{
  "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": 7.936507936507936e-06,
      "loss": 1.0932,
      "theoretical_loss": 14.920781838632275,
      "tokens_seen": 262144
    },
    {
      "epoch": 0.0,
      "learning_rate": 1.5873015873015872e-05,
      "loss": 1.0937,
      "theoretical_loss": 12.718594708127029,
      "tokens_seen": 524288
    },
    {
      "epoch": 0.0,
      "learning_rate": 2.380952380952381e-05,
      "loss": 1.0519,
      "theoretical_loss": 11.615184291350435,
      "tokens_seen": 786432
    },
    {
      "epoch": 0.0,
      "learning_rate": 3.1746031746031745e-05,
      "loss": 1.0059,
      "theoretical_loss": 10.904893169100655,
      "tokens_seen": 1048576
    },
    {
      "epoch": 0.0,
      "learning_rate": 3.968253968253968e-05,
      "loss": 0.9761,
      "theoretical_loss": 10.392029026407034,
      "tokens_seen": 1310720
    },
    {
      "epoch": 0.0,
      "learning_rate": 4.761904761904762e-05,
      "loss": 0.9521,
      "theoretical_loss": 9.996134261483984,
      "tokens_seen": 1572864
    },
    {
      "epoch": 0.0,
      "learning_rate": 5.555555555555555e-05,
      "loss": 0.9364,
      "theoretical_loss": 9.67682184172525,
      "tokens_seen": 1835008
    },
    {
      "epoch": 0.0,
      "learning_rate": 6.349206349206349e-05,
      "loss": 0.9245,
      "theoretical_loss": 9.41114487355416,
      "tokens_seen": 2097152
    },
    {
      "epoch": 0.0,
      "learning_rate": 7.142857142857142e-05,
      "loss": 0.9103,
      "theoretical_loss": 9.184905895151996,
      "tokens_seen": 2359296
    },
    {
      "epoch": 0.0,
      "learning_rate": 7.936507936507937e-05,
      "loss": 0.8933,
      "theoretical_loss": 8.988754572553061,
      "tokens_seen": 2621440
    },
    {
      "epoch": 0.0,
      "learning_rate": 8.73015873015873e-05,
      "loss": 0.8721,
      "theoretical_loss": 8.816230875422118,
      "tokens_seen": 2883584
    },
    {
      "epoch": 0.0,
      "learning_rate": 9.523809523809524e-05,
      "loss": 0.8552,
      "theoretical_loss": 8.66269920037918,
      "tokens_seen": 3145728
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00010317460317460317,
      "loss": 0.8283,
      "theoretical_loss": 8.524729102289708,
      "tokens_seen": 3407872
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0001111111111111111,
      "loss": 0.8016,
      "theoretical_loss": 8.399716359763914,
      "tokens_seen": 3670016
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00011904761904761905,
      "loss": 0.7709,
      "theoretical_loss": 8.285641004895568,
      "tokens_seen": 3932160
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00012698412698412698,
      "loss": 0.7402,
      "theoretical_loss": 8.180907195283321,
      "tokens_seen": 4194304
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0001349206349206349,
      "loss": 0.7102,
      "theoretical_loss": 8.084233979345122,
      "tokens_seen": 4456448
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00014285714285714284,
      "loss": 0.6751,
      "theoretical_loss": 7.9945788049155055,
      "tokens_seen": 4718592
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0001507936507936508,
      "loss": 0.6503,
      "theoretical_loss": 7.911082722632908,
      "tokens_seen": 4980736
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00015873015873015873,
      "loss": 0.6184,
      "theoretical_loss": 7.83303033759787,
      "tokens_seen": 5242880
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00016666666666666666,
      "loss": 0.5919,
      "theoretical_loss": 7.759820016443023,
      "tokens_seen": 5505024
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0001746031746031746,
      "loss": 0.5615,
      "theoretical_loss": 7.690941370375033,
      "tokens_seen": 5767168
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00018253968253968252,
      "loss": 0.5314,
      "theoretical_loss": 7.6259579939239845,
      "tokens_seen": 6029312
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00019047619047619048,
      "loss": 0.5049,
      "theoretical_loss": 7.564494061943624,
      "tokens_seen": 6291456
    },
    {
      "epoch": 0.0,
      "objective/train/advantage_avg": 0.19619600474834442,
      "objective/train/docs_used": 9704,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 7.333767890930176,
      "objective/train/original_loss": 7.333766937255859,
      "objective/train/theoretical_loss": 7.5062238006917354,
      "objective/train/tokens_used": 27013600,
      "objective/train/value_avg": -0.204345703125,
      "objective/train/value_loss": 0.03900672122836113,
      "objective/train/value_max": -0.19873046875,
      "objective/train/value_min": -0.251953125,
      "objective/train/value_reward_corr": -0.012004978426084556,
      "objective/train/value_std": 0.01258087158203125,
      "objective/train/weight_avg": 1.2170684337615967,
      "objective/train/weighted_lm_loss": 8.908797264099121,
      "objective/train/weights_max": 1.2865357398986816,
      "objective/train/weights_min": 0.4672624170780182,
      "theoretical_loss": 7.5062238006917354,
      "tokens_seen": 6553600
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0001984126984126984,
      "loss": 0.4798,
      "theoretical_loss": 7.5062238006917354,
      "tokens_seen": 6553600
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00020634920634920634,
      "loss": 0.4565,
      "theoretical_loss": 7.45086312850561,
      "tokens_seen": 6815744
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00021428571428571427,
      "loss": 0.4398,
      "theoretical_loss": 7.398162954262078,
      "tokens_seen": 7077888
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0002222222222222222,
      "loss": 0.4258,
      "theoretical_loss": 7.347903756717382,
      "tokens_seen": 7340032
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00023015873015873016,
      "loss": 0.4009,
      "theoretical_loss": 7.299891163694537,
      "tokens_seen": 7602176
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0002380952380952381,
      "loss": 0.391,
      "theoretical_loss": 7.253952319156202,
      "tokens_seen": 7864320
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.000246031746031746,
      "loss": 0.3753,
      "theoretical_loss": 7.2099328765932205,
      "tokens_seen": 8126464
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00025396825396825396,
      "loss": 0.366,
      "theoretical_loss": 7.167694494355343,
      "tokens_seen": 8388608
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0002619047619047619,
      "loss": 0.3573,
      "theoretical_loss": 7.127112736305475,
      "tokens_seen": 8650752
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0002698412698412698,
      "loss": 0.3476,
      "theoretical_loss": 7.0880753020982725,
      "tokens_seen": 8912896
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0002777777777777778,
      "loss": 0.3461,
      "theoretical_loss": 7.050480527300383,
      "tokens_seen": 9175040
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0002857142857142857,
      "loss": 0.3395,
      "theoretical_loss": 7.014236105786485,
      "tokens_seen": 9437184
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0002936507936507937,
      "loss": 0.3386,
      "theoretical_loss": 6.979257996300014,
      "tokens_seen": 9699328
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0003015873015873016,
      "loss": 0.3357,
      "theoretical_loss": 6.945469482441503,
      "tokens_seen": 9961472
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00030952380952380956,
      "loss": 0.3318,
      "theoretical_loss": 6.912800361140576,
      "tokens_seen": 10223616
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00031746031746031746,
      "loss": 0.33,
      "theoretical_loss": 6.881186239250335,
      "tokens_seen": 10485760
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0003253968253968254,
      "loss": 0.3287,
      "theoretical_loss": 6.8505679215514235,
      "tokens_seen": 10747904
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0003333333333333333,
      "loss": 0.3244,
      "theoretical_loss": 6.8208908763759295,
      "tokens_seen": 11010048
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0003412698412698413,
      "loss": 0.3241,
      "theoretical_loss": 6.79210476741633,
      "tokens_seen": 11272192
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0003492063492063492,
      "loss": 0.3206,
      "theoretical_loss": 6.76416304219278,
      "tokens_seen": 11534336
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00035714285714285714,
      "loss": 0.3173,
      "theoretical_loss": 6.737022569206117,
      "tokens_seen": 11796480
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00036507936507936505,
      "loss": 0.3156,
      "theoretical_loss": 6.710643317075979,
      "tokens_seen": 12058624
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.000373015873015873,
      "loss": 0.3142,
      "theoretical_loss": 6.684988070009584,
      "tokens_seen": 12320768
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00038095238095238096,
      "loss": 0.3193,
      "theoretical_loss": 6.660022174811009,
      "tokens_seen": 12582912
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0003888888888888889,
      "loss": 0.3096,
      "theoretical_loss": 6.6357133153579175,
      "tokens_seen": 12845056
    },
    {
      "epoch": 0.0,
      "objective/train/advantage_avg": 0.028042539954185486,
      "objective/train/docs_used": 12170,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 6.02435827255249,
      "objective/train/original_loss": 6.02435827255249,
      "objective/train/theoretical_loss": 6.612031311070119,
      "objective/train/tokens_used": 33567200,
      "objective/train/value_avg": -0.034576416015625,
      "objective/train/value_loss": 0.0012018069392070174,
      "objective/train/value_max": -0.01337432861328125,
      "objective/train/value_min": -0.052825927734375,
      "objective/train/value_reward_corr": -0.01474424872782136,
      "objective/train/value_std": 0.005672454833984375,
      "objective/train/weight_avg": 1.0286412239074707,
      "objective/train/weighted_lm_loss": 6.1940083503723145,
      "objective/train/weights_max": 1.054246187210083,
      "objective/train/weights_min": 0.3763794004917145,
      "theoretical_loss": 6.612031311070119,
      "tokens_seen": 13107200
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0003968253968253968,
      "loss": 0.309,
      "theoretical_loss": 6.612031311070119,
      "tokens_seen": 13107200
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0004047619047619048,
      "loss": 0.3093,
      "theoretical_loss": 6.588947936394168,
      "tokens_seen": 13369344
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0004126984126984127,
      "loss": 0.303,
      "theoretical_loss": 6.566436758747731,
      "tokens_seen": 13631488
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00042063492063492065,
      "loss": 0.3026,
      "theoretical_loss": 6.544472992721121,
      "tokens_seen": 13893632
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00042857142857142855,
      "loss": 0.3027,
      "theoretical_loss": 6.523033368632323,
      "tokens_seen": 14155776
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0004365079365079365,
      "loss": 0.3037,
      "theoretical_loss": 6.502096013785574,
      "tokens_seen": 14417920
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0004444444444444444,
      "loss": 0.2987,
      "theoretical_loss": 6.481640344999435,
      "tokens_seen": 14680064
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00045238095238095237,
      "loss": 0.3013,
      "theoretical_loss": 6.461646971154669,
      "tokens_seen": 14942208
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00046031746031746033,
      "loss": 0.3009,
      "theoretical_loss": 6.442097604670096,
      "tokens_seen": 15204352
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0004682539682539683,
      "loss": 0.291,
      "theoretical_loss": 6.422974980950157,
      "tokens_seen": 15466496
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0004761904761904762,
      "loss": 0.295,
      "theoretical_loss": 6.404262784964672,
      "tokens_seen": 15728640
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00048412698412698415,
      "loss": 0.2957,
      "theoretical_loss": 6.3859455842220765,
      "tokens_seen": 15990784
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.000492063492063492,
      "loss": 0.2955,
      "theoretical_loss": 6.368008767484675,
      "tokens_seen": 16252928
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0005,
      "loss": 0.2931,
      "theoretical_loss": 6.350438488650175,
      "tokens_seen": 16515072
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0005079365079365079,
      "loss": 0.2938,
      "theoretical_loss": 6.333221615289645,
      "tokens_seen": 16777216
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0005158730158730159,
      "loss": 0.2898,
      "theoretical_loss": 6.316345681389436,
      "tokens_seen": 17039360
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0005238095238095238,
      "loss": 0.2847,
      "theoretical_loss": 6.2997988438948465,
      "tokens_seen": 17301504
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0005317460317460317,
      "loss": 0.2833,
      "theoretical_loss": 6.283569842697203,
      "tokens_seen": 17563648
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0005396825396825396,
      "loss": 0.2873,
      "theoretical_loss": 6.26764796374462,
      "tokens_seen": 17825792
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0005476190476190477,
      "loss": 0.2833,
      "theoretical_loss": 6.25202300499066,
      "tokens_seen": 18087936
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0005555555555555556,
      "loss": 0.284,
      "theoretical_loss": 6.236685244924882,
      "tokens_seen": 18350080
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0005634920634920635,
      "loss": 0.2831,
      "theoretical_loss": 6.2216254134558024,
      "tokens_seen": 18612224
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0005714285714285714,
      "loss": 0.2838,
      "theoretical_loss": 6.206834664939976,
      "tokens_seen": 18874368
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0005793650793650794,
      "loss": 0.2854,
      "theoretical_loss": 6.192304553171669,
      "tokens_seen": 19136512
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0005873015873015874,
      "loss": 0.2811,
      "theoretical_loss": 6.178027008165916,
      "tokens_seen": 19398656
    },
    {
      "epoch": 0.01,
      "objective/train/advantage_avg": 0.023865671828389168,
      "objective/train/docs_used": 14656,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 5.437821865081787,
      "objective/train/original_loss": 5.437821388244629,
      "objective/train/theoretical_loss": 6.163994314584031,
      "objective/train/tokens_used": 40120800,
      "objective/train/value_avg": -0.032073974609375,
      "objective/train/value_loss": 0.0011683752527460456,
      "objective/train/value_max": -0.00933837890625,
      "objective/train/value_min": -0.06573486328125,
      "objective/train/value_reward_corr": 0.06998308804235678,
      "objective/train/value_std": 0.01036834716796875,
      "objective/train/weight_avg": 1.0244362354278564,
      "objective/train/weighted_lm_loss": 5.583455562591553,
      "objective/train/weights_max": 1.0648194551467896,
      "objective/train/weights_min": 0.37916356325149536,
      "theoretical_loss": 6.163994314584031,
      "tokens_seen": 19660800
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0005952380952380953,
      "loss": 0.282,
      "theoretical_loss": 6.163994314584031,
      "tokens_seen": 19660800
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0006031746031746032,
      "loss": 0.2786,
      "theoretical_loss": 6.150199091665225,
      "tokens_seen": 19922944
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0006111111111111112,
      "loss": 0.2766,
      "theoretical_loss": 6.136634274540901,
      "tokens_seen": 20185088
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0006190476190476191,
      "loss": 0.279,
      "theoretical_loss": 6.123293096819758,
      "tokens_seen": 20447232
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000626984126984127,
      "loss": 0.2761,
      "theoretical_loss": 6.1101690743422505,
      "tokens_seen": 20709376
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0006349206349206349,
      "loss": 0.2751,
      "theoretical_loss": 6.097255990012153,
      "tokens_seen": 20971520
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0006428571428571429,
      "loss": 0.2728,
      "theoretical_loss": 6.084547879621354,
      "tokens_seen": 21233664
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0006507936507936508,
      "loss": 0.2741,
      "theoretical_loss": 6.072039018591484,
      "tokens_seen": 21495808
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0006587301587301587,
      "loss": 0.2744,
      "theoretical_loss": 6.059723909562683,
      "tokens_seen": 21757952
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0006666666666666666,
      "loss": 0.2699,
      "theoretical_loss": 6.047597270765904,
      "tokens_seen": 22020096
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0006746031746031747,
      "loss": 0.271,
      "theoretical_loss": 6.035654025120612,
      "tokens_seen": 22282240
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0006825396825396826,
      "loss": 0.2683,
      "theoretical_loss": 6.023889290004692,
      "tokens_seen": 22544384
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0006904761904761905,
      "loss": 0.2678,
      "theoretical_loss": 6.012298367647816,
      "tokens_seen": 22806528
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0006984126984126984,
      "loss": 0.2651,
      "theoretical_loss": 6.000876736103618,
      "tokens_seen": 23068672
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0007063492063492064,
      "loss": 0.2683,
      "theoretical_loss": 5.989620040759641,
      "tokens_seen": 23330816
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0007142857142857143,
      "loss": 0.2624,
      "theoretical_loss": 5.978524086347409,
      "tokens_seen": 23592960
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0007222222222222222,
      "loss": 0.2659,
      "theoretical_loss": 5.967584829417934,
      "tokens_seen": 23855104
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0007301587301587301,
      "loss": 0.2616,
      "theoretical_loss": 5.956798371250791,
      "tokens_seen": 24117248
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0007380952380952381,
      "loss": 0.2583,
      "theoretical_loss": 5.9461609511673625,
      "tokens_seen": 24379392
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000746031746031746,
      "loss": 0.2625,
      "theoretical_loss": 5.935668940221127,
      "tokens_seen": 24641536
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000753968253968254,
      "loss": 0.261,
      "theoretical_loss": 5.92531883523999,
      "tokens_seen": 24903680
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0007619047619047619,
      "loss": 0.26,
      "theoretical_loss": 5.915107253197538,
      "tokens_seen": 25165824
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0007698412698412699,
      "loss": 0.2623,
      "theoretical_loss": 5.905030925891829,
      "tokens_seen": 25427968
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0007777777777777778,
      "loss": 0.2632,
      "theoretical_loss": 5.895086694911951,
      "tokens_seen": 25690112
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0007857142857142857,
      "loss": 0.2608,
      "theoretical_loss": 5.88527150687402,
      "tokens_seen": 25952256
    },
    {
      "epoch": 0.01,
      "objective/train/advantage_avg": 0.016880027949810028,
      "objective/train/docs_used": 17033,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 5.177478313446045,
      "objective/train/original_loss": 5.177477836608887,
      "objective/train/theoretical_loss": 5.8755824089096285,
      "objective/train/tokens_used": 46674400,
      "objective/train/value_avg": -0.0229949951171875,
      "objective/train/value_loss": 0.000880128238350153,
      "objective/train/value_max": -0.00821685791015625,
      "objective/train/value_min": -0.06536865234375,
      "objective/train/value_reward_corr": 0.07602733877190453,
      "objective/train/value_std": 0.00830841064453125,
      "objective/train/weight_avg": 1.017284870147705,
      "objective/train/weighted_lm_loss": 5.27687406539917,
      "objective/train/weights_max": 1.0675524473190308,
      "objective/train/weights_min": 0.3755991756916046,
      "theoretical_loss": 5.8755824089096285,
      "tokens_seen": 26214400
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0007936507936507937,
      "loss": 0.2611,
      "theoretical_loss": 5.8755824089096285,
      "tokens_seen": 26214400
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0008015873015873017,
      "loss": 0.2562,
      "theoretical_loss": 5.866016544391016,
      "tokens_seen": 26476544
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0008095238095238096,
      "loss": 0.253,
      "theoretical_loss": 5.856571148878293,
      "tokens_seen": 26738688
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0008174603174603175,
      "loss": 0.2532,
      "theoretical_loss": 5.847243546275179,
      "tokens_seen": 27000832
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0008253968253968254,
      "loss": 0.2524,
      "theoretical_loss": 5.838031145180573,
      "tokens_seen": 27262976
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0008333333333333334,
      "loss": 0.257,
      "theoretical_loss": 5.82893143542425,
      "tokens_seen": 27525120
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0008412698412698413,
      "loss": 0.253,
      "theoretical_loss": 5.81994198477569,
      "tokens_seen": 27787264
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0008492063492063492,
      "loss": 0.2507,
      "theoretical_loss": 5.811060435815881,
      "tokens_seen": 28049408
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0008571428571428571,
      "loss": 0.2532,
      "theoretical_loss": 5.802284502962563,
      "tokens_seen": 28311552
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0008650793650793651,
      "loss": 0.2512,
      "theoretical_loss": 5.793611969640068,
      "tokens_seen": 28573696
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000873015873015873,
      "loss": 0.2509,
      "theoretical_loss": 5.785040685585437,
      "tokens_seen": 28835840
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0008809523809523809,
      "loss": 0.2497,
      "theoretical_loss": 5.7765685642831155,
      "tokens_seen": 29097984
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0008888888888888888,
      "loss": 0.2538,
      "theoretical_loss": 5.768193580520972,
      "tokens_seen": 29360128
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0008968253968253968,
      "loss": 0.2495,
      "theoretical_loss": 5.759913768060882,
      "tokens_seen": 29622272
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009047619047619047,
      "loss": 0.2494,
      "theoretical_loss": 5.7517272174175496,
      "tokens_seen": 29884416
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009126984126984126,
      "loss": 0.2498,
      "theoretical_loss": 5.743632073739626,
      "tokens_seen": 30146560
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009206349206349207,
      "loss": 0.2478,
      "theoretical_loss": 5.735626534787584,
      "tokens_seen": 30408704
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009285714285714287,
      "loss": 0.2429,
      "theoretical_loss": 5.727708849003127,
      "tokens_seen": 30670848
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009365079365079366,
      "loss": 0.2443,
      "theoretical_loss": 5.719877313665254,
      "tokens_seen": 30932992
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009444444444444445,
      "loss": 0.2456,
      "theoretical_loss": 5.712130273128388,
      "tokens_seen": 31195136
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009523809523809524,
      "loss": 0.2472,
      "theoretical_loss": 5.704466117138258,
      "tokens_seen": 31457280
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009603174603174604,
      "loss": 0.2447,
      "theoretical_loss": 5.696883279221504,
      "tokens_seen": 31719424
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009682539682539683,
      "loss": 0.2421,
      "theoretical_loss": 5.689380235145171,
      "tokens_seen": 31981568
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009761904761904762,
      "loss": 0.2447,
      "theoretical_loss": 5.6819555014425305,
      "tokens_seen": 32243712
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000984126984126984,
      "loss": 0.2412,
      "theoretical_loss": 5.674607634001871,
      "tokens_seen": 32505856
    },
    {
      "epoch": 0.01,
      "objective/train/advantage_avg": 0.012133671902120113,
      "objective/train/docs_used": 19239,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 4.836267948150635,
      "objective/train/original_loss": 4.836267471313477,
      "objective/train/theoretical_loss": 5.667335226715059,
      "objective/train/tokens_used": 53228000,
      "objective/train/value_avg": -0.0186767578125,
      "objective/train/value_loss": 0.0005454737693071365,
      "objective/train/value_max": -0.00572967529296875,
      "objective/train/value_min": -0.0872802734375,
      "objective/train/value_reward_corr": 0.07709606794972773,
      "objective/train/value_std": 0.0084075927734375,
      "objective/train/weight_avg": 1.0123944282531738,
      "objective/train/weighted_lm_loss": 4.902460098266602,
      "objective/train/weights_max": 1.077897310256958,
      "objective/train/weights_min": 0.3729293644428253,
      "theoretical_loss": 5.667335226715059,
      "tokens_seen": 32768000
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000992063492063492,
      "loss": 0.2475,
      "theoretical_loss": 5.667335226715059,
      "tokens_seen": 32768000
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.001,
      "loss": 0.243,
      "theoretical_loss": 5.6601369101828904,
      "tokens_seen": 33030144
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009999197560584176,
      "loss": 0.2423,
      "theoretical_loss": 5.6530113504744435,
      "tokens_seen": 33292288
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009998395121168352,
      "loss": 0.2419,
      "theoretical_loss": 5.645957247937725,
      "tokens_seen": 33554432
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009997592681752529,
      "loss": 0.2402,
      "theoretical_loss": 5.638973336059157,
      "tokens_seen": 33816576
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009996790242336705,
      "loss": 0.235,
      "theoretical_loss": 5.632058380369512,
      "tokens_seen": 34078720
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009995987802920879,
      "loss": 0.2355,
      "theoretical_loss": 5.625211177394046,
      "tokens_seen": 34340864
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009995185363505057,
      "loss": 0.2349,
      "theoretical_loss": 5.618430553644782,
      "tokens_seen": 34603008
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000999438292408923,
      "loss": 0.2367,
      "theoretical_loss": 5.611715364652864,
      "tokens_seen": 34865152
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009993580484673407,
      "loss": 0.2314,
      "theoretical_loss": 5.605064494039176,
      "tokens_seen": 35127296
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009992778045257583,
      "loss": 0.2299,
      "theoretical_loss": 5.598476852621397,
      "tokens_seen": 35389440
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000999197560584176,
      "loss": 0.2361,
      "theoretical_loss": 5.591951377555809,
      "tokens_seen": 35651584
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009991173166425935,
      "loss": 0.2362,
      "theoretical_loss": 5.585487031512276,
      "tokens_seen": 35913728
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009990370727010112,
      "loss": 0.2358,
      "theoretical_loss": 5.579082801880871,
      "tokens_seen": 36175872
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009989568287594288,
      "loss": 0.2332,
      "theoretical_loss": 5.572737700008718,
      "tokens_seen": 36438016
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009988765848178462,
      "loss": 0.2354,
      "theoretical_loss": 5.56645076046569,
      "tokens_seen": 36700160
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009987963408762638,
      "loss": 0.2326,
      "theoretical_loss": 5.5602210403376775,
      "tokens_seen": 36962304
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009987160969346814,
      "loss": 0.2372,
      "theoretical_loss": 5.554047618546193,
      "tokens_seen": 37224448
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000998635852993099,
      "loss": 0.2413,
      "theoretical_loss": 5.547929595193182,
      "tokens_seen": 37486592
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009985556090515166,
      "loss": 0.2345,
      "theoretical_loss": 5.5418660909298945,
      "tokens_seen": 37748736
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009984753651099342,
      "loss": 0.2308,
      "theoretical_loss": 5.535856246348814,
      "tokens_seen": 38010880
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009983951211683518,
      "loss": 0.2326,
      "theoretical_loss": 5.529899221397624,
      "tokens_seen": 38273024
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009983148772267695,
      "loss": 0.2299,
      "theoretical_loss": 5.523994194814273,
      "tokens_seen": 38535168
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000998234633285187,
      "loss": 0.2315,
      "theoretical_loss": 5.518140363582252,
      "tokens_seen": 38797312
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009981543893436047,
      "loss": 0.2336,
      "theoretical_loss": 5.512336942405216,
      "tokens_seen": 39059456
    },
    {
      "epoch": 0.01,
      "objective/train/advantage_avg": 0.007620680145919323,
      "objective/train/docs_used": 21760,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 4.368919372558594,
      "objective/train/original_loss": 4.368919372558594,
      "objective/train/theoretical_loss": 5.506583163200142,
      "objective/train/tokens_used": 59781600,
      "objective/train/value_avg": -0.0172576904296875,
      "objective/train/value_loss": 0.0010155023774132133,
      "objective/train/value_max": -0.00572967529296875,
      "objective/train/value_min": -0.12042236328125,
      "objective/train/value_reward_corr": 0.19761336452653674,
      "objective/train/value_std": 0.0086517333984375,
      "objective/train/weight_avg": 1.0080623626708984,
      "objective/train/weighted_lm_loss": 4.408056735992432,
      "objective/train/weights_max": 1.1209726333618164,
      "objective/train/weights_min": 0.3737069368362427,
      "theoretical_loss": 5.506583163200142,
      "tokens_seen": 39321600
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000998074145402022,
      "loss": 0.2265,
      "theoretical_loss": 5.506583163200142,
      "tokens_seen": 39321600
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009979939014604397,
      "loss": 0.2311,
      "theoretical_loss": 5.5008782746082625,
      "tokens_seen": 39583744
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009979136575188573,
      "loss": 0.2287,
      "theoretical_loss": 5.495221541523011,
      "tokens_seen": 39845888
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000997833413577275,
      "loss": 0.2302,
      "theoretical_loss": 5.489612244634316,
      "tokens_seen": 40108032
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009977531696356925,
      "loss": 0.2287,
      "theoretical_loss": 5.48404967998854,
      "tokens_seen": 40370176
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009976729256941101,
      "loss": 0.2282,
      "theoretical_loss": 5.478533158563456,
      "tokens_seen": 40632320
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009975926817525277,
      "loss": 0.2321,
      "theoretical_loss": 5.473062005857637,
      "tokens_seen": 40894464
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009975124378109451,
      "loss": 0.2303,
      "theoretical_loss": 5.467635561493681,
      "tokens_seen": 41156608
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000997432193869363,
      "loss": 0.2289,
      "theoretical_loss": 5.462253178834744,
      "tokens_seen": 41418752
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009973519499277804,
      "loss": 0.225,
      "theoretical_loss": 5.456914224613812,
      "tokens_seen": 41680896
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009972717059861982,
      "loss": 0.228,
      "theoretical_loss": 5.451618078575256,
      "tokens_seen": 41943040
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009971914620446156,
      "loss": 0.2249,
      "theoretical_loss": 5.446364133128155,
      "tokens_seen": 42205184
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009971112181030332,
      "loss": 0.224,
      "theoretical_loss": 5.44115179301095,
      "tokens_seen": 42467328
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009970309741614508,
      "loss": 0.2268,
      "theoretical_loss": 5.435980474966981,
      "tokens_seen": 42729472
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009969507302198684,
      "loss": 0.223,
      "theoretical_loss": 5.430849607430501,
      "tokens_seen": 42991616
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000996870486278286,
      "loss": 0.2224,
      "theoretical_loss": 5.425758630222747,
      "tokens_seen": 43253760
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009967902423367037,
      "loss": 0.227,
      "theoretical_loss": 5.42070699425771,
      "tokens_seen": 43515904
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009967099983951213,
      "loss": 0.2264,
      "theoretical_loss": 5.415694161257225,
      "tokens_seen": 43778048
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009966297544535387,
      "loss": 0.2205,
      "theoretical_loss": 5.410719603475034,
      "tokens_seen": 44040192
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009965495105119565,
      "loss": 0.2239,
      "theoretical_loss": 5.405782803429483,
      "tokens_seen": 44302336
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009964692665703739,
      "loss": 0.2202,
      "theoretical_loss": 5.400883253644551,
      "tokens_seen": 44564480
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009963890226287915,
      "loss": 0.224,
      "theoretical_loss": 5.396020456398885,
      "tokens_seen": 44826624
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009963087786872091,
      "loss": 0.217,
      "theoretical_loss": 5.391193923482547,
      "tokens_seen": 45088768
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009962285347456267,
      "loss": 0.2206,
      "theoretical_loss": 5.386403175961223,
      "tokens_seen": 45350912
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009961482908040443,
      "loss": 0.2205,
      "theoretical_loss": 5.381647743947578,
      "tokens_seen": 45613056
    },
    {
      "epoch": 0.01,
      "objective/train/advantage_avg": 0.008024879731237888,
      "objective/train/docs_used": 24173,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 4.302799701690674,
      "objective/train/original_loss": 4.302799701690674,
      "objective/train/theoretical_loss": 5.37692716637954,
      "objective/train/tokens_used": 66335200,
      "objective/train/value_avg": -0.017547607421875,
      "objective/train/value_loss": 0.0007267643231898546,
      "objective/train/value_max": -0.0055999755859375,
      "objective/train/value_min": -0.1319580078125,
      "objective/train/value_reward_corr": 0.2775875654342026,
      "objective/train/value_std": 0.0094757080078125,
      "objective/train/weight_avg": 1.0083508491516113,
      "objective/train/weighted_lm_loss": 4.344386577606201,
      "objective/train/weights_max": 1.0802388191223145,
      "objective/train/weights_min": 0.23186293244361877,
      "theoretical_loss": 5.37692716637954,
      "tokens_seen": 45875200
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000996068046862462,
      "loss": 0.223,
      "theoretical_loss": 5.37692716637954,
      "tokens_seen": 45875200
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009959878029208796,
      "loss": 0.2202,
      "theoretical_loss": 5.372240990805237,
      "tokens_seen": 46137344
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009959075589792972,
      "loss": 0.2182,
      "theoretical_loss": 5.367588773174377,
      "tokens_seen": 46399488
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009958273150377146,
      "loss": 0.2236,
      "theoretical_loss": 5.36297007763582,
      "tokens_seen": 46661632
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009957470710961322,
      "loss": 0.2225,
      "theoretical_loss": 5.358384476341126,
      "tokens_seen": 46923776
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009956668271545498,
      "loss": 0.2186,
      "theoretical_loss": 5.353831549253895,
      "tokens_seen": 47185920
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009955865832129674,
      "loss": 0.2205,
      "theoretical_loss": 5.349310883964664,
      "tokens_seen": 47448064
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000995506339271385,
      "loss": 0.2193,
      "theoretical_loss": 5.344822075511196,
      "tokens_seen": 47710208
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009954260953298026,
      "loss": 0.2176,
      "theoretical_loss": 5.340364726203955,
      "tokens_seen": 47972352
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009953458513882203,
      "loss": 0.2151,
      "theoretical_loss": 5.3359384454566055,
      "tokens_seen": 48234496
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009952656074466376,
      "loss": 0.2138,
      "theoretical_loss": 5.331542849621357,
      "tokens_seen": 48496640
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009951853635050555,
      "loss": 0.2168,
      "theoretical_loss": 5.327177561828993,
      "tokens_seen": 48758784
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009951051195634729,
      "loss": 0.2189,
      "theoretical_loss": 5.32284221183342,
      "tokens_seen": 49020928
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0009950248756218905,
      "loss": 0.2206,
      "theoretical_loss": 5.318536435860599,
      "tokens_seen": 49283072
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000994944631680308,
      "loss": 0.2147,
      "theoretical_loss": 5.314259876461705,
      "tokens_seen": 49545216
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009948643877387257,
      "loss": 0.2176,
      "theoretical_loss": 5.310012182370359,
      "tokens_seen": 49807360
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009947841437971433,
      "loss": 0.214,
      "theoretical_loss": 5.305793008363841,
      "tokens_seen": 50069504
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000994703899855561,
      "loss": 0.2142,
      "theoretical_loss": 5.301602015128104,
      "tokens_seen": 50331648
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009946236559139785,
      "loss": 0.2118,
      "theoretical_loss": 5.297438869126498,
      "tokens_seen": 50593792
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009945434119723962,
      "loss": 0.2195,
      "theoretical_loss": 5.293303242472074,
      "tokens_seen": 50855936
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009944631680308138,
      "loss": 0.2132,
      "theoretical_loss": 5.289194812803347,
      "tokens_seen": 51118080
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009943829240892312,
      "loss": 0.2113,
      "theoretical_loss": 5.285113263163414,
      "tokens_seen": 51380224
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000994302680147649,
      "loss": 0.2133,
      "theoretical_loss": 5.2810582818823235,
      "tokens_seen": 51642368
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009942224362060664,
      "loss": 0.2144,
      "theoretical_loss": 5.27702956246258,
      "tokens_seen": 51904512
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000994142192264484,
      "loss": 0.2125,
      "theoretical_loss": 5.273026803467695,
      "tokens_seen": 52166656
    },
    {
      "epoch": 0.02,
      "objective/train/advantage_avg": 0.007762981578707695,
      "objective/train/docs_used": 26541,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 4.0233917236328125,
      "objective/train/original_loss": 4.023391246795654,
      "objective/train/theoretical_loss": 5.269049708413682,
      "objective/train/tokens_used": 72888800,
      "objective/train/value_avg": -0.0139312744140625,
      "objective/train/value_loss": 0.0003585839585866779,
      "objective/train/value_max": -0.004230499267578125,
      "objective/train/value_min": -0.0765380859375,
      "objective/train/value_reward_corr": 0.23784686314537432,
      "objective/train/value_std": 0.006893157958984375,
      "objective/train/weight_avg": 1.0079320669174194,
      "objective/train/weighted_lm_loss": 4.059372901916504,
      "objective/train/weights_max": 1.0704232454299927,
      "objective/train/weights_min": 0.3713054358959198,
      "theoretical_loss": 5.269049708413682,
      "tokens_seen": 52428800
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009940619483229016,
      "loss": 0.2108,
      "theoretical_loss": 5.269049708413682,
      "tokens_seen": 52428800
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009939817043813192,
      "loss": 0.2138,
      "theoretical_loss": 5.265097985663418,
      "tokens_seen": 52690944
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009939014604397368,
      "loss": 0.2121,
      "theoretical_loss": 5.261171348323755,
      "tokens_seen": 52953088
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009938212164981545,
      "loss": 0.2093,
      "theoretical_loss": 5.257269514145324,
      "tokens_seen": 53215232
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000993740972556572,
      "loss": 0.2068,
      "theoretical_loss": 5.2533922054249365,
      "tokens_seen": 53477376
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009936607286149895,
      "loss": 0.2108,
      "theoretical_loss": 5.2495391489104986,
      "tokens_seen": 53739520
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009935804846734073,
      "loss": 0.2142,
      "theoretical_loss": 5.24571007570837,
      "tokens_seen": 54001664
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009935002407318247,
      "loss": 0.2101,
      "theoretical_loss": 5.2419047211930865,
      "tokens_seen": 54263808
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009934199967902423,
      "loss": 0.2067,
      "theoretical_loss": 5.238122824919387,
      "tokens_seen": 54525952
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.00099333975284866,
      "loss": 0.2105,
      "theoretical_loss": 5.234364130536457,
      "tokens_seen": 54788096
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009932595089070775,
      "loss": 0.2097,
      "theoretical_loss": 5.230628385704337,
      "tokens_seen": 55050240
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009931792649654951,
      "loss": 0.208,
      "theoretical_loss": 5.2269153420124255,
      "tokens_seen": 55312384
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009930990210239128,
      "loss": 0.2087,
      "theoretical_loss": 5.223224754900014,
      "tokens_seen": 55574528
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009930187770823304,
      "loss": 0.2098,
      "theoretical_loss": 5.219556383578795,
      "tokens_seen": 55836672
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000992938533140748,
      "loss": 0.2064,
      "theoretical_loss": 5.215909990957291,
      "tokens_seen": 56098816
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009928582891991654,
      "loss": 0.2073,
      "theoretical_loss": 5.212285343567135,
      "tokens_seen": 56360960
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000992778045257583,
      "loss": 0.2064,
      "theoretical_loss": 5.208682211491157,
      "tokens_seen": 56623104
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009926978013160006,
      "loss": 0.2043,
      "theoretical_loss": 5.205100368293225,
      "tokens_seen": 56885248
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009926175573744182,
      "loss": 0.2038,
      "theoretical_loss": 5.201539590949796,
      "tokens_seen": 57147392
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009925373134328358,
      "loss": 0.2055,
      "theoretical_loss": 5.1979996597831,
      "tokens_seen": 57409536
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009924570694912534,
      "loss": 0.2043,
      "theoretical_loss": 5.1944803583959525,
      "tokens_seen": 57671680
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000992376825549671,
      "loss": 0.2057,
      "theoretical_loss": 5.190981473608112,
      "tokens_seen": 57933824
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009922965816080887,
      "loss": 0.2065,
      "theoretical_loss": 5.18750279539416,
      "tokens_seen": 58195968
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009922163376665063,
      "loss": 0.2025,
      "theoretical_loss": 5.184044116822849,
      "tokens_seen": 58458112
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009921360937249237,
      "loss": 0.2079,
      "theoretical_loss": 5.1806052339978965,
      "tokens_seen": 58720256
    },
    {
      "epoch": 0.02,
      "objective/train/advantage_avg": 0.005310404114425182,
      "objective/train/docs_used": 29071,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 3.910093069076538,
      "objective/train/original_loss": 3.910093069076538,
      "objective/train/theoretical_loss": 5.1771859460001455,
      "objective/train/tokens_used": 79442400,
      "objective/train/value_avg": -0.01593017578125,
      "objective/train/value_loss": 0.0005049873143434525,
      "objective/train/value_max": -0.004070281982421875,
      "objective/train/value_min": -0.09368896484375,
      "objective/train/value_reward_corr": 0.3727184924290943,
      "objective/train/value_std": 0.00925445556640625,
      "objective/train/weight_avg": 1.005548357963562,
      "objective/train/weighted_lm_loss": 3.935572624206543,
      "objective/train/weights_max": 1.0910027027130127,
      "objective/train/weights_min": 0.37153780460357666,
      "theoretical_loss": 5.1771859460001455,
      "tokens_seen": 58982400
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009920558497833415,
      "loss": 0.2002,
      "theoretical_loss": 5.1771859460001455,
      "tokens_seen": 58982400
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000991975605841759,
      "loss": 0.2064,
      "theoretical_loss": 5.1737860548311065,
      "tokens_seen": 59244544
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009918953619001765,
      "loss": 0.204,
      "theoretical_loss": 5.170405365357794,
      "tokens_seen": 59506688
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009918151179585941,
      "loss": 0.2007,
      "theoretical_loss": 5.167043685258852,
      "tokens_seen": 59768832
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009917348740170117,
      "loss": 0.2029,
      "theoretical_loss": 5.163700824971922,
      "tokens_seen": 60030976
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009916546300754293,
      "loss": 0.2062,
      "theoretical_loss": 5.160376597642223,
      "tokens_seen": 60293120
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000991574386133847,
      "loss": 0.208,
      "theoretical_loss": 5.157070819072301,
      "tokens_seen": 60555264
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009914941421922646,
      "loss": 0.2057,
      "theoretical_loss": 5.153783307672935,
      "tokens_seen": 60817408
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000991413898250682,
      "loss": 0.2046,
      "theoretical_loss": 5.150513884415149,
      "tokens_seen": 61079552
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009913336543090998,
      "loss": 0.2011,
      "theoretical_loss": 5.14726237278331,
      "tokens_seen": 61341696
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009912534103675172,
      "loss": 0.2057,
      "theoretical_loss": 5.144028598729285,
      "tokens_seen": 61603840
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009911731664259348,
      "loss": 0.2008,
      "theoretical_loss": 5.140812390627624,
      "tokens_seen": 61865984
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009910929224843524,
      "loss": 0.2033,
      "theoretical_loss": 5.137613579231737,
      "tokens_seen": 62128128
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.00099101267854277,
      "loss": 0.2025,
      "theoretical_loss": 5.134431997631053,
      "tokens_seen": 62390272
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009909324346011876,
      "loss": 0.1987,
      "theoretical_loss": 5.1312674812091235,
      "tokens_seen": 62652416
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009908521906596053,
      "loss": 0.1978,
      "theoretical_loss": 5.128119867602646,
      "tokens_seen": 62914560
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009907719467180229,
      "loss": 0.2007,
      "theoretical_loss": 5.124988996661393,
      "tokens_seen": 63176704
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009906917027764405,
      "loss": 0.1967,
      "theoretical_loss": 5.121874710409012,
      "tokens_seen": 63438848
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000990611458834858,
      "loss": 0.1992,
      "theoretical_loss": 5.118776853004677,
      "tokens_seen": 63700992
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009905312148932755,
      "loss": 0.1973,
      "theoretical_loss": 5.115695270705579,
      "tokens_seen": 63963136
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000990450970951693,
      "loss": 0.1952,
      "theoretical_loss": 5.112629811830217,
      "tokens_seen": 64225280
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009903707270101107,
      "loss": 0.1983,
      "theoretical_loss": 5.10958032672248,
      "tokens_seen": 64487424
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009902904830685283,
      "loss": 0.1941,
      "theoretical_loss": 5.106546667716508,
      "tokens_seen": 64749568
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000990210239126946,
      "loss": 0.2008,
      "theoretical_loss": 5.103528689102281,
      "tokens_seen": 65011712
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009901299951853636,
      "loss": 0.1951,
      "theoretical_loss": 5.100526247091967,
      "tokens_seen": 65273856
    },
    {
      "epoch": 0.02,
      "objective/train/advantage_avg": 0.008375543169677258,
      "objective/train/docs_used": 31526,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 3.9859790802001953,
      "objective/train/original_loss": 3.9859793186187744,
      "objective/train/theoretical_loss": 5.097539199786951,
      "objective/train/tokens_used": 85996000,
      "objective/train/value_avg": -0.0173187255859375,
      "objective/train/value_loss": 0.0005732561694458127,
      "objective/train/value_max": -0.004886627197265625,
      "objective/train/value_min": -0.1669921875,
      "objective/train/value_reward_corr": 0.3654351277123077,
      "objective/train/value_std": 0.0103912353515625,
      "objective/train/weight_avg": 1.0086482763290405,
      "objective/train/weighted_lm_loss": 4.023629665374756,
      "objective/train/weights_max": 1.1241106986999512,
      "objective/train/weights_min": 0.49856844544410706,
      "theoretical_loss": 5.097539199786951,
      "tokens_seen": 65536000
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009900497512437812,
      "loss": 0.1982,
      "theoretical_loss": 5.097539199786951,
      "tokens_seen": 65536000
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009899695073021988,
      "loss": 0.1987,
      "theoretical_loss": 5.094567407145588,
      "tokens_seen": 65798144
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009898892633606162,
      "loss": 0.1958,
      "theoretical_loss": 5.09161073095161,
      "tokens_seen": 66060288
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000989809019419034,
      "loss": 0.1928,
      "theoretical_loss": 5.0886690347832015,
      "tokens_seen": 66322432
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009897287754774514,
      "loss": 0.1945,
      "theoretical_loss": 5.0857421839827275,
      "tokens_seen": 66584576
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000989648531535869,
      "loss": 0.1912,
      "theoretical_loss": 5.082830045627072,
      "tokens_seen": 66846720
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009895682875942866,
      "loss": 0.1962,
      "theoretical_loss": 5.079932488498602,
      "tokens_seen": 67108864
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009894880436527042,
      "loss": 0.1935,
      "theoretical_loss": 5.077049383056725,
      "tokens_seen": 67371008
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009894077997111218,
      "loss": 0.1957,
      "theoretical_loss": 5.074180601410026,
      "tokens_seen": 67633152
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009893275557695395,
      "loss": 0.1884,
      "theoretical_loss": 5.0713260172889845,
      "tokens_seen": 67895296
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000989247311827957,
      "loss": 0.1908,
      "theoretical_loss": 5.068485506019231,
      "tokens_seen": 68157440
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009891670678863745,
      "loss": 0.1931,
      "theoretical_loss": 5.06565894449535,
      "tokens_seen": 68419584
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009890868239447923,
      "loss": 0.1936,
      "theoretical_loss": 5.06284621115523,
      "tokens_seen": 68681728
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009890065800032097,
      "loss": 0.198,
      "theoretical_loss": 5.060047185954893,
      "tokens_seen": 68943872
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009889263360616273,
      "loss": 0.1926,
      "theoretical_loss": 5.057261750343864,
      "tokens_seen": 69206016
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000988846092120045,
      "loss": 0.1934,
      "theoretical_loss": 5.0544897872410095,
      "tokens_seen": 69468160
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009887658481784625,
      "loss": 0.192,
      "theoretical_loss": 5.051731181010866,
      "tokens_seen": 69730304
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009886856042368801,
      "loss": 0.1905,
      "theoretical_loss": 5.048985817440432,
      "tokens_seen": 69992448
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009886053602952978,
      "loss": 0.1939,
      "theoretical_loss": 5.046253583716425,
      "tokens_seen": 70254592
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009885251163537154,
      "loss": 0.1908,
      "theoretical_loss": 5.043534368402973,
      "tokens_seen": 70516736
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000988444872412133,
      "loss": 0.19,
      "theoretical_loss": 5.040828061419762,
      "tokens_seen": 70778880
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009883646284705506,
      "loss": 0.1928,
      "theoretical_loss": 5.038134554020587,
      "tokens_seen": 71041024
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000988284384528968,
      "loss": 0.1893,
      "theoretical_loss": 5.03545373877234,
      "tokens_seen": 71303168
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009882041405873858,
      "loss": 0.1902,
      "theoretical_loss": 5.032785509534391,
      "tokens_seen": 71565312
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009881238966458032,
      "loss": 0.1882,
      "theoretical_loss": 5.030129761438376,
      "tokens_seen": 71827456
    },
    {
      "epoch": 0.02,
      "objective/train/advantage_avg": 0.006676941178739071,
      "objective/train/docs_used": 33987,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 3.692817211151123,
      "objective/train/original_loss": 3.692817211151123,
      "objective/train/theoretical_loss": 5.0274863908683685,
      "objective/train/tokens_used": 92549600,
      "objective/train/value_avg": -0.01511383056640625,
      "objective/train/value_loss": 0.0004530175356194377,
      "objective/train/value_max": -0.00426483154296875,
      "objective/train/value_min": -0.100341796875,
      "objective/train/value_reward_corr": 0.2687911557831162,
      "objective/train/value_std": 0.00835418701171875,
      "objective/train/weight_avg": 1.0068917274475098,
      "objective/train/weighted_lm_loss": 3.7194132804870605,
      "objective/train/weights_max": 1.0913752317428589,
      "objective/train/weights_min": 0.37156614661216736,
      "theoretical_loss": 5.0274863908683685,
      "tokens_seen": 72089600
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009880436527042208,
      "loss": 0.1875,
      "theoretical_loss": 5.0274863908683685,
      "tokens_seen": 72089600
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009879634087626384,
      "loss": 0.1911,
      "theoretical_loss": 5.024855295441432,
      "tokens_seen": 72351744
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000987883164821056,
      "loss": 0.189,
      "theoretical_loss": 5.022236373988544,
      "tokens_seen": 72613888
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009878029208794737,
      "loss": 0.1874,
      "theoretical_loss": 5.01962952653588,
      "tokens_seen": 72876032
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009877226769378913,
      "loss": 0.1878,
      "theoretical_loss": 5.017034654286462,
      "tokens_seen": 73138176
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000987642432996309,
      "loss": 0.19,
      "theoretical_loss": 5.0144516596021385,
      "tokens_seen": 73400320
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009875621890547263,
      "loss": 0.1827,
      "theoretical_loss": 5.011880445985916,
      "tokens_seen": 73662464
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000987481945113144,
      "loss": 0.1884,
      "theoretical_loss": 5.009320918064615,
      "tokens_seen": 73924608
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009874017011715615,
      "loss": 0.1884,
      "theoretical_loss": 5.006772981571855,
      "tokens_seen": 74186752
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009873214572299791,
      "loss": 0.1866,
      "theoretical_loss": 5.004236543331345,
      "tokens_seen": 74448896
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009872412132883967,
      "loss": 0.1851,
      "theoretical_loss": 5.001711511240506,
      "tokens_seen": 74711040
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009871609693468144,
      "loss": 0.1864,
      "theoretical_loss": 4.999197794254371,
      "tokens_seen": 74973184
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000987080725405232,
      "loss": 0.1857,
      "theoretical_loss": 4.9966953023697975,
      "tokens_seen": 75235328
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009870004814636496,
      "loss": 0.1855,
      "theoretical_loss": 4.994203946609964,
      "tokens_seen": 75497472
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000986920237522067,
      "loss": 0.1828,
      "theoretical_loss": 4.991723639009154,
      "tokens_seen": 75759616
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009868399935804848,
      "loss": 0.1839,
      "theoretical_loss": 4.989254292597813,
      "tokens_seen": 76021760
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009867597496389022,
      "loss": 0.1871,
      "theoretical_loss": 4.986795821387878,
      "tokens_seen": 76283904
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009866795056973198,
      "loss": 0.1852,
      "theoretical_loss": 4.984348140358374,
      "tokens_seen": 76546048
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009865992617557374,
      "loss": 0.1871,
      "theoretical_loss": 4.981911165441273,
      "tokens_seen": 76808192
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000986519017814155,
      "loss": 0.1845,
      "theoretical_loss": 4.979484813507599,
      "tokens_seen": 77070336
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009864387738725726,
      "loss": 0.1835,
      "theoretical_loss": 4.977069002353792,
      "tokens_seen": 77332480
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009863585299309903,
      "loss": 0.1813,
      "theoretical_loss": 4.974663650688306,
      "tokens_seen": 77594624
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009862782859894079,
      "loss": 0.1833,
      "theoretical_loss": 4.972268678118454,
      "tokens_seen": 77856768
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009861980420478253,
      "loss": 0.1796,
      "theoretical_loss": 4.969884005137479,
      "tokens_seen": 78118912
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000986117798106243,
      "loss": 0.1841,
      "theoretical_loss": 4.967509553111862,
      "tokens_seen": 78381056
    },
    {
      "epoch": 0.02,
      "objective/train/advantage_avg": 0.007857042364776134,
      "objective/train/docs_used": 36308,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 3.7754344940185547,
      "objective/train/original_loss": 3.775434970855713,
      "objective/train/theoretical_loss": 4.96514524426884,
      "objective/train/tokens_used": 99103200,
      "objective/train/value_avg": -0.01514434814453125,
      "objective/train/value_loss": 0.0004333880788180977,
      "objective/train/value_max": -0.003795623779296875,
      "objective/train/value_min": -0.1292724609375,
      "objective/train/value_reward_corr": 0.27335943213418523,
      "objective/train/value_std": 0.0078125,
      "objective/train/weight_avg": 1.0080602169036865,
      "objective/train/weighted_lm_loss": 3.8082289695739746,
      "objective/train/weights_max": 1.1106642484664917,
      "objective/train/weights_min": 0.3715236186981201,
      "theoretical_loss": 4.96514524426884,
      "tokens_seen": 78643200
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009860375541646605,
      "loss": 0.1803,
      "theoretical_loss": 4.96514524426884,
      "tokens_seen": 78643200
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009859573102230783,
      "loss": 0.1828,
      "theoretical_loss": 4.962791001684167,
      "tokens_seen": 78905344
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009858770662814957,
      "loss": 0.1812,
      "theoretical_loss": 4.960446749270055,
      "tokens_seen": 79167488
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009857968223399133,
      "loss": 0.1833,
      "theoretical_loss": 4.958112411763365,
      "tokens_seen": 79429632
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000985716578398331,
      "loss": 0.1819,
      "theoretical_loss": 4.955787914713962,
      "tokens_seen": 79691776
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009856363344567486,
      "loss": 0.1828,
      "theoretical_loss": 4.953473184473312,
      "tokens_seen": 79953920
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009855560905151662,
      "loss": 0.1815,
      "theoretical_loss": 4.951168148183246,
      "tokens_seen": 80216064
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009854758465735838,
      "loss": 0.1808,
      "theoretical_loss": 4.948872733764926,
      "tokens_seen": 80478208
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009853956026320014,
      "loss": 0.1792,
      "theoretical_loss": 4.946586869908014,
      "tokens_seen": 80740352
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009853153586904188,
      "loss": 0.1821,
      "theoretical_loss": 4.944310486060004,
      "tokens_seen": 81002496
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009852351147488366,
      "loss": 0.1818,
      "theoretical_loss": 4.942043512415751,
      "tokens_seen": 81264640
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000985154870807254,
      "loss": 0.1836,
      "theoretical_loss": 4.939785879907176,
      "tokens_seen": 81526784
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009850746268656716,
      "loss": 0.1805,
      "theoretical_loss": 4.937537520193139,
      "tokens_seen": 81788928
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009849943829240892,
      "loss": 0.181,
      "theoretical_loss": 4.93529836564949,
      "tokens_seen": 82051072
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0009849141389825069,
      "loss": 0.1797,
      "theoretical_loss": 4.933068349359283,
      "tokens_seen": 82313216
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009848338950409245,
      "loss": 0.1773,
      "theoretical_loss": 4.93084740510316,
      "tokens_seen": 82575360
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000984753651099342,
      "loss": 0.1776,
      "theoretical_loss": 4.928635467349885,
      "tokens_seen": 82837504
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009846734071577597,
      "loss": 0.1764,
      "theoretical_loss": 4.92643247124705,
      "tokens_seen": 83099648
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009845931632161773,
      "loss": 0.1805,
      "theoretical_loss": 4.924238352611924,
      "tokens_seen": 83361792
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009845129192745947,
      "loss": 0.1789,
      "theoretical_loss": 4.922053047922455,
      "tokens_seen": 83623936
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009844326753330123,
      "loss": 0.1756,
      "theoretical_loss": 4.919876494308432,
      "tokens_seen": 83886080
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00098435243139143,
      "loss": 0.1804,
      "theoretical_loss": 4.917708629542775,
      "tokens_seen": 84148224
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009842721874498475,
      "loss": 0.1778,
      "theoretical_loss": 4.915549392032985,
      "tokens_seen": 84410368
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009841919435082652,
      "loss": 0.1807,
      "theoretical_loss": 4.913398720812719,
      "tokens_seen": 84672512
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009841116995666828,
      "loss": 0.1762,
      "theoretical_loss": 4.9112565555335115,
      "tokens_seen": 84934656
    },
    {
      "epoch": 0.03,
      "objective/train/advantage_avg": 0.008176760748028755,
      "objective/train/docs_used": 38678,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 3.4570322036743164,
      "objective/train/original_loss": 3.4570322036743164,
      "objective/train/theoretical_loss": 4.909122836456632,
      "objective/train/tokens_used": 105656800,
      "objective/train/value_avg": -0.0123748779296875,
      "objective/train/value_loss": 0.00042682766797952354,
      "objective/train/value_max": -0.00313568115234375,
      "objective/train/value_min": -0.06854248046875,
      "objective/train/value_reward_corr": 0.12430929764516488,
      "objective/train/value_std": 0.005199432373046875,
      "objective/train/weight_avg": 1.0083661079406738,
      "objective/train/weighted_lm_loss": 3.4894821643829346,
      "objective/train/weights_max": 1.0679434537887573,
      "objective/train/weights_min": 0.3725539743900299,
      "theoretical_loss": 4.909122836456632,
      "tokens_seen": 85196800
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009840314556251004,
      "loss": 0.1783,
      "theoretical_loss": 4.909122836456632,
      "tokens_seen": 85196800
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009839512116835178,
      "loss": 0.1773,
      "theoretical_loss": 4.906997504445066,
      "tokens_seen": 85458944
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009838709677419356,
      "loss": 0.1792,
      "theoretical_loss": 4.904880500955633,
      "tokens_seen": 85721088
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000983790723800353,
      "loss": 0.1752,
      "theoretical_loss": 4.90277176803123,
      "tokens_seen": 85983232
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009837104798587706,
      "loss": 0.1747,
      "theoretical_loss": 4.9006712482931984,
      "tokens_seen": 86245376
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009836302359171882,
      "loss": 0.176,
      "theoretical_loss": 4.89857888493381,
      "tokens_seen": 86507520
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009835499919756058,
      "loss": 0.1777,
      "theoretical_loss": 4.896494621708882,
      "tokens_seen": 86769664
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009834697480340234,
      "loss": 0.1748,
      "theoretical_loss": 4.8944184029305,
      "tokens_seen": 87031808
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000983389504092441,
      "loss": 0.1762,
      "theoretical_loss": 4.892350173459863,
      "tokens_seen": 87293952
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009833092601508587,
      "loss": 0.1736,
      "theoretical_loss": 4.890289878700239,
      "tokens_seen": 87556096
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009832290162092763,
      "loss": 0.1736,
      "theoretical_loss": 4.888237464590028,
      "tokens_seen": 87818240
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000983148772267694,
      "loss": 0.1769,
      "theoretical_loss": 4.8861928775959464,
      "tokens_seen": 88080384
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009830685283261113,
      "loss": 0.1752,
      "theoretical_loss": 4.884156064706302,
      "tokens_seen": 88342528
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009829882843845291,
      "loss": 0.1744,
      "theoretical_loss": 4.882126973424384,
      "tokens_seen": 88604672
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009829080404429465,
      "loss": 0.1731,
      "theoretical_loss": 4.880105551761961,
      "tokens_seen": 88866816
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009828277965013641,
      "loss": 0.1783,
      "theoretical_loss": 4.87809174823286,
      "tokens_seen": 89128960
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009827475525597817,
      "loss": 0.1719,
      "theoretical_loss": 4.876085511846673,
      "tokens_seen": 89391104
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009826673086181994,
      "loss": 0.1769,
      "theoretical_loss": 4.874086792102535,
      "tokens_seen": 89653248
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000982587064676617,
      "loss": 0.1751,
      "theoretical_loss": 4.872095538983015,
      "tokens_seen": 89915392
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009825068207350346,
      "loss": 0.1779,
      "theoretical_loss": 4.870111702948094,
      "tokens_seen": 90177536
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009824265767934522,
      "loss": 0.1747,
      "theoretical_loss": 4.868135234929232,
      "tokens_seen": 90439680
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009823463328518696,
      "loss": 0.1749,
      "theoretical_loss": 4.866166086323535,
      "tokens_seen": 90701824
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009822660889102874,
      "loss": 0.1782,
      "theoretical_loss": 4.864204208988003,
      "tokens_seen": 90963968
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009821858449687048,
      "loss": 0.1738,
      "theoretical_loss": 4.86224955523387,
      "tokens_seen": 91226112
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009821056010271224,
      "loss": 0.1774,
      "theoretical_loss": 4.860302077821023,
      "tokens_seen": 91488256
    },
    {
      "epoch": 0.03,
      "objective/train/advantage_avg": 0.005717393010854721,
      "objective/train/docs_used": 40748,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 3.387242317199707,
      "objective/train/original_loss": 3.387242555618286,
      "objective/train/theoretical_loss": 4.858361729952518,
      "objective/train/tokens_used": 112210400,
      "objective/train/value_avg": -0.01448822021484375,
      "objective/train/value_loss": 0.0005706871743313968,
      "objective/train/value_max": -0.0030155181884765625,
      "objective/train/value_min": -0.09588623046875,
      "objective/train/value_reward_corr": 0.3036400333241696,
      "objective/train/value_std": 0.007572174072265625,
      "objective/train/weight_avg": 1.0059770345687866,
      "objective/train/weighted_lm_loss": 3.4078683853149414,
      "objective/train/weights_max": 1.0813897848129272,
      "objective/train/weights_min": 0.37298911809921265,
      "theoretical_loss": 4.858361729952518,
      "tokens_seen": 91750400
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00098202535708554,
      "loss": 0.1731,
      "theoretical_loss": 4.858361729952518,
      "tokens_seen": 91750400
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009819451131439577,
      "loss": 0.1724,
      "theoretical_loss": 4.856428465269159,
      "tokens_seen": 92012544
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009818648692023753,
      "loss": 0.1731,
      "theoretical_loss": 4.8545022378441836,
      "tokens_seen": 92274688
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009817846252607929,
      "loss": 0.1745,
      "theoretical_loss": 4.852583002178001,
      "tokens_seen": 92536832
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009817043813192105,
      "loss": 0.1718,
      "theoretical_loss": 4.8506707131930344,
      "tokens_seen": 92798976
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000981624137377628,
      "loss": 0.1729,
      "theoretical_loss": 4.848765326228618,
      "tokens_seen": 93061120
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009815438934360455,
      "loss": 0.1688,
      "theoretical_loss": 4.846866797035984,
      "tokens_seen": 93323264
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009814636494944631,
      "loss": 0.1717,
      "theoretical_loss": 4.844975081773322,
      "tokens_seen": 93585408
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009813834055528807,
      "loss": 0.1714,
      "theoretical_loss": 4.843090137000904,
      "tokens_seen": 93847552
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009813031616112983,
      "loss": 0.1716,
      "theoretical_loss": 4.841211919676287,
      "tokens_seen": 94109696
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000981222917669716,
      "loss": 0.1673,
      "theoretical_loss": 4.839340387149586,
      "tokens_seen": 94371840
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009811426737281336,
      "loss": 0.1707,
      "theoretical_loss": 4.837475497158817,
      "tokens_seen": 94633984
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009810624297865512,
      "loss": 0.1687,
      "theoretical_loss": 4.835617207825303,
      "tokens_seen": 94896128
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009809821858449686,
      "loss": 0.1669,
      "theoretical_loss": 4.83376547764915,
      "tokens_seen": 95158272
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009809019419033864,
      "loss": 0.1717,
      "theoretical_loss": 4.831920265504792,
      "tokens_seen": 95420416
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009808216979618038,
      "loss": 0.1723,
      "theoretical_loss": 4.830081530636594,
      "tokens_seen": 95682560
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009807414540202216,
      "loss": 0.1717,
      "theoretical_loss": 4.8282492326545245,
      "tokens_seen": 95944704
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000980661210078639,
      "loss": 0.1714,
      "theoretical_loss": 4.826423331529884,
      "tokens_seen": 96206848
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009805809661370566,
      "loss": 0.1687,
      "theoretical_loss": 4.824603787591102,
      "tokens_seen": 96468992
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009805007221954742,
      "loss": 0.1724,
      "theoretical_loss": 4.822790561519591,
      "tokens_seen": 96731136
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009804204782538919,
      "loss": 0.1713,
      "theoretical_loss": 4.8209836143456535,
      "tokens_seen": 96993280
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009803402343123095,
      "loss": 0.167,
      "theoretical_loss": 4.81918290744446,
      "tokens_seen": 97255424
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000980259990370727,
      "loss": 0.1705,
      "theoretical_loss": 4.817388402532074,
      "tokens_seen": 97517568
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009801797464291447,
      "loss": 0.176,
      "theoretical_loss": 4.815600061661536,
      "tokens_seen": 97779712
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000980099502487562,
      "loss": 0.17,
      "theoretical_loss": 4.813817847219008,
      "tokens_seen": 98041856
    },
    {
      "epoch": 0.03,
      "objective/train/advantage_avg": 0.005961360409855843,
      "objective/train/docs_used": 43181,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 3.3599514961242676,
      "objective/train/original_loss": 3.3599514961242676,
      "objective/train/theoretical_loss": 4.812041721919962,
      "objective/train/tokens_used": 118764000,
      "objective/train/value_avg": -0.01464080810546875,
      "objective/train/value_loss": 0.0005292649148032069,
      "objective/train/value_max": -0.003040313720703125,
      "objective/train/value_min": -0.2239990234375,
      "objective/train/value_reward_corr": 0.26425426870008045,
      "objective/train/value_std": 0.0106048583984375,
      "objective/train/weight_avg": 1.0062075853347778,
      "objective/train/weighted_lm_loss": 3.379601001739502,
      "objective/train/weights_max": 1.1572779417037964,
      "objective/train/weights_min": 0.37088075280189514,
      "theoretical_loss": 4.812041721919962,
      "tokens_seen": 98304000
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00098001925854598,
      "loss": 0.165,
      "theoretical_loss": 4.812041721919962,
      "tokens_seen": 98304000
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009799390146043973,
      "loss": 0.1663,
      "theoretical_loss": 4.810271648805427,
      "tokens_seen": 98566144
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000979858770662815,
      "loss": 0.17,
      "theoretical_loss": 4.8085075912383015,
      "tokens_seen": 98828288
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009797785267212325,
      "loss": 0.1685,
      "theoretical_loss": 4.806749512899687,
      "tokens_seen": 99090432
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009796982827796502,
      "loss": 0.1684,
      "theoretical_loss": 4.804997377785307,
      "tokens_seen": 99352576
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009796180388380678,
      "loss": 0.1671,
      "theoretical_loss": 4.8032511502019535,
      "tokens_seen": 99614720
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009795377948964854,
      "loss": 0.1671,
      "theoretical_loss": 4.801510794763988,
      "tokens_seen": 99876864
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000979457550954903,
      "loss": 0.1664,
      "theoretical_loss": 4.799776276389897,
      "tokens_seen": 100139008
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009793773070133206,
      "loss": 0.1693,
      "theoretical_loss": 4.798047560298882,
      "tokens_seen": 100401152
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009792970630717382,
      "loss": 0.1692,
      "theoretical_loss": 4.796324612007515,
      "tokens_seen": 100663296
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009792168191301556,
      "loss": 0.1668,
      "theoretical_loss": 4.794607397326421,
      "tokens_seen": 100925440
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009791365751885732,
      "loss": 0.1686,
      "theoretical_loss": 4.792895882357019,
      "tokens_seen": 101187584
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009790563312469908,
      "loss": 0.1683,
      "theoretical_loss": 4.791190033488302,
      "tokens_seen": 101449728
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009789760873054085,
      "loss": 0.1653,
      "theoretical_loss": 4.7894898173936635,
      "tokens_seen": 101711872
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000978895843363826,
      "loss": 0.1643,
      "theoretical_loss": 4.787795201027757,
      "tokens_seen": 101974016
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009788155994222437,
      "loss": 0.1674,
      "theoretical_loss": 4.786106151623423,
      "tokens_seen": 102236160
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009787353554806613,
      "loss": 0.1617,
      "theoretical_loss": 4.784422636688622,
      "tokens_seen": 102498304
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000978655111539079,
      "loss": 0.1637,
      "theoretical_loss": 4.782744624003442,
      "tokens_seen": 102760448
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009785748675974963,
      "loss": 0.1669,
      "theoretical_loss": 4.781072081617127,
      "tokens_seen": 103022592
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000978494623655914,
      "loss": 0.1658,
      "theoretical_loss": 4.779404977845148,
      "tokens_seen": 103284736
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009784143797143315,
      "loss": 0.1642,
      "theoretical_loss": 4.777743281266321,
      "tokens_seen": 103546880
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009783341357727491,
      "loss": 0.1616,
      "theoretical_loss": 4.776086960719956,
      "tokens_seen": 103809024
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009782538918311668,
      "loss": 0.1677,
      "theoretical_loss": 4.774435985303043,
      "tokens_seen": 104071168
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009781736478895844,
      "loss": 0.1606,
      "theoretical_loss": 4.772790324367482,
      "tokens_seen": 104333312
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000978093403948002,
      "loss": 0.1655,
      "theoretical_loss": 4.771149947517346,
      "tokens_seen": 104595456
    },
    {
      "epoch": 0.03,
      "objective/train/advantage_avg": 0.005723345559090376,
      "objective/train/docs_used": 45444,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 3.301978349685669,
      "objective/train/original_loss": 3.3019778728485107,
      "objective/train/theoretical_loss": 4.769514824606178,
      "objective/train/tokens_used": 125317600,
      "objective/train/value_avg": -0.01215362548828125,
      "objective/train/value_loss": 0.000492334773298353,
      "objective/train/value_max": -0.0025310516357421875,
      "objective/train/value_min": -0.21240234375,
      "objective/train/value_reward_corr": 0.35604829265860916,
      "objective/train/value_std": 0.007457733154296875,
      "objective/train/weight_avg": 1.005944848060608,
      "objective/train/weighted_lm_loss": 3.3237457275390625,
      "objective/train/weights_max": 1.124742031097412,
      "objective/train/weights_min": 0.3697309195995331,
      "theoretical_loss": 4.769514824606178,
      "tokens_seen": 104857600
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009780131600064196,
      "loss": 0.1616,
      "theoretical_loss": 4.769514824606178,
      "tokens_seen": 104857600
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009779329160648372,
      "loss": 0.1593,
      "theoretical_loss": 4.7678849257343305,
      "tokens_seen": 105119744
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009778526721232546,
      "loss": 0.1619,
      "theoretical_loss": 4.766260221246329,
      "tokens_seen": 105381888
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009777724281816724,
      "loss": 0.1613,
      "theoretical_loss": 4.764640681728281,
      "tokens_seen": 105644032
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009776921842400898,
      "loss": 0.1637,
      "theoretical_loss": 4.763026278005314,
      "tokens_seen": 105906176
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009776119402985074,
      "loss": 0.1604,
      "theoretical_loss": 4.761416981139046,
      "tokens_seen": 106168320
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000977531696356925,
      "loss": 0.1622,
      "theoretical_loss": 4.75981276242509,
      "tokens_seen": 106430464
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009774514524153427,
      "loss": 0.1616,
      "theoretical_loss": 4.758213593390595,
      "tokens_seen": 106692608
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009773712084737603,
      "loss": 0.1605,
      "theoretical_loss": 4.756619445791808,
      "tokens_seen": 106954752
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009772909645321779,
      "loss": 0.1627,
      "theoretical_loss": 4.755030291611678,
      "tokens_seen": 107216896
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009772107205905955,
      "loss": 0.1632,
      "theoretical_loss": 4.753446103057492,
      "tokens_seen": 107479040
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000977130476649013,
      "loss": 0.1618,
      "theoretical_loss": 4.751866852558529,
      "tokens_seen": 107741184
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009770502327074307,
      "loss": 0.1627,
      "theoretical_loss": 4.7502925127637585,
      "tokens_seen": 108003328
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009769699887658481,
      "loss": 0.162,
      "theoretical_loss": 4.7487230565395535,
      "tokens_seen": 108265472
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009768897448242657,
      "loss": 0.1609,
      "theoretical_loss": 4.747158456967452,
      "tokens_seen": 108527616
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009768095008826833,
      "loss": 0.1614,
      "theoretical_loss": 4.745598687341927,
      "tokens_seen": 108789760
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000976729256941101,
      "loss": 0.1622,
      "theoretical_loss": 4.744043721168196,
      "tokens_seen": 109051904
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009766490129995186,
      "loss": 0.1593,
      "theoretical_loss": 4.74249353216006,
      "tokens_seen": 109314048
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009765687690579362,
      "loss": 0.1589,
      "theoretical_loss": 4.740948094237761,
      "tokens_seen": 109576192
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009764885251163538,
      "loss": 0.1604,
      "theoretical_loss": 4.739407381525874,
      "tokens_seen": 109838336
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009764082811747713,
      "loss": 0.1601,
      "theoretical_loss": 4.7378713683512235,
      "tokens_seen": 110100480
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009763280372331889,
      "loss": 0.1592,
      "theoretical_loss": 4.7363400292408215,
      "tokens_seen": 110362624
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009762477932916065,
      "loss": 0.1597,
      "theoretical_loss": 4.734813338919842,
      "tokens_seen": 110624768
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000976167549350024,
      "loss": 0.1574,
      "theoretical_loss": 4.733291272309609,
      "tokens_seen": 110886912
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009760873054084417,
      "loss": 0.1573,
      "theoretical_loss": 4.731773804525616,
      "tokens_seen": 111149056
    },
    {
      "epoch": 0.03,
      "objective/train/advantage_avg": 0.005369019228965044,
      "objective/train/docs_used": 47788,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 3.1628293991088867,
      "objective/train/original_loss": 3.1628289222717285,
      "objective/train/theoretical_loss": 4.73026091087557,
      "objective/train/tokens_used": 131871200,
      "objective/train/value_avg": -0.0128021240234375,
      "objective/train/value_loss": 0.0003724124107975513,
      "objective/train/value_max": -0.0025615692138671875,
      "objective/train/value_min": -0.206787109375,
      "objective/train/value_reward_corr": 0.46127366019409083,
      "objective/train/value_std": 0.007617950439453125,
      "objective/train/weight_avg": 1.0055452585220337,
      "objective/train/weighted_lm_loss": 3.180939197540283,
      "objective/train/weights_max": 1.0874791145324707,
      "objective/train/weights_min": 0.3774433732032776,
      "theoretical_loss": 4.73026091087557,
      "tokens_seen": 111411200
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009760070614668593,
      "loss": 0.1562,
      "theoretical_loss": 4.73026091087557,
      "tokens_seen": 111411200
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009759268175252769,
      "loss": 0.1578,
      "theoretical_loss": 4.728752566857459,
      "tokens_seen": 111673344
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009758465735836945,
      "loss": 0.1577,
      "theoretical_loss": 4.72724874815764,
      "tokens_seen": 111935488
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009757663296421121,
      "loss": 0.1575,
      "theoretical_loss": 4.725749430648958,
      "tokens_seen": 112197632
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009756860857005296,
      "loss": 0.1584,
      "theoretical_loss": 4.724254590388881,
      "tokens_seen": 112459776
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009756058417589472,
      "loss": 0.1563,
      "theoretical_loss": 4.722764203617663,
      "tokens_seen": 112721920
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009755255978173648,
      "loss": 0.1581,
      "theoretical_loss": 4.7212782467565235,
      "tokens_seen": 112984064
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009754453538757823,
      "loss": 0.1575,
      "theoretical_loss": 4.719796696405858,
      "tokens_seen": 113246208
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009753651099342,
      "loss": 0.1553,
      "theoretical_loss": 4.718319529343462,
      "tokens_seen": 113508352
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009752848659926175,
      "loss": 0.1576,
      "theoretical_loss": 4.716846722522781,
      "tokens_seen": 113770496
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009752046220510353,
      "loss": 0.1545,
      "theoretical_loss": 4.715378253071181,
      "tokens_seen": 114032640
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009751243781094528,
      "loss": 0.1534,
      "theoretical_loss": 4.713914098288242,
      "tokens_seen": 114294784
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009750441341678703,
      "loss": 0.1551,
      "theoretical_loss": 4.712454235644064,
      "tokens_seen": 114556928
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000974963890226288,
      "loss": 0.155,
      "theoretical_loss": 4.710998642777606,
      "tokens_seen": 114819072
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009748836462847055,
      "loss": 0.1589,
      "theoretical_loss": 4.709547297495034,
      "tokens_seen": 115081216
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0009748034023431231,
      "loss": 0.1516,
      "theoretical_loss": 4.708100177768094,
      "tokens_seen": 115343360
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009747231584015407,
      "loss": 0.155,
      "theoretical_loss": 4.7066572617325075,
      "tokens_seen": 115605504
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009746429144599583,
      "loss": 0.1575,
      "theoretical_loss": 4.705218527686375,
      "tokens_seen": 115867648
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009745626705183758,
      "loss": 0.1577,
      "theoretical_loss": 4.703783954088612,
      "tokens_seen": 116129792
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009744824265767935,
      "loss": 0.1576,
      "theoretical_loss": 4.702353519557398,
      "tokens_seen": 116391936
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009744021826352111,
      "loss": 0.1537,
      "theoretical_loss": 4.700927202868639,
      "tokens_seen": 116654080
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009743219386936286,
      "loss": 0.1527,
      "theoretical_loss": 4.699504982954452,
      "tokens_seen": 116916224
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009742416947520463,
      "loss": 0.1534,
      "theoretical_loss": 4.698086838901676,
      "tokens_seen": 117178368
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009741614508104638,
      "loss": 0.1519,
      "theoretical_loss": 4.696672749950385,
      "tokens_seen": 117440512
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009740812068688814,
      "loss": 0.1517,
      "theoretical_loss": 4.695262695492428,
      "tokens_seen": 117702656
    },
    {
      "epoch": 0.04,
      "objective/train/advantage_avg": 0.006793029140681028,
      "objective/train/docs_used": 50256,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 3.074131488800049,
      "objective/train/original_loss": 3.0741310119628906,
      "objective/train/theoretical_loss": 4.693856655069986,
      "objective/train/tokens_used": 138424800,
      "objective/train/value_avg": -0.0142974853515625,
      "objective/train/value_loss": 0.000627185101620853,
      "objective/train/value_max": -0.00250244140625,
      "objective/train/value_min": -0.256591796875,
      "objective/train/value_reward_corr": 0.2673630119001533,
      "objective/train/value_std": 0.0104522705078125,
      "objective/train/weight_avg": 1.0070631504058838,
      "objective/train/weighted_lm_loss": 3.0982470512390137,
      "objective/train/weights_max": 1.1906453371047974,
      "objective/train/weights_min": 0.2276872992515564,
      "theoretical_loss": 4.693856655069986,
      "tokens_seen": 117964800
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.000974000962927299,
      "loss": 0.1531,
      "theoretical_loss": 4.693856655069986,
      "tokens_seen": 117964800
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009739207189857165,
      "loss": 0.1542,
      "theoretical_loss": 4.692454608374145,
      "tokens_seen": 118226944
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009738404750441343,
      "loss": 0.1509,
      "theoretical_loss": 4.6910565352434785,
      "tokens_seen": 118489088
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009737602311025518,
      "loss": 0.1512,
      "theoretical_loss": 4.689662415662662,
      "tokens_seen": 118751232
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009736799871609694,
      "loss": 0.1504,
      "theoretical_loss": 4.688272229761087,
      "tokens_seen": 119013376
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.000973599743219387,
      "loss": 0.1515,
      "theoretical_loss": 4.686885957811503,
      "tokens_seen": 119275520
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009735194992778046,
      "loss": 0.147,
      "theoretical_loss": 4.685503580228671,
      "tokens_seen": 119537664
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009734392553362221,
      "loss": 0.1475,
      "theoretical_loss": 4.684125077568028,
      "tokens_seen": 119799808
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009733590113946397,
      "loss": 0.1524,
      "theoretical_loss": 4.682750430524376,
      "tokens_seen": 120061952
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009732787674530573,
      "loss": 0.1518,
      "theoretical_loss": 4.6813796199305795,
      "tokens_seen": 120324096
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009731985235114748,
      "loss": 0.151,
      "theoretical_loss": 4.68001262675628,
      "tokens_seen": 120586240
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009731182795698925,
      "loss": 0.1465,
      "theoretical_loss": 4.678649432106627,
      "tokens_seen": 120848384
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00097303803562831,
      "loss": 0.1496,
      "theoretical_loss": 4.677290017221017,
      "tokens_seen": 121110528
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009729577916867277,
      "loss": 0.1461,
      "theoretical_loss": 4.675934363471857,
      "tokens_seen": 121372672
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009728775477451453,
      "loss": 0.1474,
      "theoretical_loss": 4.674582452363334,
      "tokens_seen": 121634816
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009727973038035629,
      "loss": 0.1463,
      "theoretical_loss": 4.673234265530201,
      "tokens_seen": 121896960
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009727170598619804,
      "loss": 0.1463,
      "theoretical_loss": 4.671889784736576,
      "tokens_seen": 122159104
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.000972636815920398,
      "loss": 0.1478,
      "theoretical_loss": 4.670548991874758,
      "tokens_seen": 122421248
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009725565719788156,
      "loss": 0.1463,
      "theoretical_loss": 4.669211868964052,
      "tokens_seen": 122683392
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009724763280372332,
      "loss": 0.1496,
      "theoretical_loss": 4.66787839814961,
      "tokens_seen": 122945536
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009723960840956508,
      "loss": 0.1477,
      "theoretical_loss": 4.666548561701285,
      "tokens_seen": 123207680
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009723158401540683,
      "loss": 0.1463,
      "theoretical_loss": 4.665222342012491,
      "tokens_seen": 123469824
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009722355962124861,
      "loss": 0.1472,
      "theoretical_loss": 4.663899721599093,
      "tokens_seen": 123731968
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009721553522709036,
      "loss": 0.1446,
      "theoretical_loss": 4.6625806830982865,
      "tokens_seen": 123994112
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009720751083293211,
      "loss": 0.1434,
      "theoretical_loss": 4.661265209267507,
      "tokens_seen": 124256256
    },
    {
      "epoch": 0.04,
      "objective/train/advantage_avg": 0.006344511639326811,
      "objective/train/docs_used": 52572,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 2.8837485313415527,
      "objective/train/original_loss": 2.8837485313415527,
      "objective/train/theoretical_loss": 4.659953282983348,
      "objective/train/tokens_used": 144978400,
      "objective/train/value_avg": -0.011322021484375,
      "objective/train/value_loss": 0.0003046532510779798,
      "objective/train/value_max": -0.001964569091796875,
      "objective/train/value_min": -0.193603515625,
      "objective/train/value_reward_corr": 0.3217214219555613,
      "objective/train/value_std": 0.007747650146484375,
      "objective/train/weight_avg": 1.0064884424209595,
      "objective/train/weighted_lm_loss": 2.9043095111846924,
      "objective/train/weights_max": 1.1398353576660156,
      "objective/train/weights_min": 0.37116101384162903,
      "theoretical_loss": 4.659953282983348,
      "tokens_seen": 124518400
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009719948643877388,
      "loss": 0.1463,
      "theoretical_loss": 4.659953282983348,
      "tokens_seen": 124518400
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009719146204461563,
      "loss": 0.1482,
      "theoretical_loss": 4.658644887240481,
      "tokens_seen": 124780544
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009718343765045739,
      "loss": 0.147,
      "theoretical_loss": 4.657340005150602,
      "tokens_seen": 125042688
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009717541325629915,
      "loss": 0.1462,
      "theoretical_loss": 4.656038619941382,
      "tokens_seen": 125304832
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009716738886214091,
      "loss": 0.1434,
      "theoretical_loss": 4.654740714955429,
      "tokens_seen": 125566976
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009715936446798266,
      "loss": 0.1433,
      "theoretical_loss": 4.653446273649259,
      "tokens_seen": 125829120
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009715134007382443,
      "loss": 0.1455,
      "theoretical_loss": 4.652155279592286,
      "tokens_seen": 126091264
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009714331567966619,
      "loss": 0.1429,
      "theoretical_loss": 4.650867716465819,
      "tokens_seen": 126353408
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009713529128550794,
      "loss": 0.1428,
      "theoretical_loss": 4.6495835680620665,
      "tokens_seen": 126615552
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009712726689134971,
      "loss": 0.1406,
      "theoretical_loss": 4.648302818283158,
      "tokens_seen": 126877696
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009711924249719146,
      "loss": 0.1459,
      "theoretical_loss": 4.64702545114017,
      "tokens_seen": 127139840
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009711121810303323,
      "loss": 0.1425,
      "theoretical_loss": 4.645751450752172,
      "tokens_seen": 127401984
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009710319370887498,
      "loss": 0.1386,
      "theoretical_loss": 4.644480801345268,
      "tokens_seen": 127664128
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009709516931471673,
      "loss": 0.1435,
      "theoretical_loss": 4.643213487251664,
      "tokens_seen": 127926272
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.000970871449205585,
      "loss": 0.1397,
      "theoretical_loss": 4.641949492908737,
      "tokens_seen": 128188416
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009707912052640026,
      "loss": 0.1406,
      "theoretical_loss": 4.640688802858113,
      "tokens_seen": 128450560
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009707109613224202,
      "loss": 0.1436,
      "theoretical_loss": 4.63943140174476,
      "tokens_seen": 128712704
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009706307173808378,
      "loss": 0.1405,
      "theoretical_loss": 4.6381772743160905,
      "tokens_seen": 128974848
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009705504734392554,
      "loss": 0.1423,
      "theoretical_loss": 4.636926405421065,
      "tokens_seen": 129236992
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009704702294976729,
      "loss": 0.1387,
      "theoretical_loss": 4.635678780009318,
      "tokens_seen": 129499136
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009703899855560905,
      "loss": 0.1392,
      "theoretical_loss": 4.634434383130284,
      "tokens_seen": 129761280
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009703097416145081,
      "loss": 0.1422,
      "theoretical_loss": 4.633193199932336,
      "tokens_seen": 130023424
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009702294976729256,
      "loss": 0.1425,
      "theoretical_loss": 4.631955215661932,
      "tokens_seen": 130285568
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009701492537313433,
      "loss": 0.1425,
      "theoretical_loss": 4.630720415662774,
      "tokens_seen": 130547712
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009700690097897609,
      "loss": 0.1353,
      "theoretical_loss": 4.629488785374969,
      "tokens_seen": 130809856
    },
    {
      "epoch": 0.04,
      "objective/train/advantage_avg": 0.005258428864181042,
      "objective/train/docs_used": 54897,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 2.703356981277466,
      "objective/train/original_loss": 2.7033567428588867,
      "objective/train/theoretical_loss": 4.628260310334209,
      "objective/train/tokens_used": 151532000,
      "objective/train/value_avg": -0.01247406005859375,
      "objective/train/value_loss": 0.0005161120789125562,
      "objective/train/value_max": -0.0016937255859375,
      "objective/train/value_min": -0.182861328125,
      "objective/train/value_reward_corr": 0.283905402927882,
      "objective/train/value_std": 0.01013946533203125,
      "objective/train/weight_avg": 1.0054808855056763,
      "objective/train/weighted_lm_loss": 2.7186498641967773,
      "objective/train/weights_max": 1.149869441986084,
      "objective/train/weights_min": 0.2252407968044281,
      "theoretical_loss": 4.628260310334209,
      "tokens_seen": 131072000
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009699887658481786,
      "loss": 0.1388,
      "theoretical_loss": 4.628260310334209,
      "tokens_seen": 131072000
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009699085219065961,
      "loss": 0.1394,
      "theoretical_loss": 4.6270349761709495,
      "tokens_seen": 131334144
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009698282779650137,
      "loss": 0.1354,
      "theoretical_loss": 4.625812768609601,
      "tokens_seen": 131596288
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009697480340234313,
      "loss": 0.1392,
      "theoretical_loss": 4.62459367346773,
      "tokens_seen": 131858432
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009696677900818488,
      "loss": 0.1383,
      "theoretical_loss": 4.623377676655271,
      "tokens_seen": 132120576
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009695875461402664,
      "loss": 0.1387,
      "theoretical_loss": 4.622164764173735,
      "tokens_seen": 132382720
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.000969507302198684,
      "loss": 0.1389,
      "theoretical_loss": 4.62095492211544,
      "tokens_seen": 132644864
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009694270582571016,
      "loss": 0.1438,
      "theoretical_loss": 4.6197481366627455,
      "tokens_seen": 132907008
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009693468143155191,
      "loss": 0.1385,
      "theoretical_loss": 4.618544394087287,
      "tokens_seen": 133169152
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009692665703739369,
      "loss": 0.1327,
      "theoretical_loss": 4.617343680749233,
      "tokens_seen": 133431296
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009691863264323544,
      "loss": 0.1351,
      "theoretical_loss": 4.61614598309653,
      "tokens_seen": 133693440
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009691060824907719,
      "loss": 0.1364,
      "theoretical_loss": 4.614951287664179,
      "tokens_seen": 133955584
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009690258385491896,
      "loss": 0.1374,
      "theoretical_loss": 4.613759581073502,
      "tokens_seen": 134217728
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009689455946076071,
      "loss": 0.1366,
      "theoretical_loss": 4.612570850031418,
      "tokens_seen": 134479872
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009688653506660247,
      "loss": 0.1337,
      "theoretical_loss": 4.611385081329736,
      "tokens_seen": 134742016
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009687851067244423,
      "loss": 0.1349,
      "theoretical_loss": 4.610202261844444,
      "tokens_seen": 135004160
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009687048627828599,
      "loss": 0.1352,
      "theoretical_loss": 4.6090223785350135,
      "tokens_seen": 135266304
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009686246188412776,
      "loss": 0.1329,
      "theoretical_loss": 4.607845418443706,
      "tokens_seen": 135528448
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009685443748996951,
      "loss": 0.1354,
      "theoretical_loss": 4.606671368694888,
      "tokens_seen": 135790592
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009684641309581127,
      "loss": 0.1341,
      "theoretical_loss": 4.6055002164943595,
      "tokens_seen": 136052736
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009683838870165303,
      "loss": 0.1295,
      "theoretical_loss": 4.604331949128672,
      "tokens_seen": 136314880
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009683036430749479,
      "loss": 0.134,
      "theoretical_loss": 4.603166553964474,
      "tokens_seen": 136577024
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009682233991333654,
      "loss": 0.1367,
      "theoretical_loss": 4.60200401844785,
      "tokens_seen": 136839168
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009681431551917831,
      "loss": 0.1308,
      "theoretical_loss": 4.6008443301036746,
      "tokens_seen": 137101312
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009680629112502006,
      "loss": 0.1338,
      "theoretical_loss": 4.5996874765349585,
      "tokens_seen": 137363456
    },
    {
      "epoch": 0.04,
      "objective/train/advantage_avg": 0.0035103908739984035,
      "objective/train/docs_used": 57180,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 2.7642529010772705,
      "objective/train/original_loss": 2.7642529010772705,
      "objective/train/theoretical_loss": 4.598533445422221,
      "objective/train/tokens_used": 158085600,
      "objective/train/value_avg": -0.01120758056640625,
      "objective/train/value_loss": 0.0006862494046799839,
      "objective/train/value_max": -0.0017547607421875,
      "objective/train/value_min": -0.2978515625,
      "objective/train/value_reward_corr": 0.42945186696590126,
      "objective/train/value_std": 0.00952911376953125,
      "objective/train/weight_avg": 1.0038180351257324,
      "objective/train/weighted_lm_loss": 2.776818037033081,
      "objective/train/weights_max": 1.2096216678619385,
      "objective/train/weights_min": 0.36997780203819275,
      "theoretical_loss": 4.598533445422221,
      "tokens_seen": 137625600
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009679826673086181,
      "loss": 0.1328,
      "theoretical_loss": 4.598533445422221,
      "tokens_seen": 137625600
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009679024233670359,
      "loss": 0.1312,
      "theoretical_loss": 4.597382224522855,
      "tokens_seen": 137887744
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009678221794254534,
      "loss": 0.1327,
      "theoretical_loss": 4.596233801670502,
      "tokens_seen": 138149888
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.000967741935483871,
      "loss": 0.1315,
      "theoretical_loss": 4.595088164774435,
      "tokens_seen": 138412032
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009676616915422886,
      "loss": 0.134,
      "theoretical_loss": 4.593945301818941,
      "tokens_seen": 138674176
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009675814476007062,
      "loss": 0.1287,
      "theoretical_loss": 4.592805200862726,
      "tokens_seen": 138936320
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009675012036591238,
      "loss": 0.1307,
      "theoretical_loss": 4.591667850038302,
      "tokens_seen": 139198464
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009674209597175413,
      "loss": 0.13,
      "theoretical_loss": 4.590533237551401,
      "tokens_seen": 139460608
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009673407157759589,
      "loss": 0.1303,
      "theoretical_loss": 4.589401351680385,
      "tokens_seen": 139722752
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009672604718343765,
      "loss": 0.1324,
      "theoretical_loss": 4.588272180775659,
      "tokens_seen": 139984896
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009671802278927941,
      "loss": 0.1279,
      "theoretical_loss": 4.587145713259102,
      "tokens_seen": 140247040
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009670999839512117,
      "loss": 0.1287,
      "theoretical_loss": 4.5860219376234905,
      "tokens_seen": 140509184
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009670197400096294,
      "loss": 0.1311,
      "theoretical_loss": 4.584900842431934,
      "tokens_seen": 140771328
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009669394960680469,
      "loss": 0.1271,
      "theoretical_loss": 4.583782416317316,
      "tokens_seen": 141033472
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009668592521264644,
      "loss": 0.1287,
      "theoretical_loss": 4.582666647981739,
      "tokens_seen": 141295616
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009667790081848821,
      "loss": 0.1268,
      "theoretical_loss": 4.581553526195974,
      "tokens_seen": 141557760
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009666987642432996,
      "loss": 0.1293,
      "theoretical_loss": 4.580443039798922,
      "tokens_seen": 141819904
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009666185203017172,
      "loss": 0.1296,
      "theoretical_loss": 4.57933517769707,
      "tokens_seen": 142082048
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009665382763601348,
      "loss": 0.1308,
      "theoretical_loss": 4.578229928863959,
      "tokens_seen": 142344192
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009664580324185524,
      "loss": 0.1281,
      "theoretical_loss": 4.57712728233966,
      "tokens_seen": 142606336
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00096637778847697,
      "loss": 0.1267,
      "theoretical_loss": 4.576027227230245,
      "tokens_seen": 142868480
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009662975445353877,
      "loss": 0.1325,
      "theoretical_loss": 4.574929752707274,
      "tokens_seen": 143130624
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009662173005938052,
      "loss": 0.1283,
      "theoretical_loss": 4.573834848007284,
      "tokens_seen": 143392768
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009661370566522228,
      "loss": 0.127,
      "theoretical_loss": 4.572742502431272,
      "tokens_seen": 143654912
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009660568127106404,
      "loss": 0.1264,
      "theoretical_loss": 4.571652705344202,
      "tokens_seen": 143917056
    },
    {
      "epoch": 0.04,
      "objective/train/advantage_avg": 0.0026565194129943848,
      "objective/train/docs_used": 59583,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 2.5324413776397705,
      "objective/train/original_loss": 2.5324416160583496,
      "objective/train/theoretical_loss": 4.570565446174504,
      "objective/train/tokens_used": 164639200,
      "objective/train/value_avg": -0.01026153564453125,
      "objective/train/value_loss": 0.0005388536374084651,
      "objective/train/value_max": -0.0011034011840820312,
      "objective/train/value_min": -0.2158203125,
      "objective/train/value_reward_corr": 0.41669318991791116,
      "objective/train/value_std": 0.0081024169921875,
      "objective/train/weight_avg": 1.0029032230377197,
