.3023,
+      "theoretical_loss": 3.396313082408834,
+      "tokens_seen": 2506293248
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024289841116995666,
+      "loss": 1.3417,
+      "theoretical_loss": 3.3962990407856837,
+      "tokens_seen": 2506424320
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024285828919916547,
+      "loss": 1.3255,
+      "theoretical_loss": 3.396285000102404,
+      "tokens_seen": 2506555392
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024281816722837427,
+      "loss": 1.2613,
+      "theoretical_loss": 3.396270960358883,
+      "tokens_seen": 2506686464
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024277804525758308,
+      "loss": 1.2712,
+      "theoretical_loss": 3.3962569215550085,
+      "tokens_seen": 2506817536
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024273792328679183,
+      "loss": 1.2881,
+      "theoretical_loss": 3.396242883690668,
+      "tokens_seen": 2506948608
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024269780131600064,
+      "loss": 1.294,
+      "theoretical_loss": 3.3962288467657507,
+      "tokens_seen": 2507079680
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024265767934520945,
+      "loss": 1.3275,
+      "theoretical_loss": 3.3962148107801435,
+      "tokens_seen": 2507210752
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024261755737441822,
+      "loss": 1.3076,
+      "theoretical_loss": 3.3962007757337354,
+      "tokens_seen": 2507341824
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024257743540362703,
+      "loss": 1.3972,
+      "theoretical_loss": 3.3961867416264138,
+      "tokens_seen": 2507472896
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024253731343283584,
+      "loss": 1.3151,
+      "theoretical_loss": 3.3961727084580664,
+      "tokens_seen": 2507603968
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024249719146204464,
+      "loss": 1.2762,
+      "theoretical_loss": 3.3961586762285827,
+      "tokens_seen": 2507735040
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024245706949125342,
+      "loss": 1.3248,
+      "theoretical_loss": 3.3961446449378494,
+      "tokens_seen": 2507866112
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002424169475204622,
+      "loss": 1.3662,
+      "theoretical_loss": 3.3961306145857555,
+      "tokens_seen": 2507997184
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.000242376825549671,
+      "loss": 1.3073,
+      "theoretical_loss": 3.3961165851721886,
+      "tokens_seen": 2508128256
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002423367035788798,
+      "loss": 1.2867,
+      "theoretical_loss": 3.3961025566970373,
+      "tokens_seen": 2508259328
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002422965816080886,
+      "loss": 1.3599,
+      "theoretical_loss": 3.3960885291601897,
+      "tokens_seen": 2508390400
+    },
+    {
+      "epoch": 0.52,
+      "objective/train/advantage_avg": 0.4928663671016693,
+      "objective/train/docs_used": 912960,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2810674905776978,
+      "objective/train/original_loss": 1.2810676097869873,
+      "objective/train/theoretical_loss": 3.3960745025615338,
+      "objective/train/tokens_used": 878980576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2432985156774521,
+      "objective/train/value_max": -0.5,
+      "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": 1.3456768989562988,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3960745025615338,
+      "tokens_seen": 2508521472
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002422564596372974,
+      "loss": 1.3306,
+      "theoretical_loss": 3.3960745025615338,
+      "tokens_seen": 2508521472
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024221633766650618,
+      "loss": 1.3051,
+      "theoretical_loss": 3.396060476900958,
+      "tokens_seen": 2508652544
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024217621569571499,
+      "loss": 1.2998,
+      "theoretical_loss": 3.396046452178351,
+      "tokens_seen": 2508783616
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024213609372492376,
+      "loss": 1.239,
+      "theoretical_loss": 3.3960324283936,
+      "tokens_seen": 2508914688
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024209597175413257,
+      "loss": 1.3136,
+      "theoretical_loss": 3.3960184055465943,
+      "tokens_seen": 2509045760
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024205584978334135,
+      "loss": 1.291,
+      "theoretical_loss": 3.3960043836372216,
+      "tokens_seen": 2509176832
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024201572781255016,
+      "loss": 1.302,
+      "theoretical_loss": 3.3959903626653705,
+      "tokens_seen": 2509307904
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024197560584175896,
+      "loss": 1.4177,
+      "theoretical_loss": 3.3959763426309295,
+      "tokens_seen": 2509438976
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024193548387096774,
+      "loss": 1.3121,
+      "theoretical_loss": 3.395962323533787,
+      "tokens_seen": 2509570048
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024189536190017655,
+      "loss": 1.2934,
+      "theoretical_loss": 3.3959483053738304,
+      "tokens_seen": 2509701120
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024185523992938533,
+      "loss": 1.3336,
+      "theoretical_loss": 3.3959342881509493,
+      "tokens_seen": 2509832192
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024181511795859413,
+      "loss": 1.1772,
+      "theoretical_loss": 3.395920271865032,
+      "tokens_seen": 2509963264
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002417749959878029,
+      "loss": 1.347,
+      "theoretical_loss": 3.3959062565159663,
+      "tokens_seen": 2510094336
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024173487401701172,
+      "loss": 1.2421,
+      "theoretical_loss": 3.395892242103641,
+      "tokens_seen": 2510225408
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024169475204622053,
+      "loss": 1.3482,
+      "theoretical_loss": 3.3958782286279448,
+      "tokens_seen": 2510356480
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002416546300754293,
+      "loss": 1.2556,
+      "theoretical_loss": 3.3958642160887655,
+      "tokens_seen": 2510487552
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002416145081046381,
+      "loss": 1.2859,
+      "theoretical_loss": 3.395850204485993,
+      "tokens_seen": 2510618624
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024157438613384692,
+      "loss": 1.2817,
+      "theoretical_loss": 3.3958361938195143,
+      "tokens_seen": 2510749696
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002415342641630557,
+      "loss": 1.338,
+      "theoretical_loss": 3.3958221840892184,
+      "tokens_seen": 2510880768
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024149414219226448,
+      "loss": 1.3132,
+      "theoretical_loss": 3.3958081752949947,
+      "tokens_seen": 2511011840
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024145402022147328,
+      "loss": 1.3193,
+      "theoretical_loss": 3.395794167436731,
+      "tokens_seen": 2511142912
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002414138982506821,
+      "loss": 1.3219,
+      "theoretical_loss": 3.3957801605143163,
+      "tokens_seen": 2511273984
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024137377627989087,
+      "loss": 1.3713,
+      "theoretical_loss": 3.3957661545276387,
+      "tokens_seen": 2511405056
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024133365430909967,
+      "loss": 1.3802,
+      "theoretical_loss": 3.3957521494765874,
+      "tokens_seen": 2511536128
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024129353233830848,
+      "loss": 1.3166,
+      "theoretical_loss": 3.395738145361051,
+      "tokens_seen": 2511667200
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024125341036751723,
+      "loss": 1.3384,
+      "theoretical_loss": 3.395724142180918,
+      "tokens_seen": 2511798272
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024121328839672604,
+      "loss": 1.4059,
+      "theoretical_loss": 3.395710139936077,
+      "tokens_seen": 2511929344
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024117316642593484,
+      "loss": 1.3938,
+      "theoretical_loss": 3.395696138626417,
+      "tokens_seen": 2512060416
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024113304445514365,
+      "loss": 1.3811,
+      "theoretical_loss": 3.3956821382518267,
+      "tokens_seen": 2512191488
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024109292248435243,
+      "loss": 1.2464,
+      "theoretical_loss": 3.395668138812195,
+      "tokens_seen": 2512322560
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024105280051356124,
+      "loss": 1.2762,
+      "theoretical_loss": 3.3956541403074105,
+      "tokens_seen": 2512453632
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024101267854277004,
+      "loss": 1.3045,
+      "theoretical_loss": 3.395640142737362,
+      "tokens_seen": 2512584704
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002409725565719788,
+      "loss": 1.3304,
+      "theoretical_loss": 3.3956261461019386,
+      "tokens_seen": 2512715776
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002409324346011876,
+      "loss": 1.4358,
+      "theoretical_loss": 3.3956121504010284,
+      "tokens_seen": 2512846848
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002408923126303964,
+      "loss": 1.3218,
+      "theoretical_loss": 3.395598155634521,
+      "tokens_seen": 2512977920
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024085219065960521,
+      "loss": 1.3036,
+      "theoretical_loss": 3.3955841618023053,
+      "tokens_seen": 2513108992
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.000240812068688814,
+      "loss": 1.4008,
+      "theoretical_loss": 3.39557016890427,
+      "tokens_seen": 2513240064
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002407719467180228,
+      "loss": 1.3857,
+      "theoretical_loss": 3.3955561769403038,
+      "tokens_seen": 2513371136
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002407318247472316,
+      "loss": 1.2744,
+      "theoretical_loss": 3.395542185910296,
+      "tokens_seen": 2513502208
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024069170277644038,
+      "loss": 1.3409,
+      "theoretical_loss": 3.395528195814135,
+      "tokens_seen": 2513633280
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024065158080564916,
+      "loss": 1.2549,
+      "theoretical_loss": 3.3955142066517103,
+      "tokens_seen": 2513764352
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024061145883485797,
+      "loss": 1.3864,
+      "theoretical_loss": 3.395500218422911,
+      "tokens_seen": 2513895424
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024057133686406678,
+      "loss": 1.3633,
+      "theoretical_loss": 3.3954862311276255,
+      "tokens_seen": 2514026496
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024053121489327556,
+      "loss": 1.3214,
+      "theoretical_loss": 3.3954722447657435,
+      "tokens_seen": 2514157568
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024049109292248436,
+      "loss": 1.262,
+      "theoretical_loss": 3.395458259337154,
+      "tokens_seen": 2514288640
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024045097095169317,
+      "loss": 1.3399,
+      "theoretical_loss": 3.3954442748417457,
+      "tokens_seen": 2514419712
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024041084898090195,
+      "loss": 1.3318,
+      "theoretical_loss": 3.395430291279408,
+      "tokens_seen": 2514550784
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024037072701011073,
+      "loss": 1.3335,
+      "theoretical_loss": 3.3954163086500295,
+      "tokens_seen": 2514681856
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024033060503931953,
+      "loss": 1.2873,
+      "theoretical_loss": 3.3954023269534996,
+      "tokens_seen": 2514812928
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024029048306852834,
+      "loss": 1.3204,
+      "theoretical_loss": 3.395388346189708,
+      "tokens_seen": 2514944000
+    },
+    {
+      "epoch": 0.52,
+      "objective/train/advantage_avg": 0.4932105839252472,
+      "objective/train/docs_used": 915361,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2162680625915527,
+      "objective/train/original_loss": 1.2162680625915527,
+      "objective/train/theoretical_loss": 3.395374366358543,
+      "objective/train/tokens_used": 885534176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.243656188249588,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505597591400146,
+      "objective/train/weighted_lm_loss": 1.2777762413024902,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.395374366358543,
+      "tokens_seen": 2515075072
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024025036109773712,
+      "loss": 1.2888,
+      "theoretical_loss": 3.395374366358543,
+      "tokens_seen": 2515075072
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024021023912694593,
+      "loss": 1.2949,
+      "theoretical_loss": 3.3953603874598945,
+      "tokens_seen": 2515206144
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024017011715615473,
+      "loss": 1.2683,
+      "theoretical_loss": 3.3953464094936514,
+      "tokens_seen": 2515337216
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002401299951853635,
+      "loss": 1.2846,
+      "theoretical_loss": 3.3953324324597025,
+      "tokens_seen": 2515468288
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024008987321457232,
+      "loss": 1.2962,
+      "theoretical_loss": 3.395318456357938,
+      "tokens_seen": 2515599360
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002400497512437811,
+      "loss": 1.3834,
+      "theoretical_loss": 3.3953044811882465,
+      "tokens_seen": 2515730432
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00024000962927298988,
+      "loss": 1.3498,
+      "theoretical_loss": 3.3952905069505173,
+      "tokens_seen": 2515861504
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00023996950730219868,
+      "loss": 1.3425,
+      "theoretical_loss": 3.39527653364464,
+      "tokens_seen": 2515992576
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002399293853314075,
+      "loss": 1.3182,
+      "theoretical_loss": 3.3952625612705036,
+      "tokens_seen": 2516123648
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002398892633606163,
+      "loss": 1.3373,
+      "theoretical_loss": 3.3952485898279976,
+      "tokens_seen": 2516254720
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023984914138982507,
+      "loss": 1.3285,
+      "theoretical_loss": 3.3952346193170113,
+      "tokens_seen": 2516385792
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023980901941903388,
+      "loss": 1.2665,
+      "theoretical_loss": 3.3952206497374346,
+      "tokens_seen": 2516516864
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023976889744824266,
+      "loss": 1.3051,
+      "theoretical_loss": 3.3952066810891557,
+      "tokens_seen": 2516647936
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023972877547745144,
+      "loss": 1.3449,
+      "theoretical_loss": 3.395192713372065,
+      "tokens_seen": 2516779008
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023968865350666024,
+      "loss": 1.2814,
+      "theoretical_loss": 3.395178746586052,
+      "tokens_seen": 2516910080
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023964853153586905,
+      "loss": 1.327,
+      "theoretical_loss": 3.3951647807310055,
+      "tokens_seen": 2517041152
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023960840956507786,
+      "loss": 1.2853,
+      "theoretical_loss": 3.395150815806815,
+      "tokens_seen": 2517172224
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023956828759428664,
+      "loss": 1.316,
+      "theoretical_loss": 3.395136851813371,
+      "tokens_seen": 2517303296
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023952816562349544,
+      "loss": 1.3352,
+      "theoretical_loss": 3.395122888750562,
+      "tokens_seen": 2517434368
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023948804365270422,
+      "loss": 1.4024,
+      "theoretical_loss": 3.3951089266182772,
+      "tokens_seen": 2517565440
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.000239447921681913,
+      "loss": 1.393,
+      "theoretical_loss": 3.395094965416407,
+      "tokens_seen": 2517696512
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002394077997111218,
+      "loss": 1.3375,
+      "theoretical_loss": 3.3950810051448412,
+      "tokens_seen": 2517827584
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002393676777403306,
+      "loss": 1.363,
+      "theoretical_loss": 3.3950670458034686,
+      "tokens_seen": 2517958656
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023932755576953942,
+      "loss": 1.3381,
+      "theoretical_loss": 3.3950530873921796,
+      "tokens_seen": 2518089728
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002392874337987482,
+      "loss": 1.2822,
+      "theoretical_loss": 3.3950391299108627,
+      "tokens_seen": 2518220800
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.000239247311827957,
+      "loss": 1.3761,
+      "theoretical_loss": 3.395025173359408,
+      "tokens_seen": 2518351872
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002392071898571658,
+      "loss": 1.2084,
+      "theoretical_loss": 3.3950112177377063,
+      "tokens_seen": 2518482944
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023916706788637456,
+      "loss": 1.3108,
+      "theoretical_loss": 3.3949972630456458,
+      "tokens_seen": 2518614016
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023912694591558337,
+      "loss": 1.267,
+      "theoretical_loss": 3.3949833092831163,
+      "tokens_seen": 2518745088
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023908682394479218,
+      "loss": 1.3627,
+      "theoretical_loss": 3.3949693564500087,
+      "tokens_seen": 2518876160
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023904670197400096,
+      "loss": 1.3503,
+      "theoretical_loss": 3.3949554045462116,
+      "tokens_seen": 2519007232
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023900658000320976,
+      "loss": 1.2904,
+      "theoretical_loss": 3.394941453571615,
+      "tokens_seen": 2519138304
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023896645803241857,
+      "loss": 1.2978,
+      "theoretical_loss": 3.394927503526109,
+      "tokens_seen": 2519269376
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023892633606162737,
+      "loss": 1.388,
+      "theoretical_loss": 3.394913554409583,
+      "tokens_seen": 2519400448
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023888621409083613,
+      "loss": 1.3125,
+      "theoretical_loss": 3.394899606221927,
+      "tokens_seen": 2519531520
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023884609212004493,
+      "loss": 1.3716,
+      "theoretical_loss": 3.394885658963031,
+      "tokens_seen": 2519662592
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023880597014925374,
+      "loss": 1.3404,
+      "theoretical_loss": 3.3948717126327845,
+      "tokens_seen": 2519793664
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023876584817846252,
+      "loss": 1.3265,
+      "theoretical_loss": 3.394857767231078,
+      "tokens_seen": 2519924736
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023872572620767132,
+      "loss": 1.2423,
+      "theoretical_loss": 3.3948438227578004,
+      "tokens_seen": 2520055808
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023868560423688013,
+      "loss": 1.3221,
+      "theoretical_loss": 3.3948298792128426,
+      "tokens_seen": 2520186880
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023864548226608894,
+      "loss": 1.3222,
+      "theoretical_loss": 3.3948159365960935,
+      "tokens_seen": 2520317952
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002386053602952977,
+      "loss": 1.3096,
+      "theoretical_loss": 3.3948019949074437,
+      "tokens_seen": 2520449024
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002385652383245065,
+      "loss": 1.3462,
+      "theoretical_loss": 3.3947880541467836,
+      "tokens_seen": 2520580096
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002385251163537153,
+      "loss": 1.3821,
+      "theoretical_loss": 3.3947741143140022,
+      "tokens_seen": 2520711168
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023848499438292408,
+      "loss": 1.3127,
+      "theoretical_loss": 3.3947601754089898,
+      "tokens_seen": 2520842240
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002384448724121329,
+      "loss": 1.3251,
+      "theoretical_loss": 3.3947462374316366,
+      "tokens_seen": 2520973312
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002384047504413417,
+      "loss": 1.3548,
+      "theoretical_loss": 3.394732300381833,
+      "tokens_seen": 2521104384
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002383646284705505,
+      "loss": 1.3643,
+      "theoretical_loss": 3.3947183642594685,
+      "tokens_seen": 2521235456
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023832450649975928,
+      "loss": 1.2853,
+      "theoretical_loss": 3.3947044290644333,
+      "tokens_seen": 2521366528
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023828438452896806,
+      "loss": 1.3817,
+      "theoretical_loss": 3.394690494796617,
+      "tokens_seen": 2521497600
+    },
+    {
+      "epoch": 0.53,
+      "objective/train/advantage_avg": 0.49406909942626953,
+      "objective/train/docs_used": 917863,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2106291055679321,
+      "objective/train/original_loss": 1.2106289863586426,
+      "objective/train/theoretical_loss": 3.394676561455911,
+      "objective/train/tokens_used": 892087776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24455241858959198,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506501197814941,
+      "objective/train/weighted_lm_loss": 1.2719998359680176,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.7788007855415344,
+      "theoretical_loss": 3.394676561455911,
+      "tokens_seen": 2521628672
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023824426255817686,
+      "loss": 1.3046,
+      "theoretical_loss": 3.394676561455911,
+      "tokens_seen": 2521628672
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023820414058738564,
+      "loss": 1.3458,
+      "theoretical_loss": 3.3946626290422044,
+      "tokens_seen": 2521759744
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023816401861659445,
+      "loss": 1.3295,
+      "theoretical_loss": 3.3946486975553873,
+      "tokens_seen": 2521890816
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023812389664580326,
+      "loss": 1.3177,
+      "theoretical_loss": 3.3946347669953507,
+      "tokens_seen": 2522021888
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023808377467501204,
+      "loss": 1.3302,
+      "theoretical_loss": 3.394620837361984,
+      "tokens_seen": 2522152960
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023804365270422084,
+      "loss": 1.3925,
+      "theoretical_loss": 3.3946069086551773,
+      "tokens_seen": 2522284032
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023800353073342962,
+      "loss": 1.293,
+      "theoretical_loss": 3.3945929808748216,
+      "tokens_seen": 2522415104
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023796340876263843,
+      "loss": 1.3708,
+      "theoretical_loss": 3.3945790540208067,
+      "tokens_seen": 2522546176
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002379232867918472,
+      "loss": 1.3433,
+      "theoretical_loss": 3.3945651280930225,
+      "tokens_seen": 2522677248
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.000237883164821056,
+      "loss": 1.3803,
+      "theoretical_loss": 3.3945512030913605,
+      "tokens_seen": 2522808320
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023784304285026482,
+      "loss": 1.2931,
+      "theoretical_loss": 3.3945372790157093,
+      "tokens_seen": 2522939392
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002378029208794736,
+      "loss": 1.3528,
+      "theoretical_loss": 3.39452335586596,
+      "tokens_seen": 2523070464
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002377627989086824,
+      "loss": 1.2859,
+      "theoretical_loss": 3.394509433642004,
+      "tokens_seen": 2523201536
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002377226769378912,
+      "loss": 1.3074,
+      "theoretical_loss": 3.39449551234373,
+      "tokens_seen": 2523332608
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002376825549671,
+      "loss": 1.3195,
+      "theoretical_loss": 3.394481591971029,
+      "tokens_seen": 2523463680
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023764243299630877,
+      "loss": 1.3815,
+      "theoretical_loss": 3.394467672523792,
+      "tokens_seen": 2523594752
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023760231102551758,
+      "loss": 1.3413,
+      "theoretical_loss": 3.394453754001908,
+      "tokens_seen": 2523725824
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023756218905472638,
+      "loss": 1.3516,
+      "theoretical_loss": 3.394439836405269,
+      "tokens_seen": 2523856896
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023752206708393516,
+      "loss": 1.4124,
+      "theoretical_loss": 3.3944259197337643,
+      "tokens_seen": 2523987968
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023748194511314397,
+      "loss": 1.3517,
+      "theoretical_loss": 3.3944120039872847,
+      "tokens_seen": 2524119040
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023744182314235277,
+      "loss": 1.3196,
+      "theoretical_loss": 3.394398089165721,
+      "tokens_seen": 2524250112
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023740170117156155,
+      "loss": 1.3745,
+      "theoretical_loss": 3.3943841752689634,
+      "tokens_seen": 2524381184
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023736157920077033,
+      "loss": 1.3521,
+      "theoretical_loss": 3.3943702622969028,
+      "tokens_seen": 2524512256
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023732145722997914,
+      "loss": 1.3166,
+      "theoretical_loss": 3.3943563502494287,
+      "tokens_seen": 2524643328
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023728133525918795,
+      "loss": 1.483,
+      "theoretical_loss": 3.394342439126433,
+      "tokens_seen": 2524774400
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023724121328839672,
+      "loss": 1.3578,
+      "theoretical_loss": 3.394328528927806,
+      "tokens_seen": 2524905472
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023720109131760553,
+      "loss": 1.3928,
+      "theoretical_loss": 3.3943146196534375,
+      "tokens_seen": 2525036544
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023716096934681434,
+      "loss": 1.3605,
+      "theoretical_loss": 3.3943007113032184,
+      "tokens_seen": 2525167616
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002371208473760231,
+      "loss": 1.4288,
+      "theoretical_loss": 3.3942868038770397,
+      "tokens_seen": 2525298688
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002370807254052319,
+      "loss": 1.3519,
+      "theoretical_loss": 3.3942728973747918,
+      "tokens_seen": 2525429760
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002370406034344407,
+      "loss": 1.3453,
+      "theoretical_loss": 3.3942589917963657,
+      "tokens_seen": 2525560832
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002370004814636495,
+      "loss": 1.2788,
+      "theoretical_loss": 3.394245087141652,
+      "tokens_seen": 2525691904
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002369603594928583,
+      "loss": 1.3965,
+      "theoretical_loss": 3.3942311834105405,
+      "tokens_seen": 2525822976
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002369202375220671,
+      "loss": 1.3489,
+      "theoretical_loss": 3.3942172806029234,
+      "tokens_seen": 2525954048
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002368801155512759,
+      "loss": 1.3767,
+      "theoretical_loss": 3.3942033787186903,
+      "tokens_seen": 2526085120
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023683999358048468,
+      "loss": 1.2849,
+      "theoretical_loss": 3.3941894777577324,
+      "tokens_seen": 2526216192
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023679987160969346,
+      "loss": 1.3366,
+      "theoretical_loss": 3.3941755777199405,
+      "tokens_seen": 2526347264
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023675974963890226,
+      "loss": 1.3054,
+      "theoretical_loss": 3.3941616786052053,
+      "tokens_seen": 2526478336
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023671962766811107,
+      "loss": 1.3583,
+      "theoretical_loss": 3.394147780413418,
+      "tokens_seen": 2526609408
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023667950569731985,
+      "loss": 1.3358,
+      "theoretical_loss": 3.3941338831444687,
+      "tokens_seen": 2526740480
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023663938372652866,
+      "loss": 1.3554,
+      "theoretical_loss": 3.3941199867982492,
+      "tokens_seen": 2526871552
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023659926175573746,
+      "loss": 1.313,
+      "theoretical_loss": 3.3941060913746495,
+      "tokens_seen": 2527002624
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023655913978494624,
+      "loss": 1.3269,
+      "theoretical_loss": 3.394092196873561,
+      "tokens_seen": 2527133696
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023651901781415502,
+      "loss": 1.3634,
+      "theoretical_loss": 3.3940783032948745,
+      "tokens_seen": 2527264768
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023647889584336383,
+      "loss": 1.2814,
+      "theoretical_loss": 3.3940644106384807,
+      "tokens_seen": 2527395840
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023643877387257263,
+      "loss": 1.3111,
+      "theoretical_loss": 3.3940505189042707,
+      "tokens_seen": 2527526912
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002363986519017814,
+      "loss": 1.2975,
+      "theoretical_loss": 3.394036628092136,
+      "tokens_seen": 2527657984
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023635852993099022,
+      "loss": 1.3254,
+      "theoretical_loss": 3.3940227382019668,
+      "tokens_seen": 2527789056
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023631840796019903,
+      "loss": 1.3181,
+      "theoretical_loss": 3.3940088492336544,
+      "tokens_seen": 2527920128
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002362782859894078,
+      "loss": 1.2911,
+      "theoretical_loss": 3.3939949611870897,
+      "tokens_seen": 2528051200
+    },
+    {
+      "epoch": 0.53,
+      "objective/train/advantage_avg": 0.4934902787208557,
+      "objective/train/docs_used": 920165,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.1657878160476685,
+      "objective/train/original_loss": 1.165787935256958,
+      "objective/train/theoretical_loss": 3.3939810740621636,
+      "objective/train/tokens_used": 898641376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24383139610290527,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050588607788086,
+      "objective/train/weighted_lm_loss": 1.2247909307479858,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3939810740621636,
+      "tokens_seen": 2528182272
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023623816401861658,
+      "loss": 1.2808,
+      "theoretical_loss": 3.3939810740621636,
+      "tokens_seen": 2528182272
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002361980420478254,
+      "loss": 1.3917,
+      "theoretical_loss": 3.393967187858768,
+      "tokens_seen": 2528313344
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002361579200770342,
+      "loss": 1.2868,
+      "theoretical_loss": 3.393953302576793,
+      "tokens_seen": 2528444416
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023611779810624298,
+      "loss": 1.3568,
+      "theoretical_loss": 3.3939394182161307,
+      "tokens_seen": 2528575488
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023607767613545178,
+      "loss": 1.344,
+      "theoretical_loss": 3.3939255347766712,
+      "tokens_seen": 2528706560
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002360375541646606,
+      "loss": 1.3219,
+      "theoretical_loss": 3.393911652258306,
+      "tokens_seen": 2528837632
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023599743219386937,
+      "loss": 1.3261,
+      "theoretical_loss": 3.393897770660926,
+      "tokens_seen": 2528968704
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023595731022307817,
+      "loss": 1.2893,
+      "theoretical_loss": 3.3938838899844237,
+      "tokens_seen": 2529099776
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023591718825228695,
+      "loss": 1.3196,
+      "theoretical_loss": 3.3938700102286887,
+      "tokens_seen": 2529230848
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023587706628149573,
+      "loss": 1.4045,
+      "theoretical_loss": 3.393856131393613,
+      "tokens_seen": 2529361920
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023583694431070454,
+      "loss": 1.3637,
+      "theoretical_loss": 3.393842253479087,
+      "tokens_seen": 2529492992
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023579682233991334,
+      "loss": 1.3468,
+      "theoretical_loss": 3.3938283764850032,
+      "tokens_seen": 2529624064
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023575670036912215,
+      "loss": 1.3298,
+      "theoretical_loss": 3.393814500411252,
+      "tokens_seen": 2529755136
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023571657839833093,
+      "loss": 1.3027,
+      "theoretical_loss": 3.3938006252577244,
+      "tokens_seen": 2529886208
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023567645642753974,
+      "loss": 1.3352,
+      "theoretical_loss": 3.393786751024313,
+      "tokens_seen": 2530017280
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023563633445674852,
+      "loss": 1.3601,
+      "theoretical_loss": 3.393772877710908,
+      "tokens_seen": 2530148352
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002355962124859573,
+      "loss": 1.3473,
+      "theoretical_loss": 3.3937590053174014,
+      "tokens_seen": 2530279424
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002355560905151661,
+      "loss": 1.3173,
+      "theoretical_loss": 3.3937451338436837,
+      "tokens_seen": 2530410496
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002355159685443749,
+      "loss": 1.3226,
+      "theoretical_loss": 3.3937312632896472,
+      "tokens_seen": 2530541568
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023547584657358371,
+      "loss": 1.3553,
+      "theoretical_loss": 3.3937173936551828,
+      "tokens_seen": 2530672640
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002354357246027925,
+      "loss": 1.3625,
+      "theoretical_loss": 3.393703524940182,
+      "tokens_seen": 2530803712
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002353956026320013,
+      "loss": 1.3348,
+      "theoretical_loss": 3.3936896571445363,
+      "tokens_seen": 2530934784
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002353554806612101,
+      "loss": 1.4069,
+      "theoretical_loss": 3.3936757902681367,
+      "tokens_seen": 2531065856
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023531535869041886,
+      "loss": 1.3225,
+      "theoretical_loss": 3.3936619243108757,
+      "tokens_seen": 2531196928
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023527523671962766,
+      "loss": 1.2948,
+      "theoretical_loss": 3.3936480592726435,
+      "tokens_seen": 2531328000
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023523511474883647,
+      "loss": 1.3054,
+      "theoretical_loss": 3.3936341951533326,
+      "tokens_seen": 2531459072
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023519499277804528,
+      "loss": 1.386,
+      "theoretical_loss": 3.3936203319528344,
+      "tokens_seen": 2531590144
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023515487080725406,
+      "loss": 1.3213,
+      "theoretical_loss": 3.39360646967104,
+      "tokens_seen": 2531721216
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023511474883646286,
+      "loss": 1.3963,
+      "theoretical_loss": 3.393592608307841,
+      "tokens_seen": 2531852288
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023507462686567167,
+      "loss": 1.3817,
+      "theoretical_loss": 3.393578747863129,
+      "tokens_seen": 2531983360
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023503450489488042,
+      "loss": 1.3496,
+      "theoretical_loss": 3.393564888336796,
+      "tokens_seen": 2532114432
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023499438292408923,
+      "loss": 1.3656,
+      "theoretical_loss": 3.3935510297287337,
+      "tokens_seen": 2532245504
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023495426095329803,
+      "loss": 1.4004,
+      "theoretical_loss": 3.393537172038833,
+      "tokens_seen": 2532376576
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002349141389825068,
+      "loss": 1.2858,
+      "theoretical_loss": 3.393523315266986,
+      "tokens_seen": 2532507648
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00023487401701171562,
+      "loss": 1.3315,
+      "theoretical_loss": 3.3935094594130844,
+      "tokens_seen": 2532638720
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023483389504092443,
+      "loss": 1.3121,
+      "theoretical_loss": 3.3934956044770197,
+      "tokens_seen": 2532769792
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023479377307013323,
+      "loss": 1.3044,
+      "theoretical_loss": 3.3934817504586836,
+      "tokens_seen": 2532900864
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023475365109934198,
+      "loss": 1.4252,
+      "theoretical_loss": 3.393467897357968,
+      "tokens_seen": 2533031936
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0002347135291285508,
+      "loss": 1.2937,
+      "theoretical_loss": 3.393454045174765,
+      "tokens_seen": 2533163008
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0002346734071577596,
+      "loss": 1.2706,
+      "theoretical_loss": 3.393440193908966,
+      "tokens_seen": 2533294080
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023463328518696838,
+      "loss": 1.346,
+      "theoretical_loss": 3.3934263435604617,
+      "tokens_seen": 2533425152
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023459316321617718,
+      "loss": 1.3321,
+      "theoretical_loss": 3.3934124941291457,
+      "tokens_seen": 2533556224
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.000234553041245386,
+      "loss": 1.3608,
+      "theoretical_loss": 3.393398645614909,
+      "tokens_seen": 2533687296
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0002345129192745948,
+      "loss": 1.3847,
+      "theoretical_loss": 3.3933847980176433,
+      "tokens_seen": 2533818368
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023447279730380357,
+      "loss": 1.3291,
+      "theoretical_loss": 3.393370951337241,
+      "tokens_seen": 2533949440
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023443267533301235,
+      "loss": 1.3224,
+      "theoretical_loss": 3.3933571055735934,
+      "tokens_seen": 2534080512
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023439255336222116,
+      "loss": 1.3216,
+      "theoretical_loss": 3.3933432607265925,
+      "tokens_seen": 2534211584
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023435243139142994,
+      "loss": 1.3157,
+      "theoretical_loss": 3.3933294167961305,
+      "tokens_seen": 2534342656
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023431230942063874,
+      "loss": 1.3621,
+      "theoretical_loss": 3.3933155737820986,
+      "tokens_seen": 2534473728
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023427218744984755,
+      "loss": 1.3044,
+      "theoretical_loss": 3.3933017316843896,
+      "tokens_seen": 2534604800
+    },
+    {
+      "epoch": 0.54,
+      "objective/train/advantage_avg": 0.493055522441864,
+      "objective/train/docs_used": 922493,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2998820543289185,
+      "objective/train/original_loss": 1.299881935119629,
+      "objective/train/theoretical_loss": 3.3932878905028954,
+      "objective/train/tokens_used": 905194976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2435069978237152,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505435466766357,
+      "objective/train/weighted_lm_loss": 1.3654881715774536,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3932878905028954,
+      "tokens_seen": 2534735872
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023423206547905636,
+      "loss": 1.3872,
+      "theoretical_loss": 3.3932878905028954,
+      "tokens_seen": 2534735872
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023419194350826514,
+      "loss": 1.3197,
+      "theoretical_loss": 3.3932740502375074,
+      "tokens_seen": 2534866944
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023415182153747392,
+      "loss": 1.2972,
+      "theoretical_loss": 3.3932602108881182,
+      "tokens_seen": 2534998016
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023411169956668272,
+      "loss": 1.3592,
+      "theoretical_loss": 3.3932463724546196,
+      "tokens_seen": 2535129088
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0002340715775958915,
+      "loss": 1.3003,
+      "theoretical_loss": 3.393232534936903,
+      "tokens_seen": 2535260160
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0002340314556251003,
+      "loss": 1.351,
+      "theoretical_loss": 3.3932186983348616,
+      "tokens_seen": 2535391232
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0002339913336543091,
+      "loss": 1.3074,
+      "theoretical_loss": 3.393204862648387,
+      "tokens_seen": 2535522304
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0002339512116835179,
+      "loss": 1.3414,
+      "theoretical_loss": 3.3931910278773705,
+      "tokens_seen": 2535653376
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0002339110897127267,
+      "loss": 1.2466,
+      "theoretical_loss": 3.3931771940217055,
+      "tokens_seen": 2535784448
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023387096774193548,
+      "loss": 1.2868,
+      "theoretical_loss": 3.3931633610812835,
+      "tokens_seen": 2535915520
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023383084577114428,
+      "loss": 1.3231,
+      "theoretical_loss": 3.3931495290559965,
+      "tokens_seen": 2536046592
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023379072380035306,
+      "loss": 1.3256,
+      "theoretical_loss": 3.3931356979457368,
+      "tokens_seen": 2536177664
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023375060182956187,
+      "loss": 1.4027,
+      "theoretical_loss": 3.3931218677503967,
+      "tokens_seen": 2536308736
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023371047985877068,
+      "loss": 1.3145,
+      "theoretical_loss": 3.3931080384698684,
+      "tokens_seen": 2536439808
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023367035788797946,
+      "loss": 1.338,
+      "theoretical_loss": 3.393094210104044,
+      "tokens_seen": 2536570880
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023363023591718826,
+      "loss": 1.244,
+      "theoretical_loss": 3.393080382652816,
+      "tokens_seen": 2536701952
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023359011394639707,
+      "loss": 1.3332,
+      "theoretical_loss": 3.3930665561160764,
+      "tokens_seen": 2536833024
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023354999197560585,
+      "loss": 1.349,
+      "theoretical_loss": 3.3930527304937175,
+      "tokens_seen": 2536964096
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023350987000481463,
+      "loss": 1.2359,
+      "theoretical_loss": 3.393038905785631,
+      "tokens_seen": 2537095168
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023346974803402343,
+      "loss": 1.3098,
+      "theoretical_loss": 3.393025081991711,
+      "tokens_seen": 2537226240
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023342962606323224,
+      "loss": 1.285,
+      "theoretical_loss": 3.393011259111848,
+      "tokens_seen": 2537357312
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023338950409244102,
+      "loss": 1.3329,
+      "theoretical_loss": 3.392997437145935,
+      "tokens_seen": 2537488384
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023334938212164982,
+      "loss": 1.2564,
+      "theoretical_loss": 3.3929836160938645,
+      "tokens_seen": 2537619456
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023330926015085863,
+      "loss": 1.3515,
+      "theoretical_loss": 3.392969795955529,
+      "tokens_seen": 2537750528
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0002332691381800674,
+      "loss": 1.3915,
+      "theoretical_loss": 3.39295597673082,
+      "tokens_seen": 2537881600
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0002332290162092762,
+      "loss": 1.2962,
+      "theoretical_loss": 3.392942158419631,
+      "tokens_seen": 2538012672
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.000233188894238485,
+      "loss": 1.2687,
+      "theoretical_loss": 3.392928341021854,
+      "tokens_seen": 2538143744
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0002331487722676938,
+      "loss": 1.3142,
+      "theoretical_loss": 3.3929145245373813,
+      "tokens_seen": 2538274816
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023310865029690258,
+      "loss": 1.2797,
+      "theoretical_loss": 3.392900708966106,
+      "tokens_seen": 2538405888
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0002330685283261114,
+      "loss": 1.2543,
+      "theoretical_loss": 3.39288689430792,
+      "tokens_seen": 2538536960
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0002330284063553202,
+      "loss": 1.2305,
+      "theoretical_loss": 3.3928730805627154,
+      "tokens_seen": 2538668032
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.000232988284384529,
+      "loss": 1.2912,
+      "theoretical_loss": 3.392859267730386,
+      "tokens_seen": 2538799104
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023294816241373775,
+      "loss": 1.3202,
+      "theoretical_loss": 3.392845455810823,
+      "tokens_seen": 2538930176
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023290804044294656,
+      "loss": 1.3297,
+      "theoretical_loss": 3.3928316448039206,
+      "tokens_seen": 2539061248
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023286791847215536,
+      "loss": 1.3548,
+      "theoretical_loss": 3.3928178347095694,
+      "tokens_seen": 2539192320
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023282779650136414,
+      "loss": 1.4097,
+      "theoretical_loss": 3.3928040255276635,
+      "tokens_seen": 2539323392
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023278767453057295,
+      "loss": 1.32,
+      "theoretical_loss": 3.392790217258095,
+      "tokens_seen": 2539454464
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023274755255978176,
+      "loss": 1.3196,
+      "theoretical_loss": 3.3927764099007565,
+      "tokens_seen": 2539585536
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023270743058899054,
+      "loss": 1.2217,
+      "theoretical_loss": 3.3927626034555405,
+      "tokens_seen": 2539716608
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023266730861819932,
+      "loss": 1.3144,
+      "theoretical_loss": 3.39274879792234,
+      "tokens_seen": 2539847680
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023262718664740812,
+      "loss": 1.3915,
+      "theoretical_loss": 3.3927349933010476,
+      "tokens_seen": 2539978752
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023258706467661693,
+      "loss": 1.2628,
+      "theoretical_loss": 3.392721189591556,
+      "tokens_seen": 2540109824
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0002325469427058257,
+      "loss": 1.4281,
+      "theoretical_loss": 3.392707386793758,
+      "tokens_seen": 2540240896
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0002325068207350345,
+      "loss": 1.3272,
+      "theoretical_loss": 3.3926935849075464,
+      "tokens_seen": 2540371968
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023246669876424332,
+      "loss": 1.3937,
+      "theoretical_loss": 3.3926797839328136,
+      "tokens_seen": 2540503040
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0002324265767934521,
+      "loss": 1.311,
+      "theoretical_loss": 3.3926659838694526,
+      "tokens_seen": 2540634112
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023238645482266088,
+      "loss": 1.2827,
+      "theoretical_loss": 3.3926521847173565,
+      "tokens_seen": 2540765184
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023234633285186968,
+      "loss": 1.2834,
+      "theoretical_loss": 3.3926383864764174,
+      "tokens_seen": 2540896256
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0002323062108810785,
+      "loss": 1.3073,
+      "theoretical_loss": 3.392624589146529,
+      "tokens_seen": 2541027328
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023226608891028727,
+      "loss": 1.3657,
+      "theoretical_loss": 3.3926107927275835,
+      "tokens_seen": 2541158400
+    },
+    {
+      "epoch": 0.54,
+      "objective/train/advantage_avg": 0.4934854507446289,
+      "objective/train/docs_used": 924855,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2131304740905762,
+      "objective/train/original_loss": 1.2131303548812866,
+      "objective/train/theoretical_loss": 3.3925969972194743,
+      "objective/train/tokens_used": 911748576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24387921392917633,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505883693695068,
+      "objective/train/weighted_lm_loss": 1.2744663953781128,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9834713935852051,
+      "theoretical_loss": 3.3925969972194743,
+      "tokens_seen": 2541289472
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023222596693949608,
+      "loss": 1.3498,
+      "theoretical_loss": 3.3925969972194743,
+      "tokens_seen": 2541289472
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023218584496870488,
+      "loss": 1.3509,
+      "theoretical_loss": 3.392583202622094,
+      "tokens_seen": 2541420544
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023214572299791366,
+      "loss": 1.249,
+      "theoretical_loss": 3.3925694089353353,
+      "tokens_seen": 2541551616
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023210560102712247,
+      "loss": 1.4007,
+      "theoretical_loss": 3.392555616159091,
+      "tokens_seen": 2541682688
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023206547905633125,
+      "loss": 1.3314,
+      "theoretical_loss": 3.392541824293255,
+      "tokens_seen": 2541813760
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023202535708554003,
+      "loss": 1.3504,
+      "theoretical_loss": 3.3925280333377197,
+      "tokens_seen": 2541944832
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023198523511474883,
+      "loss": 1.3218,
+      "theoretical_loss": 3.3925142432923776,
+      "tokens_seen": 2542075904
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023194511314395764,
+      "loss": 1.2684,
+      "theoretical_loss": 3.392500454157122,
+      "tokens_seen": 2542206976
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023190499117316645,
+      "loss": 1.3761,
+      "theoretical_loss": 3.392486665931847,
+      "tokens_seen": 2542338048
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023186486920237522,
+      "loss": 1.3479,
+      "theoretical_loss": 3.392472878616444,
+      "tokens_seen": 2542469120
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023182474723158403,
+      "loss": 1.3338,
+      "theoretical_loss": 3.392459092210807,
+      "tokens_seen": 2542600192
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0002317846252607928,
+      "loss": 1.353,
+      "theoretical_loss": 3.3924453067148286,
+      "tokens_seen": 2542731264
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0002317445032900016,
+      "loss": 1.3675,
+      "theoretical_loss": 3.3924315221284025,
+      "tokens_seen": 2542862336
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0002317043813192104,
+      "loss": 1.3852,
+      "theoretical_loss": 3.392417738451421,
+      "tokens_seen": 2542993408
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0002316642593484192,
+      "loss": 1.3104,
+      "theoretical_loss": 3.392403955683778,
+      "tokens_seen": 2543124480
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.000231624137377628,
+      "loss": 1.2551,
+      "theoretical_loss": 3.3923901738253663,
+      "tokens_seen": 2543255552
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0002315840154068368,
+      "loss": 1.3159,
+      "theoretical_loss": 3.3923763928760793,
+      "tokens_seen": 2543386624
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0002315438934360456,
+      "loss": 1.3534,
+      "theoretical_loss": 3.3923626128358095,
+      "tokens_seen": 2543517696
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023150377146525437,
+      "loss": 1.365,
+      "theoretical_loss": 3.3923488337044505,
+      "tokens_seen": 2543648768
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023146364949446315,
+      "loss": 1.366,
+      "theoretical_loss": 3.392335055481896,
+      "tokens_seen": 2543779840
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023142352752367196,
+      "loss": 1.33,
+      "theoretical_loss": 3.3923212781680387,
+      "tokens_seen": 2543910912
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023138340555288076,
+      "loss": 1.3388,
+      "theoretical_loss": 3.392307501762772,
+      "tokens_seen": 2544041984
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023134328358208957,
+      "loss": 1.3218,
+      "theoretical_loss": 3.392293726265989,
+      "tokens_seen": 2544173056
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023130316161129835,
+      "loss": 1.3107,
+      "theoretical_loss": 3.3922799516775832,
+      "tokens_seen": 2544304128
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023126303964050716,
+      "loss": 1.3371,
+      "theoretical_loss": 3.392266177997448,
+      "tokens_seen": 2544435200
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023122291766971596,
+      "loss": 1.352,
+      "theoretical_loss": 3.3922524052254763,
+      "tokens_seen": 2544566272
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023118279569892471,
+      "loss": 1.2965,
+      "theoretical_loss": 3.392238633361562,
+      "tokens_seen": 2544697344
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023114267372813352,
+      "loss": 1.2996,
+      "theoretical_loss": 3.392224862405598,
+      "tokens_seen": 2544828416
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023110255175734233,
+      "loss": 1.3738,
+      "theoretical_loss": 3.392211092357478,
+      "tokens_seen": 2544959488
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023106242978655113,
+      "loss": 1.2812,
+      "theoretical_loss": 3.392197323217095,
+      "tokens_seen": 2545090560
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0002310223078157599,
+      "loss": 1.2247,
+      "theoretical_loss": 3.392183554984343,
+      "tokens_seen": 2545221632
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023098218584496872,
+      "loss": 1.2174,
+      "theoretical_loss": 3.3921697876591153,
+      "tokens_seen": 2545352704
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023094206387417753,
+      "loss": 1.3201,
+      "theoretical_loss": 3.392156021241305,
+      "tokens_seen": 2545483776
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023090194190338628,
+      "loss": 1.3566,
+      "theoretical_loss": 3.3921422557308056,
+      "tokens_seen": 2545614848
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023086181993259508,
+      "loss": 1.2661,
+      "theoretical_loss": 3.3921284911275107,
+      "tokens_seen": 2545745920
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0002308216979618039,
+      "loss": 1.2967,
+      "theoretical_loss": 3.392114727431314,
+      "tokens_seen": 2545876992
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023078157599101267,
+      "loss": 1.2938,
+      "theoretical_loss": 3.3921009646421085,
+      "tokens_seen": 2546008064
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023074145402022148,
+      "loss": 1.3141,
+      "theoretical_loss": 3.3920872027597886,
+      "tokens_seen": 2546139136
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023070133204943028,
+      "loss": 1.3704,
+      "theoretical_loss": 3.3920734417842473,
+      "tokens_seen": 2546270208
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0002306612100786391,
+      "loss": 1.268,
+      "theoretical_loss": 3.392059681715378,
+      "tokens_seen": 2546401280
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023062108810784784,
+      "loss": 1.3166,
+      "theoretical_loss": 3.392045922553075,
+      "tokens_seen": 2546532352
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023058096613705665,
+      "loss": 1.3926,
+      "theoretical_loss": 3.392032164297231,
+      "tokens_seen": 2546663424
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023054084416626545,
+      "loss": 1.3196,
+      "theoretical_loss": 3.39201840694774,
+      "tokens_seen": 2546794496
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023050072219547423,
+      "loss": 1.2233,
+      "theoretical_loss": 3.392004650504496,
+      "tokens_seen": 2546925568
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023046060022468304,
+      "loss": 1.3544,
+      "theoretical_loss": 3.3919908949673925,
+      "tokens_seen": 2547056640
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023042047825389184,
+      "loss": 1.3374,
+      "theoretical_loss": 3.391977140336323,
+      "tokens_seen": 2547187712
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023038035628310065,
+      "loss": 1.3277,
+      "theoretical_loss": 3.391963386611181,
+      "tokens_seen": 2547318784
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023034023431230943,
+      "loss": 1.3114,
+      "theoretical_loss": 3.391949633791861,
+      "tokens_seen": 2547449856
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0002303001123415182,
+      "loss": 1.36,
+      "theoretical_loss": 3.391935881878256,
+      "tokens_seen": 2547580928
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023025999037072702,
+      "loss": 1.2499,
+      "theoretical_loss": 3.39192213087026,
+      "tokens_seen": 2547712000
+    },
+    {
+      "epoch": 0.54,
+      "objective/train/advantage_avg": 0.49075260758399963,
+      "objective/train/docs_used": 927107,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.1736656427383423,
+      "objective/train/original_loss": 1.1736655235290527,
+      "objective/train/theoretical_loss": 3.391908380767767,
+      "objective/train/tokens_used": 918302176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.241579070687294,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503032207489014,
+      "objective/train/weighted_lm_loss": 1.2327032089233398,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.391908380767767,
+      "tokens_seen": 2547843072
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0002302198683999358,
+      "loss": 1.2611,
+      "theoretical_loss": 3.391908380767767,
+      "tokens_seen": 2547843072
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0002301797464291446,
+      "loss": 1.312,
+      "theoretical_loss": 3.39189463157067,
+      "tokens_seen": 2547974144
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0002301396244583534,
+      "loss": 1.2271,
+      "theoretical_loss": 3.3918808832788643,
+      "tokens_seen": 2548105216
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023009950248756221,
+      "loss": 1.3108,
+      "theoretical_loss": 3.3918671358922423,
+      "tokens_seen": 2548236288
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.000230059380516771,
+      "loss": 1.3037,
+      "theoretical_loss": 3.3918533894106986,
+      "tokens_seen": 2548367360
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00023001925854597977,
+      "loss": 1.3124,
+      "theoretical_loss": 3.391839643834127,
+      "tokens_seen": 2548498432
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00022997913657518858,
+      "loss": 1.3816,
+      "theoretical_loss": 3.391825899162421,
+      "tokens_seen": 2548629504
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00022993901460439736,
+      "loss": 1.3239,
+      "theoretical_loss": 3.3918121553954745,
+      "tokens_seen": 2548760576
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00022989889263360616,
+      "loss": 1.3712,
+      "theoretical_loss": 3.391798412533182,
+      "tokens_seen": 2548891648
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00022985877066281497,
+      "loss": 1.3068,
+      "theoretical_loss": 3.391784670575437,
+      "tokens_seen": 2549022720
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00022981864869202375,
+      "loss": 1.2559,
+      "theoretical_loss": 3.3917709295221337,
+      "tokens_seen": 2549153792
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022977852672123256,
+      "loss": 1.2896,
+      "theoretical_loss": 3.3917571893731657,
+      "tokens_seen": 2549284864
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022973840475044136,
+      "loss": 1.3511,
+      "theoretical_loss": 3.3917434501284274,
+      "tokens_seen": 2549415936
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022969828277965014,
+      "loss": 1.3075,
+      "theoretical_loss": 3.3917297117878125,
+      "tokens_seen": 2549547008
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022965816080885892,
+      "loss": 1.2854,
+      "theoretical_loss": 3.391715974351215,
+      "tokens_seen": 2549678080
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022961803883806773,
+      "loss": 1.2965,
+      "theoretical_loss": 3.3917022378185293,
+      "tokens_seen": 2549809152
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022957791686727653,
+      "loss": 1.3152,
+      "theoretical_loss": 3.3916885021896492,
+      "tokens_seen": 2549940224
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002295377948964853,
+      "loss": 1.2958,
+      "theoretical_loss": 3.391674767464469,
+      "tokens_seen": 2550071296
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022949767292569412,
+      "loss": 1.3507,
+      "theoretical_loss": 3.3916610336428823,
+      "tokens_seen": 2550202368
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022945755095490293,
+      "loss": 1.3505,
+      "theoretical_loss": 3.391647300724784,
+      "tokens_seen": 2550333440
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002294174289841117,
+      "loss": 1.2935,
+      "theoretical_loss": 3.3916335687100676,
+      "tokens_seen": 2550464512
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022937730701332048,
+      "loss": 1.2632,
+      "theoretical_loss": 3.3916198375986273,
+      "tokens_seen": 2550595584
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002293371850425293,
+      "loss": 1.337,
+      "theoretical_loss": 3.3916061073903574,
+      "tokens_seen": 2550726656
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002292970630717381,
+      "loss": 1.2282,
+      "theoretical_loss": 3.3915923780851522,
+      "tokens_seen": 2550857728
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022925694110094688,
+      "loss": 1.3362,
+      "theoretical_loss": 3.3915786496829057,
+      "tokens_seen": 2550988800
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022921681913015568,
+      "loss": 1.3346,
+      "theoretical_loss": 3.3915649221835125,
+      "tokens_seen": 2551119872
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002291766971593645,
+      "loss": 1.3313,
+      "theoretical_loss": 3.3915511955868665,
+      "tokens_seen": 2551250944
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022913657518857327,
+      "loss": 1.2893,
+      "theoretical_loss": 3.391537469892862,
+      "tokens_seen": 2551382016
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022909645321778205,
+      "loss": 1.3082,
+      "theoretical_loss": 3.3915237451013933,
+      "tokens_seen": 2551513088
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022905633124699085,
+      "loss": 1.3259,
+      "theoretical_loss": 3.3915100212123543,
+      "tokens_seen": 2551644160
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022901620927619966,
+      "loss": 1.3192,
+      "theoretical_loss": 3.3914962982256402,
+      "tokens_seen": 2551775232
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022897608730540844,
+      "loss": 1.2825,
+      "theoretical_loss": 3.3914825761411445,
+      "tokens_seen": 2551906304
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022893596533461724,
+      "loss": 1.2171,
+      "theoretical_loss": 3.391468854958762,
+      "tokens_seen": 2552037376
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022889584336382605,
+      "loss": 1.2109,
+      "theoretical_loss": 3.391455134678387,
+      "tokens_seen": 2552168448
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022885572139303486,
+      "loss": 1.3069,
+      "theoretical_loss": 3.3914414152999144,
+      "tokens_seen": 2552299520
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002288155994222436,
+      "loss": 1.3117,
+      "theoretical_loss": 3.3914276968232375,
+      "tokens_seen": 2552430592
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022877547745145242,
+      "loss": 1.3181,
+      "theoretical_loss": 3.3914139792482514,
+      "tokens_seen": 2552561664
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022873535548066122,
+      "loss": 1.314,
+      "theoretical_loss": 3.39140026257485,
+      "tokens_seen": 2552692736
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022869523350987,
+      "loss": 1.2968,
+      "theoretical_loss": 3.3913865468029285,
+      "tokens_seen": 2552823808
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002286551115390788,
+      "loss": 1.3051,
+      "theoretical_loss": 3.3913728319323813,
+      "tokens_seen": 2552954880
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022861498956828761,
+      "loss": 1.2365,
+      "theoretical_loss": 3.391359117963102,
+      "tokens_seen": 2553085952
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002285748675974964,
+      "loss": 1.251,
+      "theoretical_loss": 3.391345404894986,
+      "tokens_seen": 2553217024
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022853474562670517,
+      "loss": 1.2449,
+      "theoretical_loss": 3.3913316927279276,
+      "tokens_seen": 2553348096
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022849462365591398,
+      "loss": 1.2641,
+      "theoretical_loss": 3.391317981461821,
+      "tokens_seen": 2553479168
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022845450168512278,
+      "loss": 1.2223,
+      "theoretical_loss": 3.3913042710965615,
+      "tokens_seen": 2553610240
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022841437971433156,
+      "loss": 1.3403,
+      "theoretical_loss": 3.391290561632043,
+      "tokens_seen": 2553741312
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022837425774354037,
+      "loss": 1.3049,
+      "theoretical_loss": 3.39127685306816,
+      "tokens_seen": 2553872384
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022833413577274918,
+      "loss": 1.29,
+      "theoretical_loss": 3.391263145404808,
+      "tokens_seen": 2554003456
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022829401380195796,
+      "loss": 1.3483,
+      "theoretical_loss": 3.391249438641881,
+      "tokens_seen": 2554134528
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022825389183116673,
+      "loss": 1.2767,
+      "theoretical_loss": 3.3912357327792737,
+      "tokens_seen": 2554265600
+    },
+    {
+      "epoch": 0.55,
+      "objective/train/advantage_avg": 0.4925300180912018,
+      "objective/train/docs_used": 929369,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2617762088775635,
+      "objective/train/original_loss": 1.2617762088775635,
+      "objective/train/theoretical_loss": 3.3912220278168803,
+      "objective/train/tokens_used": 924855776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2435230314731598,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504910945892334,
+      "objective/train/weighted_lm_loss": 1.3254189491271973,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3912220278168803,
+      "tokens_seen": 2554396672
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022821376986037554,
+      "loss": 1.2651,
+      "theoretical_loss": 3.3912220278168803,
+      "tokens_seen": 2554396672
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022817364788958435,
+      "loss": 1.2884,
+      "theoretical_loss": 3.3912083237545967,
+      "tokens_seen": 2554527744
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022813352591879313,
+      "loss": 1.3138,
+      "theoretical_loss": 3.3911946205923167,
+      "tokens_seen": 2554658816
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022809340394800193,
+      "loss": 1.2978,
+      "theoretical_loss": 3.391180918329935,
+      "tokens_seen": 2554789888
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022805328197721074,
+      "loss": 1.2443,
+      "theoretical_loss": 3.3911672169673466,
+      "tokens_seen": 2554920960
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022801316000641952,
+      "loss": 1.358,
+      "theoretical_loss": 3.391153516504446,
+      "tokens_seen": 2555052032
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022797303803562832,
+      "loss": 1.2941,
+      "theoretical_loss": 3.3911398169411284,
+      "tokens_seen": 2555183104
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002279329160648371,
+      "loss": 1.2903,
+      "theoretical_loss": 3.3911261182772883,
+      "tokens_seen": 2555314176
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022789279409404588,
+      "loss": 1.2958,
+      "theoretical_loss": 3.3911124205128207,
+      "tokens_seen": 2555445248
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002278526721232547,
+      "loss": 1.2733,
+      "theoretical_loss": 3.39109872364762,
+      "tokens_seen": 2555576320
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002278125501524635,
+      "loss": 1.3094,
+      "theoretical_loss": 3.391085027681582,
+      "tokens_seen": 2555707392
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002277724281816723,
+      "loss": 1.3103,
+      "theoretical_loss": 3.3910713326146,
+      "tokens_seen": 2555838464
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022773230621088108,
+      "loss": 1.358,
+      "theoretical_loss": 3.3910576384465703,
+      "tokens_seen": 2555969536
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002276921842400899,
+      "loss": 1.2876,
+      "theoretical_loss": 3.3910439451773873,
+      "tokens_seen": 2556100608
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022765206226929867,
+      "loss": 1.2887,
+      "theoretical_loss": 3.391030252806946,
+      "tokens_seen": 2556231680
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022761194029850745,
+      "loss": 1.2927,
+      "theoretical_loss": 3.391016561335141,
+      "tokens_seen": 2556362752
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022757181832771625,
+      "loss": 1.3281,
+      "theoretical_loss": 3.391002870761868,
+      "tokens_seen": 2556493824
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022753169635692506,
+      "loss": 1.3585,
+      "theoretical_loss": 3.3909891810870207,
+      "tokens_seen": 2556624896
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022749157438613386,
+      "loss": 1.3264,
+      "theoretical_loss": 3.390975492310495,
+      "tokens_seen": 2556755968
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022745145241534264,
+      "loss": 1.3157,
+      "theoretical_loss": 3.390961804432186,
+      "tokens_seen": 2556887040
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022741133044455145,
+      "loss": 1.2697,
+      "theoretical_loss": 3.3909481174519884,
+      "tokens_seen": 2557018112
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022737120847376026,
+      "loss": 1.3167,
+      "theoretical_loss": 3.3909344313697973,
+      "tokens_seen": 2557149184
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.000227331086502969,
+      "loss": 1.3264,
+      "theoretical_loss": 3.3909207461855075,
+      "tokens_seen": 2557280256
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022729096453217782,
+      "loss": 1.4254,
+      "theoretical_loss": 3.3909070618990147,
+      "tokens_seen": 2557411328
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022725084256138662,
+      "loss": 1.3012,
+      "theoretical_loss": 3.3908933785102136,
+      "tokens_seen": 2557542400
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022721072059059543,
+      "loss": 1.2399,
+      "theoretical_loss": 3.3908796960189993,
+      "tokens_seen": 2557673472
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002271705986198042,
+      "loss": 1.2991,
+      "theoretical_loss": 3.390866014425267,
+      "tokens_seen": 2557804544
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.000227130476649013,
+      "loss": 1.2298,
+      "theoretical_loss": 3.390852333728912,
+      "tokens_seen": 2557935616
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022709035467822182,
+      "loss": 1.2989,
+      "theoretical_loss": 3.3908386539298285,
+      "tokens_seen": 2558066688
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022705023270743057,
+      "loss": 1.3532,
+      "theoretical_loss": 3.3908249750279134,
+      "tokens_seen": 2558197760
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022701011073663938,
+      "loss": 1.2198,
+      "theoretical_loss": 3.39081129702306,
+      "tokens_seen": 2558328832
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022696998876584818,
+      "loss": 1.3279,
+      "theoretical_loss": 3.3907976199151655,
+      "tokens_seen": 2558459904
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.000226929866795057,
+      "loss": 1.3246,
+      "theoretical_loss": 3.3907839437041236,
+      "tokens_seen": 2558590976
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022688974482426577,
+      "loss": 1.2709,
+      "theoretical_loss": 3.39077026838983,
+      "tokens_seen": 2558722048
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022684962285347458,
+      "loss": 1.3027,
+      "theoretical_loss": 3.39075659397218,
+      "tokens_seen": 2558853120
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022680950088268338,
+      "loss": 1.3114,
+      "theoretical_loss": 3.390742920451069,
+      "tokens_seen": 2558984192
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022676937891189213,
+      "loss": 1.2363,
+      "theoretical_loss": 3.390729247826392,
+      "tokens_seen": 2559115264
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022672925694110094,
+      "loss": 1.3274,
+      "theoretical_loss": 3.390715576098045,
+      "tokens_seen": 2559246336
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022668913497030975,
+      "loss": 1.3181,
+      "theoretical_loss": 3.3907019052659226,
+      "tokens_seen": 2559377408
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022664901299951853,
+      "loss": 1.2712,
+      "theoretical_loss": 3.3906882353299204,
+      "tokens_seen": 2559508480
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022660889102872733,
+      "loss": 1.2768,
+      "theoretical_loss": 3.390674566289934,
+      "tokens_seen": 2559639552
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022656876905793614,
+      "loss": 1.2815,
+      "theoretical_loss": 3.390660898145858,
+      "tokens_seen": 2559770624
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022652864708714495,
+      "loss": 1.2712,
+      "theoretical_loss": 3.3906472308975886,
+      "tokens_seen": 2559901696
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022648852511635372,
+      "loss": 1.2916,
+      "theoretical_loss": 3.3906335645450216,
+      "tokens_seen": 2560032768
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002264484031455625,
+      "loss": 1.3214,
+      "theoretical_loss": 3.390619899088051,
+      "tokens_seen": 2560163840
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002264082811747713,
+      "loss": 1.2876,
+      "theoretical_loss": 3.390606234526574,
+      "tokens_seen": 2560294912
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002263681592039801,
+      "loss": 1.3088,
+      "theoretical_loss": 3.390592570860485,
+      "tokens_seen": 2560425984
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002263280372331889,
+      "loss": 1.3037,
+      "theoretical_loss": 3.3905789080896795,
+      "tokens_seen": 2560557056
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002262879152623977,
+      "loss": 1.3198,
+      "theoretical_loss": 3.390565246214053,
+      "tokens_seen": 2560688128
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002262477932916065,
+      "loss": 1.2859,
+      "theoretical_loss": 3.390551585233502,
+      "tokens_seen": 2560819200
+    },
+    {
+      "epoch": 0.55,
+      "objective/train/advantage_avg": 0.4932734966278076,
+      "objective/train/docs_used": 931622,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.0651435852050781,
+      "objective/train/original_loss": 1.065143346786499,
+      "objective/train/theoretical_loss": 3.3905379251479206,
+      "objective/train/tokens_used": 931409376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24377983808517456,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505666732788086,
+      "objective/train/weighted_lm_loss": 1.1189650297164917,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9048373699188232,
+      "theoretical_loss": 3.3905379251479206,
+      "tokens_seen": 2560950272
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002262076713208153,
+      "loss": 1.2105,
+      "theoretical_loss": 3.3905379251479206,
+      "tokens_seen": 2560950272
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022616754935002407,
+      "loss": 1.2964,
+      "theoretical_loss": 3.3905242659572052,
+      "tokens_seen": 2561081344
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022612742737923287,
+      "loss": 1.2242,
+      "theoretical_loss": 3.3905106076612515,
+      "tokens_seen": 2561212416
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022608730540844165,
+      "loss": 1.3129,
+      "theoretical_loss": 3.390496950259955,
+      "tokens_seen": 2561343488
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022604718343765046,
+      "loss": 1.3449,
+      "theoretical_loss": 3.390483293753211,
+      "tokens_seen": 2561474560
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022600706146685926,
+      "loss": 1.3584,
+      "theoretical_loss": 3.3904696381409147,
+      "tokens_seen": 2561605632
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022596693949606807,
+      "loss": 1.213,
+      "theoretical_loss": 3.390455983422963,
+      "tokens_seen": 2561736704
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022592681752527685,
+      "loss": 1.2735,
+      "theoretical_loss": 3.390442329599251,
+      "tokens_seen": 2561867776
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022588669555448563,
+      "loss": 1.3333,
+      "theoretical_loss": 3.390428676669674,
+      "tokens_seen": 2561998848
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022584657358369444,
+      "loss": 1.3062,
+      "theoretical_loss": 3.3904150246341285,
+      "tokens_seen": 2562129920
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022580645161290321,
+      "loss": 1.2326,
+      "theoretical_loss": 3.3904013734925096,
+      "tokens_seen": 2562260992
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022576632964211202,
+      "loss": 1.3023,
+      "theoretical_loss": 3.3903877232447126,
+      "tokens_seen": 2562392064
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022572620767132083,
+      "loss": 1.3106,
+      "theoretical_loss": 3.3903740738906345,
+      "tokens_seen": 2562523136
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002256860857005296,
+      "loss": 1.395,
+      "theoretical_loss": 3.3903604254301705,
+      "tokens_seen": 2562654208
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002256459637297384,
+      "loss": 1.2691,
+      "theoretical_loss": 3.3903467778632157,
+      "tokens_seen": 2562785280
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022560584175894722,
+      "loss": 1.2302,
+      "theoretical_loss": 3.390333131189667,
+      "tokens_seen": 2562916352
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.000225565719788156,
+      "loss": 1.3454,
+      "theoretical_loss": 3.39031948540942,
+      "tokens_seen": 2563047424
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022552559781736478,
+      "loss": 1.4121,
+      "theoretical_loss": 3.39030584052237,
+      "tokens_seen": 2563178496
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022548547584657358,
+      "loss": 1.3609,
+      "theoretical_loss": 3.390292196528413,
+      "tokens_seen": 2563309568
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002254453538757824,
+      "loss": 1.3217,
+      "theoretical_loss": 3.3902785534274456,
+      "tokens_seen": 2563440640
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022540523190499117,
+      "loss": 1.3403,
+      "theoretical_loss": 3.390264911219363,
+      "tokens_seen": 2563571712
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022536510993419998,
+      "loss": 1.3513,
+      "theoretical_loss": 3.390251269904061,
+      "tokens_seen": 2563702784
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022532498796340878,
+      "loss": 1.2987,
+      "theoretical_loss": 3.390237629481436,
+      "tokens_seen": 2563833856
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022528486599261756,
+      "loss": 1.3217,
+      "theoretical_loss": 3.3902239899513837,
+      "tokens_seen": 2563964928
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022524474402182634,
+      "loss": 1.2665,
+      "theoretical_loss": 3.3902103513138004,
+      "tokens_seen": 2564096000
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022520462205103515,
+      "loss": 1.3203,
+      "theoretical_loss": 3.3901967135685815,
+      "tokens_seen": 2564227072
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022516450008024395,
+      "loss": 1.3224,
+      "theoretical_loss": 3.3901830767156236,
+      "tokens_seen": 2564358144
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022512437810945273,
+      "loss": 1.432,
+      "theoretical_loss": 3.390169440754822,
+      "tokens_seen": 2564489216
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022508425613866154,
+      "loss": 1.2905,
+      "theoretical_loss": 3.390155805686074,
+      "tokens_seen": 2564620288
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022504413416787034,
+      "loss": 1.3342,
+      "theoretical_loss": 3.3901421715092743,
+      "tokens_seen": 2564751360
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022500401219707915,
+      "loss": 1.2423,
+      "theoretical_loss": 3.3901285382243196,
+      "tokens_seen": 2564882432
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002249638902262879,
+      "loss": 1.3538,
+      "theoretical_loss": 3.3901149058311058,
+      "tokens_seen": 2565013504
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002249237682554967,
+      "loss": 1.4297,
+      "theoretical_loss": 3.3901012743295293,
+      "tokens_seen": 2565144576
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022488364628470552,
+      "loss": 1.3317,
+      "theoretical_loss": 3.390087643719486,
+      "tokens_seen": 2565275648
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002248435243139143,
+      "loss": 1.3559,
+      "theoretical_loss": 3.390074014000872,
+      "tokens_seen": 2565406720
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002248034023431231,
+      "loss": 1.3544,
+      "theoretical_loss": 3.3900603851735838,
+      "tokens_seen": 2565537792
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002247632803723319,
+      "loss": 1.3677,
+      "theoretical_loss": 3.390046757237517,
+      "tokens_seen": 2565668864
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022472315840154071,
+      "loss": 1.3093,
+      "theoretical_loss": 3.390033130192568,
+      "tokens_seen": 2565799936
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022468303643074947,
+      "loss": 1.2201,
+      "theoretical_loss": 3.3900195040386336,
+      "tokens_seen": 2565931008
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022464291445995827,
+      "loss": 1.3184,
+      "theoretical_loss": 3.3900058787756095,
+      "tokens_seen": 2566062080
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022460279248916708,
+      "loss": 1.2693,
+      "theoretical_loss": 3.389992254403392,
+      "tokens_seen": 2566193152
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022456267051837586,
+      "loss": 1.3507,
+      "theoretical_loss": 3.389978630921877,
+      "tokens_seen": 2566324224
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022452254854758466,
+      "loss": 1.3327,
+      "theoretical_loss": 3.389965008330962,
+      "tokens_seen": 2566455296
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022448242657679347,
+      "loss": 1.2743,
+      "theoretical_loss": 3.3899513866305417,
+      "tokens_seen": 2566586368
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022444230460600225,
+      "loss": 1.3318,
+      "theoretical_loss": 3.3899377658205134,
+      "tokens_seen": 2566717440
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022440218263521103,
+      "loss": 1.3631,
+      "theoretical_loss": 3.389924145900773,
+      "tokens_seen": 2566848512
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022436206066441984,
+      "loss": 1.3971,
+      "theoretical_loss": 3.3899105268712173,
+      "tokens_seen": 2566979584
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022432193869362864,
+      "loss": 1.3361,
+      "theoretical_loss": 3.3898969087317425,
+      "tokens_seen": 2567110656
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022428181672283742,
+      "loss": 1.3615,
+      "theoretical_loss": 3.3898832914822448,
+      "tokens_seen": 2567241728
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022424169475204623,
+      "loss": 1.3006,
+      "theoretical_loss": 3.389869675122621,
+      "tokens_seen": 2567372800
+    },
+    {
+      "epoch": 0.56,
+      "objective/train/advantage_avg": 0.49043452739715576,
+      "objective/train/docs_used": 933971,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2099609375,
+      "objective/train/original_loss": 1.2099609375,
+      "objective/train/theoretical_loss": 3.389856059652767,
+      "objective/train/tokens_used": 937962976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24109162390232086,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0502690076828003,
+      "objective/train/weighted_lm_loss": 1.2706409692764282,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.389856059652767,
+      "tokens_seen": 2567503872
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022420157278125503,
+      "loss": 1.2744,
+      "theoretical_loss": 3.389856059652767,
+      "tokens_seen": 2567503872
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002241614508104638,
+      "loss": 1.4281,
+      "theoretical_loss": 3.3898424450725795,
+      "tokens_seen": 2567634944
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022412132883967262,
+      "loss": 1.3664,
+      "theoretical_loss": 3.389828831381955,
+      "tokens_seen": 2567766016
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002240812068688814,
+      "loss": 1.4039,
+      "theoretical_loss": 3.38981521858079,
+      "tokens_seen": 2567897088
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002240410848980902,
+      "loss": 1.2754,
+      "theoretical_loss": 3.3898016066689807,
+      "tokens_seen": 2568028160
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022400096292729898,
+      "loss": 1.3222,
+      "theoretical_loss": 3.389787995646424,
+      "tokens_seen": 2568159232
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002239608409565078,
+      "loss": 1.3492,
+      "theoretical_loss": 3.3897743855130162,
+      "tokens_seen": 2568290304
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002239207189857166,
+      "loss": 1.3998,
+      "theoretical_loss": 3.3897607762686537,
+      "tokens_seen": 2568421376
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022388059701492538,
+      "loss": 1.3539,
+      "theoretical_loss": 3.3897471679132334,
+      "tokens_seen": 2568552448
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022384047504413418,
+      "loss": 1.3916,
+      "theoretical_loss": 3.389733560446652,
+      "tokens_seen": 2568683520
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022380035307334296,
+      "loss": 1.3213,
+      "theoretical_loss": 3.389719953868805,
+      "tokens_seen": 2568814592
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022376023110255174,
+      "loss": 1.3507,
+      "theoretical_loss": 3.3897063481795904,
+      "tokens_seen": 2568945664
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022372010913176055,
+      "loss": 1.3696,
+      "theoretical_loss": 3.3896927433789044,
+      "tokens_seen": 2569076736
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022367998716096935,
+      "loss": 1.3556,
+      "theoretical_loss": 3.3896791394666432,
+      "tokens_seen": 2569207808
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022363986519017816,
+      "loss": 1.285,
+      "theoretical_loss": 3.389665536442704,
+      "tokens_seen": 2569338880
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022359974321938694,
+      "loss": 1.2659,
+      "theoretical_loss": 3.389651934306983,
+      "tokens_seen": 2569469952
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022355962124859574,
+      "loss": 1.3541,
+      "theoretical_loss": 3.3896383330593776,
+      "tokens_seen": 2569601024
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022351949927780452,
+      "loss": 1.3512,
+      "theoretical_loss": 3.389624732699784,
+      "tokens_seen": 2569732096
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002234793773070133,
+      "loss": 1.4533,
+      "theoretical_loss": 3.3896111332280983,
+      "tokens_seen": 2569863168
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002234392553362221,
+      "loss": 1.3622,
+      "theoretical_loss": 3.389597534644219,
+      "tokens_seen": 2569994240
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022339913336543092,
+      "loss": 1.3685,
+      "theoretical_loss": 3.389583936948041,
+      "tokens_seen": 2570125312
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022335901139463972,
+      "loss": 1.3193,
+      "theoretical_loss": 3.389570340139462,
+      "tokens_seen": 2570256384
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002233188894238485,
+      "loss": 1.3869,
+      "theoretical_loss": 3.3895567442183787,
+      "tokens_seen": 2570387456
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002232787674530573,
+      "loss": 1.3805,
+      "theoretical_loss": 3.3895431491846884,
+      "tokens_seen": 2570518528
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022323864548226611,
+      "loss": 1.3269,
+      "theoretical_loss": 3.389529555038287,
+      "tokens_seen": 2570649600
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022319852351147487,
+      "loss": 1.2904,
+      "theoretical_loss": 3.3895159617790718,
+      "tokens_seen": 2570780672
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022315840154068367,
+      "loss": 1.4826,
+      "theoretical_loss": 3.38950236940694,
+      "tokens_seen": 2570911744
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022311827956989248,
+      "loss": 1.3169,
+      "theoretical_loss": 3.389488777921788,
+      "tokens_seen": 2571042816
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022307815759910128,
+      "loss": 1.3395,
+      "theoretical_loss": 3.3894751873235127,
+      "tokens_seen": 2571173888
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022303803562831006,
+      "loss": 1.2815,
+      "theoretical_loss": 3.3894615976120113,
+      "tokens_seen": 2571304960
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022299791365751887,
+      "loss": 1.3769,
+      "theoretical_loss": 3.3894480087871806,
+      "tokens_seen": 2571436032
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022295779168672768,
+      "loss": 1.3794,
+      "theoretical_loss": 3.389434420848918,
+      "tokens_seen": 2571567104
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022291766971593643,
+      "loss": 1.4027,
+      "theoretical_loss": 3.3894208337971197,
+      "tokens_seen": 2571698176
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022287754774514523,
+      "loss": 1.328,
+      "theoretical_loss": 3.389407247631683,
+      "tokens_seen": 2571829248
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022283742577435404,
+      "loss": 1.404,
+      "theoretical_loss": 3.3893936623525054,
+      "tokens_seen": 2571960320
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022279730380356285,
+      "loss": 1.3821,
+      "theoretical_loss": 3.3893800779594834,
+      "tokens_seen": 2572091392
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022275718183277163,
+      "loss": 1.334,
+      "theoretical_loss": 3.3893664944525135,
+      "tokens_seen": 2572222464
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022271705986198043,
+      "loss": 1.3444,
+      "theoretical_loss": 3.389352911831494,
+      "tokens_seen": 2572353536
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022267693789118924,
+      "loss": 1.3674,
+      "theoretical_loss": 3.3893393300963215,
+      "tokens_seen": 2572484608
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022263681592039802,
+      "loss": 1.3712,
+      "theoretical_loss": 3.3893257492468924,
+      "tokens_seen": 2572615680
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002225966939496068,
+      "loss": 1.357,
+      "theoretical_loss": 3.3893121692831047,
+      "tokens_seen": 2572746752
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002225565719788156,
+      "loss": 1.3311,
+      "theoretical_loss": 3.3892985902048554,
+      "tokens_seen": 2572877824
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022251645000802438,
+      "loss": 1.2772,
+      "theoretical_loss": 3.3892850120120412,
+      "tokens_seen": 2573008896
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002224763280372332,
+      "loss": 1.3636,
+      "theoretical_loss": 3.3892714347045594,
+      "tokens_seen": 2573139968
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.000222436206066442,
+      "loss": 1.3724,
+      "theoretical_loss": 3.3892578582823076,
+      "tokens_seen": 2573271040
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002223960840956508,
+      "loss": 1.354,
+      "theoretical_loss": 3.3892442827451825,
+      "tokens_seen": 2573402112
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022235596212485958,
+      "loss": 1.33,
+      "theoretical_loss": 3.389230708093082,
+      "tokens_seen": 2573533184
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022231584015406836,
+      "loss": 1.3382,
+      "theoretical_loss": 3.3892171343259023,
+      "tokens_seen": 2573664256
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022227571818327717,
+      "loss": 1.3619,
+      "theoretical_loss": 3.3892035614435416,
+      "tokens_seen": 2573795328
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022223559621248595,
+      "loss": 1.4057,
+      "theoretical_loss": 3.3891899894458968,
+      "tokens_seen": 2573926400
+    },
+    {
+      "epoch": 0.56,
+      "objective/train/advantage_avg": 0.4935244917869568,
+      "objective/train/docs_used": 936262,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2349026203155518,
+      "objective/train/original_loss": 1.2349026203155518,
+      "objective/train/theoretical_loss": 3.3891764183328648,
+      "objective/train/tokens_used": 944516576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24397717416286469,
+      "objective/train/value_max": -0.5,
+      "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": 1.2973732948303223,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3891764183328648,
+      "tokens_seen": 2574057472
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022219547424169475,
+      "loss": 1.3087,
+      "theoretical_loss": 3.3891764183328648,
+      "tokens_seen": 2574057472
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022215535227090356,
+      "loss": 1.3007,
+      "theoretical_loss": 3.3891628481043434,
+      "tokens_seen": 2574188544
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022211523030011236,
+      "loss": 1.4112,
+      "theoretical_loss": 3.3891492787602298,
+      "tokens_seen": 2574319616
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022207510832932114,
+      "loss": 1.3129,
+      "theoretical_loss": 3.389135710300421,
+      "tokens_seen": 2574450688
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022203498635852992,
+      "loss": 1.3056,
+      "theoretical_loss": 3.389122142724815,
+      "tokens_seen": 2574581760
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022199486438773873,
+      "loss": 1.2966,
+      "theoretical_loss": 3.389108576033309,
+      "tokens_seen": 2574712832
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002219547424169475,
+      "loss": 1.4164,
+      "theoretical_loss": 3.3890950102258,
+      "tokens_seen": 2574843904
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022191462044615632,
+      "loss": 1.3391,
+      "theoretical_loss": 3.3890814453021854,
+      "tokens_seen": 2574974976
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022187449847536512,
+      "loss": 1.3689,
+      "theoretical_loss": 3.3890678812623634,
+      "tokens_seen": 2575106048
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022183437650457393,
+      "loss": 1.2823,
+      "theoretical_loss": 3.3890543181062305,
+      "tokens_seen": 2575237120
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002217942545337827,
+      "loss": 1.3172,
+      "theoretical_loss": 3.3890407558336846,
+      "tokens_seen": 2575368192
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002217541325629915,
+      "loss": 1.3942,
+      "theoretical_loss": 3.3890271944446226,
+      "tokens_seen": 2575499264
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002217140105922003,
+      "loss": 1.4026,
+      "theoretical_loss": 3.3890136339389434,
+      "tokens_seen": 2575630336
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022167388862140907,
+      "loss": 1.3231,
+      "theoretical_loss": 3.389000074316543,
+      "tokens_seen": 2575761408
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022163376665061788,
+      "loss": 1.3534,
+      "theoretical_loss": 3.3889865155773196,
+      "tokens_seen": 2575892480
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022159364467982668,
+      "loss": 1.2925,
+      "theoretical_loss": 3.388972957721171,
+      "tokens_seen": 2576023552
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022155352270903546,
+      "loss": 1.3255,
+      "theoretical_loss": 3.3889594007479946,
+      "tokens_seen": 2576154624
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022151340073824427,
+      "loss": 1.3374,
+      "theoretical_loss": 3.388945844657687,
+      "tokens_seen": 2576285696
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022147327876745308,
+      "loss": 1.3497,
+      "theoretical_loss": 3.388932289450147,
+      "tokens_seen": 2576416768
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022143315679666186,
+      "loss": 1.351,
+      "theoretical_loss": 3.388918735125272,
+      "tokens_seen": 2576547840
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022139303482587063,
+      "loss": 1.3433,
+      "theoretical_loss": 3.3889051816829596,
+      "tokens_seen": 2576678912
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022135291285507944,
+      "loss": 1.3583,
+      "theoretical_loss": 3.3888916291231066,
+      "tokens_seen": 2576809984
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022131279088428825,
+      "loss": 1.3793,
+      "theoretical_loss": 3.3888780774456118,
+      "tokens_seen": 2576941056
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022127266891349703,
+      "loss": 1.3671,
+      "theoretical_loss": 3.388864526650372,
+      "tokens_seen": 2577072128
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022123254694270583,
+      "loss": 1.3331,
+      "theoretical_loss": 3.388850976737286,
+      "tokens_seen": 2577203200
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022119242497191464,
+      "loss": 1.379,
+      "theoretical_loss": 3.3888374277062505,
+      "tokens_seen": 2577334272
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022115230300112342,
+      "loss": 1.1932,
+      "theoretical_loss": 3.3888238795571635,
+      "tokens_seen": 2577465344
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002211121810303322,
+      "loss": 1.2713,
+      "theoretical_loss": 3.3888103322899226,
+      "tokens_seen": 2577596416
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.000221072059059541,
+      "loss": 1.3104,
+      "theoretical_loss": 3.388796785904426,
+      "tokens_seen": 2577727488
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002210319370887498,
+      "loss": 1.3986,
+      "theoretical_loss": 3.388783240400571,
+      "tokens_seen": 2577858560
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002209918151179586,
+      "loss": 1.355,
+      "theoretical_loss": 3.388769695778256,
+      "tokens_seen": 2577989632
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002209516931471674,
+      "loss": 1.3631,
+      "theoretical_loss": 3.388756152037378,
+      "tokens_seen": 2578120704
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002209115711763762,
+      "loss": 1.4275,
+      "theoretical_loss": 3.388742609177836,
+      "tokens_seen": 2578251776
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.000220871449205585,
+      "loss": 1.4002,
+      "theoretical_loss": 3.388729067199526,
+      "tokens_seen": 2578382848
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022083132723479376,
+      "loss": 1.3863,
+      "theoretical_loss": 3.388715526102348,
+      "tokens_seen": 2578513920
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022079120526400257,
+      "loss": 1.3322,
+      "theoretical_loss": 3.3887019858861986,
+      "tokens_seen": 2578644992
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022075108329321137,
+      "loss": 1.3565,
+      "theoretical_loss": 3.388688446550976,
+      "tokens_seen": 2578776064
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022071096132242015,
+      "loss": 1.3152,
+      "theoretical_loss": 3.3886749080965775,
+      "tokens_seen": 2578907136
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022067083935162896,
+      "loss": 1.2878,
+      "theoretical_loss": 3.388661370522902,
+      "tokens_seen": 2579038208
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022063071738083776,
+      "loss": 1.2936,
+      "theoretical_loss": 3.3886478338298476,
+      "tokens_seen": 2579169280
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022059059541004654,
+      "loss": 1.3563,
+      "theoretical_loss": 3.3886342980173114,
+      "tokens_seen": 2579300352
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022055047343925532,
+      "loss": 1.3306,
+      "theoretical_loss": 3.3886207630851914,
+      "tokens_seen": 2579431424
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022051035146846413,
+      "loss": 1.3763,
+      "theoretical_loss": 3.3886072290333864,
+      "tokens_seen": 2579562496
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022047022949767294,
+      "loss": 1.3212,
+      "theoretical_loss": 3.3885936958617933,
+      "tokens_seen": 2579693568
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022043010752688171,
+      "loss": 1.2819,
+      "theoretical_loss": 3.388580163570311,
+      "tokens_seen": 2579824640
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022038998555609052,
+      "loss": 1.2592,
+      "theoretical_loss": 3.388566632158838,
+      "tokens_seen": 2579955712
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022034986358529933,
+      "loss": 1.2271,
+      "theoretical_loss": 3.388553101627271,
+      "tokens_seen": 2580086784
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002203097416145081,
+      "loss": 1.3495,
+      "theoretical_loss": 3.3885395719755094,
+      "tokens_seen": 2580217856
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002202696196437169,
+      "loss": 1.2626,
+      "theoretical_loss": 3.3885260432034503,
+      "tokens_seen": 2580348928
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002202294976729257,
+      "loss": 1.4407,
+      "theoretical_loss": 3.388512515310992,
+      "tokens_seen": 2580480000
+    },
+    {
+      "epoch": 0.56,
+      "objective/train/advantage_avg": 0.49107974767684937,
+      "objective/train/docs_used": 938550,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.1934958696365356,
+      "objective/train/original_loss": 1.1934958696365356,
+      "objective/train/theoretical_loss": 3.388498988298033,
+      "objective/train/tokens_used": 951070176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24186071753501892,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503374338150024,
+      "objective/train/weighted_lm_loss": 1.2534502744674683,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.388498988298033,
+      "tokens_seen": 2580611072
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002201893757021345,
+      "loss": 1.2657,
+      "theoretical_loss": 3.388498988298033,
+      "tokens_seen": 2580611072
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022014925373134328,
+      "loss": 1.3444,
+      "theoretical_loss": 3.3884854621644713,
+      "tokens_seen": 2580742144
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022010913176055208,
+      "loss": 1.3352,
+      "theoretical_loss": 3.3884719369102054,
+      "tokens_seen": 2580873216
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002200690097897609,
+      "loss": 1.3396,
+      "theoretical_loss": 3.3884584125351327,
+      "tokens_seen": 2581004288
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022002888781896967,
+      "loss": 1.3262,
+      "theoretical_loss": 3.3884448890391523,
+      "tokens_seen": 2581135360
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00021998876584817848,
+      "loss": 1.328,
+      "theoretical_loss": 3.3884313664221617,
+      "tokens_seen": 2581266432
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00021994864387738725,
+      "loss": 1.3813,
+      "theoretical_loss": 3.388417844684059,
+      "tokens_seen": 2581397504
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00021990852190659606,
+      "loss": 1.2396,
+      "theoretical_loss": 3.388404323824744,
+      "tokens_seen": 2581528576
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00021986839993580484,
+      "loss": 1.3781,
+      "theoretical_loss": 3.3883908038441133,
+      "tokens_seen": 2581659648
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00021982827796501365,
+      "loss": 1.4435,
+      "theoretical_loss": 3.3883772847420657,
+      "tokens_seen": 2581790720
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00021978815599422245,
+      "loss": 1.3777,
+      "theoretical_loss": 3.3883637665184994,
+      "tokens_seen": 2581921792
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00021974803402343123,
+      "loss": 1.3752,
+      "theoretical_loss": 3.388350249173313,
+      "tokens_seen": 2582052864
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00021970791205264004,
+      "loss": 1.327,
+      "theoretical_loss": 3.3883367327064047,
+      "tokens_seen": 2582183936
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021966779008184882,
+      "loss": 1.3595,
+      "theoretical_loss": 3.388323217117673,
+      "tokens_seen": 2582315008
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002196276681110576,
+      "loss": 1.2839,
+      "theoretical_loss": 3.3883097024070157,
+      "tokens_seen": 2582446080
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002195875461402664,
+      "loss": 1.3825,
+      "theoretical_loss": 3.388296188574332,
+      "tokens_seen": 2582577152
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002195474241694752,
+      "loss": 1.307,
+      "theoretical_loss": 3.38828267561952,
+      "tokens_seen": 2582708224
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021950730219868402,
+      "loss": 1.3253,
+      "theoretical_loss": 3.388269163542478,
+      "tokens_seen": 2582839296
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002194671802278928,
+      "loss": 1.2433,
+      "theoretical_loss": 3.3882556523431044,
+      "tokens_seen": 2582970368
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002194270582571016,
+      "loss": 1.2722,
+      "theoretical_loss": 3.388242142021298,
+      "tokens_seen": 2583101440
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002193869362863104,
+      "loss": 1.3253,
+      "theoretical_loss": 3.388228632576957,
+      "tokens_seen": 2583232512
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021934681431551916,
+      "loss": 1.2918,
+      "theoretical_loss": 3.38821512400998,
+      "tokens_seen": 2583363584
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021930669234472797,
+      "loss": 1.3015,
+      "theoretical_loss": 3.388201616320265,
+      "tokens_seen": 2583494656
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021926657037393677,
+      "loss": 1.3044,
+      "theoretical_loss": 3.3881881095077113,
+      "tokens_seen": 2583625728
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021922644840314558,
+      "loss": 1.3207,
+      "theoretical_loss": 3.3881746035722173,
+      "tokens_seen": 2583756800
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021918632643235436,
+      "loss": 1.2769,
+      "theoretical_loss": 3.388161098513681,
+      "tokens_seen": 2583887872
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021914620446156316,
+      "loss": 1.386,
+      "theoretical_loss": 3.3881475943320014,
+      "tokens_seen": 2584018944
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021910608249077197,
+      "loss": 1.3326,
+      "theoretical_loss": 3.3881340910270774,
+      "tokens_seen": 2584150016
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021906596051998072,
+      "loss": 1.3203,
+      "theoretical_loss": 3.3881205885988073,
+      "tokens_seen": 2584281088
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021902583854918953,
+      "loss": 1.3426,
+      "theoretical_loss": 3.3881070870470893,
+      "tokens_seen": 2584412160
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021898571657839834,
+      "loss": 1.3459,
+      "theoretical_loss": 3.3880935863718227,
+      "tokens_seen": 2584543232
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021894559460760714,
+      "loss": 1.2564,
+      "theoretical_loss": 3.3880800865729057,
+      "tokens_seen": 2584674304
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021890547263681592,
+      "loss": 1.2572,
+      "theoretical_loss": 3.388066587650237,
+      "tokens_seen": 2584805376
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021886535066602473,
+      "loss": 1.4025,
+      "theoretical_loss": 3.3880530896037158,
+      "tokens_seen": 2584936448
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021882522869523353,
+      "loss": 1.2487,
+      "theoretical_loss": 3.3880395924332407,
+      "tokens_seen": 2585067520
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021878510672444229,
+      "loss": 1.3181,
+      "theoretical_loss": 3.38802609613871,
+      "tokens_seen": 2585198592
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002187449847536511,
+      "loss": 1.3442,
+      "theoretical_loss": 3.3880126007200224,
+      "tokens_seen": 2585329664
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002187048627828599,
+      "loss": 1.4032,
+      "theoretical_loss": 3.3879991061770776,
+      "tokens_seen": 2585460736
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002186647408120687,
+      "loss": 1.3741,
+      "theoretical_loss": 3.3879856125097727,
+      "tokens_seen": 2585591808
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021862461884127748,
+      "loss": 1.3024,
+      "theoretical_loss": 3.3879721197180084,
+      "tokens_seen": 2585722880
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002185844968704863,
+      "loss": 1.2903,
+      "theoretical_loss": 3.387958627801682,
+      "tokens_seen": 2585853952
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002185443748996951,
+      "loss": 1.2811,
+      "theoretical_loss": 3.387945136760693,
+      "tokens_seen": 2585985024
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021850425292890388,
+      "loss": 1.3334,
+      "theoretical_loss": 3.3879316465949407,
+      "tokens_seen": 2586116096
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021846413095811265,
+      "loss": 1.3021,
+      "theoretical_loss": 3.387918157304323,
+      "tokens_seen": 2586247168
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021842400898732146,
+      "loss": 1.3202,
+      "theoretical_loss": 3.3879046688887393,
+      "tokens_seen": 2586378240
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021838388701653024,
+      "loss": 1.256,
+      "theoretical_loss": 3.3878911813480883,
+      "tokens_seen": 2586509312
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021834376504573905,
+      "loss": 1.3167,
+      "theoretical_loss": 3.3878776946822695,
+      "tokens_seen": 2586640384
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021830364307494785,
+      "loss": 1.4646,
+      "theoretical_loss": 3.387864208891181,
+      "tokens_seen": 2586771456
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021826352110415666,
+      "loss": 1.3201,
+      "theoretical_loss": 3.387850723974722,
+      "tokens_seen": 2586902528
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021822339913336544,
+      "loss": 1.4055,
+      "theoretical_loss": 3.387837239932792,
+      "tokens_seen": 2587033600
+    },
+    {
+      "epoch": 0.57,
+      "objective/train/advantage_avg": 0.491914838552475,
+      "objective/train/docs_used": 940829,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2808787822723389,
+      "objective/train/original_loss": 1.2808787822723389,
+      "objective/train/theoretical_loss": 3.387823756765289,
+      "objective/train/tokens_used": 957623776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24257121980190277,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050424575805664,
+      "objective/train/weighted_lm_loss": 1.3453412055969238,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.387823756765289,
+      "tokens_seen": 2587164672
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021818327716257422,
+      "loss": 1.3057,
+      "theoretical_loss": 3.387823756765289,
+      "tokens_seen": 2587164672
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021814315519178302,
+      "loss": 1.3476,
+      "theoretical_loss": 3.387810274472113,
+      "tokens_seen": 2587295744
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002181030332209918,
+      "loss": 1.3006,
+      "theoretical_loss": 3.3877967930531625,
+      "tokens_seen": 2587426816
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002180629112502006,
+      "loss": 1.2832,
+      "theoretical_loss": 3.3877833125083368,
+      "tokens_seen": 2587557888
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021802278927940942,
+      "loss": 1.3265,
+      "theoretical_loss": 3.3877698328375345,
+      "tokens_seen": 2587688960
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021798266730861822,
+      "loss": 1.3557,
+      "theoretical_loss": 3.387756354040655,
+      "tokens_seen": 2587820032
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.000217942545337827,
+      "loss": 1.2725,
+      "theoretical_loss": 3.3877428761175974,
+      "tokens_seen": 2587951104
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002179024233670358,
+      "loss": 1.3504,
+      "theoretical_loss": 3.3877293990682604,
+      "tokens_seen": 2588082176
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002178623013962446,
+      "loss": 1.2979,
+      "theoretical_loss": 3.3877159228925433,
+      "tokens_seen": 2588213248
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021782217942545337,
+      "loss": 1.3233,
+      "theoretical_loss": 3.3877024475903457,
+      "tokens_seen": 2588344320
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021778205745466217,
+      "loss": 1.3718,
+      "theoretical_loss": 3.3876889731615663,
+      "tokens_seen": 2588475392
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021774193548387098,
+      "loss": 1.323,
+      "theoretical_loss": 3.3876754996061047,
+      "tokens_seen": 2588606464
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021770181351307978,
+      "loss": 1.2482,
+      "theoretical_loss": 3.3876620269238593,
+      "tokens_seen": 2588737536
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021766169154228856,
+      "loss": 1.4302,
+      "theoretical_loss": 3.3876485551147297,
+      "tokens_seen": 2588868608
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021762156957149737,
+      "loss": 1.3412,
+      "theoretical_loss": 3.3876350841786156,
+      "tokens_seen": 2588999680
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021758144760070615,
+      "loss": 1.343,
+      "theoretical_loss": 3.3876216141154156,
+      "tokens_seen": 2589130752
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021754132562991493,
+      "loss": 1.3173,
+      "theoretical_loss": 3.3876081449250286,
+      "tokens_seen": 2589261824
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021750120365912373,
+      "loss": 1.3862,
+      "theoretical_loss": 3.387594676607355,
+      "tokens_seen": 2589392896
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021746108168833254,
+      "loss": 1.3406,
+      "theoretical_loss": 3.3875812091622937,
+      "tokens_seen": 2589523968
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021742095971754132,
+      "loss": 1.3877,
+      "theoretical_loss": 3.3875677425897432,
+      "tokens_seen": 2589655040
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021738083774675013,
+      "loss": 1.4005,
+      "theoretical_loss": 3.387554276889604,
+      "tokens_seen": 2589786112
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021734071577595893,
+      "loss": 1.3167,
+      "theoretical_loss": 3.3875408120617743,
+      "tokens_seen": 2589917184
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002173005938051677,
+      "loss": 1.3341,
+      "theoretical_loss": 3.387527348106154,
+      "tokens_seen": 2590048256
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002172604718343765,
+      "loss": 1.3748,
+      "theoretical_loss": 3.387513885022643,
+      "tokens_seen": 2590179328
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002172203498635853,
+      "loss": 1.3301,
+      "theoretical_loss": 3.3875004228111396,
+      "tokens_seen": 2590310400
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002171802278927941,
+      "loss": 1.3115,
+      "theoretical_loss": 3.387486961471544,
+      "tokens_seen": 2590441472
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021714010592200288,
+      "loss": 1.2999,
+      "theoretical_loss": 3.387473501003755,
+      "tokens_seen": 2590572544
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002170999839512117,
+      "loss": 1.285,
+      "theoretical_loss": 3.3874600414076728,
+      "tokens_seen": 2590703616
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002170598619804205,
+      "loss": 1.2883,
+      "theoretical_loss": 3.3874465826831965,
+      "tokens_seen": 2590834688
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002170197400096293,
+      "loss": 1.327,
+      "theoretical_loss": 3.387433124830225,
+      "tokens_seen": 2590965760
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021697961803883805,
+      "loss": 1.305,
+      "theoretical_loss": 3.3874196678486586,
+      "tokens_seen": 2591096832
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021693949606804686,
+      "loss": 1.304,
+      "theoretical_loss": 3.3874062117383965,
+      "tokens_seen": 2591227904
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021689937409725567,
+      "loss": 1.2774,
+      "theoretical_loss": 3.387392756499338,
+      "tokens_seen": 2591358976
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021685925212646445,
+      "loss": 1.3531,
+      "theoretical_loss": 3.387379302131383,
+      "tokens_seen": 2591490048
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021681913015567325,
+      "loss": 1.3971,
+      "theoretical_loss": 3.387365848634431,
+      "tokens_seen": 2591621120
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021677900818488206,
+      "loss": 1.2977,
+      "theoretical_loss": 3.3873523960083816,
+      "tokens_seen": 2591752192
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021673888621409087,
+      "loss": 1.3031,
+      "theoretical_loss": 3.387338944253134,
+      "tokens_seen": 2591883264
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021669876424329962,
+      "loss": 1.3509,
+      "theoretical_loss": 3.387325493368588,
+      "tokens_seen": 2592014336
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021665864227250842,
+      "loss": 1.3239,
+      "theoretical_loss": 3.387312043354643,
+      "tokens_seen": 2592145408
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021661852030171723,
+      "loss": 1.3622,
+      "theoretical_loss": 3.3872985942111993,
+      "tokens_seen": 2592276480
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.000216578398330926,
+      "loss": 1.3359,
+      "theoretical_loss": 3.387285145938156,
+      "tokens_seen": 2592407552
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021653827636013482,
+      "loss": 1.244,
+      "theoretical_loss": 3.3872716985354128,
+      "tokens_seen": 2592538624
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021649815438934362,
+      "loss": 1.321,
+      "theoretical_loss": 3.38725825200287,
+      "tokens_seen": 2592669696
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002164580324185524,
+      "loss": 1.3009,
+      "theoretical_loss": 3.387244806340426,
+      "tokens_seen": 2592800768
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021641791044776118,
+      "loss": 1.3161,
+      "theoretical_loss": 3.3872313615479817,
+      "tokens_seen": 2592931840
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021637778847696999,
+      "loss": 1.3704,
+      "theoretical_loss": 3.3872179176254362,
+      "tokens_seen": 2593062912
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002163376665061788,
+      "loss": 1.3282,
+      "theoretical_loss": 3.38720447457269,
+      "tokens_seen": 2593193984
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021629754453538757,
+      "loss": 1.3161,
+      "theoretical_loss": 3.3871910323896417,
+      "tokens_seen": 2593325056
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021625742256459638,
+      "loss": 1.2837,
+      "theoretical_loss": 3.3871775910761923,
+      "tokens_seen": 2593456128
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021621730059380518,
+      "loss": 1.3077,
+      "theoretical_loss": 3.3871641506322407,
+      "tokens_seen": 2593587200
+    },
+    {
+      "epoch": 0.57,
+      "objective/train/advantage_avg": 0.49432945251464844,
+      "objective/train/docs_used": 943233,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.243801474571228,
+      "objective/train/original_loss": 1.243801474571228,
+      "objective/train/theoretical_loss": 3.387150711057687,
+      "objective/train/tokens_used": 964177376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24464760720729828,
+      "objective/train/value_max": -0.5,
+      "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": 1.3067961931228638,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.387150711057687,
+      "tokens_seen": 2593718272
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021617717862301396,
+      "loss": 1.2732,
+      "theoretical_loss": 3.387150711057687,
+      "tokens_seen": 2593718272
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021613705665222277,
+      "loss": 1.3329,
+      "theoretical_loss": 3.3871372723524313,
+      "tokens_seen": 2593849344
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021609693468143155,
+      "loss": 1.3254,
+      "theoretical_loss": 3.387123834516373,
+      "tokens_seen": 2593980416
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021605681271064036,
+      "loss": 1.4059,
+      "theoretical_loss": 3.3871103975494123,
+      "tokens_seen": 2594111488
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021601669073984913,
+      "loss": 1.2719,
+      "theoretical_loss": 3.3870969614514492,
+      "tokens_seen": 2594242560
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021597656876905794,
+      "loss": 1.3767,
+      "theoretical_loss": 3.387083526222383,
+      "tokens_seen": 2594373632
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021593644679826675,
+      "loss": 1.3644,
+      "theoretical_loss": 3.3870700918621144,
+      "tokens_seen": 2594504704
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021589632482747553,
+      "loss": 1.3126,
+      "theoretical_loss": 3.3870566583705424,
+      "tokens_seen": 2594635776
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021585620285668433,
+      "loss": 1.2237,
+      "theoretical_loss": 3.387043225747568,
+      "tokens_seen": 2594766848
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002158160808858931,
+      "loss": 1.2688,
+      "theoretical_loss": 3.3870297939930905,
+      "tokens_seen": 2594897920
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021577595891510192,
+      "loss": 1.2652,
+      "theoretical_loss": 3.38701636310701,
+      "tokens_seen": 2595028992
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002157358369443107,
+      "loss": 1.3772,
+      "theoretical_loss": 3.387002933089226,
+      "tokens_seen": 2595160064
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002156957149735195,
+      "loss": 1.3267,
+      "theoretical_loss": 3.3869895039396396,
+      "tokens_seen": 2595291136
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002156555930027283,
+      "loss": 1.3988,
+      "theoretical_loss": 3.38697607565815,
+      "tokens_seen": 2595422208
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002156154710319371,
+      "loss": 1.3317,
+      "theoretical_loss": 3.386962648244658,
+      "tokens_seen": 2595553280
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002155753490611459,
+      "loss": 1.256,
+      "theoretical_loss": 3.3869492216990627,
+      "tokens_seen": 2595684352
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002155352270903547,
+      "loss": 1.2942,
+      "theoretical_loss": 3.3869357960212647,
+      "tokens_seen": 2595815424
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021549510511956345,
+      "loss": 1.3753,
+      "theoretical_loss": 3.3869223712111642,
+      "tokens_seen": 2595946496
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021545498314877226,
+      "loss": 1.4315,
+      "theoretical_loss": 3.386908947268661,
+      "tokens_seen": 2596077568
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021541486117798107,
+      "loss": 1.289,
+      "theoretical_loss": 3.3868955241936556,
+      "tokens_seen": 2596208640
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021537473920718987,
+      "loss": 1.3491,
+      "theoretical_loss": 3.3868821019860476,
+      "tokens_seen": 2596339712
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021533461723639865,
+      "loss": 1.2376,
+      "theoretical_loss": 3.3868686806457378,
+      "tokens_seen": 2596470784
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021529449526560746,
+      "loss": 1.3015,
+      "theoretical_loss": 3.3868552601726263,
+      "tokens_seen": 2596601856
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021525437329481626,
+      "loss": 1.416,
+      "theoretical_loss": 3.3868418405666127,
+      "tokens_seen": 2596732928
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021521425132402502,
+      "loss": 1.4054,
+      "theoretical_loss": 3.3868284218275977,
+      "tokens_seen": 2596864000
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021517412935323382,
+      "loss": 1.3829,
+      "theoretical_loss": 3.3868150039554816,
+      "tokens_seen": 2596995072
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021513400738244263,
+      "loss": 1.305,
+      "theoretical_loss": 3.386801586950164,
+      "tokens_seen": 2597126144
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021509388541165144,
+      "loss": 1.3606,
+      "theoretical_loss": 3.3867881708115464,
+      "tokens_seen": 2597257216
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021505376344086021,
+      "loss": 1.3102,
+      "theoretical_loss": 3.386774755539528,
+      "tokens_seen": 2597388288
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021501364147006902,
+      "loss": 1.4368,
+      "theoretical_loss": 3.386761341134009,
+      "tokens_seen": 2597519360
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021497351949927783,
+      "loss": 1.3941,
+      "theoretical_loss": 3.3867479275948904,
+      "tokens_seen": 2597650432
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021493339752848658,
+      "loss": 1.3943,
+      "theoretical_loss": 3.3867345149220722,
+      "tokens_seen": 2597781504
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021489327555769539,
+      "loss": 1.3534,
+      "theoretical_loss": 3.386721103115455,
+      "tokens_seen": 2597912576
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002148531535869042,
+      "loss": 1.4093,
+      "theoretical_loss": 3.386707692174939,
+      "tokens_seen": 2598043648
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.000214813031616113,
+      "loss": 1.3823,
+      "theoretical_loss": 3.3866942821004242,
+      "tokens_seen": 2598174720
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021477290964532178,
+      "loss": 1.4147,
+      "theoretical_loss": 3.3866808728918114,
+      "tokens_seen": 2598305792
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021473278767453058,
+      "loss": 1.3018,
+      "theoretical_loss": 3.386667464549001,
+      "tokens_seen": 2598436864
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002146926657037394,
+      "loss": 1.3955,
+      "theoretical_loss": 3.386654057071894,
+      "tokens_seen": 2598567936
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00021465254373294817,
+      "loss": 1.3441,
+      "theoretical_loss": 3.3866406504603894,
+      "tokens_seen": 2598699008
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021461242176215695,
+      "loss": 1.3072,
+      "theoretical_loss": 3.386627244714389,
+      "tokens_seen": 2598830080
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021457229979136576,
+      "loss": 1.3022,
+      "theoretical_loss": 3.3866138398337924,
+      "tokens_seen": 2598961152
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021453217782057456,
+      "loss": 1.3487,
+      "theoretical_loss": 3.386600435818501,
+      "tokens_seen": 2599092224
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021449205584978334,
+      "loss": 1.2709,
+      "theoretical_loss": 3.386587032668414,
+      "tokens_seen": 2599223296
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021445193387899215,
+      "loss": 1.3192,
+      "theoretical_loss": 3.3865736303834333,
+      "tokens_seen": 2599354368
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021441181190820095,
+      "loss": 1.3112,
+      "theoretical_loss": 3.3865602289634587,
+      "tokens_seen": 2599485440
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021437168993740973,
+      "loss": 1.2957,
+      "theoretical_loss": 3.386546828408391,
+      "tokens_seen": 2599616512
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002143315679666185,
+      "loss": 1.3758,
+      "theoretical_loss": 3.3865334287181303,
+      "tokens_seen": 2599747584
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021429144599582732,
+      "loss": 1.2898,
+      "theoretical_loss": 3.3865200298925777,
+      "tokens_seen": 2599878656
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002142513240250361,
+      "loss": 1.3859,
+      "theoretical_loss": 3.386506631931634,
+      "tokens_seen": 2600009728
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002142112020542449,
+      "loss": 1.3995,
+      "theoretical_loss": 3.386493234835199,
+      "tokens_seen": 2600140800
+    },
+    {
+      "epoch": 0.58,
+      "objective/train/advantage_avg": 0.4934045076370239,
+      "objective/train/docs_used": 945762,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.249621033668518,
+      "objective/train/original_loss": 1.2496209144592285,
+      "objective/train/theoretical_loss": 3.386479838603174,
+      "objective/train/tokens_used": 970730976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24371948838233948,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505794286727905,
+      "objective/train/weighted_lm_loss": 1.3128259181976318,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.386479838603174,
+      "tokens_seen": 2600271872
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002141710800834537,
+      "loss": 1.3098,
+      "theoretical_loss": 3.386479838603174,
+      "tokens_seen": 2600271872
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021413095811266252,
+      "loss": 1.3497,
+      "theoretical_loss": 3.3864664432354594,
+      "tokens_seen": 2600402944
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002140908361418713,
+      "loss": 1.3862,
+      "theoretical_loss": 3.386453048731956,
+      "tokens_seen": 2600534016
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021405071417108007,
+      "loss": 1.4455,
+      "theoretical_loss": 3.386439655092565,
+      "tokens_seen": 2600665088
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021401059220028888,
+      "loss": 1.3305,
+      "theoretical_loss": 3.386426262317186,
+      "tokens_seen": 2600796160
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021397047022949766,
+      "loss": 1.3045,
+      "theoretical_loss": 3.3864128704057204,
+      "tokens_seen": 2600927232
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021393034825870647,
+      "loss": 1.3774,
+      "theoretical_loss": 3.386399479358069,
+      "tokens_seen": 2601058304
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021389022628791527,
+      "loss": 1.3253,
+      "theoretical_loss": 3.3863860891741324,
+      "tokens_seen": 2601189376
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021385010431712408,
+      "loss": 1.4174,
+      "theoretical_loss": 3.386372699853811,
+      "tokens_seen": 2601320448
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021380998234633286,
+      "loss": 1.3571,
+      "theoretical_loss": 3.3863593113970065,
+      "tokens_seen": 2601451520
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021376986037554166,
+      "loss": 1.2704,
+      "theoretical_loss": 3.386345923803619,
+      "tokens_seen": 2601582592
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021372973840475044,
+      "loss": 1.4247,
+      "theoretical_loss": 3.386332537073549,
+      "tokens_seen": 2601713664
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021368961643395922,
+      "loss": 1.3189,
+      "theoretical_loss": 3.386319151206698,
+      "tokens_seen": 2601844736
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021364949446316803,
+      "loss": 1.3285,
+      "theoretical_loss": 3.386305766202967,
+      "tokens_seen": 2601975808
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021360937249237684,
+      "loss": 1.3321,
+      "theoretical_loss": 3.3862923820622566,
+      "tokens_seen": 2602106880
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021356925052158564,
+      "loss": 1.344,
+      "theoretical_loss": 3.3862789987844675,
+      "tokens_seen": 2602237952
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021352912855079442,
+      "loss": 1.26,
+      "theoretical_loss": 3.3862656163695006,
+      "tokens_seen": 2602369024
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021348900658000323,
+      "loss": 1.3221,
+      "theoretical_loss": 3.386252234817257,
+      "tokens_seen": 2602500096
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.000213448884609212,
+      "loss": 1.3807,
+      "theoretical_loss": 3.3862388541276376,
+      "tokens_seen": 2602631168
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021340876263842079,
+      "loss": 1.3907,
+      "theoretical_loss": 3.3862254743005438,
+      "tokens_seen": 2602762240
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002133686406676296,
+      "loss": 1.3315,
+      "theoretical_loss": 3.3862120953358756,
+      "tokens_seen": 2602893312
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002133285186968384,
+      "loss": 1.3253,
+      "theoretical_loss": 3.3861987172335346,
+      "tokens_seen": 2603024384
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021328839672604718,
+      "loss": 1.3508,
+      "theoretical_loss": 3.386185339993422,
+      "tokens_seen": 2603155456
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021324827475525598,
+      "loss": 1.41,
+      "theoretical_loss": 3.386171963615438,
+      "tokens_seen": 2603286528
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002132081527844648,
+      "loss": 1.3289,
+      "theoretical_loss": 3.3861585880994847,
+      "tokens_seen": 2603417600
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002131680308136736,
+      "loss": 1.369,
+      "theoretical_loss": 3.386145213445462,
+      "tokens_seen": 2603548672
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021312790884288235,
+      "loss": 1.4512,
+      "theoretical_loss": 3.386131839653272,
+      "tokens_seen": 2603679744
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021308778687209115,
+      "loss": 1.4098,
+      "theoretical_loss": 3.3861184667228157,
+      "tokens_seen": 2603810816
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021304766490129996,
+      "loss": 1.2795,
+      "theoretical_loss": 3.3861050946539932,
+      "tokens_seen": 2603941888
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021300754293050874,
+      "loss": 1.5518,
+      "theoretical_loss": 3.3860917234467065,
+      "tokens_seen": 2604072960
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021296742095971755,
+      "loss": 1.3162,
+      "theoretical_loss": 3.386078353100857,
+      "tokens_seen": 2604204032
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021292729898892635,
+      "loss": 1.3504,
+      "theoretical_loss": 3.386064983616345,
+      "tokens_seen": 2604335104
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021288717701813516,
+      "loss": 1.2729,
+      "theoretical_loss": 3.3860516149930717,
+      "tokens_seen": 2604466176
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002128470550473439,
+      "loss": 1.4019,
+      "theoretical_loss": 3.386038247230939,
+      "tokens_seen": 2604597248
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021280693307655272,
+      "loss": 1.3801,
+      "theoretical_loss": 3.3860248803298476,
+      "tokens_seen": 2604728320
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021276681110576152,
+      "loss": 1.3238,
+      "theoretical_loss": 3.3860115142896987,
+      "tokens_seen": 2604859392
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002127266891349703,
+      "loss": 1.2891,
+      "theoretical_loss": 3.385998149110394,
+      "tokens_seen": 2604990464
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002126865671641791,
+      "loss": 1.4472,
+      "theoretical_loss": 3.3859847847918343,
+      "tokens_seen": 2605121536
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021264644519338792,
+      "loss": 1.3179,
+      "theoretical_loss": 3.385971421333921,
+      "tokens_seen": 2605252608
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021260632322259672,
+      "loss": 1.298,
+      "theoretical_loss": 3.385958058736555,
+      "tokens_seen": 2605383680
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021256620125180547,
+      "loss": 1.3457,
+      "theoretical_loss": 3.3859446969996383,
+      "tokens_seen": 2605514752
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021252607928101428,
+      "loss": 1.3549,
+      "theoretical_loss": 3.385931336123072,
+      "tokens_seen": 2605645824
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002124859573102231,
+      "loss": 1.2766,
+      "theoretical_loss": 3.385917976106757,
+      "tokens_seen": 2605776896
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021244583533943187,
+      "loss": 1.2599,
+      "theoretical_loss": 3.3859046169505946,
+      "tokens_seen": 2605907968
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021240571336864067,
+      "loss": 1.3636,
+      "theoretical_loss": 3.385891258654487,
+      "tokens_seen": 2606039040
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021236559139784948,