      "objective/train/weighted_lm_loss": 2.539125442504883,
      "objective/train/weights_max": 1.1576155424118042,
      "objective/train/weights_min": 0.37208810448646545,
      "theoretical_loss": 4.570565446174504,
      "tokens_seen": 144179200
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009659765687690579,
      "loss": 0.127,
      "theoretical_loss": 4.570565446174504,
      "tokens_seen": 144179200
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009658963248274756,
      "loss": 0.1256,
      "theoretical_loss": 4.569480714413578,
      "tokens_seen": 144441344
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009658160808858931,
      "loss": 0.1276,
      "theoretical_loss": 4.568398499615305,
      "tokens_seen": 144703488
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009657358369443107,
      "loss": 0.1274,
      "theoretical_loss": 4.56731879139557,
      "tokens_seen": 144965632
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009656555930027284,
      "loss": 0.1258,
      "theoretical_loss": 4.566241579431776,
      "tokens_seen": 145227776
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009655753490611459,
      "loss": 0.1274,
      "theoretical_loss": 4.565166853462371,
      "tokens_seen": 145489920
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009654951051195635,
      "loss": 0.1289,
      "theoretical_loss": 4.564094603286375,
      "tokens_seen": 145752064
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009654148611779811,
      "loss": 0.1239,
      "theoretical_loss": 4.5630248187629245,
      "tokens_seen": 146014208
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009653346172363987,
      "loss": 0.1252,
      "theoretical_loss": 4.561957489810798,
      "tokens_seen": 146276352
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009652543732948162,
      "loss": 0.1234,
      "theoretical_loss": 4.5608926064079665,
      "tokens_seen": 146538496
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009651741293532339,
      "loss": 0.127,
      "theoretical_loss": 4.559830158591139,
      "tokens_seen": 146800640
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009650938854116514,
      "loss": 0.1215,
      "theoretical_loss": 4.558770136455316,
      "tokens_seen": 147062784
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009650136414700689,
      "loss": 0.1245,
      "theoretical_loss": 4.557712530153342,
      "tokens_seen": 147324928
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009649333975284866,
      "loss": 0.1217,
      "theoretical_loss": 4.556657329895469,
      "tokens_seen": 147587072
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009648531535869042,
      "loss": 0.125,
      "theoretical_loss": 4.5556045259489135,
      "tokens_seen": 147849216
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009647729096453219,
      "loss": 0.1258,
      "theoretical_loss": 4.554554108637437,
      "tokens_seen": 148111360
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0009646926657037394,
      "loss": 0.1232,
      "theoretical_loss": 4.553506068340907,
      "tokens_seen": 148373504
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000964612421762157,
      "loss": 0.1224,
      "theoretical_loss": 4.552460395494878,
      "tokens_seen": 148635648
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009645321778205746,
      "loss": 0.1258,
      "theoretical_loss": 4.55141708059017,
      "tokens_seen": 148897792
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009644519338789921,
      "loss": 0.1209,
      "theoretical_loss": 4.5503761141724555,
      "tokens_seen": 149159936
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009643716899374097,
      "loss": 0.1227,
      "theoretical_loss": 4.549337486841843,
      "tokens_seen": 149422080
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009642914459958273,
      "loss": 0.1231,
      "theoretical_loss": 4.548301189252473,
      "tokens_seen": 149684224
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000964211202054245,
      "loss": 0.1237,
      "theoretical_loss": 4.54726721211211,
      "tokens_seen": 149946368
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009641309581126625,
      "loss": 0.1212,
      "theoretical_loss": 4.546235546181743,
      "tokens_seen": 150208512
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009640507141710802,
      "loss": 0.118,
      "theoretical_loss": 4.545206182275189,
      "tokens_seen": 150470656
    },
    {
      "epoch": 0.05,
      "objective/train/advantage_avg": 0.002808451419696212,
      "objective/train/docs_used": 61752,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 2.444796085357666,
      "objective/train/original_loss": 2.444795846939087,
      "objective/train/theoretical_loss": 4.5441791112587016,
      "objective/train/tokens_used": 171192800,
      "objective/train/value_avg": -0.01094818115234375,
      "objective/train/value_loss": 0.0004207800084259361,
      "objective/train/value_max": -0.0013093948364257812,
      "objective/train/value_min": -0.256103515625,
      "objective/train/value_reward_corr": 0.4595803265485007,
      "objective/train/value_std": 0.00933074951171875,
      "objective/train/weight_avg": 1.0030062198638916,
      "objective/train/weighted_lm_loss": 2.4522807598114014,
      "objective/train/weights_max": 1.231403112411499,
      "objective/train/weights_min": 0.3714272677898407,
      "theoretical_loss": 4.5441791112587016,
      "tokens_seen": 150732800
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009639704702294977,
      "loss": 0.1196,
      "theoretical_loss": 4.5441791112587016,
      "tokens_seen": 150732800
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009638902262879152,
      "loss": 0.1256,
      "theoretical_loss": 4.5431543240505725,
      "tokens_seen": 150994944
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009638099823463329,
      "loss": 0.123,
      "theoretical_loss": 4.5421318116207585,
      "tokens_seen": 151257088
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009637297384047504,
      "loss": 0.1184,
      "theoretical_loss": 4.541111564990485,
      "tokens_seen": 151519232
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009636494944631681,
      "loss": 0.1206,
      "theoretical_loss": 4.540093575231879,
      "tokens_seen": 151781376
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009635692505215856,
      "loss": 0.119,
      "theoretical_loss": 4.539077833467582,
      "tokens_seen": 152043520
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009634890065800032,
      "loss": 0.1207,
      "theoretical_loss": 4.538064330870389,
      "tokens_seen": 152305664
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009634087626384209,
      "loss": 0.1227,
      "theoretical_loss": 4.537053058662869,
      "tokens_seen": 152567808
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009633285186968385,
      "loss": 0.1211,
      "theoretical_loss": 4.536044008117005,
      "tokens_seen": 152829952
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000963248274755256,
      "loss": 0.1239,
      "theoretical_loss": 4.535037170553833,
      "tokens_seen": 153092096
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009631680308136736,
      "loss": 0.1197,
      "theoretical_loss": 4.534032537343078,
      "tokens_seen": 153354240
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009630877868720912,
      "loss": 0.1227,
      "theoretical_loss": 4.533030099902803,
      "tokens_seen": 153616384
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009630075429305087,
      "loss": 0.1232,
      "theoretical_loss": 4.53202984969905,
      "tokens_seen": 153878528
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009629272989889264,
      "loss": 0.1246,
      "theoretical_loss": 4.531031778245499,
      "tokens_seen": 154140672
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009628470550473439,
      "loss": 0.1214,
      "theoretical_loss": 4.530035877103115,
      "tokens_seen": 154402816
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009627668111057615,
      "loss": 0.121,
      "theoretical_loss": 4.529042137879809,
      "tokens_seen": 154664960
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009626865671641792,
      "loss": 0.1184,
      "theoretical_loss": 4.528050552230092,
      "tokens_seen": 154927104
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009626063232225967,
      "loss": 0.1205,
      "theoretical_loss": 4.527061111854746,
      "tokens_seen": 155189248
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009625260792810143,
      "loss": 0.1217,
      "theoretical_loss": 4.526073808500481,
      "tokens_seen": 155451392
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009624458353394319,
      "loss": 0.1181,
      "theoretical_loss": 4.525088633959613,
      "tokens_seen": 155713536
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009623655913978495,
      "loss": 0.1176,
      "theoretical_loss": 4.524105580069728,
      "tokens_seen": 155975680
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009622853474562671,
      "loss": 0.1178,
      "theoretical_loss": 4.523124638713361,
      "tokens_seen": 156237824
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009622051035146847,
      "loss": 0.1206,
      "theoretical_loss": 4.522145801817673,
      "tokens_seen": 156499968
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009621248595731022,
      "loss": 0.1165,
      "theoretical_loss": 4.521169061354129,
      "tokens_seen": 156762112
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009620446156315198,
      "loss": 0.1179,
      "theoretical_loss": 4.520194409338185,
      "tokens_seen": 157024256
    },
    {
      "epoch": 0.05,
      "objective/train/advantage_avg": 0.002041465835645795,
      "objective/train/docs_used": 64097,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 2.572587728500366,
      "objective/train/original_loss": 2.572587728500366,
      "objective/train/theoretical_loss": 4.519221837828971,
      "objective/train/tokens_used": 177746400,
      "objective/train/value_avg": -0.00862884521484375,
      "objective/train/value_loss": 0.00036377867218106985,
      "objective/train/value_max": -0.0010986328125,
      "objective/train/value_min": -0.177490234375,
      "objective/train/value_reward_corr": 0.3439090032549563,
      "objective/train/value_std": 0.005954742431640625,
      "objective/train/weight_avg": 1.002203106880188,
      "objective/train/weighted_lm_loss": 2.5783188343048096,
      "objective/train/weights_max": 1.1067639589309692,
      "objective/train/weights_min": 0.22429333627223969,
      "theoretical_loss": 4.519221837828971,
      "tokens_seen": 157286400
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009619643716899374,
      "loss": 0.1204,
      "theoretical_loss": 4.519221837828971,
      "tokens_seen": 157286400
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000961884127748355,
      "loss": 0.1203,
      "theoretical_loss": 4.51825133892898,
      "tokens_seen": 157548544
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009618038838067727,
      "loss": 0.1186,
      "theoretical_loss": 4.517282904783764,
      "tokens_seen": 157810688
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009617236398651902,
      "loss": 0.1215,
      "theoretical_loss": 4.516316527581621,
      "tokens_seen": 158072832
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009616433959236078,
      "loss": 0.1222,
      "theoretical_loss": 4.515352199553295,
      "tokens_seen": 158334976
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009615631519820254,
      "loss": 0.1203,
      "theoretical_loss": 4.514389912971679,
      "tokens_seen": 158597120
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009614829080404429,
      "loss": 0.1202,
      "theoretical_loss": 4.513429660151513,
      "tokens_seen": 158859264
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009614026640988605,
      "loss": 0.1169,
      "theoretical_loss": 4.51247143344909,
      "tokens_seen": 159121408
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009613224201572781,
      "loss": 0.1199,
      "theoretical_loss": 4.511515225261961,
      "tokens_seen": 159383552
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009612421762156957,
      "loss": 0.1184,
      "theoretical_loss": 4.5105610280286506,
      "tokens_seen": 159645696
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009611619322741134,
      "loss": 0.1197,
      "theoretical_loss": 4.509608834228365,
      "tokens_seen": 159907840
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000961081688332531,
      "loss": 0.1157,
      "theoretical_loss": 4.508658636380705,
      "tokens_seen": 160169984
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009610014443909485,
      "loss": 0.1206,
      "theoretical_loss": 4.507710427045389,
      "tokens_seen": 160432128
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009609212004493661,
      "loss": 0.1179,
      "theoretical_loss": 4.50676419882197,
      "tokens_seen": 160694272
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009608409565077837,
      "loss": 0.1181,
      "theoretical_loss": 4.505819944349556,
      "tokens_seen": 160956416
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009607607125662012,
      "loss": 0.1152,
      "theoretical_loss": 4.504877656306535,
      "tokens_seen": 161218560
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009606804686246189,
      "loss": 0.1187,
      "theoretical_loss": 4.503937327410306,
      "tokens_seen": 161480704
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009606002246830364,
      "loss": 0.1185,
      "theoretical_loss": 4.502998950417004,
      "tokens_seen": 161742848
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000960519980741454,
      "loss": 0.1186,
      "theoretical_loss": 4.502062518121232,
      "tokens_seen": 162004992
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009604397367998717,
      "loss": 0.1161,
      "theoretical_loss": 4.501128023355796,
      "tokens_seen": 162267136
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009603594928582893,
      "loss": 0.116,
      "theoretical_loss": 4.500195458991443,
      "tokens_seen": 162529280
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009602792489167068,
      "loss": 0.1189,
      "theoretical_loss": 4.499264817936593,
      "tokens_seen": 162791424
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009601990049751244,
      "loss": 0.1202,
      "theoretical_loss": 4.498336093137089,
      "tokens_seen": 163053568
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000960118761033542,
      "loss": 0.1179,
      "theoretical_loss": 4.49740927757593,
      "tokens_seen": 163315712
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009600385170919595,
      "loss": 0.1162,
      "theoretical_loss": 4.496484364273021,
      "tokens_seen": 163577856
    },
    {
      "epoch": 0.05,
      "objective/train/advantage_avg": 0.004073688294738531,
      "objective/train/docs_used": 66569,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 2.1198720932006836,
      "objective/train/original_loss": 2.1198720932006836,
      "objective/train/theoretical_loss": 4.495561346284925,
      "objective/train/tokens_used": 184300000,
      "objective/train/value_avg": -0.009857177734375,
      "objective/train/value_loss": 0.00025354541139677167,
      "objective/train/value_max": -0.0008263587951660156,
      "objective/train/value_min": -0.55126953125,
      "objective/train/value_reward_corr": 0.4736301748253403,
      "objective/train/value_std": 0.00994110107421875,
      "objective/train/weight_avg": 1.004190444946289,
      "objective/train/weighted_lm_loss": 2.12888765335083,
      "objective/train/weights_max": 1.1655560731887817,
      "objective/train/weights_min": 0.36955323815345764,
      "theoretical_loss": 4.495561346284925,
      "tokens_seen": 163840000
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009599582731503772,
      "loss": 0.1154,
      "theoretical_loss": 4.495561346284925,
      "tokens_seen": 163840000
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009598780292087947,
      "loss": 0.117,
      "theoretical_loss": 4.494640216704598,
      "tokens_seen": 164102144
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009597977852672124,
      "loss": 0.1187,
      "theoretical_loss": 4.493720968661158,
      "tokens_seen": 164364288
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00095971754132563,
      "loss": 0.1155,
      "theoretical_loss": 4.492803595319623,
      "tokens_seen": 164626432
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009596372973840475,
      "loss": 0.1145,
      "theoretical_loss": 4.49188808988068,
      "tokens_seen": 164888576
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009595570534424652,
      "loss": 0.1111,
      "theoretical_loss": 4.490974445580429,
      "tokens_seen": 165150720
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009594768095008827,
      "loss": 0.1124,
      "theoretical_loss": 4.490062655690153,
      "tokens_seen": 165412864
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009593965655593003,
      "loss": 0.1125,
      "theoretical_loss": 4.489152713516077,
      "tokens_seen": 165675008
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009593163216177179,
      "loss": 0.1103,
      "theoretical_loss": 4.488244612399129,
      "tokens_seen": 165937152
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009592360776761355,
      "loss": 0.1118,
      "theoretical_loss": 4.487338345714707,
      "tokens_seen": 166199296
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000959155833734553,
      "loss": 0.1136,
      "theoretical_loss": 4.486433906872448,
      "tokens_seen": 166461440
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009590755897929706,
      "loss": 0.1148,
      "theoretical_loss": 4.485531289315997,
      "tokens_seen": 166723584
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009589953458513882,
      "loss": 0.1092,
      "theoretical_loss": 4.484630486522775,
      "tokens_seen": 166985728
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009589151019098058,
      "loss": 0.1164,
      "theoretical_loss": 4.483731492003757,
      "tokens_seen": 167247872
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009588348579682235,
      "loss": 0.1126,
      "theoretical_loss": 4.482834299303246,
      "tokens_seen": 167510016
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000958754614026641,
      "loss": 0.1125,
      "theoretical_loss": 4.481938901998647,
      "tokens_seen": 167772160
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009586743700850587,
      "loss": 0.1122,
      "theoretical_loss": 4.481045293700248,
      "tokens_seen": 168034304
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009585941261434762,
      "loss": 0.1135,
      "theoretical_loss": 4.480153468051001,
      "tokens_seen": 168296448
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009585138822018937,
      "loss": 0.1153,
      "theoretical_loss": 4.4792634187263065,
      "tokens_seen": 168558592
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009584336382603114,
      "loss": 0.1143,
      "theoretical_loss": 4.4783751394337905,
      "tokens_seen": 168820736
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009583533943187289,
      "loss": 0.1091,
      "theoretical_loss": 4.4774886239131,
      "tokens_seen": 169082880
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009582731503771465,
      "loss": 0.116,
      "theoretical_loss": 4.476603865935683,
      "tokens_seen": 169345024
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009581929064355642,
      "loss": 0.1161,
      "theoretical_loss": 4.475720859304583,
      "tokens_seen": 169607168
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009581126624939818,
      "loss": 0.1111,
      "theoretical_loss": 4.474839597854226,
      "tokens_seen": 169869312
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009580324185523993,
      "loss": 0.1142,
      "theoretical_loss": 4.473960075450218,
      "tokens_seen": 170131456
    },
    {
      "epoch": 0.05,
      "objective/train/advantage_avg": 0.004443011246621609,
      "objective/train/docs_used": 69022,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 2.34761905670166,
      "objective/train/original_loss": 2.34761905670166,
      "objective/train/theoretical_loss": 4.473082285989134,
      "objective/train/tokens_used": 190853600,
      "objective/train/value_avg": -0.009521484375,
      "objective/train/value_loss": 0.00034313887590542436,
      "objective/train/value_max": -0.001094818115234375,
      "objective/train/value_min": -0.2257080078125,
      "objective/train/value_reward_corr": 0.3206139523753071,
      "objective/train/value_std": 0.00771331787109375,
      "objective/train/weight_avg": 1.0045922994613647,
      "objective/train/weighted_lm_loss": 2.357492446899414,
      "objective/train/weights_max": 1.1233699321746826,
      "objective/train/weights_min": 0.23052240908145905,
      "theoretical_loss": 4.473082285989134,
      "tokens_seen": 170393600
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009579521746108169,
      "loss": 0.1172,
      "theoretical_loss": 4.473082285989134,
      "tokens_seen": 170393600
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009578719306692345,
      "loss": 0.114,
      "theoretical_loss": 4.472206223398325,
      "tokens_seen": 170655744
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000957791686727652,
      "loss": 0.1133,
      "theoretical_loss": 4.471331881635698,
      "tokens_seen": 170917888
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009577114427860697,
      "loss": 0.1127,
      "theoretical_loss": 4.470459254689533,
      "tokens_seen": 171180032
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009576311988444872,
      "loss": 0.1129,
      "theoretical_loss": 4.469588336578277,
      "tokens_seen": 171442176
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009575509549029048,
      "loss": 0.1119,
      "theoretical_loss": 4.468719121350343,
      "tokens_seen": 171704320
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009574707109613225,
      "loss": 0.1102,
      "theoretical_loss": 4.467851603083923,
      "tokens_seen": 171966464
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00095739046701974,
      "loss": 0.1137,
      "theoretical_loss": 4.466985775886784,
      "tokens_seen": 172228608
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009573102230781577,
      "loss": 0.1119,
      "theoretical_loss": 4.466121633896087,
      "tokens_seen": 172490752
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009572299791365752,
      "loss": 0.1152,
      "theoretical_loss": 4.465259171278182,
      "tokens_seen": 172752896
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009571497351949928,
      "loss": 0.1116,
      "theoretical_loss": 4.464398382228435,
      "tokens_seen": 173015040
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009570694912534104,
      "loss": 0.1146,
      "theoretical_loss": 4.463539260971023,
      "tokens_seen": 173277184
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000956989247311828,
      "loss": 0.1148,
      "theoretical_loss": 4.462681801758762,
      "tokens_seen": 173539328
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009569090033702455,
      "loss": 0.1142,
      "theoretical_loss": 4.461825998872914,
      "tokens_seen": 173801472
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009568287594286632,
      "loss": 0.1076,
      "theoretical_loss": 4.460971846623005,
      "tokens_seen": 174063616
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009567485154870808,
      "loss": 0.1106,
      "theoretical_loss": 4.460119339346643,
      "tokens_seen": 174325760
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009566682715454983,
      "loss": 0.1086,
      "theoretical_loss": 4.45926847140934,
      "tokens_seen": 174587904
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000956588027603916,
      "loss": 0.1096,
      "theoretical_loss": 4.45841923720433,
      "tokens_seen": 174850048
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009565077836623335,
      "loss": 0.1163,
      "theoretical_loss": 4.4575716311523905,
      "tokens_seen": 175112192
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009564275397207511,
      "loss": 0.1118,
      "theoretical_loss": 4.456725647701669,
      "tokens_seen": 175374336
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009563472957791687,
      "loss": 0.1112,
      "theoretical_loss": 4.455881281327508,
      "tokens_seen": 175636480
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009562670518375863,
      "loss": 0.1138,
      "theoretical_loss": 4.4550385265322685,
      "tokens_seen": 175898624
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009561868078960038,
      "loss": 0.1107,
      "theoretical_loss": 4.45419737784516,
      "tokens_seen": 176160768
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009561065639544214,
      "loss": 0.113,
      "theoretical_loss": 4.45335782982207,
      "tokens_seen": 176422912
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000956026320012839,
      "loss": 0.1134,
      "theoretical_loss": 4.452519877045393,
      "tokens_seen": 176685056
    },
    {
      "epoch": 0.05,
      "objective/train/advantage_avg": 0.004205191973596811,
      "objective/train/docs_used": 71338,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 2.1437151432037354,
      "objective/train/original_loss": 2.1437151432037354,
      "objective/train/theoretical_loss": 4.451683514123864,
      "objective/train/tokens_used": 197407200,
      "objective/train/value_avg": -0.009735107421875,
      "objective/train/value_loss": 0.00020696282444987446,
      "objective/train/value_max": -0.0012941360473632812,
      "objective/train/value_min": -0.50732421875,
      "objective/train/value_reward_corr": 0.42861167714862647,
      "objective/train/value_std": 0.00954437255859375,
      "objective/train/weight_avg": 1.0043038129806519,
      "objective/train/weighted_lm_loss": 2.1534533500671387,
      "objective/train/weights_max": 1.4412035942077637,
      "objective/train/weights_min": 0.3694094717502594,
      "theoretical_loss": 4.451683514123864,
      "tokens_seen": 176947200
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009559460760712567,
      "loss": 0.1134,
      "theoretical_loss": 4.451683514123864,
      "tokens_seen": 176947200
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009558658321296743,
      "loss": 0.1131,
      "theoretical_loss": 4.450848735692391,
      "tokens_seen": 177209344
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009557855881880918,
      "loss": 0.1129,
      "theoretical_loss": 4.450015536411886,
      "tokens_seen": 177471488
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009557053442465095,
      "loss": 0.1109,
      "theoretical_loss": 4.449183910969108,
      "tokens_seen": 177733632
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000955625100304927,
      "loss": 0.1089,
      "theoretical_loss": 4.448353854076494,
      "tokens_seen": 177995776
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009555448563633445,
      "loss": 0.1123,
      "theoretical_loss": 4.4475253604719995,
      "tokens_seen": 178257920
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009554646124217622,
      "loss": 0.1079,
      "theoretical_loss": 4.446698424918937,
      "tokens_seen": 178520064
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009553843684801797,
      "loss": 0.1115,
      "theoretical_loss": 4.44587304220582,
      "tokens_seen": 178782208
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009553041245385973,
      "loss": 0.1076,
      "theoretical_loss": 4.4450492071462,
      "tokens_seen": 179044352
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000955223880597015,
      "loss": 0.1146,
      "theoretical_loss": 4.444226914578513,
      "tokens_seen": 179306496
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009551436366554326,
      "loss": 0.1136,
      "theoretical_loss": 4.4434061593659235,
      "tokens_seen": 179568640
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009550633927138501,
      "loss": 0.1129,
      "theoretical_loss": 4.442586936396171,
      "tokens_seen": 179830784
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009549831487722677,
      "loss": 0.1116,
      "theoretical_loss": 4.441769240581412,
      "tokens_seen": 180092928
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009549029048306853,
      "loss": 0.1162,
      "theoretical_loss": 4.440953066858077,
      "tokens_seen": 180355072
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009548226608891028,
      "loss": 0.1104,
      "theoretical_loss": 4.4401384101867105,
      "tokens_seen": 180617216
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009547424169475205,
      "loss": 0.1108,
      "theoretical_loss": 4.439325265551826,
      "tokens_seen": 180879360
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000954662173005938,
      "loss": 0.1117,
      "theoretical_loss": 4.438513627961757,
      "tokens_seen": 181141504
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0009545819290643557,
      "loss": 0.1118,
      "theoretical_loss": 4.437703492448509,
      "tokens_seen": 181403648
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009545016851227733,
      "loss": 0.1139,
      "theoretical_loss": 4.436894854067614,
      "tokens_seen": 181665792
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009544214411811908,
      "loss": 0.1126,
      "theoretical_loss": 4.436087707897984,
      "tokens_seen": 181927936
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009543411972396085,
      "loss": 0.1126,
      "theoretical_loss": 4.435282049041769,
      "tokens_seen": 182190080
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000954260953298026,
      "loss": 0.1106,
      "theoretical_loss": 4.434477872624212,
      "tokens_seen": 182452224
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009541807093564436,
      "loss": 0.1133,
      "theoretical_loss": 4.433675173793507,
      "tokens_seen": 182714368
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009541004654148612,
      "loss": 0.1116,
      "theoretical_loss": 4.43287394772066,
      "tokens_seen": 182976512
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009540202214732788,
      "loss": 0.1147,
      "theoretical_loss": 4.43207418959935,
      "tokens_seen": 183238656
    },
    {
      "epoch": 0.06,
      "objective/train/advantage_avg": -0.000987537787295878,
      "objective/train/docs_used": 73723,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 2.2004032135009766,
      "objective/train/original_loss": 2.2004029750823975,
      "objective/train/theoretical_loss": 4.431275894645784,
      "objective/train/tokens_used": 203960800,
      "objective/train/value_avg": -0.00931549072265625,
      "objective/train/value_loss": 0.001105409930460155,
      "objective/train/value_max": -0.0009927749633789062,
      "objective/train/value_min": -0.28662109375,
      "objective/train/value_reward_corr": 0.2604573666222557,
      "objective/train/value_std": 0.00833892822265625,
      "objective/train/weight_avg": 0.9995287656784058,
      "objective/train/weighted_lm_loss": 2.196560859680176,
      "objective/train/weights_max": 1.1982022523880005,
      "objective/train/weights_min": 0.36976757645606995,
      "theoretical_loss": 4.431275894645784,
      "tokens_seen": 183500800
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009539399775316963,
      "loss": 0.1075,
      "theoretical_loss": 4.431275894645784,
      "tokens_seen": 183500800
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000953859733590114,
      "loss": 0.1111,
      "theoretical_loss": 4.43047905809857,
      "tokens_seen": 183762944
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009537794896485315,
      "loss": 0.1096,
      "theoretical_loss": 4.42968367521857,
      "tokens_seen": 184025088
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000953699245706949,
      "loss": 0.1113,
      "theoretical_loss": 4.428889741288771,
      "tokens_seen": 184287232
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009536190017653668,
      "loss": 0.1123,
      "theoretical_loss": 4.428097251614145,
      "tokens_seen": 184549376
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009535387578237843,
      "loss": 0.1104,
      "theoretical_loss": 4.427306201521524,
      "tokens_seen": 184811520
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000953458513882202,
      "loss": 0.1098,
      "theoretical_loss": 4.426516586359458,
      "tokens_seen": 185073664
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009533782699406195,
      "loss": 0.1119,
      "theoretical_loss": 4.425728401498089,
      "tokens_seen": 185335808
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009532980259990371,
      "loss": 0.1123,
      "theoretical_loss": 4.424941642329019,
      "tokens_seen": 185597952
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009532177820574547,
      "loss": 0.1109,
      "theoretical_loss": 4.42415630426518,
      "tokens_seen": 185860096
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009531375381158722,
      "loss": 0.1093,
      "theoretical_loss": 4.423372382740707,
      "tokens_seen": 186122240
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009530572941742898,
      "loss": 0.1064,
      "theoretical_loss": 4.422589873210806,
      "tokens_seen": 186384384
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009529770502327075,
      "loss": 0.1071,
      "theoretical_loss": 4.4218087711516345,
      "tokens_seen": 186646528
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009528968062911251,
      "loss": 0.1123,
      "theoretical_loss": 4.421029072060167,
      "tokens_seen": 186908672
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009528165623495426,
      "loss": 0.1095,
      "theoretical_loss": 4.420250771454078,
      "tokens_seen": 187170816
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009527363184079603,
      "loss": 0.1135,
      "theoretical_loss": 4.419473864871613,
      "tokens_seen": 187432960
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009526560744663778,
      "loss": 0.1138,
      "theoretical_loss": 4.4186983478714685,
      "tokens_seen": 187695104
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009525758305247953,
      "loss": 0.1132,
      "theoretical_loss": 4.417924216032667,
      "tokens_seen": 187957248
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000952495586583213,
      "loss": 0.1117,
      "theoretical_loss": 4.417151464954437,
      "tokens_seen": 188219392
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009524153426416305,
      "loss": 0.1107,
      "theoretical_loss": 4.416380090256095,
      "tokens_seen": 188481536
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009523350987000481,
      "loss": 0.1148,
      "theoretical_loss": 4.415610087576923,
      "tokens_seen": 188743680
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009522548547584658,
      "loss": 0.1119,
      "theoretical_loss": 4.414841452576049,
      "tokens_seen": 189005824
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009521746108168834,
      "loss": 0.1084,
      "theoretical_loss": 4.414074180932333,
      "tokens_seen": 189267968
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000952094366875301,
      "loss": 0.1087,
      "theoretical_loss": 4.413308268344249,
      "tokens_seen": 189530112
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009520141229337185,
      "loss": 0.1113,
      "theoretical_loss": 4.412543710529766,
      "tokens_seen": 189792256
    },
    {
      "epoch": 0.06,
      "objective/train/advantage_avg": 0.002621724735945463,
      "objective/train/docs_used": 76077,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 2.243046760559082,
      "objective/train/original_loss": 2.243046760559082,
      "objective/train/theoretical_loss": 4.411780503226238,
      "objective/train/tokens_used": 210514400,
      "objective/train/value_avg": -0.00859832763671875,
      "objective/train/value_loss": 0.0002450415340717882,
      "objective/train/value_max": -0.0006189346313476562,
      "objective/train/value_min": -0.332275390625,
      "objective/train/value_reward_corr": 0.4247518874973535,
      "objective/train/value_std": 0.008056640625,
      "objective/train/weight_avg": 1.002734899520874,
      "objective/train/weighted_lm_loss": 2.2489123344421387,
      "objective/train/weights_max": 1.2309222221374512,
      "objective/train/weights_min": 0.3706021308898926,
      "theoretical_loss": 4.411780503226238,
      "tokens_seen": 190054400
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009519338789921361,
      "loss": 0.1102,
      "theoretical_loss": 4.411780503226238,
      "tokens_seen": 190054400
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009518536350505537,
      "loss": 0.1125,
      "theoretical_loss": 4.4110186421902835,
      "tokens_seen": 190316544
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009517733911089713,
      "loss": 0.1074,
      "theoretical_loss": 4.4102581231976785,
      "tokens_seen": 190578688
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009516931471673888,
      "loss": 0.1119,
      "theoretical_loss": 4.409498942043237,
      "tokens_seen": 190840832
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009516129032258065,
      "loss": 0.1106,
      "theoretical_loss": 4.408741094540707,
      "tokens_seen": 191102976
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000951532659284224,
      "loss": 0.108,
      "theoretical_loss": 4.407984576522653,
      "tokens_seen": 191365120
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009514524153426416,
      "loss": 0.1106,
      "theoretical_loss": 4.407229383840347,
      "tokens_seen": 191627264
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009513721714010593,
      "loss": 0.1092,
      "theoretical_loss": 4.406475512363663,
      "tokens_seen": 191889408
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009512919274594768,
      "loss": 0.1056,
      "theoretical_loss": 4.405722957980962,
      "tokens_seen": 192151552
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009512116835178944,
      "loss": 0.1093,
      "theoretical_loss": 4.40497171659899,
      "tokens_seen": 192413696
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000951131439576312,
      "loss": 0.1061,
      "theoretical_loss": 4.404221784142768,
      "tokens_seen": 192675840
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009510511956347296,
      "loss": 0.1098,
      "theoretical_loss": 4.403473156555487,
      "tokens_seen": 192937984
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009509709516931472,
      "loss": 0.107,
      "theoretical_loss": 4.402725829798397,
      "tokens_seen": 193200128
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009508907077515648,
      "loss": 0.1066,
      "theoretical_loss": 4.4019797998507135,
      "tokens_seen": 193462272
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009508104638099823,
      "loss": 0.1042,
      "theoretical_loss": 4.401235062709502,
      "tokens_seen": 193724416
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009507302198684,
      "loss": 0.1081,
      "theoretical_loss": 4.400491614389582,
      "tokens_seen": 193986560
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009506499759268176,
      "loss": 0.1048,
      "theoretical_loss": 4.3997494509234185,
      "tokens_seen": 194248704
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009505697319852351,
      "loss": 0.1083,
      "theoretical_loss": 4.399008568361027,
      "tokens_seen": 194510848
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009504894880436528,
      "loss": 0.1097,
      "theoretical_loss": 4.398268962769867,
      "tokens_seen": 194772992
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009504092441020703,
      "loss": 0.1125,
      "theoretical_loss": 4.397530630234744,
      "tokens_seen": 195035136
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009503290001604879,
      "loss": 0.1067,
      "theoretical_loss": 4.396793566857708,
      "tokens_seen": 195297280
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009502487562189055,
      "loss": 0.1057,
      "theoretical_loss": 4.396057768757957,
      "tokens_seen": 195559424
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000950168512277323,
      "loss": 0.1076,
      "theoretical_loss": 4.395323232071737,
      "tokens_seen": 195821568
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009500882683357406,
      "loss": 0.1075,
      "theoretical_loss": 4.394589952952247,
      "tokens_seen": 196083712
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009500080243941583,
      "loss": 0.1034,
      "theoretical_loss": 4.393857927569534,
      "tokens_seen": 196345856
    },
    {
      "epoch": 0.06,
      "objective/train/advantage_avg": 0.0034133398439735174,
      "objective/train/docs_used": 78313,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 2.0345728397369385,
      "objective/train/original_loss": 2.0345726013183594,
      "objective/train/theoretical_loss": 4.393127152110409,
      "objective/train/tokens_used": 217068000,
      "objective/train/value_avg": -0.01177978515625,
      "objective/train/value_loss": 0.00046790740452706814,
      "objective/train/value_max": -0.00142669677734375,
      "objective/train/value_min": -0.64306640625,
      "objective/train/value_reward_corr": 0.5403801085285318,
      "objective/train/value_std": 0.01461029052734375,
      "objective/train/weight_avg": 1.0036334991455078,
      "objective/train/weighted_lm_loss": 2.0406572818756104,
      "objective/train/weights_max": 1.7312231063842773,
      "objective/train/weights_min": 0.3701556921005249,
      "theoretical_loss": 4.393127152110409,
      "tokens_seen": 196608000
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009499277804525759,
      "loss": 0.1063,
      "theoretical_loss": 4.393127152110409,
      "tokens_seen": 196608000
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009498475365109934,
      "loss": 0.1069,
      "theoretical_loss": 4.392397622778343,
      "tokens_seen": 196870144
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009497672925694111,
      "loss": 0.1068,
      "theoretical_loss": 4.391669335793372,
      "tokens_seen": 197132288
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009496870486278286,
      "loss": 0.1043,
      "theoretical_loss": 4.39094228739201,
      "tokens_seen": 197394432
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009496068046862462,
      "loss": 0.1086,
      "theoretical_loss": 4.390216473827143,
      "tokens_seen": 197656576
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009495265607446638,
      "loss": 0.1027,
      "theoretical_loss": 4.389491891367953,
      "tokens_seen": 197918720
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009494463168030813,
      "loss": 0.1075,
      "theoretical_loss": 4.388768536299808,
      "tokens_seen": 198180864
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000949366072861499,
      "loss": 0.1096,
      "theoretical_loss": 4.388046404924184,
      "tokens_seen": 198443008
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009492858289199166,
      "loss": 0.1058,
      "theoretical_loss": 4.387325493558566,
      "tokens_seen": 198705152
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009492055849783342,
      "loss": 0.1068,
      "theoretical_loss": 4.386605798536362,
      "tokens_seen": 198967296
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009491253410367518,
      "loss": 0.109,
      "theoretical_loss": 4.385887316206812,
      "tokens_seen": 199229440
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009490450970951693,
      "loss": 0.1029,
      "theoretical_loss": 4.385170042934896,
      "tokens_seen": 199491584
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009489648531535869,
      "loss": 0.1046,
      "theoretical_loss": 4.384453975101251,
      "tokens_seen": 199753728
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009488846092120045,
      "loss": 0.107,
      "theoretical_loss": 4.38373910910208,
      "tokens_seen": 200015872
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009488043652704221,
      "loss": 0.1056,
      "theoretical_loss": 4.383025441349063,
      "tokens_seen": 200278016
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009487241213288396,
      "loss": 0.1099,
      "theoretical_loss": 4.382312968269276,
      "tokens_seen": 200540160
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009486438773872573,
      "loss": 0.105,
      "theoretical_loss": 4.381601686305098,
      "tokens_seen": 200802304
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009485636334456749,
      "loss": 0.1039,
      "theoretical_loss": 4.38089159191413,
      "tokens_seen": 201064448
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009484833895040924,
      "loss": 0.1014,
      "theoretical_loss": 4.380182681569111,
      "tokens_seen": 201326592
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009484031455625101,
      "loss": 0.1062,
      "theoretical_loss": 4.379474951757829,
      "tokens_seen": 201588736
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009483229016209276,
      "loss": 0.107,
      "theoretical_loss": 4.378768398983042,
      "tokens_seen": 201850880
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009482426576793453,
      "loss": 0.1042,
      "theoretical_loss": 4.378063019762392,
      "tokens_seen": 202113024
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009481624137377628,
      "loss": 0.105,
      "theoretical_loss": 4.377358810628324,
      "tokens_seen": 202375168
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009480821697961804,
      "loss": 0.1045,
      "theoretical_loss": 4.3766557681280025,
      "tokens_seen": 202637312
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000948001925854598,
      "loss": 0.1059,
      "theoretical_loss": 4.375953888823233,
      "tokens_seen": 202899456
    },
    {
      "epoch": 0.06,
      "objective/train/advantage_avg": 0.0012575258733704686,
      "objective/train/docs_used": 80605,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 2.184474229812622,
      "objective/train/original_loss": 2.184473991394043,
      "objective/train/theoretical_loss": 4.375253169290376,
      "objective/train/tokens_used": 223621600,
      "objective/train/value_avg": -0.01114654541015625,
      "objective/train/value_loss": 0.000553421676158905,
      "objective/train/value_max": -0.0007014274597167969,
      "objective/train/value_min": -0.646484375,
      "objective/train/value_reward_corr": 0.5880496589440656,
      "objective/train/value_std": 0.0163116455078125,
      "objective/train/weight_avg": 1.0015053749084473,
      "objective/train/weighted_lm_loss": 2.1872315406799316,
      "objective/train/weights_max": 1.1879366636276245,
      "objective/train/weights_min": 0.3710477352142334,
      "theoretical_loss": 4.375253169290376,
      "tokens_seen": 203161600
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009479216819130155,
      "loss": 0.1046,
      "theoretical_loss": 4.375253169290376,
      "tokens_seen": 203161600
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009478414379714331,
      "loss": 0.1057,
      "theoretical_loss": 4.374553606120274,
      "tokens_seen": 203423744
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009477611940298508,
      "loss": 0.1046,
      "theoretical_loss": 4.373855195918162,
      "tokens_seen": 203685888
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009476809500882684,
      "loss": 0.1076,
      "theoretical_loss": 4.3731579353036,
      "tokens_seen": 203948032
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009476007061466859,
      "loss": 0.1036,
      "theoretical_loss": 4.372461820910382,
      "tokens_seen": 204210176
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009475204622051036,
      "loss": 0.1063,
      "theoretical_loss": 4.371766849386468,
      "tokens_seen": 204472320
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009474402182635211,
      "loss": 0.1066,
      "theoretical_loss": 4.3710730173939005,
      "tokens_seen": 204734464
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009473599743219387,
      "loss": 0.1061,
      "theoretical_loss": 4.370380321608731,
      "tokens_seen": 204996608
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009472797303803563,
      "loss": 0.1077,
      "theoretical_loss": 4.369688758720937,
      "tokens_seen": 205258752
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009471994864387738,
      "loss": 0.1037,
      "theoretical_loss": 4.368998325434355,
      "tokens_seen": 205520896
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009471192424971916,
      "loss": 0.1023,
      "theoretical_loss": 4.3683090184666,
      "tokens_seen": 205783040
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009470389985556091,
      "loss": 0.1015,
      "theoretical_loss": 4.367620834548987,
      "tokens_seen": 206045184
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009469587546140267,
      "loss": 0.104,
      "theoretical_loss": 4.3669337704264635,
      "tokens_seen": 206307328
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009468785106724443,
      "loss": 0.1068,
      "theoretical_loss": 4.366247822857533,
      "tokens_seen": 206569472
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009467982667308619,
      "loss": 0.1048,
      "theoretical_loss": 4.365562988614176,
      "tokens_seen": 206831616
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009467180227892794,
      "loss": 0.1034,
      "theoretical_loss": 4.364879264481787,
      "tokens_seen": 207093760
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000946637778847697,
      "loss": 0.1035,
      "theoretical_loss": 4.364196647259092,
      "tokens_seen": 207355904
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009465575349061146,
      "loss": 0.1061,
      "theoretical_loss": 4.363515133758084,
      "tokens_seen": 207618048
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009464772909645321,
      "loss": 0.1075,
      "theoretical_loss": 4.3628347208039475,
      "tokens_seen": 207880192
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009463970470229499,
      "loss": 0.1018,
      "theoretical_loss": 4.362155405234985,
      "tokens_seen": 208142336
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009463168030813674,
      "loss": 0.109,
      "theoretical_loss": 4.361477183902554,
      "tokens_seen": 208404480
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000946236559139785,
      "loss": 0.1052,
      "theoretical_loss": 4.360800053670989,
      "tokens_seen": 208666624
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009461563151982026,
      "loss": 0.1031,
      "theoretical_loss": 4.360124011417536,
      "tokens_seen": 208928768
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009460760712566201,
      "loss": 0.1047,
      "theoretical_loss": 4.359449054032282,
      "tokens_seen": 209190912
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009459958273150377,
      "loss": 0.1042,
      "theoretical_loss": 4.358775178418089,
      "tokens_seen": 209453056
    },
    {
      "epoch": 0.06,
      "objective/train/advantage_avg": 0.0007151683676056564,
      "objective/train/docs_used": 82917,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 2.0514161586761475,
      "objective/train/original_loss": 2.0514163970947266,
      "objective/train/theoretical_loss": 4.358102381490517,
      "objective/train/tokens_used": 230175200,
      "objective/train/value_avg": -0.01009368896484375,
      "objective/train/value_loss": 0.0004249107150826603,
      "objective/train/value_max": -0.0008263587951660156,
      "objective/train/value_min": -0.296142578125,
      "objective/train/value_reward_corr": 0.48272459031421056,
      "objective/train/value_std": 0.01027679443359375,
      "objective/train/weight_avg": 1.000907063484192,
      "objective/train/weighted_lm_loss": 2.0531537532806396,
      "objective/train/weights_max": 1.2134668827056885,
      "objective/train/weights_min": 0.370398610830307,
      "theoretical_loss": 4.358102381490517,
      "tokens_seen": 209715200
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009459155833734553,
      "loss": 0.1048,
      "theoretical_loss": 4.358102381490517,
      "tokens_seen": 209715200
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009458353394318729,
      "loss": 0.1051,
      "theoretical_loss": 4.3574306601777675,
      "tokens_seen": 209977344
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009457550954902905,
      "loss": 0.1009,
      "theoretical_loss": 4.356760011420608,
      "tokens_seen": 210239488
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009456748515487081,
      "loss": 0.1085,
      "theoretical_loss": 4.3560904321723095,
      "tokens_seen": 210501632
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009455946076071257,
      "loss": 0.1056,
      "theoretical_loss": 4.355421919398576,
      "tokens_seen": 210763776
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009455143636655433,
      "loss": 0.1054,
      "theoretical_loss": 4.354754470077481,
      "tokens_seen": 211025920
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009454341197239609,
      "loss": 0.1029,
      "theoretical_loss": 4.354088081199402,
      "tokens_seen": 211288064
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009453538757823784,
      "loss": 0.1038,
      "theoretical_loss": 4.3534227497669535,
      "tokens_seen": 211550208
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009452736318407961,
      "loss": 0.1031,
      "theoretical_loss": 4.352758472794923,
      "tokens_seen": 211812352
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009451933878992136,
      "loss": 0.1055,
      "theoretical_loss": 4.352095247310208,
      "tokens_seen": 212074496
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009451131439576312,
      "loss": 0.1042,
      "theoretical_loss": 4.351433070351748,
      "tokens_seen": 212336640
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009450329000160488,
      "loss": 0.1056,
      "theoretical_loss": 4.350771938970466,
      "tokens_seen": 212598784
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009449526560744663,
      "loss": 0.1022,
      "theoretical_loss": 4.350111850229202,
      "tokens_seen": 212860928
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000944872412132884,
      "loss": 0.101,
      "theoretical_loss": 4.34945280120265,
      "tokens_seen": 213123072
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009447921681913016,
      "loss": 0.1025,
      "theoretical_loss": 4.348794788977298,
      "tokens_seen": 213385216
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009447119242497192,
      "loss": 0.1035,
      "theoretical_loss": 4.348137810651366,
      "tokens_seen": 213647360
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009446316803081368,
      "loss": 0.1016,
      "theoretical_loss": 4.347481863334738,
      "tokens_seen": 213909504
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009445514363665544,
      "loss": 0.104,
      "theoretical_loss": 4.346826944148912,
      "tokens_seen": 214171648
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0009444711924249719,
      "loss": 0.1048,
      "theoretical_loss": 4.3461730502269305,
      "tokens_seen": 214433792
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009443909484833896,
      "loss": 0.1036,
      "theoretical_loss": 4.345520178713323,
      "tokens_seen": 214695936
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009443107045418071,
      "loss": 0.1078,
      "theoretical_loss": 4.344868326764045,
      "tokens_seen": 214958080
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009442304606002246,
      "loss": 0.1053,
      "theoretical_loss": 4.344217491546422,
      "tokens_seen": 215220224
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009441502166586424,
      "loss": 0.1035,
      "theoretical_loss": 4.343567670239084,
      "tokens_seen": 215482368
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009440699727170599,
      "loss": 0.1054,
      "theoretical_loss": 4.342918860031914,
      "tokens_seen": 215744512
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009439897287754775,
      "loss": 0.1003,
      "theoretical_loss": 4.342271058125983,
      "tokens_seen": 216006656
    },
    {
      "epoch": 0.07,
      "objective/train/advantage_avg": 0.0016028911340981722,
      "objective/train/docs_used": 85191,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 2.0327014923095703,
      "objective/train/original_loss": 2.032701253890991,
      "objective/train/theoretical_loss": 4.341624261733497,
      "objective/train/tokens_used": 236728800,
      "objective/train/value_avg": -0.00833892822265625,
      "objective/train/value_loss": 0.00031526113161817193,
      "objective/train/value_max": -0.0006666183471679688,
      "objective/train/value_min": -0.25830078125,
      "objective/train/value_reward_corr": 0.44630488005091173,
      "objective/train/value_std": 0.008331298828125,
      "objective/train/weight_avg": 1.0017452239990234,
      "objective/train/weighted_lm_loss": 2.0359926223754883,
      "objective/train/weights_max": 1.1291202306747437,
      "objective/train/weights_min": 0.36889123916625977,
      "theoretical_loss": 4.341624261733497,
      "tokens_seen": 216268800
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009439094848338951,
      "loss": 0.1044,
      "theoretical_loss": 4.341624261733497,
      "tokens_seen": 216268800
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009438292408923127,
      "loss": 0.1056,
      "theoretical_loss": 4.340978468077735,
      "tokens_seen": 216530944
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009437489969507302,
      "loss": 0.1036,
      "theoretical_loss": 4.340333674392992,
      "tokens_seen": 216793088
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009436687530091478,
      "loss": 0.105,
      "theoretical_loss": 4.339689877924531,
      "tokens_seen": 217055232
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009435885090675654,
      "loss": 0.1036,
      "theoretical_loss": 4.3390470759285105,
      "tokens_seen": 217317376
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009435082651259829,
      "loss": 0.1036,
      "theoretical_loss": 4.338405265671941,
      "tokens_seen": 217579520
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009434280211844006,
      "loss": 0.099,
      "theoretical_loss": 4.337764444432625,
      "tokens_seen": 217841664
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009433477772428182,
      "loss": 0.1036,
      "theoretical_loss": 4.337124609499101,
      "tokens_seen": 218103808
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009432675333012359,
      "loss": 0.1,
      "theoretical_loss": 4.336485758170589,
      "tokens_seen": 218365952
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009431872893596534,
      "loss": 0.1058,
      "theoretical_loss": 4.335847887756934,
      "tokens_seen": 218628096
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009431070454180709,
      "loss": 0.1014,
      "theoretical_loss": 4.335210995578553,
      "tokens_seen": 218890240
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009430268014764886,
      "loss": 0.1035,
      "theoretical_loss": 4.334575078966383,
      "tokens_seen": 219152384
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009429465575349061,
      "loss": 0.1044,
      "theoretical_loss": 4.333940135261823,
      "tokens_seen": 219414528
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009428663135933237,
      "loss": 0.1014,
      "theoretical_loss": 4.333306161816684,
      "tokens_seen": 219676672
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009427860696517413,
      "loss": 0.1056,
      "theoretical_loss": 4.332673155993131,
      "tokens_seen": 219938816
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000942705825710159,
      "loss": 0.1012,
      "theoretical_loss": 4.332041115163636,
      "tokens_seen": 220200960
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009426255817685765,
      "loss": 0.1045,
      "theoretical_loss": 4.331410036710925,
      "tokens_seen": 220463104
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009425453378269941,
      "loss": 0.1026,
      "theoretical_loss": 4.330779918027919,
      "tokens_seen": 220725248
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009424650938854117,
      "loss": 0.1044,
      "theoretical_loss": 4.330150756517692,
      "tokens_seen": 220987392
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009423848499438292,
      "loss": 0.1022,
      "theoretical_loss": 4.3295225495934115,
      "tokens_seen": 221249536
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009423046060022469,
      "loss": 0.0974,
      "theoretical_loss": 4.328895294678292,
      "tokens_seen": 221511680
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009422243620606644,
      "loss": 0.1027,
      "theoretical_loss": 4.32826898920554,
      "tokens_seen": 221773824
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009421441181190821,
      "loss": 0.1052,
      "theoretical_loss": 4.3276436306183115,
      "tokens_seen": 222035968
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009420638741774996,
      "loss": 0.0996,
      "theoretical_loss": 4.327019216369651,
      "tokens_seen": 222298112
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009419836302359171,
      "loss": 0.101,
      "theoretical_loss": 4.32639574392245,
      "tokens_seen": 222560256
    },
    {
      "epoch": 0.07,
      "objective/train/advantage_avg": 0.002008789451792836,
      "objective/train/docs_used": 87601,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.8882607221603394,
      "objective/train/original_loss": 1.8882604837417603,
      "objective/train/theoretical_loss": 4.325773210749392,
      "objective/train/tokens_used": 243282400,
      "objective/train/value_avg": -0.01070404052734375,
      "objective/train/value_loss": 0.0003292278270237148,
      "objective/train/value_max": -0.0006799697875976562,
      "objective/train/value_min": -0.82373046875,
      "objective/train/value_reward_corr": 0.602646911223948,
      "objective/train/value_std": 0.01348876953125,
      "objective/train/weight_avg": 1.002161979675293,
      "objective/train/weighted_lm_loss": 1.892277717590332,
      "objective/train/weights_max": 1.4122933149337769,
      "objective/train/weights_min": 0.370164155960083,
      "theoretical_loss": 4.325773210749392,
      "tokens_seen": 222822400
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009419033862943349,
      "loss": 0.0975,
      "theoretical_loss": 4.325773210749392,
      "tokens_seen": 222822400
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009418231423527524,
      "loss": 0.1018,
      "theoretical_loss": 4.325151614332908,
      "tokens_seen": 223084544
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00094174289841117,
      "loss": 0.1032,
      "theoretical_loss": 4.3245309521651265,
      "tokens_seen": 223346688
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009416626544695876,
      "loss": 0.103,
      "theoretical_loss": 4.323911221747817,
      "tokens_seen": 223608832
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009415824105280052,
      "loss": 0.1049,
      "theoretical_loss": 4.323292420592356,
      "tokens_seen": 223870976
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009415021665864227,
      "loss": 0.1024,
      "theoretical_loss": 4.322674546219666,
      "tokens_seen": 224133120
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009414219226448404,
      "loss": 0.1029,
      "theoretical_loss": 4.322057596160174,
      "tokens_seen": 224395264
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009413416787032579,
      "loss": 0.0989,
      "theoretical_loss": 4.321441567953762,
      "tokens_seen": 224657408
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009412614347616754,
      "loss": 0.1013,
      "theoretical_loss": 4.320826459149725,
      "tokens_seen": 224919552
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009411811908200932,
      "loss": 0.1075,
      "theoretical_loss": 4.3202122673067125,
      "tokens_seen": 225181696
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009411009468785107,
      "loss": 0.1004,
      "theoretical_loss": 4.319598989992695,
      "tokens_seen": 225443840
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009410207029369283,
      "loss": 0.1046,
      "theoretical_loss": 4.318986624784908,
      "tokens_seen": 225705984
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009409404589953459,
      "loss": 0.103,
      "theoretical_loss": 4.318375169269813,
      "tokens_seen": 225968128
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009408602150537635,
      "loss": 0.1013,
      "theoretical_loss": 4.317764621043046,
      "tokens_seen": 226230272
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009407799711121811,
      "loss": 0.0995,
      "theoretical_loss": 4.317154977709375,
      "tokens_seen": 226492416
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009406997271705986,
      "loss": 0.0999,
      "theoretical_loss": 4.3165462368826555,
      "tokens_seen": 226754560
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009406194832290162,
      "loss": 0.0995,
      "theoretical_loss": 4.315938396185782,
      "tokens_seen": 227016704
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009405392392874338,
      "loss": 0.1,
      "theoretical_loss": 4.315331453250648,
      "tokens_seen": 227278848
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009404589953458514,
      "loss": 0.0992,
      "theoretical_loss": 4.314725405718099,
      "tokens_seen": 227540992
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000940378751404269,
      "loss": 0.0997,
      "theoretical_loss": 4.314120251237887,
      "tokens_seen": 227803136
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009402985074626867,
      "loss": 0.1027,
      "theoretical_loss": 4.31351598746863,
      "tokens_seen": 228065280
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009402182635211042,
      "loss": 0.1014,
      "theoretical_loss": 4.312912612077767,
      "tokens_seen": 228327424
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009401380195795217,
      "loss": 0.1017,
      "theoretical_loss": 4.312310122741512,
      "tokens_seen": 228589568
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009400577756379394,
      "loss": 0.1026,
      "theoretical_loss": 4.311708517144817,
      "tokens_seen": 228851712
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009399775316963569,
      "loss": 0.102,
      "theoretical_loss": 4.311107792981323,
      "tokens_seen": 229113856
    },
    {
      "epoch": 0.07,
      "objective/train/advantage_avg": 0.0002358820493100211,
      "objective/train/docs_used": 90061,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 2.085197687149048,
      "objective/train/original_loss": 2.085197925567627,
      "objective/train/theoretical_loss": 4.310507947953321,
      "objective/train/tokens_used": 249836000,
      "objective/train/value_avg": -0.0097808837890625,
      "objective/train/value_loss": 0.0005574537208303809,
      "objective/train/value_max": -0.0007238388061523438,
      "objective/train/value_min": -0.52783203125,
      "objective/train/value_reward_corr": 0.7022453186963163,
      "objective/train/value_std": 0.01198577880859375,
      "objective/train/weight_avg": 1.0004955530166626,
      "objective/train/weighted_lm_loss": 2.0866754055023193,
      "objective/train/weights_max": 1.2493910789489746,
      "objective/train/weights_min": 0.37043818831443787,
      "theoretical_loss": 4.310507947953321,
      "tokens_seen": 229376000
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009398972877547745,
      "loss": 0.103,
      "theoretical_loss": 4.310507947953321,
      "tokens_seen": 229376000
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009398170438131921,
      "loss": 0.1039,
      "theoretical_loss": 4.309908979771709,
      "tokens_seen": 229638144
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009397367998716097,
      "loss": 0.1039,
      "theoretical_loss": 4.3093108861559495,
      "tokens_seen": 229900288
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009396565559300272,
      "loss": 0.1018,
      "theoretical_loss": 4.308713664834029,
      "tokens_seen": 230162432
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009395763119884449,
      "loss": 0.0956,
      "theoretical_loss": 4.308117313542413,
      "tokens_seen": 230424576
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009394960680468625,
      "loss": 0.102,
      "theoretical_loss": 4.30752183002601,
      "tokens_seen": 230686720
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009394158241052801,
      "loss": 0.0988,
      "theoretical_loss": 4.3069272120381275,
      "tokens_seen": 230948864
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009393355801636977,
      "loss": 0.102,
      "theoretical_loss": 4.30633345734043,
      "tokens_seen": 231211008
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009392553362221152,
      "loss": 0.0999,
      "theoretical_loss": 4.3057405637029,
      "tokens_seen": 231473152
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009391750922805329,
      "loss": 0.1013,
      "theoretical_loss": 4.305148528903798,
      "tokens_seen": 231735296
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009390948483389504,
      "loss": 0.1018,
      "theoretical_loss": 4.304557350729623,
      "tokens_seen": 231997440
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009390146043973679,
      "loss": 0.0998,
      "theoretical_loss": 4.303967026975072,
      "tokens_seen": 232259584
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009389343604557857,
      "loss": 0.1022,
      "theoretical_loss": 4.303377555442998,
      "tokens_seen": 232521728
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009388541165142032,
      "loss": 0.0975,
      "theoretical_loss": 4.302788933944375,
      "tokens_seen": 232783872
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009387738725726208,
      "loss": 0.1006,
      "theoretical_loss": 4.302201160298255,
      "tokens_seen": 233046016
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009386936286310384,
      "loss": 0.101,
      "theoretical_loss": 4.301614232331733,
      "tokens_seen": 233308160
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000938613384689456,
      "loss": 0.1004,
      "theoretical_loss": 4.301028147879904,
      "tokens_seen": 233570304
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009385331407478735,
      "loss": 0.0979,
      "theoretical_loss": 4.300442904785831,
      "tokens_seen": 233832448
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009384528968062911,
      "loss": 0.1012,
      "theoretical_loss": 4.299858500900495,
      "tokens_seen": 234094592
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009383726528647087,
      "loss": 0.1021,
      "theoretical_loss": 4.2992749340827725,
      "tokens_seen": 234356736
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009382924089231262,
      "loss": 0.1026,
      "theoretical_loss": 4.298692202199386,
      "tokens_seen": 234618880
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000938212164981544,
      "loss": 0.1007,
      "theoretical_loss": 4.298110303124871,
      "tokens_seen": 234881024
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009381319210399615,
      "loss": 0.102,
      "theoretical_loss": 4.29752923474154,
      "tokens_seen": 235143168
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009380516770983792,
      "loss": 0.1003,
      "theoretical_loss": 4.29694899493944,
      "tokens_seen": 235405312
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009379714331567967,
      "loss": 0.1013,
      "theoretical_loss": 4.2963695816163225,
      "tokens_seen": 235667456
    },
    {
      "epoch": 0.07,
      "objective/train/advantage_avg": 0.001282443176023662,
      "objective/train/docs_used": 92471,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 2.023468494415283,
      "objective/train/original_loss": 2.023468494415283,
      "objective/train/theoretical_loss": 4.295790992677603,
      "objective/train/tokens_used": 256389600,
      "objective/train/value_avg": -0.00916290283203125,
      "objective/train/value_loss": 0.00042869814205914736,
      "objective/train/value_max": -0.0003845691680908203,
      "objective/train/value_min": -0.7470703125,
      "objective/train/value_reward_corr": 0.596518820655821,
      "objective/train/value_std": 0.0174102783203125,
      "objective/train/weight_avg": 1.0014816522598267,
      "objective/train/weighted_lm_loss": 2.02492618560791,
      "objective/train/weights_max": 1.7296583652496338,
      "objective/train/weights_min": 0.3692615330219269,
      "theoretical_loss": 4.295790992677603,
      "tokens_seen": 235929600
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009378911892152143,
      "loss": 0.0992,
      "theoretical_loss": 4.295790992677603,
      "tokens_seen": 235929600
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009378109452736319,
      "loss": 0.1008,
      "theoretical_loss": 4.2952132260363225,
      "tokens_seen": 236191744
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009377307013320494,
      "loss": 0.0992,
      "theoretical_loss": 4.294636279613117,
      "tokens_seen": 236453888
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000937650457390467,
      "loss": 0.1033,
      "theoretical_loss": 4.294060151336178,
      "tokens_seen": 236716032
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009375702134488846,
      "loss": 0.1017,
      "theoretical_loss": 4.293484839141217,
      "tokens_seen": 236978176
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009374899695073022,
      "loss": 0.0994,
      "theoretical_loss": 4.29291034097143,
      "tokens_seen": 237240320
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009374097255657198,
      "loss": 0.0964,
      "theoretical_loss": 4.2923366547774595,
      "tokens_seen": 237502464
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009373294816241375,
      "loss": 0.0991,
      "theoretical_loss": 4.2917637785173675,
      "tokens_seen": 237764608
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000937249237682555,
      "loss": 0.1,
      "theoretical_loss": 4.291191710156591,
      "tokens_seen": 238026752
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009371689937409725,
      "loss": 0.0996,
      "theoretical_loss": 4.290620447667912,
      "tokens_seen": 238288896
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009370887497993902,
      "loss": 0.0973,
      "theoretical_loss": 4.290049989031424,
      "tokens_seen": 238551040
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009370085058578077,
      "loss": 0.1026,
      "theoretical_loss": 4.289480332234493,
      "tokens_seen": 238813184
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009369282619162254,
      "loss": 0.1009,
      "theoretical_loss": 4.288911475271731,
      "tokens_seen": 239075328
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009368480179746429,
      "loss": 0.0999,
      "theoretical_loss": 4.288343416144952,
      "tokens_seen": 239337472
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009367677740330605,
      "loss": 0.1019,
      "theoretical_loss": 4.287776152863146,
      "tokens_seen": 239599616
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009366875300914782,
      "loss": 0.1016,
      "theoretical_loss": 4.287209683442444,
      "tokens_seen": 239861760
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009366072861498957,
      "loss": 0.1001,
      "theoretical_loss": 4.286644005906081,
      "tokens_seen": 240123904
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009365270422083133,
      "loss": 0.1011,
      "theoretical_loss": 4.286079118284368,
      "tokens_seen": 240386048
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009364467982667309,
      "loss": 0.1036,
      "theoretical_loss": 4.285515018614655,
      "tokens_seen": 240648192
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009363665543251485,
      "loss": 0.0961,
      "theoretical_loss": 4.2849517049412995,
      "tokens_seen": 240910336
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000936286310383566,
      "loss": 0.0975,
      "theoretical_loss": 4.284389175315636,
      "tokens_seen": 241172480
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009362060664419837,
      "loss": 0.0993,
      "theoretical_loss": 4.283827427795939,
      "tokens_seen": 241434624
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009361258225004012,
      "loss": 0.0984,
      "theoretical_loss": 4.283266460447394,
      "tokens_seen": 241696768
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009360455785588187,
      "loss": 0.0963,
      "theoretical_loss": 4.282706271342066,
      "tokens_seen": 241958912
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009359653346172365,
      "loss": 0.1022,
      "theoretical_loss": 4.282146858558866,
      "tokens_seen": 242221056
    },
    {
      "epoch": 0.07,
      "objective/train/advantage_avg": 0.0023586128372699022,
      "objective/train/docs_used": 94878,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.9628684520721436,
      "objective/train/original_loss": 1.9628684520721436,
      "objective/train/theoretical_loss": 4.281588220183519,
      "objective/train/tokens_used": 262943200,
      "objective/train/value_avg": -0.010406494140625,
      "objective/train/value_loss": 0.00042588188080117106,
      "objective/train/value_max": -0.0003249645233154297,
      "objective/train/value_min": -0.403564453125,
      "objective/train/value_reward_corr": 0.6426931088980384,
      "objective/train/value_std": 0.01442718505859375,
      "objective/train/weight_avg": 1.0025511980056763,
      "objective/train/weighted_lm_loss": 1.9680471420288086,
      "objective/train/weights_max": 1.2811800241470337,
      "objective/train/weights_min": 0.3709741532802582,
      "theoretical_loss": 4.281588220183519,
      "tokens_seen": 242483200
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000935885090675654,
      "loss": 0.0976,
      "theoretical_loss": 4.281588220183519,
      "tokens_seen": 242483200
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009358048467340716,
      "loss": 0.0992,
      "theoretical_loss": 4.281030354308533,
      "tokens_seen": 242745344
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009357246027924892,
      "loss": 0.0984,
      "theoretical_loss": 4.280473259033169,
      "tokens_seen": 243007488
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009356443588509068,
      "loss": 0.0976,
      "theoretical_loss": 4.27991693246341,
      "tokens_seen": 243269632
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009355641149093244,
      "loss": 0.0983,
      "theoretical_loss": 4.279361372711923,
      "tokens_seen": 243531776
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009354838709677419,
      "loss": 0.1021,
      "theoretical_loss": 4.278806577898042,
      "tokens_seen": 243793920
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009354036270261595,
      "loss": 0.0976,
      "theoretical_loss": 4.278252546147724,
      "tokens_seen": 244056064
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009353233830845771,
      "loss": 0.0992,
      "theoretical_loss": 4.277699275593523,
      "tokens_seen": 244318208
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009352431391429948,
      "loss": 0.0946,
      "theoretical_loss": 4.277146764374566,
      "tokens_seen": 244580352
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009351628952014123,
      "loss": 0.0996,
      "theoretical_loss": 4.276595010636514,
      "tokens_seen": 244842496
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00093508265125983,
      "loss": 0.0991,
      "theoretical_loss": 4.276044012531534,
      "tokens_seen": 245104640
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009350024073182475,
      "loss": 0.1007,
      "theoretical_loss": 4.275493768218274,
      "tokens_seen": 245366784
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009349221633766651,
      "loss": 0.0973,
      "theoretical_loss": 4.274944275861828,
      "tokens_seen": 245628928
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009348419194350827,
      "loss": 0.0975,
      "theoretical_loss": 4.274395533633712,
      "tokens_seen": 245891072
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009347616754935002,
      "loss": 0.0955,
      "theoretical_loss": 4.273847539711825,
      "tokens_seen": 246153216
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009346814315519178,
      "loss": 0.1006,
      "theoretical_loss": 4.273300292280435,
      "tokens_seen": 246415360
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009346011876103354,
      "loss": 0.0985,
      "theoretical_loss": 4.272753789530134,
      "tokens_seen": 246677504
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000934520943668753,
      "loss": 0.0955,
      "theoretical_loss": 4.272208029657822,
      "tokens_seen": 246939648
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009344406997271707,
      "loss": 0.0968,
      "theoretical_loss": 4.271663010866669,
      "tokens_seen": 247201792
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0009343604557855883,
      "loss": 0.0999,
      "theoretical_loss": 4.2711187313660925,
      "tokens_seen": 247463936
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009342802118440058,
      "loss": 0.0981,
      "theoretical_loss": 4.270575189371727,
      "tokens_seen": 247726080
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009341999679024234,
      "loss": 0.0988,
      "theoretical_loss": 4.270032383105398,
      "tokens_seen": 247988224
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000934119723960841,
      "loss": 0.097,
      "theoretical_loss": 4.269490310795089,
      "tokens_seen": 248250368
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009340394800192585,
      "loss": 0.1021,
      "theoretical_loss": 4.268948970674917,
      "tokens_seen": 248512512
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009339592360776762,
      "loss": 0.101,
      "theoretical_loss": 4.268408360985109,
      "tokens_seen": 248774656
    },
    {
      "epoch": 0.08,
      "objective/train/advantage_avg": 5.6569744629086927e-05,
      "objective/train/docs_used": 97104,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.9197639226913452,
      "objective/train/original_loss": 1.9197638034820557,
      "objective/train/theoretical_loss": 4.267868479971966,
      "objective/train/tokens_used": 269496800,
      "objective/train/value_avg": -0.00948333740234375,
      "objective/train/value_loss": 0.0004118235665373504,
      "objective/train/value_max": -0.0005464553833007812,
      "objective/train/value_min": -0.8017578125,
      "objective/train/value_reward_corr": 0.6645390257593566,
      "objective/train/value_std": 0.015777587890625,
      "objective/train/weight_avg": 1.0002477169036865,
      "objective/train/weighted_lm_loss": 1.9196513891220093,
      "objective/train/weights_max": 1.9581830501556396,
      "objective/train/weights_min": 0.38205578923225403,
      "theoretical_loss": 4.267868479971966,
      "tokens_seen": 249036800
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009338789921360937,
      "loss": 0.0993,
      "theoretical_loss": 4.267868479971966,
      "tokens_seen": 249036800
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009337987481945113,
      "loss": 0.0985,
      "theoretical_loss": 4.267329325887841,
      "tokens_seen": 249298944
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000933718504252929,
      "loss": 0.0921,
      "theoretical_loss": 4.266790896991109,
      "tokens_seen": 249561088
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009336382603113465,
      "loss": 0.1,
      "theoretical_loss": 4.266253191546146,
      "tokens_seen": 249823232
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009335580163697641,
      "loss": 0.0989,
      "theoretical_loss": 4.265716207823292,
      "tokens_seen": 250085376
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009334777724281817,
      "loss": 0.099,
      "theoretical_loss": 4.2651799440988345,
      "tokens_seen": 250347520
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009333975284865993,
      "loss": 0.0971,
      "theoretical_loss": 4.2646443986549745,
      "tokens_seen": 250609664
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009333172845450168,
      "loss": 0.0986,
      "theoretical_loss": 4.264109569779803,
      "tokens_seen": 250871808
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009332370406034345,
      "loss": 0.1009,
      "theoretical_loss": 4.263575455767277,
      "tokens_seen": 251133952
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000933156796661852,
      "loss": 0.0992,
      "theoretical_loss": 4.263042054917186,
      "tokens_seen": 251396096
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009330765527202696,
      "loss": 0.0989,
      "theoretical_loss": 4.262509365535134,
      "tokens_seen": 251658240
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009329963087786873,
      "loss": 0.0963,
      "theoretical_loss": 4.261977385932512,
      "tokens_seen": 251920384
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009329160648371048,
      "loss": 0.0979,
      "theoretical_loss": 4.261446114426466,
      "tokens_seen": 252182528
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009328358208955225,
      "loss": 0.097,
      "theoretical_loss": 4.260915549339879,
      "tokens_seen": 252444672
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00093275557695394,
      "loss": 0.0967,
      "theoretical_loss": 4.2603856890013425,
      "tokens_seen": 252706816
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009326753330123576,
      "loss": 0.0988,
      "theoretical_loss": 4.25985653174513,
      "tokens_seen": 252968960
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009325950890707752,
      "loss": 0.0945,
      "theoretical_loss": 4.259328075911173,
      "tokens_seen": 253231104
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009325148451291927,
      "loss": 0.0985,
      "theoretical_loss": 4.258800319845038,
      "tokens_seen": 253493248
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009324346011876103,
      "loss": 0.0971,
      "theoretical_loss": 4.258273261897896,
      "tokens_seen": 253755392
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009323543572460279,
      "loss": 0.0967,
      "theoretical_loss": 4.257746900426506,
      "tokens_seen": 254017536
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009322741133044456,
      "loss": 0.097,
      "theoretical_loss": 4.25722123379318,
      "tokens_seen": 254279680
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009321938693628631,
      "loss": 0.1006,
      "theoretical_loss": 4.256696260365768,
      "tokens_seen": 254541824
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009321136254212808,
      "loss": 0.0994,
      "theoretical_loss": 4.256171978517629,
      "tokens_seen": 254803968
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009320333814796983,
      "loss": 0.0948,
      "theoretical_loss": 4.255648386627607,
      "tokens_seen": 255066112
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000931953137538116,
      "loss": 0.0982,
      "theoretical_loss": 4.255125483080007,
      "tokens_seen": 255328256
    },
    {
      "epoch": 0.08,
      "objective/train/advantage_avg": 0.0025481332559138536,
      "objective/train/docs_used": 99403,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.921761393547058,
      "objective/train/original_loss": 1.9217615127563477,
      "objective/train/theoretical_loss": 4.254603266264572,
      "objective/train/tokens_used": 276050400,
      "objective/train/value_avg": -0.008270263671875,
      "objective/train/value_loss": 0.000349750422174111,
      "objective/train/value_max": -0.0004355907440185547,
      "objective/train/value_min": -0.50537109375,
      "objective/train/value_reward_corr": 0.5029395423614991,
      "objective/train/value_std": 0.01042938232421875,
      "objective/train/weight_avg": 1.0027012825012207,
      "objective/train/weighted_lm_loss": 1.9272537231445312,
      "objective/train/weights_max": 1.5571717023849487,
      "objective/train/weights_min": 0.36996087431907654,
      "theoretical_loss": 4.254603266264572,
      "tokens_seen": 255590400
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009318728935965335,
      "loss": 0.0981,
      "theoretical_loss": 4.254603266264572,
      "tokens_seen": 255590400
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000931792649654951,
      "loss": 0.0957,
      "theoretical_loss": 4.254081734576458,
      "tokens_seen": 255852544
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009317124057133687,
      "loss": 0.0963,
      "theoretical_loss": 4.253560886416212,
      "tokens_seen": 256114688
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009316321617717862,
      "loss": 0.0968,
      "theoretical_loss": 4.253040720189746,
      "tokens_seen": 256376832
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009315519178302038,
      "loss": 0.0969,
      "theoretical_loss": 4.252521234308315,
      "tokens_seen": 256638976
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009314716738886215,
      "loss": 0.0961,
      "theoretical_loss": 4.2520024271884935,
      "tokens_seen": 256901120
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009313914299470391,
      "loss": 0.0978,
      "theoretical_loss": 4.251484297252151,
      "tokens_seen": 257163264
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009313111860054566,
      "loss": 0.1015,
      "theoretical_loss": 4.250966842926434,
      "tokens_seen": 257425408
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009312309420638742,
      "loss": 0.0991,
      "theoretical_loss": 4.250450062643734,
      "tokens_seen": 257687552
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009311506981222918,
      "loss": 0.095,
      "theoretical_loss": 4.249933954841672,
      "tokens_seen": 257949696
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009310704541807093,
      "loss": 0.1023,
      "theoretical_loss": 4.2494185179630755,
      "tokens_seen": 258211840
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000930990210239127,
      "loss": 0.0993,
      "theoretical_loss": 4.24890375045595,
      "tokens_seen": 258473984
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009309099662975445,
      "loss": 0.0972,
      "theoretical_loss": 4.248389650773463,
      "tokens_seen": 258736128
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009308297223559621,
      "loss": 0.0985,
      "theoretical_loss": 4.24787621737392,
      "tokens_seen": 258998272
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009307494784143798,
      "loss": 0.0969,
      "theoretical_loss": 4.247363448720739,
      "tokens_seen": 259260416
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009306692344727973,
      "loss": 0.0985,
      "theoretical_loss": 4.246851343282432,
      "tokens_seen": 259522560
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000930588990531215,
      "loss": 0.1008,
      "theoretical_loss": 4.246339899532582,
      "tokens_seen": 259784704
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009305087465896325,
      "loss": 0.0969,
      "theoretical_loss": 4.245829115949818,
      "tokens_seen": 260046848
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009304285026480501,
      "loss": 0.1007,
      "theoretical_loss": 4.245318991017802,
      "tokens_seen": 260308992
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009303482587064677,
      "loss": 0.098,
      "theoretical_loss": 4.244809523225195,
      "tokens_seen": 260571136
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009302680147648853,
      "loss": 0.0997,
      "theoretical_loss": 4.244300711065646,
      "tokens_seen": 260833280
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009301877708233028,
      "loss": 0.0933,
      "theoretical_loss": 4.243792553037767,
      "tokens_seen": 261095424
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009301075268817204,
      "loss": 0.0978,
      "theoretical_loss": 4.243285047645106,
      "tokens_seen": 261357568
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000930027282940138,
      "loss": 0.1023,
      "theoretical_loss": 4.242778193396136,
      "tokens_seen": 261619712
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009299470389985556,
      "loss": 0.0989,
      "theoretical_loss": 4.242271988804228,
      "tokens_seen": 261881856
    },
    {
      "epoch": 0.08,
      "objective/train/advantage_avg": 0.004064733162522316,
      "objective/train/docs_used": 101934,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 2.00811505317688,
      "objective/train/original_loss": 2.008115291595459,
      "objective/train/theoretical_loss": 4.241766432387629,
      "objective/train/tokens_used": 282604000,
      "objective/train/value_avg": -0.00791168212890625,
      "objective/train/value_loss": 0.00014644188922829926,
      "objective/train/value_max": -0.0006613731384277344,
      "objective/train/value_min": -0.193603515625,
      "objective/train/value_reward_corr": 0.33414470602713064,
      "objective/train/value_std": 0.006023406982421875,
      "objective/train/weight_avg": 1.0041327476501465,
      "objective/train/weighted_lm_loss": 2.0166854858398438,
      "objective/train/weights_max": 1.1917407512664795,
      "objective/train/weights_min": 0.37122613191604614,
      "theoretical_loss": 4.241766432387629,
      "tokens_seen": 262144000
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009298667950569733,
      "loss": 0.0953,
      "theoretical_loss": 4.241766432387629,
      "tokens_seen": 262144000
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009297865511153908,
      "loss": 0.0968,
      "theoretical_loss": 4.241261522669445,
      "tokens_seen": 262406144
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009297063071738084,
      "loss": 0.0992,
      "theoretical_loss": 4.240757258177617,
      "tokens_seen": 262668288
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000929626063232226,
      "loss": 0.0957,
      "theoretical_loss": 4.240253637444903,
      "tokens_seen": 262930432
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009295458192906435,
      "loss": 0.0951,
      "theoretical_loss": 4.239750659008854,
      "tokens_seen": 263192576
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009294655753490611,
      "loss": 0.0968,
      "theoretical_loss": 4.2392483214117975,
      "tokens_seen": 263454720
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009293853314074787,
      "loss": 0.0992,
      "theoretical_loss": 4.238746623200815,
      "tokens_seen": 263716864
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009293050874658963,
      "loss": 0.0982,
      "theoretical_loss": 4.238245562927722,
      "tokens_seen": 263979008
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000929224843524314,
      "loss": 0.0987,
      "theoretical_loss": 4.237745139149047,
      "tokens_seen": 264241152
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009291445995827316,
      "loss": 0.0959,
      "theoretical_loss": 4.237245350426015,
      "tokens_seen": 264503296
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009290643556411491,
      "loss": 0.0935,
      "theoretical_loss": 4.236746195324523,
      "tokens_seen": 264765440
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009289841116995667,
      "loss": 0.1002,
      "theoretical_loss": 4.2362476724151215,
      "tokens_seen": 265027584
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009289038677579843,
      "loss": 0.0973,
      "theoretical_loss": 4.235749780272998,
      "tokens_seen": 265289728
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009288236238164018,
      "loss": 0.1,
      "theoretical_loss": 4.235252517477956,
      "tokens_seen": 265551872
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009287433798748195,
      "loss": 0.0943,
      "theoretical_loss": 4.23475588261439,
      "tokens_seen": 265814016
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000928663135933237,
      "loss": 0.0966,
      "theoretical_loss": 4.234259874271275,
      "tokens_seen": 266076160
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009285828919916546,
      "loss": 0.0969,
      "theoretical_loss": 4.23376449104214,
      "tokens_seen": 266338304
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009285026480500723,
      "loss": 0.1002,
      "theoretical_loss": 4.233269731525055,
      "tokens_seen": 266600448
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009284224041084899,
      "loss": 0.0978,
      "theoretical_loss": 4.232775594322605,
      "tokens_seen": 266862592
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009283421601669074,
      "loss": 0.0976,
      "theoretical_loss": 4.232282078041876,
      "tokens_seen": 267124736
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000928261916225325,
      "loss": 0.0978,
      "theoretical_loss": 4.231789181294436,
      "tokens_seen": 267386880
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009281816722837426,
      "loss": 0.1,
      "theoretical_loss": 4.231296902696314,
      "tokens_seen": 267649024
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009281014283421601,
      "loss": 0.0956,
      "theoretical_loss": 4.230805240867982,
      "tokens_seen": 267911168
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009280211844005778,
      "loss": 0.097,
      "theoretical_loss": 4.230314194434336,
      "tokens_seen": 268173312
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009279409404589953,
      "loss": 0.0974,
      "theoretical_loss": 4.229823762024681,
      "tokens_seen": 268435456
    },
    {
      "epoch": 0.08,
      "objective/train/advantage_avg": 0.0018670569406822324,
      "objective/train/docs_used": 104173,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.8005046844482422,
      "objective/train/original_loss": 1.8005049228668213,
      "objective/train/theoretical_loss": 4.2293339422727065,
      "objective/train/tokens_used": 289157600,
      "objective/train/value_avg": -0.00882720947265625,
      "objective/train/value_loss": 0.0002679454628378153,
      "objective/train/value_max": -0.00035691261291503906,
      "objective/train/value_min": -0.283203125,
      "objective/train/value_reward_corr": 0.5715762562194157,
      "objective/train/value_std": 0.01021575927734375,
      "objective/train/weight_avg": 1.0019898414611816,
      "objective/train/weighted_lm_loss": 1.803560495376587,
      "objective/train/weights_max": 1.1879379749298096,
      "objective/train/weights_min": 0.3708355128765106,
      "theoretical_loss": 4.2293339422727065,
      "tokens_seen": 268697600
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000927860696517413,
      "loss": 0.0972,
      "theoretical_loss": 4.2293339422727065,
      "tokens_seen": 268697600
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009277804525758306,
      "loss": 0.0986,
      "theoretical_loss": 4.228844733816474,
      "tokens_seen": 268959744
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009277002086342481,
      "loss": 0.0948,
      "theoretical_loss": 4.228356135298394,
      "tokens_seen": 269221888
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009276199646926658,
      "loss": 0.0938,
      "theoretical_loss": 4.227868145365211,
      "tokens_seen": 269484032
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009275397207510833,
      "loss": 0.0966,
      "theoretical_loss": 4.227380762667987,
      "tokens_seen": 269746176
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009274594768095009,
      "loss": 0.0951,
      "theoretical_loss": 4.226893985862076,
      "tokens_seen": 270008320
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009273792328679185,
      "loss": 0.0985,
      "theoretical_loss": 4.226407813607116,
      "tokens_seen": 270270464
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009272989889263361,
      "loss": 0.0967,
      "theoretical_loss": 4.2259222445670055,
      "tokens_seen": 270532608
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009272187449847536,
      "loss": 0.0935,
      "theoretical_loss": 4.225437277409885,
      "tokens_seen": 270794752
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009271385010431712,
      "loss": 0.0953,
      "theoretical_loss": 4.224952910808122,
      "tokens_seen": 271056896
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009270582571015889,
      "loss": 0.0973,
      "theoretical_loss": 4.224469143438294,
      "tokens_seen": 271319040
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009269780131600064,
      "loss": 0.0998,
      "theoretical_loss": 4.223985973981171,
      "tokens_seen": 271581184
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009268977692184241,
      "loss": 0.0996,
      "theoretical_loss": 4.223503401121693,
      "tokens_seen": 271843328
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009268175252768416,
      "loss": 0.0941,
      "theoretical_loss": 4.223021423548962,
      "tokens_seen": 272105472
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009267372813352593,
      "loss": 0.0962,
      "theoretical_loss": 4.222540039956215,
      "tokens_seen": 272367616
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009266570373936768,
      "loss": 0.0953,
      "theoretical_loss": 4.222059249040814,
      "tokens_seen": 272629760
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009265767934520943,
      "loss": 0.0989,
      "theoretical_loss": 4.2215790495042285,
      "tokens_seen": 272891904
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000926496549510512,
      "loss": 0.0982,
      "theoretical_loss": 4.221099440052014,
      "tokens_seen": 273154048
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009264163055689295,
      "loss": 0.0956,
      "theoretical_loss": 4.220620419393799,
      "tokens_seen": 273416192
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009263360616273471,
      "loss": 0.0979,
      "theoretical_loss": 4.220141986243268,
      "tokens_seen": 273678336
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009262558176857648,
      "loss": 0.098,
      "theoretical_loss": 4.219664139318145,
      "tokens_seen": 273940480
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009261755737441824,
      "loss": 0.0972,
      "theoretical_loss": 4.219186877340174,
      "tokens_seen": 274202624
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009260953298025999,
      "loss": 0.0943,
      "theoretical_loss": 4.218710199035108,
      "tokens_seen": 274464768
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009260150858610175,
      "loss": 0.0941,
      "theoretical_loss": 4.218234103132686,
      "tokens_seen": 274726912
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009259348419194351,
      "loss": 0.097,
      "theoretical_loss": 4.217758588366623,
      "tokens_seen": 274989056
    },
    {
      "epoch": 0.08,
      "objective/train/advantage_avg": 0.0027189042884856462,
      "objective/train/docs_used": 106703,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.8442187309265137,
      "objective/train/original_loss": 1.8442187309265137,
      "objective/train/theoretical_loss": 4.217283653474588,
      "objective/train/tokens_used": 295711200,
      "objective/train/value_avg": -0.011016845703125,
      "objective/train/value_loss": 0.0005074041546322405,
      "objective/train/value_max": -0.0006022453308105469,
      "objective/train/value_min": -0.90625,
      "objective/train/value_reward_corr": 0.6262922111240474,
      "objective/train/value_std": 0.0198211669921875,
      "objective/train/weight_avg": 1.0029503107070923,
      "objective/train/weighted_lm_loss": 1.849584937095642,
      "objective/train/weights_max": 1.8512089252471924,
      "objective/train/weights_min": 0.36976757645606995,
      "theoretical_loss": 4.217283653474588,
      "tokens_seen": 275251200
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009258545979778526,
      "loss": 0.0936,
      "theoretical_loss": 4.217283653474588,
      "tokens_seen": 275251200
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009257743540362703,
      "loss": 0.0962,
      "theoretical_loss": 4.216809297198195,
      "tokens_seen": 275513344
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009256941100946878,
      "loss": 0.0968,
      "theoretical_loss": 4.21633551828298,
      "tokens_seen": 275775488
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009256138661531054,
      "loss": 0.0981,
      "theoretical_loss": 4.215862315478388,
      "tokens_seen": 276037632
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009255336222115231,
      "loss": 0.0992,
      "theoretical_loss": 4.2153896875377574,
      "tokens_seen": 276299776
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009254533782699407,
      "loss": 0.0979,
      "theoretical_loss": 4.214917633218304,
      "tokens_seen": 276561920
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009253731343283583,
      "loss": 0.098,
      "theoretical_loss": 4.214446151281106,
      "tokens_seen": 276824064
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009252928903867758,
      "loss": 0.0945,
      "theoretical_loss": 4.213975240491084,
      "tokens_seen": 277086208
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009252126464451934,
      "loss": 0.0964,
      "theoretical_loss": 4.213504899616995,
      "tokens_seen": 277348352
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000925132402503611,
      "loss": 0.0969,
      "theoretical_loss": 4.213035127431402,
      "tokens_seen": 277610496
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009250521585620286,
      "loss": 0.0967,
      "theoretical_loss": 4.212565922710677,
      "tokens_seen": 277872640
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009249719146204461,
      "loss": 0.0964,
      "theoretical_loss": 4.21209728423497,
      "tokens_seen": 278134784
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009248916706788639,
      "loss": 0.0984,
      "theoretical_loss": 4.2116292107882,
      "tokens_seen": 278396928
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009248114267372814,
      "loss": 0.0973,
      "theoretical_loss": 4.211161701158042,
      "tokens_seen": 278659072
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009247311827956989,
      "loss": 0.0985,
      "theoretical_loss": 4.2106947541359085,
      "tokens_seen": 278921216
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009246509388541166,
      "loss": 0.0962,
      "theoretical_loss": 4.210228368516935,
      "tokens_seen": 279183360
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009245706949125341,
      "loss": 0.1018,
      "theoretical_loss": 4.209762543099966,
      "tokens_seen": 279445504
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009244904509709517,
      "loss": 0.0971,
      "theoretical_loss": 4.209297276687541,
      "tokens_seen": 279707648
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009244102070293693,
      "loss": 0.1,
      "theoretical_loss": 4.2088325680858745,
      "tokens_seen": 279969792
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0009243299630877869,
      "loss": 0.0954,
      "theoretical_loss": 4.208368416104849,
      "tokens_seen": 280231936
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009242497191462045,
      "loss": 0.0986,
      "theoretical_loss": 4.207904819557995,
      "tokens_seen": 280494080
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000924169475204622,
      "loss": 0.0965,
      "theoretical_loss": 4.207441777262477,
      "tokens_seen": 280756224
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009240892312630397,
      "loss": 0.0958,
      "theoretical_loss": 4.206979288039081,
      "tokens_seen": 281018368
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009240089873214573,
      "loss": 0.0943,
      "theoretical_loss": 4.206517350712199,
      "tokens_seen": 281280512
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009239287433798749,
      "loss": 0.0964,
      "theoretical_loss": 4.206055964109813,
      "tokens_seen": 281542656
    },
    {
      "epoch": 0.09,
      "objective/train/advantage_avg": 0.0017802802613005042,
      "objective/train/docs_used": 109222,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.907019853591919,
      "objective/train/original_loss": 1.907020092010498,
      "objective/train/theoretical_loss": 4.205595127063485,
      "objective/train/tokens_used": 302264800,
      "objective/train/value_avg": -0.007091522216796875,
      "objective/train/value_loss": 0.00020236380805727094,
      "objective/train/value_max": -0.00042891502380371094,
      "objective/train/value_min": -0.2548828125,
      "objective/train/value_reward_corr": 0.46679314179268283,
      "objective/train/value_std": 0.0074310302734375,
      "objective/train/weight_avg": 1.0018771886825562,
      "objective/train/weighted_lm_loss": 1.9096391201019287,
      "objective/train/weights_max": 1.1684050559997559,
      "objective/train/weights_min": 0.38661831617355347,
      "theoretical_loss": 4.205595127063485,
      "tokens_seen": 281804800
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009238484994382924,
      "loss": 0.0951,
      "theoretical_loss": 4.205595127063485,
      "tokens_seen": 281804800
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009237682554967101,
      "loss": 0.0938,
      "theoretical_loss": 4.205134838408337,
      "tokens_seen": 282066944
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009236880115551276,
      "loss": 0.0959,
      "theoretical_loss": 4.20467509698304,
      "tokens_seen": 282329088
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009236077676135451,
      "loss": 0.0954,
      "theoretical_loss": 4.204215901629803,
      "tokens_seen": 282591232
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009235275236719628,
      "loss": 0.0962,
      "theoretical_loss": 4.203757251194353,
      "tokens_seen": 282853376
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009234472797303803,
      "loss": 0.0952,
      "theoretical_loss": 4.203299144525923,
      "tokens_seen": 283115520
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000923367035788798,
      "loss": 0.096,
      "theoretical_loss": 4.202841580477241,
      "tokens_seen": 283377664
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009232867918472156,
      "loss": 0.0989,
      "theoretical_loss": 4.202384557904513,
      "tokens_seen": 283639808
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009232065479056332,
      "loss": 0.0955,
      "theoretical_loss": 4.201928075667411,
      "tokens_seen": 283901952
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009231263039640507,
      "loss": 0.0949,
      "theoretical_loss": 4.201472132629057,
      "tokens_seen": 284164096
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009230460600224683,
      "loss": 0.1007,
      "theoretical_loss": 4.201016727656012,
      "tokens_seen": 284426240
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009229658160808859,
      "loss": 0.0983,
      "theoretical_loss": 4.2005618596182615,
      "tokens_seen": 284688384
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009228855721393035,
      "loss": 0.0985,
      "theoretical_loss": 4.200107527389202,
      "tokens_seen": 284950528
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009228053281977211,
      "loss": 0.0965,
      "theoretical_loss": 4.199653729845626,
      "tokens_seen": 285212672
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009227250842561386,
      "loss": 0.0965,
      "theoretical_loss": 4.199200465867714,
      "tokens_seen": 285474816
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009226448403145564,
      "loss": 0.0955,
      "theoretical_loss": 4.198747734339013,
      "tokens_seen": 285736960
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009225645963729739,
      "loss": 0.0947,
      "theoretical_loss": 4.198295534146429,
      "tokens_seen": 285999104
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009224843524313914,
      "loss": 0.095,
      "theoretical_loss": 4.197843864180214,
      "tokens_seen": 286261248
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009224041084898091,
      "loss": 0.0949,
      "theoretical_loss": 4.197392723333951,
      "tokens_seen": 286523392
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009223238645482266,
      "loss": 0.0975,
      "theoretical_loss": 4.196942110504538,
      "tokens_seen": 286785536
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009222436206066442,
      "loss": 0.0954,
      "theoretical_loss": 4.196492024592183,
      "tokens_seen": 287047680
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009221633766650618,
      "loss": 0.095,
      "theoretical_loss": 4.196042464500382,
      "tokens_seen": 287309824
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009220831327234794,
      "loss": 0.0947,
      "theoretical_loss": 4.195593429135916,
      "tokens_seen": 287571968
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009220028887818969,
      "loss": 0.1014,
      "theoretical_loss": 4.195144917408828,
      "tokens_seen": 287834112
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009219226448403146,
      "loss": 0.0949,
      "theoretical_loss": 4.194696928232417,
      "tokens_seen": 288096256
    },
    {
      "epoch": 0.09,
      "objective/train/advantage_avg": 0.002935412572696805,
      "objective/train/docs_used": 111591,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.8163905143737793,
      "objective/train/original_loss": 1.8163906335830688,
      "objective/train/theoretical_loss": 4.194249460523222,
      "objective/train/tokens_used": 308818400,
      "objective/train/value_avg": -0.00843048095703125,
      "objective/train/value_loss": 0.0001841627963585779,
      "objective/train/value_max": -0.000507354736328125,
      "objective/train/value_min": -0.318603515625,
      "objective/train/value_reward_corr": 0.49277345208210327,
      "objective/train/value_std": 0.0080413818359375,
      "objective/train/weight_avg": 1.0030207633972168,
      "objective/train/weighted_lm_loss": 1.8225624561309814,
      "objective/train/weights_max": 1.174675464630127,
      "objective/train/weights_min": 0.3693629503250122,
      "theoretical_loss": 4.194249460523222,
      "tokens_seen": 288358400
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009218424008987322,
      "loss": 0.0934,
      "theoretical_loss": 4.194249460523222,
      "tokens_seen": 288358400
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009217621569571497,
      "loss": 0.0965,
      "theoretical_loss": 4.193802513201015,
      "tokens_seen": 288620544
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009216819130155674,
      "loss": 0.0999,
      "theoretical_loss": 4.193356085188778,
      "tokens_seen": 288882688
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009216016690739849,
      "loss": 0.0984,
      "theoretical_loss": 4.1929101754127025,
      "tokens_seen": 289144832
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009215214251324026,
      "loss": 0.0957,
      "theoretical_loss": 4.192464782802167,
      "tokens_seen": 289406976
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009214411811908201,
      "loss": 0.0957,
      "theoretical_loss": 4.192019906289733,
      "tokens_seen": 289669120
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009213609372492377,
      "loss": 0.0985,
      "theoretical_loss": 4.1915755448111245,
      "tokens_seen": 289931264
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009212806933076553,
      "loss": 0.0972,
      "theoretical_loss": 4.191131697305222,
      "tokens_seen": 290193408
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009212004493660728,
      "loss": 0.0962,
      "theoretical_loss": 4.1906883627140505,
      "tokens_seen": 290455552
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009211202054244905,
      "loss": 0.0931,
      "theoretical_loss": 4.19024553998276,
      "tokens_seen": 290717696
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009210399614829081,
      "loss": 0.0957,
      "theoretical_loss": 4.189803228059623,
      "tokens_seen": 290979840
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009209597175413257,
      "loss": 0.096,
      "theoretical_loss": 4.189361425896016,
      "tokens_seen": 291241984
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009208794735997432,
      "loss": 0.0929,
      "theoretical_loss": 4.188920132446411,
      "tokens_seen": 291504128
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009207992296581609,
      "loss": 0.0937,
      "theoretical_loss": 4.188479346668359,
      "tokens_seen": 291766272
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009207189857165784,
      "loss": 0.0941,
      "theoretical_loss": 4.188039067522484,
      "tokens_seen": 292028416
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009206387417749959,
      "loss": 0.0953,
      "theoretical_loss": 4.18759929397247,
      "tokens_seen": 292290560
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009205584978334136,
      "loss": 0.0958,
      "theoretical_loss": 4.187160024985044,
      "tokens_seen": 292552704
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009204782538918311,
      "loss": 0.0973,
      "theoretical_loss": 4.1867212595299685,
      "tokens_seen": 292814848
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009203980099502489,
      "loss": 0.0958,
      "theoretical_loss": 4.186282996580034,
      "tokens_seen": 293076992
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009203177660086664,
      "loss": 0.0983,
      "theoretical_loss": 4.185845235111037,
      "tokens_seen": 293339136
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000920237522067084,
      "loss": 0.0962,
      "theoretical_loss": 4.185407974101779,
      "tokens_seen": 293601280
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009201572781255016,
      "loss": 0.0948,
      "theoretical_loss": 4.184971212534048,
      "tokens_seen": 293863424
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009200770341839191,
      "loss": 0.0959,
      "theoretical_loss": 4.184534949392611,
      "tokens_seen": 294125568
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009199967902423367,
      "loss": 0.095,
      "theoretical_loss": 4.184099183665199,
      "tokens_seen": 294387712
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009199165463007543,
      "loss": 0.0982,
      "theoretical_loss": 4.1836639143425,
      "tokens_seen": 294649856
    },
    {
      "epoch": 0.09,
      "objective/train/advantage_avg": 0.001675541396252811,
      "objective/train/docs_used": 114046,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.922358751296997,
      "objective/train/original_loss": 1.922358751296997,
      "objective/train/theoretical_loss": 4.1832291404181445,
      "objective/train/tokens_used": 315372000,
      "objective/train/value_avg": -0.00826263427734375,
      "objective/train/value_loss": 0.0002560181892476976,
      "objective/train/value_max": -0.00048041343688964844,
      "objective/train/value_min": -0.51025390625,
      "objective/train/value_reward_corr": 0.5477898481987934,
      "objective/train/value_std": 0.008514404296875,
      "objective/train/weight_avg": 1.001792550086975,
      "objective/train/weighted_lm_loss": 1.9252777099609375,
      "objective/train/weights_max": 1.2260377407073975,
      "objective/train/weights_min": 0.37242892384529114,
      "theoretical_loss": 4.1832291404181445,
      "tokens_seen": 294912000
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009198363023591719,
      "loss": 0.0955,
      "theoretical_loss": 4.1832291404181445,
      "tokens_seen": 294912000
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009197560584175894,
      "loss": 0.0958,
      "theoretical_loss": 4.182794860888696,
      "tokens_seen": 295174144
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009196758144760072,
      "loss": 0.0975,
      "theoretical_loss": 4.18236107475364,
      "tokens_seen": 295436288
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009195955705344247,
      "loss": 0.0973,
      "theoretical_loss": 4.18192778101537,
      "tokens_seen": 295698432
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009195153265928422,
      "loss": 0.0942,
      "theoretical_loss": 4.181494978679181,
      "tokens_seen": 295960576
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009194350826512599,
      "loss": 0.0951,
      "theoretical_loss": 4.181062666753256,
      "tokens_seen": 296222720
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009193548387096774,
      "loss": 0.0946,
      "theoretical_loss": 4.180630844248653,
      "tokens_seen": 296484864
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000919274594768095,
      "loss": 0.0976,
      "theoretical_loss": 4.180199510179299,
      "tokens_seen": 296747008
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009191943508265126,
      "loss": 0.0947,
      "theoretical_loss": 4.179768663561975,
      "tokens_seen": 297009152
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009191141068849302,
      "loss": 0.0924,
      "theoretical_loss": 4.1793383034163085,
      "tokens_seen": 297271296
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009190338629433478,
      "loss": 0.0939,
      "theoretical_loss": 4.178908428764759,
      "tokens_seen": 297533440
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009189536190017654,
      "loss": 0.0945,
      "theoretical_loss": 4.17847903863261,
      "tokens_seen": 297795584
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000918873375060183,
      "loss": 0.095,
      "theoretical_loss": 4.178050132047958,
      "tokens_seen": 298057728
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009187931311186006,
      "loss": 0.0937,
      "theoretical_loss": 4.177621708041703,
      "tokens_seen": 298319872
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009187128871770182,
      "loss": 0.0964,
      "theoretical_loss": 4.177193765647534,
      "tokens_seen": 298582016
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009186326432354357,
      "loss": 0.0973,
      "theoretical_loss": 4.176766303901922,
      "tokens_seen": 298844160
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009185523992938534,
      "loss": 0.0954,
      "theoretical_loss": 4.17633932184411,
      "tokens_seen": 299106304
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009184721553522709,
      "loss": 0.0951,
      "theoretical_loss": 4.1759128185161005,
      "tokens_seen": 299368448
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009183919114106885,
      "loss": 0.0917,
      "theoretical_loss": 4.175486792962646,
      "tokens_seen": 299630592
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009183116674691061,
      "loss": 0.0952,
      "theoretical_loss": 4.175061244231237,
      "tokens_seen": 299892736
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009182314235275236,
      "loss": 0.097,
      "theoretical_loss": 4.174636171372097,
      "tokens_seen": 300154880
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009181511795859412,
      "loss": 0.0914,
      "theoretical_loss": 4.174211573438166,
      "tokens_seen": 300417024
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009180709356443589,
      "loss": 0.0931,
      "theoretical_loss": 4.173787449485094,
      "tokens_seen": 300679168
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009179906917027765,
      "loss": 0.0979,
      "theoretical_loss": 4.17336379857123,
      "tokens_seen": 300941312
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009179104477611941,
      "loss": 0.0941,
      "theoretical_loss": 4.172940619757611,
      "tokens_seen": 301203456
    },
    {
      "epoch": 0.09,
      "objective/train/advantage_avg": 0.002791388425976038,
      "objective/train/docs_used": 116418,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.9814350605010986,
      "objective/train/original_loss": 1.9814350605010986,
      "objective/train/theoretical_loss": 4.172517912107954,
      "objective/train/tokens_used": 321925600,
      "objective/train/value_avg": -0.01168060302734375,
      "objective/train/value_loss": 0.0004382512124720961,
      "objective/train/value_max": -0.0003905296325683594,
      "objective/train/value_min": -0.64453125,
      "objective/train/value_reward_corr": 0.6855180405018573,
      "objective/train/value_std": 0.0270538330078125,
      "objective/train/weight_avg": 1.0029999017715454,
      "objective/train/weighted_lm_loss": 1.9875683784484863,
      "objective/train/weights_max": 1.4244129657745361,
      "objective/train/weights_min": 0.3713734447956085,
      "theoretical_loss": 4.172517912107954,
      "tokens_seen": 301465600
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009178302038196117,
      "loss": 0.0932,
      "theoretical_loss": 4.172517912107954,
      "tokens_seen": 301465600
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009177499598780292,
      "loss": 0.0923,
      "theoretical_loss": 4.172095674688645,
      "tokens_seen": 301727744
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009176697159364468,
      "loss": 0.0951,
      "theoretical_loss": 4.171673906568729,
      "tokens_seen": 301989888
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009175894719948644,
      "loss": 0.0967,
      "theoretical_loss": 4.171252606819899,
      "tokens_seen": 302252032
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009175092280532819,
      "loss": 0.0948,
      "theoretical_loss": 4.170831774516489,
      "tokens_seen": 302514176
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009174289841116997,
      "loss": 0.0958,
      "theoretical_loss": 4.170411408735461,
      "tokens_seen": 302776320
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009173487401701172,
      "loss": 0.0958,
      "theoretical_loss": 4.169991508556398,
      "tokens_seen": 303038464
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009172684962285348,
      "loss": 0.0944,
      "theoretical_loss": 4.169572073061493,
      "tokens_seen": 303300608
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009171882522869524,
      "loss": 0.0977,
      "theoretical_loss": 4.16915310133554,
      "tokens_seen": 303562752
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009171080083453699,
      "loss": 0.0962,
      "theoretical_loss": 4.1687345924659205,
      "tokens_seen": 303824896
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009170277644037875,
      "loss": 0.0937,
      "theoretical_loss": 4.168316545542602,
      "tokens_seen": 304087040
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009169475204622051,
      "loss": 0.0935,
      "theoretical_loss": 4.167898959658121,
      "tokens_seen": 304349184
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009168672765206227,
      "loss": 0.0903,
      "theoretical_loss": 4.167481833907576,
      "tokens_seen": 304611328
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009167870325790402,
      "loss": 0.098,
      "theoretical_loss": 4.16706516738862,
      "tokens_seen": 304873472
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000916706788637458,
      "loss": 0.0937,
      "theoretical_loss": 4.166648959201449,
      "tokens_seen": 305135616
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009166265446958755,
      "loss": 0.097,
      "theoretical_loss": 4.166233208448794,
      "tokens_seen": 305397760
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009165463007542931,
      "loss": 0.0962,
      "theoretical_loss": 4.165817914235908,
      "tokens_seen": 305659904
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009164660568127107,
      "loss": 0.0944,
      "theoretical_loss": 4.165403075670562,
      "tokens_seen": 305922048
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009163858128711282,
      "loss": 0.0938,
      "theoretical_loss": 4.164988691863032,
      "tokens_seen": 306184192
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009163055689295459,
      "loss": 0.0937,
      "theoretical_loss": 4.164574761926092,
      "tokens_seen": 306446336
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009162253249879634,
      "loss": 0.0943,
      "theoretical_loss": 4.164161284975005,
      "tokens_seen": 306708480
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000916145081046381,
      "loss": 0.0938,
      "theoretical_loss": 4.1637482601275115,
      "tokens_seen": 306970624
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009160648371047986,
      "loss": 0.0979,
      "theoretical_loss": 4.163335686503822,
      "tokens_seen": 307232768
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009159845931632162,
      "loss": 0.0956,
      "theoretical_loss": 4.162923563226607,
      "tokens_seen": 307494912
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009159043492216338,
      "loss": 0.093,
      "theoretical_loss": 4.1625118894209905,
      "tokens_seen": 307757056
    },
    {
      "epoch": 0.09,
      "objective/train/advantage_avg": 0.0017242436297237873,
      "objective/train/docs_used": 118729,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.8616645336151123,
      "objective/train/original_loss": 1.8616645336151123,
      "objective/train/theoretical_loss": 4.16210066421454,
      "objective/train/tokens_used": 328479200,
      "objective/train/value_avg": -0.00921630859375,
      "objective/train/value_loss": 0.0002815025800373405,
      "objective/train/value_max": -0.0003936290740966797,
      "objective/train/value_min": -0.79638671875,
      "objective/train/value_reward_corr": 0.6599847619871043,
      "objective/train/value_std": 0.0160675048828125,
      "objective/train/weight_avg": 1.0018616914749146,
      "objective/train/weighted_lm_loss": 1.86510169506073,
      "objective/train/weights_max": 2.101844549179077,
      "objective/train/weights_min": 0.36873507499694824,
      "theoretical_loss": 4.16210066421454,
      "tokens_seen": 308019200
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009158241052800514,
      "loss": 0.0932,
      "theoretical_loss": 4.16210066421454,
      "tokens_seen": 308019200
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000915743861338469,
      "loss": 0.0945,
      "theoretical_loss": 4.161689886737255,
      "tokens_seen": 308281344
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009156636173968865,
      "loss": 0.0963,
      "theoretical_loss": 4.161279556121562,
      "tokens_seen": 308543488
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009155833734553042,
      "loss": 0.0952,
      "theoretical_loss": 4.160869671502302,
      "tokens_seen": 308805632
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009155031295137217,
      "loss": 0.0959,
      "theoretical_loss": 4.160460232016725,
      "tokens_seen": 309067776
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009154228855721394,
      "loss": 0.0964,
      "theoretical_loss": 4.16005123680448,
      "tokens_seen": 309329920
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009153426416305569,
      "loss": 0.0931,
      "theoretical_loss": 4.159642685007606,
      "tokens_seen": 309592064
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009152623976889744,
      "loss": 0.0924,
      "theoretical_loss": 4.1592345757705225,
      "tokens_seen": 309854208
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009151821537473922,
      "loss": 0.0917,
      "theoretical_loss": 4.158826908240022,
      "tokens_seen": 310116352
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009151019098058097,
      "loss": 0.0928,
      "theoretical_loss": 4.158419681565265,
      "tokens_seen": 310378496
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009150216658642273,
      "loss": 0.0943,
      "theoretical_loss": 4.1580128948977615,
      "tokens_seen": 310640640
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009149414219226449,
      "loss": 0.0938,
      "theoretical_loss": 4.157606547391374,
      "tokens_seen": 310902784
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009148611779810625,
      "loss": 0.0973,
      "theoretical_loss": 4.157200638202301,
      "tokens_seen": 311164928
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00091478093403948,
      "loss": 0.0901,
      "theoretical_loss": 4.156795166489074,
      "tokens_seen": 311427072
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009147006900978976,
      "loss": 0.0948,
      "theoretical_loss": 4.156390131412543,
      "tokens_seen": 311689216
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009146204461563152,
      "loss": 0.0936,
      "theoretical_loss": 4.155985532135875,
      "tokens_seen": 311951360
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009145402022147327,
      "loss": 0.0966,
      "theoretical_loss": 4.1555813678245395,
      "tokens_seen": 312213504
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009144599582731505,
      "loss": 0.0935,
      "theoretical_loss": 4.155177637646306,
      "tokens_seen": 312475648
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000914379714331568,
      "loss": 0.0914,
      "theoretical_loss": 4.154774340771228,
      "tokens_seen": 312737792
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009142994703899856,
      "loss": 0.095,
      "theoretical_loss": 4.154371476371646,
      "tokens_seen": 312999936
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0009142192264484032,
      "loss": 0.0931,
      "theoretical_loss": 4.153969043622169,
      "tokens_seen": 313262080
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009141389825068207,
      "loss": 0.0963,
      "theoretical_loss": 4.15356704169967,
      "tokens_seen": 313524224
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009140587385652384,
      "loss": 0.0968,
      "theoretical_loss": 4.153165469783279,
      "tokens_seen": 313786368
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009139784946236559,
      "loss": 0.0969,
      "theoretical_loss": 4.152764327054376,
      "tokens_seen": 314048512
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009138982506820735,
      "loss": 0.0947,
      "theoretical_loss": 4.152363612696579,
      "tokens_seen": 314310656
    },
    {
      "epoch": 0.1,
      "objective/train/advantage_avg": 0.0024639612529426813,
      "objective/train/docs_used": 121169,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.8197975158691406,
      "objective/train/original_loss": 1.8197975158691406,
      "objective/train/theoretical_loss": 4.151963325895737,
      "objective/train/tokens_used": 335032800,
      "objective/train/value_avg": -0.0099029541015625,
      "objective/train/value_loss": 0.0005963764269836247,
      "objective/train/value_max": -0.0004029273986816406,
      "objective/train/value_min": -0.27783203125,
      "objective/train/value_reward_corr": 0.45480866686294624,
      "objective/train/value_std": 0.01166534423828125,
      "objective/train/weight_avg": 1.0026694536209106,
      "objective/train/weighted_lm_loss": 1.8254142999649048,
      "objective/train/weights_max": 1.2214922904968262,
      "objective/train/weights_min": 0.020853379741311073,
      "theoretical_loss": 4.151963325895737,
      "tokens_seen": 314572800
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009138180067404911,
      "loss": 0.0933,
      "theoretical_loss": 4.151963325895737,
      "tokens_seen": 314572800
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009137377627989088,
      "loss": 0.0944,
      "theoretical_loss": 4.151563465839927,
      "tokens_seen": 314834944
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009136575188573263,
      "loss": 0.0949,
      "theoretical_loss": 4.151164031719437,
      "tokens_seen": 315097088
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009135772749157439,
      "loss": 0.0956,
      "theoretical_loss": 4.15076502272677,
      "tokens_seen": 315359232
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009134970309741615,
      "loss": 0.0934,
      "theoretical_loss": 4.150366438056622,
      "tokens_seen": 315621376
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000913416787032579,
      "loss": 0.092,
      "theoretical_loss": 4.149968276905888,
      "tokens_seen": 315883520
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009133365430909967,
      "loss": 0.0986,
      "theoretical_loss": 4.149570538473644,
      "tokens_seen": 316145664
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009132562991494142,
      "loss": 0.0909,
      "theoretical_loss": 4.149173221961146,
      "tokens_seen": 316407808
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009131760552078318,
      "loss": 0.0936,
      "theoretical_loss": 4.1487763265718165,
      "tokens_seen": 316669952
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009130958112662494,
      "loss": 0.094,
      "theoretical_loss": 4.148379851511241,
      "tokens_seen": 316932096
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009130155673246669,
      "loss": 0.094,
      "theoretical_loss": 4.147983795987161,
      "tokens_seen": 317194240
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009129353233830846,
      "loss": 0.0934,
      "theoretical_loss": 4.14758815920946,
      "tokens_seen": 317456384
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009128550794415022,
      "loss": 0.0914,
      "theoretical_loss": 4.147192940390165,
      "tokens_seen": 317718528
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009127748354999198,
      "loss": 0.0966,
      "theoretical_loss": 4.146798138743433,
      "tokens_seen": 317980672
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009126945915583374,
      "loss": 0.09,
      "theoretical_loss": 4.146403753485544,
      "tokens_seen": 318242816
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000912614347616755,
      "loss": 0.0927,
      "theoretical_loss": 4.146009783834892,
      "tokens_seen": 318504960
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009125341036751725,
      "loss": 0.095,
      "theoretical_loss": 4.145616229011987,
      "tokens_seen": 318767104
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009124538597335902,
      "loss": 0.0958,
      "theoretical_loss": 4.145223088239432,
      "tokens_seen": 319029248
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009123736157920077,
      "loss": 0.0922,
      "theoretical_loss": 4.14483036074193,
      "tokens_seen": 319291392
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009122933718504252,
      "loss": 0.0958,
      "theoretical_loss": 4.14443804574627,
      "tokens_seen": 319553536
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000912213127908843,
      "loss": 0.0925,
      "theoretical_loss": 4.144046142481317,
      "tokens_seen": 319815680
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009121328839672605,
      "loss": 0.0965,
      "theoretical_loss": 4.143654650178012,
      "tokens_seen": 320077824
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009120526400256781,
      "loss": 0.0966,
      "theoretical_loss": 4.143263568069358,
      "tokens_seen": 320339968
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009119723960840957,
      "loss": 0.0925,
      "theoretical_loss": 4.142872895390417,
      "tokens_seen": 320602112
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009118921521425133,
      "loss": 0.0938,
      "theoretical_loss": 4.142482631378303,
      "tokens_seen": 320864256
    },
    {
      "epoch": 0.1,
      "objective/train/advantage_avg": 0.00016702228458598256,
      "objective/train/docs_used": 123582,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.8835115432739258,
      "objective/train/original_loss": 1.8835114240646362,
      "objective/train/theoretical_loss": 4.142092775272169,
      "objective/train/tokens_used": 341586400,
      "objective/train/value_avg": -0.009979248046875,
      "objective/train/value_loss": 0.000347665831213817,
      "objective/train/value_max": -0.0004355907440185547,
      "objective/train/value_min": -0.2398681640625,
      "objective/train/value_reward_corr": 0.6063551562843142,
      "objective/train/value_std": 0.012237548828125,
      "objective/train/weight_avg": 1.0003257989883423,
      "objective/train/weighted_lm_loss": 1.8840646743774414,
      "objective/train/weights_max": 1.1990368366241455,
      "objective/train/weights_min": 0.3737753629684448,
      "theoretical_loss": 4.142092775272169,
      "tokens_seen": 321126400
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009118119082009308,
      "loss": 0.0913,
      "theoretical_loss": 4.142092775272169,
      "tokens_seen": 321126400
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009117316642593484,
      "loss": 0.0893,
      "theoretical_loss": 4.141703326313209,
      "tokens_seen": 321388544
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000911651420317766,
      "loss": 0.0931,
      "theoretical_loss": 4.141314283744643,
      "tokens_seen": 321650688
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009115711763761835,
      "loss": 0.0925,
      "theoretical_loss": 4.140925646811714,
      "tokens_seen": 321912832
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009114909324346013,
      "loss": 0.0944,
      "theoretical_loss": 4.1405374147616785,
      "tokens_seen": 322174976
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009114106884930188,
      "loss": 0.0928,
      "theoretical_loss": 4.140149586843803,
      "tokens_seen": 322437120
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009113304445514365,
      "loss": 0.0937,
      "theoretical_loss": 4.139762162309355,
      "tokens_seen": 322699264
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000911250200609854,
      "loss": 0.0929,
      "theoretical_loss": 4.139375140411592,
      "tokens_seen": 322961408
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009111699566682715,
      "loss": 0.091,
      "theoretical_loss": 4.138988520405764,
      "tokens_seen": 323223552
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009110897127266892,
      "loss": 0.0919,
      "theoretical_loss": 4.138602301549097,
      "tokens_seen": 323485696
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009110094687851067,
      "loss": 0.0917,
      "theoretical_loss": 4.138216483100795,
      "tokens_seen": 323747840
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009109292248435243,
      "loss": 0.0897,
      "theoretical_loss": 4.137831064322021,
      "tokens_seen": 324009984
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009108489809019419,
      "loss": 0.0901,
      "theoretical_loss": 4.1374460444759045,
      "tokens_seen": 324272128
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009107687369603596,
      "loss": 0.092,
      "theoretical_loss": 4.137061422827525,
      "tokens_seen": 324534272
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009106884930187771,
      "loss": 0.0937,
      "theoretical_loss": 4.136677198643908,
      "tokens_seen": 324796416
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009106082490771947,
      "loss": 0.0912,
      "theoretical_loss": 4.13629337119402,
      "tokens_seen": 325058560
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009105280051356123,
      "loss": 0.0919,
      "theoretical_loss": 4.135909939748757,
      "tokens_seen": 325320704
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009104477611940298,
      "loss": 0.0915,
      "theoretical_loss": 4.135526903580946,
      "tokens_seen": 325582848
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009103675172524475,
      "loss": 0.0933,
      "theoretical_loss": 4.135144261965327,
      "tokens_seen": 325844992
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000910287273310865,
      "loss": 0.0921,
      "theoretical_loss": 4.134762014178559,
      "tokens_seen": 326107136
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009102070293692827,
      "loss": 0.092,
      "theoretical_loss": 4.134380159499204,
      "tokens_seen": 326369280
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009101267854277002,
      "loss": 0.0928,
      "theoretical_loss": 4.1339986972077245,
      "tokens_seen": 326631424
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009100465414861177,
      "loss": 0.0881,
      "theoretical_loss": 4.133617626586475,
      "tokens_seen": 326893568
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009099662975445355,
      "loss": 0.0911,
      "theoretical_loss": 4.133236946919698,
      "tokens_seen": 327155712
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000909886053602953,
      "loss": 0.0969,
      "theoretical_loss": 4.132856657493516,
      "tokens_seen": 327417856
    },
    {
      "epoch": 0.1,
      "objective/train/advantage_avg": 0.0012691118754446507,
      "objective/train/docs_used": 126013,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.943372130393982,
      "objective/train/original_loss": 1.9433720111846924,
      "objective/train/theoretical_loss": 4.132476757595925,
      "objective/train/tokens_used": 348140000,
      "objective/train/value_avg": -0.00861358642578125,
      "objective/train/value_loss": 0.0002645694767124951,
      "objective/train/value_max": -0.00031757354736328125,
      "objective/train/value_min": -0.280517578125,
      "objective/train/value_reward_corr": 0.6476616622246956,
      "objective/train/value_std": 0.01280975341796875,
      "objective/train/weight_avg": 1.0013933181762695,
      "objective/train/weighted_lm_loss": 1.9461336135864258,
      "objective/train/weights_max": 1.204296588897705,
      "objective/train/weights_min": 0.3697463870048523,
      "theoretical_loss": 4.132476757595925,
      "tokens_seen": 327680000
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009098058096613706,
      "loss": 0.095,
      "theoretical_loss": 4.132476757595925,
      "tokens_seen": 327680000
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009097255657197882,
      "loss": 0.0945,
      "theoretical_loss": 4.132097246516788,
      "tokens_seen": 327942144
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009096453217782058,
      "loss": 0.094,
      "theoretical_loss": 4.131718123547829,
      "tokens_seen": 328204288
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009095650778366233,
      "loss": 0.0928,
      "theoretical_loss": 4.131339387982628,
      "tokens_seen": 328466432
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000909484833895041,
      "loss": 0.091,
      "theoretical_loss": 4.1309610391166105,
      "tokens_seen": 328728576
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009094045899534585,
      "loss": 0.09,
      "theoretical_loss": 4.1305830762470475,
      "tokens_seen": 328990720
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000909324346011876,
      "loss": 0.0925,
      "theoretical_loss": 4.1302054986730425,
      "tokens_seen": 329252864
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009092441020702938,
      "loss": 0.0918,
      "theoretical_loss": 4.129828305695531,
      "tokens_seen": 329515008
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009091638581287113,
      "loss": 0.0915,
      "theoretical_loss": 4.129451496617269,
      "tokens_seen": 329777152
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009090836141871289,
      "loss": 0.0936,
      "theoretical_loss": 4.129075070742831,
      "tokens_seen": 330039296
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009090033702455465,
      "loss": 0.0929,
      "theoretical_loss": 4.128699027378604,
      "tokens_seen": 330301440
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009089231263039641,
      "loss": 0.0921,
      "theoretical_loss": 4.128323365832777,
      "tokens_seen": 330563584
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009088428823623817,
      "loss": 0.0936,
      "theoretical_loss": 4.127948085415338,
      "tokens_seen": 330825728
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009087626384207992,
      "loss": 0.0937,
      "theoretical_loss": 4.127573185438068,
      "tokens_seen": 331087872
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009086823944792168,
      "loss": 0.0959,
      "theoretical_loss": 4.127198665214536,
      "tokens_seen": 331350016
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009086021505376344,
      "loss": 0.0897,
      "theoretical_loss": 4.126824524060088,
      "tokens_seen": 331612160
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000908521906596052,
      "loss": 0.0925,
      "theoretical_loss": 4.126450761291847,
      "tokens_seen": 331874304
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009084416626544696,
      "loss": 0.0948,
      "theoretical_loss": 4.126077376228702,
      "tokens_seen": 332136448
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009083614187128873,
      "loss": 0.0874,
      "theoretical_loss": 4.1257043681913075,
      "tokens_seen": 332398592
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009082811747713048,
      "loss": 0.0917,
      "theoretical_loss": 4.125331736502073,
      "tokens_seen": 332660736
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009082009308297223,
      "loss": 0.0937,
      "theoretical_loss": 4.124959480485156,
      "tokens_seen": 332922880
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00090812068688814,
      "loss": 0.0899,
      "theoretical_loss": 4.124587599466462,
      "tokens_seen": 333185024
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009080404429465575,
      "loss": 0.0936,
      "theoretical_loss": 4.124216092773635,
      "tokens_seen": 333447168
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009079601990049751,
      "loss": 0.0895,
      "theoretical_loss": 4.123844959736049,
      "tokens_seen": 333709312
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009078799550633927,
      "loss": 0.091,
      "theoretical_loss": 4.123474199684807,
      "tokens_seen": 333971456
    },
    {
      "epoch": 0.1,
      "objective/train/advantage_avg": 0.002065310487523675,
      "objective/train/docs_used": 128373,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.9022668600082397,
      "objective/train/original_loss": 1.9022668600082397,
      "objective/train/theoretical_loss": 4.123103811952736,
      "objective/train/tokens_used": 354693600,
      "objective/train/value_avg": -0.00934600830078125,
      "objective/train/value_loss": 0.00017411113367415965,
      "objective/train/value_max": -0.000469207763671875,
      "objective/train/value_min": -0.1983642578125,
      "objective/train/value_reward_corr": 0.5511531255621919,
      "objective/train/value_std": 0.00977325439453125,
      "objective/train/weight_avg": 1.0021498203277588,
      "objective/train/weighted_lm_loss": 1.9055767059326172,
      "objective/train/weights_max": 1.1557458639144897,
      "objective/train/weights_min": 0.36877235770225525,
      "theoretical_loss": 4.123103811952736,
      "tokens_seen": 334233600
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009077997111218103,
      "loss": 0.0956,
      "theoretical_loss": 4.123103811952736,
      "tokens_seen": 334233600
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000907719467180228,
      "loss": 0.089,
      "theoretical_loss": 4.122733795874372,
      "tokens_seen": 334495744
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009076392232386455,
      "loss": 0.0941,
      "theoretical_loss": 4.122364150785966,
      "tokens_seen": 334757888
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009075589792970631,
      "loss": 0.0942,
      "theoretical_loss": 4.1219948760254725,
      "tokens_seen": 335020032
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009074787353554807,
      "loss": 0.093,
      "theoretical_loss": 4.121625970932542,
      "tokens_seen": 335282176
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009073984914138983,
      "loss": 0.0926,
      "theoretical_loss": 4.121257434848519,
      "tokens_seen": 335544320
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009073182474723158,
      "loss": 0.0921,
      "theoretical_loss": 4.120889267116435,
      "tokens_seen": 335806464
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009072380035307335,
      "loss": 0.0923,
      "theoretical_loss": 4.1205214670810015,
      "tokens_seen": 336068608
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000907157759589151,
      "loss": 0.0923,
      "theoretical_loss": 4.120154034088609,
      "tokens_seen": 336330752
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009070775156475685,
      "loss": 0.0922,
      "theoretical_loss": 4.119786967487314,
      "tokens_seen": 336592896
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009069972717059863,
      "loss": 0.0935,
      "theoretical_loss": 4.11942026662684,
      "tokens_seen": 336855040
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009069170277644038,
      "loss": 0.0896,
      "theoretical_loss": 4.11905393085857,
      "tokens_seen": 337117184
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009068367838228214,
      "loss": 0.0926,
      "theoretical_loss": 4.118687959535539,
      "tokens_seen": 337379328
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000906756539881239,
      "loss": 0.0906,
      "theoretical_loss": 4.118322352012429,
      "tokens_seen": 337641472
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009066762959396566,
      "loss": 0.0902,
      "theoretical_loss": 4.117957107645569,
      "tokens_seen": 337903616
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009065960519980741,
      "loss": 0.0925,
      "theoretical_loss": 4.1175922257929205,
      "tokens_seen": 338165760
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009065158080564918,
      "loss": 0.0945,
      "theoretical_loss": 4.117227705814078,
      "tokens_seen": 338427904
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009064355641149093,
      "loss": 0.0971,
      "theoretical_loss": 4.116863547070264,
      "tokens_seen": 338690048
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009063553201733269,
      "loss": 0.0927,
      "theoretical_loss": 4.116499748924319,
      "tokens_seen": 338952192
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009062750762317446,
      "loss": 0.093,
      "theoretical_loss": 4.116136310740702,
      "tokens_seen": 339214336
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009061948322901621,
      "loss": 0.0871,
      "theoretical_loss": 4.115773231885479,
      "tokens_seen": 339476480
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009061145883485798,
      "loss": 0.0904,
      "theoretical_loss": 4.115410511726323,
      "tokens_seen": 339738624
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009060343444069973,
      "loss": 0.0921,
      "theoretical_loss": 4.115048149632507,
      "tokens_seen": 340000768
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009059541004654149,
      "loss": 0.0933,
      "theoretical_loss": 4.114686144974897,
      "tokens_seen": 340262912
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009058738565238325,
      "loss": 0.0923,
      "theoretical_loss": 4.114324497125947,
      "tokens_seen": 340525056
    },
    {
      "epoch": 0.1,
      "objective/train/advantage_avg": 0.0013270352501422167,
      "objective/train/docs_used": 130755,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.773648977279663,
      "objective/train/original_loss": 1.7736492156982422,
      "objective/train/theoretical_loss": 4.113963205459697,
      "objective/train/tokens_used": 361247200,
      "objective/train/value_avg": -0.0078582763671875,
      "objective/train/value_loss": 0.0002814758918248117,
      "objective/train/value_max": -0.0004305839538574219,
      "objective/train/value_min": -0.75634765625,
      "objective/train/value_reward_corr": 0.6420233629882061,
      "objective/train/value_std": 0.01309967041015625,
      "objective/train/weight_avg": 1.0014532804489136,
      "objective/train/weighted_lm_loss": 1.775871753692627,
      "objective/train/weights_max": 1.4132821559906006,
      "objective/train/weights_min": 0.38253408670425415,
      "theoretical_loss": 4.113963205459697,
      "tokens_seen": 340787200
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00090579361258225,
      "loss": 0.0906,
      "theoretical_loss": 4.113963205459697,
      "tokens_seen": 340787200
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009057133686406676,
      "loss": 0.091,
      "theoretical_loss": 4.113602269351765,
      "tokens_seen": 341049344
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009056331246990852,
      "loss": 0.0903,
      "theoretical_loss": 4.113241688179341,
      "tokens_seen": 341311488
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009055528807575029,
      "loss": 0.093,
      "theoretical_loss": 4.1128814613211855,
      "tokens_seen": 341573632
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009054726368159204,
      "loss": 0.0896,
      "theoretical_loss": 4.1125215881576205,
      "tokens_seen": 341835776
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009053923928743381,
      "loss": 0.0899,
      "theoretical_loss": 4.112162068070525,
      "tokens_seen": 342097920
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009053121489327556,
      "loss": 0.0911,
      "theoretical_loss": 4.111802900443333,
      "tokens_seen": 342360064
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009052319049911731,
      "loss": 0.0891,
      "theoretical_loss": 4.111444084661026,
      "tokens_seen": 342622208
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009051516610495908,
      "loss": 0.0888,
      "theoretical_loss": 4.111085620110127,
      "tokens_seen": 342884352
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009050714171080083,
      "loss": 0.0914,
      "theoretical_loss": 4.110727506178697,
      "tokens_seen": 343146496
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000904991173166426,
      "loss": 0.0929,
      "theoretical_loss": 4.110369742256329,
      "tokens_seen": 343408640
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009049109292248435,
      "loss": 0.09,
      "theoretical_loss": 4.110012327734145,
      "tokens_seen": 343670784
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009048306852832611,
      "loss": 0.091,
      "theoretical_loss": 4.1096552620047895,
      "tokens_seen": 343932928
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009047504413416788,
      "loss": 0.0921,
      "theoretical_loss": 4.109298544462423,
      "tokens_seen": 344195072
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009046701974000963,
      "loss": 0.0867,
      "theoretical_loss": 4.108942174502721,
      "tokens_seen": 344457216
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009045899534585139,
      "loss": 0.0907,
      "theoretical_loss": 4.108586151522863,
      "tokens_seen": 344719360
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009045097095169315,
      "loss": 0.0932,
      "theoretical_loss": 4.1082304749215375,
      "tokens_seen": 344981504
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009044294655753491,
      "loss": 0.091,
      "theoretical_loss": 4.107875144098925,
      "tokens_seen": 345243648
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009043492216337666,
      "loss": 0.0949,
      "theoretical_loss": 4.107520158456703,
      "tokens_seen": 345505792
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009042689776921843,
      "loss": 0.0901,
      "theoretical_loss": 4.107165517398034,
      "tokens_seen": 345767936
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009041887337506018,
      "loss": 0.0926,
      "theoretical_loss": 4.106811220327568,
      "tokens_seen": 346030080
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0009041084898090193,
      "loss": 0.0906,
      "theoretical_loss": 4.10645726665143,
      "tokens_seen": 346292224
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009040282458674371,
      "loss": 0.0934,
      "theoretical_loss": 4.10610365577722,
      "tokens_seen": 346554368
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009039480019258546,
      "loss": 0.0913,
      "theoretical_loss": 4.105750387114009,
      "tokens_seen": 346816512
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009038677579842723,
      "loss": 0.0905,
      "theoretical_loss": 4.105397460072329,
      "tokens_seen": 347078656
    },
    {
      "epoch": 0.11,
      "objective/train/advantage_avg": 0.002003602683544159,
      "objective/train/docs_used": 133174,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.733391523361206,
      "objective/train/original_loss": 1.7333914041519165,
      "objective/train/theoretical_loss": 4.105044874064174,
      "objective/train/tokens_used": 367800800,
      "objective/train/value_avg": -0.0088043212890625,
      "objective/train/value_loss": 0.00023858436907175928,
      "objective/train/value_max": -0.0003998279571533203,
      "objective/train/value_min": -0.9033203125,
      "objective/train/value_reward_corr": 0.6106213393331679,
      "objective/train/value_std": 0.0132904052734375,
      "objective/train/weight_avg": 1.002118468284607,
      "objective/train/weighted_lm_loss": 1.736554741859436,
      "objective/train/weights_max": 2.2593677043914795,
      "objective/train/weights_min": 0.37059226632118225,
      "theoretical_loss": 4.105044874064174,
      "tokens_seen": 347340800
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009037875140426898,
      "loss": 0.0871,
      "theoretical_loss": 4.105044874064174,
      "tokens_seen": 347340800
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009037072701011074,
      "loss": 0.0909,
      "theoretical_loss": 4.104692628502993,
      "tokens_seen": 347602944
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000903627026159525,
      "loss": 0.0895,
      "theoretical_loss": 4.104340722803683,
      "tokens_seen": 347865088
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009035467822179425,
      "loss": 0.0903,
      "theoretical_loss": 4.103989156382589,
      "tokens_seen": 348127232
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009034665382763601,
      "loss": 0.0873,
      "theoretical_loss": 4.103637928657495,
      "tokens_seen": 348389376
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009033862943347777,
      "loss": 0.0953,
      "theoretical_loss": 4.103287039047622,
      "tokens_seen": 348651520
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009033060503931954,
      "loss": 0.0901,
      "theoretical_loss": 4.102936486973624,
      "tokens_seen": 348913664
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009032258064516129,
      "loss": 0.0896,
      "theoretical_loss": 4.102586271857579,
      "tokens_seen": 349175808
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009031455625100306,
      "loss": 0.0882,
      "theoretical_loss": 4.102236393122989,
      "tokens_seen": 349437952
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009030653185684481,
      "loss": 0.0915,
      "theoretical_loss": 4.101886850194775,
      "tokens_seen": 349700096
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009029850746268657,
      "loss": 0.0882,
      "theoretical_loss": 4.10153764249927,
      "tokens_seen": 349962240
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009029048306852833,
      "loss": 0.0884,
      "theoretical_loss": 4.1011887694642155,
      "tokens_seen": 350224384
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009028245867437008,
      "loss": 0.0897,
      "theoretical_loss": 4.100840230518759,
      "tokens_seen": 350486528
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009027443428021184,
      "loss": 0.0918,
      "theoretical_loss": 4.100492025093445,
      "tokens_seen": 350748672
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000902664098860536,
      "loss": 0.0919,
      "theoretical_loss": 4.100144152620215,
      "tokens_seen": 351010816
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009025838549189537,
      "loss": 0.0904,
      "theoretical_loss": 4.099796612532403,
      "tokens_seen": 351272960
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009025036109773713,
      "loss": 0.0898,
      "theoretical_loss": 4.0994494042647265,
      "tokens_seen": 351535104
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009024233670357889,
      "loss": 0.0904,
      "theoretical_loss": 4.099102527253285,
      "tokens_seen": 351797248
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009023431230942064,
      "loss": 0.0923,
      "theoretical_loss": 4.098755980935557,
      "tokens_seen": 352059392
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000902262879152624,
      "loss": 0.0893,
      "theoretical_loss": 4.0984097647503965,
      "tokens_seen": 352321536
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009021826352110416,
      "loss": 0.0903,
      "theoretical_loss": 4.09806387813802,
      "tokens_seen": 352583680
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009021023912694591,
      "loss": 0.0875,
      "theoretical_loss": 4.0977183205400145,
      "tokens_seen": 352845824
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009020221473278768,
      "loss": 0.0897,
      "theoretical_loss": 4.097373091399324,
      "tokens_seen": 353107968
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009019419033862943,
      "loss": 0.0896,
      "theoretical_loss": 4.097028190160249,
      "tokens_seen": 353370112
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000901861659444712,
      "loss": 0.0909,
      "theoretical_loss": 4.096683616268442,
      "tokens_seen": 353632256
    },
    {
      "epoch": 0.11,
      "objective/train/advantage_avg": 0.003045022254809737,
      "objective/train/docs_used": 135614,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.740461826324463,
      "objective/train/original_loss": 1.7404619455337524,
      "objective/train/theoretical_loss": 4.096339369170902,
      "objective/train/tokens_used": 374354400,
      "objective/train/value_avg": -0.01047515869140625,
      "objective/train/value_loss": 0.0004944366519339383,
      "objective/train/value_max": -0.00038909912109375,
      "objective/train/value_min": -0.673828125,
      "objective/train/value_reward_corr": 0.5444227902078449,
      "objective/train/value_std": 0.01522064208984375,
      "objective/train/weight_avg": 1.00326406955719,
      "objective/train/weighted_lm_loss": 1.7446165084838867,
      "objective/train/weights_max": 1.2945656776428223,
      "objective/train/weights_min": 0.3746718466281891,
      "theoretical_loss": 4.096339369170902,
      "tokens_seen": 353894400
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009017814155031296,
      "loss": 0.0898,
      "theoretical_loss": 4.096339369170902,
      "tokens_seen": 353894400
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009017011715615471,
      "loss": 0.0909,
      "theoretical_loss": 4.095995448315972,
      "tokens_seen": 354156544
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009016209276199647,
      "loss": 0.0899,
      "theoretical_loss": 4.095651853153331,
      "tokens_seen": 354418688
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009015406836783823,
      "loss": 0.0912,
      "theoretical_loss": 4.095308583133997,
      "tokens_seen": 354680832
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009014604397367999,
      "loss": 0.0915,
      "theoretical_loss": 4.094965637710314,
      "tokens_seen": 354942976
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009013801957952175,
      "loss": 0.0903,
      "theoretical_loss": 4.094623016335954,
      "tokens_seen": 355205120
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009012999518536351,
      "loss": 0.0899,
      "theoretical_loss": 4.094280718465911,
      "tokens_seen": 355467264
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009012197079120526,
      "loss": 0.09,
      "theoretical_loss": 4.093938743556496,
      "tokens_seen": 355729408
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009011394639704702,
      "loss": 0.0889,
      "theoretical_loss": 4.093597091065333,
      "tokens_seen": 355991552
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009010592200288879,
      "loss": 0.089,
      "theoretical_loss": 4.093255760451357,
      "tokens_seen": 356253696
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009009789760873054,
      "loss": 0.0883,
      "theoretical_loss": 4.092914751174808,
      "tokens_seen": 356515840
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009008987321457231,
      "loss": 0.0898,
      "theoretical_loss": 4.092574062697225,
      "tokens_seen": 356777984
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009008184882041406,
      "loss": 0.0881,
      "theoretical_loss": 4.092233694481447,
      "tokens_seen": 357040128
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009007382442625582,
      "loss": 0.0927,
      "theoretical_loss": 4.091893645991604,
      "tokens_seen": 357302272
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009006580003209758,
      "loss": 0.0913,
      "theoretical_loss": 4.091553916693115,
      "tokens_seen": 357564416
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009005777563793933,
      "loss": 0.0859,
      "theoretical_loss": 4.091214506052687,
      "tokens_seen": 357826560
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009004975124378109,
      "loss": 0.0841,
      "theoretical_loss": 4.090875413538302,
      "tokens_seen": 358088704
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009004172684962285,
      "loss": 0.0921,
      "theoretical_loss": 4.090536638619224,
      "tokens_seen": 358350848
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009003370245546462,
      "loss": 0.0872,
      "theoretical_loss": 4.090198180765989,
      "tokens_seen": 358612992
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009002567806130637,
      "loss": 0.0893,
      "theoretical_loss": 4.0898600394504,
      "tokens_seen": 358875136
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009001765366714814,
      "loss": 0.0888,
      "theoretical_loss": 4.089522214145525,
      "tokens_seen": 359137280
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009000962927298989,
      "loss": 0.0911,
      "theoretical_loss": 4.089184704325695,
      "tokens_seen": 359399424
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0009000160487883166,
      "loss": 0.0886,
      "theoretical_loss": 4.088847509466497,
      "tokens_seen": 359661568
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008999358048467341,
      "loss": 0.0885,
      "theoretical_loss": 4.088510629044771,
      "tokens_seen": 359923712
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008998555609051516,
      "loss": 0.0885,
      "theoretical_loss": 4.088174062538605,
      "tokens_seen": 360185856
    },
    {
      "epoch": 0.11,
      "objective/train/advantage_avg": 0.0014361967332661152,
      "objective/train/docs_used": 137426,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.912741780281067,
      "objective/train/original_loss": 1.9127418994903564,
      "objective/train/theoretical_loss": 4.087837809427334,
      "objective/train/tokens_used": 380908000,
      "objective/train/value_avg": -0.0081939697265625,
      "objective/train/value_loss": 0.00017770560225471854,
      "objective/train/value_max": -0.0002913475036621094,
      "objective/train/value_min": -0.2374267578125,
      "objective/train/value_reward_corr": 0.6243051078626306,
      "objective/train/value_std": 0.0115814208984375,
      "objective/train/weight_avg": 1.0015201568603516,
      "objective/train/weighted_lm_loss": 1.9144210815429688,
      "objective/train/weights_max": 1.1172159910202026,
      "objective/train/weights_min": 0.36891940236091614,
      "theoretical_loss": 4.087837809427334,
      "tokens_seen": 360448000
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008997753169635693,
      "loss": 0.093,
      "theoretical_loss": 4.087837809427334,
      "tokens_seen": 360448000
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008996950730219868,
      "loss": 0.088,
      "theoretical_loss": 4.087501869191536,
      "tokens_seen": 360710144
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008996148290804045,
      "loss": 0.0877,
      "theoretical_loss": 4.087166241313023,
      "tokens_seen": 360972288
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008995345851388221,
      "loss": 0.0925,
      "theoretical_loss": 4.086830925274842,
      "tokens_seen": 361234432
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008994543411972397,
      "loss": 0.0887,
      "theoretical_loss": 4.08649592056127,
      "tokens_seen": 361496576
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008993740972556572,
      "loss": 0.0926,
      "theoretical_loss": 4.086161226657811,
      "tokens_seen": 361758720
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008992938533140748,
      "loss": 0.0857,
      "theoretical_loss": 4.085826843051191,
      "tokens_seen": 362020864
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008992136093724924,
      "loss": 0.0915,
      "theoretical_loss": 4.0854927692293534,
      "tokens_seen": 362283008
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008991333654309099,
      "loss": 0.0891,
      "theoretical_loss": 4.085159004681458,
      "tokens_seen": 362545152
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008990531214893276,
      "loss": 0.0904,
      "theoretical_loss": 4.084825548897873,
      "tokens_seen": 362807296
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008989728775477451,
      "loss": 0.0932,
      "theoretical_loss": 4.084492401370177,
      "tokens_seen": 363069440
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008988926336061629,
      "loss": 0.0911,
      "theoretical_loss": 4.08415956159115,
      "tokens_seen": 363331584
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008988123896645804,
      "loss": 0.0894,
      "theoretical_loss": 4.083827029054773,
      "tokens_seen": 363593728
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008987321457229979,
      "loss": 0.0912,
      "theoretical_loss": 4.0834948032562215,
      "tokens_seen": 363855872
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008986519017814156,
      "loss": 0.0909,
      "theoretical_loss": 4.083162883691864,
      "tokens_seen": 364118016
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008985716578398331,
      "loss": 0.0913,
      "theoretical_loss": 4.082831269859261,
      "tokens_seen": 364380160
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008984914138982507,
      "loss": 0.0901,
      "theoretical_loss": 4.0824999612571515,
      "tokens_seen": 364642304
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008984111699566683,
      "loss": 0.0889,
      "theoretical_loss": 4.082168957385462,
      "tokens_seen": 364904448
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008983309260150859,
      "loss": 0.0917,
      "theoretical_loss": 4.081838257745293,
      "tokens_seen": 365166592
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008982506820735034,
      "loss": 0.0873,
      "theoretical_loss": 4.081507861838922,
      "tokens_seen": 365428736
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000898170438131921,
      "loss": 0.0901,
      "theoretical_loss": 4.081177769169795,
      "tokens_seen": 365690880
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008980901941903387,
      "loss": 0.0872,
      "theoretical_loss": 4.080847979242526,
      "tokens_seen": 365953024
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008980099502487562,
      "loss": 0.0898,
      "theoretical_loss": 4.0805184915628905,
      "tokens_seen": 366215168
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008979297063071739,
      "loss": 0.0891,
      "theoretical_loss": 4.080189305637827,
      "tokens_seen": 366477312
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008978494623655914,
      "loss": 0.0893,
      "theoretical_loss": 4.079860420975429,
      "tokens_seen": 366739456
    },
    {
      "epoch": 0.11,
      "objective/train/advantage_avg": 0.0024263346567749977,
      "objective/train/docs_used": 139849,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.8241596221923828,
      "objective/train/original_loss": 1.8241596221923828,
      "objective/train/theoretical_loss": 4.07953183708494,
      "objective/train/tokens_used": 387461600,
      "objective/train/value_avg": -0.0073089599609375,
      "objective/train/value_loss": 0.00016963874804787338,
      "objective/train/value_max": -0.00034999847412109375,
      "objective/train/value_min": -0.385498046875,
      "objective/train/value_reward_corr": 0.4735220244507201,
      "objective/train/value_std": 0.008209228515625,
      "objective/train/weight_avg": 1.002504587173462,
      "objective/train/weighted_lm_loss": 1.8287756443023682,
      "objective/train/weights_max": 1.386350154876709,
      "objective/train/weights_min": 0.3713734447956085,
      "theoretical_loss": 4.07953183708494,
      "tokens_seen": 367001600
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000897769218424009,
      "loss": 0.0945,
      "theoretical_loss": 4.07953183708494,
      "tokens_seen": 367001600
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008976889744824266,
      "loss": 0.0911,
      "theoretical_loss": 4.079203553476759,
      "tokens_seen": 367263744
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008976087305408441,
      "loss": 0.0914,
      "theoretical_loss": 4.078875569662424,
      "tokens_seen": 367525888
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008975284865992618,
      "loss": 0.0883,
      "theoretical_loss": 4.07854788515462,
      "tokens_seen": 367788032
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008974482426576793,
      "loss": 0.0893,
      "theoretical_loss": 4.078220499467168,
      "tokens_seen": 368050176
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000897367998716097,
      "loss": 0.0896,
      "theoretical_loss": 4.077893412115025,
      "tokens_seen": 368312320
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008972877547745146,
      "loss": 0.0872,
      "theoretical_loss": 4.077566622614281,
      "tokens_seen": 368574464
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008972075108329322,
      "loss": 0.089,
      "theoretical_loss": 4.077240130482153,
      "tokens_seen": 368836608
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008971272668913497,
      "loss": 0.0912,
      "theoretical_loss": 4.076913935236982,
      "tokens_seen": 369098752
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008970470229497674,
      "loss": 0.089,
      "theoretical_loss": 4.076588036398235,
      "tokens_seen": 369360896
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008969667790081849,
      "loss": 0.0887,
      "theoretical_loss": 4.076262433486491,
      "tokens_seen": 369623040
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008968865350666024,
      "loss": 0.0909,
      "theoretical_loss": 4.075937126023448,
      "tokens_seen": 369885184
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008968062911250201,
      "loss": 0.0886,
      "theoretical_loss": 4.075612113531915,
      "tokens_seen": 370147328
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008967260471834376,
      "loss": 0.0882,
      "theoretical_loss": 4.075287395535807,
      "tokens_seen": 370409472
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008966458032418553,
      "loss": 0.0894,
      "theoretical_loss": 4.074962971560145,
      "tokens_seen": 370671616
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008965655593002729,
      "loss": 0.0882,
      "theoretical_loss": 4.074638841131052,
      "tokens_seen": 370933760
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008964853153586905,
      "loss": 0.09,
      "theoretical_loss": 4.074315003775746,
      "tokens_seen": 371195904
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000896405071417108,
      "loss": 0.0913,
      "theoretical_loss": 4.073991459022544,
      "tokens_seen": 371458048
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008963248274755256,
      "loss": 0.0868,
      "theoretical_loss": 4.073668206400851,
      "tokens_seen": 371720192
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008962445835339432,
      "loss": 0.0952,
      "theoretical_loss": 4.0733452454411605,
      "tokens_seen": 371982336
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008961643395923608,
      "loss": 0.0901,
      "theoretical_loss": 4.0730225756750515,
      "tokens_seen": 372244480
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008960840956507784,
      "loss": 0.0874,
      "theoretical_loss": 4.072700196635185,
      "tokens_seen": 372506624
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008960038517091959,
      "loss": 0.0904,
      "theoretical_loss": 4.072378107855299,
      "tokens_seen": 372768768
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008959236077676137,
      "loss": 0.0864,
      "theoretical_loss": 4.0720563088702075,
      "tokens_seen": 373030912
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008958433638260312,
      "loss": 0.0887,
      "theoretical_loss": 4.071734799215796,
      "tokens_seen": 373293056
    },
    {
      "epoch": 0.11,
      "objective/train/advantage_avg": 0.0020079510286450386,
      "objective/train/docs_used": 142333,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.8369685411453247,
      "objective/train/original_loss": 1.8369685411453247,
      "objective/train/theoretical_loss": 4.071413578429017,
      "objective/train/tokens_used": 394015200,
      "objective/train/value_avg": -0.01097869873046875,
      "objective/train/value_loss": 0.000343000196153298,
      "objective/train/value_max": -0.00045299530029296875,
      "objective/train/value_min": -0.6923828125,
      "objective/train/value_reward_corr": 0.672175762308715,
      "objective/train/value_std": 0.0175018310546875,
      "objective/train/weight_avg": 1.0021660327911377,
      "objective/train/weighted_lm_loss": 1.8394285440444946,
      "objective/train/weights_max": 1.6616524457931519,
      "objective/train/weights_min": 0.37235790491104126,
      "theoretical_loss": 4.071413578429017,
      "tokens_seen": 373555200
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008957631198844487,
      "loss": 0.0874,
      "theoretical_loss": 4.071413578429017,
      "tokens_seen": 373555200
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008956828759428664,
      "loss": 0.0895,
      "theoretical_loss": 4.071092646047892,
      "tokens_seen": 373817344
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008956026320012839,
      "loss": 0.088,
      "theoretical_loss": 4.0707720016115,
      "tokens_seen": 374079488
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008955223880597015,
      "loss": 0.0872,
      "theoretical_loss": 4.0704516446599825,
      "tokens_seen": 374341632
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008954421441181191,
      "loss": 0.0915,
      "theoretical_loss": 4.070131574734536,
      "tokens_seen": 374603776
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008953619001765367,
      "loss": 0.0894,
      "theoretical_loss": 4.069811791377409,
      "tokens_seen": 374865920
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008952816562349542,
      "loss": 0.0883,
      "theoretical_loss": 4.0694922941319,
      "tokens_seen": 375128064
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008952014122933718,
      "loss": 0.091,
      "theoretical_loss": 4.069173082542351,
      "tokens_seen": 375390208
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008951211683517895,
      "loss": 0.0935,
      "theoretical_loss": 4.068854156154154,
      "tokens_seen": 375652352
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000895040924410207,
      "loss": 0.0902,
      "theoretical_loss": 4.068535514513734,
      "tokens_seen": 375914496
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008949606804686247,
      "loss": 0.0913,
      "theoretical_loss": 4.068217157168556,
      "tokens_seen": 376176640
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008948804365270422,
      "loss": 0.0892,
      "theoretical_loss": 4.06789908366712,
      "tokens_seen": 376438784
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008948001925854599,
      "loss": 0.0916,
      "theoretical_loss": 4.067581293558955,
      "tokens_seen": 376700928
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008947199486438774,
      "loss": 0.0913,
      "theoretical_loss": 4.0672637863946175,
      "tokens_seen": 376963072
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008946397047022949,
      "loss": 0.0882,
      "theoretical_loss": 4.0669465617256915,
      "tokens_seen": 377225216
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008945594607607126,
      "loss": 0.0874,
      "theoretical_loss": 4.06662961910478,
      "tokens_seen": 377487360
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008944792168191301,
      "loss": 0.0868,
      "theoretical_loss": 4.066312958085503,
      "tokens_seen": 377749504
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008943989728775478,
      "loss": 0.0875,
      "theoretical_loss": 4.065996578222502,
      "tokens_seen": 378011648
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008943187289359654,
      "loss": 0.0902,
      "theoretical_loss": 4.065680479071426,
      "tokens_seen": 378273792
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000894238484994383,
      "loss": 0.0869,
      "theoretical_loss": 4.0653646601889335,
      "tokens_seen": 378535936
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008941582410528005,
      "loss": 0.0885,
      "theoretical_loss": 4.065049121132693,
      "tokens_seen": 378798080
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008940779971112181,
      "loss": 0.0877,
      "theoretical_loss": 4.0647338614613755,
      "tokens_seen": 379060224
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0008939977531696357,
      "loss": 0.0859,
      "theoretical_loss": 4.0644188807346495,
      "tokens_seen": 379322368
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008939175092280532,
      "loss": 0.0903,
      "theoretical_loss": 4.064104178513186,
      "tokens_seen": 379584512
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008938372652864709,
      "loss": 0.0884,
      "theoretical_loss": 4.0637897543586465,
      "tokens_seen": 379846656
    },
    {
      "epoch": 0.12,
      "objective/train/advantage_avg": 0.0015821303240954876,
      "objective/train/docs_used": 144698,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.6598097085952759,
      "objective/train/original_loss": 1.6598095893859863,
      "objective/train/theoretical_loss": 4.063475607833687,
      "objective/train/tokens_used": 400568800,
      "objective/train/value_avg": -0.007183074951171875,
      "objective/train/value_loss": 0.00021680242207366973,
      "objective/train/value_max": -0.0001881122589111328,
      "objective/train/value_min": -0.3359375,
      "objective/train/value_reward_corr": 0.5725408863624046,
      "objective/train/value_std": 0.00899505615234375,
      "objective/train/weight_avg": 1.0016818046569824,
      "objective/train/weighted_lm_loss": 1.662369728088379,
      "objective/train/weights_max": 1.1277165412902832,
      "objective/train/weights_min": 0.373598575592041,
      "theoretical_loss": 4.063475607833687,
      "tokens_seen": 380108800
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008937570213448884,
      "loss": 0.0865,
      "theoretical_loss": 4.063475607833687,
      "tokens_seen": 380108800
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008936767774033062,
      "loss": 0.089,
      "theoretical_loss": 4.063161738501951,
      "tokens_seen": 380370944
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008935965334617237,
      "loss": 0.0886,
      "theoretical_loss": 4.0628481459280685,
      "tokens_seen": 380633088
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008935162895201413,
      "loss": 0.0867,
      "theoretical_loss": 4.062534829677653,
      "tokens_seen": 380895232
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008934360455785589,
      "loss": 0.0888,
      "theoretical_loss": 4.062221789317297,
      "tokens_seen": 381157376
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008933558016369764,
      "loss": 0.087,
      "theoretical_loss": 4.061909024414572,
      "tokens_seen": 381419520
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000893275557695394,
      "loss": 0.0875,
      "theoretical_loss": 4.061596534538021,
      "tokens_seen": 381681664
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008931953137538116,
      "loss": 0.0909,
      "theoretical_loss": 4.061284319257162,
      "tokens_seen": 381943808
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008931150698122292,
      "loss": 0.0901,
      "theoretical_loss": 4.060972378142479,
      "tokens_seen": 382205952
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008930348258706467,
      "loss": 0.0896,
      "theoretical_loss": 4.060660710765423,
      "tokens_seen": 382468096
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008929545819290645,
      "loss": 0.0909,
      "theoretical_loss": 4.060349316698408,
      "tokens_seen": 382730240
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000892874337987482,
      "loss": 0.0874,
      "theoretical_loss": 4.060038195514808,
      "tokens_seen": 382992384
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008927940940458995,
      "loss": 0.0866,
      "theoretical_loss": 4.059727346788955,
      "tokens_seen": 383254528
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008927138501043172,
      "loss": 0.087,
      "theoretical_loss": 4.059416770096134,
      "tokens_seen": 383516672
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008926336061627347,
      "loss": 0.0885,
      "theoretical_loss": 4.059106465012583,
      "tokens_seen": 383778816
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008925533622211523,
      "loss": 0.0878,
      "theoretical_loss": 4.058796431115489,
      "tokens_seen": 384040960
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008924731182795699,
      "loss": 0.0894,
      "theoretical_loss": 4.058486667982986,
      "tokens_seen": 384303104
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008923928743379875,
      "loss": 0.0873,
      "theoretical_loss": 4.058177175194148,
      "tokens_seen": 384565248
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008923126303964051,
      "loss": 0.0883,
      "theoretical_loss": 4.057867952328994,
      "tokens_seen": 384827392
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008922323864548226,
      "loss": 0.0897,
      "theoretical_loss": 4.057558998968479,
      "tokens_seen": 385089536
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008921521425132403,
      "loss": 0.0903,
      "theoretical_loss": 4.0572503146944925,
      "tokens_seen": 385351680
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008920718985716579,
      "loss": 0.0884,
      "theoretical_loss": 4.056941899089858,
      "tokens_seen": 385613824
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008919916546300755,
      "loss": 0.0881,
      "theoretical_loss": 4.056633751738328,
      "tokens_seen": 385875968
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000891911410688493,
      "loss": 0.0874,
      "theoretical_loss": 4.0563258722245825,
      "tokens_seen": 386138112
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008918311667469107,
      "loss": 0.0868,
      "theoretical_loss": 4.056018260134226,
      "tokens_seen": 386400256
    },
    {
      "epoch": 0.12,
      "objective/train/advantage_avg": 0.001992823090404272,
      "objective/train/docs_used": 146981,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.8857860565185547,
      "objective/train/original_loss": 1.8857860565185547,
      "objective/train/theoretical_loss": 4.055710915053783,
      "objective/train/tokens_used": 407122400,
      "objective/train/value_avg": -0.007904052734375,
      "objective/train/value_loss": 0.00022203872504178435,
      "objective/train/value_max": -0.00026726722717285156,
      "objective/train/value_min": -0.2406005859375,
      "objective/train/value_reward_corr": 0.5523747667992205,
      "objective/train/value_std": 0.01018524169921875,
      "objective/train/weight_avg": 1.0020960569381714,
      "objective/train/weighted_lm_loss": 1.8894469738006592,
      "objective/train/weights_max": 1.1491678953170776,
      "objective/train/weights_min": 0.3828728199005127,
      "theoretical_loss": 4.055710915053783,
      "tokens_seen": 386662400
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008917509228053282,
      "loss": 0.088,
      "theoretical_loss": 4.055710915053783,
      "tokens_seen": 386662400
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008916706788637457,
      "loss": 0.0869,
      "theoretical_loss": 4.055403836570701,
      "tokens_seen": 386924544
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008915904349221634,
      "loss": 0.0911,
      "theoretical_loss": 4.05509702427334,
      "tokens_seen": 387186688
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008915101909805809,
      "loss": 0.0864,
      "theoretical_loss": 4.054790477750974,
      "tokens_seen": 387448832
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008914299470389986,
      "loss": 0.0875,
      "theoretical_loss": 4.054484196593791,
      "tokens_seen": 387710976
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008913497030974162,
      "loss": 0.0888,
      "theoretical_loss": 4.054178180392885,
      "tokens_seen": 387973120
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008912694591558338,
      "loss": 0.0832,
      "theoretical_loss": 4.053872428740256,
      "tokens_seen": 388235264
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008911892152142514,
      "loss": 0.0889,
      "theoretical_loss": 4.053566941228809,
      "tokens_seen": 388497408
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008911089712726689,
      "loss": 0.0889,
      "theoretical_loss": 4.053261717452346,
      "tokens_seen": 388759552
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008910287273310865,
      "loss": 0.0861,
      "theoretical_loss": 4.052956757005573,
      "tokens_seen": 389021696
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008909484833895041,
      "loss": 0.086,
      "theoretical_loss": 4.0526520594840845,
      "tokens_seen": 389283840
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008908682394479217,
      "loss": 0.0902,
      "theoretical_loss": 4.052347624484373,
      "tokens_seen": 389545984
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008907879955063392,
      "loss": 0.087,
      "theoretical_loss": 4.052043451603818,
      "tokens_seen": 389808128
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000890707751564757,
      "loss": 0.0893,
      "theoretical_loss": 4.051739540440688,
      "tokens_seen": 390070272
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008906275076231745,
      "loss": 0.086,
      "theoretical_loss": 4.0514358905941386,
      "tokens_seen": 390332416
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008905472636815921,
      "loss": 0.0877,
      "theoretical_loss": 4.051132501664204,
      "tokens_seen": 390594560
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008904670197400097,
      "loss": 0.0895,
      "theoretical_loss": 4.050829373251803,
      "tokens_seen": 390856704
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008903867757984272,
      "loss": 0.0898,
      "theoretical_loss": 4.050526504958727,
      "tokens_seen": 391118848
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008903065318568448,
      "loss": 0.0846,
      "theoretical_loss": 4.050223896387647,
      "tokens_seen": 391380992
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008902262879152624,
      "loss": 0.0873,
      "theoretical_loss": 4.0499215471421035,
      "tokens_seen": 391643136
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00089014604397368,
      "loss": 0.0884,
      "theoretical_loss": 4.049619456826508,
      "tokens_seen": 391905280
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008900658000320975,
      "loss": 0.0869,
      "theoretical_loss": 4.0493176250461405,
      "tokens_seen": 392167424
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008899855560905153,
      "loss": 0.0874,
      "theoretical_loss": 4.049016051407145,
      "tokens_seen": 392429568
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008899053121489328,
      "loss": 0.0863,
      "theoretical_loss": 4.048714735516527,
      "tokens_seen": 392691712
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008898250682073504,
      "loss": 0.0904,
      "theoretical_loss": 4.048413676982155,
      "tokens_seen": 392953856
    },
    {
      "epoch": 0.12,
      "objective/train/advantage_avg": 0.0020599644631147385,
      "objective/train/docs_used": 149327,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.7652794122695923,
      "objective/train/original_loss": 1.7652791738510132,
      "objective/train/theoretical_loss": 4.048112875412752,
      "objective/train/tokens_used": 413676000,
      "objective/train/value_avg": -0.0103302001953125,
      "objective/train/value_loss": 0.00044312665704637766,
      "objective/train/value_max": -0.00033283233642578125,
      "objective/train/value_min": -0.95361328125,
      "objective/train/value_reward_corr": 0.6481849685057344,
      "objective/train/value_std": 0.0183258056640625,
      "objective/train/weight_avg": 1.002259373664856,
      "objective/train/weighted_lm_loss": 1.7693637609481812,
      "objective/train/weights_max": 1.8321107625961304,
      "objective/train/weights_min": 0.36900386214256287,
      "theoretical_loss": 4.048112875412752,
      "tokens_seen": 393216000
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000889744824265768,
      "loss": 0.0873,
      "theoretical_loss": 4.048112875412752,
      "tokens_seen": 393216000
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008896645803241855,
      "loss": 0.0889,
      "theoretical_loss": 4.0478123304179,
      "tokens_seen": 393478144
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008895843363826032,
      "loss": 0.0903,
      "theoretical_loss": 4.047512041608029,
      "tokens_seen": 393740288
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008895040924410207,
      "loss": 0.0887,
      "theoretical_loss": 4.047212008594424,
      "tokens_seen": 394002432
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008894238484994383,
      "loss": 0.0872,
      "theoretical_loss": 4.046912230989214,
      "tokens_seen": 394264576
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008893436045578559,
      "loss": 0.0895,
      "theoretical_loss": 4.0466127084053785,
      "tokens_seen": 394526720
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008892633606162734,
      "loss": 0.0879,
      "theoretical_loss": 4.046313440456733,
      "tokens_seen": 394788864
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008891831166746911,
      "loss": 0.0888,
      "theoretical_loss": 4.0460144267579405,
      "tokens_seen": 395051008
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008891028727331087,
      "loss": 0.0898,
      "theoretical_loss": 4.045715666924499,
      "tokens_seen": 395313152
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008890226287915263,
      "loss": 0.0879,
      "theoretical_loss": 4.045417160572743,
      "tokens_seen": 395575296
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008889423848499438,
      "loss": 0.0899,
      "theoretical_loss": 4.045118907319839,
      "tokens_seen": 395837440
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008888621409083615,
      "loss": 0.0865,
      "theoretical_loss": 4.04482090678379,
      "tokens_seen": 396099584
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000888781896966779,
      "loss": 0.0908,
      "theoretical_loss": 4.044523158583421,
      "tokens_seen": 396361728
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008887016530251965,
      "loss": 0.0891,
      "theoretical_loss": 4.044225662338388,
      "tokens_seen": 396623872
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008886214090836142,
      "loss": 0.0882,
      "theoretical_loss": 4.04392841766917,
      "tokens_seen": 396886016
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008885411651420317,
      "loss": 0.0861,
      "theoretical_loss": 4.043631424197067,
      "tokens_seen": 397148160
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008884609212004495,
      "loss": 0.0896,
      "theoretical_loss": 4.0433346815442,
      "tokens_seen": 397410304
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000888380677258867,
      "loss": 0.0912,
      "theoretical_loss": 4.043038189333508,
      "tokens_seen": 397672448
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008883004333172846,
      "loss": 0.0917,
      "theoretical_loss": 4.042741947188741,
      "tokens_seen": 397934592
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008882201893757022,
      "loss": 0.0874,
      "theoretical_loss": 4.042445954734466,
      "tokens_seen": 398196736
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008881399454341197,
      "loss": 0.0925,
      "theoretical_loss": 4.0421502115960575,
      "tokens_seen": 398458880
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008880597014925373,
      "loss": 0.0852,
      "theoretical_loss": 4.0418547173997,
      "tokens_seen": 398721024
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008879794575509549,
      "loss": 0.0883,
      "theoretical_loss": 4.041559471772382,
      "tokens_seen": 398983168
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008878992136093725,
      "loss": 0.0904,
      "theoretical_loss": 4.041264474341896,
      "tokens_seen": 399245312
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00088781896966779,
      "loss": 0.0888,
      "theoretical_loss": 4.040969724736838,
      "tokens_seen": 399507456
    },
    {
      "epoch": 0.12,
      "objective/train/advantage_avg": 0.00207437202334404,
      "objective/train/docs_used": 151634,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.6508839130401611,
      "objective/train/original_loss": 1.6508839130401611,
      "objective/train/theoretical_loss": 4.040675222586599,
      "objective/train/tokens_used": 420229600,
      "objective/train/value_avg": -0.00807952880859375,
      "objective/train/value_loss": 0.00020788678375538439,
      "objective/train/value_max": -0.00013136863708496094,
      "objective/train/value_min": -0.88134765625,
      "objective/train/value_reward_corr": 0.729374600933435,
      "objective/train/value_std": 0.015289306640625,
      "objective/train/weight_avg": 1.0021743774414062,
      "objective/train/weighted_lm_loss": 1.6536670923233032,
      "objective/train/weights_max": 1.6075432300567627,
      "objective/train/weights_min": 0.5353213548660278,
      "theoretical_loss": 4.040675222586599,
      "tokens_seen": 399769600
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008877387257262078,
      "loss": 0.086,
      "theoretical_loss": 4.040675222586599,
      "tokens_seen": 399769600
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008876584817846253,
      "loss": 0.0851,
      "theoretical_loss": 4.04038096752137,
      "tokens_seen": 400031744
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008875782378430429,
      "loss": 0.0882,
      "theoretical_loss": 4.040086959172136,
      "tokens_seen": 400293888
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008874979939014605,
      "loss": 0.0843,
      "theoretical_loss": 4.039793197170672,
      "tokens_seen": 400556032
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000887417749959878,
      "loss": 0.0894,
      "theoretical_loss": 4.039499681149547,
      "tokens_seen": 400818176
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008873375060182957,
      "loss": 0.0886,
      "theoretical_loss": 4.039206410742114,
      "tokens_seen": 401080320
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008872572620767132,
      "loss": 0.0864,
      "theoretical_loss": 4.038913385582515,
      "tokens_seen": 401342464
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008871770181351308,
      "loss": 0.0867,
      "theoretical_loss": 4.038620605305673,
      "tokens_seen": 401604608
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008870967741935484,
      "loss": 0.0843,
      "theoretical_loss": 4.038328069547293,
      "tokens_seen": 401866752
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000887016530251966,
      "loss": 0.0877,
      "theoretical_loss": 4.03803577794386,
      "tokens_seen": 402128896
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008869362863103836,
      "loss": 0.0886,
      "theoretical_loss": 4.037743730132635,
      "tokens_seen": 402391040
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008868560423688012,
      "loss": 0.085,
      "theoretical_loss": 4.037451925751654,
      "tokens_seen": 402653184
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008867757984272188,
      "loss": 0.0875,
      "theoretical_loss": 4.0371603644397265,
      "tokens_seen": 402915328
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008866955544856363,
      "loss": 0.0896,
      "theoretical_loss": 4.03686904583643,
      "tokens_seen": 403177472
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000886615310544054,
      "loss": 0.087,
      "theoretical_loss": 4.036577969582114,
      "tokens_seen": 403439616
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008865350666024715,
      "loss": 0.0857,
      "theoretical_loss": 4.03628713531789,
      "tokens_seen": 403701760
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008864548226608891,
      "loss": 0.089,
      "theoretical_loss": 4.035996542685638,
      "tokens_seen": 403963904
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008863745787193067,
      "loss": 0.0859,
      "theoretical_loss": 4.0357061913279955,
      "tokens_seen": 404226048
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008862943347777242,
      "loss": 0.0867,
      "theoretical_loss": 4.035416080888364,
      "tokens_seen": 404488192
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008862140908361419,
      "loss": 0.0893,
      "theoretical_loss": 4.035126211010899,
      "tokens_seen": 404750336
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008861338468945595,
      "loss": 0.0851,
      "theoretical_loss": 4.034836581340515,
      "tokens_seen": 405012480
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008860536029529771,
      "loss": 0.0866,
      "theoretical_loss": 4.034547191522877,
      "tokens_seen": 405274624
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008859733590113947,
      "loss": 0.0928,
      "theoretical_loss": 4.034258041204404,
      "tokens_seen": 405536768
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008858931150698123,
      "loss": 0.0869,
      "theoretical_loss": 4.033969130032263,
      "tokens_seen": 405798912
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008858128711282298,
      "loss": 0.0854,
      "theoretical_loss": 4.033680457654368,
      "tokens_seen": 406061056
    },
    {
      "epoch": 0.12,
      "objective/train/advantage_avg": 0.001684014918282628,
      "objective/train/docs_used": 154148,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.6580567359924316,
      "objective/train/original_loss": 1.6580567359924316,
      "objective/train/theoretical_loss": 4.0333920237193785,
      "objective/train/tokens_used": 426783200,
      "objective/train/value_avg": -0.00681304931640625,
      "objective/train/value_loss": 0.00015735704801045358,
      "objective/train/value_max": -0.0002269744873046875,
      "objective/train/value_min": -0.309326171875,
      "objective/train/value_reward_corr": 0.6036299621134384,
      "objective/train/value_std": 0.0090484619140625,
      "objective/train/weight_avg": 1.0017573833465576,
      "objective/train/weighted_lm_loss": 1.6613595485687256,
      "objective/train/weights_max": 1.1686903238296509,
      "objective/train/weights_min": 0.36919885873794556,
      "theoretical_loss": 4.0333920237193785,
      "tokens_seen": 406323200
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008857326271866474,
      "loss": 0.0855,
      "theoretical_loss": 4.0333920237193785,
      "tokens_seen": 406323200
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000885652383245065,
      "loss": 0.0818,
      "theoretical_loss": 4.0331038278766975,
      "tokens_seen": 406585344
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008855721393034825,
      "loss": 0.0891,
      "theoretical_loss": 4.032815869776471,
      "tokens_seen": 406847488
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008854918953619003,
      "loss": 0.0883,
      "theoretical_loss": 4.032528149069579,
      "tokens_seen": 407109632
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008854116514203178,
      "loss": 0.0888,
      "theoretical_loss": 4.0322406654076435,
      "tokens_seen": 407371776
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008853314074787354,
      "loss": 0.0886,
      "theoretical_loss": 4.0319534184430195,
      "tokens_seen": 407633920
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000885251163537153,
      "loss": 0.0864,
      "theoretical_loss": 4.031666407828795,
      "tokens_seen": 407896064
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008851709195955705,
      "loss": 0.0863,
      "theoretical_loss": 4.03137963321879,
      "tokens_seen": 408158208
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008850906756539881,
      "loss": 0.0886,
      "theoretical_loss": 4.0310930942675505,
      "tokens_seen": 408420352
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008850104317124057,
      "loss": 0.0875,
      "theoretical_loss": 4.030806790630353,
      "tokens_seen": 408682496
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008849301877708233,
      "loss": 0.0851,
      "theoretical_loss": 4.030520721963199,
      "tokens_seen": 408944640
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008848499438292408,
      "loss": 0.0853,
      "theoretical_loss": 4.030234887922808,
      "tokens_seen": 409206784
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008847696998876586,
      "loss": 0.0874,
      "theoretical_loss": 4.029949288166627,
      "tokens_seen": 409468928
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008846894559460761,
      "loss": 0.0822,
      "theoretical_loss": 4.0296639223528175,
      "tokens_seen": 409731072
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008846092120044937,
      "loss": 0.0858,
      "theoretical_loss": 4.029378790140261,
      "tokens_seen": 409993216
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008845289680629113,
      "loss": 0.0862,
      "theoretical_loss": 4.029093891188552,
      "tokens_seen": 410255360
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008844487241213288,
      "loss": 0.0876,
      "theoretical_loss": 4.028809225158,
      "tokens_seen": 410517504
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008843684801797465,
      "loss": 0.0856,
      "theoretical_loss": 4.028524791709621,
      "tokens_seen": 410779648
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000884288236238164,
      "loss": 0.0863,
      "theoretical_loss": 4.028240590505148,
      "tokens_seen": 411041792
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008842079922965816,
      "loss": 0.0859,
      "theoretical_loss": 4.027956621207015,
      "tokens_seen": 411303936
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008841277483549992,
      "loss": 0.0858,
      "theoretical_loss": 4.027672883478364,
      "tokens_seen": 411566080
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008840475044134169,
      "loss": 0.0892,
      "theoretical_loss": 4.027389376983041,
      "tokens_seen": 411828224
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0008839672604718344,
      "loss": 0.0893,
      "theoretical_loss": 4.02710610138559,
      "tokens_seen": 412090368
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000883887016530252,
      "loss": 0.0863,
      "theoretical_loss": 4.02682305635126,
      "tokens_seen": 412352512
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008838067725886696,
      "loss": 0.0875,
      "theoretical_loss": 4.026540241545994,
      "tokens_seen": 412614656
    },
    {
      "epoch": 0.13,
      "objective/train/advantage_avg": 0.00014230117085389793,
      "objective/train/docs_used": 156351,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.629351019859314,
      "objective/train/original_loss": 1.6293511390686035,
      "objective/train/theoretical_loss": 4.026257656636431,
      "objective/train/tokens_used": 433336800,
      "objective/train/value_avg": -0.00789642333984375,
      "objective/train/value_loss": 0.0005366242257878184,
      "objective/train/value_max": -0.00022172927856445312,
      "objective/train/value_min": -0.9091796875,
      "objective/train/value_reward_corr": 0.6416809694837178,
      "objective/train/value_std": 0.017822265625,
      "objective/train/weight_avg": 1.0003734827041626,
      "objective/train/weighted_lm_loss": 1.628688931465149,
      "objective/train/weights_max": 1.8231868743896484,
      "objective/train/weights_min": 0.36822012066841125,
      "theoretical_loss": 4.026257656636431,
      "tokens_seen": 412876800
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008837265286470871,
      "loss": 0.0856,
      "theoretical_loss": 4.026257656636431,
      "tokens_seen": 412876800
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008836462847055048,
      "loss": 0.0869,
      "theoretical_loss": 4.025975301289906,
      "tokens_seen": 413138944
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008835660407639223,
      "loss": 0.0864,
      "theoretical_loss": 4.025693175174443,
      "tokens_seen": 413401088
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00088348579682234,
      "loss": 0.0836,
      "theoretical_loss": 4.02541127795876,
      "tokens_seen": 413663232
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008834055528807575,
      "loss": 0.0861,
      "theoretical_loss": 4.02512960931226,
      "tokens_seen": 413925376
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.000883325308939175,
      "loss": 0.0831,
      "theoretical_loss": 4.024848168905035,
      "tokens_seen": 414187520
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008832450649975928,
      "loss": 0.0864,
      "theoretical_loss": 4.02456695640786,
      "tokens_seen": 414449664
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008831648210560103,
      "loss": 0.0855,
      "theoretical_loss": 4.0242859714921915,
      "tokens_seen": 414711808
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008830845771144279,
      "loss": 0.0886,
      "theoretical_loss": 4.024005213830171,
      "tokens_seen": 414973952
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008830043331728455,
      "loss": 0.0833,
      "theoretical_loss": 4.023724683094615,
      "tokens_seen": 415236096
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008829240892312631,
      "loss": 0.0833,
      "theoretical_loss": 4.023444378959019,
      "tokens_seen": 415498240
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008828438452896806,
      "loss": 0.085,
      "theoretical_loss": 4.023164301097555,
      "tokens_seen": 415760384
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008827636013480982,
      "loss": 0.0854,
      "theoretical_loss": 4.0228844491850655,
      "tokens_seen": 416022528
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008826833574065158,
      "loss": 0.0873,
      "theoretical_loss": 4.022604822897068,
      "tokens_seen": 416284672
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008826031134649333,
      "loss": 0.0872,
      "theoretical_loss": 4.0223254219097475,
      "tokens_seen": 416546816
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008825228695233511,
      "loss": 0.0875,
      "theoretical_loss": 4.022046245899958,
      "tokens_seen": 416808960
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008824426255817686,
      "loss": 0.0856,
      "theoretical_loss": 4.021767294545221,
      "tokens_seen": 417071104
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008823623816401862,
      "loss": 0.0872,
      "theoretical_loss": 4.021488567523721,
      "tokens_seen": 417333248
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008822821376986038,
      "loss": 0.087,
      "theoretical_loss": 4.021210064514305,
      "tokens_seen": 417595392
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008822018937570213,
      "loss": 0.0833,
      "theoretical_loss": 4.020931785196484,
      "tokens_seen": 417857536
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.000882121649815439,
      "loss": 0.0849,
      "theoretical_loss": 4.020653729250424,
      "tokens_seen": 418119680
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008820414058738565,
      "loss": 0.0847,
      "theoretical_loss": 4.020375896356951,
      "tokens_seen": 418381824
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008819611619322741,
      "loss": 0.0905,
      "theoretical_loss": 4.0200982861975465,
      "tokens_seen": 418643968
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008818809179906917,
      "loss": 0.0879,
      "theoretical_loss": 4.019820898454345,
      "tokens_seen": 418906112
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008818006740491094,
      "loss": 0.0838,
      "theoretical_loss": 4.019543732810134,
      "tokens_seen": 419168256
    },
    {
      "epoch": 0.13,
      "objective/train/advantage_avg": 0.0006775321089662611,
      "objective/train/docs_used": 158751,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.7296289205551147,
      "objective/train/original_loss": 1.7296291589736938,
      "objective/train/theoretical_loss": 4.019266788948352,
      "objective/train/tokens_used": 439890400,
      "objective/train/value_avg": -0.00728607177734375,
      "objective/train/value_loss": 0.0003692489117383957,
      "objective/train/value_max": -0.0002613067626953125,
      "objective/train/value_min": -0.51171875,
      "objective/train/value_reward_corr": 0.557088638783727,
      "objective/train/value_std": 0.01107025146484375,
      "objective/train/weight_avg": 1.0008379220962524,
      "objective/train/weighted_lm_loss": 1.7313810586929321,
      "objective/train/weights_max": 1.540909767150879,
      "objective/train/weights_min": 0.373028963804245,
      "theoretical_loss": 4.019266788948352,
      "tokens_seen": 419430400
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008817204301075269,
      "loss": 0.0871,
      "theoretical_loss": 4.019266788948352,
      "tokens_seen": 419430400
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008816401861659445,
      "loss": 0.0896,
      "theoretical_loss": 4.0189900665530836,
      "tokens_seen": 419692544
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008815599422243621,
      "loss": 0.0877,
      "theoretical_loss": 4.0187135653090635,
      "tokens_seen": 419954688
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008814796982827796,
      "loss": 0.0827,
      "theoretical_loss": 4.018437284901671,
      "tokens_seen": 420216832
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008813994543411973,
      "loss": 0.0896,
      "theoretical_loss": 4.018161225016926,
      "tokens_seen": 420478976
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008813192103996148,
      "loss": 0.0875,
      "theoretical_loss": 4.0178853853414935,
      "tokens_seen": 420741120
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008812389664580324,
      "loss": 0.087,
      "theoretical_loss": 4.017609765562678,
      "tokens_seen": 421003264
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00088115872251645,
      "loss": 0.0871,
      "theoretical_loss": 4.017334365368422,
      "tokens_seen": 421265408
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008810784785748677,
      "loss": 0.0888,
      "theoretical_loss": 4.017059184447303,
      "tokens_seen": 421527552
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008809982346332853,
      "loss": 0.0875,
      "theoretical_loss": 4.0167842224885355,
      "tokens_seen": 421789696
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008809179906917028,
      "loss": 0.0827,
      "theoretical_loss": 4.016509479181968,
      "tokens_seen": 422051840
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008808377467501204,
      "loss": 0.0861,
      "theoretical_loss": 4.016234954218078,
      "tokens_seen": 422313984
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.000880757502808538,
      "loss": 0.0855,
      "theoretical_loss": 4.015960647287975,
      "tokens_seen": 422576128
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008806772588669556,
      "loss": 0.0874,
      "theoretical_loss": 4.015686558083396,
      "tokens_seen": 422838272
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008805970149253731,
      "loss": 0.0878,
      "theoretical_loss": 4.015412686296704,
      "tokens_seen": 423100416
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008805167709837908,
      "loss": 0.0869,
      "theoretical_loss": 4.0151390316208895,
      "tokens_seen": 423362560
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008804365270422083,
      "loss": 0.0872,
      "theoretical_loss": 4.014865593749563,
      "tokens_seen": 423624704
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008803562831006258,
      "loss": 0.0852,
      "theoretical_loss": 4.014592372376958,
      "tokens_seen": 423886848
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008802760391590436,
      "loss": 0.0871,
      "theoretical_loss": 4.014319367197928,
      "tokens_seen": 424148992
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008801957952174611,
      "loss": 0.0905,
      "theoretical_loss": 4.014046577907946,
      "tokens_seen": 424411136
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008801155512758787,
      "loss": 0.0886,
      "theoretical_loss": 4.013774004203099,
      "tokens_seen": 424673280
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008800353073342963,
      "loss": 0.0873,
      "theoretical_loss": 4.013501645780092,
      "tokens_seen": 424935424
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008799550633927139,
      "loss": 0.0835,
      "theoretical_loss": 4.013229502336242,
      "tokens_seen": 425197568
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008798748194511314,
      "loss": 0.0863,
      "theoretical_loss": 4.0129575735694765,
      "tokens_seen": 425459712
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.000879794575509549,
      "loss": 0.0891,
      "theoretical_loss": 4.012685859178337,
      "tokens_seen": 425721856
    },
    {
      "epoch": 0.13,
      "objective/train/advantage_avg": 0.001993941143155098,
      "objective/train/docs_used": 161088,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.6738786697387695,
      "objective/train/original_loss": 1.6738786697387695,
      "objective/train/theoretical_loss": 4.012414358861969,
      "objective/train/tokens_used": 446444000,
      "objective/train/value_avg": -0.0080413818359375,
      "objective/train/value_loss": 0.00036461447598412633,
      "objective/train/value_max": -0.00036263465881347656,
      "objective/train/value_min": -0.89501953125,
      "objective/train/value_reward_corr": 0.7342816416076462,
      "objective/train/value_std": 0.018157958984375,
      "objective/train/weight_avg": 1.0021560192108154,
      "objective/train/weighted_lm_loss": 1.676504373550415,
      "objective/train/weights_max": 1.6657140254974365,
      "objective/train/weights_min": 0.3717136085033417,
      "theoretical_loss": 4.012414358861969,
      "tokens_seen": 425984000
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008797143315679666,
      "loss": 0.0842,
      "theoretical_loss": 4.012414358861969,
      "tokens_seen": 425984000
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008796340876263842,
      "loss": 0.0868,
      "theoretical_loss": 4.01214307232013,
      "tokens_seen": 426246144
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008795538436848019,
      "loss": 0.0863,
      "theoretical_loss": 4.011871999253178,
      "tokens_seen": 426508288
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008794735997432194,
      "loss": 0.0883,
      "theoretical_loss": 4.011601139362078,
      "tokens_seen": 426770432
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008793933558016371,
      "loss": 0.0852,
      "theoretical_loss": 4.011330492348397,
      "tokens_seen": 427032576
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008793131118600546,
      "loss": 0.0891,
      "theoretical_loss": 4.0110600579143,
      "tokens_seen": 427294720
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008792328679184721,
      "loss": 0.0866,
      "theoretical_loss": 4.010789835762555,
      "tokens_seen": 427556864
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008791526239768898,
      "loss": 0.0888,
      "theoretical_loss": 4.010519825596525,
      "tokens_seen": 427819008
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008790723800353073,
      "loss": 0.086,
      "theoretical_loss": 4.010250027120169,
      "tokens_seen": 428081152
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008789921360937249,
      "loss": 0.0852,
      "theoretical_loss": 4.009980440038043,
      "tokens_seen": 428343296
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008789118921521425,
      "loss": 0.0889,
      "theoretical_loss": 4.009711064055291,
      "tokens_seen": 428605440
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008788316482105602,
      "loss": 0.0852,
      "theoretical_loss": 4.009441898877652,
      "tokens_seen": 428867584
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008787514042689777,
      "loss": 0.0852,
      "theoretical_loss": 4.009172944211455,
      "tokens_seen": 429129728
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008786711603273953,
      "loss": 0.087,
      "theoretical_loss": 4.008904199763615,
      "tokens_seen": 429391872
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008785909163858129,
      "loss": 0.0851,
      "theoretical_loss": 4.008635665241635,
      "tokens_seen": 429654016
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008785106724442304,
      "loss": 0.0883,
      "theoretical_loss": 4.008367340353602,
      "tokens_seen": 429916160
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008784304285026481,
      "loss": 0.0863,
      "theoretical_loss": 4.008099224808188,
      "tokens_seen": 430178304
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008783501845610656,
      "loss": 0.0837,
      "theoretical_loss": 4.007831318314645,
      "tokens_seen": 430440448
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008782699406194833,
      "loss": 0.0846,
      "theoretical_loss": 4.00756362058281,
      "tokens_seen": 430702592
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008781896966779008,
      "loss": 0.0839,
      "theoretical_loss": 4.007296131323094,
      "tokens_seen": 430964736
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008781094527363185,
      "loss": 0.0866,
      "theoretical_loss": 4.007028850246487,
      "tokens_seen": 431226880
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008780292087947361,