+      "loss": 1.352,
+      "theoretical_loss": 3.385877901218335,
+      "tokens_seen": 2606170112
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021232546942705826,
+      "loss": 1.4203,
+      "theoretical_loss": 3.38586454464204,
+      "tokens_seen": 2606301184
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021228534745626706,
+      "loss": 1.4108,
+      "theoretical_loss": 3.3858511889255034,
+      "tokens_seen": 2606432256
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021224522548547584,
+      "loss": 1.3891,
+      "theoretical_loss": 3.385837834068627,
+      "tokens_seen": 2606563328
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021220510351468465,
+      "loss": 1.3927,
+      "theoretical_loss": 3.3858244800713115,
+      "tokens_seen": 2606694400
+    },
+    {
+      "epoch": 0.58,
+      "objective/train/advantage_avg": 0.4930676519870758,
+      "objective/train/docs_used": 948007,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3414939641952515,
+      "objective/train/original_loss": 1.341493844985962,
+      "objective/train/theoretical_loss": 3.385811126933459,
+      "objective/train/tokens_used": 977284576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24404506385326385,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050547480583191,
+      "objective/train/weighted_lm_loss": 1.4093116521835327,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.385811126933459,
+      "tokens_seen": 2606825472
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021216498154389343,
+      "loss": 1.3856,
+      "theoretical_loss": 3.385811126933459,
+      "tokens_seen": 2606825472
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021212485957310223,
+      "loss": 1.2999,
+      "theoretical_loss": 3.385797774654971,
+      "tokens_seen": 2606956544
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021208473760231104,
+      "loss": 1.2838,
+      "theoretical_loss": 3.385784423235749,
+      "tokens_seen": 2607087616
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021204461563151982,
+      "loss": 1.3743,
+      "theoretical_loss": 3.3857710726756935,
+      "tokens_seen": 2607218688
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021200449366072863,
+      "loss": 1.3792,
+      "theoretical_loss": 3.385757722974707,
+      "tokens_seen": 2607349760
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002119643716899374,
+      "loss": 1.2494,
+      "theoretical_loss": 3.385744374132691,
+      "tokens_seen": 2607480832
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002119242497191462,
+      "loss": 1.4064,
+      "theoretical_loss": 3.3857310261495472,
+      "tokens_seen": 2607611904
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.000211884127748355,
+      "loss": 1.3009,
+      "theoretical_loss": 3.3857176790251766,
+      "tokens_seen": 2607742976
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002118440057775638,
+      "loss": 1.4338,
+      "theoretical_loss": 3.385704332759481,
+      "tokens_seen": 2607874048
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002118038838067726,
+      "loss": 1.3323,
+      "theoretical_loss": 3.385690987352362,
+      "tokens_seen": 2608005120
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021176376183598138,
+      "loss": 1.3012,
+      "theoretical_loss": 3.3856776428037216,
+      "tokens_seen": 2608136192
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002117236398651902,
+      "loss": 1.3873,
+      "theoretical_loss": 3.385664299113461,
+      "tokens_seen": 2608267264
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021168351789439897,
+      "loss": 1.326,
+      "theoretical_loss": 3.3856509562814816,
+      "tokens_seen": 2608398336
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021164339592360778,
+      "loss": 1.3448,
+      "theoretical_loss": 3.3856376143076856,
+      "tokens_seen": 2608529408
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021160327395281655,
+      "loss": 1.2748,
+      "theoretical_loss": 3.385624273191975,
+      "tokens_seen": 2608660480
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021156315198202536,
+      "loss": 1.3368,
+      "theoretical_loss": 3.3856109329342505,
+      "tokens_seen": 2608791552
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021152303001123417,
+      "loss": 1.3593,
+      "theoretical_loss": 3.3855975935344147,
+      "tokens_seen": 2608922624
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021148290804044295,
+      "loss": 1.2535,
+      "theoretical_loss": 3.3855842549923687,
+      "tokens_seen": 2609053696
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021144278606965175,
+      "loss": 1.3336,
+      "theoretical_loss": 3.385570917308015,
+      "tokens_seen": 2609184768
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021140266409886056,
+      "loss": 1.3252,
+      "theoretical_loss": 3.3855575804812545,
+      "tokens_seen": 2609315840
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002113625421280693,
+      "loss": 1.3382,
+      "theoretical_loss": 3.385544244511989,
+      "tokens_seen": 2609446912
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021132242015727812,
+      "loss": 1.3944,
+      "theoretical_loss": 3.3855309094001216,
+      "tokens_seen": 2609577984
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021128229818648692,
+      "loss": 1.3028,
+      "theoretical_loss": 3.3855175751455526,
+      "tokens_seen": 2609709056
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021124217621569573,
+      "loss": 1.3621,
+      "theoretical_loss": 3.3855042417481847,
+      "tokens_seen": 2609840128
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002112020542449045,
+      "loss": 1.3818,
+      "theoretical_loss": 3.3854909092079195,
+      "tokens_seen": 2609971200
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021116193227411332,
+      "loss": 1.4146,
+      "theoretical_loss": 3.3854775775246586,
+      "tokens_seen": 2610102272
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021112181030332212,
+      "loss": 1.3553,
+      "theoretical_loss": 3.3854642466983043,
+      "tokens_seen": 2610233344
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021108168833253087,
+      "loss": 1.2972,
+      "theoretical_loss": 3.3854509167287583,
+      "tokens_seen": 2610364416
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021104156636173968,
+      "loss": 1.3291,
+      "theoretical_loss": 3.3854375876159226,
+      "tokens_seen": 2610495488
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021100144439094849,
+      "loss": 1.2847,
+      "theoretical_loss": 3.385424259359699,
+      "tokens_seen": 2610626560
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002109613224201573,
+      "loss": 1.3674,
+      "theoretical_loss": 3.3854109319599894,
+      "tokens_seen": 2610757632
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021092120044936607,
+      "loss": 1.3381,
+      "theoretical_loss": 3.385397605416696,
+      "tokens_seen": 2610888704
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021088107847857488,
+      "loss": 1.3215,
+      "theoretical_loss": 3.3853842797297204,
+      "tokens_seen": 2611019776
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021084095650778368,
+      "loss": 1.2718,
+      "theoretical_loss": 3.3853709548989652,
+      "tokens_seen": 2611150848
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021080083453699246,
+      "loss": 1.3866,
+      "theoretical_loss": 3.3853576309243323,
+      "tokens_seen": 2611281920
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021076071256620124,
+      "loss": 1.3642,
+      "theoretical_loss": 3.385344307805723,
+      "tokens_seen": 2611412992
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021072059059541005,
+      "loss": 1.3049,
+      "theoretical_loss": 3.38533098554304,
+      "tokens_seen": 2611544064
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021068046862461886,
+      "loss": 1.2305,
+      "theoretical_loss": 3.385317664136185,
+      "tokens_seen": 2611675136
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021064034665382763,
+      "loss": 1.3588,
+      "theoretical_loss": 3.38530434358506,
+      "tokens_seen": 2611806208
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021060022468303644,
+      "loss": 1.362,
+      "theoretical_loss": 3.385291023889568,
+      "tokens_seen": 2611937280
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021056010271224525,
+      "loss": 1.377,
+      "theoretical_loss": 3.38527770504961,
+      "tokens_seen": 2612068352
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021051998074145403,
+      "loss": 1.3474,
+      "theoretical_loss": 3.3852643870650887,
+      "tokens_seen": 2612199424
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002104798587706628,
+      "loss": 1.2847,
+      "theoretical_loss": 3.385251069935906,
+      "tokens_seen": 2612330496
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002104397367998716,
+      "loss": 1.2742,
+      "theoretical_loss": 3.3852377536619644,
+      "tokens_seen": 2612461568
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002103996148290804,
+      "loss": 1.3224,
+      "theoretical_loss": 3.3852244382431658,
+      "tokens_seen": 2612592640
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002103594928582892,
+      "loss": 1.4178,
+      "theoretical_loss": 3.3852111236794125,
+      "tokens_seen": 2612723712
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.000210319370887498,
+      "loss": 1.3238,
+      "theoretical_loss": 3.3851978099706064,
+      "tokens_seen": 2612854784
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002102792489167068,
+      "loss": 1.393,
+      "theoretical_loss": 3.38518449711665,
+      "tokens_seen": 2612985856
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002102391269459156,
+      "loss": 1.398,
+      "theoretical_loss": 3.3851711851174455,
+      "tokens_seen": 2613116928
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021019900497512437,
+      "loss": 1.267,
+      "theoretical_loss": 3.385157873972895,
+      "tokens_seen": 2613248000
+    },
+    {
+      "epoch": 0.58,
+      "objective/train/advantage_avg": 0.49291563034057617,
+      "objective/train/docs_used": 950332,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.214140772819519,
+      "objective/train/original_loss": 1.2141406536102295,
+      "objective/train/theoretical_loss": 3.3851445636829007,
+      "objective/train/tokens_used": 983838176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2434384971857071,
+      "objective/train/value_max": -0.5,
+      "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": 1.2754017114639282,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3851445636829007,
+      "tokens_seen": 2613379072
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021015888300433317,
+      "loss": 1.287,
+      "theoretical_loss": 3.3851445636829007,
+      "tokens_seen": 2613379072
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021011876103354195,
+      "loss": 1.3236,
+      "theoretical_loss": 3.3851312542473657,
+      "tokens_seen": 2613510144
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021007863906275076,
+      "loss": 1.3634,
+      "theoretical_loss": 3.3851179456661913,
+      "tokens_seen": 2613641216
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021003851709195957,
+      "loss": 1.3475,
+      "theoretical_loss": 3.38510463793928,
+      "tokens_seen": 2613772288
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00020999839512116837,
+      "loss": 1.3821,
+      "theoretical_loss": 3.3850913310665347,
+      "tokens_seen": 2613903360
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00020995827315037715,
+      "loss": 1.3904,
+      "theoretical_loss": 3.385078025047857,
+      "tokens_seen": 2614034432
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00020991815117958596,
+      "loss": 1.4219,
+      "theoretical_loss": 3.3850647198831503,
+      "tokens_seen": 2614165504
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00020987802920879474,
+      "loss": 1.2886,
+      "theoretical_loss": 3.385051415572316,
+      "tokens_seen": 2614296576
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00020983790723800352,
+      "loss": 1.3401,
+      "theoretical_loss": 3.3850381121152564,
+      "tokens_seen": 2614427648
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00020979778526721232,
+      "loss": 1.2981,
+      "theoretical_loss": 3.385024809511875,
+      "tokens_seen": 2614558720
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00020975766329642113,
+      "loss": 1.331,
+      "theoretical_loss": 3.385011507762073,
+      "tokens_seen": 2614689792
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00020971754132562994,
+      "loss": 1.3242,
+      "theoretical_loss": 3.3849982068657534,
+      "tokens_seen": 2614820864
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00020967741935483871,
+      "loss": 1.3646,
+      "theoretical_loss": 3.384984906822819,
+      "tokens_seen": 2614951936
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00020963729738404752,
+      "loss": 1.3602,
+      "theoretical_loss": 3.3849716076331715,
+      "tokens_seen": 2615083008
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002095971754132563,
+      "loss": 1.3253,
+      "theoretical_loss": 3.384958309296714,
+      "tokens_seen": 2615214080
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020955705344246508,
+      "loss": 1.3362,
+      "theoretical_loss": 3.384945011813349,
+      "tokens_seen": 2615345152
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020951693147167389,
+      "loss": 1.2232,
+      "theoretical_loss": 3.3849317151829785,
+      "tokens_seen": 2615476224
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002094768095008827,
+      "loss": 1.3336,
+      "theoretical_loss": 3.3849184194055058,
+      "tokens_seen": 2615607296
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002094366875300915,
+      "loss": 1.3469,
+      "theoretical_loss": 3.384905124480833,
+      "tokens_seen": 2615738368
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020939656555930028,
+      "loss": 1.3501,
+      "theoretical_loss": 3.3848918304088627,
+      "tokens_seen": 2615869440
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020935644358850908,
+      "loss": 1.3113,
+      "theoretical_loss": 3.3848785371894974,
+      "tokens_seen": 2616000512
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020931632161771786,
+      "loss": 1.2935,
+      "theoretical_loss": 3.38486524482264,
+      "tokens_seen": 2616131584
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020927619964692664,
+      "loss": 1.3289,
+      "theoretical_loss": 3.3848519533081927,
+      "tokens_seen": 2616262656
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020923607767613545,
+      "loss": 1.3709,
+      "theoretical_loss": 3.3848386626460583,
+      "tokens_seen": 2616393728
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020919595570534426,
+      "loss": 1.3739,
+      "theoretical_loss": 3.3848253728361395,
+      "tokens_seen": 2616524800
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020915583373455303,
+      "loss": 1.2544,
+      "theoretical_loss": 3.384812083878339,
+      "tokens_seen": 2616655872
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020911571176376184,
+      "loss": 1.2973,
+      "theoretical_loss": 3.38479879577256,
+      "tokens_seen": 2616786944
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020907558979297065,
+      "loss": 1.327,
+      "theoretical_loss": 3.384785508518704,
+      "tokens_seen": 2616918016
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020903546782217945,
+      "loss": 1.3337,
+      "theoretical_loss": 3.3847722221166747,
+      "tokens_seen": 2617049088
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002089953458513882,
+      "loss": 1.3434,
+      "theoretical_loss": 3.3847589365663744,
+      "tokens_seen": 2617180160
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.000208955223880597,
+      "loss": 1.2352,
+      "theoretical_loss": 3.384745651867706,
+      "tokens_seen": 2617311232
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020891510190980582,
+      "loss": 1.4135,
+      "theoretical_loss": 3.3847323680205728,
+      "tokens_seen": 2617442304
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002088749799390146,
+      "loss": 1.3422,
+      "theoretical_loss": 3.3847190850248765,
+      "tokens_seen": 2617573376
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002088348579682234,
+      "loss": 1.2756,
+      "theoretical_loss": 3.38470580288052,
+      "tokens_seen": 2617704448
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002087947359974322,
+      "loss": 1.3104,
+      "theoretical_loss": 3.3846925215874073,
+      "tokens_seen": 2617835520
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020875461402664102,
+      "loss": 1.3522,
+      "theoretical_loss": 3.38467924114544,
+      "tokens_seen": 2617966592
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020871449205584977,
+      "loss": 1.3392,
+      "theoretical_loss": 3.3846659615545214,
+      "tokens_seen": 2618097664
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020867437008505857,
+      "loss": 1.2687,
+      "theoretical_loss": 3.3846526828145542,
+      "tokens_seen": 2618228736
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020863424811426738,
+      "loss": 1.3865,
+      "theoretical_loss": 3.384639404925442,
+      "tokens_seen": 2618359808
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020859412614347616,
+      "loss": 1.258,
+      "theoretical_loss": 3.3846261278870866,
+      "tokens_seen": 2618490880
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020855400417268497,
+      "loss": 1.4024,
+      "theoretical_loss": 3.3846128516993916,
+      "tokens_seen": 2618621952
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020851388220189377,
+      "loss": 1.337,
+      "theoretical_loss": 3.38459957636226,
+      "tokens_seen": 2618753024
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020847376023110258,
+      "loss": 1.3485,
+      "theoretical_loss": 3.384586301875594,
+      "tokens_seen": 2618884096
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020843363826031136,
+      "loss": 1.3959,
+      "theoretical_loss": 3.384573028239297,
+      "tokens_seen": 2619015168
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020839351628952014,
+      "loss": 1.3718,
+      "theoretical_loss": 3.3845597554532727,
+      "tokens_seen": 2619146240
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020835339431872894,
+      "loss": 1.3666,
+      "theoretical_loss": 3.384546483517423,
+      "tokens_seen": 2619277312
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020831327234793772,
+      "loss": 1.4004,
+      "theoretical_loss": 3.384533212431651,
+      "tokens_seen": 2619408384
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020827315037714653,
+      "loss": 1.3403,
+      "theoretical_loss": 3.3845199421958605,
+      "tokens_seen": 2619539456
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020823302840635534,
+      "loss": 1.3434,
+      "theoretical_loss": 3.384506672809954,
+      "tokens_seen": 2619670528
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020819290643556411,
+      "loss": 1.3563,
+      "theoretical_loss": 3.3844934042738344,
+      "tokens_seen": 2619801600
+    },
+    {
+      "epoch": 0.59,
+      "objective/train/advantage_avg": 0.4913542568683624,
+      "objective/train/docs_used": 952663,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.21775484085083,
+      "objective/train/original_loss": 1.21775484085083,
+      "objective/train/theoretical_loss": 3.3844801365874053,
+      "objective/train/tokens_used": 990391776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2423277497291565,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503674745559692,
+      "objective/train/weighted_lm_loss": 1.2790042161941528,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9048373699188232,
+      "theoretical_loss": 3.3844801365874053,
+      "tokens_seen": 2619932672
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020815278446477292,
+      "loss": 1.3245,
+      "theoretical_loss": 3.3844801365874053,
+      "tokens_seen": 2619932672
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002081126624939817,
+      "loss": 1.2942,
+      "theoretical_loss": 3.384466869750569,
+      "tokens_seen": 2620063744
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002080725405231905,
+      "loss": 1.355,
+      "theoretical_loss": 3.3844536037632293,
+      "tokens_seen": 2620194816
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020803241855239929,
+      "loss": 1.3917,
+      "theoretical_loss": 3.3844403386252893,
+      "tokens_seen": 2620325888
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002079922965816081,
+      "loss": 1.2595,
+      "theoretical_loss": 3.3844270743366516,
+      "tokens_seen": 2620456960
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002079521746108169,
+      "loss": 1.3051,
+      "theoretical_loss": 3.38441381089722,
+      "tokens_seen": 2620588032
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020791205264002568,
+      "loss": 1.4396,
+      "theoretical_loss": 3.384400548306897,
+      "tokens_seen": 2620719104
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020787193066923448,
+      "loss": 1.3365,
+      "theoretical_loss": 3.384387286565586,
+      "tokens_seen": 2620850176
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020783180869844326,
+      "loss": 1.3762,
+      "theoretical_loss": 3.3843740256731905,
+      "tokens_seen": 2620981248
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020779168672765207,
+      "loss": 1.3074,
+      "theoretical_loss": 3.384360765629613,
+      "tokens_seen": 2621112320
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020775156475686085,
+      "loss": 1.3227,
+      "theoretical_loss": 3.384347506434758,
+      "tokens_seen": 2621243392
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020771144278606965,
+      "loss": 1.2844,
+      "theoretical_loss": 3.3843342480885275,
+      "tokens_seen": 2621374464
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020767132081527846,
+      "loss": 1.3178,
+      "theoretical_loss": 3.3843209905908256,
+      "tokens_seen": 2621505536
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020763119884448724,
+      "loss": 1.322,
+      "theoretical_loss": 3.384307733941555,
+      "tokens_seen": 2621636608
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020759107687369605,
+      "loss": 1.3924,
+      "theoretical_loss": 3.384294478140619,
+      "tokens_seen": 2621767680
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020755095490290485,
+      "loss": 1.3219,
+      "theoretical_loss": 3.384281223187921,
+      "tokens_seen": 2621898752
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020751083293211363,
+      "loss": 1.2805,
+      "theoretical_loss": 3.3842679690833646,
+      "tokens_seen": 2622029824
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002074707109613224,
+      "loss": 1.3582,
+      "theoretical_loss": 3.384254715826853,
+      "tokens_seen": 2622160896
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020743058899053122,
+      "loss": 1.3897,
+      "theoretical_loss": 3.3842414634182894,
+      "tokens_seen": 2622291968
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020739046701974002,
+      "loss": 1.3683,
+      "theoretical_loss": 3.3842282118575775,
+      "tokens_seen": 2622423040
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002073503450489488,
+      "loss": 1.2925,
+      "theoretical_loss": 3.38421496114462,
+      "tokens_seen": 2622554112
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002073102230781576,
+      "loss": 1.2884,
+      "theoretical_loss": 3.3842017112793212,
+      "tokens_seen": 2622685184
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020727010110736642,
+      "loss": 1.3978,
+      "theoretical_loss": 3.3841884622615837,
+      "tokens_seen": 2622816256
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020722997913657517,
+      "loss": 1.2972,
+      "theoretical_loss": 3.3841752140913117,
+      "tokens_seen": 2622947328
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020718985716578397,
+      "loss": 1.3378,
+      "theoretical_loss": 3.3841619667684077,
+      "tokens_seen": 2623078400
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020714973519499278,
+      "loss": 1.2238,
+      "theoretical_loss": 3.3841487202927762,
+      "tokens_seen": 2623209472
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002071096132242016,
+      "loss": 1.3333,
+      "theoretical_loss": 3.38413547466432,
+      "tokens_seen": 2623340544
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020706949125341037,
+      "loss": 1.3531,
+      "theoretical_loss": 3.3841222298829425,
+      "tokens_seen": 2623471616
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020702936928261917,
+      "loss": 1.2901,
+      "theoretical_loss": 3.384108985948548,
+      "tokens_seen": 2623602688
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020698924731182798,
+      "loss": 1.4003,
+      "theoretical_loss": 3.384095742861039,
+      "tokens_seen": 2623733760
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020694912534103673,
+      "loss": 1.3863,
+      "theoretical_loss": 3.38408250062032,
+      "tokens_seen": 2623864832
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020690900337024554,
+      "loss": 1.4431,
+      "theoretical_loss": 3.384069259226294,
+      "tokens_seen": 2623995904
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020686888139945434,
+      "loss": 1.3401,
+      "theoretical_loss": 3.384056018678865,
+      "tokens_seen": 2624126976
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020682875942866315,
+      "loss": 1.3115,
+      "theoretical_loss": 3.384042778977936,
+      "tokens_seen": 2624258048
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020678863745787193,
+      "loss": 1.312,
+      "theoretical_loss": 3.384029540123411,
+      "tokens_seen": 2624389120
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020674851548708074,
+      "loss": 1.3099,
+      "theoretical_loss": 3.384016302115193,
+      "tokens_seen": 2624520192
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020670839351628954,
+      "loss": 1.2849,
+      "theoretical_loss": 3.3840030649531867,
+      "tokens_seen": 2624651264
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020666827154549832,
+      "loss": 1.314,
+      "theoretical_loss": 3.383989828637295,
+      "tokens_seen": 2624782336
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002066281495747071,
+      "loss": 1.3898,
+      "theoretical_loss": 3.3839765931674215,
+      "tokens_seen": 2624913408
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002065880276039159,
+      "loss": 1.3195,
+      "theoretical_loss": 3.3839633585434705,
+      "tokens_seen": 2625044480
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002065479056331247,
+      "loss": 1.4007,
+      "theoretical_loss": 3.383950124765345,
+      "tokens_seen": 2625175552
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002065077836623335,
+      "loss": 1.4187,
+      "theoretical_loss": 3.38393689183295,
+      "tokens_seen": 2625306624
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002064676616915423,
+      "loss": 1.3754,
+      "theoretical_loss": 3.3839236597461877,
+      "tokens_seen": 2625437696
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002064275397207511,
+      "loss": 1.3927,
+      "theoretical_loss": 3.3839104285049624,
+      "tokens_seen": 2625568768
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020638741774995988,
+      "loss": 1.3288,
+      "theoretical_loss": 3.383897198109178,
+      "tokens_seen": 2625699840
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020634729577916866,
+      "loss": 1.3243,
+      "theoretical_loss": 3.383883968558738,
+      "tokens_seen": 2625830912
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020630717380837747,
+      "loss": 1.4277,
+      "theoretical_loss": 3.3838707398535464,
+      "tokens_seen": 2625961984
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020626705183758625,
+      "loss": 1.3692,
+      "theoretical_loss": 3.383857511993507,
+      "tokens_seen": 2626093056
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020622692986679505,
+      "loss": 1.37,
+      "theoretical_loss": 3.383844284978524,
+      "tokens_seen": 2626224128
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020618680789600386,
+      "loss": 1.3513,
+      "theoretical_loss": 3.3838310588085005,
+      "tokens_seen": 2626355200
+    },
+    {
+      "epoch": 0.59,
+      "objective/train/advantage_avg": 0.4925762712955475,
+      "objective/train/docs_used": 954702,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.1298737525939941,
+      "objective/train/original_loss": 1.1298737525939941,
+      "objective/train/theoretical_loss": 3.383817833483341,
+      "objective/train/tokens_used": 996945376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24313850700855255,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504937171936035,
+      "objective/train/weighted_lm_loss": 1.1867767572402954,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.383817833483341,
+      "tokens_seen": 2626486272
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020614668592521267,
+      "loss": 1.2888,
+      "theoretical_loss": 3.383817833483341,
+      "tokens_seen": 2626486272
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020610656395442145,
+      "loss": 1.3143,
+      "theoretical_loss": 3.383804609002949,
+      "tokens_seen": 2626617344
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020606644198363025,
+      "loss": 1.3433,
+      "theoretical_loss": 3.3837913853672283,
+      "tokens_seen": 2626748416
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020602632001283903,
+      "loss": 1.3596,
+      "theoretical_loss": 3.383778162576083,
+      "tokens_seen": 2626879488
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002059861980420478,
+      "loss": 1.3162,
+      "theoretical_loss": 3.3837649406294172,
+      "tokens_seen": 2627010560
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020594607607125662,
+      "loss": 1.3336,
+      "theoretical_loss": 3.383751719527135,
+      "tokens_seen": 2627141632
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020590595410046542,
+      "loss": 1.3739,
+      "theoretical_loss": 3.3837384992691395,
+      "tokens_seen": 2627272704
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020586583212967423,
+      "loss": 1.3268,
+      "theoretical_loss": 3.383725279855335,
+      "tokens_seen": 2627403776
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.000205825710158883,
+      "loss": 1.2485,
+      "theoretical_loss": 3.383712061285626,
+      "tokens_seen": 2627534848
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020578558818809182,
+      "loss": 1.3885,
+      "theoretical_loss": 3.383698843559916,
+      "tokens_seen": 2627665920
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002057454662173006,
+      "loss": 1.2801,
+      "theoretical_loss": 3.3836856266781092,
+      "tokens_seen": 2627796992
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020570534424650937,
+      "loss": 1.3047,
+      "theoretical_loss": 3.38367241064011,
+      "tokens_seen": 2627928064
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020566522227571818,
+      "loss": 1.4386,
+      "theoretical_loss": 3.3836591954458215,
+      "tokens_seen": 2628059136
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020562510030492699,
+      "loss": 1.3121,
+      "theoretical_loss": 3.3836459810951487,
+      "tokens_seen": 2628190208
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002055849783341358,
+      "loss": 1.3898,
+      "theoretical_loss": 3.383632767587995,
+      "tokens_seen": 2628321280
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020554485636334457,
+      "loss": 1.3352,
+      "theoretical_loss": 3.383619554924265,
+      "tokens_seen": 2628452352
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020550473439255338,
+      "loss": 1.2955,
+      "theoretical_loss": 3.3836063431038625,
+      "tokens_seen": 2628583424
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020546461242176216,
+      "loss": 1.3238,
+      "theoretical_loss": 3.3835931321266917,
+      "tokens_seen": 2628714496
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020542449045097094,
+      "loss": 1.3898,
+      "theoretical_loss": 3.3835799219926566,
+      "tokens_seen": 2628845568
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020538436848017974,
+      "loss": 1.3787,
+      "theoretical_loss": 3.3835667127016618,
+      "tokens_seen": 2628976640
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020534424650938855,
+      "loss": 1.3477,
+      "theoretical_loss": 3.383553504253611,
+      "tokens_seen": 2629107712
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020530412453859736,
+      "loss": 1.3444,
+      "theoretical_loss": 3.383540296648409,
+      "tokens_seen": 2629238784
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020526400256780613,
+      "loss": 1.3844,
+      "theoretical_loss": 3.383527089885959,
+      "tokens_seen": 2629369856
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020522388059701494,
+      "loss": 1.37,
+      "theoretical_loss": 3.383513883966166,
+      "tokens_seen": 2629500928
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020518375862622375,
+      "loss": 1.3368,
+      "theoretical_loss": 3.383500678888934,
+      "tokens_seen": 2629632000
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002051436366554325,
+      "loss": 1.2817,
+      "theoretical_loss": 3.383487474654167,
+      "tokens_seen": 2629763072
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002051035146846413,
+      "loss": 1.2501,
+      "theoretical_loss": 3.38347427126177,
+      "tokens_seen": 2629894144
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002050633927138501,
+      "loss": 1.3394,
+      "theoretical_loss": 3.3834610687116466,
+      "tokens_seen": 2630025216
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002050232707430589,
+      "loss": 1.3113,
+      "theoretical_loss": 3.3834478670037016,
+      "tokens_seen": 2630156288
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002049831487722677,
+      "loss": 1.2933,
+      "theoretical_loss": 3.383434666137839,
+      "tokens_seen": 2630287360
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002049430268014765,
+      "loss": 1.2466,
+      "theoretical_loss": 3.3834214661139628,
+      "tokens_seen": 2630418432
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002049029048306853,
+      "loss": 1.4034,
+      "theoretical_loss": 3.383408266931978,
+      "tokens_seen": 2630549504
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020486278285989406,
+      "loss": 1.3471,
+      "theoretical_loss": 3.3833950685917884,
+      "tokens_seen": 2630680576
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020482266088910287,
+      "loss": 1.3266,
+      "theoretical_loss": 3.383381871093299,
+      "tokens_seen": 2630811648
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020478253891831167,
+      "loss": 1.3502,
+      "theoretical_loss": 3.3833686744364138,
+      "tokens_seen": 2630942720
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020474241694752045,
+      "loss": 1.3715,
+      "theoretical_loss": 3.383355478621037,
+      "tokens_seen": 2631073792
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020470229497672926,
+      "loss": 1.3582,
+      "theoretical_loss": 3.3833422836470737,
+      "tokens_seen": 2631204864
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020466217300593807,
+      "loss": 1.2566,
+      "theoretical_loss": 3.3833290895144272,
+      "tokens_seen": 2631335936
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020462205103514687,
+      "loss": 1.2372,
+      "theoretical_loss": 3.383315896223003,
+      "tokens_seen": 2631467008
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00020458192906435563,
+      "loss": 1.2817,
+      "theoretical_loss": 3.3833027037727055,
+      "tokens_seen": 2631598080
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020454180709356443,
+      "loss": 1.2949,
+      "theoretical_loss": 3.3832895121634388,
+      "tokens_seen": 2631729152
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020450168512277324,
+      "loss": 1.2854,
+      "theoretical_loss": 3.3832763213951074,
+      "tokens_seen": 2631860224
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020446156315198202,
+      "loss": 1.285,
+      "theoretical_loss": 3.383263131467616,
+      "tokens_seen": 2631991296
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020442144118119082,
+      "loss": 1.2616,
+      "theoretical_loss": 3.3832499423808686,
+      "tokens_seen": 2632122368
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020438131921039963,
+      "loss": 1.2923,
+      "theoretical_loss": 3.3832367541347708,
+      "tokens_seen": 2632253440
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020434119723960844,
+      "loss": 1.4092,
+      "theoretical_loss": 3.3832235667292263,
+      "tokens_seen": 2632384512
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020430107526881721,
+      "loss": 1.3125,
+      "theoretical_loss": 3.38321038016414,
+      "tokens_seen": 2632515584
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.000204260953298026,
+      "loss": 1.2916,
+      "theoretical_loss": 3.3831971944394166,
+      "tokens_seen": 2632646656
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002042208313272348,
+      "loss": 1.2471,
+      "theoretical_loss": 3.3831840095549603,
+      "tokens_seen": 2632777728
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020418070935644358,
+      "loss": 1.2928,
+      "theoretical_loss": 3.383170825510676,
+      "tokens_seen": 2632908800
+    },
+    {
+      "epoch": 0.6,
+      "objective/train/advantage_avg": 0.4916519522666931,
+      "objective/train/docs_used": 956895,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2522923946380615,
+      "objective/train/original_loss": 1.2522923946380615,
+      "objective/train/theoretical_loss": 3.3831576423064686,
+      "objective/train/tokens_used": 1003498976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24236787855625153,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050397276878357,
+      "objective/train/weighted_lm_loss": 1.3151781558990479,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3831576423064686,
+      "tokens_seen": 2633039872
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020414058738565239,
+      "loss": 1.304,
+      "theoretical_loss": 3.3831576423064686,
+      "tokens_seen": 2633039872
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002041004654148612,
+      "loss": 1.3263,
+      "theoretical_loss": 3.383144459942242,
+      "tokens_seen": 2633170944
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020406034344406997,
+      "loss": 1.2742,
+      "theoretical_loss": 3.3831312784179017,
+      "tokens_seen": 2633302016
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020402022147327878,
+      "loss": 1.3362,
+      "theoretical_loss": 3.383118097733352,
+      "tokens_seen": 2633433088
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020398009950248756,
+      "loss": 1.3271,
+      "theoretical_loss": 3.3831049178884975,
+      "tokens_seen": 2633564160
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020393997753169636,
+      "loss": 1.284,
+      "theoretical_loss": 3.3830917388832433,
+      "tokens_seen": 2633695232
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020389985556090514,
+      "loss": 1.3474,
+      "theoretical_loss": 3.383078560717494,
+      "tokens_seen": 2633826304
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020385973359011395,
+      "loss": 1.3379,
+      "theoretical_loss": 3.3830653833911537,
+      "tokens_seen": 2633957376
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020381961161932276,
+      "loss": 1.3696,
+      "theoretical_loss": 3.383052206904128,
+      "tokens_seen": 2634088448
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020377948964853153,
+      "loss": 1.3471,
+      "theoretical_loss": 3.383039031256321,
+      "tokens_seen": 2634219520
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020373936767774034,
+      "loss": 1.333,
+      "theoretical_loss": 3.3830258564476385,
+      "tokens_seen": 2634350592
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020369924570694915,
+      "loss": 1.3151,
+      "theoretical_loss": 3.3830126824779843,
+      "tokens_seen": 2634481664
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020365912373615793,
+      "loss": 1.3315,
+      "theoretical_loss": 3.382999509347264,
+      "tokens_seen": 2634612736
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002036190017653667,
+      "loss": 1.3341,
+      "theoretical_loss": 3.3829863370553817,
+      "tokens_seen": 2634743808
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002035788797945755,
+      "loss": 1.2985,
+      "theoretical_loss": 3.382973165602243,
+      "tokens_seen": 2634874880
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020353875782378432,
+      "loss": 1.314,
+      "theoretical_loss": 3.382959994987752,
+      "tokens_seen": 2635005952
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002034986358529931,
+      "loss": 1.3771,
+      "theoretical_loss": 3.382946825211814,
+      "tokens_seen": 2635137024
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002034585138822019,
+      "loss": 1.3901,
+      "theoretical_loss": 3.382933656274334,
+      "tokens_seen": 2635268096
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002034183919114107,
+      "loss": 1.3317,
+      "theoretical_loss": 3.3829204881752166,
+      "tokens_seen": 2635399168
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002033782699406195,
+      "loss": 1.2915,
+      "theoretical_loss": 3.3829073209143674,
+      "tokens_seen": 2635530240
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020333814796982827,
+      "loss": 1.2824,
+      "theoretical_loss": 3.3828941544916904,
+      "tokens_seen": 2635661312
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020329802599903707,
+      "loss": 1.3281,
+      "theoretical_loss": 3.3828809889070914,
+      "tokens_seen": 2635792384
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020325790402824588,
+      "loss": 1.2934,
+      "theoretical_loss": 3.382867824160475,
+      "tokens_seen": 2635923456
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020321778205745466,
+      "loss": 1.3213,
+      "theoretical_loss": 3.382854660251746,
+      "tokens_seen": 2636054528
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020317766008666347,
+      "loss": 1.2085,
+      "theoretical_loss": 3.38284149718081,
+      "tokens_seen": 2636185600
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020313753811587227,
+      "loss": 1.254,
+      "theoretical_loss": 3.3828283349475714,
+      "tokens_seen": 2636316672
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020309741614508102,
+      "loss": 1.3317,
+      "theoretical_loss": 3.3828151735519354,
+      "tokens_seen": 2636447744
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020305729417428983,
+      "loss": 1.3324,
+      "theoretical_loss": 3.3828020129938077,
+      "tokens_seen": 2636578816
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020301717220349864,
+      "loss": 1.3123,
+      "theoretical_loss": 3.3827888532730928,
+      "tokens_seen": 2636709888
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020297705023270744,
+      "loss": 1.3258,
+      "theoretical_loss": 3.3827756943896956,
+      "tokens_seen": 2636840960
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020293692826191622,
+      "loss": 1.3214,
+      "theoretical_loss": 3.382762536343521,
+      "tokens_seen": 2636972032
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020289680629112503,
+      "loss": 1.3018,
+      "theoretical_loss": 3.3827493791344754,
+      "tokens_seen": 2637103104
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020285668432033384,
+      "loss": 1.2944,
+      "theoretical_loss": 3.382736222762463,
+      "tokens_seen": 2637234176
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020281656234954261,
+      "loss": 1.3701,
+      "theoretical_loss": 3.382723067227389,
+      "tokens_seen": 2637365248
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002027764403787514,
+      "loss": 1.2854,
+      "theoretical_loss": 3.382709912529158,
+      "tokens_seen": 2637496320
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002027363184079602,
+      "loss": 1.2611,
+      "theoretical_loss": 3.3826967586676764,
+      "tokens_seen": 2637627392
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.000202696196437169,
+      "loss": 1.3206,
+      "theoretical_loss": 3.382683605642849,
+      "tokens_seen": 2637758464
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020265607446637779,
+      "loss": 1.3274,
+      "theoretical_loss": 3.3826704534545806,
+      "tokens_seen": 2637889536
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002026159524955866,
+      "loss": 1.2876,
+      "theoretical_loss": 3.3826573021027766,
+      "tokens_seen": 2638020608
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002025758305247954,
+      "loss": 1.2536,
+      "theoretical_loss": 3.3826441515873427,
+      "tokens_seen": 2638151680
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020253570855400418,
+      "loss": 1.3418,
+      "theoretical_loss": 3.3826310019081833,
+      "tokens_seen": 2638282752
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020249558658321296,
+      "loss": 1.2858,
+      "theoretical_loss": 3.382617853065204,
+      "tokens_seen": 2638413824
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020245546461242176,
+      "loss": 1.3192,
+      "theoretical_loss": 3.382604705058311,
+      "tokens_seen": 2638544896
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020241534264163057,
+      "loss": 1.3334,
+      "theoretical_loss": 3.382591557887408,
+      "tokens_seen": 2638675968
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020237522067083935,
+      "loss": 1.2865,
+      "theoretical_loss": 3.3825784115524016,
+      "tokens_seen": 2638807040
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020233509870004815,
+      "loss": 1.2887,
+      "theoretical_loss": 3.382565266053197,
+      "tokens_seen": 2638938112
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020229497672925696,
+      "loss": 1.2744,
+      "theoretical_loss": 3.3825521213896987,
+      "tokens_seen": 2639069184
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020225485475846574,
+      "loss": 1.3341,
+      "theoretical_loss": 3.382538977561813,
+      "tokens_seen": 2639200256
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020221473278767452,
+      "loss": 1.2724,
+      "theoretical_loss": 3.382525834569445,
+      "tokens_seen": 2639331328
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020217461081688333,
+      "loss": 1.2768,
+      "theoretical_loss": 3.3825126924125,
+      "tokens_seen": 2639462400
+    },
+    {
+      "epoch": 0.6,
+      "objective/train/advantage_avg": 0.4917089641094208,
+      "objective/train/docs_used": 959162,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.221665382385254,
+      "objective/train/original_loss": 1.221665382385254,
+      "objective/train/theoretical_loss": 3.382499551090883,
+      "objective/train/tokens_used": 1010052576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24219606816768646,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504021644592285,
+      "objective/train/weighted_lm_loss": 1.2831776142120361,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.382499551090883,
+      "tokens_seen": 2639593472
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002021344888460921,
+      "loss": 1.2984,
+      "theoretical_loss": 3.382499551090883,
+      "tokens_seen": 2639593472
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002020943668753009,
+      "loss": 1.2571,
+      "theoretical_loss": 3.3824864106045,
+      "tokens_seen": 2639724544
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020205424490450972,
+      "loss": 1.2907,
+      "theoretical_loss": 3.3824732709532563,
+      "tokens_seen": 2639855616
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020201412293371852,
+      "loss": 1.3741,
+      "theoretical_loss": 3.3824601321370578,
+      "tokens_seen": 2639986688
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002019740009629273,
+      "loss": 1.2849,
+      "theoretical_loss": 3.382446994155809,
+      "tokens_seen": 2640117760
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002019338789921361,
+      "loss": 1.2359,
+      "theoretical_loss": 3.382433857009416,
+      "tokens_seen": 2640248832
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002018937570213449,
+      "loss": 1.2653,
+      "theoretical_loss": 3.3824207206977848,
+      "tokens_seen": 2640379904
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020185363505055367,
+      "loss": 1.3203,
+      "theoretical_loss": 3.38240758522082,
+      "tokens_seen": 2640510976
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020181351307976247,
+      "loss": 1.3025,
+      "theoretical_loss": 3.3823944505784276,
+      "tokens_seen": 2640642048
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020177339110897128,
+      "loss": 1.2376,
+      "theoretical_loss": 3.3823813167705126,
+      "tokens_seen": 2640773120
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002017332691381801,
+      "loss": 1.2747,
+      "theoretical_loss": 3.3823681837969817,
+      "tokens_seen": 2640904192
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020169314716738887,
+      "loss": 1.3979,
+      "theoretical_loss": 3.3823550516577394,
+      "tokens_seen": 2641035264
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020165302519659767,
+      "loss": 1.3057,
+      "theoretical_loss": 3.3823419203526917,
+      "tokens_seen": 2641166336
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020161290322580645,
+      "loss": 1.2615,
+      "theoretical_loss": 3.3823287898817442,
+      "tokens_seen": 2641297408
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020157278125501523,
+      "loss": 1.3027,
+      "theoretical_loss": 3.382315660244803,
+      "tokens_seen": 2641428480
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020153265928422404,
+      "loss": 1.3147,
+      "theoretical_loss": 3.382302531441773,
+      "tokens_seen": 2641559552
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020149253731343284,
+      "loss": 1.3242,
+      "theoretical_loss": 3.38228940347256,
+      "tokens_seen": 2641690624
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020145241534264165,
+      "loss": 1.3322,
+      "theoretical_loss": 3.38227627633707,
+      "tokens_seen": 2641821696
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020141229337185043,
+      "loss": 1.3577,
+      "theoretical_loss": 3.3822631500352083,
+      "tokens_seen": 2641952768
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020137217140105924,
+      "loss": 1.2262,
+      "theoretical_loss": 3.3822500245668814,
+      "tokens_seen": 2642083840
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020133204943026804,
+      "loss": 1.2111,
+      "theoretical_loss": 3.382236899931994,
+      "tokens_seen": 2642214912
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002012919274594768,
+      "loss": 1.2707,
+      "theoretical_loss": 3.3822237761304526,
+      "tokens_seen": 2642345984
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002012518054886856,
+      "loss": 1.3248,
+      "theoretical_loss": 3.3822106531621623,
+      "tokens_seen": 2642477056
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002012116835178944,
+      "loss": 1.2648,
+      "theoretical_loss": 3.3821975310270296,
+      "tokens_seen": 2642608128
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002011715615471032,
+      "loss": 1.2862,
+      "theoretical_loss": 3.3821844097249594,
+      "tokens_seen": 2642739200
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.000201131439576312,
+      "loss": 1.2818,
+      "theoretical_loss": 3.382171289255859,
+      "tokens_seen": 2642870272
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002010913176055208,
+      "loss": 1.325,
+      "theoretical_loss": 3.382158169619632,
+      "tokens_seen": 2643001344
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002010511956347296,
+      "loss": 1.3027,
+      "theoretical_loss": 3.3821450508161863,
+      "tokens_seen": 2643132416
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020101107366393836,
+      "loss": 1.2618,
+      "theoretical_loss": 3.3821319328454265,
+      "tokens_seen": 2643263488
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020097095169314716,
+      "loss": 1.3493,
+      "theoretical_loss": 3.382118815707259,
+      "tokens_seen": 2643394560
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020093082972235597,
+      "loss": 1.3411,
+      "theoretical_loss": 3.3821056994015892,
+      "tokens_seen": 2643525632
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020089070775156475,
+      "loss": 1.273,
+      "theoretical_loss": 3.382092583928324,
+      "tokens_seen": 2643656704
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020085058578077355,
+      "loss": 1.3408,
+      "theoretical_loss": 3.382079469287368,
+      "tokens_seen": 2643787776
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020081046380998236,
+      "loss": 1.3178,
+      "theoretical_loss": 3.3820663554786283,
+      "tokens_seen": 2643918848
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020077034183919117,
+      "loss": 1.2772,
+      "theoretical_loss": 3.38205324250201,
+      "tokens_seen": 2644049920
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020073021986839992,
+      "loss": 1.3563,
+      "theoretical_loss": 3.382040130357419,
+      "tokens_seen": 2644180992
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020069009789760873,
+      "loss": 1.3073,
+      "theoretical_loss": 3.382027019044762,
+      "tokens_seen": 2644312064
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020064997592681753,
+      "loss": 1.3148,
+      "theoretical_loss": 3.3820139085639447,
+      "tokens_seen": 2644443136
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002006098539560263,
+      "loss": 1.3595,
+      "theoretical_loss": 3.3820007989148726,
+      "tokens_seen": 2644574208
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020056973198523512,
+      "loss": 1.2828,
+      "theoretical_loss": 3.3819876900974526,
+      "tokens_seen": 2644705280
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020052961001444392,
+      "loss": 1.3655,
+      "theoretical_loss": 3.3819745821115896,
+      "tokens_seen": 2644836352
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020048948804365273,
+      "loss": 1.2916,
+      "theoretical_loss": 3.381961474957191,
+      "tokens_seen": 2644967424
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002004493660728615,
+      "loss": 1.2934,
+      "theoretical_loss": 3.381948368634162,
+      "tokens_seen": 2645098496
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002004092441020703,
+      "loss": 1.3434,
+      "theoretical_loss": 3.381935263142408,
+      "tokens_seen": 2645229568
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002003691221312791,
+      "loss": 1.3113,
+      "theoretical_loss": 3.381922158481837,
+      "tokens_seen": 2645360640
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020032900016048787,
+      "loss": 1.2769,
+      "theoretical_loss": 3.381909054652353,
+      "tokens_seen": 2645491712
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020028887818969668,
+      "loss": 1.2301,
+      "theoretical_loss": 3.3818959516538634,
+      "tokens_seen": 2645622784
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020024875621890549,
+      "loss": 1.2871,
+      "theoretical_loss": 3.3818828494862747,
+      "tokens_seen": 2645753856
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002002086342481143,
+      "loss": 1.2549,
+      "theoretical_loss": 3.3818697481494917,
+      "tokens_seen": 2645884928
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020016851227732307,
+      "loss": 1.2159,
+      "theoretical_loss": 3.3818566476434215,
+      "tokens_seen": 2646016000
+    },
+    {
+      "epoch": 0.6,
+      "objective/train/advantage_avg": 0.49363085627555847,
+      "objective/train/docs_used": 961281,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.1892348527908325,
+      "objective/train/original_loss": 1.1892348527908325,
+      "objective/train/theoretical_loss": 3.3818435479679705,
+      "objective/train/tokens_used": 1016606176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24398872256278992,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506035089492798,
+      "objective/train/weighted_lm_loss": 1.2493027448654175,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3818435479679705,
+      "tokens_seen": 2646147072
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020012839030653185,
+      "loss": 1.2539,
+      "theoretical_loss": 3.3818435479679705,
+      "tokens_seen": 2646147072
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020008826833574066,
+      "loss": 1.2566,
+      "theoretical_loss": 3.3818304491230435,
+      "tokens_seen": 2646278144
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020004814636494944,
+      "loss": 1.343,
+      "theoretical_loss": 3.3818173511085483,
+      "tokens_seen": 2646409216
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00020000802439415824,
+      "loss": 1.2811,
+      "theoretical_loss": 3.3818042539243907,
+      "tokens_seen": 2646540288
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00019996790242336705,
+      "loss": 1.2918,
+      "theoretical_loss": 3.3817911575704764,
+      "tokens_seen": 2646671360
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00019992778045257583,
+      "loss": 1.2951,
+      "theoretical_loss": 3.381778062046712,
+      "tokens_seen": 2646802432
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00019988765848178463,
+      "loss": 1.2961,
+      "theoretical_loss": 3.381764967353004,
+      "tokens_seen": 2646933504
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00019984753651099341,
+      "loss": 1.2872,
+      "theoretical_loss": 3.381751873489258,
+      "tokens_seen": 2647064576
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00019980741454020222,
+      "loss": 1.3297,
+      "theoretical_loss": 3.381738780455381,
+      "tokens_seen": 2647195648
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.000199767292569411,
+      "loss": 1.2787,
+      "theoretical_loss": 3.3817256882512794,
+      "tokens_seen": 2647326720
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0001997271705986198,
+      "loss": 1.3132,
+      "theoretical_loss": 3.3817125968768584,
+      "tokens_seen": 2647457792
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0001996870486278286,
+      "loss": 1.2886,
+      "theoretical_loss": 3.381699506332026,
+      "tokens_seen": 2647588864
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0001996469266570374,
+      "loss": 1.3075,
+      "theoretical_loss": 3.381686416616687,
+      "tokens_seen": 2647719936
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0001996068046862462,
+      "loss": 1.2671,
+      "theoretical_loss": 3.3816733277307494,
+      "tokens_seen": 2647851008
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.000199566682715455,
+      "loss": 1.3073,
+      "theoretical_loss": 3.381660239674118,
+      "tokens_seen": 2647982080
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00019952656074466378,
+      "loss": 1.3787,
+      "theoretical_loss": 3.3816471524467,
+      "tokens_seen": 2648113152
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019948643877387256,
+      "loss": 1.2725,
+      "theoretical_loss": 3.381634066048402,
+      "tokens_seen": 2648244224
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019944631680308137,
+      "loss": 1.2726,
+      "theoretical_loss": 3.3816209804791297,
+      "tokens_seen": 2648375296
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019940619483229017,
+      "loss": 1.2144,
+      "theoretical_loss": 3.3816078957387905,
+      "tokens_seen": 2648506368
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019936607286149895,
+      "loss": 1.3369,
+      "theoretical_loss": 3.38159481182729,
+      "tokens_seen": 2648637440
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019932595089070776,
+      "loss": 1.1988,
+      "theoretical_loss": 3.3815817287445356,
+      "tokens_seen": 2648768512
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019928582891991657,
+      "loss": 1.2532,
+      "theoretical_loss": 3.381568646490433,
+      "tokens_seen": 2648899584
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019924570694912535,
+      "loss": 1.2937,
+      "theoretical_loss": 3.3815555650648887,
+      "tokens_seen": 2649030656
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019920558497833413,
+      "loss": 1.2926,
+      "theoretical_loss": 3.38154248446781,
+      "tokens_seen": 2649161728
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019916546300754293,
+      "loss": 1.2363,
+      "theoretical_loss": 3.3815294046991027,
+      "tokens_seen": 2649292800
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019912534103675174,
+      "loss": 1.2666,
+      "theoretical_loss": 3.381516325758674,
+      "tokens_seen": 2649423872
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019908521906596052,
+      "loss": 1.3186,
+      "theoretical_loss": 3.3815032476464295,
+      "tokens_seen": 2649554944
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019904509709516932,
+      "loss": 1.2267,
+      "theoretical_loss": 3.3814901703622766,
+      "tokens_seen": 2649686016
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019900497512437813,
+      "loss": 1.2922,
+      "theoretical_loss": 3.3814770939061223,
+      "tokens_seen": 2649817088
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001989648531535869,
+      "loss": 1.291,
+      "theoretical_loss": 3.381464018277872,
+      "tokens_seen": 2649948160
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001989247311827957,
+      "loss": 1.309,
+      "theoretical_loss": 3.381450943477433,
+      "tokens_seen": 2650079232
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001988846092120045,
+      "loss": 1.3098,
+      "theoretical_loss": 3.381437869504712,
+      "tokens_seen": 2650210304
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001988444872412133,
+      "loss": 1.2562,
+      "theoretical_loss": 3.3814247963596156,
+      "tokens_seen": 2650341376
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019880436527042208,
+      "loss": 1.3352,
+      "theoretical_loss": 3.3814117240420507,
+      "tokens_seen": 2650472448
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019876424329963089,
+      "loss": 1.2617,
+      "theoretical_loss": 3.3813986525519235,
+      "tokens_seen": 2650603520
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001987241213288397,
+      "loss": 1.3478,
+      "theoretical_loss": 3.381385581889141,
+      "tokens_seen": 2650734592
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019868399935804847,
+      "loss": 1.3087,
+      "theoretical_loss": 3.38137251205361,
+      "tokens_seen": 2650865664
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019864387738725725,
+      "loss": 1.2607,
+      "theoretical_loss": 3.381359443045237,
+      "tokens_seen": 2650996736
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019860375541646606,
+      "loss": 1.3398,
+      "theoretical_loss": 3.3813463748639294,
+      "tokens_seen": 2651127808
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019856363344567486,
+      "loss": 1.2895,
+      "theoretical_loss": 3.381333307509593,
+      "tokens_seen": 2651258880
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019852351147488364,
+      "loss": 1.3206,
+      "theoretical_loss": 3.3813202409821352,
+      "tokens_seen": 2651389952
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019848338950409245,
+      "loss": 1.2409,
+      "theoretical_loss": 3.3813071752814627,
+      "tokens_seen": 2651521024
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019844326753330126,
+      "loss": 1.3067,
+      "theoretical_loss": 3.3812941104074827,
+      "tokens_seen": 2651652096
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019840314556251003,
+      "loss": 1.3234,
+      "theoretical_loss": 3.381281046360101,
+      "tokens_seen": 2651783168
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019836302359171881,
+      "loss": 1.3356,
+      "theoretical_loss": 3.3812679831392254,
+      "tokens_seen": 2651914240
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019832290162092762,
+      "loss": 1.2618,
+      "theoretical_loss": 3.3812549207447624,
+      "tokens_seen": 2652045312
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019828277965013643,
+      "loss": 1.3582,
+      "theoretical_loss": 3.381241859176619,
+      "tokens_seen": 2652176384
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001982426576793452,
+      "loss": 1.2593,
+      "theoretical_loss": 3.3812287984347016,
+      "tokens_seen": 2652307456
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.000198202535708554,
+      "loss": 1.3136,
+      "theoretical_loss": 3.3812157385189177,
+      "tokens_seen": 2652438528
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019816241373776282,
+      "loss": 1.2415,
+      "theoretical_loss": 3.3812026794291743,
+      "tokens_seen": 2652569600
+    },
+    {
+      "epoch": 0.61,
+      "objective/train/advantage_avg": 0.4926072955131531,
+      "objective/train/docs_used": 963694,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2860839366912842,
+      "objective/train/original_loss": 1.286083698272705,
+      "objective/train/theoretical_loss": 3.381189621165378,
+      "objective/train/tokens_used": 1023159776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24298864603042603,
+      "objective/train/value_max": -0.5,
+      "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": 1.351070523262024,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 1.0,
+      "theoretical_loss": 3.381189621165378,
+      "tokens_seen": 2652700672
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001981222917669716,
+      "loss": 1.3132,
+      "theoretical_loss": 3.381189621165378,
+      "tokens_seen": 2652700672
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001980821697961804,
+      "loss": 1.3134,
+      "theoretical_loss": 3.381176563727436,
+      "tokens_seen": 2652831744
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019804204782538918,
+      "loss": 1.367,
+      "theoretical_loss": 3.3811635071152546,
+      "tokens_seen": 2652962816
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019800192585459796,
+      "loss": 1.3196,
+      "theoretical_loss": 3.381150451328742,
+      "tokens_seen": 2653093888
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019796180388380677,
+      "loss": 1.2626,
+      "theoretical_loss": 3.3811373963678037,
+      "tokens_seen": 2653224960
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019792168191301557,
+      "loss": 1.2627,
+      "theoretical_loss": 3.381124342232348,
+      "tokens_seen": 2653356032
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019788155994222438,
+      "loss": 1.422,
+      "theoretical_loss": 3.3811112889222814,
+      "tokens_seen": 2653487104
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019784143797143316,
+      "loss": 1.2645,
+      "theoretical_loss": 3.381098236437511,
+      "tokens_seen": 2653618176
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019780131600064197,
+      "loss": 1.2941,
+      "theoretical_loss": 3.3810851847779437,
+      "tokens_seen": 2653749248
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019776119402985075,
+      "loss": 1.3326,
+      "theoretical_loss": 3.381072133943487,
+      "tokens_seen": 2653880320
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019772107205905952,
+      "loss": 1.3109,
+      "theoretical_loss": 3.3810590839340477,
+      "tokens_seen": 2654011392
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019768095008826833,
+      "loss": 1.3243,
+      "theoretical_loss": 3.3810460347495326,
+      "tokens_seen": 2654142464
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019764082811747714,
+      "loss": 1.3592,
+      "theoretical_loss": 3.381032986389849,
+      "tokens_seen": 2654273536
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019760070614668594,
+      "loss": 1.2845,
+      "theoretical_loss": 3.381019938854905,
+      "tokens_seen": 2654404608
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019756058417589472,
+      "loss": 1.3145,
+      "theoretical_loss": 3.3810068921446064,
+      "tokens_seen": 2654535680
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019752046220510353,
+      "loss": 1.4055,
+      "theoretical_loss": 3.380993846258861,
+      "tokens_seen": 2654666752
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001974803402343123,
+      "loss": 1.3374,
+      "theoretical_loss": 3.3809808011975755,
+      "tokens_seen": 2654797824
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001974402182635211,
+      "loss": 1.3942,
+      "theoretical_loss": 3.3809677569606578,
+      "tokens_seen": 2654928896
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001974000962927299,
+      "loss": 1.3216,
+      "theoretical_loss": 3.380954713548015,
+      "tokens_seen": 2655059968
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001973599743219387,
+      "loss": 1.2712,
+      "theoretical_loss": 3.3809416709595537,
+      "tokens_seen": 2655191040
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001973198523511475,
+      "loss": 1.3034,
+      "theoretical_loss": 3.3809286291951817,
+      "tokens_seen": 2655322112
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019727973038035629,
+      "loss": 1.277,
+      "theoretical_loss": 3.3809155882548056,
+      "tokens_seen": 2655453184
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001972396084095651,
+      "loss": 1.3238,
+      "theoretical_loss": 3.380902548138334,
+      "tokens_seen": 2655584256
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001971994864387739,
+      "loss": 1.4032,
+      "theoretical_loss": 3.380889508845673,
+      "tokens_seen": 2655715328
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019715936446798265,
+      "loss": 1.304,
+      "theoretical_loss": 3.3808764703767302,
+      "tokens_seen": 2655846400
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019711924249719146,
+      "loss": 1.3437,
+      "theoretical_loss": 3.380863432731413,
+      "tokens_seen": 2655977472
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019707912052640026,
+      "loss": 1.3593,
+      "theoretical_loss": 3.3808503959096288,
+      "tokens_seen": 2656108544
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019703899855560907,
+      "loss": 1.3357,
+      "theoretical_loss": 3.380837359911285,
+      "tokens_seen": 2656239616
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019699887658481785,
+      "loss": 1.3323,
+      "theoretical_loss": 3.380824324736288,
+      "tokens_seen": 2656370688
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019695875461402665,
+      "loss": 1.3147,
+      "theoretical_loss": 3.3808112903845466,
+      "tokens_seen": 2656501760
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019691863264323546,
+      "loss": 1.3682,
+      "theoretical_loss": 3.380798256855968,
+      "tokens_seen": 2656632832
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001968785106724442,
+      "loss": 1.3645,
+      "theoretical_loss": 3.3807852241504586,
+      "tokens_seen": 2656763904
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019683838870165302,
+      "loss": 1.3415,
+      "theoretical_loss": 3.3807721922679264,
+      "tokens_seen": 2656894976
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019679826673086183,
+      "loss": 1.2876,
+      "theoretical_loss": 3.380759161208279,
+      "tokens_seen": 2657026048
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001967581447600706,
+      "loss": 1.3564,
+      "theoretical_loss": 3.3807461309714237,
+      "tokens_seen": 2657157120
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001967180227892794,
+      "loss": 1.359,
+      "theoretical_loss": 3.3807331015572677,
+      "tokens_seen": 2657288192
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019667790081848822,
+      "loss": 1.3149,
+      "theoretical_loss": 3.380720072965719,
+      "tokens_seen": 2657419264
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019663777884769702,
+      "loss": 1.3828,
+      "theoretical_loss": 3.3807070451966847,
+      "tokens_seen": 2657550336
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001965976568769058,
+      "loss": 1.2974,
+      "theoretical_loss": 3.3806940182500727,
+      "tokens_seen": 2657681408
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019655753490611458,
+      "loss": 1.4097,
+      "theoretical_loss": 3.38068099212579,
+      "tokens_seen": 2657812480
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001965174129353234,
+      "loss": 1.253,
+      "theoretical_loss": 3.3806679668237445,
+      "tokens_seen": 2657943552
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019647729096453217,
+      "loss": 1.2465,
+      "theoretical_loss": 3.3806549423438437,
+      "tokens_seen": 2658074624
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019643716899374097,
+      "loss": 1.2792,
+      "theoretical_loss": 3.3806419186859946,
+      "tokens_seen": 2658205696
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019639704702294978,
+      "loss": 1.3237,
+      "theoretical_loss": 3.380628895850106,
+      "tokens_seen": 2658336768
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001963569250521586,
+      "loss": 1.3813,
+      "theoretical_loss": 3.3806158738360845,
+      "tokens_seen": 2658467840
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019631680308136737,
+      "loss": 1.2981,
+      "theoretical_loss": 3.380602852643838,
+      "tokens_seen": 2658598912
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019627668111057615,
+      "loss": 1.3624,
+      "theoretical_loss": 3.3805898322732744,
+      "tokens_seen": 2658729984
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019623655913978495,
+      "loss": 1.2979,
+      "theoretical_loss": 3.3805768127243008,
+      "tokens_seen": 2658861056
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019619643716899373,
+      "loss": 1.409,
+      "theoretical_loss": 3.380563793996825,
+      "tokens_seen": 2658992128
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019615631519820254,
+      "loss": 1.3165,
+      "theoretical_loss": 3.3805507760907556,
+      "tokens_seen": 2659123200
+    },
+    {
+      "epoch": 0.61,
+      "objective/train/advantage_avg": 0.49539005756378174,
+      "objective/train/docs_used": 966064,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.334479570388794,
+      "objective/train/original_loss": 1.3344796895980835,
+      "objective/train/theoretical_loss": 3.380537759005999,
+      "objective/train/tokens_used": 1029713376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24561285972595215,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0507878065109253,
+      "objective/train/weighted_lm_loss": 1.4022196531295776,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.380537759005999,
+      "tokens_seen": 2659254272
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019611619322741134,
+      "loss": 1.3419,
+      "theoretical_loss": 3.380537759005999,
+      "tokens_seen": 2659254272
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019607607125662015,
+      "loss": 1.2187,
+      "theoretical_loss": 3.3805247427424634,
+      "tokens_seen": 2659385344
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019603594928582893,
+      "loss": 1.4192,
+      "theoretical_loss": 3.380511727300057,
+      "tokens_seen": 2659516416
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001959958273150377,
+      "loss": 1.3086,
+      "theoretical_loss": 3.3804987126786865,
+      "tokens_seen": 2659647488
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019595570534424651,
+      "loss": 1.2302,
+      "theoretical_loss": 3.3804856988782603,
+      "tokens_seen": 2659778560
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001959155833734553,
+      "loss": 1.3047,
+      "theoretical_loss": 3.3804726858986864,
+      "tokens_seen": 2659909632
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001958754614026641,
+      "loss": 1.3321,
+      "theoretical_loss": 3.3804596737398724,
+      "tokens_seen": 2660040704
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001958353394318729,
+      "loss": 1.2095,
+      "theoretical_loss": 3.3804466624017255,
+      "tokens_seen": 2660171776
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019579521746108169,
+      "loss": 1.3798,
+      "theoretical_loss": 3.3804336518841542,
+      "tokens_seen": 2660302848
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001957550954902905,
+      "loss": 1.3023,
+      "theoretical_loss": 3.380420642187066,
+      "tokens_seen": 2660433920
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001957149735194993,
+      "loss": 1.3571,
+      "theoretical_loss": 3.3804076333103694,
+      "tokens_seen": 2660564992
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019567485154870808,
+      "loss": 1.2864,
+      "theoretical_loss": 3.380394625253971,
+      "tokens_seen": 2660696064
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019563472957791686,
+      "loss": 1.2637,
+      "theoretical_loss": 3.38038161801778,
+      "tokens_seen": 2660827136
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019559460760712566,
+      "loss": 1.3428,
+      "theoretical_loss": 3.380368611601703,
+      "tokens_seen": 2660958208
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019555448563633447,
+      "loss": 1.3121,
+      "theoretical_loss": 3.380355606005649,
+      "tokens_seen": 2661089280
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019551436366554325,
+      "loss": 1.3078,
+      "theoretical_loss": 3.380342601229525,
+      "tokens_seen": 2661220352
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019547424169475205,
+      "loss": 1.3259,
+      "theoretical_loss": 3.38032959727324,
+      "tokens_seen": 2661351424
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019543411972396086,
+      "loss": 1.2802,
+      "theoretical_loss": 3.380316594136701,
+      "tokens_seen": 2661482496
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019539399775316964,
+      "loss": 1.2788,
+      "theoretical_loss": 3.3803035918198163,
+      "tokens_seen": 2661613568
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019535387578237842,
+      "loss": 1.2995,
+      "theoretical_loss": 3.3802905903224936,
+      "tokens_seen": 2661744640
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019531375381158723,
+      "loss": 1.3233,
+      "theoretical_loss": 3.3802775896446415,
+      "tokens_seen": 2661875712
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019527363184079603,
+      "loss": 1.2744,
+      "theoretical_loss": 3.380264589786168,
+      "tokens_seen": 2662006784
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001952335098700048,
+      "loss": 1.2446,
+      "theoretical_loss": 3.3802515907469797,
+      "tokens_seen": 2662137856
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019519338789921362,
+      "loss": 1.3469,
+      "theoretical_loss": 3.380238592526987,
+      "tokens_seen": 2662268928
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019515326592842242,
+      "loss": 1.3596,
+      "theoretical_loss": 3.3802255951260953,
+      "tokens_seen": 2662400000
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001951131439576312,
+      "loss": 1.3218,
+      "theoretical_loss": 3.3802125985442153,
+      "tokens_seen": 2662531072
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019507302198683998,
+      "loss": 1.2703,
+      "theoretical_loss": 3.380199602781253,
+      "tokens_seen": 2662662144
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001950329000160488,
+      "loss": 1.3301,
+      "theoretical_loss": 3.380186607837117,
+      "tokens_seen": 2662793216
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001949927780452576,
+      "loss": 1.2695,
+      "theoretical_loss": 3.3801736137117167,
+      "tokens_seen": 2662924288
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019495265607446637,
+      "loss": 1.3269,
+      "theoretical_loss": 3.3801606204049586,
+      "tokens_seen": 2663055360
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019491253410367518,
+      "loss": 1.3556,
+      "theoretical_loss": 3.3801476279167515,
+      "tokens_seen": 2663186432
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019487241213288399,
+      "loss": 1.3133,
+      "theoretical_loss": 3.3801346362470035,
+      "tokens_seen": 2663317504
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019483229016209277,
+      "loss": 1.2225,
+      "theoretical_loss": 3.3801216453956227,
+      "tokens_seen": 2663448576
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019479216819130154,
+      "loss": 1.3011,
+      "theoretical_loss": 3.3801086553625175,
+      "tokens_seen": 2663579648
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019475204622051035,
+      "loss": 1.2209,
+      "theoretical_loss": 3.3800956661475956,
+      "tokens_seen": 2663710720
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019471192424971916,
+      "loss": 1.4068,
+      "theoretical_loss": 3.380082677750766,
+      "tokens_seen": 2663841792
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019467180227892794,
+      "loss": 1.3531,
+      "theoretical_loss": 3.3800696901719363,
+      "tokens_seen": 2663972864
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019463168030813674,
+      "loss": 1.3121,
+      "theoretical_loss": 3.380056703411015,
+      "tokens_seen": 2664103936
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019459155833734555,
+      "loss": 1.2624,
+      "theoretical_loss": 3.38004371746791,
+      "tokens_seen": 2664235008
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019455143636655433,
+      "loss": 1.2803,
+      "theoretical_loss": 3.38003073234253,
+      "tokens_seen": 2664366080
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001945113143957631,
+      "loss": 1.3876,
+      "theoretical_loss": 3.380017748034783,
+      "tokens_seen": 2664497152
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019447119242497191,
+      "loss": 1.3192,
+      "theoretical_loss": 3.3800047645445774,
+      "tokens_seen": 2664628224
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019443107045418072,
+      "loss": 1.2872,
+      "theoretical_loss": 3.379991781871822,
+      "tokens_seen": 2664759296
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001943909484833895,
+      "loss": 1.3883,
+      "theoretical_loss": 3.379978800016424,
+      "tokens_seen": 2664890368
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001943508265125983,
+      "loss": 1.3695,
+      "theoretical_loss": 3.3799658189782926,
+      "tokens_seen": 2665021440
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001943107045418071,
+      "loss": 1.3246,
+      "theoretical_loss": 3.379952838757336,
+      "tokens_seen": 2665152512
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001942705825710159,
+      "loss": 1.2939,
+      "theoretical_loss": 3.3799398593534624,
+      "tokens_seen": 2665283584
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019423046060022467,
+      "loss": 1.3572,
+      "theoretical_loss": 3.3799268807665808,
+      "tokens_seen": 2665414656
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019419033862943348,
+      "loss": 1.341,
+      "theoretical_loss": 3.379913902996599,
+      "tokens_seen": 2665545728
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019415021665864228,
+      "loss": 1.3461,
+      "theoretical_loss": 3.379900926043425,
+      "tokens_seen": 2665676800
+    },
+    {
+      "epoch": 0.62,
+      "objective/train/advantage_avg": 0.49288228154182434,
+      "objective/train/docs_used": 968517,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.1610095500946045,
+      "objective/train/original_loss": 1.1610095500946045,
+      "objective/train/theoretical_loss": 3.3798879499069683,
+      "objective/train/tokens_used": 1036266976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24325615167617798,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505249500274658,
+      "objective/train/weighted_lm_loss": 1.219584584236145,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3798879499069683,
+      "tokens_seen": 2665807872
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019411009468785106,
+      "loss": 1.2405,
+      "theoretical_loss": 3.3798879499069683,
+      "tokens_seen": 2665807872
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019406997271705987,
+      "loss": 1.3162,
+      "theoretical_loss": 3.379874974587137,
+      "tokens_seen": 2665938944
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019402985074626867,
+      "loss": 1.3454,
+      "theoretical_loss": 3.379862000083839,
+      "tokens_seen": 2666070016
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019398972877547745,
+      "loss": 1.3057,
+      "theoretical_loss": 3.379849026396983,
+      "tokens_seen": 2666201088
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019394960680468626,
+      "loss": 1.405,
+      "theoretical_loss": 3.379836053526478,
+      "tokens_seen": 2666332160
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019390948483389504,
+      "loss": 1.3347,
+      "theoretical_loss": 3.3798230814722316,
+      "tokens_seen": 2666463232
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019386936286310382,
+      "loss": 1.3509,
+      "theoretical_loss": 3.3798101102341533,
+      "tokens_seen": 2666594304
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019382924089231263,
+      "loss": 1.3288,
+      "theoretical_loss": 3.3797971398121516,
+      "tokens_seen": 2666725376
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019378911892152143,
+      "loss": 1.26,
+      "theoretical_loss": 3.379784170206134,
+      "tokens_seen": 2666856448
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019374899695073024,
+      "loss": 1.2829,
+      "theoretical_loss": 3.37977120141601,
+      "tokens_seen": 2666987520
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019370887497993902,
+      "loss": 1.2584,
+      "theoretical_loss": 3.379758233441688,
+      "tokens_seen": 2667118592
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019366875300914782,
+      "loss": 1.421,
+      "theoretical_loss": 3.3797452662830763,
+      "tokens_seen": 2667249664
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001936286310383566,
+      "loss": 1.2757,
+      "theoretical_loss": 3.379732299940084,
+      "tokens_seen": 2667380736
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019358850906756538,
+      "loss": 1.3105,
+      "theoretical_loss": 3.379719334412619,
+      "tokens_seen": 2667511808
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001935483870967742,
+      "loss": 1.3463,
+      "theoretical_loss": 3.3797063697005907,
+      "tokens_seen": 2667642880
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.000193508265125983,
+      "loss": 1.369,
+      "theoretical_loss": 3.3796934058039074,
+      "tokens_seen": 2667773952
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001934681431551918,
+      "loss": 1.3497,
+      "theoretical_loss": 3.379680442722478,
+      "tokens_seen": 2667905024
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019342802118440058,
+      "loss": 1.2082,
+      "theoretical_loss": 3.3796674804562103,
+      "tokens_seen": 2668036096
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019338789921360939,
+      "loss": 1.3465,
+      "theoretical_loss": 3.379654519005014,
+      "tokens_seen": 2668167168
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001933477772428182,
+      "loss": 1.2919,
+      "theoretical_loss": 3.3796415583687978,
+      "tokens_seen": 2668298240
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019330765527202694,
+      "loss": 1.3421,
+      "theoretical_loss": 3.3796285985474697,
+      "tokens_seen": 2668429312
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019326753330123575,
+      "loss": 1.2581,
+      "theoretical_loss": 3.3796156395409387,
+      "tokens_seen": 2668560384
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019322741133044456,
+      "loss": 1.3428,
+      "theoretical_loss": 3.3796026813491142,
+      "tokens_seen": 2668691456
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019318728935965336,
+      "loss": 1.3109,
+      "theoretical_loss": 3.3795897239719044,
+      "tokens_seen": 2668822528
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019314716738886214,
+      "loss": 1.2971,
+      "theoretical_loss": 3.379576767409218,
+      "tokens_seen": 2668953600
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019310704541807095,
+      "loss": 1.2226,
+      "theoretical_loss": 3.379563811660964,
+      "tokens_seen": 2669084672
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019306692344727976,
+      "loss": 1.2711,
+      "theoretical_loss": 3.379550856727051,
+      "tokens_seen": 2669215744
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001930268014764885,
+      "loss": 1.303,
+      "theoretical_loss": 3.3795379026073884,
+      "tokens_seen": 2669346816
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019298667950569731,
+      "loss": 1.2915,
+      "theoretical_loss": 3.379524949301884,
+      "tokens_seen": 2669477888
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019294655753490612,
+      "loss": 1.3127,
+      "theoretical_loss": 3.3795119968104483,
+      "tokens_seen": 2669608960
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019290643556411493,
+      "loss": 1.3407,
+      "theoretical_loss": 3.3794990451329885,
+      "tokens_seen": 2669740032
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001928663135933237,
+      "loss": 1.2225,
+      "theoretical_loss": 3.3794860942694145,
+      "tokens_seen": 2669871104
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001928261916225325,
+      "loss": 1.4358,
+      "theoretical_loss": 3.3794731442196344,
+      "tokens_seen": 2670002176
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019278606965174132,
+      "loss": 1.2921,
+      "theoretical_loss": 3.379460194983558,
+      "tokens_seen": 2670133248
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019274594768095007,
+      "loss": 1.3568,
+      "theoretical_loss": 3.3794472465610936,
+      "tokens_seen": 2670264320
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019270582571015888,
+      "loss": 1.2457,
+      "theoretical_loss": 3.3794342989521504,
+      "tokens_seen": 2670395392
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019266570373936768,
+      "loss": 1.3099,
+      "theoretical_loss": 3.3794213521566374,
+      "tokens_seen": 2670526464
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019262558176857646,
+      "loss": 1.2904,
+      "theoretical_loss": 3.3794084061744636,
+      "tokens_seen": 2670657536
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019258545979778527,
+      "loss": 1.3531,
+      "theoretical_loss": 3.379395461005538,
+      "tokens_seen": 2670788608
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019254533782699407,
+      "loss": 1.2535,
+      "theoretical_loss": 3.3793825166497693,
+      "tokens_seen": 2670919680
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019250521585620288,
+      "loss": 1.2667,
+      "theoretical_loss": 3.3793695731070668,
+      "tokens_seen": 2671050752
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019246509388541166,
+      "loss": 1.3659,
+      "theoretical_loss": 3.3793566303773392,
+      "tokens_seen": 2671181824
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019242497191462044,
+      "loss": 1.3467,
+      "theoretical_loss": 3.379343688460496,
+      "tokens_seen": 2671312896
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019238484994382925,
+      "loss": 1.3321,
+      "theoretical_loss": 3.3793307473564465,
+      "tokens_seen": 2671443968
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019234472797303802,
+      "loss": 1.3564,
+      "theoretical_loss": 3.3793178070650987,
+      "tokens_seen": 2671575040
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019230460600224683,
+      "loss": 1.281,
+      "theoretical_loss": 3.3793048675863626,
+      "tokens_seen": 2671706112
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019226448403145564,
+      "loss": 1.3389,
+      "theoretical_loss": 3.3792919289201473,
+      "tokens_seen": 2671837184
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019222436206066444,
+      "loss": 1.3286,
+      "theoretical_loss": 3.379278991066361,
+      "tokens_seen": 2671968256
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019218424008987322,
+      "loss": 1.2678,
+      "theoretical_loss": 3.379266054024914,
+      "tokens_seen": 2672099328
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.000192144118119082,
+      "loss": 1.3667,
+      "theoretical_loss": 3.3792531177957152,
+      "tokens_seen": 2672230400
+    },
+    {
+      "epoch": 0.62,
+      "objective/train/advantage_avg": 0.49203625321388245,
+      "objective/train/docs_used": 970931,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2398687601089478,
+      "objective/train/original_loss": 1.2398686408996582,
+      "objective/train/theoretical_loss": 3.3792401823786733,
+      "objective/train/tokens_used": 1042820576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24262471497058868,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050437092781067,
+      "objective/train/weighted_lm_loss": 1.3022422790527344,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9048373699188232,
+      "theoretical_loss": 3.3792401823786733,
+      "tokens_seen": 2672361472
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001921039961482908,
+      "loss": 1.29,
+      "theoretical_loss": 3.3792401823786733,
+      "tokens_seen": 2672361472
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001920638741774996,
+      "loss": 1.3398,
+      "theoretical_loss": 3.3792272477736978,
+      "tokens_seen": 2672492544
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001920237522067084,
+      "loss": 1.3129,
+      "theoretical_loss": 3.3792143139806976,
+      "tokens_seen": 2672623616
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001919836302359172,
+      "loss": 1.3381,
+      "theoretical_loss": 3.3792013809995822,
+      "tokens_seen": 2672754688
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.000191943508265126,
+      "loss": 1.3052,
+      "theoretical_loss": 3.379188448830261,
+      "tokens_seen": 2672885760
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019190338629433479,
+      "loss": 1.2824,
+      "theoretical_loss": 3.3791755174726426,
+      "tokens_seen": 2673016832
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019186326432354356,
+      "loss": 1.3923,
+      "theoretical_loss": 3.379162586926637,
+      "tokens_seen": 2673147904
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019182314235275237,
+      "loss": 1.3011,
+      "theoretical_loss": 3.3791496571921527,
+      "tokens_seen": 2673278976
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019178302038196115,
+      "loss": 1.4009,
+      "theoretical_loss": 3.3791367282690996,
+      "tokens_seen": 2673410048
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019174289841116996,
+      "loss": 1.3782,
+      "theoretical_loss": 3.379123800157387,
+      "tokens_seen": 2673541120
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019170277644037876,
+      "loss": 1.3344,
+      "theoretical_loss": 3.3791108728569235,
+      "tokens_seen": 2673672192
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019166265446958754,
+      "loss": 1.3253,
+      "theoretical_loss": 3.3790979463676196,
+      "tokens_seen": 2673803264
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019162253249879635,
+      "loss": 1.3207,
+      "theoretical_loss": 3.3790850206893834,
+      "tokens_seen": 2673934336
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019158241052800515,
+      "loss": 1.3097,
+      "theoretical_loss": 3.379072095822125,
+      "tokens_seen": 2674065408
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019154228855721393,
+      "loss": 1.3526,
+      "theoretical_loss": 3.379059171765754,
+      "tokens_seen": 2674196480
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001915021665864227,
+      "loss": 1.3708,
+      "theoretical_loss": 3.379046248520179,
+      "tokens_seen": 2674327552
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019146204461563152,
+      "loss": 1.3434,
+      "theoretical_loss": 3.3790333260853096,
+      "tokens_seen": 2674458624
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019142192264484033,
+      "loss": 1.3204,
+      "theoretical_loss": 3.3790204044610554,
+      "tokens_seen": 2674589696
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001913818006740491,
+      "loss": 1.3496,
+      "theoretical_loss": 3.3790074836473263,
+      "tokens_seen": 2674720768
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001913416787032579,
+      "loss": 1.2357,
+      "theoretical_loss": 3.378994563644031,
+      "tokens_seen": 2674851840
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019130155673246672,
+      "loss": 1.3484,
+      "theoretical_loss": 3.378981644451079,
+      "tokens_seen": 2674982912
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001912614347616755,
+      "loss": 1.2142,
+      "theoretical_loss": 3.3789687260683805,
+      "tokens_seen": 2675113984
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019122131279088428,
+      "loss": 1.2821,
+      "theoretical_loss": 3.378955808495844,
+      "tokens_seen": 2675245056
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019118119082009308,
+      "loss": 1.3116,
+      "theoretical_loss": 3.37894289173338,
+      "tokens_seen": 2675376128
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001911410688493019,
+      "loss": 1.2929,
+      "theoretical_loss": 3.3789299757808973,
+      "tokens_seen": 2675507200
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019110094687851067,
+      "loss": 1.3067,
+      "theoretical_loss": 3.3789170606383054,
+      "tokens_seen": 2675638272
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019106082490771947,
+      "loss": 1.3163,
+      "theoretical_loss": 3.3789041463055143,
+      "tokens_seen": 2675769344
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019102070293692828,
+      "loss": 1.338,
+      "theoretical_loss": 3.378891232782433,
+      "tokens_seen": 2675900416
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001909805809661371,
+      "loss": 1.3823,
+      "theoretical_loss": 3.378878320068972,
+      "tokens_seen": 2676031488
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019094045899534584,
+      "loss": 1.3258,
+      "theoretical_loss": 3.37886540816504,
+      "tokens_seen": 2676162560
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019090033702455465,
+      "loss": 1.3889,
+      "theoretical_loss": 3.378852497070547,
+      "tokens_seen": 2676293632
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019086021505376345,
+      "loss": 1.2906,
+      "theoretical_loss": 3.3788395867854026,
+      "tokens_seen": 2676424704
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019082009308297223,
+      "loss": 1.3261,
+      "theoretical_loss": 3.378826677309516,
+      "tokens_seen": 2676555776
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019077997111218104,
+      "loss": 1.2779,
+      "theoretical_loss": 3.3788137686427975,
+      "tokens_seen": 2676686848
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019073984914138984,
+      "loss": 1.3479,
+      "theoretical_loss": 3.3788008607851565,
+      "tokens_seen": 2676817920
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019069972717059862,
+      "loss": 1.3056,
+      "theoretical_loss": 3.3787879537365026,
+      "tokens_seen": 2676948992
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001906596051998074,
+      "loss": 1.3393,
+      "theoretical_loss": 3.3787750474967457,
+      "tokens_seen": 2677080064
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001906194832290162,
+      "loss": 1.2814,
+      "theoretical_loss": 3.378762142065795,
+      "tokens_seen": 2677211136
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019057936125822501,
+      "loss": 1.2346,
+      "theoretical_loss": 3.378749237443561,
+      "tokens_seen": 2677342208
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001905392392874338,
+      "loss": 1.2658,
+      "theoretical_loss": 3.3787363336299525,
+      "tokens_seen": 2677473280
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001904991173166426,
+      "loss": 1.3472,
+      "theoretical_loss": 3.3787234306248797,
+      "tokens_seen": 2677604352
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001904589953458514,
+      "loss": 1.2761,
+      "theoretical_loss": 3.378710528428253,
+      "tokens_seen": 2677735424
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019041887337506019,
+      "loss": 1.2347,
+      "theoretical_loss": 3.3786976270399816,
+      "tokens_seen": 2677866496
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019037875140426896,
+      "loss": 1.3044,
+      "theoretical_loss": 3.3786847264599746,
+      "tokens_seen": 2677997568
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019033862943347777,
+      "loss": 1.33,
+      "theoretical_loss": 3.3786718266881426,
+      "tokens_seen": 2678128640
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019029850746268658,
+      "loss": 1.3818,
+      "theoretical_loss": 3.3786589277243957,
+      "tokens_seen": 2678259712
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019025838549189536,
+      "loss": 1.3945,
+      "theoretical_loss": 3.3786460295686433,
+      "tokens_seen": 2678390784
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019021826352110416,
+      "loss": 1.3156,
+      "theoretical_loss": 3.378633132220795,
+      "tokens_seen": 2678521856
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019017814155031297,
+      "loss": 1.3328,
+      "theoretical_loss": 3.3786202356807613,
+      "tokens_seen": 2678652928
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019013801957952175,
+      "loss": 1.347,
+      "theoretical_loss": 3.3786073399484513,
+      "tokens_seen": 2678784000
+    },
+    {
+      "epoch": 0.62,
+      "objective/train/advantage_avg": 0.4920906722545624,
+      "objective/train/docs_used": 973380,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2589540481567383,
+      "objective/train/original_loss": 1.2589540481567383,
+      "objective/train/theoretical_loss": 3.3785944450237757,
+      "objective/train/tokens_used": 1049374176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24277646839618683,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504432916641235,
+      "objective/train/weighted_lm_loss": 1.3224364519119263,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3785944450237757,
+      "tokens_seen": 2678915072
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019009789760873055,
+      "loss": 1.3366,
+      "theoretical_loss": 3.3785944450237757,
+      "tokens_seen": 2678915072
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019005777563793933,
+      "loss": 1.2712,
+      "theoretical_loss": 3.378581550906644,
+      "tokens_seen": 2679046144
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00019001765366714814,
+      "loss": 1.2742,
+      "theoretical_loss": 3.3785686575969662,
+      "tokens_seen": 2679177216
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00018997753169635692,
+      "loss": 1.3021,
+      "theoretical_loss": 3.378555765094652,
+      "tokens_seen": 2679308288
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00018993740972556573,
+      "loss": 1.3175,
+      "theoretical_loss": 3.3785428733996117,
+      "tokens_seen": 2679439360
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00018989728775477453,
+      "loss": 1.2936,
+      "theoretical_loss": 3.378529982511755,
+      "tokens_seen": 2679570432
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001898571657839833,
+      "loss": 1.2943,
+      "theoretical_loss": 3.378517092430992,
+      "tokens_seen": 2679701504
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00018981704381319212,
+      "loss": 1.3783,
+      "theoretical_loss": 3.3785042031572328,
+      "tokens_seen": 2679832576
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001897769218424009,
+      "loss": 1.2106,
+      "theoretical_loss": 3.378491314690387,
+      "tokens_seen": 2679963648
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00018973679987160968,
+      "loss": 1.3541,
+      "theoretical_loss": 3.378478427030365,
+      "tokens_seen": 2680094720
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00018969667790081848,
+      "loss": 1.4363,
+      "theoretical_loss": 3.3784655401770776,
+      "tokens_seen": 2680225792
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001896565559300273,
+      "loss": 1.3445,
+      "theoretical_loss": 3.378452654130433,
+      "tokens_seen": 2680356864
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001896164339592361,
+      "loss": 1.3725,
+      "theoretical_loss": 3.3784397688903427,
+      "tokens_seen": 2680487936
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00018957631198844487,
+      "loss": 1.3094,
+      "theoretical_loss": 3.378426884456716,
+      "tokens_seen": 2680619008
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00018953619001765368,
+      "loss": 1.2808,
+      "theoretical_loss": 3.378414000829464,
+      "tokens_seen": 2680750080
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00018949606804686246,
+      "loss": 1.3327,
+      "theoretical_loss": 3.3784011180084956,
+      "tokens_seen": 2680881152
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00018945594607607124,
+      "loss": 1.2948,
+      "theoretical_loss": 3.3783882359937216,
+      "tokens_seen": 2681012224
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00018941582410528004,
+      "loss": 1.2919,
+      "theoretical_loss": 3.378375354785052,
+      "tokens_seen": 2681143296
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018937570213448885,
+      "loss": 1.3064,
+      "theoretical_loss": 3.3783624743823975,
+      "tokens_seen": 2681274368
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018933558016369766,
+      "loss": 1.3167,
+      "theoretical_loss": 3.378349594785667,
+      "tokens_seen": 2681405440
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018929545819290644,
+      "loss": 1.3442,
+      "theoretical_loss": 3.378336715994772,
+      "tokens_seen": 2681536512
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018925533622211524,
+      "loss": 1.3042,
+      "theoretical_loss": 3.3783238380096217,
+      "tokens_seen": 2681667584
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018921521425132405,
+      "loss": 1.2798,
+      "theoretical_loss": 3.378310960830127,
+      "tokens_seen": 2681798656
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001891750922805328,
+      "loss": 1.3252,
+      "theoretical_loss": 3.3782980844561976,
+      "tokens_seen": 2681929728
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001891349703097416,
+      "loss": 1.2996,
+      "theoretical_loss": 3.3782852088877444,
+      "tokens_seen": 2682060800
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018909484833895041,
+      "loss": 1.3731,
+      "theoretical_loss": 3.3782723341246768,
+      "tokens_seen": 2682191872
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018905472636815922,
+      "loss": 1.3491,
+      "theoretical_loss": 3.378259460166906,
+      "tokens_seen": 2682322944
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.000189014604397368,
+      "loss": 1.2483,
+      "theoretical_loss": 3.3782465870143414,
+      "tokens_seen": 2682454016
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001889744824265768,
+      "loss": 1.3749,
+      "theoretical_loss": 3.378233714666894,
+      "tokens_seen": 2682585088
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001889343604557856,
+      "loss": 1.2422,
+      "theoretical_loss": 3.3782208431244736,
+      "tokens_seen": 2682716160
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018889423848499436,
+      "loss": 1.2414,
+      "theoretical_loss": 3.3782079723869907,
+      "tokens_seen": 2682847232
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018885411651420317,
+      "loss": 1.3514,
+      "theoretical_loss": 3.378195102454356,
+      "tokens_seen": 2682978304
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018881399454341198,
+      "loss": 1.3256,
+      "theoretical_loss": 3.3781822333264793,
+      "tokens_seen": 2683109376
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018877387257262076,
+      "loss": 1.381,
+      "theoretical_loss": 3.378169365003271,
+      "tokens_seen": 2683240448
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018873375060182956,
+      "loss": 1.3108,
+      "theoretical_loss": 3.378156497484642,
+      "tokens_seen": 2683371520
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018869362863103837,
+      "loss": 1.3252,
+      "theoretical_loss": 3.378143630770502,
+      "tokens_seen": 2683502592
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018865350666024717,
+      "loss": 1.3148,
+      "theoretical_loss": 3.378130764860763,
+      "tokens_seen": 2683633664
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018861338468945595,
+      "loss": 1.3382,
+      "theoretical_loss": 3.378117899755333,
+      "tokens_seen": 2683764736
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018857326271866473,
+      "loss": 1.3519,
+      "theoretical_loss": 3.3781050354541238,
+      "tokens_seen": 2683895808
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018853314074787354,
+      "loss": 1.2698,
+      "theoretical_loss": 3.378092171957046,
+      "tokens_seen": 2684026880
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018849301877708232,
+      "loss": 1.3518,
+      "theoretical_loss": 3.37807930926401,
+      "tokens_seen": 2684157952
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018845289680629113,
+      "loss": 1.3011,
+      "theoretical_loss": 3.3780664473749256,
+      "tokens_seen": 2684289024
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018841277483549993,
+      "loss": 1.3469,
+      "theoretical_loss": 3.378053586289704,
+      "tokens_seen": 2684420096
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018837265286470874,
+      "loss": 1.3268,
+      "theoretical_loss": 3.3780407260082557,
+      "tokens_seen": 2684551168
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018833253089391752,
+      "loss": 1.3754,
+      "theoretical_loss": 3.3780278665304904,
+      "tokens_seen": 2684682240
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001882924089231263,
+      "loss": 1.3401,
+      "theoretical_loss": 3.3780150078563196,
+      "tokens_seen": 2684813312
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001882522869523351,
+      "loss": 1.3341,
+      "theoretical_loss": 3.3780021499856536,
+      "tokens_seen": 2684944384
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018821216498154388,
+      "loss": 1.2363,
+      "theoretical_loss": 3.3779892929184028,
+      "tokens_seen": 2685075456
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001881720430107527,
+      "loss": 1.3044,
+      "theoretical_loss": 3.377976436654478,
+      "tokens_seen": 2685206528
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001881319210399615,
+      "loss": 1.3305,
+      "theoretical_loss": 3.3779635811937894,
+      "tokens_seen": 2685337600
+    },
+    {
+      "epoch": 0.63,
+      "objective/train/advantage_avg": 0.4927713871002197,
+      "objective/train/docs_used": 975809,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.164670467376709,
+      "objective/train/original_loss": 1.164670467376709,
+      "objective/train/theoretical_loss": 3.3779507265362483,
+      "objective/train/tokens_used": 1055927776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24322357773780823,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505136251449585,
+      "objective/train/weighted_lm_loss": 1.2235887050628662,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3779507265362483,
+      "tokens_seen": 2685468672
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001880917990691703,
+      "loss": 1.2266,
+      "theoretical_loss": 3.3779507265362483,
+      "tokens_seen": 2685468672
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018805167709837908,
+      "loss": 1.3109,
+      "theoretical_loss": 3.3779378726817644,
+      "tokens_seen": 2685599744
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018801155512758786,
+      "loss": 1.4159,
+      "theoretical_loss": 3.3779250196302493,
+      "tokens_seen": 2685730816
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018797143315679667,
+      "loss": 1.3069,
+      "theoretical_loss": 3.377912167381613,
+      "tokens_seen": 2685861888
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018793131118600544,
+      "loss": 1.3338,
+      "theoretical_loss": 3.377899315935766,
+      "tokens_seen": 2685992960
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018789118921521425,
+      "loss": 1.2851,
+      "theoretical_loss": 3.3778864652926197,
+      "tokens_seen": 2686124032
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018785106724442306,
+      "loss": 1.316,
+      "theoretical_loss": 3.3778736154520845,
+      "tokens_seen": 2686255104
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018781094527363186,
+      "loss": 1.397,
+      "theoretical_loss": 3.377860766414071,
+      "tokens_seen": 2686386176
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018777082330284064,
+      "loss": 1.2877,
+      "theoretical_loss": 3.37784791817849,
+      "tokens_seen": 2686517248
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018773070133204945,
+      "loss": 1.266,
+      "theoretical_loss": 3.3778350707452525,
+      "tokens_seen": 2686648320
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018769057936125823,
+      "loss": 1.2939,
+      "theoretical_loss": 3.3778222241142686,
+      "tokens_seen": 2686779392
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.000187650457390467,
+      "loss": 1.3985,
+      "theoretical_loss": 3.3778093782854492,
+      "tokens_seen": 2686910464
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018761033541967581,
+      "loss": 1.2784,
+      "theoretical_loss": 3.377796533258706,
+      "tokens_seen": 2687041536
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018757021344888462,
+      "loss": 1.3413,
+      "theoretical_loss": 3.377783689033949,
+      "tokens_seen": 2687172608
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001875300914780934,
+      "loss": 1.3003,
+      "theoretical_loss": 3.3777708456110886,
+      "tokens_seen": 2687303680
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001874899695073022,
+      "loss": 1.2411,
+      "theoretical_loss": 3.3777580029900367,
+      "tokens_seen": 2687434752
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.000187449847536511,
+      "loss": 1.3809,
+      "theoretical_loss": 3.3777451611707034,
+      "tokens_seen": 2687565824
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001874097255657198,
+      "loss": 1.393,
+      "theoretical_loss": 3.377732320153,
+      "tokens_seen": 2687696896
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018736960359492857,
+      "loss": 1.3295,
+      "theoretical_loss": 3.377719479936837,
+      "tokens_seen": 2687827968
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018732948162413738,
+      "loss": 1.2268,
+      "theoretical_loss": 3.3777066405221254,
+      "tokens_seen": 2687959040
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018728935965334618,
+      "loss": 1.3017,
+      "theoretical_loss": 3.3776938019087765,
+      "tokens_seen": 2688090112
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018724923768255496,
+      "loss": 1.3039,
+      "theoretical_loss": 3.3776809640967,
+      "tokens_seen": 2688221184
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018720911571176377,
+      "loss": 1.3233,
+      "theoretical_loss": 3.3776681270858084,
+      "tokens_seen": 2688352256
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018716899374097257,
+      "loss": 1.4049,
+      "theoretical_loss": 3.377655290876012,
+      "tokens_seen": 2688483328
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018712887177018135,
+      "loss": 1.3134,
+      "theoretical_loss": 3.377642455467221,
+      "tokens_seen": 2688614400
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018708874979939013,
+      "loss": 1.3735,
+      "theoretical_loss": 3.3776296208593477,
+      "tokens_seen": 2688745472
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018704862782859894,
+      "loss": 1.2457,
+      "theoretical_loss": 3.377616787052302,
+      "tokens_seen": 2688876544
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018700850585780775,
+      "loss": 1.407,
+      "theoretical_loss": 3.3776039540459957,
+      "tokens_seen": 2689007616
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018696838388701652,
+      "loss": 1.2457,
+      "theoretical_loss": 3.377591121840339,
+      "tokens_seen": 2689138688
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018692826191622533,
+      "loss": 1.3907,
+      "theoretical_loss": 3.377578290435243,
+      "tokens_seen": 2689269760
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018688813994543414,
+      "loss": 1.2363,
+      "theoretical_loss": 3.37756545983062,
+      "tokens_seen": 2689400832
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018684801797464294,
+      "loss": 1.3172,
+      "theoretical_loss": 3.3775526300263796,
+      "tokens_seen": 2689531904
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001868078960038517,
+      "loss": 1.2613,
+      "theoretical_loss": 3.377539801022434,
+      "tokens_seen": 2689662976
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001867677740330605,
+      "loss": 1.2519,
+      "theoretical_loss": 3.3775269728186927,
+      "tokens_seen": 2689794048
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001867276520622693,
+      "loss": 1.3263,
+      "theoretical_loss": 3.3775141454150686,
+      "tokens_seen": 2689925120
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001866875300914781,
+      "loss": 1.3323,
+      "theoretical_loss": 3.377501318811472,
+      "tokens_seen": 2690056192
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001866474081206869,
+      "loss": 1.3805,
+      "theoretical_loss": 3.3774884930078137,
+      "tokens_seen": 2690187264
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001866072861498957,
+      "loss": 1.3364,
+      "theoretical_loss": 3.3774756680040054,
+      "tokens_seen": 2690318336
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018656716417910448,
+      "loss": 1.4125,
+      "theoretical_loss": 3.377462843799958,
+      "tokens_seen": 2690449408
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018652704220831326,
+      "loss": 1.3142,
+      "theoretical_loss": 3.3774500203955826,
+      "tokens_seen": 2690580480
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018648692023752206,
+      "loss": 1.339,
+      "theoretical_loss": 3.3774371977907904,
+      "tokens_seen": 2690711552
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018644679826673087,
+      "loss": 1.3752,
+      "theoretical_loss": 3.377424375985493,
+      "tokens_seen": 2690842624
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018640667629593965,
+      "loss": 1.3618,
+      "theoretical_loss": 3.377411554979601,
+      "tokens_seen": 2690973696
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018636655432514846,
+      "loss": 1.2928,
+      "theoretical_loss": 3.377398734773026,
+      "tokens_seen": 2691104768
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018632643235435726,
+      "loss": 1.3333,
+      "theoretical_loss": 3.377385915365679,
+      "tokens_seen": 2691235840
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018628631038356604,
+      "loss": 1.3381,
+      "theoretical_loss": 3.3773730967574718,
+      "tokens_seen": 2691366912
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018624618841277485,
+      "loss": 1.3967,
+      "theoretical_loss": 3.377360278948315,
+      "tokens_seen": 2691497984
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018620606644198363,
+      "loss": 1.349,
+      "theoretical_loss": 3.37734746193812,
+      "tokens_seen": 2691629056
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018616594447119243,
+      "loss": 1.3505,
+      "theoretical_loss": 3.3773346457267985,
+      "tokens_seen": 2691760128
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001861258225004012,
+      "loss": 1.3161,
+      "theoretical_loss": 3.377321830314262,
+      "tokens_seen": 2691891200
+    },
+    {
+      "epoch": 0.63,
+      "objective/train/advantage_avg": 0.49273455142974854,
+      "objective/train/docs_used": 978239,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.219112753868103,
+      "objective/train/original_loss": 1.219112753868103,
+      "objective/train/theoretical_loss": 3.377309015700421,
+      "objective/train/tokens_used": 1062481376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24327579140663147,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505101680755615,
+      "objective/train/weighted_lm_loss": 1.280524730682373,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.377309015700421,
+      "tokens_seen": 2692022272
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018608570052961002,
+      "loss": 1.302,
+      "theoretical_loss": 3.377309015700421,
+      "tokens_seen": 2692022272
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018604557855881883,
+      "loss": 1.3306,
+      "theoretical_loss": 3.377296201885187,
+      "tokens_seen": 2692153344
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001860054565880276,
+      "loss": 1.3264,
+      "theoretical_loss": 3.377283388868472,
+      "tokens_seen": 2692284416
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001859653346172364,
+      "loss": 1.3513,
+      "theoretical_loss": 3.3772705766501865,
+      "tokens_seen": 2692415488
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001859252126464452,
+      "loss": 1.3718,
+      "theoretical_loss": 3.3772577652302425,
+      "tokens_seen": 2692546560
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.000185885090675654,
+      "loss": 1.3699,
+      "theoretical_loss": 3.3772449546085515,
+      "tokens_seen": 2692677632
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018584496870486278,
+      "loss": 1.3489,
+      "theoretical_loss": 3.377232144785024,
+      "tokens_seen": 2692808704
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018580484673407158,
+      "loss": 1.4442,
+      "theoretical_loss": 3.377219335759573,
+      "tokens_seen": 2692939776
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001857647247632804,
+      "loss": 1.3162,
+      "theoretical_loss": 3.377206527532108,
+      "tokens_seen": 2693070848
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018572460279248917,
+      "loss": 1.3159,
+      "theoretical_loss": 3.377193720102542,
+      "tokens_seen": 2693201920
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018568448082169797,
+      "loss": 1.3604,
+      "theoretical_loss": 3.377180913470786,
+      "tokens_seen": 2693332992
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018564435885090675,
+      "loss": 1.3573,
+      "theoretical_loss": 3.3771681076367512,
+      "tokens_seen": 2693464064
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018560423688011553,
+      "loss": 1.2817,
+      "theoretical_loss": 3.3771553026003494,
+      "tokens_seen": 2693595136
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018556411490932434,
+      "loss": 1.4523,
+      "theoretical_loss": 3.3771424983614917,
+      "tokens_seen": 2693726208
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018552399293853315,
+      "loss": 1.2919,
+      "theoretical_loss": 3.3771296949200904,
+      "tokens_seen": 2693857280
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018548387096774195,
+      "loss": 1.3212,
+      "theoretical_loss": 3.377116892276056,
+      "tokens_seen": 2693988352
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018544374899695073,
+      "loss": 1.3146,
+      "theoretical_loss": 3.377104090429301,
+      "tokens_seen": 2694119424
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018540362702615954,
+      "loss": 1.3308,
+      "theoretical_loss": 3.3770912893797367,
+      "tokens_seen": 2694250496
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018536350505536834,
+      "loss": 1.3668,
+      "theoretical_loss": 3.3770784891272743,
+      "tokens_seen": 2694381568
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001853233830845771,
+      "loss": 1.3845,
+      "theoretical_loss": 3.3770656896718254,
+      "tokens_seen": 2694512640
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001852832611137859,
+      "loss": 1.3907,
+      "theoretical_loss": 3.377052891013302,
+      "tokens_seen": 2694643712
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001852431391429947,
+      "loss": 1.2975,
+      "theoretical_loss": 3.3770400931516162,
+      "tokens_seen": 2694774784
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018520301717220351,
+      "loss": 1.2969,
+      "theoretical_loss": 3.3770272960866783,
+      "tokens_seen": 2694905856
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001851628952014123,
+      "loss": 1.2971,
+      "theoretical_loss": 3.377014499818401,
+      "tokens_seen": 2695036928
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001851227732306211,
+      "loss": 1.3788,
+      "theoretical_loss": 3.3770017043466956,
+      "tokens_seen": 2695168000
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001850826512598299,
+      "loss": 1.3224,
+      "theoretical_loss": 3.3769889096714736,
+      "tokens_seen": 2695299072
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018504252928903866,
+      "loss": 1.2718,
+      "theoretical_loss": 3.376976115792647,
+      "tokens_seen": 2695430144
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018500240731824746,
+      "loss": 1.3502,
+      "theoretical_loss": 3.3769633227101274,
+      "tokens_seen": 2695561216
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018496228534745627,
+      "loss": 1.2368,
+      "theoretical_loss": 3.3769505304238265,
+      "tokens_seen": 2695692288
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018492216337666508,
+      "loss": 1.3199,
+      "theoretical_loss": 3.3769377389336563,
+      "tokens_seen": 2695823360
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018488204140587386,
+      "loss": 1.2641,
+      "theoretical_loss": 3.3769249482395276,
+      "tokens_seen": 2695954432
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018484191943508266,
+      "loss": 1.2634,
+      "theoretical_loss": 3.3769121583413537,
+      "tokens_seen": 2696085504
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018480179746429147,
+      "loss": 1.3279,
+      "theoretical_loss": 3.376899369239045,
+      "tokens_seen": 2696216576
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018476167549350022,
+      "loss": 1.4136,
+      "theoretical_loss": 3.376886580932514,
+      "tokens_seen": 2696347648
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018472155352270903,
+      "loss": 1.3463,
+      "theoretical_loss": 3.376873793421672,
+      "tokens_seen": 2696478720
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018468143155191783,
+      "loss": 1.3253,
+      "theoretical_loss": 3.3768610067064317,
+      "tokens_seen": 2696609792
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001846413095811266,
+      "loss": 1.3285,
+      "theoretical_loss": 3.3768482207867043,
+      "tokens_seen": 2696740864
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018460118761033542,
+      "loss": 1.3063,
+      "theoretical_loss": 3.3768354356624015,
+      "tokens_seen": 2696871936
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018456106563954423,
+      "loss": 1.3452,
+      "theoretical_loss": 3.376822651333436,
+      "tokens_seen": 2697003008
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018452094366875303,
+      "loss": 1.2952,
+      "theoretical_loss": 3.376809867799718,
+      "tokens_seen": 2697134080
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001844808216979618,
+      "loss": 1.2932,
+      "theoretical_loss": 3.3767970850611615,
+      "tokens_seen": 2697265152
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001844406997271706,
+      "loss": 1.2854,
+      "theoretical_loss": 3.3767843031176765,
+      "tokens_seen": 2697396224
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001844005777563794,
+      "loss": 1.3017,
+      "theoretical_loss": 3.376771521969176,
+      "tokens_seen": 2697527296
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018436045578558818,
+      "loss": 1.2705,
+      "theoretical_loss": 3.376758741615572,
+      "tokens_seen": 2697658368
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018432033381479698,
+      "loss": 1.3499,
+      "theoretical_loss": 3.376745962056776,
+      "tokens_seen": 2697789440
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001842802118440058,
+      "loss": 1.2819,
+      "theoretical_loss": 3.3767331832927003,
+      "tokens_seen": 2697920512
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001842400898732146,
+      "loss": 1.3487,
+      "theoretical_loss": 3.3767204053232565,
+      "tokens_seen": 2698051584
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018419996790242337,
+      "loss": 1.3394,
+      "theoretical_loss": 3.3767076281483566,
+      "tokens_seen": 2698182656
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018415984593163215,
+      "loss": 1.2851,
+      "theoretical_loss": 3.3766948517679127,
+      "tokens_seen": 2698313728
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018411972396084096,
+      "loss": 1.2257,
+      "theoretical_loss": 3.3766820761818375,
+      "tokens_seen": 2698444800
+    },
+    {
+      "epoch": 0.64,
+      "objective/train/advantage_avg": 0.4915187954902649,
+      "objective/train/docs_used": 980656,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.397096037864685,
+      "objective/train/original_loss": 1.3970959186553955,
+      "objective/train/theoretical_loss": 3.376669301390042,
+      "objective/train/tokens_used": 1069034976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24237337708473206,
+      "objective/train/value_max": -0.5,
+      "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": 1.4674549102783203,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.376669301390042,
+      "tokens_seen": 2698575872
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018407960199004974,
+      "loss": 1.3863,
+      "theoretical_loss": 3.376669301390042,
+      "tokens_seen": 2698575872
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018403948001925854,
+      "loss": 1.2744,
+      "theoretical_loss": 3.3766565273924387,
+      "tokens_seen": 2698706944
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018399935804846735,
+      "loss": 1.3879,
+      "theoretical_loss": 3.3766437541889394,
+      "tokens_seen": 2698838016
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018395923607767616,
+      "loss": 1.3561,
+      "theoretical_loss": 3.376630981779457,
+      "tokens_seen": 2698969088
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018391911410688494,
+      "loss": 1.3997,
+      "theoretical_loss": 3.376618210163902,
+      "tokens_seen": 2699100160
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018387899213609374,
+      "loss": 1.3016,
+      "theoretical_loss": 3.376605439342188,
+      "tokens_seen": 2699231232
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018383887016530252,
+      "loss": 1.3996,
+      "theoretical_loss": 3.3765926693142267,
+      "tokens_seen": 2699362304
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001837987481945113,
+      "loss": 1.3328,
+      "theoretical_loss": 3.37657990007993,
+      "tokens_seen": 2699493376
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001837586262237201,
+      "loss": 1.3028,
+      "theoretical_loss": 3.3765671316392103,
+      "tokens_seen": 2699624448
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018371850425292891,
+      "loss": 1.3683,
+      "theoretical_loss": 3.3765543639919793,
+      "tokens_seen": 2699755520
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018367838228213772,
+      "loss": 1.2851,
+      "theoretical_loss": 3.37654159713815,
+      "tokens_seen": 2699886592
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001836382603113465,
+      "loss": 1.3074,
+      "theoretical_loss": 3.376528831077634,
+      "tokens_seen": 2700017664
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001835981383405553,
+      "loss": 1.2854,
+      "theoretical_loss": 3.376516065810343,
+      "tokens_seen": 2700148736
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018355801636976409,
+      "loss": 1.3178,
+      "theoretical_loss": 3.37650330133619,
+      "tokens_seen": 2700279808
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018351789439897286,
+      "loss": 1.3166,
+      "theoretical_loss": 3.3764905376550867,
+      "tokens_seen": 2700410880
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018347777242818167,
+      "loss": 1.3551,
+      "theoretical_loss": 3.376477774766946,
+      "tokens_seen": 2700541952
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018343765045739048,
+      "loss": 1.3632,
+      "theoretical_loss": 3.3764650126716798,
+      "tokens_seen": 2700673024
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018339752848659926,
+      "loss": 1.3935,
+      "theoretical_loss": 3.3764522513692,
+      "tokens_seen": 2700804096
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018335740651580806,
+      "loss": 1.3546,
+      "theoretical_loss": 3.3764394908594197,
+      "tokens_seen": 2700935168
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018331728454501687,
+      "loss": 1.3692,
+      "theoretical_loss": 3.3764267311422507,
+      "tokens_seen": 2701066240
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018327716257422565,
+      "loss": 1.2453,
+      "theoretical_loss": 3.376413972217605,
+      "tokens_seen": 2701197312
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018323704060343443,
+      "loss": 1.4255,
+      "theoretical_loss": 3.3764012140853956,
+      "tokens_seen": 2701328384
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018319691863264323,
+      "loss": 1.3252,
+      "theoretical_loss": 3.3763884567455342,
+      "tokens_seen": 2701459456
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018315679666185204,
+      "loss": 1.3532,
+      "theoretical_loss": 3.3763757001979338,
+      "tokens_seen": 2701590528
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018311667469106082,
+      "loss": 1.3546,
+      "theoretical_loss": 3.376362944442506,
+      "tokens_seen": 2701721600
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018307655272026963,
+      "loss": 1.3775,
+      "theoretical_loss": 3.3763501894791634,
+      "tokens_seen": 2701852672
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018303643074947843,
+      "loss": 1.306,
+      "theoretical_loss": 3.376337435307819,
+      "tokens_seen": 2701983744
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018299630877868724,
+      "loss": 1.3319,
+      "theoretical_loss": 3.3763246819283848,
+      "tokens_seen": 2702114816
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.000182956186807896,
+      "loss": 1.2839,
+      "theoretical_loss": 3.376311929340773,
+      "tokens_seen": 2702245888
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001829160648371048,
+      "loss": 1.3109,
+      "theoretical_loss": 3.3762991775448965,
+      "tokens_seen": 2702376960
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001828759428663136,
+      "loss": 1.2946,
+      "theoretical_loss": 3.376286426540667,
+      "tokens_seen": 2702508032
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018283582089552238,
+      "loss": 1.3451,
+      "theoretical_loss": 3.3762736763279984,
+      "tokens_seen": 2702639104
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001827956989247312,
+      "loss": 1.3348,
+      "theoretical_loss": 3.3762609269068014,
+      "tokens_seen": 2702770176
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018275557695394,
+      "loss": 1.3886,
+      "theoretical_loss": 3.3762481782769895,
+      "tokens_seen": 2702901248
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001827154549831488,
+      "loss": 1.3528,
+      "theoretical_loss": 3.376235430438475,
+      "tokens_seen": 2703032320
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018267533301235755,
+      "loss": 1.2972,
+      "theoretical_loss": 3.3762226833911706,
+      "tokens_seen": 2703163392
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018263521104156636,
+      "loss": 1.3411,
+      "theoretical_loss": 3.3762099371349885,
+      "tokens_seen": 2703294464
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018259508907077517,
+      "loss": 1.2862,
+      "theoretical_loss": 3.3761971916698412,
+      "tokens_seen": 2703425536
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018255496709998394,
+      "loss": 1.3647,
+      "theoretical_loss": 3.376184446995642,
+      "tokens_seen": 2703556608
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018251484512919275,
+      "loss": 1.2609,
+      "theoretical_loss": 3.376171703112303,
+      "tokens_seen": 2703687680
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018247472315840156,
+      "loss": 1.3038,
+      "theoretical_loss": 3.376158960019736,
+      "tokens_seen": 2703818752
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018243460118761034,
+      "loss": 1.3913,
+      "theoretical_loss": 3.3761462177178547,
+      "tokens_seen": 2703949824
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018239447921681912,
+      "loss": 1.3871,
+      "theoretical_loss": 3.3761334762065713,
+      "tokens_seen": 2704080896
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018235435724602792,
+      "loss": 1.363,
+      "theoretical_loss": 3.3761207354857987,
+      "tokens_seen": 2704211968
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018231423527523673,
+      "loss": 1.3892,
+      "theoretical_loss": 3.3761079955554494,
+      "tokens_seen": 2704343040
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001822741133044455,
+      "loss": 1.409,
+      "theoretical_loss": 3.376095256415436,
+      "tokens_seen": 2704474112
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018223399133365431,
+      "loss": 1.3114,
+      "theoretical_loss": 3.3760825180656706,
+      "tokens_seen": 2704605184
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018219386936286312,
+      "loss": 1.3841,
+      "theoretical_loss": 3.376069780506067,
+      "tokens_seen": 2704736256
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001821537473920719,
+      "loss": 1.3656,
+      "theoretical_loss": 3.376057043736537,
+      "tokens_seen": 2704867328
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001821136254212807,
+      "loss": 1.2988,
+      "theoretical_loss": 3.3760443077569935,
+      "tokens_seen": 2704998400
+    },
+    {
+      "epoch": 0.64,
+      "objective/train/advantage_avg": 0.4928801357746124,
+      "objective/train/docs_used": 983087,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.4077523946762085,
+      "objective/train/original_loss": 1.407752513885498,
+      "objective/train/theoretical_loss": 3.3760315725673498,
+      "objective/train/tokens_used": 1075588576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24333065748214722,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050525188446045,
+      "objective/train/weighted_lm_loss": 1.4788596630096436,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3760315725673498,
+      "tokens_seen": 2705129472
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018207350345048948,
+      "loss": 1.4689,
+      "theoretical_loss": 3.3760315725673498,
+      "tokens_seen": 2705129472
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001820333814796983,
+      "loss": 1.3373,
+      "theoretical_loss": 3.3760188381675182,
+      "tokens_seen": 2705260544
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018199325950890707,
+      "loss": 1.3235,
+      "theoretical_loss": 3.376006104557411,
+      "tokens_seen": 2705391616
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018195313753811588,
+      "loss": 1.3173,