      "loss": 0.0861,
      "theoretical_loss": 4.006761777064557,
      "tokens_seen": 431489024
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008779489648531536,
      "loss": 0.0854,
      "theoretical_loss": 4.006494911489444,
      "tokens_seen": 431751168
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008778687209115712,
      "loss": 0.086,
      "theoretical_loss": 4.006228253233864,
      "tokens_seen": 432013312
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008777884769699888,
      "loss": 0.088,
      "theoretical_loss": 4.0059618020111,
      "tokens_seen": 432275456
    },
    {
      "epoch": 0.13,
      "objective/train/advantage_avg": 0.0015352979535236955,
      "objective/train/docs_used": 163281,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.734284520149231,
      "objective/train/original_loss": 1.7342844009399414,
      "objective/train/theoretical_loss": 4.00569555753501,
      "objective/train/tokens_used": 452997600,
      "objective/train/value_avg": -0.006992340087890625,
      "objective/train/value_loss": 0.00020126513845752925,
      "objective/train/value_max": -0.0002913475036621094,
      "objective/train/value_min": -0.54833984375,
      "objective/train/value_reward_corr": 0.5731349995282375,
      "objective/train/value_std": 0.01129913330078125,
      "objective/train/weight_avg": 1.0016310214996338,
      "objective/train/weighted_lm_loss": 1.7371350526809692,
      "objective/train/weights_max": 1.6854356527328491,
      "objective/train/weights_min": 0.3728724420070648,
      "theoretical_loss": 4.00569555753501,
      "tokens_seen": 432537600
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008777082330284064,
      "loss": 0.0868,
      "theoretical_loss": 4.00569555753501,
      "tokens_seen": 432537600
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008776279890868239,
      "loss": 0.0854,
      "theoretical_loss": 4.0054295195200185,
      "tokens_seen": 432799744
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008775477451452416,
      "loss": 0.085,
      "theoretical_loss": 4.005163687681116,
      "tokens_seen": 433061888
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008774675012036591,
      "loss": 0.085,
      "theoretical_loss": 4.00489806173386,
      "tokens_seen": 433324032
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008773872572620766,
      "loss": 0.0889,
      "theoretical_loss": 4.004632641394372,
      "tokens_seen": 433586176
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008773070133204944,
      "loss": 0.0868,
      "theoretical_loss": 4.0043674263793365,
      "tokens_seen": 433848320
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008772267693789119,
      "loss": 0.0842,
      "theoretical_loss": 4.004102416405998,
      "tokens_seen": 434110464
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008771465254373296,
      "loss": 0.0853,
      "theoretical_loss": 4.0038376111921625,
      "tokens_seen": 434372608
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008770662814957471,
      "loss": 0.0872,
      "theoretical_loss": 4.0035730104561935,
      "tokens_seen": 434634752
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008769860375541647,
      "loss": 0.0888,
      "theoretical_loss": 4.003308613917012,
      "tokens_seen": 434896896
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008769057936125823,
      "loss": 0.0873,
      "theoretical_loss": 4.003044421294094,
      "tokens_seen": 435159040
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008768255496709998,
      "loss": 0.0852,
      "theoretical_loss": 4.002780432307468,
      "tokens_seen": 435421184
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008767453057294174,
      "loss": 0.0853,
      "theoretical_loss": 4.0025166466777184,
      "tokens_seen": 435683328
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.000876665061787835,
      "loss": 0.0868,
      "theoretical_loss": 4.00225306412598,
      "tokens_seen": 435945472
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008765848178462527,
      "loss": 0.0869,
      "theoretical_loss": 4.001989684373934,
      "tokens_seen": 436207616
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008765045739046702,
      "loss": 0.0863,
      "theoretical_loss": 4.0017265071438155,
      "tokens_seen": 436469760
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008764243299630879,
      "loss": 0.0863,
      "theoretical_loss": 4.001463532158402,
      "tokens_seen": 436731904
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008763440860215054,
      "loss": 0.0864,
      "theoretical_loss": 4.001200759141019,
      "tokens_seen": 436994048
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008762638420799229,
      "loss": 0.0857,
      "theoretical_loss": 4.000938187815535,
      "tokens_seen": 437256192
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008761835981383406,
      "loss": 0.0852,
      "theoretical_loss": 4.000675817906362,
      "tokens_seen": 437518336
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008761033541967581,
      "loss": 0.0852,
      "theoretical_loss": 4.000413649138453,
      "tokens_seen": 437780480
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008760231102551757,
      "loss": 0.0871,
      "theoretical_loss": 4.000151681237301,
      "tokens_seen": 438042624
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008759428663135933,
      "loss": 0.0851,
      "theoretical_loss": 3.9998899139289392,
      "tokens_seen": 438304768
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.000875862622372011,
      "loss": 0.0855,
      "theoretical_loss": 3.999628346939934,
      "tokens_seen": 438566912
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008757823784304286,
      "loss": 0.088,
      "theoretical_loss": 3.9993669799973928,
      "tokens_seen": 438829056
    },
    {
      "epoch": 0.13,
      "objective/train/advantage_avg": 0.0022315175738185644,
      "objective/train/docs_used": 165818,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.725619912147522,
      "objective/train/original_loss": 1.725620150566101,
      "objective/train/theoretical_loss": 3.9991058128289536,
      "objective/train/tokens_used": 459551200,
      "objective/train/value_avg": -0.009033203125,
      "objective/train/value_loss": 0.0003572371497284621,
      "objective/train/value_max": -0.0001780986785888672,
      "objective/train/value_min": -0.8076171875,
      "objective/train/value_reward_corr": 0.7491131113466319,
      "objective/train/value_std": 0.020477294921875,
      "objective/train/weight_avg": 1.002398133277893,
      "objective/train/weighted_lm_loss": 1.7294703722000122,
      "objective/train/weights_max": 1.4387959241867065,
      "objective/train/weights_min": 0.41180431842803955,
      "theoretical_loss": 3.9991058128289536,
      "tokens_seen": 439091200
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008757021344888461,
      "loss": 0.0852,
      "theoretical_loss": 3.9991058128289536,
      "tokens_seen": 439091200
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008756218905472637,
      "loss": 0.0893,
      "theoretical_loss": 3.998844845162789,
      "tokens_seen": 439353344
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008755416466056813,
      "loss": 0.0869,
      "theoretical_loss": 3.998584076727604,
      "tokens_seen": 439615488
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008754614026640989,
      "loss": 0.0882,
      "theoretical_loss": 3.998323507252633,
      "tokens_seen": 439877632
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008753811587225164,
      "loss": 0.0848,
      "theoretical_loss": 3.998063136467639,
      "tokens_seen": 440139776
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008753009147809341,
      "loss": 0.0875,
      "theoretical_loss": 3.9978029641029154,
      "tokens_seen": 440401920
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008752206708393516,
      "loss": 0.0858,
      "theoretical_loss": 3.9975429898892783,
      "tokens_seen": 440664064
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008751404268977691,
      "loss": 0.0866,
      "theoretical_loss": 3.9972832135580707,
      "tokens_seen": 440926208
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008750601829561869,
      "loss": 0.0863,
      "theoretical_loss": 3.9970236348411605,
      "tokens_seen": 441188352
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008749799390146044,
      "loss": 0.0866,
      "theoretical_loss": 3.996764253470935,
      "tokens_seen": 441450496
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.000874899695073022,
      "loss": 0.0821,
      "theoretical_loss": 3.996505069180306,
      "tokens_seen": 441712640
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008748194511314396,
      "loss": 0.0869,
      "theoretical_loss": 3.9962460817027017,
      "tokens_seen": 441974784
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008747392071898572,
      "loss": 0.0861,
      "theoretical_loss": 3.995987290772071,
      "tokens_seen": 442236928
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008746589632482748,
      "loss": 0.0853,
      "theoretical_loss": 3.9957286961228786,
      "tokens_seen": 442499072
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008745787193066924,
      "loss": 0.0835,
      "theoretical_loss": 3.995470297490106,
      "tokens_seen": 442761216
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008744984753651099,
      "loss": 0.0836,
      "theoretical_loss": 3.995212094609249,
      "tokens_seen": 443023360
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008744182314235275,
      "loss": 0.0854,
      "theoretical_loss": 3.994954087216315,
      "tokens_seen": 443285504
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008743379874819452,
      "loss": 0.0881,
      "theoretical_loss": 3.994696275047825,
      "tokens_seen": 443547648
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008742577435403627,
      "loss": 0.0857,
      "theoretical_loss": 3.9944386578408113,
      "tokens_seen": 443809792
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008741774995987804,
      "loss": 0.0838,
      "theoretical_loss": 3.9941812353328126,
      "tokens_seen": 444071936
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008740972556571979,
      "loss": 0.0846,
      "theoretical_loss": 3.993924007261878,
      "tokens_seen": 444334080
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008740170117156155,
      "loss": 0.0878,
      "theoretical_loss": 3.9936669733665617,
      "tokens_seen": 444596224
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008739367677740331,
      "loss": 0.0857,
      "theoretical_loss": 3.9934101333859253,
      "tokens_seen": 444858368
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008738565238324506,
      "loss": 0.0878,
      "theoretical_loss": 3.9931534870595327,
      "tokens_seen": 445120512
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0008737762798908682,
      "loss": 0.0852,
      "theoretical_loss": 3.9928970341274517,
      "tokens_seen": 445382656
    },
    {
      "epoch": 0.13,
      "objective/train/advantage_avg": 0.0004395976720843464,
      "objective/train/docs_used": 168097,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.7950403690338135,
      "objective/train/original_loss": 1.795040249824524,
      "objective/train/theoretical_loss": 3.992640774330251,
      "objective/train/tokens_used": 466104800,
      "objective/train/value_avg": -0.00882720947265625,
      "objective/train/value_loss": 0.0003522172919474542,
      "objective/train/value_max": -0.0001823902130126953,
      "objective/train/value_min": -0.646484375,
      "objective/train/value_reward_corr": 0.7499370479136089,
      "objective/train/value_std": 0.0190887451171875,
      "objective/train/weight_avg": 1.00059974193573,
      "objective/train/weighted_lm_loss": 1.795857548713684,
      "objective/train/weights_max": 1.362145185470581,
      "objective/train/weights_min": 0.3820091187953949,
      "theoretical_loss": 3.992640774330251,
      "tokens_seen": 445644800
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008736960359492858,
      "loss": 0.0857,
      "theoretical_loss": 3.992640774330251,
      "tokens_seen": 445644800
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008736157920077035,
      "loss": 0.0836,
      "theoretical_loss": 3.9923847074090015,
      "tokens_seen": 445906944
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000873535548066121,
      "loss": 0.0831,
      "theoretical_loss": 3.9921288331052702,
      "tokens_seen": 446169088
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008734553041245387,
      "loss": 0.0851,
      "theoretical_loss": 3.991873151161124,
      "tokens_seen": 446431232
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008733750601829562,
      "loss": 0.0882,
      "theoretical_loss": 3.9916176613191263,
      "tokens_seen": 446693376
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008732948162413738,
      "loss": 0.0857,
      "theoretical_loss": 3.9913623633223354,
      "tokens_seen": 446955520
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008732145722997914,
      "loss": 0.0828,
      "theoretical_loss": 3.9911072569143036,
      "tokens_seen": 447217664
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008731343283582089,
      "loss": 0.084,
      "theoretical_loss": 3.9908523418390764,
      "tokens_seen": 447479808
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008730540844166266,
      "loss": 0.084,
      "theoretical_loss": 3.990597617841191,
      "tokens_seen": 447741952
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008729738404750441,
      "loss": 0.0838,
      "theoretical_loss": 3.9903430846656742,
      "tokens_seen": 448004096
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008728935965334618,
      "loss": 0.0868,
      "theoretical_loss": 3.990088742058043,
      "tokens_seen": 448266240
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008728133525918794,
      "loss": 0.0831,
      "theoretical_loss": 3.9898345897643024,
      "tokens_seen": 448528384
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008727331086502969,
      "loss": 0.0829,
      "theoretical_loss": 3.989580627530943,
      "tokens_seen": 448790528
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008726528647087145,
      "loss": 0.0831,
      "theoretical_loss": 3.9893268551049417,
      "tokens_seen": 449052672
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008725726207671321,
      "loss": 0.0845,
      "theoretical_loss": 3.9890732722337594,
      "tokens_seen": 449314816
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008724923768255497,
      "loss": 0.0877,
      "theoretical_loss": 3.988819878665341,
      "tokens_seen": 449576960
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008724121328839672,
      "loss": 0.0846,
      "theoretical_loss": 3.988566674148111,
      "tokens_seen": 449839104
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008723318889423849,
      "loss": 0.0849,
      "theoretical_loss": 3.988313658430978,
      "tokens_seen": 450101248
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008722516450008024,
      "loss": 0.0867,
      "theoretical_loss": 3.9880608312633274,
      "tokens_seen": 450363392
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008721714010592199,
      "loss": 0.0871,
      "theoretical_loss": 3.9878081923950237,
      "tokens_seen": 450625536
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008720911571176377,
      "loss": 0.0848,
      "theoretical_loss": 3.9875557415764087,
      "tokens_seen": 450887680
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008720109131760552,
      "loss": 0.083,
      "theoretical_loss": 3.9873034785582995,
      "tokens_seen": 451149824
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008719306692344729,
      "loss": 0.0824,
      "theoretical_loss": 3.9870514030919884,
      "tokens_seen": 451411968
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008718504252928904,
      "loss": 0.0843,
      "theoretical_loss": 3.986799514929242,
      "tokens_seen": 451674112
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000871770181351308,
      "loss": 0.087,
      "theoretical_loss": 3.9865478138222974,
      "tokens_seen": 451936256
    },
    {
      "epoch": 0.14,
      "objective/train/advantage_avg": 0.0014793577138334513,
      "objective/train/docs_used": 170598,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.7309378385543823,
      "objective/train/original_loss": 1.7309377193450928,
      "objective/train/theoretical_loss": 3.9862962995238647,
      "objective/train/tokens_used": 472658400,
      "objective/train/value_avg": -0.00775146484375,
      "objective/train/value_loss": 0.0004800742899533361,
      "objective/train/value_max": -0.00023412704467773438,
      "objective/train/value_min": -0.78662109375,
      "objective/train/value_reward_corr": 0.5517175996070474,
      "objective/train/value_std": 0.0128021240234375,
      "objective/train/weight_avg": 1.0016835927963257,
      "objective/train/weighted_lm_loss": 1.7332364320755005,
      "objective/train/weights_max": 2.0833756923675537,
      "objective/train/weights_min": 0.368600070476532,
      "theoretical_loss": 3.9862962995238647,
      "tokens_seen": 452198400
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008716899374097256,
      "loss": 0.0856,
      "theoretical_loss": 3.9862962995238647,
      "tokens_seen": 452198400
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008716096934681432,
      "loss": 0.0879,
      "theoretical_loss": 3.9860449717871234,
      "tokens_seen": 452460544
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008715294495265607,
      "loss": 0.0847,
      "theoretical_loss": 3.9857938303657217,
      "tokens_seen": 452722688
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008714492055849783,
      "loss": 0.0879,
      "theoretical_loss": 3.9855428750137754,
      "tokens_seen": 452984832
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000871368961643396,
      "loss": 0.0884,
      "theoretical_loss": 3.9852921054858665,
      "tokens_seen": 453246976
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008712887177018135,
      "loss": 0.0839,
      "theoretical_loss": 3.9850415215370445,
      "tokens_seen": 453509120
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008712084737602312,
      "loss": 0.0814,
      "theoretical_loss": 3.984791122922821,
      "tokens_seen": 453771264
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008711282298186487,
      "loss": 0.0855,
      "theoretical_loss": 3.98454090939917,
      "tokens_seen": 454033408
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008710479858770663,
      "loss": 0.0857,
      "theoretical_loss": 3.984290880722531,
      "tokens_seen": 454295552
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008709677419354839,
      "loss": 0.0831,
      "theoretical_loss": 3.9840410366498,
      "tokens_seen": 454557696
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008708874979939014,
      "loss": 0.0864,
      "theoretical_loss": 3.983791376938336,
      "tokens_seen": 454819840
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008708072540523191,
      "loss": 0.0828,
      "theoretical_loss": 3.983541901345955,
      "tokens_seen": 455081984
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008707270101107366,
      "loss": 0.0868,
      "theoretical_loss": 3.983292609630931,
      "tokens_seen": 455344128
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008706467661691543,
      "loss": 0.0872,
      "theoretical_loss": 3.9830435015519936,
      "tokens_seen": 455606272
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008705665222275719,
      "loss": 0.0855,
      "theoretical_loss": 3.982794576868328,
      "tokens_seen": 455868416
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008704862782859895,
      "loss": 0.0842,
      "theoretical_loss": 3.982545835339573,
      "tokens_seen": 456130560
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000870406034344407,
      "loss": 0.0845,
      "theoretical_loss": 3.982297276725822,
      "tokens_seen": 456392704
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008703257904028246,
      "loss": 0.0834,
      "theoretical_loss": 3.9820489007876176,
      "tokens_seen": 456654848
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008702455464612422,
      "loss": 0.0808,
      "theoretical_loss": 3.981800707285955,
      "tokens_seen": 456916992
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008701653025196597,
      "loss": 0.085,
      "theoretical_loss": 3.9815526959822787,
      "tokens_seen": 457179136
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008700850585780774,
      "loss": 0.0831,
      "theoretical_loss": 3.981304866638481,
      "tokens_seen": 457441280
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008700048146364949,
      "loss": 0.087,
      "theoretical_loss": 3.9810572190169027,
      "tokens_seen": 457703424
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008699245706949126,
      "loss": 0.0862,
      "theoretical_loss": 3.9808097528803295,
      "tokens_seen": 457965568
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008698443267533302,
      "loss": 0.0858,
      "theoretical_loss": 3.9805624679919935,
      "tokens_seen": 458227712
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008697640828117477,
      "loss": 0.0873,
      "theoretical_loss": 3.98031536411557,
      "tokens_seen": 458489856
    },
    {
      "epoch": 0.14,
      "objective/train/advantage_avg": 0.0019264193251729012,
      "objective/train/docs_used": 173009,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.6288082599639893,
      "objective/train/original_loss": 1.6288082599639893,
      "objective/train/theoretical_loss": 3.9800684410151783,
      "objective/train/tokens_used": 479212000,
      "objective/train/value_avg": -0.007114410400390625,
      "objective/train/value_loss": 0.0001740693987812847,
      "objective/train/value_max": -0.0002002716064453125,
      "objective/train/value_min": -0.42919921875,
      "objective/train/value_reward_corr": 0.5126065077385661,
      "objective/train/value_std": 0.00885772705078125,
      "objective/train/weight_avg": 1.0020055770874023,
      "objective/train/weighted_lm_loss": 1.6328537464141846,
      "objective/train/weights_max": 1.4530192613601685,
      "objective/train/weights_min": 0.3697379231452942,
      "theoretical_loss": 3.9800684410151783,
      "tokens_seen": 458752000
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008696838388701653,
      "loss": 0.0843,
      "theoretical_loss": 3.9800684410151783,
      "tokens_seen": 458752000
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008696035949285829,
      "loss": 0.082,
      "theoretical_loss": 3.979821698455379,
      "tokens_seen": 459014144
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008695233509870005,
      "loss": 0.0855,
      "theoretical_loss": 3.9795751362011735,
      "tokens_seen": 459276288
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008694431070454181,
      "loss": 0.0845,
      "theoretical_loss": 3.979328754018004,
      "tokens_seen": 459538432
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008693628631038357,
      "loss": 0.0844,
      "theoretical_loss": 3.979082551671749,
      "tokens_seen": 459800576
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008692826191622532,
      "loss": 0.0871,
      "theoretical_loss": 3.9788365289287286,
      "tokens_seen": 460062720
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008692023752206708,
      "loss": 0.0873,
      "theoretical_loss": 3.9785906855556945,
      "tokens_seen": 460324864
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008691221312790885,
      "loss": 0.0841,
      "theoretical_loss": 3.9783450213198384,
      "tokens_seen": 460587008
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000869041887337506,
      "loss": 0.0832,
      "theoretical_loss": 3.9780995359887843,
      "tokens_seen": 460849152
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008689616433959237,
      "loss": 0.0863,
      "theoretical_loss": 3.9778542293305894,
      "tokens_seen": 461111296
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008688813994543412,
      "loss": 0.086,
      "theoretical_loss": 3.977609101113744,
      "tokens_seen": 461373440
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008688011555127588,
      "loss": 0.0805,
      "theoretical_loss": 3.97736415110717,
      "tokens_seen": 461635584
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008687209115711764,
      "loss": 0.0899,
      "theoretical_loss": 3.977119379080218,
      "tokens_seen": 461897728
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000868640667629594,
      "loss": 0.088,
      "theoretical_loss": 3.9768747848026695,
      "tokens_seen": 462159872
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008685604236880115,
      "loss": 0.0832,
      "theoretical_loss": 3.9766303680447335,
      "tokens_seen": 462422016
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008684801797464291,
      "loss": 0.083,
      "theoretical_loss": 3.9763861285770457,
      "tokens_seen": 462684160
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008683999358048468,
      "loss": 0.0856,
      "theoretical_loss": 3.9761420661706683,
      "tokens_seen": 462946304
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008683196918632643,
      "loss": 0.0836,
      "theoretical_loss": 3.975898180597089,
      "tokens_seen": 463208448
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000868239447921682,
      "loss": 0.0852,
      "theoretical_loss": 3.9756544716282187,
      "tokens_seen": 463470592
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008681592039800995,
      "loss": 0.0834,
      "theoretical_loss": 3.975410939036392,
      "tokens_seen": 463732736
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008680789600385172,
      "loss": 0.0849,
      "theoretical_loss": 3.9751675825943638,
      "tokens_seen": 463994880
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008679987160969347,
      "loss": 0.0891,
      "theoretical_loss": 3.974924402075313,
      "tokens_seen": 464257024
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008679184721553522,
      "loss": 0.0858,
      "theoretical_loss": 3.9746813972528354,
      "tokens_seen": 464519168
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008678382282137699,
      "loss": 0.0871,
      "theoretical_loss": 3.9744385679009486,
      "tokens_seen": 464781312
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008677579842721874,
      "loss": 0.0829,
      "theoretical_loss": 3.9741959137940848,
      "tokens_seen": 465043456
    },
    {
      "epoch": 0.14,
      "objective/train/advantage_avg": 0.0013637046795338392,
      "objective/train/docs_used": 175471,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.55617356300354,
      "objective/train/original_loss": 1.55617356300354,
      "objective/train/theoretical_loss": 3.973953434707096,
      "objective/train/tokens_used": 485765600,
      "objective/train/value_avg": -0.00836944580078125,
      "objective/train/value_loss": 0.00027403823332861066,
      "objective/train/value_max": -0.00031256675720214844,
      "objective/train/value_min": -0.7216796875,
      "objective/train/value_reward_corr": 0.5759118889260557,
      "objective/train/value_std": 0.01276397705078125,
      "objective/train/weight_avg": 1.0014877319335938,
      "objective/train/weighted_lm_loss": 1.5572477579116821,
      "objective/train/weights_max": 1.3698445558547974,
      "objective/train/weights_min": 0.3708581030368805,
      "theoretical_loss": 3.973953434707096,
      "tokens_seen": 465305600
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008676777403306051,
      "loss": 0.0824,
      "theoretical_loss": 3.973953434707096,
      "tokens_seen": 465305600
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008675974963890227,
      "loss": 0.0837,
      "theoretical_loss": 3.973711130415248,
      "tokens_seen": 465567744
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008675172524474403,
      "loss": 0.0852,
      "theoretical_loss": 3.973469000694223,
      "tokens_seen": 465829888
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008674370085058578,
      "loss": 0.0837,
      "theoretical_loss": 3.973227045320117,
      "tokens_seen": 466092032
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008673567645642754,
      "loss": 0.085,
      "theoretical_loss": 3.9729852640694383,
      "tokens_seen": 466354176
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000867276520622693,
      "loss": 0.0853,
      "theoretical_loss": 3.972743656719107,
      "tokens_seen": 466616320
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008671962766811105,
      "loss": 0.0815,
      "theoretical_loss": 3.9725022230464537,
      "tokens_seen": 466878464
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008671160327395282,
      "loss": 0.0844,
      "theoretical_loss": 3.9722609628292207,
      "tokens_seen": 467140608
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008670357887979457,
      "loss": 0.0838,
      "theoretical_loss": 3.9720198758455574,
      "tokens_seen": 467402752
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008669555448563635,
      "loss": 0.085,
      "theoretical_loss": 3.9717789618740227,
      "tokens_seen": 467664896
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000866875300914781,
      "loss": 0.0861,
      "theoretical_loss": 3.9715382206935814,
      "tokens_seen": 467927040
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008667950569731985,
      "loss": 0.0818,
      "theoretical_loss": 3.9712976520836043,
      "tokens_seen": 468189184
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008667148130316162,
      "loss": 0.0838,
      "theoretical_loss": 3.971057255823868,
      "tokens_seen": 468451328
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008666345690900337,
      "loss": 0.0856,
      "theoretical_loss": 3.9708170316945526,
      "tokens_seen": 468713472
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008665543251484513,
      "loss": 0.0849,
      "theoretical_loss": 3.9705769794762418,
      "tokens_seen": 468975616
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008664740812068689,
      "loss": 0.0862,
      "theoretical_loss": 3.97033709894992,
      "tokens_seen": 469237760
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008663938372652865,
      "loss": 0.0846,
      "theoretical_loss": 3.970097389896975,
      "tokens_seen": 469499904
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000866313593323704,
      "loss": 0.0851,
      "theoretical_loss": 3.9698578520991936,
      "tokens_seen": 469762048
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008662333493821216,
      "loss": 0.087,
      "theoretical_loss": 3.9696184853387617,
      "tokens_seen": 470024192
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008661531054405393,
      "loss": 0.0832,
      "theoretical_loss": 3.9693792893982636,
      "tokens_seen": 470286336
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008660728614989568,
      "loss": 0.0855,
      "theoretical_loss": 3.969140264060681,
      "tokens_seen": 470548480
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008659926175573745,
      "loss": 0.0841,
      "theoretical_loss": 3.9689014091093933,
      "tokens_seen": 470810624
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000865912373615792,
      "loss": 0.0841,
      "theoretical_loss": 3.9686627243281727,
      "tokens_seen": 471072768
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008658321296742096,
      "loss": 0.0876,
      "theoretical_loss": 3.9684242095011886,
      "tokens_seen": 471334912
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008657518857326272,
      "loss": 0.0839,
      "theoretical_loss": 3.9681858644130017,
      "tokens_seen": 471597056
    },
    {
      "epoch": 0.14,
      "objective/train/advantage_avg": 0.0014818408526480198,
      "objective/train/docs_used": 177948,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.7183817625045776,
      "objective/train/original_loss": 1.718381643295288,
      "objective/train/theoretical_loss": 3.967947688848568,
      "objective/train/tokens_used": 492319200,
      "objective/train/value_avg": -0.01105499267578125,
      "objective/train/value_loss": 0.00045233668060973287,
      "objective/train/value_max": -0.00037550926208496094,
      "objective/train/value_min": -0.8310546875,
      "objective/train/value_reward_corr": 0.7132991610895623,
      "objective/train/value_std": 0.0223541259765625,
      "objective/train/weight_avg": 1.001685619354248,
      "objective/train/weighted_lm_loss": 1.7202060222625732,
      "objective/train/weights_max": 1.5177611112594604,
      "objective/train/weights_min": 0.37088075280189514,
      "theoretical_loss": 3.967947688848568,
      "tokens_seen": 471859200
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008656716417910447,
      "loss": 0.0866,
      "theoretical_loss": 3.967947688848568,
      "tokens_seen": 471859200
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008655913978494624,
      "loss": 0.0803,
      "theoretical_loss": 3.9677096825932328,
      "tokens_seen": 472121344
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008655111539078799,
      "loss": 0.0837,
      "theoretical_loss": 3.9674718454327325,
      "tokens_seen": 472383488
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008654309099662976,
      "loss": 0.0858,
      "theoretical_loss": 3.9672341771531956,
      "tokens_seen": 472645632
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008653506660247152,
      "loss": 0.0856,
      "theoretical_loss": 3.9669966775411365,
      "tokens_seen": 472907776
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008652704220831328,
      "loss": 0.0837,
      "theoretical_loss": 3.96675934638346,
      "tokens_seen": 473169920
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008651901781415503,
      "loss": 0.0859,
      "theoretical_loss": 3.9665221834674558,
      "tokens_seen": 473432064
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000865109934199968,
      "loss": 0.0837,
      "theoretical_loss": 3.9662851885808026,
      "tokens_seen": 473694208
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008650296902583855,
      "loss": 0.0843,
      "theoretical_loss": 3.966048361511562,
      "tokens_seen": 473956352
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000864949446316803,
      "loss": 0.0825,
      "theoretical_loss": 3.96581170204818,
      "tokens_seen": 474218496
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008648692023752207,
      "loss": 0.0867,
      "theoretical_loss": 3.9655752099794874,
      "tokens_seen": 474480640
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008647889584336382,
      "loss": 0.0885,
      "theoretical_loss": 3.9653388850946976,
      "tokens_seen": 474742784
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008647087144920559,
      "loss": 0.0836,
      "theoretical_loss": 3.965102727183404,
      "tokens_seen": 475004928
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008646284705504735,
      "loss": 0.0829,
      "theoretical_loss": 3.9648667360355816,
      "tokens_seen": 475267072
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008645482266088911,
      "loss": 0.0863,
      "theoretical_loss": 3.9646309114415863,
      "tokens_seen": 475529216
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008644679826673087,
      "loss": 0.0834,
      "theoretical_loss": 3.9643952531921505,
      "tokens_seen": 475791360
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008643877387257262,
      "loss": 0.0814,
      "theoretical_loss": 3.9641597610783874,
      "tokens_seen": 476053504
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008643074947841438,
      "loss": 0.0845,
      "theoretical_loss": 3.9639244348917853,
      "tokens_seen": 476315648
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008642272508425614,
      "loss": 0.0821,
      "theoretical_loss": 3.96368927442421,
      "tokens_seen": 476577792
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000864147006900979,
      "loss": 0.0851,
      "theoretical_loss": 3.9634542794679013,
      "tokens_seen": 476839936
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008640667629593965,
      "loss": 0.0828,
      "theoretical_loss": 3.963219449815475,
      "tokens_seen": 477102080
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008639865190178143,
      "loss": 0.0805,
      "theoretical_loss": 3.9629847852599207,
      "tokens_seen": 477364224
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008639062750762318,
      "loss": 0.0868,
      "theoretical_loss": 3.9627502855945984,
      "tokens_seen": 477626368
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008638260311346493,
      "loss": 0.0836,
      "theoretical_loss": 3.962515950613242,
      "tokens_seen": 477888512
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000863745787193067,
      "loss": 0.0858,
      "theoretical_loss": 3.962281780109957,
      "tokens_seen": 478150656
    },
    {
      "epoch": 0.14,
      "objective/train/advantage_avg": 0.0006618535844609141,
      "objective/train/docs_used": 180329,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.7160143852233887,
      "objective/train/original_loss": 1.7160143852233887,
      "objective/train/theoretical_loss": 3.9620477738792164,
      "objective/train/tokens_used": 498872800,
      "objective/train/value_avg": -0.006938934326171875,
      "objective/train/value_loss": 0.0002311340213054791,
      "objective/train/value_max": -0.00033926963806152344,
      "objective/train/value_min": -0.716796875,
      "objective/train/value_reward_corr": 0.5828097197150116,
      "objective/train/value_std": 0.010406494140625,
      "objective/train/weight_avg": 1.0007702112197876,
      "objective/train/weighted_lm_loss": 1.7172032594680786,
      "objective/train/weights_max": 1.2420917749404907,
      "objective/train/weights_min": 0.37445467710494995,
      "theoretical_loss": 3.9620477738792164,
      "tokens_seen": 478412800
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0008636655432514845,
      "loss": 0.0838,
      "theoretical_loss": 3.9620477738792164,
      "tokens_seen": 478412800
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008635852993099021,
      "loss": 0.085,
      "theoretical_loss": 3.9618139317158647,
      "tokens_seen": 478674944
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008635050553683197,
      "loss": 0.084,
      "theoretical_loss": 3.961580253415114,
      "tokens_seen": 478937088
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008634248114267373,
      "loss": 0.084,
      "theoretical_loss": 3.9613467387725434,
      "tokens_seen": 479199232
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008633445674851548,
      "loss": 0.0804,
      "theoretical_loss": 3.9611133875841,
      "tokens_seen": 479461376
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008632643235435724,
      "loss": 0.085,
      "theoretical_loss": 3.960880199646096,
      "tokens_seen": 479723520
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008631840796019901,
      "loss": 0.0857,
      "theoretical_loss": 3.9606471747552083,
      "tokens_seen": 479985664
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008631038356604077,
      "loss": 0.0822,
      "theoretical_loss": 3.9604143127084774,
      "tokens_seen": 480247808
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008630235917188253,
      "loss": 0.0845,
      "theoretical_loss": 3.960181613303309,
      "tokens_seen": 480509952
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008629433477772428,
      "loss": 0.0871,
      "theoretical_loss": 3.9599490763374687,
      "tokens_seen": 480772096
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008628631038356605,
      "loss": 0.0853,
      "theoretical_loss": 3.9597167016090866,
      "tokens_seen": 481034240
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000862782859894078,
      "loss": 0.0825,
      "theoretical_loss": 3.95948448891665,
      "tokens_seen": 481296384
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008627026159524955,
      "loss": 0.0835,
      "theoretical_loss": 3.959252438059009,
      "tokens_seen": 481558528
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008626223720109132,
      "loss": 0.0839,
      "theoretical_loss": 3.9590205488353707,
      "tokens_seen": 481820672
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008625421280693307,
      "loss": 0.0864,
      "theoretical_loss": 3.958788821045302,
      "tokens_seen": 482082816
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008624618841277484,
      "loss": 0.0845,
      "theoretical_loss": 3.958557254488727,
      "tokens_seen": 482344960
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000862381640186166,
      "loss": 0.0836,
      "theoretical_loss": 3.958325848965925,
      "tokens_seen": 482607104
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008623013962445836,
      "loss": 0.0829,
      "theoretical_loss": 3.958094604277532,
      "tokens_seen": 482869248
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008622211523030011,
      "loss": 0.086,
      "theoretical_loss": 3.9578635202245387,
      "tokens_seen": 483131392
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008621409083614188,
      "loss": 0.0803,
      "theoretical_loss": 3.95763259660829,
      "tokens_seen": 483393536
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008620606644198363,
      "loss": 0.0825,
      "theoretical_loss": 3.957401833230484,
      "tokens_seen": 483655680
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008619804204782538,
      "loss": 0.0829,
      "theoretical_loss": 3.957171229893171,
      "tokens_seen": 483917824
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008619001765366715,
      "loss": 0.0842,
      "theoretical_loss": 3.956940786398753,
      "tokens_seen": 484179968
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000861819932595089,
      "loss": 0.0852,
      "theoretical_loss": 3.9567105025499827,
      "tokens_seen": 484442112
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008617396886535068,
      "loss": 0.0835,
      "theoretical_loss": 3.9564803781499633,
      "tokens_seen": 484704256
    },
    {
      "epoch": 0.15,
      "objective/train/advantage_avg": 0.0013643309939652681,
      "objective/train/docs_used": 182723,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5896873474121094,
      "objective/train/original_loss": 1.5896873474121094,
      "objective/train/theoretical_loss": 3.956250413002146,
      "objective/train/tokens_used": 505426400,
      "objective/train/value_avg": -0.00983428955078125,
      "objective/train/value_loss": 0.00046001793816685677,
      "objective/train/value_max": -0.0002512931823730469,
      "objective/train/value_min": -0.94873046875,
      "objective/train/value_reward_corr": 0.692742976797035,
      "objective/train/value_std": 0.0214996337890625,
      "objective/train/weight_avg": 1.0015811920166016,
      "objective/train/weighted_lm_loss": 1.592099905014038,
      "objective/train/weights_max": 2.4389986991882324,
      "objective/train/weights_min": 0.37612104415893555,
      "theoretical_loss": 3.956250413002146,
      "tokens_seen": 484966400
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008616594447119243,
      "loss": 0.084,
      "theoretical_loss": 3.956250413002146,
      "tokens_seen": 484966400
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008615792007703419,
      "loss": 0.0847,
      "theoretical_loss": 3.9560206069103314,
      "tokens_seen": 485228544
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008614989568287595,
      "loss": 0.0855,
      "theoretical_loss": 3.9557909596786676,
      "tokens_seen": 485490688
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000861418712887177,
      "loss": 0.0836,
      "theoretical_loss": 3.9555614711116487,
      "tokens_seen": 485752832
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008613384689455946,
      "loss": 0.0837,
      "theoretical_loss": 3.9553321410141162,
      "tokens_seen": 486014976
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008612582250040122,
      "loss": 0.0828,
      "theoretical_loss": 3.9551029691912545,
      "tokens_seen": 486277120
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008611779810624298,
      "loss": 0.0839,
      "theoretical_loss": 3.954873955448594,
      "tokens_seen": 486539264
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008610977371208473,
      "loss": 0.084,
      "theoretical_loss": 3.9546450995920086,
      "tokens_seen": 486801408
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008610174931792651,
      "loss": 0.0823,
      "theoretical_loss": 3.954416401427715,
      "tokens_seen": 487063552
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008609372492376826,
      "loss": 0.0821,
      "theoretical_loss": 3.9541878607622705,
      "tokens_seen": 487325696
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008608570052961001,
      "loss": 0.0802,
      "theoretical_loss": 3.9539594774025755,
      "tokens_seen": 487587840
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008607767613545178,
      "loss": 0.0833,
      "theoretical_loss": 3.9537312511558698,
      "tokens_seen": 487849984
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008606965174129353,
      "loss": 0.0821,
      "theoretical_loss": 3.953503181829732,
      "tokens_seen": 488112128
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000860616273471353,
      "loss": 0.0838,
      "theoretical_loss": 3.9532752692320816,
      "tokens_seen": 488374272
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008605360295297705,
      "loss": 0.0807,
      "theoretical_loss": 3.9530475131711746,
      "tokens_seen": 488636416
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008604557855881881,
      "loss": 0.0823,
      "theoretical_loss": 3.9528199134556044,
      "tokens_seen": 488898560
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008603755416466057,
      "loss": 0.0838,
      "theoretical_loss": 3.9525924698943022,
      "tokens_seen": 489160704
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008602952977050232,
      "loss": 0.0829,
      "theoretical_loss": 3.952365182296533,
      "tokens_seen": 489422848
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008602150537634409,
      "loss": 0.0842,
      "theoretical_loss": 3.9521380504718975,
      "tokens_seen": 489684992
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008601348098218585,
      "loss": 0.0849,
      "theoretical_loss": 3.9519110742303325,
      "tokens_seen": 489947136
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008600545658802761,
      "loss": 0.0829,
      "theoretical_loss": 3.951684253382105,
      "tokens_seen": 490209280
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008599743219386936,
      "loss": 0.0827,
      "theoretical_loss": 3.951457587737817,
      "tokens_seen": 490471424
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008598940779971113,
      "loss": 0.0829,
      "theoretical_loss": 3.9512310771084014,
      "tokens_seen": 490733568
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008598138340555288,
      "loss": 0.0846,
      "theoretical_loss": 3.951004721305123,
      "tokens_seen": 490995712
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008597335901139463,
      "loss": 0.0828,
      "theoretical_loss": 3.950778520139576,
      "tokens_seen": 491257856
    },
    {
      "epoch": 0.15,
      "objective/train/advantage_avg": 0.0006203674129210413,
      "objective/train/docs_used": 185162,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.6717313528060913,
      "objective/train/original_loss": 1.6717313528060913,
      "objective/train/theoretical_loss": 3.9505524734236848,
      "objective/train/tokens_used": 511980000,
      "objective/train/value_avg": -0.00823974609375,
      "objective/train/value_loss": 0.00022850836103316396,
      "objective/train/value_max": -0.0002758502960205078,
      "objective/train/value_min": -0.67724609375,
      "objective/train/value_reward_corr": 0.6322221181145709,
      "objective/train/value_std": 0.0125732421875,
      "objective/train/weight_avg": 1.0007286071777344,
      "objective/train/weighted_lm_loss": 1.672489881515503,
      "objective/train/weights_max": 1.3001129627227783,
      "objective/train/weights_min": 0.372374951839447,
      "theoretical_loss": 3.9505524734236848,
      "tokens_seen": 491520000
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000859653346172364,
      "loss": 0.0839,
      "theoretical_loss": 3.9505524734236848,
      "tokens_seen": 491520000
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008595731022307815,
      "loss": 0.0811,
      "theoretical_loss": 3.950326580969703,
      "tokens_seen": 491782144
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008594928582891992,
      "loss": 0.0851,
      "theoretical_loss": 3.950100842590212,
      "tokens_seen": 492044288
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008594126143476168,
      "loss": 0.0847,
      "theoretical_loss": 3.949875258098121,
      "tokens_seen": 492306432
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008593323704060344,
      "loss": 0.0849,
      "theoretical_loss": 3.949649827306665,
      "tokens_seen": 492568576
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000859252126464452,
      "loss": 0.084,
      "theoretical_loss": 3.9494245500294047,
      "tokens_seen": 492830720
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008591718825228696,
      "loss": 0.0837,
      "theoretical_loss": 3.949199426080228,
      "tokens_seen": 493092864
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008590916385812871,
      "loss": 0.0848,
      "theoretical_loss": 3.9489744552733455,
      "tokens_seen": 493355008
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008590113946397047,
      "loss": 0.0887,
      "theoretical_loss": 3.9487496374232913,
      "tokens_seen": 493617152
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008589311506981223,
      "loss": 0.0828,
      "theoretical_loss": 3.9485249723449236,
      "tokens_seen": 493879296
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008588509067565398,
      "loss": 0.0822,
      "theoretical_loss": 3.9483004598534217,
      "tokens_seen": 494141440
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008587706628149576,
      "loss": 0.0855,
      "theoretical_loss": 3.948076099764288,
      "tokens_seen": 494403584
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008586904188733751,
      "loss": 0.089,
      "theoretical_loss": 3.947851891893343,
      "tokens_seen": 494665728
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008586101749317927,
      "loss": 0.0803,
      "theoretical_loss": 3.9476278360567303,
      "tokens_seen": 494927872
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008585299309902103,
      "loss": 0.0838,
      "theoretical_loss": 3.9474039320709107,
      "tokens_seen": 495190016
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008584496870486278,
      "loss": 0.0867,
      "theoretical_loss": 3.9471801797526633,
      "tokens_seen": 495452160
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008583694431070454,
      "loss": 0.0818,
      "theoretical_loss": 3.946956578919088,
      "tokens_seen": 495714304
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000858289199165463,
      "loss": 0.0841,
      "theoretical_loss": 3.9467331293875976,
      "tokens_seen": 495976448
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008582089552238806,
      "loss": 0.0854,
      "theoretical_loss": 3.9465098309759252,
      "tokens_seen": 496238592
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008581287112822982,
      "loss": 0.0843,
      "theoretical_loss": 3.9462866835021178,
      "tokens_seen": 496500736
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008580484673407159,
      "loss": 0.0838,
      "theoretical_loss": 3.9460636867845365,
      "tokens_seen": 496762880
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008579682233991334,
      "loss": 0.0851,
      "theoretical_loss": 3.9458408406418584,
      "tokens_seen": 497025024
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000857887979457551,
      "loss": 0.0823,
      "theoretical_loss": 3.945618144893073,
      "tokens_seen": 497287168
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008578077355159686,
      "loss": 0.0865,
      "theoretical_loss": 3.9453955993574845,
      "tokens_seen": 497549312
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008577274915743861,
      "loss": 0.083,
      "theoretical_loss": 3.945173203854707,
      "tokens_seen": 497811456
    },
    {
      "epoch": 0.15,
      "objective/train/advantage_avg": 0.00040782641735859215,
      "objective/train/docs_used": 187487,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.7300009727478027,
      "objective/train/original_loss": 1.7300009727478027,
      "objective/train/theoretical_loss": 3.9449509582046662,
      "objective/train/tokens_used": 518533600,
      "objective/train/value_avg": -0.00957489013671875,
      "objective/train/value_loss": 0.0005079260445199907,
      "objective/train/value_max": -0.00016736984252929688,
      "objective/train/value_min": -0.72998046875,
      "objective/train/value_reward_corr": 0.5804749067115648,
      "objective/train/value_std": 0.01552581787109375,
      "objective/train/weight_avg": 1.0006271600723267,
      "objective/train/weighted_lm_loss": 1.7296302318572998,
      "objective/train/weights_max": 1.716912031173706,
      "objective/train/weights_min": 0.36845633387565613,
      "theoretical_loss": 3.9449509582046662,
      "tokens_seen": 498073600
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008576472476328038,
      "loss": 0.0856,
      "theoretical_loss": 3.9449509582046662,
      "tokens_seen": 498073600
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008575670036912213,
      "loss": 0.0832,
      "theoretical_loss": 3.944728862227601,
      "tokens_seen": 498335744
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008574867597496389,
      "loss": 0.0812,
      "theoretical_loss": 3.9445069157440575,
      "tokens_seen": 498597888
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008574065158080565,
      "loss": 0.0846,
      "theoretical_loss": 3.944285118574893,
      "tokens_seen": 498860032
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000857326271866474,
      "loss": 0.0842,
      "theoretical_loss": 3.9440634705412725,
      "tokens_seen": 499122176
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008572460279248917,
      "loss": 0.0851,
      "theoretical_loss": 3.9438419714646695,
      "tokens_seen": 499384320
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008571657839833093,
      "loss": 0.0868,
      "theoretical_loss": 3.9436206211668647,
      "tokens_seen": 499646464
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008570855400417269,
      "loss": 0.0806,
      "theoretical_loss": 3.9433994194699453,
      "tokens_seen": 499908608
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008570052961001444,
      "loss": 0.0834,
      "theoretical_loss": 3.943178366196304,
      "tokens_seen": 500170752
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008569250521585621,
      "loss": 0.0847,
      "theoretical_loss": 3.942957461168639,
      "tokens_seen": 500432896
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008568448082169796,
      "loss": 0.0869,
      "theoretical_loss": 3.9427367042099544,
      "tokens_seen": 500695040
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008567645642753972,
      "loss": 0.0808,
      "theoretical_loss": 3.942516095143555,
      "tokens_seen": 500957184
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008566843203338148,
      "loss": 0.0846,
      "theoretical_loss": 3.9422956337930524,
      "tokens_seen": 501219328
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008566040763922323,
      "loss": 0.0818,
      "theoretical_loss": 3.942075319982358,
      "tokens_seen": 501481472
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008565238324506501,
      "loss": 0.0828,
      "theoretical_loss": 3.941855153535686,
      "tokens_seen": 501743616
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008564435885090676,
      "loss": 0.0837,
      "theoretical_loss": 3.9416351342775524,
      "tokens_seen": 502005760
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008563633445674852,
      "loss": 0.0825,
      "theoretical_loss": 3.9414152620327716,
      "tokens_seen": 502267904
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008562831006259028,
      "loss": 0.0798,
      "theoretical_loss": 3.941195536626461,
      "tokens_seen": 502530048
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008562028566843203,
      "loss": 0.0835,
      "theoretical_loss": 3.940975957884034,
      "tokens_seen": 502792192
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008561226127427379,
      "loss": 0.0839,
      "theoretical_loss": 3.9407565256312047,
      "tokens_seen": 503054336
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008560423688011555,
      "loss": 0.0859,
      "theoretical_loss": 3.940537239693983,
      "tokens_seen": 503316480
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008559621248595731,
      "loss": 0.0844,
      "theoretical_loss": 3.9403180998986778,
      "tokens_seen": 503578624
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008558818809179906,
      "loss": 0.0805,
      "theoretical_loss": 3.9400991060718935,
      "tokens_seen": 503840768
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008558016369764084,
      "loss": 0.0865,
      "theoretical_loss": 3.93988025804053,
      "tokens_seen": 504102912
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008557213930348259,
      "loss": 0.0846,
      "theoretical_loss": 3.9396615556317838,
      "tokens_seen": 504365056
    },
    {
      "epoch": 0.15,
      "objective/train/advantage_avg": 0.003414377337321639,
      "objective/train/docs_used": 189911,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.635353684425354,
      "objective/train/original_loss": 1.6353535652160645,
      "objective/train/theoretical_loss": 3.939442998673144,
      "objective/train/tokens_used": 525087200,
      "objective/train/value_avg": -0.0120849609375,
      "objective/train/value_loss": 0.0004840447218157351,
      "objective/train/value_max": -0.0003199577331542969,
      "objective/train/value_min": -0.7880859375,
      "objective/train/value_reward_corr": 0.7853448439202004,
      "objective/train/value_std": 0.0305023193359375,
      "objective/train/weight_avg": 1.0036317110061646,
      "objective/train/weighted_lm_loss": 1.6402772665023804,
      "objective/train/weights_max": 1.392095923423767,
      "objective/train/weights_min": 0.3688413202762604,
      "theoretical_loss": 3.939442998673144,
      "tokens_seen": 504627200
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008556411490932436,
      "loss": 0.0822,
      "theoretical_loss": 3.939442998673144,
      "tokens_seen": 504627200
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008555609051516611,
      "loss": 0.0824,
      "theoretical_loss": 3.9392245869923954,
      "tokens_seen": 504889344
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008554806612100786,
      "loss": 0.0865,
      "theoretical_loss": 3.939006320417614,
      "tokens_seen": 505151488
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008554004172684963,
      "loss": 0.0833,
      "theoretical_loss": 3.9387881987771705,
      "tokens_seen": 505413632
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008553201733269138,
      "loss": 0.0847,
      "theoretical_loss": 3.9385702218997247,
      "tokens_seen": 505675776
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008552399293853314,
      "loss": 0.0849,
      "theoretical_loss": 3.9383523896142316,
      "tokens_seen": 505937920
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000855159685443749,
      "loss": 0.0849,
      "theoretical_loss": 3.9381347017499326,
      "tokens_seen": 506200064
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008550794415021667,
      "loss": 0.0847,
      "theoretical_loss": 3.9379171581363623,
      "tokens_seen": 506462208
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008549991975605842,
      "loss": 0.0846,
      "theoretical_loss": 3.937699758603342,
      "tokens_seen": 506724352
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008549189536190018,
      "loss": 0.0845,
      "theoretical_loss": 3.937482502980985,
      "tokens_seen": 506986496
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008548387096774194,
      "loss": 0.0873,
      "theoretical_loss": 3.937265391099688,
      "tokens_seen": 507248640
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008547584657358369,
      "loss": 0.081,
      "theoretical_loss": 3.9370484227901397,
      "tokens_seen": 507510784
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008546782217942546,
      "loss": 0.0815,
      "theoretical_loss": 3.9368315978833124,
      "tokens_seen": 507772928
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008545979778526721,
      "loss": 0.0824,
      "theoretical_loss": 3.936614916210466,
      "tokens_seen": 508035072
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008545177339110897,
      "loss": 0.0811,
      "theoretical_loss": 3.9363983776031457,
      "tokens_seen": 508297216
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008544374899695073,
      "loss": 0.0846,
      "theoretical_loss": 3.936181981893182,
      "tokens_seen": 508559360
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008543572460279248,
      "loss": 0.0867,
      "theoretical_loss": 3.9359657289126875,
      "tokens_seen": 508821504
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008542770020863426,
      "loss": 0.0841,
      "theoretical_loss": 3.935749618494061,
      "tokens_seen": 509083648
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008541967581447601,
      "loss": 0.0836,
      "theoretical_loss": 3.935533650469983,
      "tokens_seen": 509345792
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008541165142031777,
      "loss": 0.0818,
      "theoretical_loss": 3.935317824673417,
      "tokens_seen": 509607936
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008540362702615953,
      "loss": 0.0842,
      "theoretical_loss": 3.935102140937608,
      "tokens_seen": 509870080
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008539560263200129,
      "loss": 0.0829,
      "theoretical_loss": 3.934886599096081,
      "tokens_seen": 510132224
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008538757823784304,
      "loss": 0.0819,
      "theoretical_loss": 3.9346711989826426,
      "tokens_seen": 510394368
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000853795538436848,
      "loss": 0.0832,
      "theoretical_loss": 3.93445594043138,
      "tokens_seen": 510656512
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008537152944952656,
      "loss": 0.082,
      "theoretical_loss": 3.9342408232766584,
      "tokens_seen": 510918656
    },
    {
      "epoch": 0.15,
      "objective/train/advantage_avg": 0.0016618784284219146,
      "objective/train/docs_used": 192198,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.7548766136169434,
      "objective/train/original_loss": 1.7548766136169434,
      "objective/train/theoretical_loss": 3.934025847353122,
      "objective/train/tokens_used": 531640800,
      "objective/train/value_avg": -0.00916290283203125,
      "objective/train/value_loss": 0.0003608835977502167,
      "objective/train/value_max": -0.0002694129943847656,
      "objective/train/value_min": -0.78173828125,
      "objective/train/value_reward_corr": 0.6032279595057155,
      "objective/train/value_std": 0.015838623046875,
      "objective/train/weight_avg": 1.0018247365951538,
      "objective/train/weighted_lm_loss": 1.7572846412658691,
      "objective/train/weights_max": 1.5906000137329102,
      "objective/train/weights_min": 0.3688792884349823,
      "theoretical_loss": 3.934025847353122,
      "tokens_seen": 511180800
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008536350505536831,
      "loss": 0.0836,
      "theoretical_loss": 3.934025847353122,
      "tokens_seen": 511180800
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0008535548066121009,
      "loss": 0.0818,
      "theoretical_loss": 3.9338110124956924,
      "tokens_seen": 511442944
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008534745626705184,
      "loss": 0.0835,
      "theoretical_loss": 3.9335963185395713,
      "tokens_seen": 511705088
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000853394318728936,
      "loss": 0.0811,
      "theoretical_loss": 3.933381765320233,
      "tokens_seen": 511967232
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008533140747873536,
      "loss": 0.0812,
      "theoretical_loss": 3.933167352673432,
      "tokens_seen": 512229376
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008532338308457711,
      "loss": 0.0834,
      "theoretical_loss": 3.9329530804351958,
      "tokens_seen": 512491520
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008531535869041887,
      "loss": 0.0855,
      "theoretical_loss": 3.9327389484418287,
      "tokens_seen": 512753664
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008530733429626063,
      "loss": 0.0804,
      "theoretical_loss": 3.9325249565299076,
      "tokens_seen": 513015808
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008529930990210239,
      "loss": 0.0843,
      "theoretical_loss": 3.932311104536285,
      "tokens_seen": 513277952
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008529128550794415,
      "loss": 0.0822,
      "theoretical_loss": 3.9320973922980844,
      "tokens_seen": 513540096
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008528326111378592,
      "loss": 0.0844,
      "theoretical_loss": 3.931883819652705,
      "tokens_seen": 513802240
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008527523671962767,
      "loss": 0.082,
      "theoretical_loss": 3.9316703864378155,
      "tokens_seen": 514064384
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008526721232546944,
      "loss": 0.082,
      "theoretical_loss": 3.9314570924913568,
      "tokens_seen": 514326528
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008525918793131119,
      "loss": 0.0834,
      "theoretical_loss": 3.9312439376515407,
      "tokens_seen": 514588672
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008525116353715294,
      "loss": 0.083,
      "theoretical_loss": 3.9310309217568493,
      "tokens_seen": 514850816
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008524313914299471,
      "loss": 0.0845,
      "theoretical_loss": 3.9308180446460343,
      "tokens_seen": 515112960
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008523511474883646,
      "loss": 0.0838,
      "theoretical_loss": 3.9306053061581165,
      "tokens_seen": 515375104
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008522709035467822,
      "loss": 0.0834,
      "theoretical_loss": 3.930392706132385,
      "tokens_seen": 515637248
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008521906596051998,
      "loss": 0.0845,
      "theoretical_loss": 3.9301802444083966,
      "tokens_seen": 515899392
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008521104156636175,
      "loss": 0.0846,
      "theoretical_loss": 3.929967920825977,
      "tokens_seen": 516161536
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000852030171722035,
      "loss": 0.0805,
      "theoretical_loss": 3.929755735225216,
      "tokens_seen": 516423680
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008519499277804526,
      "loss": 0.0829,
      "theoretical_loss": 3.9295436874464715,
      "tokens_seen": 516685824
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008518696838388702,
      "loss": 0.0836,
      "theoretical_loss": 3.929331777330366,
      "tokens_seen": 516947968
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008517894398972877,
      "loss": 0.0798,
      "theoretical_loss": 3.9291200047177886,
      "tokens_seen": 517210112
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008517091959557054,
      "loss": 0.0815,
      "theoretical_loss": 3.9289083694498905,
      "tokens_seen": 517472256
    },
    {
      "epoch": 0.16,
      "objective/train/advantage_avg": 0.001091600046493113,
      "objective/train/docs_used": 194581,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.6318016052246094,
      "objective/train/original_loss": 1.6318016052246094,
      "objective/train/theoretical_loss": 3.9286968713680883,
      "objective/train/tokens_used": 538194400,
      "objective/train/value_avg": -0.00958251953125,
      "objective/train/value_loss": 0.0004281644069124013,
      "objective/train/value_max": -0.0002453327178955078,
      "objective/train/value_min": -0.646484375,
      "objective/train/value_reward_corr": 0.7146024060573813,
      "objective/train/value_std": 0.0170135498046875,
      "objective/train/weight_avg": 1.0012816190719604,
      "objective/train/weighted_lm_loss": 1.6339389085769653,
      "objective/train/weights_max": 1.4248768091201782,
      "objective/train/weights_min": 0.36841943860054016,
      "theoretical_loss": 3.9286968713680883,
      "tokens_seen": 517734400
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008516289520141229,
      "loss": 0.0804,
      "theoretical_loss": 3.9286968713680883,
      "tokens_seen": 517734400
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008515487080725406,
      "loss": 0.0823,
      "theoretical_loss": 3.9284855103140615,
      "tokens_seen": 517996544
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008514684641309581,
      "loss": 0.0822,
      "theoretical_loss": 3.9282742861297524,
      "tokens_seen": 518258688
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008513882201893756,
      "loss": 0.0788,
      "theoretical_loss": 3.928063198657365,
      "tokens_seen": 518520832
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008513079762477934,
      "loss": 0.0808,
      "theoretical_loss": 3.9278522477393656,
      "tokens_seen": 518782976
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008512277323062109,
      "loss": 0.0824,
      "theoretical_loss": 3.9276414332184815,
      "tokens_seen": 519045120
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008511474883646285,
      "loss": 0.0818,
      "theoretical_loss": 3.927430754937699,
      "tokens_seen": 519307264
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008510672444230461,
      "loss": 0.0795,
      "theoretical_loss": 3.927220212740267,
      "tokens_seen": 519569408
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008509870004814637,
      "loss": 0.0795,
      "theoretical_loss": 3.9270098064696906,
      "tokens_seen": 519831552
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008509067565398812,
      "loss": 0.0813,
      "theoretical_loss": 3.9267995359697356,
      "tokens_seen": 520093696
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008508265125982988,
      "loss": 0.082,
      "theoretical_loss": 3.926589401084426,
      "tokens_seen": 520355840
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008507462686567164,
      "loss": 0.0803,
      "theoretical_loss": 3.9263794016580427,
      "tokens_seen": 520617984
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008506660247151339,
      "loss": 0.0798,
      "theoretical_loss": 3.9261695375351238,
      "tokens_seen": 520880128
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008505857807735517,
      "loss": 0.0839,
      "theoretical_loss": 3.9259598085604646,
      "tokens_seen": 521142272
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008505055368319692,
      "loss": 0.0809,
      "theoretical_loss": 3.925750214579116,
      "tokens_seen": 521404416
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008504252928903869,
      "loss": 0.0844,
      "theoretical_loss": 3.9255407554363835,
      "tokens_seen": 521666560
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008503450489488044,
      "loss": 0.0812,
      "theoretical_loss": 3.92533143097783,
      "tokens_seen": 521928704
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008502648050072219,
      "loss": 0.0827,
      "theoretical_loss": 3.9251222410492694,
      "tokens_seen": 522190848
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008501845610656396,
      "loss": 0.0829,
      "theoretical_loss": 3.924913185496772,
      "tokens_seen": 522452992
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008501043171240571,
      "loss": 0.0813,
      "theoretical_loss": 3.924704264166659,
      "tokens_seen": 522715136
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008500240731824747,
      "loss": 0.0816,
      "theoretical_loss": 3.9244954769055074,
      "tokens_seen": 522977280
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008499438292408923,
      "loss": 0.0843,
      "theoretical_loss": 3.924286823560144,
      "tokens_seen": 523239424
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00084986358529931,
      "loss": 0.0805,
      "theoretical_loss": 3.9240783039776472,
      "tokens_seen": 523501568
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008497833413577275,
      "loss": 0.0825,
      "theoretical_loss": 3.9238699180053485,
      "tokens_seen": 523763712
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008497030974161451,
      "loss": 0.0836,
      "theoretical_loss": 3.923661665490828,
      "tokens_seen": 524025856
    },
    {
      "epoch": 0.16,
      "objective/train/advantage_avg": 0.0016532277222722769,
      "objective/train/docs_used": 196964,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5641077756881714,
      "objective/train/original_loss": 1.5641076564788818,
      "objective/train/theoretical_loss": 3.9234535462819156,
      "objective/train/tokens_used": 544748000,
      "objective/train/value_avg": -0.00901031494140625,
      "objective/train/value_loss": 0.00023702048929408193,
      "objective/train/value_max": -0.0002434253692626953,
      "objective/train/value_min": -0.58154296875,
      "objective/train/value_reward_corr": 0.6424405895231002,
      "objective/train/value_std": 0.0155029296875,
      "objective/train/weight_avg": 1.0017648935317993,
      "objective/train/weighted_lm_loss": 1.566743016242981,
      "objective/train/weights_max": 1.7467089891433716,
      "objective/train/weights_min": 0.3871968686580658,
      "theoretical_loss": 3.9234535462819156,
      "tokens_seen": 524288000
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008496228534745627,
      "loss": 0.0815,
      "theoretical_loss": 3.9234535462819156,
      "tokens_seen": 524288000
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008495426095329802,
      "loss": 0.0821,
      "theoretical_loss": 3.923245560226693,
      "tokens_seen": 524550144
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008494623655913979,
      "loss": 0.0791,
      "theoretical_loss": 3.9230377071734885,
      "tokens_seen": 524812288
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008493821216498154,
      "loss": 0.0819,
      "theoretical_loss": 3.9228299869708794,
      "tokens_seen": 525074432
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000849301877708233,
      "loss": 0.0824,
      "theoretical_loss": 3.9226223994676923,
      "tokens_seen": 525336576
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008492216337666506,
      "loss": 0.0852,
      "theoretical_loss": 3.9224149445129983,
      "tokens_seen": 525598720
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008491413898250683,
      "loss": 0.0812,
      "theoretical_loss": 3.922207621956119,
      "tokens_seen": 525860864
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008490611458834859,
      "loss": 0.0804,
      "theoretical_loss": 3.9220004316466186,
      "tokens_seen": 526123008
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008489809019419034,
      "loss": 0.0838,
      "theoretical_loss": 3.9217933734343093,
      "tokens_seen": 526385152
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000848900658000321,
      "loss": 0.0825,
      "theoretical_loss": 3.9215864471692488,
      "tokens_seen": 526647296
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008488204140587386,
      "loss": 0.0816,
      "theoretical_loss": 3.921379652701738,
      "tokens_seen": 526909440
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008487401701171562,
      "loss": 0.0808,
      "theoretical_loss": 3.9211729898823235,
      "tokens_seen": 527171584
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008486599261755737,
      "loss": 0.082,
      "theoretical_loss": 3.920966458561794,
      "tokens_seen": 527433728
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008485796822339914,
      "loss": 0.0813,
      "theoretical_loss": 3.920760058591182,
      "tokens_seen": 527695872
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008484994382924089,
      "loss": 0.0826,
      "theoretical_loss": 3.9205537898217644,
      "tokens_seen": 527958016
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008484191943508264,
      "loss": 0.0817,
      "theoretical_loss": 3.920347652105058,
      "tokens_seen": 528220160
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008483389504092442,
      "loss": 0.0826,
      "theoretical_loss": 3.920141645292821,
      "tokens_seen": 528482304
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008482587064676617,
      "loss": 0.081,
      "theoretical_loss": 3.919935769237055,
      "tokens_seen": 528744448
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008481784625260793,
      "loss": 0.0839,
      "theoretical_loss": 3.91973002379,
      "tokens_seen": 529006592
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008480982185844969,
      "loss": 0.0818,
      "theoretical_loss": 3.919524408804137,
      "tokens_seen": 529268736
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008480179746429145,
      "loss": 0.0827,
      "theoretical_loss": 3.9193189241321873,
      "tokens_seen": 529530880
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008479377307013321,
      "loss": 0.0821,
      "theoretical_loss": 3.9191135696271098,
      "tokens_seen": 529793024
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008478574867597496,
      "loss": 0.0851,
      "theoretical_loss": 3.9189083451421025,
      "tokens_seen": 530055168
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008477772428181672,
      "loss": 0.0782,
      "theoretical_loss": 3.9187032505306023,
      "tokens_seen": 530317312
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008476969988765848,
      "loss": 0.0809,
      "theoretical_loss": 3.918498285646282,
      "tokens_seen": 530579456
    },
    {
      "epoch": 0.16,
      "objective/train/advantage_avg": 7.436873420374468e-05,
      "objective/train/docs_used": 199409,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5998988151550293,
      "objective/train/original_loss": 1.5998989343643188,
      "objective/train/theoretical_loss": 3.9182934503430538,
      "objective/train/tokens_used": 551301600,
      "objective/train/value_avg": -0.00795745849609375,
      "objective/train/value_loss": 0.00032716241548769176,
      "objective/train/value_max": -0.00034332275390625,
      "objective/train/value_min": -0.72900390625,
      "objective/train/value_reward_corr": 0.5799935782849467,
      "objective/train/value_std": 0.01358795166015625,
      "objective/train/weight_avg": 1.0002270936965942,
      "objective/train/weighted_lm_loss": 1.5996286869049072,
      "objective/train/weights_max": 1.4670954942703247,
      "objective/train/weights_min": 0.38993605971336365,
      "theoretical_loss": 3.9182934503430538,
      "tokens_seen": 530841600
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008476167549350025,
      "loss": 0.0815,
      "theoretical_loss": 3.9182934503430538,
      "tokens_seen": 530841600
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00084753651099342,
      "loss": 0.0792,
      "theoretical_loss": 3.918088744475064,
      "tokens_seen": 531103744
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008474562670518377,
      "loss": 0.084,
      "theoretical_loss": 3.9178841678966956,
      "tokens_seen": 531365888
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008473760231102552,
      "loss": 0.0841,
      "theoretical_loss": 3.9176797204625693,
      "tokens_seen": 531628032
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008472957791686727,
      "loss": 0.083,
      "theoretical_loss": 3.917475402027537,
      "tokens_seen": 531890176
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008472155352270904,
      "loss": 0.0796,
      "theoretical_loss": 3.917271212446689,
      "tokens_seen": 532152320
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008471352912855079,
      "loss": 0.082,
      "theoretical_loss": 3.917067151575348,
      "tokens_seen": 532414464
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008470550473439255,
      "loss": 0.0846,
      "theoretical_loss": 3.916863219269069,
      "tokens_seen": 532676608
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008469748034023431,
      "loss": 0.0813,
      "theoretical_loss": 3.9166594153836427,
      "tokens_seen": 532938752
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008468945594607608,
      "loss": 0.082,
      "theoretical_loss": 3.9164557397750897,
      "tokens_seen": 533200896
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008468143155191783,
      "loss": 0.0815,
      "theoretical_loss": 3.916252192299665,
      "tokens_seen": 533463040
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008467340715775959,
      "loss": 0.0816,
      "theoretical_loss": 3.9160487728138538,
      "tokens_seen": 533725184
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008466538276360135,
      "loss": 0.0821,
      "theoretical_loss": 3.9158454811743733,
      "tokens_seen": 533987328
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008465735836944311,
      "loss": 0.0847,
      "theoretical_loss": 3.915642317238171,
      "tokens_seen": 534249472
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008464933397528487,
      "loss": 0.0836,
      "theoretical_loss": 3.915439280862423,
      "tokens_seen": 534511616
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008464130958112662,
      "loss": 0.0811,
      "theoretical_loss": 3.915236371904539,
      "tokens_seen": 534773760
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008463328518696839,
      "loss": 0.0822,
      "theoretical_loss": 3.915033590222153,
      "tokens_seen": 535035904
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008462526079281014,
      "loss": 0.0824,
      "theoretical_loss": 3.914830935673132,
      "tokens_seen": 535298048
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008461723639865191,
      "loss": 0.0793,
      "theoretical_loss": 3.914628408115569,
      "tokens_seen": 535560192
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008460921200449367,
      "loss": 0.0813,
      "theoretical_loss": 3.9144260074077843,
      "tokens_seen": 535822336
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008460118761033542,
      "loss": 0.084,
      "theoretical_loss": 3.9142237334083276,
      "tokens_seen": 536084480
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008459316321617718,
      "loss": 0.0814,
      "theoretical_loss": 3.914021585975973,
      "tokens_seen": 536346624
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008458513882201894,
      "loss": 0.0803,
      "theoretical_loss": 3.9138195649697227,
      "tokens_seen": 536608768
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000845771144278607,
      "loss": 0.0823,
      "theoretical_loss": 3.9136176702488044,
      "tokens_seen": 536870912
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008456909003370245,
      "loss": 0.0789,
      "theoretical_loss": 3.91341590167267,
      "tokens_seen": 537133056
    },
    {
      "epoch": 0.16,
      "objective/train/advantage_avg": 0.0004424227518029511,
      "objective/train/docs_used": 201641,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5473430156707764,
      "objective/train/original_loss": 1.5473428964614868,
      "objective/train/theoretical_loss": 3.9132142591009975,
      "objective/train/tokens_used": 557855200,
      "objective/train/value_avg": -0.007518768310546875,
      "objective/train/value_loss": 0.00024131749523803592,
      "objective/train/value_max": -0.00021147727966308594,
      "objective/train/value_min": -0.64697265625,
      "objective/train/value_reward_corr": 0.657343727691196,
      "objective/train/value_std": 0.01261138916015625,
      "objective/train/weight_avg": 1.0005532503128052,
      "objective/train/weighted_lm_loss": 1.5476828813552856,
      "objective/train/weights_max": 1.3888442516326904,
      "objective/train/weights_min": 0.3710732161998749,
      "theoretical_loss": 3.9132142591009975,
      "tokens_seen": 537395200
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008456106563954422,
      "loss": 0.0803,
      "theoretical_loss": 3.9132142591009975,
      "tokens_seen": 537395200
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008455304124538597,
      "loss": 0.081,
      "theoretical_loss": 3.9130127423936907,
      "tokens_seen": 537657344
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008454501685122772,
      "loss": 0.0836,
      "theoretical_loss": 3.9128113514108733,
      "tokens_seen": 537919488
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000845369924570695,
      "loss": 0.0817,
      "theoretical_loss": 3.9126100860128963,
      "tokens_seen": 538181632
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008452896806291125,
      "loss": 0.0832,
      "theoretical_loss": 3.9124089460603324,
      "tokens_seen": 538443776
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008452094366875302,
      "loss": 0.0823,
      "theoretical_loss": 3.9122079314139766,
      "tokens_seen": 538705920
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008451291927459477,
      "loss": 0.0824,
      "theoretical_loss": 3.9120070419348463,
      "tokens_seen": 538968064
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008450489488043653,
      "loss": 0.0795,
      "theoretical_loss": 3.9118062774841804,
      "tokens_seen": 539230208
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008449687048627829,
      "loss": 0.0836,
      "theoretical_loss": 3.91160563792344,
      "tokens_seen": 539492352
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008448884609212004,
      "loss": 0.0825,
      "theoretical_loss": 3.911405123114305,
      "tokens_seen": 539754496
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000844808216979618,
      "loss": 0.0844,
      "theoretical_loss": 3.9112047329186783,
      "tokens_seen": 540016640
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008447279730380356,
      "loss": 0.0829,
      "theoretical_loss": 3.911004467198679,
      "tokens_seen": 540278784
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008446477290964533,
      "loss": 0.0839,
      "theoretical_loss": 3.9108043258166485,
      "tokens_seen": 540540928
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008445674851548708,
      "loss": 0.0813,
      "theoretical_loss": 3.910604308635146,
      "tokens_seen": 540803072
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008444872412132885,
      "loss": 0.0796,
      "theoretical_loss": 3.9104044155169495,
      "tokens_seen": 541065216
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000844406997271706,
      "loss": 0.0852,
      "theoretical_loss": 3.910204646325055,
      "tokens_seen": 541327360
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008443267533301235,
      "loss": 0.0845,
      "theoretical_loss": 3.9100050009226752,
      "tokens_seen": 541589504
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008442465093885412,
      "loss": 0.0826,
      "theoretical_loss": 3.9098054791732406,
      "tokens_seen": 541851648
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008441662654469587,
      "loss": 0.0833,
      "theoretical_loss": 3.909606080940399,
      "tokens_seen": 542113792
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008440860215053764,
      "loss": 0.0824,
      "theoretical_loss": 3.909406806088013,
      "tokens_seen": 542375936
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008440057775637939,
      "loss": 0.0836,
      "theoretical_loss": 3.909207654480162,
      "tokens_seen": 542638080
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008439255336222116,
      "loss": 0.0817,
      "theoretical_loss": 3.9090086259811403,
      "tokens_seen": 542900224
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008438452896806292,
      "loss": 0.082,
      "theoretical_loss": 3.908809720455457,
      "tokens_seen": 543162368
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008437650457390467,
      "loss": 0.0819,
      "theoretical_loss": 3.908610937767836,
      "tokens_seen": 543424512
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008436848017974643,
      "loss": 0.0828,
      "theoretical_loss": 3.9084122777832144,
      "tokens_seen": 543686656
    },
    {
      "epoch": 0.16,
      "objective/train/advantage_avg": 0.0003520216268952936,
      "objective/train/docs_used": 204061,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.68601655960083,
      "objective/train/original_loss": 1.6860167980194092,
      "objective/train/theoretical_loss": 3.908213740366744,
      "objective/train/tokens_used": 564408800,
      "objective/train/value_avg": -0.0084991455078125,
      "objective/train/value_loss": 0.0002631843963172287,
      "objective/train/value_max": -0.0003101825714111328,
      "objective/train/value_min": -0.2208251953125,
      "objective/train/value_reward_corr": 0.6833106642077473,
      "objective/train/value_std": 0.01250457763671875,
      "objective/train/weight_avg": 1.000474214553833,
      "objective/train/weighted_lm_loss": 1.6869690418243408,
      "objective/train/weights_max": 1.1504311561584473,
      "objective/train/weights_min": 0.3698042333126068,
      "theoretical_loss": 3.908213740366744,
      "tokens_seen": 543948800
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008436045578558819,
      "loss": 0.0831,
      "theoretical_loss": 3.908213740366744,
      "tokens_seen": 543948800
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0008435243139142995,
      "loss": 0.0826,
      "theoretical_loss": 3.908015325383788,
      "tokens_seen": 544210944
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000843444069972717,
      "loss": 0.0819,
      "theoretical_loss": 3.907817032699924,
      "tokens_seen": 544473088
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008433638260311347,
      "loss": 0.0809,
      "theoretical_loss": 3.9076188621809416,
      "tokens_seen": 544735232
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008432835820895522,
      "loss": 0.0838,
      "theoretical_loss": 3.9074208136928408,
      "tokens_seen": 544997376
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008432033381479699,
      "loss": 0.083,
      "theoretical_loss": 3.907222887101834,
      "tokens_seen": 545259520
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008431230942063875,
      "loss": 0.0854,
      "theoretical_loss": 3.9070250822743446,
      "tokens_seen": 545521664
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000843042850264805,
      "loss": 0.0844,
      "theoretical_loss": 3.906827399077006,
      "tokens_seen": 545783808
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008429626063232226,
      "loss": 0.0827,
      "theoretical_loss": 3.9066298373766615,
      "tokens_seen": 546045952
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008428823623816402,
      "loss": 0.0831,
      "theoretical_loss": 3.9064323970403656,
      "tokens_seen": 546308096
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008428021184400578,
      "loss": 0.0832,
      "theoretical_loss": 3.9062350779353787,
      "tokens_seen": 546570240
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008427218744984754,
      "loss": 0.0825,
      "theoretical_loss": 3.906037879929174,
      "tokens_seen": 546832384
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000842641630556893,
      "loss": 0.0813,
      "theoretical_loss": 3.90584080288943,
      "tokens_seen": 547094528
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008425613866153105,
      "loss": 0.083,
      "theoretical_loss": 3.905643846684034,
      "tokens_seen": 547356672
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008424811426737282,
      "loss": 0.08,
      "theoretical_loss": 3.9054470111810815,
      "tokens_seen": 547618816
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008424008987321458,
      "loss": 0.0811,
      "theoretical_loss": 3.9052502962488735,
      "tokens_seen": 547880960
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008423206547905633,
      "loss": 0.0826,
      "theoretical_loss": 3.9050537017559197,
      "tokens_seen": 548143104
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000842240410848981,
      "loss": 0.0827,
      "theoretical_loss": 3.904857227570934,
      "tokens_seen": 548405248
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008421601669073985,
      "loss": 0.0788,
      "theoretical_loss": 3.904660873562837,
      "tokens_seen": 548667392
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008420799229658161,
      "loss": 0.0814,
      "theoretical_loss": 3.9044646396007545,
      "tokens_seen": 548929536
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008419996790242337,
      "loss": 0.0818,
      "theoretical_loss": 3.9042685255540177,
      "tokens_seen": 549191680
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008419194350826512,
      "loss": 0.0825,
      "theoretical_loss": 3.9040725312921616,
      "tokens_seen": 549453824
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008418391911410688,
      "loss": 0.0825,
      "theoretical_loss": 3.9038766566849263,
      "tokens_seen": 549715968
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008417589471994864,
      "loss": 0.0823,
      "theoretical_loss": 3.903680901602254,
      "tokens_seen": 549978112
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008416787032579041,
      "loss": 0.0811,
      "theoretical_loss": 3.9034852659142913,
      "tokens_seen": 550240256
    },
    {
      "epoch": 0.17,
      "objective/train/advantage_avg": 0.0010044012451544404,
      "objective/train/docs_used": 206347,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.6453677415847778,
      "objective/train/original_loss": 1.6453678607940674,
      "objective/train/theoretical_loss": 3.9032897494913876,
      "objective/train/tokens_used": 570962400,
      "objective/train/value_avg": -0.0096435546875,
      "objective/train/value_loss": 0.0002447115839459002,
      "objective/train/value_max": -0.00025916099548339844,
      "objective/train/value_min": -0.59619140625,
      "objective/train/value_reward_corr": 0.7088037903523504,
      "objective/train/value_std": 0.01555633544921875,
      "objective/train/weight_avg": 1.0011190176010132,
      "objective/train/weighted_lm_loss": 1.6466437578201294,
      "objective/train/weights_max": 1.6212894916534424,
      "objective/train/weights_min": 0.3786143362522125,
      "theoretical_loss": 3.9032897494913876,
      "tokens_seen": 550502400
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008415984593163216,
      "loss": 0.0808,
      "theoretical_loss": 3.9032897494913876,
      "tokens_seen": 550502400
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008415182153747393,
      "loss": 0.0809,
      "theoretical_loss": 3.9030943522040946,
      "tokens_seen": 550764544
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008414379714331568,
      "loss": 0.0847,
      "theoretical_loss": 3.902899073923166,
      "tokens_seen": 551026688
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008413577274915744,
      "loss": 0.0825,
      "theoretical_loss": 3.902703914519557,
      "tokens_seen": 551288832
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000841277483549992,
      "loss": 0.0825,
      "theoretical_loss": 3.9025088738644236,
      "tokens_seen": 551550976
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008411972396084095,
      "loss": 0.0796,
      "theoretical_loss": 3.9023139518291243,
      "tokens_seen": 551813120
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008411169956668272,
      "loss": 0.0806,
      "theoretical_loss": 3.902119148285216,
      "tokens_seen": 552075264
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008410367517252447,
      "loss": 0.0824,
      "theoretical_loss": 3.9019244631044563,
      "tokens_seen": 552337408
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008409565077836624,
      "loss": 0.0798,
      "theoretical_loss": 3.9017298961588027,
      "tokens_seen": 552599552
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00084087626384208,
      "loss": 0.0815,
      "theoretical_loss": 3.901535447320412,
      "tokens_seen": 552861696
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008407960199004975,
      "loss": 0.0809,
      "theoretical_loss": 3.901341116461639,
      "tokens_seen": 553123840
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008407157759589151,
      "loss": 0.0804,
      "theoretical_loss": 3.9011469034550372,
      "tokens_seen": 553385984
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008406355320173327,
      "loss": 0.0816,
      "theoretical_loss": 3.900952808173358,
      "tokens_seen": 553648128
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008405552880757503,
      "loss": 0.0834,
      "theoretical_loss": 3.900758830489551,
      "tokens_seen": 553910272
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008404750441341678,
      "loss": 0.0827,
      "theoretical_loss": 3.900564970276762,
      "tokens_seen": 554172416
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008403948001925855,
      "loss": 0.0809,
      "theoretical_loss": 3.9003712274083346,
      "tokens_seen": 554434560
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000840314556251003,
      "loss": 0.0822,
      "theoretical_loss": 3.9001776017578074,
      "tokens_seen": 554696704
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008402343123094207,
      "loss": 0.0812,
      "theoretical_loss": 3.899984093198916,
      "tokens_seen": 554958848
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008401540683678383,
      "loss": 0.0798,
      "theoretical_loss": 3.899790701605592,
      "tokens_seen": 555220992
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008400738244262558,
      "loss": 0.0807,
      "theoretical_loss": 3.899597426851961,
      "tokens_seen": 555483136
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008399935804846735,
      "loss": 0.078,
      "theoretical_loss": 3.899404268812343,
      "tokens_seen": 555745280
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000839913336543091,
      "loss": 0.0803,
      "theoretical_loss": 3.8992112273612545,
      "tokens_seen": 556007424
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008398330926015086,
      "loss": 0.0821,
      "theoretical_loss": 3.8990183023734044,
      "tokens_seen": 556269568
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008397528486599262,
      "loss": 0.0813,
      "theoretical_loss": 3.8988254937236952,
      "tokens_seen": 556531712
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008396726047183438,
      "loss": 0.0815,
      "theoretical_loss": 3.8986328012872233,
      "tokens_seen": 556793856
    },
    {
      "epoch": 0.17,
      "objective/train/advantage_avg": 0.002579332096502185,
      "objective/train/docs_used": 208740,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5783114433288574,
      "objective/train/original_loss": 1.5783112049102783,
      "objective/train/theoretical_loss": 3.8984402249392778,
      "objective/train/tokens_used": 577516000,
      "objective/train/value_avg": -0.00783538818359375,
      "objective/train/value_loss": 0.0001847637031460181,
      "objective/train/value_max": -0.0002868175506591797,
      "objective/train/value_min": -0.235107421875,
      "objective/train/value_reward_corr": 0.47073112766333935,
      "objective/train/value_std": 0.00836944580078125,
      "objective/train/weight_avg": 1.00266432762146,
      "objective/train/weighted_lm_loss": 1.5828243494033813,
      "objective/train/weights_max": 1.1927765607833862,
      "objective/train/weights_min": 0.36875197291374207,
      "theoretical_loss": 3.8984402249392778,
      "tokens_seen": 557056000
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008395923607767613,
      "loss": 0.079,
      "theoretical_loss": 3.8984402249392778,
      "tokens_seen": 557056000
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000839512116835179,
      "loss": 0.0781,
      "theoretical_loss": 3.8982477645553395,
      "tokens_seen": 557318144
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008394318728935966,
      "loss": 0.0796,
      "theoretical_loss": 3.898055420011082,
      "tokens_seen": 557580288
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008393516289520141,
      "loss": 0.0779,
      "theoretical_loss": 3.8978631911823705,
      "tokens_seen": 557842432
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008392713850104318,
      "loss": 0.0799,
      "theoretical_loss": 3.8976710779452612,
      "tokens_seen": 558104576
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008391911410688493,
      "loss": 0.0811,
      "theoretical_loss": 3.8974790801760015,
      "tokens_seen": 558366720
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008391108971272669,
      "loss": 0.0834,
      "theoretical_loss": 3.897287197751029,
      "tokens_seen": 558628864
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008390306531856845,
      "loss": 0.0828,
      "theoretical_loss": 3.897095430546971,
      "tokens_seen": 558891008
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000838950409244102,
      "loss": 0.083,
      "theoretical_loss": 3.896903778440646,
      "tokens_seen": 559153152
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008388701653025197,
      "loss": 0.0822,
      "theoretical_loss": 3.896712241309061,
      "tokens_seen": 559415296
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008387899213609372,
      "loss": 0.083,
      "theoretical_loss": 3.896520819029411,
      "tokens_seen": 559677440
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008387096774193549,
      "loss": 0.0816,
      "theoretical_loss": 3.896329511479082,
      "tokens_seen": 559939584
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008386294334777725,
      "loss": 0.0789,
      "theoretical_loss": 3.8961383185356455,
      "tokens_seen": 560201728
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008385491895361901,
      "loss": 0.0789,
      "theoretical_loss": 3.895947240076862,
      "tokens_seen": 560463872
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008384689455946076,
      "loss": 0.0782,
      "theoretical_loss": 3.895756275980681,
      "tokens_seen": 560726016
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008383887016530252,
      "loss": 0.0817,
      "theoretical_loss": 3.895565426125237,
      "tokens_seen": 560988160
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008383084577114428,
      "loss": 0.0809,
      "theoretical_loss": 3.8953746903888513,
      "tokens_seen": 561250304
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008382282137698603,
      "loss": 0.0785,
      "theoretical_loss": 3.895184068650033,
      "tokens_seen": 561512448
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000838147969828278,
      "loss": 0.0814,
      "theoretical_loss": 3.8949935607874764,
      "tokens_seen": 561774592
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008380677258866955,
      "loss": 0.0803,
      "theoretical_loss": 3.8948031666800613,
      "tokens_seen": 562036736
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008379874819451132,
      "loss": 0.0835,
      "theoretical_loss": 3.8946128862068528,
      "tokens_seen": 562298880
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008379072380035308,
      "loss": 0.0802,
      "theoretical_loss": 3.8944227192471006,
      "tokens_seen": 562561024
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008378269940619483,
      "loss": 0.0818,
      "theoretical_loss": 3.8942326656802395,
      "tokens_seen": 562823168
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000837746750120366,
      "loss": 0.0825,
      "theoretical_loss": 3.894042725385888,
      "tokens_seen": 563085312
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008376665061787835,
      "loss": 0.0834,
      "theoretical_loss": 3.893852898243849,
      "tokens_seen": 563347456
    },
    {
      "epoch": 0.17,
      "objective/train/advantage_avg": 0.0005252464907243848,
      "objective/train/docs_used": 211161,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.6151103973388672,
      "objective/train/original_loss": 1.6151103973388672,
      "objective/train/theoretical_loss": 3.8936631841341076,
      "objective/train/tokens_used": 584069600,
      "objective/train/value_avg": -0.0084381103515625,
      "objective/train/value_loss": 0.00027146536740474403,
      "objective/train/value_max": -0.0002613067626953125,
      "objective/train/value_min": -0.5791015625,
      "objective/train/value_reward_corr": 0.630790701887311,
      "objective/train/value_std": 0.01227569580078125,
      "objective/train/weight_avg": 1.0006484985351562,
      "objective/train/weighted_lm_loss": 1.6155797243118286,
      "objective/train/weights_max": 1.255659818649292,
      "objective/train/weights_min": 0.3694954514503479,
      "theoretical_loss": 3.8936631841341076,
      "tokens_seen": 563609600
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008375862622372011,
      "loss": 0.0798,
      "theoretical_loss": 3.8936631841341076,
      "tokens_seen": 563609600
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008375060182956187,
      "loss": 0.0812,
      "theoretical_loss": 3.893473582936833,
      "tokens_seen": 563871744
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008374257743540363,
      "loss": 0.0826,
      "theoretical_loss": 3.8932840945323774,
      "tokens_seen": 564133888
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008373455304124538,
      "loss": 0.0806,
      "theoretical_loss": 3.8930947188012737,
      "tokens_seen": 564396032
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008372652864708715,
      "loss": 0.0783,
      "theoretical_loss": 3.8929054556242377,
      "tokens_seen": 564658176
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008371850425292891,
      "loss": 0.0787,
      "theoretical_loss": 3.892716304882167,
      "tokens_seen": 564920320
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008371047985877066,
      "loss": 0.0811,
      "theoretical_loss": 3.892527266456141,
      "tokens_seen": 565182464
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008370245546461243,
      "loss": 0.0828,
      "theoretical_loss": 3.8923383402274174,
      "tokens_seen": 565444608
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008369443107045418,
      "loss": 0.0819,
      "theoretical_loss": 3.8921495260774375,
      "tokens_seen": 565706752
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008368640667629594,
      "loss": 0.0809,
      "theoretical_loss": 3.8919608238878216,
      "tokens_seen": 565968896
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000836783822821377,
      "loss": 0.077,
      "theoretical_loss": 3.891772233540369,
      "tokens_seen": 566231040
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008367035788797946,
      "loss": 0.0798,
      "theoretical_loss": 3.8915837549170584,
      "tokens_seen": 566493184
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008366233349382121,
      "loss": 0.079,
      "theoretical_loss": 3.89139538790005,
      "tokens_seen": 566755328
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008365430909966297,
      "loss": 0.0827,
      "theoretical_loss": 3.8912071323716795,
      "tokens_seen": 567017472
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008364628470550474,
      "loss": 0.0811,
      "theoretical_loss": 3.8910189882144626,
      "tokens_seen": 567279616
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000836382603113465,
      "loss": 0.0831,
      "theoretical_loss": 3.8908309553110936,
      "tokens_seen": 567541760
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008363023591718826,
      "loss": 0.0802,
      "theoretical_loss": 3.8906430335444426,
      "tokens_seen": 567803904
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008362221152303001,
      "loss": 0.0813,
      "theoretical_loss": 3.8904552227975593,
      "tokens_seen": 568066048
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008361418712887178,
      "loss": 0.082,
      "theoretical_loss": 3.8902675229536685,
      "tokens_seen": 568328192
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008360616273471353,
      "loss": 0.0823,
      "theoretical_loss": 3.8900799338961725,
      "tokens_seen": 568590336
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008359813834055528,
      "loss": 0.0832,
      "theoretical_loss": 3.8898924555086496,
      "tokens_seen": 568852480
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008359011394639705,
      "loss": 0.0809,
      "theoretical_loss": 3.8897050876748542,
      "tokens_seen": 569114624
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000835820895522388,
      "loss": 0.0821,
      "theoretical_loss": 3.8895178302787166,
      "tokens_seen": 569376768
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008357406515808057,
      "loss": 0.0814,
      "theoretical_loss": 3.8893306832043404,
      "tokens_seen": 569638912
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008356604076392233,
      "loss": 0.0833,
      "theoretical_loss": 3.8891436463360076,
      "tokens_seen": 569901056
    },
    {
      "epoch": 0.17,
      "objective/train/advantage_avg": 0.0007684892625547945,
      "objective/train/docs_used": 213119,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.636327862739563,
      "objective/train/original_loss": 1.6363275051116943,
      "objective/train/theoretical_loss": 3.8889567195581716,
      "objective/train/tokens_used": 590623200,
      "objective/train/value_avg": -0.00799560546875,
      "objective/train/value_loss": 0.00021641695639118552,
      "objective/train/value_max": -0.00025916099548339844,
      "objective/train/value_min": -0.65771484375,
      "objective/train/value_reward_corr": 0.6434202611001032,
      "objective/train/value_std": 0.0118408203125,
      "objective/train/weight_avg": 1.000868797302246,
      "objective/train/weighted_lm_loss": 1.6379215717315674,
      "objective/train/weights_max": 1.409537672996521,
      "objective/train/weights_min": 0.36900386214256287,
      "theoretical_loss": 3.8889567195581716,
      "tokens_seen": 570163200
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008355801636976409,
      "loss": 0.0818,
      "theoretical_loss": 3.8889567195581716,
      "tokens_seen": 570163200
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008354999197560584,
      "loss": 0.0787,
      "theoretical_loss": 3.8887699027554614,
      "tokens_seen": 570425344
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000835419675814476,
      "loss": 0.0802,
      "theoretical_loss": 3.8885831958126786,
      "tokens_seen": 570687488
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008353394318728936,
      "loss": 0.0823,
      "theoretical_loss": 3.8883965986148015,
      "tokens_seen": 570949632
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008352591879313111,
      "loss": 0.079,
      "theoretical_loss": 3.888210111046978,
      "tokens_seen": 571211776
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008351789439897288,
      "loss": 0.0822,
      "theoretical_loss": 3.8880237329945295,
      "tokens_seen": 571473920
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008350987000481463,
      "loss": 0.0836,
      "theoretical_loss": 3.887837464342952,
      "tokens_seen": 571736064
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008350184561065641,
      "loss": 0.0825,
      "theoretical_loss": 3.8876513049779113,
      "tokens_seen": 571998208
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008349382121649816,
      "loss": 0.0826,
      "theoretical_loss": 3.887465254785246,
      "tokens_seen": 572260352
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008348579682233991,
      "loss": 0.0836,
      "theoretical_loss": 3.887279313650967,
      "tokens_seen": 572522496
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008347777242818168,
      "loss": 0.0831,
      "theoretical_loss": 3.8870934814612546,
      "tokens_seen": 572784640
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008346974803402343,
      "loss": 0.0782,
      "theoretical_loss": 3.886907758102461,
      "tokens_seen": 573046784
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008346172363986519,
      "loss": 0.0829,
      "theoretical_loss": 3.8867221434611094,
      "tokens_seen": 573308928
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008345369924570695,
      "loss": 0.0817,
      "theoretical_loss": 3.8865366374238914,
      "tokens_seen": 573571072
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008344567485154871,
      "loss": 0.0816,
      "theoretical_loss": 3.88635123987767,
      "tokens_seen": 573833216
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008343765045739046,
      "loss": 0.0823,
      "theoretical_loss": 3.8861659507094766,
      "tokens_seen": 574095360
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008342962606323223,
      "loss": 0.0813,
      "theoretical_loss": 3.885980769806513,
      "tokens_seen": 574357504
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008342160166907399,
      "loss": 0.0844,
      "theoretical_loss": 3.8857956970561487,
      "tokens_seen": 574619648
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008341357727491574,
      "loss": 0.0785,
      "theoretical_loss": 3.8856107323459215,
      "tokens_seen": 574881792
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008340555288075751,
      "loss": 0.0819,
      "theoretical_loss": 3.8854258755635387,
      "tokens_seen": 575143936
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008339752848659926,
      "loss": 0.0829,
      "theoretical_loss": 3.885241126596874,
      "tokens_seen": 575406080
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008338950409244103,
      "loss": 0.0799,
      "theoretical_loss": 3.885056485333969,
      "tokens_seen": 575668224
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008338147969828278,
      "loss": 0.0828,
      "theoretical_loss": 3.884871951663034,
      "tokens_seen": 575930368
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008337345530412454,
      "loss": 0.0778,
      "theoretical_loss": 3.8846875254724442,
      "tokens_seen": 576192512
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000833654309099663,
      "loss": 0.0797,
      "theoretical_loss": 3.8845032066507414,
      "tokens_seen": 576454656
    },
    {
      "epoch": 0.17,
      "objective/train/advantage_avg": 0.0010946927359327674,
      "objective/train/docs_used": 215374,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.6169906854629517,
      "objective/train/original_loss": 1.6169908046722412,
      "objective/train/theoretical_loss": 3.884318995086635,
      "objective/train/tokens_used": 597176800,
      "objective/train/value_avg": -0.0078277587890625,
      "objective/train/value_loss": 0.0002228701050626114,
      "objective/train/value_max": -0.00022172927856445312,
      "objective/train/value_min": -0.49267578125,
      "objective/train/value_reward_corr": 0.7077948649588294,
      "objective/train/value_std": 0.01322174072265625,
      "objective/train/weight_avg": 1.0012009143829346,
      "objective/train/weighted_lm_loss": 1.6180263757705688,
      "objective/train/weights_max": 1.1417571306228638,
      "objective/train/weights_min": 0.3976996839046478,
      "theoretical_loss": 3.884318995086635,
      "tokens_seen": 576716800
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008335740651580805,
      "loss": 0.0791,
      "theoretical_loss": 3.884318995086635,
      "tokens_seen": 576716800
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008334938212164982,
      "loss": 0.0814,
      "theoretical_loss": 3.8841348906689985,
      "tokens_seen": 576978944
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0008334135772749158,
      "loss": 0.0828,
      "theoretical_loss": 3.8839508932868725,
      "tokens_seen": 577241088
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008333333333333334,
      "loss": 0.0803,
      "theoretical_loss": 3.8837670028294626,
      "tokens_seen": 577503232
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008332530893917509,
      "loss": 0.0822,
      "theoretical_loss": 3.883583219186138,
      "tokens_seen": 577765376
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008331728454501686,
      "loss": 0.082,
      "theoretical_loss": 3.8833995422464342,
      "tokens_seen": 578027520
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008330926015085861,
      "loss": 0.0818,
      "theoretical_loss": 3.88321597190005,
      "tokens_seen": 578289664
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008330123575670036,
      "loss": 0.0822,
      "theoretical_loss": 3.883032508036848,
      "tokens_seen": 578551808
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008329321136254213,
      "loss": 0.0793,
      "theoretical_loss": 3.882849150546856,
      "tokens_seen": 578813952
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008328518696838388,
      "loss": 0.0829,
      "theoretical_loss": 3.8826658993202625,
      "tokens_seen": 579076096
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008327716257422565,
      "loss": 0.0814,
      "theoretical_loss": 3.8824827542474214,
      "tokens_seen": 579338240
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008326913818006741,
      "loss": 0.0819,
      "theoretical_loss": 3.882299715218848,
      "tokens_seen": 579600384
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008326111378590917,
      "loss": 0.0813,
      "theoretical_loss": 3.8821167821252196,
      "tokens_seen": 579862528
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008325308939175093,
      "loss": 0.0828,
      "theoretical_loss": 3.8819339548573772,
      "tokens_seen": 580124672
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008324506499759268,
      "loss": 0.0815,
      "theoretical_loss": 3.881751233306322,
      "tokens_seen": 580386816
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008323704060343444,
      "loss": 0.0814,
      "theoretical_loss": 3.881568617363218,
      "tokens_seen": 580648960
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000832290162092762,
      "loss": 0.0824,
      "theoretical_loss": 3.881386106919389,
      "tokens_seen": 580911104
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008322099181511796,
      "loss": 0.0826,
      "theoretical_loss": 3.88120370186632,
      "tokens_seen": 581173248
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008321296742095971,
      "loss": 0.0802,
      "theoretical_loss": 3.881021402095657,
      "tokens_seen": 581435392
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008320494302680149,
      "loss": 0.0807,
      "theoretical_loss": 3.880839207499205,
      "tokens_seen": 581697536
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008319691863264324,
      "loss": 0.082,
      "theoretical_loss": 3.880657117968931,
      "tokens_seen": 581959680
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008318889423848499,
      "loss": 0.0844,
      "theoretical_loss": 3.880475133396959,
      "tokens_seen": 582221824
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008318086984432676,
      "loss": 0.0812,
      "theoretical_loss": 3.8802932536755748,
      "tokens_seen": 582483968
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008317284545016851,
      "loss": 0.0833,
      "theoretical_loss": 3.880111478697221,
      "tokens_seen": 582746112
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008316482105601027,
      "loss": 0.085,
      "theoretical_loss": 3.8799298083545004,
      "tokens_seen": 583008256
    },
    {
      "epoch": 0.18,
      "objective/train/advantage_avg": 0.0015177363529801369,
      "objective/train/docs_used": 217803,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.666736364364624,
      "objective/train/original_loss": 1.6667362451553345,
      "objective/train/theoretical_loss": 3.879748242540173,
      "objective/train/tokens_used": 603730400,
      "objective/train/value_avg": -0.011322021484375,
      "objective/train/value_loss": 0.0003124627110082656,
      "objective/train/value_max": -0.00027370452880859375,
      "objective/train/value_min": -0.642578125,
      "objective/train/value_reward_corr": 0.6764778613382025,
      "objective/train/value_std": 0.01788330078125,
      "objective/train/weight_avg": 1.0016603469848633,
      "objective/train/weighted_lm_loss": 1.668639898300171,
      "objective/train/weights_max": 1.3257246017456055,
      "objective/train/weights_min": 0.36870554089546204,
      "theoretical_loss": 3.879748242540173,
      "tokens_seen": 583270400
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008315679666185203,
      "loss": 0.0814,
      "theoretical_loss": 3.879748242540173,
      "tokens_seen": 583270400
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008314877226769379,
      "loss": 0.0839,
      "theoretical_loss": 3.8795667811471573,
      "tokens_seen": 583532544
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008314074787353555,
      "loss": 0.082,
      "theoretical_loss": 3.8793854240685306,
      "tokens_seen": 583794688
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000831327234793773,
      "loss": 0.0817,
      "theoretical_loss": 3.879204171197525,
      "tokens_seen": 584056832
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008312469908521907,
      "loss": 0.081,
      "theoretical_loss": 3.879023022427533,
      "tokens_seen": 584318976
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008311667469106083,
      "loss": 0.0814,
      "theoretical_loss": 3.878841977652101,
      "tokens_seen": 584581120
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008310865029690259,
      "loss": 0.0815,
      "theoretical_loss": 3.8786610367649343,
      "tokens_seen": 584843264
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008310062590274434,
      "loss": 0.0826,
      "theoretical_loss": 3.8784801996598928,
      "tokens_seen": 585105408
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008309260150858611,
      "loss": 0.0819,
      "theoretical_loss": 3.878299466230992,
      "tokens_seen": 585367552
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008308457711442786,
      "loss": 0.0808,
      "theoretical_loss": 3.8781188363724057,
      "tokens_seen": 585629696
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008307655272026961,
      "loss": 0.0807,
      "theoretical_loss": 3.87793830997846,
      "tokens_seen": 585891840
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008306852832611138,
      "loss": 0.0803,
      "theoretical_loss": 3.8777578869436384,
      "tokens_seen": 586153984
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008306050393195313,
      "loss": 0.0817,
      "theoretical_loss": 3.8775775671625765,
      "tokens_seen": 586416128
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000830524795377949,
      "loss": 0.082,
      "theoretical_loss": 3.8773973505300674,
      "tokens_seen": 586678272
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008304445514363666,
      "loss": 0.08,
      "theoretical_loss": 3.877217236941055,
      "tokens_seen": 586940416
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008303643074947842,
      "loss": 0.0833,
      "theoretical_loss": 3.877037226290641,
      "tokens_seen": 587202560
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008302840635532017,
      "loss": 0.0798,
      "theoretical_loss": 3.8768573184740767,
      "tokens_seen": 587464704
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008302038196116194,
      "loss": 0.0804,
      "theoretical_loss": 3.87667751338677,
      "tokens_seen": 587726848
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008301235756700369,
      "loss": 0.0799,
      "theoretical_loss": 3.8764978109242794,
      "tokens_seen": 587988992
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008300433317284545,
      "loss": 0.0819,
      "theoretical_loss": 3.8763182109823173,
      "tokens_seen": 588251136
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008299630877868721,
      "loss": 0.0761,
      "theoretical_loss": 3.8761387134567475,
      "tokens_seen": 588513280
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008298828438452896,
      "loss": 0.081,
      "theoretical_loss": 3.8759593182435874,
      "tokens_seen": 588775424
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008298025999037074,
      "loss": 0.0844,
      "theoretical_loss": 3.875780025239005,
      "tokens_seen": 589037568
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008297223559621249,
      "loss": 0.0793,
      "theoretical_loss": 3.8756008343393202,
      "tokens_seen": 589299712
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008296421120205425,
      "loss": 0.0832,
      "theoretical_loss": 3.8754217454410043,
      "tokens_seen": 589561856
    },
    {
      "epoch": 0.18,
      "objective/train/advantage_avg": 0.000526767922565341,
      "objective/train/docs_used": 220165,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.7038625478744507,
      "objective/train/original_loss": 1.7038625478744507,
      "objective/train/theoretical_loss": 3.875242758440679,
      "objective/train/tokens_used": 610284000,
      "objective/train/value_avg": -0.00890350341796875,
      "objective/train/value_loss": 0.0003305371792521328,
      "objective/train/value_max": -0.0002378225326538086,
      "objective/train/value_min": -0.6884765625,
      "objective/train/value_reward_corr": 0.6204838794811066,
      "objective/train/value_std": 0.0142974853515625,
      "objective/train/weight_avg": 1.000672459602356,
      "objective/train/weighted_lm_loss": 1.7044023275375366,
      "objective/train/weights_max": 1.3236627578735352,
      "objective/train/weights_min": 0.3694390654563904,
      "theoretical_loss": 3.875242758440679,
      "tokens_seen": 589824000
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008295618680789601,
      "loss": 0.0818,
      "theoretical_loss": 3.875242758440679,
      "tokens_seen": 589824000
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008294816241373776,
      "loss": 0.0807,
      "theoretical_loss": 3.875063873235117,
      "tokens_seen": 590086144
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008294013801957952,
      "loss": 0.085,
      "theoretical_loss": 3.874885089721242,
      "tokens_seen": 590348288
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008293211362542128,
      "loss": 0.0778,
      "theoretical_loss": 3.8747064077961264,
      "tokens_seen": 590610432
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008292408923126304,
      "loss": 0.0787,
      "theoretical_loss": 3.874527827356994,
      "tokens_seen": 590872576
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008291606483710479,
      "loss": 0.0828,
      "theoretical_loss": 3.8743493483012172,
      "tokens_seen": 591134720
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008290804044294657,
      "loss": 0.0803,
      "theoretical_loss": 3.874170970526317,
      "tokens_seen": 591396864
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008290001604878832,
      "loss": 0.0824,
      "theoretical_loss": 3.873992693929965,
      "tokens_seen": 591659008
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008289199165463007,
      "loss": 0.0781,
      "theoretical_loss": 3.8738145184099797,
      "tokens_seen": 591921152
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008288396726047184,
      "loss": 0.0793,
      "theoretical_loss": 3.8736364438643296,
      "tokens_seen": 592183296
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008287594286631359,
      "loss": 0.0831,
      "theoretical_loss": 3.87345847019113,
      "tokens_seen": 592445440
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008286791847215536,
      "loss": 0.0787,
      "theoretical_loss": 3.8732805972886446,
      "tokens_seen": 592707584
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008285989407799711,
      "loss": 0.0791,
      "theoretical_loss": 3.873102825055285,
      "tokens_seen": 592969728
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008285186968383887,
      "loss": 0.0793,
      "theoretical_loss": 3.87292515338961,
      "tokens_seen": 593231872
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008284384528968063,
      "loss": 0.08,
      "theoretical_loss": 3.872747582190324,
      "tokens_seen": 593494016
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008283582089552239,
      "loss": 0.0813,
      "theoretical_loss": 3.8725701113562794,
      "tokens_seen": 593756160
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008282779650136415,
      "loss": 0.0791,
      "theoretical_loss": 3.8723927407864758,
      "tokens_seen": 594018304
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008281977210720591,
      "loss": 0.0814,
      "theoretical_loss": 3.8722154703800573,
      "tokens_seen": 594280448
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008281174771304767,
      "loss": 0.082,
      "theoretical_loss": 3.8720383000363148,
      "tokens_seen": 594542592
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008280372331888942,
      "loss": 0.0807,
      "theoretical_loss": 3.871861229654684,
      "tokens_seen": 594804736
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008279569892473119,
      "loss": 0.0776,
      "theoretical_loss": 3.8716842591347476,
      "tokens_seen": 595066880
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008278767453057294,
      "loss": 0.0818,
      "theoretical_loss": 3.871507388376231,
      "tokens_seen": 595329024
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008277965013641469,
      "loss": 0.0792,
      "theoretical_loss": 3.871330617279006,
      "tokens_seen": 595591168
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008277162574225646,
      "loss": 0.0812,
      "theoretical_loss": 3.8711539457430897,
      "tokens_seen": 595853312
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008276360134809821,
      "loss": 0.0793,
      "theoretical_loss": 3.87097737366864,
      "tokens_seen": 596115456
    },
    {
      "epoch": 0.18,
      "objective/train/advantage_avg": 0.0013384453486651182,
      "objective/train/docs_used": 222574,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5089752674102783,
      "objective/train/original_loss": 1.5089752674102783,
      "objective/train/theoretical_loss": 3.870800900955963,
      "objective/train/tokens_used": 616837600,
      "objective/train/value_avg": -0.0093841552734375,
      "objective/train/value_loss": 0.0004224953299853951,
      "objective/train/value_max": -0.00018525123596191406,
      "objective/train/value_min": -0.80712890625,
      "objective/train/value_reward_corr": 0.6443973910376418,
      "objective/train/value_std": 0.0171661376953125,
      "objective/train/weight_avg": 1.0015227794647217,
      "objective/train/weighted_lm_loss": 1.5108550786972046,
      "objective/train/weights_max": 1.870554804801941,
      "objective/train/weights_min": 0.3708694577217102,
      "theoretical_loss": 3.870800900955963,
      "tokens_seen": 596377600
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008275557695393999,
      "loss": 0.079,
      "theoretical_loss": 3.870800900955963,
      "tokens_seen": 596377600
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008274755255978174,
      "loss": 0.0812,
      "theoretical_loss": 3.8706245275055062,
      "tokens_seen": 596639744
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000827395281656235,
      "loss": 0.0796,
      "theoretical_loss": 3.8704482532178606,
      "tokens_seen": 596901888
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008273150377146526,
      "loss": 0.0836,
      "theoretical_loss": 3.8702720779937607,
      "tokens_seen": 597164032
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008272347937730702,
      "loss": 0.079,
      "theoretical_loss": 3.8700960017340833,
      "tokens_seen": 597426176
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008271545498314877,
      "loss": 0.0797,
      "theoretical_loss": 3.8699200243398493,
      "tokens_seen": 597688320
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008270743058899053,
      "loss": 0.0845,
      "theoretical_loss": 3.8697441457122204,
      "tokens_seen": 597950464
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008269940619483229,
      "loss": 0.0819,
      "theoretical_loss": 3.8695683657525013,
      "tokens_seen": 598212608
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008269138180067404,
      "loss": 0.0823,
      "theoretical_loss": 3.8693926843621376,
      "tokens_seen": 598474752
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008268335740651582,
      "loss": 0.0802,
      "theoretical_loss": 3.8692171014427177,
      "tokens_seen": 598736896
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008267533301235757,
      "loss": 0.0794,
      "theoretical_loss": 3.86904161689597,
      "tokens_seen": 598999040
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008266730861819933,
      "loss": 0.0828,
      "theoretical_loss": 3.868866230623766,
      "tokens_seen": 599261184
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008265928422404109,
      "loss": 0.0806,
      "theoretical_loss": 3.8686909425281146,
      "tokens_seen": 599523328
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008265125982988284,
      "loss": 0.081,
      "theoretical_loss": 3.8685157525111684,
      "tokens_seen": 599785472
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000826432354357246,
      "loss": 0.0814,
      "theoretical_loss": 3.8683406604752184,
      "tokens_seen": 600047616
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008263521104156636,
      "loss": 0.0827,
      "theoretical_loss": 3.868165666322696,
      "tokens_seen": 600309760
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008262718664740812,
      "loss": 0.0826,
      "theoretical_loss": 3.8679907699561733,
      "tokens_seen": 600571904
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008261916225324988,
      "loss": 0.0819,
      "theoretical_loss": 3.86781597127836,
      "tokens_seen": 600834048
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008261113785909165,
      "loss": 0.0824,
      "theoretical_loss": 3.867641270192107,
      "tokens_seen": 601096192
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000826031134649334,
      "loss": 0.0831,
      "theoretical_loss": 3.867466666600402,
      "tokens_seen": 601358336
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008259508907077516,
      "loss": 0.0815,
      "theoretical_loss": 3.867292160406373,
      "tokens_seen": 601620480
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008258706467661692,
      "loss": 0.0796,
      "theoretical_loss": 3.8671177515132857,
      "tokens_seen": 601882624
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008257904028245867,
      "loss": 0.0821,
      "theoretical_loss": 3.866943439824545,
      "tokens_seen": 602144768
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008257101588830044,
      "loss": 0.0797,
      "theoretical_loss": 3.8667692252436914,
      "tokens_seen": 602406912
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008256299149414219,
      "loss": 0.0818,
      "theoretical_loss": 3.8665951076744056,
      "tokens_seen": 602669056
    },
    {
      "epoch": 0.18,
      "objective/train/advantage_avg": 0.0010131277376785874,
      "objective/train/docs_used": 224736,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.561912178993225,
      "objective/train/original_loss": 1.5619122982025146,
      "objective/train/theoretical_loss": 3.866421087020504,
      "objective/train/tokens_used": 623391200,
      "objective/train/value_avg": -0.006183624267578125,
      "objective/train/value_loss": 0.00039571512024849653,
      "objective/train/value_max": -0.00018966197967529297,
      "objective/train/value_min": -0.95263671875,
      "objective/train/value_reward_corr": 0.493842246751972,
      "objective/train/value_std": 0.01300811767578125,
      "objective/train/weight_avg": 1.0011917352676392,
      "objective/train/weighted_lm_loss": 1.5638322830200195,
      "objective/train/weights_max": 2.0086898803710938,
      "objective/train/weights_min": 0.36912351846694946,
      "theoretical_loss": 3.866421087020504,
      "tokens_seen": 602931200
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008255496709998395,
      "loss": 0.0792,
      "theoretical_loss": 3.866421087020504,
      "tokens_seen": 602931200
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008254694270582571,
      "loss": 0.0765,
      "theoretical_loss": 3.8662471631859407,
      "tokens_seen": 603193344
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008253891831166747,
      "loss": 0.0819,
      "theoretical_loss": 3.866073336074807,
      "tokens_seen": 603455488
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008253089391750923,
      "loss": 0.0785,
      "theoretical_loss": 3.8658996055913297,
      "tokens_seen": 603717632
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008252286952335099,
      "loss": 0.0833,
      "theoretical_loss": 3.8657259716398737,
      "tokens_seen": 603979776
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008251484512919275,
      "loss": 0.0798,
      "theoretical_loss": 3.8655524341249388,
      "tokens_seen": 604241920
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000825068207350345,
      "loss": 0.0825,
      "theoretical_loss": 3.865378992951161,
      "tokens_seen": 604504064
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008249879634087627,
      "loss": 0.0808,
      "theoretical_loss": 3.865205648023311,
      "tokens_seen": 604766208
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008249077194671802,
      "loss": 0.0818,
      "theoretical_loss": 3.8650323992462963,
      "tokens_seen": 605028352
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008248274755255978,
      "loss": 0.0803,
      "theoretical_loss": 3.8648592465251586,
      "tokens_seen": 605290496
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008247472315840154,
      "loss": 0.0813,
      "theoretical_loss": 3.864686189765075,
      "tokens_seen": 605552640
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000824666987642433,
      "loss": 0.0815,
      "theoretical_loss": 3.864513228871357,
      "tokens_seen": 605814784
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008245867437008507,
      "loss": 0.0804,
      "theoretical_loss": 3.8643403637494504,
      "tokens_seen": 606076928
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008245064997592682,
      "loss": 0.0794,
      "theoretical_loss": 3.8641675943049343,
      "tokens_seen": 606339072
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008244262558176858,
      "loss": 0.0827,
      "theoretical_loss": 3.863994920443523,
      "tokens_seen": 606601216
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008243460118761034,
      "loss": 0.0811,
      "theoretical_loss": 3.8638223420710647,
      "tokens_seen": 606863360
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000824265767934521,
      "loss": 0.0786,
      "theoretical_loss": 3.863649859093538,
      "tokens_seen": 607125504
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008241855239929385,
      "loss": 0.0793,
      "theoretical_loss": 3.863477471417059,
      "tokens_seen": 607387648
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008241052800513561,
      "loss": 0.0818,
      "theoretical_loss": 3.8633051789478734,
      "tokens_seen": 607649792
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008240250361097737,
      "loss": 0.0838,
      "theoretical_loss": 3.8631329815923605,
      "tokens_seen": 607911936
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008239447921681912,
      "loss": 0.0787,
      "theoretical_loss": 3.862960879257032,
      "tokens_seen": 608174080
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000823864548226609,
      "loss": 0.0828,
      "theoretical_loss": 3.8627888718485313,
      "tokens_seen": 608436224
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008237843042850265,
      "loss": 0.0792,
      "theoretical_loss": 3.862616959273635,
      "tokens_seen": 608698368
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008237040603434442,
      "loss": 0.0799,
      "theoretical_loss": 3.8624451414392498,
      "tokens_seen": 608960512
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008236238164018617,
      "loss": 0.0782,
      "theoretical_loss": 3.8622734182524154,
      "tokens_seen": 609222656
    },
    {
      "epoch": 0.18,
      "objective/train/advantage_avg": 0.0017172880470752716,
      "objective/train/docs_used": 227108,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5490598678588867,
      "objective/train/original_loss": 1.5490598678588867,
      "objective/train/theoretical_loss": 3.8621017896203007,
      "objective/train/tokens_used": 629944800,
      "objective/train/value_avg": -0.007549285888671875,
      "objective/train/value_loss": 0.00015352350601460785,
      "objective/train/value_max": -0.0002378225326538086,
      "objective/train/value_min": -0.423095703125,
      "objective/train/value_reward_corr": 0.6703752245048074,
      "objective/train/value_std": 0.011810302734375,
      "objective/train/weight_avg": 1.0017893314361572,
      "objective/train/weighted_lm_loss": 1.5520132780075073,
      "objective/train/weights_max": 1.3398975133895874,
      "objective/train/weights_min": 0.3707548677921295,
      "theoretical_loss": 3.8621017896203007,
      "tokens_seen": 609484800
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008235435724602792,
      "loss": 0.0801,
      "theoretical_loss": 3.8621017896203007,
      "tokens_seen": 609484800
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008234633285186969,
      "loss": 0.0791,
      "theoretical_loss": 3.8619302554502077,
      "tokens_seen": 609746944
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0008233830845771144,
      "loss": 0.0783,
      "theoretical_loss": 3.8617588156495666,
      "tokens_seen": 610009088
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000823302840635532,
      "loss": 0.0826,
      "theoretical_loss": 3.861587470125941,
      "tokens_seen": 610271232
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008232225966939496,
      "loss": 0.0807,
      "theoretical_loss": 3.8614162187870216,
      "tokens_seen": 610533376
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008231423527523673,
      "loss": 0.0801,
      "theoretical_loss": 3.861245061540631,
      "tokens_seen": 610795520
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008230621088107848,
      "loss": 0.0803,
      "theoretical_loss": 3.8610739982947218,
      "tokens_seen": 611057664
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008229818648692024,
      "loss": 0.077,
      "theoretical_loss": 3.8609030289573747,
      "tokens_seen": 611319808
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00082290162092762,
      "loss": 0.0822,
      "theoretical_loss": 3.8607321534368007,
      "tokens_seen": 611581952
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008228213769860375,
      "loss": 0.077,
      "theoretical_loss": 3.8605613716413396,
      "tokens_seen": 611844096
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008227411330444552,
      "loss": 0.082,
      "theoretical_loss": 3.860390683479459,
      "tokens_seen": 612106240
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008226608891028727,
      "loss": 0.0775,
      "theoretical_loss": 3.860220088859757,
      "tokens_seen": 612368384
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008225806451612903,
      "loss": 0.0771,
      "theoretical_loss": 3.860049587690958,
      "tokens_seen": 612630528
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008225004012197079,
      "loss": 0.0796,
      "theoretical_loss": 3.8598791798819154,
      "tokens_seen": 612892672
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008224201572781254,
      "loss": 0.0767,
      "theoretical_loss": 3.859708865341611,
      "tokens_seen": 613154816
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008223399133365432,
      "loss": 0.0815,
      "theoretical_loss": 3.8595386439791532,
      "tokens_seen": 613416960
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008222596693949607,
      "loss": 0.0792,
      "theoretical_loss": 3.859368515703778,
      "tokens_seen": 613679104
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008221794254533783,
      "loss": 0.0777,
      "theoretical_loss": 3.859198480424849,
      "tokens_seen": 613941248
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008220991815117959,
      "loss": 0.0783,
      "theoretical_loss": 3.859028538051856,
      "tokens_seen": 614203392
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008220189375702135,
      "loss": 0.0797,
      "theoretical_loss": 3.858858688494416,
      "tokens_seen": 614465536
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000821938693628631,
      "loss": 0.0789,
      "theoretical_loss": 3.8586889316622726,
      "tokens_seen": 614727680
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008218584496870486,
      "loss": 0.0758,
      "theoretical_loss": 3.8585192674652955,
      "tokens_seen": 614989824
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008217782057454662,
      "loss": 0.0785,
      "theoretical_loss": 3.8583496958134793,
      "tokens_seen": 615251968
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008216979618038837,
      "loss": 0.0766,
      "theoretical_loss": 3.8581802166169457,
      "tokens_seen": 615514112
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008216177178623015,
      "loss": 0.0798,
      "theoretical_loss": 3.8580108297859415,
      "tokens_seen": 615776256
    },
    {
      "epoch": 0.19,
      "objective/train/advantage_avg": -0.00047024516970850527,
      "objective/train/docs_used": 229528,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.600695252418518,
      "objective/train/original_loss": 1.6006953716278076,
      "objective/train/theoretical_loss": 3.857841535230839,
      "objective/train/tokens_used": 636498400,
      "objective/train/value_avg": -0.01091766357421875,
      "objective/train/value_loss": 0.00041529766167514026,
      "objective/train/value_max": -0.0002033710479736328,
      "objective/train/value_min": -0.474853515625,
      "objective/train/value_reward_corr": 0.8363627909970652,
      "objective/train/value_std": 0.0230560302734375,
      "objective/train/weight_avg": 0.9997208714485168,
      "objective/train/weighted_lm_loss": 1.6006137132644653,
      "objective/train/weights_max": 1.2145042419433594,
      "objective/train/weights_min": 0.23296229541301727,
      "theoretical_loss": 3.857841535230839,
      "tokens_seen": 616038400
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000821537473920719,
      "loss": 0.0775,
      "theoretical_loss": 3.857841535230839,
      "tokens_seen": 616038400
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008214572299791366,
      "loss": 0.0789,
      "theoretical_loss": 3.8576723328621347,
      "tokens_seen": 616300544
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008213769860375542,
      "loss": 0.0791,
      "theoretical_loss": 3.8575032225904513,
      "tokens_seen": 616562688
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008212967420959717,
      "loss": 0.0773,
      "theoretical_loss": 3.8573342043265346,
      "tokens_seen": 616824832
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008212164981543894,
      "loss": 0.083,
      "theoretical_loss": 3.857165277981256,
      "tokens_seen": 617086976
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008211362542128069,
      "loss": 0.0808,
      "theoretical_loss": 3.8569964434656105,
      "tokens_seen": 617349120
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008210560102712245,
      "loss": 0.0807,
      "theoretical_loss": 3.856827700690718,
      "tokens_seen": 617611264
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008209757663296422,
      "loss": 0.0794,
      "theoretical_loss": 3.8566590495678192,
      "tokens_seen": 617873408
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008208955223880598,
      "loss": 0.0816,
      "theoretical_loss": 3.856490490008282,
      "tokens_seen": 618135552
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008208152784464773,
      "loss": 0.0807,
      "theoretical_loss": 3.856322021923595,
      "tokens_seen": 618397696
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000820735034504895,
      "loss": 0.0798,
      "theoretical_loss": 3.8561536452253713,
      "tokens_seen": 618659840
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008206547905633125,
      "loss": 0.0769,
      "theoretical_loss": 3.855985359825346,
      "tokens_seen": 618921984
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00082057454662173,
      "loss": 0.0824,
      "theoretical_loss": 3.855817165635377,
      "tokens_seen": 619184128
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008204943026801477,
      "loss": 0.0818,
      "theoretical_loss": 3.8556490625674447,
      "tokens_seen": 619446272
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008204140587385652,
      "loss": 0.0812,
      "theoretical_loss": 3.855481050533651,
      "tokens_seen": 619708416
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008203338147969828,
      "loss": 0.0813,
      "theoretical_loss": 3.8553131294462206,
      "tokens_seen": 619970560
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008202535708554004,
      "loss": 0.0813,
      "theoretical_loss": 3.8551452992175,
      "tokens_seen": 620232704
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008201733269138181,
      "loss": 0.0813,
      "theoretical_loss": 3.8549775597599556,
      "tokens_seen": 620494848
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008200930829722356,
      "loss": 0.0786,
      "theoretical_loss": 3.8548099109861775,
      "tokens_seen": 620756992
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008200128390306532,
      "loss": 0.0808,
      "theoretical_loss": 3.8546423528088747,
      "tokens_seen": 621019136
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008199325950890708,
      "loss": 0.0801,
      "theoretical_loss": 3.8544748851408777,
      "tokens_seen": 621281280
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008198523511474884,
      "loss": 0.078,
      "theoretical_loss": 3.8543075078951388,
      "tokens_seen": 621543424
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000819772107205906,
      "loss": 0.078,
      "theoretical_loss": 3.8541402209847284,
      "tokens_seen": 621805568
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008196918632643235,
      "loss": 0.0815,
      "theoretical_loss": 3.8539730243228387,
      "tokens_seen": 622067712
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008196116193227412,
      "loss": 0.0774,
      "theoretical_loss": 3.8538059178227817,
      "tokens_seen": 622329856
    },
    {
      "epoch": 0.19,
      "objective/train/advantage_avg": 0.00011773208098020405,
      "objective/train/docs_used": 231894,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.6076971292495728,
      "objective/train/original_loss": 1.6076970100402832,
      "objective/train/theoretical_loss": 3.8536389013979893,
      "objective/train/tokens_used": 643052000,
      "objective/train/value_avg": -0.0093994140625,
      "objective/train/value_loss": 0.00031335835228674114,
      "objective/train/value_max": -0.00016868114471435547,
      "objective/train/value_min": -0.638671875,
      "objective/train/value_reward_corr": 0.6932073575502529,
      "objective/train/value_std": 0.01538848876953125,
      "objective/train/weight_avg": 1.000260829925537,
      "objective/train/weighted_lm_loss": 1.6074447631835938,
      "objective/train/weights_max": 1.6733942031860352,
      "objective/train/weights_min": 0.37257105112075806,
      "theoretical_loss": 3.8536389013979893,
      "tokens_seen": 622592000
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008195313753811587,
      "loss": 0.0831,
      "theoretical_loss": 3.8536389013979893,
      "tokens_seen": 622592000
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008194511314395762,
      "loss": 0.0791,
      "theoretical_loss": 3.8534719749620114,
      "tokens_seen": 622854144
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000819370887497994,
      "loss": 0.0785,
      "theoretical_loss": 3.8533051384285195,
      "tokens_seen": 623116288
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008192906435564115,
      "loss": 0.0811,
      "theoretical_loss": 3.853138391711303,
      "tokens_seen": 623378432
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008192103996148291,
      "loss": 0.079,
      "theoretical_loss": 3.852971734724269,
      "tokens_seen": 623640576
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008191301556732467,
      "loss": 0.0785,
      "theoretical_loss": 3.8528051673814456,
      "tokens_seen": 623902720
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008190499117316643,
      "loss": 0.0806,
      "theoretical_loss": 3.8526386895969775,
      "tokens_seen": 624164864
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008189696677900818,
      "loss": 0.0821,
      "theoretical_loss": 3.8524723012851294,
      "tokens_seen": 624427008
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008188894238484994,
      "loss": 0.0807,
      "theoretical_loss": 3.852306002360282,
      "tokens_seen": 624689152
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000818809179906917,
      "loss": 0.0811,
      "theoretical_loss": 3.852139792736936,
      "tokens_seen": 624951296
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008187289359653345,
      "loss": 0.0784,
      "theoretical_loss": 3.8519736723297067,
      "tokens_seen": 625213440
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008186486920237523,
      "loss": 0.0789,
      "theoretical_loss": 3.8518076410533304,
      "tokens_seen": 625475584
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008185684480821698,
      "loss": 0.0786,
      "theoretical_loss": 3.8516416988226574,
      "tokens_seen": 625737728
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008184882041405875,
      "loss": 0.079,
      "theoretical_loss": 3.851475845552658,
      "tokens_seen": 625999872
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000818407960199005,
      "loss": 0.0818,
      "theoretical_loss": 3.8513100811584158,
      "tokens_seen": 626262016
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008183277162574225,
      "loss": 0.0815,
      "theoretical_loss": 3.851144405555134,
      "tokens_seen": 626524160
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008182474723158402,
      "loss": 0.0787,
      "theoretical_loss": 3.85097881865813,
      "tokens_seen": 626786304
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008181672283742577,
      "loss": 0.0807,
      "theoretical_loss": 3.850813320382839,
      "tokens_seen": 627048448
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008180869844326753,
      "loss": 0.0809,
      "theoretical_loss": 3.8506479106448115,
      "tokens_seen": 627310592
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000818006740491093,
      "loss": 0.0823,
      "theoretical_loss": 3.8504825893597134,
      "tokens_seen": 627572736
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008179264965495106,
      "loss": 0.0802,
      "theoretical_loss": 3.850317356443326,
      "tokens_seen": 627834880
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008178462526079281,
      "loss": 0.0787,
      "theoretical_loss": 3.8501522118115465,
      "tokens_seen": 628097024
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008177660086663458,
      "loss": 0.0824,
      "theoretical_loss": 3.8499871553803873,
      "tokens_seen": 628359168
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008176857647247633,
      "loss": 0.0804,
      "theoretical_loss": 3.8498221870659743,
      "tokens_seen": 628621312
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008176055207831808,
      "loss": 0.0761,
      "theoretical_loss": 3.8496573067845503,
      "tokens_seen": 628883456
    },
    {
      "epoch": 0.19,
      "objective/train/advantage_avg": 0.0010866678785532713,
      "objective/train/docs_used": 234355,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5178707838058472,
      "objective/train/original_loss": 1.5178709030151367,
      "objective/train/theoretical_loss": 3.8494925144524705,
      "objective/train/tokens_used": 649605600,
      "objective/train/value_avg": -0.0084991455078125,
      "objective/train/value_loss": 0.0005936273955740035,
      "objective/train/value_max": -0.00015115737915039062,
      "objective/train/value_min": -0.94189453125,
      "objective/train/value_reward_corr": 0.5752126539490294,
      "objective/train/value_std": 0.0167694091796875,
      "objective/train/weight_avg": 1.001339316368103,
      "objective/train/weighted_lm_loss": 1.5189374685287476,
      "objective/train/weights_max": 1.8356927633285522,
      "objective/train/weights_min": 0.36808282136917114,
      "theoretical_loss": 3.8494925144524705,
      "tokens_seen": 629145600
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008175252768415985,
      "loss": 0.081,
      "theoretical_loss": 3.8494925144524705,
      "tokens_seen": 629145600
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000817445032900016,
      "loss": 0.0801,
      "theoretical_loss": 3.849327809986206,
      "tokens_seen": 629407744
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008173647889584337,
      "loss": 0.0767,
      "theoretical_loss": 3.8491631933023407,
      "tokens_seen": 629669888
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008172845450168512,
      "loss": 0.0814,
      "theoretical_loss": 3.8489986643175733,
      "tokens_seen": 629932032
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008172043010752689,
      "loss": 0.08,
      "theoretical_loss": 3.8488342229487156,
      "tokens_seen": 630194176
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008171240571336865,
      "loss": 0.0819,
      "theoretical_loss": 3.8486698691126935,
      "tokens_seen": 630456320
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000817043813192104,
      "loss": 0.0806,
      "theoretical_loss": 3.8485056027265454,
      "tokens_seen": 630718464
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008169635692505216,
      "loss": 0.0798,
      "theoretical_loss": 3.848341423707423,
      "tokens_seen": 630980608
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008168833253089392,
      "loss": 0.0788,
      "theoretical_loss": 3.8481773319725914,
      "tokens_seen": 631242752
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008168030813673568,
      "loss": 0.0829,
      "theoretical_loss": 3.8480133274394275,
      "tokens_seen": 631504896
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008167228374257743,
      "loss": 0.076,
      "theoretical_loss": 3.8478494100254217,
      "tokens_seen": 631767040
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000816642593484192,
      "loss": 0.0823,
      "theoretical_loss": 3.847685579648176,
      "tokens_seen": 632029184
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008165623495426095,
      "loss": 0.0821,
      "theoretical_loss": 3.847521836225404,
      "tokens_seen": 632291328
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000816482105601027,
      "loss": 0.0825,
      "theoretical_loss": 3.8473581796749317,
      "tokens_seen": 632553472
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008164018616594448,
      "loss": 0.077,
      "theoretical_loss": 3.8471946099146983,
      "tokens_seen": 632815616
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008163216177178623,
      "loss": 0.079,
      "theoretical_loss": 3.847031126862751,
      "tokens_seen": 633077760
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008162413737762799,
      "loss": 0.0788,
      "theoretical_loss": 3.8468677304372507,
      "tokens_seen": 633339904
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008161611298346975,
      "loss": 0.0777,
      "theoretical_loss": 3.8467044205564704,
      "tokens_seen": 633602048
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008160808858931151,
      "loss": 0.0799,
      "theoretical_loss": 3.846541197138791,
      "tokens_seen": 633864192
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008160006419515327,
      "loss": 0.0836,
      "theoretical_loss": 3.8463780601027056,
      "tokens_seen": 634126336
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008159203980099502,
      "loss": 0.0784,
      "theoretical_loss": 3.846215009366819,
      "tokens_seen": 634388480
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008158401540683678,
      "loss": 0.0809,
      "theoretical_loss": 3.846052044849843,
      "tokens_seen": 634650624
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008157599101267855,
      "loss": 0.0793,
      "theoretical_loss": 3.845889166470604,
      "tokens_seen": 634912768
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008156796661852031,
      "loss": 0.08,
      "theoretical_loss": 3.8457263741480343,
      "tokens_seen": 635174912
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008155994222436206,
      "loss": 0.0787,
      "theoretical_loss": 3.845563667801178,
      "tokens_seen": 635437056
    },
    {
      "epoch": 0.19,
      "objective/train/advantage_avg": 0.0005161234876140952,
      "objective/train/docs_used": 236526,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5930426120758057,
      "objective/train/original_loss": 1.5930426120758057,
      "objective/train/theoretical_loss": 3.8454010473491884,
      "objective/train/tokens_used": 656159200,
      "objective/train/value_avg": -0.007137298583984375,
      "objective/train/value_loss": 0.00014894510968588293,
      "objective/train/value_max": -0.00016224384307861328,
      "objective/train/value_min": -0.2247314453125,
      "objective/train/value_reward_corr": 0.6557688653083844,
      "objective/train/value_std": 0.01020050048828125,
      "objective/train/weight_avg": 1.0005874633789062,
      "objective/train/weighted_lm_loss": 1.5939151048660278,
      "objective/train/weights_max": 1.1805914640426636,
      "objective/train/weights_min": 0.3701472282409668,
      "theoretical_loss": 3.8454010473491884,
      "tokens_seen": 635699200
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008155191783020383,
      "loss": 0.0807,
      "theoretical_loss": 3.8454010473491884,
      "tokens_seen": 635699200
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008154389343604558,
      "loss": 0.0791,
      "theoretical_loss": 3.845238512711327,
      "tokens_seen": 635961344
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008153586904188733,
      "loss": 0.0815,
      "theoretical_loss": 3.845076063806966,
      "tokens_seen": 636223488
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000815278446477291,
      "loss": 0.0828,
      "theoretical_loss": 3.844913700555586,
      "tokens_seen": 636485632
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008151982025357085,
      "loss": 0.0779,
      "theoretical_loss": 3.8447514228767763,
      "tokens_seen": 636747776
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008151179585941261,
      "loss": 0.0827,
      "theoretical_loss": 3.844589230690234,
      "tokens_seen": 637009920
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008150377146525437,
      "loss": 0.0793,
      "theoretical_loss": 3.844427123915766,
      "tokens_seen": 637272064
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008149574707109614,
      "loss": 0.0779,
      "theoretical_loss": 3.8442651024732863,
      "tokens_seen": 637534208
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000814877226769379,
      "loss": 0.078,
      "theoretical_loss": 3.8441031662828173,
      "tokens_seen": 637796352
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008147969828277966,
      "loss": 0.0796,
      "theoretical_loss": 3.843941315264489,
      "tokens_seen": 638058496
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008147167388862141,
      "loss": 0.0786,
      "theoretical_loss": 3.8437795493385387,
      "tokens_seen": 638320640
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008146364949446317,
      "loss": 0.0812,
      "theoretical_loss": 3.8436178684253126,
      "tokens_seen": 638582784
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008145562510030493,
      "loss": 0.0816,
      "theoretical_loss": 3.843456272445262,
      "tokens_seen": 638844928
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008144760070614668,
      "loss": 0.0778,
      "theoretical_loss": 3.8432947613189468,
      "tokens_seen": 639107072
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008143957631198845,
      "loss": 0.0821,
      "theoretical_loss": 3.8431333349670336,
      "tokens_seen": 639369216
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000814315519178302,
      "loss": 0.0785,
      "theoretical_loss": 3.842971993310294,
      "tokens_seen": 639631360
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008142352752367197,
      "loss": 0.0789,
      "theoretical_loss": 3.8428107362696085,
      "tokens_seen": 639893504
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008141550312951373,
      "loss": 0.0778,
      "theoretical_loss": 3.842649563765962,
      "tokens_seen": 640155648
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008140747873535548,
      "loss": 0.0815,
      "theoretical_loss": 3.8424884757204474,
      "tokens_seen": 640417792
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008139945434119724,
      "loss": 0.0797,
      "theoretical_loss": 3.8423274720542606,
      "tokens_seen": 640679936
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00081391429947039,
      "loss": 0.0784,
      "theoretical_loss": 3.842166552688706,
      "tokens_seen": 640942080
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008138340555288076,
      "loss": 0.0803,
      "theoretical_loss": 3.8420057175451934,
      "tokens_seen": 641204224
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008137538115872251,
      "loss": 0.0807,
      "theoretical_loss": 3.841844966545236,
      "tokens_seen": 641466368
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008136735676456428,
      "loss": 0.08,
      "theoretical_loss": 3.841684299610453,
      "tokens_seen": 641728512
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008135933237040603,
      "loss": 0.0785,
      "theoretical_loss": 3.8415237166625698,
      "tokens_seen": 641990656
    },
    {
      "epoch": 0.19,
      "objective/train/advantage_avg": 0.00029344300855882466,
      "objective/train/docs_used": 239059,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.705520510673523,
      "objective/train/original_loss": 1.7055203914642334,
      "objective/train/theoretical_loss": 3.8413632176234147,
      "objective/train/tokens_used": 662712800,
      "objective/train/value_avg": -0.007610321044921875,
      "objective/train/value_loss": 0.0002510190533939749,
      "objective/train/value_max": -0.00013768672943115234,
      "objective/train/value_min": -0.431396484375,
      "objective/train/value_reward_corr": 0.6089659714411729,
      "objective/train/value_std": 0.0107879638671875,
      "objective/train/weight_avg": 1.000407338142395,
      "objective/train/weighted_lm_loss": 1.704552412033081,
      "objective/train/weights_max": 1.1669175624847412,
      "objective/train/weights_min": 0.24382223188877106,
      "theoretical_loss": 3.8413632176234147,
      "tokens_seen": 642252800
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000813513079762478,
      "loss": 0.0791,
      "theoretical_loss": 3.8413632176234147,
      "tokens_seen": 642252800
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008134328358208956,
      "loss": 0.0832,
      "theoretical_loss": 3.8412028024149225,
      "tokens_seen": 642514944
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008133525918793131,
      "loss": 0.0788,
      "theoretical_loss": 3.841042470959131,
      "tokens_seen": 642777088
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008132723479377308,
      "loss": 0.0788,
      "theoretical_loss": 3.840882223178183,
      "tokens_seen": 643039232
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0008131921039961483,
      "loss": 0.0809,
      "theoretical_loss": 3.8407220589943254,
      "tokens_seen": 643301376
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008131118600545659,
      "loss": 0.0775,
      "theoretical_loss": 3.8405619783299083,
      "tokens_seen": 643563520
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008130316161129835,
      "loss": 0.0781,
      "theoretical_loss": 3.8404019811073864,
      "tokens_seen": 643825664
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000812951372171401,
      "loss": 0.0784,
      "theoretical_loss": 3.840242067249317,
      "tokens_seen": 644087808
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008128711282298186,
      "loss": 0.0782,
      "theoretical_loss": 3.840082236678362,
      "tokens_seen": 644349952
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008127908842882363,
      "loss": 0.0797,
      "theoretical_loss": 3.8399224893172854,
      "tokens_seen": 644612096
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008127106403466539,
      "loss": 0.0789,
      "theoretical_loss": 3.839762825088955,
      "tokens_seen": 644874240
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008126303964050714,
      "loss": 0.0828,
      "theoretical_loss": 3.8396032439163394,
      "tokens_seen": 645136384
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008125501524634891,
      "loss": 0.0803,
      "theoretical_loss": 3.8394437457225132,
      "tokens_seen": 645398528
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008124699085219066,
      "loss": 0.0831,
      "theoretical_loss": 3.839284330430651,
      "tokens_seen": 645660672
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008123896645803241,
      "loss": 0.0806,
      "theoretical_loss": 3.8391249979640305,
      "tokens_seen": 645922816
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008123094206387418,
      "loss": 0.0798,
      "theoretical_loss": 3.8389657482460313,
      "tokens_seen": 646184960
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008122291766971593,
      "loss": 0.0771,
      "theoretical_loss": 3.838806581200134,
      "tokens_seen": 646447104
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000812148932755577,
      "loss": 0.0806,
      "theoretical_loss": 3.838647496749924,
      "tokens_seen": 646709248
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008120686888139945,
      "loss": 0.0815,
      "theoretical_loss": 3.8384884948190847,
      "tokens_seen": 646971392
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008119884448724122,
      "loss": 0.0805,
      "theoretical_loss": 3.838329575331403,
      "tokens_seen": 647233536
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008119082009308298,
      "loss": 0.0795,
      "theoretical_loss": 3.8381707382107657,
      "tokens_seen": 647495680
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008118279569892473,
      "loss": 0.0834,
      "theoretical_loss": 3.838011983381162,
      "tokens_seen": 647757824
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008117477130476649,
      "loss": 0.0788,
      "theoretical_loss": 3.8378533107666817,
      "tokens_seen": 648019968
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008116674691060825,
      "loss": 0.0831,
      "theoretical_loss": 3.8376947202915144,
      "tokens_seen": 648282112
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008115872251645001,
      "loss": 0.0797,
      "theoretical_loss": 3.8375362118799505,
      "tokens_seen": 648544256
    },
    {
      "epoch": 0.2,
      "objective/train/advantage_avg": 0.0012510496890172362,
      "objective/train/docs_used": 241544,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5836241245269775,
      "objective/train/original_loss": 1.5836241245269775,
      "objective/train/theoretical_loss": 3.8373777854563813,
      "objective/train/tokens_used": 669266400,
      "objective/train/value_avg": -0.00818634033203125,
      "objective/train/value_loss": 0.00038118340307846665,
      "objective/train/value_max": -0.0001150369644165039,
      "objective/train/value_min": -0.9169921875,
      "objective/train/value_reward_corr": 0.6984226450131933,
      "objective/train/value_std": 0.016143798828125,
      "objective/train/weight_avg": 1.0014175176620483,
      "objective/train/weighted_lm_loss": 1.5856133699417114,
      "objective/train/weights_max": 1.2598052024841309,
      "objective/train/weights_min": 0.25575754046440125,
      "theoretical_loss": 3.8373777854563813,
      "tokens_seen": 648806400
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008115069812229176,
      "loss": 0.0792,
      "theoretical_loss": 3.8373777854563813,
      "tokens_seen": 648806400
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008114267372813353,
      "loss": 0.0814,
      "theoretical_loss": 3.837219440945298,
      "tokens_seen": 649068544
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008113464933397528,
      "loss": 0.0798,
      "theoretical_loss": 3.8370611782712922,
      "tokens_seen": 649330688
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008112662493981705,
      "loss": 0.0821,
      "theoretical_loss": 3.8369029973590543,
      "tokens_seen": 649592832
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008111860054565881,
      "loss": 0.0793,
      "theoretical_loss": 3.836744898133376,
      "tokens_seen": 649854976
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008111057615150056,
      "loss": 0.0786,
      "theoretical_loss": 3.8365868805191456,
      "tokens_seen": 650117120
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008110255175734233,
      "loss": 0.0798,
      "theoretical_loss": 3.836428944441354,
      "tokens_seen": 650379264
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008109452736318408,
      "loss": 0.0803,
      "theoretical_loss": 3.8362710898250896,
      "tokens_seen": 650641408
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008108650296902584,
      "loss": 0.0817,
      "theoretical_loss": 3.83611331659554,
      "tokens_seen": 650903552
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000810784785748676,
      "loss": 0.0762,
      "theoretical_loss": 3.8359556246779913,
      "tokens_seen": 651165696
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008107045418070936,
      "loss": 0.0826,
      "theoretical_loss": 3.8357980139978283,
      "tokens_seen": 651427840
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008106242978655111,
      "loss": 0.0823,
      "theoretical_loss": 3.8356404844805354,
      "tokens_seen": 651689984
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008105440539239288,
      "loss": 0.0812,
      "theoretical_loss": 3.835483036051694,
      "tokens_seen": 651952128
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008104638099823464,
      "loss": 0.0799,
      "theoretical_loss": 3.835325668636983,
      "tokens_seen": 652214272
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008103835660407639,
      "loss": 0.0815,
      "theoretical_loss": 3.8351683821621814,
      "tokens_seen": 652476416
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008103033220991816,
      "loss": 0.0816,
      "theoretical_loss": 3.8350111765531647,
      "tokens_seen": 652738560
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008102230781575991,
      "loss": 0.0796,
      "theoretical_loss": 3.834854051735906,
      "tokens_seen": 653000704
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008101428342160167,
      "loss": 0.0823,
      "theoretical_loss": 3.8346970076364757,
      "tokens_seen": 653262848
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008100625902744343,
      "loss": 0.0795,
      "theoretical_loss": 3.8345400441810424,
      "tokens_seen": 653524992
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008099823463328518,
      "loss": 0.0816,
      "theoretical_loss": 3.8343831612958703,
      "tokens_seen": 653787136
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008099021023912694,
      "loss": 0.0815,
      "theoretical_loss": 3.834226358907322,
      "tokens_seen": 654049280
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000809821858449687,
      "loss": 0.0816,
      "theoretical_loss": 3.8340696369418565,
      "tokens_seen": 654311424
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008097416145081047,
      "loss": 0.0783,
      "theoretical_loss": 3.833912995326029,
      "tokens_seen": 654573568
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008096613705665223,
      "loss": 0.0789,
      "theoretical_loss": 3.8337564339864914,
      "tokens_seen": 654835712
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008095811266249399,
      "loss": 0.0798,
      "theoretical_loss": 3.8335999528499913,
      "tokens_seen": 655097856
    },
    {
      "debugging/Compilability": 0.9523809523809523,
      "debugging/distinct-1-grams": 0.7472213669987466,
      "debugging/entropy-1-grams": 5.482485850115401,
      "debugging/length": 435.9047619047619,
      "debugging/num_segments": 21,
      "debugging/raw_token_scores_avg": 0.0068328846246004105,
      "debugging/raw_token_scores_std": 0.01969163864850998,
      "debugging/score": 0.013429834759321626,
      "debugging/score_std": 0.01710154615280685,
      "epoch": 0.2,
      "objective/train/advantage_avg": 0.0014720445033162832,
      "objective/train/docs_used": 243976,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.65055251121521,
      "objective/train/original_loss": 1.65055251121521,
      "objective/train/theoretical_loss": 3.833443551843374,
      "objective/train/tokens_used": 675820000,
      "objective/train/value_avg": -0.00830841064453125,
      "objective/train/value_loss": 0.00020881656382698566,
      "objective/train/value_max": -0.00023055076599121094,
      "objective/train/value_min": -0.321533203125,
      "objective/train/value_reward_corr": 0.6838538864516404,
      "objective/train/value_std": 0.01381683349609375,
      "objective/train/weight_avg": 1.0015703439712524,
      "objective/train/weighted_lm_loss": 1.6532535552978516,
      "objective/train/weights_max": 1.1742674112319946,
      "objective/train/weights_min": 0.3706134557723999,
      "theoretical_loss": 3.833443551843374,
      "tokens_seen": 655360000
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008095008826833574,
      "loss": 0.0823,
      "theoretical_loss": 3.833443551843374,
      "tokens_seen": 655360000
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000809420638741775,
      "loss": 0.0799,
      "theoretical_loss": 3.8332872308935793,
      "tokens_seen": 655622144
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008093403948001926,
      "loss": 0.0809,
      "theoretical_loss": 3.833130989927643,
      "tokens_seen": 655884288
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008092601508586101,
      "loss": 0.0794,
      "theoretical_loss": 3.8329748288726972,
      "tokens_seen": 656146432
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008091799069170278,
      "loss": 0.0833,
      "theoretical_loss": 3.8328187476559687,
      "tokens_seen": 656408576
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008090996629754453,
      "loss": 0.0818,
      "theoretical_loss": 3.83266274620478,
      "tokens_seen": 656670720
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000809019419033863,
      "loss": 0.0781,
      "theoretical_loss": 3.832506824446549,
      "tokens_seen": 656932864
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008089391750922806,
      "loss": 0.0791,
      "theoretical_loss": 3.832350982308788,
      "tokens_seen": 657195008
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008088589311506981,
      "loss": 0.0781,
      "theoretical_loss": 3.8321952197191043,
      "tokens_seen": 657457152
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008087786872091157,
      "loss": 0.0813,
      "theoretical_loss": 3.8320395366052,
      "tokens_seen": 657719296
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008086984432675333,
      "loss": 0.0791,
      "theoretical_loss": 3.8318839328948715,
      "tokens_seen": 657981440
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008086181993259509,
      "loss": 0.0802,
      "theoretical_loss": 3.8317284085160095,
      "tokens_seen": 658243584
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008085379553843684,
      "loss": 0.0802,
      "theoretical_loss": 3.8315729633965994,
      "tokens_seen": 658505728
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008084577114427861,
      "loss": 0.0783,
      "theoretical_loss": 3.8314175974647195,
      "tokens_seen": 658767872
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008083774675012036,
      "loss": 0.0803,
      "theoretical_loss": 3.831262310648544,
      "tokens_seen": 659030016
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008082972235596214,
      "loss": 0.0813,
      "theoretical_loss": 3.831107102876338,
      "tokens_seen": 659292160
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008082169796180389,
      "loss": 0.0825,
      "theoretical_loss": 3.830951974076463,
      "tokens_seen": 659554304
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008081367356764564,
      "loss": 0.0821,
      "theoretical_loss": 3.830796924177371,
      "tokens_seen": 659816448
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008080564917348741,
      "loss": 0.079,
      "theoretical_loss": 3.830641953107609,
      "tokens_seen": 660078592
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008079762477932916,
      "loss": 0.0803,
      "theoretical_loss": 3.8304870607958175,
      "tokens_seen": 660340736
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008078960038517092,
      "loss": 0.0804,
      "theoretical_loss": 3.8303322471707286,
      "tokens_seen": 660602880
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008078157599101268,
      "loss": 0.0782,
      "theoretical_loss": 3.830177512161167,
      "tokens_seen": 660865024
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008077355159685444,
      "loss": 0.0815,
      "theoretical_loss": 3.8300228556960523,
      "tokens_seen": 661127168
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008076552720269619,
      "loss": 0.0774,
      "theoretical_loss": 3.829868277704393,
      "tokens_seen": 661389312
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008075750280853796,
      "loss": 0.0806,
      "theoretical_loss": 3.829713778115293,
      "tokens_seen": 661651456
    },
    {
      "epoch": 0.2,
      "objective/train/advantage_avg": 0.00017224658222403377,
      "objective/train/docs_used": 246530,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.7009872198104858,
      "objective/train/original_loss": 1.7009871006011963,
      "objective/train/theoretical_loss": 3.8295593568579465,
      "objective/train/tokens_used": 682373600,
      "objective/train/value_avg": -0.01042938232421875,
      "objective/train/value_loss": 0.00039018821553327143,
      "objective/train/value_max": -0.0001634359359741211,
      "objective/train/value_min": -0.291748046875,
      "objective/train/value_reward_corr": 0.6474135841444115,
      "objective/train/value_std": 0.01617431640625,
      "objective/train/weight_avg": 1.0003470182418823,
      "objective/train/weighted_lm_loss": 1.7011173963546753,
      "objective/train/weights_max": 1.2405765056610107,
      "objective/train/weights_min": 0.3682652413845062,
      "theoretical_loss": 3.8295593568579465,
      "tokens_seen": 661913600
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008074947841437972,
      "loss": 0.0814,
      "theoretical_loss": 3.8295593568579465,
      "tokens_seen": 661913600
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008074145402022147,
      "loss": 0.0811,
      "theoretical_loss": 3.8294050138616402,
      "tokens_seen": 662175744
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008073342962606324,
      "loss": 0.0806,
      "theoretical_loss": 3.8292507490557526,
      "tokens_seen": 662437888
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008072540523190499,
      "loss": 0.0764,
      "theoretical_loss": 3.8290965623697537,
      "tokens_seen": 662700032
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008071738083774676,
      "loss": 0.078,
      "theoretical_loss": 3.8289424537332053,
      "tokens_seen": 662962176
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008070935644358851,
      "loss": 0.0825,
      "theoretical_loss": 3.82878842307576,
      "tokens_seen": 663224320
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008070133204943026,
      "loss": 0.0794,
      "theoretical_loss": 3.828634470327162,
      "tokens_seen": 663486464
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008069330765527203,
      "loss": 0.0805,
      "theoretical_loss": 3.8284805954172474,
      "tokens_seen": 663748608
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008068528326111379,
      "loss": 0.081,
      "theoretical_loss": 3.828326798275941,
      "tokens_seen": 664010752
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008067725886695555,
      "loss": 0.0793,
      "theoretical_loss": 3.82817307883326,
      "tokens_seen": 664272896
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008066923447279731,
      "loss": 0.0795,
      "theoretical_loss": 3.8280194370193112,
      "tokens_seen": 664535040
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008066121007863907,
      "loss": 0.0796,
      "theoretical_loss": 3.827865872764293,
      "tokens_seen": 664797184
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008065318568448082,
      "loss": 0.0799,
      "theoretical_loss": 3.8277123859984936,
      "tokens_seen": 665059328
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008064516129032258,
      "loss": 0.0811,
      "theoretical_loss": 3.8275589766522895,
      "tokens_seen": 665321472
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008063713689616434,
      "loss": 0.0794,
      "theoretical_loss": 3.8274056446561504,
      "tokens_seen": 665583616
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008062911250200609,
      "loss": 0.0768,
      "theoretical_loss": 3.827252389940633,
      "tokens_seen": 665845760
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008062108810784786,
      "loss": 0.0787,
      "theoretical_loss": 3.827099212436386,
      "tokens_seen": 666107904
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008061306371368961,
      "loss": 0.0801,
      "theoretical_loss": 3.826946112074145,
      "tokens_seen": 666370048
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008060503931953138,
      "loss": 0.0789,
      "theoretical_loss": 3.826793088784737,
      "tokens_seen": 666632192
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008059701492537314,
      "loss": 0.0816,
      "theoretical_loss": 3.826640142499077,
      "tokens_seen": 666894336
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008058899053121489,
      "loss": 0.0782,
      "theoretical_loss": 3.8264872731481705,
      "tokens_seen": 667156480
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008058096613705666,
      "loss": 0.0785,
      "theoretical_loss": 3.8263344806631103,
      "tokens_seen": 667418624
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008057294174289841,
      "loss": 0.0802,
      "theoretical_loss": 3.8261817649750784,
      "tokens_seen": 667680768
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008056491734874017,
      "loss": 0.0778,
      "theoretical_loss": 3.8260291260153463,
      "tokens_seen": 667942912
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008055689295458193,
      "loss": 0.0796,
      "theoretical_loss": 3.8258765637152727,
      "tokens_seen": 668205056
    },
    {
      "epoch": 0.2,
      "objective/train/advantage_avg": 0.0008796442998573184,
      "objective/train/docs_used": 248744,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.6819145679473877,
      "objective/train/original_loss": 1.6819145679473877,
      "objective/train/theoretical_loss": 3.8257240780063055,
      "objective/train/tokens_used": 688927200,
      "objective/train/value_avg": -0.007282257080078125,
      "objective/train/value_loss": 0.00022768820053897798,
      "objective/train/value_max": -0.00021314620971679688,
      "objective/train/value_min": -0.29150390625,
      "objective/train/value_reward_corr": 0.70498133906636,
      "objective/train/value_std": 0.01126861572265625,
      "objective/train/weight_avg": 1.0009862184524536,
      "objective/train/weighted_lm_loss": 1.6842213869094849,
      "objective/train/weights_max": 1.2005013227462769,
      "objective/train/weights_min": 0.37098827958106995,
      "theoretical_loss": 3.8257240780063055,
      "tokens_seen": 668467200
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008054886856042369,
      "loss": 0.0805,
      "theoretical_loss": 3.8257240780063055,
      "tokens_seen": 668467200
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008054084416626544,
      "loss": 0.0779,
      "theoretical_loss": 3.8255716688199803,
      "tokens_seen": 668729344
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008053281977210722,
      "loss": 0.0788,
      "theoretical_loss": 3.825419336087921,
      "tokens_seen": 668991488
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008052479537794897,
      "loss": 0.0786,
      "theoretical_loss": 3.825267079741839,
      "tokens_seen": 669253632
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008051677098379072,
      "loss": 0.0792,
      "theoretical_loss": 3.825114899713533,
      "tokens_seen": 669515776
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008050874658963249,
      "loss": 0.0808,
      "theoretical_loss": 3.8249627959348915,
      "tokens_seen": 669777920
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008050072219547424,
      "loss": 0.0788,
      "theoretical_loss": 3.824810768337887,
      "tokens_seen": 670040064
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00080492697801316,
      "loss": 0.0798,
      "theoretical_loss": 3.8246588168545816,
      "tokens_seen": 670302208
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008048467340715776,
      "loss": 0.0745,
      "theoretical_loss": 3.824506941417125,
      "tokens_seen": 670564352
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008047664901299952,
      "loss": 0.0809,
      "theoretical_loss": 3.824355141957752,
      "tokens_seen": 670826496
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008046862461884128,
      "loss": 0.0798,
      "theoretical_loss": 3.8242034184087847,
      "tokens_seen": 671088640
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008046060022468304,
      "loss": 0.0772,
      "theoretical_loss": 3.824051770702633,
      "tokens_seen": 671350784
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000804525758305248,
      "loss": 0.0819,
      "theoretical_loss": 3.823900198771792,
      "tokens_seen": 671612928
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008044455143636656,
      "loss": 0.0765,
      "theoretical_loss": 3.823748702548845,
      "tokens_seen": 671875072
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008043652704220832,
      "loss": 0.0772,
      "theoretical_loss": 3.823597281966459,
      "tokens_seen": 672137216
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008042850264805007,
      "loss": 0.0785,
      "theoretical_loss": 3.8234459369573894,
      "tokens_seen": 672399360
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008042047825389184,
      "loss": 0.0792,
      "theoretical_loss": 3.8232946674544763,
      "tokens_seen": 672661504
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008041245385973359,
      "loss": 0.0802,
      "theoretical_loss": 3.8231434733906458,
      "tokens_seen": 672923648
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008040442946557534,
      "loss": 0.0799,
      "theoretical_loss": 3.822992354698911,
      "tokens_seen": 673185792
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008039640507141711,
      "loss": 0.0783,
      "theoretical_loss": 3.822841311312368,
      "tokens_seen": 673447936
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008038838067725887,
      "loss": 0.0791,
      "theoretical_loss": 3.8226903431642008,
      "tokens_seen": 673710080
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008038035628310063,
      "loss": 0.0789,
      "theoretical_loss": 3.8225394501876764,
      "tokens_seen": 673972224
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008037233188894239,
      "loss": 0.0798,
      "theoretical_loss": 3.822388632316149,
      "tokens_seen": 674234368
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008036430749478415,
      "loss": 0.0776,
      "theoretical_loss": 3.822237889483057,
      "tokens_seen": 674496512
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000803562831006259,
      "loss": 0.0791,
      "theoretical_loss": 3.822087221621923,
      "tokens_seen": 674758656
    },
    {
      "epoch": 0.2,
      "objective/train/advantage_avg": 0.00036516584805212915,
      "objective/train/docs_used": 251058,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5383315086364746,
      "objective/train/original_loss": 1.5383315086364746,
      "objective/train/theoretical_loss": 3.821936628666355,
      "objective/train/tokens_used": 695480800,
      "objective/train/value_avg": -0.0069732666015625,
      "objective/train/value_loss": 0.00019810259982477874,
      "objective/train/value_max": -0.0002269744873046875,
      "objective/train/value_min": -0.9658203125,
      "objective/train/value_reward_corr": 0.6702206529549529,
      "objective/train/value_std": 0.0112762451171875,
      "objective/train/weight_avg": 1.0004541873931885,
      "objective/train/weighted_lm_loss": 1.5389385223388672,
      "objective/train/weights_max": 1.1621463298797607,
      "objective/train/weights_min": 0.23926687240600586,
      "theoretical_loss": 3.821936628666355,
      "tokens_seen": 675020800
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008034825870646766,
      "loss": 0.075,
      "theoretical_loss": 3.821936628666355,
      "tokens_seen": 675020800
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008034023431230942,
      "loss": 0.0775,
      "theoretical_loss": 3.821786110550045,
      "tokens_seen": 675282944
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008033220991815118,
      "loss": 0.0766,
      "theoretical_loss": 3.8216356672067704,
      "tokens_seen": 675545088
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008032418552399294,
      "loss": 0.0791,
      "theoretical_loss": 3.821485298570391,
      "tokens_seen": 675807232
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000803161611298347,
      "loss": 0.0771,
      "theoretical_loss": 3.8213350045748533,
      "tokens_seen": 676069376
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0008030813673567647,
      "loss": 0.0784,
      "theoretical_loss": 3.821184785154186,
      "tokens_seen": 676331520
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008030011234151822,
      "loss": 0.0789,
      "theoretical_loss": 3.8210346402425013,
      "tokens_seen": 676593664
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008029208794735997,
      "loss": 0.0787,
      "theoretical_loss": 3.820884569773997,
      "tokens_seen": 676855808
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008028406355320174,
      "loss": 0.0768,
      "theoretical_loss": 3.8207345736829526,
      "tokens_seen": 677117952
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008027603915904349,
      "loss": 0.0778,
      "theoretical_loss": 3.820584651903732,
      "tokens_seen": 677380096
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008026801476488525,
      "loss": 0.0782,
      "theoretical_loss": 3.820434804370782,
      "tokens_seen": 677642240
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008025999037072701,
      "loss": 0.0778,
      "theoretical_loss": 3.820285031018633,
      "tokens_seen": 677904384
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008025196597656877,
      "loss": 0.0785,
      "theoretical_loss": 3.8201353317818985,
      "tokens_seen": 678166528
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008024394158241052,
      "loss": 0.0764,
      "theoretical_loss": 3.819985706595274,
      "tokens_seen": 678428672
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008023591718825229,
      "loss": 0.0815,
      "theoretical_loss": 3.8198361553935385,
      "tokens_seen": 678690816
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008022789279409405,
      "loss": 0.0812,
      "theoretical_loss": 3.8196866781115526,
      "tokens_seen": 678952960
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000802198683999358,
      "loss": 0.0828,
      "theoretical_loss": 3.8195372746842615,
      "tokens_seen": 679215104
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008021184400577757,
      "loss": 0.0766,
      "theoretical_loss": 3.8193879450466905,
      "tokens_seen": 679477248
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008020381961161932,
      "loss": 0.0797,
      "theoretical_loss": 3.819238689133948,
      "tokens_seen": 679739392
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008019579521746109,
      "loss": 0.0793,
      "theoretical_loss": 3.819089506881225,
      "tokens_seen": 680001536
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008018777082330284,
      "loss": 0.0789,
      "theoretical_loss": 3.8189403982237935,
      "tokens_seen": 680263680
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000801797464291446,
      "loss": 0.0772,
      "theoretical_loss": 3.818791363097008,
      "tokens_seen": 680525824
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008017172203498636,
      "loss": 0.0787,
      "theoretical_loss": 3.818642401436304,
      "tokens_seen": 680787968
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008016369764082812,
      "loss": 0.0791,
      "theoretical_loss": 3.8184935131771987,
      "tokens_seen": 681050112
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008015567324666988,
      "loss": 0.0784,
      "theoretical_loss": 3.8183446982552915,
      "tokens_seen": 681312256
    },
    {
      "epoch": 0.21,
      "objective/train/advantage_avg": 0.0005019235541112721,
      "objective/train/docs_used": 253498,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.601072072982788,
      "objective/train/original_loss": 1.6010721921920776,
      "objective/train/theoretical_loss": 3.818195956606262,
      "objective/train/tokens_used": 702034400,
      "objective/train/value_avg": -0.006389617919921875,
      "objective/train/value_loss": 0.00024236510216724128,
      "objective/train/value_max": -0.0001366138458251953,
      "objective/train/value_min": -0.63427734375,
      "objective/train/value_reward_corr": 0.4653045941913805,
      "objective/train/value_std": 0.0096893310546875,
      "objective/train/weight_avg": 1.0006093978881836,
      "objective/train/weighted_lm_loss": 1.6022027730941772,
      "objective/train/weights_max": 1.7492560148239136,
      "objective/train/weights_min": 0.37081003189086914,
      "theoretical_loss": 3.818195956606262,
      "tokens_seen": 681574400
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008014764885251164,
      "loss": 0.0804,
      "theoretical_loss": 3.818195956606262,
      "tokens_seen": 681574400
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000801396244583534,
      "loss": 0.0799,
      "theoretical_loss": 3.8180472881658707,
      "tokens_seen": 681836544
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008013160006419515,
      "loss": 0.0805,
      "theoretical_loss": 3.817898692869961,
      "tokens_seen": 682098688
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008012357567003692,
      "loss": 0.0809,
      "theoretical_loss": 3.817750170654455,
      "tokens_seen": 682360832
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008011555127587867,
      "loss": 0.0793,
      "theoretical_loss": 3.8176017214553575,
      "tokens_seen": 682622976
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008010752688172042,
      "loss": 0.0801,
      "theoretical_loss": 3.8174533452087513,
      "tokens_seen": 682885120
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008009950248756219,
      "loss": 0.0771,
      "theoretical_loss": 3.817305041850802,
      "tokens_seen": 683147264
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008009147809340394,
      "loss": 0.0804,
      "theoretical_loss": 3.817156811317754,
      "tokens_seen": 683409408
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008008345369924572,
      "loss": 0.0799,
      "theoretical_loss": 3.8170086535459333,
      "tokens_seen": 683671552
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008007542930508747,
      "loss": 0.0797,
      "theoretical_loss": 3.8168605684717454,
      "tokens_seen": 683933696
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008006740491092923,
      "loss": 0.0796,
      "theoretical_loss": 3.816712556031675,
      "tokens_seen": 684195840
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008005938051677099,
      "loss": 0.0788,
      "theoretical_loss": 3.816564616162287,
      "tokens_seen": 684457984
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008005135612261274,
      "loss": 0.0776,
      "theoretical_loss": 3.8164167488002265,
      "tokens_seen": 684720128
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000800433317284545,
      "loss": 0.0795,
      "theoretical_loss": 3.816268953882218,
      "tokens_seen": 684982272
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008003530733429626,
      "loss": 0.0775,
      "theoretical_loss": 3.8161212313450648,
      "tokens_seen": 685244416
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008002728294013802,
      "loss": 0.079,
      "theoretical_loss": 3.81597358112565,
      "tokens_seen": 685506560
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008001925854597977,
      "loss": 0.0796,
      "theoretical_loss": 3.815826003160935,
      "tokens_seen": 685768704
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0008001123415182155,
      "loss": 0.0809,
      "theoretical_loss": 3.815678497387962,
      "tokens_seen": 686030848
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000800032097576633,
      "loss": 0.0785,
      "theoretical_loss": 3.8155310637438506,
      "tokens_seen": 686292992
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007999518536350505,
      "loss": 0.0787,
      "theoretical_loss": 3.8153837021657995,
      "tokens_seen": 686555136
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007998716096934682,
      "loss": 0.0774,
      "theoretical_loss": 3.8152364125910863,
      "tokens_seen": 686817280
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007997913657518857,
      "loss": 0.0779,
      "theoretical_loss": 3.8150891949570664,
      "tokens_seen": 687079424
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007997111218103033,
      "loss": 0.078,
      "theoretical_loss": 3.814942049201175,
      "tokens_seen": 687341568
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007996308778687209,
      "loss": 0.0812,
      "theoretical_loss": 3.8147949752609236,
      "tokens_seen": 687603712
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007995506339271385,
      "loss": 0.0787,
      "theoretical_loss": 3.814647973073903,
      "tokens_seen": 687865856
    },
    {
      "epoch": 0.21,
      "objective/train/advantage_avg": 0.0018313382752239704,
      "objective/train/docs_used": 255827,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5675907135009766,
      "objective/train/original_loss": 1.5675907135009766,
      "objective/train/theoretical_loss": 3.8145010425777826,
      "objective/train/tokens_used": 708588000,
      "objective/train/value_avg": -0.01007080078125,
      "objective/train/value_loss": 0.00023178788251243532,
      "objective/train/value_max": -0.0001499652862548828,
      "objective/train/value_min": -0.5947265625,
      "objective/train/value_reward_corr": 0.7783312114068648,
      "objective/train/value_std": 0.018707275390625,
      "objective/train/weight_avg": 1.0019445419311523,
      "objective/train/weighted_lm_loss": 1.5707687139511108,
      "objective/train/weights_max": 1.2117313146591187,
      "objective/train/weights_min": 0.6250301003456116,
      "theoretical_loss": 3.8145010425777826,
      "tokens_seen": 688128000
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007994703899855562,
      "loss": 0.0784,
      "theoretical_loss": 3.8145010425777826,
      "tokens_seen": 688128000
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007993901460439737,
      "loss": 0.0793,
      "theoretical_loss": 3.814354183710308,
      "tokens_seen": 688390144
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007993099021023913,
      "loss": 0.0793,
      "theoretical_loss": 3.8142073964093046,
      "tokens_seen": 688652288
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007992296581608089,
      "loss": 0.0781,
      "theoretical_loss": 3.8140606806126733,
      "tokens_seen": 688914432
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007991494142192265,
      "loss": 0.0792,
      "theoretical_loss": 3.813914036258393,
      "tokens_seen": 689176576
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000799069170277644,
      "loss": 0.0786,
      "theoretical_loss": 3.813767463284522,
      "tokens_seen": 689438720
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007989889263360617,
      "loss": 0.0793,
      "theoretical_loss": 3.8136209616291934,
      "tokens_seen": 689700864
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007989086823944792,
      "loss": 0.0794,
      "theoretical_loss": 3.813474531230618,
      "tokens_seen": 689963008
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007988284384528968,
      "loss": 0.0782,
      "theoretical_loss": 3.8133281720270835,
      "tokens_seen": 690225152
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007987481945113144,
      "loss": 0.0802,
      "theoretical_loss": 3.8131818839569562,
      "tokens_seen": 690487296
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000798667950569732,
      "loss": 0.0804,
      "theoretical_loss": 3.8130356669586765,
      "tokens_seen": 690749440
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007985877066281496,
      "loss": 0.0765,
      "theoretical_loss": 3.8128895209707627,
      "tokens_seen": 691011584
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007985074626865672,
      "loss": 0.0817,
      "theoretical_loss": 3.81274344593181,
      "tokens_seen": 691273728
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007984272187449848,
      "loss": 0.0773,
      "theoretical_loss": 3.8125974417804893,
      "tokens_seen": 691535872
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007983469748034023,
      "loss": 0.0765,
      "theoretical_loss": 3.8124515084555477,
      "tokens_seen": 691798016
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00079826673086182,
      "loss": 0.0782,
      "theoretical_loss": 3.8123056458958087,
      "tokens_seen": 692060160
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007981864869202375,
      "loss": 0.0776,
      "theoretical_loss": 3.812159854040172,
      "tokens_seen": 692322304
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007981062429786551,
      "loss": 0.0793,
      "theoretical_loss": 3.8120141328276125,
      "tokens_seen": 692584448
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007980259990370727,
      "loss": 0.0808,
      "theoretical_loss": 3.8118684821971813,
      "tokens_seen": 692846592
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007979457550954902,
      "loss": 0.0803,
      "theoretical_loss": 3.8117229020880057,
      "tokens_seen": 693108736
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000797865511153908,
      "loss": 0.08,
      "theoretical_loss": 3.811577392439287,
      "tokens_seen": 693370880
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007977852672123255,
      "loss": 0.08,
      "theoretical_loss": 3.8114319531903025,
      "tokens_seen": 693633024
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007977050232707431,
      "loss": 0.0802,
      "theoretical_loss": 3.811286584280406,
      "tokens_seen": 693895168
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007976247793291607,
      "loss": 0.0797,
      "theoretical_loss": 3.8111412856490245,
      "tokens_seen": 694157312
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007975445353875782,
      "loss": 0.08,
      "theoretical_loss": 3.810996057235661,
      "tokens_seen": 694419456
    },
    {
      "epoch": 0.21,
      "objective/train/advantage_avg": 0.001043860800564289,
      "objective/train/docs_used": 258341,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.6668003797531128,
      "objective/train/original_loss": 1.6668004989624023,
      "objective/train/theoretical_loss": 3.810850898979894,
      "objective/train/tokens_used": 715141600,
      "objective/train/value_avg": -0.0081634521484375,
      "objective/train/value_loss": 0.00030596103169955313,
      "objective/train/value_max": -0.00021827220916748047,
      "objective/train/value_min": -0.765625,
      "objective/train/value_reward_corr": 0.5807898030559664,
      "objective/train/value_std": 0.01239776611328125,
      "objective/train/weight_avg": 1.001173973083496,
      "objective/train/weighted_lm_loss": 1.668200969696045,
      "objective/train/weights_max": 1.4868050813674927,
      "objective/train/weights_min": 0.22793063521385193,
      "theoretical_loss": 3.810850898979894,
      "tokens_seen": 694681600
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007974642914459958,
      "loss": 0.0797,
      "theoretical_loss": 3.810850898979894,
      "tokens_seen": 694681600
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007973840475044134,
      "loss": 0.0806,
      "theoretical_loss": 3.810705810821375,
      "tokens_seen": 694943744
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000797303803562831,
      "loss": 0.0794,
      "theoretical_loss": 3.8105607926998326,
      "tokens_seen": 695205888
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007972235596212485,
      "loss": 0.0782,
      "theoretical_loss": 3.810415844555067,
      "tokens_seen": 695468032
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007971433156796663,
      "loss": 0.0775,
      "theoretical_loss": 3.8102709663269554,
      "tokens_seen": 695730176
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007970630717380838,
      "loss": 0.0788,
      "theoretical_loss": 3.810126157955448,
      "tokens_seen": 695992320
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007969828277965014,
      "loss": 0.0777,
      "theoretical_loss": 3.809981419380569,
      "tokens_seen": 696254464
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000796902583854919,
      "loss": 0.0796,
      "theoretical_loss": 3.809836750542418,
      "tokens_seen": 696516608
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007968223399133365,
      "loss": 0.08,
      "theoretical_loss": 3.8096921513811663,
      "tokens_seen": 696778752
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007967420959717542,
      "loss": 0.08,
      "theoretical_loss": 3.809547621837061,
      "tokens_seen": 697040896
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007966618520301717,
      "loss": 0.0763,
      "theoretical_loss": 3.809403161850423,
      "tokens_seen": 697303040
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007965816080885893,
      "loss": 0.0822,
      "theoretical_loss": 3.8092587713616446,
      "tokens_seen": 697565184
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000796501364147007,
      "loss": 0.0808,
      "theoretical_loss": 3.809114450311193,
      "tokens_seen": 697827328
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007964211202054245,
      "loss": 0.0796,
      "theoretical_loss": 3.808970198639609,
      "tokens_seen": 698089472
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007963408762638421,
      "loss": 0.0778,
      "theoretical_loss": 3.808826016287507,
      "tokens_seen": 698351616
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007962606323222597,
      "loss": 0.0802,
      "theoretical_loss": 3.8086819031955725,
      "tokens_seen": 698613760
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007961803883806773,
      "loss": 0.0786,
      "theoretical_loss": 3.8085378593045665,
      "tokens_seen": 698875904
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007961001444390948,
      "loss": 0.0816,
      "theoretical_loss": 3.8083938845553202,
      "tokens_seen": 699138048
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007960199004975125,
      "loss": 0.0778,
      "theoretical_loss": 3.80824997888874,
      "tokens_seen": 699400192
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00079593965655593,
      "loss": 0.0807,
      "theoretical_loss": 3.8081061422458036,
      "tokens_seen": 699662336
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007958594126143476,
      "loss": 0.0784,
      "theoretical_loss": 3.8079623745675613,
      "tokens_seen": 699924480
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007957791686727652,
      "loss": 0.0752,
      "theoretical_loss": 3.8078186757951364,
      "tokens_seen": 700186624
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007956989247311828,
      "loss": 0.0805,
      "theoretical_loss": 3.8076750458697237,
      "tokens_seen": 700448768
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007956186807896005,
      "loss": 0.0788,
      "theoretical_loss": 3.807531484732591,
      "tokens_seen": 700710912
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000795538436848018,
      "loss": 0.0793,
      "theoretical_loss": 3.8073879923250775,
      "tokens_seen": 700973056
    },
    {
      "epoch": 0.21,
      "objective/train/advantage_avg": 0.0004075298202224076,
      "objective/train/docs_used": 260260,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5826354026794434,
      "objective/train/original_loss": 1.5826352834701538,
      "objective/train/theoretical_loss": 3.8072445685885947,
      "objective/train/tokens_used": 721695200,
      "objective/train/value_avg": -0.00783538818359375,
      "objective/train/value_loss": 0.00022773307864554226,
      "objective/train/value_max": -0.00013136863708496094,
      "objective/train/value_min": -0.63134765625,
      "objective/train/value_reward_corr": 0.635312677935756,
      "objective/train/value_std": 0.01226806640625,
      "objective/train/weight_avg": 1.0005120038986206,
      "objective/train/weighted_lm_loss": 1.5832562446594238,
      "objective/train/weights_max": 1.4071308374404907,
      "objective/train/weights_min": 0.36930587887763977,
      "theoretical_loss": 3.8072445685885947,
      "tokens_seen": 701235200
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007954581929064356,
      "loss": 0.0774,
      "theoretical_loss": 3.8072445685885947,
      "tokens_seen": 701235200
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007953779489648532,
      "loss": 0.0784,
      "theoretical_loss": 3.807101213464625,
      "tokens_seen": 701497344
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007952977050232708,
      "loss": 0.08,
      "theoretical_loss": 3.8069579268947242,
      "tokens_seen": 701759488
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007952174610816883,
      "loss": 0.0812,
      "theoretical_loss": 3.806814708820519,
      "tokens_seen": 702021632
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007951372171401059,
      "loss": 0.0758,
      "theoretical_loss": 3.806671559183706,
      "tokens_seen": 702283776
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007950569731985235,
      "loss": 0.0778,
      "theoretical_loss": 3.806528477926056,
      "tokens_seen": 702545920
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000794976729256941,
      "loss": 0.081,
      "theoretical_loss": 3.806385464989409,
      "tokens_seen": 702808064
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007948964853153588,
      "loss": 0.0806,
      "theoretical_loss": 3.806242520315676,
      "tokens_seen": 703070208
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007948162413737763,
      "loss": 0.0789,
      "theoretical_loss": 3.806099643846841,
      "tokens_seen": 703332352
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007947359974321939,
      "loss": 0.0789,
      "theoretical_loss": 3.8059568355249564,
      "tokens_seen": 703594496
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007946557534906115,
      "loss": 0.0786,
      "theoretical_loss": 3.8058140952921478,
      "tokens_seen": 703856640
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000794575509549029,
      "loss": 0.0794,
      "theoretical_loss": 3.805671423090609,
      "tokens_seen": 704118784
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007944952656074467,
      "loss": 0.0779,
      "theoretical_loss": 3.805528818862607,
      "tokens_seen": 704380928
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007944150216658642,
      "loss": 0.0777,
      "theoretical_loss": 3.8053862825504776,
      "tokens_seen": 704643072
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007943347777242818,
      "loss": 0.0805,
      "theoretical_loss": 3.8052438140966265,
      "tokens_seen": 704905216
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007942545337826995,
      "loss": 0.08,
      "theoretical_loss": 3.8051014134435315,
      "tokens_seen": 705167360
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007941742898411171,
      "loss": 0.0751,
      "theoretical_loss": 3.804959080533739,
      "tokens_seen": 705429504
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007940940458995346,
      "loss": 0.0788,
      "theoretical_loss": 3.8048168153098656,
      "tokens_seen": 705691648
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007940138019579522,
      "loss": 0.0784,
      "theoretical_loss": 3.8046746177145985,
      "tokens_seen": 705953792
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007939335580163698,
      "loss": 0.0796,
      "theoretical_loss": 3.804532487690694,
      "tokens_seen": 706215936
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007938533140747873,
      "loss": 0.078,
      "theoretical_loss": 3.8043904251809786,
      "tokens_seen": 706478080
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000793773070133205,
      "loss": 0.0793,
      "theoretical_loss": 3.8042484301283475,
      "tokens_seen": 706740224
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007936928261916225,
      "loss": 0.0805,
      "theoretical_loss": 3.8041065024757668,
      "tokens_seen": 707002368
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007936125822500401,
      "loss": 0.0788,
      "theoretical_loss": 3.8039646421662705,
      "tokens_seen": 707264512
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007935323383084578,
      "loss": 0.078,
      "theoretical_loss": 3.8038228491429624,
      "tokens_seen": 707526656
    },
    {
      "epoch": 0.21,
      "objective/train/advantage_avg": -0.0002181452582590282,
      "objective/train/docs_used": 262675,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.6911776065826416,
      "objective/train/original_loss": 1.6911776065826416,
      "objective/train/theoretical_loss": 3.8036811233490164,
      "objective/train/tokens_used": 728248800,
      "objective/train/value_avg": -0.0099334716796875,
      "objective/train/value_loss": 0.0003097241569776088,
      "objective/train/value_max": -0.00015723705291748047,
      "objective/train/value_min": -0.83837890625,
      "objective/train/value_reward_corr": 0.6940126679848521,
      "objective/train/value_std": 0.0159912109375,
      "objective/train/weight_avg": 0.9999237656593323,
      "objective/train/weighted_lm_loss": 1.6909995079040527,
      "objective/train/weights_max": 1.2376219034194946,
      "objective/train/weights_min": 0.3753642737865448,
      "theoretical_loss": 3.8036811233490164,
      "tokens_seen": 707788800
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007934520943668753,
      "loss": 0.08,
      "theoretical_loss": 3.8036811233490164,
      "tokens_seen": 707788800
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007933718504252929,
      "loss": 0.0775,
      "theoretical_loss": 3.803539464727673,
      "tokens_seen": 708050944
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007932916064837105,
      "loss": 0.0812,
      "theoretical_loss": 3.803397873222244,
      "tokens_seen": 708313088
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007932113625421281,
      "loss": 0.08,
      "theoretical_loss": 3.8032563487761095,
      "tokens_seen": 708575232
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007931311186005457,
      "loss": 0.0772,
      "theoretical_loss": 3.8031148913327177,
      "tokens_seen": 708837376
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007930508746589633,
      "loss": 0.0804,
      "theoretical_loss": 3.8029735008355843,
      "tokens_seen": 709099520
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0007929706307173808,
      "loss": 0.0799,
      "theoretical_loss": 3.8028321772282965,
      "tokens_seen": 709361664
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007928903867757984,
      "loss": 0.0789,
      "theoretical_loss": 3.8026909204545065,
      "tokens_seen": 709623808
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000792810142834216,
      "loss": 0.0811,
      "theoretical_loss": 3.8025497304579376,
      "tokens_seen": 709885952
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007927298988926336,
      "loss": 0.0782,
      "theoretical_loss": 3.80240860718238,
      "tokens_seen": 710148096
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007926496549510513,
      "loss": 0.0827,
      "theoretical_loss": 3.802267550571691,
      "tokens_seen": 710410240
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007925694110094688,
      "loss": 0.0818,
      "theoretical_loss": 3.802126560569798,
      "tokens_seen": 710672384
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007924891670678864,
      "loss": 0.0812,
      "theoretical_loss": 3.801985637120694,
      "tokens_seen": 710934528
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000792408923126304,
      "loss": 0.0785,
      "theoretical_loss": 3.801844780168441,
      "tokens_seen": 711196672
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007923286791847216,
      "loss": 0.0815,
      "theoretical_loss": 3.8017039896571685,
      "tokens_seen": 711458816
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007922484352431391,
      "loss": 0.0784,
      "theoretical_loss": 3.8015632655310734,
      "tokens_seen": 711720960
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007921681913015567,
      "loss": 0.08,
      "theoretical_loss": 3.8014226077344198,
      "tokens_seen": 711983104
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007920879473599743,
      "loss": 0.0775,
      "theoretical_loss": 3.8012820162115393,
      "tokens_seen": 712245248
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007920077034183918,
      "loss": 0.077,
      "theoretical_loss": 3.801141490906831,
      "tokens_seen": 712507392
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007919274594768096,
      "loss": 0.0777,
      "theoretical_loss": 3.80100103176476,
      "tokens_seen": 712769536
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007918472155352271,
      "loss": 0.0814,
      "theoretical_loss": 3.8008606387298594,
      "tokens_seen": 713031680
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007917669715936448,
      "loss": 0.0764,
      "theoretical_loss": 3.80072031174673,
      "tokens_seen": 713293824
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007916867276520623,
      "loss": 0.0802,
      "theoretical_loss": 3.800580050760036,
      "tokens_seen": 713555968
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007916064837104798,
      "loss": 0.0763,
      "theoretical_loss": 3.800439855714512,
      "tokens_seen": 713818112
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007915262397688975,
      "loss": 0.0798,
      "theoretical_loss": 3.8002997265549574,
      "tokens_seen": 714080256
    },
    {
      "epoch": 0.22,
      "objective/train/advantage_avg": 0.0012540040770545602,
      "objective/train/docs_used": 265036,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5773979425430298,
      "objective/train/original_loss": 1.5773980617523193,
      "objective/train/theoretical_loss": 3.8001596632262387,
      "objective/train/tokens_used": 734802400,
      "objective/train/value_avg": -0.01009368896484375,
      "objective/train/value_loss": 0.00023382111976388842,
      "objective/train/value_max": -0.00015115737915039062,
      "objective/train/value_min": -0.39892578125,
      "objective/train/value_reward_corr": 0.6356383073725016,
      "objective/train/value_std": 0.01374053955078125,
      "objective/train/weight_avg": 1.001361608505249,
      "objective/train/weighted_lm_loss": 1.57895028591156,
      "objective/train/weights_max": 1.2423335313796997,
      "objective/train/weights_min": 0.3686169385910034,
      "theoretical_loss": 3.8001596632262387,
      "tokens_seen": 714342400
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000791445995827315,
      "loss": 0.0799,
      "theoretical_loss": 3.8001596632262387,
      "tokens_seen": 714342400
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007913657518857326,
      "loss": 0.0776,
      "theoretical_loss": 3.8000196656732874,
      "tokens_seen": 714604544
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007912855079441503,
      "loss": 0.0778,
      "theoretical_loss": 3.7998797338411032,
      "tokens_seen": 714866688
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007912052640025679,
      "loss": 0.0794,
      "theoretical_loss": 3.7997398676747496,
      "tokens_seen": 715128832
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007911250200609854,
      "loss": 0.0795,
      "theoretical_loss": 3.7996000671193593,
      "tokens_seen": 715390976
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000791044776119403,
      "loss": 0.0774,
      "theoretical_loss": 3.7994603321201277,
      "tokens_seen": 715653120
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007909645321778206,
      "loss": 0.0785,
      "theoretical_loss": 3.7993206626223177,