+      "theoretical_loss": 3.375993371736942,
+      "tokens_seen": 2705522688
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018191301556732468,
+      "loss": 1.2855,
+      "theoretical_loss": 3.375980639706023,
+      "tokens_seen": 2705653760
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018187289359653346,
+      "loss": 1.2822,
+      "theoretical_loss": 3.3759679084645677,
+      "tokens_seen": 2705784832
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018183277162574227,
+      "loss": 1.3589,
+      "theoretical_loss": 3.3759551780124877,
+      "tokens_seen": 2705915904
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018179264965495105,
+      "loss": 1.3593,
+      "theoretical_loss": 3.375942448349697,
+      "tokens_seen": 2706046976
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018175252768415985,
+      "loss": 1.2736,
+      "theoretical_loss": 3.375929719476108,
+      "tokens_seen": 2706178048
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018171240571336863,
+      "loss": 1.3123,
+      "theoretical_loss": 3.3759169913916334,
+      "tokens_seen": 2706309120
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018167228374257744,
+      "loss": 1.3028,
+      "theoretical_loss": 3.375904264096186,
+      "tokens_seen": 2706440192
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018163216177178625,
+      "loss": 1.365,
+      "theoretical_loss": 3.375891537589679,
+      "tokens_seen": 2706571264
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018159203980099502,
+      "loss": 1.3992,
+      "theoretical_loss": 3.3758788118720258,
+      "tokens_seen": 2706702336
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018155191783020383,
+      "loss": 1.4035,
+      "theoretical_loss": 3.3758660869431383,
+      "tokens_seen": 2706833408
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018151179585941264,
+      "loss": 1.424,
+      "theoretical_loss": 3.3758533628029292,
+      "tokens_seen": 2706964480
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001814716738886214,
+      "loss": 1.3559,
+      "theoretical_loss": 3.3758406394513125,
+      "tokens_seen": 2707095552
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001814315519178302,
+      "loss": 1.445,
+      "theoretical_loss": 3.3758279168882006,
+      "tokens_seen": 2707226624
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.000181391429947039,
+      "loss": 1.307,
+      "theoretical_loss": 3.375815195113507,
+      "tokens_seen": 2707357696
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001813513079762478,
+      "loss": 1.4498,
+      "theoretical_loss": 3.3758024741271435,
+      "tokens_seen": 2707488768
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001813111860054566,
+      "loss": 1.4021,
+      "theoretical_loss": 3.3757897539290243,
+      "tokens_seen": 2707619840
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001812710640346654,
+      "loss": 1.3217,
+      "theoretical_loss": 3.3757770345190616,
+      "tokens_seen": 2707750912
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001812309420638742,
+      "loss": 1.3259,
+      "theoretical_loss": 3.3757643158971686,
+      "tokens_seen": 2707881984
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018119082009308295,
+      "loss": 1.2999,
+      "theoretical_loss": 3.3757515980632586,
+      "tokens_seen": 2708013056
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018115069812229176,
+      "loss": 1.3074,
+      "theoretical_loss": 3.375738881017245,
+      "tokens_seen": 2708144128
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018111057615150057,
+      "loss": 1.4333,
+      "theoretical_loss": 3.3757261647590395,
+      "tokens_seen": 2708275200
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018107045418070937,
+      "loss": 1.3738,
+      "theoretical_loss": 3.3757134492885568,
+      "tokens_seen": 2708406272
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018103033220991815,
+      "loss": 1.4095,
+      "theoretical_loss": 3.375700734605709,
+      "tokens_seen": 2708537344
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018099021023912696,
+      "loss": 1.3383,
+      "theoretical_loss": 3.3756880207104087,
+      "tokens_seen": 2708668416
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018095008826833576,
+      "loss": 1.2574,
+      "theoretical_loss": 3.3756753076025703,
+      "tokens_seen": 2708799488
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018090996629754452,
+      "loss": 1.4571,
+      "theoretical_loss": 3.3756625952821064,
+      "tokens_seen": 2708930560
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018086984432675332,
+      "loss": 1.423,
+      "theoretical_loss": 3.37564988374893,
+      "tokens_seen": 2709061632
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018082972235596213,
+      "loss": 1.4137,
+      "theoretical_loss": 3.3756371730029544,
+      "tokens_seen": 2709192704
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018078960038517093,
+      "loss": 1.3949,
+      "theoretical_loss": 3.375624463044093,
+      "tokens_seen": 2709323776
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001807494784143797,
+      "loss": 1.3768,
+      "theoretical_loss": 3.375611753872258,
+      "tokens_seen": 2709454848
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018070935644358852,
+      "loss": 1.3851,
+      "theoretical_loss": 3.375599045487364,
+      "tokens_seen": 2709585920
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018066923447279733,
+      "loss": 1.4291,
+      "theoretical_loss": 3.3755863378893234,
+      "tokens_seen": 2709716992
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001806291125020061,
+      "loss": 1.3633,
+      "theoretical_loss": 3.375573631078049,
+      "tokens_seen": 2709848064
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018058899053121488,
+      "loss": 1.2996,
+      "theoretical_loss": 3.375560925053455,
+      "tokens_seen": 2709979136
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001805488685604237,
+      "loss": 1.3431,
+      "theoretical_loss": 3.375548219815454,
+      "tokens_seen": 2710110208
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018050874658963247,
+      "loss": 1.3886,
+      "theoretical_loss": 3.3755355153639597,
+      "tokens_seen": 2710241280
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018046862461884128,
+      "loss": 1.3662,
+      "theoretical_loss": 3.375522811698885,
+      "tokens_seen": 2710372352
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018042850264805008,
+      "loss": 1.3524,
+      "theoretical_loss": 3.3755101088201434,
+      "tokens_seen": 2710503424
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001803883806772589,
+      "loss": 1.3094,
+      "theoretical_loss": 3.375497406727648,
+      "tokens_seen": 2710634496
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018034825870646767,
+      "loss": 1.4436,
+      "theoretical_loss": 3.3754847054213126,
+      "tokens_seen": 2710765568
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018030813673567645,
+      "loss": 1.3799,
+      "theoretical_loss": 3.3754720049010505,
+      "tokens_seen": 2710896640
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018026801476488525,
+      "loss": 1.3816,
+      "theoretical_loss": 3.375459305166774,
+      "tokens_seen": 2711027712
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018022789279409403,
+      "loss": 1.3495,
+      "theoretical_loss": 3.3754466062183974,
+      "tokens_seen": 2711158784
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018018777082330284,
+      "loss": 1.2444,
+      "theoretical_loss": 3.375433908055834,
+      "tokens_seen": 2711289856
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018014764885251165,
+      "loss": 1.389,
+      "theoretical_loss": 3.375421210678997,
+      "tokens_seen": 2711420928
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018010752688172045,
+      "loss": 1.4601,
+      "theoretical_loss": 3.3754085140878005,
+      "tokens_seen": 2711552000
+    },
+    {
+      "epoch": 0.64,
+      "objective/train/advantage_avg": 0.49413126707077026,
+      "objective/train/docs_used": 985372,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.286076307296753,
+      "objective/train/original_loss": 1.286076307296753,
+      "objective/train/theoretical_loss": 3.375395818282157,
+      "objective/train/tokens_used": 1082142176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24463817477226257,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506569147109985,
+      "objective/train/weighted_lm_loss": 1.3512893915176392,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9048373699188232,
+      "theoretical_loss": 3.375395818282157,
+      "tokens_seen": 2711683072
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018006740491092923,
+      "loss": 1.3663,
+      "theoretical_loss": 3.375395818282157,
+      "tokens_seen": 2711683072
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.000180027282940138,
+      "loss": 1.3197,
+      "theoretical_loss": 3.3753831232619795,
+      "tokens_seen": 2711814144
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00017998716096934682,
+      "loss": 1.323,
+      "theoretical_loss": 3.375370429027183,
+      "tokens_seen": 2711945216
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001799470389985556,
+      "loss": 1.3328,
+      "theoretical_loss": 3.37535773557768,
+      "tokens_seen": 2712076288
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001799069170277644,
+      "loss": 1.3423,
+      "theoretical_loss": 3.3753450429133838,
+      "tokens_seen": 2712207360
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001798667950569732,
+      "loss": 1.3609,
+      "theoretical_loss": 3.375332351034208,
+      "tokens_seen": 2712338432
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00017982667308618201,
+      "loss": 1.3646,
+      "theoretical_loss": 3.375319659940067,
+      "tokens_seen": 2712469504
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001797865511153908,
+      "loss": 1.3827,
+      "theoretical_loss": 3.3753069696308735,
+      "tokens_seen": 2712600576
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001797464291445996,
+      "loss": 1.5161,
+      "theoretical_loss": 3.375294280106541,
+      "tokens_seen": 2712731648
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00017970630717380838,
+      "loss": 1.3611,
+      "theoretical_loss": 3.3752815913669827,
+      "tokens_seen": 2712862720
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00017966618520301716,
+      "loss": 1.3722,
+      "theoretical_loss": 3.3752689034121133,
+      "tokens_seen": 2712993792
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00017962606323222596,
+      "loss": 1.307,
+      "theoretical_loss": 3.375256216241845,
+      "tokens_seen": 2713124864
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00017958594126143477,
+      "loss": 1.2862,
+      "theoretical_loss": 3.3752435298560926,
+      "tokens_seen": 2713255936
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00017954581929064358,
+      "loss": 1.4525,
+      "theoretical_loss": 3.375230844254769,
+      "tokens_seen": 2713387008
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00017950569731985236,
+      "loss": 1.3907,
+      "theoretical_loss": 3.3752181594377877,
+      "tokens_seen": 2713518080
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00017946557534906116,
+      "loss": 1.2867,
+      "theoretical_loss": 3.375205475405063,
+      "tokens_seen": 2713649152
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00017942545337826994,
+      "loss": 1.4737,
+      "theoretical_loss": 3.375192792156508,
+      "tokens_seen": 2713780224
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00017938533140747872,
+      "loss": 1.3972,
+      "theoretical_loss": 3.375180109692036,
+      "tokens_seen": 2713911296
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00017934520943668753,
+      "loss": 1.3455,
+      "theoretical_loss": 3.375167428011562,
+      "tokens_seen": 2714042368
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00017930508746589633,
+      "loss": 1.3289,
+      "theoretical_loss": 3.3751547471149985,
+      "tokens_seen": 2714173440
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001792649654951051,
+      "loss": 1.331,
+      "theoretical_loss": 3.3751420670022596,
+      "tokens_seen": 2714304512
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017922484352431392,
+      "loss": 1.3574,
+      "theoretical_loss": 3.3751293876732587,
+      "tokens_seen": 2714435584
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017918472155352273,
+      "loss": 1.4511,
+      "theoretical_loss": 3.3751167091279095,
+      "tokens_seen": 2714566656
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017914459958273153,
+      "loss": 1.3666,
+      "theoretical_loss": 3.3751040313661265,
+      "tokens_seen": 2714697728
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017910447761194028,
+      "loss": 1.3947,
+      "theoretical_loss": 3.3750913543878225,
+      "tokens_seen": 2714828800
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001790643556411491,
+      "loss": 1.3382,
+      "theoretical_loss": 3.375078678192912,
+      "tokens_seen": 2714959872
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001790242336703579,
+      "loss": 1.333,
+      "theoretical_loss": 3.375066002781308,
+      "tokens_seen": 2715090944
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017898411169956668,
+      "loss": 1.3735,
+      "theoretical_loss": 3.3750533281529247,
+      "tokens_seen": 2715222016
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017894398972877548,
+      "loss": 1.2138,
+      "theoretical_loss": 3.3750406543076763,
+      "tokens_seen": 2715353088
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001789038677579843,
+      "loss": 1.3508,
+      "theoretical_loss": 3.375027981245476,
+      "tokens_seen": 2715484160
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001788637457871931,
+      "loss": 1.3543,
+      "theoretical_loss": 3.3750153089662382,
+      "tokens_seen": 2715615232
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017882362381640185,
+      "loss": 1.2234,
+      "theoretical_loss": 3.375002637469876,
+      "tokens_seen": 2715746304
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017878350184561065,
+      "loss": 1.3419,
+      "theoretical_loss": 3.3749899667563037,
+      "tokens_seen": 2715877376
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017874337987481946,
+      "loss": 1.2982,
+      "theoretical_loss": 3.3749772968254352,
+      "tokens_seen": 2716008448
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017870325790402824,
+      "loss": 1.3061,
+      "theoretical_loss": 3.3749646276771843,
+      "tokens_seen": 2716139520
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017866313593323704,
+      "loss": 1.3396,
+      "theoretical_loss": 3.374951959311465,
+      "tokens_seen": 2716270592
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017862301396244585,
+      "loss": 1.256,
+      "theoretical_loss": 3.3749392917281913,
+      "tokens_seen": 2716401664
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017858289199165466,
+      "loss": 1.3281,
+      "theoretical_loss": 3.3749266249272765,
+      "tokens_seen": 2716532736
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001785427700208634,
+      "loss": 1.3227,
+      "theoretical_loss": 3.374913958908635,
+      "tokens_seen": 2716663808
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017850264805007222,
+      "loss": 1.2895,
+      "theoretical_loss": 3.374901293672181,
+      "tokens_seen": 2716794880
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017846252607928102,
+      "loss": 1.3258,
+      "theoretical_loss": 3.374888629217828,
+      "tokens_seen": 2716925952
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001784224041084898,
+      "loss": 1.384,
+      "theoretical_loss": 3.37487596554549,
+      "tokens_seen": 2717057024
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001783822821376986,
+      "loss": 1.3353,
+      "theoretical_loss": 3.3748633026550814,
+      "tokens_seen": 2717188096
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017834216016690741,
+      "loss": 1.3394,
+      "theoretical_loss": 3.3748506405465157,
+      "tokens_seen": 2717319168
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001783020381961162,
+      "loss": 1.3522,
+      "theoretical_loss": 3.3748379792197074,
+      "tokens_seen": 2717450240
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.000178261916225325,
+      "loss": 1.3508,
+      "theoretical_loss": 3.3748253186745703,
+      "tokens_seen": 2717581312
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017822179425453378,
+      "loss": 1.3347,
+      "theoretical_loss": 3.374812658911018,
+      "tokens_seen": 2717712384
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017818167228374259,
+      "loss": 1.2438,
+      "theoretical_loss": 3.3747999999289657,
+      "tokens_seen": 2717843456
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017814155031295136,
+      "loss": 1.3931,
+      "theoretical_loss": 3.374787341728326,
+      "tokens_seen": 2717974528
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017810142834216017,
+      "loss": 1.3056,
+      "theoretical_loss": 3.3747746843090143,
+      "tokens_seen": 2718105600
+    },
+    {
+      "epoch": 0.65,
+      "objective/train/advantage_avg": 0.4934532642364502,
+      "objective/train/docs_used": 987833,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3577929735183716,
+      "objective/train/original_loss": 1.357792854309082,
+      "objective/train/theoretical_loss": 3.3747620276709442,
+      "objective/train/tokens_used": 1088695776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24385176599025726,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505850315093994,
+      "objective/train/weighted_lm_loss": 1.4265449047088623,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3747620276709442,
+      "tokens_seen": 2718236672
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017806130637136898,
+      "loss": 1.3331,
+      "theoretical_loss": 3.3747620276709442,
+      "tokens_seen": 2718236672
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017802118440057776,
+      "loss": 1.3582,
+      "theoretical_loss": 3.3747493718140293,
+      "tokens_seen": 2718367744
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017798106242978656,
+      "loss": 1.3515,
+      "theoretical_loss": 3.3747367167381848,
+      "tokens_seen": 2718498816
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017794094045899534,
+      "loss": 1.2387,
+      "theoretical_loss": 3.374724062443324,
+      "tokens_seen": 2718629888
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017790081848820415,
+      "loss": 1.3031,
+      "theoretical_loss": 3.374711408929361,
+      "tokens_seen": 2718760960
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017786069651741293,
+      "loss": 1.2537,
+      "theoretical_loss": 3.3746987561962105,
+      "tokens_seen": 2718892032
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017782057454662173,
+      "loss": 1.3942,
+      "theoretical_loss": 3.3746861042437866,
+      "tokens_seen": 2719023104
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017778045257583054,
+      "loss": 1.3264,
+      "theoretical_loss": 3.3746734530720026,
+      "tokens_seen": 2719154176
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017774033060503932,
+      "loss": 1.2932,
+      "theoretical_loss": 3.3746608026807743,
+      "tokens_seen": 2719285248
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017770020863424813,
+      "loss": 1.3308,
+      "theoretical_loss": 3.3746481530700145,
+      "tokens_seen": 2719416320
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001776600866634569,
+      "loss": 1.2955,
+      "theoretical_loss": 3.374635504239638,
+      "tokens_seen": 2719547392
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001776199646926657,
+      "loss": 1.3962,
+      "theoretical_loss": 3.3746228561895597,
+      "tokens_seen": 2719678464
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001775798427218745,
+      "loss": 1.3608,
+      "theoretical_loss": 3.3746102089196928,
+      "tokens_seen": 2719809536
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001775397207510833,
+      "loss": 1.3405,
+      "theoretical_loss": 3.3745975624299516,
+      "tokens_seen": 2719940608
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001774995987802921,
+      "loss": 1.3915,
+      "theoretical_loss": 3.3745849167202513,
+      "tokens_seen": 2720071680
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017745947680950088,
+      "loss": 1.4282,
+      "theoretical_loss": 3.3745722717905053,
+      "tokens_seen": 2720202752
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001774193548387097,
+      "loss": 1.3639,
+      "theoretical_loss": 3.3745596276406284,
+      "tokens_seen": 2720333824
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001773792328679185,
+      "loss": 1.2871,
+      "theoretical_loss": 3.374546984270535,
+      "tokens_seen": 2720464896
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017733911089712725,
+      "loss": 1.3853,
+      "theoretical_loss": 3.374534341680139,
+      "tokens_seen": 2720595968
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017729898892633605,
+      "loss": 1.3027,
+      "theoretical_loss": 3.374521699869355,
+      "tokens_seen": 2720727040
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017725886695554486,
+      "loss": 1.3362,
+      "theoretical_loss": 3.3745090588380977,
+      "tokens_seen": 2720858112
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017721874498475367,
+      "loss": 1.4022,
+      "theoretical_loss": 3.374496418586281,
+      "tokens_seen": 2720989184
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017717862301396244,
+      "loss": 1.3292,
+      "theoretical_loss": 3.3744837791138194,
+      "tokens_seen": 2721120256
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017713850104317125,
+      "loss": 1.2836,
+      "theoretical_loss": 3.3744711404206273,
+      "tokens_seen": 2721251328
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017709837907238006,
+      "loss": 1.4156,
+      "theoretical_loss": 3.374458502506619,
+      "tokens_seen": 2721382400
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001770582571015888,
+      "loss": 1.2836,
+      "theoretical_loss": 3.3744458653717095,
+      "tokens_seen": 2721513472
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017701813513079762,
+      "loss": 1.3697,
+      "theoretical_loss": 3.374433229015813,
+      "tokens_seen": 2721644544
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017697801316000642,
+      "loss": 1.378,
+      "theoretical_loss": 3.3744205934388436,
+      "tokens_seen": 2721775616
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017693789118921523,
+      "loss": 1.3498,
+      "theoretical_loss": 3.3744079586407163,
+      "tokens_seen": 2721906688
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.000176897769218424,
+      "loss": 1.3061,
+      "theoretical_loss": 3.374395324621345,
+      "tokens_seen": 2722037760
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017685764724763281,
+      "loss": 1.4022,
+      "theoretical_loss": 3.374382691380645,
+      "tokens_seen": 2722168832
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017681752527684162,
+      "loss": 1.3962,
+      "theoretical_loss": 3.37437005891853,
+      "tokens_seen": 2722299904
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001767774033060504,
+      "loss": 1.2812,
+      "theoretical_loss": 3.374357427234915,
+      "tokens_seen": 2722430976
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017673728133525918,
+      "loss": 1.3392,
+      "theoretical_loss": 3.374344796329714,
+      "tokens_seen": 2722562048
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017669715936446798,
+      "loss": 1.26,
+      "theoretical_loss": 3.3743321662028425,
+      "tokens_seen": 2722693120
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001766570373936768,
+      "loss": 1.4551,
+      "theoretical_loss": 3.3743195368542143,
+      "tokens_seen": 2722824192
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017661691542288557,
+      "loss": 1.2968,
+      "theoretical_loss": 3.374306908283744,
+      "tokens_seen": 2722955264
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017657679345209438,
+      "loss": 1.2717,
+      "theoretical_loss": 3.374294280491347,
+      "tokens_seen": 2723086336
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017653667148130318,
+      "loss": 1.3758,
+      "theoretical_loss": 3.3742816534769373,
+      "tokens_seen": 2723217408
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017649654951051196,
+      "loss": 1.2992,
+      "theoretical_loss": 3.3742690272404294,
+      "tokens_seen": 2723348480
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017645642753972074,
+      "loss": 1.2299,
+      "theoretical_loss": 3.3742564017817376,
+      "tokens_seen": 2723479552
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017641630556892955,
+      "loss": 1.2982,
+      "theoretical_loss": 3.374243777100778,
+      "tokens_seen": 2723610624
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017637618359813833,
+      "loss": 1.3656,
+      "theoretical_loss": 3.3742311531974636,
+      "tokens_seen": 2723741696
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017633606162734713,
+      "loss": 1.3027,
+      "theoretical_loss": 3.3742185300717105,
+      "tokens_seen": 2723872768
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017629593965655594,
+      "loss": 1.3278,
+      "theoretical_loss": 3.3742059077234323,
+      "tokens_seen": 2724003840
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017625581768576475,
+      "loss": 1.3632,
+      "theoretical_loss": 3.3741932861525443,
+      "tokens_seen": 2724134912
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017621569571497352,
+      "loss": 1.3864,
+      "theoretical_loss": 3.3741806653589608,
+      "tokens_seen": 2724265984
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001761755737441823,
+      "loss": 1.3929,
+      "theoretical_loss": 3.3741680453425973,
+      "tokens_seen": 2724397056
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001761354517733911,
+      "loss": 1.3118,
+      "theoretical_loss": 3.3741554261033677,
+      "tokens_seen": 2724528128
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001760953298025999,
+      "loss": 1.2936,
+      "theoretical_loss": 3.374142807641187,
+      "tokens_seen": 2724659200
+    },
+    {
+      "epoch": 0.65,
+      "objective/train/advantage_avg": 0.49091410636901855,
+      "objective/train/docs_used": 990115,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.303934097290039,
+      "objective/train/original_loss": 1.3039339780807495,
+      "objective/train/theoretical_loss": 3.37413018995597,
+      "objective/train/tokens_used": 1095249376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2416989505290985,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503201484680176,
+      "objective/train/weighted_lm_loss": 1.3694597482681274,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9048373699188232,
+      "theoretical_loss": 3.37413018995597,
+      "tokens_seen": 2724790272
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001760552078318087,
+      "loss": 1.3549,
+      "theoretical_loss": 3.37413018995597,
+      "tokens_seen": 2724790272
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001760150858610175,
+      "loss": 1.3121,
+      "theoretical_loss": 3.3741175730476316,
+      "tokens_seen": 2724921344
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001759749638902263,
+      "loss": 1.2672,
+      "theoretical_loss": 3.374104956916087,
+      "tokens_seen": 2725052416
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001759348419194351,
+      "loss": 1.3438,
+      "theoretical_loss": 3.37409234156125,
+      "tokens_seen": 2725183488
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001758947199486439,
+      "loss": 1.3271,
+      "theoretical_loss": 3.3740797269830365,
+      "tokens_seen": 2725314560
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017585459797785267,
+      "loss": 1.3399,
+      "theoretical_loss": 3.3740671131813604,
+      "tokens_seen": 2725445632
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017581447600706145,
+      "loss": 1.37,
+      "theoretical_loss": 3.374054500156137,
+      "tokens_seen": 2725576704
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017577435403627026,
+      "loss": 1.3051,
+      "theoretical_loss": 3.3740418879072815,
+      "tokens_seen": 2725707776
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017573423206547907,
+      "loss": 1.358,
+      "theoretical_loss": 3.3740292764347086,
+      "tokens_seen": 2725838848
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017569411009468787,
+      "loss": 1.2471,
+      "theoretical_loss": 3.3740166657383326,
+      "tokens_seen": 2725969920
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017565398812389665,
+      "loss": 1.3817,
+      "theoretical_loss": 3.374004055818069,
+      "tokens_seen": 2726100992
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017561386615310546,
+      "loss": 1.3101,
+      "theoretical_loss": 3.3739914466738328,
+      "tokens_seen": 2726232064
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017557374418231424,
+      "loss": 1.2372,
+      "theoretical_loss": 3.3739788383055385,
+      "tokens_seen": 2726363136
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017553362221152302,
+      "loss": 1.3647,
+      "theoretical_loss": 3.3739662307131013,
+      "tokens_seen": 2726494208
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017549350024073182,
+      "loss": 1.3129,
+      "theoretical_loss": 3.373953623896436,
+      "tokens_seen": 2726625280
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017545337826994063,
+      "loss": 1.395,
+      "theoretical_loss": 3.373941017855458,
+      "tokens_seen": 2726756352
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017541325629914943,
+      "loss": 1.294,
+      "theoretical_loss": 3.3739284125900815,
+      "tokens_seen": 2726887424
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001753731343283582,
+      "loss": 1.3164,
+      "theoretical_loss": 3.3739158081002225,
+      "tokens_seen": 2727018496
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017533301235756702,
+      "loss": 1.3307,
+      "theoretical_loss": 3.3739032043857953,
+      "tokens_seen": 2727149568
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001752928903867758,
+      "loss": 1.2308,
+      "theoretical_loss": 3.373890601446715,
+      "tokens_seen": 2727280640
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017525276841598458,
+      "loss": 1.3245,
+      "theoretical_loss": 3.373877999282897,
+      "tokens_seen": 2727411712
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017521264644519338,
+      "loss": 1.3131,
+      "theoretical_loss": 3.373865397894256,
+      "tokens_seen": 2727542784
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001751725244744022,
+      "loss": 1.3475,
+      "theoretical_loss": 3.3738527972807075,
+      "tokens_seen": 2727673856
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017513240250361097,
+      "loss": 1.3045,
+      "theoretical_loss": 3.373840197442166,
+      "tokens_seen": 2727804928
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017509228053281978,
+      "loss": 1.3307,
+      "theoretical_loss": 3.373827598378547,
+      "tokens_seen": 2727936000
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017505215856202858,
+      "loss": 1.3562,
+      "theoretical_loss": 3.3738150000897655,
+      "tokens_seen": 2728067072
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001750120365912374,
+      "loss": 1.2867,
+      "theoretical_loss": 3.3738024025757367,
+      "tokens_seen": 2728198144
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017497191462044614,
+      "loss": 1.3044,
+      "theoretical_loss": 3.3737898058363753,
+      "tokens_seen": 2728329216
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017493179264965495,
+      "loss": 1.3855,
+      "theoretical_loss": 3.373777209871597,
+      "tokens_seen": 2728460288
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017489167067886375,
+      "loss": 1.3023,
+      "theoretical_loss": 3.373764614681317,
+      "tokens_seen": 2728591360
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017485154870807253,
+      "loss": 1.3298,
+      "theoretical_loss": 3.3737520202654503,
+      "tokens_seen": 2728722432
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017481142673728134,
+      "loss": 1.4036,
+      "theoretical_loss": 3.373739426623912,
+      "tokens_seen": 2728853504
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017477130476649015,
+      "loss": 1.341,
+      "theoretical_loss": 3.373726833756617,
+      "tokens_seen": 2728984576
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017473118279569895,
+      "loss": 1.3954,
+      "theoretical_loss": 3.373714241663481,
+      "tokens_seen": 2729115648
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001746910608249077,
+      "loss": 1.3009,
+      "theoretical_loss": 3.3737016503444193,
+      "tokens_seen": 2729246720
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001746509388541165,
+      "loss": 1.3305,
+      "theoretical_loss": 3.373689059799347,
+      "tokens_seen": 2729377792
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017461081688332532,
+      "loss": 1.3195,
+      "theoretical_loss": 3.373676470028179,
+      "tokens_seen": 2729508864
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001745706949125341,
+      "loss": 1.3417,
+      "theoretical_loss": 3.373663881030831,
+      "tokens_seen": 2729639936
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001745305729417429,
+      "loss": 1.2981,
+      "theoretical_loss": 3.373651292807218,
+      "tokens_seen": 2729771008
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001744904509709517,
+      "loss": 1.335,
+      "theoretical_loss": 3.373638705357256,
+      "tokens_seen": 2729902080
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017445032900016051,
+      "loss": 1.3271,
+      "theoretical_loss": 3.373626118680859,
+      "tokens_seen": 2730033152
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001744102070293693,
+      "loss": 1.3518,
+      "theoretical_loss": 3.3736135327779433,
+      "tokens_seen": 2730164224
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017437008505857807,
+      "loss": 1.3389,
+      "theoretical_loss": 3.3736009476484243,
+      "tokens_seen": 2730295296
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017432996308778688,
+      "loss": 1.3097,
+      "theoretical_loss": 3.373588363292217,
+      "tokens_seen": 2730426368
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017428984111699566,
+      "loss": 1.3781,
+      "theoretical_loss": 3.373575779709237,
+      "tokens_seen": 2730557440
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00017424971914620446,
+      "loss": 1.2812,
+      "theoretical_loss": 3.373563196899399,
+      "tokens_seen": 2730688512
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017420959717541327,
+      "loss": 1.316,
+      "theoretical_loss": 3.3735506148626193,
+      "tokens_seen": 2730819584
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017416947520462205,
+      "loss": 1.3125,
+      "theoretical_loss": 3.373538033598813,
+      "tokens_seen": 2730950656
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017412935323383086,
+      "loss": 1.3074,
+      "theoretical_loss": 3.373525453107895,
+      "tokens_seen": 2731081728
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017408923126303964,
+      "loss": 1.315,
+      "theoretical_loss": 3.3735128733897817,
+      "tokens_seen": 2731212800
+    },
+    {
+      "epoch": 0.66,
+      "objective/train/advantage_avg": 0.49284446239471436,
+      "objective/train/docs_used": 992345,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3238368034362793,
+      "objective/train/original_loss": 1.3238368034362793,
+      "objective/train/theoretical_loss": 3.3735002944443875,
+      "objective/train/tokens_used": 1101802976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2435806691646576,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050522804260254,
+      "objective/train/weighted_lm_loss": 1.390689730644226,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3735002944443875,
+      "tokens_seen": 2731343872
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017404910929224844,
+      "loss": 1.3618,
+      "theoretical_loss": 3.3735002944443875,
+      "tokens_seen": 2731343872
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017400898732145722,
+      "loss": 1.3411,
+      "theoretical_loss": 3.3734877162716286,
+      "tokens_seen": 2731474944
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017396886535066603,
+      "loss": 1.3065,
+      "theoretical_loss": 3.37347513887142,
+      "tokens_seen": 2731606016
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017392874337987483,
+      "loss": 1.3756,
+      "theoretical_loss": 3.3734625622436774,
+      "tokens_seen": 2731737088
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001738886214090836,
+      "loss": 1.2685,
+      "theoretical_loss": 3.3734499863883167,
+      "tokens_seen": 2731868160
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017384849943829242,
+      "loss": 1.3115,
+      "theoretical_loss": 3.373437411305253,
+      "tokens_seen": 2731999232
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001738083774675012,
+      "loss": 1.3754,
+      "theoretical_loss": 3.3734248369944013,
+      "tokens_seen": 2732130304
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017376825549671,
+      "loss": 1.3302,
+      "theoretical_loss": 3.3734122634556782,
+      "tokens_seen": 2732261376
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017372813352591878,
+      "loss": 1.3373,
+      "theoretical_loss": 3.3733996906889985,
+      "tokens_seen": 2732392448
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001736880115551276,
+      "loss": 1.3185,
+      "theoretical_loss": 3.373387118694278,
+      "tokens_seen": 2732523520
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001736478895843364,
+      "loss": 1.3686,
+      "theoretical_loss": 3.3733745474714323,
+      "tokens_seen": 2732654592
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017360776761354518,
+      "loss": 1.3591,
+      "theoretical_loss": 3.373361977020377,
+      "tokens_seen": 2732785664
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017356764564275398,
+      "loss": 1.316,
+      "theoretical_loss": 3.3733494073410277,
+      "tokens_seen": 2732916736
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001735275236719628,
+      "loss": 1.3863,
+      "theoretical_loss": 3.3733368384333,
+      "tokens_seen": 2733047808
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017348740170117157,
+      "loss": 1.2988,
+      "theoretical_loss": 3.3733242702971094,
+      "tokens_seen": 2733178880
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017344727973038035,
+      "loss": 1.3493,
+      "theoretical_loss": 3.3733117029323716,
+      "tokens_seen": 2733309952
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017340715775958915,
+      "loss": 1.2031,
+      "theoretical_loss": 3.3732991363390026,
+      "tokens_seen": 2733441024
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017336703578879796,
+      "loss": 1.3837,
+      "theoretical_loss": 3.3732865705169175,
+      "tokens_seen": 2733572096
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017332691381800674,
+      "loss": 1.3586,
+      "theoretical_loss": 3.3732740054660324,
+      "tokens_seen": 2733703168
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017328679184721555,
+      "loss": 1.3387,
+      "theoretical_loss": 3.3732614411862625,
+      "tokens_seen": 2733834240
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017324666987642435,
+      "loss": 1.3357,
+      "theoretical_loss": 3.3732488776775242,
+      "tokens_seen": 2733965312
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001732065479056331,
+      "loss": 1.3906,
+      "theoretical_loss": 3.373236314939733,
+      "tokens_seen": 2734096384
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001731664259348419,
+      "loss": 1.2556,
+      "theoretical_loss": 3.3732237529728044,
+      "tokens_seen": 2734227456
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017312630396405072,
+      "loss": 1.3323,
+      "theoretical_loss": 3.373211191776654,
+      "tokens_seen": 2734358528
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017308618199325952,
+      "loss": 1.3587,
+      "theoretical_loss": 3.373198631351198,
+      "tokens_seen": 2734489600
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001730460600224683,
+      "loss": 1.3637,
+      "theoretical_loss": 3.373186071696352,
+      "tokens_seen": 2734620672
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001730059380516771,
+      "loss": 1.2952,
+      "theoretical_loss": 3.373173512812032,
+      "tokens_seen": 2734751744
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017296581608088591,
+      "loss": 1.2421,
+      "theoretical_loss": 3.3731609546981534,
+      "tokens_seen": 2734882816
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017292569411009467,
+      "loss": 1.3322,
+      "theoretical_loss": 3.3731483973546323,
+      "tokens_seen": 2735013888
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017288557213930347,
+      "loss": 1.2454,
+      "theoretical_loss": 3.3731358407813845,
+      "tokens_seen": 2735144960
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017284545016851228,
+      "loss": 1.2845,
+      "theoretical_loss": 3.373123284978326,
+      "tokens_seen": 2735276032
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017280532819772109,
+      "loss": 1.3362,
+      "theoretical_loss": 3.373110729945372,
+      "tokens_seen": 2735407104
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017276520622692986,
+      "loss": 1.3125,
+      "theoretical_loss": 3.373098175682439,
+      "tokens_seen": 2735538176
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017272508425613867,
+      "loss": 1.2993,
+      "theoretical_loss": 3.3730856221894427,
+      "tokens_seen": 2735669248
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017268496228534748,
+      "loss": 1.2663,
+      "theoretical_loss": 3.373073069466299,
+      "tokens_seen": 2735800320
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017264484031455626,
+      "loss": 1.3826,
+      "theoretical_loss": 3.373060517512924,
+      "tokens_seen": 2735931392
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017260471834376504,
+      "loss": 1.2399,
+      "theoretical_loss": 3.373047966329233,
+      "tokens_seen": 2736062464
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017256459637297384,
+      "loss": 1.3604,
+      "theoretical_loss": 3.3730354159151426,
+      "tokens_seen": 2736193536
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017252447440218265,
+      "loss": 1.2869,
+      "theoretical_loss": 3.3730228662705684,
+      "tokens_seen": 2736324608
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017248435243139143,
+      "loss": 1.2684,
+      "theoretical_loss": 3.3730103173954267,
+      "tokens_seen": 2736455680
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017244423046060023,
+      "loss": 1.3057,
+      "theoretical_loss": 3.3729977692896327,
+      "tokens_seen": 2736586752
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017240410848980904,
+      "loss": 1.2541,
+      "theoretical_loss": 3.3729852219531034,
+      "tokens_seen": 2736717824
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017236398651901782,
+      "loss": 1.2997,
+      "theoretical_loss": 3.3729726753857543,
+      "tokens_seen": 2736848896
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001723238645482266,
+      "loss": 1.4019,
+      "theoretical_loss": 3.3729601295875016,
+      "tokens_seen": 2736979968
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001722837425774354,
+      "loss": 1.3442,
+      "theoretical_loss": 3.372947584558261,
+      "tokens_seen": 2737111040
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017224362060664418,
+      "loss": 1.366,
+      "theoretical_loss": 3.3729350402979486,
+      "tokens_seen": 2737242112
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.000172203498635853,
+      "loss": 1.3262,
+      "theoretical_loss": 3.3729224968064804,
+      "tokens_seen": 2737373184
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001721633766650618,
+      "loss": 1.4588,
+      "theoretical_loss": 3.372909954083773,
+      "tokens_seen": 2737504256
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001721232546942706,
+      "loss": 1.3134,
+      "theoretical_loss": 3.3728974121297415,
+      "tokens_seen": 2737635328
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017208313272347938,
+      "loss": 1.2683,
+      "theoretical_loss": 3.3728848709443033,
+      "tokens_seen": 2737766400
+    },
+    {
+      "epoch": 0.66,
+      "objective/train/advantage_avg": 0.4949920177459717,
+      "objective/train/docs_used": 994721,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.1496843099594116,
+      "objective/train/original_loss": 1.149684190750122,
+      "objective/train/theoretical_loss": 3.3728723305273736,
+      "objective/train/tokens_used": 1108356576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2452174872159958,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050745964050293,
+      "objective/train/weighted_lm_loss": 1.2079073190689087,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3728723305273736,
+      "tokens_seen": 2737897472
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001720430107526882,
+      "loss": 1.2569,
+      "theoretical_loss": 3.3728723305273736,
+      "tokens_seen": 2737897472
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017200288878189697,
+      "loss": 1.3243,
+      "theoretical_loss": 3.3728597908788682,
+      "tokens_seen": 2738028544
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017196276681110575,
+      "loss": 1.3055,
+      "theoretical_loss": 3.3728472519987043,
+      "tokens_seen": 2738159616
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017192264484031455,
+      "loss": 1.3214,
+      "theoretical_loss": 3.3728347138867973,
+      "tokens_seen": 2738290688
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017188252286952336,
+      "loss": 1.2422,
+      "theoretical_loss": 3.372822176543064,
+      "tokens_seen": 2738421760
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017184240089873217,
+      "loss": 1.2882,
+      "theoretical_loss": 3.3728096399674197,
+      "tokens_seen": 2738552832
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017180227892794094,
+      "loss": 1.2331,
+      "theoretical_loss": 3.3727971041597806,
+      "tokens_seen": 2738683904
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017176215695714975,
+      "loss": 1.371,
+      "theoretical_loss": 3.372784569120064,
+      "tokens_seen": 2738814976
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017172203498635853,
+      "loss": 1.3015,
+      "theoretical_loss": 3.372772034848185,
+      "tokens_seen": 2738946048
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001716819130155673,
+      "loss": 1.2584,
+      "theoretical_loss": 3.3727595013440608,
+      "tokens_seen": 2739077120
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017164179104477612,
+      "loss": 1.3941,
+      "theoretical_loss": 3.372746968607607,
+      "tokens_seen": 2739208192
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017160166907398492,
+      "loss": 1.3796,
+      "theoretical_loss": 3.3727344366387393,
+      "tokens_seen": 2739339264
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017156154710319373,
+      "loss": 1.3414,
+      "theoretical_loss": 3.372721905437375,
+      "tokens_seen": 2739470336
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001715214251324025,
+      "loss": 1.2582,
+      "theoretical_loss": 3.37270937500343,
+      "tokens_seen": 2739601408
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017148130316161131,
+      "loss": 1.353,
+      "theoretical_loss": 3.3726968453368205,
+      "tokens_seen": 2739732480
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001714411811908201,
+      "loss": 1.3164,
+      "theoretical_loss": 3.372684316437463,
+      "tokens_seen": 2739863552
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017140105922002887,
+      "loss": 1.3528,
+      "theoretical_loss": 3.3726717883052735,
+      "tokens_seen": 2739994624
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017136093724923768,
+      "loss": 1.3605,
+      "theoretical_loss": 3.3726592609401687,
+      "tokens_seen": 2740125696
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017132081527844648,
+      "loss": 1.4096,
+      "theoretical_loss": 3.3726467343420645,
+      "tokens_seen": 2740256768
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017128069330765526,
+      "loss": 1.2759,
+      "theoretical_loss": 3.3726342085108776,
+      "tokens_seen": 2740387840
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017124057133686407,
+      "loss": 1.2754,
+      "theoretical_loss": 3.372621683446525,
+      "tokens_seen": 2740518912
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017120044936607288,
+      "loss": 1.3106,
+      "theoretical_loss": 3.3726091591489213,
+      "tokens_seen": 2740649984
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017116032739528168,
+      "loss": 1.3327,
+      "theoretical_loss": 3.3725966356179846,
+      "tokens_seen": 2740781056
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017112020542449044,
+      "loss": 1.228,
+      "theoretical_loss": 3.3725841128536302,
+      "tokens_seen": 2740912128
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017108008345369924,
+      "loss": 1.2355,
+      "theoretical_loss": 3.3725715908557756,
+      "tokens_seen": 2741043200
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017103996148290805,
+      "loss": 1.394,
+      "theoretical_loss": 3.372559069624336,
+      "tokens_seen": 2741174272
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017099983951211683,
+      "loss": 1.3162,
+      "theoretical_loss": 3.372546549159229,
+      "tokens_seen": 2741305344
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017095971754132563,
+      "loss": 1.3256,
+      "theoretical_loss": 3.37253402946037,
+      "tokens_seen": 2741436416
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017091959557053444,
+      "loss": 1.3155,
+      "theoretical_loss": 3.372521510527676,
+      "tokens_seen": 2741567488
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017087947359974325,
+      "loss": 1.3714,
+      "theoretical_loss": 3.3725089923610643,
+      "tokens_seen": 2741698560
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.000170839351628952,
+      "loss": 1.2811,
+      "theoretical_loss": 3.37249647496045,
+      "tokens_seen": 2741829632
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001707992296581608,
+      "loss": 1.3088,
+      "theoretical_loss": 3.37248395832575,
+      "tokens_seen": 2741960704
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001707591076873696,
+      "loss": 1.2517,
+      "theoretical_loss": 3.3724714424568814,
+      "tokens_seen": 2742091776
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001707189857165784,
+      "loss": 1.2944,
+      "theoretical_loss": 3.3724589273537604,
+      "tokens_seen": 2742222848
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001706788637457872,
+      "loss": 1.3009,
+      "theoretical_loss": 3.3724464130163034,
+      "tokens_seen": 2742353920
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.000170638741774996,
+      "loss": 1.3043,
+      "theoretical_loss": 3.372433899444427,
+      "tokens_seen": 2742484992
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001705986198042048,
+      "loss": 1.2762,
+      "theoretical_loss": 3.372421386638048,
+      "tokens_seen": 2742616064
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017055849783341356,
+      "loss": 1.2956,
+      "theoretical_loss": 3.372408874597083,
+      "tokens_seen": 2742747136
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017051837586262237,
+      "loss": 1.3136,
+      "theoretical_loss": 3.372396363321448,
+      "tokens_seen": 2742878208
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017047825389183117,
+      "loss": 1.3216,
+      "theoretical_loss": 3.3723838528110606,
+      "tokens_seen": 2743009280
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017043813192103995,
+      "loss": 1.2252,
+      "theoretical_loss": 3.3723713430658364,
+      "tokens_seen": 2743140352
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017039800995024876,
+      "loss": 1.3222,
+      "theoretical_loss": 3.372358834085693,
+      "tokens_seen": 2743271424
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017035788797945757,
+      "loss": 1.318,
+      "theoretical_loss": 3.3723463258705464,
+      "tokens_seen": 2743402496
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017031776600866637,
+      "loss": 1.3294,
+      "theoretical_loss": 3.3723338184203135,
+      "tokens_seen": 2743533568
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017027764403787515,
+      "loss": 1.1954,
+      "theoretical_loss": 3.372321311734911,
+      "tokens_seen": 2743664640
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017023752206708393,
+      "loss": 1.319,
+      "theoretical_loss": 3.3723088058142556,
+      "tokens_seen": 2743795712
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017019740009629274,
+      "loss": 1.2945,
+      "theoretical_loss": 3.372296300658264,
+      "tokens_seen": 2743926784
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017015727812550152,
+      "loss": 1.2331,
+      "theoretical_loss": 3.372283796266853,
+      "tokens_seen": 2744057856
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017011715615471032,
+      "loss": 1.298,
+      "theoretical_loss": 3.372271292639939,
+      "tokens_seen": 2744188928
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00017007703418391913,
+      "loss": 1.265,
+      "theoretical_loss": 3.372258789777439,
+      "tokens_seen": 2744320000
+    },
+    {
+      "epoch": 0.66,
+      "objective/train/advantage_avg": 0.4928019344806671,
+      "objective/train/docs_used": 997363,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3268628120422363,
+      "objective/train/original_loss": 1.3268628120422363,
+      "objective/train/theoretical_loss": 3.37224628767927,
+      "objective/train/tokens_used": 1114910176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2432902753353119,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505170822143555,
+      "objective/train/weighted_lm_loss": 1.393796682357788,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.37224628767927,
+      "tokens_seen": 2744451072
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001700369122131279,
+      "loss": 1.3762,
+      "theoretical_loss": 3.37224628767927,
+      "tokens_seen": 2744451072
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001699967902423367,
+      "loss": 1.2785,
+      "theoretical_loss": 3.372233786345348,
+      "tokens_seen": 2744582144
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001699566682715455,
+      "loss": 1.3084,
+      "theoretical_loss": 3.3722212857755904,
+      "tokens_seen": 2744713216
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001699165463007543,
+      "loss": 1.2922,
+      "theoretical_loss": 3.3722087859699146,
+      "tokens_seen": 2744844288
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00016987642432996308,
+      "loss": 1.2659,
+      "theoretical_loss": 3.3721962869282356,
+      "tokens_seen": 2744975360
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00016983630235917188,
+      "loss": 1.3237,
+      "theoretical_loss": 3.372183788650472,
+      "tokens_seen": 2745106432
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001697961803883807,
+      "loss": 1.3608,
+      "theoretical_loss": 3.37217129113654,
+      "tokens_seen": 2745237504
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00016975605841758947,
+      "loss": 1.3509,
+      "theoretical_loss": 3.3721587943863565,
+      "tokens_seen": 2745368576
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00016971593644679828,
+      "loss": 1.346,
+      "theoretical_loss": 3.372146298399838,
+      "tokens_seen": 2745499648
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00016967581447600708,
+      "loss": 1.2994,
+      "theoretical_loss": 3.3721338031769017,
+      "tokens_seen": 2745630720
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00016963569250521586,
+      "loss": 1.3676,
+      "theoretical_loss": 3.372121308717465,
+      "tokens_seen": 2745761792
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00016959557053442464,
+      "loss": 1.2421,
+      "theoretical_loss": 3.372108815021444,
+      "tokens_seen": 2745892864
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00016955544856363345,
+      "loss": 1.3151,
+      "theoretical_loss": 3.3720963220887556,
+      "tokens_seen": 2746023936
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00016951532659284225,
+      "loss": 1.2278,
+      "theoretical_loss": 3.372083829919317,
+      "tokens_seen": 2746155008
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00016947520462205103,
+      "loss": 1.3374,
+      "theoretical_loss": 3.372071338513045,
+      "tokens_seen": 2746286080
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00016943508265125984,
+      "loss": 1.3377,
+      "theoretical_loss": 3.3720588478698574,
+      "tokens_seen": 2746417152
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00016939496068046865,
+      "loss": 1.2871,
+      "theoretical_loss": 3.37204635798967,
+      "tokens_seen": 2746548224
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00016935483870967742,
+      "loss": 1.333,
+      "theoretical_loss": 3.3720338688724008,
+      "tokens_seen": 2746679296
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001693147167388862,
+      "loss": 1.2709,
+      "theoretical_loss": 3.3720213805179657,
+      "tokens_seen": 2746810368
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.000169274594768095,
+      "loss": 1.3422,
+      "theoretical_loss": 3.3720088929262824,
+      "tokens_seen": 2746941440
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00016923447279730382,
+      "loss": 1.2747,
+      "theoretical_loss": 3.371996406097268,
+      "tokens_seen": 2747072512
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001691943508265126,
+      "loss": 1.3602,
+      "theoretical_loss": 3.3719839200308392,
+      "tokens_seen": 2747203584
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001691542288557214,
+      "loss": 1.2808,
+      "theoretical_loss": 3.3719714347269134,
+      "tokens_seen": 2747334656
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001691141068849302,
+      "loss": 1.2331,
+      "theoretical_loss": 3.371958950185407,
+      "tokens_seen": 2747465728
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016907398491413896,
+      "loss": 1.2966,
+      "theoretical_loss": 3.3719464664062375,
+      "tokens_seen": 2747596800
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016903386294334777,
+      "loss": 1.295,
+      "theoretical_loss": 3.3719339833893223,
+      "tokens_seen": 2747727872
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016899374097255657,
+      "loss": 1.2446,
+      "theoretical_loss": 3.3719215011345782,
+      "tokens_seen": 2747858944
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016895361900176538,
+      "loss": 1.3504,
+      "theoretical_loss": 3.3719090196419224,
+      "tokens_seen": 2747990016
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016891349703097416,
+      "loss": 1.3747,
+      "theoretical_loss": 3.3718965389112716,
+      "tokens_seen": 2748121088
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016887337506018296,
+      "loss": 1.3137,
+      "theoretical_loss": 3.371884058942544,
+      "tokens_seen": 2748252160
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016883325308939177,
+      "loss": 1.3225,
+      "theoretical_loss": 3.371871579735655,
+      "tokens_seen": 2748383232
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016879313111860055,
+      "loss": 1.2751,
+      "theoretical_loss": 3.3718591012905237,
+      "tokens_seen": 2748514304
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016875300914780933,
+      "loss": 1.3045,
+      "theoretical_loss": 3.3718466236070657,
+      "tokens_seen": 2748645376
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016871288717701814,
+      "loss": 1.2866,
+      "theoretical_loss": 3.3718341466851993,
+      "tokens_seen": 2748776448
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016867276520622694,
+      "loss": 1.2823,
+      "theoretical_loss": 3.3718216705248407,
+      "tokens_seen": 2748907520
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016863264323543572,
+      "loss": 1.3083,
+      "theoretical_loss": 3.371809195125908,
+      "tokens_seen": 2749038592
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016859252126464453,
+      "loss": 1.2974,
+      "theoretical_loss": 3.3717967204883186,
+      "tokens_seen": 2749169664
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016855239929385333,
+      "loss": 1.3483,
+      "theoretical_loss": 3.3717842466119885,
+      "tokens_seen": 2749300736
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001685122773230621,
+      "loss": 1.2841,
+      "theoretical_loss": 3.371771773496836,
+      "tokens_seen": 2749431808
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001684721553522709,
+      "loss": 1.3538,
+      "theoretical_loss": 3.371759301142778,
+      "tokens_seen": 2749562880
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001684320333814797,
+      "loss": 1.3099,
+      "theoretical_loss": 3.3717468295497324,
+      "tokens_seen": 2749693952
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001683919114106885,
+      "loss": 1.3306,
+      "theoretical_loss": 3.371734358717615,
+      "tokens_seen": 2749825024
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016835178943989728,
+      "loss": 1.3411,
+      "theoretical_loss": 3.371721888646345,
+      "tokens_seen": 2749956096
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001683116674691061,
+      "loss": 1.3241,
+      "theoretical_loss": 3.371709419335838,
+      "tokens_seen": 2750087168
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001682715454983149,
+      "loss": 1.3827,
+      "theoretical_loss": 3.3716969507860126,
+      "tokens_seen": 2750218240
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016823142352752368,
+      "loss": 1.2803,
+      "theoretical_loss": 3.371684482996786,
+      "tokens_seen": 2750349312
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016819130155673246,
+      "loss": 1.3855,
+      "theoretical_loss": 3.371672015968074,
+      "tokens_seen": 2750480384
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016815117958594126,
+      "loss": 1.4145,
+      "theoretical_loss": 3.371659549699796,
+      "tokens_seen": 2750611456
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016811105761515004,
+      "loss": 1.4455,
+      "theoretical_loss": 3.3716470841918684,
+      "tokens_seen": 2750742528
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016807093564435885,
+      "loss": 1.2352,
+      "theoretical_loss": 3.371634619444209,
+      "tokens_seen": 2750873600
+    },
+    {
+      "epoch": 0.67,
+      "objective/train/advantage_avg": 0.49175482988357544,
+      "objective/train/docs_used": 999260,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2141786813735962,
+      "objective/train/original_loss": 1.2141788005828857,
+      "objective/train/theoretical_loss": 3.3716221554567345,
+      "objective/train/tokens_used": 1121463776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24237994849681854,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504076480865479,
+      "objective/train/weighted_lm_loss": 1.2753417491912842,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3716221554567345,
+      "tokens_seen": 2751004672
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016803081367356765,
+      "loss": 1.3257,
+      "theoretical_loss": 3.3716221554567345,
+      "tokens_seen": 2751004672
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016799069170277646,
+      "loss": 1.2505,
+      "theoretical_loss": 3.3716096922293635,
+      "tokens_seen": 2751135744
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016795056973198524,
+      "loss": 1.2563,
+      "theoretical_loss": 3.371597229762012,
+      "tokens_seen": 2751266816
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016791044776119405,
+      "loss": 1.303,
+      "theoretical_loss": 3.3715847680545985,
+      "tokens_seen": 2751397888
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016787032579040282,
+      "loss": 1.2904,
+      "theoretical_loss": 3.3715723071070403,
+      "tokens_seen": 2751528960
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001678302038196116,
+      "loss": 1.2918,
+      "theoretical_loss": 3.3715598469192543,
+      "tokens_seen": 2751660032
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001677900818488204,
+      "loss": 1.2371,
+      "theoretical_loss": 3.371547387491159,
+      "tokens_seen": 2751791104
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016774995987802922,
+      "loss": 1.2848,
+      "theoretical_loss": 3.371534928822671,
+      "tokens_seen": 2751922176
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016770983790723802,
+      "loss": 1.3335,
+      "theoretical_loss": 3.3715224709137077,
+      "tokens_seen": 2752053248
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001676697159364468,
+      "loss": 1.3639,
+      "theoretical_loss": 3.3715100137641874,
+      "tokens_seen": 2752184320
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001676295939656556,
+      "loss": 1.3092,
+      "theoretical_loss": 3.371497557374027,
+      "tokens_seen": 2752315392
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001675894719948644,
+      "loss": 1.2788,
+      "theoretical_loss": 3.371485101743145,
+      "tokens_seen": 2752446464
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016754935002407317,
+      "loss": 1.2885,
+      "theoretical_loss": 3.3714726468714575,
+      "tokens_seen": 2752577536
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016750922805328197,
+      "loss": 1.4468,
+      "theoretical_loss": 3.3714601927588834,
+      "tokens_seen": 2752708608
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016746910608249078,
+      "loss": 1.3902,
+      "theoretical_loss": 3.37144773940534,
+      "tokens_seen": 2752839680
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016742898411169959,
+      "loss": 1.2703,
+      "theoretical_loss": 3.371435286810744,
+      "tokens_seen": 2752970752
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016738886214090836,
+      "loss": 1.3449,
+      "theoretical_loss": 3.371422834975014,
+      "tokens_seen": 2753101824
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016734874017011717,
+      "loss": 1.291,
+      "theoretical_loss": 3.371410383898067,
+      "tokens_seen": 2753232896
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016730861819932598,
+      "loss": 1.3379,
+      "theoretical_loss": 3.3713979335798214,
+      "tokens_seen": 2753363968
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016726849622853473,
+      "loss": 1.2483,
+      "theoretical_loss": 3.3713854840201947,
+      "tokens_seen": 2753495040
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016722837425774354,
+      "loss": 1.4242,
+      "theoretical_loss": 3.371373035219104,
+      "tokens_seen": 2753626112
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016718825228695234,
+      "loss": 1.3623,
+      "theoretical_loss": 3.3713605871764667,
+      "tokens_seen": 2753757184
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016714813031616112,
+      "loss": 1.3335,
+      "theoretical_loss": 3.3713481398922016,
+      "tokens_seen": 2753888256
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016710800834536993,
+      "loss": 1.2693,
+      "theoretical_loss": 3.371335693366226,
+      "tokens_seen": 2754019328
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016706788637457873,
+      "loss": 1.2932,
+      "theoretical_loss": 3.3713232475984567,
+      "tokens_seen": 2754150400
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016702776440378754,
+      "loss": 1.3527,
+      "theoretical_loss": 3.371310802588813,
+      "tokens_seen": 2754281472
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001669876424329963,
+      "loss": 1.2937,
+      "theoretical_loss": 3.3712983583372114,
+      "tokens_seen": 2754412544
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001669475204622051,
+      "loss": 1.29,
+      "theoretical_loss": 3.3712859148435705,
+      "tokens_seen": 2754543616
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001669073984914139,
+      "loss": 1.24,
+      "theoretical_loss": 3.3712734721078075,
+      "tokens_seen": 2754674688
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016686727652062268,
+      "loss": 1.3499,
+      "theoretical_loss": 3.37126103012984,
+      "tokens_seen": 2754805760
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001668271545498315,
+      "loss": 1.3488,
+      "theoretical_loss": 3.3712485889095865,
+      "tokens_seen": 2754936832
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001667870325790403,
+      "loss": 1.3101,
+      "theoretical_loss": 3.3712361484469646,
+      "tokens_seen": 2755067904
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001667469106082491,
+      "loss": 1.3115,
+      "theoretical_loss": 3.371223708741892,
+      "tokens_seen": 2755198976
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016670678863745785,
+      "loss": 1.3755,
+      "theoretical_loss": 3.371211269794286,
+      "tokens_seen": 2755330048
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016666666666666666,
+      "loss": 1.4054,
+      "theoretical_loss": 3.3711988316040653,
+      "tokens_seen": 2755461120
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016662654469587547,
+      "loss": 1.3776,
+      "theoretical_loss": 3.371186394171148,
+      "tokens_seen": 2755592192
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016658642272508425,
+      "loss": 1.3454,
+      "theoretical_loss": 3.3711739574954507,
+      "tokens_seen": 2755723264
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016654630075429305,
+      "loss": 1.2999,
+      "theoretical_loss": 3.371161521576892,
+      "tokens_seen": 2755854336
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016650617878350186,
+      "loss": 1.3129,
+      "theoretical_loss": 3.37114908641539,
+      "tokens_seen": 2755985408
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016646605681271067,
+      "loss": 1.275,
+      "theoretical_loss": 3.3711366520108625,
+      "tokens_seen": 2756116480
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016642593484191944,
+      "loss": 1.3676,
+      "theoretical_loss": 3.371124218363227,
+      "tokens_seen": 2756247552
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016638581287112822,
+      "loss": 1.2786,
+      "theoretical_loss": 3.3711117854724018,
+      "tokens_seen": 2756378624
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016634569090033703,
+      "loss": 1.2982,
+      "theoretical_loss": 3.3710993533383053,
+      "tokens_seen": 2756509696
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001663055689295458,
+      "loss": 1.3361,
+      "theoretical_loss": 3.3710869219608544,
+      "tokens_seen": 2756640768
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016626544695875462,
+      "loss": 1.3392,
+      "theoretical_loss": 3.371074491339968,
+      "tokens_seen": 2756771840
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016622532498796342,
+      "loss": 1.3112,
+      "theoretical_loss": 3.3710620614755635,
+      "tokens_seen": 2756902912
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016618520301717223,
+      "loss": 1.3148,
+      "theoretical_loss": 3.3710496323675594,
+      "tokens_seen": 2757033984
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.000166145081046381,
+      "loss": 1.3039,
+      "theoretical_loss": 3.3710372040158734,
+      "tokens_seen": 2757165056
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001661049590755898,
+      "loss": 1.3666,
+      "theoretical_loss": 3.3710247764204233,
+      "tokens_seen": 2757296128
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001660648371047986,
+      "loss": 1.3955,
+      "theoretical_loss": 3.371012349581128,
+      "tokens_seen": 2757427200
+    },
+    {
+      "epoch": 0.67,
+      "objective/train/advantage_avg": 0.49231553077697754,
+      "objective/train/docs_used": 1001599,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.402485728263855,
+      "objective/train/original_loss": 1.4024856090545654,
+      "objective/train/theoretical_loss": 3.3709999234979042,
+      "objective/train/tokens_used": 1128017376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2428082376718521,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504659414291382,
+      "objective/train/weighted_lm_loss": 1.4730373620986938,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3709999234979042,
+      "tokens_seen": 2757558272
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016602471513400737,
+      "loss": 1.3935,
+      "theoretical_loss": 3.3709999234979042,
+      "tokens_seen": 2757558272
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016598459316321618,
+      "loss": 1.3327,
+      "theoretical_loss": 3.3709874981706713,
+      "tokens_seen": 2757689344
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016594447119242498,
+      "loss": 1.3229,
+      "theoretical_loss": 3.3709750735993467,
+      "tokens_seen": 2757820416
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016590434922163376,
+      "loss": 1.3283,
+      "theoretical_loss": 3.3709626497838485,
+      "tokens_seen": 2757951488
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016586422725084257,
+      "loss": 1.255,
+      "theoretical_loss": 3.370950226724095,
+      "tokens_seen": 2758082560
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016582410528005135,
+      "loss": 1.2976,
+      "theoretical_loss": 3.370937804420004,
+      "tokens_seen": 2758213632
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016578398330926016,
+      "loss": 1.2808,
+      "theoretical_loss": 3.3709253828714942,
+      "tokens_seen": 2758344704
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016574386133846894,
+      "loss": 1.3021,
+      "theoretical_loss": 3.3709129620784832,
+      "tokens_seen": 2758475776
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016570373936767774,
+      "loss": 1.3324,
+      "theoretical_loss": 3.3709005420408893,
+      "tokens_seen": 2758606848
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016566361739688655,
+      "loss": 1.3803,
+      "theoretical_loss": 3.370888122758631,
+      "tokens_seen": 2758737920
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016562349542609533,
+      "loss": 1.3686,
+      "theoretical_loss": 3.3708757042316257,
+      "tokens_seen": 2758868992
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016558337345530413,
+      "loss": 1.4018,
+      "theoretical_loss": 3.3708632864597927,
+      "tokens_seen": 2759000064
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016554325148451294,
+      "loss": 1.3548,
+      "theoretical_loss": 3.370850869443049,
+      "tokens_seen": 2759131136
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016550312951372172,
+      "loss": 1.2663,
+      "theoretical_loss": 3.370838453181314,
+      "tokens_seen": 2759262208
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001654630075429305,
+      "loss": 1.2895,
+      "theoretical_loss": 3.370826037674505,
+      "tokens_seen": 2759393280
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001654228855721393,
+      "loss": 1.3421,
+      "theoretical_loss": 3.3708136229225403,
+      "tokens_seen": 2759524352
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001653827636013481,
+      "loss": 1.2313,
+      "theoretical_loss": 3.370801208925339,
+      "tokens_seen": 2759655424
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001653426416305569,
+      "loss": 1.3098,
+      "theoretical_loss": 3.3707887956828184,
+      "tokens_seen": 2759786496
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001653025196597657,
+      "loss": 1.3264,
+      "theoretical_loss": 3.3707763831948974,
+      "tokens_seen": 2759917568
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001652623976889745,
+      "loss": 1.2998,
+      "theoretical_loss": 3.370763971461494,
+      "tokens_seen": 2760048640
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016522227571818328,
+      "loss": 1.2818,
+      "theoretical_loss": 3.370751560482527,
+      "tokens_seen": 2760179712
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016518215374739206,
+      "loss": 1.2937,
+      "theoretical_loss": 3.3707391502579136,
+      "tokens_seen": 2760310784
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016514203177660087,
+      "loss": 1.3356,
+      "theoretical_loss": 3.3707267407875734,
+      "tokens_seen": 2760441856
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016510190980580967,
+      "loss": 1.2855,
+      "theoretical_loss": 3.370714332071424,
+      "tokens_seen": 2760572928
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016506178783501845,
+      "loss": 1.2803,
+      "theoretical_loss": 3.370701924109384,
+      "tokens_seen": 2760704000
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016502166586422726,
+      "loss": 1.4032,
+      "theoretical_loss": 3.3706895169013715,
+      "tokens_seen": 2760835072
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016498154389343607,
+      "loss": 1.359,
+      "theoretical_loss": 3.3706771104473052,
+      "tokens_seen": 2760966144
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016494142192264484,
+      "loss": 1.2609,
+      "theoretical_loss": 3.3706647047471034,
+      "tokens_seen": 2761097216
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016490129995185362,
+      "loss": 1.2991,
+      "theoretical_loss": 3.3706522998006845,
+      "tokens_seen": 2761228288
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016486117798106243,
+      "loss": 1.2906,
+      "theoretical_loss": 3.370639895607967,
+      "tokens_seen": 2761359360
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016482105601027124,
+      "loss": 1.3861,
+      "theoretical_loss": 3.3706274921688695,
+      "tokens_seen": 2761490432
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016478093403948002,
+      "loss": 1.379,
+      "theoretical_loss": 3.3706150894833096,
+      "tokens_seen": 2761621504
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016474081206868882,
+      "loss": 1.3454,
+      "theoretical_loss": 3.370602687551207,
+      "tokens_seen": 2761752576
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016470069009789763,
+      "loss": 1.3134,
+      "theoretical_loss": 3.3705902863724795,
+      "tokens_seen": 2761883648
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001646605681271064,
+      "loss": 1.335,
+      "theoretical_loss": 3.370577885947045,
+      "tokens_seen": 2762014720
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016462044615631519,
+      "loss": 1.3132,
+      "theoretical_loss": 3.370565486274823,
+      "tokens_seen": 2762145792
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.000164580324185524,
+      "loss": 1.3061,
+      "theoretical_loss": 3.3705530873557317,
+      "tokens_seen": 2762276864
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001645402022147328,
+      "loss": 1.3249,
+      "theoretical_loss": 3.3705406891896894,
+      "tokens_seen": 2762407936
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016450008024394158,
+      "loss": 1.3258,
+      "theoretical_loss": 3.370528291776615,
+      "tokens_seen": 2762539008
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016445995827315038,
+      "loss": 1.307,
+      "theoretical_loss": 3.370515895116427,
+      "tokens_seen": 2762670080
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001644198363023592,
+      "loss": 1.2795,
+      "theoretical_loss": 3.3705034992090432,
+      "tokens_seen": 2762801152
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016437971433156797,
+      "loss": 1.4106,
+      "theoretical_loss": 3.3704911040543832,
+      "tokens_seen": 2762932224
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016433959236077675,
+      "loss": 1.3891,
+      "theoretical_loss": 3.370478709652365,
+      "tokens_seen": 2763063296
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016429947038998556,
+      "loss": 1.3515,
+      "theoretical_loss": 3.370466316002908,
+      "tokens_seen": 2763194368
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016425934841919436,
+      "loss": 1.254,
+      "theoretical_loss": 3.3704539231059294,
+      "tokens_seen": 2763325440
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016421922644840314,
+      "loss": 1.2444,
+      "theoretical_loss": 3.3704415309613487,
+      "tokens_seen": 2763456512
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016417910447761195,
+      "loss": 1.2567,
+      "theoretical_loss": 3.370429139569085,
+      "tokens_seen": 2763587584
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016413898250682075,
+      "loss": 1.2663,
+      "theoretical_loss": 3.3704167489290557,
+      "tokens_seen": 2763718656
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016409886053602953,
+      "loss": 1.2997,
+      "theoretical_loss": 3.3704043590411805,
+      "tokens_seen": 2763849728
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016405873856523834,
+      "loss": 1.3062,
+      "theoretical_loss": 3.3703919699053775,
+      "tokens_seen": 2763980800
+    },
+    {
+      "epoch": 0.68,
+      "objective/train/advantage_avg": 0.4913116693496704,
+      "objective/train/docs_used": 1003856,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2214455604553223,
+      "objective/train/original_loss": 1.2214454412460327,
+      "objective/train/theoretical_loss": 3.370379581521566,
+      "objective/train/tokens_used": 1134570976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24182650446891785,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503605604171753,
+      "objective/train/weighted_lm_loss": 1.2829574346542358,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.370379581521566,
+      "tokens_seen": 2764111872
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016401861659444712,
+      "loss": 1.3476,
+      "theoretical_loss": 3.370379581521566,
+      "tokens_seen": 2764111872
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001639784946236559,
+      "loss": 1.3267,
+      "theoretical_loss": 3.3703671938896638,
+      "tokens_seen": 2764242944
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001639383726528647,
+      "loss": 1.279,
+      "theoretical_loss": 3.3703548070095906,
+      "tokens_seen": 2764374016
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001638982506820735,
+      "loss": 1.3602,
+      "theoretical_loss": 3.370342420881264,
+      "tokens_seen": 2764505088
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016385812871128232,
+      "loss": 1.387,
+      "theoretical_loss": 3.3703300355046037,
+      "tokens_seen": 2764636160
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001638180067404911,
+      "loss": 1.3498,
+      "theoretical_loss": 3.3703176508795285,
+      "tokens_seen": 2764767232
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001637778847696999,
+      "loss": 1.2857,
+      "theoretical_loss": 3.3703052670059566,
+      "tokens_seen": 2764898304
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016373776279890868,
+      "loss": 1.3376,
+      "theoretical_loss": 3.370292883883807,
+      "tokens_seen": 2765029376
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016369764082811746,
+      "loss": 1.3088,
+      "theoretical_loss": 3.3702805015129984,
+      "tokens_seen": 2765160448
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016365751885732627,
+      "loss": 1.3915,
+      "theoretical_loss": 3.3702681198934497,
+      "tokens_seen": 2765291520
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016361739688653507,
+      "loss": 1.3866,
+      "theoretical_loss": 3.3702557390250796,
+      "tokens_seen": 2765422592
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016357727491574388,
+      "loss": 1.334,
+      "theoretical_loss": 3.370243358907807,
+      "tokens_seen": 2765553664
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016353715294495266,
+      "loss": 1.3588,
+      "theoretical_loss": 3.370230979541551,
+      "tokens_seen": 2765684736
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016349703097416146,
+      "loss": 1.3218,
+      "theoretical_loss": 3.37021860092623,
+      "tokens_seen": 2765815808
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016345690900337024,
+      "loss": 1.3147,
+      "theoretical_loss": 3.3702062230617633,
+      "tokens_seen": 2765946880
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016341678703257902,
+      "loss": 1.3869,
+      "theoretical_loss": 3.3701938459480694,
+      "tokens_seen": 2766077952
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016337666506178783,
+      "loss": 1.4266,
+      "theoretical_loss": 3.370181469585067,
+      "tokens_seen": 2766209024
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016333654309099664,
+      "loss": 1.2508,
+      "theoretical_loss": 3.3701690939726756,
+      "tokens_seen": 2766340096
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016329642112020544,
+      "loss": 1.3526,
+      "theoretical_loss": 3.3701567191108137,
+      "tokens_seen": 2766471168
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016325629914941422,
+      "loss": 1.2731,
+      "theoretical_loss": 3.3701443449994004,
+      "tokens_seen": 2766602240
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016321617717862303,
+      "loss": 1.3262,
+      "theoretical_loss": 3.370131971638355,
+      "tokens_seen": 2766733312
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016317605520783183,
+      "loss": 1.297,
+      "theoretical_loss": 3.370119599027596,
+      "tokens_seen": 2766864384
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016313593323704059,
+      "loss": 1.3172,
+      "theoretical_loss": 3.370107227167042,
+      "tokens_seen": 2766995456
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001630958112662494,
+      "loss": 1.3375,
+      "theoretical_loss": 3.3700948560566126,
+      "tokens_seen": 2767126528
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001630556892954582,
+      "loss": 1.2703,
+      "theoretical_loss": 3.370082485696227,
+      "tokens_seen": 2767257600
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016301556732466698,
+      "loss": 1.3395,
+      "theoretical_loss": 3.3700701160858033,
+      "tokens_seen": 2767388672
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016297544535387578,
+      "loss": 1.284,
+      "theoretical_loss": 3.3700577472252613,
+      "tokens_seen": 2767519744
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001629353233830846,
+      "loss": 1.3069,
+      "theoretical_loss": 3.3700453791145195,
+      "tokens_seen": 2767650816
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001628952014122934,
+      "loss": 1.3456,
+      "theoretical_loss": 3.370033011753497,
+      "tokens_seen": 2767781888
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016285507944150215,
+      "loss": 1.3437,
+      "theoretical_loss": 3.3700206451421137,
+      "tokens_seen": 2767912960
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016281495747071096,
+      "loss": 1.3487,
+      "theoretical_loss": 3.3700082792802877,
+      "tokens_seen": 2768044032
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016277483549991976,
+      "loss": 1.3227,
+      "theoretical_loss": 3.369995914167938,
+      "tokens_seen": 2768175104
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016273471352912854,
+      "loss": 1.3132,
+      "theoretical_loss": 3.369983549804985,
+      "tokens_seen": 2768306176
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016269459155833735,
+      "loss": 1.2996,
+      "theoretical_loss": 3.3699711861913464,
+      "tokens_seen": 2768437248
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016265446958754615,
+      "loss": 1.3611,
+      "theoretical_loss": 3.3699588233269413,
+      "tokens_seen": 2768568320
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016261434761675496,
+      "loss": 1.4106,
+      "theoretical_loss": 3.36994646121169,
+      "tokens_seen": 2768699392
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016257422564596374,
+      "loss": 1.1924,
+      "theoretical_loss": 3.3699340998455107,
+      "tokens_seen": 2768830464
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016253410367517252,
+      "loss": 1.3058,
+      "theoretical_loss": 3.369921739228323,
+      "tokens_seen": 2768961536
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016249398170438132,
+      "loss": 1.3623,
+      "theoretical_loss": 3.369909379360046,
+      "tokens_seen": 2769092608
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001624538597335901,
+      "loss": 1.3169,
+      "theoretical_loss": 3.3698970202405985,
+      "tokens_seen": 2769223680
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001624137377627989,
+      "loss": 1.3313,
+      "theoretical_loss": 3.3698846618698997,
+      "tokens_seen": 2769354752
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016237361579200772,
+      "loss": 1.2923,
+      "theoretical_loss": 3.3698723042478695,
+      "tokens_seen": 2769485824
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016233349382121652,
+      "loss": 1.3959,
+      "theoretical_loss": 3.3698599473744264,
+      "tokens_seen": 2769616896
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001622933718504253,
+      "loss": 1.3965,
+      "theoretical_loss": 3.3698475912494903,
+      "tokens_seen": 2769747968
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016225324987963408,
+      "loss": 1.3313,
+      "theoretical_loss": 3.36983523587298,
+      "tokens_seen": 2769879040
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001622131279088429,
+      "loss": 1.3103,
+      "theoretical_loss": 3.369822881244814,
+      "tokens_seen": 2770010112
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016217300593805167,
+      "loss": 1.3282,
+      "theoretical_loss": 3.3698105273649133,
+      "tokens_seen": 2770141184
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016213288396726047,
+      "loss": 1.3709,
+      "theoretical_loss": 3.369798174233196,
+      "tokens_seen": 2770272256
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016209276199646928,
+      "loss": 1.2935,
+      "theoretical_loss": 3.3697858218495815,
+      "tokens_seen": 2770403328
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016205264002567809,
+      "loss": 1.3427,
+      "theoretical_loss": 3.3697734702139894,
+      "tokens_seen": 2770534400
+    },
+    {
+      "epoch": 0.68,
+      "objective/train/advantage_avg": 0.4930559992790222,
+      "objective/train/docs_used": 1006223,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.1650590896606445,
+      "objective/train/original_loss": 1.1650590896606445,
+      "objective/train/theoretical_loss": 3.3697611193263386,
+      "objective/train/tokens_used": 1141124576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24372906982898712,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505446195602417,
+      "objective/train/weighted_lm_loss": 1.2238843441009521,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3697611193263386,
+      "tokens_seen": 2770665472
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016201251805488686,
+      "loss": 1.2904,
+      "theoretical_loss": 3.3697611193263386,
+      "tokens_seen": 2770665472
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016197239608409564,
+      "loss": 1.3273,
+      "theoretical_loss": 3.369748769186549,
+      "tokens_seen": 2770796544
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016193227411330445,
+      "loss": 1.4137,
+      "theoretical_loss": 3.3697364197945396,
+      "tokens_seen": 2770927616
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016189215214251323,
+      "loss": 1.3476,
+      "theoretical_loss": 3.36972407115023,
+      "tokens_seen": 2771058688
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016185203017172204,
+      "loss": 1.2904,
+      "theoretical_loss": 3.369711723253539,
+      "tokens_seen": 2771189760
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016181190820093084,
+      "loss": 1.3155,
+      "theoretical_loss": 3.3696993761043865,
+      "tokens_seen": 2771320832
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016177178623013962,
+      "loss": 1.3008,
+      "theoretical_loss": 3.3696870297026917,
+      "tokens_seen": 2771451904
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016173166425934843,
+      "loss": 1.3667,
+      "theoretical_loss": 3.369674684048374,
+      "tokens_seen": 2771582976
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016169154228855723,
+      "loss": 1.3394,
+      "theoretical_loss": 3.3696623391413527,
+      "tokens_seen": 2771714048
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.000161651420317766,
+      "loss": 1.2862,
+      "theoretical_loss": 3.3696499949815477,
+      "tokens_seen": 2771845120
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001616112983469748,
+      "loss": 1.2665,
+      "theoretical_loss": 3.369637651568878,
+      "tokens_seen": 2771976192
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001615711763761836,
+      "loss": 1.3267,
+      "theoretical_loss": 3.369625308903263,
+      "tokens_seen": 2772107264
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001615310544053924,
+      "loss": 1.3333,
+      "theoretical_loss": 3.3696129669846226,
+      "tokens_seen": 2772238336
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016149093243460118,
+      "loss": 1.2635,
+      "theoretical_loss": 3.369600625812876,
+      "tokens_seen": 2772369408
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016145081046381,
+      "loss": 1.2567,
+      "theoretical_loss": 3.3695882853879424,
+      "tokens_seen": 2772500480
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001614106884930188,
+      "loss": 1.2829,
+      "theoretical_loss": 3.369575945709742,
+      "tokens_seen": 2772631552
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016137056652222758,
+      "loss": 1.2498,
+      "theoretical_loss": 3.3695636067781938,
+      "tokens_seen": 2772762624
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016133044455143635,
+      "loss": 1.2558,
+      "theoretical_loss": 3.3695512685932174,
+      "tokens_seen": 2772893696
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016129032258064516,
+      "loss": 1.3592,
+      "theoretical_loss": 3.3695389311547324,
+      "tokens_seen": 2773024768
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016125020060985397,
+      "loss": 1.2819,
+      "theoretical_loss": 3.3695265944626587,
+      "tokens_seen": 2773155840
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016121007863906275,
+      "loss": 1.3206,
+      "theoretical_loss": 3.369514258516915,
+      "tokens_seen": 2773286912
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016116995666827155,
+      "loss": 1.318,
+      "theoretical_loss": 3.3695019233174217,
+      "tokens_seen": 2773417984
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016112983469748036,
+      "loss": 1.3158,
+      "theoretical_loss": 3.3694895888640986,
+      "tokens_seen": 2773549056
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001610897127266891,
+      "loss": 1.3041,
+      "theoretical_loss": 3.369477255156864,
+      "tokens_seen": 2773680128
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016104959075589792,
+      "loss": 1.3529,
+      "theoretical_loss": 3.3694649221956388,
+      "tokens_seen": 2773811200
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016100946878510672,
+      "loss": 1.2909,
+      "theoretical_loss": 3.369452589980342,
+      "tokens_seen": 2773942272
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016096934681431553,
+      "loss": 1.3613,
+      "theoretical_loss": 3.369440258510893,
+      "tokens_seen": 2774073344
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001609292248435243,
+      "loss": 1.3417,
+      "theoretical_loss": 3.3694279277872123,
+      "tokens_seen": 2774204416
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016088910287273312,
+      "loss": 1.3505,
+      "theoretical_loss": 3.369415597809219,
+      "tokens_seen": 2774335488
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016084898090194192,
+      "loss": 1.3371,
+      "theoretical_loss": 3.369403268576833,
+      "tokens_seen": 2774466560
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001608088589311507,
+      "loss": 1.3034,
+      "theoretical_loss": 3.369390940089974,
+      "tokens_seen": 2774597632
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016076873696035948,
+      "loss": 1.2423,
+      "theoretical_loss": 3.3693786123485614,
+      "tokens_seen": 2774728704
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001607286149895683,
+      "loss": 1.3192,
+      "theoretical_loss": 3.369366285352515,
+      "tokens_seen": 2774859776
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001606884930187771,
+      "loss": 1.3218,
+      "theoretical_loss": 3.3693539591017547,
+      "tokens_seen": 2774990848
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016064837104798587,
+      "loss": 1.3425,
+      "theoretical_loss": 3.3693416335962,
+      "tokens_seen": 2775121920
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016060824907719468,
+      "loss": 1.2382,
+      "theoretical_loss": 3.369329308835771,
+      "tokens_seen": 2775252992
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016056812710640348,
+      "loss": 1.232,
+      "theoretical_loss": 3.3693169848203874,
+      "tokens_seen": 2775384064
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016052800513561226,
+      "loss": 1.2826,
+      "theoretical_loss": 3.3693046615499687,
+      "tokens_seen": 2775515136
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016048788316482104,
+      "loss": 1.3069,
+      "theoretical_loss": 3.369292339024435,
+      "tokens_seen": 2775646208
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016044776119402985,
+      "loss": 1.2329,
+      "theoretical_loss": 3.369280017243706,
+      "tokens_seen": 2775777280
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016040763922323866,
+      "loss": 1.3103,
+      "theoretical_loss": 3.3692676962077015,
+      "tokens_seen": 2775908352
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016036751725244744,
+      "loss": 1.2492,
+      "theoretical_loss": 3.369255375916341,
+      "tokens_seen": 2776039424
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016032739528165624,
+      "loss": 1.294,
+      "theoretical_loss": 3.369243056369545,
+      "tokens_seen": 2776170496
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016028727331086505,
+      "loss": 1.3255,
+      "theoretical_loss": 3.369230737567233,
+      "tokens_seen": 2776301568
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016024715134007383,
+      "loss": 1.3325,
+      "theoretical_loss": 3.3692184195093247,
+      "tokens_seen": 2776432640
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001602070293692826,
+      "loss": 1.3304,
+      "theoretical_loss": 3.36920610219574,
+      "tokens_seen": 2776563712
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001601669073984914,
+      "loss": 1.3268,
+      "theoretical_loss": 3.369193785626399,
+      "tokens_seen": 2776694784
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00016012678542770022,
+      "loss": 1.2586,
+      "theoretical_loss": 3.369181469801222,
+      "tokens_seen": 2776825856
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.000160086663456909,
+      "loss": 1.2952,
+      "theoretical_loss": 3.369169154720128,
+      "tokens_seen": 2776956928
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001600465414861178,
+      "loss": 1.3365,
+      "theoretical_loss": 3.369156840383037,
+      "tokens_seen": 2777088000
+    },
+    {
+      "epoch": 0.68,
+      "objective/train/advantage_avg": 0.4946650564670563,
+      "objective/train/docs_used": 1008708,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2027205228805542,
+      "objective/train/original_loss": 1.2027205228805542,
+      "objective/train/theoretical_loss": 3.3691445267898703,
+      "objective/train/tokens_used": 1147678176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24509866535663605,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0507124662399292,
+      "objective/train/weighted_lm_loss": 1.2636467218399048,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3691445267898703,
+      "tokens_seen": 2777219072
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001600064195153266,
+      "loss": 1.2833,
+      "theoretical_loss": 3.3691445267898703,
+      "tokens_seen": 2777219072
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001599662975445354,
+      "loss": 1.279,
+      "theoretical_loss": 3.3691322139405466,
+      "tokens_seen": 2777350144
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001599261755737442,
+      "loss": 1.283,
+      "theoretical_loss": 3.3691199018349858,
+      "tokens_seen": 2777481216
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00015988605360295298,
+      "loss": 1.3521,
+      "theoretical_loss": 3.3691075904731083,
+      "tokens_seen": 2777612288
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00015984593163216175,
+      "loss": 1.4023,
+      "theoretical_loss": 3.369095279854834,
+      "tokens_seen": 2777743360
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00015980580966137056,
+      "loss": 1.3501,
+      "theoretical_loss": 3.369082969980083,
+      "tokens_seen": 2777874432
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00015976568769057937,
+      "loss": 1.2556,
+      "theoretical_loss": 3.369070660848775,
+      "tokens_seen": 2778005504
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00015972556571978817,
+      "loss": 1.2767,
+      "theoretical_loss": 3.3690583524608306,
+      "tokens_seen": 2778136576
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00015968544374899695,
+      "loss": 1.242,
+      "theoretical_loss": 3.3690460448161694,
+      "tokens_seen": 2778267648
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00015964532177820576,
+      "loss": 1.3704,
+      "theoretical_loss": 3.369033737914711,
+      "tokens_seen": 2778398720
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00015960519980741454,
+      "loss": 1.2492,
+      "theoretical_loss": 3.3690214317563765,
+      "tokens_seen": 2778529792
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00015956507783662332,
+      "loss": 1.3552,
+      "theoretical_loss": 3.3690091263410857,
+      "tokens_seen": 2778660864
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00015952495586583212,
+      "loss": 1.3183,
+      "theoretical_loss": 3.368996821668758,
+      "tokens_seen": 2778791936
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00015948483389504093,
+      "loss": 1.3447,
+      "theoretical_loss": 3.3689845177393143,
+      "tokens_seen": 2778923008
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00015944471192424974,
+      "loss": 1.319,
+      "theoretical_loss": 3.368972214552674,
+      "tokens_seen": 2779054080
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00015940458995345852,
+      "loss": 1.2667,
+      "theoretical_loss": 3.368959912108758,
+      "tokens_seen": 2779185152
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00015936446798266732,
+      "loss": 1.3283,
+      "theoretical_loss": 3.368947610407486,
+      "tokens_seen": 2779316224
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00015932434601187613,
+      "loss": 1.3259,
+      "theoretical_loss": 3.368935309448778,
+      "tokens_seen": 2779447296
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00015928422404108488,
+      "loss": 1.354,
+      "theoretical_loss": 3.3689230092325544,
+      "tokens_seen": 2779578368
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001592441020702937,
+      "loss": 1.3337,
+      "theoretical_loss": 3.3689107097587354,
+      "tokens_seen": 2779709440
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001592039800995025,
+      "loss": 1.2724,
+      "theoretical_loss": 3.368898411027241,
+      "tokens_seen": 2779840512
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001591638581287113,
+      "loss": 1.2698,
+      "theoretical_loss": 3.3688861130379912,
+      "tokens_seen": 2779971584
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00015912373615792008,
+      "loss": 1.3323,
+      "theoretical_loss": 3.3688738157909075,
+      "tokens_seen": 2780102656
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015908361418712888,
+      "loss": 1.2269,
+      "theoretical_loss": 3.3688615192859084,
+      "tokens_seen": 2780233728
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001590434922163377,
+      "loss": 1.3519,
+      "theoretical_loss": 3.368849223522915,
+      "tokens_seen": 2780364800
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015900337024554644,
+      "loss": 1.3192,
+      "theoretical_loss": 3.3688369285018473,
+      "tokens_seen": 2780495872
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015896324827475525,
+      "loss": 1.3588,
+      "theoretical_loss": 3.368824634222626,
+      "tokens_seen": 2780626944
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015892312630396406,
+      "loss": 1.272,
+      "theoretical_loss": 3.368812340685171,
+      "tokens_seen": 2780758016
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015888300433317283,
+      "loss": 1.2904,
+      "theoretical_loss": 3.3688000478894025,
+      "tokens_seen": 2780889088
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015884288236238164,
+      "loss": 1.2932,
+      "theoretical_loss": 3.368787755835241,
+      "tokens_seen": 2781020160
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015880276039159045,
+      "loss": 1.3531,
+      "theoretical_loss": 3.368775464522607,
+      "tokens_seen": 2781151232
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015876263842079925,
+      "loss": 1.318,
+      "theoretical_loss": 3.3687631739514203,
+      "tokens_seen": 2781282304
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.000158722516450008,
+      "loss": 1.3002,
+      "theoretical_loss": 3.3687508841216016,
+      "tokens_seen": 2781413376
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001586823944792168,
+      "loss": 1.334,
+      "theoretical_loss": 3.3687385950330713,
+      "tokens_seen": 2781544448
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015864227250842562,
+      "loss": 1.2913,
+      "theoretical_loss": 3.3687263066857493,
+      "tokens_seen": 2781675520
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001586021505376344,
+      "loss": 1.3143,
+      "theoretical_loss": 3.368714019079557,
+      "tokens_seen": 2781806592
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001585620285668432,
+      "loss": 1.3485,
+      "theoretical_loss": 3.3687017322144133,
+      "tokens_seen": 2781937664
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.000158521906596052,
+      "loss": 1.3353,
+      "theoretical_loss": 3.3686894460902397,
+      "tokens_seen": 2782068736
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015848178462526082,
+      "loss": 1.3434,
+      "theoretical_loss": 3.368677160706956,
+      "tokens_seen": 2782199808
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001584416626544696,
+      "loss": 1.3112,
+      "theoretical_loss": 3.3686648760644835,
+      "tokens_seen": 2782330880
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015840154068367837,
+      "loss": 1.3138,
+      "theoretical_loss": 3.368652592162742,
+      "tokens_seen": 2782461952
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015836141871288718,
+      "loss": 1.3428,
+      "theoretical_loss": 3.3686403090016515,
+      "tokens_seen": 2782593024
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015832129674209596,
+      "loss": 1.3028,
+      "theoretical_loss": 3.368628026581133,
+      "tokens_seen": 2782724096
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015828117477130477,
+      "loss": 1.3559,
+      "theoretical_loss": 3.368615744901107,
+      "tokens_seen": 2782855168
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015824105280051357,
+      "loss": 1.3221,
+      "theoretical_loss": 3.3686034639614943,
+      "tokens_seen": 2782986240
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015820093082972238,
+      "loss": 1.3418,
+      "theoretical_loss": 3.3685911837622147,
+      "tokens_seen": 2783117312
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015816080885893116,
+      "loss": 1.2763,
+      "theoretical_loss": 3.368578904303189,
+      "tokens_seen": 2783248384
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015812068688813994,
+      "loss": 1.2957,
+      "theoretical_loss": 3.3685666255843376,
+      "tokens_seen": 2783379456
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015808056491734874,
+      "loss": 1.3221,
+      "theoretical_loss": 3.3685543476055817,
+      "tokens_seen": 2783510528
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015804044294655752,
+      "loss": 1.4586,
+      "theoretical_loss": 3.368542070366841,
+      "tokens_seen": 2783641600
+    },
+    {
+      "epoch": 0.69,
+      "objective/train/advantage_avg": 0.491060346364975,
+      "objective/train/docs_used": 1011173,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.1923192739486694,
+      "objective/train/original_loss": 1.192319393157959,
+      "objective/train/theoretical_loss": 3.3685297938680363,
+      "objective/train/tokens_used": 1154231776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24215741455554962,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503370761871338,
+      "objective/train/weighted_lm_loss": 1.2522099018096924,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3685297938680363,
+      "tokens_seen": 2783772672
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015800032097576633,
+      "loss": 1.2736,
+      "theoretical_loss": 3.3685297938680363,
+      "tokens_seen": 2783772672
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015796019900497514,
+      "loss": 1.3471,
+      "theoretical_loss": 3.368517518109088,
+      "tokens_seen": 2783903744
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015792007703418394,
+      "loss": 1.2869,
+      "theoretical_loss": 3.368505243089917,
+      "tokens_seen": 2784034816
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015787995506339272,
+      "loss": 1.3531,
+      "theoretical_loss": 3.368492968810444,
+      "tokens_seen": 2784165888
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001578398330926015,
+      "loss": 1.2589,
+      "theoretical_loss": 3.3684806952705895,
+      "tokens_seen": 2784296960
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001577997111218103,
+      "loss": 1.304,
+      "theoretical_loss": 3.368468422470274,
+      "tokens_seen": 2784428032
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015775958915101909,
+      "loss": 1.3435,
+      "theoretical_loss": 3.3684561504094175,
+      "tokens_seen": 2784559104
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001577194671802279,
+      "loss": 1.2699,
+      "theoretical_loss": 3.368443879087942,
+      "tokens_seen": 2784690176
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001576793452094367,
+      "loss": 1.3064,
+      "theoretical_loss": 3.3684316085057673,
+      "tokens_seen": 2784821248
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015763922323864548,
+      "loss": 1.2655,
+      "theoretical_loss": 3.368419338662814,
+      "tokens_seen": 2784952320
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015759910126785428,
+      "loss": 1.371,
+      "theoretical_loss": 3.3684070695590034,
+      "tokens_seen": 2785083392
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001575589792970631,
+      "loss": 1.3465,
+      "theoretical_loss": 3.3683948011942553,
+      "tokens_seen": 2785214464
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015751885732627187,
+      "loss": 1.2349,
+      "theoretical_loss": 3.368382533568491,
+      "tokens_seen": 2785345536
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015747873535548065,
+      "loss": 1.3439,
+      "theoretical_loss": 3.3683702666816315,
+      "tokens_seen": 2785476608
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015743861338468946,
+      "loss": 1.3448,
+      "theoretical_loss": 3.368358000533597,
+      "tokens_seen": 2785607680
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015739849141389826,
+      "loss": 1.353,
+      "theoretical_loss": 3.368345735124308,
+      "tokens_seen": 2785738752
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015735836944310704,
+      "loss": 1.2978,
+      "theoretical_loss": 3.368333470453686,
+      "tokens_seen": 2785869824
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015731824747231585,
+      "loss": 1.3718,
+      "theoretical_loss": 3.3683212065216512,
+      "tokens_seen": 2786000896
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015727812550152465,
+      "loss": 1.3369,
+      "theoretical_loss": 3.368308943328125,
+      "tokens_seen": 2786131968
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015723800353073343,
+      "loss": 1.3913,
+      "theoretical_loss": 3.3682966808730272,
+      "tokens_seen": 2786263040
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001571978815599422,
+      "loss": 1.2821,
+      "theoretical_loss": 3.368284419156279,
+      "tokens_seen": 2786394112
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015715775958915102,
+      "loss": 1.3172,
+      "theoretical_loss": 3.3682721581778017,
+      "tokens_seen": 2786525184
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015711763761835982,
+      "loss": 1.3241,
+      "theoretical_loss": 3.368259897937516,
+      "tokens_seen": 2786656256
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001570775156475686,
+      "loss": 1.3157,
+      "theoretical_loss": 3.368247638435342,
+      "tokens_seen": 2786787328
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001570373936767774,
+      "loss": 1.4241,
+      "theoretical_loss": 3.368235379671202,
+      "tokens_seen": 2786918400
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015699727170598622,
+      "loss": 1.3761,
+      "theoretical_loss": 3.3682231216450154,
+      "tokens_seen": 2787049472
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015695714973519502,
+      "loss": 1.372,
+      "theoretical_loss": 3.3682108643567035,
+      "tokens_seen": 2787180544
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015691702776440377,
+      "loss": 1.2961,
+      "theoretical_loss": 3.368198607806187,
+      "tokens_seen": 2787311616
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015687690579361258,
+      "loss": 1.3511,
+      "theoretical_loss": 3.368186351993388,
+      "tokens_seen": 2787442688
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001568367838228214,
+      "loss": 1.3127,
+      "theoretical_loss": 3.368174096918226,
+      "tokens_seen": 2787573760
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015679666185203017,
+      "loss": 1.3081,
+      "theoretical_loss": 3.3681618425806223,
+      "tokens_seen": 2787704832
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015675653988123897,
+      "loss": 1.3318,
+      "theoretical_loss": 3.368149588980498,
+      "tokens_seen": 2787835904
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015671641791044778,
+      "loss": 1.2663,
+      "theoretical_loss": 3.3681373361177744,
+      "tokens_seen": 2787966976
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015667629593965656,
+      "loss": 1.3267,
+      "theoretical_loss": 3.3681250839923718,
+      "tokens_seen": 2788098048
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015663617396886534,
+      "loss": 1.2994,
+      "theoretical_loss": 3.368112832604212,
+      "tokens_seen": 2788229120
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015659605199807414,
+      "loss": 1.3329,
+      "theoretical_loss": 3.3681005819532146,
+      "tokens_seen": 2788360192
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015655593002728295,
+      "loss": 1.3723,
+      "theoretical_loss": 3.368088332039302,
+      "tokens_seen": 2788491264
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015651580805649173,
+      "loss": 1.3819,
+      "theoretical_loss": 3.368076082862394,
+      "tokens_seen": 2788622336
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015647568608570054,
+      "loss": 1.3249,
+      "theoretical_loss": 3.368063834422413,
+      "tokens_seen": 2788753408
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015643556411490934,
+      "loss": 1.3334,
+      "theoretical_loss": 3.368051586719279,
+      "tokens_seen": 2788884480
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015639544214411812,
+      "loss": 1.3386,
+      "theoretical_loss": 3.3680393397529134,
+      "tokens_seen": 2789015552
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001563553201733269,
+      "loss": 1.3496,
+      "theoretical_loss": 3.3680270935232373,
+      "tokens_seen": 2789146624
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001563151982025357,
+      "loss": 1.2867,
+      "theoretical_loss": 3.3680148480301715,
+      "tokens_seen": 2789277696
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001562750762317445,
+      "loss": 1.3204,
+      "theoretical_loss": 3.368002603273637,
+      "tokens_seen": 2789408768
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001562349542609533,
+      "loss": 1.2813,
+      "theoretical_loss": 3.3679903592535556,
+      "tokens_seen": 2789539840
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001561948322901621,
+      "loss": 1.4016,
+      "theoretical_loss": 3.367978115969848,
+      "tokens_seen": 2789670912
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001561547103193709,
+      "loss": 1.3357,
+      "theoretical_loss": 3.3679658734224347,
+      "tokens_seen": 2789801984
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015611458834857968,
+      "loss": 1.2844,
+      "theoretical_loss": 3.367953631611238,
+      "tokens_seen": 2789933056
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001560744663777885,
+      "loss": 1.2936,
+      "theoretical_loss": 3.3679413905361786,
+      "tokens_seen": 2790064128
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015603434440699727,
+      "loss": 1.3418,
+      "theoretical_loss": 3.3679291501971766,
+      "tokens_seen": 2790195200
+    },
+    {
+      "epoch": 0.69,
+      "objective/train/advantage_avg": 0.4916509985923767,
+      "objective/train/docs_used": 1013591,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3807640075683594,
+      "objective/train/original_loss": 1.3807637691497803,
+      "objective/train/theoretical_loss": 3.367916910594155,
+      "objective/train/tokens_used": 1160785376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2423628568649292,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050397276878357,
+      "objective/train/weighted_lm_loss": 1.4503434896469116,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.367916910594155,
+      "tokens_seen": 2790326272
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015599422243620608,
+      "loss": 1.4034,
+      "theoretical_loss": 3.367916910594155,
+      "tokens_seen": 2790326272
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015595410046541485,
+      "loss": 1.3456,
+      "theoretical_loss": 3.3679046717270333,
+      "tokens_seen": 2790457344
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015591397849462366,
+      "loss": 1.3126,
+      "theoretical_loss": 3.367892433595734,
+      "tokens_seen": 2790588416
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015587385652383247,
+      "loss": 1.2852,
+      "theoretical_loss": 3.3678801962001774,
+      "tokens_seen": 2790719488
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015583373455304125,
+      "loss": 1.3992,
+      "theoretical_loss": 3.3678679595402854,
+      "tokens_seen": 2790850560
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015579361258225005,
+      "loss": 1.3362,
+      "theoretical_loss": 3.3678557236159783,
+      "tokens_seen": 2790981632
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015575349061145883,
+      "loss": 1.3508,
+      "theoretical_loss": 3.3678434884271784,
+      "tokens_seen": 2791112704
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001557133686406676,
+      "loss": 1.3573,
+      "theoretical_loss": 3.3678312539738062,
+      "tokens_seen": 2791243776
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015567324666987642,
+      "loss": 1.2756,
+      "theoretical_loss": 3.3678190202557836,
+      "tokens_seen": 2791374848
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015563312469908522,
+      "loss": 1.2731,
+      "theoretical_loss": 3.3678067872730315,
+      "tokens_seen": 2791505920
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015559300272829403,
+      "loss": 1.3397,
+      "theoretical_loss": 3.367794555025471,
+      "tokens_seen": 2791636992
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001555528807575028,
+      "loss": 1.3425,
+      "theoretical_loss": 3.3677823235130235,
+      "tokens_seen": 2791768064
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015551275878671162,
+      "loss": 1.394,
+      "theoretical_loss": 3.3677700927356105,
+      "tokens_seen": 2791899136
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001554726368159204,
+      "loss": 1.3178,
+      "theoretical_loss": 3.3677578626931535,
+      "tokens_seen": 2792030208
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015543251484512917,
+      "loss": 1.3665,
+      "theoretical_loss": 3.367745633385573,
+      "tokens_seen": 2792161280
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015539239287433798,
+      "loss": 1.3747,
+      "theoretical_loss": 3.3677334048127916,
+      "tokens_seen": 2792292352
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001553522709035468,
+      "loss": 1.309,
+      "theoretical_loss": 3.3677211769747295,
+      "tokens_seen": 2792423424
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001553121489327556,
+      "loss": 1.3173,
+      "theoretical_loss": 3.3677089498713086,
+      "tokens_seen": 2792554496
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015527202696196437,
+      "loss": 1.3134,
+      "theoretical_loss": 3.3676967235024504,
+      "tokens_seen": 2792685568
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015523190499117318,
+      "loss": 1.3149,
+      "theoretical_loss": 3.367684497868076,
+      "tokens_seen": 2792816640
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015519178302038198,
+      "loss": 1.3674,
+      "theoretical_loss": 3.367672272968107,
+      "tokens_seen": 2792947712
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015515166104959074,
+      "loss": 1.2948,
+      "theoretical_loss": 3.367660048802465,
+      "tokens_seen": 2793078784
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015511153907879954,
+      "loss": 1.3088,
+      "theoretical_loss": 3.367647825371071,
+      "tokens_seen": 2793209856
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015507141710800835,
+      "loss": 1.2803,
+      "theoretical_loss": 3.3676356026738468,
+      "tokens_seen": 2793340928
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015503129513721716,
+      "loss": 1.3089,
+      "theoretical_loss": 3.3676233807107137,
+      "tokens_seen": 2793472000
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015499117316642594,
+      "loss": 1.3515,
+      "theoretical_loss": 3.3676111594815925,
+      "tokens_seen": 2793603072
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015495105119563474,
+      "loss": 1.2759,
+      "theoretical_loss": 3.367598938986406,
+      "tokens_seen": 2793734144
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015491092922484355,
+      "loss": 1.3221,
+      "theoretical_loss": 3.3675867192250752,
+      "tokens_seen": 2793865216
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001548708072540523,
+      "loss": 1.2913,
+      "theoretical_loss": 3.367574500197521,
+      "tokens_seen": 2793996288
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001548306852832611,
+      "loss": 1.3238,
+      "theoretical_loss": 3.3675622819036657,
+      "tokens_seen": 2794127360
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001547905633124699,
+      "loss": 1.3346,
+      "theoretical_loss": 3.3675500643434306,
+      "tokens_seen": 2794258432
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001547504413416787,
+      "loss": 1.2491,
+      "theoretical_loss": 3.367537847516737,
+      "tokens_seen": 2794389504
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001547103193708875,
+      "loss": 1.3146,
+      "theoretical_loss": 3.3675256314235065,
+      "tokens_seen": 2794520576
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001546701974000963,
+      "loss": 1.3167,
+      "theoretical_loss": 3.3675134160636606,
+      "tokens_seen": 2794651648
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001546300754293051,
+      "loss": 1.3308,
+      "theoretical_loss": 3.3675012014371215,
+      "tokens_seen": 2794782720
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001545899534585139,
+      "loss": 1.3717,
+      "theoretical_loss": 3.36748898754381,
+      "tokens_seen": 2794913792
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015454983148772267,
+      "loss": 1.3688,
+      "theoretical_loss": 3.367476774383648,
+      "tokens_seen": 2795044864
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015450970951693148,
+      "loss": 1.2965,
+      "theoretical_loss": 3.3674645619565573,
+      "tokens_seen": 2795175936
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015446958754614025,
+      "loss": 1.3457,
+      "theoretical_loss": 3.3674523502624596,
+      "tokens_seen": 2795307008
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015442946557534906,
+      "loss": 1.3522,
+      "theoretical_loss": 3.367440139301276,
+      "tokens_seen": 2795438080
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015438934360455787,
+      "loss": 1.4071,
+      "theoretical_loss": 3.3674279290729285,
+      "tokens_seen": 2795569152
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015434922163376667,
+      "loss": 1.3562,
+      "theoretical_loss": 3.367415719577339,
+      "tokens_seen": 2795700224
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015430909966297545,
+      "loss": 1.3294,
+      "theoretical_loss": 3.3674035108144285,
+      "tokens_seen": 2795831296
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015426897769218423,
+      "loss": 1.3446,
+      "theoretical_loss": 3.367391302784119,
+      "tokens_seen": 2795962368
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015422885572139304,
+      "loss": 1.3547,
+      "theoretical_loss": 3.367379095486333,
+      "tokens_seen": 2796093440
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015418873375060182,
+      "loss": 1.3155,
+      "theoretical_loss": 3.367366888920991,
+      "tokens_seen": 2796224512
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015414861177981062,
+      "loss": 1.3429,
+      "theoretical_loss": 3.3673546830880157,
+      "tokens_seen": 2796355584
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015410848980901943,
+      "loss": 1.366,
+      "theoretical_loss": 3.3673424779873278,
+      "tokens_seen": 2796486656
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00015406836783822824,
+      "loss": 1.3429,
+      "theoretical_loss": 3.3673302736188497,
+      "tokens_seen": 2796617728
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015402824586743702,
+      "loss": 1.3347,
+      "theoretical_loss": 3.3673180699825034,
+      "tokens_seen": 2796748800
+    },
+    {
+      "epoch": 0.7,
+      "objective/train/advantage_avg": 0.4920945465564728,
+      "objective/train/docs_used": 1015946,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2084870338439941,
+      "objective/train/original_loss": 1.2084869146347046,
+      "objective/train/theoretical_loss": 3.36730586707821,
+      "objective/train/tokens_used": 1167338976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24325260519981384,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504460334777832,
+      "objective/train/weighted_lm_loss": 1.2693443298339844,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9048373699188232,
+      "theoretical_loss": 3.36730586707821,
+      "tokens_seen": 2796879872
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001539881238966458,
+      "loss": 1.3247,
+      "theoretical_loss": 3.36730586707821,
+      "tokens_seen": 2796879872
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001539480019258546,
+      "loss": 1.3138,
+      "theoretical_loss": 3.3672936649058918,
+      "tokens_seen": 2797010944
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015390787995506338,
+      "loss": 1.423,
+      "theoretical_loss": 3.3672814634654706,
+      "tokens_seen": 2797142016
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001538677579842722,
+      "loss": 1.3173,
+      "theoretical_loss": 3.3672692627568677,
+      "tokens_seen": 2797273088
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.000153827636013481,
+      "loss": 1.3693,
+      "theoretical_loss": 3.367257062780005,
+      "tokens_seen": 2797404160
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001537875140426898,
+      "loss": 1.4178,
+      "theoretical_loss": 3.367244863534805,
+      "tokens_seen": 2797535232
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015374739207189858,
+      "loss": 1.3513,
+      "theoretical_loss": 3.3672326650211892,
+      "tokens_seen": 2797666304
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015370727010110738,
+      "loss": 1.3425,
+      "theoretical_loss": 3.367220467239079,
+      "tokens_seen": 2797797376
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015366714813031616,
+      "loss": 1.2741,
+      "theoretical_loss": 3.3672082701883967,
+      "tokens_seen": 2797928448
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015362702615952494,
+      "loss": 1.3194,
+      "theoretical_loss": 3.3671960738690645,
+      "tokens_seen": 2798059520
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015358690418873375,
+      "loss": 1.2978,
+      "theoretical_loss": 3.3671838782810037,
+      "tokens_seen": 2798190592
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015354678221794256,
+      "loss": 1.322,
+      "theoretical_loss": 3.3671716834241368,
+      "tokens_seen": 2798321664
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015350666024715133,
+      "loss": 1.3297,
+      "theoretical_loss": 3.3671594892983845,
+      "tokens_seen": 2798452736
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015346653827636014,
+      "loss": 1.3994,
+      "theoretical_loss": 3.36714729590367,
+      "tokens_seen": 2798583808
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015342641630556895,
+      "loss": 1.2753,
+      "theoretical_loss": 3.367135103239915,
+      "tokens_seen": 2798714880
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015338629433477773,
+      "loss": 1.3747,
+      "theoretical_loss": 3.3671229113070407,
+      "tokens_seen": 2798845952
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001533461723639865,
+      "loss": 1.2957,
+      "theoretical_loss": 3.36711072010497,
+      "tokens_seen": 2798977024
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001533060503931953,
+      "loss": 1.3258,
+      "theoretical_loss": 3.3670985296336244,
+      "tokens_seen": 2799108096
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015326592842240412,
+      "loss": 1.2727,
+      "theoretical_loss": 3.3670863398929263,
+      "tokens_seen": 2799239168
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001532258064516129,
+      "loss": 1.3289,
+      "theoretical_loss": 3.367074150882797,
+      "tokens_seen": 2799370240
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001531856844808217,
+      "loss": 1.3465,
+      "theoretical_loss": 3.367061962603159,
+      "tokens_seen": 2799501312
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001531455625100305,
+      "loss": 1.2948,
+      "theoretical_loss": 3.367049775053934,
+      "tokens_seen": 2799632384
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001531054405392393,
+      "loss": 1.3121,
+      "theoretical_loss": 3.3670375882350445,
+      "tokens_seen": 2799763456
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015306531856844807,
+      "loss": 1.3015,
+      "theoretical_loss": 3.367025402146412,
+      "tokens_seen": 2799894528
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015302519659765687,
+      "loss": 1.2459,
+      "theoretical_loss": 3.367013216787959,
+      "tokens_seen": 2800025600
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015298507462686568,
+      "loss": 1.3244,
+      "theoretical_loss": 3.3670010321596076,
+      "tokens_seen": 2800156672
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015294495265607446,
+      "loss": 1.2941,
+      "theoretical_loss": 3.3669888482612795,
+      "tokens_seen": 2800287744
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015290483068528327,
+      "loss": 1.3448,
+      "theoretical_loss": 3.3669766650928974,
+      "tokens_seen": 2800418816
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015286470871449207,
+      "loss": 1.2854,
+      "theoretical_loss": 3.3669644826543825,
+      "tokens_seen": 2800549888
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015282458674370088,
+      "loss": 1.3264,
+      "theoretical_loss": 3.3669523009456577,
+      "tokens_seen": 2800680960
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015278446477290963,
+      "loss": 1.2606,
+      "theoretical_loss": 3.3669401199666447,
+      "tokens_seen": 2800812032
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015274434280211844,
+      "loss": 1.2655,
+      "theoretical_loss": 3.366927939717266,
+      "tokens_seen": 2800943104
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015270422083132724,
+      "loss": 1.3391,
+      "theoretical_loss": 3.3669157601974433,
+      "tokens_seen": 2801074176
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015266409886053602,
+      "loss": 1.3407,
+      "theoretical_loss": 3.366903581407099,
+      "tokens_seen": 2801205248
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015262397688974483,
+      "loss": 1.2888,
+      "theoretical_loss": 3.3668914033461554,
+      "tokens_seen": 2801336320
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015258385491895364,
+      "loss": 1.2888,
+      "theoretical_loss": 3.3668792260145346,
+      "tokens_seen": 2801467392
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015254373294816242,
+      "loss": 1.3701,
+      "theoretical_loss": 3.366867049412159,
+      "tokens_seen": 2801598464
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001525036109773712,
+      "loss": 1.3062,
+      "theoretical_loss": 3.36685487353895,
+      "tokens_seen": 2801729536
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015246348900658,
+      "loss": 1.3849,
+      "theoretical_loss": 3.366842698394831,
+      "tokens_seen": 2801860608
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001524233670357888,
+      "loss": 1.3005,
+      "theoretical_loss": 3.3668305239797234,
+      "tokens_seen": 2801991680
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015238324506499759,
+      "loss": 1.3216,
+      "theoretical_loss": 3.3668183502935496,
+      "tokens_seen": 2802122752
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001523431230942064,
+      "loss": 1.3062,
+      "theoretical_loss": 3.3668061773362323,
+      "tokens_seen": 2802253824
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001523030011234152,
+      "loss": 1.3049,
+      "theoretical_loss": 3.3667940051076934,
+      "tokens_seen": 2802384896
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015226287915262398,
+      "loss": 1.3854,
+      "theoretical_loss": 3.366781833607855,
+      "tokens_seen": 2802515968
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015222275718183278,
+      "loss": 1.3561,
+      "theoretical_loss": 3.36676966283664,
+      "tokens_seen": 2802647040
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015218263521104156,
+      "loss": 1.3113,
+      "theoretical_loss": 3.36675749279397,
+      "tokens_seen": 2802778112
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015214251324025037,
+      "loss": 1.3865,
+      "theoretical_loss": 3.3667453234797677,
+      "tokens_seen": 2802909184
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015210239126945915,
+      "loss": 1.2671,
+      "theoretical_loss": 3.3667331548939554,
+      "tokens_seen": 2803040256
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015206226929866796,
+      "loss": 1.3336,
+      "theoretical_loss": 3.3667209870364556,
+      "tokens_seen": 2803171328
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015202214732787676,
+      "loss": 1.2981,
+      "theoretical_loss": 3.3667088199071906,
+      "tokens_seen": 2803302400
+    },
+    {
+      "epoch": 0.7,
+      "objective/train/advantage_avg": 0.49035370349884033,
+      "objective/train/docs_used": 1018418,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2396246194839478,
+      "objective/train/original_loss": 1.2396245002746582,
+      "objective/train/theoretical_loss": 3.3666966535060827,
+      "objective/train/tokens_used": 1173892576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24127620458602905,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0502619743347168,
+      "objective/train/weighted_lm_loss": 1.3019874095916748,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3666966535060827,
+      "tokens_seen": 2803433472
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015198202535708554,
+      "loss": 1.3469,
+      "theoretical_loss": 3.3666966535060827,
+      "tokens_seen": 2803433472
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015194190338629435,
+      "loss": 1.2722,
+      "theoretical_loss": 3.366684487833054,
+      "tokens_seen": 2803564544
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015190178141550313,
+      "loss": 1.2433,
+      "theoretical_loss": 3.3666723228880273,
+      "tokens_seen": 2803695616
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015186165944471193,
+      "loss": 1.3571,
+      "theoretical_loss": 3.366660158670925,
+      "tokens_seen": 2803826688
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001518215374739207,
+      "loss": 1.3929,
+      "theoretical_loss": 3.366647995181669,
+      "tokens_seen": 2803957760
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015178141550312952,
+      "loss": 1.3336,
+      "theoretical_loss": 3.366635832420182,
+      "tokens_seen": 2804088832
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015174129353233832,
+      "loss": 1.3764,
+      "theoretical_loss": 3.366623670386387,
+      "tokens_seen": 2804219904
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001517011715615471,
+      "loss": 1.3269,
+      "theoretical_loss": 3.366611509080206,
+      "tokens_seen": 2804350976
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001516610495907559,
+      "loss": 1.3138,
+      "theoretical_loss": 3.3665993485015613,
+      "tokens_seen": 2804482048
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001516209276199647,
+      "loss": 1.3189,
+      "theoretical_loss": 3.3665871886503753,
+      "tokens_seen": 2804613120
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015158080564917347,
+      "loss": 1.2964,
+      "theoretical_loss": 3.366575029526571,
+      "tokens_seen": 2804744192
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015154068367838227,
+      "loss": 1.3024,
+      "theoretical_loss": 3.366562871130071,
+      "tokens_seen": 2804875264
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015150056170759108,
+      "loss": 1.3249,
+      "theoretical_loss": 3.3665507134607964,
+      "tokens_seen": 2805006336
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001514604397367999,
+      "loss": 1.338,
+      "theoretical_loss": 3.3665385565186714,
+      "tokens_seen": 2805137408
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015142031776600867,
+      "loss": 1.344,
+      "theoretical_loss": 3.366526400303618,
+      "tokens_seen": 2805268480
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015138019579521747,
+      "loss": 1.3798,
+      "theoretical_loss": 3.3665142448155585,
+      "tokens_seen": 2805399552
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015134007382442628,
+      "loss": 1.3148,
+      "theoretical_loss": 3.3665020900544156,
+      "tokens_seen": 2805530624
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015129995185363503,
+      "loss": 1.3278,
+      "theoretical_loss": 3.3664899360201117,
+      "tokens_seen": 2805661696
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015125982988284384,
+      "loss": 1.2961,
+      "theoretical_loss": 3.36647778271257,
+      "tokens_seen": 2805792768
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015121970791205264,
+      "loss": 1.338,
+      "theoretical_loss": 3.3664656301317124,
+      "tokens_seen": 2805923840
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015117958594126145,
+      "loss": 1.3331,
+      "theoretical_loss": 3.3664534782774616,
+      "tokens_seen": 2806054912
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015113946397047023,
+      "loss": 1.3557,
+      "theoretical_loss": 3.36644132714974,
+      "tokens_seen": 2806185984
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015109934199967904,
+      "loss": 1.3691,
+      "theoretical_loss": 3.3664291767484715,
+      "tokens_seen": 2806317056
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015105922002888784,
+      "loss": 1.2787,
+      "theoretical_loss": 3.3664170270735774,
+      "tokens_seen": 2806448128
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001510190980580966,
+      "loss": 1.2962,
+      "theoretical_loss": 3.366404878124981,
+      "tokens_seen": 2806579200
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001509789760873054,
+      "loss": 1.3255,
+      "theoretical_loss": 3.366392729902605,
+      "tokens_seen": 2806710272
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001509388541165142,
+      "loss": 1.3034,
+      "theoretical_loss": 3.366380582406371,
+      "tokens_seen": 2806841344
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.000150898732145723,
+      "loss": 1.2789,
+      "theoretical_loss": 3.366368435636203,
+      "tokens_seen": 2806972416
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001508586101749318,
+      "loss": 1.2903,
+      "theoretical_loss": 3.3663562895920234,
+      "tokens_seen": 2807103488
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001508184882041406,
+      "loss": 1.3094,
+      "theoretical_loss": 3.3663441442737545,
+      "tokens_seen": 2807234560
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001507783662333494,
+      "loss": 1.3154,
+      "theoretical_loss": 3.3663319996813197,
+      "tokens_seen": 2807365632
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015073824426255816,
+      "loss": 1.3695,
+      "theoretical_loss": 3.366319855814641,
+      "tokens_seen": 2807496704
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015069812229176696,
+      "loss": 1.4027,
+      "theoretical_loss": 3.3663077126736414,
+      "tokens_seen": 2807627776
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015065800032097577,
+      "loss": 1.3571,
+      "theoretical_loss": 3.3662955702582438,
+      "tokens_seen": 2807758848
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015061787835018455,
+      "loss": 1.3023,
+      "theoretical_loss": 3.366283428568371,
+      "tokens_seen": 2807889920
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015057775637939335,
+      "loss": 1.3076,
+      "theoretical_loss": 3.366271287603946,
+      "tokens_seen": 2808020992
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015053763440860216,
+      "loss": 1.3002,
+      "theoretical_loss": 3.3662591473648904,
+      "tokens_seen": 2808152064
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015049751243781097,
+      "loss": 1.34,
+      "theoretical_loss": 3.3662470078511286,
+      "tokens_seen": 2808283136
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015045739046701975,
+      "loss": 1.2968,
+      "theoretical_loss": 3.3662348690625823,
+      "tokens_seen": 2808414208
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015041726849622853,
+      "loss": 1.368,
+      "theoretical_loss": 3.366222730999175,
+      "tokens_seen": 2808545280
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015037714652543733,
+      "loss": 1.3651,
+      "theoretical_loss": 3.3662105936608295,
+      "tokens_seen": 2808676352
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001503370245546461,
+      "loss": 1.3423,
+      "theoretical_loss": 3.366198457047468,
+      "tokens_seen": 2808807424
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015029690258385492,
+      "loss": 1.344,
+      "theoretical_loss": 3.366186321159014,
+      "tokens_seen": 2808938496
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015025678061306372,
+      "loss": 1.2928,
+      "theoretical_loss": 3.3661741859953898,
+      "tokens_seen": 2809069568
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015021665864227253,
+      "loss": 1.2381,
+      "theoretical_loss": 3.366162051556519,
+      "tokens_seen": 2809200640
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001501765366714813,
+      "loss": 1.2963,
+      "theoretical_loss": 3.3661499178423244,
+      "tokens_seen": 2809331712
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001501364147006901,
+      "loss": 1.3954,
+      "theoretical_loss": 3.3661377848527287,
+      "tokens_seen": 2809462784
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001500962927298989,
+      "loss": 1.3487,
+      "theoretical_loss": 3.3661256525876544,
+      "tokens_seen": 2809593856
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015005617075910767,
+      "loss": 1.3365,
+      "theoretical_loss": 3.3661135210470254,
+      "tokens_seen": 2809724928
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00015001604878831648,
+      "loss": 1.2916,
+      "theoretical_loss": 3.3661013902307637,
+      "tokens_seen": 2809856000
+    },
+    {
+      "epoch": 0.7,
+      "objective/train/advantage_avg": 0.4916039705276489,
+      "objective/train/docs_used": 1020904,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3345210552215576,
+      "objective/train/original_loss": 1.3345211744308472,
+      "objective/train/theoretical_loss": 3.366089260138793,
+      "objective/train/tokens_used": 1180446176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2421063482761383,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503911972045898,
+      "objective/train/weighted_lm_loss": 1.4015673398971558,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.366089260138793,
+      "tokens_seen": 2809987072
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001499759268175253,
+      "loss": 1.3745,
+      "theoretical_loss": 3.366089260138793,
+      "tokens_seen": 2809987072
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001499358048467341,
+      "loss": 1.3297,
+      "theoretical_loss": 3.3660771307710355,
+      "tokens_seen": 2810118144
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00014989568287594287,
+      "loss": 1.335,
+      "theoretical_loss": 3.3660650021274146,
+      "tokens_seen": 2810249216
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00014985556090515168,
+      "loss": 1.2878,
+      "theoretical_loss": 3.366052874207854,
+      "tokens_seen": 2810380288
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00014981543893436046,
+      "loss": 1.2811,
+      "theoretical_loss": 3.3660407470122755,
+      "tokens_seen": 2810511360
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00014977531696356924,
+      "loss": 1.3756,
+      "theoretical_loss": 3.3660286205406025,
+      "tokens_seen": 2810642432
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00014973519499277804,
+      "loss": 1.3291,
+      "theoretical_loss": 3.3660164947927584,
+      "tokens_seen": 2810773504
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00014969507302198685,
+      "loss": 1.3193,
+      "theoretical_loss": 3.3660043697686666,
+      "tokens_seen": 2810904576
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00014965495105119563,
+      "loss": 1.3192,
+      "theoretical_loss": 3.3659922454682487,
+      "tokens_seen": 2811035648
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00014961482908040444,
+      "loss": 1.3898,
+      "theoretical_loss": 3.365980121891429,
+      "tokens_seen": 2811166720
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00014957470710961324,
+      "loss": 1.4056,
+      "theoretical_loss": 3.3659679990381304,
+      "tokens_seen": 2811297792
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00014953458513882202,
+      "loss": 1.3228,
+      "theoretical_loss": 3.365955876908276,
+      "tokens_seen": 2811428864
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001494944631680308,
+      "loss": 1.4332,
+      "theoretical_loss": 3.365943755501788,
+      "tokens_seen": 2811559936
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001494543411972396,
+      "loss": 1.299,
+      "theoretical_loss": 3.365931634818591,
+      "tokens_seen": 2811691008
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001494142192264484,
+      "loss": 1.2742,
+      "theoretical_loss": 3.3659195148586067,
+      "tokens_seen": 2811822080
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001493740972556572,
+      "loss": 1.3218,
+      "theoretical_loss": 3.3659073956217593,
+      "tokens_seen": 2811953152
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.000149333975284866,
+      "loss": 1.3135,
+      "theoretical_loss": 3.3658952771079718,
+      "tokens_seen": 2812084224
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001492938533140748,
+      "loss": 1.3086,
+      "theoretical_loss": 3.3658831593171668,
+      "tokens_seen": 2812215296
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00014925373134328358,
+      "loss": 1.3324,
+      "theoretical_loss": 3.3658710422492675,
+      "tokens_seen": 2812346368
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00014921360937249236,
+      "loss": 1.4174,
+      "theoretical_loss": 3.365858925904198,
+      "tokens_seen": 2812477440
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00014917348740170117,
+      "loss": 1.2963,
+      "theoretical_loss": 3.3658468102818806,
+      "tokens_seen": 2812608512
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00014913336543090998,
+      "loss": 1.4299,
+      "theoretical_loss": 3.365834695382239,
+      "tokens_seen": 2812739584
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00014909324346011875,
+      "loss": 1.3474,
+      "theoretical_loss": 3.3658225812051956,
+      "tokens_seen": 2812870656
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00014905312148932756,
+      "loss": 1.3625,
+      "theoretical_loss": 3.365810467750675,
+      "tokens_seen": 2813001728
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00014901299951853637,
+      "loss": 1.3058,
+      "theoretical_loss": 3.3657983550185993,
+      "tokens_seen": 2813132800
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014897287754774517,
+      "loss": 1.3483,
+      "theoretical_loss": 3.365786243008892,
+      "tokens_seen": 2813263872
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014893275557695393,
+      "loss": 1.3238,
+      "theoretical_loss": 3.3657741317214764,
+      "tokens_seen": 2813394944
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014889263360616273,
+      "loss": 1.3859,
+      "theoretical_loss": 3.3657620211562764,
+      "tokens_seen": 2813526016
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014885251163537154,
+      "loss": 1.3281,
+      "theoretical_loss": 3.3657499113132143,
+      "tokens_seen": 2813657088
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014881238966458032,
+      "loss": 1.2964,
+      "theoretical_loss": 3.365737802192214,
+      "tokens_seen": 2813788160
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014877226769378912,
+      "loss": 1.3337,
+      "theoretical_loss": 3.365725693793199,
+      "tokens_seen": 2813919232
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014873214572299793,
+      "loss": 1.2537,
+      "theoretical_loss": 3.365713586116092,
+      "tokens_seen": 2814050304
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014869202375220674,
+      "loss": 1.3494,
+      "theoretical_loss": 3.365701479160817,
+      "tokens_seen": 2814181376
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001486519017814155,
+      "loss": 1.3577,
+      "theoretical_loss": 3.3656893729272968,
+      "tokens_seen": 2814312448
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001486117798106243,
+      "loss": 1.2975,
+      "theoretical_loss": 3.365677267415455,
+      "tokens_seen": 2814443520
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001485716578398331,
+      "loss": 1.2778,
+      "theoretical_loss": 3.3656651626252145,
+      "tokens_seen": 2814574592
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014853153586904188,
+      "loss": 1.3289,
+      "theoretical_loss": 3.3656530585564997,
+      "tokens_seen": 2814705664
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001484914138982507,
+      "loss": 1.3774,
+      "theoretical_loss": 3.3656409552092335,
+      "tokens_seen": 2814836736
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001484512919274595,
+      "loss": 1.3054,
+      "theoretical_loss": 3.3656288525833387,
+      "tokens_seen": 2814967808
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014841116995666827,
+      "loss": 1.2546,
+      "theoretical_loss": 3.3656167506787393,
+      "tokens_seen": 2815098880
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014837104798587705,
+      "loss": 1.327,
+      "theoretical_loss": 3.365604649495359,
+      "tokens_seen": 2815229952
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014833092601508586,
+      "loss": 1.3679,
+      "theoretical_loss": 3.3655925490331207,
+      "tokens_seen": 2815361024
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014829080404429466,
+      "loss": 1.3617,
+      "theoretical_loss": 3.365580449291948,
+      "tokens_seen": 2815492096
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014825068207350344,
+      "loss": 1.4055,
+      "theoretical_loss": 3.365568350271765,
+      "tokens_seen": 2815623168
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014821056010271225,
+      "loss": 1.3288,
+      "theoretical_loss": 3.365556251972494,
+      "tokens_seen": 2815754240
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014817043813192106,
+      "loss": 1.4266,
+      "theoretical_loss": 3.365544154394059,
+      "tokens_seen": 2815885312
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014813031616112983,
+      "loss": 1.3289,
+      "theoretical_loss": 3.3655320575363845,
+      "tokens_seen": 2816016384
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014809019419033864,
+      "loss": 1.2303,
+      "theoretical_loss": 3.3655199613993925,
+      "tokens_seen": 2816147456
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014805007221954742,
+      "loss": 1.3065,
+      "theoretical_loss": 3.365507865983007,
+      "tokens_seen": 2816278528
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014800995024875623,
+      "loss": 1.2705,
+      "theoretical_loss": 3.365495771287152,
+      "tokens_seen": 2816409600
+    },
+    {
+      "epoch": 0.71,
+      "objective/train/advantage_avg": 0.4903794825077057,
+      "objective/train/docs_used": 1023297,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.344061255455017,
+      "objective/train/original_loss": 1.344061017036438,
+      "objective/train/theoretical_loss": 3.3654836773117505,
+      "objective/train/tokens_used": 1186999776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24107243120670319,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0502634048461914,
+      "objective/train/weighted_lm_loss": 1.411332607269287,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3654836773117505,
+      "tokens_seen": 2816540672
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.000147969828277965,
+      "loss": 1.4321,
+      "theoretical_loss": 3.3654836773117505,
+      "tokens_seen": 2816540672
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001479297063071738,
+      "loss": 1.3825,
+      "theoretical_loss": 3.3654715840567264,
+      "tokens_seen": 2816671744
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014788958433638262,
+      "loss": 1.3151,
+      "theoretical_loss": 3.365459491522003,
+      "tokens_seen": 2816802816
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001478494623655914,
+      "loss": 1.3457,
+      "theoretical_loss": 3.3654473997075045,
+      "tokens_seen": 2816933888
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001478093403948002,
+      "loss": 1.3497,
+      "theoretical_loss": 3.365435308613154,
+      "tokens_seen": 2817064960
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014776921842400898,
+      "loss": 1.2856,
+      "theoretical_loss": 3.365423218238875,
+      "tokens_seen": 2817196032
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001477290964532178,
+      "loss": 1.3434,
+      "theoretical_loss": 3.365411128584591,
+      "tokens_seen": 2817327104
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014768897448242657,
+      "loss": 1.2971,
+      "theoretical_loss": 3.3653990396502262,
+      "tokens_seen": 2817458176
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014764885251163538,
+      "loss": 1.3948,
+      "theoretical_loss": 3.3653869514357035,
+      "tokens_seen": 2817589248
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014760873054084418,
+      "loss": 1.2861,
+      "theoretical_loss": 3.3653748639409473,
+      "tokens_seen": 2817720320
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014756860857005296,
+      "loss": 1.3769,
+      "theoretical_loss": 3.365362777165881,
+      "tokens_seen": 2817851392
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014752848659926177,
+      "loss": 1.3864,
+      "theoretical_loss": 3.3653506911104283,
+      "tokens_seen": 2817982464
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014748836462847057,
+      "loss": 1.2867,
+      "theoretical_loss": 3.3653386057745127,
+      "tokens_seen": 2818113536
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014744824265767933,
+      "loss": 1.4026,
+      "theoretical_loss": 3.365326521158058,
+      "tokens_seen": 2818244608
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014740812068688813,
+      "loss": 1.379,
+      "theoretical_loss": 3.365314437260988,
+      "tokens_seen": 2818375680
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014736799871609694,
+      "loss": 1.3027,
+      "theoretical_loss": 3.3653023540832265,
+      "tokens_seen": 2818506752
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014732787674530574,
+      "loss": 1.2449,
+      "theoretical_loss": 3.365290271624697,
+      "tokens_seen": 2818637824
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014728775477451452,
+      "loss": 1.3653,
+      "theoretical_loss": 3.3652781898853235,
+      "tokens_seen": 2818768896
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014724763280372333,
+      "loss": 1.3211,
+      "theoretical_loss": 3.365266108865029,
+      "tokens_seen": 2818899968
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014720751083293214,
+      "loss": 1.3482,
+      "theoretical_loss": 3.3652540285637382,
+      "tokens_seen": 2819031040
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001471673888621409,
+      "loss": 1.3027,
+      "theoretical_loss": 3.3652419489813745,
+      "tokens_seen": 2819162112
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001471272668913497,
+      "loss": 1.35,
+      "theoretical_loss": 3.365229870117862,
+      "tokens_seen": 2819293184
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001470871449205585,
+      "loss": 1.4019,
+      "theoretical_loss": 3.365217791973124,
+      "tokens_seen": 2819424256
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001470470229497673,
+      "loss": 1.3005,
+      "theoretical_loss": 3.3652057145470846,
+      "tokens_seen": 2819555328
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014700690097897609,
+      "loss": 1.3375,
+      "theoretical_loss": 3.365193637839668,
+      "tokens_seen": 2819686400
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001469667790081849,
+      "loss": 1.3124,
+      "theoretical_loss": 3.3651815618507968,
+      "tokens_seen": 2819817472
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001469266570373937,
+      "loss": 1.3508,
+      "theoretical_loss": 3.3651694865803963,
+      "tokens_seen": 2819948544
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014688653506660245,
+      "loss": 1.2602,
+      "theoretical_loss": 3.3651574120283896,
+      "tokens_seen": 2820079616
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014684641309581126,
+      "loss": 1.3443,
+      "theoretical_loss": 3.3651453381947007,
+      "tokens_seen": 2820210688
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014680629112502006,
+      "loss": 1.3246,
+      "theoretical_loss": 3.365133265079253,
+      "tokens_seen": 2820341760
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014676616915422887,
+      "loss": 1.2714,
+      "theoretical_loss": 3.3651211926819715,
+      "tokens_seen": 2820472832
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014672604718343765,
+      "loss": 1.3345,
+      "theoretical_loss": 3.3651091210027793,
+      "tokens_seen": 2820603904
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014668592521264646,
+      "loss": 1.3652,
+      "theoretical_loss": 3.3650970500416006,
+      "tokens_seen": 2820734976
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014664580324185526,
+      "loss": 1.3309,
+      "theoretical_loss": 3.3650849797983593,
+      "tokens_seen": 2820866048
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014660568127106404,
+      "loss": 1.2919,
+      "theoretical_loss": 3.3650729102729793,
+      "tokens_seen": 2820997120
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014656555930027282,
+      "loss": 1.308,
+      "theoretical_loss": 3.365060841465384,
+      "tokens_seen": 2821128192
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014652543732948163,
+      "loss": 1.3403,
+      "theoretical_loss": 3.3650487733754986,
+      "tokens_seen": 2821259264
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001464853153586904,
+      "loss": 1.2691,
+      "theoretical_loss": 3.365036706003246,
+      "tokens_seen": 2821390336
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001464451933878992,
+      "loss": 1.3352,
+      "theoretical_loss": 3.365024639348551,
+      "tokens_seen": 2821521408
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014640507141710802,
+      "loss": 1.3592,
+      "theoretical_loss": 3.3650125734113367,
+      "tokens_seen": 2821652480
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014636494944631682,
+      "loss": 1.2617,
+      "theoretical_loss": 3.365000508191528,
+      "tokens_seen": 2821783552
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001463248274755256,
+      "loss": 1.3426,
+      "theoretical_loss": 3.3649884436890485,
+      "tokens_seen": 2821914624
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014628470550473438,
+      "loss": 1.3959,
+      "theoretical_loss": 3.364976379903822,
+      "tokens_seen": 2822045696
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001462445835339432,
+      "loss": 1.3395,
+      "theoretical_loss": 3.3649643168357732,
+      "tokens_seen": 2822176768
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014620446156315197,
+      "loss": 1.3318,
+      "theoretical_loss": 3.3649522544848254,
+      "tokens_seen": 2822307840
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014616433959236077,
+      "loss": 1.3558,
+      "theoretical_loss": 3.364940192850903,
+      "tokens_seen": 2822438912
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014612421762156958,
+      "loss": 1.3339,
+      "theoretical_loss": 3.3649281319339304,
+      "tokens_seen": 2822569984
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001460840956507784,
+      "loss": 1.3719,
+      "theoretical_loss": 3.3649160717338313,
+      "tokens_seen": 2822701056
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014604397367998717,
+      "loss": 1.2661,
+      "theoretical_loss": 3.36490401225053,
+      "tokens_seen": 2822832128
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014600385170919595,
+      "loss": 1.3449,
+      "theoretical_loss": 3.3648919534839505,
+      "tokens_seen": 2822963200
+    },
+    {
+      "epoch": 0.71,
+      "objective/train/advantage_avg": 0.49297645688056946,
+      "objective/train/docs_used": 1025771,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2796722650527954,
+      "objective/train/original_loss": 1.279672384262085,
+      "objective/train/theoretical_loss": 3.364879895434017,
+      "objective/train/tokens_used": 1193553376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2434253841638565,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505352020263672,
+      "objective/train/weighted_lm_loss": 1.3443247079849243,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.364879895434017,
+      "tokens_seen": 2823094272
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014596372973840475,
+      "loss": 1.3969,
+      "theoretical_loss": 3.364879895434017,
+      "tokens_seen": 2823094272
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014592360776761353,
+      "loss": 1.3942,
+      "theoretical_loss": 3.364867838100653,
+      "tokens_seen": 2823225344
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014588348579682234,
+      "loss": 1.3513,
+      "theoretical_loss": 3.3648557814837843,
+      "tokens_seen": 2823356416
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014584336382603114,
+      "loss": 1.3623,
+      "theoretical_loss": 3.3648437255833334,
+      "tokens_seen": 2823487488
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014580324185523995,
+      "loss": 1.3824,
+      "theoretical_loss": 3.3648316703992256,
+      "tokens_seen": 2823618560
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014576311988444873,
+      "loss": 1.3608,
+      "theoretical_loss": 3.364819615931384,
+      "tokens_seen": 2823749632
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014572299791365754,
+      "loss": 1.2506,
+      "theoretical_loss": 3.3648075621797338,
+      "tokens_seen": 2823880704
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014568287594286631,
+      "loss": 1.3224,
+      "theoretical_loss": 3.3647955091441983,
+      "tokens_seen": 2824011776
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001456427539720751,
+      "loss": 1.4048,
+      "theoretical_loss": 3.3647834568247026,
+      "tokens_seen": 2824142848
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001456026320012839,
+      "loss": 1.3519,
+      "theoretical_loss": 3.3647714052211706,
+      "tokens_seen": 2824273920
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001455625100304927,
+      "loss": 1.335,
+      "theoretical_loss": 3.3647593543335264,
+      "tokens_seen": 2824404992
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014552238805970149,
+      "loss": 1.2599,
+      "theoretical_loss": 3.364747304161694,
+      "tokens_seen": 2824536064
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001454822660889103,
+      "loss": 1.3334,
+      "theoretical_loss": 3.3647352547055984,
+      "tokens_seen": 2824667136
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001454421441181191,
+      "loss": 1.2393,
+      "theoretical_loss": 3.3647232059651633,
+      "tokens_seen": 2824798208
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014540202214732788,
+      "loss": 1.3648,
+      "theoretical_loss": 3.3647111579403135,
+      "tokens_seen": 2824929280
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014536190017653666,
+      "loss": 1.3242,
+      "theoretical_loss": 3.3646991106309727,
+      "tokens_seen": 2825060352
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014532177820574546,
+      "loss": 1.2754,
+      "theoretical_loss": 3.364687064037066,
+      "tokens_seen": 2825191424
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014528165623495427,
+      "loss": 1.2466,
+      "theoretical_loss": 3.3646750181585166,
+      "tokens_seen": 2825322496
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014524153426416305,
+      "loss": 1.369,
+      "theoretical_loss": 3.3646629729952497,
+      "tokens_seen": 2825453568
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014520141229337185,
+      "loss": 1.3328,
+      "theoretical_loss": 3.3646509285471895,
+      "tokens_seen": 2825584640
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014516129032258066,
+      "loss": 1.3755,
+      "theoretical_loss": 3.36463888481426,
+      "tokens_seen": 2825715712
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014512116835178947,
+      "loss": 1.2859,
+      "theoretical_loss": 3.364626841796386,
+      "tokens_seen": 2825846784
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014508104638099822,
+      "loss": 1.3719,
+      "theoretical_loss": 3.3646147994934914,
+      "tokens_seen": 2825977856
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014504092441020703,
+      "loss": 1.3,
+      "theoretical_loss": 3.3646027579055016,
+      "tokens_seen": 2826108928
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014500080243941583,
+      "loss": 1.2707,
+      "theoretical_loss": 3.36459071703234,
+      "tokens_seen": 2826240000
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001449606804686246,
+      "loss": 1.3068,
+      "theoretical_loss": 3.364578676873931,
+      "tokens_seen": 2826371072
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014492055849783342,
+      "loss": 1.2799,
+      "theoretical_loss": 3.3645666374301997,
+      "tokens_seen": 2826502144
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014488043652704222,
+      "loss": 1.3323,
+      "theoretical_loss": 3.36455459870107,
+      "tokens_seen": 2826633216
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014484031455625103,
+      "loss": 1.3771,
+      "theoretical_loss": 3.3645425606864667,
+      "tokens_seen": 2826764288
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014480019258545978,
+      "loss": 1.2912,
+      "theoretical_loss": 3.364530523386314,
+      "tokens_seen": 2826895360
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001447600706146686,
+      "loss": 1.2796,
+      "theoretical_loss": 3.364518486800537,
+      "tokens_seen": 2827026432
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001447199486438774,
+      "loss": 1.3109,
+      "theoretical_loss": 3.364506450929059,
+      "tokens_seen": 2827157504
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014467982667308617,
+      "loss": 1.3313,
+      "theoretical_loss": 3.364494415771806,
+      "tokens_seen": 2827288576
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014463970470229498,
+      "loss": 1.29,
+      "theoretical_loss": 3.364482381328701,
+      "tokens_seen": 2827419648
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001445995827315038,
+      "loss": 1.3176,
+      "theoretical_loss": 3.3644703475996693,
+      "tokens_seen": 2827550720
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001445594607607126,
+      "loss": 1.3662,
+      "theoretical_loss": 3.364458314584635,
+      "tokens_seen": 2827681792
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014451933878992135,
+      "loss": 1.2809,
+      "theoretical_loss": 3.3644462822835237,
+      "tokens_seen": 2827812864
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014447921681913015,
+      "loss": 1.4034,
+      "theoretical_loss": 3.364434250696259,
+      "tokens_seen": 2827943936
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014443909484833896,
+      "loss": 1.3384,
+      "theoretical_loss": 3.3644222198227656,
+      "tokens_seen": 2828075008
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014439897287754774,
+      "loss": 1.3472,
+      "theoretical_loss": 3.3644101896629683,
+      "tokens_seen": 2828206080
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014435885090675654,
+      "loss": 1.3694,
+      "theoretical_loss": 3.3643981602167914,
+      "tokens_seen": 2828337152
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014431872893596535,
+      "loss": 1.2828,
+      "theoretical_loss": 3.3643861314841597,
+      "tokens_seen": 2828468224
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014427860696517413,
+      "loss": 1.2765,
+      "theoretical_loss": 3.364374103464998,
+      "tokens_seen": 2828599296
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014423848499438294,
+      "loss": 1.2676,
+      "theoretical_loss": 3.3643620761592303,
+      "tokens_seen": 2828730368
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014419836302359171,
+      "loss": 1.3177,
+      "theoretical_loss": 3.3643500495667817,
+      "tokens_seen": 2828861440
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00014415824105280052,
+      "loss": 1.3523,
+      "theoretical_loss": 3.3643380236875773,
+      "tokens_seen": 2828992512
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001441181190820093,
+      "loss": 1.3755,
+      "theoretical_loss": 3.364325998521541,
+      "tokens_seen": 2829123584
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001440779971112181,
+      "loss": 1.3689,
+      "theoretical_loss": 3.3643139740685974,
+      "tokens_seen": 2829254656
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001440378751404269,
+      "loss": 1.3195,
+      "theoretical_loss": 3.364301950328672,
+      "tokens_seen": 2829385728
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001439977531696357,
+      "loss": 1.3129,
+      "theoretical_loss": 3.3642899273016886,
+      "tokens_seen": 2829516800
+    },
+    {
+      "epoch": 0.71,
+      "objective/train/advantage_avg": 0.4901696443557739,
+      "objective/train/docs_used": 1028281,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.233358383178711,
+      "objective/train/original_loss": 1.233358383178711,
+      "objective/train/theoretical_loss": 3.3642779049875724,
+      "objective/train/tokens_used": 1200106976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.241465762257576,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0502444505691528,
+      "objective/train/weighted_lm_loss": 1.2952136993408203,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3642779049875724,
+      "tokens_seen": 2829647872
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001439576311988445,
+      "loss": 1.3234,
+      "theoretical_loss": 3.3642779049875724,
+      "tokens_seen": 2829647872
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014391750922805328,
+      "loss": 1.262,
+      "theoretical_loss": 3.3642658833862478,
+      "tokens_seen": 2829778944
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014387738725726208,
+      "loss": 1.3441,
+      "theoretical_loss": 3.36425386249764,
+      "tokens_seen": 2829910016
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014383726528647086,
+      "loss": 1.3607,
+      "theoretical_loss": 3.3642418423216736,
+      "tokens_seen": 2830041088
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014379714331567967,
+      "loss": 1.3039,
+      "theoretical_loss": 3.364229822858273,
+      "tokens_seen": 2830172160
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014375702134488848,
+      "loss": 1.35,
+      "theoretical_loss": 3.364217804107363,
+      "tokens_seen": 2830303232
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014371689937409725,
+      "loss": 1.3724,
+      "theoretical_loss": 3.3642057860688688,
+      "tokens_seen": 2830434304
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014367677740330606,
+      "loss": 1.2618,
+      "theoretical_loss": 3.364193768742715,
+      "tokens_seen": 2830565376
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014363665543251484,
+      "loss": 1.2689,
+      "theoretical_loss": 3.364181752128826,
+      "tokens_seen": 2830696448
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014359653346172365,
+      "loss": 1.328,
+      "theoretical_loss": 3.3641697362271272,
+      "tokens_seen": 2830827520
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014355641149093243,
+      "loss": 1.2889,
+      "theoretical_loss": 3.364157721037543,
+      "tokens_seen": 2830958592
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014351628952014123,
+      "loss": 1.26,
+      "theoretical_loss": 3.3641457065599987,
+      "tokens_seen": 2831089664
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014347616754935004,
+      "loss": 1.4071,
+      "theoretical_loss": 3.3641336927944185,
+      "tokens_seen": 2831220736
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014343604557855882,
+      "loss": 1.3572,
+      "theoretical_loss": 3.3641216797407276,
+      "tokens_seen": 2831351808
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014339592360776762,
+      "loss": 1.3708,
+      "theoretical_loss": 3.364109667398851,
+      "tokens_seen": 2831482880
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014335580163697643,
+      "loss": 1.3474,
+      "theoretical_loss": 3.3640976557687132,
+      "tokens_seen": 2831613952
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014331567966618518,
+      "loss": 1.3931,
+      "theoretical_loss": 3.3640856448502396,
+      "tokens_seen": 2831745024
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.000143275557695394,
+      "loss": 1.3569,
+      "theoretical_loss": 3.3640736346433546,
+      "tokens_seen": 2831876096
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0001432354357246028,
+      "loss": 1.3063,
+      "theoretical_loss": 3.3640616251479836,
+      "tokens_seen": 2832007168
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0001431953137538116,
+      "loss": 1.3756,
+      "theoretical_loss": 3.364049616364051,
+      "tokens_seen": 2832138240
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014315519178302038,
+      "loss": 1.3725,
+      "theoretical_loss": 3.364037608291482,
+      "tokens_seen": 2832269312
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0001431150698122292,
+      "loss": 1.3186,
+      "theoretical_loss": 3.3640256009302014,
+      "tokens_seen": 2832400384
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.000143074947841438,
+      "loss": 1.3178,
+      "theoretical_loss": 3.3640135942801344,
+      "tokens_seen": 2832531456
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014303482587064675,
+      "loss": 1.3299,
+      "theoretical_loss": 3.3640015883412056,
+      "tokens_seen": 2832662528
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014299470389985555,
+      "loss": 1.3212,
+      "theoretical_loss": 3.3639895831133404,
+      "tokens_seen": 2832793600
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014295458192906436,
+      "loss": 1.3201,
+      "theoretical_loss": 3.3639775785964634,
+      "tokens_seen": 2832924672
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014291445995827316,
+      "loss": 1.358,
+      "theoretical_loss": 3.3639655747905,
+      "tokens_seen": 2833055744
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014287433798748194,
+      "loss": 1.3788,
+      "theoretical_loss": 3.363953571695375,
+      "tokens_seen": 2833186816
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014283421601669075,
+      "loss": 1.3094,
+      "theoretical_loss": 3.363941569311013,
+      "tokens_seen": 2833317888
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014279409404589956,
+      "loss": 1.3288,
+      "theoretical_loss": 3.36392956763734,
+      "tokens_seen": 2833448960
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014275397207510833,
+      "loss": 1.3087,
+      "theoretical_loss": 3.3639175666742807,
+      "tokens_seen": 2833580032
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014271385010431711,
+      "loss": 1.291,
+      "theoretical_loss": 3.363905566421759,
+      "tokens_seen": 2833711104
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014267372813352592,
+      "loss": 1.3904,
+      "theoretical_loss": 3.363893566879702,
+      "tokens_seen": 2833842176
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014263360616273473,
+      "loss": 1.3208,
+      "theoretical_loss": 3.363881568048033,
+      "tokens_seen": 2833973248
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0001425934841919435,
+      "loss": 1.371,
+      "theoretical_loss": 3.3638695699266785,
+      "tokens_seen": 2834104320
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0001425533622211523,
+      "loss": 1.3515,
+      "theoretical_loss": 3.3638575725155624,
+      "tokens_seen": 2834235392
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014251324025036112,
+      "loss": 1.3806,
+      "theoretical_loss": 3.36384557581461,
+      "tokens_seen": 2834366464
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0001424731182795699,
+      "loss": 1.4087,
+      "theoretical_loss": 3.3638335798237473,
+      "tokens_seen": 2834497536
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014243299630877868,
+      "loss": 1.3507,
+      "theoretical_loss": 3.3638215845428987,
+      "tokens_seen": 2834628608
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014239287433798748,
+      "loss": 1.3653,
+      "theoretical_loss": 3.3638095899719898,
+      "tokens_seen": 2834759680
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014235275236719626,
+      "loss": 1.3832,
+      "theoretical_loss": 3.363797596110945,
+      "tokens_seen": 2834890752
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014231263039640507,
+      "loss": 1.3337,
+      "theoretical_loss": 3.36378560295969,
+      "tokens_seen": 2835021824
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014227250842561388,
+      "loss": 1.3902,
+      "theoretical_loss": 3.3637736105181504,
+      "tokens_seen": 2835152896
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014223238645482268,
+      "loss": 1.3363,
+      "theoretical_loss": 3.363761618786251,
+      "tokens_seen": 2835283968
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014219226448403146,
+      "loss": 1.3563,
+      "theoretical_loss": 3.3637496277639163,
+      "tokens_seen": 2835415040
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014215214251324024,
+      "loss": 1.3131,
+      "theoretical_loss": 3.363737637451073,
+      "tokens_seen": 2835546112
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014211202054244905,
+      "loss": 1.3937,
+      "theoretical_loss": 3.3637256478476445,
+      "tokens_seen": 2835677184
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014207189857165783,
+      "loss": 1.251,
+      "theoretical_loss": 3.3637136589535577,
+      "tokens_seen": 2835808256
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014203177660086663,
+      "loss": 1.3582,
+      "theoretical_loss": 3.363701670768737,
+      "tokens_seen": 2835939328
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014199165463007544,
+      "loss": 1.3251,
+      "theoretical_loss": 3.3636896832931074,
+      "tokens_seen": 2836070400
+    },
+    {
+      "epoch": 0.72,
+      "objective/train/advantage_avg": 0.4904578924179077,
+      "objective/train/docs_used": 1030811,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.4069328308105469,
+      "objective/train/original_loss": 1.4069327116012573,
+      "objective/train/theoretical_loss": 3.3636776965265955,
+      "objective/train/tokens_used": 1206660576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24115115404129028,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.05027174949646,
+      "objective/train/weighted_lm_loss": 1.477555751800537,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9048373699188232,
+      "theoretical_loss": 3.3636776965265955,
+      "tokens_seen": 2836201472
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014195153265928424,
+      "loss": 1.3841,
+      "theoretical_loss": 3.3636776965265955,
+      "tokens_seen": 2836201472
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014191141068849302,
+      "loss": 1.2899,
+      "theoretical_loss": 3.363665710469125,
+      "tokens_seen": 2836332544
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014187128871770183,
+      "loss": 1.3094,
+      "theoretical_loss": 3.363653725120622,
+      "tokens_seen": 2836463616
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0001418311667469106,
+      "loss": 1.2982,
+      "theoretical_loss": 3.363641740481012,
+      "tokens_seen": 2836594688
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0001417910447761194,
+      "loss": 1.3837,
+      "theoretical_loss": 3.36362975655022,
+      "tokens_seen": 2836725760
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0001417509228053282,
+      "loss": 1.4204,
+      "theoretical_loss": 3.363617773328171,
+      "tokens_seen": 2836856832
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.000141710800834537,
+      "loss": 1.3429,
+      "theoretical_loss": 3.3636057908147907,
+      "tokens_seen": 2836987904
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0001416706788637458,
+      "loss": 1.3735,
+      "theoretical_loss": 3.3635938090100046,
+      "tokens_seen": 2837118976
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014163055689295459,
+      "loss": 1.2873,
+      "theoretical_loss": 3.363581827913738,
+      "tokens_seen": 2837250048
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0001415904349221634,
+      "loss": 1.3787,
+      "theoretical_loss": 3.363569847525916,
+      "tokens_seen": 2837381120
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014155031295137217,
+      "loss": 1.2844,
+      "theoretical_loss": 3.363557867846464,
+      "tokens_seen": 2837512192
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014151019098058095,
+      "loss": 1.3466,
+      "theoretical_loss": 3.363545888875308,
+      "tokens_seen": 2837643264
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014147006900978976,
+      "loss": 1.2862,
+      "theoretical_loss": 3.3635339106123725,
+      "tokens_seen": 2837774336
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014142994703899856,
+      "loss": 1.4085,
+      "theoretical_loss": 3.3635219330575836,
+      "tokens_seen": 2837905408
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014138982506820734,
+      "loss": 1.3629,
+      "theoretical_loss": 3.3635099562108666,
+      "tokens_seen": 2838036480
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014134970309741615,
+      "loss": 1.342,
+      "theoretical_loss": 3.363497980072147,
+      "tokens_seen": 2838167552
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014130958112662496,
+      "loss": 1.2388,
+      "theoretical_loss": 3.36348600464135,
+      "tokens_seen": 2838298624
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014126945915583373,
+      "loss": 1.301,
+      "theoretical_loss": 3.363474029918401,
+      "tokens_seen": 2838429696
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014122933718504251,
+      "loss": 1.3191,
+      "theoretical_loss": 3.363462055903226,
+      "tokens_seen": 2838560768
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014118921521425132,
+      "loss": 1.28,
+      "theoretical_loss": 3.3634500825957496,
+      "tokens_seen": 2838691840
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014114909324346013,
+      "loss": 1.2699,
+      "theoretical_loss": 3.363438109995898,
+      "tokens_seen": 2838822912
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0001411089712726689,
+      "loss": 1.2484,
+      "theoretical_loss": 3.363426138103597,
+      "tokens_seen": 2838953984
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0001410688493018777,
+      "loss": 1.3258,
+      "theoretical_loss": 3.3634141669187714,
+      "tokens_seen": 2839085056
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014102872733108652,
+      "loss": 1.3675,
+      "theoretical_loss": 3.363402196441347,
+      "tokens_seen": 2839216128
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014098860536029532,
+      "loss": 1.3061,
+      "theoretical_loss": 3.363390226671249,
+      "tokens_seen": 2839347200
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014094848338950408,
+      "loss": 1.351,
+      "theoretical_loss": 3.363378257608404,
+      "tokens_seen": 2839478272
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014090836141871288,
+      "loss": 1.3488,
+      "theoretical_loss": 3.3633662892527365,
+      "tokens_seen": 2839609344
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0001408682394479217,
+      "loss": 1.2763,
+      "theoretical_loss": 3.3633543216041724,
+      "tokens_seen": 2839740416
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014082811747713047,
+      "loss": 1.3164,
+      "theoretical_loss": 3.3633423546626373,
+      "tokens_seen": 2839871488
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014078799550633927,
+      "loss": 1.263,
+      "theoretical_loss": 3.363330388428057,
+      "tokens_seen": 2840002560
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014074787353554808,
+      "loss": 1.4178,
+      "theoretical_loss": 3.363318422900357,
+      "tokens_seen": 2840133632
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0001407077515647569,
+      "loss": 1.3228,
+      "theoretical_loss": 3.363306458079463,
+      "tokens_seen": 2840264704
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014066762959396564,
+      "loss": 1.3786,
+      "theoretical_loss": 3.3632944939653,
+      "tokens_seen": 2840395776
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014062750762317445,
+      "loss": 1.2681,
+      "theoretical_loss": 3.3632825305577945,
+      "tokens_seen": 2840526848
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014058738565238325,
+      "loss": 1.3103,
+      "theoretical_loss": 3.3632705678568717,
+      "tokens_seen": 2840657920
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014054726368159203,
+      "loss": 1.3524,
+      "theoretical_loss": 3.363258605862457,
+      "tokens_seen": 2840788992
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014050714171080084,
+      "loss": 1.2814,
+      "theoretical_loss": 3.3632466445744775,
+      "tokens_seen": 2840920064
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014046701974000964,
+      "loss": 1.2937,
+      "theoretical_loss": 3.363234683992857,
+      "tokens_seen": 2841051136
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014042689776921845,
+      "loss": 1.3769,
+      "theoretical_loss": 3.3632227241175223,
+      "tokens_seen": 2841182208
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014038677579842723,
+      "loss": 1.3758,
+      "theoretical_loss": 3.363210764948399,
+      "tokens_seen": 2841313280
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.000140346653827636,
+      "loss": 1.368,
+      "theoretical_loss": 3.3631988064854124,
+      "tokens_seen": 2841444352
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014030653185684481,
+      "loss": 1.3819,
+      "theoretical_loss": 3.3631868487284886,
+      "tokens_seen": 2841575424
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0001402664098860536,
+      "loss": 1.2544,
+      "theoretical_loss": 3.3631748916775535,
+      "tokens_seen": 2841706496
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0001402262879152624,
+      "loss": 1.3449,
+      "theoretical_loss": 3.3631629353325323,
+      "tokens_seen": 2841837568
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0001401861659444712,
+      "loss": 1.3099,
+      "theoretical_loss": 3.3631509796933514,
+      "tokens_seen": 2841968640
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014014604397367999,
+      "loss": 1.3069,
+      "theoretical_loss": 3.3631390247599358,
+      "tokens_seen": 2842099712
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0001401059220028888,
+      "loss": 1.3861,
+      "theoretical_loss": 3.3631270705322125,
+      "tokens_seen": 2842230784
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014006580003209757,
+      "loss": 1.3373,
+      "theoretical_loss": 3.363115117010106,
+      "tokens_seen": 2842361856
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00014002567806130638,
+      "loss": 1.407,
+      "theoretical_loss": 3.363103164193543,
+      "tokens_seen": 2842492928
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00013998555609051516,
+      "loss": 1.3215,
+      "theoretical_loss": 3.3630912120824488,
+      "tokens_seen": 2842624000
+    },
+    {
+      "epoch": 0.72,
+      "objective/train/advantage_avg": 0.49206626415252686,
+      "objective/train/docs_used": 1033097,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2449837923049927,
+      "objective/train/original_loss": 1.2449836730957031,
+      "objective/train/theoretical_loss": 3.3630792606767494,
+      "objective/train/tokens_used": 1213214176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24266310036182404,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504401922225952,
+      "objective/train/weighted_lm_loss": 1.3077586889266968,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3630792606767494,
+      "tokens_seen": 2842755072
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00013994543411972396,
+      "loss": 1.3006,
+      "theoretical_loss": 3.3630792606767494,
+      "tokens_seen": 2842755072
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00013990531214893277,
+      "loss": 1.3328,
+      "theoretical_loss": 3.363067309976371,
+      "tokens_seen": 2842886144
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00013986519017814155,
+      "loss": 1.339,
+      "theoretical_loss": 3.3630553599812387,
+      "tokens_seen": 2843017216
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00013982506820735036,
+      "loss": 1.294,
+      "theoretical_loss": 3.3630434106912794,
+      "tokens_seen": 2843148288
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00013978494623655913,
+      "loss": 1.3346,
+      "theoretical_loss": 3.363031462106418,
+      "tokens_seen": 2843279360
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00013974482426576794,
+      "loss": 1.2914,
+      "theoretical_loss": 3.363019514226581,
+      "tokens_seen": 2843410432
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00013970470229497672,
+      "loss": 1.3262,
+      "theoretical_loss": 3.3630075670516937,
+      "tokens_seen": 2843541504
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00013966458032418553,
+      "loss": 1.345,
+      "theoretical_loss": 3.3629956205816827,
+      "tokens_seen": 2843672576
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00013962445835339433,
+      "loss": 1.293,
+      "theoretical_loss": 3.3629836748164736,
+      "tokens_seen": 2843803648
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0001395843363826031,
+      "loss": 1.263,
+      "theoretical_loss": 3.3629717297559925,
+      "tokens_seen": 2843934720
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00013954421441181192,
+      "loss": 1.3387,
+      "theoretical_loss": 3.3629597854001654,
+      "tokens_seen": 2844065792
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00013950409244102072,
+      "loss": 1.3286,
+      "theoretical_loss": 3.362947841748918,
+      "tokens_seen": 2844196864
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00013946397047022948,
+      "loss": 1.3374,
+      "theoretical_loss": 3.362935898802176,
+      "tokens_seen": 2844327936
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00013942384849943828,
+      "loss": 1.3518,
+      "theoretical_loss": 3.362923956559866,
+      "tokens_seen": 2844459008
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0001393837265286471,
+      "loss": 1.2858,
+      "theoretical_loss": 3.3629120150219136,
+      "tokens_seen": 2844590080
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0001393436045578559,
+      "loss": 1.3555,
+      "theoretical_loss": 3.362900074188245,
+      "tokens_seen": 2844721152
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00013930348258706467,
+      "loss": 1.2746,
+      "theoretical_loss": 3.362888134058786,
+      "tokens_seen": 2844852224
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00013926336061627348,
+      "loss": 1.3121,
+      "theoretical_loss": 3.362876194633463,
+      "tokens_seen": 2844983296
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0001392232386454823,
+      "loss": 1.3501,
+      "theoretical_loss": 3.3628642559122017,
+      "tokens_seen": 2845114368
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00013918311667469104,
+      "loss": 1.3405,
+      "theoretical_loss": 3.3628523178949283,
+      "tokens_seen": 2845245440
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00013914299470389985,
+      "loss": 1.2903,
+      "theoretical_loss": 3.362840380581569,
+      "tokens_seen": 2845376512
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00013910287273310865,
+      "loss": 1.3169,
+      "theoretical_loss": 3.3628284439720497,
+      "tokens_seen": 2845507584
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00013906275076231746,
+      "loss": 1.3264,
+      "theoretical_loss": 3.362816508066296,
+      "tokens_seen": 2845638656
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00013902262879152624,
+      "loss": 1.3115,
+      "theoretical_loss": 3.3628045728642353,
+      "tokens_seen": 2845769728
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00013898250682073504,
+      "loss": 1.3093,
+      "theoretical_loss": 3.3627926383657925,
+      "tokens_seen": 2845900800
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00013894238484994385,
+      "loss": 1.3694,
+      "theoretical_loss": 3.362780704570894,
+      "tokens_seen": 2846031872
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0001389022628791526,
+      "loss": 1.3251,
+      "theoretical_loss": 3.362768771479466,
+      "tokens_seen": 2846162944
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001388621409083614,
+      "loss": 1.2558,
+      "theoretical_loss": 3.3627568390914346,
+      "tokens_seen": 2846294016
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013882201893757021,
+      "loss": 1.3397,
+      "theoretical_loss": 3.362744907406726,
+      "tokens_seen": 2846425088
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013878189696677902,
+      "loss": 1.2567,
+      "theoretical_loss": 3.3627329764252667,
+      "tokens_seen": 2846556160
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001387417749959878,
+      "loss": 1.36,
+      "theoretical_loss": 3.3627210461469823,
+      "tokens_seen": 2846687232
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001387016530251966,
+      "loss": 1.3453,
+      "theoretical_loss": 3.362709116571799,
+      "tokens_seen": 2846818304
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001386615310544054,
+      "loss": 1.3109,
+      "theoretical_loss": 3.3626971876996437,
+      "tokens_seen": 2846949376
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001386214090836142,
+      "loss": 1.3157,
+      "theoretical_loss": 3.362685259530442,
+      "tokens_seen": 2847080448
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013858128711282297,
+      "loss": 1.2651,
+      "theoretical_loss": 3.3626733320641202,
+      "tokens_seen": 2847211520
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013854116514203178,
+      "loss": 1.3451,
+      "theoretical_loss": 3.3626614053006043,
+      "tokens_seen": 2847342592
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013850104317124058,
+      "loss": 1.3693,
+      "theoretical_loss": 3.3626494792398214,
+      "tokens_seen": 2847473664
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013846092120044936,
+      "loss": 1.3192,
+      "theoretical_loss": 3.3626375538816964,
+      "tokens_seen": 2847604736
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013842079922965817,
+      "loss": 1.2968,
+      "theoretical_loss": 3.362625629226157,
+      "tokens_seen": 2847735808
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013838067725886698,
+      "loss": 1.2782,
+      "theoretical_loss": 3.3626137052731284,
+      "tokens_seen": 2847866880
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013834055528807575,
+      "loss": 1.3017,
+      "theoretical_loss": 3.3626017820225376,
+      "tokens_seen": 2847997952
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013830043331728453,
+      "loss": 1.3102,
+      "theoretical_loss": 3.36258985947431,
+      "tokens_seen": 2848129024
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013826031134649334,
+      "loss": 1.2344,
+      "theoretical_loss": 3.362577937628373,
+      "tokens_seen": 2848260096
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013822018937570212,
+      "loss": 1.2744,
+      "theoretical_loss": 3.3625660164846525,
+      "tokens_seen": 2848391168
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013818006740491093,
+      "loss": 1.3384,
+      "theoretical_loss": 3.362554096043074,
+      "tokens_seen": 2848522240
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013813994543411973,
+      "loss": 1.377,
+      "theoretical_loss": 3.3625421763035654,
+      "tokens_seen": 2848653312
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013809982346332854,
+      "loss": 1.3627,
+      "theoretical_loss": 3.3625302572660516,
+      "tokens_seen": 2848784384
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013805970149253732,
+      "loss": 1.3662,
+      "theoretical_loss": 3.3625183389304603,
+      "tokens_seen": 2848915456
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013801957952174612,
+      "loss": 1.3036,
+      "theoretical_loss": 3.3625064212967164,
+      "tokens_seen": 2849046528
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001379794575509549,
+      "loss": 1.3166,
+      "theoretical_loss": 3.362494504364747,
+      "tokens_seen": 2849177600
+    },
+    {
+      "epoch": 0.73,
+      "objective/train/advantage_avg": 0.49387094378471375,
+      "objective/train/docs_used": 1035566,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.207085371017456,
+      "objective/train/original_loss": 1.2070854902267456,
+      "objective/train/theoretical_loss": 3.3624825881344793,
+      "objective/train/tokens_used": 1219767776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2443949431180954,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506294965744019,
+      "objective/train/weighted_lm_loss": 1.2680890560150146,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.7788007855415344,
+      "theoretical_loss": 3.3624825881344793,
+      "tokens_seen": 2849308672
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013793933558016368,
+      "loss": 1.3098,
+      "theoretical_loss": 3.3624825881344793,
+      "tokens_seen": 2849308672
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001378992136093725,
+      "loss": 1.336,
+      "theoretical_loss": 3.362470672605838,
+      "tokens_seen": 2849439744
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001378590916385813,
+      "loss": 1.3432,
+      "theoretical_loss": 3.3624587577787515,
+      "tokens_seen": 2849570816
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001378189696677901,
+      "loss": 1.3233,
+      "theoretical_loss": 3.3624468436531445,
+      "tokens_seen": 2849701888
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013777884769699888,
+      "loss": 1.2805,
+      "theoretical_loss": 3.362434930228944,
+      "tokens_seen": 2849832960
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001377387257262077,
+      "loss": 1.4097,
+      "theoretical_loss": 3.3624230175060768,
+      "tokens_seen": 2849964032
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013769860375541647,
+      "loss": 1.2966,
+      "theoretical_loss": 3.362411105484469,
+      "tokens_seen": 2850095104
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013765848178462525,
+      "loss": 1.3143,
+      "theoretical_loss": 3.3623991941640474,
+      "tokens_seen": 2850226176
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013761835981383405,
+      "loss": 1.2815,
+      "theoretical_loss": 3.362387283544738,
+      "tokens_seen": 2850357248
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013757823784304286,
+      "loss": 1.3562,
+      "theoretical_loss": 3.3623753736264677,
+      "tokens_seen": 2850488320
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013753811587225166,
+      "loss": 1.3587,
+      "theoretical_loss": 3.362363464409163,
+      "tokens_seen": 2850619392
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013749799390146044,
+      "loss": 1.3627,
+      "theoretical_loss": 3.3623515558927504,
+      "tokens_seen": 2850750464
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013745787193066925,
+      "loss": 1.3708,
+      "theoretical_loss": 3.362339648077156,
+      "tokens_seen": 2850881536
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013741774995987803,
+      "loss": 1.3142,
+      "theoretical_loss": 3.3623277409623067,
+      "tokens_seen": 2851012608
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001373776279890868,
+      "loss": 1.2578,
+      "theoretical_loss": 3.362315834548129,
+      "tokens_seen": 2851143680
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013733750601829561,
+      "loss": 1.3506,
+      "theoretical_loss": 3.36230392883455,
+      "tokens_seen": 2851274752
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013729738404750442,
+      "loss": 1.3858,
+      "theoretical_loss": 3.362292023821495,
+      "tokens_seen": 2851405824
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001372572620767132,
+      "loss": 1.4081,
+      "theoretical_loss": 3.362280119508892,
+      "tokens_seen": 2851536896
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.000137217140105922,
+      "loss": 1.3654,
+      "theoretical_loss": 3.362268215896666,
+      "tokens_seen": 2851667968
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001371770181351308,
+      "loss": 1.2625,
+      "theoretical_loss": 3.3622563129847456,
+      "tokens_seen": 2851799040
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013713689616433962,
+      "loss": 1.3037,
+      "theoretical_loss": 3.3622444107730556,
+      "tokens_seen": 2851930112
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013709677419354837,
+      "loss": 1.3438,
+      "theoretical_loss": 3.3622325092615237,
+      "tokens_seen": 2852061184
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013705665222275718,
+      "loss": 1.396,
+      "theoretical_loss": 3.3622206084500763,
+      "tokens_seen": 2852192256
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013701653025196598,
+      "loss": 1.2664,
+      "theoretical_loss": 3.3622087083386396,
+      "tokens_seen": 2852323328
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013697640828117476,
+      "loss": 1.2769,
+      "theoretical_loss": 3.362196808927141,
+      "tokens_seen": 2852454400
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013693628631038357,
+      "loss": 1.3167,
+      "theoretical_loss": 3.3621849102155066,
+      "tokens_seen": 2852585472
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013689616433959238,
+      "loss": 1.3262,
+      "theoretical_loss": 3.3621730122036633,
+      "tokens_seen": 2852716544
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013685604236880118,
+      "loss": 1.3718,
+      "theoretical_loss": 3.3621611148915376,
+      "tokens_seen": 2852847616
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013681592039800993,
+      "loss": 1.3746,
+      "theoretical_loss": 3.3621492182790567,
+      "tokens_seen": 2852978688
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013677579842721874,
+      "loss": 1.3645,
+      "theoretical_loss": 3.362137322366147,
+      "tokens_seen": 2853109760
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013673567645642755,
+      "loss": 1.3064,
+      "theoretical_loss": 3.3621254271527348,
+      "tokens_seen": 2853240832
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013669555448563633,
+      "loss": 1.2932,
+      "theoretical_loss": 3.3621135326387472,
+      "tokens_seen": 2853371904
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013665543251484513,
+      "loss": 1.3069,
+      "theoretical_loss": 3.3621016388241114,
+      "tokens_seen": 2853502976
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013661531054405394,
+      "loss": 1.3345,
+      "theoretical_loss": 3.3620897457087535,
+      "tokens_seen": 2853634048
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013657518857326274,
+      "loss": 1.2829,
+      "theoretical_loss": 3.362077853292601,
+      "tokens_seen": 2853765120
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001365350666024715,
+      "loss": 1.2898,
+      "theoretical_loss": 3.3620659615755795,
+      "tokens_seen": 2853896192
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001364949446316803,
+      "loss": 1.2031,
+      "theoretical_loss": 3.362054070557617,
+      "tokens_seen": 2854027264
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001364548226608891,
+      "loss": 1.3487,
+      "theoretical_loss": 3.3620421802386398,
+      "tokens_seen": 2854158336
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001364147006900979,
+      "loss": 1.3549,
+      "theoretical_loss": 3.3620302906185744,
+      "tokens_seen": 2854289408
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001363745787193067,
+      "loss": 1.3358,
+      "theoretical_loss": 3.3620184016973482,
+      "tokens_seen": 2854420480
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001363344567485155,
+      "loss": 1.3763,
+      "theoretical_loss": 3.362006513474888,
+      "tokens_seen": 2854551552
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001362943347777243,
+      "loss": 1.3963,
+      "theoretical_loss": 3.36199462595112,
+      "tokens_seen": 2854682624
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013625421280693309,
+      "loss": 1.2425,
+      "theoretical_loss": 3.361982739125972,
+      "tokens_seen": 2854813696
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013621409083614187,
+      "loss": 1.2924,
+      "theoretical_loss": 3.3619708529993697,
+      "tokens_seen": 2854944768
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013617396886535067,
+      "loss": 1.3383,
+      "theoretical_loss": 3.3619589675712414,
+      "tokens_seen": 2855075840
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013613384689455945,
+      "loss": 1.3282,
+      "theoretical_loss": 3.3619470828415126,
+      "tokens_seen": 2855206912
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013609372492376826,
+      "loss": 1.3424,
+      "theoretical_loss": 3.361935198810111,
+      "tokens_seen": 2855337984
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013605360295297706,
+      "loss": 1.3501,
+      "theoretical_loss": 3.3619233154769637,
+      "tokens_seen": 2855469056
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013601348098218584,
+      "loss": 1.4069,
+      "theoretical_loss": 3.361911432841997,
+      "tokens_seen": 2855600128
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013597335901139465,
+      "loss": 1.298,
+      "theoretical_loss": 3.361899550905138,
+      "tokens_seen": 2855731200
+    },
+    {
+      "epoch": 0.73,
+      "objective/train/advantage_avg": 0.4928230941295624,
+      "objective/train/docs_used": 1037860,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3380557298660278,
+      "objective/train/original_loss": 1.3380557298660278,
+      "objective/train/theoretical_loss": 3.361887669666314,
+      "objective/train/tokens_used": 1226321376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2434956133365631,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505201816558838,
+      "objective/train/weighted_lm_loss": 1.4055039882659912,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.361887669666314,
+      "tokens_seen": 2855862272
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013593323704060343,
+      "loss": 1.3532,
+      "theoretical_loss": 3.361887669666314,
+      "tokens_seen": 2855862272
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013589311506981223,
+      "loss": 1.3617,
+      "theoretical_loss": 3.3618757891254516,
+      "tokens_seen": 2855993344
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013585299309902101,
+      "loss": 1.4015,
+      "theoretical_loss": 3.361863909282478,
+      "tokens_seen": 2856124416
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013581287112822982,
+      "loss": 1.3602,
+      "theoretical_loss": 3.36185203013732,
+      "tokens_seen": 2856255488
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013577274915743863,
+      "loss": 1.3516,
+      "theoretical_loss": 3.3618401516899046,
+      "tokens_seen": 2856386560
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001357326271866474,
+      "loss": 1.3427,
+      "theoretical_loss": 3.3618282739401586,
+      "tokens_seen": 2856517632
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001356925052158562,
+      "loss": 1.3412,
+      "theoretical_loss": 3.3618163968880097,
+      "tokens_seen": 2856648704
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013565238324506502,
+      "loss": 1.3446,
+      "theoretical_loss": 3.3618045205333846,
+      "tokens_seen": 2856779776
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001356122612742738,
+      "loss": 1.3193,
+      "theoretical_loss": 3.36179264487621,
+      "tokens_seen": 2856910848
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013557213930348258,
+      "loss": 1.3983,
+      "theoretical_loss": 3.3617807699164133,
+      "tokens_seen": 2857041920
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013553201733269138,
+      "loss": 1.3956,
+      "theoretical_loss": 3.3617688956539213,
+      "tokens_seen": 2857172992
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001354918953619002,
+      "loss": 1.3744,
+      "theoretical_loss": 3.3617570220886615,
+      "tokens_seen": 2857304064
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013545177339110897,
+      "loss": 1.3192,
+      "theoretical_loss": 3.3617451492205603,
+      "tokens_seen": 2857435136
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013541165142031777,
+      "loss": 1.2655,
+      "theoretical_loss": 3.361733277049545,
+      "tokens_seen": 2857566208
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013537152944952658,
+      "loss": 1.2602,
+      "theoretical_loss": 3.3617214055755436,
+      "tokens_seen": 2857697280
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013533140747873533,
+      "loss": 1.3295,
+      "theoretical_loss": 3.3617095347984822,
+      "tokens_seen": 2857828352
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013529128550794414,
+      "loss": 1.2927,
+      "theoretical_loss": 3.361697664718288,
+      "tokens_seen": 2857959424
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013525116353715295,
+      "loss": 1.345,
+      "theoretical_loss": 3.361685795334888,
+      "tokens_seen": 2858090496
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013521104156636175,
+      "loss": 1.2704,
+      "theoretical_loss": 3.3616739266482107,
+      "tokens_seen": 2858221568
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013517091959557053,
+      "loss": 1.2604,
+      "theoretical_loss": 3.3616620586581814,
+      "tokens_seen": 2858352640
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013513079762477934,
+      "loss": 1.3211,
+      "theoretical_loss": 3.3616501913647285,
+      "tokens_seen": 2858483712
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013509067565398814,
+      "loss": 1.3656,
+      "theoretical_loss": 3.361638324767779,
+      "tokens_seen": 2858614784
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001350505536831969,
+      "loss": 1.3091,
+      "theoretical_loss": 3.3616264588672595,
+      "tokens_seen": 2858745856
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001350104317124057,
+      "loss": 1.2957,
+      "theoretical_loss": 3.3616145936630977,
+      "tokens_seen": 2858876928
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001349703097416145,
+      "loss": 1.3079,
+      "theoretical_loss": 3.3616027291552206,
+      "tokens_seen": 2859008000
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013493018777082331,
+      "loss": 1.2929,
+      "theoretical_loss": 3.3615908653435556,
+      "tokens_seen": 2859139072
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001348900658000321,
+      "loss": 1.3118,
+      "theoretical_loss": 3.36157900222803,
+      "tokens_seen": 2859270144
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001348499438292409,
+      "loss": 1.3538,
+      "theoretical_loss": 3.3615671398085705,
+      "tokens_seen": 2859401216
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001348098218584497,
+      "loss": 1.2571,
+      "theoretical_loss": 3.361555278085105,
+      "tokens_seen": 2859532288
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013476969988765849,
+      "loss": 1.3053,
+      "theoretical_loss": 3.36154341705756,
+      "tokens_seen": 2859663360
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013472957791686727,
+      "loss": 1.4071,
+      "theoretical_loss": 3.361531556725864,
+      "tokens_seen": 2859794432
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013468945594607607,
+      "loss": 1.338,
+      "theoretical_loss": 3.3615196970899435,
+      "tokens_seen": 2859925504
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013464933397528488,
+      "loss": 1.3707,
+      "theoretical_loss": 3.3615078381497256,
+      "tokens_seen": 2860056576
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013460921200449366,
+      "loss": 1.3819,
+      "theoretical_loss": 3.361495979905138,
+      "tokens_seen": 2860187648
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013456909003370246,
+      "loss": 1.3788,
+      "theoretical_loss": 3.361484122356108,
+      "tokens_seen": 2860318720
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013452896806291127,
+      "loss": 1.2986,
+      "theoretical_loss": 3.3614722655025626,
+      "tokens_seen": 2860449792
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013448884609212005,
+      "loss": 1.2865,
+      "theoretical_loss": 3.3614604093444296,
+      "tokens_seen": 2860580864
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013444872412132883,
+      "loss": 1.2722,
+      "theoretical_loss": 3.361448553881636,
+      "tokens_seen": 2860711936
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013440860215053763,
+      "loss": 1.3807,
+      "theoretical_loss": 3.361436699114109,
+      "tokens_seen": 2860843008
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013436848017974644,
+      "loss": 1.3227,
+      "theoretical_loss": 3.3614248450417765,
+      "tokens_seen": 2860974080
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013432835820895522,
+      "loss": 1.316,
+      "theoretical_loss": 3.361412991664566,
+      "tokens_seen": 2861105152
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013428823623816403,
+      "loss": 1.3517,
+      "theoretical_loss": 3.361401138982404,
+      "tokens_seen": 2861236224
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013424811426737283,
+      "loss": 1.3152,
+      "theoretical_loss": 3.3613892869952187,
+      "tokens_seen": 2861367296
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001342079922965816,
+      "loss": 1.3956,
+      "theoretical_loss": 3.361377435702937,
+      "tokens_seen": 2861498368
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001341678703257904,
+      "loss": 1.3275,
+      "theoretical_loss": 3.361365585105487,
+      "tokens_seen": 2861629440
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001341277483549992,
+      "loss": 1.3369,
+      "theoretical_loss": 3.361353735202796,
+      "tokens_seen": 2861760512
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013408762638420798,
+      "loss": 1.3042,
+      "theoretical_loss": 3.3613418859947903,
+      "tokens_seen": 2861891584
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013404750441341678,
+      "loss": 1.2437,
+      "theoretical_loss": 3.3613300374813986,
+      "tokens_seen": 2862022656
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001340073824426256,
+      "loss": 1.3408,
+      "theoretical_loss": 3.361318189662548,
+      "tokens_seen": 2862153728
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001339672604718344,
+      "loss": 1.3004,
+      "theoretical_loss": 3.361306342538166,
+      "tokens_seen": 2862284800
+    },
+    {
+      "epoch": 0.73,
+      "objective/train/advantage_avg": 0.4913198947906494,
+      "objective/train/docs_used": 1040389,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3099042177200317,
+      "objective/train/original_loss": 1.3099043369293213,
+      "objective/train/theoretical_loss": 3.3612944961081803,
+      "objective/train/tokens_used": 1232874976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2418699711561203,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503617525100708,
+      "objective/train/weighted_lm_loss": 1.3758071660995483,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3612944961081803,
+      "tokens_seen": 2862415872
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013392713850104317,
+      "loss": 1.3414,
+      "theoretical_loss": 3.3612944961081803,
+      "tokens_seen": 2862415872
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013388701653025198,
+      "loss": 1.2677,
+      "theoretical_loss": 3.3612826503725186,
+      "tokens_seen": 2862546944
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00013384689455946076,
+      "loss": 1.2986,
+      "theoretical_loss": 3.3612708053311073,
+      "tokens_seen": 2862678016
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013380677258866954,
+      "loss": 1.3232,
+      "theoretical_loss": 3.361258960983875,
+      "tokens_seen": 2862809088
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013376665061787835,
+      "loss": 1.312,
+      "theoretical_loss": 3.3612471173307488,
+      "tokens_seen": 2862940160
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013372652864708715,
+      "loss": 1.3837,
+      "theoretical_loss": 3.3612352743716563,
+      "tokens_seen": 2863071232
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013368640667629596,
+      "loss": 1.3586,
+      "theoretical_loss": 3.3612234321065246,
+      "tokens_seen": 2863202304
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013364628470550474,
+      "loss": 1.2302,
+      "theoretical_loss": 3.3612115905352824,
+      "tokens_seen": 2863333376
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013360616273471354,
+      "loss": 1.3594,
+      "theoretical_loss": 3.3611997496578567,
+      "tokens_seen": 2863464448
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013356604076392232,
+      "loss": 1.278,
+      "theoretical_loss": 3.361187909474175,
+      "tokens_seen": 2863595520
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001335259187931311,
+      "loss": 1.3827,
+      "theoretical_loss": 3.3611760699841646,
+      "tokens_seen": 2863726592
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001334857968223399,
+      "loss": 1.3358,
+      "theoretical_loss": 3.361164231187754,
+      "tokens_seen": 2863857664
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013344567485154871,
+      "loss": 1.3064,
+      "theoretical_loss": 3.3611523930848697,
+      "tokens_seen": 2863988736
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013340555288075752,
+      "loss": 1.3533,
+      "theoretical_loss": 3.36114055567544,
+      "tokens_seen": 2864119808
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001333654309099663,
+      "loss": 1.2991,
+      "theoretical_loss": 3.3611287189593924,
+      "tokens_seen": 2864250880
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001333253089391751,
+      "loss": 1.3359,
+      "theoretical_loss": 3.361116882936655,
+      "tokens_seen": 2864381952
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001332851869683839,
+      "loss": 1.3496,
+      "theoretical_loss": 3.361105047607155,
+      "tokens_seen": 2864513024
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013324506499759266,
+      "loss": 1.269,
+      "theoretical_loss": 3.36109321297082,
+      "tokens_seen": 2864644096
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013320494302680147,
+      "loss": 1.4207,
+      "theoretical_loss": 3.3610813790275778,
+      "tokens_seen": 2864775168
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013316482105601028,
+      "loss": 1.3477,
+      "theoretical_loss": 3.3610695457773563,
+      "tokens_seen": 2864906240
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013312469908521906,
+      "loss": 1.4005,
+      "theoretical_loss": 3.361057713220083,
+      "tokens_seen": 2865037312
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013308457711442786,
+      "loss": 1.3157,
+      "theoretical_loss": 3.361045881355686,
+      "tokens_seen": 2865168384
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013304445514363667,
+      "loss": 1.3335,
+      "theoretical_loss": 3.3610340501840925,
+      "tokens_seen": 2865299456
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013300433317284548,
+      "loss": 1.3484,
+      "theoretical_loss": 3.36102221970523,
+      "tokens_seen": 2865430528
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013296421120205423,
+      "loss": 1.2496,
+      "theoretical_loss": 3.3610103899190276,
+      "tokens_seen": 2865561600
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013292408923126303,
+      "loss": 1.4148,
+      "theoretical_loss": 3.3609985608254114,
+      "tokens_seen": 2865692672
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013288396726047184,
+      "loss": 1.3483,
+      "theoretical_loss": 3.3609867324243106,
+      "tokens_seen": 2865823744
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013284384528968062,
+      "loss": 1.2871,
+      "theoretical_loss": 3.3609749047156523,
+      "tokens_seen": 2865954816
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013280372331888943,
+      "loss": 1.3203,
+      "theoretical_loss": 3.3609630776993638,
+      "tokens_seen": 2866085888
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013276360134809823,
+      "loss": 1.3048,
+      "theoretical_loss": 3.360951251375374,
+      "tokens_seen": 2866216960
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013272347937730704,
+      "loss": 1.2685,
+      "theoretical_loss": 3.36093942574361,
+      "tokens_seen": 2866348032
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001326833574065158,
+      "loss": 1.3574,
+      "theoretical_loss": 3.3609276008039997,
+      "tokens_seen": 2866479104
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001326432354357246,
+      "loss": 1.3397,
+      "theoretical_loss": 3.360915776556471,
+      "tokens_seen": 2866610176
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001326031134649334,
+      "loss": 1.4053,
+      "theoretical_loss": 3.3609039530009523,
+      "tokens_seen": 2866741248
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013256299149414218,
+      "loss": 1.3196,
+      "theoretical_loss": 3.3608921301373704,
+      "tokens_seen": 2866872320
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.000132522869523351,
+      "loss": 1.355,
+      "theoretical_loss": 3.3608803079656537,
+      "tokens_seen": 2867003392
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001324827475525598,
+      "loss": 1.2825,
+      "theoretical_loss": 3.3608684864857308,
+      "tokens_seen": 2867134464
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001324426255817686,
+      "loss": 1.3391,
+      "theoretical_loss": 3.3608566656975283,
+      "tokens_seen": 2867265536
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013240250361097738,
+      "loss": 1.3899,
+      "theoretical_loss": 3.360844845600975,
+      "tokens_seen": 2867396608
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013236238164018616,
+      "loss": 1.2673,
+      "theoretical_loss": 3.3608330261959987,
+      "tokens_seen": 2867527680
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013232225966939497,
+      "loss": 1.2685,
+      "theoretical_loss": 3.3608212074825268,
+      "tokens_seen": 2867658752
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013228213769860375,
+      "loss": 1.3075,
+      "theoretical_loss": 3.3608093894604876,
+      "tokens_seen": 2867789824
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013224201572781255,
+      "loss": 1.3798,
+      "theoretical_loss": 3.3607975721298096,
+      "tokens_seen": 2867920896
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013220189375702136,
+      "loss": 1.3353,
+      "theoretical_loss": 3.36078575549042,
+      "tokens_seen": 2868051968
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013216177178623016,
+      "loss": 1.343,
+      "theoretical_loss": 3.3607739395422467,
+      "tokens_seen": 2868183040
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013212164981543894,
+      "loss": 1.389,
+      "theoretical_loss": 3.3607621242852184,
+      "tokens_seen": 2868314112
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013208152784464772,
+      "loss": 1.323,
+      "theoretical_loss": 3.360750309719262,
+      "tokens_seen": 2868445184
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013204140587385653,
+      "loss": 1.2677,
+      "theoretical_loss": 3.360738495844307,
+      "tokens_seen": 2868576256
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001320012839030653,
+      "loss": 1.3215,
+      "theoretical_loss": 3.36072668266028,
+      "tokens_seen": 2868707328
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013196116193227411,
+      "loss": 1.3834,
+      "theoretical_loss": 3.36071487016711,
+      "tokens_seen": 2868838400
+    },
+    {
+      "epoch": 0.74,
+      "objective/train/advantage_avg": 0.4911837875843048,
+      "objective/train/docs_used": 1042847,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3018237352371216,
+      "objective/train/original_loss": 1.301823616027832,
+      "objective/train/theoretical_loss": 3.3607030583647246,
+      "objective/train/tokens_used": 1239428576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24194809794425964,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503482818603516,
+      "objective/train/weighted_lm_loss": 1.3672696352005005,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9048373699188232,
+      "theoretical_loss": 3.3607030583647246,
+      "tokens_seen": 2868969472
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013192103996148292,
+      "loss": 1.3829,
+      "theoretical_loss": 3.3607030583647246,
+      "tokens_seen": 2868969472
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001318809179906917,
+      "loss": 1.3545,
+      "theoretical_loss": 3.360691247253052,
+      "tokens_seen": 2869100544
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001318407960199005,
+      "loss": 1.3294,
+      "theoretical_loss": 3.3606794368320196,
+      "tokens_seen": 2869231616
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013180067404910929,
+      "loss": 1.3098,
+      "theoretical_loss": 3.3606676271015568,
+      "tokens_seen": 2869362688
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001317605520783181,
+      "loss": 1.3179,
+      "theoretical_loss": 3.3606558180615904,
+      "tokens_seen": 2869493760
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013172043010752687,
+      "loss": 1.3441,
+      "theoretical_loss": 3.360644009712049,
+      "tokens_seen": 2869624832
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013168030813673568,
+      "loss": 1.3268,
+      "theoretical_loss": 3.360632202052861,
+      "tokens_seen": 2869755904
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013164018616594448,
+      "loss": 1.3312,
+      "theoretical_loss": 3.360620395083954,
+      "tokens_seen": 2869886976
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013160006419515326,
+      "loss": 1.286,
+      "theoretical_loss": 3.360608588805256,
+      "tokens_seen": 2870018048
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013155994222436207,
+      "loss": 1.3607,
+      "theoretical_loss": 3.360596783216696,
+      "tokens_seen": 2870149120
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013151982025357088,
+      "loss": 1.3263,
+      "theoretical_loss": 3.360584978318202,
+      "tokens_seen": 2870280192
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013147969828277965,
+      "loss": 1.3111,
+      "theoretical_loss": 3.360573174109701,
+      "tokens_seen": 2870411264
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013143957631198843,
+      "loss": 1.3825,
+      "theoretical_loss": 3.360561370591122,
+      "tokens_seen": 2870542336
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013139945434119724,
+      "loss": 1.3765,
+      "theoretical_loss": 3.3605495677623933,
+      "tokens_seen": 2870673408
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013135933237040605,
+      "loss": 1.303,
+      "theoretical_loss": 3.3605377656234428,
+      "tokens_seen": 2870804480
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013131921039961483,
+      "loss": 1.271,
+      "theoretical_loss": 3.3605259641741987,
+      "tokens_seen": 2870935552
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013127908842882363,
+      "loss": 1.2625,
+      "theoretical_loss": 3.3605141634145896,
+      "tokens_seen": 2871066624
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013123896645803244,
+      "loss": 1.3263,
+      "theoretical_loss": 3.360502363344543,
+      "tokens_seen": 2871197696
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001311988444872412,
+      "loss": 1.3867,
+      "theoretical_loss": 3.360490563963988,
+      "tokens_seen": 2871328768
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013115872251645,
+      "loss": 1.2848,
+      "theoretical_loss": 3.3604787652728523,
+      "tokens_seen": 2871459840
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001311186005456588,
+      "loss": 1.2946,
+      "theoretical_loss": 3.360466967271064,
+      "tokens_seen": 2871590912
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001310784785748676,
+      "loss": 1.2663,
+      "theoretical_loss": 3.360455169958552,
+      "tokens_seen": 2871721984
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001310383566040764,
+      "loss": 1.3409,
+      "theoretical_loss": 3.3604433733352437,
+      "tokens_seen": 2871853056
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001309982346332852,
+      "loss": 1.3545,
+      "theoretical_loss": 3.360431577401068,
+      "tokens_seen": 2871984128
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.000130958112662494,
+      "loss": 1.3368,
+      "theoretical_loss": 3.360419782155953,
+      "tokens_seen": 2872115200
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013091799069170278,
+      "loss": 1.3331,
+      "theoretical_loss": 3.360407987599827,
+      "tokens_seen": 2872246272
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013087786872091156,
+      "loss": 1.2527,
+      "theoretical_loss": 3.360396193732619,
+      "tokens_seen": 2872377344
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013083774675012037,
+      "loss": 1.2911,
+      "theoretical_loss": 3.3603844005542562,
+      "tokens_seen": 2872508416
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013079762477932917,
+      "loss": 1.26,
+      "theoretical_loss": 3.360372608064668,
+      "tokens_seen": 2872639488
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013075750280853795,
+      "loss": 1.3387,
+      "theoretical_loss": 3.3603608162637815,
+      "tokens_seen": 2872770560
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013071738083774676,
+      "loss": 1.3143,
+      "theoretical_loss": 3.3603490251515256,
+      "tokens_seen": 2872901632
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013067725886695556,
+      "loss": 1.2731,
+      "theoretical_loss": 3.3603372347278295,
+      "tokens_seen": 2873032704
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013063713689616434,
+      "loss": 1.3213,
+      "theoretical_loss": 3.3603254449926205,
+      "tokens_seen": 2873163776
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013059701492537312,
+      "loss": 1.3137,
+      "theoretical_loss": 3.3603136559458275,
+      "tokens_seen": 2873294848
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013055689295458193,
+      "loss": 1.2378,
+      "theoretical_loss": 3.3603018675873786,
+      "tokens_seen": 2873425920
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013051677098379073,
+      "loss": 1.3936,
+      "theoretical_loss": 3.3602900799172026,
+      "tokens_seen": 2873556992
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013047664901299951,
+      "loss": 1.3619,
+      "theoretical_loss": 3.3602782929352277,
+      "tokens_seen": 2873688064
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013043652704220832,
+      "loss": 1.3107,
+      "theoretical_loss": 3.3602665066413824,
+      "tokens_seen": 2873819136
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013039640507141713,
+      "loss": 1.2905,
+      "theoretical_loss": 3.360254721035595,
+      "tokens_seen": 2873950208
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001303562831006259,
+      "loss": 1.3688,
+      "theoretical_loss": 3.3602429361177943,
+      "tokens_seen": 2874081280
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013031616112983468,
+      "loss": 1.2691,
+      "theoretical_loss": 3.360231151887908,
+      "tokens_seen": 2874212352
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001302760391590435,
+      "loss": 1.3367,
+      "theoretical_loss": 3.3602193683458657,
+      "tokens_seen": 2874343424
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001302359171882523,
+      "loss": 1.3486,
+      "theoretical_loss": 3.3602075854915947,
+      "tokens_seen": 2874474496
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013019579521746108,
+      "loss": 1.227,
+      "theoretical_loss": 3.3601958033250243,
+      "tokens_seen": 2874605568
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013015567324666988,
+      "loss": 1.3035,
+      "theoretical_loss": 3.3601840218460834,
+      "tokens_seen": 2874736640
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001301155512758787,
+      "loss": 1.332,
+      "theoretical_loss": 3.3601722410546992,
+      "tokens_seen": 2874867712
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013007542930508747,
+      "loss": 1.3245,
+      "theoretical_loss": 3.360160460950801,
+      "tokens_seen": 2874998784
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00013003530733429627,
+      "loss": 1.4062,
+      "theoretical_loss": 3.3601486815343176,
+      "tokens_seen": 2875129856
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00012999518536350505,
+      "loss": 1.3975,
+      "theoretical_loss": 3.360136902805177,
+      "tokens_seen": 2875260928
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00012995506339271383,
+      "loss": 1.3727,
+      "theoretical_loss": 3.360125124763308,
+      "tokens_seen": 2875392000
+    },
+    {
+      "epoch": 0.74,
+      "objective/train/advantage_avg": 0.49373599886894226,
+      "objective/train/docs_used": 1045259,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.1182292699813843,
+      "objective/train/original_loss": 1.1182291507720947,
+      "objective/train/theoretical_loss": 3.3601133474086398,
+      "objective/train/tokens_used": 1245982176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2441120445728302,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050614595413208,
+      "objective/train/weighted_lm_loss": 1.174704909324646,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3601133474086398,
+      "tokens_seen": 2875523072
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00012991494142192264,
+      "loss": 1.2541,
+      "theoretical_loss": 3.3601133474086398,
+      "tokens_seen": 2875523072
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00012987481945113145,
+      "loss": 1.3429,
+      "theoretical_loss": 3.3601015707410995,
+      "tokens_seen": 2875654144
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00012983469748034025,
+      "loss": 1.3528,
+      "theoretical_loss": 3.3600897947606168,
+      "tokens_seen": 2875785216
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00012979457550954903,
+      "loss": 1.3254,
+      "theoretical_loss": 3.36007801946712,
+      "tokens_seen": 2875916288
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00012975445353875784,
+      "loss": 1.3413,
+      "theoretical_loss": 3.3600662448605383,
+      "tokens_seen": 2876047360
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00012971433156796662,
+      "loss": 1.362,
+      "theoretical_loss": 3.360054470940799,
+      "tokens_seen": 2876178432
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001296742095971754,
+      "loss": 1.3521,
+      "theoretical_loss": 3.360042697707832,
+      "tokens_seen": 2876309504
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001296340876263842,
+      "loss": 1.336,
+      "theoretical_loss": 3.3600309251615657,
+      "tokens_seen": 2876440576
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.000129593965655593,
+      "loss": 1.3788,
+      "theoretical_loss": 3.360019153301928,
+      "tokens_seen": 2876571648
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00012955384368480181,
+      "loss": 1.2919,
+      "theoretical_loss": 3.3600073821288485,
+      "tokens_seen": 2876702720
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001295137217140106,
+      "loss": 1.3505,
+      "theoretical_loss": 3.3599956116422556,
+      "tokens_seen": 2876833792
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001294735997432194,
+      "loss": 1.3357,
+      "theoretical_loss": 3.3599838418420775,
+      "tokens_seen": 2876964864
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00012943347777242818,
+      "loss": 1.4101,
+      "theoretical_loss": 3.3599720727282434,
+      "tokens_seen": 2877095936
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00012939335580163696,
+      "loss": 1.3061,
+      "theoretical_loss": 3.3599603043006816,
+      "tokens_seen": 2877227008
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00012935323383084577,
+      "loss": 1.3528,
+      "theoretical_loss": 3.3599485365593216,
+      "tokens_seen": 2877358080
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00012931311186005457,
+      "loss": 1.3116,
+      "theoretical_loss": 3.3599367695040914,
+      "tokens_seen": 2877489152
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00012927298988926338,
+      "loss": 1.3284,
+      "theoretical_loss": 3.35992500313492,
+      "tokens_seen": 2877620224
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00012923286791847216,
+      "loss": 1.3417,
+      "theoretical_loss": 3.359913237451736,
+      "tokens_seen": 2877751296
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00012919274594768096,
+      "loss": 1.2948,
+      "theoretical_loss": 3.3599014724544687,
+      "tokens_seen": 2877882368
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00012915262397688977,
+      "loss": 1.3733,
+      "theoretical_loss": 3.3598897081430463,
+      "tokens_seen": 2878013440
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00012911250200609852,
+      "loss": 1.3629,
+      "theoretical_loss": 3.3598779445173976,
+      "tokens_seen": 2878144512
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00012907238003530733,
+      "loss": 1.3669,
+      "theoretical_loss": 3.359866181577452,
+      "tokens_seen": 2878275584
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00012903225806451613,
+      "loss": 1.3451,
+      "theoretical_loss": 3.3598544193231374,
+      "tokens_seen": 2878406656
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001289921360937249,
+      "loss": 1.3008,
+      "theoretical_loss": 3.3598426577543834,
+      "tokens_seen": 2878537728
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00012895201412293372,
+      "loss": 1.3834,
+      "theoretical_loss": 3.3598308968711184,
+      "tokens_seen": 2878668800
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00012891189215214253,
+      "loss": 1.3618,
+      "theoretical_loss": 3.3598191366732713,
+      "tokens_seen": 2878799872
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00012887177018135133,
+      "loss": 1.2489,
+      "theoretical_loss": 3.359807377160771,
+      "tokens_seen": 2878930944
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00012883164821056008,
+      "loss": 1.2969,
+      "theoretical_loss": 3.3597956183335462,
+      "tokens_seen": 2879062016
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001287915262397689,
+      "loss": 1.3686,
+      "theoretical_loss": 3.359783860191526,
+      "tokens_seen": 2879193088
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001287514042689777,
+      "loss": 1.3181,
+      "theoretical_loss": 3.35977210273464,
+      "tokens_seen": 2879324160
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012871128229818648,
+      "loss": 1.3739,
+      "theoretical_loss": 3.3597603459628154,
+      "tokens_seen": 2879455232
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012867116032739528,
+      "loss": 1.3102,
+      "theoretical_loss": 3.3597485898759825,
+      "tokens_seen": 2879586304
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001286310383566041,
+      "loss": 1.2577,
+      "theoretical_loss": 3.3597368344740692,
+      "tokens_seen": 2879717376
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001285909163858129,
+      "loss": 1.3233,
+      "theoretical_loss": 3.3597250797570055,
+      "tokens_seen": 2879848448
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012855079441502167,
+      "loss": 1.355,
+      "theoretical_loss": 3.3597133257247194,
+      "tokens_seen": 2879979520
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012851067244423045,
+      "loss": 1.3453,
+      "theoretical_loss": 3.3597015723771406,
+      "tokens_seen": 2880110592
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012847055047343926,
+      "loss": 1.2945,
+      "theoretical_loss": 3.3596898197141973,
+      "tokens_seen": 2880241664
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012843042850264804,
+      "loss": 1.3334,
+      "theoretical_loss": 3.3596780677358193,
+      "tokens_seen": 2880372736
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012839030653185685,
+      "loss": 1.3544,
+      "theoretical_loss": 3.3596663164419347,
+      "tokens_seen": 2880503808
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012835018456106565,
+      "loss": 1.3298,
+      "theoretical_loss": 3.3596545658324732,
+      "tokens_seen": 2880634880
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012831006259027446,
+      "loss": 1.269,
+      "theoretical_loss": 3.359642815907364,
+      "tokens_seen": 2880765952
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012826994061948324,
+      "loss": 1.2957,
+      "theoretical_loss": 3.3596310666665348,
+      "tokens_seen": 2880897024
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012822981864869202,
+      "loss": 1.269,
+      "theoretical_loss": 3.3596193181099157,
+      "tokens_seen": 2881028096
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012818969667790082,
+      "loss": 1.3292,
+      "theoretical_loss": 3.3596075702374355,
+      "tokens_seen": 2881159168
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001281495747071096,
+      "loss": 1.3713,
+      "theoretical_loss": 3.3595958230490233,
+      "tokens_seen": 2881290240
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001281094527363184,
+      "loss": 1.3267,
+      "theoretical_loss": 3.359584076544608,
+      "tokens_seen": 2881421312
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012806933076552721,
+      "loss": 1.3256,
+      "theoretical_loss": 3.3595723307241188,
+      "tokens_seen": 2881552384
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.000128029208794736,
+      "loss": 1.3513,
+      "theoretical_loss": 3.3595605855874844,
+      "tokens_seen": 2881683456
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001279890868239448,
+      "loss": 1.2965,
+      "theoretical_loss": 3.3595488411346346,
+      "tokens_seen": 2881814528
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012794896485315358,
+      "loss": 1.3243,
+      "theoretical_loss": 3.359537097365498,
+      "tokens_seen": 2881945600
+    },
+    {
+      "epoch": 0.75,
+      "objective/train/advantage_avg": 0.4919630289077759,
+      "objective/train/docs_used": 1047629,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2433931827545166,
+      "objective/train/original_loss": 1.2433931827545166,
+      "objective/train/theoretical_loss": 3.359525354280003,
+      "objective/train/tokens_used": 1252535776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24281856417655945,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504306554794312,
+      "objective/train/weighted_lm_loss": 1.3060230016708374,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.359525354280003,
+      "tokens_seen": 2882076672
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012790884288236239,
+      "loss": 1.2724,
+      "theoretical_loss": 3.359525354280003,
+      "tokens_seen": 2882076672
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012786872091157116,
+      "loss": 1.2913,
+      "theoretical_loss": 3.359513611878081,
+      "tokens_seen": 2882207744
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012782859894077997,
+      "loss": 1.3817,
+      "theoretical_loss": 3.3595018701596584,
+      "tokens_seen": 2882338816
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012778847696998878,
+      "loss": 1.2183,
+      "theoretical_loss": 3.359490129124666,
+      "tokens_seen": 2882469888
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012774835499919756,
+      "loss": 1.325,
+      "theoretical_loss": 3.3594783887730326,
+      "tokens_seen": 2882600960
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012770823302840636,
+      "loss": 1.3467,
+      "theoretical_loss": 3.359466649104687,
+      "tokens_seen": 2882732032
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012766811105761517,
+      "loss": 1.3796,
+      "theoretical_loss": 3.3594549101195588,
+      "tokens_seen": 2882863104
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012762798908682395,
+      "loss": 1.3424,
+      "theoretical_loss": 3.359443171817577,
+      "tokens_seen": 2882994176
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012758786711603273,
+      "loss": 1.3232,
+      "theoretical_loss": 3.359431434198671,
+      "tokens_seen": 2883125248
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012754774514524153,
+      "loss": 1.341,
+      "theoretical_loss": 3.3594196972627697,
+      "tokens_seen": 2883256320
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012750762317445034,
+      "loss": 1.2562,
+      "theoretical_loss": 3.359407961009802,
+      "tokens_seen": 2883387392
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012746750120365912,
+      "loss": 1.4082,
+      "theoretical_loss": 3.359396225439698,
+      "tokens_seen": 2883518464
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012742737923286793,
+      "loss": 1.308,
+      "theoretical_loss": 3.359384490552386,
+      "tokens_seen": 2883649536
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012738725726207673,
+      "loss": 1.3607,
+      "theoretical_loss": 3.3593727563477964,
+      "tokens_seen": 2883780608
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001273471352912855,
+      "loss": 1.2878,
+      "theoretical_loss": 3.359361022825857,
+      "tokens_seen": 2883911680
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001273070133204943,
+      "loss": 1.3165,
+      "theoretical_loss": 3.3593492899864987,
+      "tokens_seen": 2884042752
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001272668913497031,
+      "loss": 1.3347,
+      "theoretical_loss": 3.359337557829649,
+      "tokens_seen": 2884173824
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001272267693789119,
+      "loss": 1.3586,
+      "theoretical_loss": 3.3593258263552386,
+      "tokens_seen": 2884304896
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012718664740812068,
+      "loss": 1.2885,
+      "theoretical_loss": 3.359314095563196,
+      "tokens_seen": 2884435968
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001271465254373295,
+      "loss": 1.2844,
+      "theoretical_loss": 3.359302365453451,
+      "tokens_seen": 2884567040
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001271064034665383,
+      "loss": 1.3338,
+      "theoretical_loss": 3.3592906360259325,
+      "tokens_seen": 2884698112
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012706628149574705,
+      "loss": 1.3036,
+      "theoretical_loss": 3.35927890728057,
+      "tokens_seen": 2884829184
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012702615952495585,
+      "loss": 1.2997,
+      "theoretical_loss": 3.359267179217293,
+      "tokens_seen": 2884960256
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012698603755416466,
+      "loss": 1.2049,
+      "theoretical_loss": 3.359255451836031,
+      "tokens_seen": 2885091328
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012694591558337347,
+      "loss": 1.3191,
+      "theoretical_loss": 3.359243725136712,
+      "tokens_seen": 2885222400
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012690579361258225,
+      "loss": 1.4005,
+      "theoretical_loss": 3.3592319991192676,
+      "tokens_seen": 2885353472
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012686567164179105,
+      "loss": 1.3197,
+      "theoretical_loss": 3.3592202737836256,
+      "tokens_seen": 2885484544
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012682554967099986,
+      "loss": 1.3268,
+      "theoretical_loss": 3.3592085491297157,
+      "tokens_seen": 2885615616
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012678542770020864,
+      "loss": 1.3164,
+      "theoretical_loss": 3.359196825157467,
+      "tokens_seen": 2885746688
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012674530572941742,
+      "loss": 1.354,
+      "theoretical_loss": 3.35918510186681,
+      "tokens_seen": 2885877760
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012670518375862622,
+      "loss": 1.2449,
+      "theoretical_loss": 3.359173379257673,
+      "tokens_seen": 2886008832
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012666506178783503,
+      "loss": 1.3404,
+      "theoretical_loss": 3.359161657329986,
+      "tokens_seen": 2886139904
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001266249398170438,
+      "loss": 1.3026,
+      "theoretical_loss": 3.3591499360836785,
+      "tokens_seen": 2886270976
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012658481784625261,
+      "loss": 1.337,
+      "theoretical_loss": 3.359138215518679,
+      "tokens_seen": 2886402048
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012654469587546142,
+      "loss": 1.3061,
+      "theoretical_loss": 3.3591264956349187,
+      "tokens_seen": 2886533120
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001265045739046702,
+      "loss": 1.2952,
+      "theoretical_loss": 3.3591147764323255,
+      "tokens_seen": 2886664192
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012646445193387898,
+      "loss": 1.3123,
+      "theoretical_loss": 3.3591030579108296,
+      "tokens_seen": 2886795264
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012642432996308779,
+      "loss": 1.3706,
+      "theoretical_loss": 3.35909134007036,
+      "tokens_seen": 2886926336
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001263842079922966,
+      "loss": 1.3376,
+      "theoretical_loss": 3.359079622910847,
+      "tokens_seen": 2887057408
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012634408602150537,
+      "loss": 1.2911,
+      "theoretical_loss": 3.3590679064322195,
+      "tokens_seen": 2887188480
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012630396405071418,
+      "loss": 1.2864,
+      "theoretical_loss": 3.3590561906344076,
+      "tokens_seen": 2887319552
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012626384207992298,
+      "loss": 1.3254,
+      "theoretical_loss": 3.35904447551734,
+      "tokens_seen": 2887450624
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012622372010913176,
+      "loss": 1.3172,
+      "theoretical_loss": 3.3590327610809467,
+      "tokens_seen": 2887581696
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012618359813834057,
+      "loss": 1.3261,
+      "theoretical_loss": 3.3590210473251574,
+      "tokens_seen": 2887712768
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012614347616754935,
+      "loss": 1.2833,
+      "theoretical_loss": 3.3590093342499014,
+      "tokens_seen": 2887843840
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012610335419675815,
+      "loss": 1.3394,
+      "theoretical_loss": 3.3589976218551083,
+      "tokens_seen": 2887974912
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012606323222596693,
+      "loss": 1.2683,
+      "theoretical_loss": 3.3589859101407082,
+      "tokens_seen": 2888105984
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012602311025517574,
+      "loss": 1.4215,
+      "theoretical_loss": 3.35897419910663,
+      "tokens_seen": 2888237056
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012598298828438455,
+      "loss": 1.294,
+      "theoretical_loss": 3.3589624887528036,
+      "tokens_seen": 2888368128
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012594286631359333,
+      "loss": 1.3277,
+      "theoretical_loss": 3.3589507790791586,
+      "tokens_seen": 2888499200
+    },
+    {
+      "epoch": 0.75,
+      "objective/train/advantage_avg": 0.49336034059524536,
+      "objective/train/docs_used": 1050131,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3315789699554443,
+      "objective/train/original_loss": 1.3315792083740234,
+      "objective/train/theoretical_loss": 3.3589390700856248,
+      "objective/train/tokens_used": 1259089376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24374200403690338,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505751371383667,
+      "objective/train/weighted_lm_loss": 1.398840069770813,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3589390700856248,
+      "tokens_seen": 2888630272
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012590274434280213,
+      "loss": 1.3706,
+      "theoretical_loss": 3.3589390700856248,
+      "tokens_seen": 2888630272
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001258626223720109,
+      "loss": 1.3853,
+      "theoretical_loss": 3.3589273617721314,
+      "tokens_seen": 2888761344
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001258225004012197,
+      "loss": 1.3895,
+      "theoretical_loss": 3.3589156541386083,
+      "tokens_seen": 2888892416
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001257823784304285,
+      "loss": 1.2914,
+      "theoretical_loss": 3.3589039471849853,
+      "tokens_seen": 2889023488
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001257422564596373,
+      "loss": 1.3523,
+      "theoretical_loss": 3.3588922409111923,
+      "tokens_seen": 2889154560
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001257021344888461,
+      "loss": 1.335,
+      "theoretical_loss": 3.3588805353171582,
+      "tokens_seen": 2889285632
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001256620125180549,
+      "loss": 1.2985,
+      "theoretical_loss": 3.3588688304028134,
+      "tokens_seen": 2889416704
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001256218905472637,
+      "loss": 1.3197,
+      "theoretical_loss": 3.358857126168087,
+      "tokens_seen": 2889547776
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012558176857647247,
+      "loss": 1.3372,
+      "theoretical_loss": 3.3588454226129096,
+      "tokens_seen": 2889678848
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012554164660568125,
+      "loss": 1.3097,
+      "theoretical_loss": 3.35883371973721,
+      "tokens_seen": 2889809920
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012550152463489006,
+      "loss": 1.309,
+      "theoretical_loss": 3.3588220175409185,
+      "tokens_seen": 2889940992
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012546140266409887,
+      "loss": 1.3281,
+      "theoretical_loss": 3.3588103160239644,
+      "tokens_seen": 2890072064
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012542128069330767,
+      "loss": 1.3032,
+      "theoretical_loss": 3.358798615186278,
+      "tokens_seen": 2890203136
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012538115872251645,
+      "loss": 1.292,
+      "theoretical_loss": 3.358786915027789,
+      "tokens_seen": 2890334208
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012534103675172526,
+      "loss": 1.3452,
+      "theoretical_loss": 3.3587752155484263,
+      "tokens_seen": 2890465280
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012530091478093406,
+      "loss": 1.3074,
+      "theoretical_loss": 3.3587635167481205,
+      "tokens_seen": 2890596352
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012526079281014282,
+      "loss": 1.3484,
+      "theoretical_loss": 3.3587518186268017,
+      "tokens_seen": 2890727424
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012522067083935162,
+      "loss": 1.2324,
+      "theoretical_loss": 3.358740121184399,
+      "tokens_seen": 2890858496
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012518054886856043,
+      "loss": 1.3898,
+      "theoretical_loss": 3.3587284244208426,
+      "tokens_seen": 2890989568
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012514042689776923,
+      "loss": 1.2846,
+      "theoretical_loss": 3.358716728336062,
+      "tokens_seen": 2891120640
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012510030492697801,
+      "loss": 1.2872,
+      "theoretical_loss": 3.3587050329299872,
+      "tokens_seen": 2891251712
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012506018295618682,
+      "loss": 1.3295,
+      "theoretical_loss": 3.3586933382025483,
+      "tokens_seen": 2891382784
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012502006098539563,
+      "loss": 1.3308,
+      "theoretical_loss": 3.3586816441536747,
+      "tokens_seen": 2891513856
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001249799390146044,
+      "loss": 1.3322,
+      "theoretical_loss": 3.3586699507832964,
+      "tokens_seen": 2891644928
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001249398170438132,
+      "loss": 1.2959,
+      "theoretical_loss": 3.358658258091344,
+      "tokens_seen": 2891776000
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.000124899695073022,
+      "loss": 1.335,
+      "theoretical_loss": 3.3586465660777467,
+      "tokens_seen": 2891907072
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012485957310223077,
+      "loss": 1.2858,
+      "theoretical_loss": 3.3586348747424344,
+      "tokens_seen": 2892038144
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012481945113143958,
+      "loss": 1.3194,
+      "theoretical_loss": 3.358623184085337,
+      "tokens_seen": 2892169216
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012477932916064838,
+      "loss": 1.336,
+      "theoretical_loss": 3.358611494106384,
+      "tokens_seen": 2892300288
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012473920718985716,
+      "loss": 1.3233,
+      "theoretical_loss": 3.3585998048055066,
+      "tokens_seen": 2892431360
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012469908521906597,
+      "loss": 1.3142,
+      "theoretical_loss": 3.358588116182634,
+      "tokens_seen": 2892562432
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012465896324827477,
+      "loss": 1.3433,
+      "theoretical_loss": 3.3585764282376958,
+      "tokens_seen": 2892693504
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012461884127748355,
+      "loss": 1.3516,
+      "theoretical_loss": 3.3585647409706225,
+      "tokens_seen": 2892824576
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012457871930669233,
+      "loss": 1.3744,
+      "theoretical_loss": 3.358553054381344,
+      "tokens_seen": 2892955648
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012453859733590114,
+      "loss": 1.2385,
+      "theoretical_loss": 3.35854136846979,
+      "tokens_seen": 2893086720
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012449847536510995,
+      "loss": 1.3256,
+      "theoretical_loss": 3.358529683235891,
+      "tokens_seen": 2893217792
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012445835339431873,
+      "loss": 1.3805,
+      "theoretical_loss": 3.3585179986795763,
+      "tokens_seen": 2893348864
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012441823142352753,
+      "loss": 1.37,
+      "theoretical_loss": 3.3585063148007768,
+      "tokens_seen": 2893479936
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001243781094527363,
+      "loss": 1.288,
+      "theoretical_loss": 3.3584946315994215,
+      "tokens_seen": 2893611008
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012433798748194512,
+      "loss": 1.3054,
+      "theoretical_loss": 3.3584829490754413,
+      "tokens_seen": 2893742080
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001242978655111539,
+      "loss": 1.3187,
+      "theoretical_loss": 3.358471267228766,
+      "tokens_seen": 2893873152
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001242577435403627,
+      "loss": 1.2615,
+      "theoretical_loss": 3.3584595860593254,
+      "tokens_seen": 2894004224
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001242176215695715,
+      "loss": 1.2733,
+      "theoretical_loss": 3.3584479055670498,
+      "tokens_seen": 2894135296
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001241774995987803,
+      "loss": 1.2831,
+      "theoretical_loss": 3.358436225751869,
+      "tokens_seen": 2894266368
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001241373776279891,
+      "loss": 1.3579,
+      "theoretical_loss": 3.3584245466137137,
+      "tokens_seen": 2894397440
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012409725565719787,
+      "loss": 1.3671,
+      "theoretical_loss": 3.3584128681525134,
+      "tokens_seen": 2894528512
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012405713368640668,
+      "loss": 1.3339,
+      "theoretical_loss": 3.3584011903681983,
+      "tokens_seen": 2894659584
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012401701171561546,
+      "loss": 1.3654,
+      "theoretical_loss": 3.3583895132606987,
+      "tokens_seen": 2894790656
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012397688974482427,
+      "loss": 1.3781,
+      "theoretical_loss": 3.358377836829945,
+      "tokens_seen": 2894921728
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012393676777403307,
+      "loss": 1.3236,
+      "theoretical_loss": 3.3583661610758666,
+      "tokens_seen": 2895052800
+    },
+    {
+      "epoch": 0.75,
+      "objective/train/advantage_avg": 0.49467605352401733,
+      "objective/train/docs_used": 1052611,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2405346632003784,
+      "objective/train/original_loss": 1.2405346632003784,
+      "objective/train/theoretical_loss": 3.3583544859983947,
+      "objective/train/tokens_used": 1265642976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24509792029857635,
+      "objective/train/value_max": -0.5,
+      "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": 1.3033971786499023,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3583544859983947,
+      "tokens_seen": 2895183872
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012389664580324185,
+      "loss": 1.2953,
+      "theoretical_loss": 3.3583544859983947,
+      "tokens_seen": 2895183872
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012385652383245066,
+      "loss": 1.2979,
+      "theoretical_loss": 3.358342811597458,
+      "tokens_seen": 2895314944
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012381640186165944,
+      "loss": 1.3056,
+      "theoretical_loss": 3.3583311378729883,
+      "tokens_seen": 2895446016
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00012377627989086824,
+      "loss": 1.3445,
+      "theoretical_loss": 3.3583194648249144,
+      "tokens_seen": 2895577088
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012373615792007702,
+      "loss": 1.294,
+      "theoretical_loss": 3.3583077924531675,
+      "tokens_seen": 2895708160
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012369603594928583,
+      "loss": 1.2771,
+      "theoretical_loss": 3.358296120757677,
+      "tokens_seen": 2895839232
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012365591397849463,
+      "loss": 1.3038,
+      "theoretical_loss": 3.358284449738374,
+      "tokens_seen": 2895970304
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012361579200770341,
+      "loss": 1.3245,
+      "theoretical_loss": 3.358272779395188,
+      "tokens_seen": 2896101376
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012357567003691222,
+      "loss": 1.3469,
+      "theoretical_loss": 3.358261109728049,
+      "tokens_seen": 2896232448
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.000123535548066121,
+      "loss": 1.3206,
+      "theoretical_loss": 3.358249440736888,
+      "tokens_seen": 2896363520
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0001234954260953298,
+      "loss": 1.2991,
+      "theoretical_loss": 3.358237772421635,
+      "tokens_seen": 2896494592
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0001234553041245386,
+      "loss": 1.3752,
+      "theoretical_loss": 3.3582261047822204,
+      "tokens_seen": 2896625664
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0001234151821537474,
+      "loss": 1.3084,
+      "theoretical_loss": 3.358214437818574,
+      "tokens_seen": 2896756736
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0001233750601829562,
+      "loss": 1.3443,
+      "theoretical_loss": 3.358202771530627,
+      "tokens_seen": 2896887808
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012333493821216498,
+      "loss": 1.2471,
+      "theoretical_loss": 3.3581911059183085,
+      "tokens_seen": 2897018880
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012329481624137378,
+      "loss": 1.3642,
+      "theoretical_loss": 3.3581794409815493,
+      "tokens_seen": 2897149952
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012325469427058256,
+      "loss": 1.2535,
+      "theoretical_loss": 3.35816777672028,
+      "tokens_seen": 2897281024
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012321457229979137,
+      "loss": 1.3354,
+      "theoretical_loss": 3.358156113134431,
+      "tokens_seen": 2897412096
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012317445032900017,
+      "loss": 1.2632,
+      "theoretical_loss": 3.3581444502239317,
+      "tokens_seen": 2897543168
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012313432835820895,
+      "loss": 1.2772,
+      "theoretical_loss": 3.3581327879887137,
+      "tokens_seen": 2897674240
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012309420638741776,
+      "loss": 1.274,
+      "theoretical_loss": 3.3581211264287067,
+      "tokens_seen": 2897805312
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012305408441662654,
+      "loss": 1.3061,
+      "theoretical_loss": 3.358109465543841,
+      "tokens_seen": 2897936384
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012301396244583535,
+      "loss": 1.3617,
+      "theoretical_loss": 3.3580978053340473,
+      "tokens_seen": 2898067456
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012297384047504412,
+      "loss": 1.3017,
+      "theoretical_loss": 3.3580861457992555,
+      "tokens_seen": 2898198528
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012293371850425293,
+      "loss": 1.3792,
+      "theoretical_loss": 3.3580744869393966,
+      "tokens_seen": 2898329600
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012289359653346174,
+      "loss": 1.3068,
+      "theoretical_loss": 3.3580628287544005,
+      "tokens_seen": 2898460672
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012285347456267052,
+      "loss": 1.2818,
+      "theoretical_loss": 3.358051171244198,
+      "tokens_seen": 2898591744
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012281335259187932,
+      "loss": 1.2819,
+      "theoretical_loss": 3.35803951440872,
+      "tokens_seen": 2898722816
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0001227732306210881,
+      "loss": 1.3081,
+      "theoretical_loss": 3.3580278582478953,
+      "tokens_seen": 2898853888
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0001227331086502969,
+      "loss": 1.2929,
+      "theoretical_loss": 3.3580162027616556,
+      "tokens_seen": 2898984960
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0001226929866795057,
+      "loss": 1.3682,
+      "theoretical_loss": 3.358004547949932,
+      "tokens_seen": 2899116032
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0001226528647087145,
+      "loss": 1.3029,
+      "theoretical_loss": 3.357992893812653,
+      "tokens_seen": 2899247104
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0001226127427379233,
+      "loss": 1.2302,
+      "theoretical_loss": 3.3579812403497504,
+      "tokens_seen": 2899378176
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012257262076713208,
+      "loss": 1.3722,
+      "theoretical_loss": 3.357969587561155,
+      "tokens_seen": 2899509248
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012253249879634089,
+      "loss": 1.2862,
+      "theoretical_loss": 3.3579579354467963,
+      "tokens_seen": 2899640320
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012249237682554966,
+      "loss": 1.288,
+      "theoretical_loss": 3.357946284006606,
+      "tokens_seen": 2899771392
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012245225485475847,
+      "loss": 1.3305,
+      "theoretical_loss": 3.357934633240513,
+      "tokens_seen": 2899902464
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012241213288396728,
+      "loss": 1.4064,
+      "theoretical_loss": 3.357922983148449,
+      "tokens_seen": 2900033536
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012237201091317606,
+      "loss": 1.3022,
+      "theoretical_loss": 3.357911333730345,
+      "tokens_seen": 2900164608
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012233188894238486,
+      "loss": 1.3569,
+      "theoretical_loss": 3.35789968498613,
+      "tokens_seen": 2900295680
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012229176697159364,
+      "loss": 1.3294,
+      "theoretical_loss": 3.357888036915736,
+      "tokens_seen": 2900426752
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012225164500080245,
+      "loss": 1.2844,
+      "theoretical_loss": 3.3578763895190926,
+      "tokens_seen": 2900557824
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012221152303001123,
+      "loss": 1.3532,
+      "theoretical_loss": 3.3578647427961306,
+      "tokens_seen": 2900688896
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012217140105922003,
+      "loss": 1.3507,
+      "theoretical_loss": 3.3578530967467812,
+      "tokens_seen": 2900819968
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012213127908842884,
+      "loss": 1.3323,
+      "theoretical_loss": 3.3578414513709745,
+      "tokens_seen": 2900951040
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012209115711763762,
+      "loss": 1.2972,
+      "theoretical_loss": 3.3578298066686414,
+      "tokens_seen": 2901082112
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012205103514684641,
+      "loss": 1.2341,
+      "theoretical_loss": 3.357818162639712,
+      "tokens_seen": 2901213184
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012201091317605522,
+      "loss": 1.3904,
+      "theoretical_loss": 3.357806519284117,
+      "tokens_seen": 2901344256
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012197079120526401,
+      "loss": 1.3286,
+      "theoretical_loss": 3.3577948766017878,
+      "tokens_seen": 2901475328
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012193066923447279,
+      "loss": 1.2897,
+      "theoretical_loss": 3.3577832345926546,
+      "tokens_seen": 2901606400
+    },
+    {
+      "epoch": 0.76,
+      "objective/train/advantage_avg": 0.490834504365921,
+      "objective/train/docs_used": 1055165,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3241896629333496,
+      "objective/train/original_loss": 1.3241896629333496,
+      "objective/train/theoretical_loss": 3.357771593256648,
+      "objective/train/tokens_used": 1272196576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24156969785690308,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503114461898804,
+      "objective/train/weighted_lm_loss": 1.3907438516616821,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.357771593256648,
+      "tokens_seen": 2901737472
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0001218905472636816,
+      "loss": 1.3412,
+      "theoretical_loss": 3.357771593256648,
+      "tokens_seen": 2901737472
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012185042529289039,
+      "loss": 1.2971,
+      "theoretical_loss": 3.3577599525936987,
+      "tokens_seen": 2901868544
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012181030332209918,
+      "loss": 1.4314,
+      "theoretical_loss": 3.357748312603737,
+      "tokens_seen": 2901999616
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012177018135130798,
+      "loss": 1.3674,
+      "theoretical_loss": 3.3577366732866945,
+      "tokens_seen": 2902130688
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012173005938051678,
+      "loss": 1.3021,
+      "theoretical_loss": 3.357725034642501,
+      "tokens_seen": 2902261760
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012168993740972557,
+      "loss": 1.3824,
+      "theoretical_loss": 3.357713396671088,
+      "tokens_seen": 2902392832
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012164981543893435,
+      "loss": 1.3998,
+      "theoretical_loss": 3.357701759372386,
+      "tokens_seen": 2902523904
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012160969346814316,
+      "loss": 1.3328,
+      "theoretical_loss": 3.357690122746325,
+      "tokens_seen": 2902654976
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012156957149735195,
+      "loss": 1.254,
+      "theoretical_loss": 3.3576784867928366,
+      "tokens_seen": 2902786048
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012152944952656076,
+      "loss": 1.2909,
+      "theoretical_loss": 3.3576668515118513,
+      "tokens_seen": 2902917120
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012148932755576954,
+      "loss": 1.3591,
+      "theoretical_loss": 3.3576552169033,
+      "tokens_seen": 2903048192
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012144920558497833,
+      "loss": 1.3495,
+      "theoretical_loss": 3.357643582967113,
+      "tokens_seen": 2903179264
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012140908361418714,
+      "loss": 1.3171,
+      "theoretical_loss": 3.357631949703222,
+      "tokens_seen": 2903310336
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012136896164339592,
+      "loss": 1.2361,
+      "theoretical_loss": 3.357620317111557,
+      "tokens_seen": 2903441408
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012132883967260472,
+      "loss": 1.3194,
+      "theoretical_loss": 3.357608685192049,
+      "tokens_seen": 2903572480
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012128871770181352,
+      "loss": 1.3261,
+      "theoretical_loss": 3.357597053944629,
+      "tokens_seen": 2903703552
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012124859573102232,
+      "loss": 1.3443,
+      "theoretical_loss": 3.3575854233692275,
+      "tokens_seen": 2903834624
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0001212084737602311,
+      "loss": 1.285,
+      "theoretical_loss": 3.3575737934657757,
+      "tokens_seen": 2903965696
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0001211683517894399,
+      "loss": 1.3531,
+      "theoretical_loss": 3.3575621642342046,
+      "tokens_seen": 2904096768
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0001211282298186487,
+      "loss": 1.3279,
+      "theoretical_loss": 3.357550535674444,
+      "tokens_seen": 2904227840
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012108810784785749,
+      "loss": 1.2987,
+      "theoretical_loss": 3.3575389077864264,
+      "tokens_seen": 2904358912
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012104798587706629,
+      "loss": 1.332,
+      "theoretical_loss": 3.357527280570081,
+      "tokens_seen": 2904489984
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012100786390627508,
+      "loss": 1.3014,
+      "theoretical_loss": 3.3575156540253404,
+      "tokens_seen": 2904621056
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012096774193548387,
+      "loss": 1.2864,
+      "theoretical_loss": 3.357504028152134,
+      "tokens_seen": 2904752128
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012092761996469266,
+      "loss": 1.3291,
+      "theoretical_loss": 3.3574924029503936,
+      "tokens_seen": 2904883200
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012088749799390146,
+      "loss": 1.3506,
+      "theoretical_loss": 3.35748077842005,
+      "tokens_seen": 2905014272
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012084737602311026,