      "tokens_seen": 715915264
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007908842882362381,
      "loss": 0.0787,
      "theoretical_loss": 3.799181058571258,
      "tokens_seen": 716177408
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007908040442946558,
      "loss": 0.0754,
      "theoretical_loss": 3.7990415199123424,
      "tokens_seen": 716439552
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007907238003530733,
      "loss": 0.0804,
      "theoretical_loss": 3.79890204659103,
      "tokens_seen": 716701696
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000790643556411491,
      "loss": 0.0781,
      "theoretical_loss": 3.7987626385528466,
      "tokens_seen": 716963840
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007905633124699085,
      "loss": 0.0815,
      "theoretical_loss": 3.798623295743382,
      "tokens_seen": 717225984
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000790483068528326,
      "loss": 0.0817,
      "theoretical_loss": 3.798484018108291,
      "tokens_seen": 717488128
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007904028245867438,
      "loss": 0.078,
      "theoretical_loss": 3.7983448055932953,
      "tokens_seen": 717750272
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007903225806451613,
      "loss": 0.0819,
      "theoretical_loss": 3.79820565814418,
      "tokens_seen": 718012416
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007902423367035789,
      "loss": 0.0782,
      "theoretical_loss": 3.798066575706795,
      "tokens_seen": 718274560
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007901620927619965,
      "loss": 0.0779,
      "theoretical_loss": 3.797927558227056,
      "tokens_seen": 718536704
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007900818488204141,
      "loss": 0.077,
      "theoretical_loss": 3.7977886056509433,
      "tokens_seen": 718798848
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007900016048788316,
      "loss": 0.078,
      "theoretical_loss": 3.797649717924502,
      "tokens_seen": 719060992
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007899213609372492,
      "loss": 0.0789,
      "theoretical_loss": 3.797510894993839,
      "tokens_seen": 719323136
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007898411169956668,
      "loss": 0.078,
      "theoretical_loss": 3.79737213680513,
      "tokens_seen": 719585280
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007897608730540844,
      "loss": 0.0787,
      "theoretical_loss": 3.797233443304612,
      "tokens_seen": 719847424
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007896806291125021,
      "loss": 0.0801,
      "theoretical_loss": 3.7970948144385868,
      "tokens_seen": 720109568
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007896003851709196,
      "loss": 0.0761,
      "theoretical_loss": 3.796956250153421,
      "tokens_seen": 720371712
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007895201412293372,
      "loss": 0.0796,
      "theoretical_loss": 3.796817750395544,
      "tokens_seen": 720633856
    },
    {
      "epoch": 0.22,
      "objective/train/advantage_avg": 0.0021576914004981518,
      "objective/train/docs_used": 267624,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.453436255455017,
      "objective/train/original_loss": 1.4534361362457275,
      "objective/train/theoretical_loss": 3.7966793151114504,
      "objective/train/tokens_used": 741356000,
      "objective/train/value_avg": -0.00920867919921875,
      "objective/train/value_loss": 0.0003623893717303872,
      "objective/train/value_max": -0.0002148151397705078,
      "objective/train/value_min": -0.8974609375,
      "objective/train/value_reward_corr": 0.59584870633317,
      "objective/train/value_std": 0.0142364501953125,
      "objective/train/weight_avg": 1.0023139715194702,
      "objective/train/weighted_lm_loss": 1.456945538520813,
      "objective/train/weights_max": 1.6503322124481201,
      "objective/train/weights_min": 0.2744547724723816,
      "theoretical_loss": 3.7966793151114504,
      "tokens_seen": 720896000
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007894398972877548,
      "loss": 0.0765,
      "theoretical_loss": 3.7966793151114504,
      "tokens_seen": 720896000
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007893596533461724,
      "loss": 0.0766,
      "theoretical_loss": 3.796540944247698,
      "tokens_seen": 721158144
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00078927940940459,
      "loss": 0.0778,
      "theoretical_loss": 3.796402637750908,
      "tokens_seen": 721420288
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007891991654630075,
      "loss": 0.0806,
      "theoretical_loss": 3.796264395567766,
      "tokens_seen": 721682432
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007891189215214251,
      "loss": 0.0784,
      "theoretical_loss": 3.7961262176450195,
      "tokens_seen": 721944576
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007890386775798428,
      "loss": 0.0797,
      "theoretical_loss": 3.795988103929482,
      "tokens_seen": 722206720
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007889584336382604,
      "loss": 0.0758,
      "theoretical_loss": 3.7958500543680276,
      "tokens_seen": 722468864
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007888781896966779,
      "loss": 0.0788,
      "theoretical_loss": 3.795712068907596,
      "tokens_seen": 722731008
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007887979457550956,
      "loss": 0.0822,
      "theoretical_loss": 3.795574147495188,
      "tokens_seen": 722993152
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007887177018135131,
      "loss": 0.0806,
      "theoretical_loss": 3.795436290077868,
      "tokens_seen": 723255296
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007886374578719306,
      "loss": 0.0809,
      "theoretical_loss": 3.795298496602765,
      "tokens_seen": 723517440
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007885572139303483,
      "loss": 0.0776,
      "theoretical_loss": 3.795160767017068,
      "tokens_seen": 723779584
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007884769699887658,
      "loss": 0.0785,
      "theoretical_loss": 3.795023101268031,
      "tokens_seen": 724041728
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007883967260471834,
      "loss": 0.0826,
      "theoretical_loss": 3.7948854993029695,
      "tokens_seen": 724303872
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000788316482105601,
      "loss": 0.0766,
      "theoretical_loss": 3.7947479610692616,
      "tokens_seen": 724566016
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007882362381640187,
      "loss": 0.0783,
      "theoretical_loss": 3.794610486514348,
      "tokens_seen": 724828160
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007881559942224363,
      "loss": 0.0788,
      "theoretical_loss": 3.7944730755857323,
      "tokens_seen": 725090304
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007880757502808538,
      "loss": 0.0772,
      "theoretical_loss": 3.794335728230979,
      "tokens_seen": 725352448
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007879955063392714,
      "loss": 0.0779,
      "theoretical_loss": 3.7941984443977157,
      "tokens_seen": 725614592
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000787915262397689,
      "loss": 0.0761,
      "theoretical_loss": 3.7940612240336327,
      "tokens_seen": 725876736
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007878350184561066,
      "loss": 0.0785,
      "theoretical_loss": 3.793924067086481,
      "tokens_seen": 726138880
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007877547745145241,
      "loss": 0.078,
      "theoretical_loss": 3.793786973504073,
      "tokens_seen": 726401024
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007876745305729418,
      "loss": 0.077,
      "theoretical_loss": 3.7936499432342847,
      "tokens_seen": 726663168
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007875942866313593,
      "loss": 0.0779,
      "theoretical_loss": 3.7935129762250526,
      "tokens_seen": 726925312
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007875140426897769,
      "loss": 0.0811,
      "theoretical_loss": 3.7933760724243752,
      "tokens_seen": 727187456
    },
    {
      "epoch": 0.22,
      "objective/train/advantage_avg": 0.001988545060157776,
      "objective/train/docs_used": 269904,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5989689826965332,
      "objective/train/original_loss": 1.5989689826965332,
      "objective/train/theoretical_loss": 3.7932392317803116,
      "objective/train/tokens_used": 747909600,
      "objective/train/value_avg": -0.0074920654296875,
      "objective/train/value_loss": 0.00017211709928233176,
      "objective/train/value_max": -0.0002397298812866211,
      "objective/train/value_min": -0.310302734375,
      "objective/train/value_reward_corr": 0.5813086882183613,
      "objective/train/value_std": 0.0099639892578125,
      "objective/train/weight_avg": 1.0020697116851807,
      "objective/train/weighted_lm_loss": 1.6021634340286255,
      "objective/train/weights_max": 1.3142492771148682,
      "objective/train/weights_min": 0.36833229660987854,
      "theoretical_loss": 3.7932392317803116,
      "tokens_seen": 727449600
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007874337987481946,
      "loss": 0.0795,
      "theoretical_loss": 3.7932392317803116,
      "tokens_seen": 727449600
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007873535548066121,
      "loss": 0.077,
      "theoretical_loss": 3.7931024542409837,
      "tokens_seen": 727711744
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007872733108650297,
      "loss": 0.0782,
      "theoretical_loss": 3.7929657397545733,
      "tokens_seen": 727973888
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007871930669234473,
      "loss": 0.0766,
      "theoretical_loss": 3.792829088269324,
      "tokens_seen": 728236032
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007871128229818649,
      "loss": 0.0768,
      "theoretical_loss": 3.792692499733541,
      "tokens_seen": 728498176
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007870325790402824,
      "loss": 0.0784,
      "theoretical_loss": 3.7925559740955896,
      "tokens_seen": 728760320
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007869523350987,
      "loss": 0.0775,
      "theoretical_loss": 3.7924195113038968,
      "tokens_seen": 729022464
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007868720911571176,
      "loss": 0.0786,
      "theoretical_loss": 3.7922831113069493,
      "tokens_seen": 729284608
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007867918472155353,
      "loss": 0.0761,
      "theoretical_loss": 3.792146774053296,
      "tokens_seen": 729546752
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007867116032739529,
      "loss": 0.0767,
      "theoretical_loss": 3.792010499491545,
      "tokens_seen": 729808896
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007866313593323704,
      "loss": 0.0752,
      "theoretical_loss": 3.7918742875703657,
      "tokens_seen": 730071040
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007865511153907881,
      "loss": 0.0762,
      "theoretical_loss": 3.7917381382384883,
      "tokens_seen": 730333184
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007864708714492056,
      "loss": 0.0767,
      "theoretical_loss": 3.791602051444703,
      "tokens_seen": 730595328
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007863906275076232,
      "loss": 0.075,
      "theoretical_loss": 3.791466027137859,
      "tokens_seen": 730857472
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007863103835660408,
      "loss": 0.0781,
      "theoretical_loss": 3.7913300652668678,
      "tokens_seen": 731119616
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007862301396244583,
      "loss": 0.0776,
      "theoretical_loss": 3.7911941657807002,
      "tokens_seen": 731381760
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007861498956828759,
      "loss": 0.0752,
      "theoretical_loss": 3.7910583286283854,
      "tokens_seen": 731643904
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007860696517412936,
      "loss": 0.0783,
      "theoretical_loss": 3.7909225537590157,
      "tokens_seen": 731906048
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007859894077997112,
      "loss": 0.0766,
      "theoretical_loss": 3.790786841121739,
      "tokens_seen": 732168192
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007859091638581287,
      "loss": 0.0765,
      "theoretical_loss": 3.7906511906657676,
      "tokens_seen": 732430336
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007858289199165464,
      "loss": 0.0764,
      "theoretical_loss": 3.7905156023403697,
      "tokens_seen": 732692480
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007857486759749639,
      "loss": 0.0781,
      "theoretical_loss": 3.7903800760948743,
      "tokens_seen": 732954624
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007856684320333814,
      "loss": 0.0776,
      "theoretical_loss": 3.790244611878671,
      "tokens_seen": 733216768
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007855881880917991,
      "loss": 0.0778,
      "theoretical_loss": 3.790109209641206,
      "tokens_seen": 733478912
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007855079441502166,
      "loss": 0.0768,
      "theoretical_loss": 3.7899738693319875,
      "tokens_seen": 733741056
    },
    {
      "epoch": 0.22,
      "objective/train/advantage_avg": 0.001371144549921155,
      "objective/train/docs_used": 272255,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.6625961065292358,
      "objective/train/original_loss": 1.6625961065292358,
      "objective/train/theoretical_loss": 3.7898385909005814,
      "objective/train/tokens_used": 754463200,
      "objective/train/value_avg": -0.007289886474609375,
      "objective/train/value_loss": 0.00022098013141658157,
      "objective/train/value_max": -8.547306060791016e-05,
      "objective/train/value_min": -0.287353515625,
      "objective/train/value_reward_corr": 0.5914789303006973,
      "objective/train/value_std": 0.0107879638671875,
      "objective/train/weight_avg": 1.0014703273773193,
      "objective/train/weighted_lm_loss": 1.6649521589279175,
      "objective/train/weights_max": 1.150571584701538,
      "objective/train/weights_min": 0.3686802387237549,
      "theoretical_loss": 3.7898385909005814,
      "tokens_seen": 734003200
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007854277002086343,
      "loss": 0.08,
      "theoretical_loss": 3.7898385909005814,
      "tokens_seen": 734003200
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007853474562670519,
      "loss": 0.0773,
      "theoretical_loss": 3.7897033742966135,
      "tokens_seen": 734265344
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007852672123254695,
      "loss": 0.0774,
      "theoretical_loss": 3.789568219469767,
      "tokens_seen": 734527488
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007851869683838871,
      "loss": 0.0798,
      "theoretical_loss": 3.789433126369786,
      "tokens_seen": 734789632
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007851067244423046,
      "loss": 0.0765,
      "theoretical_loss": 3.7892980949464716,
      "tokens_seen": 735051776
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007850264805007222,
      "loss": 0.0779,
      "theoretical_loss": 3.7891631251496856,
      "tokens_seen": 735313920
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007849462365591398,
      "loss": 0.0772,
      "theoretical_loss": 3.7890282169293465,
      "tokens_seen": 735576064
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007848659926175574,
      "loss": 0.0762,
      "theoretical_loss": 3.7888933702354324,
      "tokens_seen": 735838208
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007847857486759749,
      "loss": 0.0792,
      "theoretical_loss": 3.7887585850179786,
      "tokens_seen": 736100352
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007847055047343926,
      "loss": 0.0784,
      "theoretical_loss": 3.788623861227081,
      "tokens_seen": 736362496
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007846252607928101,
      "loss": 0.0751,
      "theoretical_loss": 3.7884891988128926,
      "tokens_seen": 736624640
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007845450168512277,
      "loss": 0.0778,
      "theoretical_loss": 3.7883545977256228,
      "tokens_seen": 736886784
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007844647729096454,
      "loss": 0.0781,
      "theoretical_loss": 3.7882200579155416,
      "tokens_seen": 737148928
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007843845289680629,
      "loss": 0.0788,
      "theoretical_loss": 3.788085579332977,
      "tokens_seen": 737411072
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007843042850264806,
      "loss": 0.0772,
      "theoretical_loss": 3.787951161928312,
      "tokens_seen": 737673216
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007842240410848981,
      "loss": 0.0782,
      "theoretical_loss": 3.7878168056519916,
      "tokens_seen": 737935360
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007841437971433157,
      "loss": 0.0766,
      "theoretical_loss": 3.787682510454515,
      "tokens_seen": 738197504
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007840635532017333,
      "loss": 0.0777,
      "theoretical_loss": 3.7875482762864405,
      "tokens_seen": 738459648
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007839833092601508,
      "loss": 0.0757,
      "theoretical_loss": 3.787414103098384,
      "tokens_seen": 738721792
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007839030653185684,
      "loss": 0.0765,
      "theoretical_loss": 3.7872799908410193,
      "tokens_seen": 738983936
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007838228213769861,
      "loss": 0.0761,
      "theoretical_loss": 3.787145939465076,
      "tokens_seen": 739246080
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007837425774354037,
      "loss": 0.0771,
      "theoretical_loss": 3.7870119489213425,
      "tokens_seen": 739508224
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007836623334938212,
      "loss": 0.0794,
      "theoretical_loss": 3.786878019160664,
      "tokens_seen": 739770368
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007835820895522389,
      "loss": 0.0761,
      "theoretical_loss": 3.7867441501339427,
      "tokens_seen": 740032512
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007835018456106564,
      "loss": 0.0781,
      "theoretical_loss": 3.7866103417921373,
      "tokens_seen": 740294656
    },
    {
      "epoch": 0.22,
      "objective/train/advantage_avg": 0.0015989760868251324,
      "objective/train/docs_used": 274540,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.6382532119750977,
      "objective/train/original_loss": 1.6382529735565186,
      "objective/train/theoretical_loss": 3.786476594086265,
      "objective/train/tokens_used": 761016800,
      "objective/train/value_avg": -0.0094757080078125,
      "objective/train/value_loss": 0.0003717961080837995,
      "objective/train/value_max": -0.00016605854034423828,
      "objective/train/value_min": -0.59716796875,
      "objective/train/value_reward_corr": 0.6324069891531287,
      "objective/train/value_std": 0.0156707763671875,
      "objective/train/weight_avg": 1.0017673969268799,
      "objective/train/weighted_lm_loss": 1.6408437490463257,
      "objective/train/weights_max": 1.6781408786773682,
      "objective/train/weights_min": 0.37242892384529114,
      "theoretical_loss": 3.786476594086265,
      "tokens_seen": 740556800
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007834216016690739,
      "loss": 0.0818,
      "theoretical_loss": 3.786476594086265,
      "tokens_seen": 740556800
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007833413577274916,
      "loss": 0.076,
      "theoretical_loss": 3.7863429069673984,
      "tokens_seen": 740818944
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007832611137859091,
      "loss": 0.0773,
      "theoretical_loss": 3.7862092803866663,
      "tokens_seen": 741081088
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007831808698443267,
      "loss": 0.0743,
      "theoretical_loss": 3.786075714295257,
      "tokens_seen": 741343232
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007831006259027444,
      "loss": 0.0797,
      "theoretical_loss": 3.7859422086444123,
      "tokens_seen": 741605376
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000783020381961162,
      "loss": 0.0777,
      "theoretical_loss": 3.7858087633854325,
      "tokens_seen": 741867520
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007829401380195796,
      "loss": 0.077,
      "theoretical_loss": 3.785675378469673,
      "tokens_seen": 742129664
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0007828598940779972,
      "loss": 0.0771,
      "theoretical_loss": 3.7855420538485474,
      "tokens_seen": 742391808
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007827796501364147,
      "loss": 0.0786,
      "theoretical_loss": 3.7854087894735233,
      "tokens_seen": 742653952
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007826994061948323,
      "loss": 0.0749,
      "theoretical_loss": 3.7852755852961257,
      "tokens_seen": 742916096
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007826191622532499,
      "loss": 0.0789,
      "theoretical_loss": 3.785142441267936,
      "tokens_seen": 743178240
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007825389183116674,
      "loss": 0.0776,
      "theoretical_loss": 3.7850093573405905,
      "tokens_seen": 743440384
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007824586743700851,
      "loss": 0.076,
      "theoretical_loss": 3.7848763334657827,
      "tokens_seen": 743702528
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007823784304285027,
      "loss": 0.0805,
      "theoretical_loss": 3.7847433695952617,
      "tokens_seen": 743964672
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007822981864869203,
      "loss": 0.0783,
      "theoretical_loss": 3.7846104656808306,
      "tokens_seen": 744226816
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007822179425453379,
      "loss": 0.0757,
      "theoretical_loss": 3.7844776216743505,
      "tokens_seen": 744488960
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007821376986037554,
      "loss": 0.077,
      "theoretical_loss": 3.784344837527737,
      "tokens_seen": 744751104
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000782057454662173,
      "loss": 0.0759,
      "theoretical_loss": 3.784212113192961,
      "tokens_seen": 745013248
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007819772107205906,
      "loss": 0.0773,
      "theoretical_loss": 3.7840794486220495,
      "tokens_seen": 745275392
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007818969667790082,
      "loss": 0.0777,
      "theoretical_loss": 3.783946843767084,
      "tokens_seen": 745537536
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007818167228374257,
      "loss": 0.0785,
      "theoretical_loss": 3.783814298580203,
      "tokens_seen": 745799680
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007817364788958434,
      "loss": 0.0772,
      "theoretical_loss": 3.7836818130135974,
      "tokens_seen": 746061824
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000781656234954261,
      "loss": 0.0756,
      "theoretical_loss": 3.783549387019515,
      "tokens_seen": 746323968
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007815759910126786,
      "loss": 0.0779,
      "theoretical_loss": 3.7834170205502584,
      "tokens_seen": 746586112
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007814957470710962,
      "loss": 0.0764,
      "theoretical_loss": 3.783284713558186,
      "tokens_seen": 746848256
    },
    {
      "epoch": 0.23,
      "objective/train/advantage_avg": 0.001808889559470117,
      "objective/train/docs_used": 277055,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.6275434494018555,
      "objective/train/original_loss": 1.627543330192566,
      "objective/train/theoretical_loss": 3.783152465995708,
      "objective/train/tokens_used": 767570400,
      "objective/train/value_avg": -0.01079559326171875,
      "objective/train/value_loss": 0.0003174249141011387,
      "objective/train/value_max": -0.00019109249114990234,
      "objective/train/value_min": -0.4189453125,
      "objective/train/value_reward_corr": 0.6689077552958134,
      "objective/train/value_std": 0.01580810546875,
      "objective/train/weight_avg": 1.001956582069397,
      "objective/train/weighted_lm_loss": 1.6307531595230103,
      "objective/train/weights_max": 1.1777127981185913,
      "objective/train/weights_min": 0.3695334792137146,
      "theoretical_loss": 3.783152465995708,
      "tokens_seen": 747110400
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007814155031295137,
      "loss": 0.079,
      "theoretical_loss": 3.783152465995708,
      "tokens_seen": 747110400
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007813352591879314,
      "loss": 0.077,
      "theoretical_loss": 3.7830202778152935,
      "tokens_seen": 747372544
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007812550152463489,
      "loss": 0.0748,
      "theoretical_loss": 3.7828881489694632,
      "tokens_seen": 747634688
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007811747713047665,
      "loss": 0.0756,
      "theoretical_loss": 3.7827560794107926,
      "tokens_seen": 747896832
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007810945273631841,
      "loss": 0.0765,
      "theoretical_loss": 3.7826240690919137,
      "tokens_seen": 748158976
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007810142834216016,
      "loss": 0.0792,
      "theoretical_loss": 3.7824921179655115,
      "tokens_seen": 748421120
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007809340394800192,
      "loss": 0.0773,
      "theoretical_loss": 3.782360225984325,
      "tokens_seen": 748683264
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007808537955384369,
      "loss": 0.0769,
      "theoretical_loss": 3.782228393101149,
      "tokens_seen": 748945408
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007807735515968545,
      "loss": 0.0775,
      "theoretical_loss": 3.78209661926883,
      "tokens_seen": 749207552
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000780693307655272,
      "loss": 0.0771,
      "theoretical_loss": 3.781964904440271,
      "tokens_seen": 749469696
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007806130637136897,
      "loss": 0.0795,
      "theoretical_loss": 3.7818332485684283,
      "tokens_seen": 749731840
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007805328197721072,
      "loss": 0.0772,
      "theoretical_loss": 3.781701651606311,
      "tokens_seen": 749993984
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007804525758305248,
      "loss": 0.0793,
      "theoretical_loss": 3.7815701135069846,
      "tokens_seen": 750256128
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007803723318889424,
      "loss": 0.0765,
      "theoretical_loss": 3.7814386342235653,
      "tokens_seen": 750518272
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007802920879473599,
      "loss": 0.0726,
      "theoretical_loss": 3.7813072137092254,
      "tokens_seen": 750780416
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007802118440057776,
      "loss": 0.0749,
      "theoretical_loss": 3.7811758519171894,
      "tokens_seen": 751042560
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007801316000641952,
      "loss": 0.0788,
      "theoretical_loss": 3.781044548800736,
      "tokens_seen": 751304704
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007800513561226128,
      "loss": 0.0794,
      "theoretical_loss": 3.7809133043131973,
      "tokens_seen": 751566848
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007799711121810304,
      "loss": 0.0784,
      "theoretical_loss": 3.7807821184079584,
      "tokens_seen": 751828992
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000779890868239448,
      "loss": 0.0775,
      "theoretical_loss": 3.780650991038459,
      "tokens_seen": 752091136
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007798106242978655,
      "loss": 0.0773,
      "theoretical_loss": 3.7805199221581893,
      "tokens_seen": 752353280
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007797303803562831,
      "loss": 0.0796,
      "theoretical_loss": 3.7803889117206957,
      "tokens_seen": 752615424
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007796501364147007,
      "loss": 0.0788,
      "theoretical_loss": 3.7802579596795756,
      "tokens_seen": 752877568
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007795698924731182,
      "loss": 0.0753,
      "theoretical_loss": 3.7801270659884807,
      "tokens_seen": 753139712
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007794896485315359,
      "loss": 0.0779,
      "theoretical_loss": 3.7799962306011143,
      "tokens_seen": 753401856
    },
    {
      "epoch": 0.23,
      "objective/train/advantage_avg": 0.000318468373734504,
      "objective/train/docs_used": 279492,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5279823541641235,
      "objective/train/original_loss": 1.527982473373413,
      "objective/train/theoretical_loss": 3.779865453471234,
      "objective/train/tokens_used": 774124000,
      "objective/train/value_avg": -0.009857177734375,
      "objective/train/value_loss": 0.0003417794650886208,
      "objective/train/value_max": -0.0001767873764038086,
      "objective/train/value_min": -0.51904296875,
      "objective/train/value_reward_corr": 0.737366687695921,
      "objective/train/value_std": 0.01702880859375,
      "objective/train/weight_avg": 1.0004727840423584,
      "objective/train/weighted_lm_loss": 1.528694748878479,
      "objective/train/weights_max": 1.1768189668655396,
      "objective/train/weights_min": 0.3686802387237549,
      "theoretical_loss": 3.779865453471234,
      "tokens_seen": 753664000
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007794094045899534,
      "loss": 0.079,
      "theoretical_loss": 3.779865453471234,
      "tokens_seen": 753664000
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007793291606483711,
      "loss": 0.0776,
      "theoretical_loss": 3.7797347345526484,
      "tokens_seen": 753926144
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007792489167067887,
      "loss": 0.0775,
      "theoretical_loss": 3.7796040737992205,
      "tokens_seen": 754188288
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007791686727652062,
      "loss": 0.0782,
      "theoretical_loss": 3.7794734711648648,
      "tokens_seen": 754450432
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007790884288236239,
      "loss": 0.0791,
      "theoretical_loss": 3.779342926603549,
      "tokens_seen": 754712576
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007790081848820414,
      "loss": 0.0767,
      "theoretical_loss": 3.7792124400692924,
      "tokens_seen": 754974720
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000778927940940459,
      "loss": 0.0761,
      "theoretical_loss": 3.7790820115161674,
      "tokens_seen": 755236864
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007788476969988766,
      "loss": 0.079,
      "theoretical_loss": 3.778951640898298,
      "tokens_seen": 755499008
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007787674530572942,
      "loss": 0.0779,
      "theoretical_loss": 3.7788213281698617,
      "tokens_seen": 755761152
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007786872091157117,
      "loss": 0.0758,
      "theoretical_loss": 3.778691073285086,
      "tokens_seen": 756023296
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007786069651741294,
      "loss": 0.0796,
      "theoretical_loss": 3.7785608761982523,
      "tokens_seen": 756285440
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000778526721232547,
      "loss": 0.0763,
      "theoretical_loss": 3.7784307368636934,
      "tokens_seen": 756547584
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007784464772909645,
      "loss": 0.0754,
      "theoretical_loss": 3.7783006552357934,
      "tokens_seen": 756809728
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007783662333493822,
      "loss": 0.0791,
      "theoretical_loss": 3.7781706312689893,
      "tokens_seen": 757071872
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007782859894077997,
      "loss": 0.0778,
      "theoretical_loss": 3.7780406649177696,
      "tokens_seen": 757334016
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007782057454662173,
      "loss": 0.0771,
      "theoretical_loss": 3.777910756136673,
      "tokens_seen": 757596160
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007781255015246349,
      "loss": 0.0779,
      "theoretical_loss": 3.777780904880292,
      "tokens_seen": 757858304
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007780452575830524,
      "loss": 0.0791,
      "theoretical_loss": 3.777651111103269,
      "tokens_seen": 758120448
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007779650136414702,
      "loss": 0.0767,
      "theoretical_loss": 3.777521374760298,
      "tokens_seen": 758382592
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007778847696998877,
      "loss": 0.0794,
      "theoretical_loss": 3.7773916958061253,
      "tokens_seen": 758644736
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007778045257583053,
      "loss": 0.081,
      "theoretical_loss": 3.777262074195548,
      "tokens_seen": 758906880
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007777242818167229,
      "loss": 0.079,
      "theoretical_loss": 3.777132509883413,
      "tokens_seen": 759169024
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007776440378751405,
      "loss": 0.0786,
      "theoretical_loss": 3.7770030028246215,
      "tokens_seen": 759431168
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000777563793933558,
      "loss": 0.0776,
      "theoretical_loss": 3.7768735529741226,
      "tokens_seen": 759693312
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007774835499919756,
      "loss": 0.0756,
      "theoretical_loss": 3.776744160286918,
      "tokens_seen": 759955456
    },
    {
      "epoch": 0.23,
      "objective/train/advantage_avg": -0.00012553480337373912,
      "objective/train/docs_used": 281753,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.536434531211853,
      "objective/train/original_loss": 1.5364346504211426,
      "objective/train/theoretical_loss": 3.77661482471806,
      "objective/train/tokens_used": 780677600,
      "objective/train/value_avg": -0.009521484375,
      "objective/train/value_loss": 0.0005207026842981577,
      "objective/train/value_max": -0.00010311603546142578,
      "objective/train/value_min": -0.755859375,
      "objective/train/value_reward_corr": 0.614135921117643,
      "objective/train/value_std": 0.01593017578125,
      "objective/train/weight_avg": 1.0001031160354614,
      "objective/train/weighted_lm_loss": 1.5369855165481567,
      "objective/train/weights_max": 1.5545350313186646,
      "objective/train/weights_min": 0.26852282881736755,
      "theoretical_loss": 3.77661482471806,
      "tokens_seen": 760217600
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007774033060503932,
      "loss": 0.0804,
      "theoretical_loss": 3.77661482471806,
      "tokens_seen": 760217600
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007773230621088107,
      "loss": 0.0797,
      "theoretical_loss": 3.776485546222651,
      "tokens_seen": 760479744
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007772428181672284,
      "loss": 0.0783,
      "theoretical_loss": 3.776356324755847,
      "tokens_seen": 760741888
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000777162574225646,
      "loss": 0.0766,
      "theoretical_loss": 3.7762271602728497,
      "tokens_seen": 761004032
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007770823302840636,
      "loss": 0.0784,
      "theoretical_loss": 3.7760980527289156,
      "tokens_seen": 761266176
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007770020863424812,
      "loss": 0.0756,
      "theoretical_loss": 3.77596900207935,
      "tokens_seen": 761528320
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007769218424008987,
      "loss": 0.076,
      "theoretical_loss": 3.775840008279509,
      "tokens_seen": 761790464
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007768415984593163,
      "loss": 0.0785,
      "theoretical_loss": 3.7757110712847997,
      "tokens_seen": 762052608
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007767613545177339,
      "loss": 0.0791,
      "theoretical_loss": 3.775582191050678,
      "tokens_seen": 762314752
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007766811105761515,
      "loss": 0.0781,
      "theoretical_loss": 3.775453367532651,
      "tokens_seen": 762576896
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007766008666345691,
      "loss": 0.0781,
      "theoretical_loss": 3.775324600686276,
      "tokens_seen": 762839040
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007765206226929867,
      "loss": 0.0774,
      "theoretical_loss": 3.7751958904671614,
      "tokens_seen": 763101184
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007764403787514042,
      "loss": 0.0785,
      "theoretical_loss": 3.7750672368309623,
      "tokens_seen": 763363328
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000776360134809822,
      "loss": 0.0812,
      "theoretical_loss": 3.7749386397333873,
      "tokens_seen": 763625472
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007762798908682395,
      "loss": 0.0782,
      "theoretical_loss": 3.774810099130193,
      "tokens_seen": 763887616
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000776199646926657,
      "loss": 0.0785,
      "theoretical_loss": 3.7746816149771862,
      "tokens_seen": 764149760
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007761194029850747,
      "loss": 0.0772,
      "theoretical_loss": 3.7745531872302234,
      "tokens_seen": 764411904
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007760391590434922,
      "loss": 0.078,
      "theoretical_loss": 3.774424815845211,
      "tokens_seen": 764674048
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007759589151019098,
      "loss": 0.0778,
      "theoretical_loss": 3.774296500778105,
      "tokens_seen": 764936192
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007758786711603274,
      "loss": 0.0783,
      "theoretical_loss": 3.77416824198491,
      "tokens_seen": 765198336
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000775798427218745,
      "loss": 0.0762,
      "theoretical_loss": 3.7740400394216813,
      "tokens_seen": 765460480
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007757181832771625,
      "loss": 0.081,
      "theoretical_loss": 3.7739118930445223,
      "tokens_seen": 765722624
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007756379393355802,
      "loss": 0.0798,
      "theoretical_loss": 3.7737838028095867,
      "tokens_seen": 765984768
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007755576953939978,
      "loss": 0.0771,
      "theoretical_loss": 3.773655768673077,
      "tokens_seen": 766246912
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007754774514524153,
      "loss": 0.0804,
      "theoretical_loss": 3.7735277905912445,
      "tokens_seen": 766509056
    },
    {
      "epoch": 0.23,
      "objective/train/advantage_avg": 0.0010230033658444881,
      "objective/train/docs_used": 283988,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5113848447799683,
      "objective/train/original_loss": 1.5113848447799683,
      "objective/train/theoretical_loss": 3.773399868520391,
      "objective/train/tokens_used": 787231200,
      "objective/train/value_avg": -0.009429931640625,
      "objective/train/value_loss": 0.00033447632449679077,
      "objective/train/value_max": -0.00022518634796142578,
      "objective/train/value_min": -0.59326171875,
      "objective/train/value_reward_corr": 0.6839028954132548,
      "objective/train/value_std": 0.0179901123046875,
      "objective/train/weight_avg": 1.0011742115020752,
      "objective/train/weighted_lm_loss": 1.5128288269042969,
      "objective/train/weights_max": 1.5810500383377075,
      "objective/train/weights_min": 0.370164155960083,
      "theoretical_loss": 3.773399868520391,
      "tokens_seen": 766771200
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000775397207510833,
      "loss": 0.0784,
      "theoretical_loss": 3.773399868520391,
      "tokens_seen": 766771200
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007753169635692505,
      "loss": 0.0784,
      "theoretical_loss": 3.7732720024168644,
      "tokens_seen": 767033344
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007752367196276682,
      "loss": 0.0771,
      "theoretical_loss": 3.773144192237065,
      "tokens_seen": 767295488
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007751564756860857,
      "loss": 0.078,
      "theoretical_loss": 3.7730164379374402,
      "tokens_seen": 767557632
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007750762317445032,
      "loss": 0.0775,
      "theoretical_loss": 3.772888739474485,
      "tokens_seen": 767819776
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000774995987802921,
      "loss": 0.0768,
      "theoretical_loss": 3.772761096804745,
      "tokens_seen": 768081920
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007749157438613385,
      "loss": 0.0777,
      "theoretical_loss": 3.7726335098848143,
      "tokens_seen": 768344064
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007748354999197561,
      "loss": 0.0773,
      "theoretical_loss": 3.772505978671334,
      "tokens_seen": 768606208
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007747552559781737,
      "loss": 0.081,
      "theoretical_loss": 3.772378503120996,
      "tokens_seen": 768868352
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007746750120365913,
      "loss": 0.0785,
      "theoretical_loss": 3.7722510831905387,
      "tokens_seen": 769130496
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007745947680950088,
      "loss": 0.0791,
      "theoretical_loss": 3.7721237188367494,
      "tokens_seen": 769392640
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007745145241534264,
      "loss": 0.0773,
      "theoretical_loss": 3.771996410016464,
      "tokens_seen": 769654784
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000774434280211844,
      "loss": 0.0756,
      "theoretical_loss": 3.7718691566865665,
      "tokens_seen": 769916928
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007743540362702615,
      "loss": 0.0781,
      "theoretical_loss": 3.7717419588039887,
      "tokens_seen": 770179072
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007742737923286792,
      "loss": 0.0779,
      "theoretical_loss": 3.7716148163257115,
      "tokens_seen": 770441216
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007741935483870968,
      "loss": 0.0748,
      "theoretical_loss": 3.7714877292087623,
      "tokens_seen": 770703360
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007741133044455145,
      "loss": 0.0772,
      "theoretical_loss": 3.7713606974102167,
      "tokens_seen": 770965504
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000774033060503932,
      "loss": 0.0807,
      "theoretical_loss": 3.7712337208872,
      "tokens_seen": 771227648
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007739528165623495,
      "loss": 0.0764,
      "theoretical_loss": 3.7711067995968826,
      "tokens_seen": 771489792
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007738725726207672,
      "loss": 0.0797,
      "theoretical_loss": 3.770979933496485,
      "tokens_seen": 771751936
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007737923286791847,
      "loss": 0.0769,
      "theoretical_loss": 3.770853122543274,
      "tokens_seen": 772014080
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007737120847376023,
      "loss": 0.0752,
      "theoretical_loss": 3.770726366694564,
      "tokens_seen": 772276224
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007736318407960199,
      "loss": 0.0788,
      "theoretical_loss": 3.7705996659077172,
      "tokens_seen": 772538368
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007735515968544375,
      "loss": 0.0765,
      "theoretical_loss": 3.770473020140143,
      "tokens_seen": 772800512
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000773471352912855,
      "loss": 0.0816,
      "theoretical_loss": 3.770346429349299,
      "tokens_seen": 773062656
    },
    {
      "epoch": 0.23,
      "objective/train/advantage_avg": 0.0005200718878768384,
      "objective/train/docs_used": 286255,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5496317148208618,
      "objective/train/original_loss": 1.5496317148208618,
      "objective/train/theoretical_loss": 3.7702198934926896,
      "objective/train/tokens_used": 793784800,
      "objective/train/value_avg": -0.00848388671875,
      "objective/train/value_loss": 0.0003523084451444447,
      "objective/train/value_max": -0.0001302957534790039,
      "objective/train/value_min": -0.4052734375,
      "objective/train/value_reward_corr": 0.5911265747381873,
      "objective/train/value_std": 0.01352691650390625,
      "objective/train/weight_avg": 1.000672698020935,
      "objective/train/weighted_lm_loss": 1.550767421722412,
      "objective/train/weights_max": 1.2783743143081665,
      "objective/train/weights_min": 0.36907142400741577,
      "theoretical_loss": 3.7702198934926896,
      "tokens_seen": 773324800
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007733911089712728,
      "loss": 0.0771,
      "theoretical_loss": 3.7702198934926896,
      "tokens_seen": 773324800
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007733108650296903,
      "loss": 0.0765,
      "theoretical_loss": 3.7700934125278653,
      "tokens_seen": 773586944
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007732306210881078,
      "loss": 0.0802,
      "theoretical_loss": 3.7699669864124266,
      "tokens_seen": 773849088
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007731503771465255,
      "loss": 0.0774,
      "theoretical_loss": 3.769840615104018,
      "tokens_seen": 774111232
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000773070133204943,
      "loss": 0.0806,
      "theoretical_loss": 3.7697142985603325,
      "tokens_seen": 774373376
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007729898892633606,
      "loss": 0.0798,
      "theoretical_loss": 3.76958803673911,
      "tokens_seen": 774635520
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007729096453217782,
      "loss": 0.0797,
      "theoretical_loss": 3.7694618295981375,
      "tokens_seen": 774897664
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007728294013801958,
      "loss": 0.0777,
      "theoretical_loss": 3.769335677095248,
      "tokens_seen": 775159808
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0007727491574386135,
      "loss": 0.0804,
      "theoretical_loss": 3.769209579188323,
      "tokens_seen": 775421952
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000772668913497031,
      "loss": 0.0778,
      "theoretical_loss": 3.7690835358352883,
      "tokens_seen": 775684096
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007725886695554486,
      "loss": 0.0804,
      "theoretical_loss": 3.7689575469941183,
      "tokens_seen": 775946240
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007725084256138662,
      "loss": 0.079,
      "theoretical_loss": 3.768831612622833,
      "tokens_seen": 776208384
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007724281816722838,
      "loss": 0.0773,
      "theoretical_loss": 3.7687057326794986,
      "tokens_seen": 776470528
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007723479377307013,
      "loss": 0.0805,
      "theoretical_loss": 3.768579907122229,
      "tokens_seen": 776732672
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000772267693789119,
      "loss": 0.078,
      "theoretical_loss": 3.768454135909183,
      "tokens_seen": 776994816
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007721874498475365,
      "loss": 0.078,
      "theoretical_loss": 3.768328418998567,
      "tokens_seen": 777256960
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000772107205905954,
      "loss": 0.0777,
      "theoretical_loss": 3.7682027563486327,
      "tokens_seen": 777519104
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007720269619643718,
      "loss": 0.0805,
      "theoretical_loss": 3.768077147917678,
      "tokens_seen": 777781248
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007719467180227893,
      "loss": 0.0771,
      "theoretical_loss": 3.7679515936640477,
      "tokens_seen": 778043392
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007718664740812069,
      "loss": 0.0757,
      "theoretical_loss": 3.7678260935461316,
      "tokens_seen": 778305536
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007717862301396245,
      "loss": 0.074,
      "theoretical_loss": 3.767700647522366,
      "tokens_seen": 778567680
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007717059861980421,
      "loss": 0.0735,
      "theoretical_loss": 3.7675752555512334,
      "tokens_seen": 778829824
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007716257422564597,
      "loss": 0.0782,
      "theoretical_loss": 3.7674499175912617,
      "tokens_seen": 779091968
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007715454983148772,
      "loss": 0.0774,
      "theoretical_loss": 3.767324633601024,
      "tokens_seen": 779354112
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007714652543732948,
      "loss": 0.0786,
      "theoretical_loss": 3.7671994035391405,
      "tokens_seen": 779616256
    },
    {
      "epoch": 0.24,
      "objective/train/advantage_avg": 0.0016698199324309826,
      "objective/train/docs_used": 288522,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.6363639831542969,
      "objective/train/original_loss": 1.6363639831542969,
      "objective/train/theoretical_loss": 3.767074227364275,
      "objective/train/tokens_used": 800338400,
      "objective/train/value_avg": -0.00970458984375,
      "objective/train/value_loss": 0.0003810957132373005,
      "objective/train/value_max": -0.00016224384307861328,
      "objective/train/value_min": -0.76025390625,
      "objective/train/value_reward_corr": 0.7845589972125679,
      "objective/train/value_std": 0.0241241455078125,
      "objective/train/weight_avg": 1.0018428564071655,
      "objective/train/weighted_lm_loss": 1.6393061876296997,
      "objective/train/weights_max": 1.5092625617980957,
      "objective/train/weights_min": 0.36901721358299255,
      "theoretical_loss": 3.767074227364275,
      "tokens_seen": 779878400
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007713850104317124,
      "loss": 0.0804,
      "theoretical_loss": 3.767074227364275,
      "tokens_seen": 779878400
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00077130476649013,
      "loss": 0.0804,
      "theoretical_loss": 3.7669491050351396,
      "tokens_seen": 780140544
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007712245225485476,
      "loss": 0.0792,
      "theoretical_loss": 3.7668240365104895,
      "tokens_seen": 780402688
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007711442786069653,
      "loss": 0.0796,
      "theoretical_loss": 3.7666990217491265,
      "tokens_seen": 780664832
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007710640346653828,
      "loss": 0.0777,
      "theoretical_loss": 3.7665740607098974,
      "tokens_seen": 780926976
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007709837907238003,
      "loss": 0.077,
      "theoretical_loss": 3.7664491533516946,
      "tokens_seen": 781189120
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000770903546782218,
      "loss": 0.0771,
      "theoretical_loss": 3.766324299633455,
      "tokens_seen": 781451264
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007708233028406355,
      "loss": 0.0786,
      "theoretical_loss": 3.766199499514162,
      "tokens_seen": 781713408
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007707430588990531,
      "loss": 0.0753,
      "theoretical_loss": 3.7660747529528424,
      "tokens_seen": 781975552
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007706628149574707,
      "loss": 0.0789,
      "theoretical_loss": 3.76595005990857,
      "tokens_seen": 782237696
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007705825710158883,
      "loss": 0.0751,
      "theoretical_loss": 3.7658254203404615,
      "tokens_seen": 782499840
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007705023270743058,
      "loss": 0.0771,
      "theoretical_loss": 3.7657008342076796,
      "tokens_seen": 782761984
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007704220831327236,
      "loss": 0.076,
      "theoretical_loss": 3.765576301469433,
      "tokens_seen": 783024128
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007703418391911411,
      "loss": 0.0775,
      "theoretical_loss": 3.7654518220849726,
      "tokens_seen": 783286272
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007702615952495587,
      "loss": 0.0768,
      "theoretical_loss": 3.7653273960135962,
      "tokens_seen": 783548416
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007701813513079763,
      "loss": 0.0809,
      "theoretical_loss": 3.765203023214645,
      "tokens_seen": 783810560
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007701011073663938,
      "loss": 0.0754,
      "theoretical_loss": 3.7650787036475055,
      "tokens_seen": 784072704
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007700208634248115,
      "loss": 0.0765,
      "theoretical_loss": 3.764954437271608,
      "tokens_seen": 784334848
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000769940619483229,
      "loss": 0.0718,
      "theoretical_loss": 3.7648302240464284,
      "tokens_seen": 784596992
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007698603755416466,
      "loss": 0.0762,
      "theoretical_loss": 3.764706063931486,
      "tokens_seen": 784859136
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007697801316000643,
      "loss": 0.0783,
      "theoretical_loss": 3.764581956886345,
      "tokens_seen": 785121280
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007696998876584818,
      "loss": 0.0777,
      "theoretical_loss": 3.7644579028706135,
      "tokens_seen": 785383424
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007696196437168994,
      "loss": 0.0756,
      "theoretical_loss": 3.764333901843944,
      "tokens_seen": 785645568
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000769539399775317,
      "loss": 0.0771,
      "theoretical_loss": 3.764209953766033,
      "tokens_seen": 785907712
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007694591558337346,
      "loss": 0.0753,
      "theoretical_loss": 3.7640860585966207,
      "tokens_seen": 786169856
    },
    {
      "epoch": 0.24,
      "objective/train/advantage_avg": 0.0009356484515592456,
      "objective/train/docs_used": 290810,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.523367166519165,
      "objective/train/original_loss": 1.523367166519165,
      "objective/train/theoretical_loss": 3.763962216295493,
      "objective/train/tokens_used": 806892000,
      "objective/train/value_avg": -0.007610321044921875,
      "objective/train/value_loss": 0.0002046849112957716,
      "objective/train/value_max": -0.0001838207244873047,
      "objective/train/value_min": -0.347900390625,
      "objective/train/value_reward_corr": 0.6702344519054119,
      "objective/train/value_std": 0.01255035400390625,
      "objective/train/weight_avg": 1.0010277032852173,
      "objective/train/weighted_lm_loss": 1.5253008604049683,
      "objective/train/weights_max": 1.1150742769241333,
      "objective/train/weights_min": 0.3685617744922638,
      "theoretical_loss": 3.763962216295493,
      "tokens_seen": 786432000
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007693789118921521,
      "loss": 0.0757,
      "theoretical_loss": 3.763962216295493,
      "tokens_seen": 786432000
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007692986679505698,
      "loss": 0.0761,
      "theoretical_loss": 3.7638384268224776,
      "tokens_seen": 786694144
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007692184240089873,
      "loss": 0.0773,
      "theoretical_loss": 3.7637146901374474,
      "tokens_seen": 786956288
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007691381800674048,
      "loss": 0.0785,
      "theoretical_loss": 3.7635910062003193,
      "tokens_seen": 787218432
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007690579361258225,
      "loss": 0.0765,
      "theoretical_loss": 3.7634673749710523,
      "tokens_seen": 787480576
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00076897769218424,
      "loss": 0.0753,
      "theoretical_loss": 3.763343796409651,
      "tokens_seen": 787742720
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007688974482426578,
      "loss": 0.076,
      "theoretical_loss": 3.7632202704761637,
      "tokens_seen": 788004864
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007688172043010753,
      "loss": 0.0756,
      "theoretical_loss": 3.7630967971306797,
      "tokens_seen": 788267008
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007687369603594929,
      "loss": 0.0767,
      "theoretical_loss": 3.762973376333335,
      "tokens_seen": 788529152
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007686567164179105,
      "loss": 0.0775,
      "theoretical_loss": 3.7628500080443077,
      "tokens_seen": 788791296
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000768576472476328,
      "loss": 0.077,
      "theoretical_loss": 3.7627266922238185,
      "tokens_seen": 789053440
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007684962285347456,
      "loss": 0.0768,
      "theoretical_loss": 3.762603428832133,
      "tokens_seen": 789315584
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007684159845931632,
      "loss": 0.0751,
      "theoretical_loss": 3.7624802178295584,
      "tokens_seen": 789577728
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007683357406515808,
      "loss": 0.0778,
      "theoretical_loss": 3.762357059176447,
      "tokens_seen": 789839872
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007682554967099984,
      "loss": 0.0773,
      "theoretical_loss": 3.762233952833193,
      "tokens_seen": 790102016
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007681752527684161,
      "loss": 0.0798,
      "theoretical_loss": 3.7621108987602336,
      "tokens_seen": 790364160
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007680950088268336,
      "loss": 0.0764,
      "theoretical_loss": 3.76198789691805,
      "tokens_seen": 790626304
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007680147648852511,
      "loss": 0.0798,
      "theoretical_loss": 3.7618649472671652,
      "tokens_seen": 790888448
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007679345209436688,
      "loss": 0.0781,
      "theoretical_loss": 3.761742049768146,
      "tokens_seen": 791150592
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007678542770020863,
      "loss": 0.0789,
      "theoretical_loss": 3.761619204381602,
      "tokens_seen": 791412736
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000767774033060504,
      "loss": 0.0751,
      "theoretical_loss": 3.7614964110681846,
      "tokens_seen": 791674880
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007676937891189215,
      "loss": 0.0769,
      "theoretical_loss": 3.761373669788589,
      "tokens_seen": 791937024
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007676135451773391,
      "loss": 0.0765,
      "theoretical_loss": 3.7612509805035526,
      "tokens_seen": 792199168
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007675333012357568,
      "loss": 0.075,
      "theoretical_loss": 3.761128343173856,
      "tokens_seen": 792461312
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007674530572941743,
      "loss": 0.0755,
      "theoretical_loss": 3.7610057577603215,
      "tokens_seen": 792723456
    },
    {
      "epoch": 0.24,
      "objective/train/advantage_avg": 0.0012367293238639832,
      "objective/train/docs_used": 293128,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5767234563827515,
      "objective/train/original_loss": 1.576723575592041,
      "objective/train/theoretical_loss": 3.7608832242238144,
      "objective/train/tokens_used": 813445600,
      "objective/train/value_avg": -0.0086669921875,
      "objective/train/value_loss": 0.00022169639123603702,
      "objective/train/value_max": -0.0002472400665283203,
      "objective/train/value_min": -0.58544921875,
      "objective/train/value_reward_corr": 0.6488604890647144,
      "objective/train/value_std": 0.0127410888671875,
      "objective/train/weight_avg": 1.001336932182312,
      "objective/train/weighted_lm_loss": 1.5789175033569336,
      "objective/train/weights_max": 1.3074339628219604,
      "objective/train/weights_min": 0.36864086985588074,
      "theoretical_loss": 3.7608832242238144,
      "tokens_seen": 792985600
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007673728133525919,
      "loss": 0.0782,
      "theoretical_loss": 3.7608832242238144,
      "tokens_seen": 792985600
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007672925694110095,
      "loss": 0.0778,
      "theoretical_loss": 3.7607607425252416,
      "tokens_seen": 793247744
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007672123254694271,
      "loss": 0.0788,
      "theoretical_loss": 3.7606383126255536,
      "tokens_seen": 793509888
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007671320815278446,
      "loss": 0.0792,
      "theoretical_loss": 3.760515934485743,
      "tokens_seen": 793772032
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007670518375862623,
      "loss": 0.0776,
      "theoretical_loss": 3.760393608066843,
      "tokens_seen": 794034176
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007669715936446798,
      "loss": 0.0759,
      "theoretical_loss": 3.760271333329932,
      "tokens_seen": 794296320
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007668913497030974,
      "loss": 0.0779,
      "theoretical_loss": 3.7601491102361275,
      "tokens_seen": 794558464
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000766811105761515,
      "loss": 0.0776,
      "theoretical_loss": 3.7600269387465914,
      "tokens_seen": 794820608
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007667308618199326,
      "loss": 0.078,
      "theoretical_loss": 3.759904818822525,
      "tokens_seen": 795082752
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007666506178783502,
      "loss": 0.0787,
      "theoretical_loss": 3.759782750425175,
      "tokens_seen": 795344896
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007665703739367678,
      "loss": 0.076,
      "theoretical_loss": 3.759660733515826,
      "tokens_seen": 795607040
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007664901299951854,
      "loss": 0.0765,
      "theoretical_loss": 3.7595387680558088,
      "tokens_seen": 795869184
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000766409886053603,
      "loss": 0.0791,
      "theoretical_loss": 3.759416854006492,
      "tokens_seen": 796131328
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007663296421120206,
      "loss": 0.0793,
      "theoretical_loss": 3.7592949913292886,
      "tokens_seen": 796393472
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007662493981704381,
      "loss": 0.0786,
      "theoretical_loss": 3.759173179985652,
      "tokens_seen": 796655616
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007661691542288557,
      "loss": 0.0747,
      "theoretical_loss": 3.7590514199370775,
      "tokens_seen": 796917760
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007660889102872733,
      "loss": 0.0735,
      "theoretical_loss": 3.758929711145101,
      "tokens_seen": 797179904
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007660086663456909,
      "loss": 0.0756,
      "theoretical_loss": 3.758808053571302,
      "tokens_seen": 797442048
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007659284224041086,
      "loss": 0.0805,
      "theoretical_loss": 3.7586864471772996,
      "tokens_seen": 797704192
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007658481784625261,
      "loss": 0.0771,
      "theoretical_loss": 3.758564891924755,
      "tokens_seen": 797966336
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007657679345209437,
      "loss": 0.0792,
      "theoretical_loss": 3.758443387775371,
      "tokens_seen": 798228480
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007656876905793613,
      "loss": 0.0762,
      "theoretical_loss": 3.7583219346908905,
      "tokens_seen": 798490624
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007656074466377788,
      "loss": 0.077,
      "theoretical_loss": 3.758200532633099,
      "tokens_seen": 798752768
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007655272026961964,
      "loss": 0.0767,
      "theoretical_loss": 3.7580791815638213,
      "tokens_seen": 799014912
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000765446958754614,
      "loss": 0.0782,
      "theoretical_loss": 3.7579578814449253,
      "tokens_seen": 799277056
    },
    {
      "epoch": 0.24,
      "objective/train/advantage_avg": 0.0012315761996433139,
      "objective/train/docs_used": 295724,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5393258333206177,
      "objective/train/original_loss": 1.5393258333206177,
      "objective/train/theoretical_loss": 3.7578366322383188,
      "objective/train/tokens_used": 819999200,
      "objective/train/value_avg": -0.007904052734375,
      "objective/train/value_loss": 0.00038409262197092175,
      "objective/train/value_max": -0.00012433528900146484,
      "objective/train/value_min": -0.87158203125,
      "objective/train/value_reward_corr": 0.6161805024099198,
      "objective/train/value_std": 0.0164794921875,
      "objective/train/weight_avg": 1.0014045238494873,
      "objective/train/weighted_lm_loss": 1.5412237644195557,
      "objective/train/weights_max": 1.616646409034729,
      "objective/train/weights_min": 0.3682824373245239,
      "theoretical_loss": 3.7578366322383188,
      "tokens_seen": 799539200
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007653667148130316,
      "loss": 0.0778,
      "theoretical_loss": 3.7578366322383188,
      "tokens_seen": 799539200
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007652864708714491,
      "loss": 0.0762,
      "theoretical_loss": 3.7577154339059504,
      "tokens_seen": 799801344
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007652062269298669,
      "loss": 0.0768,
      "theoretical_loss": 3.7575942864098106,
      "tokens_seen": 800063488
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007651259829882844,
      "loss": 0.0755,
      "theoretical_loss": 3.75747318971193,
      "tokens_seen": 800325632
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000765045739046702,
      "loss": 0.0757,
      "theoretical_loss": 3.7573521437743795,
      "tokens_seen": 800587776
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007649654951051196,
      "loss": 0.0775,
      "theoretical_loss": 3.7572311485592715,
      "tokens_seen": 800849920
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007648852511635371,
      "loss": 0.0754,
      "theoretical_loss": 3.7571102040287596,
      "tokens_seen": 801112064
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007648050072219548,
      "loss": 0.0803,
      "theoretical_loss": 3.7569893101450367,
      "tokens_seen": 801374208
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007647247632803723,
      "loss": 0.0804,
      "theoretical_loss": 3.756868466870337,
      "tokens_seen": 801636352
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007646445193387899,
      "loss": 0.078,
      "theoretical_loss": 3.7567476741669346,
      "tokens_seen": 801898496
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007645642753972076,
      "loss": 0.0778,
      "theoretical_loss": 3.756626931997145,
      "tokens_seen": 802160640
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007644840314556251,
      "loss": 0.0745,
      "theoretical_loss": 3.7565062403233234,
      "tokens_seen": 802422784
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007644037875140427,
      "loss": 0.0742,
      "theoretical_loss": 3.7563855991078654,
      "tokens_seen": 802684928
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007643235435724603,
      "loss": 0.0725,
      "theoretical_loss": 3.7562650083132074,
      "tokens_seen": 802947072
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007642432996308779,
      "loss": 0.0782,
      "theoretical_loss": 3.756144467901825,
      "tokens_seen": 803209216
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007641630556892954,
      "loss": 0.0777,
      "theoretical_loss": 3.756023977836235,
      "tokens_seen": 803471360
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007640828117477131,
      "loss": 0.0767,
      "theoretical_loss": 3.755903538078994,
      "tokens_seen": 803733504
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007640025678061306,
      "loss": 0.0803,
      "theoretical_loss": 3.7557831485926982,
      "tokens_seen": 803995648
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007639223238645483,
      "loss": 0.0786,
      "theoretical_loss": 3.7556628093399835,
      "tokens_seen": 804257792
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007638420799229659,
      "loss": 0.0802,
      "theoretical_loss": 3.7555425202835275,
      "tokens_seen": 804519936
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007637618359813834,
      "loss": 0.0736,
      "theoretical_loss": 3.7554222813860463,
      "tokens_seen": 804782080
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007636815920398011,
      "loss": 0.0769,
      "theoretical_loss": 3.7553020926102954,
      "tokens_seen": 805044224
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007636013480982186,
      "loss": 0.0786,
      "theoretical_loss": 3.755181953919071,
      "tokens_seen": 805306368
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007635211041566362,
      "loss": 0.0769,
      "theoretical_loss": 3.755061865275209,
      "tokens_seen": 805568512
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007634408602150538,
      "loss": 0.0779,
      "theoretical_loss": 3.754941826641584,
      "tokens_seen": 805830656
    },
    {
      "epoch": 0.24,
      "objective/train/advantage_avg": 0.0008438312797807157,
      "objective/train/docs_used": 297892,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4582135677337646,
      "objective/train/original_loss": 1.458213448524475,
      "objective/train/theoretical_loss": 3.754821837981112,
      "objective/train/tokens_used": 826552800,
      "objective/train/value_avg": -0.005859375,
      "objective/train/value_loss": 0.0003289504675194621,
      "objective/train/value_max": -0.00010973215103149414,
      "objective/train/value_min": -0.73828125,
      "objective/train/value_reward_corr": 0.4976893790214529,
      "objective/train/value_std": 0.0099029541015625,
      "objective/train/weight_avg": 1.0009607076644897,
      "objective/train/weighted_lm_loss": 1.4596952199935913,
      "objective/train/weights_max": 1.2155424356460571,
      "objective/train/weights_min": 0.05370466783642769,
      "theoretical_loss": 3.754821837981112,
      "tokens_seen": 806092800
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007633606162734714,
      "loss": 0.0753,
      "theoretical_loss": 3.754821837981112,
      "tokens_seen": 806092800
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007632803723318889,
      "loss": 0.0759,
      "theoretical_loss": 3.7547018992567462,
      "tokens_seen": 806354944
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007632001283903065,
      "loss": 0.079,
      "theoretical_loss": 3.7545820104314815,
      "tokens_seen": 806617088
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007631198844487241,
      "loss": 0.0746,
      "theoretical_loss": 3.7544621714683517,
      "tokens_seen": 806879232
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007630396405071417,
      "loss": 0.0763,
      "theoretical_loss": 3.754342382330428,
      "tokens_seen": 807141376
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007629593965655594,
      "loss": 0.0746,
      "theoretical_loss": 3.7542226429808236,
      "tokens_seen": 807403520
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007628791526239769,
      "loss": 0.0742,
      "theoretical_loss": 3.7541029533826893,
      "tokens_seen": 807665664
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007627989086823945,
      "loss": 0.0753,
      "theoretical_loss": 3.7539833134992158,
      "tokens_seen": 807927808
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007627186647408121,
      "loss": 0.0783,
      "theoretical_loss": 3.753863723293634,
      "tokens_seen": 808189952
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0007626384207992296,
      "loss": 0.0766,
      "theoretical_loss": 3.7537441827292106,
      "tokens_seen": 808452096
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007625581768576473,
      "loss": 0.077,
      "theoretical_loss": 3.753624691769255,
      "tokens_seen": 808714240
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007624779329160648,
      "loss": 0.079,
      "theoretical_loss": 3.7535052503771142,
      "tokens_seen": 808976384
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007623976889744824,
      "loss": 0.0756,
      "theoretical_loss": 3.7533858585161735,
      "tokens_seen": 809238528
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007623174450329001,
      "loss": 0.0796,
      "theoretical_loss": 3.753266516149858,
      "tokens_seen": 809500672
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007622372010913177,
      "loss": 0.0766,
      "theoretical_loss": 3.7531472232416316,
      "tokens_seen": 809762816
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007621569571497352,
      "loss": 0.0736,
      "theoretical_loss": 3.7530279797549957,
      "tokens_seen": 810024960
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007620767132081528,
      "loss": 0.0748,
      "theoretical_loss": 3.752908785653492,
      "tokens_seen": 810287104
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007619964692665704,
      "loss": 0.0756,
      "theoretical_loss": 3.7527896409007004,
      "tokens_seen": 810549248
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007619162253249879,
      "loss": 0.078,
      "theoretical_loss": 3.7526705454602394,
      "tokens_seen": 810811392
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007618359813834056,
      "loss": 0.077,
      "theoretical_loss": 3.752551499295766,
      "tokens_seen": 811073536
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007617557374418231,
      "loss": 0.0766,
      "theoretical_loss": 3.7524325023709757,
      "tokens_seen": 811335680
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007616754935002407,
      "loss": 0.0736,
      "theoretical_loss": 3.7523135546496023,
      "tokens_seen": 811597824
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007615952495586584,
      "loss": 0.0772,
      "theoretical_loss": 3.7521946560954182,
      "tokens_seen": 811859968
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007615150056170759,
      "loss": 0.0767,
      "theoretical_loss": 3.7520758066722344,
      "tokens_seen": 812122112
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007614347616754936,
      "loss": 0.075,
      "theoretical_loss": 3.7519570063438996,
      "tokens_seen": 812384256
    },
    {
      "epoch": 0.25,
      "objective/train/advantage_avg": 0.0014068076852709055,
      "objective/train/docs_used": 300102,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5233573913574219,
      "objective/train/original_loss": 1.5233573913574219,
      "objective/train/theoretical_loss": 3.7518382550743024,
      "objective/train/tokens_used": 833106400,
      "objective/train/value_avg": -0.006847381591796875,
      "objective/train/value_loss": 0.00014167482731863856,
      "objective/train/value_max": -0.0001150369644165039,
      "objective/train/value_min": -0.31005859375,
      "objective/train/value_reward_corr": 0.5786646527156373,
      "objective/train/value_std": 0.01003265380859375,
      "objective/train/weight_avg": 1.0014715194702148,
      "objective/train/weighted_lm_loss": 1.5256245136260986,
      "objective/train/weights_max": 1.1971921920776367,
      "objective/train/weights_min": 0.36875900626182556,
      "theoretical_loss": 3.7518382550743024,
      "tokens_seen": 812646400
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007613545177339111,
      "loss": 0.0763,
      "theoretical_loss": 3.7518382550743024,
      "tokens_seen": 812646400
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007612742737923287,
      "loss": 0.0777,
      "theoretical_loss": 3.7517195528273666,
      "tokens_seen": 812908544
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007611940298507463,
      "loss": 0.0758,
      "theoretical_loss": 3.751600899567057,
      "tokens_seen": 813170688
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007611137859091639,
      "loss": 0.0724,
      "theoretical_loss": 3.7514822952573743,
      "tokens_seen": 813432832
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007610335419675814,
      "loss": 0.0772,
      "theoretical_loss": 3.7513637398623603,
      "tokens_seen": 813694976
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007609532980259991,
      "loss": 0.0756,
      "theoretical_loss": 3.751245233346091,
      "tokens_seen": 813957120
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007608730540844167,
      "loss": 0.0768,
      "theoretical_loss": 3.7511267756726823,
      "tokens_seen": 814219264
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007607928101428342,
      "loss": 0.0783,
      "theoretical_loss": 3.7510083668062886,
      "tokens_seen": 814481408
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007607125662012519,
      "loss": 0.075,
      "theoretical_loss": 3.750890006711101,
      "tokens_seen": 814743552
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007606323222596694,
      "loss": 0.075,
      "theoretical_loss": 3.7507716953513492,
      "tokens_seen": 815005696
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000760552078318087,
      "loss": 0.0784,
      "theoretical_loss": 3.7506534326912995,
      "tokens_seen": 815267840
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007604718343765046,
      "loss": 0.0769,
      "theoretical_loss": 3.7505352186952567,
      "tokens_seen": 815529984
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007603915904349222,
      "loss": 0.0738,
      "theoretical_loss": 3.7504170533275634,
      "tokens_seen": 815792128
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007603113464933397,
      "loss": 0.0732,
      "theoretical_loss": 3.7502989365526,
      "tokens_seen": 816054272
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007602311025517573,
      "loss": 0.0762,
      "theoretical_loss": 3.7501808683347826,
      "tokens_seen": 816316416
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000760150858610175,
      "loss": 0.0751,
      "theoretical_loss": 3.7500628486385668,
      "tokens_seen": 816578560
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007600706146685926,
      "loss": 0.0744,
      "theoretical_loss": 3.7499448774284447,
      "tokens_seen": 816840704
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007599903707270102,
      "loss": 0.0771,
      "theoretical_loss": 3.749826954668946,
      "tokens_seen": 817102848
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007599101267854277,
      "loss": 0.077,
      "theoretical_loss": 3.7497090803246387,
      "tokens_seen": 817364992
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007598298828438454,
      "loss": 0.0749,
      "theoretical_loss": 3.7495912543601246,
      "tokens_seen": 817627136
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007597496389022629,
      "loss": 0.0775,
      "theoretical_loss": 3.7494734767400475,
      "tokens_seen": 817889280
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007596693949606804,
      "loss": 0.0766,
      "theoretical_loss": 3.7493557474290853,
      "tokens_seen": 818151424
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007595891510190981,
      "loss": 0.0743,
      "theoretical_loss": 3.7492380663919533,
      "tokens_seen": 818413568
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007595089070775156,
      "loss": 0.0762,
      "theoretical_loss": 3.7491204335934043,
      "tokens_seen": 818675712
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007594286631359332,
      "loss": 0.076,
      "theoretical_loss": 3.7490028489982286,
      "tokens_seen": 818937856
    },
    {
      "epoch": 0.25,
      "objective/train/advantage_avg": -2.009487616305705e-05,
      "objective/train/docs_used": 302476,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5615952014923096,
      "objective/train/original_loss": 1.5615952014923096,
      "objective/train/theoretical_loss": 3.7488853125712525,
      "objective/train/tokens_used": 839660000,
      "objective/train/value_avg": -0.0078582763671875,
      "objective/train/value_loss": 0.00040608173003420234,
      "objective/train/value_max": -0.0001323223114013672,
      "objective/train/value_min": -0.56494140625,
      "objective/train/value_reward_corr": 0.58299361157949,
      "objective/train/value_std": 0.01151275634765625,
      "objective/train/weight_avg": 1.0001566410064697,
      "objective/train/weighted_lm_loss": 1.563400387763977,
      "objective/train/weights_max": 1.6039460897445679,
      "objective/train/weights_min": 0.3728724420070648,
      "theoretical_loss": 3.7488853125712525,
      "tokens_seen": 819200000
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007593484191943509,
      "loss": 0.0727,
      "theoretical_loss": 3.7488853125712525,
      "tokens_seen": 819200000
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007592681752527685,
      "loss": 0.0738,
      "theoretical_loss": 3.7487678242773406,
      "tokens_seen": 819462144
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000759187931311186,
      "loss": 0.0772,
      "theoretical_loss": 3.748650384081392,
      "tokens_seen": 819724288
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007591076873696036,
      "loss": 0.0771,
      "theoretical_loss": 3.7485329919483448,
      "tokens_seen": 819986432
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007590274434280212,
      "loss": 0.0746,
      "theoretical_loss": 3.7484156478431734,
      "tokens_seen": 820248576
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007589471994864387,
      "loss": 0.075,
      "theoretical_loss": 3.748298351730888,
      "tokens_seen": 820510720
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007588669555448564,
      "loss": 0.0757,
      "theoretical_loss": 3.748181103576537,
      "tokens_seen": 820772864
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007587867116032739,
      "loss": 0.0773,
      "theoretical_loss": 3.7480639033452032,
      "tokens_seen": 821035008
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007587064676616916,
      "loss": 0.078,
      "theoretical_loss": 3.747946751002009,
      "tokens_seen": 821297152
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007586262237201092,
      "loss": 0.0784,
      "theoretical_loss": 3.747829646512109,
      "tokens_seen": 821559296
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007585459797785267,
      "loss": 0.0763,
      "theoretical_loss": 3.747712589840699,
      "tokens_seen": 821821440
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007584657358369444,
      "loss": 0.0751,
      "theoretical_loss": 3.7475955809530084,
      "tokens_seen": 822083584
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007583854918953619,
      "loss": 0.0731,
      "theoretical_loss": 3.747478619814303,
      "tokens_seen": 822345728
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007583052479537795,
      "loss": 0.0792,
      "theoretical_loss": 3.7473617063898863,
      "tokens_seen": 822607872
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007582250040121971,
      "loss": 0.0767,
      "theoretical_loss": 3.747244840645097,
      "tokens_seen": 822870016
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007581447600706147,
      "loss": 0.0751,
      "theoretical_loss": 3.7471280225453096,
      "tokens_seen": 823132160
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007580645161290322,
      "loss": 0.0764,
      "theoretical_loss": 3.747011252055936,
      "tokens_seen": 823394304
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007579842721874498,
      "loss": 0.075,
      "theoretical_loss": 3.746894529142424,
      "tokens_seen": 823656448
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007579040282458675,
      "loss": 0.0762,
      "theoretical_loss": 3.746777853770256,
      "tokens_seen": 823918592
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000757823784304285,
      "loss": 0.0751,
      "theoretical_loss": 3.746661225904953,
      "tokens_seen": 824180736
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007577435403627027,
      "loss": 0.0776,
      "theoretical_loss": 3.746544645512069,
      "tokens_seen": 824442880
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007576632964211202,
      "loss": 0.0753,
      "theoretical_loss": 3.7464281125571963,
      "tokens_seen": 824705024
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007575830524795379,
      "loss": 0.0762,
      "theoretical_loss": 3.7463116270059618,
      "tokens_seen": 824967168
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007575028085379554,
      "loss": 0.0748,
      "theoretical_loss": 3.7461951888240286,
      "tokens_seen": 825229312
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000757422564596373,
      "loss": 0.0769,
      "theoretical_loss": 3.7460787979770958,
      "tokens_seen": 825491456
    },
    {
      "epoch": 0.25,
      "objective/train/advantage_avg": -6.741825927747414e-05,
      "objective/train/docs_used": 304950,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5155413150787354,
      "objective/train/original_loss": 1.5155411958694458,
      "objective/train/theoretical_loss": 3.745962454430897,
      "objective/train/tokens_used": 846213600,
      "objective/train/value_avg": -0.007843017578125,
      "objective/train/value_loss": 0.0001769889786373824,
      "objective/train/value_max": -0.00011682510375976562,
      "objective/train/value_min": -0.30322265625,
      "objective/train/value_reward_corr": 0.7321892783075874,
      "objective/train/value_std": 0.01239776611328125,
      "objective/train/weight_avg": 1.000016212463379,
      "objective/train/weighted_lm_loss": 1.5158941745758057,
      "objective/train/weights_max": 1.2777711153030396,
      "objective/train/weights_min": 0.37355300784111023,
      "theoretical_loss": 3.745962454430897,
      "tokens_seen": 825753600
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007573423206547906,
      "loss": 0.0764,
      "theoretical_loss": 3.745962454430897,
      "tokens_seen": 825753600
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007572620767132081,
      "loss": 0.0785,
      "theoretical_loss": 3.745846158151204,
      "tokens_seen": 826015744
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007571818327716257,
      "loss": 0.0765,
      "theoretical_loss": 3.7457299091038214,
      "tokens_seen": 826277888
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007571015888300434,
      "loss": 0.0748,
      "theoretical_loss": 3.745613707254591,
      "tokens_seen": 826540032
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000757021344888461,
      "loss": 0.0763,
      "theoretical_loss": 3.7454975525693897,
      "tokens_seen": 826802176
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007569411009468785,
      "loss": 0.076,
      "theoretical_loss": 3.7453814450141305,
      "tokens_seen": 827064320
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007568608570052962,
      "loss": 0.0806,
      "theoretical_loss": 3.7452653845547603,
      "tokens_seen": 827326464
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007567806130637137,
      "loss": 0.0798,
      "theoretical_loss": 3.745149371157263,
      "tokens_seen": 827588608
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007567003691221312,
      "loss": 0.0778,
      "theoretical_loss": 3.7450334047876574,
      "tokens_seen": 827850752
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007566201251805489,
      "loss": 0.0779,
      "theoretical_loss": 3.744917485411997,
      "tokens_seen": 828112896
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007565398812389664,
      "loss": 0.0803,
      "theoretical_loss": 3.744801612996371,
      "tokens_seen": 828375040
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000756459637297384,
      "loss": 0.0766,
      "theoretical_loss": 3.744685787506903,
      "tokens_seen": 828637184
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007563793933558017,
      "loss": 0.0775,
      "theoretical_loss": 3.7445700089097533,
      "tokens_seen": 828899328
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007562991494142193,
      "loss": 0.0764,
      "theoretical_loss": 3.7444542771711165,
      "tokens_seen": 829161472
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007562189054726369,
      "loss": 0.0745,
      "theoretical_loss": 3.744338592257222,
      "tokens_seen": 829423616
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007561386615310544,
      "loss": 0.076,
      "theoretical_loss": 3.744222954134334,
      "tokens_seen": 829685760
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000756058417589472,
      "loss": 0.0747,
      "theoretical_loss": 3.7441073627687524,
      "tokens_seen": 829947904
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007559781736478896,
      "loss": 0.0769,
      "theoretical_loss": 3.743991818126812,
      "tokens_seen": 830210048
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007558979297063072,
      "loss": 0.0761,
      "theoretical_loss": 3.7438763201748815,
      "tokens_seen": 830472192
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007558176857647247,
      "loss": 0.0765,
      "theoretical_loss": 3.743760868879365,
      "tokens_seen": 830734336
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007557374418231424,
      "loss": 0.0748,
      "theoretical_loss": 3.743645464206702,
      "tokens_seen": 830996480
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00075565719788156,
      "loss": 0.0757,
      "theoretical_loss": 3.743530106123365,
      "tokens_seen": 831258624
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007555769539399775,
      "loss": 0.0756,
      "theoretical_loss": 3.7434147945958642,
      "tokens_seen": 831520768
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007554967099983952,
      "loss": 0.0766,
      "theoretical_loss": 3.7432995295907405,
      "tokens_seen": 831782912
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007554164660568127,
      "loss": 0.0781,
      "theoretical_loss": 3.7431843110745726,
      "tokens_seen": 832045056
    },
    {
      "epoch": 0.25,
      "objective/train/advantage_avg": 0.0015261891530826688,
      "objective/train/docs_used": 307306,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4608768224716187,
      "objective/train/original_loss": 1.4608770608901978,
      "objective/train/theoretical_loss": 3.743069139013972,
      "objective/train/tokens_used": 852767200,
      "objective/train/value_avg": -0.009246826171875,
      "objective/train/value_loss": 0.00019339752907399088,
      "objective/train/value_max": -0.00014889240264892578,
      "objective/train/value_min": -0.43701171875,
      "objective/train/value_reward_corr": 0.673779683354758,
      "objective/train/value_std": 0.0131378173828125,
      "objective/train/weight_avg": 1.0016170740127563,
      "objective/train/weighted_lm_loss": 1.4626826047897339,
      "objective/train/weights_max": 1.353039264678955,
      "objective/train/weights_min": 0.37851035594940186,
      "theoretical_loss": 3.743069139013972,
      "tokens_seen": 832307200
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007553362221152303,
      "loss": 0.075,
      "theoretical_loss": 3.743069139013972,
      "tokens_seen": 832307200
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007552559781736479,
      "loss": 0.076,
      "theoretical_loss": 3.742954013375586,
      "tokens_seen": 832569344
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007551757342320655,
      "loss": 0.0768,
      "theoretical_loss": 3.742838934126094,
      "tokens_seen": 832831488
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000755095490290483,
      "loss": 0.0764,
      "theoretical_loss": 3.742723901232213,
      "tokens_seen": 833093632
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007550152463489006,
      "loss": 0.0764,
      "theoretical_loss": 3.742608914660692,
      "tokens_seen": 833355776
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007549350024073182,
      "loss": 0.078,
      "theoretical_loss": 3.742493974378314,
      "tokens_seen": 833617920
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007548547584657359,
      "loss": 0.0751,
      "theoretical_loss": 3.742379080351899,
      "tokens_seen": 833880064
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007547745145241535,
      "loss": 0.079,
      "theoretical_loss": 3.7422642325482975,
      "tokens_seen": 834142208
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000754694270582571,
      "loss": 0.0808,
      "theoretical_loss": 3.742149430934398,
      "tokens_seen": 834404352
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007546140266409887,
      "loss": 0.0806,
      "theoretical_loss": 3.7420346754771208,
      "tokens_seen": 834666496
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007545337826994062,
      "loss": 0.0763,
      "theoretical_loss": 3.7419199661434197,
      "tokens_seen": 834928640
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007544535387578238,
      "loss": 0.078,
      "theoretical_loss": 3.7418053029002842,
      "tokens_seen": 835190784
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007543732948162414,
      "loss": 0.0738,
      "theoretical_loss": 3.7416906857147367,
      "tokens_seen": 835452928
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007542930508746589,
      "loss": 0.0758,
      "theoretical_loss": 3.741576114553835,
      "tokens_seen": 835715072
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007542128069330765,
      "loss": 0.0775,
      "theoretical_loss": 3.7414615893846683,
      "tokens_seen": 835977216
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007541325629914942,
      "loss": 0.0766,
      "theoretical_loss": 3.741347110174362,
      "tokens_seen": 836239360
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007540523190499118,
      "loss": 0.0775,
      "theoretical_loss": 3.741232676890074,
      "tokens_seen": 836501504
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007539720751083293,
      "loss": 0.0782,
      "theoretical_loss": 3.7411182894989965,
      "tokens_seen": 836763648
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000753891831166747,
      "loss": 0.0778,
      "theoretical_loss": 3.7410039479683546,
      "tokens_seen": 837025792
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007538115872251645,
      "loss": 0.0758,
      "theoretical_loss": 3.740889652265408,
      "tokens_seen": 837287936
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007537313432835821,
      "loss": 0.0772,
      "theoretical_loss": 3.7407754023574507,
      "tokens_seen": 837550080
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007536510993419997,
      "loss": 0.0767,
      "theoretical_loss": 3.7406611982118076,
      "tokens_seen": 837812224
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007535708554004172,
      "loss": 0.0768,
      "theoretical_loss": 3.74054703979584,
      "tokens_seen": 838074368
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000753490611458835,
      "loss": 0.0782,
      "theoretical_loss": 3.7404329270769403,
      "tokens_seen": 838336512
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007534103675172525,
      "loss": 0.0758,
      "theoretical_loss": 3.740318860022537,
      "tokens_seen": 838598656
    },
    {
      "epoch": 0.25,
      "objective/train/advantage_avg": 0.0015410750638693571,
      "objective/train/docs_used": 309674,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.561414361000061,
      "objective/train/original_loss": 1.5614142417907715,
      "objective/train/theoretical_loss": 3.7402048386000892,
      "objective/train/tokens_used": 859320800,
      "objective/train/value_avg": -0.01053619384765625,
      "objective/train/value_loss": 0.0005841842503286898,
      "objective/train/value_max": -0.0002846717834472656,
      "objective/train/value_min": -0.48974609375,
      "objective/train/value_reward_corr": 0.5773049027811739,
      "objective/train/value_std": 0.016204833984375,
      "objective/train/weight_avg": 1.0017683506011963,
      "objective/train/weighted_lm_loss": 1.563459038734436,
      "objective/train/weights_max": 1.4500267505645752,
      "objective/train/weights_min": 0.05458702892065048,
      "theoretical_loss": 3.7402048386000892,
      "tokens_seen": 838860800
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007533301235756701,
      "loss": 0.0754,
      "theoretical_loss": 3.7402048386000892,
      "tokens_seen": 838860800
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007532498796340877,
      "loss": 0.0739,
      "theoretical_loss": 3.740090862777091,
      "tokens_seen": 839122944
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007531696356925052,
      "loss": 0.0778,
      "theoretical_loss": 3.7399769325210697,
      "tokens_seen": 839385088
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007530893917509228,
      "loss": 0.0771,
      "theoretical_loss": 3.7398630477995853,
      "tokens_seen": 839647232
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007530091478093404,
      "loss": 0.0748,
      "theoretical_loss": 3.7397492085802315,
      "tokens_seen": 839909376
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000752928903867758,
      "loss": 0.0764,
      "theoretical_loss": 3.739635414830635,
      "tokens_seen": 840171520
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007528486599261755,
      "loss": 0.076,
      "theoretical_loss": 3.7395216665184554,
      "tokens_seen": 840433664
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007527684159845932,
      "loss": 0.0792,
      "theoretical_loss": 3.739407963611386,
      "tokens_seen": 840695808
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007526881720430108,
      "loss": 0.074,
      "theoretical_loss": 3.739294306077152,
      "tokens_seen": 840957952
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0007526079281014283,
      "loss": 0.0785,
      "theoretical_loss": 3.7391806938835126,
      "tokens_seen": 841220096
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000752527684159846,
      "loss": 0.0755,
      "theoretical_loss": 3.7390671269982603,
      "tokens_seen": 841482240
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007524474402182635,
      "loss": 0.0761,
      "theoretical_loss": 3.7389536053892187,
      "tokens_seen": 841744384
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007523671962766812,
      "loss": 0.0741,
      "theoretical_loss": 3.738840129024246,
      "tokens_seen": 842006528
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007522869523350987,
      "loss": 0.0771,
      "theoretical_loss": 3.738726697871233,
      "tokens_seen": 842268672
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007522067083935163,
      "loss": 0.0795,
      "theoretical_loss": 3.738613311898103,
      "tokens_seen": 842530816
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007521264644519339,
      "loss": 0.0762,
      "theoretical_loss": 3.7384999710728106,
      "tokens_seen": 842792960
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007520462205103514,
      "loss": 0.0768,
      "theoretical_loss": 3.738386675363346,
      "tokens_seen": 843055104
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000751965976568769,
      "loss": 0.0763,
      "theoretical_loss": 3.738273424737729,
      "tokens_seen": 843317248
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007518857326271867,
      "loss": 0.0792,
      "theoretical_loss": 3.7381602191640146,
      "tokens_seen": 843579392
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007518054886856043,
      "loss": 0.0769,
      "theoretical_loss": 3.738047058610289,
      "tokens_seen": 843841536
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007517252447440218,
      "loss": 0.0767,
      "theoretical_loss": 3.7379339430446707,
      "tokens_seen": 844103680
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007516450008024395,
      "loss": 0.075,
      "theoretical_loss": 3.7378208724353117,
      "tokens_seen": 844365824
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000751564756860857,
      "loss": 0.0776,
      "theoretical_loss": 3.7377078467503955,
      "tokens_seen": 844627968
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007514845129192746,
      "loss": 0.0779,
      "theoretical_loss": 3.737594865958138,
      "tokens_seen": 844890112
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007514042689776922,
      "loss": 0.0759,
      "theoretical_loss": 3.7374819300267883,
      "tokens_seen": 845152256
    },
    {
      "epoch": 0.26,
      "objective/train/advantage_avg": 0.0018314362969249487,
      "objective/train/docs_used": 312167,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4999068975448608,
      "objective/train/original_loss": 1.4999067783355713,
      "objective/train/theoretical_loss": 3.7373690389246272,
      "objective/train/tokens_used": 865874400,
      "objective/train/value_avg": -0.00806427001953125,
      "objective/train/value_loss": 0.00013757370470557362,
      "objective/train/value_max": -0.0001323223114013672,
      "objective/train/value_min": -0.31982421875,
      "objective/train/value_reward_corr": 0.6737398459252089,
      "objective/train/value_std": 0.01104736328125,
      "objective/train/weight_avg": 1.0018991231918335,
      "objective/train/weighted_lm_loss": 1.5027458667755127,
      "objective/train/weights_max": 1.1910135746002197,
      "objective/train/weights_min": 0.6367841958999634,
      "theoretical_loss": 3.7373690389246272,
      "tokens_seen": 845414400
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007513240250361097,
      "loss": 0.0768,
      "theoretical_loss": 3.7373690389246272,
      "tokens_seen": 845414400
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007512437810945275,
      "loss": 0.0783,
      "theoretical_loss": 3.737256192619967,
      "tokens_seen": 845676544
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000751163537152945,
      "loss": 0.0751,
      "theoretical_loss": 3.737143391081154,
      "tokens_seen": 845938688
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007510832932113626,
      "loss": 0.076,
      "theoretical_loss": 3.7370306342765653,
      "tokens_seen": 846200832
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007510030492697802,
      "loss": 0.0736,
      "theoretical_loss": 3.73691792217461,
      "tokens_seen": 846462976
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007509228053281978,
      "loss": 0.0778,
      "theoretical_loss": 3.7368052547437305,
      "tokens_seen": 846725120
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007508425613866153,
      "loss": 0.077,
      "theoretical_loss": 3.7366926319524003,
      "tokens_seen": 846987264
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007507623174450329,
      "loss": 0.0742,
      "theoretical_loss": 3.736580053769125,
      "tokens_seen": 847249408
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007506820735034505,
      "loss": 0.0727,
      "theoretical_loss": 3.736467520162442,
      "tokens_seen": 847511552
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000750601829561868,
      "loss": 0.0769,
      "theoretical_loss": 3.736355031100922,
      "tokens_seen": 847773696
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007505215856202858,
      "loss": 0.0763,
      "theoretical_loss": 3.7362425865531654,
      "tokens_seen": 848035840
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007504413416787033,
      "loss": 0.0773,
      "theoretical_loss": 3.736130186487806,
      "tokens_seen": 848297984
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007503610977371209,
      "loss": 0.0748,
      "theoretical_loss": 3.736017830873508,
      "tokens_seen": 848560128
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007502808537955385,
      "loss": 0.0769,
      "theoretical_loss": 3.7359055196789694,
      "tokens_seen": 848822272
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000750200609853956,
      "loss": 0.0747,
      "theoretical_loss": 3.7357932528729183,
      "tokens_seen": 849084416
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007501203659123736,
      "loss": 0.0785,
      "theoretical_loss": 3.7356810304241144,
      "tokens_seen": 849346560
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007500401219707912,
      "loss": 0.0781,
      "theoretical_loss": 3.73556885230135,
      "tokens_seen": 849608704
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007499598780292088,
      "loss": 0.0755,
      "theoretical_loss": 3.735456718473449,
      "tokens_seen": 849870848
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007498796340876264,
      "loss": 0.0716,
      "theoretical_loss": 3.7353446289092647,
      "tokens_seen": 850132992
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000749799390146044,
      "loss": 0.0779,
      "theoretical_loss": 3.7352325835776856,
      "tokens_seen": 850395136
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007497191462044616,
      "loss": 0.0772,
      "theoretical_loss": 3.7351205824476277,
      "tokens_seen": 850657280
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007496389022628792,
      "loss": 0.079,
      "theoretical_loss": 3.7350086254880415,
      "tokens_seen": 850919424
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007495586583212968,
      "loss": 0.0742,
      "theoretical_loss": 3.734896712667907,
      "tokens_seen": 851181568
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007494784143797143,
      "loss": 0.0753,
      "theoretical_loss": 3.734784843956236,
      "tokens_seen": 851443712
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000749398170438132,
      "loss": 0.0745,
      "theoretical_loss": 3.7346730193220727,
      "tokens_seen": 851705856
    },
    {
      "epoch": 0.26,
      "objective/train/advantage_avg": 0.0010687694884836674,
      "objective/train/docs_used": 314587,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.526440143585205,
      "objective/train/original_loss": 1.526440143585205,
      "objective/train/theoretical_loss": 3.7345612387344906,
      "objective/train/tokens_used": 872428000,
      "objective/train/value_avg": -0.00872802734375,
      "objective/train/value_loss": 0.0003840262070298195,
      "objective/train/value_max": -0.0002779960632324219,
      "objective/train/value_min": -0.62109375,
      "objective/train/value_reward_corr": 0.6480027416305001,
      "objective/train/value_std": 0.016021728515625,
      "objective/train/weight_avg": 1.0012387037277222,
      "objective/train/weighted_lm_loss": 1.527116060256958,
      "objective/train/weights_max": 1.4146990776062012,
      "objective/train/weights_min": 0.370330810546875,
      "theoretical_loss": 3.7345612387344906,
      "tokens_seen": 851968000
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007493179264965495,
      "loss": 0.0755,
      "theoretical_loss": 3.7345612387344906,
      "tokens_seen": 851968000
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007492376825549671,
      "loss": 0.078,
      "theoretical_loss": 3.734449502162596,
      "tokens_seen": 852230144
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007491574386133847,
      "loss": 0.0772,
      "theoretical_loss": 3.7343378095755257,
      "tokens_seen": 852492288
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007490771946718022,
      "loss": 0.0743,
      "theoretical_loss": 3.7342261609424483,
      "tokens_seen": 852754432
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007489969507302198,
      "loss": 0.0787,
      "theoretical_loss": 3.7341145562325613,
      "tokens_seen": 853016576
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007489167067886375,
      "loss": 0.0761,
      "theoretical_loss": 3.734002995415096,
      "tokens_seen": 853278720
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007488364628470551,
      "loss": 0.0771,
      "theoretical_loss": 3.7338914784593134,
      "tokens_seen": 853540864
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007487562189054726,
      "loss": 0.0754,
      "theoretical_loss": 3.733780005334505,
      "tokens_seen": 853803008
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007486759749638903,
      "loss": 0.0786,
      "theoretical_loss": 3.733668576009995,
      "tokens_seen": 854065152
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007485957310223078,
      "loss": 0.0764,
      "theoretical_loss": 3.733557190455136,
      "tokens_seen": 854327296
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007485154870807254,
      "loss": 0.074,
      "theoretical_loss": 3.733445848639313,
      "tokens_seen": 854589440
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000748435243139143,
      "loss": 0.0753,
      "theoretical_loss": 3.733334550531942,
      "tokens_seen": 854851584
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007483549991975605,
      "loss": 0.0776,
      "theoretical_loss": 3.7332232961024694,
      "tokens_seen": 855113728
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007482747552559783,
      "loss": 0.0758,
      "theoretical_loss": 3.7331120853203714,
      "tokens_seen": 855375872
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007481945113143958,
      "loss": 0.0749,
      "theoretical_loss": 3.733000918155156,
      "tokens_seen": 855638016
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007481142673728134,
      "loss": 0.0725,
      "theoretical_loss": 3.7328897945763617,
      "tokens_seen": 855900160
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000748034023431231,
      "loss": 0.078,
      "theoretical_loss": 3.7327787145535574,
      "tokens_seen": 856162304
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007479537794896486,
      "loss": 0.0736,
      "theoretical_loss": 3.732667678056342,
      "tokens_seen": 856424448
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007478735355480661,
      "loss": 0.0729,
      "theoretical_loss": 3.732556685054346,
      "tokens_seen": 856686592
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007477932916064837,
      "loss": 0.0744,
      "theoretical_loss": 3.7324457355172296,
      "tokens_seen": 856948736
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007477130476649013,
      "loss": 0.0746,
      "theoretical_loss": 3.7323348294146843,
      "tokens_seen": 857210880
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007476328037233188,
      "loss": 0.0757,
      "theoretical_loss": 3.73222396671643,
      "tokens_seen": 857473024
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007475525597817365,
      "loss": 0.0766,
      "theoretical_loss": 3.73211314739222,
      "tokens_seen": 857735168
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000747472315840154,
      "loss": 0.0748,
      "theoretical_loss": 3.732002371411835,
      "tokens_seen": 857997312
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007473920718985718,
      "loss": 0.0738,
      "theoretical_loss": 3.7318916387450876,
      "tokens_seen": 858259456
    },
    {
      "epoch": 0.26,
      "objective/train/advantage_avg": 0.0009312600013799965,
      "objective/train/docs_used": 317142,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5528024435043335,
      "objective/train/original_loss": 1.5528024435043335,
      "objective/train/theoretical_loss": 3.7317809493618204,
      "objective/train/tokens_used": 878981600,
      "objective/train/value_avg": -0.00759124755859375,
      "objective/train/value_loss": 0.0003197310143150389,
      "objective/train/value_max": -0.00022339820861816406,
      "objective/train/value_min": -0.65625,
      "objective/train/value_reward_corr": 0.5907966960002585,
      "objective/train/value_std": 0.01306915283203125,
      "objective/train/weight_avg": 1.0010712146759033,
      "objective/train/weighted_lm_loss": 1.5542237758636475,
      "objective/train/weights_max": 1.8819797039031982,
      "objective/train/weights_min": 0.3681640923023224,
      "theoretical_loss": 3.7317809493618204,
      "tokens_seen": 858521600
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007473118279569893,
      "loss": 0.0775,
      "theoretical_loss": 3.7317809493618204,
      "tokens_seen": 858521600