+      "loss": 1.3041,
+      "theoretical_loss": 3.357469154561034,
+      "tokens_seen": 2905145344
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012080725405231906,
+      "loss": 1.3105,
+      "theoretical_loss": 3.357457531373276,
+      "tokens_seen": 2905276416
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012076713208152785,
+      "loss": 1.3076,
+      "theoretical_loss": 3.357445908856708,
+      "tokens_seen": 2905407488
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012072701011073664,
+      "loss": 1.3144,
+      "theoretical_loss": 3.35743428701126,
+      "tokens_seen": 2905538560
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012068688813994543,
+      "loss": 1.3223,
+      "theoretical_loss": 3.357422665836864,
+      "tokens_seen": 2905669632
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012064676616915424,
+      "loss": 1.3406,
+      "theoretical_loss": 3.35741104533345,
+      "tokens_seen": 2905800704
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012060664419836302,
+      "loss": 1.3033,
+      "theoretical_loss": 3.3573994255009496,
+      "tokens_seen": 2905931776
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012056652222757183,
+      "loss": 1.3637,
+      "theoretical_loss": 3.3573878063392937,
+      "tokens_seen": 2906062848
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012052640025678062,
+      "loss": 1.3422,
+      "theoretical_loss": 3.357376187848413,
+      "tokens_seen": 2906193920
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0001204862782859894,
+      "loss": 1.2916,
+      "theoretical_loss": 3.357364570028239,
+      "tokens_seen": 2906324992
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0001204461563151982,
+      "loss": 1.3105,
+      "theoretical_loss": 3.3573529528787023,
+      "tokens_seen": 2906456064
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.000120406034344407,
+      "loss": 1.297,
+      "theoretical_loss": 3.357341336399734,
+      "tokens_seen": 2906587136
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0001203659123736158,
+      "loss": 1.3364,
+      "theoretical_loss": 3.3573297205912658,
+      "tokens_seen": 2906718208
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012032579040282458,
+      "loss": 1.3401,
+      "theoretical_loss": 3.3573181054532277,
+      "tokens_seen": 2906849280
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012028566843203339,
+      "loss": 1.341,
+      "theoretical_loss": 3.357306490985552,
+      "tokens_seen": 2906980352
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012024554646124218,
+      "loss": 1.282,
+      "theoretical_loss": 3.3572948771881688,
+      "tokens_seen": 2907111424
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012020542449045097,
+      "loss": 1.2982,
+      "theoretical_loss": 3.3572832640610093,
+      "tokens_seen": 2907242496
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012016530251965977,
+      "loss": 1.269,
+      "theoretical_loss": 3.3572716516040053,
+      "tokens_seen": 2907373568
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012012518054886856,
+      "loss": 1.3294,
+      "theoretical_loss": 3.357260039817087,
+      "tokens_seen": 2907504640
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012008505857807737,
+      "loss": 1.2938,
+      "theoretical_loss": 3.3572484287001867,
+      "tokens_seen": 2907635712
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012004493660728616,
+      "loss": 1.3365,
+      "theoretical_loss": 3.357236818253234,
+      "tokens_seen": 2907766784
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00012000481463649494,
+      "loss": 1.3341,
+      "theoretical_loss": 3.3572252084761613,
+      "tokens_seen": 2907897856
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00011996469266570374,
+      "loss": 1.287,
+      "theoretical_loss": 3.357213599368899,
+      "tokens_seen": 2908028928
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00011992457069491254,
+      "loss": 1.2679,
+      "theoretical_loss": 3.357201990931379,
+      "tokens_seen": 2908160000
+    },
+    {
+      "epoch": 0.76,
+      "objective/train/advantage_avg": 0.49267274141311646,
+      "objective/train/docs_used": 1057433,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2761567831039429,
+      "objective/train/original_loss": 1.2761569023132324,
+      "objective/train/theoretical_loss": 3.357190383163532,
+      "objective/train/tokens_used": 1278750176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24308837950229645,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050503134727478,
+      "objective/train/weighted_lm_loss": 1.3406473398208618,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.357190383163532,
+      "tokens_seen": 2908291072
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00011988444872412133,
+      "loss": 1.2947,
+      "theoretical_loss": 3.357190383163532,
+      "tokens_seen": 2908291072
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00011984432675333012,
+      "loss": 1.3468,
+      "theoretical_loss": 3.357178776065289,
+      "tokens_seen": 2908422144
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00011980420478253893,
+      "loss": 1.2866,
+      "theoretical_loss": 3.3571671696365817,
+      "tokens_seen": 2908553216
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00011976408281174772,
+      "loss": 1.2941,
+      "theoretical_loss": 3.357155563877341,
+      "tokens_seen": 2908684288
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0001197239608409565,
+      "loss": 1.2594,
+      "theoretical_loss": 3.3571439587874985,
+      "tokens_seen": 2908815360
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0001196838388701653,
+      "loss": 1.3037,
+      "theoretical_loss": 3.3571323543669847,
+      "tokens_seen": 2908946432
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0001196437168993741,
+      "loss": 1.2319,
+      "theoretical_loss": 3.3571207506157315,
+      "tokens_seen": 2909077504
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0001196035949285829,
+      "loss": 1.3764,
+      "theoretical_loss": 3.35710914753367,
+      "tokens_seen": 2909208576
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00011956347295779169,
+      "loss": 1.3483,
+      "theoretical_loss": 3.357097545120731,
+      "tokens_seen": 2909339648
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00011952335098700048,
+      "loss": 1.2827,
+      "theoretical_loss": 3.3570859433768465,
+      "tokens_seen": 2909470720
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00011948322901620928,
+      "loss": 1.3308,
+      "theoretical_loss": 3.357074342301947,
+      "tokens_seen": 2909601792
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00011944310704541806,
+      "loss": 1.3734,
+      "theoretical_loss": 3.3570627418959647,
+      "tokens_seen": 2909732864
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00011940298507462687,
+      "loss": 1.2655,
+      "theoretical_loss": 3.3570511421588303,
+      "tokens_seen": 2909863936
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00011936286310383566,
+      "loss": 1.2636,
+      "theoretical_loss": 3.357039543090475,
+      "tokens_seen": 2909995008
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00011932274113304447,
+      "loss": 1.3712,
+      "theoretical_loss": 3.357027944690831,
+      "tokens_seen": 2910126080
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00011928261916225325,
+      "loss": 1.2403,
+      "theoretical_loss": 3.357016346959828,
+      "tokens_seen": 2910257152
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00011924249719146204,
+      "loss": 1.4366,
+      "theoretical_loss": 3.357004749897399,
+      "tokens_seen": 2910388224
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00011920237522067085,
+      "loss": 1.2626,
+      "theoretical_loss": 3.3569931535034745,
+      "tokens_seen": 2910519296
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00011916225324987964,
+      "loss": 1.3095,
+      "theoretical_loss": 3.356981557777986,
+      "tokens_seen": 2910650368
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00011912213127908843,
+      "loss": 1.3471,
+      "theoretical_loss": 3.356969962720865,
+      "tokens_seen": 2910781440
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00011908200930829723,
+      "loss": 1.3694,
+      "theoretical_loss": 3.3569583683320428,
+      "tokens_seen": 2910912512
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00011904188733750602,
+      "loss": 1.3889,
+      "theoretical_loss": 3.3569467746114503,
+      "tokens_seen": 2911043584
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00011900176536671481,
+      "loss": 1.3709,
+      "theoretical_loss": 3.3569351815590203,
+      "tokens_seen": 2911174656
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0001189616433959236,
+      "loss": 1.3159,
+      "theoretical_loss": 3.3569235891746825,
+      "tokens_seen": 2911305728
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00011892152142513241,
+      "loss": 1.306,
+      "theoretical_loss": 3.3569119974583694,
+      "tokens_seen": 2911436800
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0001188813994543412,
+      "loss": 1.3294,
+      "theoretical_loss": 3.3569004064100123,
+      "tokens_seen": 2911567872
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00011884127748355,
+      "loss": 1.2544,
+      "theoretical_loss": 3.3568888160295423,
+      "tokens_seen": 2911698944
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00011880115551275879,
+      "loss": 1.3003,
+      "theoretical_loss": 3.3568772263168913,
+      "tokens_seen": 2911830016
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00011876103354196758,
+      "loss": 1.2975,
+      "theoretical_loss": 3.3568656372719903,
+      "tokens_seen": 2911961088
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00011872091157117639,
+      "loss": 1.3285,
+      "theoretical_loss": 3.356854048894771,
+      "tokens_seen": 2912092160
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011868078960038517,
+      "loss": 1.4317,
+      "theoretical_loss": 3.3568424611851646,
+      "tokens_seen": 2912223232
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011864066762959397,
+      "loss": 1.324,
+      "theoretical_loss": 3.356830874143103,
+      "tokens_seen": 2912354304
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011860054565880277,
+      "loss": 1.3475,
+      "theoretical_loss": 3.356819287768518,
+      "tokens_seen": 2912485376
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011856042368801154,
+      "loss": 1.3338,
+      "theoretical_loss": 3.3568077020613405,
+      "tokens_seen": 2912616448
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011852030171722035,
+      "loss": 1.2853,
+      "theoretical_loss": 3.356796117021502,
+      "tokens_seen": 2912747520
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011848017974642914,
+      "loss": 1.3321,
+      "theoretical_loss": 3.3567845326489345,
+      "tokens_seen": 2912878592
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011844005777563795,
+      "loss": 1.2698,
+      "theoretical_loss": 3.356772948943569,
+      "tokens_seen": 2913009664
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011839993580484673,
+      "loss": 1.2248,
+      "theoretical_loss": 3.3567613659053372,
+      "tokens_seen": 2913140736
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011835981383405554,
+      "loss": 1.3469,
+      "theoretical_loss": 3.3567497835341715,
+      "tokens_seen": 2913271808
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011831969186326433,
+      "loss": 1.353,
+      "theoretical_loss": 3.356738201830002,
+      "tokens_seen": 2913402880
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011827956989247312,
+      "loss": 1.2815,
+      "theoretical_loss": 3.3567266207927613,
+      "tokens_seen": 2913533952
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011823944792168191,
+      "loss": 1.3929,
+      "theoretical_loss": 3.356715040422381,
+      "tokens_seen": 2913665024
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0001181993259508907,
+      "loss": 1.3184,
+      "theoretical_loss": 3.3567034607187924,
+      "tokens_seen": 2913796096
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011815920398009951,
+      "loss": 1.301,
+      "theoretical_loss": 3.356691881681927,
+      "tokens_seen": 2913927168
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011811908200930829,
+      "loss": 1.2729,
+      "theoretical_loss": 3.3566803033117165,
+      "tokens_seen": 2914058240
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0001180789600385171,
+      "loss": 1.3019,
+      "theoretical_loss": 3.3566687256080927,
+      "tokens_seen": 2914189312
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011803883806772589,
+      "loss": 1.2869,
+      "theoretical_loss": 3.3566571485709873,
+      "tokens_seen": 2914320384
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011799871609693468,
+      "loss": 1.3176,
+      "theoretical_loss": 3.356645572200332,
+      "tokens_seen": 2914451456
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011795859412614348,
+      "loss": 1.3739,
+      "theoretical_loss": 3.356633996496058,
+      "tokens_seen": 2914582528
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011791847215535227,
+      "loss": 1.2662,
+      "theoretical_loss": 3.3566224214580975,
+      "tokens_seen": 2914713600
+    },
+    {
+      "epoch": 0.77,
+      "objective/train/advantage_avg": 0.49482154846191406,
+      "objective/train/docs_used": 1059748,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3139622211456299,
+      "objective/train/original_loss": 1.3139622211456299,
+      "objective/train/theoretical_loss": 3.3566108470863814,
+      "objective/train/tokens_used": 1285303776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24505549669265747,
+      "objective/train/value_max": -0.5,
+      "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": 1.3805732727050781,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3566108470863814,
+      "tokens_seen": 2914844672
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011787835018456108,
+      "loss": 1.3549,
+      "theoretical_loss": 3.3566108470863814,
+      "tokens_seen": 2914844672
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011783822821376987,
+      "loss": 1.3605,
+      "theoretical_loss": 3.3565992733808425,
+      "tokens_seen": 2914975744
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011779810624297865,
+      "loss": 1.2938,
+      "theoretical_loss": 3.3565877003414117,
+      "tokens_seen": 2915106816
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011775798427218745,
+      "loss": 1.2828,
+      "theoretical_loss": 3.356576127968021,
+      "tokens_seen": 2915237888
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011771786230139625,
+      "loss": 1.4409,
+      "theoretical_loss": 3.3565645562606026,
+      "tokens_seen": 2915368960
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011767774033060505,
+      "loss": 1.3456,
+      "theoretical_loss": 3.3565529852190874,
+      "tokens_seen": 2915500032
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011763761835981383,
+      "loss": 1.3788,
+      "theoretical_loss": 3.356541414843407,
+      "tokens_seen": 2915631104
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011759749638902264,
+      "loss": 1.3191,
+      "theoretical_loss": 3.356529845133494,
+      "tokens_seen": 2915762176
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011755737441823143,
+      "loss": 1.3831,
+      "theoretical_loss": 3.3565182760892798,
+      "tokens_seen": 2915893248
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011751725244744021,
+      "loss": 1.2387,
+      "theoretical_loss": 3.3565067077106963,
+      "tokens_seen": 2916024320
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011747713047664902,
+      "loss": 1.3395,
+      "theoretical_loss": 3.356495139997675,
+      "tokens_seen": 2916155392
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011743700850585781,
+      "loss": 1.3231,
+      "theoretical_loss": 3.356483572950148,
+      "tokens_seen": 2916286464
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011739688653506662,
+      "loss": 1.2516,
+      "theoretical_loss": 3.356472006568047,
+      "tokens_seen": 2916417536
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0001173567645642754,
+      "loss": 1.2751,
+      "theoretical_loss": 3.3564604408513037,
+      "tokens_seen": 2916548608
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011731664259348419,
+      "loss": 1.309,
+      "theoretical_loss": 3.35644887579985,
+      "tokens_seen": 2916679680
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.000117276520622693,
+      "loss": 1.3751,
+      "theoretical_loss": 3.3564373114136177,
+      "tokens_seen": 2916810752
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011723639865190179,
+      "loss": 1.2646,
+      "theoretical_loss": 3.3564257476925388,
+      "tokens_seen": 2916941824
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011719627668111058,
+      "loss": 1.3319,
+      "theoretical_loss": 3.3564141846365447,
+      "tokens_seen": 2917072896
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011715615471031937,
+      "loss": 1.3456,
+      "theoretical_loss": 3.356402622245568,
+      "tokens_seen": 2917203968
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011711603273952818,
+      "loss": 1.3143,
+      "theoretical_loss": 3.35639106051954,
+      "tokens_seen": 2917335040
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011707591076873696,
+      "loss": 1.317,
+      "theoretical_loss": 3.3563794994583924,
+      "tokens_seen": 2917466112
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011703578879794575,
+      "loss": 1.3625,
+      "theoretical_loss": 3.356367939062058,
+      "tokens_seen": 2917597184
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011699566682715456,
+      "loss": 1.2961,
+      "theoretical_loss": 3.3563563793304683,
+      "tokens_seen": 2917728256
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011695554485636335,
+      "loss": 1.3433,
+      "theoretical_loss": 3.356344820263555,
+      "tokens_seen": 2917859328
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011691542288557214,
+      "loss": 1.325,
+      "theoretical_loss": 3.35633326186125,
+      "tokens_seen": 2917990400
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011687530091478094,
+      "loss": 1.3629,
+      "theoretical_loss": 3.356321704123485,
+      "tokens_seen": 2918121472
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011683517894398973,
+      "loss": 1.3602,
+      "theoretical_loss": 3.3563101470501926,
+      "tokens_seen": 2918252544
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011679505697319853,
+      "loss": 1.2831,
+      "theoretical_loss": 3.3562985906413045,
+      "tokens_seen": 2918383616
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011675493500240731,
+      "loss": 1.3512,
+      "theoretical_loss": 3.3562870348967526,
+      "tokens_seen": 2918514688
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011671481303161612,
+      "loss": 1.3101,
+      "theoretical_loss": 3.356275479816469,
+      "tokens_seen": 2918645760
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011667469106082491,
+      "loss": 1.3395,
+      "theoretical_loss": 3.3562639254003854,
+      "tokens_seen": 2918776832
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0001166345690900337,
+      "loss": 1.3802,
+      "theoretical_loss": 3.3562523716484343,
+      "tokens_seen": 2918907904
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0001165944471192425,
+      "loss": 1.2886,
+      "theoretical_loss": 3.356240818560547,
+      "tokens_seen": 2919038976
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011655432514845129,
+      "loss": 1.3302,
+      "theoretical_loss": 3.3562292661366557,
+      "tokens_seen": 2919170048
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0001165142031776601,
+      "loss": 1.3868,
+      "theoretical_loss": 3.3562177143766934,
+      "tokens_seen": 2919301120
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011647408120686888,
+      "loss": 1.3306,
+      "theoretical_loss": 3.356206163280591,
+      "tokens_seen": 2919432192
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011643395923607768,
+      "loss": 1.3652,
+      "theoretical_loss": 3.356194612848281,
+      "tokens_seen": 2919563264
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011639383726528648,
+      "loss": 1.3254,
+      "theoretical_loss": 3.356183063079695,
+      "tokens_seen": 2919694336
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011635371529449527,
+      "loss": 1.3422,
+      "theoretical_loss": 3.3561715139747657,
+      "tokens_seen": 2919825408
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011631359332370406,
+      "loss": 1.2937,
+      "theoretical_loss": 3.356159965533425,
+      "tokens_seen": 2919956480
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011627347135291285,
+      "loss": 1.3105,
+      "theoretical_loss": 3.356148417755605,
+      "tokens_seen": 2920087552
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011623334938212166,
+      "loss": 1.3626,
+      "theoretical_loss": 3.3561368706412376,
+      "tokens_seen": 2920218624
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011619322741133044,
+      "loss": 1.3959,
+      "theoretical_loss": 3.3561253241902547,
+      "tokens_seen": 2920349696
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011615310544053925,
+      "loss": 1.3621,
+      "theoretical_loss": 3.356113778402589,
+      "tokens_seen": 2920480768
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011611298346974804,
+      "loss": 1.3708,
+      "theoretical_loss": 3.3561022332781723,
+      "tokens_seen": 2920611840
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011607286149895683,
+      "loss": 1.321,
+      "theoretical_loss": 3.356090688816937,
+      "tokens_seen": 2920742912
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011603273952816562,
+      "loss": 1.324,
+      "theoretical_loss": 3.3560791450188145,
+      "tokens_seen": 2920873984
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011599261755737442,
+      "loss": 1.3581,
+      "theoretical_loss": 3.356067601883738,
+      "tokens_seen": 2921005056
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011595249558658322,
+      "loss": 1.3266,
+      "theoretical_loss": 3.356056059411639,
+      "tokens_seen": 2921136128
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011591237361579202,
+      "loss": 1.2754,
+      "theoretical_loss": 3.3560445176024496,
+      "tokens_seen": 2921267200
+    },
+    {
+      "epoch": 0.77,
+      "objective/train/advantage_avg": 0.4905661940574646,
+      "objective/train/docs_used": 1062024,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3939625024795532,
+      "objective/train/original_loss": 1.3939623832702637,
+      "objective/train/theoretical_loss": 3.3560329764561025,
+      "objective/train/tokens_used": 1291857376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24155829846858978,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0502846240997314,
+      "objective/train/weighted_lm_loss": 1.464076042175293,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3560329764561025,
+      "tokens_seen": 2921398272
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0001158722516450008,
+      "loss": 1.3659,
+      "theoretical_loss": 3.3560329764561025,
+      "tokens_seen": 2921398272
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0001158321296742096,
+      "loss": 1.337,
+      "theoretical_loss": 3.3560214359725298,
+      "tokens_seen": 2921529344
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0001157920077034184,
+      "loss": 1.2919,
+      "theoretical_loss": 3.3560098961516633,
+      "tokens_seen": 2921660416
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011575188573262719,
+      "loss": 1.3068,
+      "theoretical_loss": 3.3559983569934353,
+      "tokens_seen": 2921791488
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011571176376183598,
+      "loss": 1.3327,
+      "theoretical_loss": 3.3559868184977786,
+      "tokens_seen": 2921922560
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011567164179104479,
+      "loss": 1.3419,
+      "theoretical_loss": 3.3559752806646244,
+      "tokens_seen": 2922053632
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011563151982025358,
+      "loss": 1.3339,
+      "theoretical_loss": 3.3559637434939056,
+      "tokens_seen": 2922184704
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011559139784946236,
+      "loss": 1.3462,
+      "theoretical_loss": 3.355952206985555,
+      "tokens_seen": 2922315776
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011555127587867116,
+      "loss": 1.2444,
+      "theoretical_loss": 3.355940671139504,
+      "tokens_seen": 2922446848
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011551115390787996,
+      "loss": 1.3895,
+      "theoretical_loss": 3.355929135955685,
+      "tokens_seen": 2922577920
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011547103193708876,
+      "loss": 1.3521,
+      "theoretical_loss": 3.3559176014340304,
+      "tokens_seen": 2922708992
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011543090996629754,
+      "loss": 1.3614,
+      "theoretical_loss": 3.3559060675744727,
+      "tokens_seen": 2922840064
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011539078799550633,
+      "loss": 1.3282,
+      "theoretical_loss": 3.3558945343769437,
+      "tokens_seen": 2922971136
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011535066602471514,
+      "loss": 1.2938,
+      "theoretical_loss": 3.3558830018413763,
+      "tokens_seen": 2923102208
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011531054405392392,
+      "loss": 1.3236,
+      "theoretical_loss": 3.355871469967703,
+      "tokens_seen": 2923233280
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011527042208313273,
+      "loss": 1.356,
+      "theoretical_loss": 3.355859938755855,
+      "tokens_seen": 2923364352
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011523030011234152,
+      "loss": 1.3091,
+      "theoretical_loss": 3.355848408205766,
+      "tokens_seen": 2923495424
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011519017814155033,
+      "loss": 1.3244,
+      "theoretical_loss": 3.3558368783173673,
+      "tokens_seen": 2923626496
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0001151500561707591,
+      "loss": 1.3974,
+      "theoretical_loss": 3.355825349090592,
+      "tokens_seen": 2923757568
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0001151099341999679,
+      "loss": 1.3372,
+      "theoretical_loss": 3.3558138205253716,
+      "tokens_seen": 2923888640
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0001150698122291767,
+      "loss": 1.2855,
+      "theoretical_loss": 3.3558022926216395,
+      "tokens_seen": 2924019712
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0001150296902583855,
+      "loss": 1.2981,
+      "theoretical_loss": 3.3557907653793277,
+      "tokens_seen": 2924150784
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011498956828759429,
+      "loss": 1.4074,
+      "theoretical_loss": 3.3557792387983683,
+      "tokens_seen": 2924281856
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011494944631680308,
+      "loss": 1.3687,
+      "theoretical_loss": 3.355767712878694,
+      "tokens_seen": 2924412928
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011490932434601187,
+      "loss": 1.4727,
+      "theoretical_loss": 3.355756187620237,
+      "tokens_seen": 2924544000
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011486920237522068,
+      "loss": 1.3028,
+      "theoretical_loss": 3.3557446630229304,
+      "tokens_seen": 2924675072
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011482908040442946,
+      "loss": 1.3619,
+      "theoretical_loss": 3.355733139086706,
+      "tokens_seen": 2924806144
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011478895843363827,
+      "loss": 1.4061,
+      "theoretical_loss": 3.3557216158114964,
+      "tokens_seen": 2924937216
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011474883646284706,
+      "loss": 1.3931,
+      "theoretical_loss": 3.355710093197234,
+      "tokens_seen": 2925068288
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011470871449205585,
+      "loss": 1.3836,
+      "theoretical_loss": 3.3556985712438516,
+      "tokens_seen": 2925199360
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011466859252126464,
+      "loss": 1.2792,
+      "theoretical_loss": 3.355687049951281,
+      "tokens_seen": 2925330432
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011462847055047344,
+      "loss": 1.286,
+      "theoretical_loss": 3.3556755293194556,
+      "tokens_seen": 2925461504
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011458834857968224,
+      "loss": 1.3937,
+      "theoretical_loss": 3.3556640093483074,
+      "tokens_seen": 2925592576
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011454822660889102,
+      "loss": 1.3844,
+      "theoretical_loss": 3.3556524900377687,
+      "tokens_seen": 2925723648
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011450810463809983,
+      "loss": 1.4107,
+      "theoretical_loss": 3.3556409713877726,
+      "tokens_seen": 2925854720
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011446798266730862,
+      "loss": 1.3461,
+      "theoretical_loss": 3.355629453398251,
+      "tokens_seen": 2925985792
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011442786069651743,
+      "loss": 1.3206,
+      "theoretical_loss": 3.355617936069137,
+      "tokens_seen": 2926116864
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011438773872572621,
+      "loss": 1.2987,
+      "theoretical_loss": 3.3556064194003627,
+      "tokens_seen": 2926247936
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.000114347616754935,
+      "loss": 1.369,
+      "theoretical_loss": 3.3555949033918613,
+      "tokens_seen": 2926379008
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011430749478414381,
+      "loss": 1.2507,
+      "theoretical_loss": 3.3555833880435646,
+      "tokens_seen": 2926510080
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011426737281335259,
+      "loss": 1.3917,
+      "theoretical_loss": 3.3555718733554056,
+      "tokens_seen": 2926641152
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011422725084256139,
+      "loss": 1.2972,
+      "theoretical_loss": 3.355560359327317,
+      "tokens_seen": 2926772224
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011418712887177019,
+      "loss": 1.3211,
+      "theoretical_loss": 3.3555488459592313,
+      "tokens_seen": 2926903296
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011414700690097898,
+      "loss": 1.3298,
+      "theoretical_loss": 3.355537333251081,
+      "tokens_seen": 2927034368
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011410688493018777,
+      "loss": 1.3189,
+      "theoretical_loss": 3.3555258212027983,
+      "tokens_seen": 2927165440
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011406676295939656,
+      "loss": 1.3562,
+      "theoretical_loss": 3.3555143098143168,
+      "tokens_seen": 2927296512
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011402664098860537,
+      "loss": 1.3129,
+      "theoretical_loss": 3.355502799085569,
+      "tokens_seen": 2927427584
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011398651901781416,
+      "loss": 1.2954,
+      "theoretical_loss": 3.3554912890164865,
+      "tokens_seen": 2927558656
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011394639704702294,
+      "loss": 1.3394,
+      "theoretical_loss": 3.355479779607003,
+      "tokens_seen": 2927689728
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011390627507623175,
+      "loss": 1.2706,
+      "theoretical_loss": 3.3554682708570507,
+      "tokens_seen": 2927820800
+    },
+    {
+      "epoch": 0.77,
+      "objective/train/advantage_avg": 0.4944932162761688,
+      "objective/train/docs_used": 1064246,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2802696228027344,
+      "objective/train/original_loss": 1.2802696228027344,
+      "objective/train/theoretical_loss": 3.3554567627665626,
+      "objective/train/tokens_used": 1298410976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24476492404937744,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506936311721802,
+      "objective/train/weighted_lm_loss": 1.3451051712036133,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3554567627665626,
+      "tokens_seen": 2927951872
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011386615310544054,
+      "loss": 1.3362,
+      "theoretical_loss": 3.3554567627665626,
+      "tokens_seen": 2927951872
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011382603113464933,
+      "loss": 1.3309,
+      "theoretical_loss": 3.355445255335471,
+      "tokens_seen": 2928082944
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011378590916385813,
+      "loss": 1.3522,
+      "theoretical_loss": 3.3554337485637094,
+      "tokens_seen": 2928214016
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011374578719306693,
+      "loss": 1.2429,
+      "theoretical_loss": 3.3554222424512092,
+      "tokens_seen": 2928345088
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00011370566522227573,
+      "loss": 1.3317,
+      "theoretical_loss": 3.3554107369979045,
+      "tokens_seen": 2928476160
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0001136655432514845,
+      "loss": 1.3421,
+      "theoretical_loss": 3.3553992322037267,
+      "tokens_seen": 2928607232
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011362542128069331,
+      "loss": 1.3128,
+      "theoretical_loss": 3.3553877280686097,
+      "tokens_seen": 2928738304
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.0001135852993099021,
+      "loss": 1.3491,
+      "theoretical_loss": 3.355376224592486,
+      "tokens_seen": 2928869376
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011354517733911091,
+      "loss": 1.347,
+      "theoretical_loss": 3.3553647217752878,
+      "tokens_seen": 2929000448
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011350505536831969,
+      "loss": 1.3692,
+      "theoretical_loss": 3.3553532196169487,
+      "tokens_seen": 2929131520
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.0001134649333975285,
+      "loss": 1.2657,
+      "theoretical_loss": 3.3553417181174003,
+      "tokens_seen": 2929262592
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011342481142673729,
+      "loss": 1.3231,
+      "theoretical_loss": 3.3553302172765767,
+      "tokens_seen": 2929393664
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011338468945594607,
+      "loss": 1.363,
+      "theoretical_loss": 3.3553187170944097,
+      "tokens_seen": 2929524736
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011334456748515487,
+      "loss": 1.2515,
+      "theoretical_loss": 3.3553072175708327,
+      "tokens_seen": 2929655808
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011330444551436367,
+      "loss": 1.3557,
+      "theoretical_loss": 3.3552957187057784,
+      "tokens_seen": 2929786880
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011326432354357247,
+      "loss": 1.3003,
+      "theoretical_loss": 3.355284220499179,
+      "tokens_seen": 2929917952
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011322420157278125,
+      "loss": 1.2909,
+      "theoretical_loss": 3.3552727229509687,
+      "tokens_seen": 2930049024
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011318407960199004,
+      "loss": 1.3344,
+      "theoretical_loss": 3.3552612260610792,
+      "tokens_seen": 2930180096
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011314395763119885,
+      "loss": 1.2885,
+      "theoretical_loss": 3.3552497298294437,
+      "tokens_seen": 2930311168
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011310383566040764,
+      "loss": 1.2858,
+      "theoretical_loss": 3.355238234255995,
+      "tokens_seen": 2930442240
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011306371368961644,
+      "loss": 1.3603,
+      "theoretical_loss": 3.3552267393406665,
+      "tokens_seen": 2930573312
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011302359171882523,
+      "loss": 1.3379,
+      "theoretical_loss": 3.3552152450833903,
+      "tokens_seen": 2930704384
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011298346974803404,
+      "loss": 1.3264,
+      "theoretical_loss": 3.3552037514840998,
+      "tokens_seen": 2930835456
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011294334777724281,
+      "loss": 1.3564,
+      "theoretical_loss": 3.355192258542728,
+      "tokens_seen": 2930966528
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011290322580645161,
+      "loss": 1.2795,
+      "theoretical_loss": 3.355180766259207,
+      "tokens_seen": 2931097600
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011286310383566041,
+      "loss": 1.4272,
+      "theoretical_loss": 3.3551692746334707,
+      "tokens_seen": 2931228672
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.0001128229818648692,
+      "loss": 1.3313,
+      "theoretical_loss": 3.355157783665452,
+      "tokens_seen": 2931359744
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.000112782859894078,
+      "loss": 1.3363,
+      "theoretical_loss": 3.355146293355083,
+      "tokens_seen": 2931490816
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011274273792328679,
+      "loss": 1.2525,
+      "theoretical_loss": 3.355134803702297,
+      "tokens_seen": 2931621888
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011270261595249558,
+      "loss": 1.3643,
+      "theoretical_loss": 3.3551233147070274,
+      "tokens_seen": 2931752960
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011266249398170439,
+      "loss": 1.3055,
+      "theoretical_loss": 3.3551118263692072,
+      "tokens_seen": 2931884032
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011262237201091317,
+      "loss": 1.4041,
+      "theoretical_loss": 3.355100338688769,
+      "tokens_seen": 2932015104
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011258225004012198,
+      "loss": 1.2436,
+      "theoretical_loss": 3.355088851665646,
+      "tokens_seen": 2932146176
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011254212806933077,
+      "loss": 1.2608,
+      "theoretical_loss": 3.3550773652997705,
+      "tokens_seen": 2932277248
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011250200609853958,
+      "loss": 1.2964,
+      "theoretical_loss": 3.3550658795910766,
+      "tokens_seen": 2932408320
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011246188412774835,
+      "loss": 1.2851,
+      "theoretical_loss": 3.3550543945394966,
+      "tokens_seen": 2932539392
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011242176215695715,
+      "loss": 1.2825,
+      "theoretical_loss": 3.355042910144964,
+      "tokens_seen": 2932670464
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011238164018616595,
+      "loss": 1.3892,
+      "theoretical_loss": 3.355031426407412,
+      "tokens_seen": 2932801536
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011234151821537473,
+      "loss": 1.2802,
+      "theoretical_loss": 3.3550199433267727,
+      "tokens_seen": 2932932608
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011230139624458354,
+      "loss": 1.2754,
+      "theoretical_loss": 3.35500846090298,
+      "tokens_seen": 2933063680
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011226127427379233,
+      "loss": 1.3065,
+      "theoretical_loss": 3.354996979135967,
+      "tokens_seen": 2933194752
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011222115230300112,
+      "loss": 1.274,
+      "theoretical_loss": 3.354985498025666,
+      "tokens_seen": 2933325824
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011218103033220992,
+      "loss": 1.3968,
+      "theoretical_loss": 3.354974017572011,
+      "tokens_seen": 2933456896
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011214090836141871,
+      "loss": 1.3347,
+      "theoretical_loss": 3.3549625377749344,
+      "tokens_seen": 2933587968
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011210078639062752,
+      "loss": 1.3263,
+      "theoretical_loss": 3.35495105863437,
+      "tokens_seen": 2933719040
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011206066441983631,
+      "loss": 1.2998,
+      "theoretical_loss": 3.3549395801502504,
+      "tokens_seen": 2933850112
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.0001120205424490451,
+      "loss": 1.3189,
+      "theoretical_loss": 3.3549281023225093,
+      "tokens_seen": 2933981184
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.0001119804204782539,
+      "loss": 1.3868,
+      "theoretical_loss": 3.354916625151079,
+      "tokens_seen": 2934112256
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011194029850746269,
+      "loss": 1.35,
+      "theoretical_loss": 3.3549051486358934,
+      "tokens_seen": 2934243328
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011190017653667148,
+      "loss": 1.2856,
+      "theoretical_loss": 3.3548936727768854,
+      "tokens_seen": 2934374400
+    },
+    {
+      "epoch": 0.78,
+      "objective/train/advantage_avg": 0.4888792037963867,
+      "objective/train/docs_used": 1066769,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3082714080810547,
+      "objective/train/original_loss": 1.3082714080810547,
+      "objective/train/theoretical_loss": 3.354882197573988,
+      "objective/train/tokens_used": 1304964576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.23990017175674438,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0501073598861694,
+      "objective/train/weighted_lm_loss": 1.3736592531204224,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.354882197573988,
+      "tokens_seen": 2934505472
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011186005456588027,
+      "loss": 1.3286,
+      "theoretical_loss": 3.354882197573988,
+      "tokens_seen": 2934505472
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011181993259508908,
+      "loss": 1.2853,
+      "theoretical_loss": 3.3548707230271346,
+      "tokens_seen": 2934636544
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011177981062429787,
+      "loss": 1.289,
+      "theoretical_loss": 3.3548592491362585,
+      "tokens_seen": 2934767616
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011173968865350665,
+      "loss": 1.3446,
+      "theoretical_loss": 3.3548477759012925,
+      "tokens_seen": 2934898688
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011169956668271546,
+      "loss": 1.3384,
+      "theoretical_loss": 3.35483630332217,
+      "tokens_seen": 2935029760
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011165944471192425,
+      "loss": 1.2934,
+      "theoretical_loss": 3.3548248313988243,
+      "tokens_seen": 2935160832
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011161932274113306,
+      "loss": 1.2968,
+      "theoretical_loss": 3.354813360131189,
+      "tokens_seen": 2935291904
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011157920077034184,
+      "loss": 1.2545,
+      "theoretical_loss": 3.3548018895191967,
+      "tokens_seen": 2935422976
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011153907879955064,
+      "loss": 1.2335,
+      "theoretical_loss": 3.354790419562781,
+      "tokens_seen": 2935554048
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011149895682875944,
+      "loss": 1.4127,
+      "theoretical_loss": 3.354778950261875,
+      "tokens_seen": 2935685120
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011145883485796821,
+      "loss": 1.3676,
+      "theoretical_loss": 3.354767481616412,
+      "tokens_seen": 2935816192
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011141871288717702,
+      "loss": 1.291,
+      "theoretical_loss": 3.354756013626325,
+      "tokens_seen": 2935947264
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011137859091638581,
+      "loss": 1.2311,
+      "theoretical_loss": 3.354744546291548,
+      "tokens_seen": 2936078336
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011133846894559462,
+      "loss": 1.2573,
+      "theoretical_loss": 3.354733079612014,
+      "tokens_seen": 2936209408
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.0001112983469748034,
+      "loss": 1.3057,
+      "theoretical_loss": 3.3547216135876563,
+      "tokens_seen": 2936340480
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011125822500401219,
+      "loss": 1.2491,
+      "theoretical_loss": 3.354710148218408,
+      "tokens_seen": 2936471552
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.000111218103033221,
+      "loss": 1.2654,
+      "theoretical_loss": 3.3546986835042025,
+      "tokens_seen": 2936602624
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011117798106242979,
+      "loss": 1.3454,
+      "theoretical_loss": 3.3546872194449735,
+      "tokens_seen": 2936733696
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011113785909163858,
+      "loss": 1.301,
+      "theoretical_loss": 3.3546757560406535,
+      "tokens_seen": 2936864768
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011109773712084738,
+      "loss": 1.2551,
+      "theoretical_loss": 3.3546642932911768,
+      "tokens_seen": 2936995840
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011105761515005618,
+      "loss": 1.3242,
+      "theoretical_loss": 3.3546528311964767,
+      "tokens_seen": 2937126912
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011101749317926496,
+      "loss": 1.2885,
+      "theoretical_loss": 3.354641369756486,
+      "tokens_seen": 2937257984
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011097737120847375,
+      "loss": 1.3822,
+      "theoretical_loss": 3.3546299089711384,
+      "tokens_seen": 2937389056
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011093724923768256,
+      "loss": 1.3407,
+      "theoretical_loss": 3.3546184488403674,
+      "tokens_seen": 2937520128
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011089712726689135,
+      "loss": 1.363,
+      "theoretical_loss": 3.3546069893641066,
+      "tokens_seen": 2937651200
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011085700529610015,
+      "loss": 1.2913,
+      "theoretical_loss": 3.3545955305422885,
+      "tokens_seen": 2937782272
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011081688332530894,
+      "loss": 1.291,
+      "theoretical_loss": 3.3545840723748475,
+      "tokens_seen": 2937913344
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011077676135451773,
+      "loss": 1.3234,
+      "theoretical_loss": 3.3545726148617163,
+      "tokens_seen": 2938044416
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011073663938372654,
+      "loss": 1.3438,
+      "theoretical_loss": 3.354561158002829,
+      "tokens_seen": 2938175488
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011069651741293532,
+      "loss": 1.3861,
+      "theoretical_loss": 3.3545497017981187,
+      "tokens_seen": 2938306560
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011065639544214412,
+      "loss": 1.3818,
+      "theoretical_loss": 3.354538246247519,
+      "tokens_seen": 2938437632
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011061627347135292,
+      "loss": 1.3126,
+      "theoretical_loss": 3.3545267913509638,
+      "tokens_seen": 2938568704
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011057615150056171,
+      "loss": 1.3023,
+      "theoretical_loss": 3.354515337108386,
+      "tokens_seen": 2938699776
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.0001105360295297705,
+      "loss": 1.3126,
+      "theoretical_loss": 3.3545038835197185,
+      "tokens_seen": 2938830848
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.0001104959075589793,
+      "loss": 1.2826,
+      "theoretical_loss": 3.3544924305848958,
+      "tokens_seen": 2938961920
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.0001104557855881881,
+      "loss": 1.3533,
+      "theoretical_loss": 3.3544809783038514,
+      "tokens_seen": 2939092992
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011041566361739688,
+      "loss": 1.2844,
+      "theoretical_loss": 3.3544695266765183,
+      "tokens_seen": 2939224064
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011037554164660569,
+      "loss": 1.3606,
+      "theoretical_loss": 3.354458075702831,
+      "tokens_seen": 2939355136
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011033541967581448,
+      "loss": 1.3555,
+      "theoretical_loss": 3.3544466253827214,
+      "tokens_seen": 2939486208
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011029529770502327,
+      "loss": 1.3302,
+      "theoretical_loss": 3.3544351757161244,
+      "tokens_seen": 2939617280
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011025517573423206,
+      "loss": 1.3849,
+      "theoretical_loss": 3.354423726702973,
+      "tokens_seen": 2939748352
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011021505376344086,
+      "loss": 1.3133,
+      "theoretical_loss": 3.3544122783432013,
+      "tokens_seen": 2939879424
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011017493179264966,
+      "loss": 1.316,
+      "theoretical_loss": 3.3544008306367425,
+      "tokens_seen": 2940010496
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011013480982185846,
+      "loss": 1.3117,
+      "theoretical_loss": 3.35438938358353,
+      "tokens_seen": 2940141568
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011009468785106725,
+      "loss": 1.3198,
+      "theoretical_loss": 3.354377937183498,
+      "tokens_seen": 2940272640
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011005456588027604,
+      "loss": 1.293,
+      "theoretical_loss": 3.354366491436579,
+      "tokens_seen": 2940403712
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00011001444390948483,
+      "loss": 1.269,
+      "theoretical_loss": 3.3543550463427083,
+      "tokens_seen": 2940534784
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00010997432193869363,
+      "loss": 1.3067,
+      "theoretical_loss": 3.354343601901818,
+      "tokens_seen": 2940665856
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00010993419996790242,
+      "loss": 1.3136,
+      "theoretical_loss": 3.3543321581138423,
+      "tokens_seen": 2940796928
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00010989407799711123,
+      "loss": 1.3104,
+      "theoretical_loss": 3.354320714978715,
+      "tokens_seen": 2940928000
+    },
+    {
+      "epoch": 0.78,
+      "objective/train/advantage_avg": 0.4926430284976959,
+      "objective/train/docs_used": 1069310,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2792717218399048,
+      "objective/train/original_loss": 1.2792718410491943,
+      "objective/train/theoretical_loss": 3.35430927249637,
+      "objective/train/tokens_used": 1311518176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24302135407924652,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504997968673706,
+      "objective/train/weighted_lm_loss": 1.3438254594802856,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.35430927249637,
+      "tokens_seen": 2941059072
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00010985395602632002,
+      "loss": 1.3385,
+      "theoretical_loss": 3.35430927249637,
+      "tokens_seen": 2941059072
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.0001098138340555288,
+      "loss": 1.2607,
+      "theoretical_loss": 3.35429783066674,
+      "tokens_seen": 2941190144
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.0001097737120847376,
+      "loss": 1.3436,
+      "theoretical_loss": 3.3542863894897597,
+      "tokens_seen": 2941321216
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.0001097335901139464,
+      "loss": 1.3022,
+      "theoretical_loss": 3.3542749489653625,
+      "tokens_seen": 2941452288
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.0001096934681431552,
+      "loss": 1.3673,
+      "theoretical_loss": 3.354263509093482,
+      "tokens_seen": 2941583360
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00010965334617236398,
+      "loss": 1.3429,
+      "theoretical_loss": 3.3542520698740517,
+      "tokens_seen": 2941714432
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00010961322420157279,
+      "loss": 1.2402,
+      "theoretical_loss": 3.3542406313070057,
+      "tokens_seen": 2941845504
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00010957310223078158,
+      "loss": 1.3064,
+      "theoretical_loss": 3.3542291933922774,
+      "tokens_seen": 2941976576
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00010953298025999036,
+      "loss": 1.3513,
+      "theoretical_loss": 3.3542177561298008,
+      "tokens_seen": 2942107648
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00010949285828919917,
+      "loss": 1.3901,
+      "theoretical_loss": 3.3542063195195095,
+      "tokens_seen": 2942238720
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00010945273631840796,
+      "loss": 1.3281,
+      "theoretical_loss": 3.3541948835613375,
+      "tokens_seen": 2942369792
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00010941261434761677,
+      "loss": 1.242,
+      "theoretical_loss": 3.354183448255218,
+      "tokens_seen": 2942500864
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00010937249237682555,
+      "loss": 1.2935,
+      "theoretical_loss": 3.3541720136010857,
+      "tokens_seen": 2942631936
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00010933237040603435,
+      "loss": 1.3298,
+      "theoretical_loss": 3.3541605795988736,
+      "tokens_seen": 2942763008
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00010929224843524314,
+      "loss": 1.3064,
+      "theoretical_loss": 3.3541491462485156,
+      "tokens_seen": 2942894080
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00010925212646445194,
+      "loss": 1.2616,
+      "theoretical_loss": 3.354137713549946,
+      "tokens_seen": 2943025152
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00010921200449366073,
+      "loss": 1.3197,
+      "theoretical_loss": 3.354126281503098,
+      "tokens_seen": 2943156224
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00010917188252286952,
+      "loss": 1.3051,
+      "theoretical_loss": 3.354114850107906,
+      "tokens_seen": 2943287296
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00010913176055207833,
+      "loss": 1.3752,
+      "theoretical_loss": 3.354103419364303,
+      "tokens_seen": 2943418368
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00010909163858128711,
+      "loss": 1.3094,
+      "theoretical_loss": 3.3540919892722236,
+      "tokens_seen": 2943549440
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.0001090515166104959,
+      "loss": 1.3321,
+      "theoretical_loss": 3.3540805598316012,
+      "tokens_seen": 2943680512
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00010901139463970471,
+      "loss": 1.3093,
+      "theoretical_loss": 3.3540691310423707,
+      "tokens_seen": 2943811584
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.0001089712726689135,
+      "loss": 1.3636,
+      "theoretical_loss": 3.3540577029044645,
+      "tokens_seen": 2943942656
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.0001089311506981223,
+      "loss": 1.4286,
+      "theoretical_loss": 3.3540462754178173,
+      "tokens_seen": 2944073728
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00010889102872733109,
+      "loss": 1.3191,
+      "theoretical_loss": 3.354034848582363,
+      "tokens_seen": 2944204800
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00010885090675653989,
+      "loss": 1.4032,
+      "theoretical_loss": 3.3540234223980354,
+      "tokens_seen": 2944335872
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00010881078478574869,
+      "loss": 1.3879,
+      "theoretical_loss": 3.354011996864768,
+      "tokens_seen": 2944466944
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00010877066281495746,
+      "loss": 1.4092,
+      "theoretical_loss": 3.3540005719824952,
+      "tokens_seen": 2944598016
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00010873054084416627,
+      "loss": 1.3653,
+      "theoretical_loss": 3.353989147751151,
+      "tokens_seen": 2944729088
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00010869041887337506,
+      "loss": 1.2952,
+      "theoretical_loss": 3.3539777241706696,
+      "tokens_seen": 2944860160
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00010865029690258386,
+      "loss": 1.334,
+      "theoretical_loss": 3.353966301240984,
+      "tokens_seen": 2944991232
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00010861017493179265,
+      "loss": 1.3402,
+      "theoretical_loss": 3.3539548789620284,
+      "tokens_seen": 2945122304
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010857005296100144,
+      "loss": 1.3876,
+      "theoretical_loss": 3.3539434573337377,
+      "tokens_seen": 2945253376
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010852993099021025,
+      "loss": 1.3603,
+      "theoretical_loss": 3.3539320363560448,
+      "tokens_seen": 2945384448
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010848980901941903,
+      "loss": 1.3086,
+      "theoretical_loss": 3.3539206160288844,
+      "tokens_seen": 2945515520
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010844968704862783,
+      "loss": 1.294,
+      "theoretical_loss": 3.3539091963521903,
+      "tokens_seen": 2945646592
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010840956507783663,
+      "loss": 1.3554,
+      "theoretical_loss": 3.353897777325896,
+      "tokens_seen": 2945777664
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010836944310704543,
+      "loss": 1.2292,
+      "theoretical_loss": 3.3538863589499366,
+      "tokens_seen": 2945908736
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010832932113625421,
+      "loss": 1.3768,
+      "theoretical_loss": 3.353874941224245,
+      "tokens_seen": 2946039808
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.000108289199165463,
+      "loss": 1.3189,
+      "theoretical_loss": 3.353863524148756,
+      "tokens_seen": 2946170880
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010824907719467181,
+      "loss": 1.3566,
+      "theoretical_loss": 3.3538521077234034,
+      "tokens_seen": 2946301952
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010820895522388059,
+      "loss": 1.3325,
+      "theoretical_loss": 3.353840691948121,
+      "tokens_seen": 2946433024
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0001081688332530894,
+      "loss": 1.2718,
+      "theoretical_loss": 3.353829276822843,
+      "tokens_seen": 2946564096
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010812871128229819,
+      "loss": 1.3323,
+      "theoretical_loss": 3.353817862347504,
+      "tokens_seen": 2946695168
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010808858931150698,
+      "loss": 1.3484,
+      "theoretical_loss": 3.3538064485220374,
+      "tokens_seen": 2946826240
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010804846734071577,
+      "loss": 1.2889,
+      "theoretical_loss": 3.3537950353463777,
+      "tokens_seen": 2946957312
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010800834536992457,
+      "loss": 1.3361,
+      "theoretical_loss": 3.3537836228204587,
+      "tokens_seen": 2947088384
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010796822339913337,
+      "loss": 1.3224,
+      "theoretical_loss": 3.353772210944215,
+      "tokens_seen": 2947219456
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010792810142834217,
+      "loss": 1.2424,
+      "theoretical_loss": 3.3537607997175805,
+      "tokens_seen": 2947350528
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010788797945755096,
+      "loss": 1.3133,
+      "theoretical_loss": 3.353749389140489,
+      "tokens_seen": 2947481600
+    },
+    {
+      "epoch": 0.79,
+      "objective/train/advantage_avg": 0.49143001437187195,
+      "objective/train/docs_used": 1071710,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3553062677383423,
+      "objective/train/original_loss": 1.3553060293197632,
+      "objective/train/theoretical_loss": 3.3537379792128745,
+      "objective/train/tokens_used": 1318071776,
+      "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": 1.0503734350204468,
+      "objective/train/weighted_lm_loss": 1.423631191253662,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3537379792128745,
+      "tokens_seen": 2947612672
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010784785748675975,
+      "loss": 1.3447,
+      "theoretical_loss": 3.3537379792128745,
+      "tokens_seen": 2947612672
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010780773551596854,
+      "loss": 1.2333,
+      "theoretical_loss": 3.353726569934672,
+      "tokens_seen": 2947743744
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010776761354517735,
+      "loss": 1.3302,
+      "theoretical_loss": 3.353715161305815,
+      "tokens_seen": 2947874816
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010772749157438613,
+      "loss": 1.2552,
+      "theoretical_loss": 3.353703753326238,
+      "tokens_seen": 2948005888
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010768736960359494,
+      "loss": 1.3213,
+      "theoretical_loss": 3.3536923459958756,
+      "tokens_seen": 2948136960
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010764724763280373,
+      "loss": 1.3752,
+      "theoretical_loss": 3.353680939314661,
+      "tokens_seen": 2948268032
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010760712566201251,
+      "loss": 1.3712,
+      "theoretical_loss": 3.353669533282529,
+      "tokens_seen": 2948399104
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010756700369122131,
+      "loss": 1.2781,
+      "theoretical_loss": 3.353658127899413,
+      "tokens_seen": 2948530176
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010752688172043011,
+      "loss": 1.3158,
+      "theoretical_loss": 3.353646723165249,
+      "tokens_seen": 2948661248
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010748675974963891,
+      "loss": 1.315,
+      "theoretical_loss": 3.3536353190799693,
+      "tokens_seen": 2948792320
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010744663777884769,
+      "loss": 1.2221,
+      "theoretical_loss": 3.3536239156435093,
+      "tokens_seen": 2948923392
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0001074065158080565,
+      "loss": 1.3265,
+      "theoretical_loss": 3.3536125128558028,
+      "tokens_seen": 2949054464
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010736639383726529,
+      "loss": 1.4626,
+      "theoretical_loss": 3.3536011107167845,
+      "tokens_seen": 2949185536
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010732627186647408,
+      "loss": 1.2331,
+      "theoretical_loss": 3.353589709226388,
+      "tokens_seen": 2949316608
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010728614989568288,
+      "loss": 1.3433,
+      "theoretical_loss": 3.353578308384548,
+      "tokens_seen": 2949447680
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010724602792489167,
+      "loss": 1.3063,
+      "theoretical_loss": 3.353566908191199,
+      "tokens_seen": 2949578752
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010720590595410048,
+      "loss": 1.2554,
+      "theoretical_loss": 3.3535555086462745,
+      "tokens_seen": 2949709824
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010716578398330926,
+      "loss": 1.367,
+      "theoretical_loss": 3.35354410974971,
+      "tokens_seen": 2949840896
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010712566201251805,
+      "loss": 1.2486,
+      "theoretical_loss": 3.3535327115014386,
+      "tokens_seen": 2949971968
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010708554004172685,
+      "loss": 1.2613,
+      "theoretical_loss": 3.3535213139013953,
+      "tokens_seen": 2950103040
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010704541807093565,
+      "loss": 1.2561,
+      "theoretical_loss": 3.3535099169495144,
+      "tokens_seen": 2950234112
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010700529610014444,
+      "loss": 1.3494,
+      "theoretical_loss": 3.35349852064573,
+      "tokens_seen": 2950365184
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010696517412935323,
+      "loss": 1.2937,
+      "theoretical_loss": 3.353487124989977,
+      "tokens_seen": 2950496256
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010692505215856204,
+      "loss": 1.3448,
+      "theoretical_loss": 3.3534757299821893,
+      "tokens_seen": 2950627328
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010688493018777083,
+      "loss": 1.2894,
+      "theoretical_loss": 3.3534643356223013,
+      "tokens_seen": 2950758400
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010684480821697961,
+      "loss": 1.3546,
+      "theoretical_loss": 3.3534529419102475,
+      "tokens_seen": 2950889472
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010680468624618842,
+      "loss": 1.3346,
+      "theoretical_loss": 3.3534415488459617,
+      "tokens_seen": 2951020544
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010676456427539721,
+      "loss": 1.2979,
+      "theoretical_loss": 3.353430156429379,
+      "tokens_seen": 2951151616
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.000106724442304606,
+      "loss": 1.2778,
+      "theoretical_loss": 3.353418764660434,
+      "tokens_seen": 2951282688
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0001066843203338148,
+      "loss": 1.3717,
+      "theoretical_loss": 3.353407373539061,
+      "tokens_seen": 2951413760
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010664419836302359,
+      "loss": 1.2985,
+      "theoretical_loss": 3.3533959830651936,
+      "tokens_seen": 2951544832
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0001066040763922324,
+      "loss": 1.3513,
+      "theoretical_loss": 3.353384593238767,
+      "tokens_seen": 2951675904
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010656395442144117,
+      "loss": 1.2882,
+      "theoretical_loss": 3.3533732040597157,
+      "tokens_seen": 2951806976
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010652383245064998,
+      "loss": 1.3197,
+      "theoretical_loss": 3.3533618155279736,
+      "tokens_seen": 2951938048
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010648371047985877,
+      "loss": 1.3562,
+      "theoretical_loss": 3.353350427643476,
+      "tokens_seen": 2952069120
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010644358850906758,
+      "loss": 1.2776,
+      "theoretical_loss": 3.3533390404061567,
+      "tokens_seen": 2952200192
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010640346653827636,
+      "loss": 1.3853,
+      "theoretical_loss": 3.3533276538159504,
+      "tokens_seen": 2952331264
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010636334456748515,
+      "loss": 1.2473,
+      "theoretical_loss": 3.3533162678727915,
+      "tokens_seen": 2952462336
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010632322259669396,
+      "loss": 1.3265,
+      "theoretical_loss": 3.3533048825766145,
+      "tokens_seen": 2952593408
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010628310062590274,
+      "loss": 1.2735,
+      "theoretical_loss": 3.3532934979273543,
+      "tokens_seen": 2952724480
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010624297865511154,
+      "loss": 1.3901,
+      "theoretical_loss": 3.353282113924945,
+      "tokens_seen": 2952855552
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010620285668432034,
+      "loss": 1.371,
+      "theoretical_loss": 3.3532707305693217,
+      "tokens_seen": 2952986624
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010616273471352913,
+      "loss": 1.3659,
+      "theoretical_loss": 3.353259347860418,
+      "tokens_seen": 2953117696
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010612261274273792,
+      "loss": 1.2356,
+      "theoretical_loss": 3.353247965798169,
+      "tokens_seen": 2953248768
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010608249077194671,
+      "loss": 1.375,
+      "theoretical_loss": 3.3532365843825094,
+      "tokens_seen": 2953379840
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010604236880115552,
+      "loss": 1.3108,
+      "theoretical_loss": 3.353225203613374,
+      "tokens_seen": 2953510912
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010600224683036431,
+      "loss": 1.3368,
+      "theoretical_loss": 3.3532138234906963,
+      "tokens_seen": 2953641984
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0001059621248595731,
+      "loss": 1.3046,
+      "theoretical_loss": 3.353202444014412,
+      "tokens_seen": 2953773056
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0001059220028887819,
+      "loss": 1.3784,
+      "theoretical_loss": 3.3531910651844554,
+      "tokens_seen": 2953904128
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010588188091799069,
+      "loss": 1.3873,
+      "theoretical_loss": 3.353179687000761,
+      "tokens_seen": 2954035200
+    },
+    {
+      "epoch": 0.79,
+      "objective/train/advantage_avg": 0.494637668132782,
+      "objective/train/docs_used": 1074062,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.257887840270996,
+      "objective/train/original_loss": 1.257887840270996,
+      "objective/train/theoretical_loss": 3.353168309463263,
+      "objective/train/tokens_used": 1324625376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24505630135536194,
+      "objective/train/value_max": -0.5,
+      "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": 1.3216443061828613,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.353168309463263,
+      "tokens_seen": 2954166272
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010584175894719948,
+      "loss": 1.3261,
+      "theoretical_loss": 3.353168309463263,
+      "tokens_seen": 2954166272
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010580163697640828,
+      "loss": 1.3007,
+      "theoretical_loss": 3.3531569325718973,
+      "tokens_seen": 2954297344
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010576151500561708,
+      "loss": 1.2978,
+      "theoretical_loss": 3.3531455563265973,
+      "tokens_seen": 2954428416
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010572139303482588,
+      "loss": 1.3034,
+      "theoretical_loss": 3.353134180727298,
+      "tokens_seen": 2954559488
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010568127106403466,
+      "loss": 1.3558,
+      "theoretical_loss": 3.353122805773934,
+      "tokens_seen": 2954690560
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010564114909324346,
+      "loss": 1.3557,
+      "theoretical_loss": 3.3531114314664405,
+      "tokens_seen": 2954821632
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010560102712245225,
+      "loss": 1.356,
+      "theoretical_loss": 3.3531000578047516,
+      "tokens_seen": 2954952704
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010556090515166106,
+      "loss": 1.3329,
+      "theoretical_loss": 3.353088684788802,
+      "tokens_seen": 2955083776
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010552078318086984,
+      "loss": 1.3445,
+      "theoretical_loss": 3.3530773124185274,
+      "tokens_seen": 2955214848
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010548066121007865,
+      "loss": 1.3065,
+      "theoretical_loss": 3.3530659406938605,
+      "tokens_seen": 2955345920
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010544053923928744,
+      "loss": 1.3317,
+      "theoretical_loss": 3.353054569614738,
+      "tokens_seen": 2955476992
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010540041726849623,
+      "loss": 1.2504,
+      "theoretical_loss": 3.3530431991810934,
+      "tokens_seen": 2955608064
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010536029529770502,
+      "loss": 1.2596,
+      "theoretical_loss": 3.3530318293928625,
+      "tokens_seen": 2955739136
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010532017332691382,
+      "loss": 1.2934,
+      "theoretical_loss": 3.3530204602499785,
+      "tokens_seen": 2955870208
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010528005135612262,
+      "loss": 1.2871,
+      "theoretical_loss": 3.353009091752378,
+      "tokens_seen": 2956001280
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0001052399293853314,
+      "loss": 1.3121,
+      "theoretical_loss": 3.3529977238999944,
+      "tokens_seen": 2956132352
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0001051998074145402,
+      "loss": 1.2672,
+      "theoretical_loss": 3.352986356692763,
+      "tokens_seen": 2956263424
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.000105159685443749,
+      "loss": 1.4026,
+      "theoretical_loss": 3.352974990130618,
+      "tokens_seen": 2956394496
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0001051195634729578,
+      "loss": 1.3819,
+      "theoretical_loss": 3.352963624213495,
+      "tokens_seen": 2956525568
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010507944150216659,
+      "loss": 1.3508,
+      "theoretical_loss": 3.352952258941329,
+      "tokens_seen": 2956656640
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010503931953137538,
+      "loss": 1.3004,
+      "theoretical_loss": 3.3529408943140537,
+      "tokens_seen": 2956787712
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010499919756058419,
+      "loss": 1.3498,
+      "theoretical_loss": 3.3529295303316045,
+      "tokens_seen": 2956918784
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010495907558979298,
+      "loss": 1.3273,
+      "theoretical_loss": 3.3529181669939168,
+      "tokens_seen": 2957049856
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010491895361900176,
+      "loss": 1.3515,
+      "theoretical_loss": 3.3529068043009245,
+      "tokens_seen": 2957180928
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010487883164821056,
+      "loss": 1.3654,
+      "theoretical_loss": 3.3528954422525628,
+      "tokens_seen": 2957312000
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010483870967741936,
+      "loss": 1.3505,
+      "theoretical_loss": 3.3528840808487668,
+      "tokens_seen": 2957443072
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010479858770662815,
+      "loss": 1.308,
+      "theoretical_loss": 3.3528727200894712,
+      "tokens_seen": 2957574144
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010475846573583694,
+      "loss": 1.307,
+      "theoretical_loss": 3.352861359974611,
+      "tokens_seen": 2957705216
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010471834376504575,
+      "loss": 1.2638,
+      "theoretical_loss": 3.3528500005041204,
+      "tokens_seen": 2957836288
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010467822179425454,
+      "loss": 1.3402,
+      "theoretical_loss": 3.352838641677935,
+      "tokens_seen": 2957967360
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010463809982346332,
+      "loss": 1.2491,
+      "theoretical_loss": 3.35282728349599,
+      "tokens_seen": 2958098432
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010459797785267213,
+      "loss": 1.2681,
+      "theoretical_loss": 3.3528159259582195,
+      "tokens_seen": 2958229504
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010455785588188092,
+      "loss": 1.2274,
+      "theoretical_loss": 3.352804569064559,
+      "tokens_seen": 2958360576
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010451773391108973,
+      "loss": 1.3584,
+      "theoretical_loss": 3.352793212814943,
+      "tokens_seen": 2958491648
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0001044776119402985,
+      "loss": 1.3322,
+      "theoretical_loss": 3.352781857209307,
+      "tokens_seen": 2958622720
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0001044374899695073,
+      "loss": 1.3655,
+      "theoretical_loss": 3.3527705022475853,
+      "tokens_seen": 2958753792
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0001043973679987161,
+      "loss": 1.3155,
+      "theoretical_loss": 3.3527591479297136,
+      "tokens_seen": 2958884864
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010435724602792488,
+      "loss": 1.3212,
+      "theoretical_loss": 3.3527477942556265,
+      "tokens_seen": 2959015936
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010431712405713369,
+      "loss": 1.2743,
+      "theoretical_loss": 3.3527364412252583,
+      "tokens_seen": 2959147008
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010427700208634248,
+      "loss": 1.3535,
+      "theoretical_loss": 3.3527250888385454,
+      "tokens_seen": 2959278080
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010423688011555129,
+      "loss": 1.3144,
+      "theoretical_loss": 3.3527137370954216,
+      "tokens_seen": 2959409152
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010419675814476007,
+      "loss": 1.3876,
+      "theoretical_loss": 3.3527023859958223,
+      "tokens_seen": 2959540224
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010415663617396886,
+      "loss": 1.3073,
+      "theoretical_loss": 3.352691035539683,
+      "tokens_seen": 2959671296
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010411651420317767,
+      "loss": 1.2935,
+      "theoretical_loss": 3.352679685726938,
+      "tokens_seen": 2959802368
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010407639223238646,
+      "loss": 1.4394,
+      "theoretical_loss": 3.3526683365575227,
+      "tokens_seen": 2959933440
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010403627026159525,
+      "loss": 1.3348,
+      "theoretical_loss": 3.352656988031372,
+      "tokens_seen": 2960064512
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010399614829080405,
+      "loss": 1.2949,
+      "theoretical_loss": 3.3526456401484213,
+      "tokens_seen": 2960195584
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010395602632001284,
+      "loss": 1.3223,
+      "theoretical_loss": 3.352634292908605,
+      "tokens_seen": 2960326656
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010391590434922163,
+      "loss": 1.3506,
+      "theoretical_loss": 3.3526229463118593,
+      "tokens_seen": 2960457728
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010387578237843042,
+      "loss": 1.2818,
+      "theoretical_loss": 3.3526116003581183,
+      "tokens_seen": 2960588800
+    },
+    {
+      "debugging/Compilability": 1.0,
+      "debugging/distinct-1-grams": 0.7459160326454821,
+      "debugging/entropy-1-grams": 5.28426573657414,
+      "debugging/length": 471.5,
+      "debugging/num_segments": 12,
+      "debugging/raw_token_scores_avg": 0.005323128309100866,
+      "debugging/raw_token_scores_std": 0.015938615426421165,
+      "debugging/score": 0.006897032404886403,
+      "debugging/score_std": 0.00540563131039379,
+      "epoch": 0.79,
+      "objective/train/advantage_avg": 0.49467670917510986,
+      "objective/train/docs_used": 1076444,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2147760391235352,
+      "objective/train/original_loss": 1.2147760391235352,
+      "objective/train/theoretical_loss": 3.3526002550473173,
+      "objective/train/tokens_used": 1331178976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24495922029018402,
+      "objective/train/value_max": -0.5,
+      "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": 1.2763735055923462,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3526002550473173,
+      "tokens_seen": 2960719872
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010383566040763923,
+      "loss": 1.2988,
+      "theoretical_loss": 3.3526002550473173,
+      "tokens_seen": 2960719872
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010379553843684802,
+      "loss": 1.3122,
+      "theoretical_loss": 3.3525889103793913,
+      "tokens_seen": 2960850944
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010375541646605682,
+      "loss": 1.3463,
+      "theoretical_loss": 3.3525775663542756,
+      "tokens_seen": 2960982016
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010371529449526561,
+      "loss": 1.3428,
+      "theoretical_loss": 3.3525662229719058,
+      "tokens_seen": 2961113088
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0001036751725244744,
+      "loss": 1.3368,
+      "theoretical_loss": 3.3525548802322165,
+      "tokens_seen": 2961244160
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010363505055368321,
+      "loss": 1.2109,
+      "theoretical_loss": 3.3525435381351425,
+      "tokens_seen": 2961375232
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00010359492858289199,
+      "loss": 1.3796,
+      "theoretical_loss": 3.35253219668062,
+      "tokens_seen": 2961506304
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0001035548066121008,
+      "loss": 1.2762,
+      "theoretical_loss": 3.352520855868583,
+      "tokens_seen": 2961637376
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010351468464130959,
+      "loss": 1.3473,
+      "theoretical_loss": 3.352509515698967,
+      "tokens_seen": 2961768448
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010347456267051837,
+      "loss": 1.2794,
+      "theoretical_loss": 3.352498176171708,
+      "tokens_seen": 2961899520
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010343444069972717,
+      "loss": 1.3627,
+      "theoretical_loss": 3.3524868372867402,
+      "tokens_seen": 2962030592
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010339431872893596,
+      "loss": 1.3734,
+      "theoretical_loss": 3.352475499043999,
+      "tokens_seen": 2962161664
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010335419675814477,
+      "loss": 1.2281,
+      "theoretical_loss": 3.3524641614434203,
+      "tokens_seen": 2962292736
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010331407478735355,
+      "loss": 1.3939,
+      "theoretical_loss": 3.3524528244849385,
+      "tokens_seen": 2962423808
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010327395281656236,
+      "loss": 1.2797,
+      "theoretical_loss": 3.3524414881684894,
+      "tokens_seen": 2962554880
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010323383084577115,
+      "loss": 1.3155,
+      "theoretical_loss": 3.3524301524940077,
+      "tokens_seen": 2962685952
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010319370887497994,
+      "loss": 1.2293,
+      "theoretical_loss": 3.352418817461429,
+      "tokens_seen": 2962817024
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010315358690418873,
+      "loss": 1.3064,
+      "theoretical_loss": 3.3524074830706883,
+      "tokens_seen": 2962948096
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010311346493339753,
+      "loss": 1.2102,
+      "theoretical_loss": 3.352396149321721,
+      "tokens_seen": 2963079168
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010307334296260633,
+      "loss": 1.2414,
+      "theoretical_loss": 3.352384816214463,
+      "tokens_seen": 2963210240
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010303322099181513,
+      "loss": 1.2999,
+      "theoretical_loss": 3.352373483748848,
+      "tokens_seen": 2963341312
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0001029930990210239,
+      "loss": 1.3347,
+      "theoretical_loss": 3.3523621519248135,
+      "tokens_seen": 2963472384
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010295297705023271,
+      "loss": 1.2765,
+      "theoretical_loss": 3.3523508207422927,
+      "tokens_seen": 2963603456
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0001029128550794415,
+      "loss": 1.2936,
+      "theoretical_loss": 3.352339490201222,
+      "tokens_seen": 2963734528
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0001028727331086503,
+      "loss": 1.2925,
+      "theoretical_loss": 3.3523281603015365,
+      "tokens_seen": 2963865600
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010283261113785909,
+      "loss": 1.2907,
+      "theoretical_loss": 3.3523168310431717,
+      "tokens_seen": 2963996672
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0001027924891670679,
+      "loss": 1.3096,
+      "theoretical_loss": 3.352305502426063,
+      "tokens_seen": 2964127744
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010275236719627669,
+      "loss": 1.3086,
+      "theoretical_loss": 3.352294174450145,
+      "tokens_seen": 2964258816
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010271224522548547,
+      "loss": 1.4007,
+      "theoretical_loss": 3.352282847115354,
+      "tokens_seen": 2964389888
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010267212325469427,
+      "loss": 1.2648,
+      "theoretical_loss": 3.352271520421625,
+      "tokens_seen": 2964520960
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010263200128390307,
+      "loss": 1.3085,
+      "theoretical_loss": 3.352260194368893,
+      "tokens_seen": 2964652032
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010259187931311187,
+      "loss": 1.2892,
+      "theoretical_loss": 3.352248868957094,
+      "tokens_seen": 2964783104
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010255175734232065,
+      "loss": 1.2535,
+      "theoretical_loss": 3.3522375441861634,
+      "tokens_seen": 2964914176
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010251163537152945,
+      "loss": 1.2705,
+      "theoretical_loss": 3.352226220056036,
+      "tokens_seen": 2965045248
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010247151340073825,
+      "loss": 1.3824,
+      "theoretical_loss": 3.3522148965666476,
+      "tokens_seen": 2965176320
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010243139142994703,
+      "loss": 1.2601,
+      "theoretical_loss": 3.352203573717934,
+      "tokens_seen": 2965307392
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010239126945915584,
+      "loss": 1.1982,
+      "theoretical_loss": 3.3521922515098295,
+      "tokens_seen": 2965438464
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010235114748836463,
+      "loss": 1.3206,
+      "theoretical_loss": 3.3521809299422705,
+      "tokens_seen": 2965569536
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010231102551757344,
+      "loss": 1.2304,
+      "theoretical_loss": 3.3521696090151925,
+      "tokens_seen": 2965700608
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010227090354678222,
+      "loss": 1.2781,
+      "theoretical_loss": 3.3521582887285306,
+      "tokens_seen": 2965831680
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010223078157599101,
+      "loss": 1.286,
+      "theoretical_loss": 3.3521469690822205,
+      "tokens_seen": 2965962752
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010219065960519981,
+      "loss": 1.2341,
+      "theoretical_loss": 3.3521356500761974,
+      "tokens_seen": 2966093824
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010215053763440861,
+      "loss": 1.2764,
+      "theoretical_loss": 3.352124331710397,
+      "tokens_seen": 2966224896
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0001021104156636174,
+      "loss": 1.4066,
+      "theoretical_loss": 3.3521130139847544,
+      "tokens_seen": 2966355968
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010207029369282619,
+      "loss": 1.2744,
+      "theoretical_loss": 3.3521016968992057,
+      "tokens_seen": 2966487040
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010203017172203499,
+      "loss": 1.2888,
+      "theoretical_loss": 3.352090380453686,
+      "tokens_seen": 2966618112
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010199004975124378,
+      "loss": 1.35,
+      "theoretical_loss": 3.352079064648131,
+      "tokens_seen": 2966749184
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010194992778045257,
+      "loss": 1.331,
+      "theoretical_loss": 3.3520677494824764,
+      "tokens_seen": 2966880256
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010190980580966138,
+      "loss": 1.367,
+      "theoretical_loss": 3.3520564349566575,
+      "tokens_seen": 2967011328
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010186968383887017,
+      "loss": 1.334,
+      "theoretical_loss": 3.35204512107061,
+      "tokens_seen": 2967142400
+    },
+    {
+      "epoch": 0.8,
+      "objective/train/advantage_avg": 0.4933345317840576,
+      "objective/train/docs_used": 1078969,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.257318139076233,
+      "objective/train/original_loss": 1.2573180198669434,
+      "objective/train/theoretical_loss": 3.3520338078242693,
+      "objective/train/tokens_used": 1337732576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24369396269321442,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050572395324707,
+      "objective/train/weighted_lm_loss": 1.3208434581756592,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3520338078242693,
+      "tokens_seen": 2967273472
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010182956186807896,
+      "loss": 1.3527,
+      "theoretical_loss": 3.3520338078242693,
+      "tokens_seen": 2967273472
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010178943989728776,
+      "loss": 1.3361,
+      "theoretical_loss": 3.352022495217571,
+      "tokens_seen": 2967404544
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010174931792649655,
+      "loss": 1.2751,
+      "theoretical_loss": 3.3520111832504504,
+      "tokens_seen": 2967535616
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010170919595570535,
+      "loss": 1.284,
+      "theoretical_loss": 3.351999871922844,
+      "tokens_seen": 2967666688
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010166907398491413,
+      "loss": 1.2296,
+      "theoretical_loss": 3.3519885612346867,
+      "tokens_seen": 2967797760
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010162895201412294,
+      "loss": 1.3538,
+      "theoretical_loss": 3.3519772511859145,
+      "tokens_seen": 2967928832
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010158883004333173,
+      "loss": 1.3272,
+      "theoretical_loss": 3.351965941776462,
+      "tokens_seen": 2968059904
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010154870807254051,
+      "loss": 1.2407,
+      "theoretical_loss": 3.3519546330062666,
+      "tokens_seen": 2968190976
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010150858610174932,
+      "loss": 1.2844,
+      "theoretical_loss": 3.351943324875262,
+      "tokens_seen": 2968322048
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010146846413095811,
+      "loss": 1.3444,
+      "theoretical_loss": 3.3519320173833855,
+      "tokens_seen": 2968453120
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010142834216016692,
+      "loss": 1.3382,
+      "theoretical_loss": 3.351920710530572,
+      "tokens_seen": 2968584192
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0001013882201893757,
+      "loss": 1.2997,
+      "theoretical_loss": 3.351909404316757,
+      "tokens_seen": 2968715264
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0001013480982185845,
+      "loss": 1.2805,
+      "theoretical_loss": 3.3518980987418763,
+      "tokens_seen": 2968846336
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0001013079762477933,
+      "loss": 1.2504,
+      "theoretical_loss": 3.351886793805866,
+      "tokens_seen": 2968977408
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010126785427700209,
+      "loss": 1.3166,
+      "theoretical_loss": 3.3518754895086613,
+      "tokens_seen": 2969108480
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010122773230621088,
+      "loss": 1.3347,
+      "theoretical_loss": 3.3518641858501983,
+      "tokens_seen": 2969239552
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010118761033541967,
+      "loss": 1.307,
+      "theoretical_loss": 3.3518528828304124,
+      "tokens_seen": 2969370624
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010114748836462848,
+      "loss": 1.2989,
+      "theoretical_loss": 3.351841580449239,
+      "tokens_seen": 2969501696
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010110736639383726,
+      "loss": 1.3058,
+      "theoretical_loss": 3.3518302787066148,
+      "tokens_seen": 2969632768
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010106724442304605,
+      "loss": 1.3065,
+      "theoretical_loss": 3.351818977602475,
+      "tokens_seen": 2969763840
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010102712245225486,
+      "loss": 1.2798,
+      "theoretical_loss": 3.351807677136755,
+      "tokens_seen": 2969894912
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010098700048146365,
+      "loss": 1.3856,
+      "theoretical_loss": 3.351796377309391,
+      "tokens_seen": 2970025984
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010094687851067244,
+      "loss": 1.3177,
+      "theoretical_loss": 3.351785078120319,
+      "tokens_seen": 2970157056
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010090675653988124,
+      "loss": 1.2703,
+      "theoretical_loss": 3.3517737795694744,
+      "tokens_seen": 2970288128
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010086663456909004,
+      "loss": 1.2656,
+      "theoretical_loss": 3.351762481656793,
+      "tokens_seen": 2970419200
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010082651259829884,
+      "loss": 1.2963,
+      "theoretical_loss": 3.3517511843822105,
+      "tokens_seen": 2970550272
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010078639062750762,
+      "loss": 1.2741,
+      "theoretical_loss": 3.351739887745663,
+      "tokens_seen": 2970681344
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010074626865671642,
+      "loss": 1.2618,
+      "theoretical_loss": 3.351728591747086,
+      "tokens_seen": 2970812416
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010070614668592521,
+      "loss": 1.3117,
+      "theoretical_loss": 3.3517172963864157,
+      "tokens_seen": 2970943488
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010066602471513402,
+      "loss": 1.2175,
+      "theoretical_loss": 3.3517060016635876,
+      "tokens_seen": 2971074560
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0001006259027443428,
+      "loss": 1.2973,
+      "theoretical_loss": 3.3516947075785377,
+      "tokens_seen": 2971205632
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0001005857807735516,
+      "loss": 1.348,
+      "theoretical_loss": 3.351683414131202,
+      "tokens_seen": 2971336704
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0001005456588027604,
+      "loss": 1.3317,
+      "theoretical_loss": 3.351672121321516,
+      "tokens_seen": 2971467776
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010050553683196918,
+      "loss": 1.3656,
+      "theoretical_loss": 3.351660829149416,
+      "tokens_seen": 2971598848
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010046541486117798,
+      "loss": 1.2811,
+      "theoretical_loss": 3.3516495376148376,
+      "tokens_seen": 2971729920
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010042529289038678,
+      "loss": 1.2924,
+      "theoretical_loss": 3.351638246717717,
+      "tokens_seen": 2971860992
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010038517091959558,
+      "loss": 1.2923,
+      "theoretical_loss": 3.3516269564579892,
+      "tokens_seen": 2971992064
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010034504894880436,
+      "loss": 1.2965,
+      "theoretical_loss": 3.351615666835591,
+      "tokens_seen": 2972123136
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010030492697801316,
+      "loss": 1.2539,
+      "theoretical_loss": 3.3516043778504585,
+      "tokens_seen": 2972254208
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010026480500722196,
+      "loss": 1.3665,
+      "theoretical_loss": 3.3515930895025265,
+      "tokens_seen": 2972385280
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010022468303643075,
+      "loss": 1.2505,
+      "theoretical_loss": 3.351581801791732,
+      "tokens_seen": 2972516352
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010018456106563955,
+      "loss": 1.3426,
+      "theoretical_loss": 3.351570514718011,
+      "tokens_seen": 2972647424
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010014443909484834,
+      "loss": 1.3172,
+      "theoretical_loss": 3.3515592282812987,
+      "tokens_seen": 2972778496
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010010431712405715,
+      "loss": 1.367,
+      "theoretical_loss": 3.351547942481531,
+      "tokens_seen": 2972909568
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010006419515326593,
+      "loss": 1.2772,
+      "theoretical_loss": 3.351536657318645,
+      "tokens_seen": 2973040640
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00010002407318247472,
+      "loss": 1.2489,
+      "theoretical_loss": 3.3515253727925756,
+      "tokens_seen": 2973171712
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.998395121168352e-05,
+      "loss": 1.3421,
+      "theoretical_loss": 3.3515140889032593,
+      "tokens_seen": 2973302784
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.994382924089232e-05,
+      "loss": 1.4,
+      "theoretical_loss": 3.3515028056506315,
+      "tokens_seen": 2973433856
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.990370727010111e-05,
+      "loss": 1.2991,
+      "theoretical_loss": 3.351491523034629,
+      "tokens_seen": 2973564928
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.98635852993099e-05,
+      "loss": 1.2641,
+      "theoretical_loss": 3.351480241055188,
+      "tokens_seen": 2973696000
+    },
+    {
+      "epoch": 0.8,
+      "objective/train/advantage_avg": 0.49271485209465027,
+      "objective/train/docs_used": 1081215,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.205015778541565,
+      "objective/train/original_loss": 1.205015778541565,
+      "objective/train/theoretical_loss": 3.3514689597122436,
+      "objective/train/tokens_used": 1344286176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24346449971199036,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505092144012451,
+      "objective/train/weighted_lm_loss": 1.2658560276031494,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3514689597122436,
+      "tokens_seen": 2973827072
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.98234633285187e-05,
+      "loss": 1.3573,
+      "theoretical_loss": 3.3514689597122436,
+      "tokens_seen": 2973827072
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.97833413577275e-05,
+      "loss": 1.2439,
+      "theoretical_loss": 3.351457679005732,
+      "tokens_seen": 2973958144
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.974321938693628e-05,
+      "loss": 1.3016,
+      "theoretical_loss": 3.3514463989355896,
+      "tokens_seen": 2974089216
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.970309741614509e-05,
+      "loss": 1.3322,
+      "theoretical_loss": 3.351435119501753,
+      "tokens_seen": 2974220288
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.966297544535388e-05,
+      "loss": 1.3276,
+      "theoretical_loss": 3.351423840704157,
+      "tokens_seen": 2974351360
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.962285347456267e-05,
+      "loss": 1.266,
+      "theoretical_loss": 3.3514125625427384,
+      "tokens_seen": 2974482432
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.958273150377147e-05,
+      "loss": 1.3232,
+      "theoretical_loss": 3.3514012850174333,
+      "tokens_seen": 2974613504
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.954260953298026e-05,
+      "loss": 1.311,
+      "theoretical_loss": 3.3513900081281776,
+      "tokens_seen": 2974744576
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.950248756218906e-05,
+      "loss": 1.2721,
+      "theoretical_loss": 3.3513787318749078,
+      "tokens_seen": 2974875648
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.946236559139784e-05,
+      "loss": 1.2701,
+      "theoretical_loss": 3.3513674562575595,
+      "tokens_seen": 2975006720
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.942224362060665e-05,
+      "loss": 1.3832,
+      "theoretical_loss": 3.351356181276069,
+      "tokens_seen": 2975137792
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.938212164981544e-05,
+      "loss": 1.3921,
+      "theoretical_loss": 3.3513449069303727,
+      "tokens_seen": 2975268864
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.934199967902424e-05,
+      "loss": 1.2064,
+      "theoretical_loss": 3.3513336332204062,
+      "tokens_seen": 2975399936
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.930187770823303e-05,
+      "loss": 1.2952,
+      "theoretical_loss": 3.351322360146106,
+      "tokens_seen": 2975531008
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.926175573744182e-05,
+      "loss": 1.3144,
+      "theoretical_loss": 3.3513110877074084,
+      "tokens_seen": 2975662080
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.922163376665063e-05,
+      "loss": 1.3131,
+      "theoretical_loss": 3.351299815904249,
+      "tokens_seen": 2975793152
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.918151179585941e-05,
+      "loss": 1.3634,
+      "theoretical_loss": 3.351288544736565,
+      "tokens_seen": 2975924224
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.914138982506821e-05,
+      "loss": 1.3548,
+      "theoretical_loss": 3.3512772742042913,
+      "tokens_seen": 2976055296
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.9101267854277e-05,
+      "loss": 1.3452,
+      "theoretical_loss": 3.3512660043073654,
+      "tokens_seen": 2976186368
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.90611458834858e-05,
+      "loss": 1.2438,
+      "theoretical_loss": 3.3512547350457225,
+      "tokens_seen": 2976317440
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.902102391269459e-05,
+      "loss": 1.3163,
+      "theoretical_loss": 3.351243466419299,
+      "tokens_seen": 2976448512
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.898090194190338e-05,
+      "loss": 1.3379,
+      "theoretical_loss": 3.3512321984280313,
+      "tokens_seen": 2976579584
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.894077997111219e-05,
+      "loss": 1.3302,
+      "theoretical_loss": 3.3512209310718557,
+      "tokens_seen": 2976710656
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.890065800032098e-05,
+      "loss": 1.3146,
+      "theoretical_loss": 3.3512096643507086,
+      "tokens_seen": 2976841728
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.886053602952976e-05,
+      "loss": 1.2611,
+      "theoretical_loss": 3.3511983982645264,
+      "tokens_seen": 2976972800
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.882041405873857e-05,
+      "loss": 1.302,
+      "theoretical_loss": 3.3511871328132443,
+      "tokens_seen": 2977103872
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.878029208794736e-05,
+      "loss": 1.2265,
+      "theoretical_loss": 3.3511758679967993,
+      "tokens_seen": 2977234944
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.874017011715615e-05,
+      "loss": 1.338,
+      "theoretical_loss": 3.3511646038151275,
+      "tokens_seen": 2977366016
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.870004814636495e-05,
+      "loss": 1.2756,
+      "theoretical_loss": 3.3511533402681657,
+      "tokens_seen": 2977497088
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.865992617557375e-05,
+      "loss": 1.3426,
+      "theoretical_loss": 3.3511420773558496,
+      "tokens_seen": 2977628160
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.861980420478255e-05,
+      "loss": 1.2977,
+      "theoretical_loss": 3.351130815078116,
+      "tokens_seen": 2977759232
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.857968223399133e-05,
+      "loss": 1.281,
+      "theoretical_loss": 3.3511195534349008,
+      "tokens_seen": 2977890304
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.853956026320013e-05,
+      "loss": 1.2571,
+      "theoretical_loss": 3.35110829242614,
+      "tokens_seen": 2978021376
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.849943829240892e-05,
+      "loss": 1.3335,
+      "theoretical_loss": 3.3510970320517712,
+      "tokens_seen": 2978152448
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.845931632161773e-05,
+      "loss": 1.3179,
+      "theoretical_loss": 3.3510857723117295,
+      "tokens_seen": 2978283520
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.841919435082651e-05,
+      "loss": 1.2881,
+      "theoretical_loss": 3.3510745132059516,
+      "tokens_seen": 2978414592
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.83790723800353e-05,
+      "loss": 1.3278,
+      "theoretical_loss": 3.3510632547343744,
+      "tokens_seen": 2978545664
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.833895040924411e-05,
+      "loss": 1.3183,
+      "theoretical_loss": 3.3510519968969334,
+      "tokens_seen": 2978676736
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.82988284384529e-05,
+      "loss": 1.2874,
+      "theoretical_loss": 3.3510407396935653,
+      "tokens_seen": 2978807808
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.82587064676617e-05,
+      "loss": 1.2303,
+      "theoretical_loss": 3.351029483124207,
+      "tokens_seen": 2978938880
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.821858449687049e-05,
+      "loss": 1.3364,
+      "theoretical_loss": 3.3510182271887943,
+      "tokens_seen": 2979069952
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.81784625260793e-05,
+      "loss": 1.3012,
+      "theoretical_loss": 3.351006971887264,
+      "tokens_seen": 2979201024
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.813834055528807e-05,
+      "loss": 1.3597,
+      "theoretical_loss": 3.350995717219552,
+      "tokens_seen": 2979332096
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.809821858449687e-05,
+      "loss": 1.3007,
+      "theoretical_loss": 3.350984463185595,
+      "tokens_seen": 2979463168
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.805809661370567e-05,
+      "loss": 1.3673,
+      "theoretical_loss": 3.3509732097853298,
+      "tokens_seen": 2979594240
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.801797464291446e-05,
+      "loss": 1.3191,
+      "theoretical_loss": 3.3509619570186926,
+      "tokens_seen": 2979725312
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.797785267212326e-05,
+      "loss": 1.3595,
+      "theoretical_loss": 3.3509507048856193,
+      "tokens_seen": 2979856384
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.793773070133205e-05,
+      "loss": 1.3472,
+      "theoretical_loss": 3.3509394533860473,
+      "tokens_seen": 2979987456
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.789760873054084e-05,
+      "loss": 1.3518,
+      "theoretical_loss": 3.3509282025199125,
+      "tokens_seen": 2980118528
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.785748675974965e-05,
+      "loss": 1.3569,
+      "theoretical_loss": 3.350916952287151,
+      "tokens_seen": 2980249600
+    },
+    {
+      "epoch": 0.81,
+      "objective/train/advantage_avg": 0.48958030343055725,
+      "objective/train/docs_used": 1083727,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2750203609466553,
+      "objective/train/original_loss": 1.2750202417373657,
+      "objective/train/theoretical_loss": 3.3509057026877005,
+      "objective/train/tokens_used": 1350839776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24089105427265167,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050182580947876,
+      "objective/train/weighted_lm_loss": 1.3389700651168823,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3509057026877005,
+      "tokens_seen": 2980380672
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.781736478895843e-05,
+      "loss": 1.3506,
+      "theoretical_loss": 3.3509057026877005,
+      "tokens_seen": 2980380672
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.777724281816723e-05,
+      "loss": 1.2952,
+      "theoretical_loss": 3.3508944537214966,
+      "tokens_seen": 2980511744
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.773712084737603e-05,
+      "loss": 1.3527,
+      "theoretical_loss": 3.350883205388476,
+      "tokens_seen": 2980642816
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.769699887658482e-05,
+      "loss": 1.3681,
+      "theoretical_loss": 3.3508719576885753,
+      "tokens_seen": 2980773888
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.765687690579361e-05,
+      "loss": 1.2578,
+      "theoretical_loss": 3.3508607106217303,
+      "tokens_seen": 2980904960
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.76167549350024e-05,
+      "loss": 1.2518,
+      "theoretical_loss": 3.3508494641878785,
+      "tokens_seen": 2981036032
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.757663296421121e-05,
+      "loss": 1.2354,
+      "theoretical_loss": 3.3508382183869565,
+      "tokens_seen": 2981167104
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.753651099341999e-05,
+      "loss": 1.3354,
+      "theoretical_loss": 3.3508269732189,
+      "tokens_seen": 2981298176
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.74963890226288e-05,
+      "loss": 1.3535,
+      "theoretical_loss": 3.3508157286836466,
+      "tokens_seen": 2981429248
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.745626705183759e-05,
+      "loss": 1.3442,
+      "theoretical_loss": 3.350804484781132,
+      "tokens_seen": 2981560320
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.741614508104638e-05,
+      "loss": 1.2949,
+      "theoretical_loss": 3.350793241511293,
+      "tokens_seen": 2981691392
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.737602311025518e-05,
+      "loss": 1.3468,
+      "theoretical_loss": 3.350781998874066,
+      "tokens_seen": 2981822464
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.733590113946397e-05,
+      "loss": 1.2838,
+      "theoretical_loss": 3.3507707568693883,
+      "tokens_seen": 2981953536
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.729577916867277e-05,
+      "loss": 1.3974,
+      "theoretical_loss": 3.350759515497196,
+      "tokens_seen": 2982084608
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.725565719788155e-05,
+      "loss": 1.3598,
+      "theoretical_loss": 3.350748274757426,
+      "tokens_seen": 2982215680
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.721553522709036e-05,
+      "loss": 1.3125,
+      "theoretical_loss": 3.3507370346500145,
+      "tokens_seen": 2982346752
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.717541325629915e-05,
+      "loss": 1.2347,
+      "theoretical_loss": 3.3507257951748985,
+      "tokens_seen": 2982477824
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.713529128550795e-05,
+      "loss": 1.2981,
+      "theoretical_loss": 3.350714556332014,
+      "tokens_seen": 2982608896
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.709516931471674e-05,
+      "loss": 1.3772,
+      "theoretical_loss": 3.3507033181212993,
+      "tokens_seen": 2982739968
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.705504734392553e-05,
+      "loss": 1.2813,
+      "theoretical_loss": 3.350692080542689,
+      "tokens_seen": 2982871040
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.701492537313434e-05,
+      "loss": 1.2741,
+      "theoretical_loss": 3.350680843596121,
+      "tokens_seen": 2983002112
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.697480340234313e-05,
+      "loss": 1.2942,
+      "theoretical_loss": 3.3506696072815316,
+      "tokens_seen": 2983133184
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.693468143155191e-05,
+      "loss": 1.2697,
+      "theoretical_loss": 3.3506583715988576,
+      "tokens_seen": 2983264256
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.689455946076072e-05,
+      "loss": 1.2765,
+      "theoretical_loss": 3.350647136548036,
+      "tokens_seen": 2983395328
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.685443748996951e-05,
+      "loss": 1.2296,
+      "theoretical_loss": 3.350635902129003,
+      "tokens_seen": 2983526400
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.68143155191783e-05,
+      "loss": 1.3368,
+      "theoretical_loss": 3.3506246683416956,
+      "tokens_seen": 2983657472
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.67741935483871e-05,
+      "loss": 1.2649,
+      "theoretical_loss": 3.35061343518605,
+      "tokens_seen": 2983788544
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.67340715775959e-05,
+      "loss": 1.359,
+      "theoretical_loss": 3.350602202662004,
+      "tokens_seen": 2983919616
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.669394960680469e-05,
+      "loss": 1.2491,
+      "theoretical_loss": 3.350590970769493,
+      "tokens_seen": 2984050688
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.665382763601347e-05,
+      "loss": 1.2907,
+      "theoretical_loss": 3.350579739508455,
+      "tokens_seen": 2984181760
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.661370566522228e-05,
+      "loss": 1.2786,
+      "theoretical_loss": 3.350568508878826,
+      "tokens_seen": 2984312832
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.657358369443107e-05,
+      "loss": 1.3095,
+      "theoretical_loss": 3.3505572788805433,
+      "tokens_seen": 2984443904
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.653346172363988e-05,
+      "loss": 1.3108,
+      "theoretical_loss": 3.3505460495135435,
+      "tokens_seen": 2984574976
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.649333975284866e-05,
+      "loss": 1.3233,
+      "theoretical_loss": 3.3505348207777628,
+      "tokens_seen": 2984706048
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.645321778205746e-05,
+      "loss": 1.2552,
+      "theoretical_loss": 3.350523592673139,
+      "tokens_seen": 2984837120
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.641309581126626e-05,
+      "loss": 1.3285,
+      "theoretical_loss": 3.350512365199608,
+      "tokens_seen": 2984968192
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.637297384047504e-05,
+      "loss": 1.2025,
+      "theoretical_loss": 3.3505011383571075,
+      "tokens_seen": 2985099264
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.633285186968384e-05,
+      "loss": 1.2711,
+      "theoretical_loss": 3.3504899121455733,
+      "tokens_seen": 2985230336
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.629272989889263e-05,
+      "loss": 1.2923,
+      "theoretical_loss": 3.350478686564943,
+      "tokens_seen": 2985361408
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.625260792810144e-05,
+      "loss": 1.327,
+      "theoretical_loss": 3.3504674616151533,
+      "tokens_seen": 2985492480
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.621248595731022e-05,
+      "loss": 1.2958,
+      "theoretical_loss": 3.3504562372961413,
+      "tokens_seen": 2985623552
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.617236398651901e-05,
+      "loss": 1.2042,
+      "theoretical_loss": 3.350445013607843,
+      "tokens_seen": 2985754624
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.613224201572782e-05,
+      "loss": 1.3385,
+      "theoretical_loss": 3.350433790550196,
+      "tokens_seen": 2985885696
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.609212004493661e-05,
+      "loss": 1.2847,
+      "theoretical_loss": 3.350422568123137,
+      "tokens_seen": 2986016768
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.60519980741454e-05,
+      "loss": 1.2516,
+      "theoretical_loss": 3.350411346326603,
+      "tokens_seen": 2986147840
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.60118761033542e-05,
+      "loss": 1.3285,
+      "theoretical_loss": 3.350400125160531,
+      "tokens_seen": 2986278912
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.5971754132563e-05,
+      "loss": 1.2868,
+      "theoretical_loss": 3.3503889046248574,
+      "tokens_seen": 2986409984
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.593163216177178e-05,
+      "loss": 1.2403,
+      "theoretical_loss": 3.3503776847195192,
+      "tokens_seen": 2986541056
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.589151019098058e-05,
+      "loss": 1.3426,
+      "theoretical_loss": 3.3503664654444543,
+      "tokens_seen": 2986672128
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.585138822018938e-05,
+      "loss": 1.2998,
+      "theoretical_loss": 3.3503552467995985,
+      "tokens_seen": 2986803200
+    },
+    {
+      "epoch": 0.81,
+      "objective/train/advantage_avg": 0.49376311898231506,
+      "objective/train/docs_used": 1086010,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.305530309677124,
+      "objective/train/original_loss": 1.3055301904678345,
+      "objective/train/theoretical_loss": 3.350344028784889,
+      "objective/train/tokens_used": 1357393376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24445341527462006,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506192445755005,
+      "objective/train/weighted_lm_loss": 1.3716784715652466,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.350344028784889,
+      "tokens_seen": 2986934272
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.581126624939817e-05,
+      "loss": 1.3294,
+      "theoretical_loss": 3.350344028784889,
+      "tokens_seen": 2986934272
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.577114427860697e-05,
+      "loss": 1.2812,
+      "theoretical_loss": 3.3503328114002633,
+      "tokens_seen": 2987065344
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.573102230781576e-05,
+      "loss": 1.3736,
+      "theoretical_loss": 3.3503215946456573,
+      "tokens_seen": 2987196416
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.569090033702455e-05,
+      "loss": 1.3051,
+      "theoretical_loss": 3.350310378521009,
+      "tokens_seen": 2987327488
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.565077836623336e-05,
+      "loss": 1.2593,
+      "theoretical_loss": 3.3502991630262553,
+      "tokens_seen": 2987458560
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.561065639544214e-05,
+      "loss": 1.3554,
+      "theoretical_loss": 3.3502879481613324,
+      "tokens_seen": 2987589632
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.557053442465094e-05,
+      "loss": 1.3032,
+      "theoretical_loss": 3.3502767339261785,
+      "tokens_seen": 2987720704
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.553041245385974e-05,
+      "loss": 1.3225,
+      "theoretical_loss": 3.3502655203207294,
+      "tokens_seen": 2987851776
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.549029048306854e-05,
+      "loss": 1.2834,
+      "theoretical_loss": 3.3502543073449225,
+      "tokens_seen": 2987982848
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.545016851227732e-05,
+      "loss": 1.2942,
+      "theoretical_loss": 3.3502430949986954,
+      "tokens_seen": 2988113920
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.541004654148612e-05,
+      "loss": 1.2602,
+      "theoretical_loss": 3.3502318832819844,
+      "tokens_seen": 2988244992
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.536992457069492e-05,
+      "loss": 1.3578,
+      "theoretical_loss": 3.350220672194727,
+      "tokens_seen": 2988376064
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.53298025999037e-05,
+      "loss": 1.203,
+      "theoretical_loss": 3.3502094617368603,
+      "tokens_seen": 2988507136
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.528968062911251e-05,
+      "loss": 1.3249,
+      "theoretical_loss": 3.3501982519083207,
+      "tokens_seen": 2988638208
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.52495586583213e-05,
+      "loss": 1.2786,
+      "theoretical_loss": 3.350187042709046,
+      "tokens_seen": 2988769280
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.520943668753009e-05,
+      "loss": 1.3257,
+      "theoretical_loss": 3.350175834138973,
+      "tokens_seen": 2988900352
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.516931471673889e-05,
+      "loss": 1.2404,
+      "theoretical_loss": 3.350164626198039,
+      "tokens_seen": 2989031424
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.512919274594768e-05,
+      "loss": 1.2724,
+      "theoretical_loss": 3.3501534188861806,
+      "tokens_seen": 2989162496
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.508907077515648e-05,
+      "loss": 1.2988,
+      "theoretical_loss": 3.3501422122033357,
+      "tokens_seen": 2989293568
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.504894880436528e-05,
+      "loss": 1.3796,
+      "theoretical_loss": 3.3501310061494403,
+      "tokens_seen": 2989424640
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.500882683357407e-05,
+      "loss": 1.2255,
+      "theoretical_loss": 3.350119800724433,
+      "tokens_seen": 2989555712
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.496870486278286e-05,
+      "loss": 1.3259,
+      "theoretical_loss": 3.3501085959282495,
+      "tokens_seen": 2989686784
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.492858289199166e-05,
+      "loss": 1.3327,
+      "theoretical_loss": 3.3500973917608277,
+      "tokens_seen": 2989817856
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.488846092120045e-05,
+      "loss": 1.2659,
+      "theoretical_loss": 3.3500861882221047,
+      "tokens_seen": 2989948928
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.484833895040924e-05,
+      "loss": 1.2877,
+      "theoretical_loss": 3.350074985312017,
+      "tokens_seen": 2990080000
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.480821697961805e-05,
+      "loss": 1.3726,
+      "theoretical_loss": 3.350063783030503,
+      "tokens_seen": 2990211072
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.476809500882684e-05,
+      "loss": 1.2927,
+      "theoretical_loss": 3.350052581377499,
+      "tokens_seen": 2990342144
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.472797303803562e-05,
+      "loss": 1.3162,
+      "theoretical_loss": 3.3500413803529425,
+      "tokens_seen": 2990473216
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.468785106724443e-05,
+      "loss": 1.2149,
+      "theoretical_loss": 3.3500301799567707,
+      "tokens_seen": 2990604288
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.464772909645322e-05,
+      "loss": 1.294,
+      "theoretical_loss": 3.3500189801889206,
+      "tokens_seen": 2990735360
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.460760712566202e-05,
+      "loss": 1.4149,
+      "theoretical_loss": 3.3500077810493294,
+      "tokens_seen": 2990866432
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.45674851548708e-05,
+      "loss": 1.2932,
+      "theoretical_loss": 3.3499965825379348,
+      "tokens_seen": 2990997504
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.452736318407961e-05,
+      "loss": 1.2903,
+      "theoretical_loss": 3.3499853846546737,
+      "tokens_seen": 2991128576
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.44872412132884e-05,
+      "loss": 1.1704,
+      "theoretical_loss": 3.349974187399483,
+      "tokens_seen": 2991259648
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.444711924249718e-05,
+      "loss": 1.2783,
+      "theoretical_loss": 3.3499629907723003,
+      "tokens_seen": 2991390720
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.440699727170599e-05,
+      "loss": 1.3106,
+      "theoretical_loss": 3.3499517947730633,
+      "tokens_seen": 2991521792
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.436687530091478e-05,
+      "loss": 1.2441,
+      "theoretical_loss": 3.3499405994017084,
+      "tokens_seen": 2991652864
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.432675333012359e-05,
+      "loss": 1.3664,
+      "theoretical_loss": 3.3499294046581736,
+      "tokens_seen": 2991783936
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.428663135933237e-05,
+      "loss": 1.3022,
+      "theoretical_loss": 3.3499182105423957,
+      "tokens_seen": 2991915008
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.424650938854116e-05,
+      "loss": 1.2687,
+      "theoretical_loss": 3.3499070170543126,
+      "tokens_seen": 2992046080
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.420638741774997e-05,
+      "loss": 1.2335,
+      "theoretical_loss": 3.349895824193861,
+      "tokens_seen": 2992177152
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.416626544695876e-05,
+      "loss": 1.2755,
+      "theoretical_loss": 3.349884631960978,
+      "tokens_seen": 2992308224
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.412614347616755e-05,
+      "loss": 1.2682,
+      "theoretical_loss": 3.3498734403556023,
+      "tokens_seen": 2992439296
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.408602150537634e-05,
+      "loss": 1.2102,
+      "theoretical_loss": 3.3498622493776695,
+      "tokens_seen": 2992570368
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.404589953458515e-05,
+      "loss": 1.2855,
+      "theoretical_loss": 3.349851059027118,
+      "tokens_seen": 2992701440
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.400577756379393e-05,
+      "loss": 1.2824,
+      "theoretical_loss": 3.349839869303885,
+      "tokens_seen": 2992832512
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.396565559300272e-05,
+      "loss": 1.208,
+      "theoretical_loss": 3.3498286802079074,
+      "tokens_seen": 2992963584
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.392553362221153e-05,
+      "loss": 1.3221,
+      "theoretical_loss": 3.3498174917391235,
+      "tokens_seen": 2993094656
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.388541165142032e-05,
+      "loss": 1.2607,
+      "theoretical_loss": 3.3498063038974695,
+      "tokens_seen": 2993225728
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.384528968062911e-05,
+      "loss": 1.248,
+      "theoretical_loss": 3.349795116682884,
+      "tokens_seen": 2993356800
+    },
+    {
+      "epoch": 0.81,
+      "objective/train/advantage_avg": 0.49041640758514404,
+      "objective/train/docs_used": 1088408,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2716584205627441,
+      "objective/train/original_loss": 1.2716584205627441,
+      "objective/train/theoretical_loss": 3.3497839300953034,
+      "objective/train/tokens_used": 1363946976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2411275953054428,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0502674579620361,
+      "objective/train/weighted_lm_loss": 1.335556149482727,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3497839300953034,
+      "tokens_seen": 2993487872
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.380516770983791e-05,
+      "loss": 1.4139,
+      "theoretical_loss": 3.3497839300953034,
+      "tokens_seen": 2993487872
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.37650457390467e-05,
+      "loss": 1.3573,
+      "theoretical_loss": 3.3497727441346656,
+      "tokens_seen": 2993618944
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.37249237682555e-05,
+      "loss": 1.3723,
+      "theoretical_loss": 3.3497615588009078,
+      "tokens_seen": 2993750016
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.368480179746429e-05,
+      "loss": 1.2883,
+      "theoretical_loss": 3.3497503740939676,
+      "tokens_seen": 2993881088
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.364467982667309e-05,
+      "loss": 1.3457,
+      "theoretical_loss": 3.3497391900137825,
+      "tokens_seen": 2994012160
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.360455785588188e-05,
+      "loss": 1.2904,
+      "theoretical_loss": 3.3497280065602895,
+      "tokens_seen": 2994143232
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.356443588509068e-05,
+      "loss": 1.3908,
+      "theoretical_loss": 3.3497168237334267,
+      "tokens_seen": 2994274304
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.352431391429947e-05,
+      "loss": 1.1966,
+      "theoretical_loss": 3.349705641533131,
+      "tokens_seen": 2994405376
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.348419194350826e-05,
+      "loss": 1.3333,
+      "theoretical_loss": 3.3496944599593403,
+      "tokens_seen": 2994536448
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.344406997271707e-05,
+      "loss": 1.2728,
+      "theoretical_loss": 3.349683279011992,
+      "tokens_seen": 2994667520
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.340394800192585e-05,
+      "loss": 1.3487,
+      "theoretical_loss": 3.3496720986910233,
+      "tokens_seen": 2994798592
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.336382603113465e-05,
+      "loss": 1.3185,
+      "theoretical_loss": 3.3496609189963715,
+      "tokens_seen": 2994929664
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.332370406034345e-05,
+      "loss": 1.4126,
+      "theoretical_loss": 3.349649739927975,
+      "tokens_seen": 2995060736
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.328358208955224e-05,
+      "loss": 1.3322,
+      "theoretical_loss": 3.3496385614857704,
+      "tokens_seen": 2995191808
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.324346011876103e-05,
+      "loss": 1.3781,
+      "theoretical_loss": 3.349627383669696,
+      "tokens_seen": 2995322880
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.320333814796983e-05,
+      "loss": 1.2789,
+      "theoretical_loss": 3.3496162064796886,
+      "tokens_seen": 2995453952
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.316321617717863e-05,
+      "loss": 1.3181,
+      "theoretical_loss": 3.3496050299156863,
+      "tokens_seen": 2995585024
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.312309420638742e-05,
+      "loss": 1.2895,
+      "theoretical_loss": 3.3495938539776264,
+      "tokens_seen": 2995716096
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.308297223559622e-05,
+      "loss": 1.3383,
+      "theoretical_loss": 3.3495826786654463,
+      "tokens_seen": 2995847168
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.304285026480501e-05,
+      "loss": 1.2786,
+      "theoretical_loss": 3.3495715039790843,
+      "tokens_seen": 2995978240
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.30027282940138e-05,
+      "loss": 1.2974,
+      "theoretical_loss": 3.349560329918477,
+      "tokens_seen": 2996109312
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.29626063232226e-05,
+      "loss": 1.2521,
+      "theoretical_loss": 3.349549156483562,
+      "tokens_seen": 2996240384
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.292248435243139e-05,
+      "loss": 1.245,
+      "theoretical_loss": 3.3495379836742782,
+      "tokens_seen": 2996371456
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.28823623816402e-05,
+      "loss": 1.3031,
+      "theoretical_loss": 3.349526811490562,
+      "tokens_seen": 2996502528
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.284224041084899e-05,
+      "loss": 1.3477,
+      "theoretical_loss": 3.349515639932351,
+      "tokens_seen": 2996633600
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.280211844005777e-05,
+      "loss": 1.2992,
+      "theoretical_loss": 3.3495044689995837,
+      "tokens_seen": 2996764672
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.276199646926657e-05,
+      "loss": 1.3857,
+      "theoretical_loss": 3.3494932986921966,
+      "tokens_seen": 2996895744
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.272187449847537e-05,
+      "loss": 1.2581,
+      "theoretical_loss": 3.3494821290101284,
+      "tokens_seen": 2997026816
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.268175252768417e-05,
+      "loss": 1.2731,
+      "theoretical_loss": 3.349470959953316,
+      "tokens_seen": 2997157888
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.264163055689295e-05,
+      "loss": 1.3138,
+      "theoretical_loss": 3.3494597915216975,
+      "tokens_seen": 2997288960
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.260150858610176e-05,
+      "loss": 1.3587,
+      "theoretical_loss": 3.3494486237152103,
+      "tokens_seen": 2997420032
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.256138661531055e-05,
+      "loss": 1.3166,
+      "theoretical_loss": 3.349437456533792,
+      "tokens_seen": 2997551104
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.252126464451933e-05,
+      "loss": 1.2695,
+      "theoretical_loss": 3.349426289977381,
+      "tokens_seen": 2997682176
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.248114267372814e-05,
+      "loss": 1.3041,
+      "theoretical_loss": 3.349415124045914,
+      "tokens_seen": 2997813248
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.244102070293693e-05,
+      "loss": 1.2973,
+      "theoretical_loss": 3.3494039587393294,
+      "tokens_seen": 2997944320
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.240089873214573e-05,
+      "loss": 1.2973,
+      "theoretical_loss": 3.3493927940575645,
+      "tokens_seen": 2998075392
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.236077676135451e-05,
+      "loss": 1.2074,
+      "theoretical_loss": 3.349381630000557,
+      "tokens_seen": 2998206464
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.23206547905633e-05,
+      "loss": 1.3441,
+      "theoretical_loss": 3.3493704665682453,
+      "tokens_seen": 2998337536
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.228053281977211e-05,
+      "loss": 1.3678,
+      "theoretical_loss": 3.3493593037605667,
+      "tokens_seen": 2998468608
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.22404108489809e-05,
+      "loss": 1.3155,
+      "theoretical_loss": 3.3493481415774586,
+      "tokens_seen": 2998599680
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.22002888781897e-05,
+      "loss": 1.3249,
+      "theoretical_loss": 3.349336980018859,
+      "tokens_seen": 2998730752
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.216016690739849e-05,
+      "loss": 1.3074,
+      "theoretical_loss": 3.349325819084706,
+      "tokens_seen": 2998861824
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.21200449366073e-05,
+      "loss": 1.2705,
+      "theoretical_loss": 3.349314658774937,
+      "tokens_seen": 2998992896
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.207992296581608e-05,
+      "loss": 1.2788,
+      "theoretical_loss": 3.3493034990894897,
+      "tokens_seen": 2999123968
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.203980099502487e-05,
+      "loss": 1.3253,
+      "theoretical_loss": 3.3492923400283026,
+      "tokens_seen": 2999255040
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.199967902423368e-05,
+      "loss": 1.2654,
+      "theoretical_loss": 3.3492811815913126,
+      "tokens_seen": 2999386112
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.195955705344247e-05,
+      "loss": 1.2481,
+      "theoretical_loss": 3.3492700237784576,
+      "tokens_seen": 2999517184
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.191943508265126e-05,
+      "loss": 1.2158,
+      "theoretical_loss": 3.3492588665896763,
+      "tokens_seen": 2999648256
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.187931311186005e-05,
+      "loss": 1.2446,
+      "theoretical_loss": 3.3492477100249056,
+      "tokens_seen": 2999779328
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.183919114106886e-05,
+      "loss": 1.3248,
+      "theoretical_loss": 3.349236554084084,
+      "tokens_seen": 2999910400
+    },
+    {
+      "epoch": 0.82,
+      "objective/train/advantage_avg": 0.4921075701713562,
+      "objective/train/docs_used": 1090669,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3833339214324951,
+      "objective/train/original_loss": 1.3833339214324951,
+      "objective/train/theoretical_loss": 3.3492253987671483,
+      "objective/train/tokens_used": 1370500576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24271728098392487,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504447221755981,
+      "objective/train/weighted_lm_loss": 1.4532510042190552,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3492253987671483,
+      "tokens_seen": 3000041472
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.179906917027765e-05,
+      "loss": 1.3641,
+      "theoretical_loss": 3.3492253987671483,
+      "tokens_seen": 3000041472
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.175894719948643e-05,
+      "loss": 1.2219,
+      "theoretical_loss": 3.3492142440740373,
+      "tokens_seen": 3000172544
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.171882522869524e-05,
+      "loss": 1.2688,
+      "theoretical_loss": 3.349203090004689,
+      "tokens_seen": 3000303616
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.167870325790403e-05,
+      "loss": 1.3316,
+      "theoretical_loss": 3.3491919365590412,
+      "tokens_seen": 3000434688
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.163858128711282e-05,
+      "loss": 1.3606,
+      "theoretical_loss": 3.349180783737031,
+      "tokens_seen": 3000565760
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.159845931632162e-05,
+      "loss": 1.2781,
+      "theoretical_loss": 3.3491696315385964,
+      "tokens_seen": 3000696832
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.155833734553041e-05,
+      "loss": 1.195,
+      "theoretical_loss": 3.3491584799636764,
+      "tokens_seen": 3000827904
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.151821537473922e-05,
+      "loss": 1.3333,
+      "theoretical_loss": 3.349147329012208,
+      "tokens_seen": 3000958976
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.1478093403948e-05,
+      "loss": 1.4202,
+      "theoretical_loss": 3.349136178684129,
+      "tokens_seen": 3001090048
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.14379714331568e-05,
+      "loss": 1.2272,
+      "theoretical_loss": 3.349125028979378,
+      "tokens_seen": 3001221120
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.13978494623656e-05,
+      "loss": 1.3668,
+      "theoretical_loss": 3.3491138798978923,
+      "tokens_seen": 3001352192
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.13577274915744e-05,
+      "loss": 1.1886,
+      "theoretical_loss": 3.34910273143961,
+      "tokens_seen": 3001483264
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.131760552078318e-05,
+      "loss": 1.224,
+      "theoretical_loss": 3.3490915836044692,
+      "tokens_seen": 3001614336
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.127748354999197e-05,
+      "loss": 1.2308,
+      "theoretical_loss": 3.3490804363924083,
+      "tokens_seen": 3001745408
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.123736157920078e-05,
+      "loss": 1.3452,
+      "theoretical_loss": 3.3490692898033645,
+      "tokens_seen": 3001876480
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.119723960840956e-05,
+      "loss": 1.3297,
+      "theoretical_loss": 3.3490581438372757,
+      "tokens_seen": 3002007552
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.115711763761836e-05,
+      "loss": 1.3455,
+      "theoretical_loss": 3.3490469984940807,
+      "tokens_seen": 3002138624
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.111699566682716e-05,
+      "loss": 1.3547,
+      "theoretical_loss": 3.3490358537737173,
+      "tokens_seen": 3002269696
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.107687369603595e-05,
+      "loss": 1.3028,
+      "theoretical_loss": 3.3490247096761223,
+      "tokens_seen": 3002400768
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.103675172524474e-05,
+      "loss": 1.3943,
+      "theoretical_loss": 3.3490135662012355,
+      "tokens_seen": 3002531840
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.099662975445354e-05,
+      "loss": 1.3217,
+      "theoretical_loss": 3.3490024233489937,
+      "tokens_seen": 3002662912
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.095650778366234e-05,
+      "loss": 1.3629,
+      "theoretical_loss": 3.3489912811193356,
+      "tokens_seen": 3002793984
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.091638581287113e-05,
+      "loss": 1.3137,
+      "theoretical_loss": 3.348980139512199,
+      "tokens_seen": 3002925056
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.087626384207993e-05,
+      "loss": 1.2449,
+      "theoretical_loss": 3.3489689985275217,
+      "tokens_seen": 3003056128
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.083614187128872e-05,
+      "loss": 1.3775,
+      "theoretical_loss": 3.348957858165242,
+      "tokens_seen": 3003187200
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.079601990049751e-05,
+      "loss": 1.2954,
+      "theoretical_loss": 3.3489467184252977,
+      "tokens_seen": 3003318272
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.075589792970632e-05,
+      "loss": 1.3971,
+      "theoretical_loss": 3.3489355793076276,
+      "tokens_seen": 3003449344
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.07157759589151e-05,
+      "loss": 1.2931,
+      "theoretical_loss": 3.3489244408121692,
+      "tokens_seen": 3003580416
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.06756539881239e-05,
+      "loss": 1.3245,
+      "theoretical_loss": 3.3489133029388607,
+      "tokens_seen": 3003711488
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.06355320173327e-05,
+      "loss": 1.2462,
+      "theoretical_loss": 3.34890216568764,
+      "tokens_seen": 3003842560
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.059541004654148e-05,
+      "loss": 1.3132,
+      "theoretical_loss": 3.3488910290584455,
+      "tokens_seen": 3003973632
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.055528807575028e-05,
+      "loss": 1.3047,
+      "theoretical_loss": 3.3488798930512154,
+      "tokens_seen": 3004104704
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.051516610495908e-05,
+      "loss": 1.2559,
+      "theoretical_loss": 3.3488687576658873,
+      "tokens_seen": 3004235776
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.047504413416788e-05,