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007472315840154068,
      "loss": 0.0751,
      "theoretical_loss": 3.7316703032319056,
      "tokens_seen": 858783744
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007471513400738245,
      "loss": 0.0786,
      "theoretical_loss": 3.7315597003252474,
      "tokens_seen": 859045888
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000747071096132242,
      "loss": 0.0784,
      "theoretical_loss": 3.731449140611777,
      "tokens_seen": 859308032
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007469908521906596,
      "loss": 0.0748,
      "theoretical_loss": 3.7313386240614577,
      "tokens_seen": 859570176
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007469106082490772,
      "loss": 0.0761,
      "theoretical_loss": 3.7312281506442835,
      "tokens_seen": 859832320
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007468303643074948,
      "loss": 0.0757,
      "theoretical_loss": 3.7311177203302766,
      "tokens_seen": 860094464
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007467501203659124,
      "loss": 0.0766,
      "theoretical_loss": 3.7310073330894906,
      "tokens_seen": 860356608
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00074666987642433,
      "loss": 0.0739,
      "theoretical_loss": 3.730896988892008,
      "tokens_seen": 860618752
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007465896324827476,
      "loss": 0.0808,
      "theoretical_loss": 3.7307866877079414,
      "tokens_seen": 860880896
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007465093885411651,
      "loss": 0.0756,
      "theoretical_loss": 3.730676429507435,
      "tokens_seen": 861143040
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007464291445995828,
      "loss": 0.0736,
      "theoretical_loss": 3.730566214260659,
      "tokens_seen": 861405184
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007463489006580003,
      "loss": 0.0759,
      "theoretical_loss": 3.730456041937817,
      "tokens_seen": 861667328
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007462686567164179,
      "loss": 0.0766,
      "theoretical_loss": 3.730345912509141,
      "tokens_seen": 861929472
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007461884127748355,
      "loss": 0.074,
      "theoretical_loss": 3.7302358259448924,
      "tokens_seen": 862191616
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000746108168833253,
      "loss": 0.0776,
      "theoretical_loss": 3.730125782215362,
      "tokens_seen": 862453760
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007460279248916708,
      "loss": 0.0766,
      "theoretical_loss": 3.730015781290872,
      "tokens_seen": 862715904
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007459476809500883,
      "loss": 0.0783,
      "theoretical_loss": 3.729905823141771,
      "tokens_seen": 862978048
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007458674370085059,
      "loss": 0.0739,
      "theoretical_loss": 3.729795907738441,
      "tokens_seen": 863240192
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007457871930669235,
      "loss": 0.0746,
      "theoretical_loss": 3.729686035051291,
      "tokens_seen": 863502336
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007457069491253411,
      "loss": 0.0761,
      "theoretical_loss": 3.7295762050507593,
      "tokens_seen": 863764480
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007456267051837586,
      "loss": 0.0754,
      "theoretical_loss": 3.7294664177073145,
      "tokens_seen": 864026624
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007455464612421762,
      "loss": 0.0773,
      "theoretical_loss": 3.7293566729914547,
      "tokens_seen": 864288768
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007454662173005938,
      "loss": 0.0756,
      "theoretical_loss": 3.7292469708737066,
      "tokens_seen": 864550912
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007453859733590113,
      "loss": 0.0764,
      "theoretical_loss": 3.729137311324627,
      "tokens_seen": 864813056
    },
    {
      "epoch": 0.26,
      "objective/train/advantage_avg": -0.00015239574713632464,
      "objective/train/docs_used": 319421,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4767508506774902,
      "objective/train/original_loss": 1.4767508506774902,
      "objective/train/theoretical_loss": 3.7290276943148015,
      "objective/train/tokens_used": 885535200,
      "objective/train/value_avg": -0.00782012939453125,
      "objective/train/value_loss": 0.0005899901734665036,
      "objective/train/value_max": -0.00014889240264892578,
      "objective/train/value_min": -0.95166015625,
      "objective/train/value_reward_corr": 0.5327929289583835,
      "objective/train/value_std": 0.0201263427734375,
      "objective/train/weight_avg": 1.0001168251037598,
      "objective/train/weighted_lm_loss": 1.4773733615875244,
      "objective/train/weights_max": 2.5354816913604736,
      "objective/train/weights_min": 0.23555737733840942,
      "theoretical_loss": 3.7290276943148015,
      "tokens_seen": 865075200
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000745305729417429,
      "loss": 0.073,
      "theoretical_loss": 3.7290276943148015,
      "tokens_seen": 865075200
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007452254854758466,
      "loss": 0.0777,
      "theoretical_loss": 3.7289181198148458,
      "tokens_seen": 865337344
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007451452415342642,
      "loss": 0.0771,
      "theoretical_loss": 3.7288085877954025,
      "tokens_seen": 865599488
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007450649975926818,
      "loss": 0.0765,
      "theoretical_loss": 3.728699098227146,
      "tokens_seen": 865861632
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007449847536510994,
      "loss": 0.0757,
      "theoretical_loss": 3.728589651080779,
      "tokens_seen": 866123776
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000744904509709517,
      "loss": 0.0753,
      "theoretical_loss": 3.728480246327032,
      "tokens_seen": 866385920
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007448242657679345,
      "loss": 0.0744,
      "theoretical_loss": 3.7283708839366656,
      "tokens_seen": 866648064
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007447440218263521,
      "loss": 0.0766,
      "theoretical_loss": 3.72826156388047,
      "tokens_seen": 866910208
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007446637778847697,
      "loss": 0.0762,
      "theoretical_loss": 3.728152286129263,
      "tokens_seen": 867172352
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007445835339431873,
      "loss": 0.0744,
      "theoretical_loss": 3.728043050653893,
      "tokens_seen": 867434496
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007445032900016049,
      "loss": 0.0797,
      "theoretical_loss": 3.7279338574252354,
      "tokens_seen": 867696640
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007444230460600226,
      "loss": 0.0779,
      "theoretical_loss": 3.7278247064141956,
      "tokens_seen": 867958784
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007443428021184401,
      "loss": 0.0771,
      "theoretical_loss": 3.7277155975917076,
      "tokens_seen": 868220928
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007442625581768576,
      "loss": 0.0779,
      "theoretical_loss": 3.7276065309287345,
      "tokens_seen": 868483072
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007441823142352753,
      "loss": 0.0774,
      "theoretical_loss": 3.727497506396267,
      "tokens_seen": 868745216
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007441020702936928,
      "loss": 0.075,
      "theoretical_loss": 3.7273885239653266,
      "tokens_seen": 869007360
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007440218263521104,
      "loss": 0.0752,
      "theoretical_loss": 3.727279583606961,
      "tokens_seen": 869269504
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000743941582410528,
      "loss": 0.0763,
      "theoretical_loss": 3.727170685292248,
      "tokens_seen": 869531648
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007438613384689456,
      "loss": 0.0769,
      "theoretical_loss": 3.7270618289922943,
      "tokens_seen": 869793792
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007437810945273631,
      "loss": 0.0752,
      "theoretical_loss": 3.7269530146782337,
      "tokens_seen": 870055936
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007437008505857808,
      "loss": 0.0808,
      "theoretical_loss": 3.72684424232123,
      "tokens_seen": 870318080
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007436206066441984,
      "loss": 0.077,
      "theoretical_loss": 3.7267355118924748,
      "tokens_seen": 870580224
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000743540362702616,
      "loss": 0.0752,
      "theoretical_loss": 3.726626823363188,
      "tokens_seen": 870842368
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007434601187610336,
      "loss": 0.0754,
      "theoretical_loss": 3.7265181767046176,
      "tokens_seen": 871104512
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007433798748194511,
      "loss": 0.0772,
      "theoretical_loss": 3.726409571888042,
      "tokens_seen": 871366656
    },
    {
      "epoch": 0.26,
      "objective/train/advantage_avg": 0.0006801082054153085,
      "objective/train/docs_used": 321822,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.6108815670013428,
      "objective/train/original_loss": 1.6108815670013428,
      "objective/train/theoretical_loss": 3.7263010088847652,
      "objective/train/tokens_used": 892088800,
      "objective/train/value_avg": -0.00757598876953125,
      "objective/train/value_loss": 0.00016946658433880657,
      "objective/train/value_max": -0.00016224384307861328,
      "objective/train/value_min": -0.28515625,
      "objective/train/value_reward_corr": 0.7011885448576807,
      "objective/train/value_std": 0.01305389404296875,
      "objective/train/weight_avg": 1.0007609128952026,
      "objective/train/weighted_lm_loss": 1.6116539239883423,
      "objective/train/weights_max": 1.1948398351669312,
      "objective/train/weights_min": 0.39612501859664917,
      "theoretical_loss": 3.7263010088847652,
      "tokens_seen": 871628800
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007432996308778688,
      "loss": 0.0783,
      "theoretical_loss": 3.7263010088847652,
      "tokens_seen": 871628800
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007432193869362863,
      "loss": 0.0751,
      "theoretical_loss": 3.726192487666121,
      "tokens_seen": 871890944
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007431391429947038,
      "loss": 0.0762,
      "theoretical_loss": 3.7260840082034714,
      "tokens_seen": 872153088
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007430588990531216,
      "loss": 0.0758,
      "theoretical_loss": 3.7259755704682065,
      "tokens_seen": 872415232
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007429786551115391,
      "loss": 0.0765,
      "theoretical_loss": 3.7258671744317446,
      "tokens_seen": 872677376
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007428984111699567,
      "loss": 0.0767,
      "theoretical_loss": 3.725758820065531,
      "tokens_seen": 872939520
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007428181672283743,
      "loss": 0.0759,
      "theoretical_loss": 3.725650507341042,
      "tokens_seen": 873201664
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007427379232867919,
      "loss": 0.0773,
      "theoretical_loss": 3.7255422362297788,
      "tokens_seen": 873463808
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007426576793452094,
      "loss": 0.0758,
      "theoretical_loss": 3.7254340067032725,
      "tokens_seen": 873725952
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000742577435403627,
      "loss": 0.0765,
      "theoretical_loss": 3.7253258187330816,
      "tokens_seen": 873988096
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0007424971914620446,
      "loss": 0.0761,
      "theoretical_loss": 3.7252176722907926,
      "tokens_seen": 874250240
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007424169475204621,
      "loss": 0.0786,
      "theoretical_loss": 3.725109567348021,
      "tokens_seen": 874512384
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007423367035788799,
      "loss": 0.0777,
      "theoretical_loss": 3.725001503876408,
      "tokens_seen": 874774528
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007422564596372974,
      "loss": 0.0753,
      "theoretical_loss": 3.7248934818476247,
      "tokens_seen": 875036672
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007421762156957151,
      "loss": 0.0756,
      "theoretical_loss": 3.7247855012333693,
      "tokens_seen": 875298816
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007420959717541326,
      "loss": 0.0758,
      "theoretical_loss": 3.7246775620053665,
      "tokens_seen": 875560960
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007420157278125502,
      "loss": 0.0766,
      "theoretical_loss": 3.724569664135372,
      "tokens_seen": 875823104
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007419354838709678,
      "loss": 0.0761,
      "theoretical_loss": 3.7244618075951657,
      "tokens_seen": 876085248
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007418552399293853,
      "loss": 0.0779,
      "theoretical_loss": 3.7243539923565576,
      "tokens_seen": 876347392
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007417749959878029,
      "loss": 0.0773,
      "theoretical_loss": 3.724246218391384,
      "tokens_seen": 876609536
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007416947520462205,
      "loss": 0.0779,
      "theoretical_loss": 3.7241384856715096,
      "tokens_seen": 876871680
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007416145081046381,
      "loss": 0.0754,
      "theoretical_loss": 3.724030794168826,
      "tokens_seen": 877133824
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007415342641630557,
      "loss": 0.0752,
      "theoretical_loss": 3.723923143855253,
      "tokens_seen": 877395968
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007414540202214734,
      "loss": 0.0744,
      "theoretical_loss": 3.723815534702738,
      "tokens_seen": 877658112
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007413737762798909,
      "loss": 0.0784,
      "theoretical_loss": 3.7237079666832553,
      "tokens_seen": 877920256
    },
    {
      "epoch": 0.27,
      "objective/train/advantage_avg": 9.239621431333944e-05,
      "objective/train/docs_used": 324331,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5191644430160522,
      "objective/train/original_loss": 1.5191644430160522,
      "objective/train/theoretical_loss": 3.7236004397688065,
      "objective/train/tokens_used": 898642400,
      "objective/train/value_avg": -0.01021575927734375,
      "objective/train/value_loss": 0.0004930080031044781,
      "objective/train/value_max": -0.0001596212387084961,
      "objective/train/value_min": -0.9365234375,
      "objective/train/value_reward_corr": 0.6092888427252324,
      "objective/train/value_std": 0.017242431640625,
      "objective/train/weight_avg": 1.0003056526184082,
      "objective/train/weighted_lm_loss": 1.51966392993927,
      "objective/train/weights_max": 1.5377529859542847,
      "objective/train/weights_min": 0.36856982111930847,
      "theoretical_loss": 3.7236004397688065,
      "tokens_seen": 878182400
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007412935323383084,
      "loss": 0.0779,
      "theoretical_loss": 3.7236004397688065,
      "tokens_seen": 878182400
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007412132883967261,
      "loss": 0.0754,
      "theoretical_loss": 3.723492953931421,
      "tokens_seen": 878444544
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007411330444551436,
      "loss": 0.0758,
      "theoretical_loss": 3.7233855091431565,
      "tokens_seen": 878706688
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007410528005135613,
      "loss": 0.0759,
      "theoretical_loss": 3.723278105376096,
      "tokens_seen": 878968832
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007409725565719788,
      "loss": 0.0747,
      "theoretical_loss": 3.723170742602351,
      "tokens_seen": 879230976
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007408923126303964,
      "loss": 0.0772,
      "theoretical_loss": 3.7230634207940607,
      "tokens_seen": 879493120
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007408120686888141,
      "loss": 0.0771,
      "theoretical_loss": 3.7229561399233906,
      "tokens_seen": 879755264
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007407318247472316,
      "loss": 0.0789,
      "theoretical_loss": 3.7228488999625338,
      "tokens_seen": 880017408
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007406515808056492,
      "loss": 0.0773,
      "theoretical_loss": 3.722741700883711,
      "tokens_seen": 880279552
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007405713368640668,
      "loss": 0.0755,
      "theoretical_loss": 3.7226345426591694,
      "tokens_seen": 880541696
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007404910929224844,
      "loss": 0.073,
      "theoretical_loss": 3.722527425261183,
      "tokens_seen": 880803840
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007404108489809019,
      "loss": 0.0782,
      "theoretical_loss": 3.7224203486620535,
      "tokens_seen": 881065984
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007403306050393196,
      "loss": 0.0775,
      "theoretical_loss": 3.7223133128341104,
      "tokens_seen": 881328128
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007402503610977371,
      "loss": 0.0751,
      "theoretical_loss": 3.722206317749708,
      "tokens_seen": 881590272
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007401701171561546,
      "loss": 0.0777,
      "theoretical_loss": 3.722099363381229,
      "tokens_seen": 881852416
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007400898732145724,
      "loss": 0.0767,
      "theoretical_loss": 3.7219924497010837,
      "tokens_seen": 882114560
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007400096292729899,
      "loss": 0.0767,
      "theoretical_loss": 3.721885576681708,
      "tokens_seen": 882376704
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007399293853314075,
      "loss": 0.0778,
      "theoretical_loss": 3.7217787442955643,
      "tokens_seen": 882638848
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007398491413898251,
      "loss": 0.0765,
      "theoretical_loss": 3.721671952515144,
      "tokens_seen": 882900992
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007397688974482427,
      "loss": 0.0761,
      "theoretical_loss": 3.7215652013129628,
      "tokens_seen": 883163136
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007396886535066603,
      "loss": 0.0756,
      "theoretical_loss": 3.7214584906615644,
      "tokens_seen": 883425280
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007396084095650778,
      "loss": 0.0732,
      "theoretical_loss": 3.7213518205335196,
      "tokens_seen": 883687424
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007395281656234954,
      "loss": 0.0764,
      "theoretical_loss": 3.721245190901425,
      "tokens_seen": 883949568
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.000739447921681913,
      "loss": 0.0768,
      "theoretical_loss": 3.721138601737904,
      "tokens_seen": 884211712
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007393676777403307,
      "loss": 0.0753,
      "theoretical_loss": 3.721032053015607,
      "tokens_seen": 884473856
    },
    {
      "epoch": 0.27,
      "objective/train/advantage_avg": 0.0008710287511348724,
      "objective/train/docs_used": 326745,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.574365496635437,
      "objective/train/original_loss": 1.5743653774261475,
      "objective/train/theoretical_loss": 3.720925544707211,
      "objective/train/tokens_used": 905196000,
      "objective/train/value_avg": -0.007781982421875,
      "objective/train/value_loss": 0.00020630114886444062,
      "objective/train/value_max": -0.00011235475540161133,
      "objective/train/value_min": -0.50830078125,
      "objective/train/value_reward_corr": 0.6715271500942702,
      "objective/train/value_std": 0.013824462890625,
      "objective/train/weight_avg": 1.0009685754776,
      "objective/train/weighted_lm_loss": 1.576448917388916,
      "objective/train/weights_max": 1.275231957435608,
      "objective/train/weights_min": 0.3771900534629822,
      "theoretical_loss": 3.720925544707211,
      "tokens_seen": 884736000
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007392874337987482,
      "loss": 0.0758,
      "theoretical_loss": 3.720925544707211,
      "tokens_seen": 884736000
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007392071898571659,
      "loss": 0.0778,
      "theoretical_loss": 3.720819076785419,
      "tokens_seen": 884998144
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007391269459155834,
      "loss": 0.074,
      "theoretical_loss": 3.720712649222961,
      "tokens_seen": 885260288
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007390467019740009,
      "loss": 0.075,
      "theoretical_loss": 3.720606261992593,
      "tokens_seen": 885522432
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007389664580324186,
      "loss": 0.0775,
      "theoretical_loss": 3.7204999150670988,
      "tokens_seen": 885784576
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007388862140908361,
      "loss": 0.0753,
      "theoretical_loss": 3.7203936084192866,
      "tokens_seen": 886046720
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007388059701492537,
      "loss": 0.0772,
      "theoretical_loss": 3.720287342021992,
      "tokens_seen": 886308864
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007387257262076713,
      "loss": 0.0745,
      "theoretical_loss": 3.720181115848078,
      "tokens_seen": 886571008
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.000738645482266089,
      "loss": 0.074,
      "theoretical_loss": 3.7200749298704316,
      "tokens_seen": 886833152
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007385652383245065,
      "loss": 0.0778,
      "theoretical_loss": 3.7199687840619675,
      "tokens_seen": 887095296
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007384849943829242,
      "loss": 0.0766,
      "theoretical_loss": 3.719862678395627,
      "tokens_seen": 887357440
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007384047504413417,
      "loss": 0.0755,
      "theoretical_loss": 3.719756612844377,
      "tokens_seen": 887619584
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007383245064997593,
      "loss": 0.0737,
      "theoretical_loss": 3.7196505873812105,
      "tokens_seen": 887881728
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007382442625581769,
      "loss": 0.0764,
      "theoretical_loss": 3.7195446019791465,
      "tokens_seen": 888143872
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007381640186165944,
      "loss": 0.0755,
      "theoretical_loss": 3.7194386566112314,
      "tokens_seen": 888406016
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007380837746750121,
      "loss": 0.0737,
      "theoretical_loss": 3.7193327512505356,
      "tokens_seen": 888668160
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007380035307334296,
      "loss": 0.0773,
      "theoretical_loss": 3.7192268858701576,
      "tokens_seen": 888930304
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007379232867918472,
      "loss": 0.0742,
      "theoretical_loss": 3.7191210604432205,
      "tokens_seen": 889192448
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007378430428502649,
      "loss": 0.0774,
      "theoretical_loss": 3.7190152749428735,
      "tokens_seen": 889454592
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007377627989086824,
      "loss": 0.0758,
      "theoretical_loss": 3.7189095293422927,
      "tokens_seen": 889716736
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007376825549671,
      "loss": 0.0772,
      "theoretical_loss": 3.71880382361468,
      "tokens_seen": 889978880
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007376023110255176,
      "loss": 0.0766,
      "theoretical_loss": 3.7186981577332614,
      "tokens_seen": 890241024
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007375220670839352,
      "loss": 0.0735,
      "theoretical_loss": 3.718592531671291,
      "tokens_seen": 890503168
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007374418231423527,
      "loss": 0.0759,
      "theoretical_loss": 3.7184869454020477,
      "tokens_seen": 890765312
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007373615792007704,
      "loss": 0.0747,
      "theoretical_loss": 3.7183813988988357,
      "tokens_seen": 891027456
    },
    {
      "epoch": 0.27,
      "objective/train/advantage_avg": 0.0005592244560830295,
      "objective/train/docs_used": 329220,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.468513011932373,
      "objective/train/original_loss": 1.468513011932373,
      "objective/train/theoretical_loss": 3.7182758921349865,
      "objective/train/tokens_used": 911749600,
      "objective/train/value_avg": -0.00971221923828125,
      "objective/train/value_loss": 0.0005271669942885637,
      "objective/train/value_max": -0.00017535686492919922,
      "objective/train/value_min": -0.74658203125,
      "objective/train/value_reward_corr": 0.6262416599375313,
      "objective/train/value_std": 0.0177154541015625,
      "objective/train/weight_avg": 1.0007938146591187,
      "objective/train/weighted_lm_loss": 1.4691156148910522,
      "objective/train/weights_max": 1.5486183166503906,
      "objective/train/weights_min": 0.36942073702812195,
      "theoretical_loss": 3.7182758921349865,
      "tokens_seen": 891289600
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007372813352591879,
      "loss": 0.078,
      "theoretical_loss": 3.7182758921349865,
      "tokens_seen": 891289600
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007372010913176055,
      "loss": 0.076,
      "theoretical_loss": 3.718170425083856,
      "tokens_seen": 891551744
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007371208473760232,
      "loss": 0.0755,
      "theoretical_loss": 3.718064997718826,
      "tokens_seen": 891813888
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007370406034344407,
      "loss": 0.0754,
      "theoretical_loss": 3.7179596100133034,
      "tokens_seen": 892076032
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007369603594928584,
      "loss": 0.078,
      "theoretical_loss": 3.7178542619407233,
      "tokens_seen": 892338176
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007368801155512759,
      "loss": 0.0785,
      "theoretical_loss": 3.7177489534745427,
      "tokens_seen": 892600320
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007367998716096935,
      "loss": 0.0786,
      "theoretical_loss": 3.717643684588247,
      "tokens_seen": 892862464
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007367196276681111,
      "loss": 0.0766,
      "theoretical_loss": 3.7175384552553457,
      "tokens_seen": 893124608
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007366393837265286,
      "loss": 0.0739,
      "theoretical_loss": 3.7174332654493742,
      "tokens_seen": 893386752
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007365591397849462,
      "loss": 0.0766,
      "theoretical_loss": 3.717328115143894,
      "tokens_seen": 893648896
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007364788958433638,
      "loss": 0.0766,
      "theoretical_loss": 3.717223004312491,
      "tokens_seen": 893911040
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007363986519017815,
      "loss": 0.0767,
      "theoretical_loss": 3.717117932928777,
      "tokens_seen": 894173184
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.000736318407960199,
      "loss": 0.0752,
      "theoretical_loss": 3.7170129009663886,
      "tokens_seen": 894435328
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007362381640186167,
      "loss": 0.0777,
      "theoretical_loss": 3.716907908398989,
      "tokens_seen": 894697472
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007361579200770342,
      "loss": 0.0756,
      "theoretical_loss": 3.7168029552002655,
      "tokens_seen": 894959616
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007360776761354517,
      "loss": 0.0764,
      "theoretical_loss": 3.716698041343931,
      "tokens_seen": 895221760
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007359974321938694,
      "loss": 0.0779,
      "theoretical_loss": 3.716593166803724,
      "tokens_seen": 895483904
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007359171882522869,
      "loss": 0.0734,
      "theoretical_loss": 3.7164883315534087,
      "tokens_seen": 895746048
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007358369443107046,
      "loss": 0.0773,
      "theoretical_loss": 3.7163835355667723,
      "tokens_seen": 896008192
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007357567003691221,
      "loss": 0.074,
      "theoretical_loss": 3.7162787788176295,
      "tokens_seen": 896270336
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007356764564275397,
      "loss": 0.0762,
      "theoretical_loss": 3.716174061279819,
      "tokens_seen": 896532480
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007355962124859574,
      "loss": 0.0739,
      "theoretical_loss": 3.7160693829272047,
      "tokens_seen": 896794624
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.000735515968544375,
      "loss": 0.0746,
      "theoretical_loss": 3.715964743733676,
      "tokens_seen": 897056768
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007354357246027925,
      "loss": 0.0778,
      "theoretical_loss": 3.7158601436731464,
      "tokens_seen": 897318912
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007353554806612101,
      "loss": 0.0749,
      "theoretical_loss": 3.715755582719556,
      "tokens_seen": 897581056
    },
    {
      "epoch": 0.27,
      "objective/train/advantage_avg": 0.0012793752830475569,
      "objective/train/docs_used": 331577,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5417509078979492,
      "objective/train/original_loss": 1.5417510271072388,
      "objective/train/theoretical_loss": 3.7156510608468674,
      "objective/train/tokens_used": 918303200,
      "objective/train/value_avg": -0.007205963134765625,
      "objective/train/value_loss": 0.0003468697250355035,
      "objective/train/value_max": -8.684396743774414e-05,
      "objective/train/value_min": -0.890625,
      "objective/train/value_reward_corr": 0.5998975412340294,
      "objective/train/value_std": 0.014312744140625,
      "objective/train/weight_avg": 1.0014266967773438,
      "objective/train/weighted_lm_loss": 1.5438824892044067,
      "objective/train/weights_max": 1.7019463777542114,
      "objective/train/weights_min": 0.2276768684387207,
      "theoretical_loss": 3.7156510608468674,
      "tokens_seen": 897843200
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007352752367196277,
      "loss": 0.072,
      "theoretical_loss": 3.7156510608468674,
      "tokens_seen": 897843200
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007351949927780452,
      "loss": 0.0769,
      "theoretical_loss": 3.7155465780290706,
      "tokens_seen": 898105344
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007351147488364629,
      "loss": 0.0755,
      "theoretical_loss": 3.7154421342401793,
      "tokens_seen": 898367488
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007350345048948804,
      "loss": 0.0757,
      "theoretical_loss": 3.7153377294542325,
      "tokens_seen": 898629632
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.000734954260953298,
      "loss": 0.0756,
      "theoretical_loss": 3.715233363645293,
      "tokens_seen": 898891776
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007348740170117157,
      "loss": 0.0757,
      "theoretical_loss": 3.7151290367874497,
      "tokens_seen": 899153920
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007347937730701332,
      "loss": 0.0744,
      "theoretical_loss": 3.715024748854815,
      "tokens_seen": 899416064
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007347135291285509,
      "loss": 0.0753,
      "theoretical_loss": 3.714920499821528,
      "tokens_seen": 899678208
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007346332851869684,
      "loss": 0.0746,
      "theoretical_loss": 3.7148162896617505,
      "tokens_seen": 899940352
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.000734553041245386,
      "loss": 0.074,
      "theoretical_loss": 3.714712118349669,
      "tokens_seen": 900202496
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007344727973038036,
      "loss": 0.0788,
      "theoretical_loss": 3.7146079858594976,
      "tokens_seen": 900464640
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007343925533622212,
      "loss": 0.0753,
      "theoretical_loss": 3.714503892165471,
      "tokens_seen": 900726784
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007343123094206387,
      "loss": 0.0758,
      "theoretical_loss": 3.714399837241851,
      "tokens_seen": 900988928
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007342320654790563,
      "loss": 0.0748,
      "theoretical_loss": 3.7142958210629233,
      "tokens_seen": 901251072
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.000734151821537474,
      "loss": 0.0769,
      "theoretical_loss": 3.714191843602998,
      "tokens_seen": 901513216
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007340715775958915,
      "loss": 0.0769,
      "theoretical_loss": 3.7140879048364104,
      "tokens_seen": 901775360
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007339913336543092,
      "loss": 0.0756,
      "theoretical_loss": 3.7139840047375183,
      "tokens_seen": 902037504
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007339110897127267,
      "loss": 0.0759,
      "theoretical_loss": 3.713880143280707,
      "tokens_seen": 902299648
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007338308457711443,
      "loss": 0.0758,
      "theoretical_loss": 3.713776320440383,
      "tokens_seen": 902561792
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007337506018295619,
      "loss": 0.0756,
      "theoretical_loss": 3.7136725361909795,
      "tokens_seen": 902823936
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007336703578879794,
      "loss": 0.0763,
      "theoretical_loss": 3.713568790506953,
      "tokens_seen": 903086080
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.000733590113946397,
      "loss": 0.0745,
      "theoretical_loss": 3.7134650833627854,
      "tokens_seen": 903348224
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007335098700048146,
      "loss": 0.0735,
      "theoretical_loss": 3.7133614147329808,
      "tokens_seen": 903610368
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007334296260632322,
      "loss": 0.0775,
      "theoretical_loss": 3.7132577845920696,
      "tokens_seen": 903872512
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007333493821216499,
      "loss": 0.076,
      "theoretical_loss": 3.7131541929146055,
      "tokens_seen": 904134656
    },
    {
      "epoch": 0.27,
      "objective/train/advantage_avg": -9.983734344132245e-05,
      "objective/train/docs_used": 333943,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4496132135391235,
      "objective/train/original_loss": 1.449613094329834,
      "objective/train/theoretical_loss": 3.713050639675166,
      "objective/train/tokens_used": 924856800,
      "objective/train/value_avg": -0.00740814208984375,
      "objective/train/value_loss": 0.00036323920357972383,
      "objective/train/value_max": -0.00014650821685791016,
      "objective/train/value_min": -0.87744140625,
      "objective/train/value_reward_corr": 0.7248527212737947,
      "objective/train/value_std": 0.014312744140625,
      "objective/train/weight_avg": 1.000061273574829,
      "objective/train/weighted_lm_loss": 1.4491292238235474,
      "objective/train/weights_max": 1.8285359144210815,
      "objective/train/weights_min": 0.3715519905090332,
      "theoretical_loss": 3.713050639675166,
      "tokens_seen": 904396800
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007332691381800675,
      "loss": 0.0729,
      "theoretical_loss": 3.713050639675166,
      "tokens_seen": 904396800
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.000733188894238485,
      "loss": 0.0731,
      "theoretical_loss": 3.712947124848354,
      "tokens_seen": 904658944
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007331086502969026,
      "loss": 0.0771,
      "theoretical_loss": 3.7128436484087954,
      "tokens_seen": 904921088
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007330284063553202,
      "loss": 0.0748,
      "theoretical_loss": 3.712740210331141,
      "tokens_seen": 905183232
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007329481624137377,
      "loss": 0.0769,
      "theoretical_loss": 3.712636810590065,
      "tokens_seen": 905445376
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007328679184721554,
      "loss": 0.0736,
      "theoretical_loss": 3.7125334491602664,
      "tokens_seen": 905707520
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007327876745305729,
      "loss": 0.0743,
      "theoretical_loss": 3.712430126016467,
      "tokens_seen": 905969664
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007327074305889905,
      "loss": 0.0751,
      "theoretical_loss": 3.7123268411334136,
      "tokens_seen": 906231808
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007326271866474082,
      "loss": 0.0736,
      "theoretical_loss": 3.7122235944858772,
      "tokens_seen": 906493952
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007325469427058258,
      "loss": 0.0734,
      "theoretical_loss": 3.712120386048652,
      "tokens_seen": 906756096
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007324666987642433,
      "loss": 0.0758,
      "theoretical_loss": 3.712017215796556,
      "tokens_seen": 907018240
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0007323864548226609,
      "loss": 0.0744,
      "theoretical_loss": 3.7119140837044315,
      "tokens_seen": 907280384
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007323062108810785,
      "loss": 0.0735,
      "theoretical_loss": 3.7118109897471445,
      "tokens_seen": 907542528
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000732225966939496,
      "loss": 0.0732,
      "theoretical_loss": 3.7117079338995858,
      "tokens_seen": 907804672
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007321457229979137,
      "loss": 0.0767,
      "theoretical_loss": 3.7116049161366673,
      "tokens_seen": 908066816
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007320654790563312,
      "loss": 0.0727,
      "theoretical_loss": 3.7115019364333275,
      "tokens_seen": 908328960
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000731985235114749,
      "loss": 0.0761,
      "theoretical_loss": 3.7113989947645276,
      "tokens_seen": 908591104
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007319049911731665,
      "loss": 0.0754,
      "theoretical_loss": 3.711296091105252,
      "tokens_seen": 908853248
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000731824747231584,
      "loss": 0.0722,
      "theoretical_loss": 3.7111932254305096,
      "tokens_seen": 909115392
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007317445032900017,
      "loss": 0.0747,
      "theoretical_loss": 3.7110903977153313,
      "tokens_seen": 909377536
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007316642593484192,
      "loss": 0.0757,
      "theoretical_loss": 3.710987607934774,
      "tokens_seen": 909639680
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007315840154068368,
      "loss": 0.0759,
      "theoretical_loss": 3.7108848560639167,
      "tokens_seen": 909901824
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007315037714652544,
      "loss": 0.0753,
      "theoretical_loss": 3.7107821420778615,
      "tokens_seen": 910163968
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000731423527523672,
      "loss": 0.0744,
      "theoretical_loss": 3.7106794659517357,
      "tokens_seen": 910426112
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007313432835820895,
      "loss": 0.0729,
      "theoretical_loss": 3.7105768276606885,
      "tokens_seen": 910688256
    },
    {
      "epoch": 0.28,
      "objective/train/advantage_avg": -0.001721195294521749,
      "objective/train/docs_used": 336141,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5599461793899536,
      "objective/train/original_loss": 1.5599462985992432,
      "objective/train/theoretical_loss": 3.710474227179893,
      "objective/train/tokens_used": 931410400,
      "objective/train/value_avg": -0.006168365478515625,
      "objective/train/value_loss": 0.00024355393543373793,
      "objective/train/value_max": -0.0001398324966430664,
      "objective/train/value_min": -0.4716796875,
      "objective/train/value_reward_corr": 0.6900471509482724,
      "objective/train/value_std": 0.0086822509765625,
      "objective/train/weight_avg": 0.9983928203582764,
      "objective/train/weighted_lm_loss": 1.5585672855377197,
      "objective/train/weights_max": 1.1376217603683472,
      "objective/train/weights_min": 0.3742261826992035,
      "theoretical_loss": 3.710474227179893,
      "tokens_seen": 910950400
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007312630396405071,
      "loss": 0.0743,
      "theoretical_loss": 3.710474227179893,
      "tokens_seen": 910950400
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007311827956989248,
      "loss": 0.0741,
      "theoretical_loss": 3.710371664484547,
      "tokens_seen": 911212544
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007311025517573423,
      "loss": 0.0714,
      "theoretical_loss": 3.710269139549869,
      "tokens_seen": 911474688
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00073102230781576,
      "loss": 0.0724,
      "theoretical_loss": 3.7101666523511034,
      "tokens_seen": 911736832
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007309420638741775,
      "loss": 0.075,
      "theoretical_loss": 3.710064202863517,
      "tokens_seen": 911998976
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007308618199325952,
      "loss": 0.077,
      "theoretical_loss": 3.7099617910623994,
      "tokens_seen": 912261120
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007307815759910127,
      "loss": 0.0751,
      "theoretical_loss": 3.7098594169230648,
      "tokens_seen": 912523264
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007307013320494302,
      "loss": 0.0732,
      "theoretical_loss": 3.7097570804208497,
      "tokens_seen": 912785408
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007306210881078479,
      "loss": 0.074,
      "theoretical_loss": 3.709654781531113,
      "tokens_seen": 913047552
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007305408441662654,
      "loss": 0.0752,
      "theoretical_loss": 3.709552520229239,
      "tokens_seen": 913309696
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000730460600224683,
      "loss": 0.074,
      "theoretical_loss": 3.7094502964906337,
      "tokens_seen": 913571840
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007303803562831007,
      "loss": 0.0741,
      "theoretical_loss": 3.709348110290726,
      "tokens_seen": 913833984
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007303001123415183,
      "loss": 0.0742,
      "theoretical_loss": 3.7092459616049682,
      "tokens_seen": 914096128
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007302198683999358,
      "loss": 0.0745,
      "theoretical_loss": 3.709143850408837,
      "tokens_seen": 914358272
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007301396244583534,
      "loss": 0.074,
      "theoretical_loss": 3.7090417766778305,
      "tokens_seen": 914620416
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000730059380516771,
      "loss": 0.0736,
      "theoretical_loss": 3.7089397403874704,
      "tokens_seen": 914882560
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007299791365751885,
      "loss": 0.0742,
      "theoretical_loss": 3.708837741513301,
      "tokens_seen": 915144704
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007298988926336062,
      "loss": 0.0751,
      "theoretical_loss": 3.7087357800308904,
      "tokens_seen": 915406848
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007298186486920237,
      "loss": 0.0734,
      "theoretical_loss": 3.708633855915829,
      "tokens_seen": 915668992
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007297384047504413,
      "loss": 0.0739,
      "theoretical_loss": 3.708531969143731,
      "tokens_seen": 915931136
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000729658160808859,
      "loss": 0.0766,
      "theoretical_loss": 3.708430119690232,
      "tokens_seen": 916193280
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007295779168672765,
      "loss": 0.0748,
      "theoretical_loss": 3.708328307530991,
      "tokens_seen": 916455424
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007294976729256942,
      "loss": 0.0736,
      "theoretical_loss": 3.7082265326416914,
      "tokens_seen": 916717568
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007294174289841117,
      "loss": 0.0732,
      "theoretical_loss": 3.708124794998037,
      "tokens_seen": 916979712
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007293371850425293,
      "loss": 0.0722,
      "theoretical_loss": 3.708023094575756,
      "tokens_seen": 917241856
    },
    {
      "epoch": 0.28,
      "objective/train/advantage_avg": 0.0011513980571180582,
      "objective/train/docs_used": 338425,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.586716890335083,
      "objective/train/original_loss": 1.5867170095443726,
      "objective/train/theoretical_loss": 3.7079214313505986,
      "objective/train/tokens_used": 937964000,
      "objective/train/value_avg": -0.0078125,
      "objective/train/value_loss": 0.00020614459936041385,
      "objective/train/value_max": -0.00014650821685791016,
      "objective/train/value_min": -0.955078125,
      "objective/train/value_reward_corr": 0.6817206994700864,
      "objective/train/value_std": 0.014862060546875,
      "objective/train/weight_avg": 1.0012516975402832,
      "objective/train/weighted_lm_loss": 1.588771939277649,
      "objective/train/weights_max": 2.5229907035827637,
      "objective/train/weights_min": 0.39287444949150085,
      "theoretical_loss": 3.7079214313505986,
      "tokens_seen": 917504000
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007292569411009469,
      "loss": 0.0756,
      "theoretical_loss": 3.7079214313505986,
      "tokens_seen": 917504000
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007291766971593645,
      "loss": 0.073,
      "theoretical_loss": 3.707819805298338,
      "tokens_seen": 917766144
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000729096453217782,
      "loss": 0.0756,
      "theoretical_loss": 3.7077182163947704,
      "tokens_seen": 918028288
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007290162092761998,
      "loss": 0.0748,
      "theoretical_loss": 3.7076166646157134,
      "tokens_seen": 918290432
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007289359653346173,
      "loss": 0.075,
      "theoretical_loss": 3.7075151499370094,
      "tokens_seen": 918552576
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007288557213930348,
      "loss": 0.0768,
      "theoretical_loss": 3.7074136723345212,
      "tokens_seen": 918814720
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007287754774514525,
      "loss": 0.0768,
      "theoretical_loss": 3.707312231784136,
      "tokens_seen": 919076864
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00072869523350987,
      "loss": 0.0745,
      "theoretical_loss": 3.7072108282617617,
      "tokens_seen": 919339008
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007286149895682876,
      "loss": 0.0756,
      "theoretical_loss": 3.7071094617433307,
      "tokens_seen": 919601152
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007285347456267052,
      "loss": 0.0748,
      "theoretical_loss": 3.707008132204796,
      "tokens_seen": 919863296
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007284545016851228,
      "loss": 0.0734,
      "theoretical_loss": 3.7069068396221345,
      "tokens_seen": 920125440
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007283742577435403,
      "loss": 0.0732,
      "theoretical_loss": 3.706805583971345,
      "tokens_seen": 920387584
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007282940138019579,
      "loss": 0.0756,
      "theoretical_loss": 3.7067043652284495,
      "tokens_seen": 920649728
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007282137698603756,
      "loss": 0.0809,
      "theoretical_loss": 3.7066031833694906,
      "tokens_seen": 920911872
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007281335259187932,
      "loss": 0.0763,
      "theoretical_loss": 3.7065020383705347,
      "tokens_seen": 921174016
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007280532819772108,
      "loss": 0.0719,
      "theoretical_loss": 3.70640093020767,
      "tokens_seen": 921436160
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007279730380356283,
      "loss": 0.0762,
      "theoretical_loss": 3.7062998588570073,
      "tokens_seen": 921698304
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000727892794094046,
      "loss": 0.074,
      "theoretical_loss": 3.7061988242946793,
      "tokens_seen": 921960448
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007278125501524635,
      "loss": 0.0744,
      "theoretical_loss": 3.7060978264968423,
      "tokens_seen": 922222592
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000727732306210881,
      "loss": 0.0749,
      "theoretical_loss": 3.705996865439672,
      "tokens_seen": 922484736
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007276520622692987,
      "loss": 0.0727,
      "theoretical_loss": 3.7058959410993695,
      "tokens_seen": 922746880
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007275718183277162,
      "loss": 0.0743,
      "theoretical_loss": 3.7057950534521558,
      "tokens_seen": 923009024
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007274915743861338,
      "loss": 0.0763,
      "theoretical_loss": 3.705694202474275,
      "tokens_seen": 923271168
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007274113304445515,
      "loss": 0.0742,
      "theoretical_loss": 3.7055933881419936,
      "tokens_seen": 923533312
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007273310865029691,
      "loss": 0.0705,
      "theoretical_loss": 3.7054926104315995,
      "tokens_seen": 923795456
    },
    {
      "epoch": 0.28,
      "objective/train/advantage_avg": 0.0009234403842128813,
      "objective/train/docs_used": 340768,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5103869438171387,
      "objective/train/original_loss": 1.5103867053985596,
      "objective/train/theoretical_loss": 3.705391869319403,
      "objective/train/tokens_used": 944517600,
      "objective/train/value_avg": -0.00710296630859375,
      "objective/train/value_loss": 0.00018672032456379384,
      "objective/train/value_max": -0.00013136863708496094,
      "objective/train/value_min": -0.277099609375,
      "objective/train/value_reward_corr": 0.6575917614537756,
      "objective/train/value_std": 0.01128387451171875,
      "objective/train/weight_avg": 1.001011610031128,
      "objective/train/weighted_lm_loss": 1.5110077857971191,
      "objective/train/weights_max": 1.3192977905273438,
      "objective/train/weights_min": 0.3727017939090729,
      "theoretical_loss": 3.705391869319403,
      "tokens_seen": 924057600
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007272508425613866,
      "loss": 0.0732,
      "theoretical_loss": 3.705391869319403,
      "tokens_seen": 924057600
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007271705986198042,
      "loss": 0.0717,
      "theoretical_loss": 3.7052911647817357,
      "tokens_seen": 924319744
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007270903546782218,
      "loss": 0.0752,
      "theoretical_loss": 3.7051904967949527,
      "tokens_seen": 924581888
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007270101107366394,
      "loss": 0.0711,
      "theoretical_loss": 3.7050898653354296,
      "tokens_seen": 924844032
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000726929866795057,
      "loss": 0.0745,
      "theoretical_loss": 3.7049892703795653,
      "tokens_seen": 925106176
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007268496228534745,
      "loss": 0.0739,
      "theoretical_loss": 3.70488871190378,
      "tokens_seen": 925368320
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007267693789118923,
      "loss": 0.0737,
      "theoretical_loss": 3.704788189884515,
      "tokens_seen": 925630464
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007266891349703098,
      "loss": 0.0753,
      "theoretical_loss": 3.7046877042982347,
      "tokens_seen": 925892608
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007266088910287273,
      "loss": 0.0757,
      "theoretical_loss": 3.7045872551214254,
      "tokens_seen": 926154752
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000726528647087145,
      "loss": 0.0719,
      "theoretical_loss": 3.704486842330594,
      "tokens_seen": 926416896
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007264484031455625,
      "loss": 0.074,
      "theoretical_loss": 3.7043864659022696,
      "tokens_seen": 926679040
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007263681592039801,
      "loss": 0.0714,
      "theoretical_loss": 3.704286125813004,
      "tokens_seen": 926941184
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007262879152623977,
      "loss": 0.074,
      "theoretical_loss": 3.7041858220393706,
      "tokens_seen": 927203328
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007262076713208153,
      "loss": 0.0731,
      "theoretical_loss": 3.704085554557964,
      "tokens_seen": 927465472
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007261274273792328,
      "loss": 0.0728,
      "theoretical_loss": 3.703985323345399,
      "tokens_seen": 927727616
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007260471834376505,
      "loss": 0.0705,
      "theoretical_loss": 3.7038851283783156,
      "tokens_seen": 927989760
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000725966939496068,
      "loss": 0.075,
      "theoretical_loss": 3.7037849696333724,
      "tokens_seen": 928251904
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007258866955544856,
      "loss": 0.0765,
      "theoretical_loss": 3.703684847087251,
      "tokens_seen": 928514048
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007258064516129033,
      "loss": 0.0759,
      "theoretical_loss": 3.7035847607166534,
      "tokens_seen": 928776192
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007257262076713208,
      "loss": 0.0734,
      "theoretical_loss": 3.703484710498306,
      "tokens_seen": 929038336
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007256459637297385,
      "loss": 0.0801,
      "theoretical_loss": 3.703384696408953,
      "tokens_seen": 929300480
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000725565719788156,
      "loss": 0.0735,
      "theoretical_loss": 3.7032847184253628,
      "tokens_seen": 929562624
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007254854758465736,
      "loss": 0.0732,
      "theoretical_loss": 3.7031847765243233,
      "tokens_seen": 929824768
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007254052319049912,
      "loss": 0.0713,
      "theoretical_loss": 3.7030848706826465,
      "tokens_seen": 930086912
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007253249879634087,
      "loss": 0.0752,
      "theoretical_loss": 3.702985000877163,
      "tokens_seen": 930349056
    },
    {
      "epoch": 0.28,
      "objective/train/advantage_avg": 0.0017973899375647306,
      "objective/train/docs_used": 343003,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5838013887405396,
      "objective/train/original_loss": 1.5838017463684082,
      "objective/train/theoretical_loss": 3.7028851670847267,
      "objective/train/tokens_used": 951071200,
      "objective/train/value_avg": -0.009552001953125,
      "objective/train/value_loss": 0.00028080277843400836,
      "objective/train/value_max": -0.00020182132720947266,
      "objective/train/value_min": -0.52685546875,
      "objective/train/value_reward_corr": 0.687662785571624,
      "objective/train/value_std": 0.0157623291015625,
      "objective/train/weight_avg": 1.001927137374878,
      "objective/train/weighted_lm_loss": 1.586591362953186,
      "objective/train/weights_max": 1.4974627494812012,
      "objective/train/weights_min": 0.3688567578792572,
      "theoretical_loss": 3.7028851670847267,
      "tokens_seen": 930611200
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007252447440218264,
      "loss": 0.0745,
      "theoretical_loss": 3.7028851670847267,
      "tokens_seen": 930611200
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000725164500080244,
      "loss": 0.0729,
      "theoretical_loss": 3.7027853692822124,
      "tokens_seen": 930873344
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007250842561386616,
      "loss": 0.076,
      "theoretical_loss": 3.702685607446516,
      "tokens_seen": 931135488
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007250040121970791,
      "loss": 0.076,
      "theoretical_loss": 3.7025858815545543,
      "tokens_seen": 931397632
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007249237682554968,
      "loss": 0.0746,
      "theoretical_loss": 3.7024861915832665,
      "tokens_seen": 931659776
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007248435243139143,
      "loss": 0.0739,
      "theoretical_loss": 3.7023865375096126,
      "tokens_seen": 931921920
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007247632803723318,
      "loss": 0.0757,
      "theoretical_loss": 3.7022869193105734,
      "tokens_seen": 932184064
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007246830364307495,
      "loss": 0.0728,
      "theoretical_loss": 3.702187336963151,
      "tokens_seen": 932446208
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000724602792489167,
      "loss": 0.0741,
      "theoretical_loss": 3.70208779044437,
      "tokens_seen": 932708352
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007245225485475848,
      "loss": 0.0719,
      "theoretical_loss": 3.7019882797312746,
      "tokens_seen": 932970496
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007244423046060023,
      "loss": 0.0794,
      "theoretical_loss": 3.701888804800931,
      "tokens_seen": 933232640
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007243620606644199,
      "loss": 0.071,
      "theoretical_loss": 3.701789365630426,
      "tokens_seen": 933494784
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007242818167228375,
      "loss": 0.0728,
      "theoretical_loss": 3.701689962196868,
      "tokens_seen": 933756928
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000724201572781255,
      "loss": 0.0769,
      "theoretical_loss": 3.701590594477387,
      "tokens_seen": 934019072
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007241213288396726,
      "loss": 0.0766,
      "theoretical_loss": 3.701491262449131,
      "tokens_seen": 934281216
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007240410848980902,
      "loss": 0.0764,
      "theoretical_loss": 3.7013919660892736,
      "tokens_seen": 934543360
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007239608409565078,
      "loss": 0.0743,
      "theoretical_loss": 3.701292705375006,
      "tokens_seen": 934805504
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007238805970149253,
      "loss": 0.0755,
      "theoretical_loss": 3.701193480283542,
      "tokens_seen": 935067648
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000723800353073343,
      "loss": 0.0718,
      "theoretical_loss": 3.701094290792116,
      "tokens_seen": 935329792
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007237201091317606,
      "loss": 0.0755,
      "theoretical_loss": 3.7009951368779825,
      "tokens_seen": 935591936
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007236398651901781,
      "loss": 0.0729,
      "theoretical_loss": 3.700896018518418,
      "tokens_seen": 935854080
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007235596212485958,
      "loss": 0.0756,
      "theoretical_loss": 3.700796935690719,
      "tokens_seen": 936116224
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007234793773070133,
      "loss": 0.0736,
      "theoretical_loss": 3.700697888372204,
      "tokens_seen": 936378368
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007233991333654309,
      "loss": 0.0746,
      "theoretical_loss": 3.7005988765402114,
      "tokens_seen": 936640512
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007233188894238485,
      "loss": 0.0735,
      "theoretical_loss": 3.700499900172101,
      "tokens_seen": 936902656
    },
    {
      "epoch": 0.28,
      "objective/train/advantage_avg": 0.0006104443455114961,
      "objective/train/docs_used": 344763,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4431649446487427,
      "objective/train/original_loss": 1.4431649446487427,
      "objective/train/theoretical_loss": 3.700400959245252,
      "objective/train/tokens_used": 957624800,
      "objective/train/value_avg": -0.01019287109375,
      "objective/train/value_loss": 0.00036883429856970906,
      "objective/train/value_max": -0.0001252889633178711,
      "objective/train/value_min": -0.8720703125,
      "objective/train/value_reward_corr": 0.8053598602672474,
      "objective/train/value_std": 0.0275726318359375,
      "objective/train/weight_avg": 1.000779628753662,
      "objective/train/weighted_lm_loss": 1.4440263509750366,
      "objective/train/weights_max": 1.7109980583190918,
      "objective/train/weights_min": 0.3686527907848358,
      "theoretical_loss": 3.700400959245252,
      "tokens_seen": 937164800
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007232386454822661,
      "loss": 0.0714,
      "theoretical_loss": 3.700400959245252,
      "tokens_seen": 937164800
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007231584015406837,
      "loss": 0.0727,
      "theoretical_loss": 3.7003020537370657,
      "tokens_seen": 937426944
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007230781575991013,
      "loss": 0.0735,
      "theoretical_loss": 3.7002031836249643,
      "tokens_seen": 937689088
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007229979136575189,
      "loss": 0.0755,
      "theoretical_loss": 3.7001043488863896,
      "tokens_seen": 937951232
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007229176697159365,
      "loss": 0.0751,
      "theoretical_loss": 3.7000055494988047,
      "tokens_seen": 938213376
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007228374257743541,
      "loss": 0.0768,
      "theoretical_loss": 3.6999067854396936,
      "tokens_seen": 938475520
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007227571818327716,
      "loss": 0.0713,
      "theoretical_loss": 3.6998080566865608,
      "tokens_seen": 938737664
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007226769378911893,
      "loss": 0.0738,
      "theoretical_loss": 3.6997093632169307,
      "tokens_seen": 938999808
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007225966939496068,
      "loss": 0.0765,
      "theoretical_loss": 3.699610705008349,
      "tokens_seen": 939261952
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007225164500080244,
      "loss": 0.0705,
      "theoretical_loss": 3.6995120820383818,
      "tokens_seen": 939524096
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000722436206066442,
      "loss": 0.0741,
      "theoretical_loss": 3.6994134942846157,
      "tokens_seen": 939786240
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007223559621248595,
      "loss": 0.0758,
      "theoretical_loss": 3.6993149417246576,
      "tokens_seen": 940048384
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0007222757181832772,
      "loss": 0.0735,
      "theoretical_loss": 3.699216424336135,
      "tokens_seen": 940310528
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007221954742416948,
      "loss": 0.0745,
      "theoretical_loss": 3.6991179420966964,
      "tokens_seen": 940572672
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007221152303001124,
      "loss": 0.0728,
      "theoretical_loss": 3.69901949498401,
      "tokens_seen": 940834816
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007220349863585299,
      "loss": 0.0759,
      "theoretical_loss": 3.6989210829757644,
      "tokens_seen": 941096960
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007219547424169476,
      "loss": 0.0756,
      "theoretical_loss": 3.6988227060496692,
      "tokens_seen": 941359104
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007218744984753651,
      "loss": 0.0761,
      "theoretical_loss": 3.6987243641834535,
      "tokens_seen": 941621248
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007217942545337827,
      "loss": 0.0743,
      "theoretical_loss": 3.6986260573548675,
      "tokens_seen": 941883392
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007217140105922003,
      "loss": 0.0746,
      "theoretical_loss": 3.698527785541682,
      "tokens_seen": 942145536
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007216337666506178,
      "loss": 0.0741,
      "theoretical_loss": 3.6984295487216867,
      "tokens_seen": 942407680
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007215535227090356,
      "loss": 0.0745,
      "theoretical_loss": 3.6983313468726924,
      "tokens_seen": 942669824
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007214732787674531,
      "loss": 0.0772,
      "theoretical_loss": 3.6982331799725303,
      "tokens_seen": 942931968
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007213930348258707,
      "loss": 0.0733,
      "theoretical_loss": 3.6981350479990525,
      "tokens_seen": 943194112
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007213127908842883,
      "loss": 0.078,
      "theoretical_loss": 3.6980369509301285,
      "tokens_seen": 943456256
    },
    {
      "epoch": 0.29,
      "objective/train/advantage_avg": 0.0005385353579185903,
      "objective/train/docs_used": 346959,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3849838972091675,
      "objective/train/original_loss": 1.384983777999878,
      "objective/train/theoretical_loss": 3.6979388887436517,
      "objective/train/tokens_used": 964178400,
      "objective/train/value_avg": -0.008209228515625,
      "objective/train/value_loss": 0.00034320203121751547,
      "objective/train/value_max": -5.8770179748535156e-05,
      "objective/train/value_min": -0.97412109375,
      "objective/train/value_reward_corr": 0.7214583684226242,
      "objective/train/value_std": 0.0195465087890625,
      "objective/train/weight_avg": 1.0007014274597168,
      "objective/train/weighted_lm_loss": 1.3859144449234009,
      "objective/train/weights_max": 2.094869613647461,
      "objective/train/weights_min": 0.3750951290130615,
      "theoretical_loss": 3.6979388887436517,
      "tokens_seen": 943718400
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007212325469427058,
      "loss": 0.0757,
      "theoretical_loss": 3.6979388887436517,
      "tokens_seen": 943718400
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007211523030011234,
      "loss": 0.0757,
      "theoretical_loss": 3.697840861417533,
      "tokens_seen": 943980544
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000721072059059541,
      "loss": 0.0754,
      "theoretical_loss": 3.697742868929704,
      "tokens_seen": 944242688
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007209918151179586,
      "loss": 0.0744,
      "theoretical_loss": 3.6976449112581173,
      "tokens_seen": 944504832
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007209115711763761,
      "loss": 0.0744,
      "theoretical_loss": 3.697546988380744,
      "tokens_seen": 944766976
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007208313272347939,
      "loss": 0.076,
      "theoretical_loss": 3.697449100275577,
      "tokens_seen": 945029120
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007207510832932114,
      "loss": 0.0726,
      "theoretical_loss": 3.6973512469206278,
      "tokens_seen": 945291264
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000720670839351629,
      "loss": 0.0732,
      "theoretical_loss": 3.6972534282939282,
      "tokens_seen": 945553408
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007205905954100466,
      "loss": 0.0726,
      "theoretical_loss": 3.6971556443735314,
      "tokens_seen": 945815552
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007205103514684641,
      "loss": 0.0709,
      "theoretical_loss": 3.697057895137508,
      "tokens_seen": 946077696
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007204301075268818,
      "loss": 0.077,
      "theoretical_loss": 3.696960180563951,
      "tokens_seen": 946339840
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007203498635852993,
      "loss": 0.0751,
      "theoretical_loss": 3.6968625006309717,
      "tokens_seen": 946601984
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007202696196437169,
      "loss": 0.0745,
      "theoretical_loss": 3.6967648553167014,
      "tokens_seen": 946864128
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007201893757021345,
      "loss": 0.0737,
      "theoretical_loss": 3.696667244599292,
      "tokens_seen": 947126272
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000720109131760552,
      "loss": 0.0758,
      "theoretical_loss": 3.6965696684569154,
      "tokens_seen": 947388416
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007200288878189697,
      "loss": 0.0759,
      "theoretical_loss": 3.6964721268677616,
      "tokens_seen": 947650560
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007199486438773873,
      "loss": 0.0743,
      "theoretical_loss": 3.696374619810043,
      "tokens_seen": 947912704
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007198683999358049,
      "loss": 0.0713,
      "theoretical_loss": 3.6962771472619886,
      "tokens_seen": 948174848
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007197881559942224,
      "loss": 0.0743,
      "theoretical_loss": 3.69617970920185,
      "tokens_seen": 948436992
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007197079120526401,
      "loss": 0.0767,
      "theoretical_loss": 3.6960823056078973,
      "tokens_seen": 948699136
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007196276681110576,
      "loss": 0.0766,
      "theoretical_loss": 3.6959849364584203,
      "tokens_seen": 948961280
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007195474241694752,
      "loss": 0.0752,
      "theoretical_loss": 3.695887601731728,
      "tokens_seen": 949223424
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007194671802278928,
      "loss": 0.0764,
      "theoretical_loss": 3.69579030140615,
      "tokens_seen": 949485568
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007193869362863103,
      "loss": 0.075,
      "theoretical_loss": 3.6956930354600352,
      "tokens_seen": 949747712
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007193066923447281,
      "loss": 0.0748,
      "theoretical_loss": 3.6955958038717522,
      "tokens_seen": 950009856
    },
    {
      "epoch": 0.29,
      "objective/train/advantage_avg": 0.000604399829171598,
      "objective/train/docs_used": 349336,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.432031512260437,
      "objective/train/original_loss": 1.432031512260437,
      "objective/train/theoretical_loss": 3.695498606619688,
      "objective/train/tokens_used": 970732000,
      "objective/train/value_avg": -0.007843017578125,
      "objective/train/value_loss": 0.00045492048957385123,
      "objective/train/value_max": -0.00015842914581298828,
      "objective/train/value_min": -0.95703125,
      "objective/train/value_reward_corr": 0.5685032250536571,
      "objective/train/value_std": 0.01549530029296875,
      "objective/train/weight_avg": 1.0007842779159546,
      "objective/train/weighted_lm_loss": 1.4333014488220215,
      "objective/train/weights_max": 2.597604990005493,
      "objective/train/weights_min": 0.05960186570882797,
      "theoretical_loss": 3.695498606619688,
      "tokens_seen": 950272000
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007192264484031456,
      "loss": 0.0746,
      "theoretical_loss": 3.695498606619688,
      "tokens_seen": 950272000
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007191462044615632,
      "loss": 0.0727,
      "theoretical_loss": 3.6954014436822513,
      "tokens_seen": 950534144
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007190659605199808,
      "loss": 0.0749,
      "theoretical_loss": 3.695304315037868,
      "tokens_seen": 950796288
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007189857165783984,
      "loss": 0.0744,
      "theoretical_loss": 3.6952072206649857,
      "tokens_seen": 951058432
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007189054726368159,
      "loss": 0.075,
      "theoretical_loss": 3.695110160542069,
      "tokens_seen": 951320576
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007188252286952335,
      "loss": 0.0741,
      "theoretical_loss": 3.6950131346476054,
      "tokens_seen": 951582720
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007187449847536511,
      "loss": 0.0753,
      "theoretical_loss": 3.694916142960098,
      "tokens_seen": 951844864
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007186647408120686,
      "loss": 0.0723,
      "theoretical_loss": 3.6948191854580728,
      "tokens_seen": 952107008
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007185844968704864,
      "loss": 0.0736,
      "theoretical_loss": 3.694722262120072,
      "tokens_seen": 952369152
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007185042529289039,
      "loss": 0.0751,
      "theoretical_loss": 3.6946253729246594,
      "tokens_seen": 952631296
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007184240089873215,
      "loss": 0.0732,
      "theoretical_loss": 3.6945285178504172,
      "tokens_seen": 952893440
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007183437650457391,
      "loss": 0.0727,
      "theoretical_loss": 3.694431696875948,
      "tokens_seen": 953155584
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007182635211041566,
      "loss": 0.0777,
      "theoretical_loss": 3.6943349099798715,
      "tokens_seen": 953417728
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007181832771625743,
      "loss": 0.0706,
      "theoretical_loss": 3.6942381571408287,
      "tokens_seen": 953679872
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007181030332209918,
      "loss": 0.0759,
      "theoretical_loss": 3.6941414383374793,
      "tokens_seen": 953942016
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007180227892794094,
      "loss": 0.0725,
      "theoretical_loss": 3.6940447535485026,
      "tokens_seen": 954204160
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000717942545337827,
      "loss": 0.0734,
      "theoretical_loss": 3.6939481027525956,
      "tokens_seen": 954466304
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007178623013962447,
      "loss": 0.0737,
      "theoretical_loss": 3.6938514859284766,
      "tokens_seen": 954728448
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007177820574546622,
      "loss": 0.0737,
      "theoretical_loss": 3.6937549030548813,
      "tokens_seen": 954990592
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007177018135130798,
      "loss": 0.0748,
      "theoretical_loss": 3.693658354110565,
      "tokens_seen": 955252736
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007176215695714974,
      "loss": 0.0738,
      "theoretical_loss": 3.6935618390743032,
      "tokens_seen": 955514880
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007175413256299149,
      "loss": 0.0742,
      "theoretical_loss": 3.6934653579248886,
      "tokens_seen": 955777024
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007174610816883326,
      "loss": 0.0727,
      "theoretical_loss": 3.693368910641135,
      "tokens_seen": 956039168
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007173808377467501,
      "loss": 0.076,
      "theoretical_loss": 3.693272497201874,
      "tokens_seen": 956301312
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007173005938051677,
      "loss": 0.0776,
      "theoretical_loss": 3.6931761175859554,
      "tokens_seen": 956563456
    },
    {
      "epoch": 0.29,
      "objective/train/advantage_avg": 0.0019286867463961244,
      "objective/train/docs_used": 351696,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4690124988555908,
      "objective/train/original_loss": 1.4690124988555908,
      "objective/train/theoretical_loss": 3.693079771772251,
      "objective/train/tokens_used": 977285600,
      "objective/train/value_avg": -0.008026123046875,
      "objective/train/value_loss": 0.0001780358434189111,
      "objective/train/value_max": -0.00020503997802734375,
      "objective/train/value_min": -0.70703125,
      "objective/train/value_reward_corr": 0.6057208029602728,
      "objective/train/value_std": 0.01194000244140625,
      "objective/train/weight_avg": 1.0020111799240112,
      "objective/train/weighted_lm_loss": 1.4720525741577148,
      "objective/train/weights_max": 1.3528940677642822,
      "objective/train/weights_min": 0.36884480714797974,
      "theoretical_loss": 3.693079771772251,
      "tokens_seen": 956825600
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007172203498635853,
      "loss": 0.0746,
      "theoretical_loss": 3.693079771772251,
      "tokens_seen": 956825600
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007171401059220028,
      "loss": 0.0763,
      "theoretical_loss": 3.692983459739649,
      "tokens_seen": 957087744
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007170598619804205,
      "loss": 0.0776,
      "theoretical_loss": 3.6928871814670563,
      "tokens_seen": 957349888
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007169796180388381,
      "loss": 0.0762,
      "theoretical_loss": 3.6927909369334007,
      "tokens_seen": 957612032
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007168993740972557,
      "loss": 0.0724,
      "theoretical_loss": 3.6926947261176277,
      "tokens_seen": 957874176
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007168191301556733,
      "loss": 0.076,
      "theoretical_loss": 3.692598548998702,
      "tokens_seen": 958136320
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007167388862140909,
      "loss": 0.077,
      "theoretical_loss": 3.692502405555606,
      "tokens_seen": 958398464
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007166586422725084,
      "loss": 0.0769,
      "theoretical_loss": 3.692406295767344,
      "tokens_seen": 958660608
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007165783983309261,
      "loss": 0.0752,
      "theoretical_loss": 3.692310219612936,
      "tokens_seen": 958922752
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007164981543893436,
      "loss": 0.0741,
      "theoretical_loss": 3.6922141770714214,
      "tokens_seen": 959184896
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007164179104477611,
      "loss": 0.0736,
      "theoretical_loss": 3.6921181681218602,
      "tokens_seen": 959447040
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007163376665061789,
      "loss": 0.0753,
      "theoretical_loss": 3.6920221927433294,
      "tokens_seen": 959709184
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007162574225645964,
      "loss": 0.0771,
      "theoretical_loss": 3.691926250914925,
      "tokens_seen": 959971328
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000716177178623014,
      "loss": 0.0751,
      "theoretical_loss": 3.691830342615763,
      "tokens_seen": 960233472
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007160969346814316,
      "loss": 0.0743,
      "theoretical_loss": 3.6917344678249755,
      "tokens_seen": 960495616
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007160166907398492,
      "loss": 0.0733,
      "theoretical_loss": 3.6916386265217156,
      "tokens_seen": 960757760
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007159364467982667,
      "loss": 0.0727,
      "theoretical_loss": 3.6915428186851553,
      "tokens_seen": 961019904
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007158562028566843,
      "loss": 0.0751,
      "theoretical_loss": 3.6914470442944824,
      "tokens_seen": 961282048
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007157759589151019,
      "loss": 0.0722,
      "theoretical_loss": 3.691351303328907,
      "tokens_seen": 961544192
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007156957149735194,
      "loss": 0.0753,
      "theoretical_loss": 3.691255595767654,
      "tokens_seen": 961806336
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007156154710319372,
      "loss": 0.0765,
      "theoretical_loss": 3.6911599215899704,
      "tokens_seen": 962068480
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007155352270903547,
      "loss": 0.0757,
      "theoretical_loss": 3.6910642807751195,
      "tokens_seen": 962330624
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007154549831487724,
      "loss": 0.075,
      "theoretical_loss": 3.6909686733023843,
      "tokens_seen": 962592768
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007153747392071899,
      "loss": 0.0721,
      "theoretical_loss": 3.690873099151065,
      "tokens_seen": 962854912
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007152944952656074,
      "loss": 0.0729,
      "theoretical_loss": 3.690777558300482,
      "tokens_seen": 963117056
    },
    {
      "epoch": 0.29,
      "objective/train/advantage_avg": 0.0009449372300878167,
      "objective/train/docs_used": 353969,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4017493724822998,
      "objective/train/original_loss": 1.4017493724822998,
      "objective/train/theoretical_loss": 3.690682050729972,
      "objective/train/tokens_used": 983839200,
      "objective/train/value_avg": -0.00891876220703125,
      "objective/train/value_loss": 0.00023926046560518444,
      "objective/train/value_max": -0.0001926422119140625,
      "objective/train/value_min": -0.299072265625,
      "objective/train/value_reward_corr": 0.619741688664302,
      "objective/train/value_std": 0.0124359130859375,
      "objective/train/weight_avg": 1.0010510683059692,
      "objective/train/weighted_lm_loss": 1.403052568435669,
      "objective/train/weights_max": 1.2246274948120117,
      "objective/train/weights_min": 0.36947569251060486,
      "theoretical_loss": 3.690682050729972,
      "tokens_seen": 963379200
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007152142513240251,
      "loss": 0.0734,
      "theoretical_loss": 3.690682050729972,
      "tokens_seen": 963379200
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007151340073824426,
      "loss": 0.0755,
      "theoretical_loss": 3.6905865764188923,
      "tokens_seen": 963641344
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007150537634408602,
      "loss": 0.0721,
      "theoretical_loss": 3.6904911353466177,
      "tokens_seen": 963903488
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007149735194992778,
      "loss": 0.0754,
      "theoretical_loss": 3.690395727492541,
      "tokens_seen": 964165632
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007148932755576955,
      "loss": 0.0732,
      "theoretical_loss": 3.690300352836074,
      "tokens_seen": 964427776
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000714813031616113,
      "loss": 0.0769,
      "theoretical_loss": 3.690205011356646,
      "tokens_seen": 964689920
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007147327876745306,
      "loss": 0.0753,
      "theoretical_loss": 3.6901097030337056,
      "tokens_seen": 964952064
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007146525437329482,
      "loss": 0.0727,
      "theoretical_loss": 3.6900144278467204,
      "tokens_seen": 965214208
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007145722997913657,
      "loss": 0.0755,
      "theoretical_loss": 3.6899191857751736,
      "tokens_seen": 965476352
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007144920558497834,
      "loss": 0.0722,
      "theoretical_loss": 3.6898239767985688,
      "tokens_seen": 965738496
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007144118119082009,
      "loss": 0.0753,
      "theoretical_loss": 3.689728800896428,
      "tokens_seen": 966000640
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007143315679666186,
      "loss": 0.0735,
      "theoretical_loss": 3.68963365804829,
      "tokens_seen": 966262784
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007142513240250361,
      "loss": 0.076,
      "theoretical_loss": 3.689538548233713,
      "tokens_seen": 966524928
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007141710800834536,
      "loss": 0.0775,
      "theoretical_loss": 3.6894434714322726,
      "tokens_seen": 966787072
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007140908361418714,
      "loss": 0.073,
      "theoretical_loss": 3.689348427623563,
      "tokens_seen": 967049216
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007140105922002889,
      "loss": 0.0725,
      "theoretical_loss": 3.689253416787197,
      "tokens_seen": 967311360
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007139303482587065,
      "loss": 0.0725,
      "theoretical_loss": 3.6891584389028047,
      "tokens_seen": 967573504
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007138501043171241,
      "loss": 0.0746,
      "theoretical_loss": 3.689063493950034,
      "tokens_seen": 967835648
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007137698603755417,
      "loss": 0.0736,
      "theoretical_loss": 3.6889685819085525,
      "tokens_seen": 968097792
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007136896164339592,
      "loss": 0.0744,
      "theoretical_loss": 3.688873702758044,
      "tokens_seen": 968359936
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007136093724923768,
      "loss": 0.073,
      "theoretical_loss": 3.688778856478211,
      "tokens_seen": 968622080
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007135291285507944,
      "loss": 0.0737,
      "theoretical_loss": 3.6886840430487746,
      "tokens_seen": 968884224
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007134488846092119,
      "loss": 0.0726,
      "theoretical_loss": 3.688589262449474,
      "tokens_seen": 969146368
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007133686406676297,
      "loss": 0.0753,
      "theoretical_loss": 3.6884945146600643,
      "tokens_seen": 969408512
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007132883967260472,
      "loss": 0.0739,
      "theoretical_loss": 3.6883997996603215,
      "tokens_seen": 969670656
    },
    {
      "epoch": 0.29,
      "objective/train/advantage_avg": 0.001449979841709137,
      "objective/train/docs_used": 356486,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4616808891296387,
      "objective/train/original_loss": 1.4616808891296387,
      "objective/train/theoretical_loss": 3.688305117430038,
      "objective/train/tokens_used": 990392800,
      "objective/train/value_avg": -0.0089111328125,
      "objective/train/value_loss": 0.00036317447666078806,
      "objective/train/value_max": -0.00010150671005249023,
      "objective/train/value_min": -0.70361328125,
      "objective/train/value_reward_corr": 0.6299372029237911,
      "objective/train/value_std": 0.017303466796875,
      "objective/train/weight_avg": 1.0016149282455444,
      "objective/train/weighted_lm_loss": 1.4640196561813354,
      "objective/train/weights_max": 1.4167561531066895,
      "objective/train/weights_min": 0.3715973496437073,
      "theoretical_loss": 3.688305117430038,
      "tokens_seen": 969932800
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007132081527844648,
      "loss": 0.0745,
      "theoretical_loss": 3.688305117430038,
      "tokens_seen": 969932800
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007131279088428824,
      "loss": 0.0743,
      "theoretical_loss": 3.688210467949023,
      "tokens_seen": 970194944
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007130476649013,
      "loss": 0.0741,
      "theoretical_loss": 3.6881158511971055,
      "tokens_seen": 970457088
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007129674209597176,
      "loss": 0.0745,
      "theoretical_loss": 3.6880212671541326,
      "tokens_seen": 970719232
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007128871770181351,
      "loss": 0.0751,
      "theoretical_loss": 3.687926715799967,
      "tokens_seen": 970981376
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007128069330765527,
      "loss": 0.0746,
      "theoretical_loss": 3.687832197114491,
      "tokens_seen": 971243520
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007127266891349703,
      "loss": 0.0759,
      "theoretical_loss": 3.687737711077605,
      "tokens_seen": 971505664
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000712646445193388,
      "loss": 0.0738,
      "theoretical_loss": 3.687643257669225,
      "tokens_seen": 971767808
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007125662012518055,
      "loss": 0.0748,
      "theoretical_loss": 3.6875488368692877,
      "tokens_seen": 972029952
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007124859573102232,
      "loss": 0.0737,
      "theoretical_loss": 3.687454448657745,
      "tokens_seen": 972292096
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007124057133686407,
      "loss": 0.0739,
      "theoretical_loss": 3.687360093014568,
      "tokens_seen": 972554240
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007123254694270582,
      "loss": 0.0744,
      "theoretical_loss": 3.687265769919745,
      "tokens_seen": 972816384
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007122452254854759,
      "loss": 0.0751,
      "theoretical_loss": 3.6871714793532826,
      "tokens_seen": 973078528
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0007121649815438934,
      "loss": 0.077,
      "theoretical_loss": 3.687077221295203,
      "tokens_seen": 973340672
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000712084737602311,
      "loss": 0.0724,
      "theoretical_loss": 3.6869829957255496,
      "tokens_seen": 973602816
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007120044936607286,
      "loss": 0.0734,
      "theoretical_loss": 3.68688880262438,
      "tokens_seen": 973864960
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007119242497191462,
      "loss": 0.0728,
      "theoretical_loss": 3.6867946419717716,
      "tokens_seen": 974127104
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007118440057775638,
      "loss": 0.0726,
      "theoretical_loss": 3.6867005137478177,
      "tokens_seen": 974389248
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007117637618359814,
      "loss": 0.0769,
      "theoretical_loss": 3.686606417932631,
      "tokens_seen": 974651392
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000711683517894399,
      "loss": 0.0729,
      "theoretical_loss": 3.6865123545063403,
      "tokens_seen": 974913536
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007116032739528166,
      "loss": 0.0769,
      "theoretical_loss": 3.686418323449093,
      "tokens_seen": 975175680
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007115230300112342,
      "loss": 0.0764,
      "theoretical_loss": 3.6863243247410526,
      "tokens_seen": 975437824
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007114427860696517,
      "loss": 0.0734,
      "theoretical_loss": 3.686230358362401,
      "tokens_seen": 975699968
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007113625421280694,
      "loss": 0.0738,
      "theoretical_loss": 3.686136424293338,
      "tokens_seen": 975962112
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007112822981864869,
      "loss": 0.0742,
      "theoretical_loss": 3.68604252251408,
      "tokens_seen": 976224256
    },
    {
      "epoch": 0.3,
      "objective/train/advantage_avg": -0.001066173892468214,
      "objective/train/docs_used": 358882,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4502272605895996,
      "objective/train/original_loss": 1.4502272605895996,
      "objective/train/theoretical_loss": 3.6859486530048615,
      "objective/train/tokens_used": 996946400,
      "objective/train/value_avg": -0.011871337890625,
      "objective/train/value_loss": 0.0007070415304042399,
      "objective/train/value_max": -5.692243576049805e-05,
      "objective/train/value_min": -0.36865234375,
      "objective/train/value_reward_corr": 0.9649324601528265,
      "objective/train/value_std": 0.037567138671875,
      "objective/train/weight_avg": 0.9992637634277344,
      "objective/train/weighted_lm_loss": 1.4516780376434326,
      "objective/train/weights_max": 1.3549121618270874,
      "objective/train/weights_min": 0.36916786432266235,
      "theoretical_loss": 3.6859486530048615,
      "tokens_seen": 976486400
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007112020542449044,
      "loss": 0.0734,
      "theoretical_loss": 3.6859486530048615,
      "tokens_seen": 976486400
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007111218103033222,
      "loss": 0.0734,
      "theoretical_loss": 3.685854815745933,
      "tokens_seen": 976748544
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007110415663617397,
      "loss": 0.0743,
      "theoretical_loss": 3.6857610107175645,
      "tokens_seen": 977010688
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007109613224201573,
      "loss": 0.077,
      "theoretical_loss": 3.6856672379000415,
      "tokens_seen": 977272832
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007108810784785749,
      "loss": 0.0763,
      "theoretical_loss": 3.6855734972736682,
      "tokens_seen": 977534976
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007108008345369925,
      "loss": 0.0722,
      "theoretical_loss": 3.685479788818766,
      "tokens_seen": 977797120
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00071072059059541,
      "loss": 0.0754,
      "theoretical_loss": 3.6853861125156717,
      "tokens_seen": 978059264
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007106403466538276,
      "loss": 0.0744,
      "theoretical_loss": 3.6852924683447412,
      "tokens_seen": 978321408
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007105601027122452,
      "loss": 0.0723,
      "theoretical_loss": 3.6851988562863482,
      "tokens_seen": 978583552
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007104798587706628,
      "loss": 0.0747,
      "theoretical_loss": 3.6851052763208823,
      "tokens_seen": 978845696
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007103996148290805,
      "loss": 0.0719,
      "theoretical_loss": 3.6850117284287505,
      "tokens_seen": 979107840
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000710319370887498,
      "loss": 0.0708,
      "theoretical_loss": 3.6849182125903774,
      "tokens_seen": 979369984
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007102391269459157,
      "loss": 0.0764,
      "theoretical_loss": 3.6848247287862046,
      "tokens_seen": 979632128
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007101588830043332,
      "loss": 0.0758,
      "theoretical_loss": 3.684731276996691,
      "tokens_seen": 979894272
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007100786390627508,
      "loss": 0.0747,
      "theoretical_loss": 3.684637857202312,
      "tokens_seen": 980156416
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007099983951211684,
      "loss": 0.0739,
      "theoretical_loss": 3.684544469383562,
      "tokens_seen": 980418560
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007099181511795859,
      "loss": 0.0724,
      "theoretical_loss": 3.6844511135209497,
      "tokens_seen": 980680704
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007098379072380035,
      "loss": 0.0731,
      "theoretical_loss": 3.684357789595003,
      "tokens_seen": 980942848
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007097576632964211,
      "loss": 0.0755,
      "theoretical_loss": 3.684264497586266,
      "tokens_seen": 981204992
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007096774193548388,
      "loss": 0.0752,
      "theoretical_loss": 3.684171237475301,
      "tokens_seen": 981467136
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007095971754132563,
      "loss": 0.0752,
      "theoretical_loss": 3.6840780092426852,
      "tokens_seen": 981729280
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000709516931471674,
      "loss": 0.0736,
      "theoretical_loss": 3.6839848128690145,
      "tokens_seen": 981991424
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007094366875300915,
      "loss": 0.076,
      "theoretical_loss": 3.683891648334901,
      "tokens_seen": 982253568
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000709356443588509,
      "loss": 0.0751,
      "theoretical_loss": 3.6837985156209743,
      "tokens_seen": 982515712
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007092761996469267,
      "loss": 0.0773,
      "theoretical_loss": 3.683705414707881,
      "tokens_seen": 982777856
    },
    {
      "epoch": 0.3,
      "objective/train/advantage_avg": 0.0013032013084739447,
      "objective/train/docs_used": 361383,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.466055154800415,
      "objective/train/original_loss": 1.466055154800415,
      "objective/train/theoretical_loss": 3.6836123455762837,
      "objective/train/tokens_used": 1003500000,
      "objective/train/value_avg": -0.007678985595703125,
      "objective/train/value_loss": 0.00017884932458400726,
      "objective/train/value_max": -0.00016868114471435547,
      "objective/train/value_min": -0.6962890625,
      "objective/train/value_reward_corr": 0.6831618657773055,
      "objective/train/value_std": 0.0128631591796875,
      "objective/train/weight_avg": 1.0013844966888428,
      "objective/train/weighted_lm_loss": 1.4686609506607056,
      "objective/train/weights_max": 1.261329174041748,
      "objective/train/weights_min": 0.37812936305999756,
      "theoretical_loss": 3.6836123455762837,
      "tokens_seen": 983040000
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007091959557053442,
      "loss": 0.0733,
      "theoretical_loss": 3.6836123455762837,
      "tokens_seen": 983040000
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007091157117637619,
      "loss": 0.0749,
      "theoretical_loss": 3.683519308206863,
      "tokens_seen": 983302144
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007090354678221794,
      "loss": 0.075,
      "theoretical_loss": 3.683426302580316,
      "tokens_seen": 983564288
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000708955223880597,
      "loss": 0.0761,
      "theoretical_loss": 3.683333328677356,
      "tokens_seen": 983826432
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007088749799390147,
      "loss": 0.0706,
      "theoretical_loss": 3.6832403864787144,
      "tokens_seen": 984088576
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007087947359974322,
      "loss": 0.0748,
      "theoretical_loss": 3.683147475965139,
      "tokens_seen": 984350720
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007087144920558498,
      "loss": 0.0758,
      "theoretical_loss": 3.683054597117393,
      "tokens_seen": 984612864
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007086342481142674,
      "loss": 0.0756,
      "theoretical_loss": 3.6829617499162595,
      "tokens_seen": 984875008
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000708554004172685,
      "loss": 0.075,
      "theoretical_loss": 3.6828689343425345,
      "tokens_seen": 985137152
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007084737602311025,
      "loss": 0.074,
      "theoretical_loss": 3.682776150377034,
      "tokens_seen": 985399296
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007083935162895202,
      "loss": 0.0762,
      "theoretical_loss": 3.682683398000589,
      "tokens_seen": 985661440
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007083132723479377,
      "loss": 0.0771,
      "theoretical_loss": 3.6825906771940478,
      "tokens_seen": 985923584
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007082330284063552,
      "loss": 0.0737,
      "theoretical_loss": 3.682497987938275,
      "tokens_seen": 986185728
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000708152784464773,
      "loss": 0.0731,
      "theoretical_loss": 3.682405330214153,
      "tokens_seen": 986447872
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007080725405231905,
      "loss": 0.0716,
      "theoretical_loss": 3.682312704002579,
      "tokens_seen": 986710016
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007079922965816082,
      "loss": 0.0761,
      "theoretical_loss": 3.6822201092844686,
      "tokens_seen": 986972160
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007079120526400257,
      "loss": 0.0734,
      "theoretical_loss": 3.682127546040753,
      "tokens_seen": 987234304
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007078318086984433,
      "loss": 0.075,
      "theoretical_loss": 3.6820350142523806,
      "tokens_seen": 987496448
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007077515647568609,
      "loss": 0.0748,
      "theoretical_loss": 3.6819425139003155,
      "tokens_seen": 987758592
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007076713208152784,
      "loss": 0.0761,
      "theoretical_loss": 3.6818500449655396,
      "tokens_seen": 988020736
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000707591076873696,
      "loss": 0.0722,
      "theoretical_loss": 3.6817576074290503,
      "tokens_seen": 988282880
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007075108329321136,
      "loss": 0.0724,
      "theoretical_loss": 3.681665201271862,
      "tokens_seen": 988545024
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007074305889905313,
      "loss": 0.0772,
      "theoretical_loss": 3.681572826475006,
      "tokens_seen": 988807168
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007073503450489488,
      "loss": 0.0751,
      "theoretical_loss": 3.681480483019529,
      "tokens_seen": 989069312
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007072701011073665,
      "loss": 0.0755,
      "theoretical_loss": 3.6813881708864953,
      "tokens_seen": 989331456
    },
    {
      "epoch": 0.3,
      "objective/train/advantage_avg": 0.0020013097673654556,
      "objective/train/docs_used": 363941,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4069637060165405,
      "objective/train/original_loss": 1.406963586807251,
      "objective/train/theoretical_loss": 3.681295890056985,
      "objective/train/tokens_used": 1010053600,
      "objective/train/value_avg": -0.00994873046875,
      "objective/train/value_loss": 0.00019873856217600405,
      "objective/train/value_max": -0.00013554096221923828,
      "objective/train/value_min": -0.890625,
      "objective/train/value_reward_corr": 0.7167314143110259,
      "objective/train/value_std": 0.01451873779296875,
      "objective/train/weight_avg": 1.0020908117294312,
      "objective/train/weighted_lm_loss": 1.4100981950759888,
      "objective/train/weights_max": 1.2156908512115479,
      "objective/train/weights_min": 0.36818236112594604,
      "theoretical_loss": 3.681295890056985,
      "tokens_seen": 989593600
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000707189857165784,
      "loss": 0.074,
      "theoretical_loss": 3.681295890056985,
      "tokens_seen": 989593600
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007071096132242016,
      "loss": 0.0717,
      "theoretical_loss": 3.681203640512095,
      "tokens_seen": 989855744
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007070293692826192,
      "loss": 0.0757,
      "theoretical_loss": 3.681111422232937,
      "tokens_seen": 990117888
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007069491253410367,
      "loss": 0.0763,
      "theoretical_loss": 3.681019235200643,
      "tokens_seen": 990380032
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007068688813994543,
      "loss": 0.0747,
      "theoretical_loss": 3.680927079396357,
      "tokens_seen": 990642176
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007067886374578719,
      "loss": 0.0734,
      "theoretical_loss": 3.680834954801242,
      "tokens_seen": 990904320
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007067083935162896,
      "loss": 0.0734,
      "theoretical_loss": 3.6807428613964763,
      "tokens_seen": 991166464
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007066281495747072,
      "loss": 0.0777,
      "theoretical_loss": 3.6806507991632555,
      "tokens_seen": 991428608
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007065479056331248,
      "loss": 0.0744,
      "theoretical_loss": 3.68055876808279,
      "tokens_seen": 991690752
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007064676616915423,
      "loss": 0.079,
      "theoretical_loss": 3.680466768136308,
      "tokens_seen": 991952896
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007063874177499599,
      "loss": 0.0758,
      "theoretical_loss": 3.680374799305053,
      "tokens_seen": 992215040
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007063071738083775,
      "loss": 0.0742,
      "theoretical_loss": 3.6802828615702845,
      "tokens_seen": 992477184
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000706226929866795,
      "loss": 0.0764,
      "theoretical_loss": 3.6801909549132796,
      "tokens_seen": 992739328
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007061466859252127,
      "loss": 0.0757,
      "theoretical_loss": 3.6800990793153305,
      "tokens_seen": 993001472
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007060664419836302,
      "loss": 0.0749,
      "theoretical_loss": 3.6800072347577455,
      "tokens_seen": 993263616
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007059861980420478,
      "loss": 0.0743,
      "theoretical_loss": 3.6799154212218506,
      "tokens_seen": 993525760
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007059059541004655,
      "loss": 0.0726,
      "theoretical_loss": 3.679823638688985,
      "tokens_seen": 993787904
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000705825710158883,
      "loss": 0.0741,
      "theoretical_loss": 3.679731887140508,
      "tokens_seen": 994050048
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007057454662173006,
      "loss": 0.071,
      "theoretical_loss": 3.6796401665577916,
      "tokens_seen": 994312192
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007056652222757182,
      "loss": 0.0766,
      "theoretical_loss": 3.679548476922225,
      "tokens_seen": 994574336
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007055849783341358,
      "loss": 0.0749,
      "theoretical_loss": 3.6794568182152143,
      "tokens_seen": 994836480
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007055047343925533,
      "loss": 0.0756,
      "theoretical_loss": 3.6793651904181806,
      "tokens_seen": 995098624
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000705424490450971,
      "loss": 0.075,
      "theoretical_loss": 3.679273593512563,
      "tokens_seen": 995360768
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007053442465093885,
      "loss": 0.0767,
      "theoretical_loss": 3.679182027479812,
      "tokens_seen": 995622912
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007052640025678061,
      "loss": 0.0766,
      "theoretical_loss": 3.6790904923014005,
      "tokens_seen": 995885056
    },
    {
      "epoch": 0.3,
      "objective/train/advantage_avg": 0.001141717773862183,
      "objective/train/docs_used": 366329,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4647668600082397,
      "objective/train/original_loss": 1.4647668600082397,
      "objective/train/theoretical_loss": 3.6789989879588125,
      "objective/train/tokens_used": 1016607200,
      "objective/train/value_avg": -0.00992584228515625,
      "objective/train/value_loss": 0.00031617286731489,
      "objective/train/value_max": -0.00011414289474487305,
      "objective/train/value_min": -0.49951171875,
      "objective/train/value_reward_corr": 0.7429311902922219,
      "objective/train/value_std": 0.0218963623046875,
      "objective/train/weight_avg": 1.0012768507003784,
      "objective/train/weighted_lm_loss": 1.465766429901123,
      "objective/train/weights_max": 1.3314474821090698,
      "objective/train/weights_min": 0.2260430008172989,
      "theoretical_loss": 3.6789989879588125,
      "tokens_seen": 996147200
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007051837586262238,
      "loss": 0.0741,
      "theoretical_loss": 3.6789989879588125,
      "tokens_seen": 996147200
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007051035146846413,
      "loss": 0.0743,
      "theoretical_loss": 3.6789075144335497,
      "tokens_seen": 996409344
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000705023270743059,
      "loss": 0.073,
      "theoretical_loss": 3.6788160717071303,
      "tokens_seen": 996671488
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007049430268014765,
      "loss": 0.0771,
      "theoretical_loss": 3.678724659761087,
      "tokens_seen": 996933632
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007048627828598941,
      "loss": 0.0732,
      "theoretical_loss": 3.6786332785769695,
      "tokens_seen": 997195776
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007047825389183117,
      "loss": 0.0734,
      "theoretical_loss": 3.678541928136344,
      "tokens_seen": 997457920
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007047022949767292,
      "loss": 0.0757,
      "theoretical_loss": 3.6784506084207904,
      "tokens_seen": 997720064
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007046220510351468,
      "loss": 0.0779,
      "theoretical_loss": 3.6783593194119066,
      "tokens_seen": 997982208
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007045418070935644,
      "loss": 0.076,
      "theoretical_loss": 3.6782680610913054,
      "tokens_seen": 998244352
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007044615631519821,
      "loss": 0.079,
      "theoretical_loss": 3.6781768334406157,
      "tokens_seen": 998506496
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007043813192103996,
      "loss": 0.0736,
      "theoretical_loss": 3.678085636441482,
      "tokens_seen": 998768640
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007043010752688173,
      "loss": 0.0782,
      "theoretical_loss": 3.677994470075565,
      "tokens_seen": 999030784
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007042208313272348,
      "loss": 0.0749,
      "theoretical_loss": 3.6779033343245406,
      "tokens_seen": 999292928
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007041405873856524,
      "loss": 0.0755,
      "theoretical_loss": 3.677812229170101,
      "tokens_seen": 999555072
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00070406034344407,
      "loss": 0.0731,
      "theoretical_loss": 3.677721154593953,
      "tokens_seen": 999817216
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007039800995024875,
      "loss": 0.0756,
      "theoretical_loss": 3.6776301105778213,
      "tokens_seen": 1000079360
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007038998555609052,
      "loss": 0.0759,
      "theoretical_loss": 3.6775390971034447,
      "tokens_seen": 1000341504
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007038196116193227,
      "loss": 0.0742,
      "theoretical_loss": 3.6774481141525777,
      "tokens_seen": 1000603648
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007037393676777404,
      "loss": 0.075,
      "theoretical_loss": 3.6773571617069907,
      "tokens_seen": 1000865792
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000703659123736158,
      "loss": 0.0729,
      "theoretical_loss": 3.6772662397484703,
      "tokens_seen": 1001127936
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007035788797945756,
      "loss": 0.0705,
      "theoretical_loss": 3.6771753482588183,
      "tokens_seen": 1001390080
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007034986358529931,
      "loss": 0.0741,
      "theoretical_loss": 3.6770844872198523,
      "tokens_seen": 1001652224
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007034183919114107,
      "loss": 0.0733,
      "theoretical_loss": 3.6769936566134045,
      "tokens_seen": 1001914368
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007033381479698283,
      "loss": 0.0756,
      "theoretical_loss": 3.676902856421324,
      "tokens_seen": 1002176512
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007032579040282458,
      "loss": 0.0758,
      "theoretical_loss": 3.6768120866254757,
      "tokens_seen": 1002438656
    },
    {
      "epoch": 0.3,
      "objective/train/advantage_avg": -1.4584275049855933e-05,
      "objective/train/docs_used": 368716,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5315086841583252,