+      "loss": 1.265,
+      "theoretical_loss": 3.3488576229024,
+      "tokens_seen": 3004366848
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.043492216337666e-05,
+      "loss": 1.2769,
+      "theoretical_loss": 3.3488464887606915,
+      "tokens_seen": 3004497920
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.039480019258547e-05,
+      "loss": 1.2834,
+      "theoretical_loss": 3.3488353552406993,
+      "tokens_seen": 3004628992
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.035467822179426e-05,
+      "loss": 1.2855,
+      "theoretical_loss": 3.3488242223423623,
+      "tokens_seen": 3004760064
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.031455625100305e-05,
+      "loss": 1.245,
+      "theoretical_loss": 3.348813090065619,
+      "tokens_seen": 3004891136
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.027443428021185e-05,
+      "loss": 1.2285,
+      "theoretical_loss": 3.348801958410406,
+      "tokens_seen": 3005022208
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.023431230942064e-05,
+      "loss": 1.2931,
+      "theoretical_loss": 3.3487908273766633,
+      "tokens_seen": 3005153280
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.019419033862944e-05,
+      "loss": 1.2706,
+      "theoretical_loss": 3.348779696964328,
+      "tokens_seen": 3005284352
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.015406836783822e-05,
+      "loss": 1.282,
+      "theoretical_loss": 3.3487685671733383,
+      "tokens_seen": 3005415424
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.011394639704702e-05,
+      "loss": 1.3151,
+      "theoretical_loss": 3.3487574380036333,
+      "tokens_seen": 3005546496
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.007382442625582e-05,
+      "loss": 1.2796,
+      "theoretical_loss": 3.3487463094551506,
+      "tokens_seen": 3005677568
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.003370245546462e-05,
+      "loss": 1.3123,
+      "theoretical_loss": 3.3487351815278283,
+      "tokens_seen": 3005808640
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.999358048467341e-05,
+      "loss": 1.2206,
+      "theoretical_loss": 3.348724054221605,
+      "tokens_seen": 3005939712
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.99534585138822e-05,
+      "loss": 1.2041,
+      "theoretical_loss": 3.3487129275364187,
+      "tokens_seen": 3006070784
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.991333654309101e-05,
+      "loss": 1.2105,
+      "theoretical_loss": 3.348701801472208,
+      "tokens_seen": 3006201856
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.98732145722998e-05,
+      "loss": 1.2907,
+      "theoretical_loss": 3.3486906760289106,
+      "tokens_seen": 3006332928
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.983309260150858e-05,
+      "loss": 1.297,
+      "theoretical_loss": 3.3486795512064655,
+      "tokens_seen": 3006464000
+    },
+    {
+      "epoch": 0.82,
+      "objective/train/advantage_avg": 0.49275365471839905,
+      "objective/train/docs_used": 1093144,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2475663423538208,
+      "objective/train/original_loss": 1.2475664615631104,
+      "objective/train/theoretical_loss": 3.34866842700481,
+      "objective/train/tokens_used": 1377054176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2435685098171234,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505136251449585,
+      "objective/train/weighted_lm_loss": 1.3106735944747925,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9048373699188232,
+      "theoretical_loss": 3.34866842700481,
+      "tokens_seen": 3006595072
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.979297063071739e-05,
+      "loss": 1.306,
+      "theoretical_loss": 3.34866842700481,
+      "tokens_seen": 3006595072
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.975284865992618e-05,
+      "loss": 1.277,
+      "theoretical_loss": 3.3486573034238836,
+      "tokens_seen": 3006726144
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.971272668913497e-05,
+      "loss": 1.3262,
+      "theoretical_loss": 3.348646180463624,
+      "tokens_seen": 3006857216
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.967260471834376e-05,
+      "loss": 1.2497,
+      "theoretical_loss": 3.3486350581239694,
+      "tokens_seen": 3006988288
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.963248274755256e-05,
+      "loss": 1.2438,
+      "theoretical_loss": 3.3486239364048584,
+      "tokens_seen": 3007119360
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.959236077676136e-05,
+      "loss": 1.3152,
+      "theoretical_loss": 3.348612815306229,
+      "tokens_seen": 3007250432
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.955223880597014e-05,
+      "loss": 1.2044,
+      "theoretical_loss": 3.3486016948280195,
+      "tokens_seen": 3007381504
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.951211683517895e-05,
+      "loss": 1.3367,
+      "theoretical_loss": 3.348590574970169,
+      "tokens_seen": 3007512576
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.947199486438774e-05,
+      "loss": 1.3417,
+      "theoretical_loss": 3.3485794557326147,
+      "tokens_seen": 3007643648
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.943187289359655e-05,
+      "loss": 1.2379,
+      "theoretical_loss": 3.348568337115296,
+      "tokens_seen": 3007774720
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.939175092280533e-05,
+      "loss": 1.3289,
+      "theoretical_loss": 3.3485572191181507,
+      "tokens_seen": 3007905792
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.935162895201412e-05,
+      "loss": 1.2697,
+      "theoretical_loss": 3.3485461017411176,
+      "tokens_seen": 3008036864
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.931150698122293e-05,
+      "loss": 1.2623,
+      "theoretical_loss": 3.348534984984135,
+      "tokens_seen": 3008167936
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.92713850104317e-05,
+      "loss": 1.2723,
+      "theoretical_loss": 3.348523868847141,
+      "tokens_seen": 3008299008
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.923126303964051e-05,
+      "loss": 1.2207,
+      "theoretical_loss": 3.348512753330074,
+      "tokens_seen": 3008430080
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.91911410688493e-05,
+      "loss": 1.246,
+      "theoretical_loss": 3.3485016384328725,
+      "tokens_seen": 3008561152
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.91510190980581e-05,
+      "loss": 1.3052,
+      "theoretical_loss": 3.3484905241554754,
+      "tokens_seen": 3008692224
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.911089712726689e-05,
+      "loss": 1.3311,
+      "theoretical_loss": 3.3484794104978204,
+      "tokens_seen": 3008823296
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.907077515647568e-05,
+      "loss": 1.2972,
+      "theoretical_loss": 3.3484682974598465,
+      "tokens_seen": 3008954368
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.903065318568449e-05,
+      "loss": 1.2208,
+      "theoretical_loss": 3.348457185041492,
+      "tokens_seen": 3009085440
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.899053121489328e-05,
+      "loss": 1.3353,
+      "theoretical_loss": 3.3484460732426955,
+      "tokens_seen": 3009216512
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.895040924410207e-05,
+      "loss": 1.2922,
+      "theoretical_loss": 3.3484349620633953,
+      "tokens_seen": 3009347584
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.891028727331087e-05,
+      "loss": 1.3045,
+      "theoretical_loss": 3.3484238515035294,
+      "tokens_seen": 3009478656
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.887016530251966e-05,
+      "loss": 1.2621,
+      "theoretical_loss": 3.3484127415630374,
+      "tokens_seen": 3009609728
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.883004333172845e-05,
+      "loss": 1.2827,
+      "theoretical_loss": 3.3484016322418566,
+      "tokens_seen": 3009740800
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.878992136093724e-05,
+      "loss": 1.3225,
+      "theoretical_loss": 3.3483905235399263,
+      "tokens_seen": 3009871872
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.874979939014605e-05,
+      "loss": 1.2844,
+      "theoretical_loss": 3.3483794154571846,
+      "tokens_seen": 3010002944
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.870967741935484e-05,
+      "loss": 1.2348,
+      "theoretical_loss": 3.348368307993571,
+      "tokens_seen": 3010134016
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.866955544856362e-05,
+      "loss": 1.2669,
+      "theoretical_loss": 3.348357201149023,
+      "tokens_seen": 3010265088
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.862943347777243e-05,
+      "loss": 1.3294,
+      "theoretical_loss": 3.3483460949234787,
+      "tokens_seen": 3010396160
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.858931150698122e-05,
+      "loss": 1.2647,
+      "theoretical_loss": 3.348334989316878,
+      "tokens_seen": 3010527232
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.854918953619003e-05,
+      "loss": 1.356,
+      "theoretical_loss": 3.3483238843291585,
+      "tokens_seen": 3010658304
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.850906756539881e-05,
+      "loss": 1.2747,
+      "theoretical_loss": 3.3483127799602594,
+      "tokens_seen": 3010789376
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.846894559460761e-05,
+      "loss": 1.2881,
+      "theoretical_loss": 3.3483016762101188,
+      "tokens_seen": 3010920448
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.842882362381641e-05,
+      "loss": 1.3361,
+      "theoretical_loss": 3.3482905730786756,
+      "tokens_seen": 3011051520
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 8.83887016530252e-05,
+      "loss": 1.2445,
+      "theoretical_loss": 3.3482794705658683,
+      "tokens_seen": 3011182592
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.834857968223399e-05,
+      "loss": 1.2335,
+      "theoretical_loss": 3.3482683686716355,
+      "tokens_seen": 3011313664
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.830845771144279e-05,
+      "loss": 1.2968,
+      "theoretical_loss": 3.3482572673959154,
+      "tokens_seen": 3011444736
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.826833574065159e-05,
+      "loss": 1.3229,
+      "theoretical_loss": 3.3482461667386474,
+      "tokens_seen": 3011575808
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.822821376986037e-05,
+      "loss": 1.3111,
+      "theoretical_loss": 3.348235066699769,
+      "tokens_seen": 3011706880
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.818809179906916e-05,
+      "loss": 1.2845,
+      "theoretical_loss": 3.3482239672792207,
+      "tokens_seen": 3011837952
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.814796982827797e-05,
+      "loss": 1.3066,
+      "theoretical_loss": 3.3482128684769394,
+      "tokens_seen": 3011969024
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.810784785748676e-05,
+      "loss": 1.2635,
+      "theoretical_loss": 3.348201770292864,
+      "tokens_seen": 3012100096
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.806772588669556e-05,
+      "loss": 1.2394,
+      "theoretical_loss": 3.3481906727269344,
+      "tokens_seen": 3012231168
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.802760391590435e-05,
+      "loss": 1.2749,
+      "theoretical_loss": 3.348179575779088,
+      "tokens_seen": 3012362240
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.798748194511315e-05,
+      "loss": 1.2271,
+      "theoretical_loss": 3.3481684794492637,
+      "tokens_seen": 3012493312
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.794735997432195e-05,
+      "loss": 1.3629,
+      "theoretical_loss": 3.3481573837374006,
+      "tokens_seen": 3012624384
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.790723800353073e-05,
+      "loss": 1.3136,
+      "theoretical_loss": 3.3481462886434374,
+      "tokens_seen": 3012755456
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.786711603273953e-05,
+      "loss": 1.2975,
+      "theoretical_loss": 3.3481351941673125,
+      "tokens_seen": 3012886528
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.782699406194833e-05,
+      "loss": 1.3568,
+      "theoretical_loss": 3.3481241003089646,
+      "tokens_seen": 3013017600
+    },
+    {
+      "epoch": 0.83,
+      "objective/train/advantage_avg": 0.4914937913417816,
+      "objective/train/docs_used": 1095529,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2421298027038574,
+      "objective/train/original_loss": 1.2421298027038574,
+      "objective/train/theoretical_loss": 3.3481130070683327,
+      "objective/train/tokens_used": 1383607776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24214236438274384,
+      "objective/train/value_max": -0.5,
+      "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": 1.3046658039093018,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3481130070683327,
+      "tokens_seen": 3013148672
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.778687209115712e-05,
+      "loss": 1.2655,
+      "theoretical_loss": 3.3481130070683327,
+      "tokens_seen": 3013148672
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.774675012036591e-05,
+      "loss": 1.3144,
+      "theoretical_loss": 3.3481019144453557,
+      "tokens_seen": 3013279744
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.770662814957472e-05,
+      "loss": 1.2556,
+      "theoretical_loss": 3.3480908224399712,
+      "tokens_seen": 3013410816
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.766650617878351e-05,
+      "loss": 1.277,
+      "theoretical_loss": 3.3480797310521195,
+      "tokens_seen": 3013541888
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.762638420799229e-05,
+      "loss": 1.294,
+      "theoretical_loss": 3.3480686402817383,
+      "tokens_seen": 3013672960
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.75862622372011e-05,
+      "loss": 1.2828,
+      "theoretical_loss": 3.348057550128767,
+      "tokens_seen": 3013804032
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.754614026640989e-05,
+      "loss": 1.3209,
+      "theoretical_loss": 3.348046460593144,
+      "tokens_seen": 3013935104
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.75060182956187e-05,
+      "loss": 1.3112,
+      "theoretical_loss": 3.348035371674808,
+      "tokens_seen": 3014066176
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.746589632482747e-05,
+      "loss": 1.3431,
+      "theoretical_loss": 3.3480242833736984,
+      "tokens_seen": 3014197248
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.742577435403627e-05,
+      "loss": 1.2745,
+      "theoretical_loss": 3.348013195689753,
+      "tokens_seen": 3014328320
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.738565238324507e-05,
+      "loss": 1.366,
+      "theoretical_loss": 3.3480021086229117,
+      "tokens_seen": 3014459392
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.734553041245385e-05,
+      "loss": 1.3051,
+      "theoretical_loss": 3.347991022173113,
+      "tokens_seen": 3014590464
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.730540844166266e-05,
+      "loss": 1.2738,
+      "theoretical_loss": 3.3479799363402956,
+      "tokens_seen": 3014721536
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.726528647087145e-05,
+      "loss": 1.3123,
+      "theoretical_loss": 3.347968851124398,
+      "tokens_seen": 3014852608
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.722516450008026e-05,
+      "loss": 1.2844,
+      "theoretical_loss": 3.3479577665253593,
+      "tokens_seen": 3014983680
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.718504252928904e-05,
+      "loss": 1.3721,
+      "theoretical_loss": 3.3479466825431183,
+      "tokens_seen": 3015114752
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.714492055849783e-05,
+      "loss": 1.3314,
+      "theoretical_loss": 3.3479355991776143,
+      "tokens_seen": 3015245824
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.710479858770664e-05,
+      "loss": 1.2652,
+      "theoretical_loss": 3.347924516428786,
+      "tokens_seen": 3015376896
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.706467661691543e-05,
+      "loss": 1.3031,
+      "theoretical_loss": 3.3479134342965717,
+      "tokens_seen": 3015507968
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.702455464612422e-05,
+      "loss": 1.3834,
+      "theoretical_loss": 3.347902352780911,
+      "tokens_seen": 3015639040
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.698443267533301e-05,
+      "loss": 1.2046,
+      "theoretical_loss": 3.3478912718817426,
+      "tokens_seen": 3015770112
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.69443107045418e-05,
+      "loss": 1.2673,
+      "theoretical_loss": 3.3478801915990055,
+      "tokens_seen": 3015901184
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.69041887337506e-05,
+      "loss": 1.2278,
+      "theoretical_loss": 3.3478691119326385,
+      "tokens_seen": 3016032256
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.686406676295939e-05,
+      "loss": 1.2311,
+      "theoretical_loss": 3.34785803288258,
+      "tokens_seen": 3016163328
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.68239447921682e-05,
+      "loss": 1.2506,
+      "theoretical_loss": 3.34784695444877,
+      "tokens_seen": 3016294400
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.678382282137699e-05,
+      "loss": 1.3667,
+      "theoretical_loss": 3.3478358766311467,
+      "tokens_seen": 3016425472
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.674370085058578e-05,
+      "loss": 1.2932,
+      "theoretical_loss": 3.3478247994296493,
+      "tokens_seen": 3016556544
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.670357887979458e-05,
+      "loss": 1.3061,
+      "theoretical_loss": 3.3478137228442164,
+      "tokens_seen": 3016687616
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.666345690900337e-05,
+      "loss": 1.2822,
+      "theoretical_loss": 3.3478026468747877,
+      "tokens_seen": 3016818688
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.662333493821218e-05,
+      "loss": 1.3154,
+      "theoretical_loss": 3.3477915715213014,
+      "tokens_seen": 3016949760
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.658321296742095e-05,
+      "loss": 1.293,
+      "theoretical_loss": 3.347780496783697,
+      "tokens_seen": 3017080832
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.654309099662976e-05,
+      "loss": 1.2666,
+      "theoretical_loss": 3.3477694226619135,
+      "tokens_seen": 3017211904
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.650296902583855e-05,
+      "loss": 1.2558,
+      "theoretical_loss": 3.3477583491558898,
+      "tokens_seen": 3017342976
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.646284705504733e-05,
+      "loss": 1.1904,
+      "theoretical_loss": 3.3477472762655647,
+      "tokens_seen": 3017474048
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.642272508425614e-05,
+      "loss": 1.2814,
+      "theoretical_loss": 3.3477362039908773,
+      "tokens_seen": 3017605120
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.638260311346493e-05,
+      "loss": 1.3326,
+      "theoretical_loss": 3.347725132331767,
+      "tokens_seen": 3017736192
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.634248114267374e-05,
+      "loss": 1.378,
+      "theoretical_loss": 3.3477140612881726,
+      "tokens_seen": 3017867264
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.630235917188252e-05,
+      "loss": 1.2057,
+      "theoretical_loss": 3.3477029908600326,
+      "tokens_seen": 3017998336
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.626223720109132e-05,
+      "loss": 1.2904,
+      "theoretical_loss": 3.347691921047287,
+      "tokens_seen": 3018129408
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.622211523030012e-05,
+      "loss": 1.2914,
+      "theoretical_loss": 3.3476808518498746,
+      "tokens_seen": 3018260480
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.618199325950891e-05,
+      "loss": 1.3224,
+      "theoretical_loss": 3.347669783267734,
+      "tokens_seen": 3018391552
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.61418712887177e-05,
+      "loss": 1.3078,
+      "theoretical_loss": 3.347658715300805,
+      "tokens_seen": 3018522624
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.61017493179265e-05,
+      "loss": 1.2936,
+      "theoretical_loss": 3.347647647949026,
+      "tokens_seen": 3018653696
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.60616273471353e-05,
+      "loss": 1.2318,
+      "theoretical_loss": 3.347636581212336,
+      "tokens_seen": 3018784768
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.60215053763441e-05,
+      "loss": 1.2762,
+      "theoretical_loss": 3.3476255150906753,
+      "tokens_seen": 3018915840
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.598138340555287e-05,
+      "loss": 1.3719,
+      "theoretical_loss": 3.3476144495839817,
+      "tokens_seen": 3019046912
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.594126143476168e-05,
+      "loss": 1.2374,
+      "theoretical_loss": 3.3476033846921953,
+      "tokens_seen": 3019177984
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.590113946397047e-05,
+      "loss": 1.3553,
+      "theoretical_loss": 3.3475923204152545,
+      "tokens_seen": 3019309056
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.586101749317927e-05,
+      "loss": 1.2809,
+      "theoretical_loss": 3.3475812567530987,
+      "tokens_seen": 3019440128
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.582089552238806e-05,
+      "loss": 1.341,
+      "theoretical_loss": 3.3475701937056677,
+      "tokens_seen": 3019571200
+    },
+    {
+      "epoch": 0.83,
+      "objective/train/advantage_avg": 0.492309033870697,
+      "objective/train/docs_used": 1097674,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2276958227157593,
+      "objective/train/original_loss": 1.2276957035064697,
+      "objective/train/theoretical_loss": 3.347559131272899,
+      "objective/train/tokens_used": 1390161376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24281731247901917,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504653453826904,
+      "objective/train/weighted_lm_loss": 1.2894574403762817,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.347559131272899,
+      "tokens_seen": 3019702272
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.578077355159686e-05,
+      "loss": 1.2793,
+      "theoretical_loss": 3.347559131272899,
+      "tokens_seen": 3019702272
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.574065158080566e-05,
+      "loss": 1.345,
+      "theoretical_loss": 3.3475480694547337,
+      "tokens_seen": 3019833344
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.570052961001444e-05,
+      "loss": 1.3261,
+      "theoretical_loss": 3.34753700825111,
+      "tokens_seen": 3019964416
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.566040763922324e-05,
+      "loss": 1.2612,
+      "theoretical_loss": 3.3475259476619668,
+      "tokens_seen": 3020095488
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.562028566843204e-05,
+      "loss": 1.3024,
+      "theoretical_loss": 3.347514887687244,
+      "tokens_seen": 3020226560
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.558016369764084e-05,
+      "loss": 1.3099,
+      "theoretical_loss": 3.3475038283268805,
+      "tokens_seen": 3020357632
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.554004172684962e-05,
+      "loss": 1.2759,
+      "theoretical_loss": 3.3474927695808154,
+      "tokens_seen": 3020488704
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.549991975605841e-05,
+      "loss": 1.2949,
+      "theoretical_loss": 3.3474817114489883,
+      "tokens_seen": 3020619776
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.545979778526722e-05,
+      "loss": 1.2892,
+      "theoretical_loss": 3.3474706539313384,
+      "tokens_seen": 3020750848
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.5419675814476e-05,
+      "loss": 1.2794,
+      "theoretical_loss": 3.3474595970278043,
+      "tokens_seen": 3020881920
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.53795538436848e-05,
+      "loss": 1.2937,
+      "theoretical_loss": 3.3474485407383257,
+      "tokens_seen": 3021012992
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.53394318728936e-05,
+      "loss": 1.3316,
+      "theoretical_loss": 3.3474374850628417,
+      "tokens_seen": 3021144064
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.52993099021024e-05,
+      "loss": 1.3103,
+      "theoretical_loss": 3.347426430001292,
+      "tokens_seen": 3021275136
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.525918793131118e-05,
+      "loss": 1.3577,
+      "theoretical_loss": 3.3474153755536156,
+      "tokens_seen": 3021406208
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.521906596051998e-05,
+      "loss": 1.267,
+      "theoretical_loss": 3.3474043217197513,
+      "tokens_seen": 3021537280
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.517894398972878e-05,
+      "loss": 1.2915,
+      "theoretical_loss": 3.3473932684996393,
+      "tokens_seen": 3021668352
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.513882201893758e-05,
+      "loss": 1.3243,
+      "theoretical_loss": 3.3473822158932185,
+      "tokens_seen": 3021799424
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.509870004814637e-05,
+      "loss": 1.3372,
+      "theoretical_loss": 3.3473711639004278,
+      "tokens_seen": 3021930496
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.505857807735516e-05,
+      "loss": 1.2483,
+      "theoretical_loss": 3.347360112521207,
+      "tokens_seen": 3022061568
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.501845610656395e-05,
+      "loss": 1.2929,
+      "theoretical_loss": 3.3473490617554953,
+      "tokens_seen": 3022192640
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.497833413577275e-05,
+      "loss": 1.3106,
+      "theoretical_loss": 3.3473380116032323,
+      "tokens_seen": 3022323712
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.493821216498154e-05,
+      "loss": 1.2829,
+      "theoretical_loss": 3.347326962064357,
+      "tokens_seen": 3022454784
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.489809019419035e-05,
+      "loss": 1.2704,
+      "theoretical_loss": 3.3473159131388086,
+      "tokens_seen": 3022585856
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.485796822339914e-05,
+      "loss": 1.2686,
+      "theoretical_loss": 3.347304864826527,
+      "tokens_seen": 3022716928
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.481784625260793e-05,
+      "loss": 1.2515,
+      "theoretical_loss": 3.347293817127451,
+      "tokens_seen": 3022848000
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.477772428181672e-05,
+      "loss": 1.2658,
+      "theoretical_loss": 3.34728277004152,
+      "tokens_seen": 3022979072
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.473760231102552e-05,
+      "loss": 1.265,
+      "theoretical_loss": 3.3472717235686744,
+      "tokens_seen": 3023110144
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.469748034023432e-05,
+      "loss": 1.3042,
+      "theoretical_loss": 3.3472606777088525,
+      "tokens_seen": 3023241216
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.46573583694431e-05,
+      "loss": 1.34,
+      "theoretical_loss": 3.347249632461994,
+      "tokens_seen": 3023372288
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.461723639865191e-05,
+      "loss": 1.2322,
+      "theoretical_loss": 3.347238587828038,
+      "tokens_seen": 3023503360
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.45771144278607e-05,
+      "loss": 1.3214,
+      "theoretical_loss": 3.347227543806925,
+      "tokens_seen": 3023634432
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.453699245706948e-05,
+      "loss": 1.2856,
+      "theoretical_loss": 3.347216500398593,
+      "tokens_seen": 3023765504
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.449687048627829e-05,
+      "loss": 1.2397,
+      "theoretical_loss": 3.347205457602983,
+      "tokens_seen": 3023896576
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.445674851548708e-05,
+      "loss": 1.2417,
+      "theoretical_loss": 3.347194415420033,
+      "tokens_seen": 3024027648
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.441662654469589e-05,
+      "loss": 1.2889,
+      "theoretical_loss": 3.347183373849683,
+      "tokens_seen": 3024158720
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.437650457390466e-05,
+      "loss": 1.2589,
+      "theoretical_loss": 3.3471723328918728,
+      "tokens_seen": 3024289792
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.433638260311347e-05,
+      "loss": 1.2825,
+      "theoretical_loss": 3.3471612925465415,
+      "tokens_seen": 3024420864
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.429626063232226e-05,
+      "loss": 1.3428,
+      "theoretical_loss": 3.3471502528136283,
+      "tokens_seen": 3024551936
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.425613866153106e-05,
+      "loss": 1.3725,
+      "theoretical_loss": 3.3471392136930733,
+      "tokens_seen": 3024683008
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.421601669073985e-05,
+      "loss": 1.2187,
+      "theoretical_loss": 3.347128175184816,
+      "tokens_seen": 3024814080
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.417589471994864e-05,
+      "loss": 1.3369,
+      "theoretical_loss": 3.347117137288796,
+      "tokens_seen": 3024945152
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.413577274915745e-05,
+      "loss": 1.3261,
+      "theoretical_loss": 3.3471061000049516,
+      "tokens_seen": 3025076224
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.409565077836623e-05,
+      "loss": 1.2644,
+      "theoretical_loss": 3.3470950633332235,
+      "tokens_seen": 3025207296
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.405552880757502e-05,
+      "loss": 1.2934,
+      "theoretical_loss": 3.347084027273551,
+      "tokens_seen": 3025338368
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.401540683678383e-05,
+      "loss": 1.2767,
+      "theoretical_loss": 3.347072991825874,
+      "tokens_seen": 3025469440
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.397528486599262e-05,
+      "loss": 1.4153,
+      "theoretical_loss": 3.347061956990131,
+      "tokens_seen": 3025600512
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.393516289520141e-05,
+      "loss": 1.2219,
+      "theoretical_loss": 3.347050922766263,
+      "tokens_seen": 3025731584
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.38950409244102e-05,
+      "loss": 1.3468,
+      "theoretical_loss": 3.347039889154208,
+      "tokens_seen": 3025862656
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.385491895361901e-05,
+      "loss": 1.2619,
+      "theoretical_loss": 3.3470288561539068,
+      "tokens_seen": 3025993728
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.38147969828278e-05,
+      "loss": 1.3463,
+      "theoretical_loss": 3.3470178237652983,
+      "tokens_seen": 3026124800
+    },
+    {
+      "epoch": 0.83,
+      "objective/train/advantage_avg": 0.4922446012496948,
+      "objective/train/docs_used": 1099986,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2712173461914062,
+      "objective/train/original_loss": 1.2712173461914062,
+      "objective/train/theoretical_loss": 3.3470067919883224,
+      "objective/train/tokens_used": 1396714976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.242795929312706,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504587888717651,
+      "objective/train/weighted_lm_loss": 1.3350766897201538,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3470067919883224,
+      "tokens_seen": 3026255872
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.377467501203658e-05,
+      "loss": 1.3256,
+      "theoretical_loss": 3.3470067919883224,
+      "tokens_seen": 3026255872
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.373455304124539e-05,
+      "loss": 1.3752,
+      "theoretical_loss": 3.3469957608229186,
+      "tokens_seen": 3026386944
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.369443107045418e-05,
+      "loss": 1.372,
+      "theoretical_loss": 3.3469847302690265,
+      "tokens_seen": 3026518016
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.365430909966299e-05,
+      "loss": 1.364,
+      "theoretical_loss": 3.346973700326586,
+      "tokens_seen": 3026649088
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.361418712887177e-05,
+      "loss": 1.271,
+      "theoretical_loss": 3.346962670995536,
+      "tokens_seen": 3026780160
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.357406515808056e-05,
+      "loss": 1.3316,
+      "theoretical_loss": 3.346951642275817,
+      "tokens_seen": 3026911232
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.353394318728937e-05,
+      "loss": 1.2531,
+      "theoretical_loss": 3.3469406141673685,
+      "tokens_seen": 3027042304
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.349382121649815e-05,
+      "loss": 1.3176,
+      "theoretical_loss": 3.3469295866701296,
+      "tokens_seen": 3027173376
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.345369924570695e-05,
+      "loss": 1.3392,
+      "theoretical_loss": 3.34691855978404,
+      "tokens_seen": 3027304448
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.341357727491575e-05,
+      "loss": 1.3458,
+      "theoretical_loss": 3.3469075335090404,
+      "tokens_seen": 3027435520
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 8.337345530412455e-05,
+      "loss": 1.3159,
+      "theoretical_loss": 3.346896507845069,
+      "tokens_seen": 3027566592
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.333333333333333e-05,
+      "loss": 1.3205,
+      "theoretical_loss": 3.3468854827920667,
+      "tokens_seen": 3027697664
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.329321136254212e-05,
+      "loss": 1.3097,
+      "theoretical_loss": 3.3468744583499728,
+      "tokens_seen": 3027828736
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.325308939175093e-05,
+      "loss": 1.3411,
+      "theoretical_loss": 3.3468634345187267,
+      "tokens_seen": 3027959808
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.321296742095972e-05,
+      "loss": 1.2783,
+      "theoretical_loss": 3.3468524112982685,
+      "tokens_seen": 3028090880
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.317284545016852e-05,
+      "loss": 1.2358,
+      "theoretical_loss": 3.3468413886885373,
+      "tokens_seen": 3028221952
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.313272347937731e-05,
+      "loss": 1.2273,
+      "theoretical_loss": 3.3468303666894736,
+      "tokens_seen": 3028353024
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.309260150858611e-05,
+      "loss": 1.2906,
+      "theoretical_loss": 3.346819345301017,
+      "tokens_seen": 3028484096
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.30524795377949e-05,
+      "loss": 1.2204,
+      "theoretical_loss": 3.346808324523107,
+      "tokens_seen": 3028615168
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.301235756700369e-05,
+      "loss": 1.3192,
+      "theoretical_loss": 3.3467973043556833,
+      "tokens_seen": 3028746240
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.297223559621249e-05,
+      "loss": 1.3887,
+      "theoretical_loss": 3.346786284798686,
+      "tokens_seen": 3028877312
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.293211362542129e-05,
+      "loss": 1.3486,
+      "theoretical_loss": 3.3467752658520546,
+      "tokens_seen": 3029008384
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.289199165463008e-05,
+      "loss": 1.3119,
+      "theoretical_loss": 3.346764247515729,
+      "tokens_seen": 3029139456
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.285186968383887e-05,
+      "loss": 1.3191,
+      "theoretical_loss": 3.3467532297896487,
+      "tokens_seen": 3029270528
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.281174771304766e-05,
+      "loss": 1.2904,
+      "theoretical_loss": 3.346742212673754,
+      "tokens_seen": 3029401600
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.277162574225647e-05,
+      "loss": 1.3387,
+      "theoretical_loss": 3.3467311961679846,
+      "tokens_seen": 3029532672
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.273150377146525e-05,
+      "loss": 1.274,
+      "theoretical_loss": 3.34672018027228,
+      "tokens_seen": 3029663744
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.269138180067406e-05,
+      "loss": 1.2241,
+      "theoretical_loss": 3.34670916498658,
+      "tokens_seen": 3029794816
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.265125982988285e-05,
+      "loss": 1.3312,
+      "theoretical_loss": 3.346698150310825,
+      "tokens_seen": 3029925888
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.261113785909164e-05,
+      "loss": 1.3225,
+      "theoretical_loss": 3.346687136244954,
+      "tokens_seen": 3030056960
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.257101588830043e-05,
+      "loss": 1.2655,
+      "theoretical_loss": 3.3466761227889084,
+      "tokens_seen": 3030188032
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.253089391750923e-05,
+      "loss": 1.3144,
+      "theoretical_loss": 3.346665109942626,
+      "tokens_seen": 3030319104
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.249077194671803e-05,
+      "loss": 1.3242,
+      "theoretical_loss": 3.3466540977060477,
+      "tokens_seen": 3030450176
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.245064997592681e-05,
+      "loss": 1.2465,
+      "theoretical_loss": 3.346643086079114,
+      "tokens_seen": 3030581248
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.241052800513562e-05,
+      "loss": 1.3495,
+      "theoretical_loss": 3.3466320750617635,
+      "tokens_seen": 3030712320
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.237040603434441e-05,
+      "loss": 1.3437,
+      "theoretical_loss": 3.3466210646539367,
+      "tokens_seen": 3030843392
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.23302840635532e-05,
+      "loss": 1.3436,
+      "theoretical_loss": 3.346610054855574,
+      "tokens_seen": 3030974464
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.2290162092762e-05,
+      "loss": 1.2944,
+      "theoretical_loss": 3.3465990456666144,
+      "tokens_seen": 3031105536
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.225004012197079e-05,
+      "loss": 1.2745,
+      "theoretical_loss": 3.3465880370869985,
+      "tokens_seen": 3031236608
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.22099181511796e-05,
+      "loss": 1.332,
+      "theoretical_loss": 3.346577029116666,
+      "tokens_seen": 3031367680
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.216979618038837e-05,
+      "loss": 1.4074,
+      "theoretical_loss": 3.3465660217555566,
+      "tokens_seen": 3031498752
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.212967420959718e-05,
+      "loss": 1.3083,
+      "theoretical_loss": 3.3465550150036103,
+      "tokens_seen": 3031629824
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.208955223880597e-05,
+      "loss": 1.3525,
+      "theoretical_loss": 3.346544008860768,
+      "tokens_seen": 3031760896
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.204943026801477e-05,
+      "loss": 1.2953,
+      "theoretical_loss": 3.346533003326968,
+      "tokens_seen": 3031891968
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.200930829722356e-05,
+      "loss": 1.4001,
+      "theoretical_loss": 3.3465219984021513,
+      "tokens_seen": 3032023040
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.196918632643235e-05,
+      "loss": 1.3813,
+      "theoretical_loss": 3.346510994086258,
+      "tokens_seen": 3032154112
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.192906435564116e-05,
+      "loss": 1.2438,
+      "theoretical_loss": 3.3464999903792276,
+      "tokens_seen": 3032285184
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.188894238484995e-05,
+      "loss": 1.3696,
+      "theoretical_loss": 3.3464889872810004,
+      "tokens_seen": 3032416256
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.184882041405873e-05,
+      "loss": 1.2671,
+      "theoretical_loss": 3.3464779847915165,
+      "tokens_seen": 3032547328
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.180869844326754e-05,
+      "loss": 1.3526,
+      "theoretical_loss": 3.3464669829107154,
+      "tokens_seen": 3032678400
+    },
+    {
+      "epoch": 0.84,
+      "objective/train/advantage_avg": 0.4925715923309326,
+      "objective/train/docs_used": 1102459,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2340601682662964,
+      "objective/train/original_loss": 1.2340600490570068,
+      "objective/train/theoretical_loss": 3.3464559816385373,
+      "objective/train/tokens_used": 1403268576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24304991960525513,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050492763519287,
+      "objective/train/weighted_lm_loss": 1.2962785959243774,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3464559816385373,
+      "tokens_seen": 3032809472
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.176857647247633e-05,
+      "loss": 1.3114,
+      "theoretical_loss": 3.3464559816385373,
+      "tokens_seen": 3032809472
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.172845450168512e-05,
+      "loss": 1.2788,
+      "theoretical_loss": 3.346444980974923,
+      "tokens_seen": 3032940544
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.168833253089391e-05,
+      "loss": 1.2989,
+      "theoretical_loss": 3.3464339809198114,
+      "tokens_seen": 3033071616
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.164821056010272e-05,
+      "loss": 1.2712,
+      "theoretical_loss": 3.346422981473143,
+      "tokens_seen": 3033202688
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.160808858931151e-05,
+      "loss": 1.2617,
+      "theoretical_loss": 3.346411982634858,
+      "tokens_seen": 3033333760
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.156796661852029e-05,
+      "loss": 1.3031,
+      "theoretical_loss": 3.3464009844048968,
+      "tokens_seen": 3033464832
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.15278446477291e-05,
+      "loss": 1.2818,
+      "theoretical_loss": 3.3463899867831985,
+      "tokens_seen": 3033595904
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.148772267693789e-05,
+      "loss": 1.3332,
+      "theoretical_loss": 3.346378989769704,
+      "tokens_seen": 3033726976
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.14476007061467e-05,
+      "loss": 1.3349,
+      "theoretical_loss": 3.3463679933643533,
+      "tokens_seen": 3033858048
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.140747873535548e-05,
+      "loss": 1.3132,
+      "theoretical_loss": 3.346356997567086,
+      "tokens_seen": 3033989120
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.136735676456427e-05,
+      "loss": 1.3596,
+      "theoretical_loss": 3.3463460023778424,
+      "tokens_seen": 3034120192
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.132723479377308e-05,
+      "loss": 1.3352,
+      "theoretical_loss": 3.3463350077965632,
+      "tokens_seen": 3034251264
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.128711282298187e-05,
+      "loss": 1.2769,
+      "theoretical_loss": 3.3463240138231876,
+      "tokens_seen": 3034382336
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.124699085219066e-05,
+      "loss": 1.3654,
+      "theoretical_loss": 3.3463130204576563,
+      "tokens_seen": 3034513408
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.120686888139945e-05,
+      "loss": 1.2457,
+      "theoretical_loss": 3.3463020276999096,
+      "tokens_seen": 3034644480
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.116674691060826e-05,
+      "loss": 1.3775,
+      "theoretical_loss": 3.3462910355498874,
+      "tokens_seen": 3034775552
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.112662493981704e-05,
+      "loss": 1.2394,
+      "theoretical_loss": 3.3462800440075293,
+      "tokens_seen": 3034906624
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.108650296902583e-05,
+      "loss": 1.3315,
+      "theoretical_loss": 3.3462690530727763,
+      "tokens_seen": 3035037696
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.104638099823464e-05,
+      "loss": 1.2977,
+      "theoretical_loss": 3.3462580627455685,
+      "tokens_seen": 3035168768
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.100625902744343e-05,
+      "loss": 1.3559,
+      "theoretical_loss": 3.346247073025846,
+      "tokens_seen": 3035299840
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.096613705665222e-05,
+      "loss": 1.317,
+      "theoretical_loss": 3.3462360839135483,
+      "tokens_seen": 3035430912
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.092601508586102e-05,
+      "loss": 1.3336,
+      "theoretical_loss": 3.346225095408616,
+      "tokens_seen": 3035561984
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.088589311506981e-05,
+      "loss": 1.316,
+      "theoretical_loss": 3.34621410751099,
+      "tokens_seen": 3035693056
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.084577114427862e-05,
+      "loss": 1.3333,
+      "theoretical_loss": 3.34620312022061,
+      "tokens_seen": 3035824128
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.08056491734874e-05,
+      "loss": 1.405,
+      "theoretical_loss": 3.346192133537416,
+      "tokens_seen": 3035955200
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.07655272026962e-05,
+      "loss": 1.3522,
+      "theoretical_loss": 3.346181147461348,
+      "tokens_seen": 3036086272
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.0725405231905e-05,
+      "loss": 1.3807,
+      "theoretical_loss": 3.3461701619923474,
+      "tokens_seen": 3036217344
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.068528326111379e-05,
+      "loss": 1.2608,
+      "theoretical_loss": 3.346159177130353,
+      "tokens_seen": 3036348416
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.064516129032258e-05,
+      "loss": 1.3652,
+      "theoretical_loss": 3.3461481928753063,
+      "tokens_seen": 3036479488
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.060503931953137e-05,
+      "loss": 1.3554,
+      "theoretical_loss": 3.3461372092271473,
+      "tokens_seen": 3036610560
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.056491734874018e-05,
+      "loss": 1.3881,
+      "theoretical_loss": 3.3461262261858153,
+      "tokens_seen": 3036741632
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.052479537794896e-05,
+      "loss": 1.3138,
+      "theoretical_loss": 3.346115243751252,
+      "tokens_seen": 3036872704
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.048467340715777e-05,
+      "loss": 1.2471,
+      "theoretical_loss": 3.3461042619233963,
+      "tokens_seen": 3037003776
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.044455143636656e-05,
+      "loss": 1.348,
+      "theoretical_loss": 3.3460932807021897,
+      "tokens_seen": 3037134848
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.040442946557535e-05,
+      "loss": 1.3697,
+      "theoretical_loss": 3.3460823000875717,
+      "tokens_seen": 3037265920
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.036430749478414e-05,
+      "loss": 1.2999,
+      "theoretical_loss": 3.346071320079483,
+      "tokens_seen": 3037396992
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.032418552399294e-05,
+      "loss": 1.2793,
+      "theoretical_loss": 3.346060340677864,
+      "tokens_seen": 3037528064
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.028406355320174e-05,
+      "loss": 1.3401,
+      "theoretical_loss": 3.3460493618826552,
+      "tokens_seen": 3037659136
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.024394158241052e-05,
+      "loss": 1.358,
+      "theoretical_loss": 3.346038383693796,
+      "tokens_seen": 3037790208
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.020381961161933e-05,
+      "loss": 1.3112,
+      "theoretical_loss": 3.3460274061112276,
+      "tokens_seen": 3037921280
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.016369764082812e-05,
+      "loss": 1.3091,
+      "theoretical_loss": 3.3460164291348904,
+      "tokens_seen": 3038052352
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.012357567003691e-05,
+      "loss": 1.3179,
+      "theoretical_loss": 3.3460054527647243,
+      "tokens_seen": 3038183424
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.00834536992457e-05,
+      "loss": 1.2983,
+      "theoretical_loss": 3.3459944770006698,
+      "tokens_seen": 3038314496
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.00433317284545e-05,
+      "loss": 1.2666,
+      "theoretical_loss": 3.3459835018426674,
+      "tokens_seen": 3038445568
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 8.00032097576633e-05,
+      "loss": 1.2088,
+      "theoretical_loss": 3.3459725272906575,
+      "tokens_seen": 3038576640
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.99630877868721e-05,
+      "loss": 1.2979,
+      "theoretical_loss": 3.345961553344581,
+      "tokens_seen": 3038707712
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.992296581608088e-05,
+      "loss": 1.2842,
+      "theoretical_loss": 3.345950580004377,
+      "tokens_seen": 3038838784
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.988284384528968e-05,
+      "loss": 1.2999,
+      "theoretical_loss": 3.345939607269987,
+      "tokens_seen": 3038969856
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.984272187449848e-05,
+      "loss": 1.2578,
+      "theoretical_loss": 3.345928635141351,
+      "tokens_seen": 3039100928
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.980259990370727e-05,
+      "loss": 1.3633,
+      "theoretical_loss": 3.34591766361841,
+      "tokens_seen": 3039232000
+    },
+    {
+      "epoch": 0.84,
+      "objective/train/advantage_avg": 0.49323543906211853,
+      "objective/train/docs_used": 1104383,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.1944180727005005,
+      "objective/train/original_loss": 1.1944180727005005,
+      "objective/train/theoretical_loss": 3.3459066927011034,
+      "objective/train/tokens_used": 1409822176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24392791092395782,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505635738372803,
+      "objective/train/weighted_lm_loss": 1.254874348640442,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3459066927011034,
+      "tokens_seen": 3039363072
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.976247793291606e-05,
+      "loss": 1.2611,
+      "theoretical_loss": 3.3459066927011034,
+      "tokens_seen": 3039363072
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.972235596212487e-05,
+      "loss": 1.2893,
+      "theoretical_loss": 3.345895722389373,
+      "tokens_seen": 3039494144
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.968223399133366e-05,
+      "loss": 1.3689,
+      "theoretical_loss": 3.345884752683158,
+      "tokens_seen": 3039625216
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.964211202054244e-05,
+      "loss": 1.2942,
+      "theoretical_loss": 3.345873783582399,
+      "tokens_seen": 3039756288
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.960199004975125e-05,
+      "loss": 1.3433,
+      "theoretical_loss": 3.3458628150870378,
+      "tokens_seen": 3039887360
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.956186807896004e-05,
+      "loss": 1.2669,
+      "theoretical_loss": 3.345851847197013,
+      "tokens_seen": 3040018432
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.952174610816885e-05,
+      "loss": 1.249,
+      "theoretical_loss": 3.345840879912267,
+      "tokens_seen": 3040149504
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.948162413737762e-05,
+      "loss": 1.3318,
+      "theoretical_loss": 3.345829913232739,
+      "tokens_seen": 3040280576
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.944150216658642e-05,
+      "loss": 1.2493,
+      "theoretical_loss": 3.3458189471583695,
+      "tokens_seen": 3040411648
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.940138019579522e-05,
+      "loss": 1.288,
+      "theoretical_loss": 3.3458079816890995,
+      "tokens_seen": 3040542720
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.9361258225004e-05,
+      "loss": 1.305,
+      "theoretical_loss": 3.3457970168248696,
+      "tokens_seen": 3040673792
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.932113625421281e-05,
+      "loss": 1.3632,
+      "theoretical_loss": 3.3457860525656202,
+      "tokens_seen": 3040804864
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.92810142834216e-05,
+      "loss": 1.3121,
+      "theoretical_loss": 3.345775088911292,
+      "tokens_seen": 3040935936
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.924089231263041e-05,
+      "loss": 1.2624,
+      "theoretical_loss": 3.345764125861825,
+      "tokens_seen": 3041067008
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.920077034183919e-05,
+      "loss": 1.3837,
+      "theoretical_loss": 3.34575316341716,
+      "tokens_seen": 3041198080
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.916064837104798e-05,
+      "loss": 1.3372,
+      "theoretical_loss": 3.345742201577238,
+      "tokens_seen": 3041329152
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.912052640025679e-05,
+      "loss": 1.2685,
+      "theoretical_loss": 3.345731240341999,
+      "tokens_seen": 3041460224
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.908040442946558e-05,
+      "loss": 1.3192,
+      "theoretical_loss": 3.345720279711384,
+      "tokens_seen": 3041591296
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.904028245867437e-05,
+      "loss": 1.3237,
+      "theoretical_loss": 3.3457093196853336,
+      "tokens_seen": 3041722368
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.900016048788316e-05,
+      "loss": 1.41,
+      "theoretical_loss": 3.345698360263788,
+      "tokens_seen": 3041853440
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.896003851709197e-05,
+      "loss": 1.2714,
+      "theoretical_loss": 3.345687401446688,
+      "tokens_seen": 3041984512
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.891991654630075e-05,
+      "loss": 1.2667,
+      "theoretical_loss": 3.345676443233975,
+      "tokens_seen": 3042115584
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.887979457550954e-05,
+      "loss": 1.4775,
+      "theoretical_loss": 3.345665485625588,
+      "tokens_seen": 3042246656
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.883967260471835e-05,
+      "loss": 1.2956,
+      "theoretical_loss": 3.345654528621469,
+      "tokens_seen": 3042377728
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.879955063392714e-05,
+      "loss": 1.3453,
+      "theoretical_loss": 3.345643572221558,
+      "tokens_seen": 3042508800
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.875942866313593e-05,
+      "loss": 1.3495,
+      "theoretical_loss": 3.345632616425796,
+      "tokens_seen": 3042639872
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.871930669234473e-05,
+      "loss": 1.2167,
+      "theoretical_loss": 3.3456216612341234,
+      "tokens_seen": 3042770944
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.867918472155352e-05,
+      "loss": 1.4142,
+      "theoretical_loss": 3.345610706646481,
+      "tokens_seen": 3042902016
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.863906275076233e-05,
+      "loss": 1.3206,
+      "theoretical_loss": 3.34559975266281,
+      "tokens_seen": 3043033088
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.85989407799711e-05,
+      "loss": 1.3849,
+      "theoretical_loss": 3.3455887992830498,
+      "tokens_seen": 3043164160
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.855881880917991e-05,
+      "loss": 1.2148,
+      "theoretical_loss": 3.345577846507142,
+      "tokens_seen": 3043295232
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.85186968383887e-05,
+      "loss": 1.2923,
+      "theoretical_loss": 3.345566894335027,
+      "tokens_seen": 3043426304
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.847857486759751e-05,
+      "loss": 1.3764,
+      "theoretical_loss": 3.345555942766646,
+      "tokens_seen": 3043557376
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.843845289680629e-05,
+      "loss": 1.3191,
+      "theoretical_loss": 3.3455449918019395,
+      "tokens_seen": 3043688448
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.839833092601508e-05,
+      "loss": 1.2794,
+      "theoretical_loss": 3.345534041440848,
+      "tokens_seen": 3043819520
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.835820895522389e-05,
+      "loss": 1.3409,
+      "theoretical_loss": 3.3455230916833125,
+      "tokens_seen": 3043950592
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 7.831808698443267e-05,
+      "loss": 1.3315,
+      "theoretical_loss": 3.345512142529273,
+      "tokens_seen": 3044081664
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.827796501364148e-05,
+      "loss": 1.2711,
+      "theoretical_loss": 3.345501193978672,
+      "tokens_seen": 3044212736
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.823784304285027e-05,
+      "loss": 1.322,
+      "theoretical_loss": 3.345490246031448,
+      "tokens_seen": 3044343808
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.819772107205906e-05,
+      "loss": 1.3398,
+      "theoretical_loss": 3.3454792986875432,
+      "tokens_seen": 3044474880
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.815759910126785e-05,
+      "loss": 1.3606,
+      "theoretical_loss": 3.3454683519468986,
+      "tokens_seen": 3044605952
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.811747713047665e-05,
+      "loss": 1.2751,
+      "theoretical_loss": 3.3454574058094537,
+      "tokens_seen": 3044737024
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.807735515968545e-05,
+      "loss": 1.3139,
+      "theoretical_loss": 3.3454464602751504,
+      "tokens_seen": 3044868096
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.803723318889425e-05,
+      "loss": 1.3406,
+      "theoretical_loss": 3.3454355153439295,
+      "tokens_seen": 3044999168
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.799711121810304e-05,
+      "loss": 1.3736,
+      "theoretical_loss": 3.345424571015731,
+      "tokens_seen": 3045130240
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.795698924731183e-05,
+      "loss": 1.313,
+      "theoretical_loss": 3.345413627290496,
+      "tokens_seen": 3045261312
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.791686727652062e-05,
+      "loss": 1.2447,
+      "theoretical_loss": 3.345402684168166,
+      "tokens_seen": 3045392384
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.787674530572942e-05,
+      "loss": 1.1938,
+      "theoretical_loss": 3.3453917416486814,
+      "tokens_seen": 3045523456
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.783662333493821e-05,
+      "loss": 1.3597,
+      "theoretical_loss": 3.345380799731983,
+      "tokens_seen": 3045654528
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.779650136414702e-05,
+      "loss": 1.3043,
+      "theoretical_loss": 3.345369858418011,
+      "tokens_seen": 3045785600
+    },
+    {
+      "epoch": 0.85,
+      "objective/train/advantage_avg": 0.4917280972003937,
+      "objective/train/docs_used": 1106703,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2214179039001465,
+      "objective/train/original_loss": 1.221417784690857,
+      "objective/train/theoretical_loss": 3.345358917706708,
+      "objective/train/tokens_used": 1416375776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24272267520427704,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050406813621521,
+      "objective/train/weighted_lm_loss": 1.2829310894012451,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.345358917706708,
+      "tokens_seen": 3045916672
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.775637939335581e-05,
+      "loss": 1.3311,
+      "theoretical_loss": 3.345358917706708,
+      "tokens_seen": 3045916672
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.771625742256459e-05,
+      "loss": 1.3269,
+      "theoretical_loss": 3.3453479775980135,
+      "tokens_seen": 3046047744
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.76761354517734e-05,
+      "loss": 1.3131,
+      "theoretical_loss": 3.345337038091868,
+      "tokens_seen": 3046178816
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.763601348098219e-05,
+      "loss": 1.3068,
+      "theoretical_loss": 3.345326099188214,
+      "tokens_seen": 3046309888
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.759589151019099e-05,
+      "loss": 1.2924,
+      "theoretical_loss": 3.345315160886991,
+      "tokens_seen": 3046440960
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.755576953939977e-05,
+      "loss": 1.3093,
+      "theoretical_loss": 3.3453042231881405,
+      "tokens_seen": 3046572032
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.751564756860858e-05,
+      "loss": 1.3161,
+      "theoretical_loss": 3.3452932860916027,
+      "tokens_seen": 3046703104
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.747552559781737e-05,
+      "loss": 1.3137,
+      "theoretical_loss": 3.34528234959732,
+      "tokens_seen": 3046834176
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.743540362702615e-05,
+      "loss": 1.3512,
+      "theoretical_loss": 3.3452714137052317,
+      "tokens_seen": 3046965248
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.739528165623496e-05,
+      "loss": 1.2911,
+      "theoretical_loss": 3.34526047841528,
+      "tokens_seen": 3047096320
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.735515968544375e-05,
+      "loss": 1.3462,
+      "theoretical_loss": 3.3452495437274052,
+      "tokens_seen": 3047227392
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.731503771465256e-05,
+      "loss": 1.3574,
+      "theoretical_loss": 3.3452386096415485,
+      "tokens_seen": 3047358464
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.727491574386133e-05,
+      "loss": 1.3121,
+      "theoretical_loss": 3.3452276761576507,
+      "tokens_seen": 3047489536
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.723479377307013e-05,
+      "loss": 1.3142,
+      "theoretical_loss": 3.3452167432756528,
+      "tokens_seen": 3047620608
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.719467180227893e-05,
+      "loss": 1.3385,
+      "theoretical_loss": 3.3452058109954956,
+      "tokens_seen": 3047751680
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.715454983148773e-05,
+      "loss": 1.3796,
+      "theoretical_loss": 3.3451948793171207,
+      "tokens_seen": 3047882752
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.711442786069652e-05,
+      "loss": 1.3447,
+      "theoretical_loss": 3.345183948240468,
+      "tokens_seen": 3048013824
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.707430588990531e-05,
+      "loss": 1.3373,
+      "theoretical_loss": 3.3451730177654797,
+      "tokens_seen": 3048144896
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.703418391911412e-05,
+      "loss": 1.3305,
+      "theoretical_loss": 3.3451620878920965,
+      "tokens_seen": 3048275968
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.69940619483229e-05,
+      "loss": 1.2733,
+      "theoretical_loss": 3.3451511586202587,
+      "tokens_seen": 3048407040
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.695393997753169e-05,
+      "loss": 1.2834,
+      "theoretical_loss": 3.3451402299499082,
+      "tokens_seen": 3048538112
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.69138180067405e-05,
+      "loss": 1.3167,
+      "theoretical_loss": 3.345129301880986,
+      "tokens_seen": 3048669184
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.687369603594929e-05,
+      "loss": 1.3304,
+      "theoretical_loss": 3.345118374413432,
+      "tokens_seen": 3048800256
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.683357406515808e-05,
+      "loss": 1.3348,
+      "theoretical_loss": 3.3451074475471887,
+      "tokens_seen": 3048931328
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.679345209436687e-05,
+      "loss": 1.2954,
+      "theoretical_loss": 3.345096521282197,
+      "tokens_seen": 3049062400
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.675333012357567e-05,
+      "loss": 1.2316,
+      "theoretical_loss": 3.3450855956183965,
+      "tokens_seen": 3049193472
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.671320815278447e-05,
+      "loss": 1.3137,
+      "theoretical_loss": 3.34507467055573,
+      "tokens_seen": 3049324544
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.667308618199325e-05,
+      "loss": 1.3308,
+      "theoretical_loss": 3.3450637460941377,
+      "tokens_seen": 3049455616
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.663296421120206e-05,
+      "loss": 1.3294,
+      "theoretical_loss": 3.3450528222335607,
+      "tokens_seen": 3049586688
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.659284224041085e-05,
+      "loss": 1.2807,
+      "theoretical_loss": 3.3450418989739408,
+      "tokens_seen": 3049717760
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.655272026961964e-05,
+      "loss": 1.3049,
+      "theoretical_loss": 3.3450309763152184,
+      "tokens_seen": 3049848832
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.651259829882844e-05,
+      "loss": 1.2667,
+      "theoretical_loss": 3.3450200542573345,
+      "tokens_seen": 3049979904
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.647247632803723e-05,
+      "loss": 1.3108,
+      "theoretical_loss": 3.345009132800231,
+      "tokens_seen": 3050110976
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.643235435724604e-05,
+      "loss": 1.2706,
+      "theoretical_loss": 3.3449982119438486,
+      "tokens_seen": 3050242048
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.639223238645482e-05,
+      "loss": 1.3363,
+      "theoretical_loss": 3.3449872916881285,
+      "tokens_seen": 3050373120
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.635211041566362e-05,
+      "loss": 1.2837,
+      "theoretical_loss": 3.3449763720330115,
+      "tokens_seen": 3050504192
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.631198844487241e-05,
+      "loss": 1.3479,
+      "theoretical_loss": 3.344965452978439,
+      "tokens_seen": 3050635264
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.627186647408121e-05,
+      "loss": 1.3109,
+      "theoretical_loss": 3.3449545345243523,
+      "tokens_seen": 3050766336
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.623174450329e-05,
+      "loss": 1.3284,
+      "theoretical_loss": 3.344943616670693,
+      "tokens_seen": 3050897408
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.619162253249879e-05,
+      "loss": 1.2096,
+      "theoretical_loss": 3.3449326994174013,
+      "tokens_seen": 3051028480
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.61515005617076e-05,
+      "loss": 1.3453,
+      "theoretical_loss": 3.3449217827644193,
+      "tokens_seen": 3051159552
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.611137859091639e-05,
+      "loss": 1.3665,
+      "theoretical_loss": 3.344910866711688,
+      "tokens_seen": 3051290624
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.607125662012518e-05,
+      "loss": 1.2757,
+      "theoretical_loss": 3.3448999512591477,
+      "tokens_seen": 3051421696
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.603113464933398e-05,
+      "loss": 1.1914,
+      "theoretical_loss": 3.344889036406741,
+      "tokens_seen": 3051552768
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.599101267854277e-05,
+      "loss": 1.2898,
+      "theoretical_loss": 3.344878122154408,
+      "tokens_seen": 3051683840
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.595089070775156e-05,
+      "loss": 1.289,
+      "theoretical_loss": 3.3448672085020905,
+      "tokens_seen": 3051814912
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.591076873696036e-05,
+      "loss": 1.3186,
+      "theoretical_loss": 3.3448562954497296,
+      "tokens_seen": 3051945984
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.587064676616916e-05,
+      "loss": 1.3542,
+      "theoretical_loss": 3.344845382997267,
+      "tokens_seen": 3052077056
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.583052479537795e-05,
+      "loss": 1.3381,
+      "theoretical_loss": 3.3448344711446434,
+      "tokens_seen": 3052208128
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.579040282458673e-05,
+      "loss": 1.2556,
+      "theoretical_loss": 3.3448235598918004,
+      "tokens_seen": 3052339200
+    },
+    {
+      "epoch": 0.85,
+      "objective/train/advantage_avg": 0.49233606457710266,
+      "objective/train/docs_used": 1109048,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2499102354049683,
+      "objective/train/original_loss": 1.2499102354049683,
+      "objective/train/theoretical_loss": 3.344812649238679,
+      "objective/train/tokens_used": 1422929376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24310192465782166,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504695177078247,
+      "objective/train/weighted_lm_loss": 1.312795639038086,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.344812649238679,
+      "tokens_seen": 3052470272
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.575028085379554e-05,
+      "loss": 1.3064,
+      "theoretical_loss": 3.344812649238679,
+      "tokens_seen": 3052470272
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.571015888300433e-05,
+      "loss": 1.3113,
+      "theoretical_loss": 3.3448017391852205,
+      "tokens_seen": 3052601344
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.567003691221314e-05,
+      "loss": 1.376,
+      "theoretical_loss": 3.3447908297313664,
+      "tokens_seen": 3052732416
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.562991494142192e-05,
+      "loss": 1.3142,
+      "theoretical_loss": 3.344779920877058,
+      "tokens_seen": 3052863488
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.558979297063073e-05,
+      "loss": 1.1986,
+      "theoretical_loss": 3.3447690126222365,
+      "tokens_seen": 3052994560
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.554967099983952e-05,
+      "loss": 1.2653,
+      "theoretical_loss": 3.344758104966843,
+      "tokens_seen": 3053125632
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.55095490290483e-05,
+      "loss": 1.342,
+      "theoretical_loss": 3.344747197910819,
+      "tokens_seen": 3053256704
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.54694270582571e-05,
+      "loss": 1.297,
+      "theoretical_loss": 3.3447362914541063,
+      "tokens_seen": 3053387776
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.54293050874659e-05,
+      "loss": 1.325,
+      "theoretical_loss": 3.344725385596646,
+      "tokens_seen": 3053518848
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.53891831166747e-05,
+      "loss": 1.3553,
+      "theoretical_loss": 3.3447144803383795,
+      "tokens_seen": 3053649920
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.534906114588348e-05,
+      "loss": 1.261,
+      "theoretical_loss": 3.3447035756792474,
+      "tokens_seen": 3053780992
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.530893917509227e-05,
+      "loss": 1.3189,
+      "theoretical_loss": 3.3446926716191916,
+      "tokens_seen": 3053912064
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.526881720430108e-05,
+      "loss": 1.3081,
+      "theoretical_loss": 3.344681768158154,
+      "tokens_seen": 3054043136
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.522869523350987e-05,
+      "loss": 1.25,
+      "theoretical_loss": 3.344670865296075,
+      "tokens_seen": 3054174208
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.518857326271867e-05,
+      "loss": 1.2543,
+      "theoretical_loss": 3.344659963032897,
+      "tokens_seen": 3054305280
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.514845129192746e-05,
+      "loss": 1.3038,
+      "theoretical_loss": 3.3446490613685604,
+      "tokens_seen": 3054436352
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.510832932113627e-05,
+      "loss": 1.2532,
+      "theoretical_loss": 3.3446381603030075,
+      "tokens_seen": 3054567424
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.506820735034504e-05,
+      "loss": 1.2796,
+      "theoretical_loss": 3.3446272598361793,
+      "tokens_seen": 3054698496
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.502808537955384e-05,
+      "loss": 1.2264,
+      "theoretical_loss": 3.344616359968017,
+      "tokens_seen": 3054829568
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.498796340876264e-05,
+      "loss": 1.2353,
+      "theoretical_loss": 3.3446054606984625,
+      "tokens_seen": 3054960640
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.494784143797144e-05,
+      "loss": 1.2689,
+      "theoretical_loss": 3.3445945620274573,
+      "tokens_seen": 3055091712
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.490771946718023e-05,
+      "loss": 1.2479,
+      "theoretical_loss": 3.344583663954942,
+      "tokens_seen": 3055222784
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.486759749638902e-05,
+      "loss": 1.2825,
+      "theoretical_loss": 3.344572766480859,
+      "tokens_seen": 3055353856
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.482747552559781e-05,
+      "loss": 1.3233,
+      "theoretical_loss": 3.34456186960515,
+      "tokens_seen": 3055484928
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.478735355480662e-05,
+      "loss": 1.3049,
+      "theoretical_loss": 3.344550973327755,
+      "tokens_seen": 3055616000
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.47472315840154e-05,
+      "loss": 1.3475,
+      "theoretical_loss": 3.3445400776486167,
+      "tokens_seen": 3055747072
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.47071096132242e-05,
+      "loss": 1.208,
+      "theoretical_loss": 3.344529182567676,
+      "tokens_seen": 3055878144
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.4666987642433e-05,
+      "loss": 1.2819,
+      "theoretical_loss": 3.3445182880848754,
+      "tokens_seen": 3056009216
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.462686567164179e-05,
+      "loss": 1.3186,
+      "theoretical_loss": 3.3445073942001553,
+      "tokens_seen": 3056140288
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.458674370085058e-05,
+      "loss": 1.2866,
+      "theoretical_loss": 3.3444965009134577,
+      "tokens_seen": 3056271360
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.454662173005938e-05,
+      "loss": 1.346,
+      "theoretical_loss": 3.344485608224724,
+      "tokens_seen": 3056402432
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.450649975926818e-05,
+      "loss": 1.2544,
+      "theoretical_loss": 3.3444747161338957,
+      "tokens_seen": 3056533504
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.446637778847696e-05,
+      "loss": 1.2336,
+      "theoretical_loss": 3.3444638246409144,
+      "tokens_seen": 3056664576
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.442625581768577e-05,
+      "loss": 1.2847,
+      "theoretical_loss": 3.344452933745722,
+      "tokens_seen": 3056795648
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.438613384689456e-05,
+      "loss": 1.3135,
+      "theoretical_loss": 3.3444420434482596,
+      "tokens_seen": 3056926720
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.434601187610337e-05,
+      "loss": 1.2977,
+      "theoretical_loss": 3.3444311537484683,
+      "tokens_seen": 3057057792
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.430588990531215e-05,
+      "loss": 1.2831,
+      "theoretical_loss": 3.344420264646291,
+      "tokens_seen": 3057188864
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.426576793452094e-05,
+      "loss": 1.2156,
+      "theoretical_loss": 3.3444093761416687,
+      "tokens_seen": 3057319936
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.422564596372975e-05,
+      "loss": 1.2623,
+      "theoretical_loss": 3.344398488234542,
+      "tokens_seen": 3057451008
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.418552399293853e-05,
+      "loss": 1.333,
+      "theoretical_loss": 3.344387600924854,
+      "tokens_seen": 3057582080
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.414540202214733e-05,
+      "loss": 1.2072,
+      "theoretical_loss": 3.3443767142125456,
+      "tokens_seen": 3057713152
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.410528005135612e-05,
+      "loss": 1.2207,
+      "theoretical_loss": 3.3443658280975583,
+      "tokens_seen": 3057844224
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.406515808056492e-05,
+      "loss": 1.296,
+      "theoretical_loss": 3.344354942579834,
+      "tokens_seen": 3057975296
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.402503610977371e-05,
+      "loss": 1.2743,
+      "theoretical_loss": 3.344344057659314,
+      "tokens_seen": 3058106368
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.39849141389825e-05,
+      "loss": 1.2352,
+      "theoretical_loss": 3.344333173335941,
+      "tokens_seen": 3058237440
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.394479216819131e-05,
+      "loss": 1.2636,
+      "theoretical_loss": 3.344322289609655,
+      "tokens_seen": 3058368512
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.39046701974001e-05,
+      "loss": 1.3535,
+      "theoretical_loss": 3.344311406480399,
+      "tokens_seen": 3058499584
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.38645482266089e-05,
+      "loss": 1.3817,
+      "theoretical_loss": 3.3443005239481134,
+      "tokens_seen": 3058630656
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.382442625581769e-05,
+      "loss": 1.1443,
+      "theoretical_loss": 3.3442896420127415,
+      "tokens_seen": 3058761728
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.378430428502648e-05,
+      "loss": 1.3005,
+      "theoretical_loss": 3.3442787606742232,
+      "tokens_seen": 3058892800
+    },
+    {
+      "epoch": 0.85,
+      "objective/train/advantage_avg": 0.49478521943092346,
+      "objective/train/docs_used": 1111330,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.1516021490097046,
+      "objective/train/original_loss": 1.151602029800415,
+      "objective/train/theoretical_loss": 3.344267879932502,
+      "objective/train/tokens_used": 1429482976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24507014453411102,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0507245063781738,
+      "objective/train/weighted_lm_loss": 1.2100579738616943,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.344267879932502,
+      "tokens_seen": 3059023872
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.374418231423529e-05,
+      "loss": 1.2264,
+      "theoretical_loss": 3.344267879932502,
+      "tokens_seen": 3059023872
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.370406034344407e-05,
+      "loss": 1.3286,
+      "theoretical_loss": 3.3442569997875182,
+      "tokens_seen": 3059154944
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.366393837265287e-05,
+      "loss": 1.2911,
+      "theoretical_loss": 3.344246120239214,
+      "tokens_seen": 3059286016
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.362381640186166e-05,
+      "loss": 1.3277,
+      "theoretical_loss": 3.3442352412875316,
+      "tokens_seen": 3059417088
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.358369443107044e-05,
+      "loss": 1.235,
+      "theoretical_loss": 3.3442243629324118,
+      "tokens_seen": 3059548160
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.354357246027925e-05,
+      "loss": 1.2499,
+      "theoretical_loss": 3.344213485173797,
+      "tokens_seen": 3059679232
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.350345048948804e-05,
+      "loss": 1.2561,
+      "theoretical_loss": 3.3442026080116283,
+      "tokens_seen": 3059810304
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.346332851869685e-05,
+      "loss": 1.2667,
+      "theoretical_loss": 3.3441917314458482,
+      "tokens_seen": 3059941376
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.342320654790563e-05,
+      "loss": 1.291,
+      "theoretical_loss": 3.344180855476398,
+      "tokens_seen": 3060072448
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.338308457711443e-05,
+      "loss": 1.3625,
+      "theoretical_loss": 3.34416998010322,
+      "tokens_seen": 3060203520
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.334296260632323e-05,
+      "loss": 1.2297,
+      "theoretical_loss": 3.344159105326255,
+      "tokens_seen": 3060334592
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.330284063553202e-05,
+      "loss": 1.303,
+      "theoretical_loss": 3.3441482311454456,
+      "tokens_seen": 3060465664
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 7.326271866474081e-05,
+      "loss": 1.3018,
+      "theoretical_loss": 3.3441373575607334,
+      "tokens_seen": 3060596736
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.32225966939496e-05,
+      "loss": 1.2407,
+      "theoretical_loss": 3.34412648457206,
+      "tokens_seen": 3060727808
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.318247472315841e-05,
+      "loss": 1.3458,
+      "theoretical_loss": 3.344115612179367,
+      "tokens_seen": 3060858880
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.314235275236719e-05,
+      "loss": 1.2348,
+      "theoretical_loss": 3.344104740382597,
+      "tokens_seen": 3060989952
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.310223078157598e-05,
+      "loss": 1.2591,
+      "theoretical_loss": 3.3440938691816915,
+      "tokens_seen": 3061121024
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.306210881078479e-05,
+      "loss": 1.281,
+      "theoretical_loss": 3.344082998576592,
+      "tokens_seen": 3061252096
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.302198683999358e-05,
+      "loss": 1.1591,
+      "theoretical_loss": 3.3440721285672406,
+      "tokens_seen": 3061383168
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.298186486920238e-05,
+      "loss": 1.3315,
+      "theoretical_loss": 3.344061259153579,
+      "tokens_seen": 3061514240
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.294174289841117e-05,
+      "loss": 1.311,
+      "theoretical_loss": 3.344050390335549,
+      "tokens_seen": 3061645312
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.290162092761998e-05,
+      "loss": 1.2923,
+      "theoretical_loss": 3.3440395221130927,
+      "tokens_seen": 3061776384
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.286149895682877e-05,
+      "loss": 1.2868,
+      "theoretical_loss": 3.344028654486152,
+      "tokens_seen": 3061907456
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.282137698603755e-05,
+      "loss": 1.3886,
+      "theoretical_loss": 3.3440177874546686,
+      "tokens_seen": 3062038528
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.278125501524635e-05,
+      "loss": 1.2974,
+      "theoretical_loss": 3.3440069210185843,
+      "tokens_seen": 3062169600
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.274113304445515e-05,
+      "loss": 1.2575,
+      "theoretical_loss": 3.343996055177841,
+      "tokens_seen": 3062300672
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.270101107366394e-05,
+      "loss": 1.3601,
+      "theoretical_loss": 3.343985189932381,
+      "tokens_seen": 3062431744
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.266088910287273e-05,
+      "loss": 1.3313,
+      "theoretical_loss": 3.343974325282146,
+      "tokens_seen": 3062562816
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.262076713208152e-05,
+      "loss": 1.4277,
+      "theoretical_loss": 3.343963461227078,
+      "tokens_seen": 3062693888
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.258064516129033e-05,
+      "loss": 1.2733,
+      "theoretical_loss": 3.3439525977671183,
+      "tokens_seen": 3062824960
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.254052319049911e-05,
+      "loss": 1.3062,
+      "theoretical_loss": 3.34394173490221,
+      "tokens_seen": 3062956032
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.250040121970792e-05,
+      "loss": 1.2627,
+      "theoretical_loss": 3.343930872632294,
+      "tokens_seen": 3063087104
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.246027924891671e-05,
+      "loss": 1.3065,
+      "theoretical_loss": 3.343920010957312,
+      "tokens_seen": 3063218176
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.242015727812552e-05,
+      "loss": 1.3296,
+      "theoretical_loss": 3.3439091498772076,
+      "tokens_seen": 3063349248
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.23800353073343e-05,
+      "loss": 1.3581,
+      "theoretical_loss": 3.3438982893919214,
+      "tokens_seen": 3063480320
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.233991333654309e-05,
+      "loss": 1.3018,
+      "theoretical_loss": 3.3438874295013954,
+      "tokens_seen": 3063611392
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.22997913657519e-05,
+      "loss": 1.3345,
+      "theoretical_loss": 3.3438765702055724,
+      "tokens_seen": 3063742464
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.225966939496067e-05,
+      "loss": 1.2913,
+      "theoretical_loss": 3.3438657115043933,
+      "tokens_seen": 3063873536
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.221954742416948e-05,
+      "loss": 1.2897,
+      "theoretical_loss": 3.343854853397801,
+      "tokens_seen": 3064004608
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.217942545337827e-05,
+      "loss": 1.3742,
+      "theoretical_loss": 3.3438439958857376,
+      "tokens_seen": 3064135680
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.213930348258706e-05,
+      "loss": 1.3205,
+      "theoretical_loss": 3.3438331389681446,
+      "tokens_seen": 3064266752
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.209918151179586e-05,
+      "loss": 1.2441,
+      "theoretical_loss": 3.3438222826449637,
+      "tokens_seen": 3064397824
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.205905954100465e-05,
+      "loss": 1.2285,
+      "theoretical_loss": 3.3438114269161376,
+      "tokens_seen": 3064528896
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.201893757021346e-05,
+      "loss": 1.3246,
+      "theoretical_loss": 3.3438005717816086,
+      "tokens_seen": 3064659968
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.197881559942225e-05,
+      "loss": 1.2576,
+      "theoretical_loss": 3.343789717241318,
+      "tokens_seen": 3064791040
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.193869362863104e-05,
+      "loss": 1.2565,
+      "theoretical_loss": 3.3437788632952077,
+      "tokens_seen": 3064922112
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.189857165783983e-05,
+      "loss": 1.3231,
+      "theoretical_loss": 3.3437680099432208,
+      "tokens_seen": 3065053184
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.185844968704863e-05,
+      "loss": 1.3227,
+      "theoretical_loss": 3.3437571571852986,
+      "tokens_seen": 3065184256
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.181832771625742e-05,
+      "loss": 1.2534,
+      "theoretical_loss": 3.343746305021383,
+      "tokens_seen": 3065315328
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.177820574546621e-05,
+      "loss": 1.2925,
+      "theoretical_loss": 3.343735453451417,
+      "tokens_seen": 3065446400
+    },
+    {
+      "epoch": 0.86,
+      "objective/train/advantage_avg": 0.49466797709465027,
+      "objective/train/docs_used": 1113675,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2124992609024048,
+      "objective/train/original_loss": 1.2124991416931152,
+      "objective/train/theoretical_loss": 3.343724602475342,
+      "objective/train/tokens_used": 1436036576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24493803083896637,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050711989402771,
+      "objective/train/weighted_lm_loss": 1.2739607095718384,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.343724602475342,
+      "tokens_seen": 3065577472
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.173808377467502e-05,
+      "loss": 1.3092,
+      "theoretical_loss": 3.343724602475342,
+      "tokens_seen": 3065577472
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.169796180388381e-05,
+      "loss": 1.2887,
+      "theoretical_loss": 3.3437137520931004,
+      "tokens_seen": 3065708544
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.165783983309259e-05,
+      "loss": 1.3165,
+      "theoretical_loss": 3.343702902304634,
+      "tokens_seen": 3065839616
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.16177178623014e-05,
+      "loss": 1.3093,
+      "theoretical_loss": 3.3436920531098853,
+      "tokens_seen": 3065970688
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.157759589151019e-05,
+      "loss": 1.3655,
+      "theoretical_loss": 3.343681204508796,
+      "tokens_seen": 3066101760
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.1537473920719e-05,
+      "loss": 1.2449,
+      "theoretical_loss": 3.3436703565013084,
+      "tokens_seen": 3066232832
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.149735194992778e-05,
+      "loss": 1.3005,
+      "theoretical_loss": 3.343659509087365,
+      "tokens_seen": 3066363904
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.145722997913658e-05,
+      "loss": 1.2752,
+      "theoretical_loss": 3.3436486622669075,
+      "tokens_seen": 3066494976
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.141710800834537e-05,
+      "loss": 1.2236,
+      "theoretical_loss": 3.343637816039878,
+      "tokens_seen": 3066626048
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.137698603755417e-05,
+      "loss": 1.3068,
+      "theoretical_loss": 3.3436269704062194,
+      "tokens_seen": 3066757120
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.133686406676296e-05,
+      "loss": 1.2931,
+      "theoretical_loss": 3.343616125365873,
+      "tokens_seen": 3066888192
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.129674209597175e-05,
+      "loss": 1.2854,
+      "theoretical_loss": 3.3436052809187817,
+      "tokens_seen": 3067019264
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.125662012518056e-05,
+      "loss": 1.328,
+      "theoretical_loss": 3.343594437064887,
+      "tokens_seen": 3067150336
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.121649815438934e-05,
+      "loss": 1.3025,
+      "theoretical_loss": 3.343583593804132,
+      "tokens_seen": 3067281408
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.117637618359813e-05,
+      "loss": 1.3112,
+      "theoretical_loss": 3.343572751136458,
+      "tokens_seen": 3067412480
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.113625421280694e-05,
+      "loss": 1.3035,
+      "theoretical_loss": 3.3435619090618074,
+      "tokens_seen": 3067543552
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.109613224201573e-05,
+      "loss": 1.2865,
+      "theoretical_loss": 3.3435510675801234,
+      "tokens_seen": 3067674624
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.105601027122452e-05,
+      "loss": 1.2067,
+      "theoretical_loss": 3.3435402266913465,
+      "tokens_seen": 3067805696
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.101588830043332e-05,
+      "loss": 1.2402,
+      "theoretical_loss": 3.3435293863954207,
+      "tokens_seen": 3067936768
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.097576632964212e-05,
+      "loss": 1.3124,
+      "theoretical_loss": 3.3435185466922874,
+      "tokens_seen": 3068067840
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.093564435885091e-05,
+      "loss": 1.266,
+      "theoretical_loss": 3.343507707581889,
+      "tokens_seen": 3068198912
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.08955223880597e-05,
+      "loss": 1.2652,
+      "theoretical_loss": 3.343496869064167,
+      "tokens_seen": 3068329984
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.08554004172685e-05,
+      "loss": 1.3797,
+      "theoretical_loss": 3.343486031139065,
+      "tokens_seen": 3068461056
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.081527844647729e-05,
+      "loss": 1.313,
+      "theoretical_loss": 3.3434751938065244,
+      "tokens_seen": 3068592128
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.077515647568609e-05,
+      "loss": 1.3134,
+      "theoretical_loss": 3.343464357066488,
+      "tokens_seen": 3068723200
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.073503450489488e-05,
+      "loss": 1.2528,
+      "theoretical_loss": 3.3434535209188976,
+      "tokens_seen": 3068854272
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.069491253410367e-05,
+      "loss": 1.2327,
+      "theoretical_loss": 3.343442685363696,
+      "tokens_seen": 3068985344
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.065479056331248e-05,
+      "loss": 1.3094,
+      "theoretical_loss": 3.343431850400825,
+      "tokens_seen": 3069116416
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.061466859252126e-05,
+      "loss": 1.326,
+      "theoretical_loss": 3.3434210160302276,
+      "tokens_seen": 3069247488
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.057454662173006e-05,
+      "loss": 1.2891,
+      "theoretical_loss": 3.3434101822518456,
+      "tokens_seen": 3069378560
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.053442465093886e-05,
+      "loss": 1.307,
+      "theoretical_loss": 3.3433993490656215,
+      "tokens_seen": 3069509632
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.049430268014766e-05,
+      "loss": 1.2729,
+      "theoretical_loss": 3.3433885164714976,
+      "tokens_seen": 3069640704
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.045418070935644e-05,
+      "loss": 1.351,
+      "theoretical_loss": 3.343377684469416,
+      "tokens_seen": 3069771776
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.041405873856523e-05,
+      "loss": 1.3565,
+      "theoretical_loss": 3.3433668530593197,
+      "tokens_seen": 3069902848
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.037393676777404e-05,
+      "loss": 1.3565,
+      "theoretical_loss": 3.3433560222411507,
+      "tokens_seen": 3070033920
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.033381479698282e-05,
+      "loss": 1.3291,
+      "theoretical_loss": 3.3433451920148514,
+      "tokens_seen": 3070164992
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.029369282619163e-05,
+      "loss": 1.3401,
+      "theoretical_loss": 3.343334362380364,
+      "tokens_seen": 3070296064
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.025357085540042e-05,
+      "loss": 1.2881,
+      "theoretical_loss": 3.3433235333376317,
+      "tokens_seen": 3070427136
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.021344888460923e-05,
+      "loss": 1.3162,
+      "theoretical_loss": 3.3433127048865954,
+      "tokens_seen": 3070558208
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.0173326913818e-05,
+      "loss": 1.3053,
+      "theoretical_loss": 3.343301877027199,
+      "tokens_seen": 3070689280
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.01332049430268e-05,
+      "loss": 1.3173,
+      "theoretical_loss": 3.3432910497593844,
+      "tokens_seen": 3070820352
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.00930829722356e-05,
+      "loss": 1.349,
+      "theoretical_loss": 3.343280223083094,
+      "tokens_seen": 3070951424
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.00529610014444e-05,
+      "loss": 1.2926,
+      "theoretical_loss": 3.34326939699827,
+      "tokens_seen": 3071082496
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 7.001283903065319e-05,
+      "loss": 1.2902,
+      "theoretical_loss": 3.343258571504855,
+      "tokens_seen": 3071213568
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.997271705986198e-05,
+      "loss": 1.3082,
+      "theoretical_loss": 3.343247746602792,
+      "tokens_seen": 3071344640
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.993259508907077e-05,
+      "loss": 1.2746,
+      "theoretical_loss": 3.3432369222920224,
+      "tokens_seen": 3071475712
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.989247311827957e-05,
+      "loss": 1.3099,
+      "theoretical_loss": 3.3432260985724898,
+      "tokens_seen": 3071606784
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.985235114748836e-05,
+      "loss": 1.3885,
+      "theoretical_loss": 3.3432152754441358,
+      "tokens_seen": 3071737856
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.981222917669717e-05,
+      "loss": 1.3112,
+      "theoretical_loss": 3.343204452906903,
+      "tokens_seen": 3071868928
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.977210720590596e-05,
+      "loss": 1.3624,
+      "theoretical_loss": 3.3431936309607346,
+      "tokens_seen": 3072000000
+    },
+    {
+      "epoch": 0.86,
+      "objective/train/advantage_avg": 0.49419450759887695,
+      "objective/train/docs_used": 1116127,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.1892919540405273,
+      "objective/train/original_loss": 1.189292073249817,
+      "objective/train/theoretical_loss": 3.3431828096055725,
+      "objective/train/tokens_used": 1442590176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24454230070114136,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050662636756897,
+      "objective/train/weighted_lm_loss": 1.2494804859161377,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3431828096055725,
+      "tokens_seen": 3072131072
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.973198523511474e-05,
+      "loss": 1.2967,
+      "theoretical_loss": 3.3431828096055725,
+      "tokens_seen": 3072131072
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.969186326432354e-05,
+      "loss": 1.2853,
+      "theoretical_loss": 3.343171988841359,
+      "tokens_seen": 3072262144
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.965174129353234e-05,
+      "loss": 1.3286,
+      "theoretical_loss": 3.343161168668037,
+      "tokens_seen": 3072393216
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.961161932274114e-05,
+      "loss": 1.2588,
+      "theoretical_loss": 3.3431503490855494,
+      "tokens_seen": 3072524288
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.957149735194992e-05,
+      "loss": 1.3072,
+      "theoretical_loss": 3.343139530093838,
+      "tokens_seen": 3072655360
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.953137538115873e-05,
+      "loss": 1.3188,
+      "theoretical_loss": 3.3431287116928456,
+      "tokens_seen": 3072786432
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.949125341036752e-05,
+      "loss": 1.3292,
+      "theoretical_loss": 3.3431178938825146,
+      "tokens_seen": 3072917504
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.94511314395763e-05,
+      "loss": 1.2738,
+      "theoretical_loss": 3.3431070766627884,
+      "tokens_seen": 3073048576
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.941100946878511e-05,
+      "loss": 1.2404,
+      "theoretical_loss": 3.3430962600336085,
+      "tokens_seen": 3073179648
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.93708874979939e-05,
+      "loss": 1.3315,
+      "theoretical_loss": 3.343085443994918,
+      "tokens_seen": 3073310720
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.93307655272027e-05,
+      "loss": 1.2365,
+      "theoretical_loss": 3.3430746285466593,
+      "tokens_seen": 3073441792
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.929064355641149e-05,
+      "loss": 1.3483,
+      "theoretical_loss": 3.343063813688775,
+      "tokens_seen": 3073572864
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.925052158562029e-05,
+      "loss": 1.3574,
+      "theoretical_loss": 3.343052999421208,
+      "tokens_seen": 3073703936
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.921039961482908e-05,
+      "loss": 1.2239,
+      "theoretical_loss": 3.3430421857439008,
+      "tokens_seen": 3073835008
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.917027764403788e-05,
+      "loss": 1.3062,
+      "theoretical_loss": 3.3430313726567955,
+      "tokens_seen": 3073966080
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.913015567324667e-05,
+      "loss": 1.3134,
+      "theoretical_loss": 3.343020560159835,
+      "tokens_seen": 3074097152
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.909003370245546e-05,
+      "loss": 1.3088,
+      "theoretical_loss": 3.3430097482529626,
+      "tokens_seen": 3074228224
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.904991173166427e-05,
+      "loss": 1.2499,
+      "theoretical_loss": 3.3429989369361204,
+      "tokens_seen": 3074359296
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.900978976087306e-05,
+      "loss": 1.3431,
+      "theoretical_loss": 3.3429881262092507,
+      "tokens_seen": 3074490368
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.896966779008184e-05,
+      "loss": 1.2911,
+      "theoretical_loss": 3.3429773160722966,
+      "tokens_seen": 3074621440
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.892954581929065e-05,
+      "loss": 1.302,
+      "theoretical_loss": 3.3429665065252006,
+      "tokens_seen": 3074752512
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.888942384849944e-05,
+      "loss": 1.2103,
+      "theoretical_loss": 3.3429556975679056,
+      "tokens_seen": 3074883584
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.884930187770823e-05,
+      "loss": 1.2905,
+      "theoretical_loss": 3.342944889200354,
+      "tokens_seen": 3075014656
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.880917990691703e-05,
+      "loss": 1.3484,
+      "theoretical_loss": 3.3429340814224884,
+      "tokens_seen": 3075145728
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.876905793612583e-05,
+      "loss": 1.319,
+      "theoretical_loss": 3.342923274234252,
+      "tokens_seen": 3075276800
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.872893596533462e-05,
+      "loss": 1.2987,
+      "theoretical_loss": 3.3429124676355872,
+      "tokens_seen": 3075407872
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.86888139945434e-05,
+      "loss": 1.3054,
+      "theoretical_loss": 3.3429016616264366,
+      "tokens_seen": 3075538944
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.864869202375221e-05,
+      "loss": 1.3423,
+      "theoretical_loss": 3.342890856206743,
+      "tokens_seen": 3075670016
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.8608570052961e-05,
+      "loss": 1.2717,
+      "theoretical_loss": 3.342880051376449,
+      "tokens_seen": 3075801088
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.856844808216981e-05,
+      "loss": 1.2143,
+      "theoretical_loss": 3.3428692471354977,
+      "tokens_seen": 3075932160
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.852832611137859e-05,
+      "loss": 1.2858,
+      "theoretical_loss": 3.3428584434838315,
+      "tokens_seen": 3076063232
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.848820414058738e-05,
+      "loss": 1.2964,
+      "theoretical_loss": 3.3428476404213936,
+      "tokens_seen": 3076194304
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.844808216979619e-05,
+      "loss": 1.3483,
+      "theoretical_loss": 3.342836837948126,
+      "tokens_seen": 3076325376
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.840796019900497e-05,
+      "loss": 1.276,
+      "theoretical_loss": 3.3428260360639723,
+      "tokens_seen": 3076456448
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.836783822821377e-05,
+      "loss": 1.3032,
+      "theoretical_loss": 3.342815234768875,
+      "tokens_seen": 3076587520
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.832771625742257e-05,
+      "loss": 1.2142,
+      "theoretical_loss": 3.3428044340627765,
+      "tokens_seen": 3076718592
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.828759428663137e-05,
+      "loss": 1.2456,
+      "theoretical_loss": 3.34279363394562,
+      "tokens_seen": 3076849664
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.824747231584015e-05,
+      "loss": 1.2948,
+      "theoretical_loss": 3.342782834417348,
+      "tokens_seen": 3076980736
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.820735034504894e-05,
+      "loss": 1.2742,
+      "theoretical_loss": 3.3427720354779034,
+      "tokens_seen": 3077111808
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.816722837425775e-05,
+      "loss": 1.3261,
+      "theoretical_loss": 3.3427612371272293,
+      "tokens_seen": 3077242880
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.812710640346654e-05,
+      "loss": 1.2982,
+      "theoretical_loss": 3.3427504393652683,
+      "tokens_seen": 3077373952
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.808698443267534e-05,
+      "loss": 1.3202,
+      "theoretical_loss": 3.342739642191963,
+      "tokens_seen": 3077505024
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.804686246188413e-05,
+      "loss": 1.2985,
+      "theoretical_loss": 3.342728845607257,
+      "tokens_seen": 3077636096
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.800674049109292e-05,
+      "loss": 1.3323,
+      "theoretical_loss": 3.342718049611092,
+      "tokens_seen": 3077767168
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.796661852030171e-05,
+      "loss": 1.2972,
+      "theoretical_loss": 3.342707254203412,
+      "tokens_seen": 3077898240
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.792649654951051e-05,
+      "loss": 1.3325,
+      "theoretical_loss": 3.3426964593841593,
+      "tokens_seen": 3078029312
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.788637457871931e-05,
+      "loss": 1.2712,
+      "theoretical_loss": 3.3426856651532764,
+      "tokens_seen": 3078160384
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.78462526079281e-05,
+      "loss": 1.3075,
+      "theoretical_loss": 3.342674871510707,
+      "tokens_seen": 3078291456
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.78061306371369e-05,
+      "loss": 1.2524,
+      "theoretical_loss": 3.3426640784563935,
+      "tokens_seen": 3078422528
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.776600866634569e-05,
+      "loss": 1.3327,
+      "theoretical_loss": 3.342653285990279,
+      "tokens_seen": 3078553600
+    },
+    {
+      "epoch": 0.87,
+      "objective/train/advantage_avg": 0.4918553829193115,
+      "objective/train/docs_used": 1118389,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2797116041183472,
+      "objective/train/original_loss": 1.2797117233276367,
+      "objective/train/theoretical_loss": 3.342642494112306,
+      "objective/train/tokens_used": 1449143776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2423219084739685,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050417423248291,
+      "objective/train/weighted_lm_loss": 1.344075083732605,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.342642494112306,
+      "tokens_seen": 3078684672
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.772588669555448e-05,
+      "loss": 1.3345,
+      "theoretical_loss": 3.342642494112306,
+      "tokens_seen": 3078684672
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.768576472476329e-05,
+      "loss": 1.2731,
+      "theoretical_loss": 3.3426317028224175,
+      "tokens_seen": 3078815744
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.764564275397207e-05,
+      "loss": 1.25,
+      "theoretical_loss": 3.342620912120557,
+      "tokens_seen": 3078946816
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.760552078318088e-05,
+      "loss": 1.2101,
+      "theoretical_loss": 3.342610122006667,
+      "tokens_seen": 3079077888
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.756539881238967e-05,
+      "loss": 1.3152,
+      "theoretical_loss": 3.3425993324806904,
+      "tokens_seen": 3079208960
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.752527684159845e-05,
+      "loss": 1.3316,
+      "theoretical_loss": 3.34258854354257,
+      "tokens_seen": 3079340032
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.748515487080725e-05,
+      "loss": 1.2684,
+      "theoretical_loss": 3.3425777551922495,
+      "tokens_seen": 3079471104
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.744503290001605e-05,
+      "loss": 1.3205,
+      "theoretical_loss": 3.342566967429671,
+      "tokens_seen": 3079602176
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.740491092922485e-05,
+      "loss": 1.3321,
+      "theoretical_loss": 3.3425561802547783,
+      "tokens_seen": 3079733248
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.736478895843363e-05,
+      "loss": 1.3148,
+      "theoretical_loss": 3.342545393667513,
+      "tokens_seen": 3079864320
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.732466698764244e-05,
+      "loss": 1.3137,
+      "theoretical_loss": 3.3425346076678197,
+      "tokens_seen": 3079995392
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.728454501685123e-05,
+      "loss": 1.2983,
+      "theoretical_loss": 3.3425238222556404,
+      "tokens_seen": 3080126464
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.724442304606002e-05,
+      "loss": 1.3478,
+      "theoretical_loss": 3.342513037430918,
+      "tokens_seen": 3080257536
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.720430107526882e-05,
+      "loss": 1.2445,
+      "theoretical_loss": 3.3425022531935964,
+      "tokens_seen": 3080388608
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.716417910447761e-05,
+      "loss": 1.3261,
+      "theoretical_loss": 3.342491469543618,
+      "tokens_seen": 3080519680
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.712405713368642e-05,
+      "loss": 1.3326,
+      "theoretical_loss": 3.3424806864809256,
+      "tokens_seen": 3080650752
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.70839351628952e-05,
+      "loss": 1.2581,
+      "theoretical_loss": 3.342469904005463,
+      "tokens_seen": 3080781824
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.704381319210399e-05,
+      "loss": 1.3008,
+      "theoretical_loss": 3.342459122117172,
+      "tokens_seen": 3080912896
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.70036912213128e-05,
+      "loss": 1.3095,
+      "theoretical_loss": 3.342448340815997,
+      "tokens_seen": 3081043968
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.696356925052159e-05,
+      "loss": 1.3572,
+      "theoretical_loss": 3.3424375601018803,
+      "tokens_seen": 3081175040
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.692344727973038e-05,
+      "loss": 1.3022,
+      "theoretical_loss": 3.342426779974765,
+      "tokens_seen": 3081306112
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.688332530893917e-05,
+      "loss": 1.3476,
+      "theoretical_loss": 3.3424160004345946,
+      "tokens_seen": 3081437184
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.684320333814798e-05,
+      "loss": 1.3066,
+      "theoretical_loss": 3.342405221481312,
+      "tokens_seen": 3081568256
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.680308136735677e-05,
+      "loss": 1.3808,
+      "theoretical_loss": 3.3423944431148596,
+      "tokens_seen": 3081699328
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.676295939656555e-05,
+      "loss": 1.2601,
+      "theoretical_loss": 3.342383665335181,
+      "tokens_seen": 3081830400
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.672283742577436e-05,
+      "loss": 1.2637,
+      "theoretical_loss": 3.34237288814222,
+      "tokens_seen": 3081961472
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.668271545498315e-05,
+      "loss": 1.2024,
+      "theoretical_loss": 3.342362111535919,
+      "tokens_seen": 3082092544
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.664259348419196e-05,
+      "loss": 1.2794,
+      "theoretical_loss": 3.3423513355162204,
+      "tokens_seen": 3082223616
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.660247151340074e-05,
+      "loss": 1.2538,
+      "theoretical_loss": 3.342340560083069,
+      "tokens_seen": 3082354688
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.656234954260953e-05,
+      "loss": 1.2988,
+      "theoretical_loss": 3.3423297852364064,
+      "tokens_seen": 3082485760
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.652222757181833e-05,
+      "loss": 1.3311,
+      "theoretical_loss": 3.342319010976177,
+      "tokens_seen": 3082616832
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.648210560102711e-05,
+      "loss": 1.2834,
+      "theoretical_loss": 3.3423082373023227,
+      "tokens_seen": 3082747904
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.644198363023592e-05,
+      "loss": 1.2938,
+      "theoretical_loss": 3.342297464214788,
+      "tokens_seen": 3082878976
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.640186165944471e-05,
+      "loss": 1.2471,
+      "theoretical_loss": 3.3422866917135146,
+      "tokens_seen": 3083010048
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.636173968865352e-05,
+      "loss": 1.3135,
+      "theoretical_loss": 3.342275919798447,
+      "tokens_seen": 3083141120
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.63216177178623e-05,
+      "loss": 1.2877,
+      "theoretical_loss": 3.3422651484695276,
+      "tokens_seen": 3083272192
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.628149574707109e-05,
+      "loss": 1.2952,
+      "theoretical_loss": 3.3422543777266998,
+      "tokens_seen": 3083403264
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.62413737762799e-05,
+      "loss": 1.2918,
+      "theoretical_loss": 3.342243607569907,
+      "tokens_seen": 3083534336
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.620125180548869e-05,
+      "loss": 1.3234,
+      "theoretical_loss": 3.342232837999092,
+      "tokens_seen": 3083665408
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.616112983469748e-05,
+      "loss": 1.2734,
+      "theoretical_loss": 3.3422220690141984,
+      "tokens_seen": 3083796480
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.612100786390628e-05,
+      "loss": 1.2379,
+      "theoretical_loss": 3.342211300615169,
+      "tokens_seen": 3083927552
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.608088589311508e-05,
+      "loss": 1.3241,
+      "theoretical_loss": 3.3422005328019475,
+      "tokens_seen": 3084058624
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.604076392232386e-05,
+      "loss": 1.2744,
+      "theoretical_loss": 3.3421897655744766,
+      "tokens_seen": 3084189696
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.600064195153265e-05,
+      "loss": 1.3107,
+      "theoretical_loss": 3.3421789989327,
+      "tokens_seen": 3084320768
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.596051998074146e-05,
+      "loss": 1.3411,
+      "theoretical_loss": 3.342168232876561,
+      "tokens_seen": 3084451840
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.592039800995025e-05,
+      "loss": 1.2309,
+      "theoretical_loss": 3.3421574674060026,
+      "tokens_seen": 3084582912
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.588027603915905e-05,
+      "loss": 1.307,
+      "theoretical_loss": 3.342146702520968,
+      "tokens_seen": 3084713984
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.584015406836784e-05,
+      "loss": 1.3482,
+      "theoretical_loss": 3.342135938221401,
+      "tokens_seen": 3084845056
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.580003209757663e-05,
+      "loss": 1.3028,
+      "theoretical_loss": 3.342125174507244,
+      "tokens_seen": 3084976128
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.575991012678544e-05,
+      "loss": 1.3174,
+      "theoretical_loss": 3.342114411378441,
+      "tokens_seen": 3085107200
+    },
+    {
+      "epoch": 0.87,
+      "objective/train/advantage_avg": 0.49401023983955383,
+      "objective/train/docs_used": 1120785,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.288630723953247,
+      "objective/train/original_loss": 1.288630723953247,
+      "objective/train/theoretical_loss": 3.342103648834935,
+      "objective/train/tokens_used": 1455697376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24442297220230103,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506436824798584,
+      "objective/train/weighted_lm_loss": 1.353763222694397,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.342103648834935,
+      "tokens_seen": 3085238272
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.571978815599422e-05,
+      "loss": 1.2987,
+      "theoretical_loss": 3.342103648834935,
+      "tokens_seen": 3085238272
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.567966618520302e-05,
+      "loss": 1.3336,
+      "theoretical_loss": 3.3420928868766695,
+      "tokens_seen": 3085369344
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.563954421441182e-05,
+      "loss": 1.2712,
+      "theoretical_loss": 3.3420821255035875,
+      "tokens_seen": 3085500416
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.55994222436206e-05,
+      "loss": 1.3656,
+      "theoretical_loss": 3.342071364715633,
+      "tokens_seen": 3085631488
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.55593002728294e-05,
+      "loss": 1.2867,
+      "theoretical_loss": 3.3420606045127483,
+      "tokens_seen": 3085762560
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.55191783020382e-05,
+      "loss": 1.344,
+      "theoretical_loss": 3.342049844894878,
+      "tokens_seen": 3085893632
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.5479056331247e-05,
+      "loss": 1.2939,
+      "theoretical_loss": 3.3420390858619644,
+      "tokens_seen": 3086024704
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.543893436045578e-05,
+      "loss": 1.2599,
+      "theoretical_loss": 3.342028327413951,
+      "tokens_seen": 3086155776
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.539881238966459e-05,
+      "loss": 1.265,
+      "theoretical_loss": 3.3420175695507814,
+      "tokens_seen": 3086286848
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.535869041887338e-05,
+      "loss": 1.32,
+      "theoretical_loss": 3.342006812272399,
+      "tokens_seen": 3086417920
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.531856844808217e-05,
+      "loss": 1.2389,
+      "theoretical_loss": 3.3419960555787473,
+      "tokens_seen": 3086548992
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.527844647729096e-05,
+      "loss": 1.253,
+      "theoretical_loss": 3.3419852994697696,
+      "tokens_seen": 3086680064
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.523832450649976e-05,
+      "loss": 1.3371,
+      "theoretical_loss": 3.341974543945409,
+      "tokens_seen": 3086811136
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.519820253570856e-05,
+      "loss": 1.2902,
+      "theoretical_loss": 3.341963789005609,
+      "tokens_seen": 3086942208
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.515808056491734e-05,
+      "loss": 1.2173,
+      "theoretical_loss": 3.341953034650314,
+      "tokens_seen": 3087073280
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.511795859412615e-05,
+      "loss": 1.1987,
+      "theoretical_loss": 3.3419422808794654,
+      "tokens_seen": 3087204352
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.507783662333494e-05,
+      "loss": 1.3348,
+      "theoretical_loss": 3.341931527693008,
+      "tokens_seen": 3087335424
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.503771465254373e-05,
+      "loss": 1.3153,
+      "theoretical_loss": 3.3419207750908857,
+      "tokens_seen": 3087466496
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.499759268175253e-05,
+      "loss": 1.2562,
+      "theoretical_loss": 3.3419100230730407,
+      "tokens_seen": 3087597568
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.495747071096132e-05,
+      "loss": 1.2903,
+      "theoretical_loss": 3.341899271639417,
+      "tokens_seen": 3087728640
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.491734874017013e-05,
+      "loss": 1.3205,
+      "theoretical_loss": 3.341888520789958,
+      "tokens_seen": 3087859712
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.487722676937892e-05,
+      "loss": 1.2369,
+      "theoretical_loss": 3.3418777705246074,
+      "tokens_seen": 3087990784
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.48371047985877e-05,
+      "loss": 1.2677,
+      "theoretical_loss": 3.341867020843308,
+      "tokens_seen": 3088121856
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.47969828277965e-05,
+      "loss": 1.2676,
+      "theoretical_loss": 3.3418562717460043,
+      "tokens_seen": 3088252928
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.47568608570053e-05,
+      "loss": 1.3063,
+      "theoretical_loss": 3.341845523232639,
+      "tokens_seen": 3088384000
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.471673888621409e-05,
+      "loss": 1.248,
+      "theoretical_loss": 3.3418347753031563,
+      "tokens_seen": 3088515072
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.467661691542288e-05,
+      "loss": 1.246,
+      "theoretical_loss": 3.341824027957499,
+      "tokens_seen": 3088646144
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.463649494463169e-05,
+      "loss": 1.3438,
+      "theoretical_loss": 3.341813281195611,
+      "tokens_seen": 3088777216
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.459637297384048e-05,
+      "loss": 1.3187,
+      "theoretical_loss": 3.3418025350174356,
+      "tokens_seen": 3088908288
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.455625100304926e-05,
+      "loss": 1.3214,
+      "theoretical_loss": 3.341791789422916,
+      "tokens_seen": 3089039360
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.451612903225807e-05,
+      "loss": 1.2417,
+      "theoretical_loss": 3.341781044411997,
+      "tokens_seen": 3089170432
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.447600706146686e-05,
+      "loss": 1.2765,
+      "theoretical_loss": 3.3417702999846206,
+      "tokens_seen": 3089301504
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.443588509067567e-05,
+      "loss": 1.2677,
+      "theoretical_loss": 3.3417595561407314,
+      "tokens_seen": 3089432576
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.439576311988445e-05,
+      "loss": 1.3221,
+      "theoretical_loss": 3.3417488128802724,
+      "tokens_seen": 3089563648
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.435564114909324e-05,
+      "loss": 1.3106,
+      "theoretical_loss": 3.3417380702031876,
+      "tokens_seen": 3089694720
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.431551917830204e-05,
+      "loss": 1.3225,
+      "theoretical_loss": 3.34172732810942,
+      "tokens_seen": 3089825792
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.427539720751084e-05,
+      "loss": 1.3813,
+      "theoretical_loss": 3.3417165865989142,
+      "tokens_seen": 3089956864
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.423527523671963e-05,
+      "loss": 1.2428,
+      "theoretical_loss": 3.341705845671613,
+      "tokens_seen": 3090087936
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.419515326592842e-05,
+      "loss": 1.3269,
+      "theoretical_loss": 3.34169510532746,
+      "tokens_seen": 3090219008
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.415503129513723e-05,
+      "loss": 1.2255,
+      "theoretical_loss": 3.341684365566399,
+      "tokens_seen": 3090350080
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.411490932434601e-05,
+      "loss": 1.2456,
+      "theoretical_loss": 3.3416736263883733,
+      "tokens_seen": 3090481152
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.40747873535548e-05,
+      "loss": 1.2286,
+      "theoretical_loss": 3.3416628877933268,
+      "tokens_seen": 3090612224
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.403466538276361e-05,
+      "loss": 1.2892,
+      "theoretical_loss": 3.3416521497812033,
+      "tokens_seen": 3090743296
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.39945434119724e-05,
+      "loss": 1.2865,
+      "theoretical_loss": 3.3416414123519464,
+      "tokens_seen": 3090874368
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.395442144118119e-05,
+      "loss": 1.1796,
+      "theoretical_loss": 3.3416306755054994,
+      "tokens_seen": 3091005440
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.391429947038999e-05,
+      "loss": 1.2541,
+      "theoretical_loss": 3.3416199392418062,
+      "tokens_seen": 3091136512
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.387417749959878e-05,
+      "loss": 1.3285,
+      "theoretical_loss": 3.3416092035608105,
+      "tokens_seen": 3091267584
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.383405552880758e-05,
+      "loss": 1.2642,
+      "theoretical_loss": 3.341598468462456,
+      "tokens_seen": 3091398656
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.379393355801636e-05,
+      "loss": 1.229,
+      "theoretical_loss": 3.3415877339466857,
+      "tokens_seen": 3091529728
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.375381158722517e-05,
+      "loss": 1.3175,
+      "theoretical_loss": 3.3415770000134444,
+      "tokens_seen": 3091660800
+    },
+    {
+      "epoch": 0.87,
+      "objective/train/advantage_avg": 0.49364814162254333,
+      "objective/train/docs_used": 1123259,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.234599232673645,
+      "objective/train/original_loss": 1.234598994255066,
+      "objective/train/theoretical_loss": 3.341566266662675,
+      "objective/train/tokens_used": 1462250976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24397891759872437,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506051778793335,
+      "objective/train/weighted_lm_loss": 1.297003984451294,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.341566266662675,
+      "tokens_seen": 3091791872
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.371368961643396e-05,
+      "loss": 1.2979,
+      "theoretical_loss": 3.341566266662675,
+      "tokens_seen": 3091791872
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.367356764564276e-05,
+      "loss": 1.2899,
+      "theoretical_loss": 3.3415555338943213,
+      "tokens_seen": 3091922944
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.363344567485155e-05,
+      "loss": 1.392,
+      "theoretical_loss": 3.3415448017083276,
+      "tokens_seen": 3092054016
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.359332370406034e-05,
+      "loss": 1.3754,
+      "theoretical_loss": 3.341534070104637,
+      "tokens_seen": 3092185088
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.355320173326915e-05,
+      "loss": 1.2844,
+      "theoretical_loss": 3.341523339083193,
+      "tokens_seen": 3092316160
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.351307976247793e-05,
+      "loss": 1.3141,
+      "theoretical_loss": 3.3415126086439404,
+      "tokens_seen": 3092447232
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.347295779168673e-05,
+      "loss": 1.313,
+      "theoretical_loss": 3.341501878786822,
+      "tokens_seen": 3092578304
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.343283582089553e-05,
+      "loss": 1.3016,
+      "theoretical_loss": 3.3414911495117816,
+      "tokens_seen": 3092709376
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.339271385010432e-05,
+      "loss": 1.3463,
+      "theoretical_loss": 3.341480420818763,
+      "tokens_seen": 3092840448
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.335259187931311e-05,
+      "loss": 1.3172,
+      "theoretical_loss": 3.3414696927077108,
+      "tokens_seen": 3092971520
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.33124699085219e-05,
+      "loss": 1.2846,
+      "theoretical_loss": 3.3414589651785676,
+      "tokens_seen": 3093102592
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.327234793773071e-05,
+      "loss": 1.2541,
+      "theoretical_loss": 3.341448238231278,
+      "tokens_seen": 3093233664
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.323222596693949e-05,
+      "loss": 1.2912,
+      "theoretical_loss": 3.3414375118657853,
+      "tokens_seen": 3093364736
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.31921039961483e-05,
+      "loss": 1.3425,
+      "theoretical_loss": 3.3414267860820335,
+      "tokens_seen": 3093495808
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.315198202535709e-05,
+      "loss": 1.3746,
+      "theoretical_loss": 3.3414160608799666,
+      "tokens_seen": 3093626880
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.311186005456588e-05,
+      "loss": 1.3119,
+      "theoretical_loss": 3.341405336259528,
+      "tokens_seen": 3093757952
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.307173808377467e-05,
+      "loss": 1.3971,
+      "theoretical_loss": 3.341394612220662,
+      "tokens_seen": 3093889024
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.303161611298347e-05,
+      "loss": 1.3211,
+      "theoretical_loss": 3.3413838887633114,
+      "tokens_seen": 3094020096
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.299149414219227e-05,
+      "loss": 1.3971,
+      "theoretical_loss": 3.3413731658874215,
+      "tokens_seen": 3094151168
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.295137217140107e-05,
+      "loss": 1.3062,
+      "theoretical_loss": 3.341362443592935,
+      "tokens_seen": 3094282240
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.291125020060985e-05,
+      "loss": 1.2237,
+      "theoretical_loss": 3.3413517218797963,
+      "tokens_seen": 3094413312
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.287112822981865e-05,
+      "loss": 1.2128,
+      "theoretical_loss": 3.341341000747949,
+      "tokens_seen": 3094544384
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.283100625902744e-05,
+      "loss": 1.3292,
+      "theoretical_loss": 3.3413302801973375,
+      "tokens_seen": 3094675456
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.279088428823624e-05,
+      "loss": 1.2594,
+      "theoretical_loss": 3.341319560227905,
+      "tokens_seen": 3094806528
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.275076231744503e-05,
+      "loss": 1.2657,
+      "theoretical_loss": 3.3413088408395955,
+      "tokens_seen": 3094937600
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.271064034665384e-05,
+      "loss": 1.2986,
+      "theoretical_loss": 3.341298122032353,
+      "tokens_seen": 3095068672
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.267051837586263e-05,
+      "loss": 1.3342,
+      "theoretical_loss": 3.3412874038061218,
+      "tokens_seen": 3095199744
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.263039640507141e-05,
+      "loss": 1.2938,
+      "theoretical_loss": 3.341276686160845,
+      "tokens_seen": 3095330816
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.259027443428021e-05,
+      "loss": 1.4141,
+      "theoretical_loss": 3.341265969096467,
+      "tokens_seen": 3095461888
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.255015246348901e-05,
+      "loss": 1.247,
+      "theoretical_loss": 3.3412552526129318,
+      "tokens_seen": 3095592960
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.251003049269781e-05,
+      "loss": 1.2807,
+      "theoretical_loss": 3.3412445367101835,
+      "tokens_seen": 3095724032
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.24699085219066e-05,
+      "loss": 1.2459,
+      "theoretical_loss": 3.341233821388165,
+      "tokens_seen": 3095855104
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.242978655111539e-05,
+      "loss": 1.3361,
+      "theoretical_loss": 3.3412231066468214,
+      "tokens_seen": 3095986176
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.238966458032419e-05,
+      "loss": 1.3773,
+      "theoretical_loss": 3.341212392486096,
+      "tokens_seen": 3096117248
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.234954260953298e-05,
+      "loss": 1.3395,
+      "theoretical_loss": 3.341201678905933,
+      "tokens_seen": 3096248320
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.230942063874178e-05,
+      "loss": 1.2579,
+      "theoretical_loss": 3.341190965906277,
+      "tokens_seen": 3096379392
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.226929866795057e-05,
+      "loss": 1.2529,
+      "theoretical_loss": 3.3411802534870705,
+      "tokens_seen": 3096510464
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.222917669715936e-05,
+      "loss": 1.2928,
+      "theoretical_loss": 3.3411695416482585,
+      "tokens_seen": 3096641536
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.218905472636816e-05,
+      "loss": 1.2939,
+      "theoretical_loss": 3.341158830389785,
+      "tokens_seen": 3096772608
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.214893275557695e-05,
+      "loss": 1.2421,
+      "theoretical_loss": 3.341148119711594,
+      "tokens_seen": 3096903680
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.210881078478575e-05,
+      "loss": 1.2589,
+      "theoretical_loss": 3.341137409613629,
+      "tokens_seen": 3097034752
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.206868881399455e-05,
+      "loss": 1.3334,
+      "theoretical_loss": 3.341126700095834,
+      "tokens_seen": 3097165824
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.202856684320334e-05,
+      "loss": 1.2951,
+      "theoretical_loss": 3.3411159911581536,
+      "tokens_seen": 3097296896
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.198844487241213e-05,
+      "loss": 1.3247,
+      "theoretical_loss": 3.3411052828005317,
+      "tokens_seen": 3097427968
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.194832290162093e-05,
+      "loss": 1.2177,
+      "theoretical_loss": 3.341094575022912,
+      "tokens_seen": 3097559040
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.190820093082972e-05,
+      "loss": 1.3342,
+      "theoretical_loss": 3.3410838678252386,
+      "tokens_seen": 3097690112
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.186807896003851e-05,
+      "loss": 1.3304,
+      "theoretical_loss": 3.341073161207456,
+      "tokens_seen": 3097821184
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.182795698924732e-05,
+      "loss": 1.3126,
+      "theoretical_loss": 3.341062455169508,
+      "tokens_seen": 3097952256
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.178783501845611e-05,
+      "loss": 1.3273,
+      "theoretical_loss": 3.3410517497113386,
+      "tokens_seen": 3098083328
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.17477130476649e-05,
+      "loss": 1.3729,
+      "theoretical_loss": 3.3410410448328913,
+      "tokens_seen": 3098214400
+    },
+    {
+      "epoch": 0.88,
+      "objective/train/advantage_avg": 0.49290502071380615,
+      "objective/train/docs_used": 1125757,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2949086427688599,
+      "objective/train/original_loss": 1.2949085235595703,
+      "objective/train/theoretical_loss": 3.3410303405341115,
+      "objective/train/tokens_used": 1468804576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24394862353801727,
+      "objective/train/value_max": -0.5,
+      "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": 1.3603086471557617,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3410303405341115,
+      "tokens_seen": 3098345472
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.17075910768737e-05,
+      "loss": 1.2937,
+      "theoretical_loss": 3.3410303405341115,
+      "tokens_seen": 3098345472
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.166746910608249e-05,
+      "loss": 1.3037,
+      "theoretical_loss": 3.3410196368149423,
+      "tokens_seen": 3098476544
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.162734713529128e-05,
+      "loss": 1.2717,
+      "theoretical_loss": 3.3410089336753277,
+      "tokens_seen": 3098607616
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.158722516450009e-05,
+      "loss": 1.3144,
+      "theoretical_loss": 3.3409982311152127,
+      "tokens_seen": 3098738688
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.154710319370888e-05,
+      "loss": 1.3389,
+      "theoretical_loss": 3.340987529134541,
+      "tokens_seen": 3098869760
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.150698122291767e-05,
+      "loss": 1.2607,
+      "theoretical_loss": 3.340976827733256,
+      "tokens_seen": 3099000832
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.146685925212647e-05,
+      "loss": 1.3231,
+      "theoretical_loss": 3.3409661269113027,
+      "tokens_seen": 3099131904
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.142673728133526e-05,
+      "loss": 1.3945,
+      "theoretical_loss": 3.340955426668625,
+      "tokens_seen": 3099262976
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.138661531054405e-05,
+      "loss": 1.2902,
+      "theoretical_loss": 3.340944727005167,
+      "tokens_seen": 3099394048
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.134649333975284e-05,
+      "loss": 1.3517,
+      "theoretical_loss": 3.340934027920873,
+      "tokens_seen": 3099525120
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.130637136896165e-05,
+      "loss": 1.34,
+      "theoretical_loss": 3.340923329415687,
+      "tokens_seen": 3099656192
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.126624939817044e-05,
+      "loss": 1.294,
+      "theoretical_loss": 3.3409126314895534,
+      "tokens_seen": 3099787264
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.122612742737924e-05,
+      "loss": 1.3324,
+      "theoretical_loss": 3.3409019341424155,
+      "tokens_seen": 3099918336
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.118600545658803e-05,
+      "loss": 1.358,
+      "theoretical_loss": 3.340891237374219,
+      "tokens_seen": 3100049408
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.114588348579682e-05,
+      "loss": 1.2841,
+      "theoretical_loss": 3.340880541184907,
+      "tokens_seen": 3100180480
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.110576151500561e-05,
+      "loss": 1.3006,
+      "theoretical_loss": 3.3408698455744235,
+      "tokens_seen": 3100311552
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.106563954421442e-05,
+      "loss": 1.2813,
+      "theoretical_loss": 3.3408591505427134,
+      "tokens_seen": 3100442624
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.1025517573423206e-05,
+      "loss": 1.3027,
+      "theoretical_loss": 3.340848456089721,
+      "tokens_seen": 3100573696
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.0985395602632006e-05,
+      "loss": 1.3352,
+      "theoretical_loss": 3.34083776221539,
+      "tokens_seen": 3100704768
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.09452736318408e-05,
+      "loss": 1.3404,
+      "theoretical_loss": 3.3408270689196646,
+      "tokens_seen": 3100835840
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.090515166104959e-05,
+      "loss": 1.253,
+      "theoretical_loss": 3.34081637620249,
+      "tokens_seen": 3100966912
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.086502969025839e-05,
+      "loss": 1.2781,
+      "theoretical_loss": 3.3408056840638087,
+      "tokens_seen": 3101097984
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.082490771946718e-05,
+      "loss": 1.2869,
+      "theoretical_loss": 3.3407949925035667,
+      "tokens_seen": 3101229056
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.0784785748675976e-05,
+      "loss": 1.2495,
+      "theoretical_loss": 3.3407843015217074,
+      "tokens_seen": 3101360128
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.074466377788477e-05,
+      "loss": 1.227,
+      "theoretical_loss": 3.3407736111181747,
+      "tokens_seen": 3101491200
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.070454180709357e-05,
+      "loss": 1.252,
+      "theoretical_loss": 3.3407629212929137,
+      "tokens_seen": 3101622272
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.066441983630236e-05,
+      "loss": 1.2964,
+      "theoretical_loss": 3.3407522320458685,
+      "tokens_seen": 3101753344
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.062429786551116e-05,
+      "loss": 1.2755,
+      "theoretical_loss": 3.3407415433769834,
+      "tokens_seen": 3101884416
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.058417589471995e-05,
+      "loss": 1.3424,
+      "theoretical_loss": 3.3407308552862025,
+      "tokens_seen": 3102015488
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.0544053923928746e-05,
+      "loss": 1.2654,
+      "theoretical_loss": 3.34072016777347,
+      "tokens_seen": 3102146560
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.050393195313754e-05,
+      "loss": 1.2336,
+      "theoretical_loss": 3.34070948083873,
+      "tokens_seen": 3102277632
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.046380998234633e-05,
+      "loss": 1.3349,
+      "theoretical_loss": 3.3406987944819275,
+      "tokens_seen": 3102408704
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.042368801155513e-05,
+      "loss": 1.3191,
+      "theoretical_loss": 3.3406881087030067,
+      "tokens_seen": 3102539776
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.0383566040763924e-05,
+      "loss": 1.3483,
+      "theoretical_loss": 3.3406774235019117,
+      "tokens_seen": 3102670848
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.034344406997272e-05,
+      "loss": 1.2719,
+      "theoretical_loss": 3.340666738878587,
+      "tokens_seen": 3102801920
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.030332209918151e-05,
+      "loss": 1.2267,
+      "theoretical_loss": 3.3406560548329765,
+      "tokens_seen": 3102932992
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.026320012839031e-05,
+      "loss": 1.2014,
+      "theoretical_loss": 3.3406453713650253,
+      "tokens_seen": 3103064064
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.02230781575991e-05,
+      "loss": 1.3458,
+      "theoretical_loss": 3.340634688474677,
+      "tokens_seen": 3103195136
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.01829561868079e-05,
+      "loss": 1.308,
+      "theoretical_loss": 3.340624006161877,
+      "tokens_seen": 3103326208
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.0142834216016694e-05,
+      "loss": 1.2715,
+      "theoretical_loss": 3.3406133244265686,
+      "tokens_seen": 3103457280
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.010271224522549e-05,
+      "loss": 1.2563,
+      "theoretical_loss": 3.3406026432686966,
+      "tokens_seen": 3103588352
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.006259027443428e-05,
+      "loss": 1.2955,
+      "theoretical_loss": 3.3405919626882055,
+      "tokens_seen": 3103719424
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 6.002246830364308e-05,
+      "loss": 1.2106,
+      "theoretical_loss": 3.34058128268504,
+      "tokens_seen": 3103850496
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.998234633285187e-05,
+      "loss": 1.4049,
+      "theoretical_loss": 3.340570603259144,
+      "tokens_seen": 3103981568
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.9942224362060665e-05,
+      "loss": 1.2631,
+      "theoretical_loss": 3.340559924410462,
+      "tokens_seen": 3104112640
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.9902102391269464e-05,
+      "loss": 1.2877,
+      "theoretical_loss": 3.3405492461389388,
+      "tokens_seen": 3104243712
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.986198042047825e-05,
+      "loss": 1.2705,
+      "theoretical_loss": 3.340538568444518,
+      "tokens_seen": 3104374784
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.982185844968705e-05,
+      "loss": 1.3012,
+      "theoretical_loss": 3.3405278913271452,
+      "tokens_seen": 3104505856
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.978173647889584e-05,
+      "loss": 1.2309,
+      "theoretical_loss": 3.340517214786764,
+      "tokens_seen": 3104636928
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.974161450810464e-05,
+      "loss": 1.253,
+      "theoretical_loss": 3.3405065388233193,
+      "tokens_seen": 3104768000
+    },
+    {
+      "epoch": 0.88,
+      "objective/train/advantage_avg": 0.4935816526412964,
+      "objective/train/docs_used": 1128234,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.21914541721344,
+      "objective/train/original_loss": 1.2191452980041504,
+      "objective/train/theoretical_loss": 3.340495863436755,
+      "objective/train/tokens_used": 1475358176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24392729997634888,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.05059814453125,
+      "objective/train/weighted_lm_loss": 1.2807223796844482,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.340495863436755,
+      "tokens_seen": 3104899072
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.9701492537313435e-05,
+      "loss": 1.3737,
+      "theoretical_loss": 3.340495863436755,
+      "tokens_seen": 3104899072
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.9661370566522234e-05,
+      "loss": 1.2255,
+      "theoretical_loss": 3.3404851886270164,
+      "tokens_seen": 3105030144
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.962124859573102e-05,
+      "loss": 1.3042,
+      "theoretical_loss": 3.3404745143940473,
+      "tokens_seen": 3105161216
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.958112662493982e-05,
+      "loss": 1.3459,
+      "theoretical_loss": 3.340463840737793,
+      "tokens_seen": 3105292288
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.954100465414861e-05,
+      "loss": 1.3564,
+      "theoretical_loss": 3.3404531676581968,
+      "tokens_seen": 3105423360
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.9500882683357405e-05,
+      "loss": 1.3112,
+      "theoretical_loss": 3.3404424951552043,
+      "tokens_seen": 3105554432
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.9460760712566205e-05,
+      "loss": 1.2616,
+      "theoretical_loss": 3.340431823228759,
+      "tokens_seen": 3105685504
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.9420638741775e-05,
+      "loss": 1.4088,
+      "theoretical_loss": 3.3404211518788065,
+      "tokens_seen": 3105816576
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.938051677098379e-05,
+      "loss": 1.2208,
+      "theoretical_loss": 3.3404104811052906,
+      "tokens_seen": 3105947648
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.934039480019258e-05,
+      "loss": 1.3281,
+      "theoretical_loss": 3.3403998109081563,
+      "tokens_seen": 3106078720
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.930027282940138e-05,
+      "loss": 1.2914,
+      "theoretical_loss": 3.340389141287348,
+      "tokens_seen": 3106209792
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.9260150858610175e-05,
+      "loss": 1.2971,
+      "theoretical_loss": 3.34037847224281,
+      "tokens_seen": 3106340864
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.9220028887818975e-05,
+      "loss": 1.2687,
+      "theoretical_loss": 3.340367803774487,
+      "tokens_seen": 3106471936
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.917990691702777e-05,
+      "loss": 1.2859,
+      "theoretical_loss": 3.3403571358823236,
+      "tokens_seen": 3106603008
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.913978494623656e-05,
+      "loss": 1.2862,
+      "theoretical_loss": 3.3403464685662643,
+      "tokens_seen": 3106734080
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.909966297544535e-05,
+      "loss": 1.3266,
+      "theoretical_loss": 3.340335801826254,
+      "tokens_seen": 3106865152
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.9059541004654146e-05,
+      "loss": 1.3382,
+      "theoretical_loss": 3.340325135662237,
+      "tokens_seen": 3106996224
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.9019419033862945e-05,
+      "loss": 1.2247,
+      "theoretical_loss": 3.3403144700741576,
+      "tokens_seen": 3107127296
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.897929706307174e-05,
+      "loss": 1.3033,
+      "theoretical_loss": 3.340303805061961,
+      "tokens_seen": 3107258368
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.893917509228054e-05,
+      "loss": 1.3086,
+      "theoretical_loss": 3.340293140625592,
+      "tokens_seen": 3107389440
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.8899053121489324e-05,
+      "loss": 1.2617,
+      "theoretical_loss": 3.3402824767649943,
+      "tokens_seen": 3107520512
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.885893115069812e-05,
+      "loss": 1.3607,
+      "theoretical_loss": 3.340271813480113,
+      "tokens_seen": 3107651584
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.8818809179906916e-05,
+      "loss": 1.3173,
+      "theoretical_loss": 3.3402611507708935,
+      "tokens_seen": 3107782656
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.8778687209115716e-05,
+      "loss": 1.383,
+      "theoretical_loss": 3.340250488637279,
+      "tokens_seen": 3107913728
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.873856523832451e-05,
+      "loss": 1.3265,
+      "theoretical_loss": 3.340239827079215,
+      "tokens_seen": 3108044800
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.869844326753331e-05,
+      "loss": 1.2885,
+      "theoretical_loss": 3.340229166096646,
+      "tokens_seen": 3108175872
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.8658321296742094e-05,
+      "loss": 1.2552,
+      "theoretical_loss": 3.3402185056895166,
+      "tokens_seen": 3108306944
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.861819932595089e-05,
+      "loss": 1.2693,
+      "theoretical_loss": 3.3402078458577718,
+      "tokens_seen": 3108438016
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.8578077355159686e-05,
+      "loss": 1.3164,
+      "theoretical_loss": 3.3401971866013564,
+      "tokens_seen": 3108569088
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.853795538436848e-05,
+      "loss": 1.3045,
+      "theoretical_loss": 3.3401865279202143,
+      "tokens_seen": 3108700160
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.849783341357728e-05,
+      "loss": 1.3227,
+      "theoretical_loss": 3.3401758698142907,
+      "tokens_seen": 3108831232
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.845771144278607e-05,
+      "loss": 1.3391,
+      "theoretical_loss": 3.34016521228353,
+      "tokens_seen": 3108962304
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.8417589471994864e-05,
+      "loss": 1.2621,
+      "theoretical_loss": 3.340154555327878,
+      "tokens_seen": 3109093376
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.837746750120366e-05,
+      "loss": 1.296,
+      "theoretical_loss": 3.340143898947278,
+      "tokens_seen": 3109224448
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.8337345530412456e-05,
+      "loss": 1.4093,
+      "theoretical_loss": 3.3401332431416755,
+      "tokens_seen": 3109355520
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.829722355962125e-05,
+      "loss": 1.3094,
+      "theoretical_loss": 3.340122587911015,
+      "tokens_seen": 3109486592
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.825710158883005e-05,
+      "loss": 1.3203,
+      "theoretical_loss": 3.340111933255241,
+      "tokens_seen": 3109617664
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.821697961803884e-05,
+      "loss": 1.3197,
+      "theoretical_loss": 3.340101279174299,
+      "tokens_seen": 3109748736
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.8176857647247634e-05,
+      "loss": 1.3483,
+      "theoretical_loss": 3.3400906256681333,
+      "tokens_seen": 3109879808
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.813673567645643e-05,
+      "loss": 1.3179,
+      "theoretical_loss": 3.3400799727366883,
+      "tokens_seen": 3110010880
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.809661370566522e-05,
+      "loss": 1.2633,
+      "theoretical_loss": 3.3400693203799094,
+      "tokens_seen": 3110141952
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.805649173487402e-05,
+      "loss": 1.3338,
+      "theoretical_loss": 3.3400586685977416,
+      "tokens_seen": 3110273024
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.801636976408281e-05,
+      "loss": 1.2583,
+      "theoretical_loss": 3.340048017390129,
+      "tokens_seen": 3110404096
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.797624779329161e-05,
+      "loss": 1.4092,
+      "theoretical_loss": 3.3400373667570165,
+      "tokens_seen": 3110535168
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.79361258225004e-05,
+      "loss": 1.2687,
+      "theoretical_loss": 3.3400267166983486,
+      "tokens_seen": 3110666240
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.78960038517092e-05,
+      "loss": 1.3156,
+      "theoretical_loss": 3.340016067214071,
+      "tokens_seen": 3110797312
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.785588188091799e-05,
+      "loss": 1.2713,
+      "theoretical_loss": 3.3400054183041283,
+      "tokens_seen": 3110928384
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.781575991012679e-05,
+      "loss": 1.3348,
+      "theoretical_loss": 3.339994769968465,
+      "tokens_seen": 3111059456
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.777563793933558e-05,
+      "loss": 1.3256,
+      "theoretical_loss": 3.339984122207026,
+      "tokens_seen": 3111190528
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.773551596854438e-05,
+      "loss": 1.2488,
+      "theoretical_loss": 3.339973475019756,
+      "tokens_seen": 3111321600
+    },
+    {
+      "epoch": 0.89,
+      "objective/train/advantage_avg": 0.49214473366737366,
+      "objective/train/docs_used": 1130437,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2263256311416626,
+      "objective/train/original_loss": 1.2263256311416626,
+      "objective/train/theoretical_loss": 3.3399628284066,
+      "objective/train/tokens_used": 1481911776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24270327389240265,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504482984542847,
+      "objective/train/weighted_lm_loss": 1.288198709487915,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3399628284066,
+      "tokens_seen": 3111452672
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.769539399775317e-05,
+      "loss": 1.3625,
+      "theoretical_loss": 3.3399628284066,
+      "tokens_seen": 3111452672
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.765527202696196e-05,
+      "loss": 1.3122,
+      "theoretical_loss": 3.3399521823675036,
+      "tokens_seen": 3111583744
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.761515005617076e-05,
+      "loss": 1.234,
+      "theoretical_loss": 3.3399415369024106,
+      "tokens_seen": 3111714816
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.757502808537955e-05,
+      "loss": 1.281,
+      "theoretical_loss": 3.339930892011266,
+      "tokens_seen": 3111845888
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.753490611458835e-05,
+      "loss": 1.2985,
+      "theoretical_loss": 3.339920247694015,
+      "tokens_seen": 3111976960
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.7494784143797145e-05,
+      "loss": 1.2892,
+      "theoretical_loss": 3.339909603950603,
+      "tokens_seen": 3112108032
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.745466217300594e-05,
+      "loss": 1.2499,
+      "theoretical_loss": 3.3398989607809737,
+      "tokens_seen": 3112239104
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.741454020221473e-05,
+      "loss": 1.4004,
+      "theoretical_loss": 3.3398883181850723,
+      "tokens_seen": 3112370176
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.737441823142353e-05,
+      "loss": 1.3673,
+      "theoretical_loss": 3.339877676162845,
+      "tokens_seen": 3112501248
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.733429626063232e-05,
+      "loss": 1.2153,
+      "theoretical_loss": 3.339867034714235,
+      "tokens_seen": 3112632320
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.729417428984112e-05,
+      "loss": 1.3385,
+      "theoretical_loss": 3.339856393839188,
+      "tokens_seen": 3112763392
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.7254052319049915e-05,
+      "loss": 1.2876,
+      "theoretical_loss": 3.3398457535376496,
+      "tokens_seen": 3112894464
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.7213930348258714e-05,
+      "loss": 1.366,
+      "theoretical_loss": 3.3398351138095634,
+      "tokens_seen": 3113025536
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.71738083774675e-05,
+      "loss": 1.3463,
+      "theoretical_loss": 3.3398244746548755,
+      "tokens_seen": 3113156608
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.713368640667629e-05,
+      "loss": 1.3364,
+      "theoretical_loss": 3.3398138360735303,
+      "tokens_seen": 3113287680
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.709356443588509e-05,
+      "loss": 1.3643,
+      "theoretical_loss": 3.3398031980654728,
+      "tokens_seen": 3113418752
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.7053442465093885e-05,
+      "loss": 1.2455,
+      "theoretical_loss": 3.3397925606306478,
+      "tokens_seen": 3113549824
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.7013320494302685e-05,
+      "loss": 1.3211,
+      "theoretical_loss": 3.3397819237690007,
+      "tokens_seen": 3113680896
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.697319852351147e-05,
+      "loss": 1.3366,
+      "theoretical_loss": 3.339771287480476,
+      "tokens_seen": 3113811968
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.693307655272027e-05,
+      "loss": 1.2726,
+      "theoretical_loss": 3.33976065176502,
+      "tokens_seen": 3113943040
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.689295458192906e-05,
+      "loss": 1.3832,
+      "theoretical_loss": 3.3397500166225753,
+      "tokens_seen": 3114074112
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.685283261113786e-05,
+      "loss": 1.3082,
+      "theoretical_loss": 3.339739382053089,
+      "tokens_seen": 3114205184
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.6812710640346655e-05,
+      "loss": 1.356,
+      "theoretical_loss": 3.3397287480565057,
+      "tokens_seen": 3114336256
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.6772588669555455e-05,
+      "loss": 1.2955,
+      "theoretical_loss": 3.3397181146327695,
+      "tokens_seen": 3114467328
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.673246669876425e-05,
+      "loss": 1.3165,
+      "theoretical_loss": 3.3397074817818266,
+      "tokens_seen": 3114598400
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.6692344727973034e-05,
+      "loss": 1.3389,
+      "theoretical_loss": 3.3396968495036212,
+      "tokens_seen": 3114729472
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.665222275718183e-05,
+      "loss": 1.2444,
+      "theoretical_loss": 3.339686217798099,
+      "tokens_seen": 3114860544
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.6612100786390626e-05,
+      "loss": 1.322,
+      "theoretical_loss": 3.339675586665204,
+      "tokens_seen": 3114991616
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.6571978815599425e-05,
+      "loss": 1.3249,
+      "theoretical_loss": 3.339664956104883,
+      "tokens_seen": 3115122688
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.653185684480822e-05,
+      "loss": 1.2968,
+      "theoretical_loss": 3.3396543261170795,
+      "tokens_seen": 3115253760
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.649173487401702e-05,
+      "loss": 1.3435,
+      "theoretical_loss": 3.3396436967017396,
+      "tokens_seen": 3115384832
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.6451612903225804e-05,
+      "loss": 1.3313,
+      "theoretical_loss": 3.3396330678588075,
+      "tokens_seen": 3115515904
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.64114909324346e-05,
+      "loss": 1.342,
+      "theoretical_loss": 3.339622439588229,
+      "tokens_seen": 3115646976
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.6371368961643396e-05,
+      "loss": 1.3026,
+      "theoretical_loss": 3.3396118118899487,
+      "tokens_seen": 3115778048
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.6331246990852196e-05,
+      "loss": 1.3487,
+      "theoretical_loss": 3.3396011847639118,
+      "tokens_seen": 3115909120
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.629112502006099e-05,
+      "loss": 1.3596,
+      "theoretical_loss": 3.339590558210064,
+      "tokens_seen": 3116040192
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.625100304926979e-05,
+      "loss": 1.3247,
+      "theoretical_loss": 3.3395799322283497,
+      "tokens_seen": 3116171264
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.6210881078478574e-05,
+      "loss": 1.3276,
+      "theoretical_loss": 3.3395693068187144,
+      "tokens_seen": 3116302336
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.6170759107687367e-05,
+      "loss": 1.3176,
+      "theoretical_loss": 3.339558681981103,
+      "tokens_seen": 3116433408
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.6130637136896166e-05,
+      "loss": 1.327,
+      "theoretical_loss": 3.3395480577154615,
+      "tokens_seen": 3116564480
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.609051516610496e-05,
+      "loss": 1.4045,
+      "theoretical_loss": 3.3395374340217336,
+      "tokens_seen": 3116695552
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.605039319531376e-05,
+      "loss": 1.3831,
+      "theoretical_loss": 3.3395268108998657,
+      "tokens_seen": 3116826624
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.601027122452255e-05,
+      "loss": 1.2725,
+      "theoretical_loss": 3.339516188349802,
+      "tokens_seen": 3116957696
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.5970149253731344e-05,
+      "loss": 1.3836,
+      "theoretical_loss": 3.3395055663714888,
+      "tokens_seen": 3117088768
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.5930027282940137e-05,
+      "loss": 1.3061,
+      "theoretical_loss": 3.33949494496487,
+      "tokens_seen": 3117219840
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.5889905312148936e-05,
+      "loss": 1.3016,
+      "theoretical_loss": 3.3394843241298924,
+      "tokens_seen": 3117350912
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.584978334135773e-05,
+      "loss": 1.3316,
+      "theoretical_loss": 3.3394737038664997,
+      "tokens_seen": 3117481984
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.580966137056653e-05,
+      "loss": 1.3781,
+      "theoretical_loss": 3.3394630841746373,
+      "tokens_seen": 3117613056
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.576953939977532e-05,
+      "loss": 1.3845,
+      "theoretical_loss": 3.3394524650542516,
+      "tokens_seen": 3117744128
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.572941742898411e-05,
+      "loss": 1.3514,
+      "theoretical_loss": 3.3394418465052866,
+      "tokens_seen": 3117875200
+    },
+    {
+      "epoch": 0.89,
+      "objective/train/advantage_avg": 0.4936418831348419,
+      "objective/train/docs_used": 1132901,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3795562982559204,
+      "objective/train/original_loss": 1.37955641746521,
+      "objective/train/theoretical_loss": 3.3394312285276877,
+      "objective/train/tokens_used": 1488465376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2440977841615677,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506051778793335,
+      "objective/train/weighted_lm_loss": 1.449318289756775,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3394312285276877,
+      "tokens_seen": 3118006272
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.568929545819291e-05,
+      "loss": 1.45,
+      "theoretical_loss": 3.3394312285276877,
+      "tokens_seen": 3118006272
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.56491734874017e-05,
+      "loss": 1.4195,
+      "theoretical_loss": 3.3394206111214007,
+      "tokens_seen": 3118137344
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.56090515166105e-05,
+      "loss": 1.3338,
+      "theoretical_loss": 3.3394099942863704,
+      "tokens_seen": 3118268416
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.556892954581929e-05,
+      "loss": 1.3565,
+      "theoretical_loss": 3.3393993780225424,
+      "tokens_seen": 3118399488
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.552880757502809e-05,
+      "loss": 1.3649,
+      "theoretical_loss": 3.339388762329862,
+      "tokens_seen": 3118530560
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.548868560423688e-05,
+      "loss": 1.3629,
+      "theoretical_loss": 3.3393781472082735,
+      "tokens_seen": 3118661632
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.544856363344568e-05,
+      "loss": 1.3474,
+      "theoretical_loss": 3.3393675326577235,
+      "tokens_seen": 3118792704
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.540844166265447e-05,
+      "loss": 1.3992,
+      "theoretical_loss": 3.339356918678156,
+      "tokens_seen": 3118923776
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.536831969186327e-05,
+      "loss": 1.383,
+      "theoretical_loss": 3.3393463052695176,
+      "tokens_seen": 3119054848
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.532819772107206e-05,
+      "loss": 1.3093,
+      "theoretical_loss": 3.339335692431753,
+      "tokens_seen": 3119185920
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.5288075750280855e-05,
+      "loss": 1.3782,
+      "theoretical_loss": 3.339325080164807,
+      "tokens_seen": 3119316992
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.524795377948965e-05,
+      "loss": 1.3349,
+      "theoretical_loss": 3.3393144684686256,
+      "tokens_seen": 3119448064
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.520783180869844e-05,
+      "loss": 1.281,
+      "theoretical_loss": 3.339303857343154,
+      "tokens_seen": 3119579136
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.516770983790724e-05,
+      "loss": 1.3255,
+      "theoretical_loss": 3.3392932467883374,
+      "tokens_seen": 3119710208
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.512758786711603e-05,
+      "loss": 1.3033,
+      "theoretical_loss": 3.3392826368041213,
+      "tokens_seen": 3119841280
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.508746589632483e-05,
+      "loss": 1.3503,
+      "theoretical_loss": 3.3392720273904506,
+      "tokens_seen": 3119972352
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.5047343925533625e-05,
+      "loss": 1.38,
+      "theoretical_loss": 3.339261418547271,
+      "tokens_seen": 3120103424
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.500722195474242e-05,
+      "loss": 1.4086,
+      "theoretical_loss": 3.339250810274528,
+      "tokens_seen": 3120234496
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.496709998395121e-05,
+      "loss": 1.3303,
+      "theoretical_loss": 3.339240202572167,
+      "tokens_seen": 3120365568
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.492697801316001e-05,
+      "loss": 1.3402,
+      "theoretical_loss": 3.339229595440133,
+      "tokens_seen": 3120496640
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.48868560423688e-05,
+      "loss": 1.3917,
+      "theoretical_loss": 3.3392189888783714,
+      "tokens_seen": 3120627712
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.48467340715776e-05,
+      "loss": 1.3708,
+      "theoretical_loss": 3.339208382886828,
+      "tokens_seen": 3120758784
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.4806612100786395e-05,
+      "loss": 1.3838,
+      "theoretical_loss": 3.339197777465448,
+      "tokens_seen": 3120889856
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.476649012999518e-05,
+      "loss": 1.3295,
+      "theoretical_loss": 3.3391871726141766,
+      "tokens_seen": 3121020928
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.472636815920398e-05,
+      "loss": 1.335,
+      "theoretical_loss": 3.3391765683329595,
+      "tokens_seen": 3121152000
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.468624618841277e-05,
+      "loss": 1.3704,
+      "theoretical_loss": 3.339165964621742,
+      "tokens_seen": 3121283072
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.464612421762157e-05,
+      "loss": 1.3591,
+      "theoretical_loss": 3.339155361480469,
+      "tokens_seen": 3121414144
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.4606002246830365e-05,
+      "loss": 1.3578,
+      "theoretical_loss": 3.339144758909087,
+      "tokens_seen": 3121545216
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.4565880276039165e-05,
+      "loss": 1.4307,
+      "theoretical_loss": 3.339134156907541,
+      "tokens_seen": 3121676288
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.452575830524795e-05,
+      "loss": 1.3118,
+      "theoretical_loss": 3.339123555475776,
+      "tokens_seen": 3121807360
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.448563633445675e-05,
+      "loss": 1.3257,
+      "theoretical_loss": 3.3391129546137375,
+      "tokens_seen": 3121938432
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.444551436366554e-05,
+      "loss": 1.3166,
+      "theoretical_loss": 3.339102354321372,
+      "tokens_seen": 3122069504
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.440539239287434e-05,
+      "loss": 1.4123,
+      "theoretical_loss": 3.3390917545986234,
+      "tokens_seen": 3122200576
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.4365270422083135e-05,
+      "loss": 1.3119,
+      "theoretical_loss": 3.339081155445439,
+      "tokens_seen": 3122331648
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.432514845129193e-05,
+      "loss": 1.3748,
+      "theoretical_loss": 3.3390705568617625,
+      "tokens_seen": 3122462720
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.428502648050072e-05,
+      "loss": 1.3842,
+      "theoretical_loss": 3.3390599588475407,
+      "tokens_seen": 3122593792
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.4244904509709514e-05,
+      "loss": 1.3491,
+      "theoretical_loss": 3.339049361402718,
+      "tokens_seen": 3122724864
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.420478253891831e-05,
+      "loss": 1.3316,
+      "theoretical_loss": 3.339038764527241,
+      "tokens_seen": 3122855936
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.4164660568127106e-05,
+      "loss": 1.3997,
+      "theoretical_loss": 3.3390281682210547,
+      "tokens_seen": 3122987008
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.4124538597335905e-05,
+      "loss": 1.3557,
+      "theoretical_loss": 3.3390175724841047,
+      "tokens_seen": 3123118080
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.40844166265447e-05,
+      "loss": 1.329,
+      "theoretical_loss": 3.3390069773163362,
+      "tokens_seen": 3123249152
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.404429465575349e-05,
+      "loss": 1.3068,
+      "theoretical_loss": 3.338996382717695,
+      "tokens_seen": 3123380224
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.4004172684962284e-05,
+      "loss": 1.3029,
+      "theoretical_loss": 3.3389857886881273,
+      "tokens_seen": 3123511296
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.396405071417108e-05,
+      "loss": 1.3088,
+      "theoretical_loss": 3.3389751952275777,
+      "tokens_seen": 3123642368
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.3923928743379876e-05,
+      "loss": 1.4135,
+      "theoretical_loss": 3.3389646023359916,
+      "tokens_seen": 3123773440
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.3883806772588675e-05,
+      "loss": 1.3079,
+      "theoretical_loss": 3.3389540100133157,
+      "tokens_seen": 3123904512
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.384368480179747e-05,
+      "loss": 1.3813,
+      "theoretical_loss": 3.3389434182594946,
+      "tokens_seen": 3124035584
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.3803562831006254e-05,
+      "loss": 1.3703,
+      "theoretical_loss": 3.3389328270744745,
+      "tokens_seen": 3124166656
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.3763440860215054e-05,
+      "loss": 1.2059,
+      "theoretical_loss": 3.3389222364582003,
+      "tokens_seen": 3124297728
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.3723318889423846e-05,
+      "loss": 1.3273,
+      "theoretical_loss": 3.3389116464106183,
+      "tokens_seen": 3124428800
+    },
+    {
+      "epoch": 0.89,
+      "objective/train/advantage_avg": 0.49282315373420715,
+      "objective/train/docs_used": 1135223,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2427654266357422,
+      "objective/train/original_loss": 1.242765188217163,
+      "objective/train/theoretical_loss": 3.338901056931674,
+      "objective/train/tokens_used": 1495018976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24352626502513885,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050520420074463,
+      "objective/train/weighted_lm_loss": 1.3054494857788086,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.7788007855415344,
+      "theoretical_loss": 3.338901056931674,
+      "tokens_seen": 3124559872
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.3683196918632646e-05,
+      "loss": 1.3084,
+      "theoretical_loss": 3.338901056931674,
+      "tokens_seen": 3124559872
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.364307494784144e-05,
+      "loss": 1.3289,
+      "theoretical_loss": 3.3388904680213125,
+      "tokens_seen": 3124690944
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.360295297705024e-05,
+      "loss": 1.2488,
+      "theoretical_loss": 3.33887987967948,
+      "tokens_seen": 3124822016
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.3562831006259024e-05,
+      "loss": 1.3174,
+      "theoretical_loss": 3.3388692919061214,
+      "tokens_seen": 3124953088
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.3522709035467824e-05,
+      "loss": 1.3086,
+      "theoretical_loss": 3.3388587047011837,
+      "tokens_seen": 3125084160
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.3482587064676617e-05,
+      "loss": 1.3556,
+      "theoretical_loss": 3.338848118064611,
+      "tokens_seen": 3125215232
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.3442465093885416e-05,
+      "loss": 1.4009,
+      "theoretical_loss": 3.33883753199635,
+      "tokens_seen": 3125346304
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.340234312309421e-05,
+      "loss": 1.283,
+      "theoretical_loss": 3.3388269464963454,
+      "tokens_seen": 3125477376
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.3362221152303e-05,
+      "loss": 1.3628,
+      "theoretical_loss": 3.338816361564544,
+      "tokens_seen": 3125608448
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.3322099181511794e-05,
+      "loss": 1.29,
+      "theoretical_loss": 3.338805777200891,
+      "tokens_seen": 3125739520
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.328197721072059e-05,
+      "loss": 1.3098,
+      "theoretical_loss": 3.3387951934053315,
+      "tokens_seen": 3125870592
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.324185523992939e-05,
+      "loss": 1.3203,
+      "theoretical_loss": 3.338784610177812,
+      "tokens_seen": 3126001664
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.320173326913818e-05,
+      "loss": 1.2578,
+      "theoretical_loss": 3.338774027518278,
+      "tokens_seen": 3126132736
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.316161129834698e-05,
+      "loss": 1.3565,
+      "theoretical_loss": 3.338763445426675,
+      "tokens_seen": 3126263808
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.312148932755577e-05,
+      "loss": 1.345,
+      "theoretical_loss": 3.3387528639029487,
+      "tokens_seen": 3126394880
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.3081367356764564e-05,
+      "loss": 1.317,
+      "theoretical_loss": 3.338742282947045,
+      "tokens_seen": 3126525952
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.304124538597336e-05,
+      "loss": 1.3516,
+      "theoretical_loss": 3.3387317025589094,
+      "tokens_seen": 3126657024
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.300112341518216e-05,
+      "loss": 1.3891,
+      "theoretical_loss": 3.3387211227384883,
+      "tokens_seen": 3126788096
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.296100144439095e-05,
+      "loss": 1.382,
+      "theoretical_loss": 3.3387105434857265,
+      "tokens_seen": 3126919168
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.292087947359974e-05,
+      "loss": 1.3036,
+      "theoretical_loss": 3.33869996480057,
+      "tokens_seen": 3127050240
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.288075750280854e-05,
+      "loss": 1.3152,
+      "theoretical_loss": 3.338689386682965,
+      "tokens_seen": 3127181312
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.284063553201733e-05,
+      "loss": 1.3592,
+      "theoretical_loss": 3.3386788091328574,
+      "tokens_seen": 3127312384
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.280051356122613e-05,
+      "loss": 1.3431,
+      "theoretical_loss": 3.338668232150192,
+      "tokens_seen": 3127443456
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.276039159043492e-05,
+      "loss": 1.3368,
+      "theoretical_loss": 3.3386576557349157,
+      "tokens_seen": 3127574528
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.272026961964372e-05,
+      "loss": 1.3634,
+      "theoretical_loss": 3.338647079886973,
+      "tokens_seen": 3127705600
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.268014764885251e-05,
+      "loss": 1.3816,
+      "theoretical_loss": 3.338636504606311,
+      "tokens_seen": 3127836672
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.264002567806131e-05,
+      "loss": 1.3453,
+      "theoretical_loss": 3.338625929892875,
+      "tokens_seen": 3127967744
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.25999037072701e-05,
+      "loss": 1.2471,
+      "theoretical_loss": 3.3386153557466103,
+      "tokens_seen": 3128098816
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.25597817364789e-05,
+      "loss": 1.285,
+      "theoretical_loss": 3.3386047821674634,
+      "tokens_seen": 3128229888
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.251965976568769e-05,
+      "loss": 1.2848,
+      "theoretical_loss": 3.33859420915538,
+      "tokens_seen": 3128360960
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.247953779489649e-05,
+      "loss": 1.4036,
+      "theoretical_loss": 3.3385836367103057,
+      "tokens_seen": 3128492032
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.243941582410528e-05,
+      "loss": 1.3115,
+      "theoretical_loss": 3.3385730648321865,
+      "tokens_seen": 3128623104
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.2399293853314075e-05,
+      "loss": 1.2935,
+      "theoretical_loss": 3.338562493520968,
+      "tokens_seen": 3128754176
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.2359171882522875e-05,
+      "loss": 1.3382,
+      "theoretical_loss": 3.3385519227765967,
+      "tokens_seen": 3128885248
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.231904991173166e-05,
+      "loss": 1.3246,
+      "theoretical_loss": 3.3385413525990177,
+      "tokens_seen": 3129016320
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.227892794094046e-05,
+      "loss": 1.3633,
+      "theoretical_loss": 3.3385307829881774,
+      "tokens_seen": 3129147392
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.223880597014925e-05,
+      "loss": 1.2855,
+      "theoretical_loss": 3.338520213944021,
+      "tokens_seen": 3129278464
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.219868399935805e-05,
+      "loss": 1.262,
+      "theoretical_loss": 3.3385096454664955,
+      "tokens_seen": 3129409536
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.2158562028566845e-05,
+      "loss": 1.3107,
+      "theoretical_loss": 3.3384990775555456,
+      "tokens_seen": 3129540608
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.2118440057775645e-05,
+      "loss": 1.323,
+      "theoretical_loss": 3.338488510211118,
+      "tokens_seen": 3129671680
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.207831808698443e-05,
+      "loss": 1.3634,
+      "theoretical_loss": 3.3384779434331584,
+      "tokens_seen": 3129802752
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.203819611619323e-05,
+      "loss": 1.4209,
+      "theoretical_loss": 3.3384673772216127,
+      "tokens_seen": 3129933824
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.199807414540202e-05,
+      "loss": 1.412,
+      "theoretical_loss": 3.3384568115764264,
+      "tokens_seen": 3130064896
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.1957952174610816e-05,
+      "loss": 1.2679,
+      "theoretical_loss": 3.338446246497546,
+      "tokens_seen": 3130195968
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.1917830203819615e-05,
+      "loss": 1.3691,
+      "theoretical_loss": 3.3384356819849175,
+      "tokens_seen": 3130327040
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.187770823302841e-05,
+      "loss": 1.3926,
+      "theoretical_loss": 3.3384251180384865,
+      "tokens_seen": 3130458112
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.18375862622372e-05,
+      "loss": 1.2909,
+      "theoretical_loss": 3.3384145546581987,
+      "tokens_seen": 3130589184
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.1797464291445994e-05,
+      "loss": 1.2891,
+      "theoretical_loss": 3.3384039918440007,
+      "tokens_seen": 3130720256
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.175734232065479e-05,
+      "loss": 1.3398,
+      "theoretical_loss": 3.338393429595838,
+      "tokens_seen": 3130851328
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.1717220349863586e-05,
+      "loss": 1.322,
+      "theoretical_loss": 3.3383828679136567,
+      "tokens_seen": 3130982400
+    },
+    {
+      "epoch": 0.9,
+      "objective/train/advantage_avg": 0.4953627288341522,
+      "objective/train/docs_used": 1137606,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.1683340072631836,
+      "objective/train/original_loss": 1.168333888053894,
+      "objective/train/theoretical_loss": 3.338372306797403,
+      "objective/train/tokens_used": 1501572576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24559558928012848,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050784945487976,
+      "objective/train/weighted_lm_loss": 1.2276848554611206,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.338372306797403,
+      "tokens_seen": 3131113472
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.1677098379072385e-05,
+      "loss": 1.2893,
+      "theoretical_loss": 3.338372306797403,
+      "tokens_seen": 3131113472
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.163697640828118e-05,
+      "loss": 1.2853,
+      "theoretical_loss": 3.3383617462470223,
+      "tokens_seen": 3131244544
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.159685443748997e-05,
+      "loss": 1.3509,
+      "theoretical_loss": 3.3383511862624617,
+      "tokens_seen": 3131375616
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.1556732466698764e-05,
+      "loss": 1.2329,
+      "theoretical_loss": 3.338340626843666,
+      "tokens_seen": 3131506688
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.151661049590756e-05,
+      "loss": 1.3505,
+      "theoretical_loss": 3.3383300679905816,
+      "tokens_seen": 3131637760
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.1476488525116356e-05,
+      "loss": 1.3207,
+      "theoretical_loss": 3.3383195097031546,
+      "tokens_seen": 3131768832
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.143636655432515e-05,
+      "loss": 1.2888,
+      "theoretical_loss": 3.3383089519813316,
+      "tokens_seen": 3131899904
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.139624458353395e-05,
+      "loss": 1.2455,
+      "theoretical_loss": 3.3382983948250575,
+      "tokens_seen": 3132030976
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.1356122612742734e-05,
+      "loss": 1.3214,
+      "theoretical_loss": 3.3382878382342795,
+      "tokens_seen": 3132162048
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.1316000641951534e-05,
+      "loss": 1.2887,
+      "theoretical_loss": 3.3382772822089426,
+      "tokens_seen": 3132293120
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.1275878671160326e-05,
+      "loss": 1.282,
+      "theoretical_loss": 3.3382667267489934,
+      "tokens_seen": 3132424192
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.1235756700369126e-05,
+      "loss": 1.3259,
+      "theoretical_loss": 3.3382561718543777,
+      "tokens_seen": 3132555264
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.119563472957792e-05,
+      "loss": 1.3029,
+      "theoretical_loss": 3.338245617525042,
+      "tokens_seen": 3132686336
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.115551275878672e-05,
+      "loss": 1.3608,
+      "theoretical_loss": 3.338235063760932,
+      "tokens_seen": 3132817408
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.1115390787995504e-05,
+      "loss": 1.3061,
+      "theoretical_loss": 3.338224510561994,
+      "tokens_seen": 3132948480
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.1075268817204304e-05,
+      "loss": 1.371,
+      "theoretical_loss": 3.338213957928174,
+      "tokens_seen": 3133079552
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.1035146846413096e-05,
+      "loss": 1.2773,
+      "theoretical_loss": 3.338203405859418,
+      "tokens_seen": 3133210624
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.099502487562189e-05,
+      "loss": 1.4331,
+      "theoretical_loss": 3.3381928543556727,
+      "tokens_seen": 3133341696
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.095490290483069e-05,
+      "loss": 1.3399,
+      "theoretical_loss": 3.338182303416883,
+      "tokens_seen": 3133472768
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.091478093403948e-05,
+      "loss": 1.2545,
+      "theoretical_loss": 3.338171753042996,
+      "tokens_seen": 3133603840
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.0874658963248274e-05,
+      "loss": 1.3566,
+      "theoretical_loss": 3.3381612032339576,
+      "tokens_seen": 3133734912
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.083453699245707e-05,
+      "loss": 1.4399,
+      "theoretical_loss": 3.338150653989714,
+      "tokens_seen": 3133865984
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.0794415021665867e-05,
+      "loss": 1.3523,
+      "theoretical_loss": 3.3381401053102113,
+      "tokens_seen": 3133997056
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.075429305087466e-05,
+      "loss": 1.3161,
+      "theoretical_loss": 3.338129557195395,
+      "tokens_seen": 3134128128
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.071417108008346e-05,
+      "loss": 1.3342,
+      "theoretical_loss": 3.3381190096452125,
+      "tokens_seen": 3134259200
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.067404910929225e-05,
+      "loss": 1.4007,
+      "theoretical_loss": 3.3381084626596094,
+      "tokens_seen": 3134390272
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.0633927138501044e-05,
+      "loss": 1.3526,
+      "theoretical_loss": 3.338097916238531,
+      "tokens_seen": 3134521344
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.059380516770984e-05,
+      "loss": 1.2722,
+      "theoretical_loss": 3.338087370381925,
+      "tokens_seen": 3134652416
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.055368319691863e-05,
+      "loss": 1.322,
+      "theoretical_loss": 3.3380768250897366,
+      "tokens_seen": 3134783488
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.051356122612743e-05,
+      "loss": 1.261,
+      "theoretical_loss": 3.3380662803619123,
+      "tokens_seen": 3134914560
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.047343925533622e-05,
+      "loss": 1.3972,
+      "theoretical_loss": 3.338055736198398,
+      "tokens_seen": 3135045632
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.043331728454502e-05,
+      "loss": 1.4298,
+      "theoretical_loss": 3.338045192599141,
+      "tokens_seen": 3135176704
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.039319531375381e-05,
+      "loss": 1.3664,
+      "theoretical_loss": 3.3380346495640856,
+      "tokens_seen": 3135307776
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.035307334296261e-05,
+      "loss": 1.2855,
+      "theoretical_loss": 3.3380241070931795,
+      "tokens_seen": 3135438848
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.03129513721714e-05,
+      "loss": 1.3557,
+      "theoretical_loss": 3.338013565186369,
+      "tokens_seen": 3135569920
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.02728294013802e-05,
+      "loss": 1.2896,
+      "theoretical_loss": 3.338003023843599,
+      "tokens_seen": 3135700992
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.023270743058899e-05,
+      "loss": 1.302,
+      "theoretical_loss": 3.3379924830648173,
+      "tokens_seen": 3135832064
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.019258545979779e-05,
+      "loss": 1.3286,
+      "theoretical_loss": 3.3379819428499693,
+      "tokens_seen": 3135963136
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.015246348900658e-05,
+      "loss": 1.2581,
+      "theoretical_loss": 3.3379714031990013,
+      "tokens_seen": 3136094208
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.011234151821538e-05,
+      "loss": 1.2981,
+      "theoretical_loss": 3.33796086411186,
+      "tokens_seen": 3136225280
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.007221954742417e-05,
+      "loss": 1.3223,
+      "theoretical_loss": 3.337950325588491,
+      "tokens_seen": 3136356352
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.003209757663296e-05,
+      "loss": 1.3163,
+      "theoretical_loss": 3.3379397876288412,
+      "tokens_seen": 3136487424
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.999197560584176e-05,
+      "loss": 1.314,
+      "theoretical_loss": 3.3379292502328566,
+      "tokens_seen": 3136618496
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.9951853635050555e-05,
+      "loss": 1.2754,
+      "theoretical_loss": 3.3379187134004837,
+      "tokens_seen": 3136749568
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.991173166425935e-05,
+      "loss": 1.2224,
+      "theoretical_loss": 3.337908177131668,
+      "tokens_seen": 3136880640
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.987160969346814e-05,
+      "loss": 1.2646,
+      "theoretical_loss": 3.3378976414263573,
+      "tokens_seen": 3137011712
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.983148772267694e-05,
+      "loss": 1.2622,
+      "theoretical_loss": 3.3378871062844966,
+      "tokens_seen": 3137142784
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.979136575188573e-05,
+      "loss": 1.3207,
+      "theoretical_loss": 3.337876571706033,
+      "tokens_seen": 3137273856
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.975124378109453e-05,
+      "loss": 1.2521,
+      "theoretical_loss": 3.337866037690912,
+      "tokens_seen": 3137404928
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.9711121810303325e-05,
+      "loss": 1.3008,
+      "theoretical_loss": 3.337855504239081,
+      "tokens_seen": 3137536000
+    },
+    {
+      "epoch": 0.9,
+      "objective/train/advantage_avg": 0.4934064745903015,
+      "objective/train/docs_used": 1140026,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2878319025039673,
+      "objective/train/original_loss": 1.2878319025039673,
+      "objective/train/theoretical_loss": 3.337844971350486,
+      "objective/train/tokens_used": 1508126176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24387255311012268,
+      "objective/train/value_max": -0.5,
+      "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": 1.3529466390609741,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.337844971350486,
+      "tokens_seen": 3137667072
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.967099983951212e-05,
+      "loss": 1.316,
+      "theoretical_loss": 3.337844971350486,
+      "tokens_seen": 3137667072
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.963087786872091e-05,
+      "loss": 1.3183,
+      "theoretical_loss": 3.3378344390250727,
+      "tokens_seen": 3137798144
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.9590755897929703e-05,
+      "loss": 1.3164,
+      "theoretical_loss": 3.337823907262788,
+      "tokens_seen": 3137929216
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.95506339271385e-05,
+      "loss": 1.3267,
+      "theoretical_loss": 3.3378133760635786,
+      "tokens_seen": 3138060288
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.9510511956347296e-05,
+      "loss": 1.2815,
+      "theoretical_loss": 3.3378028454273903,
+      "tokens_seen": 3138191360
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.9470389985556095e-05,
+      "loss": 1.3606,
+      "theoretical_loss": 3.3377923153541698,
+      "tokens_seen": 3138322432
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.943026801476488e-05,
+      "loss": 1.2929,
+      "theoretical_loss": 3.3377817858438634,
+      "tokens_seen": 3138453504
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.939014604397368e-05,
+      "loss": 1.3452,
+      "theoretical_loss": 3.337771256896417,
+      "tokens_seen": 3138584576
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.9350024073182473e-05,
+      "loss": 1.2954,
+      "theoretical_loss": 3.3377607285117783,
+      "tokens_seen": 3138715648
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.930990210239127e-05,
+      "loss": 1.3276,
+      "theoretical_loss": 3.3377502006898925,
+      "tokens_seen": 3138846720
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.9269780131600066e-05,
+      "loss": 1.3121,
+      "theoretical_loss": 3.3377396734307063,
+      "tokens_seen": 3138977792
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.9229658160808865e-05,
+      "loss": 1.317,
+      "theoretical_loss": 3.3377291467341665,
+      "tokens_seen": 3139108864
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.918953619001765e-05,
+      "loss": 1.291,
+      "theoretical_loss": 3.337718620600219,
+      "tokens_seen": 3139239936
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.914941421922645e-05,
+      "loss": 1.3543,
+      "theoretical_loss": 3.3377080950288107,
+      "tokens_seen": 3139371008
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.9109292248435244e-05,
+      "loss": 1.3513,
+      "theoretical_loss": 3.337697570019888,
+      "tokens_seen": 3139502080
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.9069170277644036e-05,
+      "loss": 1.3307,
+      "theoretical_loss": 3.3376870455733973,
+      "tokens_seen": 3139633152
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.9029048306852836e-05,
+      "loss": 1.3547,
+      "theoretical_loss": 3.3376765216892847,
+      "tokens_seen": 3139764224
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.898892633606163e-05,
+      "loss": 1.3069,
+      "theoretical_loss": 3.3376659983674974,
+      "tokens_seen": 3139895296
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.894880436527042e-05,
+      "loss": 1.3574,
+      "theoretical_loss": 3.3376554756079813,
+      "tokens_seen": 3140026368
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.8908682394479214e-05,
+      "loss": 1.2314,
+      "theoretical_loss": 3.337644953410683,
+      "tokens_seen": 3140157440
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.8868560423688014e-05,
+      "loss": 1.388,
+      "theoretical_loss": 3.337634431775549,
+      "tokens_seen": 3140288512
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.8828438452896806e-05,
+      "loss": 1.25,
+      "theoretical_loss": 3.337623910702526,
+      "tokens_seen": 3140419584
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.8788316482105606e-05,
+      "loss": 1.3032,
+      "theoretical_loss": 3.3376133901915606,
+      "tokens_seen": 3140550656
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.87481945113144e-05,
+      "loss": 1.2577,
+      "theoretical_loss": 3.337602870242599,
+      "tokens_seen": 3140681728
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.870807254052319e-05,
+      "loss": 1.2866,
+      "theoretical_loss": 3.3375923508555876,
+      "tokens_seen": 3140812800
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.8667950569731984e-05,
+      "loss": 1.3838,
+      "theoretical_loss": 3.337581832030473,
+      "tokens_seen": 3140943872
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.862782859894078e-05,
+      "loss": 1.3404,
+      "theoretical_loss": 3.337571313767202,
+      "tokens_seen": 3141074944
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.8587706628149576e-05,
+      "loss": 1.3083,
+      "theoretical_loss": 3.3375607960657216,
+      "tokens_seen": 3141206016
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.854758465735837e-05,
+      "loss": 1.3357,
+      "theoretical_loss": 3.337550278925977,
+      "tokens_seen": 3141337088
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.850746268656717e-05,
+      "loss": 1.351,
+      "theoretical_loss": 3.337539762347916,
+      "tokens_seen": 3141468160
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.8467340715775955e-05,
+      "loss": 1.3192,
+      "theoretical_loss": 3.337529246331485,
+      "tokens_seen": 3141599232
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.8427218744984754e-05,
+      "loss": 1.3903,
+      "theoretical_loss": 3.3375187308766296,
+      "tokens_seen": 3141730304
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.838709677419355e-05,
+      "loss": 1.3097,
+      "theoretical_loss": 3.3375082159832976,
+      "tokens_seen": 3141861376
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.8346974803402347e-05,
+      "loss": 1.3155,
+      "theoretical_loss": 3.337497701651435,
+      "tokens_seen": 3141992448
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.830685283261114e-05,
+      "loss": 1.3093,
+      "theoretical_loss": 3.3374871878809884,
+      "tokens_seen": 3142123520
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.826673086181994e-05,
+      "loss": 1.2787,
+      "theoretical_loss": 3.3374766746719047,
+      "tokens_seen": 3142254592
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.822660889102873e-05,
+      "loss": 1.362,
+      "theoretical_loss": 3.33746616202413,
+      "tokens_seen": 3142385664
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.818648692023752e-05,
+      "loss": 1.2471,
+      "theoretical_loss": 3.337455649937611,
+      "tokens_seen": 3142516736
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.814636494944632e-05,
+      "loss": 1.3289,
+      "theoretical_loss": 3.337445138412295,
+      "tokens_seen": 3142647808
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.810624297865511e-05,
+      "loss": 1.2846,
+      "theoretical_loss": 3.337434627448128,
+      "tokens_seen": 3142778880
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.806612100786391e-05,
+      "loss": 1.2819,
+      "theoretical_loss": 3.337424117045057,
+      "tokens_seen": 3142909952
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.80259990370727e-05,
+      "loss": 1.2874,
+      "theoretical_loss": 3.337413607203028,
+      "tokens_seen": 3143041024
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.79858770662815e-05,
+      "loss": 1.4387,
+      "theoretical_loss": 3.337403097921989,
+      "tokens_seen": 3143172096
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.794575509549029e-05,
+      "loss": 1.3027,
+      "theoretical_loss": 3.337392589201885,
+      "tokens_seen": 3143303168
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.790563312469909e-05,
+      "loss": 1.2937,
+      "theoretical_loss": 3.3373820810426635,
+      "tokens_seen": 3143434240
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.786551115390788e-05,
+      "loss": 1.2592,
+      "theoretical_loss": 3.3373715734442713,
+      "tokens_seen": 3143565312
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.782538918311668e-05,
+      "loss": 1.2812,
+      "theoretical_loss": 3.3373610664066553,
+      "tokens_seen": 3143696384
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.778526721232547e-05,
+      "loss": 1.352,
+      "theoretical_loss": 3.3373505599297615,
+      "tokens_seen": 3143827456
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.774514524153427e-05,
+      "loss": 1.3555,
+      "theoretical_loss": 3.3373400540135365,
+      "tokens_seen": 3143958528
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.770502327074306e-05,
+      "loss": 1.3511,
+      "theoretical_loss": 3.337329548657928,
+      "tokens_seen": 3144089600
+    },
+    {
+      "epoch": 0.91,
+      "objective/train/advantage_avg": 0.49394115805625916,
+      "objective/train/docs_used": 1142279,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2193377017974854,
+      "objective/train/original_loss": 1.2193375825881958,
+      "objective/train/theoretical_loss": 3.337319043862882,
+      "objective/train/tokens_used": 1514679776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2442944496870041,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506361722946167,
+      "objective/train/weighted_lm_loss": 1.2810492515563965,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.337319043862882,
+      "tokens_seen": 3144220672
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.766490129995185e-05,
+      "loss": 1.3121,
+      "theoretical_loss": 3.337319043862882,
+      "tokens_seen": 3144220672
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.762477932916065e-05,
+      "loss": 1.2157,
+      "theoretical_loss": 3.3373085396283453,
+      "tokens_seen": 3144351744
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.758465735836944e-05,
+      "loss": 1.3537,
+      "theoretical_loss": 3.337298035954265,
+      "tokens_seen": 3144482816
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.754453538757824e-05,
+      "loss": 1.2737,
+      "theoretical_loss": 3.337287532840587,
+      "tokens_seen": 3144613888
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.7504413416787035e-05,
+      "loss": 1.3518,
+      "theoretical_loss": 3.337277030287259,
+      "tokens_seen": 3144744960
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.746429144599583e-05,
+      "loss": 1.3342,
+      "theoretical_loss": 3.337266528294227,
+      "tokens_seen": 3144876032
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.742416947520462e-05,
+      "loss": 1.295,
+      "theoretical_loss": 3.3372560268614384,
+      "tokens_seen": 3145007104
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.738404750441342e-05,
+      "loss": 1.3648,
+      "theoretical_loss": 3.3372455259888394,
+      "tokens_seen": 3145138176
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.734392553362221e-05,
+      "loss": 1.306,
+      "theoretical_loss": 3.337235025676377,
+      "tokens_seen": 3145269248
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.730380356283101e-05,
+      "loss": 1.2734,
+      "theoretical_loss": 3.3372245259239985,
+      "tokens_seen": 3145400320
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.7263681592039805e-05,
+      "loss": 1.2586,
+      "theoretical_loss": 3.3372140267316497,
+      "tokens_seen": 3145531392
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.722355962124859e-05,
+      "loss": 1.2915,
+      "theoretical_loss": 3.3372035280992782,
+      "tokens_seen": 3145662464
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.718343765045739e-05,
+      "loss": 1.304,
+      "theoretical_loss": 3.33719303002683,
+      "tokens_seen": 3145793536
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.714331567966618e-05,
+      "loss": 1.3599,
+      "theoretical_loss": 3.337182532514253,
+      "tokens_seen": 3145924608
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.710319370887498e-05,
+      "loss": 1.3101,
+      "theoretical_loss": 3.337172035561493,
+      "tokens_seen": 3146055680
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.7063071738083776e-05,
+      "loss": 1.32,
+      "theoretical_loss": 3.3371615391684974,
+      "tokens_seen": 3146186752
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.7022949767292575e-05,
+      "loss": 1.2887,
+      "theoretical_loss": 3.337151043335213,
+      "tokens_seen": 3146317824
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.698282779650136e-05,
+      "loss": 1.2581,
+      "theoretical_loss": 3.3371405480615866,
+      "tokens_seen": 3146448896
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.694270582571016e-05,
+      "loss": 1.3386,
+      "theoretical_loss": 3.3371300533475647,
+      "tokens_seen": 3146579968
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.6902583854918953e-05,
+      "loss": 1.2685,
+      "theoretical_loss": 3.3371195591930944,
+      "tokens_seen": 3146711040
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.686246188412775e-05,
+      "loss": 1.269,
+      "theoretical_loss": 3.3371090655981224,
+      "tokens_seen": 3146842112
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.6822339913336546e-05,
+      "loss": 1.2735,
+      "theoretical_loss": 3.337098572562596,
+      "tokens_seen": 3146973184
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.678221794254534e-05,
+      "loss": 1.3007,
+      "theoretical_loss": 3.337088080086462,
+      "tokens_seen": 3147104256
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.674209597175413e-05,
+      "loss": 1.3125,
+      "theoretical_loss": 3.337077588169667,
+      "tokens_seen": 3147235328
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.6701974000962924e-05,
+      "loss": 1.3305,
+      "theoretical_loss": 3.3370670968121576,
+      "tokens_seen": 3147366400
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.6661852030171723e-05,
+      "loss": 1.3297,
+      "theoretical_loss": 3.3370566060138818,
+      "tokens_seen": 3147497472
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.6621730059380516e-05,
+      "loss": 1.38,
+      "theoretical_loss": 3.3370461157747853,
+      "tokens_seen": 3147628544
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.6581608088589316e-05,
+      "loss": 1.3043,
+      "theoretical_loss": 3.3370356260948153,
+      "tokens_seen": 3147759616
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.654148611779811e-05,
+      "loss": 1.2683,
+      "theoretical_loss": 3.3370251369739194,
+      "tokens_seen": 3147890688
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.65013641470069e-05,
+      "loss": 1.2934,
+      "theoretical_loss": 3.337014648412044,
+      "tokens_seen": 3148021760
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.6461242176215694e-05,
+      "loss": 1.268,
+      "theoretical_loss": 3.337004160409136,
+      "tokens_seen": 3148152832
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.6421120205424494e-05,
+      "loss": 1.2592,
+      "theoretical_loss": 3.336993672965142,
+      "tokens_seen": 3148283904
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.6380998234633286e-05,
+      "loss": 1.2926,
+      "theoretical_loss": 3.33698318608001,
+      "tokens_seen": 3148414976
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.6340876263842086e-05,
+      "loss": 1.3025,
+      "theoretical_loss": 3.336972699753686,
+      "tokens_seen": 3148546048
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.630075429305088e-05,
+      "loss": 1.2739,
+      "theoretical_loss": 3.336962213986117,
+      "tokens_seen": 3148677120
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.6260632322259665e-05,
+      "loss": 1.2737,
+      "theoretical_loss": 3.336951728777251,
+      "tokens_seen": 3148808192
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.6220510351468464e-05,
+      "loss": 1.3571,
+      "theoretical_loss": 3.3369412441270336,
+      "tokens_seen": 3148939264
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.618038838067726e-05,
+      "loss": 1.3785,
+      "theoretical_loss": 3.3369307600354126,
+      "tokens_seen": 3149070336
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.6140266409886056e-05,
+      "loss": 1.328,
+      "theoretical_loss": 3.336920276502335,
+      "tokens_seen": 3149201408
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.610014443909485e-05,
+      "loss": 1.3496,
+      "theoretical_loss": 3.3369097935277474,
+      "tokens_seen": 3149332480
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.606002246830365e-05,
+      "loss": 1.3841,
+      "theoretical_loss": 3.336899311111597,
+      "tokens_seen": 3149463552
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.6019900497512435e-05,
+      "loss": 1.3227,
+      "theoretical_loss": 3.336888829253831,
+      "tokens_seen": 3149594624
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.5979778526721234e-05,
+      "loss": 1.2853,
+      "theoretical_loss": 3.336878347954396,
+      "tokens_seen": 3149725696
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.593965655593003e-05,
+      "loss": 1.381,
+      "theoretical_loss": 3.3368678672132392,
+      "tokens_seen": 3149856768
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.5899534585138826e-05,
+      "loss": 1.2889,
+      "theoretical_loss": 3.336857387030308,
+      "tokens_seen": 3149987840
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.585941261434762e-05,
+      "loss": 1.2851,
+      "theoretical_loss": 3.3368469074055493,
+      "tokens_seen": 3150118912
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.581929064355641e-05,
+      "loss": 1.3828,
+      "theoretical_loss": 3.3368364283389096,
+      "tokens_seen": 3150249984
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.5779168672765205e-05,
+      "loss": 1.2512,
+      "theoretical_loss": 3.3368259498303363,
+      "tokens_seen": 3150381056
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.5739046701974e-05,
+      "loss": 1.303,
+      "theoretical_loss": 3.336815471879777,
+      "tokens_seen": 3150512128
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.56989247311828e-05,
+      "loss": 1.3927,
+      "theoretical_loss": 3.336804994487178,
+      "tokens_seen": 3150643200
+    },
+    {
+      "epoch": 0.91,
+      "objective/train/advantage_avg": 0.49141839146614075,
+      "objective/train/docs_used": 1144741,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.1860675811767578,
+      "objective/train/original_loss": 1.1860677003860474,
+      "objective/train/theoretical_loss": 3.3367945176524865,
+      "objective/train/tokens_used": 1521233376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24208040535449982,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503724813461304,
+      "objective/train/weighted_lm_loss": 1.245784878730774,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3367945176524865,
+      "tokens_seen": 3150774272
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.565880276039159e-05,
+      "loss": 1.2786,
+      "theoretical_loss": 3.3367945176524865,
+      "tokens_seen": 3150774272
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.561868078960039e-05,
+      "loss": 1.355,
+      "theoretical_loss": 3.33678404137565,
+      "tokens_seen": 3150905344
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.557855881880918e-05,
+      "loss": 1.1983,
+      "theoretical_loss": 3.3367735656566153,
+      "tokens_seen": 3151036416
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.5538436848017975e-05,
+      "loss": 1.3552,
+      "theoretical_loss": 3.3367630904953294,
+      "tokens_seen": 3151167488
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.549831487722677e-05,
+      "loss": 1.3189,
+      "theoretical_loss": 3.3367526158917395,
+      "tokens_seen": 3151298560
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.545819290643557e-05,
+      "loss": 1.2478,
+      "theoretical_loss": 3.336742141845793,
+      "tokens_seen": 3151429632
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.541807093564436e-05,
+      "loss": 1.3858,
+      "theoretical_loss": 3.336731668357437,
+      "tokens_seen": 3151560704
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.537794896485316e-05,
+      "loss": 1.3415,
+      "theoretical_loss": 3.336721195426618,
+      "tokens_seen": 3151691776
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.533782699406195e-05,
+      "loss": 1.3027,
+      "theoretical_loss": 3.3367107230532835,
+      "tokens_seen": 3151822848
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.529770502327074e-05,
+      "loss": 1.2866,
+      "theoretical_loss": 3.3367002512373807,
+      "tokens_seen": 3151953920
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.525758305247954e-05,
+      "loss": 1.3291,
+      "theoretical_loss": 3.336689779978857,
+      "tokens_seen": 3152084992
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.521746108168833e-05,
+      "loss": 1.2695,
+      "theoretical_loss": 3.336679309277659,
+      "tokens_seen": 3152216064
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.517733911089713e-05,
+      "loss": 1.3157,
+      "theoretical_loss": 3.3366688391337345,
+      "tokens_seen": 3152347136
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.513721714010592e-05,
+      "loss": 1.2855,
+      "theoretical_loss": 3.33665836954703,
+      "tokens_seen": 3152478208
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.509709516931472e-05,
+      "loss": 1.2573,
+      "theoretical_loss": 3.336647900517493,
+      "tokens_seen": 3152609280
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.505697319852351e-05,
+      "loss": 1.3846,
+      "theoretical_loss": 3.3366374320450713,
+      "tokens_seen": 3152740352
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.501685122773231e-05,
+      "loss": 1.3077,
+      "theoretical_loss": 3.336626964129711,
+      "tokens_seen": 3152871424
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.49767292569411e-05,
+      "loss": 1.4317,
+      "theoretical_loss": 3.33661649677136,
+      "tokens_seen": 3153002496
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.49366072861499e-05,
+      "loss": 1.2383,
+      "theoretical_loss": 3.336606029969965,
+      "tokens_seen": 3153133568
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.489648531535869e-05,
+      "loss": 1.2966,
+      "theoretical_loss": 3.3365955637254734,
+      "tokens_seen": 3153264640
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.4856363344567486e-05,
+      "loss": 1.313,
+      "theoretical_loss": 3.336585098037833,
+      "tokens_seen": 3153395712
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.481624137377628e-05,
+      "loss": 1.3277,
+      "theoretical_loss": 3.3365746329069905,
+      "tokens_seen": 3153526784
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.477611940298507e-05,
+      "loss": 1.3293,
+      "theoretical_loss": 3.3365641683328926,
+      "tokens_seen": 3153657856
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.473599743219387e-05,
+      "loss": 1.3121,
+      "theoretical_loss": 3.336553704315488,
+      "tokens_seen": 3153788928
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.469587546140266e-05,
+      "loss": 1.3585,
+      "theoretical_loss": 3.3365432408547226,
+      "tokens_seen": 3153920000
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.465575349061146e-05,
+      "loss": 1.2747,
+      "theoretical_loss": 3.336532777950544,
+      "tokens_seen": 3154051072
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.4615631519820256e-05,
+      "loss": 1.3507,
+      "theoretical_loss": 3.3365223156029002,
+      "tokens_seen": 3154182144
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.457550954902905e-05,
+      "loss": 1.2254,
+      "theoretical_loss": 3.3365118538117375,
+      "tokens_seen": 3154313216
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.453538757823784e-05,
+      "loss": 1.2943,
+      "theoretical_loss": 3.3365013925770035,
+      "tokens_seen": 3154444288
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.449526560744664e-05,
+      "loss": 1.2955,
+      "theoretical_loss": 3.3364909318986458,
+      "tokens_seen": 3154575360
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.445514363665543e-05,
+      "loss": 1.3118,
+      "theoretical_loss": 3.3364804717766114,
+      "tokens_seen": 3154706432
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.4415021665864226e-05,
+      "loss": 1.349,
+      "theoretical_loss": 3.3364700122108473,
+      "tokens_seen": 3154837504
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.4374899695073026e-05,
+      "loss": 1.3607,
+      "theoretical_loss": 3.3364595532013017,
+      "tokens_seen": 3154968576
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.433477772428181e-05,
+      "loss": 1.3788,
+      "theoretical_loss": 3.336449094747921,
+      "tokens_seen": 3155099648
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.429465575349061e-05,
+      "loss": 1.2563,
+      "theoretical_loss": 3.3364386368506533,
+      "tokens_seen": 3155230720
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.4254533782699404e-05,
+      "loss": 1.3248,
+      "theoretical_loss": 3.3364281795094453,
+      "tokens_seen": 3155361792
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.4214411811908203e-05,
+      "loss": 1.2854,
+      "theoretical_loss": 3.336417722724244,
+      "tokens_seen": 3155492864
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.4174289841116996e-05,
+      "loss": 1.3894,
+      "theoretical_loss": 3.336407266494998,
+      "tokens_seen": 3155623936
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.4134167870325796e-05,
+      "loss": 1.2683,
+      "theoretical_loss": 3.336396810821654,
+      "tokens_seen": 3155755008
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.409404589953458e-05,
+      "loss": 1.3269,
+      "theoretical_loss": 3.336386355704159,
+      "tokens_seen": 3155886080
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.405392392874338e-05,
+      "loss": 1.2997,
+      "theoretical_loss": 3.336375901142461,
+      "tokens_seen": 3156017152
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.4013801957952174e-05,
+      "loss": 1.2634,
+      "theoretical_loss": 3.3363654471365067,
+      "tokens_seen": 3156148224
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.3973679987160974e-05,
+      "loss": 1.2506,
+      "theoretical_loss": 3.336354993686244,
+      "tokens_seen": 3156279296
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.3933558016369766e-05,
+      "loss": 1.2991,
+      "theoretical_loss": 3.33634454079162,
+      "tokens_seen": 3156410368
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.389343604557856e-05,
+      "loss": 1.2823,
+      "theoretical_loss": 3.3363340884525825,
+      "tokens_seen": 3156541440
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.385331407478736e-05,
+      "loss": 1.2851,
+      "theoretical_loss": 3.3363236366690785,
+      "tokens_seen": 3156672512
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.3813192103996145e-05,
+      "loss": 1.4071,
+      "theoretical_loss": 3.3363131854410555,
+      "tokens_seen": 3156803584
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.3773070133204944e-05,
+      "loss": 1.2484,
+      "theoretical_loss": 3.336302734768461,
+      "tokens_seen": 3156934656
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.373294816241374e-05,
+      "loss": 1.3567,
+      "theoretical_loss": 3.3362922846512424,
+      "tokens_seen": 3157065728
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.3692826191622536e-05,
+      "loss": 1.3542,
+      "theoretical_loss": 3.3362818350893466,
+      "tokens_seen": 3157196800
+    },
+    {
+      "epoch": 0.91,
+      "objective/train/advantage_avg": 0.49196669459342957,
+      "objective/train/docs_used": 1147172,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3383793830871582,
+      "objective/train/original_loss": 1.3383792638778687,
+      "objective/train/theoretical_loss": 3.3362713860827222,
+      "objective/train/tokens_used": 1527786976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24270577728748322,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504305362701416,
+      "objective/train/weighted_lm_loss": 1.4058023691177368,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3362713860827222,
+      "tokens_seen": 3157327872
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.365270422083133e-05,
+      "loss": 1.3739,
+      "theoretical_loss": 3.3362713860827222,
+      "tokens_seen": 3157327872
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.361258225004013e-05,
+      "loss": 1.3529,
+      "theoretical_loss": 3.3362609376313155,
+      "tokens_seen": 3157458944
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.3572460279248915e-05,
+      "loss": 1.32,
+      "theoretical_loss": 3.336250489735075,
+      "tokens_seen": 3157590016
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.3532338308457714e-05,
+      "loss": 1.3365,
+      "theoretical_loss": 3.336240042393947,
+      "tokens_seen": 3157721088
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.349221633766651e-05,
+      "loss": 1.3052,
+      "theoretical_loss": 3.33622959560788,
+      "tokens_seen": 3157852160
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.34520943668753e-05,
+      "loss": 1.2704,
+      "theoretical_loss": 3.3362191493768205,
+      "tokens_seen": 3157983232
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.34119723960841e-05,
+      "loss": 1.3096,
+      "theoretical_loss": 3.336208703700717,
+      "tokens_seen": 3158114304
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.337185042529289e-05,
+      "loss": 1.3784,
+      "theoretical_loss": 3.3361982585795165,
+      "tokens_seen": 3158245376
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.3331728454501685e-05,
+      "loss": 1.27,
+      "theoretical_loss": 3.336187814013166,
+      "tokens_seen": 3158376448
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.329160648371048e-05,
+      "loss": 1.2597,
+      "theoretical_loss": 3.336177370001614,
+      "tokens_seen": 3158507520
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.325148451291928e-05,
+      "loss": 1.3233,
+      "theoretical_loss": 3.3361669265448075,
+      "tokens_seen": 3158638592
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.321136254212807e-05,
+      "loss": 1.3079,
+      "theoretical_loss": 3.336156483642694,
+      "tokens_seen": 3158769664
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.317124057133687e-05,
+      "loss": 1.2742,
+      "theoretical_loss": 3.3361460412952204,
+      "tokens_seen": 3158900736
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.313111860054566e-05,
+      "loss": 1.3022,
+      "theoretical_loss": 3.3361355995023354,
+      "tokens_seen": 3159031808
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.3090996629754455e-05,
+      "loss": 1.3171,
+      "theoretical_loss": 3.336125158263986,
+      "tokens_seen": 3159162880
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.305087465896325e-05,
+      "loss": 1.3091,
+      "theoretical_loss": 3.33611471758012,
+      "tokens_seen": 3159293952
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.301075268817205e-05,
+      "loss": 1.313,
+      "theoretical_loss": 3.3361042774506844,
+      "tokens_seen": 3159425024
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.297063071738084e-05,
+      "loss": 1.3264,
+      "theoretical_loss": 3.3360938378756266,
+      "tokens_seen": 3159556096
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.293050874658963e-05,
+      "loss": 1.2659,
+      "theoretical_loss": 3.3360833988548952,
+      "tokens_seen": 3159687168
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.289038677579843e-05,
+      "loss": 1.3515,
+      "theoretical_loss": 3.3360729603884374,
+      "tokens_seen": 3159818240
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.285026480500722e-05,
+      "loss": 1.3375,
+      "theoretical_loss": 3.3360625224762,
+      "tokens_seen": 3159949312
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.281014283421602e-05,
+      "loss": 1.3297,
+      "theoretical_loss": 3.3360520851181317,
+      "tokens_seen": 3160080384
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.277002086342481e-05,
+      "loss": 1.2632,
+      "theoretical_loss": 3.3360416483141795,
+      "tokens_seen": 3160211456
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.272989889263361e-05,
+      "loss": 1.4334,
+      "theoretical_loss": 3.3360312120642908,
+      "tokens_seen": 3160342528
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.26897769218424e-05,
+      "loss": 1.1993,
+      "theoretical_loss": 3.3360207763684135,
+      "tokens_seen": 3160473600
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.26496549510512e-05,
+      "loss": 1.318,
+      "theoretical_loss": 3.3360103412264954,
+      "tokens_seen": 3160604672
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.260953298025999e-05,
+      "loss": 1.2816,
+      "theoretical_loss": 3.3359999066384836,
+      "tokens_seen": 3160735744
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.256941100946879e-05,
+      "loss": 1.2813,
+      "theoretical_loss": 3.3359894726043264,
+      "tokens_seen": 3160866816
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.252928903867758e-05,
+      "loss": 1.3083,
+      "theoretical_loss": 3.3359790391239708,
+      "tokens_seen": 3160997888
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.248916706788637e-05,
+      "loss": 1.2876,
+      "theoretical_loss": 3.335968606197365,
+      "tokens_seen": 3161128960
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.244904509709517e-05,
+      "loss": 1.2725,
+      "theoretical_loss": 3.3359581738244564,
+      "tokens_seen": 3161260032
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.2408923126303965e-05,
+      "loss": 1.2675,
+      "theoretical_loss": 3.3359477420051924,
+      "tokens_seen": 3161391104
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.236880115551276e-05,
+      "loss": 1.2756,
+      "theoretical_loss": 3.3359373107395207,
+      "tokens_seen": 3161522176
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.232867918472155e-05,
+      "loss": 1.3465,
+      "theoretical_loss": 3.3359268800273894,
+      "tokens_seen": 3161653248
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.228855721393035e-05,
+      "loss": 1.3767,
+      "theoretical_loss": 3.335916449868746,
+      "tokens_seen": 3161784320
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.224843524313914e-05,
+      "loss": 1.3357,
+      "theoretical_loss": 3.3359060202635384,
+      "tokens_seen": 3161915392
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.220831327234794e-05,
+      "loss": 1.376,
+      "theoretical_loss": 3.3358955912117136,
+      "tokens_seen": 3162046464
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.2168191301556736e-05,
+      "loss": 1.3997,
+      "theoretical_loss": 3.33588516271322,
+      "tokens_seen": 3162177536
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.212806933076553e-05,
+      "loss": 1.2832,
+      "theoretical_loss": 3.335874734768005,
+      "tokens_seen": 3162308608
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.208794735997432e-05,
+      "loss": 1.3025,
+      "theoretical_loss": 3.335864307376016,
+      "tokens_seen": 3162439680
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.2047825389183114e-05,
+      "loss": 1.2856,
+      "theoretical_loss": 3.3358538805372016,
+      "tokens_seen": 3162570752
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.200770341839191e-05,
+      "loss": 1.3309,
+      "theoretical_loss": 3.335843454251509,
+      "tokens_seen": 3162701824
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.1967581447600706e-05,
+      "loss": 1.3297,
+      "theoretical_loss": 3.335833028518886,
+      "tokens_seen": 3162832896
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.1927459476809506e-05,
+      "loss": 1.2662,
+      "theoretical_loss": 3.3358226033392797,
+      "tokens_seen": 3162963968
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.188733750601829e-05,
+      "loss": 1.3195,
+      "theoretical_loss": 3.3358121787126387,
+      "tokens_seen": 3163095040
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.184721553522709e-05,
+      "loss": 1.3735,
+      "theoretical_loss": 3.3358017546389105,
+      "tokens_seen": 3163226112
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.1807093564435884e-05,
+      "loss": 1.2349,
+      "theoretical_loss": 3.335791331118043,
+      "tokens_seen": 3163357184
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.1766971593644683e-05,
+      "loss": 1.2811,
+      "theoretical_loss": 3.3357809081499843,
+      "tokens_seen": 3163488256
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.1726849622853476e-05,
+      "loss": 1.3111,
+      "theoretical_loss": 3.335770485734681,
+      "tokens_seen": 3163619328
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.1686727652062276e-05,
+      "loss": 1.3223,
+      "theoretical_loss": 3.335760063872082,
+      "tokens_seen": 3163750400
+    },
+    {
+      "epoch": 0.92,
+      "objective/train/advantage_avg": 0.49249267578125,
+      "objective/train/docs_used": 1149760,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3125483989715576,
+      "objective/train/original_loss": 1.3125483989715576,
+      "objective/train/theoretical_loss": 3.3357496425621345,
+      "objective/train/tokens_used": 1534340576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24339713156223297,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.05048668384552,
+      "objective/train/weighted_lm_loss": 1.3787648677825928,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3357496425621345,
+      "tokens_seen": 3163881472
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.164660568127106e-05,
+      "loss": 1.3459,
+      "theoretical_loss": 3.3357496425621345,
+      "tokens_seen": 3163881472
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.160648371047986e-05,
+      "loss": 1.2648,
+      "theoretical_loss": 3.3357392218047863,
+      "tokens_seen": 3164012544
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.1566361739688654e-05,
+      "loss": 1.3082,
+      "theoretical_loss": 3.335728801599986,
+      "tokens_seen": 3164143616
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.152623976889745e-05,
+      "loss": 1.2527,
+      "theoretical_loss": 3.3357183819476806,
+      "tokens_seen": 3164274688
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.1486117798106246e-05,
+      "loss": 1.2632,
+      "theoretical_loss": 3.3357079628478177,
+      "tokens_seen": 3164405760
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.144599582731504e-05,
+      "loss": 1.3127,
+      "theoretical_loss": 3.335697544300346,
+      "tokens_seen": 3164536832
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.140587385652383e-05,
+      "loss": 1.2978,
+      "theoretical_loss": 3.335687126305213,
+      "tokens_seen": 3164667904
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.1365751885732624e-05,
+      "loss": 1.2485,
+      "theoretical_loss": 3.335676708862366,
+      "tokens_seen": 3164798976
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.1325629914941424e-05,
+      "loss": 1.2763,
+      "theoretical_loss": 3.3356662919717537,
+      "tokens_seen": 3164930048
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.128550794415022e-05,
+      "loss": 1.3285,
+      "theoretical_loss": 3.335655875633323,
+      "tokens_seen": 3165061120
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.1245385973359016e-05,
+      "loss": 1.2638,
+      "theoretical_loss": 3.335645459847023,
+      "tokens_seen": 3165192192
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.120526400256781e-05,
+      "loss": 1.2634,
+      "theoretical_loss": 3.335635044612801,
+      "tokens_seen": 3165323264
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.11651420317766e-05,
+      "loss": 1.2657,
+      "theoretical_loss": 3.3356246299306043,
+      "tokens_seen": 3165454336
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.1125020060985395e-05,
+      "loss": 1.2302,
+      "theoretical_loss": 3.3356142158003816,
+      "tokens_seen": 3165585408
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.108489809019419e-05,
+      "loss": 1.2647,
+      "theoretical_loss": 3.33560380222208,
+      "tokens_seen": 3165716480
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.104477611940299e-05,
+      "loss": 1.295,
+      "theoretical_loss": 3.3355933891956484,
+      "tokens_seen": 3165847552
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.100465414861178e-05,
+      "loss": 1.2766,
+      "theoretical_loss": 3.335582976721034,
+      "tokens_seen": 3165978624
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.096453217782058e-05,
+      "loss": 1.2946,
+      "theoretical_loss": 3.3355725647981846,
+      "tokens_seen": 3166109696
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.0924410207029365e-05,
+      "loss": 1.3556,
+      "theoretical_loss": 3.3355621534270483,
+      "tokens_seen": 3166240768
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.0884288236238165e-05,
+      "loss": 1.327,
+      "theoretical_loss": 3.3355517426075734,
+      "tokens_seen": 3166371840
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.084416626544696e-05,
+      "loss": 1.2775,
+      "theoretical_loss": 3.335541332339708,
+      "tokens_seen": 3166502912
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.080404429465576e-05,
+      "loss": 1.2988,
+      "theoretical_loss": 3.335530922623399,
+      "tokens_seen": 3166633984
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.076392232386455e-05,
+      "loss": 1.3908,
+      "theoretical_loss": 3.335520513458595,
+      "tokens_seen": 3166765056
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.072380035307335e-05,
+      "loss": 1.2841,
+      "theoretical_loss": 3.335510104845244,
+      "tokens_seen": 3166896128
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.0683678382282135e-05,
+      "loss": 1.2758,
+      "theoretical_loss": 3.3354996967832937,
+      "tokens_seen": 3167027200
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.0643556411490935e-05,
+      "loss": 1.2499,
+      "theoretical_loss": 3.335489289272693,
+      "tokens_seen": 3167158272
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.060343444069973e-05,
+      "loss": 1.3366,
+      "theoretical_loss": 3.3354788823133883,
+      "tokens_seen": 3167289344
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.056331246990852e-05,
+      "loss": 1.2847,
+      "theoretical_loss": 3.3354684759053286,
+      "tokens_seen": 3167420416
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.052319049911732e-05,
+      "loss": 1.3733,
+      "theoretical_loss": 3.3354580700484617,
+      "tokens_seen": 3167551488
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.048306852832611e-05,
+      "loss": 1.3201,
+      "theoretical_loss": 3.335447664742736,
+      "tokens_seen": 3167682560
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.0442946557534905e-05,
+      "loss": 1.2699,
+      "theoretical_loss": 3.3354372599880984,
+      "tokens_seen": 3167813632
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.04028245867437e-05,
+      "loss": 1.2886,
+      "theoretical_loss": 3.335426855784498,
+      "tokens_seen": 3167944704
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.03627026159525e-05,
+      "loss": 1.3055,
+      "theoretical_loss": 3.3354164521318825,
+      "tokens_seen": 3168075776
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.032258064516129e-05,
+      "loss": 1.3133,
+      "theoretical_loss": 3.3354060490301998,
+      "tokens_seen": 3168206848
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.028245867437009e-05,
+      "loss": 1.3714,
+      "theoretical_loss": 3.335395646479398,
+      "tokens_seen": 3168337920
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.024233670357888e-05,
+      "loss": 1.3302,
+      "theoretical_loss": 3.3353852444794247,
+      "tokens_seen": 3168468992
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.0202214732787675e-05,
+      "loss": 1.4003,
+      "theoretical_loss": 3.335374843030229,
+      "tokens_seen": 3168600064
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.016209276199647e-05,
+      "loss": 1.3702,
+      "theoretical_loss": 3.335364442131758,
+      "tokens_seen": 3168731136
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.012197079120526e-05,
+      "loss": 1.2814,
+      "theoretical_loss": 3.3353540417839604,
+      "tokens_seen": 3168862208
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.008184882041406e-05,
+      "loss": 1.3697,
+      "theoretical_loss": 3.3353436419867837,
+      "tokens_seen": 3168993280
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.004172684962285e-05,
+      "loss": 1.3713,
+      "theoretical_loss": 3.3353332427401767,
+      "tokens_seen": 3169124352
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.000160487883165e-05,
+      "loss": 1.4164,
+      "theoretical_loss": 3.3353228440440867,
+      "tokens_seen": 3169255424
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.996148290804044e-05,
+      "loss": 1.2992,
+      "theoretical_loss": 3.335312445898462,
+      "tokens_seen": 3169386496
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.992136093724924e-05,
+      "loss": 1.2832,
+      "theoretical_loss": 3.3353020483032507,
+      "tokens_seen": 3169517568
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.988123896645803e-05,
+      "loss": 1.3296,
+      "theoretical_loss": 3.3352916512584017,
+      "tokens_seen": 3169648640
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.984111699566683e-05,
+      "loss": 1.2903,
+      "theoretical_loss": 3.3352812547638617,
+      "tokens_seen": 3169779712
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.980099502487562e-05,
+      "loss": 1.3279,
+      "theoretical_loss": 3.33527085881958,
+      "tokens_seen": 3169910784
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.976087305408442e-05,
+      "loss": 1.4094,
+      "theoretical_loss": 3.335260463425504,
+      "tokens_seen": 3170041856
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.972075108329321e-05,
+      "loss": 1.2709,
+      "theoretical_loss": 3.3352500685815825,
+      "tokens_seen": 3170172928
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.9680629112502e-05,
+      "loss": 1.3224,
+      "theoretical_loss": 3.335239674287763,
+      "tokens_seen": 3170304000
+    },
+    {
+      "epoch": 0.92,
+      "objective/train/advantage_avg": 0.49287331104278564,
+      "objective/train/docs_used": 1152259,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2082154750823975,
+      "objective/train/original_loss": 1.2082154750823975,
+      "objective/train/theoretical_loss": 3.335229280543994,
+      "objective/train/tokens_used": 1540894176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24373306334018707,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505263805389404,
+      "objective/train/weighted_lm_loss": 1.269112467765808,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.8607079386711121,
+      "theoretical_loss": 3.335229280543994,
+      "tokens_seen": 3170435072
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.96405071417108e-05,
+      "loss": 1.2992,
+      "theoretical_loss": 3.335229280543994,
+      "tokens_seen": 3170435072
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.9600385170919594e-05,
+      "loss": 1.4342,
+      "theoretical_loss": 3.335218887350223,
+      "tokens_seen": 3170566144
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.956026320012839e-05,
+      "loss": 1.2315,
+      "theoretical_loss": 3.335208494706399,
+      "tokens_seen": 3170697216
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.9520141229337186e-05,
+      "loss": 1.3219,
+      "theoretical_loss": 3.3351981026124706,
+      "tokens_seen": 3170828288
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.9480019258545986e-05,
+      "loss": 1.3052,
+      "theoretical_loss": 3.3351877110683845,
+      "tokens_seen": 3170959360
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.943989728775477e-05,
+      "loss": 1.3476,
+      "theoretical_loss": 3.3351773200740897,
+      "tokens_seen": 3171090432
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.939977531696357e-05,
+      "loss": 1.2859,
+      "theoretical_loss": 3.3351669296295348,
+      "tokens_seen": 3171221504
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.9359653346172364e-05,
+      "loss": 1.334,
+      "theoretical_loss": 3.335156539734667,
+      "tokens_seen": 3171352576
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.931953137538116e-05,
+      "loss": 1.3206,
+      "theoretical_loss": 3.335146150389435,
+      "tokens_seen": 3171483648
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.9279409404589956e-05,
+      "loss": 1.3393,
+      "theoretical_loss": 3.3351357615937873,
+      "tokens_seen": 3171614720
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.9239287433798756e-05,
+      "loss": 1.3024,
+      "theoretical_loss": 3.335125373347672,
+      "tokens_seen": 3171745792
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.919916546300754e-05,
+      "loss": 1.2924,
+      "theoretical_loss": 3.335114985651037,
+      "tokens_seen": 3171876864
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.9159043492216334e-05,
+      "loss": 1.259,
+      "theoretical_loss": 3.335104598503831,
+      "tokens_seen": 3172007936
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.9118921521425134e-05,
+      "loss": 1.3133,
+      "theoretical_loss": 3.3350942119060014,
+      "tokens_seen": 3172139008
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.907879955063393e-05,
+      "loss": 1.3351,
+      "theoretical_loss": 3.3350838258574975,
+      "tokens_seen": 3172270080
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.9038677579842726e-05,
+      "loss": 1.3186,
+      "theoretical_loss": 3.3350734403582667,
+      "tokens_seen": 3172401152
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.899855560905152e-05,
+      "loss": 1.2535,
+      "theoretical_loss": 3.3350630554082574,
+      "tokens_seen": 3172532224
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.895843363826031e-05,
+      "loss": 1.3109,
+      "theoretical_loss": 3.3350526710074186,
+      "tokens_seen": 3172663296
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.8918311667469104e-05,
+      "loss": 1.4074,
+      "theoretical_loss": 3.335042287155698,
+      "tokens_seen": 3172794368
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.8878189696677904e-05,
+      "loss": 1.2976,
+      "theoretical_loss": 3.335031903853044,
+      "tokens_seen": 3172925440
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.88380677258867e-05,
+      "loss": 1.3052,
+      "theoretical_loss": 3.3350215210994043,
+      "tokens_seen": 3173056512
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.8797945755095496e-05,
+      "loss": 1.2864,
+      "theoretical_loss": 3.335011138894728,
+      "tokens_seen": 3173187584
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.875782378430429e-05,
+      "loss": 1.3021,
+      "theoretical_loss": 3.335000757238963,
+      "tokens_seen": 3173318656
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.8717701813513075e-05,
+      "loss": 1.3256,
+      "theoretical_loss": 3.3349903761320574,
+      "tokens_seen": 3173449728
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.8677579842721875e-05,
+      "loss": 1.3001,
+      "theoretical_loss": 3.3349799955739607,
+      "tokens_seen": 3173580800
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.863745787193067e-05,
+      "loss": 1.3407,
+      "theoretical_loss": 3.33496961556462,
+      "tokens_seen": 3173711872
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.859733590113947e-05,
+      "loss": 1.298,
+      "theoretical_loss": 3.3349592361039835,
+      "tokens_seen": 3173842944
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.855721393034826e-05,
+      "loss": 1.3324,
+      "theoretical_loss": 3.3349488571920003,
+      "tokens_seen": 3173974016
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.851709195955706e-05,
+      "loss": 1.3522,
+      "theoretical_loss": 3.3349384788286187,
+      "tokens_seen": 3174105088
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.8476969988765845e-05,
+      "loss": 1.2747,
+      "theoretical_loss": 3.3349281010137863,
+      "tokens_seen": 3174236160
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.8436848017974645e-05,
+      "loss": 1.3307,
+      "theoretical_loss": 3.3349177237474525,
+      "tokens_seen": 3174367232
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.839672604718344e-05,
+      "loss": 1.2558,
+      "theoretical_loss": 3.334907347029565,
+      "tokens_seen": 3174498304
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.835660407639224e-05,
+      "loss": 1.2761,
+      "theoretical_loss": 3.334896970860072,
+      "tokens_seen": 3174629376
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.831648210560103e-05,
+      "loss": 1.3076,
+      "theoretical_loss": 3.3348865952389226,
+      "tokens_seen": 3174760448
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.827636013480982e-05,
+      "loss": 1.3514,
+      "theoretical_loss": 3.3348762201660644,
+      "tokens_seen": 3174891520
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.8236238164018615e-05,
+      "loss": 1.261,
+      "theoretical_loss": 3.3348658456414464,
+      "tokens_seen": 3175022592
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.819611619322741e-05,
+      "loss": 1.2686,
+      "theoretical_loss": 3.3348554716650165,
+      "tokens_seen": 3175153664
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.815599422243621e-05,
+      "loss": 1.281,
+      "theoretical_loss": 3.3348450982367237,
+      "tokens_seen": 3175284736
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.8115872251645e-05,
+      "loss": 1.3334,
+      "theoretical_loss": 3.334834725356516,
+      "tokens_seen": 3175415808
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.80757502808538e-05,
+      "loss": 1.338,
+      "theoretical_loss": 3.3348243530243415,
+      "tokens_seen": 3175546880
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.803562831006259e-05,
+      "loss": 1.2881,
+      "theoretical_loss": 3.3348139812401496,
+      "tokens_seen": 3175677952
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.7995506339271385e-05,
+      "loss": 1.3706,
+      "theoretical_loss": 3.3348036100038883,
+      "tokens_seen": 3175809024
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.795538436848018e-05,
+      "loss": 1.3446,
+      "theoretical_loss": 3.334793239315505,
+      "tokens_seen": 3175940096
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.791526239768898e-05,
+      "loss": 1.3195,
+      "theoretical_loss": 3.33478286917495,
+      "tokens_seen": 3176071168
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.787514042689777e-05,
+      "loss": 1.3019,
+      "theoretical_loss": 3.3347724995821704,
+      "tokens_seen": 3176202240
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.783501845610657e-05,
+      "loss": 1.2254,
+      "theoretical_loss": 3.334762130537115,
+      "tokens_seen": 3176333312
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.779489648531536e-05,
+      "loss": 1.2904,
+      "theoretical_loss": 3.334751762039732,
+      "tokens_seen": 3176464384
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.775477451452415e-05,
+      "loss": 1.2873,
+      "theoretical_loss": 3.3347413940899706,
+      "tokens_seen": 3176595456
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.771465254373295e-05,
+      "loss": 1.2827,
+      "theoretical_loss": 3.3347310266877788,
+      "tokens_seen": 3176726528
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.767453057294174e-05,
+      "loss": 1.3291,
+      "theoretical_loss": 3.3347206598331054,
+      "tokens_seen": 3176857600
+    },
+    {
+      "epoch": 0.93,
+      "objective/train/advantage_avg": 0.4929877519607544,
+      "objective/train/docs_used": 1154349,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2559075355529785,
+      "objective/train/original_loss": 1.2559075355529785,
+      "objective/train/theoretical_loss": 3.3347102935258985,
+      "objective/train/tokens_used": 1547447776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24340681731700897,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505361557006836,
+      "objective/train/weighted_lm_loss": 1.319364309310913,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3347102935258985,
+      "tokens_seen": 3176988672
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.763440860215054e-05,
+      "loss": 1.3357,
+      "theoretical_loss": 3.3347102935258985,
+      "tokens_seen": 3176988672
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.759428663135933e-05,
+      "loss": 1.3408,
+      "theoretical_loss": 3.3346999277661067,
+      "tokens_seen": 3177119744
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.755416466056813e-05,
+      "loss": 1.2987,
+      "theoretical_loss": 3.3346895625536788,
+      "tokens_seen": 3177250816
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.751404268977692e-05,
+      "loss": 1.3252,
+      "theoretical_loss": 3.334679197888563,
+      "tokens_seen": 3177381888
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.747392071898572e-05,
+      "loss": 1.3666,
+      "theoretical_loss": 3.3346688337707078,
+      "tokens_seen": 3177512960
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.743379874819451e-05,
+      "loss": 1.3162,
+      "theoretical_loss": 3.334658470200062,
+      "tokens_seen": 3177644032
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.739367677740331e-05,
+      "loss": 1.2942,
+      "theoretical_loss": 3.334648107176574,
+      "tokens_seen": 3177775104
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.73535548066121e-05,
+      "loss": 1.3392,
+      "theoretical_loss": 3.3346377447001925,
+      "tokens_seen": 3177906176
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.7313432835820896e-05,
+      "loss": 1.3318,
+      "theoretical_loss": 3.334627382770866,
+      "tokens_seen": 3178037248
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.727331086502969e-05,
+      "loss": 1.304,
+      "theoretical_loss": 3.334617021388542,
+      "tokens_seen": 3178168320
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.723318889423848e-05,
+      "loss": 1.3014,
+      "theoretical_loss": 3.3346066605531712,
+      "tokens_seen": 3178299392
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.719306692344728e-05,
+      "loss": 1.2949,
+      "theoretical_loss": 3.3345963002647006,
+      "tokens_seen": 3178430464
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.7152944952656074e-05,
+      "loss": 1.2786,
+      "theoretical_loss": 3.334585940523079,
+      "tokens_seen": 3178561536
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.711282298186487e-05,
+      "loss": 1.2795,
+      "theoretical_loss": 3.3345755813282554,
+      "tokens_seen": 3178692608
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.7072701011073666e-05,
+      "loss": 1.3372,
+      "theoretical_loss": 3.334565222680178,
+      "tokens_seen": 3178823680
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.703257904028246e-05,
+      "loss": 1.3354,
+      "theoretical_loss": 3.334554864578796,
+      "tokens_seen": 3178954752
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.699245706949125e-05,
+      "loss": 1.2745,
+      "theoretical_loss": 3.3345445070240567,
+      "tokens_seen": 3179085824
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.695233509870005e-05,
+      "loss": 1.3235,
+      "theoretical_loss": 3.3345341500159105,
+      "tokens_seen": 3179216896
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.6912213127908844e-05,
+      "loss": 1.2899,
+      "theoretical_loss": 3.3345237935543044,
+      "tokens_seen": 3179347968
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.687209115711764e-05,
+      "loss": 1.3863,
+      "theoretical_loss": 3.3345134376391883,
+      "tokens_seen": 3179479040
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.6831969186326436e-05,
+      "loss": 1.281,
+      "theoretical_loss": 3.3345030822705097,
+      "tokens_seen": 3179610112
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.679184721553522e-05,
+      "loss": 1.3512,
+      "theoretical_loss": 3.334492727448218,
+      "tokens_seen": 3179741184
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.675172524474402e-05,
+      "loss": 1.3647,
+      "theoretical_loss": 3.334482373172262,
+      "tokens_seen": 3179872256
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.6711603273952814e-05,
+      "loss": 1.3458,
+      "theoretical_loss": 3.33447201944259,
+      "tokens_seen": 3180003328
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.6671481303161614e-05,
+      "loss": 1.3263,
+      "theoretical_loss": 3.3344616662591506,
+      "tokens_seen": 3180134400
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.6631359332370407e-05,
+      "loss": 1.3988,
+      "theoretical_loss": 3.3344513136218925,
+      "tokens_seen": 3180265472
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.6591237361579206e-05,
+      "loss": 1.3295,
+      "theoretical_loss": 3.334440961530764,
+      "tokens_seen": 3180396544
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.655111539078799e-05,
+      "loss": 1.3035,
+      "theoretical_loss": 3.334430609985715,
+      "tokens_seen": 3180527616
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.651099341999679e-05,
+      "loss": 1.3326,
+      "theoretical_loss": 3.334420258986693,
+      "tokens_seen": 3180658688
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.6470871449205584e-05,
+      "loss": 1.3204,
+      "theoretical_loss": 3.3344099085336474,
+      "tokens_seen": 3180789760
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.6430749478414384e-05,
+      "loss": 1.2129,
+      "theoretical_loss": 3.3343995586265263,
+      "tokens_seen": 3180920832
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.639062750762318e-05,
+      "loss": 1.2606,
+      "theoretical_loss": 3.3343892092652787,
+      "tokens_seen": 3181051904
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.635050553683197e-05,
+      "loss": 1.34,
+      "theoretical_loss": 3.3343788604498537,
+      "tokens_seen": 3181182976
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.631038356604076e-05,
+      "loss": 1.3167,
+      "theoretical_loss": 3.334368512180199,
+      "tokens_seen": 3181314048
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.6270261595249555e-05,
+      "loss": 1.3285,
+      "theoretical_loss": 3.3343581644562645,
+      "tokens_seen": 3181445120
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.6230139624458354e-05,
+      "loss": 1.2971,
+      "theoretical_loss": 3.3343478172779983,
+      "tokens_seen": 3181576192
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.619001765366715e-05,
+      "loss": 1.3473,
+      "theoretical_loss": 3.33433747064535,
+      "tokens_seen": 3181707264
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.614989568287595e-05,
+      "loss": 1.2933,
+      "theoretical_loss": 3.334327124558267,
+      "tokens_seen": 3181838336
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.610977371208474e-05,
+      "loss": 1.2894,
+      "theoretical_loss": 3.3343167790166985,
+      "tokens_seen": 3181969408
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.606965174129353e-05,
+      "loss": 1.3229,
+      "theoretical_loss": 3.3343064340205935,
+      "tokens_seen": 3182100480
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.6029529770502325e-05,
+      "loss": 1.3417,
+      "theoretical_loss": 3.334296089569901,
+      "tokens_seen": 3182231552
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.5989407799711125e-05,
+      "loss": 1.2572,
+      "theoretical_loss": 3.3342857456645696,
+      "tokens_seen": 3182362624
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.594928582891992e-05,
+      "loss": 1.3394,
+      "theoretical_loss": 3.3342754023045478,
+      "tokens_seen": 3182493696
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.590916385812871e-05,
+      "loss": 1.3893,
+      "theoretical_loss": 3.334265059489785,
+      "tokens_seen": 3182624768
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.586904188733751e-05,
+      "loss": 1.3554,
+      "theoretical_loss": 3.3342547172202295,
+      "tokens_seen": 3182755840
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.5828919916546296e-05,
+      "loss": 1.3332,
+      "theoretical_loss": 3.33424437549583,
+      "tokens_seen": 3182886912
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.5788797945755095e-05,
+      "loss": 1.2834,
+      "theoretical_loss": 3.3342340343165358,
+      "tokens_seen": 3183017984
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.574867597496389e-05,
+      "loss": 1.3309,
+      "theoretical_loss": 3.3342236936822953,
+      "tokens_seen": 3183149056
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.570855400417269e-05,
+      "loss": 1.2817,
+      "theoretical_loss": 3.334213353593057,
+      "tokens_seen": 3183280128
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.566843203338148e-05,
+      "loss": 1.3524,
+      "theoretical_loss": 3.334203014048771,
+      "tokens_seen": 3183411200
+    },
+    {
+      "epoch": 0.93,
+      "objective/train/advantage_avg": 0.49445295333862305,
+      "objective/train/docs_used": 1156704,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2962783575057983,
+      "objective/train/original_loss": 1.296278476715088,
+      "objective/train/theoretical_loss": 3.334192675049385,
+      "objective/train/tokens_used": 1554001376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24486693739891052,
+      "objective/train/value_max": -0.5,
+      "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": 1.3619087934494019,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.334192675049385,
+      "tokens_seen": 3183542272
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.562831006259028e-05,
+      "loss": 1.3697,
+      "theoretical_loss": 3.334192675049385,
+      "tokens_seen": 3183542272
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.5588188091799066e-05,
+      "loss": 1.3066,
+      "theoretical_loss": 3.3341823365948486,
+      "tokens_seen": 3183673344
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.5548066121007865e-05,
+      "loss": 1.3201,
+      "theoretical_loss": 3.33417199868511,
+      "tokens_seen": 3183804416
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.550794415021666e-05,
+      "loss": 1.2648,
+      "theoretical_loss": 3.3341616613201186,
+      "tokens_seen": 3183935488
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.546782217942546e-05,
+      "loss": 1.2794,
+      "theoretical_loss": 3.334151324499823,
+      "tokens_seen": 3184066560
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.542770020863425e-05,
+      "loss": 1.2686,
+      "theoretical_loss": 3.3341409882241715,
+      "tokens_seen": 3184197632
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.538757823784304e-05,
+      "loss": 1.3532,
+      "theoretical_loss": 3.3341306524931142,
+      "tokens_seen": 3184328704
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.5347456267051836e-05,
+      "loss": 1.3647,
+      "theoretical_loss": 3.334120317306599,
+      "tokens_seen": 3184459776
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.530733429626063e-05,
+      "loss": 1.3153,
+      "theoretical_loss": 3.3341099826645757,
+      "tokens_seen": 3184590848
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.526721232546943e-05,
+      "loss": 1.3209,
+      "theoretical_loss": 3.3340996485669923,
+      "tokens_seen": 3184721920
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.522709035467822e-05,
+      "loss": 1.2939,
+      "theoretical_loss": 3.334089315013798,
+      "tokens_seen": 3184852992
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.518696838388702e-05,
+      "loss": 1.3008,
+      "theoretical_loss": 3.334078982004942,
+      "tokens_seen": 3184984064
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.514684641309581e-05,
+      "loss": 1.3004,
+      "theoretical_loss": 3.334068649540373,
+      "tokens_seen": 3185115136
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.510672444230461e-05,
+      "loss": 1.341,
+      "theoretical_loss": 3.3340583176200402,
+      "tokens_seen": 3185246208
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.50666024715134e-05,
+      "loss": 1.3234,
+      "theoretical_loss": 3.334047986243892,
+      "tokens_seen": 3185377280
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.50264805007222e-05,
+      "loss": 1.3755,
+      "theoretical_loss": 3.3340376554118776,
+      "tokens_seen": 3185508352
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.498635852993099e-05,
+      "loss": 1.3434,
+      "theoretical_loss": 3.3340273251239463,
+      "tokens_seen": 3185639424
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.4946236559139784e-05,
+      "loss": 1.3222,
+      "theoretical_loss": 3.3340169953800465,
+      "tokens_seen": 3185770496
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.490611458834858e-05,
+      "loss": 1.3137,
+      "theoretical_loss": 3.3340066661801275,
+      "tokens_seen": 3185901568
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.486599261755737e-05,
+      "loss": 1.3708,
+      "theoretical_loss": 3.3339963375241384,
+      "tokens_seen": 3186032640
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.482587064676617e-05,
+      "loss": 1.3026,
+      "theoretical_loss": 3.333986009412028,
+      "tokens_seen": 3186163712
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.478574867597496e-05,
+      "loss": 1.3452,
+      "theoretical_loss": 3.333975681843745,
+      "tokens_seen": 3186294784
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.474562670518376e-05,
+      "loss": 1.3703,
+      "theoretical_loss": 3.3339653548192394,
+      "tokens_seen": 3186425856
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.4705504734392554e-05,
+      "loss": 1.351,
+      "theoretical_loss": 3.3339550283384587,
+      "tokens_seen": 3186556928
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.466538276360135e-05,
+      "loss": 1.2455,
+      "theoretical_loss": 3.3339447024013533,
+      "tokens_seen": 3186688000
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.4625260792810146e-05,
+      "loss": 1.3387,
+      "theoretical_loss": 3.3339343770078713,
+      "tokens_seen": 3186819072
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.458513882201894e-05,
+      "loss": 1.3451,
+      "theoretical_loss": 3.333924052157962,
+      "tokens_seen": 3186950144
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.454501685122773e-05,
+      "loss": 1.315,
+      "theoretical_loss": 3.3339137278515745,
+      "tokens_seen": 3187081216
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.450489488043653e-05,
+      "loss": 1.3303,
+      "theoretical_loss": 3.333903404088658,
+      "tokens_seen": 3187212288
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.4464772909645324e-05,
+      "loss": 1.3631,
+      "theoretical_loss": 3.333893080869161,
+      "tokens_seen": 3187343360
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.4424650938854116e-05,
+      "loss": 1.2931,
+      "theoretical_loss": 3.333882758193033,
+      "tokens_seen": 3187474432
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.4384528968062916e-05,
+      "loss": 1.3502,
+      "theoretical_loss": 3.333872436060223,
+      "tokens_seen": 3187605504
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.43444069972717e-05,
+      "loss": 1.3773,
+      "theoretical_loss": 3.3338621144706804,
+      "tokens_seen": 3187736576
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.43042850264805e-05,
+      "loss": 1.3481,
+      "theoretical_loss": 3.3338517934243534,
+      "tokens_seen": 3187867648
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.4264163055689294e-05,
+      "loss": 1.3614,
+      "theoretical_loss": 3.3338414729211916,
+      "tokens_seen": 3187998720
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.4224041084898094e-05,
+      "loss": 1.3483,
+      "theoretical_loss": 3.333831152961144,
+      "tokens_seen": 3188129792
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.4183919114106887e-05,
+      "loss": 1.3051,
+      "theoretical_loss": 3.33382083354416,
+      "tokens_seen": 3188260864
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.4143797143315686e-05,
+      "loss": 1.3338,
+      "theoretical_loss": 3.3338105146701884,
+      "tokens_seen": 3188391936
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.410367517252447e-05,
+      "loss": 1.3211,
+      "theoretical_loss": 3.3338001963391783,
+      "tokens_seen": 3188523008
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.406355320173327e-05,
+      "loss": 1.2639,
+      "theoretical_loss": 3.333789878551079,
+      "tokens_seen": 3188654080
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.4023431230942064e-05,
+      "loss": 1.3194,
+      "theoretical_loss": 3.3337795613058394,
+      "tokens_seen": 3188785152
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.398330926015086e-05,
+      "loss": 1.263,
+      "theoretical_loss": 3.333769244603409,
+      "tokens_seen": 3188916224
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.394318728935966e-05,
+      "loss": 1.257,
+      "theoretical_loss": 3.333758928443736,
+      "tokens_seen": 3189047296
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.390306531856845e-05,
+      "loss": 1.2833,
+      "theoretical_loss": 3.3337486128267706,
+      "tokens_seen": 3189178368
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.386294334777724e-05,
+      "loss": 1.3122,
+      "theoretical_loss": 3.333738297752461,
+      "tokens_seen": 3189309440
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.3822821376986035e-05,
+      "loss": 1.267,
+      "theoretical_loss": 3.333727983220758,
+      "tokens_seen": 3189440512
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.3782699406194834e-05,
+      "loss": 1.3789,
+      "theoretical_loss": 3.3337176692316084,
+      "tokens_seen": 3189571584
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.374257743540363e-05,
+      "loss": 1.3313,
+      "theoretical_loss": 3.3337073557849632,
+      "tokens_seen": 3189702656
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.370245546461243e-05,
+      "loss": 1.3007,
+      "theoretical_loss": 3.333697042880771,
+      "tokens_seen": 3189833728
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.366233349382122e-05,
+      "loss": 1.3849,
+      "theoretical_loss": 3.3336867305189806,
+      "tokens_seen": 3189964800
+    },
+    {
+      "epoch": 0.93,
+      "objective/train/advantage_avg": 0.4909904897212982,
+      "objective/train/docs_used": 1159129,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2911938428878784,
+      "objective/train/original_loss": 1.2911938428878784,
+      "objective/train/theoretical_loss": 3.3336764186995422,
+      "objective/train/tokens_used": 1560554976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.241747185587883,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050328016281128,
+      "objective/train/weighted_lm_loss": 1.356105923652649,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3336764186995422,
+      "tokens_seen": 3190095872
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.362221152303001e-05,
+      "loss": 1.3534,
+      "theoretical_loss": 3.3336764186995422,
+      "tokens_seen": 3190095872
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.3582089552238805e-05,
+      "loss": 1.2772,
+      "theoretical_loss": 3.333666107422404,
+      "tokens_seen": 3190226944
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.35419675814476e-05,
+      "loss": 1.3333,
+      "theoretical_loss": 3.3336557966875153,
+      "tokens_seen": 3190358016
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.35018456106564e-05,
+      "loss": 1.3468,
+      "theoretical_loss": 3.333645486494826,
+      "tokens_seen": 3190489088
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.346172363986519e-05,
+      "loss": 1.2674,
+      "theoretical_loss": 3.3336351768442847,
+      "tokens_seen": 3190620160
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.342160166907399e-05,
+      "loss": 1.3248,
+      "theoretical_loss": 3.3336248677358404,
+      "tokens_seen": 3190751232
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.3381479698282775e-05,
+      "loss": 1.2415,
+      "theoretical_loss": 3.333614559169443,
+      "tokens_seen": 3190882304
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.3341357727491575e-05,
+      "loss": 1.3127,
+      "theoretical_loss": 3.333604251145042,
+      "tokens_seen": 3191013376
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.330123575670037e-05,
+      "loss": 1.3598,
+      "theoretical_loss": 3.3335939436625854,
+      "tokens_seen": 3191144448
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.326111378590917e-05,
+      "loss": 1.3013,
+      "theoretical_loss": 3.3335836367220235,
+      "tokens_seen": 3191275520
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.322099181511796e-05,
+      "loss": 1.3354,
+      "theoretical_loss": 3.333573330323305,
+      "tokens_seen": 3191406592
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.318086984432676e-05,
+      "loss": 1.3187,
+      "theoretical_loss": 3.3335630244663794,
+      "tokens_seen": 3191537664
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.3140747873535546e-05,
+      "loss": 1.2806,
+      "theoretical_loss": 3.333552719151196,
+      "tokens_seen": 3191668736
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.3100625902744345e-05,
+      "loss": 1.2987,
+      "theoretical_loss": 3.3335424143777037,
+      "tokens_seen": 3191799808
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.306050393195314e-05,
+      "loss": 1.3908,
+      "theoretical_loss": 3.3335321101458524,
+      "tokens_seen": 3191930880
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.302038196116193e-05,
+      "loss": 1.2543,
+      "theoretical_loss": 3.3335218064555914,
+      "tokens_seen": 3192061952
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.298025999037073e-05,
+      "loss": 1.2574,
+      "theoretical_loss": 3.333511503306869,
+      "tokens_seen": 3192193024
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.294013801957952e-05,
+      "loss": 1.3178,
+      "theoretical_loss": 3.3335012006996356,
+      "tokens_seen": 3192324096
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.2900016048788316e-05,
+      "loss": 1.3009,
+      "theoretical_loss": 3.33349089863384,
+      "tokens_seen": 3192455168
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 3.285989407799711e-05,
+      "loss": 1.2887,
+      "theoretical_loss": 3.333480597109432,
+      "tokens_seen": 3192586240
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.281977210720591e-05,
+      "loss": 1.3598,
+      "theoretical_loss": 3.33347029612636,
+      "tokens_seen": 3192717312
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.27796501364147e-05,
+      "loss": 1.3399,
+      "theoretical_loss": 3.3334599956845743,
+      "tokens_seen": 3192848384
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.27395281656235e-05,
+      "loss": 1.3162,
+      "theoretical_loss": 3.333449695784023,
+      "tokens_seen": 3192979456
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.269940619483229e-05,
+      "loss": 1.2608,
+      "theoretical_loss": 3.3334393964246574,
+      "tokens_seen": 3193110528
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.2659284224041086e-05,
+      "loss": 1.2824,
+      "theoretical_loss": 3.333429097606425,
+      "tokens_seen": 3193241600
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.261916225324988e-05,
+      "loss": 1.3035,
+      "theoretical_loss": 3.333418799329276,
+      "tokens_seen": 3193372672
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.257904028245867e-05,
+      "loss": 1.3701,
+      "theoretical_loss": 3.3334085015931594,
+      "tokens_seen": 3193503744
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.253891831166747e-05,
+      "loss": 1.3661,
+      "theoretical_loss": 3.3333982043980255,
+      "tokens_seen": 3193634816
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.2498796340876264e-05,
+      "loss": 1.293,
+      "theoretical_loss": 3.3333879077438224,
+      "tokens_seen": 3193765888
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.245867437008506e-05,
+      "loss": 1.3094,
+      "theoretical_loss": 3.3333776116305,
+      "tokens_seen": 3193896960
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.241855239929385e-05,
+      "loss": 1.2704,
+      "theoretical_loss": 3.333367316058008,
+      "tokens_seen": 3194028032
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.237843042850265e-05,
+      "loss": 1.2959,
+      "theoretical_loss": 3.3333570210262957,
+      "tokens_seen": 3194159104
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.233830845771144e-05,
+      "loss": 1.2686,
+      "theoretical_loss": 3.333346726535312,
+      "tokens_seen": 3194290176
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.229818648692024e-05,
+      "loss": 1.2956,
+      "theoretical_loss": 3.333336432585007,
+      "tokens_seen": 3194421248
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.2258064516129034e-05,
+      "loss": 1.3043,
+      "theoretical_loss": 3.3333261391753295,
+      "tokens_seen": 3194552320
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.221794254533783e-05,
+      "loss": 1.4272,
+      "theoretical_loss": 3.3333158463062293,
+      "tokens_seen": 3194683392
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.217782057454662e-05,
+      "loss": 1.3356,
+      "theoretical_loss": 3.333305553977656,
+      "tokens_seen": 3194814464
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.213769860375542e-05,
+      "loss": 1.2564,
+      "theoretical_loss": 3.3332952621895586,
+      "tokens_seen": 3194945536
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.209757663296421e-05,
+      "loss": 1.3436,
+      "theoretical_loss": 3.333284970941887,
+      "tokens_seen": 3195076608
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.2057454662173004e-05,
+      "loss": 1.3089,
+      "theoretical_loss": 3.33327468023459,
+      "tokens_seen": 3195207680
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.2017332691381804e-05,
+      "loss": 1.2935,
+      "theoretical_loss": 3.3332643900676175,
+      "tokens_seen": 3195338752
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.1977210720590596e-05,
+      "loss": 1.2705,
+      "theoretical_loss": 3.3332541004409193,
+      "tokens_seen": 3195469824
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.193708874979939e-05,
+      "loss": 1.3117,
+      "theoretical_loss": 3.333243811354444,
+      "tokens_seen": 3195600896
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.189696677900818e-05,
+      "loss": 1.3208,
+      "theoretical_loss": 3.333233522808142,
+      "tokens_seen": 3195731968
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.185684480821698e-05,
+      "loss": 1.3582,
+      "theoretical_loss": 3.333223234801962,
+      "tokens_seen": 3195863040
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.1816722837425774e-05,
+      "loss": 1.2787,
+      "theoretical_loss": 3.3332129473358543,
+      "tokens_seen": 3195994112
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.1776600866634574e-05,
+      "loss": 1.2824,
+      "theoretical_loss": 3.3332026604097678,
+      "tokens_seen": 3196125184
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.1736478895843366e-05,
+      "loss": 1.3142,
+      "theoretical_loss": 3.333192374023652,
+      "tokens_seen": 3196256256
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.169635692505216e-05,
+      "loss": 1.353,
+      "theoretical_loss": 3.3331820881774568,
+      "tokens_seen": 3196387328
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.165623495426095e-05,
+      "loss": 1.3147,
+      "theoretical_loss": 3.3331718028711315,
+      "tokens_seen": 3196518400
+    },
+    {
+      "epoch": 0.94,
+      "objective/train/advantage_avg": 0.49434036016464233,
+      "objective/train/docs_used": 1161767,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.216107726097107,
+      "objective/train/original_loss": 1.2161078453063965,
+      "objective/train/theoretical_loss": 3.3331615181046255,
+      "objective/train/tokens_used": 1567108576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24468639492988586,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506778955459595,
+      "objective/train/weighted_lm_loss": 1.2777843475341797,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3331615181046255,
+      "tokens_seen": 3196649472
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.1616112983469745e-05,
+      "loss": 1.2523,
+      "theoretical_loss": 3.3331615181046255,
+      "tokens_seen": 3196649472
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.1575991012678544e-05,
+      "loss": 1.2273,
+      "theoretical_loss": 3.3331512338778886,
+      "tokens_seen": 3196780544
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.153586904188734e-05,
+      "loss": 1.2905,
+      "theoretical_loss": 3.33314095019087,
+      "tokens_seen": 3196911616
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.1495747071096137e-05,
+      "loss": 1.2382,
+      "theoretical_loss": 3.3331306670435197,
+      "tokens_seen": 3197042688
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.145562510030492e-05,
+      "loss": 1.309,
+      "theoretical_loss": 3.3331203844357873,
+      "tokens_seen": 3197173760
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.141550312951372e-05,
+      "loss": 1.3147,
+      "theoretical_loss": 3.3331101023676215,
+      "tokens_seen": 3197304832
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.1375381158722515e-05,
+      "loss": 1.255,
+      "theoretical_loss": 3.3330998208389726,
+      "tokens_seen": 3197435904
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.1335259187931314e-05,
+      "loss": 1.2869,
+      "theoretical_loss": 3.3330895398497904,
+      "tokens_seen": 3197566976
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.129513721714011e-05,
+      "loss": 1.315,
+      "theoretical_loss": 3.333079259400024,
+      "tokens_seen": 3197698048
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.125501524634891e-05,
+      "loss": 1.2982,
+      "theoretical_loss": 3.3330689794896227,
+      "tokens_seen": 3197829120
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.121489327555769e-05,
+      "loss": 1.3113,
+      "theoretical_loss": 3.3330587001185368,
+      "tokens_seen": 3197960192
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.117477130476649e-05,
+      "loss": 1.4002,
+      "theoretical_loss": 3.3330484212867155,
+      "tokens_seen": 3198091264
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.1134649333975285e-05,
+      "loss": 1.2643,
+      "theoretical_loss": 3.3330381429941083,
+      "tokens_seen": 3198222336
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.109452736318408e-05,
+      "loss": 1.3016,
+      "theoretical_loss": 3.3330278652406653,
+      "tokens_seen": 3198353408
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.105440539239288e-05,
+      "loss": 1.3055,
+      "theoretical_loss": 3.333017588026336,
+      "tokens_seen": 3198484480
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.101428342160167e-05,
+      "loss": 1.3548,
+      "theoretical_loss": 3.3330073113510696,
+      "tokens_seen": 3198615552
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.097416145081046e-05,
+      "loss": 1.2703,
+      "theoretical_loss": 3.3329970352148166,
+      "tokens_seen": 3198746624
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.0934039480019255e-05,
+      "loss": 1.326,
+      "theoretical_loss": 3.3329867596175253,
+      "tokens_seen": 3198877696
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.0893917509228055e-05,
+      "loss": 1.2659,
+      "theoretical_loss": 3.3329764845591465,
+      "tokens_seen": 3199008768
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.085379553843685e-05,
+      "loss": 1.2976,
+      "theoretical_loss": 3.3329662100396296,
+      "tokens_seen": 3199139840
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.081367356764564e-05,
+      "loss": 1.2873,
+      "theoretical_loss": 3.332955936058924,
+      "tokens_seen": 3199270912
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.077355159685444e-05,
+      "loss": 1.3348,
+      "theoretical_loss": 3.332945662616979,
+      "tokens_seen": 3199401984
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.073342962606323e-05,
+      "loss": 1.2454,
+      "theoretical_loss": 3.3329353897137457,
+      "tokens_seen": 3199533056
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.0693307655272026e-05,
+      "loss": 1.2739,
+      "theoretical_loss": 3.332925117349172,
+      "tokens_seen": 3199664128
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.0653185684480825e-05,
+      "loss": 1.3466,
+      "theoretical_loss": 3.332914845523209,
+      "tokens_seen": 3199795200
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.061306371368962e-05,
+      "loss": 1.2942,
+      "theoretical_loss": 3.3329045742358057,
+      "tokens_seen": 3199926272
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.057294174289841e-05,
+      "loss": 1.2553,
+      "theoretical_loss": 3.332894303486912,
+      "tokens_seen": 3200057344
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.053281977210721e-05,
+      "loss": 1.3593,
+      "theoretical_loss": 3.3328840332764775,
+      "tokens_seen": 3200188416
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.0492697801316003e-05,
+      "loss": 1.3287,
+      "theoretical_loss": 3.3328737636044523,
+      "tokens_seen": 3200319488
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.0452575830524796e-05,
+      "loss": 1.3238,
+      "theoretical_loss": 3.3328634944707853,
+      "tokens_seen": 3200450560
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.041245385973359e-05,
+      "loss": 1.3368,
+      "theoretical_loss": 3.3328532258754273,
+      "tokens_seen": 3200581632
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.0372331888942384e-05,
+      "loss": 1.2495,
+      "theoretical_loss": 3.332842957818327,
+      "tokens_seen": 3200712704
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.033220991815118e-05,
+      "loss": 1.3313,
+      "theoretical_loss": 3.332832690299435,
+      "tokens_seen": 3200843776
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.0292087947359973e-05,
+      "loss": 1.2791,
+      "theoretical_loss": 3.3328224233187007,
+      "tokens_seen": 3200974848
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.025196597656877e-05,
+      "loss": 1.364,
+      "theoretical_loss": 3.3328121568760736,
+      "tokens_seen": 3201105920
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.0211844005777566e-05,
+      "loss": 1.2857,
+      "theoretical_loss": 3.3328018909715036,
+      "tokens_seen": 3201236992
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.017172203498636e-05,
+      "loss": 1.2929,
+      "theoretical_loss": 3.332791625604941,
+      "tokens_seen": 3201368064
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.0131600064195155e-05,
+      "loss": 1.3483,
+      "theoretical_loss": 3.332781360776335,
+      "tokens_seen": 3201499136
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.009147809340395e-05,
+      "loss": 1.2964,
+      "theoretical_loss": 3.332771096485635,
+      "tokens_seen": 3201630208
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.0051356122612743e-05,
+      "loss": 1.3246,
+      "theoretical_loss": 3.332760832732792,
+      "tokens_seen": 3201761280
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 3.001123415182154e-05,
+      "loss": 1.2622,
+      "theoretical_loss": 3.332750569517755,
+      "tokens_seen": 3201892352
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.9971112181030332e-05,
+      "loss": 1.3214,
+      "theoretical_loss": 3.332740306840474,
+      "tokens_seen": 3202023424
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.9930990210239125e-05,
+      "loss": 1.3567,
+      "theoretical_loss": 3.3327300447008987,
+      "tokens_seen": 3202154496
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.989086823944792e-05,
+      "loss": 1.2814,
+      "theoretical_loss": 3.332719783098979,
+      "tokens_seen": 3202285568
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.9850746268656717e-05,
+      "loss": 1.2228,
+      "theoretical_loss": 3.3327095220346648,
+      "tokens_seen": 3202416640
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.981062429786551e-05,
+      "loss": 1.3508,
+      "theoretical_loss": 3.3326992615079054,
+      "tokens_seen": 3202547712
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.9770502327074306e-05,
+      "loss": 1.283,
+      "theoretical_loss": 3.332689001518651,
+      "tokens_seen": 3202678784
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.9730380356283102e-05,
+      "loss": 1.1774,
+      "theoretical_loss": 3.3326787420668524,
+      "tokens_seen": 3202809856
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.9690258385491895e-05,
+      "loss": 1.3412,
+      "theoretical_loss": 3.332668483152458,
+      "tokens_seen": 3202940928
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.965013641470069e-05,
+      "loss": 1.28,
+      "theoretical_loss": 3.332658224775418,
+      "tokens_seen": 3203072000
+    },
+    {
+      "epoch": 0.94,
+      "objective/train/advantage_avg": 0.49100571870803833,
+      "objective/train/docs_used": 1164087,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2862893342971802,
+      "objective/train/original_loss": 1.2862892150878906,
+      "objective/train/theoretical_loss": 3.332647966935683,
+      "objective/train/tokens_used": 1573662176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24167996644973755,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503292083740234,
+      "objective/train/weighted_lm_loss": 1.3508278131484985,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.332647966935683,
+      "tokens_seen": 3203203072
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.9610014443909487e-05,
+      "loss": 1.3148,
+      "theoretical_loss": 3.332647966935683,
+      "tokens_seen": 3203203072
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.956989247311828e-05,
+      "loss": 1.2272,
+      "theoretical_loss": 3.332637709633202,
+      "tokens_seen": 3203334144
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.9529770502327073e-05,
+      "loss": 1.2974,
+      "theoretical_loss": 3.3326274528679254,
+      "tokens_seen": 3203465216
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.948964853153587e-05,
+      "loss": 1.2374,
+      "theoretical_loss": 3.332617196639803,
+      "tokens_seen": 3203596288
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.9449526560744662e-05,
+      "loss": 1.3032,
+      "theoretical_loss": 3.332606940948785,
+      "tokens_seen": 3203727360
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.9409404589953458e-05,
+      "loss": 1.2643,
+      "theoretical_loss": 3.33259668579482,
+      "tokens_seen": 3203858432
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.9369282619162254e-05,
+      "loss": 1.2959,
+      "theoretical_loss": 3.3325864311778597,
+      "tokens_seen": 3203989504
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.9329160648371047e-05,
+      "loss": 1.3441,
+      "theoretical_loss": 3.332576177097853,
+      "tokens_seen": 3204120576
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.9289038677579843e-05,
+      "loss": 1.3151,
+      "theoretical_loss": 3.33256592355475,
+      "tokens_seen": 3204251648
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.924891670678864e-05,
+      "loss": 1.3306,
+      "theoretical_loss": 3.3325556705485,
+      "tokens_seen": 3204382720
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.9208794735997432e-05,
+      "loss": 1.3073,
+      "theoretical_loss": 3.3325454180790537,
+      "tokens_seen": 3204513792
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.9168672765206228e-05,
+      "loss": 1.2934,
+      "theoretical_loss": 3.332535166146361,
+      "tokens_seen": 3204644864
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.9128550794415024e-05,
+      "loss": 1.3075,
+      "theoretical_loss": 3.332524914750372,
+      "tokens_seen": 3204775936
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.9088428823623817e-05,
+      "loss": 1.339,
+      "theoretical_loss": 3.332514663891036,
+      "tokens_seen": 3204907008
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.904830685283261e-05,
+      "loss": 1.3367,
+      "theoretical_loss": 3.3325044135683037,
+      "tokens_seen": 3205038080
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.9008184882041406e-05,
+      "loss": 1.3149,
+      "theoretical_loss": 3.3324941637821244,
+      "tokens_seen": 3205169152
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.89680629112502e-05,
+      "loss": 1.2643,
+      "theoretical_loss": 3.3324839145324487,
+      "tokens_seen": 3205300224
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.8927940940458995e-05,
+      "loss": 1.3177,
+      "theoretical_loss": 3.332473665819226,
+      "tokens_seen": 3205431296
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.888781896966779e-05,
+      "loss": 1.2789,
+      "theoretical_loss": 3.3324634176424066,
+      "tokens_seen": 3205562368
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.8847696998876584e-05,
+      "loss": 1.3592,
+      "theoretical_loss": 3.3324531700019406,
+      "tokens_seen": 3205693440
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.880757502808538e-05,
+      "loss": 1.2916,
+      "theoretical_loss": 3.3324429228977777,
+      "tokens_seen": 3205824512
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.8767453057294176e-05,
+      "loss": 1.2177,
+      "theoretical_loss": 3.3324326763298684,
+      "tokens_seen": 3205955584
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.872733108650297e-05,
+      "loss": 1.2816,
+      "theoretical_loss": 3.3324224302981618,
+      "tokens_seen": 3206086656
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.8687209115711765e-05,
+      "loss": 1.285,
+      "theoretical_loss": 3.332412184802609,
+      "tokens_seen": 3206217728
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.864708714492056e-05,
+      "loss": 1.2877,
+      "theoretical_loss": 3.332401939843159,
+      "tokens_seen": 3206348800
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.8606965174129357e-05,
+      "loss": 1.3077,
+      "theoretical_loss": 3.3323916954197625,
+      "tokens_seen": 3206479872
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.8566843203338146e-05,
+      "loss": 1.2642,
+      "theoretical_loss": 3.3323814515323695,
+      "tokens_seen": 3206610944
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.8526721232546943e-05,
+      "loss": 1.281,
+      "theoretical_loss": 3.33237120818093,
+      "tokens_seen": 3206742016
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.8486599261755735e-05,
+      "loss": 1.2538,
+      "theoretical_loss": 3.332360965365394,
+      "tokens_seen": 3206873088
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.844647729096453e-05,
+      "loss": 1.2436,
+      "theoretical_loss": 3.3323507230857112,
+      "tokens_seen": 3207004160
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.8406355320173328e-05,
+      "loss": 1.2639,
+      "theoretical_loss": 3.3323404813418325,
+      "tokens_seen": 3207135232
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.8366233349382124e-05,
+      "loss": 1.328,
+      "theoretical_loss": 3.3323302401337074,
+      "tokens_seen": 3207266304
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.8326111378590917e-05,
+      "loss": 1.2387,
+      "theoretical_loss": 3.3323199994612858,
+      "tokens_seen": 3207397376
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.8285989407799713e-05,
+      "loss": 1.3312,
+      "theoretical_loss": 3.3323097593245183,
+      "tokens_seen": 3207528448
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.824586743700851e-05,
+      "loss": 1.3305,
+      "theoretical_loss": 3.3322995197233545,
+      "tokens_seen": 3207659520
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.82057454662173e-05,
+      "loss": 1.223,
+      "theoretical_loss": 3.332289280657745,
+      "tokens_seen": 3207790592
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.8165623495426098e-05,
+      "loss": 1.2655,
+      "theoretical_loss": 3.33227904212764,
+      "tokens_seen": 3207921664
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.8125501524634894e-05,
+      "loss": 1.3457,
+      "theoretical_loss": 3.3322688041329886,
+      "tokens_seen": 3208052736
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.8085379553843683e-05,
+      "loss": 1.3404,
+      "theoretical_loss": 3.3322585666737417,
+      "tokens_seen": 3208183808
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.804525758305248e-05,
+      "loss": 1.2089,
+      "theoretical_loss": 3.3322483297498495,
+      "tokens_seen": 3208314880
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.8005135612261276e-05,
+      "loss": 1.1747,
+      "theoretical_loss": 3.332238093361262,
+      "tokens_seen": 3208445952
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.7965013641470068e-05,
+      "loss": 1.3217,
+      "theoretical_loss": 3.332227857507929,
+      "tokens_seen": 3208577024
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.7924891670678864e-05,
+      "loss": 1.3154,
+      "theoretical_loss": 3.332217622189802,
+      "tokens_seen": 3208708096
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.788476969988766e-05,
+      "loss": 1.2183,
+      "theoretical_loss": 3.3322073874068288,
+      "tokens_seen": 3208839168
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.7844647729096453e-05,
+      "loss": 1.3271,
+      "theoretical_loss": 3.332197153158962,
+      "tokens_seen": 3208970240
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.780452575830525e-05,
+      "loss": 1.2766,
+      "theoretical_loss": 3.3321869194461495,
+      "tokens_seen": 3209101312
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.7764403787514046e-05,
+      "loss": 1.233,
+      "theoretical_loss": 3.3321766862683435,
+      "tokens_seen": 3209232384
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.772428181672284e-05,
+      "loss": 1.2884,
+      "theoretical_loss": 3.3321664536254927,
+      "tokens_seen": 3209363456
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.7684159845931635e-05,
+      "loss": 1.3147,
+      "theoretical_loss": 3.3321562215175486,
+      "tokens_seen": 3209494528
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.7644037875140427e-05,
+      "loss": 1.2951,
+      "theoretical_loss": 3.3321459899444603,
+      "tokens_seen": 3209625600
+    },
+    {
+      "epoch": 0.95,
+      "objective/train/advantage_avg": 0.4925563931465149,
+      "objective/train/docs_used": 1166593,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2651240825653076,
+      "objective/train/original_loss": 1.2651240825653076,
+      "objective/train/theoretical_loss": 3.332135758906178,
+      "objective/train/tokens_used": 1580215776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2432008981704712,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504920482635498,
+      "objective/train/weighted_lm_loss": 1.3289883136749268,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.332135758906178,
+      "tokens_seen": 3209756672
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.760391590434922e-05,
+      "loss": 1.3189,
+      "theoretical_loss": 3.332135758906178,
+      "tokens_seen": 3209756672
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.7563793933558016e-05,
+      "loss": 1.2864,
+      "theoretical_loss": 3.3321255284026527,
+      "tokens_seen": 3209887744
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.7523671962766812e-05,
+      "loss": 1.2569,
+      "theoretical_loss": 3.332115298433834,
+      "tokens_seen": 3210018816
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.7483549991975605e-05,
+      "loss": 1.2681,
+      "theoretical_loss": 3.3321050689996725,
+      "tokens_seen": 3210149888
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.74434280211844e-05,
+      "loss": 1.2107,
+      "theoretical_loss": 3.3320948401001185,
+      "tokens_seen": 3210280960
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.7403306050393197e-05,
+      "loss": 1.2951,
+      "theoretical_loss": 3.3320846117351213,
+      "tokens_seen": 3210412032
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.736318407960199e-05,
+      "loss": 1.3198,
+      "theoretical_loss": 3.3320743839046325,
+      "tokens_seen": 3210543104
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.7323062108810786e-05,
+      "loss": 1.195,
+      "theoretical_loss": 3.3320641566086016,
+      "tokens_seen": 3210674176
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.7282940138019582e-05,
+      "loss": 1.2777,
+      "theoretical_loss": 3.3320539298469787,
+      "tokens_seen": 3210805248
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.7242818167228375e-05,
+      "loss": 1.3473,
+      "theoretical_loss": 3.332043703619714,
+      "tokens_seen": 3210936320
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.720269619643717e-05,
+      "loss": 1.2557,
+      "theoretical_loss": 3.3320334779267586,
+      "tokens_seen": 3211067392
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.7162574225645964e-05,
+      "loss": 1.2772,
+      "theoretical_loss": 3.332023252768062,
+      "tokens_seen": 3211198464
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.7122452254854757e-05,
+      "loss": 1.259,
+      "theoretical_loss": 3.332013028143575,
+      "tokens_seen": 3211329536
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.7082330284063553e-05,
+      "loss": 1.3519,
+      "theoretical_loss": 3.3320028040532472,
+      "tokens_seen": 3211460608
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.704220831327235e-05,
+      "loss": 1.3715,
+      "theoretical_loss": 3.3319925804970296,
+      "tokens_seen": 3211591680
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.7002086342481142e-05,
+      "loss": 1.2651,
+      "theoretical_loss": 3.331982357474872,
+      "tokens_seen": 3211722752
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.6961964371689938e-05,
+      "loss": 1.2791,
+      "theoretical_loss": 3.331972134986725,
+      "tokens_seen": 3211853824
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.6921842400898734e-05,
+      "loss": 1.3058,
+      "theoretical_loss": 3.3319619130325386,
+      "tokens_seen": 3211984896
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.6881720430107527e-05,
+      "loss": 1.2956,
+      "theoretical_loss": 3.3319516916122636,
+      "tokens_seen": 3212115968
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.6841598459316323e-05,
+      "loss": 1.3236,
+      "theoretical_loss": 3.3319414707258503,
+      "tokens_seen": 3212247040
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.680147648852512e-05,
+      "loss": 1.2813,
+      "theoretical_loss": 3.3319312503732483,
+      "tokens_seen": 3212378112
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.6761354517733912e-05,
+      "loss": 1.2623,
+      "theoretical_loss": 3.331921030554409,
+      "tokens_seen": 3212509184
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.6721232546942708e-05,
+      "loss": 1.3516,
+      "theoretical_loss": 3.3319108112692817,
+      "tokens_seen": 3212640256
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.66811105761515e-05,
+      "loss": 1.3222,
+      "theoretical_loss": 3.3319005925178176,
+      "tokens_seen": 3212771328
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.6640988605360294e-05,
+      "loss": 1.337,
+      "theoretical_loss": 3.3318903742999666,
+      "tokens_seen": 3212902400
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.660086663456909e-05,
+      "loss": 1.3808,
+      "theoretical_loss": 3.331880156615679,
+      "tokens_seen": 3213033472
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.6560744663777886e-05,
+      "loss": 1.3306,
+      "theoretical_loss": 3.3318699394649056,
+      "tokens_seen": 3213164544
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.652062269298668e-05,
+      "loss": 1.3951,
+      "theoretical_loss": 3.3318597228475966,
+      "tokens_seen": 3213295616
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.6480500722195475e-05,
+      "loss": 1.2965,
+      "theoretical_loss": 3.331849506763702,
+      "tokens_seen": 3213426688
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.644037875140427e-05,
+      "loss": 1.3096,
+      "theoretical_loss": 3.331839291213173,
+      "tokens_seen": 3213557760
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.6400256780613064e-05,
+      "loss": 1.2463,
+      "theoretical_loss": 3.3318290761959592,
+      "tokens_seen": 3213688832
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.636013480982186e-05,
+      "loss": 1.2747,
+      "theoretical_loss": 3.3318188617120112,
+      "tokens_seen": 3213819904
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.6320012839030656e-05,
+      "loss": 1.2953,
+      "theoretical_loss": 3.33180864776128,
+      "tokens_seen": 3213950976
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.627989086823945e-05,
+      "loss": 1.2964,
+      "theoretical_loss": 3.3317984343437153,
+      "tokens_seen": 3214082048
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.6239768897448245e-05,
+      "loss": 1.2811,
+      "theoretical_loss": 3.331788221459268,
+      "tokens_seen": 3214213120
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.6199646926657038e-05,
+      "loss": 1.3724,
+      "theoretical_loss": 3.3317780091078877,
+      "tokens_seen": 3214344192
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.615952495586583e-05,
+      "loss": 1.2704,
+      "theoretical_loss": 3.331767797289526,
+      "tokens_seen": 3214475264
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.6119402985074626e-05,
+      "loss": 1.3102,
+      "theoretical_loss": 3.331757586004133,
+      "tokens_seen": 3214606336
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.6079281014283423e-05,
+      "loss": 1.2915,
+      "theoretical_loss": 3.331747375251658,
+      "tokens_seen": 3214737408
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.6039159043492215e-05,
+      "loss": 1.3185,
+      "theoretical_loss": 3.3317371650320533,
+      "tokens_seen": 3214868480
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.599903707270101e-05,
+      "loss": 1.2421,
+      "theoretical_loss": 3.3317269553452684,
+      "tokens_seen": 3214999552
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.5958915101909808e-05,
+      "loss": 1.1732,
+      "theoretical_loss": 3.331716746191254,
+      "tokens_seen": 3215130624
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.59187931311186e-05,
+      "loss": 1.2865,
+      "theoretical_loss": 3.33170653756996,
+      "tokens_seen": 3215261696
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.5878671160327397e-05,
+      "loss": 1.2813,
+      "theoretical_loss": 3.3316963294813378,
+      "tokens_seen": 3215392768
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.5838549189536193e-05,
+      "loss": 1.2449,
+      "theoretical_loss": 3.331686121925337,
+      "tokens_seen": 3215523840
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.5798427218744985e-05,
+      "loss": 1.302,
+      "theoretical_loss": 3.3316759149019086,
+      "tokens_seen": 3215654912
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.575830524795378e-05,
+      "loss": 1.2882,
+      "theoretical_loss": 3.3316657084110037,
+      "tokens_seen": 3215785984
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.5718183277162574e-05,
+      "loss": 1.2928,
+      "theoretical_loss": 3.3316555024525716,
+      "tokens_seen": 3215917056
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.5678061306371367e-05,
+      "loss": 1.2825,
+      "theoretical_loss": 3.3316452970265633,
+      "tokens_seen": 3216048128
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.5637939335580163e-05,
+      "loss": 1.3171,
+      "theoretical_loss": 3.331635092132929,
+      "tokens_seen": 3216179200
+    },
+    {
+      "epoch": 0.95,
+      "objective/train/advantage_avg": 0.49511581659317017,
+      "objective/train/docs_used": 1168814,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.0947701930999756,
+      "objective/train/original_loss": 1.0947703123092651,
+      "objective/train/theoretical_loss": 3.3316248877716204,
+      "objective/train/tokens_used": 1586769376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24550645053386688,
+      "objective/train/value_max": -0.5,
+      "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": 1.150239109992981,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9048373699188232,
+      "theoretical_loss": 3.3316248877716204,
+      "tokens_seen": 3216310272
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.559781736478896e-05,
+      "loss": 1.1922,
+      "theoretical_loss": 3.3316248877716204,
+      "tokens_seen": 3216310272
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.5557695393997752e-05,
+      "loss": 1.3718,
+      "theoretical_loss": 3.331614683942587,
+      "tokens_seen": 3216441344
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.5517573423206548e-05,
+      "loss": 1.256,
+      "theoretical_loss": 3.3316044806457796,
+      "tokens_seen": 3216572416
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.5477451452415344e-05,
+      "loss": 1.3187,
+      "theoretical_loss": 3.331594277881149,
+      "tokens_seen": 3216703488
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.5437329481624137e-05,
+      "loss": 1.3446,
+      "theoretical_loss": 3.3315840756486454,
+      "tokens_seen": 3216834560
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.5397207510832933e-05,
+      "loss": 1.2733,
+      "theoretical_loss": 3.3315738739482197,
+      "tokens_seen": 3216965632
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.535708554004173e-05,
+      "loss": 1.32,
+      "theoretical_loss": 3.331563672779822,
+      "tokens_seen": 3217096704
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.5316963569250522e-05,
+      "loss": 1.3502,
+      "theoretical_loss": 3.331553472143403,
+      "tokens_seen": 3217227776
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.5276841598459315e-05,
+      "loss": 1.2653,
+      "theoretical_loss": 3.331543272038914,
+      "tokens_seen": 3217358848
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.523671962766811e-05,
+      "loss": 1.2238,
+      "theoretical_loss": 3.3315330724663044,
+      "tokens_seen": 3217489920
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.5196597656876904e-05,
+      "loss": 1.3116,
+      "theoretical_loss": 3.331522873425526,
+      "tokens_seen": 3217620992
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.51564756860857e-05,
+      "loss": 1.3179,
+      "theoretical_loss": 3.3315126749165285,
+      "tokens_seen": 3217752064
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.5116353715294496e-05,
+      "loss": 1.2697,
+      "theoretical_loss": 3.331502476939263,
+      "tokens_seen": 3217883136
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.507623174450329e-05,
+      "loss": 1.3304,
+      "theoretical_loss": 3.33149227949368,
+      "tokens_seen": 3218014208
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.5036109773712085e-05,
+      "loss": 1.2695,
+      "theoretical_loss": 3.33148208257973,
+      "tokens_seen": 3218145280
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.499598780292088e-05,
+      "loss": 1.2757,
+      "theoretical_loss": 3.331471886197364,
+      "tokens_seen": 3218276352
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.4955865832129674e-05,
+      "loss": 1.2885,
+      "theoretical_loss": 3.331461690346532,
+      "tokens_seen": 3218407424
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.491574386133847e-05,
+      "loss": 1.3334,
+      "theoretical_loss": 3.331451495027185,
+      "tokens_seen": 3218538496
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.4875621890547266e-05,
+      "loss": 1.2396,
+      "theoretical_loss": 3.331441300239274,
+      "tokens_seen": 3218669568
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.483549991975606e-05,
+      "loss": 1.2167,
+      "theoretical_loss": 3.331431105982749,
+      "tokens_seen": 3218800640
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.4795377948964852e-05,
+      "loss": 1.3434,
+      "theoretical_loss": 3.331420912257561,
+      "tokens_seen": 3218931712
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.4755255978173648e-05,
+      "loss": 1.3038,
+      "theoretical_loss": 3.331410719063661,
+      "tokens_seen": 3219062784
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.471513400738244e-05,
+      "loss": 1.2525,
+      "theoretical_loss": 3.331400526400999,
+      "tokens_seen": 3219193856
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.4675012036591237e-05,
+      "loss": 1.3624,
+      "theoretical_loss": 3.3313903342695266,
+      "tokens_seen": 3219324928
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.4634890065800033e-05,
+      "loss": 1.25,
+      "theoretical_loss": 3.3313801426691936,
+      "tokens_seen": 3219456000
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.4594768095008826e-05,
+      "loss": 1.2736,
+      "theoretical_loss": 3.331369951599951,
+      "tokens_seen": 3219587072
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.4554646124217622e-05,
+      "loss": 1.2663,
+      "theoretical_loss": 3.331359761061749,
+      "tokens_seen": 3219718144
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.4514524153426418e-05,
+      "loss": 1.2614,
+      "theoretical_loss": 3.3313495710545395,
+      "tokens_seen": 3219849216
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.447440218263521e-05,
+      "loss": 1.2794,
+      "theoretical_loss": 3.3313393815782724,
+      "tokens_seen": 3219980288
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.4434280211844007e-05,
+      "loss": 1.2778,
+      "theoretical_loss": 3.3313291926328987,
+      "tokens_seen": 3220111360
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.4394158241052803e-05,
+      "loss": 1.2813,
+      "theoretical_loss": 3.331319004218369,
+      "tokens_seen": 3220242432
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.4354036270261596e-05,
+      "loss": 1.2539,
+      "theoretical_loss": 3.3313088163346336,
+      "tokens_seen": 3220373504
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.431391429947039e-05,
+      "loss": 1.2359,
+      "theoretical_loss": 3.331298628981644,
+      "tokens_seen": 3220504576
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.4273792328679185e-05,
+      "loss": 1.3655,
+      "theoretical_loss": 3.3312884421593507,
+      "tokens_seen": 3220635648
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.4233670357887977e-05,
+      "loss": 1.3414,
+      "theoretical_loss": 3.3312782558677045,
+      "tokens_seen": 3220766720
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.4193548387096773e-05,
+      "loss": 1.3808,
+      "theoretical_loss": 3.331268070106656,
+      "tokens_seen": 3220897792
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.415342641630557e-05,
+      "loss": 1.2785,
+      "theoretical_loss": 3.3312578848761563,
+      "tokens_seen": 3221028864
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.4113304445514366e-05,
+      "loss": 1.2304,
+      "theoretical_loss": 3.3312477001761556,
+      "tokens_seen": 3221159936
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.407318247472316e-05,
+      "loss": 1.3958,
+      "theoretical_loss": 3.331237516006605,
+      "tokens_seen": 3221291008
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.4033060503931955e-05,
+      "loss": 1.2201,
+      "theoretical_loss": 3.331227332367455,
+      "tokens_seen": 3221422080
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.399293853314075e-05,
+      "loss": 1.3273,
+      "theoretical_loss": 3.331217149258657,
+      "tokens_seen": 3221553152
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.3952816562349544e-05,
+      "loss": 1.3052,
+      "theoretical_loss": 3.3312069666801616,
+      "tokens_seen": 3221684224
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.391269459155834e-05,
+      "loss": 1.2817,
+      "theoretical_loss": 3.3311967846319197,
+      "tokens_seen": 3221815296
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.3872572620767136e-05,
+      "loss": 1.3138,
+      "theoretical_loss": 3.3311866031138813,
+      "tokens_seen": 3221946368
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.3832450649975925e-05,
+      "loss": 1.278,
+      "theoretical_loss": 3.3311764221259983,
+      "tokens_seen": 3222077440
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.379232867918472e-05,
+      "loss": 1.3529,
+      "theoretical_loss": 3.331166241668221,
+      "tokens_seen": 3222208512
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.3752206708393518e-05,
+      "loss": 1.2683,
+      "theoretical_loss": 3.3311560617405003,
+      "tokens_seen": 3222339584
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.371208473760231e-05,
+      "loss": 1.2264,
+      "theoretical_loss": 3.331145882342787,
+      "tokens_seen": 3222470656
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.3671962766811106e-05,
+      "loss": 1.266,
+      "theoretical_loss": 3.3311357034750317,
+      "tokens_seen": 3222601728
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.3631840796019903e-05,
+      "loss": 1.3127,
+      "theoretical_loss": 3.3311255251371863,
+      "tokens_seen": 3222732800
+    },
+    {
+      "epoch": 0.95,
+      "objective/train/advantage_avg": 0.49344173073768616,
+      "objective/train/docs_used": 1170779,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2451012134552002,
+      "objective/train/original_loss": 1.2451010942459106,
+      "objective/train/theoretical_loss": 3.3311153473292006,
+      "objective/train/tokens_used": 1593322976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24387894570827484,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505839586257935,
+      "objective/train/weighted_lm_loss": 1.308040738105774,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3311153473292006,
+      "tokens_seen": 3222863872
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.3591718825228695e-05,
+      "loss": 1.2949,
+      "theoretical_loss": 3.3311153473292006,
+      "tokens_seen": 3222863872
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.355159685443749e-05,
+      "loss": 1.3009,
+      "theoretical_loss": 3.3311051700510257,
+      "tokens_seen": 3222994944
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.3511474883646288e-05,
+      "loss": 1.3483,
+      "theoretical_loss": 3.3310949933026124,
+      "tokens_seen": 3223126016
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.347135291285508e-05,
+      "loss": 1.3829,
+      "theoretical_loss": 3.331084817083912,
+      "tokens_seen": 3223257088
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.3431230942063876e-05,
+      "loss": 1.3285,
+      "theoretical_loss": 3.331074641394875,
+      "tokens_seen": 3223388160
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.339110897127267e-05,
+      "loss": 1.3314,
+      "theoretical_loss": 3.3310644662354525,
+      "tokens_seen": 3223519232
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.3350987000481462e-05,
+      "loss": 1.2187,
+      "theoretical_loss": 3.331054291605595,
+      "tokens_seen": 3223650304
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.3310865029690258e-05,
+      "loss": 1.342,
+      "theoretical_loss": 3.3310441175052543,
+      "tokens_seen": 3223781376
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.3270743058899054e-05,
+      "loss": 1.1793,
+      "theoretical_loss": 3.3310339439343806,
+      "tokens_seen": 3223912448
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.3230621088107847e-05,
+      "loss": 1.2775,
+      "theoretical_loss": 3.331023770892925,
+      "tokens_seen": 3224043520
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.3190499117316643e-05,
+      "loss": 1.3003,
+      "theoretical_loss": 3.331013598380838,
+      "tokens_seen": 3224174592
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.315037714652544e-05,
+      "loss": 1.2735,
+      "theoretical_loss": 3.3310034263980715,
+      "tokens_seen": 3224305664
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.3110255175734232e-05,
+      "loss": 1.3241,
+      "theoretical_loss": 3.330993254944576,
+      "tokens_seen": 3224436736
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.3070133204943028e-05,
+      "loss": 1.3477,
+      "theoretical_loss": 3.3309830840203016,
+      "tokens_seen": 3224567808
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.3030011234151824e-05,
+      "loss": 1.3735,
+      "theoretical_loss": 3.3309729136252004,
+      "tokens_seen": 3224698880
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.2989889263360617e-05,
+      "loss": 1.3181,
+      "theoretical_loss": 3.330962743759223,
+      "tokens_seen": 3224829952
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.2949767292569413e-05,
+      "loss": 1.319,
+      "theoretical_loss": 3.3309525744223203,
+      "tokens_seen": 3224961024
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.2909645321778206e-05,
+      "loss": 1.2841,
+      "theoretical_loss": 3.3309424056144437,
+      "tokens_seen": 3225092096
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.2869523350987e-05,
+      "loss": 1.2765,
+      "theoretical_loss": 3.330932237335543,
+      "tokens_seen": 3225223168
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.2829401380195795e-05,
+      "loss": 1.3686,
+      "theoretical_loss": 3.3309220695855704,
+      "tokens_seen": 3225354240
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.278927940940459e-05,
+      "loss": 1.2789,
+      "theoretical_loss": 3.3309119023644764,
+      "tokens_seen": 3225485312
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.2749157438613384e-05,
+      "loss": 1.2968,
+      "theoretical_loss": 3.330901735672212,
+      "tokens_seen": 3225616384
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.270903546782218e-05,
+      "loss": 1.3,
+      "theoretical_loss": 3.3308915695087284,
+      "tokens_seen": 3225747456
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.2668913497030976e-05,
+      "loss": 1.3392,
+      "theoretical_loss": 3.3308814038739762,
+      "tokens_seen": 3225878528
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.262879152623977e-05,
+      "loss": 1.3248,
+      "theoretical_loss": 3.330871238767907,
+      "tokens_seen": 3226009600
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.2588669555448565e-05,
+      "loss": 1.2764,
+      "theoretical_loss": 3.330861074190471,
+      "tokens_seen": 3226140672
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.254854758465736e-05,
+      "loss": 1.2887,
+      "theoretical_loss": 3.33085091014162,
+      "tokens_seen": 3226271744
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.2508425613866154e-05,
+      "loss": 1.382,
+      "theoretical_loss": 3.330840746621305,
+      "tokens_seen": 3226402816
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.246830364307495e-05,
+      "loss": 1.2978,
+      "theoretical_loss": 3.330830583629477,
+      "tokens_seen": 3226533888
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.2428181672283743e-05,
+      "loss": 1.3602,
+      "theoretical_loss": 3.3308204211660866,
+      "tokens_seen": 3226664960
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.2388059701492536e-05,
+      "loss": 1.275,
+      "theoretical_loss": 3.330810259231085,
+      "tokens_seen": 3226796032
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.234793773070133e-05,
+      "loss": 1.2573,
+      "theoretical_loss": 3.3308000978244237,
+      "tokens_seen": 3226927104
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.2307815759910128e-05,
+      "loss": 1.3682,
+      "theoretical_loss": 3.3307899369460534,
+      "tokens_seen": 3227058176
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.226769378911892e-05,
+      "loss": 1.3388,
+      "theoretical_loss": 3.330779776595925,
+      "tokens_seen": 3227189248
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.2227571818327717e-05,
+      "loss": 1.3351,
+      "theoretical_loss": 3.33076961677399,
+      "tokens_seen": 3227320320
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.2187449847536513e-05,
+      "loss": 1.2477,
+      "theoretical_loss": 3.3307594574802,
+      "tokens_seen": 3227451392
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.2147327876745306e-05,
+      "loss": 1.288,
+      "theoretical_loss": 3.3307492987145046,
+      "tokens_seen": 3227582464
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.2107205905954102e-05,
+      "loss": 1.2988,
+      "theoretical_loss": 3.3307391404768563,
+      "tokens_seen": 3227713536
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.2067083935162898e-05,
+      "loss": 1.311,
+      "theoretical_loss": 3.330728982767205,
+      "tokens_seen": 3227844608
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.202696196437169e-05,
+      "loss": 1.3495,
+      "theoretical_loss": 3.330718825585503,
+      "tokens_seen": 3227975680
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.1986839993580487e-05,
+      "loss": 1.3723,
+      "theoretical_loss": 3.3307086689317007,
+      "tokens_seen": 3228106752
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.194671802278928e-05,
+      "loss": 1.2766,
+      "theoretical_loss": 3.33069851280575,
+      "tokens_seen": 3228237824
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.1906596051998072e-05,
+      "loss": 1.3252,
+      "theoretical_loss": 3.3306883572076007,
+      "tokens_seen": 3228368896
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.186647408120687e-05,
+      "loss": 1.218,
+      "theoretical_loss": 3.330678202137205,
+      "tokens_seen": 3228499968
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.1826352110415665e-05,
+      "loss": 1.3111,
+      "theoretical_loss": 3.3306680475945134,
+      "tokens_seen": 3228631040
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.1786230139624457e-05,
+      "loss": 1.3513,
+      "theoretical_loss": 3.330657893579478,
+      "tokens_seen": 3228762112
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.1746108168833253e-05,
+      "loss": 1.3869,
+      "theoretical_loss": 3.330647740092049,
+      "tokens_seen": 3228893184
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.170598619804205e-05,
+      "loss": 1.2431,
+      "theoretical_loss": 3.3306375871321783,
+      "tokens_seen": 3229024256
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.1665864227250842e-05,
+      "loss": 1.2395,
+      "theoretical_loss": 3.3306274346998164,
+      "tokens_seen": 3229155328
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.162574225645964e-05,
+      "loss": 1.3213,
+      "theoretical_loss": 3.3306172827949148,
+      "tokens_seen": 3229286400
+    },
+    {
+      "epoch": 0.96,
+      "objective/train/advantage_avg": 0.4933333694934845,
+      "objective/train/docs_used": 1173331,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.1824135780334473,
+      "objective/train/original_loss": 1.1824135780334473,
+      "objective/train/theoretical_loss": 3.330607131417425,
+      "objective/train/tokens_used": 1599876576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24372342228889465,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050572395324707,
+      "objective/train/weighted_lm_loss": 1.2420932054519653,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.330607131417425,
+      "tokens_seen": 3229417472
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.1585620285668435e-05,
+      "loss": 1.2938,
+      "theoretical_loss": 3.330607131417425,
+      "tokens_seen": 3229417472
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.1545498314877227e-05,
+      "loss": 1.3076,
+      "theoretical_loss": 3.3305969805672975,
+      "tokens_seen": 3229548544
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.1505376344086024e-05,
+      "loss": 1.3341,
+      "theoretical_loss": 3.330586830244484,
+      "tokens_seen": 3229679616
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.1465254373294816e-05,
+      "loss": 1.2318,
+      "theoretical_loss": 3.330576680448936,
+      "tokens_seen": 3229810688
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.142513240250361e-05,
+      "loss": 1.1866,
+      "theoretical_loss": 3.3305665311806036,
+      "tokens_seen": 3229941760
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.1385010431712405e-05,
+      "loss": 1.2966,
+      "theoretical_loss": 3.3305563824394393,
+      "tokens_seen": 3230072832
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.13448884609212e-05,
+      "loss": 1.2774,
+      "theoretical_loss": 3.3305462342253938,
+      "tokens_seen": 3230203904
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.1304766490129994e-05,
+      "loss": 1.3508,
+      "theoretical_loss": 3.3305360865384177,
+      "tokens_seen": 3230334976
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.126464451933879e-05,
+      "loss": 1.2625,
+      "theoretical_loss": 3.3305259393784636,
+      "tokens_seen": 3230466048
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.1224522548547586e-05,
+      "loss": 1.3014,
+      "theoretical_loss": 3.3305157927454814,
+      "tokens_seen": 3230597120
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.118440057775638e-05,
+      "loss": 1.3093,
+      "theoretical_loss": 3.3305056466394234,
+      "tokens_seen": 3230728192
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.1144278606965175e-05,
+      "loss": 1.2504,
+      "theoretical_loss": 3.3304955010602404,
+      "tokens_seen": 3230859264
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.110415663617397e-05,
+      "loss": 1.3545,
+      "theoretical_loss": 3.330485356007883,
+      "tokens_seen": 3230990336
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.1064034665382764e-05,
+      "loss": 1.2994,
+      "theoretical_loss": 3.330475211482304,
+      "tokens_seen": 3231121408
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.1023912694591557e-05,
+      "loss": 1.3423,
+      "theoretical_loss": 3.3304650674834533,
+      "tokens_seen": 3231252480
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.0983790723800353e-05,
+      "loss": 1.3205,
+      "theoretical_loss": 3.3304549240112826,
+      "tokens_seen": 3231383552
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.0943668753009146e-05,
+      "loss": 1.2717,
+      "theoretical_loss": 3.3304447810657436,
+      "tokens_seen": 3231514624
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.0903546782217942e-05,
+      "loss": 1.3041,
+      "theoretical_loss": 3.3304346386467873,
+      "tokens_seen": 3231645696
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.0863424811426738e-05,
+      "loss": 1.3196,
+      "theoretical_loss": 3.330424496754365,
+      "tokens_seen": 3231776768
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.082330284063553e-05,
+      "loss": 1.3752,
+      "theoretical_loss": 3.3304143553884282,
+      "tokens_seen": 3231907840
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.0783180869844327e-05,
+      "loss": 1.2933,
+      "theoretical_loss": 3.3304042145489277,
+      "tokens_seen": 3232038912
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.0743058899053123e-05,
+      "loss": 1.2548,
+      "theoretical_loss": 3.3303940742358153,
+      "tokens_seen": 3232169984
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.0702936928261916e-05,
+      "loss": 1.2374,
+      "theoretical_loss": 3.330383934449042,
+      "tokens_seen": 3232301056
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.0662814957470712e-05,
+      "loss": 1.3418,
+      "theoretical_loss": 3.3303737951885597,
+      "tokens_seen": 3232432128
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.0622692986679508e-05,
+      "loss": 1.3287,
+      "theoretical_loss": 3.330363656454319,
+      "tokens_seen": 3232563200
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.05825710158883e-05,
+      "loss": 1.352,
+      "theoretical_loss": 3.3303535182462722,
+      "tokens_seen": 3232694272
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.0542449045097094e-05,
+      "loss": 1.2904,
+      "theoretical_loss": 3.3303433805643694,
+      "tokens_seen": 3232825344
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.050232707430589e-05,
+      "loss": 1.3647,
+      "theoretical_loss": 3.330333243408563,
+      "tokens_seen": 3232956416
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.0462205103514683e-05,
+      "loss": 1.2632,
+      "theoretical_loss": 3.330323106778804,
+      "tokens_seen": 3233087488
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.042208313272348e-05,
+      "loss": 1.3235,
+      "theoretical_loss": 3.330312970675044,
+      "tokens_seen": 3233218560
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.0381961161932275e-05,
+      "loss": 1.2964,
+      "theoretical_loss": 3.3303028350972337,
+      "tokens_seen": 3233349632
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.0341839191141068e-05,
+      "loss": 1.3034,
+      "theoretical_loss": 3.3302927000453257,
+      "tokens_seen": 3233480704
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.0301717220349864e-05,
+      "loss": 1.286,
+      "theoretical_loss": 3.33028256551927,
+      "tokens_seen": 3233611776
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.026159524955866e-05,
+      "loss": 1.2418,
+      "theoretical_loss": 3.3302724315190186,
+      "tokens_seen": 3233742848
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.0221473278767453e-05,
+      "loss": 1.2812,
+      "theoretical_loss": 3.3302622980445236,
+      "tokens_seen": 3233873920
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.018135130797625e-05,
+      "loss": 1.3142,
+      "theoretical_loss": 3.3302521650957355,
+      "tokens_seen": 3234004992
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.0141229337185045e-05,
+      "loss": 1.2815,
+      "theoretical_loss": 3.330242032672606,
+      "tokens_seen": 3234136064
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.0101107366393838e-05,
+      "loss": 1.2958,
+      "theoretical_loss": 3.330231900775087,
+      "tokens_seen": 3234267136
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.006098539560263e-05,
+      "loss": 1.2598,
+      "theoretical_loss": 3.3302217694031286,
+      "tokens_seen": 3234398208
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.0020863424811427e-05,
+      "loss": 1.3216,
+      "theoretical_loss": 3.330211638556684,
+      "tokens_seen": 3234529280
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.998074145402022e-05,
+      "loss": 1.2913,
+      "theoretical_loss": 3.330201508235703,
+      "tokens_seen": 3234660352
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.9940619483229015e-05,
+      "loss": 1.3276,
+      "theoretical_loss": 3.3301913784401385,
+      "tokens_seen": 3234791424
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.990049751243781e-05,
+      "loss": 1.3119,
+      "theoretical_loss": 3.3301812491699407,
+      "tokens_seen": 3234922496
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.9860375541646604e-05,
+      "loss": 1.3466,
+      "theoretical_loss": 3.330171120425062,
+      "tokens_seen": 3235053568
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.98202535708554e-05,
+      "loss": 1.3104,
+      "theoretical_loss": 3.3301609922054536,
+      "tokens_seen": 3235184640
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.9780131600064197e-05,
+      "loss": 1.2571,
+      "theoretical_loss": 3.330150864511067,
+      "tokens_seen": 3235315712
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.9740009629272993e-05,
+      "loss": 1.3853,
+      "theoretical_loss": 3.3301407373418535,
+      "tokens_seen": 3235446784
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.9699887658481786e-05,
+      "loss": 1.2301,
+      "theoretical_loss": 3.3301306106977644,
+      "tokens_seen": 3235577856
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.965976568769058e-05,
+      "loss": 1.3635,
+      "theoretical_loss": 3.3301204845787513,
+      "tokens_seen": 3235708928
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.9619643716899378e-05,
+      "loss": 1.2969,
+      "theoretical_loss": 3.3301103589847667,
+      "tokens_seen": 3235840000
+    },
+    {
+      "epoch": 0.96,
+      "objective/train/advantage_avg": 0.4917212426662445,
+      "objective/train/docs_used": 1175644,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2475812435150146,
+      "objective/train/original_loss": 1.2475812435150146,
+      "objective/train/theoretical_loss": 3.330100233915761,
+      "objective/train/tokens_used": 1606430176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2426784783601761,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504058599472046,
+      "objective/train/weighted_lm_loss": 1.3104125261306763,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.330100233915761,
+      "tokens_seen": 3235971072
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.9579521746108167e-05,
+      "loss": 1.299,
+      "theoretical_loss": 3.330100233915761,
+      "tokens_seen": 3235971072
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.9539399775316963e-05,
+      "loss": 1.289,
+      "theoretical_loss": 3.3300901093716857,
+      "tokens_seen": 3236102144
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.949927780452576e-05,
+      "loss": 1.324,
+      "theoretical_loss": 3.330079985352493,
+      "tokens_seen": 3236233216
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.9459155833734552e-05,
+      "loss": 1.249,
+      "theoretical_loss": 3.3300698618581337,
+      "tokens_seen": 3236364288
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.941903386294335e-05,
+      "loss": 1.3223,
+      "theoretical_loss": 3.3300597388885604,
+      "tokens_seen": 3236495360
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.9378911892152144e-05,
+      "loss": 1.2974,
+      "theoretical_loss": 3.3300496164437234,
+      "tokens_seen": 3236626432
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.9338789921360937e-05,
+      "loss": 1.229,
+      "theoretical_loss": 3.330039494523575,
+      "tokens_seen": 3236757504
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.9298667950569733e-05,
+      "loss": 1.2995,
+      "theoretical_loss": 3.330029373128067,
+      "tokens_seen": 3236888576
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.925854597977853e-05,
+      "loss": 1.3124,
+      "theoretical_loss": 3.3300192522571495,
+      "tokens_seen": 3237019648
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.9218424008987322e-05,
+      "loss": 1.2842,
+      "theoretical_loss": 3.330009131910776,
+      "tokens_seen": 3237150720
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.917830203819612e-05,
+      "loss": 1.2322,
+      "theoretical_loss": 3.329999012088897,
+      "tokens_seen": 3237281792
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.913818006740491e-05,
+      "loss": 1.2714,
+      "theoretical_loss": 3.329988892791464,
+      "tokens_seen": 3237412864
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.9098058096613704e-05,
+      "loss": 1.3508,
+      "theoretical_loss": 3.3299787740184295,
+      "tokens_seen": 3237543936
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.90579361258225e-05,
+      "loss": 1.3634,
+      "theoretical_loss": 3.329968655769744,
+      "tokens_seen": 3237675008
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.9017814155031296e-05,
+      "loss": 1.3756,
+      "theoretical_loss": 3.3299585380453594,
+      "tokens_seen": 3237806080
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.897769218424009e-05,
+      "loss": 1.2253,
+      "theoretical_loss": 3.329948420845228,
+      "tokens_seen": 3237937152
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.8937570213448885e-05,
+      "loss": 1.3198,
+      "theoretical_loss": 3.3299383041693003,
+      "tokens_seen": 3238068224
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.889744824265768e-05,
+      "loss": 1.2813,
+      "theoretical_loss": 3.329928188017529,
+      "tokens_seen": 3238199296
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.8857326271866474e-05,
+      "loss": 1.2827,
+      "theoretical_loss": 3.329918072389865,
+      "tokens_seen": 3238330368
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.881720430107527e-05,
+      "loss": 1.2964,
+      "theoretical_loss": 3.32990795728626,
+      "tokens_seen": 3238461440
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.8777082330284066e-05,
+      "loss": 1.3245,
+      "theoretical_loss": 3.329897842706666,
+      "tokens_seen": 3238592512
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.873696035949286e-05,
+      "loss": 1.3435,
+      "theoretical_loss": 3.3298877286510344,
+      "tokens_seen": 3238723584
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.8696838388701655e-05,
+      "loss": 1.3641,
+      "theoretical_loss": 3.329877615119317,
+      "tokens_seen": 3238854656
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.8656716417910448e-05,
+      "loss": 1.2148,
+      "theoretical_loss": 3.3298675021114654,
+      "tokens_seen": 3238985728
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.861659444711924e-05,
+      "loss": 1.3158,
+      "theoretical_loss": 3.3298573896274313,
+      "tokens_seen": 3239116800
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.8576472476328037e-05,
+      "loss": 1.3248,
+      "theoretical_loss": 3.3298472776671666,
+      "tokens_seen": 3239247872
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.8536350505536833e-05,
+      "loss": 1.3554,
+      "theoretical_loss": 3.329837166230622,
+      "tokens_seen": 3239378944
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.8496228534745626e-05,
+      "loss": 1.3532,
+      "theoretical_loss": 3.3298270553177503,
+      "tokens_seen": 3239510016
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.8456106563954422e-05,
+      "loss": 1.2727,
+      "theoretical_loss": 3.3298169449285027,
+      "tokens_seen": 3239641088
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.8415984593163218e-05,
+      "loss": 1.2528,
+      "theoretical_loss": 3.329806835062831,
+      "tokens_seen": 3239772160
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.837586262237201e-05,
+      "loss": 1.2666,
+      "theoretical_loss": 3.329796725720687,
+      "tokens_seen": 3239903232
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.8335740651580807e-05,
+      "loss": 1.2918,
+      "theoretical_loss": 3.3297866169020223,
+      "tokens_seen": 3240034304
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.8295618680789603e-05,
+      "loss": 1.3374,
+      "theoretical_loss": 3.3297765086067885,
+      "tokens_seen": 3240165376
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.8255496709998396e-05,
+      "loss": 1.3373,
+      "theoretical_loss": 3.329766400834937,
+      "tokens_seen": 3240296448
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.8215374739207192e-05,
+      "loss": 1.3982,
+      "theoretical_loss": 3.3297562935864207,
+      "tokens_seen": 3240427520
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.8175252768415985e-05,
+      "loss": 1.3263,
+      "theoretical_loss": 3.32974618686119,
+      "tokens_seen": 3240558592
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.8135130797624777e-05,
+      "loss": 1.3125,
+      "theoretical_loss": 3.3297360806591976,
+      "tokens_seen": 3240689664
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.8095008826833574e-05,
+      "loss": 1.2695,
+      "theoretical_loss": 3.329725974980395,
+      "tokens_seen": 3240820736
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.805488685604237e-05,
+      "loss": 1.2275,
+      "theoretical_loss": 3.3297158698247338,
+      "tokens_seen": 3240951808
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.8014764885251163e-05,
+      "loss": 1.3199,
+      "theoretical_loss": 3.3297057651921653,
+      "tokens_seen": 3241082880
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.797464291445996e-05,
+      "loss": 1.3208,
+      "theoretical_loss": 3.3296956610826425,
+      "tokens_seen": 3241213952
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.7934520943668755e-05,
+      "loss": 1.3023,
+      "theoretical_loss": 3.329685557496116,
+      "tokens_seen": 3241345024
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.7894398972877548e-05,
+      "loss": 1.2064,
+      "theoretical_loss": 3.329675454432538,
+      "tokens_seen": 3241476096
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.7854277002086344e-05,
+      "loss": 1.2121,
+      "theoretical_loss": 3.329665351891861,
+      "tokens_seen": 3241607168
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.781415503129514e-05,
+      "loss": 1.3226,
+      "theoretical_loss": 3.3296552498740355,
+      "tokens_seen": 3241738240
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.7774033060503933e-05,
+      "loss": 1.3049,
+      "theoretical_loss": 3.329645148379014,
+      "tokens_seen": 3241869312
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.773391108971273e-05,
+      "loss": 1.2453,
+      "theoretical_loss": 3.3296350474067484,
+      "tokens_seen": 3242000384
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.769378911892152e-05,
+      "loss": 1.2607,
+      "theoretical_loss": 3.3296249469571904,
+      "tokens_seen": 3242131456
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.7653667148130314e-05,
+      "loss": 1.2367,
+      "theoretical_loss": 3.3296148470302915,
+      "tokens_seen": 3242262528
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.761354517733911e-05,
+      "loss": 1.2667,
+      "theoretical_loss": 3.329604747626004,
+      "tokens_seen": 3242393600
+    },
+    {
+      "epoch": 0.97,
+      "objective/train/advantage_avg": 0.49251583218574524,
+      "objective/train/docs_used": 1177712,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2686760425567627,
+      "objective/train/original_loss": 1.2686760425567627,
+      "objective/train/theoretical_loss": 3.32959464874428,
+      "objective/train/tokens_used": 1612983776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24321727454662323,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504882335662842,
+      "objective/train/weighted_lm_loss": 1.3326996564865112,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.32959464874428,
+      "tokens_seen": 3242524672
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.7573423206547907e-05,
+      "loss": 1.293,
+      "theoretical_loss": 3.32959464874428,
+      "tokens_seen": 3242524672
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.75333012357567e-05,
+      "loss": 1.3277,
+      "theoretical_loss": 3.32958455038507,
+      "tokens_seen": 3242655744
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.7493179264965495e-05,
+      "loss": 1.397,
+      "theoretical_loss": 3.329574452548327,
+      "tokens_seen": 3242786816
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.745305729417429e-05,
+      "loss": 1.2837,
+      "theoretical_loss": 3.3295643552340026,
+      "tokens_seen": 3242917888
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.7412935323383084e-05,
+      "loss": 1.2673,
+      "theoretical_loss": 3.329554258442049,
+      "tokens_seen": 3243048960
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.737281335259188e-05,
+      "loss": 1.4001,
+      "theoretical_loss": 3.329544162172417,
+      "tokens_seen": 3243180032
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.7332691381800677e-05,
+      "loss": 1.2816,
+      "theoretical_loss": 3.32953406642506,
+      "tokens_seen": 3243311104
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.729256941100947e-05,
+      "loss": 1.2807,
+      "theoretical_loss": 3.3295239711999285,
+      "tokens_seen": 3243442176
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.7252447440218265e-05,
+      "loss": 1.3718,
+      "theoretical_loss": 3.329513876496975,
+      "tokens_seen": 3243573248
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.7212325469427058e-05,
+      "loss": 1.2502,
+      "theoretical_loss": 3.3295037823161513,
+      "tokens_seen": 3243704320
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.717220349863585e-05,
+      "loss": 1.3099,
+      "theoretical_loss": 3.3294936886574096,
+      "tokens_seen": 3243835392
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.7132081527844647e-05,
+      "loss": 1.299,
+      "theoretical_loss": 3.329483595520701,
+      "tokens_seen": 3243966464
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.7091959557053443e-05,
+      "loss": 1.2446,
+      "theoretical_loss": 3.3294735029059783,
+      "tokens_seen": 3244097536
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.7051837586262236e-05,
+      "loss": 1.2835,
+      "theoretical_loss": 3.329463410813193,
+      "tokens_seen": 3244228608
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.7011715615471032e-05,
+      "loss": 1.2785,
+      "theoretical_loss": 3.329453319242297,
+      "tokens_seen": 3244359680
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.697159364467983e-05,
+      "loss": 1.2651,
+      "theoretical_loss": 3.3294432281932425,
+      "tokens_seen": 3244490752
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.693147167388862e-05,
+      "loss": 1.2766,
+      "theoretical_loss": 3.329433137665981,
+      "tokens_seen": 3244621824
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.6891349703097417e-05,
+      "loss": 1.3311,
+      "theoretical_loss": 3.3294230476604647,
+      "tokens_seen": 3244752896
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.6851227732306213e-05,
+      "loss": 1.2903,
+      "theoretical_loss": 3.3294129581766456,
+      "tokens_seen": 3244883968
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.6811105761515006e-05,
+      "loss": 1.3399,
+      "theoretical_loss": 3.3294028692144755,
+      "tokens_seen": 3245015040
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.67709837907238e-05,
+      "loss": 1.335,
+      "theoretical_loss": 3.3293927807739063,
+      "tokens_seen": 3245146112
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.6730861819932595e-05,
+      "loss": 1.3313,
+      "theoretical_loss": 3.3293826928548906,
+      "tokens_seen": 3245277184
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.6690739849141388e-05,
+      "loss": 1.3365,
+      "theoretical_loss": 3.3293726054573796,
+      "tokens_seen": 3245408256
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.6650617878350184e-05,
+      "loss": 1.282,
+      "theoretical_loss": 3.3293625185813256,
+      "tokens_seen": 3245539328
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.661049590755898e-05,
+      "loss": 1.2627,
+      "theoretical_loss": 3.3293524322266803,
+      "tokens_seen": 3245670400
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.6570373936767773e-05,
+      "loss": 1.3097,
+      "theoretical_loss": 3.329342346393396,
+      "tokens_seen": 3245801472
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.653025196597657e-05,
+      "loss": 1.2541,
+      "theoretical_loss": 3.329332261081425,
+      "tokens_seen": 3245932544
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.6490129995185365e-05,
+      "loss": 1.3493,
+      "theoretical_loss": 3.329322176290719,
+      "tokens_seen": 3246063616
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.6450008024394158e-05,
+      "loss": 1.2365,
+      "theoretical_loss": 3.3293120920212296,
+      "tokens_seen": 3246194688
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.6409886053602954e-05,
+      "loss": 1.2399,
+      "theoretical_loss": 3.3293020082729092,
+      "tokens_seen": 3246325760
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.636976408281175e-05,
+      "loss": 1.3128,
+      "theoretical_loss": 3.32929192504571,
+      "tokens_seen": 3246456832
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.6329642112020543e-05,
+      "loss": 1.284,
+      "theoretical_loss": 3.3292818423395834,
+      "tokens_seen": 3246587904
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.6289520141229336e-05,
+      "loss": 1.2618,
+      "theoretical_loss": 3.3292717601544823,
+      "tokens_seen": 3246718976
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.6249398170438132e-05,
+      "loss": 1.319,
+      "theoretical_loss": 3.329261678490358,
+      "tokens_seen": 3246850048
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.6209276199646925e-05,
+      "loss": 1.3603,
+      "theoretical_loss": 3.329251597347163,
+      "tokens_seen": 3246981120
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.616915422885572e-05,
+      "loss": 1.2889,
+      "theoretical_loss": 3.3292415167248492,
+      "tokens_seen": 3247112192
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.6129032258064517e-05,
+      "loss": 1.4072,
+      "theoretical_loss": 3.3292314366233686,
+      "tokens_seen": 3247243264
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.608891028727331e-05,
+      "loss": 1.3349,
+      "theoretical_loss": 3.3292213570426736,
+      "tokens_seen": 3247374336
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.6048788316482106e-05,
+      "loss": 1.3006,
+      "theoretical_loss": 3.329211277982716,
+      "tokens_seen": 3247505408
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.6008666345690902e-05,
+      "loss": 1.2847,
+      "theoretical_loss": 3.329201199443448,
+      "tokens_seen": 3247636480
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.5968544374899695e-05,
+      "loss": 1.3843,
+      "theoretical_loss": 3.329191121424821,
+      "tokens_seen": 3247767552
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.592842240410849e-05,
+      "loss": 1.2692,
+      "theoretical_loss": 3.3291810439267886,
+      "tokens_seen": 3247898624
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.5888300433317287e-05,
+      "loss": 1.3193,
+      "theoretical_loss": 3.329170966949301,
+      "tokens_seen": 3248029696
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.584817846252608e-05,
+      "loss": 1.3231,
+      "theoretical_loss": 3.329160890492312,
+      "tokens_seen": 3248160768
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.5808056491734872e-05,
+      "loss": 1.3344,
+      "theoretical_loss": 3.3291508145557724,
+      "tokens_seen": 3248291840
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.576793452094367e-05,
+      "loss": 1.3376,
+      "theoretical_loss": 3.3291407391396355,
+      "tokens_seen": 3248422912
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.572781255015246e-05,
+      "loss": 1.2923,
+      "theoretical_loss": 3.329130664243853,
+      "tokens_seen": 3248553984
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.5687690579361257e-05,
+      "loss": 1.3267,
+      "theoretical_loss": 3.3291205898683764,
+      "tokens_seen": 3248685056
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.5647568608570054e-05,
+      "loss": 1.2697,
+      "theoretical_loss": 3.3291105160131584,
+      "tokens_seen": 3248816128
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.5607446637778846e-05,
+      "loss": 1.3333,
+      "theoretical_loss": 3.329100442678151,
+      "tokens_seen": 3248947200
+    },
+    {
+      "epoch": 0.97,
+      "objective/train/advantage_avg": 0.4940483272075653,
+      "objective/train/docs_used": 1180177,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2070366144180298,
+      "objective/train/original_loss": 1.2070364952087402,
+      "objective/train/theoretical_loss": 3.3290903698633065,
+      "objective/train/tokens_used": 1619537376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24436669051647186,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506471395492554,
+      "objective/train/weighted_lm_loss": 1.2680847644805908,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3290903698633065,
+      "tokens_seen": 3249078272
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.5567324666987642e-05,
+      "loss": 1.3403,
+      "theoretical_loss": 3.3290903698633065,
+      "tokens_seen": 3249078272
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.552720269619644e-05,
+      "loss": 1.3484,
+      "theoretical_loss": 3.329080297568577,
+      "tokens_seen": 3249209344
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.548708072540523e-05,
+      "loss": 1.3079,
+      "theoretical_loss": 3.329070225793915,
+      "tokens_seen": 3249340416
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.5446958754614027e-05,
+      "loss": 1.3473,
+      "theoretical_loss": 3.3290601545392717,
+      "tokens_seen": 3249471488
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.540683678382282e-05,
+      "loss": 1.2859,
+      "theoretical_loss": 3.3290500838046,
+      "tokens_seen": 3249602560
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.5366714813031616e-05,
+      "loss": 1.3412,
+      "theoretical_loss": 3.3290400135898524,
+      "tokens_seen": 3249733632
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.5326592842240413e-05,
+      "loss": 1.2393,
+      "theoretical_loss": 3.32902994389498,
+      "tokens_seen": 3249864704
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.5286470871449205e-05,
+      "loss": 1.3592,
+      "theoretical_loss": 3.329019874719936,
+      "tokens_seen": 3249995776
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.5246348900658001e-05,
+      "loss": 1.3093,
+      "theoretical_loss": 3.329009806064673,
+      "tokens_seen": 3250126848
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.5206226929866794e-05,
+      "loss": 1.294,
+      "theoretical_loss": 3.3289997379291414,
+      "tokens_seen": 3250257920
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.516610495907559e-05,
+      "loss": 1.2952,
+      "theoretical_loss": 3.3289896703132946,
+      "tokens_seen": 3250388992
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.5125982988284385e-05,
+      "loss": 1.1866,
+      "theoretical_loss": 3.3289796032170855,
+      "tokens_seen": 3250520064
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.508586101749318e-05,
+      "loss": 1.2938,
+      "theoretical_loss": 3.328969536640465,
+      "tokens_seen": 3250651136
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.5045739046701975e-05,
+      "loss": 1.36,
+      "theoretical_loss": 3.328959470583386,
+      "tokens_seen": 3250782208
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.500561707591077e-05,
+      "loss": 1.343,
+      "theoretical_loss": 3.3289494050458,
+      "tokens_seen": 3250913280
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.4965495105119563e-05,
+      "loss": 1.359,
+      "theoretical_loss": 3.3289393400276603,
+      "tokens_seen": 3251044352
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.4925373134328359e-05,
+      "loss": 1.3128,
+      "theoretical_loss": 3.328929275528919,
+      "tokens_seen": 3251175424
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.4885251163537153e-05,
+      "loss": 1.2842,
+      "theoretical_loss": 3.3289192115495276,
+      "tokens_seen": 3251306496
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.4845129192745948e-05,
+      "loss": 1.3076,
+      "theoretical_loss": 3.328909148089439,
+      "tokens_seen": 3251437568
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.4805007221954744e-05,
+      "loss": 1.2615,
+      "theoretical_loss": 3.3288990851486053,
+      "tokens_seen": 3251568640
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.4764885251163536e-05,
+      "loss": 1.2725,
+      "theoretical_loss": 3.3288890227269787,
+      "tokens_seen": 3251699712
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.4724763280372331e-05,
+      "loss": 1.2882,
+      "theoretical_loss": 3.3288789608245115,
+      "tokens_seen": 3251830784
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.4684641309581127e-05,
+      "loss": 1.3663,
+      "theoretical_loss": 3.328868899441156,
+      "tokens_seen": 3251961856
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.4644519338789922e-05,
+      "loss": 1.2777,
+      "theoretical_loss": 3.328858838576865,
+      "tokens_seen": 3252092928
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.4604397367998716e-05,
+      "loss": 1.2297,
+      "theoretical_loss": 3.32884877823159,
+      "tokens_seen": 3252224000
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.4564275397207512e-05,
+      "loss": 1.2641,
+      "theoretical_loss": 3.328838718405284,
+      "tokens_seen": 3252355072
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.4524153426416305e-05,
+      "loss": 1.3886,
+      "theoretical_loss": 3.3288286590978986,
+      "tokens_seen": 3252486144
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.44840314556251e-05,
+      "loss": 1.3025,
+      "theoretical_loss": 3.3288186003093863,
+      "tokens_seen": 3252617216
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.4443909484833895e-05,
+      "loss": 1.3759,
+      "theoretical_loss": 3.3288085420397,
+      "tokens_seen": 3252748288
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.440378751404269e-05,
+      "loss": 1.224,
+      "theoretical_loss": 3.3287984842887917,
+      "tokens_seen": 3252879360
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.4363665543251484e-05,
+      "loss": 1.2938,
+      "theoretical_loss": 3.3287884270566135,
+      "tokens_seen": 3253010432
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.432354357246028e-05,
+      "loss": 1.353,
+      "theoretical_loss": 3.3287783703431177,
+      "tokens_seen": 3253141504
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.4283421601669073e-05,
+      "loss": 1.3244,
+      "theoretical_loss": 3.3287683141482574,
+      "tokens_seen": 3253272576
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.4243299630877868e-05,
+      "loss": 1.3116,
+      "theoretical_loss": 3.3287582584719844,
+      "tokens_seen": 3253403648
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.4203177660086664e-05,
+      "loss": 1.2713,
+      "theoretical_loss": 3.328748203314251,
+      "tokens_seen": 3253534720
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.4163055689295458e-05,
+      "loss": 1.3749,
+      "theoretical_loss": 3.3287381486750096,
+      "tokens_seen": 3253665792
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.4122933718504254e-05,
+      "loss": 1.287,
+      "theoretical_loss": 3.328728094554213,
+      "tokens_seen": 3253796864
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.4082811747713049e-05,
+      "loss": 1.2551,
+      "theoretical_loss": 3.328718040951813,
+      "tokens_seen": 3253927936
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.4042689776921842e-05,
+      "loss": 1.2616,
+      "theoretical_loss": 3.328707987867762,
+      "tokens_seen": 3254059008
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.4002567806130638e-05,
+      "loss": 1.4004,
+      "theoretical_loss": 3.3286979353020127,
+      "tokens_seen": 3254190080
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.3962445835339432e-05,
+      "loss": 1.3858,
+      "theoretical_loss": 3.328687883254518,
+      "tokens_seen": 3254321152
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.3922323864548227e-05,
+      "loss": 1.3183,
+      "theoretical_loss": 3.3286778317252295,
+      "tokens_seen": 3254452224
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.3882201893757023e-05,
+      "loss": 1.3413,
+      "theoretical_loss": 3.3286677807141,
+      "tokens_seen": 3254583296
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.3842079922965817e-05,
+      "loss": 1.2997,
+      "theoretical_loss": 3.3286577302210816,
+      "tokens_seen": 3254714368
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.380195795217461e-05,
+      "loss": 1.314,
+      "theoretical_loss": 3.3286476802461267,
+      "tokens_seen": 3254845440
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.3761835981383406e-05,
+      "loss": 1.3862,
+      "theoretical_loss": 3.3286376307891885,
+      "tokens_seen": 3254976512
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.37217140105922e-05,
+      "loss": 1.2642,
+      "theoretical_loss": 3.3286275818502182,
+      "tokens_seen": 3255107584
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.3681592039800995e-05,
+      "loss": 1.3461,
+      "theoretical_loss": 3.3286175334291697,
+      "tokens_seen": 3255238656
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.3641470069009791e-05,
+      "loss": 1.2693,
+      "theoretical_loss": 3.3286074855259944,
+      "tokens_seen": 3255369728
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.3601348098218586e-05,
+      "loss": 1.2807,
+      "theoretical_loss": 3.328597438140645,
+      "tokens_seen": 3255500800
+    },
+    {
+      "epoch": 0.97,
+      "objective/train/advantage_avg": 0.4918806552886963,
+      "objective/train/docs_used": 1182603,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.289878010749817,
+      "objective/train/original_loss": 1.2898778915405273,
+      "objective/train/theoretical_loss": 3.328587391273074,
+      "objective/train/tokens_used": 1626090976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24297910928726196,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050423264503479,
+      "objective/train/weighted_lm_loss": 1.3548123836517334,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9048373699188232,
+      "theoretical_loss": 3.328587391273074,
+      "tokens_seen": 3255631872
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.3561226127427378e-05,
+      "loss": 1.3227,
+      "theoretical_loss": 3.328587391273074,
+      "tokens_seen": 3255631872
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.3521104156636175e-05,
+      "loss": 1.3016,
+      "theoretical_loss": 3.328577344923234,
+      "tokens_seen": 3255762944
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.3480982185844969e-05,
+      "loss": 1.1814,
+      "theoretical_loss": 3.3285672990910773,
+      "tokens_seen": 3255894016
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.3440860215053763e-05,
+      "loss": 1.3189,
+      "theoretical_loss": 3.3285572537765566,
+      "tokens_seen": 3256025088
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.340073824426256e-05,
+      "loss": 1.4003,
+      "theoretical_loss": 3.328547208979624,
+      "tokens_seen": 3256156160
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.3360616273471354e-05,
+      "loss": 1.2605,
+      "theoretical_loss": 3.328537164700233,
+      "tokens_seen": 3256287232
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.3320494302680147e-05,
+      "loss": 1.3118,
+      "theoretical_loss": 3.3285271209383347,
+      "tokens_seen": 3256418304
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.3280372331888943e-05,
+      "loss": 1.3487,
+      "theoretical_loss": 3.328517077693882,
+      "tokens_seen": 3256549376
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.3240250361097737e-05,
+      "loss": 1.3491,
+      "theoretical_loss": 3.328507034966828,
+      "tokens_seen": 3256680448
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.3200128390306532e-05,
+      "loss": 1.401,
+      "theoretical_loss": 3.3284969927571253,
+      "tokens_seen": 3256811520
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.3160006419515328e-05,
+      "loss": 1.3472,
+      "theoretical_loss": 3.328486951064726,
+      "tokens_seen": 3256942592
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.3119884448724122e-05,
+      "loss": 1.2614,
+      "theoretical_loss": 3.328476909889582,
+      "tokens_seen": 3257073664
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.3079762477932915e-05,
+      "loss": 1.2789,
+      "theoretical_loss": 3.3284668692316473,
+      "tokens_seen": 3257204736
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.3039640507141711e-05,
+      "loss": 1.3345,
+      "theoretical_loss": 3.3284568290908734,
+      "tokens_seen": 3257335808
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.2999518536350506e-05,
+      "loss": 1.2884,
+      "theoretical_loss": 3.328446789467213,
+      "tokens_seen": 3257466880
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.29593965655593e-05,
+      "loss": 1.332,
+      "theoretical_loss": 3.328436750360619,
+      "tokens_seen": 3257597952
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.2919274594768096e-05,
+      "loss": 1.309,
+      "theoretical_loss": 3.3284267117710433,
+      "tokens_seen": 3257729024
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.287915262397689e-05,
+      "loss": 1.3001,
+      "theoretical_loss": 3.328416673698439,
+      "tokens_seen": 3257860096
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.2839030653185684e-05,
+      "loss": 1.3254,
+      "theoretical_loss": 3.328406636142759,
+      "tokens_seen": 3257991168
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.279890868239448e-05,
+      "loss": 1.3309,
+      "theoretical_loss": 3.3283965991039555,
+      "tokens_seen": 3258122240
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.2758786711603274e-05,
+      "loss": 1.279,
+      "theoretical_loss": 3.328386562581981,
+      "tokens_seen": 3258253312
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.2718664740812069e-05,
+      "loss": 1.2869,
+      "theoretical_loss": 3.328376526576788,
+      "tokens_seen": 3258384384
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.2678542770020865e-05,
+      "loss": 1.2693,
+      "theoretical_loss": 3.328366491088329,
+      "tokens_seen": 3258515456
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.2638420799229657e-05,
+      "loss": 1.2607,
+      "theoretical_loss": 3.3283564561165577,
+      "tokens_seen": 3258646528
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.2598298828438452e-05,
+      "loss": 1.2063,
+      "theoretical_loss": 3.3283464216614256,
+      "tokens_seen": 3258777600
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.2558176857647248e-05,
+      "loss": 1.2541,
+      "theoretical_loss": 3.328336387722885,
+      "tokens_seen": 3258908672
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.2518054886856043e-05,
+      "loss": 1.2543,
+      "theoretical_loss": 3.32832635430089,
+      "tokens_seen": 3259039744
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.2477932916064837e-05,
+      "loss": 1.2623,
+      "theoretical_loss": 3.328316321395392,
+      "tokens_seen": 3259170816
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.2437810945273633e-05,
+      "loss": 1.2869,
+      "theoretical_loss": 3.328306289006344,
+      "tokens_seen": 3259301888
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.2397688974482426e-05,
+      "loss": 1.3125,
+      "theoretical_loss": 3.328296257133699,
+      "tokens_seen": 3259432960
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.235756700369122e-05,
+      "loss": 1.2304,
+      "theoretical_loss": 3.328286225777409,
+      "tokens_seen": 3259564032
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.2317445032900016e-05,
+      "loss": 1.3619,
+      "theoretical_loss": 3.3282761949374273,
+      "tokens_seen": 3259695104
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.2277323062108811e-05,
+      "loss": 1.3689,
+      "theoretical_loss": 3.328266164613706,
+      "tokens_seen": 3259826176
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.2237201091317605e-05,
+      "loss": 1.2415,
+      "theoretical_loss": 3.3282561348061983,
+      "tokens_seen": 3259957248
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.2197079120526401e-05,
+      "loss": 1.2831,
+      "theoretical_loss": 3.3282461055148564,
+      "tokens_seen": 3260088320
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.2156957149735194e-05,
+      "loss": 1.2203,
+      "theoretical_loss": 3.3282360767396333,
+      "tokens_seen": 3260219392
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.2116835178943989e-05,
+      "loss": 1.2621,
+      "theoretical_loss": 3.3282260484804818,
+      "tokens_seen": 3260350464
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.2076713208152785e-05,
+      "loss": 1.1816,
+      "theoretical_loss": 3.328216020737354,
+      "tokens_seen": 3260481536
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.203659123736158e-05,
+      "loss": 1.2483,
+      "theoretical_loss": 3.3282059935102035,
+      "tokens_seen": 3260612608
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.1996469266570375e-05,
+      "loss": 1.2966,
+      "theoretical_loss": 3.328195966798982,
+      "tokens_seen": 3260743680
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.195634729577917e-05,
+      "loss": 1.2436,
+      "theoretical_loss": 3.328185940603643,
+      "tokens_seen": 3260874752
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.1916225324987963e-05,
+      "loss": 1.2274,
+      "theoretical_loss": 3.328175914924139,
+      "tokens_seen": 3261005824
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.1876103354196759e-05,
+      "loss": 1.3266,
+      "theoretical_loss": 3.3281658897604225,
+      "tokens_seen": 3261136896
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.1835981383405553e-05,
+      "loss": 1.2364,
+      "theoretical_loss": 3.3281558651124463,
+      "tokens_seen": 3261267968
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.1795859412614348e-05,
+      "loss": 1.2342,
+      "theoretical_loss": 3.328145840980164,
+      "tokens_seen": 3261399040
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.1755737441823144e-05,
+      "loss": 1.2782,
+      "theoretical_loss": 3.3281358173635267,
+      "tokens_seen": 3261530112
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.1715615471031938e-05,
+      "loss": 1.2691,
+      "theoretical_loss": 3.3281257942624887,
+      "tokens_seen": 3261661184
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.1675493500240731e-05,
+      "loss": 1.2749,
+      "theoretical_loss": 3.328115771677002,
+      "tokens_seen": 3261792256
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.1635371529449527e-05,
+      "loss": 1.2946,
+      "theoretical_loss": 3.328105749607019,
+      "tokens_seen": 3261923328
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.1595249558658322e-05,
+      "loss": 1.2599,
+      "theoretical_loss": 3.328095728052493,
+      "tokens_seen": 3262054400
+    },
+    {
+      "epoch": 0.98,
+      "objective/train/advantage_avg": 0.49270233511924744,
+      "objective/train/docs_used": 1184948,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.3193860054016113,
+      "objective/train/original_loss": 1.3193860054016113,
+      "objective/train/theoretical_loss": 3.3280857070133774,
+      "objective/train/tokens_used": 1632644576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2431263029575348,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050506353378296,
+      "objective/train/weighted_lm_loss": 1.3858903646469116,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3280857070133774,
+      "tokens_seen": 3262185472
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.1555127587867116e-05,
+      "loss": 1.3525,
+      "theoretical_loss": 3.3280857070133774,
+      "tokens_seen": 3262185472
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.1515005617075912e-05,
+      "loss": 1.3444,
+      "theoretical_loss": 3.328075686489624,
+      "tokens_seen": 3262316544
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.1474883646284707e-05,
+      "loss": 1.3238,
+      "theoretical_loss": 3.328065666481186,
+      "tokens_seen": 3262447616
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.14347616754935e-05,
+      "loss": 1.2917,
+      "theoretical_loss": 3.328055646988016,
+      "tokens_seen": 3262578688
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.1394639704702296e-05,
+      "loss": 1.2564,
+      "theoretical_loss": 3.3280456280100665,
+      "tokens_seen": 3262709760
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.135451773391109e-05,
+      "loss": 1.3871,
+      "theoretical_loss": 3.3280356095472916,
+      "tokens_seen": 3262840832
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.1314395763119884e-05,
+      "loss": 1.3531,
+      "theoretical_loss": 3.328025591599643,
+      "tokens_seen": 3262971904
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.127427379232868e-05,
+      "loss": 1.175,
+      "theoretical_loss": 3.328015574167073,
+      "tokens_seen": 3263102976
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.1234151821537475e-05,
+      "loss": 1.2577,
+      "theoretical_loss": 3.328005557249536,
+      "tokens_seen": 3263234048
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.1194029850746268e-05,
+      "loss": 1.2863,
+      "theoretical_loss": 3.3279955408469837,
+      "tokens_seen": 3263365120
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.1153907879955064e-05,
+      "loss": 1.2352,
+      "theoretical_loss": 3.3279855249593697,
+      "tokens_seen": 3263496192
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.1113785909163858e-05,
+      "loss": 1.2912,
+      "theoretical_loss": 3.327975509586646,
+      "tokens_seen": 3263627264
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.1073663938372653e-05,
+      "loss": 1.2847,
+      "theoretical_loss": 3.327965494728766,
+      "tokens_seen": 3263758336
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.1033541967581449e-05,
+      "loss": 1.2991,
+      "theoretical_loss": 3.3279554803856826,
+      "tokens_seen": 3263889408
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.0993419996790243e-05,
+      "loss": 1.3077,
+      "theoretical_loss": 3.327945466557348,
+      "tokens_seen": 3264020480
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.0953298025999036e-05,
+      "loss": 1.3081,
+      "theoretical_loss": 3.327935453243716,
+      "tokens_seen": 3264151552
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.0913176055207832e-05,
+      "loss": 1.3006,
+      "theoretical_loss": 3.327925440444739,
+      "tokens_seen": 3264282624
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.0873054084416627e-05,
+      "loss": 1.3142,
+      "theoretical_loss": 3.32791542816037,
+      "tokens_seen": 3264413696
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.0832932113625421e-05,
+      "loss": 1.3206,
+      "theoretical_loss": 3.3279054163905615,
+      "tokens_seen": 3264544768
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.0792810142834217e-05,
+      "loss": 1.298,
+      "theoretical_loss": 3.327895405135267,
+      "tokens_seen": 3264675840
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.0752688172043012e-05,
+      "loss": 1.3406,
+      "theoretical_loss": 3.327885394394439,
+      "tokens_seen": 3264806912
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.0712566201251805e-05,
+      "loss": 1.328,
+      "theoretical_loss": 3.3278753841680304,
+      "tokens_seen": 3264937984
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.06724442304606e-05,
+      "loss": 1.2551,
+      "theoretical_loss": 3.3278653744559947,
+      "tokens_seen": 3265069056
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.0632322259669395e-05,
+      "loss": 1.2658,
+      "theoretical_loss": 3.327855365258284,
+      "tokens_seen": 3265200128
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.059220028887819e-05,
+      "loss": 1.3097,
+      "theoretical_loss": 3.327845356574852,
+      "tokens_seen": 3265331200
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.0552078318086986e-05,
+      "loss": 1.3448,
+      "theoretical_loss": 3.3278353484056504,
+      "tokens_seen": 3265462272
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.0511956347295778e-05,
+      "loss": 1.3517,
+      "theoretical_loss": 3.3278253407506337,
+      "tokens_seen": 3265593344
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.0471834376504573e-05,
+      "loss": 1.2686,
+      "theoretical_loss": 3.3278153336097542,
+      "tokens_seen": 3265724416
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.0431712405713369e-05,
+      "loss": 1.2258,
+      "theoretical_loss": 3.3278053269829644,
+      "tokens_seen": 3265855488
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.0391590434922163e-05,
+      "loss": 1.2748,
+      "theoretical_loss": 3.3277953208702176,
+      "tokens_seen": 3265986560
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.0351468464130958e-05,
+      "loss": 1.3457,
+      "theoretical_loss": 3.327785315271467,
+      "tokens_seen": 3266117632
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.0311346493339754e-05,
+      "loss": 1.215,
+      "theoretical_loss": 3.3277753101866656,
+      "tokens_seen": 3266248704
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.0271224522548547e-05,
+      "loss": 1.3401,
+      "theoretical_loss": 3.327765305615766,
+      "tokens_seen": 3266379776
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.0231102551757341e-05,
+      "loss": 1.2882,
+      "theoretical_loss": 3.327755301558721,
+      "tokens_seen": 3266510848
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.0190980580966137e-05,
+      "loss": 1.2753,
+      "theoretical_loss": 3.327745298015484,
+      "tokens_seen": 3266641920
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.0150858610174932e-05,
+      "loss": 1.3259,
+      "theoretical_loss": 3.327735294986008,
+      "tokens_seen": 3266772992
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.0110736639383726e-05,
+      "loss": 1.3917,
+      "theoretical_loss": 3.3277252924702463,
+      "tokens_seen": 3266904064
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.0070614668592522e-05,
+      "loss": 1.3033,
+      "theoretical_loss": 3.3277152904681513,
+      "tokens_seen": 3267035136
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.0030492697801315e-05,
+      "loss": 1.2868,
+      "theoretical_loss": 3.327705288979676,
+      "tokens_seen": 3267166208
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 9.99037072701011e-06,
+      "loss": 1.2179,
+      "theoretical_loss": 3.327695288004774,
+      "tokens_seen": 3267297280
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 9.950248756218906e-06,
+      "loss": 1.3263,
+      "theoretical_loss": 3.3276852875433978,
+      "tokens_seen": 3267428352
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 9.9101267854277e-06,
+      "loss": 1.2743,
+      "theoretical_loss": 3.3276752875955005,
+      "tokens_seen": 3267559424
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 9.870004814636496e-06,
+      "loss": 1.3036,
+      "theoretical_loss": 3.3276652881610356,
+      "tokens_seen": 3267690496
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 9.82988284384529e-06,
+      "loss": 1.3152,
+      "theoretical_loss": 3.3276552892399556,
+      "tokens_seen": 3267821568
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 9.789760873054084e-06,
+      "loss": 1.3634,
+      "theoretical_loss": 3.327645290832214,
+      "tokens_seen": 3267952640
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 9.74963890226288e-06,
+      "loss": 1.2776,
+      "theoretical_loss": 3.3276352929377633,
+      "tokens_seen": 3268083712
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 9.709516931471674e-06,
+      "loss": 1.3035,
+      "theoretical_loss": 3.327625295556557,
+      "tokens_seen": 3268214784
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 9.669394960680469e-06,
+      "loss": 1.3173,
+      "theoretical_loss": 3.3276152986885483,
+      "tokens_seen": 3268345856
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 9.629272989889265e-06,
+      "loss": 1.3102,
+      "theoretical_loss": 3.32760530233369,
+      "tokens_seen": 3268476928
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 9.58915101909806e-06,
+      "loss": 1.1957,
+      "theoretical_loss": 3.3275953064919346,
+      "tokens_seen": 3268608000
+    },
+    {
+      "epoch": 0.98,
+      "objective/train/advantage_avg": 0.49059271812438965,
+      "objective/train/docs_used": 1187457,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.234603762626648,
+      "objective/train/original_loss": 1.2346038818359375,
+      "objective/train/theoretical_loss": 3.327585311163236,
+      "objective/train/tokens_used": 1639198176,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24149785935878754,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0502870082855225,
+      "objective/train/weighted_lm_loss": 1.2965948581695557,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.327585311163236,
+      "tokens_seen": 3268739072
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 9.549029048306852e-06,
+      "loss": 1.3177,
+      "theoretical_loss": 3.327585311163236,
+      "tokens_seen": 3268739072
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 9.508907077515648e-06,
+      "loss": 1.3063,
+      "theoretical_loss": 3.3275753163475477,
+      "tokens_seen": 3268870144
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 9.468785106724443e-06,
+      "loss": 1.3436,
+      "theoretical_loss": 3.327565322044822,
+      "tokens_seen": 3269001216
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 9.428663135933237e-06,
+      "loss": 1.2239,
+      "theoretical_loss": 3.327555328255012,
+      "tokens_seen": 3269132288
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 9.388541165142033e-06,
+      "loss": 1.4545,
+      "theoretical_loss": 3.327545334978071,
+      "tokens_seen": 3269263360
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 9.348419194350828e-06,
+      "loss": 1.2966,
+      "theoretical_loss": 3.327535342213952,
+      "tokens_seen": 3269394432
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 9.30829722355962e-06,
+      "loss": 1.2541,
+      "theoretical_loss": 3.3275253499626083,
+      "tokens_seen": 3269525504
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 9.268175252768416e-06,
+      "loss": 1.3018,
+      "theoretical_loss": 3.3275153582239936,
+      "tokens_seen": 3269656576
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 9.228053281977211e-06,
+      "loss": 1.3535,
+      "theoretical_loss": 3.32750536699806,
+      "tokens_seen": 3269787648
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 9.187931311186005e-06,
+      "loss": 1.2537,
+      "theoretical_loss": 3.327495376284761,
+      "tokens_seen": 3269918720
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 9.147809340394802e-06,
+      "loss": 1.2903,
+      "theoretical_loss": 3.32748538608405,
+      "tokens_seen": 3270049792
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 9.107687369603596e-06,
+      "loss": 1.2475,
+      "theoretical_loss": 3.3274753963958794,
+      "tokens_seen": 3270180864
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 9.067565398812389e-06,
+      "loss": 1.2325,
+      "theoretical_loss": 3.327465407220204,
+      "tokens_seen": 3270311936
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 9.027443428021185e-06,
+      "loss": 1.2113,
+      "theoretical_loss": 3.3274554185569754,
+      "tokens_seen": 3270443008
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 8.98732145722998e-06,
+      "loss": 1.2795,
+      "theoretical_loss": 3.327445430406147,
+      "tokens_seen": 3270574080
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 8.947199486438774e-06,
+      "loss": 1.2276,
+      "theoretical_loss": 3.3274354427676727,
+      "tokens_seen": 3270705152
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 8.90707751564757e-06,
+      "loss": 1.2877,
+      "theoretical_loss": 3.327425455641505,
+      "tokens_seen": 3270836224
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 8.866955544856364e-06,
+      "loss": 1.3277,
+      "theoretical_loss": 3.3274154690275974,
+      "tokens_seen": 3270967296
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 8.826833574065157e-06,
+      "loss": 1.2603,
+      "theoretical_loss": 3.3274054829259034,
+      "tokens_seen": 3271098368
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 8.786711603273953e-06,
+      "loss": 1.2521,
+      "theoretical_loss": 3.3273954973363753,
+      "tokens_seen": 3271229440
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 8.746589632482748e-06,
+      "loss": 1.2923,
+      "theoretical_loss": 3.3273855122589673,
+      "tokens_seen": 3271360512
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 8.706467661691542e-06,
+      "loss": 1.297,
+      "theoretical_loss": 3.327375527693632,
+      "tokens_seen": 3271491584
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 8.666345690900338e-06,
+      "loss": 1.2835,
+      "theoretical_loss": 3.327365543640323,
+      "tokens_seen": 3271622656
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 8.626223720109133e-06,
+      "loss": 1.325,
+      "theoretical_loss": 3.3273555600989932,
+      "tokens_seen": 3271753728
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 8.586101749317925e-06,
+      "loss": 1.2312,
+      "theoretical_loss": 3.327345577069596,
+      "tokens_seen": 3271884800
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 8.545979778526722e-06,
+      "loss": 1.1541,
+      "theoretical_loss": 3.3273355945520846,
+      "tokens_seen": 3272015872
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 8.505857807735516e-06,
+      "loss": 1.3083,
+      "theoretical_loss": 3.327325612546413,
+      "tokens_seen": 3272146944
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 8.46573583694431e-06,
+      "loss": 1.263,
+      "theoretical_loss": 3.3273156310525325,
+      "tokens_seen": 3272278016
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 8.425613866153107e-06,
+      "loss": 1.373,
+      "theoretical_loss": 3.3273056500703984,
+      "tokens_seen": 3272409088
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 8.3854918953619e-06,
+      "loss": 1.3495,
+      "theoretical_loss": 3.327295669599963,
+      "tokens_seen": 3272540160
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 8.345369924570694e-06,
+      "loss": 1.3124,
+      "theoretical_loss": 3.3272856896411795,
+      "tokens_seen": 3272671232
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 8.30524795377949e-06,
+      "loss": 1.3361,
+      "theoretical_loss": 3.3272757101940016,
+      "tokens_seen": 3272802304
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 8.265125982988284e-06,
+      "loss": 1.3014,
+      "theoretical_loss": 3.3272657312583824,
+      "tokens_seen": 3272933376
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 8.225004012197079e-06,
+      "loss": 1.2298,
+      "theoretical_loss": 3.3272557528342754,
+      "tokens_seen": 3273064448
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 8.184882041405875e-06,
+      "loss": 1.2626,
+      "theoretical_loss": 3.3272457749216335,
+      "tokens_seen": 3273195520
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 8.144760070614668e-06,
+      "loss": 1.3008,
+      "theoretical_loss": 3.32723579752041,
+      "tokens_seen": 3273326592
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 8.104638099823462e-06,
+      "loss": 1.3103,
+      "theoretical_loss": 3.3272258206305585,
+      "tokens_seen": 3273457664
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 8.064516129032258e-06,
+      "loss": 1.3422,
+      "theoretical_loss": 3.3272158442520325,
+      "tokens_seen": 3273588736
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 8.024394158241053e-06,
+      "loss": 1.2566,
+      "theoretical_loss": 3.3272058683847847,
+      "tokens_seen": 3273719808
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 7.984272187449847e-06,
+      "loss": 1.3898,
+      "theoretical_loss": 3.327195893028769,
+      "tokens_seen": 3273850880
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 7.944150216658643e-06,
+      "loss": 1.3271,
+      "theoretical_loss": 3.3271859181839383,
+      "tokens_seen": 3273981952
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 7.904028245867436e-06,
+      "loss": 1.2701,
+      "theoretical_loss": 3.3271759438502464,
+      "tokens_seen": 3274113024
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 7.86390627507623e-06,
+      "loss": 1.2247,
+      "theoretical_loss": 3.3271659700276466,
+      "tokens_seen": 3274244096
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 7.823784304285027e-06,
+      "loss": 1.3663,
+      "theoretical_loss": 3.3271559967160913,
+      "tokens_seen": 3274375168
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 7.783662333493821e-06,
+      "loss": 1.2893,
+      "theoretical_loss": 3.3271460239155353,
+      "tokens_seen": 3274506240
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 7.743540362702616e-06,
+      "loss": 1.3012,
+      "theoretical_loss": 3.327136051625931,
+      "tokens_seen": 3274637312
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 7.70341839191141e-06,
+      "loss": 1.2404,
+      "theoretical_loss": 3.3271260798472317,
+      "tokens_seen": 3274768384
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 7.663296421120206e-06,
+      "loss": 1.3204,
+      "theoretical_loss": 3.327116108579392,
+      "tokens_seen": 3274899456
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 7.623174450329001e-06,
+      "loss": 1.3334,
+      "theoretical_loss": 3.327106137822364,
+      "tokens_seen": 3275030528
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 7.583052479537795e-06,
+      "loss": 1.3708,
+      "theoretical_loss": 3.327096167576101,
+      "tokens_seen": 3275161600
+    },
+    {
+      "epoch": 0.98,
+      "objective/train/advantage_avg": 0.49024778604507446,
+      "objective/train/docs_used": 1189805,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2095701694488525,
+      "objective/train/original_loss": 1.2095701694488525,
+      "objective/train/theoretical_loss": 3.3270861978405573,
+      "objective/train/tokens_used": 1645751776,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24122464656829834,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0502511262893677,
+      "objective/train/weighted_lm_loss": 1.270178198814392,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3270861978405573,
+      "tokens_seen": 3275292672
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 7.54293050874659e-06,
+      "loss": 1.2502,
+      "theoretical_loss": 3.3270861978405573,
+      "tokens_seen": 3275292672
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 7.502808537955385e-06,
+      "loss": 1.3157,
+      "theoretical_loss": 3.327076228615686,
+      "tokens_seen": 3275423744
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 7.462686567164179e-06,
+      "loss": 1.2905,
+      "theoretical_loss": 3.3270662599014402,
+      "tokens_seen": 3275554816
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 7.422564596372974e-06,
+      "loss": 1.2811,
+      "theoretical_loss": 3.3270562916977737,
+      "tokens_seen": 3275685888
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 7.382442625581768e-06,
+      "loss": 1.3045,
+      "theoretical_loss": 3.3270463240046393,
+      "tokens_seen": 3275816960
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 7.3423206547905635e-06,
+      "loss": 1.3185,
+      "theoretical_loss": 3.327036356821991,
+      "tokens_seen": 3275948032
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 7.302198683999358e-06,
+      "loss": 1.3355,
+      "theoretical_loss": 3.3270263901497827,
+      "tokens_seen": 3276079104
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 7.2620767132081524e-06,
+      "loss": 1.2666,
+      "theoretical_loss": 3.327016423987967,
+      "tokens_seen": 3276210176
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 7.221954742416948e-06,
+      "loss": 1.3624,
+      "theoretical_loss": 3.3270064583364976,
+      "tokens_seen": 3276341248
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 7.181832771625742e-06,
+      "loss": 1.1546,
+      "theoretical_loss": 3.3269964931953275,
+      "tokens_seen": 3276472320
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 7.141710800834537e-06,
+      "loss": 1.3167,
+      "theoretical_loss": 3.3269865285644107,
+      "tokens_seen": 3276603392
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 7.101588830043332e-06,
+      "loss": 1.26,
+      "theoretical_loss": 3.326976564443701,
+      "tokens_seen": 3276734464
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 7.061466859252127e-06,
+      "loss": 1.2261,
+      "theoretical_loss": 3.326966600833151,
+      "tokens_seen": 3276865536
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 7.021344888460921e-06,
+      "loss": 1.2781,
+      "theoretical_loss": 3.3269566377327147,
+      "tokens_seen": 3276996608
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 6.981222917669716e-06,
+      "loss": 1.2607,
+      "theoretical_loss": 3.3269466751423455,
+      "tokens_seen": 3277127680
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 6.941100946878511e-06,
+      "loss": 1.3108,
+      "theoretical_loss": 3.326936713061997,
+      "tokens_seen": 3277258752
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 6.900978976087305e-06,
+      "loss": 1.2726,
+      "theoretical_loss": 3.326926751491622,
+      "tokens_seen": 3277389824
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 6.8608570052961e-06,
+      "loss": 1.2883,
+      "theoretical_loss": 3.3269167904311754,
+      "tokens_seen": 3277520896
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 6.820735034504896e-06,
+      "loss": 1.2339,
+      "theoretical_loss": 3.3269068298806093,
+      "tokens_seen": 3277651968
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 6.780613063713689e-06,
+      "loss": 1.279,
+      "theoretical_loss": 3.326896869839878,
+      "tokens_seen": 3277783040
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 6.7404910929224845e-06,
+      "loss": 1.3045,
+      "theoretical_loss": 3.326886910308935,
+      "tokens_seen": 3277914112
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 6.70036912213128e-06,
+      "loss": 1.3315,
+      "theoretical_loss": 3.326876951287733,
+      "tokens_seen": 3278045184
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 6.660247151340073e-06,
+      "loss": 1.284,
+      "theoretical_loss": 3.3268669927762264,
+      "tokens_seen": 3278176256
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 6.620125180548869e-06,
+      "loss": 1.3156,
+      "theoretical_loss": 3.3268570347743687,
+      "tokens_seen": 3278307328
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 6.580003209757664e-06,
+      "loss": 1.2925,
+      "theoretical_loss": 3.326847077282113,
+      "tokens_seen": 3278438400
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 6.539881238966458e-06,
+      "loss": 1.3032,
+      "theoretical_loss": 3.3268371202994134,
+      "tokens_seen": 3278569472
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 6.499759268175253e-06,
+      "loss": 1.3017,
+      "theoretical_loss": 3.326827163826223,
+      "tokens_seen": 3278700544
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 6.459637297384048e-06,
+      "loss": 1.2731,
+      "theoretical_loss": 3.3268172078624954,
+      "tokens_seen": 3278831616
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 6.419515326592842e-06,
+      "loss": 1.2958,
+      "theoretical_loss": 3.3268072524081846,
+      "tokens_seen": 3278962688
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 6.379393355801637e-06,
+      "loss": 1.288,
+      "theoretical_loss": 3.3267972974632434,
+      "tokens_seen": 3279093760
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 6.339271385010432e-06,
+      "loss": 1.358,
+      "theoretical_loss": 3.3267873430276262,
+      "tokens_seen": 3279224832
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 6.299149414219226e-06,
+      "loss": 1.3279,
+      "theoretical_loss": 3.3267773891012857,
+      "tokens_seen": 3279355904
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 6.259027443428021e-06,
+      "loss": 1.2948,
+      "theoretical_loss": 3.326767435684176,
+      "tokens_seen": 3279486976
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 6.2189054726368165e-06,
+      "loss": 1.2214,
+      "theoretical_loss": 3.326757482776251,
+      "tokens_seen": 3279618048
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 6.17878350184561e-06,
+      "loss": 1.184,
+      "theoretical_loss": 3.3267475303774643,
+      "tokens_seen": 3279749120
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 6.1386615310544054e-06,
+      "loss": 1.4182,
+      "theoretical_loss": 3.326737578487769,
+      "tokens_seen": 3279880192
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 6.098539560263201e-06,
+      "loss": 1.2806,
+      "theoretical_loss": 3.3267276271071187,
+      "tokens_seen": 3280011264
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 6.058417589471994e-06,
+      "loss": 1.3876,
+      "theoretical_loss": 3.326717676235467,
+      "tokens_seen": 3280142336
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 6.01829561868079e-06,
+      "loss": 1.3363,
+      "theoretical_loss": 3.326707725872768,
+      "tokens_seen": 3280273408
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5.978173647889585e-06,
+      "loss": 1.2099,
+      "theoretical_loss": 3.3266977760189755,
+      "tokens_seen": 3280404480
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5.938051677098379e-06,
+      "loss": 1.3709,
+      "theoretical_loss": 3.3266878266740423,
+      "tokens_seen": 3280535552
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5.897929706307174e-06,
+      "loss": 1.2585,
+      "theoretical_loss": 3.3266778778379225,
+      "tokens_seen": 3280666624
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5.857807735515969e-06,
+      "loss": 1.2955,
+      "theoretical_loss": 3.3266679295105703,
+      "tokens_seen": 3280797696
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5.8176857647247636e-06,
+      "loss": 1.2203,
+      "theoretical_loss": 3.326657981691938,
+      "tokens_seen": 3280928768
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5.777563793933558e-06,
+      "loss": 1.3335,
+      "theoretical_loss": 3.3266480343819804,
+      "tokens_seen": 3281059840
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5.737441823142353e-06,
+      "loss": 1.3314,
+      "theoretical_loss": 3.326638087580651,
+      "tokens_seen": 3281190912
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5.697319852351148e-06,
+      "loss": 1.2275,
+      "theoretical_loss": 3.326628141287903,
+      "tokens_seen": 3281321984
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5.657197881559942e-06,
+      "loss": 1.3085,
+      "theoretical_loss": 3.326618195503691,
+      "tokens_seen": 3281453056
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5.6170759107687375e-06,
+      "loss": 1.2799,
+      "theoretical_loss": 3.3266082502279675,
+      "tokens_seen": 3281584128
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5.576953939977532e-06,
+      "loss": 1.2954,
+      "theoretical_loss": 3.3265983054606867,
+      "tokens_seen": 3281715200
+    },
+    {
+      "epoch": 0.99,
+      "objective/train/advantage_avg": 0.49482399225234985,
+      "objective/train/docs_used": 1192226,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2442878484725952,
+      "objective/train/original_loss": 1.2442878484725952,
+      "objective/train/theoretical_loss": 3.326588361201803,
+      "objective/train/tokens_used": 1652305376,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24509567022323608,
+      "objective/train/value_max": -0.5,
+      "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": 1.3073484897613525,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.326588361201803,
+      "tokens_seen": 3281846272
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5.536831969186326e-06,
+      "loss": 1.2828,
+      "theoretical_loss": 3.326588361201803,
+      "tokens_seen": 3281846272
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5.496709998395122e-06,
+      "loss": 1.3724,
+      "theoretical_loss": 3.3265784174512687,
+      "tokens_seen": 3281977344
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5.456588027603916e-06,
+      "loss": 1.2628,
+      "theoretical_loss": 3.3265684742090387,
+      "tokens_seen": 3282108416
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5.416466056812711e-06,
+      "loss": 1.2773,
+      "theoretical_loss": 3.326558531475066,
+      "tokens_seen": 3282239488
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5.376344086021506e-06,
+      "loss": 1.288,
+      "theoretical_loss": 3.326548589249305,
+      "tokens_seen": 3282370560
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5.3362221152303e-06,
+      "loss": 1.3161,
+      "theoretical_loss": 3.3265386475317094,
+      "tokens_seen": 3282501632
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5.296100144439095e-06,
+      "loss": 1.251,
+      "theoretical_loss": 3.326528706322232,
+      "tokens_seen": 3282632704
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5.255978173647889e-06,
+      "loss": 1.3421,
+      "theoretical_loss": 3.3265187656208273,
+      "tokens_seen": 3282763776
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5.2158562028566845e-06,
+      "loss": 1.315,
+      "theoretical_loss": 3.3265088254274486,
+      "tokens_seen": 3282894848
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5.175734232065479e-06,
+      "loss": 1.3134,
+      "theoretical_loss": 3.3264988857420503,
+      "tokens_seen": 3283025920
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5.135612261274273e-06,
+      "loss": 1.2971,
+      "theoretical_loss": 3.326488946564586,
+      "tokens_seen": 3283156992
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5.095490290483069e-06,
+      "loss": 1.3204,
+      "theoretical_loss": 3.326479007895009,
+      "tokens_seen": 3283288064
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5.055368319691863e-06,
+      "loss": 1.2824,
+      "theoretical_loss": 3.3264690697332737,
+      "tokens_seen": 3283419136
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5.015246348900658e-06,
+      "loss": 1.3263,
+      "theoretical_loss": 3.326459132079333,
+      "tokens_seen": 3283550208
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 4.975124378109453e-06,
+      "loss": 1.2412,
+      "theoretical_loss": 3.3264491949331414,
+      "tokens_seen": 3283681280
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 4.935002407318248e-06,
+      "loss": 1.2932,
+      "theoretical_loss": 3.326439258294653,
+      "tokens_seen": 3283812352
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 4.894880436527042e-06,
+      "loss": 1.2848,
+      "theoretical_loss": 3.3264293221638206,
+      "tokens_seen": 3283943424
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 4.854758465735837e-06,
+      "loss": 1.1764,
+      "theoretical_loss": 3.326419386540598,
+      "tokens_seen": 3284074496
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 4.814636494944632e-06,
+      "loss": 1.1454,
+      "theoretical_loss": 3.3264094514249405,
+      "tokens_seen": 3284205568
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 4.774514524153426e-06,
+      "loss": 1.2297,
+      "theoretical_loss": 3.326399516816801,
+      "tokens_seen": 3284336640
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 4.734392553362221e-06,
+      "loss": 1.2434,
+      "theoretical_loss": 3.3263895827161325,
+      "tokens_seen": 3284467712
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 4.6942705825710166e-06,
+      "loss": 1.3144,
+      "theoretical_loss": 3.32637964912289,
+      "tokens_seen": 3284598784
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 4.65414861177981e-06,
+      "loss": 1.3278,
+      "theoretical_loss": 3.3263697160370267,
+      "tokens_seen": 3284729856
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 4.6140266409886055e-06,
+      "loss": 1.3051,
+      "theoretical_loss": 3.326359783458497,
+      "tokens_seen": 3284860928
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 4.573904670197401e-06,
+      "loss": 1.1683,
+      "theoretical_loss": 3.3263498513872545,
+      "tokens_seen": 3284992000
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 4.533782699406194e-06,
+      "loss": 1.3273,
+      "theoretical_loss": 3.3263399198232526,
+      "tokens_seen": 3285123072
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 4.49366072861499e-06,
+      "loss": 1.2407,
+      "theoretical_loss": 3.3263299887664455,
+      "tokens_seen": 3285254144
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 4.453538757823785e-06,
+      "loss": 1.3014,
+      "theoretical_loss": 3.3263200582167873,
+      "tokens_seen": 3285385216
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 4.4134167870325786e-06,
+      "loss": 1.294,
+      "theoretical_loss": 3.3263101281742316,
+      "tokens_seen": 3285516288
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 4.373294816241374e-06,
+      "loss": 1.3597,
+      "theoretical_loss": 3.326300198638732,
+      "tokens_seen": 3285647360
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 4.333172845450169e-06,
+      "loss": 1.2956,
+      "theoretical_loss": 3.3262902696102428,
+      "tokens_seen": 3285778432
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 4.293050874658963e-06,
+      "loss": 1.3083,
+      "theoretical_loss": 3.3262803410887183,
+      "tokens_seen": 3285909504
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 4.252928903867758e-06,
+      "loss": 1.244,
+      "theoretical_loss": 3.3262704130741114,
+      "tokens_seen": 3286040576
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 4.212806933076553e-06,
+      "loss": 1.2479,
+      "theoretical_loss": 3.326260485566377,
+      "tokens_seen": 3286171648
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 4.172684962285347e-06,
+      "loss": 1.3284,
+      "theoretical_loss": 3.326250558565468,
+      "tokens_seen": 3286302720
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 4.132562991494142e-06,
+      "loss": 1.28,
+      "theoretical_loss": 3.3262406320713387,
+      "tokens_seen": 3286433792
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 4.0924410207029375e-06,
+      "loss": 1.2703,
+      "theoretical_loss": 3.3262307060839436,
+      "tokens_seen": 3286564864
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 4.052319049911731e-06,
+      "loss": 1.2907,
+      "theoretical_loss": 3.326220780603236,
+      "tokens_seen": 3286695936
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 4.012197079120526e-06,
+      "loss": 1.2508,
+      "theoretical_loss": 3.32621085562917,
+      "tokens_seen": 3286827008
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 3.972075108329322e-06,
+      "loss": 1.3656,
+      "theoretical_loss": 3.3262009311616993,
+      "tokens_seen": 3286958080
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 3.931953137538115e-06,
+      "loss": 1.2698,
+      "theoretical_loss": 3.3261910072007783,
+      "tokens_seen": 3287089152
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 3.891831166746911e-06,
+      "loss": 1.3001,
+      "theoretical_loss": 3.3261810837463606,
+      "tokens_seen": 3287220224
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 3.851709195955705e-06,
+      "loss": 1.2468,
+      "theoretical_loss": 3.3261711607984004,
+      "tokens_seen": 3287351296
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 3.8115872251645004e-06,
+      "loss": 1.2234,
+      "theoretical_loss": 3.3261612383568515,
+      "tokens_seen": 3287482368
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 3.771465254373295e-06,
+      "loss": 1.2651,
+      "theoretical_loss": 3.326151316421668,
+      "tokens_seen": 3287613440
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 3.7313432835820897e-06,
+      "loss": 1.2307,
+      "theoretical_loss": 3.326141394992803,
+      "tokens_seen": 3287744512
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 3.691221312790884e-06,
+      "loss": 1.249,
+      "theoretical_loss": 3.326131474070212,
+      "tokens_seen": 3287875584
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 3.651099341999679e-06,
+      "loss": 1.2279,
+      "theoretical_loss": 3.326121553653848,
+      "tokens_seen": 3288006656
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 3.610977371208474e-06,
+      "loss": 1.2967,
+      "theoretical_loss": 3.3261116337436656,
+      "tokens_seen": 3288137728
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 3.5708554004172683e-06,
+      "loss": 1.2191,
+      "theoretical_loss": 3.326101714339618,
+      "tokens_seen": 3288268800
+    },
+    {
+      "epoch": 0.99,
+      "objective/train/advantage_avg": 0.4924720823764801,
+      "objective/train/docs_used": 1194491,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.2059017419815063,
+      "objective/train/original_loss": 1.2059016227722168,
+      "objective/train/theoretical_loss": 3.3260917954416596,
+      "objective/train/tokens_used": 1658858976,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24301908910274506,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504827499389648,
+      "objective/train/weighted_lm_loss": 1.2667547464370728,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.3260917954416596,
+      "tokens_seen": 3288399872
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 3.5307334296260636e-06,
+      "loss": 1.2506,
+      "theoretical_loss": 3.3260917954416596,
+      "tokens_seen": 3288399872
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 3.490611458834858e-06,
+      "loss": 1.2472,
+      "theoretical_loss": 3.3260818770497447,
+      "tokens_seen": 3288530944
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 3.4504894880436525e-06,
+      "loss": 1.3624,
+      "theoretical_loss": 3.326071959163827,
+      "tokens_seen": 3288662016
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 3.410367517252448e-06,
+      "loss": 1.4071,
+      "theoretical_loss": 3.3260620417838602,
+      "tokens_seen": 3288793088
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 3.3702455464612422e-06,
+      "loss": 1.301,
+      "theoretical_loss": 3.326052124909799,
+      "tokens_seen": 3288924160
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 3.3301235756700367e-06,
+      "loss": 1.2761,
+      "theoretical_loss": 3.326042208541597,
+      "tokens_seen": 3289055232
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 3.290001604878832e-06,
+      "loss": 1.3424,
+      "theoretical_loss": 3.3260322926792085,
+      "tokens_seen": 3289186304
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 3.2498796340876264e-06,
+      "loss": 1.2409,
+      "theoretical_loss": 3.3260223773225874,
+      "tokens_seen": 3289317376
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 3.209757663296421e-06,
+      "loss": 1.2707,
+      "theoretical_loss": 3.3260124624716876,
+      "tokens_seen": 3289448448
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 3.169635692505216e-06,
+      "loss": 1.2532,
+      "theoretical_loss": 3.3260025481264637,
+      "tokens_seen": 3289579520
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 3.1295137217140106e-06,
+      "loss": 1.3387,
+      "theoretical_loss": 3.325992634286869,
+      "tokens_seen": 3289710592
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 3.089391750922805e-06,
+      "loss": 1.2348,
+      "theoretical_loss": 3.325982720952858,
+      "tokens_seen": 3289841664
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 3.0492697801316004e-06,
+      "loss": 1.3006,
+      "theoretical_loss": 3.325972808124385,
+      "tokens_seen": 3289972736
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 3.009147809340395e-06,
+      "loss": 1.2686,
+      "theoretical_loss": 3.3259628958014034,
+      "tokens_seen": 3290103808
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 2.9690258385491897e-06,
+      "loss": 1.3085,
+      "theoretical_loss": 3.3259529839838677,
+      "tokens_seen": 3290234880
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 2.9289038677579846e-06,
+      "loss": 1.2535,
+      "theoretical_loss": 3.3259430726717323,
+      "tokens_seen": 3290365952
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 2.888781896966779e-06,
+      "loss": 1.2822,
+      "theoretical_loss": 3.325933161864951,
+      "tokens_seen": 3290497024
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 2.848659926175574e-06,
+      "loss": 1.2874,
+      "theoretical_loss": 3.325923251563477,
+      "tokens_seen": 3290628096
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 2.8085379553843687e-06,
+      "loss": 1.3205,
+      "theoretical_loss": 3.3259133417672664,
+      "tokens_seen": 3290759168
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 2.768415984593163e-06,
+      "loss": 1.2614,
+      "theoretical_loss": 3.3259034324762715,
+      "tokens_seen": 3290890240
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 2.728294013801958e-06,
+      "loss": 1.2704,
+      "theoretical_loss": 3.3258935236904477,
+      "tokens_seen": 3291021312
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 2.688172043010753e-06,
+      "loss": 1.2658,
+      "theoretical_loss": 3.325883615409748,
+      "tokens_seen": 3291152384
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 2.6480500722195474e-06,
+      "loss": 1.279,
+      "theoretical_loss": 3.325873707634128,
+      "tokens_seen": 3291283456
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 2.6079281014283423e-06,
+      "loss": 1.315,
+      "theoretical_loss": 3.32586380036354,
+      "tokens_seen": 3291414528
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 2.5678061306371367e-06,
+      "loss": 1.2791,
+      "theoretical_loss": 3.325853893597939,
+      "tokens_seen": 3291545600
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 2.5276841598459316e-06,
+      "loss": 1.2677,
+      "theoretical_loss": 3.32584398733728,
+      "tokens_seen": 3291676672
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 2.4875621890547264e-06,
+      "loss": 1.2612,
+      "theoretical_loss": 3.325834081581516,
+      "tokens_seen": 3291807744
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 2.447440218263521e-06,
+      "loss": 1.2082,
+      "theoretical_loss": 3.3258241763306016,
+      "tokens_seen": 3291938816
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 2.407318247472316e-06,
+      "loss": 1.2763,
+      "theoretical_loss": 3.3258142715844907,
+      "tokens_seen": 3292069888
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 2.3671962766811106e-06,
+      "loss": 1.2705,
+      "theoretical_loss": 3.325804367343138,
+      "tokens_seen": 3292200960
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 2.327074305889905e-06,
+      "loss": 1.3229,
+      "theoretical_loss": 3.3257944636064973,
+      "tokens_seen": 3292332032
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 2.2869523350987004e-06,
+      "loss": 1.2952,
+      "theoretical_loss": 3.325784560374523,
+      "tokens_seen": 3292463104
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 2.246830364307495e-06,
+      "loss": 1.3724,
+      "theoretical_loss": 3.325774657647169,
+      "tokens_seen": 3292594176
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 2.2067083935162893e-06,
+      "loss": 1.2603,
+      "theoretical_loss": 3.3257647554243897,
+      "tokens_seen": 3292725248
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 2.1665864227250846e-06,
+      "loss": 1.273,
+      "theoretical_loss": 3.3257548537061394,
+      "tokens_seen": 3292856320
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 2.126464451933879e-06,
+      "loss": 1.2705,
+      "theoretical_loss": 3.325744952492372,
+      "tokens_seen": 3292987392
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 2.0863424811426735e-06,
+      "loss": 1.1594,
+      "theoretical_loss": 3.325735051783042,
+      "tokens_seen": 3293118464
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 2.0462205103514688e-06,
+      "loss": 1.2979,
+      "theoretical_loss": 3.3257251515781037,
+      "tokens_seen": 3293249536
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 2.006098539560263e-06,
+      "loss": 1.2209,
+      "theoretical_loss": 3.325715251877511,
+      "tokens_seen": 3293380608
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.9659765687690577e-06,
+      "loss": 1.2841,
+      "theoretical_loss": 3.325705352681218,
+      "tokens_seen": 3293511680
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.9258545979778525e-06,
+      "loss": 1.2547,
+      "theoretical_loss": 3.32569545398918,
+      "tokens_seen": 3293642752
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.8857326271866474e-06,
+      "loss": 1.3619,
+      "theoretical_loss": 3.32568555580135,
+      "tokens_seen": 3293773824
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.845610656395442e-06,
+      "loss": 1.2982,
+      "theoretical_loss": 3.325675658117683,
+      "tokens_seen": 3293904896
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.805488685604237e-06,
+      "loss": 1.2085,
+      "theoretical_loss": 3.325665760938133,
+      "tokens_seen": 3294035968
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.7653667148130318e-06,
+      "loss": 1.2829,
+      "theoretical_loss": 3.3256558642626537,
+      "tokens_seen": 3294167040
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.7252447440218263e-06,
+      "loss": 1.3034,
+      "theoretical_loss": 3.3256459680912003,
+      "tokens_seen": 3294298112
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.6851227732306211e-06,
+      "loss": 1.3435,
+      "theoretical_loss": 3.325636072423727,
+      "tokens_seen": 3294429184
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.645000802439416e-06,
+      "loss": 1.2296,
+      "theoretical_loss": 3.3256261772601876,
+      "tokens_seen": 3294560256
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.6048788316482104e-06,
+      "loss": 1.2283,
+      "theoretical_loss": 3.3256162826005364,
+      "tokens_seen": 3294691328
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.5647568608570053e-06,
+      "loss": 1.2854,
+      "theoretical_loss": 3.325606388444728,
+      "tokens_seen": 3294822400
+    },
+    {
+      "epoch": 1.0,
+      "objective/train/advantage_avg": 0.4926229417324066,
+      "objective/train/docs_used": 1196765,
+      "objective/train/instantaneous_batch_size": 64,
+      "objective/train/instantaneous_microbatch_size": 65536,
+      "objective/train/lm_loss": 1.166650414466858,
+      "objective/train/original_loss": 1.166650414466858,
+      "objective/train/theoretical_loss": 3.325596494792717,
+      "objective/train/tokens_used": 1665412576,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24339522421360016,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050499677658081,
+      "objective/train/weighted_lm_loss": 1.225674033164978,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 3.325596494792717,
+      "tokens_seen": 3294953472
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.5246348900658002e-06,
+      "loss": 1.3226,
+      "theoretical_loss": 3.325596494792717,
+      "tokens_seen": 3294953472
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.4845129192745948e-06,
+      "loss": 1.3186,
+      "theoretical_loss": 3.325586601644457,
+      "tokens_seen": 3295084544
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.4443909484833895e-06,
+      "loss": 1.2148,
+      "theoretical_loss": 3.325576708999902,
+      "tokens_seen": 3295215616
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.4042689776921844e-06,
+      "loss": 1.2793,
+      "theoretical_loss": 3.325566816859008,
+      "tokens_seen": 3295346688
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.364147006900979e-06,
+      "loss": 1.2784,
+      "theoretical_loss": 3.3255569252217274,
+      "tokens_seen": 3295477760
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.3240250361097737e-06,
+      "loss": 1.2645,
+      "theoretical_loss": 3.325547034088016,
+      "tokens_seen": 3295608832
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.2839030653185684e-06,
+      "loss": 1.2477,
+      "theoretical_loss": 3.325537143457827,
+      "tokens_seen": 3295739904
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.2437810945273632e-06,
+      "loss": 1.2585,
+      "theoretical_loss": 3.3255272533311158,
+      "tokens_seen": 3295870976
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.203659123736158e-06,
+      "loss": 1.3062,
+      "theoretical_loss": 3.325517363707836,
+      "tokens_seen": 3296002048
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.1635371529449525e-06,
+      "loss": 1.3083,
+      "theoretical_loss": 3.3255074745879423,
+      "tokens_seen": 3296133120
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.1234151821537474e-06,
+      "loss": 1.2942,
+      "theoretical_loss": 3.325497585971389,
+      "tokens_seen": 3296264192
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.0832932113625423e-06,
+      "loss": 1.3123,
+      "theoretical_loss": 3.3254876978581303,
+      "tokens_seen": 3296395264
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.0431712405713367e-06,
+      "loss": 1.3352,
+      "theoretical_loss": 3.3254778102481204,
+      "tokens_seen": 3296526336
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.0030492697801316e-06,
+      "loss": 1.2987,
+      "theoretical_loss": 3.3254679231413142,
+      "tokens_seen": 3296657408
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 9.629272989889263e-07,
+      "loss": 1.3163,
+      "theoretical_loss": 3.325458036537666,
+      "tokens_seen": 3296788480
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 9.22805328197721e-07,
+      "loss": 1.3349,
+      "theoretical_loss": 3.32544815043713,
+      "tokens_seen": 3296919552
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 8.826833574065159e-07,
+      "loss": 1.2707,
+      "theoretical_loss": 3.3254382648396605,
+      "tokens_seen": 3297050624
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 8.425613866153106e-07,
+      "loss": 1.2712,
+      "theoretical_loss": 3.325428379745212,
+      "tokens_seen": 3297181696
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 8.024394158241052e-07,
+      "loss": 1.2873,
+      "theoretical_loss": 3.3254184951537393,
+      "tokens_seen": 3297312768
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 7.623174450329001e-07,
+      "loss": 1.2996,
+      "theoretical_loss": 3.325408611065196,
+      "tokens_seen": 3297443840
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 7.221954742416948e-07,
+      "loss": 1.267,
+      "theoretical_loss": 3.3253987274795374,
+      "tokens_seen": 3297574912
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 6.820735034504895e-07,
+      "loss": 1.2445,
+      "theoretical_loss": 3.3253888443967172,
+      "tokens_seen": 3297705984
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 6.419515326592842e-07,
+      "loss": 1.2272,
+      "theoretical_loss": 3.3253789618166905,
+      "tokens_seen": 3297837056
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 6.01829561868079e-07,
+      "loss": 1.2925,
+      "theoretical_loss": 3.325369079739411,
+      "tokens_seen": 3297968128
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 5.617075910768737e-07,
+      "loss": 1.2917,
+      "theoretical_loss": 3.3253591981648336,
+      "tokens_seen": 3298099200
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 5.215856202856684e-07,
+      "loss": 1.3077,
+      "theoretical_loss": 3.325349317092913,
+      "tokens_seen": 3298230272
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 4.814636494944631e-07,
+      "loss": 1.2171,
+      "theoretical_loss": 3.325339436523603,
+      "tokens_seen": 3298361344
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 4.4134167870325795e-07,
+      "loss": 1.2613,
+      "theoretical_loss": 3.3253295564568583,
+      "tokens_seen": 3298492416
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 4.012197079120526e-07,
+      "loss": 1.3622,
+      "theoretical_loss": 3.325319676892634,
+      "tokens_seen": 3298623488
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 3.610977371208474e-07,
+      "loss": 1.2959,
+      "theoretical_loss": 3.325309797830884,
+      "tokens_seen": 3298754560
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 3.209757663296421e-07,
+      "loss": 1.2886,
+      "theoretical_loss": 3.325299919271562,
+      "tokens_seen": 3298885632
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 2.8085379553843685e-07,
+      "loss": 1.2627,
+      "theoretical_loss": 3.325290041214624,
+      "tokens_seen": 3299016704
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 2.4073182474723157e-07,
+      "loss": 1.3106,
+      "theoretical_loss": 3.3252801636600235,
+      "tokens_seen": 3299147776
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 2.006098539560263e-07,
+      "loss": 1.2343,
+      "theoretical_loss": 3.3252702866077155,
+      "tokens_seen": 3299278848
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.6048788316482104e-07,
+      "loss": 1.3334,
+      "theoretical_loss": 3.3252604100576546,
+      "tokens_seen": 3299409920
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.2036591237361578e-07,
+      "loss": 1.2939,
+      "theoretical_loss": 3.3252505340097946,
+      "tokens_seen": 3299540992
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 8.024394158241052e-08,
+      "loss": 1.3045,
+      "theoretical_loss": 3.3252406584640903,
+      "tokens_seen": 3299672064
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 4.012197079120526e-08,
+      "loss": 1.2735,
+      "theoretical_loss": 3.325230783420497,
+      "tokens_seen": 3299803136
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.0,
+      "loss": 1.255,
+      "theoretical_loss": 3.325220908878968,
+      "tokens_seen": 3299934208
+    }
+  ],
+  "max_steps": 12588,
+  "num_train_epochs": 9223372036854775807,
+  "total_flos": 8.420284921943163e+17,
+  "trial_name": null,
+  "trial_params": null
+}