      "objective/train/original_loss": 1.5315088033676147,
      "objective/train/theoretical_loss": 3.6767213472077387,
      "objective/train/tokens_used": 1023160800,
      "objective/train/value_avg": -0.0098419189453125,
      "objective/train/value_loss": 0.0005082013085484505,
      "objective/train/value_max": -9.244680404663086e-05,
      "objective/train/value_min": -0.75244140625,
      "objective/train/value_reward_corr": 0.6553984384647921,
      "objective/train/value_std": 0.019561767578125,
      "objective/train/weight_avg": 1.0001962184906006,
      "objective/train/weighted_lm_loss": 1.5310980081558228,
      "objective/train/weights_max": 1.727001667022705,
      "objective/train/weights_min": 0.06900446861982346,
      "theoretical_loss": 3.6767213472077387,
      "tokens_seen": 1002700800
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007031776600866635,
      "loss": 0.0743,
      "theoretical_loss": 3.6767213472077387,
      "tokens_seen": 1002700800
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000703097416145081,
      "loss": 0.0729,
      "theoretical_loss": 3.676630638150008,
      "tokens_seen": 1002962944
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007030171722034986,
      "loss": 0.072,
      "theoretical_loss": 3.6765399594341943,
      "tokens_seen": 1003225088
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007029369282619163,
      "loss": 0.0758,
      "theoretical_loss": 3.676449311042225,
      "tokens_seen": 1003487232
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007028566843203338,
      "loss": 0.0788,
      "theoretical_loss": 3.6763586929560415,
      "tokens_seen": 1003749376
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007027764403787515,
      "loss": 0.0757,
      "theoretical_loss": 3.6762681051576003,
      "tokens_seen": 1004011520
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000702696196437169,
      "loss": 0.074,
      "theoretical_loss": 3.6761775476288747,
      "tokens_seen": 1004273664
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007026159524955866,
      "loss": 0.0751,
      "theoretical_loss": 3.6760870203518525,
      "tokens_seen": 1004535808
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007025357085540042,
      "loss": 0.0715,
      "theoretical_loss": 3.6759965233085383,
      "tokens_seen": 1004797952
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007024554646124218,
      "loss": 0.0748,
      "theoretical_loss": 3.6759060564809496,
      "tokens_seen": 1005060096
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007023752206708393,
      "loss": 0.0749,
      "theoretical_loss": 3.6758156198511216,
      "tokens_seen": 1005322240
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007022949767292569,
      "loss": 0.0759,
      "theoretical_loss": 3.675725213401104,
      "tokens_seen": 1005584384
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007022147327876746,
      "loss": 0.0758,
      "theoretical_loss": 3.6756348371129617,
      "tokens_seen": 1005846528
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007021344888460921,
      "loss": 0.0775,
      "theoretical_loss": 3.6755444909687744,
      "tokens_seen": 1006108672
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0007020542449045098,
      "loss": 0.0745,
      "theoretical_loss": 3.675454174950639,
      "tokens_seen": 1006370816
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0007019740009629273,
      "loss": 0.0733,
      "theoretical_loss": 3.675363889040666,
      "tokens_seen": 1006632960
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0007018937570213449,
      "loss": 0.075,
      "theoretical_loss": 3.675273633220981,
      "tokens_seen": 1006895104
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0007018135130797625,
      "loss": 0.0745,
      "theoretical_loss": 3.6751834074737264,
      "tokens_seen": 1007157248
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00070173326913818,
      "loss": 0.0761,
      "theoretical_loss": 3.675093211781059,
      "tokens_seen": 1007419392
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0007016530251965977,
      "loss": 0.0773,
      "theoretical_loss": 3.67500304612515,
      "tokens_seen": 1007681536
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0007015727812550152,
      "loss": 0.0757,
      "theoretical_loss": 3.674912910488187,
      "tokens_seen": 1007943680
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0007014925373134329,
      "loss": 0.076,
      "theoretical_loss": 3.6748228048523726,
      "tokens_seen": 1008205824
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0007014122933718505,
      "loss": 0.0784,
      "theoretical_loss": 3.674732729199924,
      "tokens_seen": 1008467968
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0007013320494302681,
      "loss": 0.0741,
      "theoretical_loss": 3.674642683513074,
      "tokens_seen": 1008730112
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0007012518054886856,
      "loss": 0.0764,
      "theoretical_loss": 3.674552667774071,
      "tokens_seen": 1008992256
    },
    {
      "epoch": 0.31,
      "objective/train/advantage_avg": 0.0006658255588263273,
      "objective/train/docs_used": 370943,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.50751793384552,
      "objective/train/original_loss": 1.50751793384552,
      "objective/train/theoretical_loss": 3.6744626819651773,
      "objective/train/tokens_used": 1029714400,
      "objective/train/value_avg": -0.007167816162109375,
      "objective/train/value_loss": 0.00024274643510580063,
      "objective/train/value_max": -0.000110626220703125,
      "objective/train/value_min": -0.354736328125,
      "objective/train/value_reward_corr": 0.6235155171541196,
      "objective/train/value_std": 0.01125335693359375,
      "objective/train/weight_avg": 1.0007777214050293,
      "objective/train/weighted_lm_loss": 1.5075055360794067,
      "objective/train/weights_max": 1.4258047342300415,
      "objective/train/weights_min": 0.39044809341430664,
      "theoretical_loss": 3.6744626819651773,
      "tokens_seen": 1009254400
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0007011715615471032,
      "loss": 0.073,
      "theoretical_loss": 3.6744626819651773,
      "tokens_seen": 1009254400
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0007010913176055208,
      "loss": 0.0745,
      "theoretical_loss": 3.674372726068671,
      "tokens_seen": 1009516544
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0007010110736639383,
      "loss": 0.0737,
      "theoretical_loss": 3.6742828000668464,
      "tokens_seen": 1009778688
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.000700930829722356,
      "loss": 0.0745,
      "theoretical_loss": 3.6741929039420103,
      "tokens_seen": 1010040832
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0007008505857807735,
      "loss": 0.0739,
      "theoretical_loss": 3.6741030376764865,
      "tokens_seen": 1010302976
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0007007703418391912,
      "loss": 0.0764,
      "theoretical_loss": 3.674013201252614,
      "tokens_seen": 1010565120
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0007006900978976088,
      "loss": 0.0749,
      "theoretical_loss": 3.6739233946527454,
      "tokens_seen": 1010827264
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0007006098539560264,
      "loss": 0.0727,
      "theoretical_loss": 3.6738336178592492,
      "tokens_seen": 1011089408
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0007005296100144439,
      "loss": 0.0792,
      "theoretical_loss": 3.6737438708545094,
      "tokens_seen": 1011351552
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0007004493660728615,
      "loss": 0.0749,
      "theoretical_loss": 3.673654153620924,
      "tokens_seen": 1011613696
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0007003691221312791,
      "loss": 0.0782,
      "theoretical_loss": 3.673564466140906,
      "tokens_seen": 1011875840
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0007002888781896967,
      "loss": 0.0746,
      "theoretical_loss": 3.6734748083968842,
      "tokens_seen": 1012137984
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0007002086342481143,
      "loss": 0.076,
      "theoretical_loss": 3.6733851803713016,
      "tokens_seen": 1012400128
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0007001283903065318,
      "loss": 0.0759,
      "theoretical_loss": 3.673295582046616,
      "tokens_seen": 1012662272
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0007000481463649496,
      "loss": 0.0755,
      "theoretical_loss": 3.6732060134053013,
      "tokens_seen": 1012924416
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006999679024233671,
      "loss": 0.076,
      "theoretical_loss": 3.673116474429844,
      "tokens_seen": 1013186560
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006998876584817846,
      "loss": 0.0775,
      "theoretical_loss": 3.673026965102748,
      "tokens_seen": 1013448704
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006998074145402023,
      "loss": 0.0738,
      "theoretical_loss": 3.67293748540653,
      "tokens_seen": 1013710848
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006997271705986198,
      "loss": 0.0737,
      "theoretical_loss": 3.672848035323723,
      "tokens_seen": 1013972992
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006996469266570374,
      "loss": 0.0745,
      "theoretical_loss": 3.6727586148368743,
      "tokens_seen": 1014235136
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.000699566682715455,
      "loss": 0.0753,
      "theoretical_loss": 3.672669223928545,
      "tokens_seen": 1014497280
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006994864387738726,
      "loss": 0.0744,
      "theoretical_loss": 3.672579862581313,
      "tokens_seen": 1014759424
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006994061948322901,
      "loss": 0.0746,
      "theoretical_loss": 3.672490530777769,
      "tokens_seen": 1015021568
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006993259508907077,
      "loss": 0.0751,
      "theoretical_loss": 3.6724012285005196,
      "tokens_seen": 1015283712
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006992457069491254,
      "loss": 0.0763,
      "theoretical_loss": 3.6723119557321864,
      "tokens_seen": 1015545856
    },
    {
      "epoch": 0.31,
      "objective/train/advantage_avg": 0.0006244329269975424,
      "objective/train/docs_used": 372984,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4833120107650757,
      "objective/train/original_loss": 1.4833121299743652,
      "objective/train/theoretical_loss": 3.6722227124554045,
      "objective/train/tokens_used": 1036268000,
      "objective/train/value_avg": -0.007251739501953125,
      "objective/train/value_loss": 0.00037429071380756795,
      "objective/train/value_max": -8.094310760498047e-05,
      "objective/train/value_min": -0.8642578125,
      "objective/train/value_reward_corr": 0.6612852682359046,
      "objective/train/value_std": 0.01514434814453125,
      "objective/train/weight_avg": 1.0007858276367188,
      "objective/train/weighted_lm_loss": 1.484993577003479,
      "objective/train/weights_max": 1.7426632642745972,
      "objective/train/weights_min": 0.3694700598716736,
      "theoretical_loss": 3.6722227124554045,
      "tokens_seen": 1015808000
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006991654630075429,
      "loss": 0.0756,
      "theoretical_loss": 3.6722227124554045,
      "tokens_seen": 1015808000
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006990852190659606,
      "loss": 0.0727,
      "theoretical_loss": 3.6721334986528236,
      "tokens_seen": 1016070144
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006990049751243781,
      "loss": 0.0766,
      "theoretical_loss": 3.6720443143071106,
      "tokens_seen": 1016332288
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006989247311827958,
      "loss": 0.0743,
      "theoretical_loss": 3.671955159400943,
      "tokens_seen": 1016594432
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006988444872412133,
      "loss": 0.0763,
      "theoretical_loss": 3.6718660339170173,
      "tokens_seen": 1016856576
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006987642432996308,
      "loss": 0.0758,
      "theoretical_loss": 3.6717769378380414,
      "tokens_seen": 1017118720
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006986839993580485,
      "loss": 0.0748,
      "theoretical_loss": 3.671687871146739,
      "tokens_seen": 1017380864
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.000698603755416466,
      "loss": 0.0731,
      "theoretical_loss": 3.6715988338258487,
      "tokens_seen": 1017643008
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006985235114748837,
      "loss": 0.0745,
      "theoretical_loss": 3.6715098258581236,
      "tokens_seen": 1017905152
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006984432675333013,
      "loss": 0.0761,
      "theoretical_loss": 3.6714208472263303,
      "tokens_seen": 1018167296
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006983630235917189,
      "loss": 0.0738,
      "theoretical_loss": 3.6713318979132517,
      "tokens_seen": 1018429440
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006982827796501364,
      "loss": 0.0752,
      "theoretical_loss": 3.671242977901683,
      "tokens_seen": 1018691584
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.000698202535708554,
      "loss": 0.0762,
      "theoretical_loss": 3.671154087174436,
      "tokens_seen": 1018953728
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006981222917669716,
      "loss": 0.0762,
      "theoretical_loss": 3.6710652257143366,
      "tokens_seen": 1019215872
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006980420478253891,
      "loss": 0.0755,
      "theoretical_loss": 3.6709763935042243,
      "tokens_seen": 1019478016
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006979618038838068,
      "loss": 0.0759,
      "theoretical_loss": 3.670887590526953,
      "tokens_seen": 1019740160
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006978815599422243,
      "loss": 0.0753,
      "theoretical_loss": 3.6707988167653927,
      "tokens_seen": 1020002304
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006978013160006421,
      "loss": 0.0749,
      "theoretical_loss": 3.670710072202426,
      "tokens_seen": 1020264448
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006977210720590596,
      "loss": 0.0762,
      "theoretical_loss": 3.670621356820951,
      "tokens_seen": 1020526592
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006976408281174772,
      "loss": 0.0736,
      "theoretical_loss": 3.67053267060388,
      "tokens_seen": 1020788736
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006975605841758948,
      "loss": 0.0761,
      "theoretical_loss": 3.6704440135341394,
      "tokens_seen": 1021050880
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006974803402343123,
      "loss": 0.0755,
      "theoretical_loss": 3.6703553855946702,
      "tokens_seen": 1021313024
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006974000962927299,
      "loss": 0.0732,
      "theoretical_loss": 3.6702667867684275,
      "tokens_seen": 1021575168
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006973198523511475,
      "loss": 0.0725,
      "theoretical_loss": 3.670178217038381,
      "tokens_seen": 1021837312
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006972396084095651,
      "loss": 0.0737,
      "theoretical_loss": 3.670089676387515,
      "tokens_seen": 1022099456
    },
    {
      "epoch": 0.31,
      "objective/train/advantage_avg": 0.000926980166696012,
      "objective/train/docs_used": 375283,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5110857486724854,
      "objective/train/original_loss": 1.5110857486724854,
      "objective/train/theoretical_loss": 3.6700011647988275,
      "objective/train/tokens_used": 1042821600,
      "objective/train/value_avg": -0.01114654541015625,
      "objective/train/value_loss": 0.00015908897330518812,
      "objective/train/value_max": -9.608268737792969e-05,
      "objective/train/value_min": -0.3095703125,
      "objective/train/value_reward_corr": 0.9276620899819128,
      "objective/train/value_std": 0.02862548828125,
      "objective/train/weight_avg": 1.0010052919387817,
      "objective/train/weighted_lm_loss": 1.5121495723724365,
      "objective/train/weights_max": 1.142175316810608,
      "objective/train/weights_min": 0.7194867134094238,
      "theoretical_loss": 3.6700011647988275,
      "tokens_seen": 1022361600
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006971593644679826,
      "loss": 0.0772,
      "theoretical_loss": 3.6700011647988275,
      "tokens_seen": 1022361600
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006970791205264004,
      "loss": 0.0725,
      "theoretical_loss": 3.6699126822553314,
      "tokens_seen": 1022623744
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006969988765848179,
      "loss": 0.0735,
      "theoretical_loss": 3.669824228740053,
      "tokens_seen": 1022885888
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006969186326432354,
      "loss": 0.0773,
      "theoretical_loss": 3.6697358042360344,
      "tokens_seen": 1023148032
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006968383887016531,
      "loss": 0.0755,
      "theoretical_loss": 3.6696474087263296,
      "tokens_seen": 1023410176
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006967581447600706,
      "loss": 0.0773,
      "theoretical_loss": 3.6695590421940096,
      "tokens_seen": 1023672320
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006966779008184882,
      "loss": 0.076,
      "theoretical_loss": 3.6694707046221575,
      "tokens_seen": 1023934464
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006965976568769058,
      "loss": 0.074,
      "theoretical_loss": 3.669382395993871,
      "tokens_seen": 1024196608
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006965174129353234,
      "loss": 0.0753,
      "theoretical_loss": 3.669294116292263,
      "tokens_seen": 1024458752
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.000696437168993741,
      "loss": 0.0748,
      "theoretical_loss": 3.6692058655004605,
      "tokens_seen": 1024720896
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006963569250521585,
      "loss": 0.0766,
      "theoretical_loss": 3.669117643601602,
      "tokens_seen": 1024983040
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006962766811105762,
      "loss": 0.0718,
      "theoretical_loss": 3.6690294505788446,
      "tokens_seen": 1025245184
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006961964371689938,
      "loss": 0.0737,
      "theoretical_loss": 3.668941286415355,
      "tokens_seen": 1025507328
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006961161932274114,
      "loss": 0.0729,
      "theoretical_loss": 3.668853151094318,
      "tokens_seen": 1025769472
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006960359492858289,
      "loss": 0.0727,
      "theoretical_loss": 3.6687650445989295,
      "tokens_seen": 1026031616
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006959557053442466,
      "loss": 0.0731,
      "theoretical_loss": 3.6686769669124004,
      "tokens_seen": 1026293760
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006958754614026641,
      "loss": 0.0744,
      "theoretical_loss": 3.6685889180179565,
      "tokens_seen": 1026555904
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006957952174610816,
      "loss": 0.0741,
      "theoretical_loss": 3.668500897898837,
      "tokens_seen": 1026818048
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006957149735194993,
      "loss": 0.0753,
      "theoretical_loss": 3.668412906538295,
      "tokens_seen": 1027080192
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006956347295779168,
      "loss": 0.076,
      "theoretical_loss": 3.6683249439195977,
      "tokens_seen": 1027342336
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006955544856363345,
      "loss": 0.0751,
      "theoretical_loss": 3.668237010026026,
      "tokens_seen": 1027604480
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006954742416947521,
      "loss": 0.0752,
      "theoretical_loss": 3.668149104840876,
      "tokens_seen": 1027866624
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006953939977531697,
      "loss": 0.0747,
      "theoretical_loss": 3.6680612283474567,
      "tokens_seen": 1028128768
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006953137538115872,
      "loss": 0.0742,
      "theoretical_loss": 3.667973380529091,
      "tokens_seen": 1028390912
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006952335098700048,
      "loss": 0.0743,
      "theoretical_loss": 3.6678855613691157,
      "tokens_seen": 1028653056
    },
    {
      "epoch": 0.31,
      "objective/train/advantage_avg": 0.0015299071092158556,
      "objective/train/docs_used": 377682,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4341442584991455,
      "objective/train/original_loss": 1.4341442584991455,
      "objective/train/theoretical_loss": 3.667797770850883,
      "objective/train/tokens_used": 1049375200,
      "objective/train/value_avg": -0.00940704345703125,
      "objective/train/value_loss": 0.00021609137183986604,
      "objective/train/value_max": -0.0001214742660522461,
      "objective/train/value_min": -0.354248046875,
      "objective/train/value_reward_corr": 0.8563500147459093,
      "objective/train/value_std": 0.0274658203125,
      "objective/train/weight_avg": 1.0016289949417114,
      "objective/train/weighted_lm_loss": 1.4355690479278564,
      "objective/train/weights_max": 1.3001129627227783,
      "objective/train/weights_min": 0.36912843585014343,
      "theoretical_loss": 3.667797770850883,
      "tokens_seen": 1028915200
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006951532659284224,
      "loss": 0.0732,
      "theoretical_loss": 3.667797770850883,
      "tokens_seen": 1028915200
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00069507302198684,
      "loss": 0.0749,
      "theoretical_loss": 3.667710008957756,
      "tokens_seen": 1029177344
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006949927780452576,
      "loss": 0.0775,
      "theoretical_loss": 3.667622275673115,
      "tokens_seen": 1029439488
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006949125341036751,
      "loss": 0.0747,
      "theoretical_loss": 3.667534570980353,
      "tokens_seen": 1029701632
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006948322901620929,
      "loss": 0.0761,
      "theoretical_loss": 3.667446894862876,
      "tokens_seen": 1029963776
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006947520462205104,
      "loss": 0.073,
      "theoretical_loss": 3.667359247304104,
      "tokens_seen": 1030225920
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006946718022789279,
      "loss": 0.0739,
      "theoretical_loss": 3.667271628287472,
      "tokens_seen": 1030488064
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006945915583373456,
      "loss": 0.0751,
      "theoretical_loss": 3.6671840377964275,
      "tokens_seen": 1030750208
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006945113143957631,
      "loss": 0.0754,
      "theoretical_loss": 3.667096475814433,
      "tokens_seen": 1031012352
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006944310704541807,
      "loss": 0.0746,
      "theoretical_loss": 3.6670089423249643,
      "tokens_seen": 1031274496
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006943508265125983,
      "loss": 0.0784,
      "theoretical_loss": 3.6669214373115104,
      "tokens_seen": 1031536640
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006942705825710159,
      "loss": 0.0745,
      "theoretical_loss": 3.6668339607575744,
      "tokens_seen": 1031798784
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006941903386294334,
      "loss": 0.0726,
      "theoretical_loss": 3.6667465126466743,
      "tokens_seen": 1032060928
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006941100946878512,
      "loss": 0.0708,
      "theoretical_loss": 3.6666590929623393,
      "tokens_seen": 1032323072
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006940298507462687,
      "loss": 0.0717,
      "theoretical_loss": 3.666571701688115,
      "tokens_seen": 1032585216
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006939496068046863,
      "loss": 0.0767,
      "theoretical_loss": 3.6664843388075594,
      "tokens_seen": 1032847360
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006938693628631039,
      "loss": 0.0705,
      "theoretical_loss": 3.6663970043042435,
      "tokens_seen": 1033109504
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006937891189215214,
      "loss": 0.0717,
      "theoretical_loss": 3.6663096981617533,
      "tokens_seen": 1033371648
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006937088749799391,
      "loss": 0.0716,
      "theoretical_loss": 3.6662224203636886,
      "tokens_seen": 1033633792
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006936286310383566,
      "loss": 0.0744,
      "theoretical_loss": 3.6661351708936616,
      "tokens_seen": 1033895936
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006935483870967742,
      "loss": 0.0693,
      "theoretical_loss": 3.6660479497352982,
      "tokens_seen": 1034158080
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006934681431551918,
      "loss": 0.0753,
      "theoretical_loss": 3.665960756872239,
      "tokens_seen": 1034420224
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006933878992136093,
      "loss": 0.0745,
      "theoretical_loss": 3.6658735922881376,
      "tokens_seen": 1034682368
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.000693307655272027,
      "loss": 0.0744,
      "theoretical_loss": 3.665786455966661,
      "tokens_seen": 1034944512
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006932274113304446,
      "loss": 0.0747,
      "theoretical_loss": 3.6656993478914903,
      "tokens_seen": 1035206656
    },
    {
      "epoch": 0.31,
      "objective/train/advantage_avg": -7.297027332242578e-05,
      "objective/train/docs_used": 379908,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4424421787261963,
      "objective/train/original_loss": 1.4424422979354858,
      "objective/train/theoretical_loss": 3.6656122680463197,
      "objective/train/tokens_used": 1055928800,
      "objective/train/value_avg": -0.00970458984375,
      "objective/train/value_loss": 0.0001646455639274791,
      "objective/train/value_max": -6.711483001708984e-05,
      "objective/train/value_min": -0.310302734375,
      "objective/train/value_reward_corr": 0.9264396814702418,
      "objective/train/value_std": 0.0283050537109375,
      "objective/train/weight_avg": 1.0000029802322388,
      "objective/train/weighted_lm_loss": 1.4421523809432983,
      "objective/train/weights_max": 1.2352874279022217,
      "objective/train/weights_min": 0.3707435429096222,
      "theoretical_loss": 3.6656122680463197,
      "tokens_seen": 1035468800
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006931471673888622,
      "loss": 0.0754,
      "theoretical_loss": 3.6656122680463197,
      "tokens_seen": 1035468800
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006930669234472797,
      "loss": 0.0729,
      "theoretical_loss": 3.6655252164148564,
      "tokens_seen": 1035730944
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006929866795056974,
      "loss": 0.0758,
      "theoretical_loss": 3.6654381929808233,
      "tokens_seen": 1035993088
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006929064355641149,
      "loss": 0.0735,
      "theoretical_loss": 3.6653511977279534,
      "tokens_seen": 1036255232
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006928261916225324,
      "loss": 0.0746,
      "theoretical_loss": 3.6652642306399965,
      "tokens_seen": 1036517376
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006927459476809501,
      "loss": 0.0749,
      "theoretical_loss": 3.6651772917007137,
      "tokens_seen": 1036779520
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006926657037393676,
      "loss": 0.0754,
      "theoretical_loss": 3.6650903808938806,
      "tokens_seen": 1037041664
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006925854597977854,
      "loss": 0.0752,
      "theoretical_loss": 3.6650034982032857,
      "tokens_seen": 1037303808
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006925052158562029,
      "loss": 0.0734,
      "theoretical_loss": 3.664916643612732,
      "tokens_seen": 1037565952
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006924249719146205,
      "loss": 0.0729,
      "theoretical_loss": 3.6648298171060345,
      "tokens_seen": 1037828096
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006923447279730381,
      "loss": 0.0726,
      "theoretical_loss": 3.6647430186670222,
      "tokens_seen": 1038090240
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006922644840314556,
      "loss": 0.0734,
      "theoretical_loss": 3.6646562482795373,
      "tokens_seen": 1038352384
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006921842400898732,
      "loss": 0.0721,
      "theoretical_loss": 3.664569505927436,
      "tokens_seen": 1038614528
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006921039961482908,
      "loss": 0.0736,
      "theoretical_loss": 3.664482791594588,
      "tokens_seen": 1038876672
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006920237522067084,
      "loss": 0.0738,
      "theoretical_loss": 3.664396105264875,
      "tokens_seen": 1039138816
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0006919435082651259,
      "loss": 0.0735,
      "theoretical_loss": 3.6643094469221933,
      "tokens_seen": 1039400960
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006918632643235437,
      "loss": 0.0734,
      "theoretical_loss": 3.664222816550452,
      "tokens_seen": 1039663104
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006917830203819612,
      "loss": 0.0723,
      "theoretical_loss": 3.6641362141335727,
      "tokens_seen": 1039925248
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006917027764403787,
      "loss": 0.0731,
      "theoretical_loss": 3.6640496396554925,
      "tokens_seen": 1040187392
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006916225324987964,
      "loss": 0.0753,
      "theoretical_loss": 3.6639630931001594,
      "tokens_seen": 1040449536
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006915422885572139,
      "loss": 0.0737,
      "theoretical_loss": 3.6638765744515367,
      "tokens_seen": 1040711680
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006914620446156316,
      "loss": 0.076,
      "theoretical_loss": 3.6637900836935993,
      "tokens_seen": 1040973824
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006913818006740491,
      "loss": 0.0736,
      "theoretical_loss": 3.6637036208103364,
      "tokens_seen": 1041235968
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006913015567324667,
      "loss": 0.0711,
      "theoretical_loss": 3.663617185785749,
      "tokens_seen": 1041498112
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006912213127908843,
      "loss": 0.0724,
      "theoretical_loss": 3.6635307786038536,
      "tokens_seen": 1041760256
    },
    {
      "epoch": 0.32,
      "objective/train/advantage_avg": 0.0013375055277720094,
      "objective/train/docs_used": 382133,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4263135194778442,
      "objective/train/original_loss": 1.4263136386871338,
      "objective/train/theoretical_loss": 3.663444399248678,
      "objective/train/tokens_used": 1062482400,
      "objective/train/value_avg": -0.006931304931640625,
      "objective/train/value_loss": 0.00021545964409597218,
      "objective/train/value_max": -0.00010150671005249023,
      "objective/train/value_min": -0.296875,
      "objective/train/value_reward_corr": 0.5908098049024169,
      "objective/train/value_std": 0.01110076904296875,
      "objective/train/weight_avg": 1.0014337301254272,
      "objective/train/weighted_lm_loss": 1.4280571937561035,
      "objective/train/weights_max": 1.1778565645217896,
      "objective/train/weights_min": 0.3711100220680237,
      "theoretical_loss": 3.663444399248678,
      "tokens_seen": 1042022400
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000691141068849302,
      "loss": 0.0719,
      "theoretical_loss": 3.663444399248678,
      "tokens_seen": 1042022400
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006910608249077195,
      "loss": 0.0754,
      "theoretical_loss": 3.6633580477042633,
      "tokens_seen": 1042284544
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006909805809661371,
      "loss": 0.0733,
      "theoretical_loss": 3.663271723954665,
      "tokens_seen": 1042546688
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006909003370245547,
      "loss": 0.0739,
      "theoretical_loss": 3.6631854279839513,
      "tokens_seen": 1042808832
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006908200930829722,
      "loss": 0.0703,
      "theoretical_loss": 3.6630991597762024,
      "tokens_seen": 1043070976
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006907398491413899,
      "loss": 0.0742,
      "theoretical_loss": 3.6630129193155128,
      "tokens_seen": 1043333120
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006906596051998074,
      "loss": 0.0733,
      "theoretical_loss": 3.6629267065859894,
      "tokens_seen": 1043595264
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000690579361258225,
      "loss": 0.0755,
      "theoretical_loss": 3.662840521571753,
      "tokens_seen": 1043857408
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006904991173166426,
      "loss": 0.074,
      "theoretical_loss": 3.662754364256937,
      "tokens_seen": 1044119552
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006904188733750601,
      "loss": 0.0773,
      "theoretical_loss": 3.662668234625688,
      "tokens_seen": 1044381696
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006903386294334778,
      "loss": 0.0719,
      "theoretical_loss": 3.6625821326621653,
      "tokens_seen": 1044643840
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006902583854918954,
      "loss": 0.0761,
      "theoretical_loss": 3.6624960583505404,
      "tokens_seen": 1044905984
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000690178141550313,
      "loss": 0.0717,
      "theoretical_loss": 3.662410011675001,
      "tokens_seen": 1045168128
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006900978976087306,
      "loss": 0.0748,
      "theoretical_loss": 3.6623239926197444,
      "tokens_seen": 1045430272
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006900176536671482,
      "loss": 0.0741,
      "theoretical_loss": 3.6622380011689826,
      "tokens_seen": 1045692416
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006899374097255657,
      "loss": 0.0735,
      "theoretical_loss": 3.66215203730694,
      "tokens_seen": 1045954560
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006898571657839833,
      "loss": 0.0728,
      "theoretical_loss": 3.6620661010178543,
      "tokens_seen": 1046216704
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006897769218424009,
      "loss": 0.0737,
      "theoretical_loss": 3.6619801922859763,
      "tokens_seen": 1046478848
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006896966779008184,
      "loss": 0.0726,
      "theoretical_loss": 3.661894311095568,
      "tokens_seen": 1046740992
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006896164339592362,
      "loss": 0.0751,
      "theoretical_loss": 3.6618084574309075,
      "tokens_seen": 1047003136
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006895361900176537,
      "loss": 0.0726,
      "theoretical_loss": 3.6617226312762834,
      "tokens_seen": 1047265280
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006894559460760713,
      "loss": 0.0724,
      "theoretical_loss": 3.6616368326159976,
      "tokens_seen": 1047527424
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006893757021344889,
      "loss": 0.0747,
      "theoretical_loss": 3.6615510614343654,
      "tokens_seen": 1047789568
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006892954581929064,
      "loss": 0.0713,
      "theoretical_loss": 3.661465317715715,
      "tokens_seen": 1048051712
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000689215214251324,
      "loss": 0.0729,
      "theoretical_loss": 3.6613796014443865,
      "tokens_seen": 1048313856
    },
    {
      "epoch": 0.32,
      "objective/train/advantage_avg": 0.000845851085614413,
      "objective/train/docs_used": 384505,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4578531980514526,
      "objective/train/original_loss": 1.457853078842163,
      "objective/train/theoretical_loss": 3.661293912604734,
      "objective/train/tokens_used": 1069036000,
      "objective/train/value_avg": -0.00762176513671875,
      "objective/train/value_loss": 0.0002575255639385432,
      "objective/train/value_max": -6.812810897827148e-05,
      "objective/train/value_min": -0.26220703125,
      "objective/train/value_reward_corr": 0.5918434184109678,
      "objective/train/value_std": 0.01096343994140625,
      "objective/train/weight_avg": 1.0009592771530151,
      "objective/train/weighted_lm_loss": 1.459302544593811,
      "objective/train/weights_max": 1.1559641361236572,
      "objective/train/weights_min": 0.368507981300354,
      "theoretical_loss": 3.661293912604734,
      "tokens_seen": 1048576000
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006891349703097416,
      "loss": 0.0725,
      "theoretical_loss": 3.661293912604734,
      "tokens_seen": 1048576000
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006890547263681592,
      "loss": 0.0737,
      "theoretical_loss": 3.661208251181124,
      "tokens_seen": 1048838144
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006889744824265767,
      "loss": 0.072,
      "theoretical_loss": 3.6611226171579356,
      "tokens_seen": 1049100288
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006888942384849945,
      "loss": 0.0739,
      "theoretical_loss": 3.6610370105195607,
      "tokens_seen": 1049362432
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000688813994543412,
      "loss": 0.0753,
      "theoretical_loss": 3.660951431250405,
      "tokens_seen": 1049624576
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006887337506018296,
      "loss": 0.071,
      "theoretical_loss": 3.6608658793348847,
      "tokens_seen": 1049886720
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006886535066602472,
      "loss": 0.074,
      "theoretical_loss": 3.6607803547574314,
      "tokens_seen": 1050148864
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006885732627186647,
      "loss": 0.0754,
      "theoretical_loss": 3.660694857502487,
      "tokens_seen": 1050411008
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006884930187770824,
      "loss": 0.0729,
      "theoretical_loss": 3.660609387554509,
      "tokens_seen": 1050673152
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006884127748354999,
      "loss": 0.0761,
      "theoretical_loss": 3.6605239448979647,
      "tokens_seen": 1050935296
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006883325308939175,
      "loss": 0.0751,
      "theoretical_loss": 3.660438529517336,
      "tokens_seen": 1051197440
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006882522869523351,
      "loss": 0.076,
      "theoretical_loss": 3.660353141397116,
      "tokens_seen": 1051459584
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006881720430107528,
      "loss": 0.0765,
      "theoretical_loss": 3.6602677805218127,
      "tokens_seen": 1051721728
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006880917990691703,
      "loss": 0.0754,
      "theoretical_loss": 3.660182446875944,
      "tokens_seen": 1051983872
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006880115551275879,
      "loss": 0.0699,
      "theoretical_loss": 3.6600971404440434,
      "tokens_seen": 1052246016
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006879313111860055,
      "loss": 0.0713,
      "theoretical_loss": 3.660011861210654,
      "tokens_seen": 1052508160
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000687851067244423,
      "loss": 0.0712,
      "theoretical_loss": 3.659926609160334,
      "tokens_seen": 1052770304
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006877708233028407,
      "loss": 0.0701,
      "theoretical_loss": 3.6598413842776534,
      "tokens_seen": 1053032448
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006876905793612582,
      "loss": 0.0729,
      "theoretical_loss": 3.6597561865471935,
      "tokens_seen": 1053294592
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006876103354196759,
      "loss": 0.0774,
      "theoretical_loss": 3.6596710159535504,
      "tokens_seen": 1053556736
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006875300914780934,
      "loss": 0.0753,
      "theoretical_loss": 3.659585872481331,
      "tokens_seen": 1053818880
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006874498475365109,
      "loss": 0.0752,
      "theoretical_loss": 3.659500756115156,
      "tokens_seen": 1054081024
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006873696035949287,
      "loss": 0.0736,
      "theoretical_loss": 3.659415666839658,
      "tokens_seen": 1054343168
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006872893596533462,
      "loss": 0.0724,
      "theoretical_loss": 3.6593306046394813,
      "tokens_seen": 1054605312
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006872091157117638,
      "loss": 0.075,
      "theoretical_loss": 3.6592455694992854,
      "tokens_seen": 1054867456
    },
    {
      "epoch": 0.32,
      "objective/train/advantage_avg": 0.001419214648194611,
      "objective/train/docs_used": 386814,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3335970640182495,
      "objective/train/original_loss": 1.33359694480896,
      "objective/train/theoretical_loss": 3.659160561403739,
      "objective/train/tokens_used": 1075589600,
      "objective/train/value_avg": -0.007778167724609375,
      "objective/train/value_loss": 0.00025554379681125283,
      "objective/train/value_max": -0.00013875961303710938,
      "objective/train/value_min": -0.865234375,
      "objective/train/value_reward_corr": 0.7009334463254417,
      "objective/train/value_std": 0.0123443603515625,
      "objective/train/weight_avg": 1.0015344619750977,
      "objective/train/weighted_lm_loss": 1.3353464603424072,
      "objective/train/weights_max": 1.3698445558547974,
      "objective/train/weights_min": 0.3807870149612427,
      "theoretical_loss": 3.659160561403739,
      "tokens_seen": 1055129600
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006871288717701814,
      "loss": 0.07,
      "theoretical_loss": 3.659160561403739,
      "tokens_seen": 1055129600
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000687048627828599,
      "loss": 0.0718,
      "theoretical_loss": 3.6590755803375252,
      "tokens_seen": 1055391744
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006869683838870165,
      "loss": 0.0735,
      "theoretical_loss": 3.65899062628534,
      "tokens_seen": 1055653888
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006868881399454341,
      "loss": 0.073,
      "theoretical_loss": 3.65890569923189,
      "tokens_seen": 1055916032
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006868078960038517,
      "loss": 0.0743,
      "theoretical_loss": 3.658820799161896,
      "tokens_seen": 1056178176
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006867276520622692,
      "loss": 0.0749,
      "theoretical_loss": 3.65873592606009,
      "tokens_seen": 1056440320
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000686647408120687,
      "loss": 0.0707,
      "theoretical_loss": 3.658651079911218,
      "tokens_seen": 1056702464
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006865671641791045,
      "loss": 0.0735,
      "theoretical_loss": 3.658566260700036,
      "tokens_seen": 1056964608
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006864869202375221,
      "loss": 0.0735,
      "theoretical_loss": 3.658481468411315,
      "tokens_seen": 1057226752
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006864066762959397,
      "loss": 0.0715,
      "theoretical_loss": 3.6583967030298368,
      "tokens_seen": 1057488896
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006863264323543572,
      "loss": 0.0746,
      "theoretical_loss": 3.6583119645403954,
      "tokens_seen": 1057751040
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006862461884127749,
      "loss": 0.0709,
      "theoretical_loss": 3.658227252927799,
      "tokens_seen": 1058013184
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006861659444711924,
      "loss": 0.0704,
      "theoretical_loss": 3.6581425681768653,
      "tokens_seen": 1058275328
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00068608570052961,
      "loss": 0.072,
      "theoretical_loss": 3.6580579102724267,
      "tokens_seen": 1058537472
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006860054565880276,
      "loss": 0.0734,
      "theoretical_loss": 3.657973279199327,
      "tokens_seen": 1058799616
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006859252126464453,
      "loss": 0.0734,
      "theoretical_loss": 3.6578886749424226,
      "tokens_seen": 1059061760
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006858449687048628,
      "loss": 0.077,
      "theoretical_loss": 3.657804097486581,
      "tokens_seen": 1059323904
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006857647247632804,
      "loss": 0.0705,
      "theoretical_loss": 3.657719546816685,
      "tokens_seen": 1059586048
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000685684480821698,
      "loss": 0.0709,
      "theoretical_loss": 3.657635022917626,
      "tokens_seen": 1059848192
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006856042368801155,
      "loss": 0.0715,
      "theoretical_loss": 3.65755052577431,
      "tokens_seen": 1060110336
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006855239929385332,
      "loss": 0.071,
      "theoretical_loss": 3.657466055371654,
      "tokens_seen": 1060372480
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006854437489969507,
      "loss": 0.0721,
      "theoretical_loss": 3.657381611694588,
      "tokens_seen": 1060634624
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006853635050553683,
      "loss": 0.0731,
      "theoretical_loss": 3.6572971947280544,
      "tokens_seen": 1060896768
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006852832611137859,
      "loss": 0.0764,
      "theoretical_loss": 3.6572128044570067,
      "tokens_seen": 1061158912
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006852030171722034,
      "loss": 0.0724,
      "theoretical_loss": 3.657128440866412,
      "tokens_seen": 1061421056
    },
    {
      "epoch": 0.32,
      "objective/train/advantage_avg": 0.0005861930549144745,
      "objective/train/docs_used": 389244,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5149002075195312,
      "objective/train/original_loss": 1.5149000883102417,
      "objective/train/theoretical_loss": 3.6570441039412485,
      "objective/train/tokens_used": 1082143200,
      "objective/train/value_avg": -0.00844573974609375,
      "objective/train/value_loss": 0.00028737890534102917,
      "objective/train/value_max": -9.918212890625e-05,
      "objective/train/value_min": -0.37060546875,
      "objective/train/value_reward_corr": 0.6168363103967418,
      "objective/train/value_std": 0.01305389404296875,
      "objective/train/weight_avg": 1.0007121562957764,
      "objective/train/weighted_lm_loss": 1.5156128406524658,
      "objective/train/weights_max": 1.3453842401504517,
      "objective/train/weights_min": 0.36841732263565063,
      "theoretical_loss": 3.6570441039412485,
      "tokens_seen": 1061683200
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006851227732306211,
      "loss": 0.073,
      "theoretical_loss": 3.6570441039412485,
      "tokens_seen": 1061683200
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006850425292890387,
      "loss": 0.0734,
      "theoretical_loss": 3.6569597936665064,
      "tokens_seen": 1061945344
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006849622853474563,
      "loss": 0.0753,
      "theoretical_loss": 3.6568755100271897,
      "tokens_seen": 1062207488
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006848820414058739,
      "loss": 0.0727,
      "theoretical_loss": 3.656791253008313,
      "tokens_seen": 1062469632
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006848017974642915,
      "loss": 0.0738,
      "theoretical_loss": 3.6567070225949028,
      "tokens_seen": 1062731776
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000684721553522709,
      "loss": 0.0727,
      "theoretical_loss": 3.656622818771999,
      "tokens_seen": 1062993920
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006846413095811267,
      "loss": 0.0711,
      "theoretical_loss": 3.6565386415246524,
      "tokens_seen": 1063256064
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006845610656395442,
      "loss": 0.0719,
      "theoretical_loss": 3.6564544908379273,
      "tokens_seen": 1063518208
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006844808216979617,
      "loss": 0.0723,
      "theoretical_loss": 3.6563703666968985,
      "tokens_seen": 1063780352
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006844005777563795,
      "loss": 0.0708,
      "theoretical_loss": 3.656286269086653,
      "tokens_seen": 1064042496
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000684320333814797,
      "loss": 0.0734,
      "theoretical_loss": 3.6562021979922923,
      "tokens_seen": 1064304640
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006842400898732146,
      "loss": 0.0752,
      "theoretical_loss": 3.6561181533989267,
      "tokens_seen": 1064566784
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006841598459316322,
      "loss": 0.0731,
      "theoretical_loss": 3.6560341352916796,
      "tokens_seen": 1064828928
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006840796019900498,
      "loss": 0.0729,
      "theoretical_loss": 3.655950143655688,
      "tokens_seen": 1065091072
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006839993580484673,
      "loss": 0.0733,
      "theoretical_loss": 3.655866178476098,
      "tokens_seen": 1065353216
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006839191141068849,
      "loss": 0.0729,
      "theoretical_loss": 3.65578223973807,
      "tokens_seen": 1065615360
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006838388701653025,
      "loss": 0.077,
      "theoretical_loss": 3.6556983274267765,
      "tokens_seen": 1065877504
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006837586262237201,
      "loss": 0.0728,
      "theoretical_loss": 3.6556144415273994,
      "tokens_seen": 1066139648
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006836783822821378,
      "loss": 0.0735,
      "theoretical_loss": 3.655530582025136,
      "tokens_seen": 1066401792
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006835981383405553,
      "loss": 0.072,
      "theoretical_loss": 3.6554467489051925,
      "tokens_seen": 1066663936
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000683517894398973,
      "loss": 0.0746,
      "theoretical_loss": 3.6553629421527885,
      "tokens_seen": 1066926080
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006834376504573905,
      "loss": 0.0744,
      "theoretical_loss": 3.655279161753156,
      "tokens_seen": 1067188224
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000683357406515808,
      "loss": 0.0713,
      "theoretical_loss": 3.6551954076915374,
      "tokens_seen": 1067450368
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006832771625742257,
      "loss": 0.0727,
      "theoretical_loss": 3.655111679953188,
      "tokens_seen": 1067712512
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006831969186326432,
      "loss": 0.0737,
      "theoretical_loss": 3.6550279785233757,
      "tokens_seen": 1067974656
    },
    {
      "epoch": 0.32,
      "objective/train/advantage_avg": 0.001482620951719582,
      "objective/train/docs_used": 391774,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.458082914352417,
      "objective/train/original_loss": 1.458082914352417,
      "objective/train/theoretical_loss": 3.654944303387378,
      "objective/train/tokens_used": 1088696800,
      "objective/train/value_avg": -0.00849151611328125,
      "objective/train/value_loss": 0.00018974917475134134,
      "objective/train/value_max": -0.00010472536087036133,
      "objective/train/value_min": -0.64208984375,
      "objective/train/value_reward_corr": 0.7253735317584146,
      "objective/train/value_std": 0.0149993896484375,
      "objective/train/weight_avg": 1.0015714168548584,
      "objective/train/weighted_lm_loss": 1.4602124691009521,
      "objective/train/weights_max": 1.292288899421692,
      "objective/train/weights_min": 0.36969563364982605,
      "theoretical_loss": 3.654944303387378,
      "tokens_seen": 1068236800
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006831166746910608,
      "loss": 0.0736,
      "theoretical_loss": 3.654944303387378,
      "tokens_seen": 1068236800
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006830364307494784,
      "loss": 0.073,
      "theoretical_loss": 3.654860654530486,
      "tokens_seen": 1068498944
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006829561868078961,
      "loss": 0.0741,
      "theoretical_loss": 3.6547770319380026,
      "tokens_seen": 1068761088
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006828759428663136,
      "loss": 0.0712,
      "theoretical_loss": 3.6546934355952425,
      "tokens_seen": 1069023232
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006827956989247312,
      "loss": 0.0743,
      "theoretical_loss": 3.6546098654875303,
      "tokens_seen": 1069285376
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006827154549831488,
      "loss": 0.0743,
      "theoretical_loss": 3.654526321600205,
      "tokens_seen": 1069547520
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006826352110415663,
      "loss": 0.0728,
      "theoretical_loss": 3.6544428039186165,
      "tokens_seen": 1069809664
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000682554967099984,
      "loss": 0.0715,
      "theoretical_loss": 3.6543593124281264,
      "tokens_seen": 1070071808
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006824747231584015,
      "loss": 0.0738,
      "theoretical_loss": 3.654275847114107,
      "tokens_seen": 1070333952
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006823944792168192,
      "loss": 0.0761,
      "theoretical_loss": 3.6541924079619443,
      "tokens_seen": 1070596096
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006823142352752367,
      "loss": 0.0741,
      "theoretical_loss": 3.654108994957034,
      "tokens_seen": 1070858240
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006822339913336542,
      "loss": 0.075,
      "theoretical_loss": 3.654025608084786,
      "tokens_seen": 1071120384
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000682153747392072,
      "loss": 0.0743,
      "theoretical_loss": 3.653942247330619,
      "tokens_seen": 1071382528
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006820735034504895,
      "loss": 0.0725,
      "theoretical_loss": 3.653858912679966,
      "tokens_seen": 1071644672
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006819932595089071,
      "loss": 0.0735,
      "theoretical_loss": 3.6537756041182696,
      "tokens_seen": 1071906816
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006819130155673247,
      "loss": 0.0738,
      "theoretical_loss": 3.6536923216309862,
      "tokens_seen": 1072168960
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0006818327716257423,
      "loss": 0.0743,
      "theoretical_loss": 3.653609065203582,
      "tokens_seen": 1072431104
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006817525276841598,
      "loss": 0.0724,
      "theoretical_loss": 3.6535258348215356,
      "tokens_seen": 1072693248
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006816722837425775,
      "loss": 0.0747,
      "theoretical_loss": 3.653442630470337,
      "tokens_seen": 1072955392
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000681592039800995,
      "loss": 0.074,
      "theoretical_loss": 3.653359452135488,
      "tokens_seen": 1073217536
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006815117958594125,
      "loss": 0.0735,
      "theoretical_loss": 3.653276299802503,
      "tokens_seen": 1073479680
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006814315519178303,
      "loss": 0.0736,
      "theoretical_loss": 3.6531931734569056,
      "tokens_seen": 1073741824
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006813513079762478,
      "loss": 0.0724,
      "theoretical_loss": 3.6531100730842336,
      "tokens_seen": 1074003968
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006812710640346655,
      "loss": 0.0746,
      "theoretical_loss": 3.653026998670035,
      "tokens_seen": 1074266112
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000681190820093083,
      "loss": 0.0733,
      "theoretical_loss": 3.652943950199869,
      "tokens_seen": 1074528256
    },
    {
      "epoch": 0.33,
      "objective/train/advantage_avg": 0.0008113806834444404,
      "objective/train/docs_used": 394219,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4396977424621582,
      "objective/train/original_loss": 1.4396977424621582,
      "objective/train/theoretical_loss": 3.652860927659307,
      "objective/train/tokens_used": 1095250400,
      "objective/train/value_avg": -0.007663726806640625,
      "objective/train/value_loss": 0.00024871714413166046,
      "objective/train/value_max": -0.00011146068572998047,
      "objective/train/value_min": -0.373046875,
      "objective/train/value_reward_corr": 0.6646306086042744,
      "objective/train/value_std": 0.01288604736328125,
      "objective/train/weight_avg": 1.0009232759475708,
      "objective/train/weighted_lm_loss": 1.4412873983383179,
      "objective/train/weights_max": 1.1857911348342896,
      "objective/train/weights_min": 0.3686436712741852,
      "theoretical_loss": 3.652860927659307,
      "tokens_seen": 1074790400
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006811105761515006,
      "loss": 0.0734,
      "theoretical_loss": 3.652860927659307,
      "tokens_seen": 1074790400
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006810303322099182,
      "loss": 0.0727,
      "theoretical_loss": 3.6527779310339326,
      "tokens_seen": 1075052544
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006809500882683357,
      "loss": 0.0742,
      "theoretical_loss": 3.652694960309339,
      "tokens_seen": 1075314688
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006808698443267533,
      "loss": 0.0751,
      "theoretical_loss": 3.6526120154711332,
      "tokens_seen": 1075576832
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006807896003851709,
      "loss": 0.072,
      "theoretical_loss": 3.6525290965049324,
      "tokens_seen": 1075838976
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006807093564435886,
      "loss": 0.0738,
      "theoretical_loss": 3.652446203396365,
      "tokens_seen": 1076101120
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006806291125020061,
      "loss": 0.0729,
      "theoretical_loss": 3.6523633361310717,
      "tokens_seen": 1076363264
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006805488685604238,
      "loss": 0.0737,
      "theoretical_loss": 3.6522804946947045,
      "tokens_seen": 1076625408
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006804686246188413,
      "loss": 0.0728,
      "theoretical_loss": 3.6521976790729265,
      "tokens_seen": 1076887552
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006803883806772588,
      "loss": 0.0727,
      "theoretical_loss": 3.652114889251412,
      "tokens_seen": 1077149696
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006803081367356765,
      "loss": 0.0763,
      "theoretical_loss": 3.6520321252158485,
      "tokens_seen": 1077411840
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000680227892794094,
      "loss": 0.0724,
      "theoretical_loss": 3.651949386951933,
      "tokens_seen": 1077673984
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006801476488525116,
      "loss": 0.0729,
      "theoretical_loss": 3.6518666744453734,
      "tokens_seen": 1077936128
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006800674049109292,
      "loss": 0.073,
      "theoretical_loss": 3.651783987681892,
      "tokens_seen": 1078198272
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006799871609693469,
      "loss": 0.0734,
      "theoretical_loss": 3.6517013266472187,
      "tokens_seen": 1078460416
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006799069170277645,
      "loss": 0.0736,
      "theoretical_loss": 3.651618691327098,
      "tokens_seen": 1078722560
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000679826673086182,
      "loss": 0.0749,
      "theoretical_loss": 3.651536081707284,
      "tokens_seen": 1078984704
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006797464291445996,
      "loss": 0.0758,
      "theoretical_loss": 3.651453497773543,
      "tokens_seen": 1079246848
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006796661852030172,
      "loss": 0.0746,
      "theoretical_loss": 3.6513709395116516,
      "tokens_seen": 1079508992
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006795859412614348,
      "loss": 0.0714,
      "theoretical_loss": 3.651288406907399,
      "tokens_seen": 1079771136
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006795056973198523,
      "loss": 0.0737,
      "theoretical_loss": 3.6512058999465844,
      "tokens_seen": 1080033280
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00067942545337827,
      "loss": 0.0732,
      "theoretical_loss": 3.6511234186150197,
      "tokens_seen": 1080295424
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006793452094366875,
      "loss": 0.0742,
      "theoretical_loss": 3.6510409628985263,
      "tokens_seen": 1080557568
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000679264965495105,
      "loss": 0.0733,
      "theoretical_loss": 3.6509585327829392,
      "tokens_seen": 1080819712
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006791847215535228,
      "loss": 0.073,
      "theoretical_loss": 3.6508761282541027,
      "tokens_seen": 1081081856
    },
    {
      "epoch": 0.33,
      "objective/train/advantage_avg": 0.0017559973057359457,
      "objective/train/docs_used": 396489,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5023612976074219,
      "objective/train/original_loss": 1.5023612976074219,
      "objective/train/theoretical_loss": 3.6507937492978733,
      "objective/train/tokens_used": 1101804000,
      "objective/train/value_avg": -0.0075531005859375,
      "objective/train/value_loss": 0.0002941191487479955,
      "objective/train/value_max": -0.00011962652206420898,
      "objective/train/value_min": -0.9130859375,
      "objective/train/value_reward_corr": 0.6127776569660509,
      "objective/train/value_std": 0.0134735107421875,
      "objective/train/weight_avg": 1.0018857717514038,
      "objective/train/weighted_lm_loss": 1.5048304796218872,
      "objective/train/weights_max": 1.7268811464309692,
      "objective/train/weights_min": 0.3711496591567993,
      "theoretical_loss": 3.6507937492978733,
      "tokens_seen": 1081344000
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006791044776119403,
      "loss": 0.0738,
      "theoretical_loss": 3.6507937492978733,
      "tokens_seen": 1081344000
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006790242336703579,
      "loss": 0.0712,
      "theoretical_loss": 3.6507113959001183,
      "tokens_seen": 1081606144
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006789439897287755,
      "loss": 0.077,
      "theoretical_loss": 3.6506290680467166,
      "tokens_seen": 1081868288
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006788637457871931,
      "loss": 0.0739,
      "theoretical_loss": 3.650546765723558,
      "tokens_seen": 1082130432
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006787835018456106,
      "loss": 0.0731,
      "theoretical_loss": 3.650464488916544,
      "tokens_seen": 1082392576
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006787032579040283,
      "loss": 0.0746,
      "theoretical_loss": 3.650382237611587,
      "tokens_seen": 1082654720
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006786230139624458,
      "loss": 0.0763,
      "theoretical_loss": 3.65030001179461,
      "tokens_seen": 1082916864
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006785427700208634,
      "loss": 0.0735,
      "theoretical_loss": 3.650217811451548,
      "tokens_seen": 1083179008
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006784625260792811,
      "loss": 0.0728,
      "theoretical_loss": 3.650135636568347,
      "tokens_seen": 1083441152
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006783822821376986,
      "loss": 0.0734,
      "theoretical_loss": 3.6500534871309642,
      "tokens_seen": 1083703296
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006783020381961163,
      "loss": 0.0736,
      "theoretical_loss": 3.649971363125368,
      "tokens_seen": 1083965440
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006782217942545338,
      "loss": 0.0738,
      "theoretical_loss": 3.6498892645375367,
      "tokens_seen": 1084227584
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006781415503129514,
      "loss": 0.0727,
      "theoretical_loss": 3.649807191353462,
      "tokens_seen": 1084489728
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000678061306371369,
      "loss": 0.0742,
      "theoretical_loss": 3.6497251435591442,
      "tokens_seen": 1084751872
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006779810624297865,
      "loss": 0.0777,
      "theoretical_loss": 3.6496431211405973,
      "tokens_seen": 1085014016
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006779008184882041,
      "loss": 0.0714,
      "theoretical_loss": 3.649561124083844,
      "tokens_seen": 1085276160
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006778205745466217,
      "loss": 0.0752,
      "theoretical_loss": 3.6494791523749193,
      "tokens_seen": 1085538304
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006777403306050394,
      "loss": 0.0739,
      "theoretical_loss": 3.6493972059998696,
      "tokens_seen": 1085800448
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006776600866634569,
      "loss": 0.0734,
      "theoretical_loss": 3.649315284944751,
      "tokens_seen": 1086062592
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006775798427218746,
      "loss": 0.072,
      "theoretical_loss": 3.649233389195632,
      "tokens_seen": 1086324736
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006774995987802921,
      "loss": 0.0735,
      "theoretical_loss": 3.6491515187385914,
      "tokens_seen": 1086586880
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006774193548387097,
      "loss": 0.071,
      "theoretical_loss": 3.649069673559719,
      "tokens_seen": 1086849024
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006773391108971273,
      "loss": 0.0731,
      "theoretical_loss": 3.648987853645116,
      "tokens_seen": 1087111168
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006772588669555448,
      "loss": 0.0715,
      "theoretical_loss": 3.648906058980894,
      "tokens_seen": 1087373312
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006771786230139625,
      "loss": 0.0735,
      "theoretical_loss": 3.6488242895531764,
      "tokens_seen": 1087635456
    },
    {
      "epoch": 0.33,
      "objective/train/advantage_avg": 0.00033141745370812714,
      "objective/train/docs_used": 398862,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3994327783584595,
      "objective/train/original_loss": 1.39943265914917,
      "objective/train/theoretical_loss": 3.6487425453480973,
      "objective/train/tokens_used": 1108357600,
      "objective/train/value_avg": -0.01074981689453125,
      "objective/train/value_loss": 0.0003049577062483877,
      "objective/train/value_max": -0.0001499652862548828,
      "objective/train/value_min": -0.64697265625,
      "objective/train/value_reward_corr": 0.7564707177647516,
      "objective/train/value_std": 0.0179290771484375,
      "objective/train/weight_avg": 1.0004724264144897,
      "objective/train/weighted_lm_loss": 1.4000054597854614,
      "objective/train/weights_max": 1.461032509803772,
      "objective/train/weights_min": 0.3686436712741852,
      "theoretical_loss": 3.6487425453480973,
      "tokens_seen": 1087897600
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00067709837907238,
      "loss": 0.0722,
      "theoretical_loss": 3.6487425453480973,
      "tokens_seen": 1087897600
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006770181351307977,
      "loss": 0.0712,
      "theoretical_loss": 3.648660826351801,
      "tokens_seen": 1088159744
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006769378911892153,
      "loss": 0.0736,
      "theoretical_loss": 3.6485791325504437,
      "tokens_seen": 1088421888
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006768576472476328,
      "loss": 0.0729,
      "theoretical_loss": 3.648497463930192,
      "tokens_seen": 1088684032
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006767774033060504,
      "loss": 0.0729,
      "theoretical_loss": 3.6484158204772235,
      "tokens_seen": 1088946176
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000676697159364468,
      "loss": 0.0749,
      "theoretical_loss": 3.648334202177727,
      "tokens_seen": 1089208320
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006766169154228856,
      "loss": 0.0762,
      "theoretical_loss": 3.648252609017902,
      "tokens_seen": 1089470464
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006765366714813031,
      "loss": 0.0741,
      "theoretical_loss": 3.648171040983959,
      "tokens_seen": 1089732608
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006764564275397208,
      "loss": 0.0762,
      "theoretical_loss": 3.648089498062119,
      "tokens_seen": 1089994752
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006763761835981383,
      "loss": 0.0752,
      "theoretical_loss": 3.648007980238614,
      "tokens_seen": 1090256896
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006762959396565558,
      "loss": 0.0723,
      "theoretical_loss": 3.6479264874996877,
      "tokens_seen": 1090519040
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006762156957149736,
      "loss": 0.0728,
      "theoretical_loss": 3.6478450198315926,
      "tokens_seen": 1090781184
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006761354517733911,
      "loss": 0.0753,
      "theoretical_loss": 3.6477635772205947,
      "tokens_seen": 1091043328
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006760552078318088,
      "loss": 0.075,
      "theoretical_loss": 3.647682159652969,
      "tokens_seen": 1091305472
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006759749638902263,
      "loss": 0.074,
      "theoretical_loss": 3.647600767115002,
      "tokens_seen": 1091567616
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006758947199486439,
      "loss": 0.0758,
      "theoretical_loss": 3.6475193995929907,
      "tokens_seen": 1091829760
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006758144760070615,
      "loss": 0.0736,
      "theoretical_loss": 3.6474380570732423,
      "tokens_seen": 1092091904
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000675734232065479,
      "loss": 0.0749,
      "theoretical_loss": 3.6473567395420767,
      "tokens_seen": 1092354048
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006756539881238966,
      "loss": 0.0728,
      "theoretical_loss": 3.647275446985822,
      "tokens_seen": 1092616192
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006755737441823142,
      "loss": 0.074,
      "theoretical_loss": 3.64719417939082,
      "tokens_seen": 1092878336
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006754935002407319,
      "loss": 0.0759,
      "theoretical_loss": 3.6471129367434205,
      "tokens_seen": 1093140480
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006754132562991494,
      "loss": 0.0715,
      "theoretical_loss": 3.647031719029985,
      "tokens_seen": 1093402624
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006753330123575671,
      "loss": 0.073,
      "theoretical_loss": 3.646950526236887,
      "tokens_seen": 1093664768
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006752527684159846,
      "loss": 0.0741,
      "theoretical_loss": 3.6468693583505085,
      "tokens_seen": 1093926912
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006751725244744022,
      "loss": 0.0759,
      "theoretical_loss": 3.646788215357244,
      "tokens_seen": 1094189056
    },
    {
      "epoch": 0.33,
      "objective/train/advantage_avg": 0.000770938815549016,
      "objective/train/docs_used": 401268,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4965838193893433,
      "objective/train/original_loss": 1.4965837001800537,
      "objective/train/theoretical_loss": 3.646707097243498,
      "objective/train/tokens_used": 1114911200,
      "objective/train/value_avg": -0.00963592529296875,
      "objective/train/value_loss": 0.0003326190635561943,
      "objective/train/value_max": -0.0001634359359741211,
      "objective/train/value_min": -0.9462890625,
      "objective/train/value_reward_corr": 0.7103235067802546,
      "objective/train/value_std": 0.0166168212890625,
      "objective/train/weight_avg": 1.0009206533432007,
      "objective/train/weighted_lm_loss": 1.4972987174987793,
      "objective/train/weights_max": 1.833900809288025,
      "objective/train/weights_min": 0.36999329924583435,
      "theoretical_loss": 3.646707097243498,
      "tokens_seen": 1094451200
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006750922805328198,
      "loss": 0.0734,
      "theoretical_loss": 3.646707097243498,
      "tokens_seen": 1094451200
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006750120365912373,
      "loss": 0.0727,
      "theoretical_loss": 3.646626003995685,
      "tokens_seen": 1094713344
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000674931792649655,
      "loss": 0.0757,
      "theoretical_loss": 3.6465449356002315,
      "tokens_seen": 1094975488
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006748515487080725,
      "loss": 0.0712,
      "theoretical_loss": 3.646463892043574,
      "tokens_seen": 1095237632
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006747713047664902,
      "loss": 0.0736,
      "theoretical_loss": 3.6463828733121586,
      "tokens_seen": 1095499776
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006746910608249078,
      "loss": 0.0746,
      "theoretical_loss": 3.6463018793924453,
      "tokens_seen": 1095761920
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006746108168833254,
      "loss": 0.0712,
      "theoretical_loss": 3.6462209102709,
      "tokens_seen": 1096024064
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006745305729417429,
      "loss": 0.0721,
      "theoretical_loss": 3.6461399659340037,
      "tokens_seen": 1096286208
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006744503290001605,
      "loss": 0.0738,
      "theoretical_loss": 3.6460590463682454,
      "tokens_seen": 1096548352
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006743700850585781,
      "loss": 0.0727,
      "theoretical_loss": 3.6459781515601244,
      "tokens_seen": 1096810496
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006742898411169956,
      "loss": 0.0749,
      "theoretical_loss": 3.6458972814961528,
      "tokens_seen": 1097072640
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006742095971754133,
      "loss": 0.0723,
      "theoretical_loss": 3.6458164361628516,
      "tokens_seen": 1097334784
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006741293532338308,
      "loss": 0.0721,
      "theoretical_loss": 3.645735615546752,
      "tokens_seen": 1097596928
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006740491092922485,
      "loss": 0.0751,
      "theoretical_loss": 3.645654819634397,
      "tokens_seen": 1097859072
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006739688653506661,
      "loss": 0.0735,
      "theoretical_loss": 3.6455740484123407,
      "tokens_seen": 1098121216
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006738886214090836,
      "loss": 0.0713,
      "theoretical_loss": 3.645493301867145,
      "tokens_seen": 1098383360
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006738083774675012,
      "loss": 0.0749,
      "theoretical_loss": 3.6454125799853854,
      "tokens_seen": 1098645504
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006737281335259188,
      "loss": 0.0735,
      "theoretical_loss": 3.645331882753645,
      "tokens_seen": 1098907648
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006736478895843364,
      "loss": 0.0732,
      "theoretical_loss": 3.6452512101585195,
      "tokens_seen": 1099169792
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000673567645642754,
      "loss": 0.0701,
      "theoretical_loss": 3.645170562186615,
      "tokens_seen": 1099431936
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006734874017011716,
      "loss": 0.0742,
      "theoretical_loss": 3.6450899388245466,
      "tokens_seen": 1099694080
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006734071577595891,
      "loss": 0.0747,
      "theoretical_loss": 3.645009340058941,
      "tokens_seen": 1099956224
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006733269138180067,
      "loss": 0.0734,
      "theoretical_loss": 3.644928765876436,
      "tokens_seen": 1100218368
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006732466698764244,
      "loss": 0.0722,
      "theoretical_loss": 3.644848216263678,
      "tokens_seen": 1100480512
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006731664259348419,
      "loss": 0.075,
      "theoretical_loss": 3.6447676912073255,
      "tokens_seen": 1100742656
    },
    {
      "epoch": 0.33,
      "objective/train/advantage_avg": -0.001446638605557382,
      "objective/train/docs_used": 403530,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5791977643966675,
      "objective/train/original_loss": 1.5791977643966675,
      "objective/train/theoretical_loss": 3.6446871906940466,
      "objective/train/tokens_used": 1121464800,
      "objective/train/value_avg": -0.00872802734375,
      "objective/train/value_loss": 0.00039117602864280343,
      "objective/train/value_max": -0.00010389089584350586,
      "objective/train/value_min": -0.5576171875,
      "objective/train/value_reward_corr": 0.8311069137177848,
      "objective/train/value_std": 0.01605224609375,
      "objective/train/weight_avg": 0.998737633228302,
      "objective/train/weighted_lm_loss": 1.5767664909362793,
      "objective/train/weights_max": 1.3273746967315674,
      "objective/train/weights_min": 0.368156373500824,
      "theoretical_loss": 3.6446871906940466,
      "tokens_seen": 1101004800
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006730861819932596,
      "loss": 0.0757,
      "theoretical_loss": 3.6446871906940466,
      "tokens_seen": 1101004800
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006730059380516771,
      "loss": 0.072,
      "theoretical_loss": 3.6446067147105197,
      "tokens_seen": 1101266944
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006729256941100947,
      "loss": 0.0713,
      "theoretical_loss": 3.644526263243433,
      "tokens_seen": 1101529088
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006728454501685123,
      "loss": 0.0749,
      "theoretical_loss": 3.644445836279488,
      "tokens_seen": 1101791232
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006727652062269298,
      "loss": 0.0697,
      "theoretical_loss": 3.644365433805393,
      "tokens_seen": 1102053376
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006726849622853474,
      "loss": 0.0721,
      "theoretical_loss": 3.6442850558078685,
      "tokens_seen": 1102315520
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000672604718343765,
      "loss": 0.0766,
      "theoretical_loss": 3.6442047022736452,
      "tokens_seen": 1102577664
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006725244744021827,
      "loss": 0.0759,
      "theoretical_loss": 3.644124373189464,
      "tokens_seen": 1102839808
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006724442304606002,
      "loss": 0.0744,
      "theoretical_loss": 3.644044068542076,
      "tokens_seen": 1103101952
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006723639865190179,
      "loss": 0.0728,
      "theoretical_loss": 3.643963788318242,
      "tokens_seen": 1103364096
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006722837425774354,
      "loss": 0.0723,
      "theoretical_loss": 3.6438835325047356,
      "tokens_seen": 1103626240
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006722034986358531,
      "loss": 0.0729,
      "theoretical_loss": 3.6438033010883375,
      "tokens_seen": 1103888384
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006721232546942706,
      "loss": 0.0745,
      "theoretical_loss": 3.643723094055841,
      "tokens_seen": 1104150528
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006720430107526881,
      "loss": 0.0767,
      "theoretical_loss": 3.643642911394048,
      "tokens_seen": 1104412672
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006719627668111058,
      "loss": 0.0734,
      "theoretical_loss": 3.643562753089772,
      "tokens_seen": 1104674816
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0006718825228695233,
      "loss": 0.0733,
      "theoretical_loss": 3.6434826191298364,
      "tokens_seen": 1104936960
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000671802278927941,
      "loss": 0.0726,
      "theoretical_loss": 3.6434025095010747,
      "tokens_seen": 1105199104
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006717220349863586,
      "loss": 0.0763,
      "theoretical_loss": 3.6433224241903304,
      "tokens_seen": 1105461248
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006716417910447762,
      "loss": 0.0749,
      "theoretical_loss": 3.643242363184458,
      "tokens_seen": 1105723392
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006715615471031937,
      "loss": 0.0741,
      "theoretical_loss": 3.6431623264703212,
      "tokens_seen": 1105985536
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006714813031616113,
      "loss": 0.0711,
      "theoretical_loss": 3.6430823140347943,
      "tokens_seen": 1106247680
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006714010592200289,
      "loss": 0.0726,
      "theoretical_loss": 3.643002325864763,
      "tokens_seen": 1106509824
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006713208152784464,
      "loss": 0.0726,
      "theoretical_loss": 3.6429223619471207,
      "tokens_seen": 1106771968
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006712405713368641,
      "loss": 0.0749,
      "theoretical_loss": 3.6428424222687736,
      "tokens_seen": 1107034112
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006711603273952816,
      "loss": 0.074,
      "theoretical_loss": 3.642762506816636,
      "tokens_seen": 1107296256
    },
    {
      "epoch": 0.34,
      "objective/train/advantage_avg": 0.0007008896209299564,
      "objective/train/docs_used": 405946,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.543431043624878,
      "objective/train/original_loss": 1.5434308052062988,
      "objective/train/theoretical_loss": 3.642682615577634,
      "objective/train/tokens_used": 1128018400,
      "objective/train/value_avg": -0.0092010498046875,
      "objective/train/value_loss": 0.00045838035293854773,
      "objective/train/value_max": -7.969141006469727e-05,
      "objective/train/value_min": -0.953125,
      "objective/train/value_reward_corr": 0.5853789003053751,
      "objective/train/value_std": 0.0174102783203125,
      "objective/train/weight_avg": 1.000909447669983,
      "objective/train/weighted_lm_loss": 1.5446616411209106,
      "objective/train/weights_max": 2.1671571731567383,
      "objective/train/weights_min": 0.3689032196998596,
      "theoretical_loss": 3.642682615577634,
      "tokens_seen": 1107558400
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006710800834536994,
      "loss": 0.0717,
      "theoretical_loss": 3.642682615577634,
      "tokens_seen": 1107558400
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006709998395121169,
      "loss": 0.0757,
      "theoretical_loss": 3.6426027485387023,
      "tokens_seen": 1107820544
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006709195955705344,
      "loss": 0.0733,
      "theoretical_loss": 3.6425229056867865,
      "tokens_seen": 1108082688
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006708393516289521,
      "loss": 0.0723,
      "theoretical_loss": 3.642443087008844,
      "tokens_seen": 1108344832
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006707591076873696,
      "loss": 0.0716,
      "theoretical_loss": 3.6423632924918383,
      "tokens_seen": 1108606976
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006706788637457872,
      "loss": 0.0728,
      "theoretical_loss": 3.6422835221227468,
      "tokens_seen": 1108869120
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006705986198042048,
      "loss": 0.0728,
      "theoretical_loss": 3.6422037758885555,
      "tokens_seen": 1109131264
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006705183758626224,
      "loss": 0.0743,
      "theoretical_loss": 3.6421240537762607,
      "tokens_seen": 1109393408
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006704381319210399,
      "loss": 0.0772,
      "theoretical_loss": 3.6420443557728674,
      "tokens_seen": 1109655552
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006703578879794575,
      "loss": 0.0741,
      "theoretical_loss": 3.6419646818653932,
      "tokens_seen": 1109917696
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006702776440378752,
      "loss": 0.076,
      "theoretical_loss": 3.641885032040864,
      "tokens_seen": 1110179840
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006701974000962927,
      "loss": 0.0751,
      "theoretical_loss": 3.6418054062863163,
      "tokens_seen": 1110441984
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006701171561547104,
      "loss": 0.073,
      "theoretical_loss": 3.6417258045887966,
      "tokens_seen": 1110704128
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006700369122131279,
      "loss": 0.0744,
      "theoretical_loss": 3.641646226935361,
      "tokens_seen": 1110966272
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006699566682715455,
      "loss": 0.0759,
      "theoretical_loss": 3.641566673313076,
      "tokens_seen": 1111228416
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006698764243299631,
      "loss": 0.0776,
      "theoretical_loss": 3.6414871437090186,
      "tokens_seen": 1111490560
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006697961803883806,
      "loss": 0.0752,
      "theoretical_loss": 3.641407638110275,
      "tokens_seen": 1111752704
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006697159364467983,
      "loss": 0.0753,
      "theoretical_loss": 3.641328156503942,
      "tokens_seen": 1112014848
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006696356925052158,
      "loss": 0.0757,
      "theoretical_loss": 3.6412486988771255,
      "tokens_seen": 1112276992
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006695554485636335,
      "loss": 0.0727,
      "theoretical_loss": 3.6411692652169423,
      "tokens_seen": 1112539136
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006694752046220511,
      "loss": 0.0749,
      "theoretical_loss": 3.641089855510518,
      "tokens_seen": 1112801280
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006693949606804687,
      "loss": 0.0726,
      "theoretical_loss": 3.64101046974499,
      "tokens_seen": 1113063424
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006693147167388862,
      "loss": 0.074,
      "theoretical_loss": 3.640931107907504,
      "tokens_seen": 1113325568
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006692344727973039,
      "loss": 0.0767,
      "theoretical_loss": 3.6408517699852165,
      "tokens_seen": 1113587712
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006691542288557214,
      "loss": 0.0779,
      "theoretical_loss": 3.640772455965293,
      "tokens_seen": 1113849856
    },
    {
      "epoch": 0.34,
      "objective/train/advantage_avg": 0.0008538411930203438,
      "objective/train/docs_used": 408181,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3940370082855225,
      "objective/train/original_loss": 1.3940370082855225,
      "objective/train/theoretical_loss": 3.64069316583491,
      "objective/train/tokens_used": 1134572000,
      "objective/train/value_avg": -0.0080718994140625,
      "objective/train/value_loss": 0.0002601642336230725,
      "objective/train/value_max": -8.684396743774414e-05,
      "objective/train/value_min": -0.57763671875,
      "objective/train/value_reward_corr": 0.6369542776258966,
      "objective/train/value_std": 0.01348876953125,
      "objective/train/weight_avg": 1.0009734630584717,
      "objective/train/weighted_lm_loss": 1.3951406478881836,
      "objective/train/weights_max": 1.348272681236267,
      "objective/train/weights_min": 0.36883002519607544,
      "theoretical_loss": 3.64069316583491,
      "tokens_seen": 1114112000
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006690739849141389,
      "loss": 0.0725,
      "theoretical_loss": 3.64069316583491,
      "tokens_seen": 1114112000
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006689937409725566,
      "loss": 0.075,
      "theoretical_loss": 3.640613899581253,
      "tokens_seen": 1114374144
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006689134970309741,
      "loss": 0.0775,
      "theoretical_loss": 3.6405346571915187,
      "tokens_seen": 1114636288
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006688332530893918,
      "loss": 0.0755,
      "theoretical_loss": 3.6404554386529115,
      "tokens_seen": 1114898432
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006687530091478094,
      "loss": 0.0721,
      "theoretical_loss": 3.640376243952648,
      "tokens_seen": 1115160576
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000668672765206227,
      "loss": 0.0721,
      "theoretical_loss": 3.640297073077953,
      "tokens_seen": 1115422720
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006685925212646445,
      "loss": 0.0777,
      "theoretical_loss": 3.640217926016061,
      "tokens_seen": 1115684864
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006685122773230621,
      "loss": 0.0741,
      "theoretical_loss": 3.6401388027542185,
      "tokens_seen": 1115947008
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006684320333814797,
      "loss": 0.0707,
      "theoretical_loss": 3.6400597032796798,
      "tokens_seen": 1116209152
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006683517894398973,
      "loss": 0.0752,
      "theoretical_loss": 3.6399806275797095,
      "tokens_seen": 1116471296
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006682715454983149,
      "loss": 0.0761,
      "theoretical_loss": 3.639901575641582,
      "tokens_seen": 1116733440
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006681913015567324,
      "loss": 0.0739,
      "theoretical_loss": 3.6398225474525816,
      "tokens_seen": 1116995584
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006681110576151502,
      "loss": 0.0725,
      "theoretical_loss": 3.639743543000003,
      "tokens_seen": 1117257728
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006680308136735677,
      "loss": 0.0732,
      "theoretical_loss": 3.6396645622711494,
      "tokens_seen": 1117519872
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006679505697319852,
      "loss": 0.0764,
      "theoretical_loss": 3.639585605253335,
      "tokens_seen": 1117782016
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006678703257904029,
      "loss": 0.0736,
      "theoretical_loss": 3.639506671933882,
      "tokens_seen": 1118044160
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006677900818488204,
      "loss": 0.0739,
      "theoretical_loss": 3.639427762300125,
      "tokens_seen": 1118306304
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000667709837907238,
      "loss": 0.0747,
      "theoretical_loss": 3.6393488763394064,
      "tokens_seen": 1118568448
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006676295939656556,
      "loss": 0.0737,
      "theoretical_loss": 3.639270014039078,
      "tokens_seen": 1118830592
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006675493500240732,
      "loss": 0.0739,
      "theoretical_loss": 3.6391911753865034,
      "tokens_seen": 1119092736
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006674691060824907,
      "loss": 0.0753,
      "theoretical_loss": 3.639112360369054,
      "tokens_seen": 1119354880
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006673888621409083,
      "loss": 0.0764,
      "theoretical_loss": 3.6390335689741113,
      "tokens_seen": 1119617024
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000667308618199326,
      "loss": 0.0751,
      "theoretical_loss": 3.638954801189067,
      "tokens_seen": 1119879168
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006672283742577436,
      "loss": 0.0758,
      "theoretical_loss": 3.6388760570013226,
      "tokens_seen": 1120141312
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006671481303161612,
      "loss": 0.0729,
      "theoretical_loss": 3.6387973363982877,
      "tokens_seen": 1120403456
    },
    {
      "epoch": 0.34,
      "objective/train/advantage_avg": 0.00035512945032678545,
      "objective/train/docs_used": 410712,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4525707960128784,
      "objective/train/original_loss": 1.4525706768035889,
      "objective/train/theoretical_loss": 3.638718639367384,
      "objective/train/tokens_used": 1141125600,
      "objective/train/value_avg": -0.00772857666015625,
      "objective/train/value_loss": 0.00021194637520238757,
      "objective/train/value_max": -5.3882598876953125e-05,
      "objective/train/value_min": -0.6298828125,
      "objective/train/value_reward_corr": 0.7386635939472923,
      "objective/train/value_std": 0.014801025390625,
      "objective/train/weight_avg": 1.0004568099975586,
      "objective/train/weighted_lm_loss": 1.453499674797058,
      "objective/train/weights_max": 1.5431358814239502,
      "objective/train/weights_min": 0.541466236114502,
      "theoretical_loss": 3.638718639367384,
      "tokens_seen": 1120665600
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006670678863745787,
      "loss": 0.0727,
      "theoretical_loss": 3.638718639367384,
      "tokens_seen": 1120665600
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006669876424329964,
      "loss": 0.0738,
      "theoretical_loss": 3.638639965896041,
      "tokens_seen": 1120927744
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006669073984914139,
      "loss": 0.0765,
      "theoretical_loss": 3.638561315971698,
      "tokens_seen": 1121189888
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006668271545498314,
      "loss": 0.0732,
      "theoretical_loss": 3.638482689581805,
      "tokens_seen": 1121452032
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006667469106082491,
      "loss": 0.0724,
      "theoretical_loss": 3.6384040867138214,
      "tokens_seen": 1121714176
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006666666666666666,
      "loss": 0.0765,
      "theoretical_loss": 3.6383255073552148,
      "tokens_seen": 1121976320
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006665864227250843,
      "loss": 0.0752,
      "theoretical_loss": 3.638246951493463,
      "tokens_seen": 1122238464
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006665061787835019,
      "loss": 0.0739,
      "theoretical_loss": 3.6381684191160555,
      "tokens_seen": 1122500608
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006664259348419195,
      "loss": 0.0725,
      "theoretical_loss": 3.638089910210488,
      "tokens_seen": 1122762752
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000666345690900337,
      "loss": 0.0762,
      "theoretical_loss": 3.638011424764269,
      "tokens_seen": 1123024896
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006662654469587546,
      "loss": 0.0711,
      "theoretical_loss": 3.6379329627649137,
      "tokens_seen": 1123287040
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006661852030171722,
      "loss": 0.0732,
      "theoretical_loss": 3.6378545241999487,
      "tokens_seen": 1123549184
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006661049590755897,
      "loss": 0.0746,
      "theoretical_loss": 3.637776109056909,
      "tokens_seen": 1123811328
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006660247151340074,
      "loss": 0.0737,
      "theoretical_loss": 3.6376977173233405,
      "tokens_seen": 1124073472
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006659444711924249,
      "loss": 0.0729,
      "theoretical_loss": 3.6376193489867976,
      "tokens_seen": 1124335616
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006658642272508427,
      "loss": 0.0754,
      "theoretical_loss": 3.6375410040348446,
      "tokens_seen": 1124597760
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006657839833092602,
      "loss": 0.071,
      "theoretical_loss": 3.637462682455055,
      "tokens_seen": 1124859904
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006657037393676778,
      "loss": 0.0737,
      "theoretical_loss": 3.6373843842350118,
      "tokens_seen": 1125122048
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006656234954260954,
      "loss": 0.0752,
      "theoretical_loss": 3.637306109362308,
      "tokens_seen": 1125384192
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006655432514845129,
      "loss": 0.0735,
      "theoretical_loss": 3.6372278578245454,
      "tokens_seen": 1125646336
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006654630075429305,
      "loss": 0.0732,
      "theoretical_loss": 3.6371496296093357,
      "tokens_seen": 1125908480
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006653827636013481,
      "loss": 0.0746,
      "theoretical_loss": 3.6370714247043003,
      "tokens_seen": 1126170624
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006653025196597657,
      "loss": 0.0736,
      "theoretical_loss": 3.6369932430970695,
      "tokens_seen": 1126432768
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006652222757181832,
      "loss": 0.077,
      "theoretical_loss": 3.6369150847752834,
      "tokens_seen": 1126694912
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000665142031776601,
      "loss": 0.0746,
      "theoretical_loss": 3.6368369497265913,
      "tokens_seen": 1126957056
    },
    {
      "epoch": 0.34,
      "objective/train/advantage_avg": -0.0004652831412386149,
      "objective/train/docs_used": 413086,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3972619771957397,
      "objective/train/original_loss": 1.3972619771957397,
      "objective/train/theoretical_loss": 3.6367588379386513,
      "objective/train/tokens_used": 1147679200,
      "objective/train/value_avg": -0.0119476318359375,
      "objective/train/value_loss": 0.0004500811919569969,
      "objective/train/value_max": -0.00012242794036865234,
      "objective/train/value_min": -0.638671875,
      "objective/train/value_reward_corr": 0.8368038617349711,
      "objective/train/value_std": 0.02716064453125,
      "objective/train/weight_avg": 0.9997386932373047,
      "objective/train/weighted_lm_loss": 1.3970850706100464,
      "objective/train/weights_max": 1.5594542026519775,
      "objective/train/weights_min": 0.36921927332878113,
      "theoretical_loss": 3.6367588379386513,
      "tokens_seen": 1127219200
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006650617878350185,
      "loss": 0.0732,
      "theoretical_loss": 3.6367588379386513,
      "tokens_seen": 1127219200
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000664981543893436,
      "loss": 0.0754,
      "theoretical_loss": 3.636680749399133,
      "tokens_seen": 1127481344
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006649012999518537,
      "loss": 0.0744,
      "theoretical_loss": 3.6366026840957133,
      "tokens_seen": 1127743488
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006648210560102712,
      "loss": 0.076,
      "theoretical_loss": 3.636524642016079,
      "tokens_seen": 1128005632
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006647408120686889,
      "loss": 0.0743,
      "theoretical_loss": 3.636446623147927,
      "tokens_seen": 1128267776
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006646605681271064,
      "loss": 0.0761,
      "theoretical_loss": 3.6363686274789626,
      "tokens_seen": 1128529920
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000664580324185524,
      "loss": 0.0731,
      "theoretical_loss": 3.6362906549969014,
      "tokens_seen": 1128792064
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006645000802439416,
      "loss": 0.0741,
      "theoretical_loss": 3.6362127056894673,
      "tokens_seen": 1129054208
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006644198363023591,
      "loss": 0.0751,
      "theoretical_loss": 3.6361347795443955,
      "tokens_seen": 1129316352
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006643395923607768,
      "loss": 0.0739,
      "theoretical_loss": 3.636056876549427,
      "tokens_seen": 1129578496
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006642593484191944,
      "loss": 0.0738,
      "theoretical_loss": 3.6359789966923164,
      "tokens_seen": 1129840640
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000664179104477612,
      "loss": 0.0773,
      "theoretical_loss": 3.6359011399608243,
      "tokens_seen": 1130102784
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006640988605360295,
      "loss": 0.0748,
      "theoretical_loss": 3.6358233063427225,
      "tokens_seen": 1130364928
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006640186165944472,
      "loss": 0.0717,
      "theoretical_loss": 3.635745495825791,
      "tokens_seen": 1130627072
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006639383726528647,
      "loss": 0.0721,
      "theoretical_loss": 3.63566770839782,
      "tokens_seen": 1130889216
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006638581287112822,
      "loss": 0.0742,
      "theoretical_loss": 3.6355899440466075,
      "tokens_seen": 1131151360
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006637778847696999,
      "loss": 0.0733,
      "theoretical_loss": 3.635512202759964,
      "tokens_seen": 1131413504
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006636976408281174,
      "loss": 0.0738,
      "theoretical_loss": 3.635434484525704,
      "tokens_seen": 1131675648
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006636173968865351,
      "loss": 0.0759,
      "theoretical_loss": 3.6353567893316567,
      "tokens_seen": 1131937792
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006635371529449527,
      "loss": 0.0737,
      "theoretical_loss": 3.6352791171656573,
      "tokens_seen": 1132199936
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006634569090033703,
      "loss": 0.0742,
      "theoretical_loss": 3.635201468015551,
      "tokens_seen": 1132462080
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006633766650617879,
      "loss": 0.074,
      "theoretical_loss": 3.635123841869193,
      "tokens_seen": 1132724224
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006632964211202054,
      "loss": 0.0751,
      "theoretical_loss": 3.6350462387144464,
      "tokens_seen": 1132986368
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000663216177178623,
      "loss": 0.0747,
      "theoretical_loss": 3.634968658539184,
      "tokens_seen": 1133248512
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006631359332370406,
      "loss": 0.0768,
      "theoretical_loss": 3.6348911013312883,
      "tokens_seen": 1133510656
    },
    {
      "epoch": 0.34,
      "objective/train/advantage_avg": 0.0009396148379892111,
      "objective/train/docs_used": 415442,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4944310188293457,
      "objective/train/original_loss": 1.4944311380386353,
      "objective/train/theoretical_loss": 3.6348135670786506,
      "objective/train/tokens_used": 1154232800,
      "objective/train/value_avg": -0.0083770751953125,
      "objective/train/value_loss": 0.00016622063412796706,
      "objective/train/value_max": -7.545948028564453e-05,
      "objective/train/value_min": -0.54296875,
      "objective/train/value_reward_corr": 0.7126787421781615,
      "objective/train/value_std": 0.01410675048828125,
      "objective/train/weight_avg": 1.0010205507278442,
      "objective/train/weighted_lm_loss": 1.4958319664001465,
      "objective/train/weights_max": 1.4884049892425537,
      "objective/train/weights_min": 0.5564592480659485,
      "theoretical_loss": 3.6348135670786506,
      "tokens_seen": 1133772800
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006630556892954582,
      "loss": 0.0735,
      "theoretical_loss": 3.6348135670786506,
      "tokens_seen": 1133772800
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006629754453538757,
      "loss": 0.0746,
      "theoretical_loss": 3.6347360557691712,
      "tokens_seen": 1134034944
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006628952014122935,
      "loss": 0.0737,
      "theoretical_loss": 3.63465856739076,
      "tokens_seen": 1134297088
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000662814957470711,
      "loss": 0.0745,
      "theoretical_loss": 3.634581101931336,
      "tokens_seen": 1134559232
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006627347135291286,
      "loss": 0.0753,
      "theoretical_loss": 3.6345036593788276,
      "tokens_seen": 1134821376
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006626544695875462,
      "loss": 0.0758,
      "theoretical_loss": 3.6344262397211704,
      "tokens_seen": 1135083520
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006625742256459637,
      "loss": 0.0753,
      "theoretical_loss": 3.6343488429463124,
      "tokens_seen": 1135345664
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006624939817043813,
      "loss": 0.076,
      "theoretical_loss": 3.634271469042208,
      "tokens_seen": 1135607808
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006624137377627989,
      "loss": 0.0727,
      "theoretical_loss": 3.634194117996822,
      "tokens_seen": 1135869952
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006623334938212165,
      "loss": 0.0736,
      "theoretical_loss": 3.634116789798129,
      "tokens_seen": 1136132096
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000662253249879634,
      "loss": 0.0768,
      "theoretical_loss": 3.6340394844341097,
      "tokens_seen": 1136394240
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006621730059380518,
      "loss": 0.0768,
      "theoretical_loss": 3.6339622018927575,
      "tokens_seen": 1136656384
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006620927619964693,
      "loss": 0.0738,
      "theoretical_loss": 3.633884942162073,
      "tokens_seen": 1136918528
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006620125180548869,
      "loss": 0.0733,
      "theoretical_loss": 3.6338077052300664,
      "tokens_seen": 1137180672
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006619322741133045,
      "loss": 0.0725,
      "theoretical_loss": 3.633730491084756,
      "tokens_seen": 1137442816
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000661852030171722,
      "loss": 0.0746,
      "theoretical_loss": 3.6336532997141706,
      "tokens_seen": 1137704960
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006617717862301397,
      "loss": 0.0779,
      "theoretical_loss": 3.6335761311063473,
      "tokens_seen": 1137967104
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0006616915422885572,
      "loss": 0.0763,
      "theoretical_loss": 3.633498985249332,
      "tokens_seen": 1138229248
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006616112983469748,
      "loss": 0.0738,
      "theoretical_loss": 3.63342186213118,
      "tokens_seen": 1138491392
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006615310544053924,
      "loss": 0.0735,
      "theoretical_loss": 3.6333447617399557,
      "tokens_seen": 1138753536
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00066145081046381,
      "loss": 0.0751,
      "theoretical_loss": 3.6332676840637324,
      "tokens_seen": 1139015680
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006613705665222276,
      "loss": 0.0758,
      "theoretical_loss": 3.633190629090592,
      "tokens_seen": 1139277824
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006612903225806452,
      "loss": 0.075,
      "theoretical_loss": 3.6331135968086263,
      "tokens_seen": 1139539968
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006612100786390628,
      "loss": 0.0738,
      "theoretical_loss": 3.633036587205935,
      "tokens_seen": 1139802112
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006611298346974803,
      "loss": 0.0745,
      "theoretical_loss": 3.6329596002706275,
      "tokens_seen": 1140064256
    },
    {
      "epoch": 0.35,
      "objective/train/advantage_avg": 0.0012790506007149816,
      "objective/train/docs_used": 417664,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.491121530532837,
      "objective/train/original_loss": 1.491121530532837,
      "objective/train/theoretical_loss": 3.632882635990822,
      "objective/train/tokens_used": 1160786400,
      "objective/train/value_avg": -0.006160736083984375,
      "objective/train/value_loss": 0.00011857429490191862,
      "objective/train/value_max": -8.684396743774414e-05,
      "objective/train/value_min": -0.5185546875,
      "objective/train/value_reward_corr": 0.7576070544182378,
      "objective/train/value_std": 0.01470184326171875,
      "objective/train/weight_avg": 1.0013362169265747,
      "objective/train/weighted_lm_loss": 1.4934520721435547,
      "objective/train/weights_max": 1.1257035732269287,
      "objective/train/weights_min": 0.4068576395511627,
      "theoretical_loss": 3.632882635990822,
      "tokens_seen": 1140326400
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000661049590755898,
      "loss": 0.0743,
      "theoretical_loss": 3.632882635990822,
      "tokens_seen": 1140326400
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006609693468143155,
      "loss": 0.0763,
      "theoretical_loss": 3.632805694354646,
      "tokens_seen": 1140588544
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006608891028727331,
      "loss": 0.0751,
      "theoretical_loss": 3.6327287753502358,
      "tokens_seen": 1140850688
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006608088589311507,
      "loss": 0.0752,
      "theoretical_loss": 3.632651878965735,
      "tokens_seen": 1141112832
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006607286149895682,
      "loss": 0.0735,
      "theoretical_loss": 3.632575005189299,
      "tokens_seen": 1141374976
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000660648371047986,
      "loss": 0.0763,
      "theoretical_loss": 3.6324981540090895,
      "tokens_seen": 1141637120
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006605681271064035,
      "loss": 0.0741,
      "theoretical_loss": 3.6324213254132793,
      "tokens_seen": 1141899264
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006604878831648211,
      "loss": 0.0758,
      "theoretical_loss": 3.632344519390049,
      "tokens_seen": 1142161408
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006604076392232387,
      "loss": 0.0725,
      "theoretical_loss": 3.632267735927588,
      "tokens_seen": 1142423552
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006603273952816562,
      "loss": 0.0745,
      "theoretical_loss": 3.632190975014094,
      "tokens_seen": 1142685696
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006602471513400738,
      "loss": 0.0721,
      "theoretical_loss": 3.6321142366377757,
      "tokens_seen": 1142947840
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006601669073984914,
      "loss": 0.0719,
      "theoretical_loss": 3.6320375207868483,
      "tokens_seen": 1143209984
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000660086663456909,
      "loss": 0.073,
      "theoretical_loss": 3.6319608274495376,
      "tokens_seen": 1143472128
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006600064195153265,
      "loss": 0.0744,
      "theoretical_loss": 3.6318841566140767,
      "tokens_seen": 1143734272
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006599261755737443,
      "loss": 0.0722,
      "theoretical_loss": 3.63180750826871,
      "tokens_seen": 1143996416
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006598459316321618,
      "loss": 0.0701,
      "theoretical_loss": 3.6317308824016874,
      "tokens_seen": 1144258560
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006597656876905794,
      "loss": 0.0707,
      "theoretical_loss": 3.6316542790012702,
      "tokens_seen": 1144520704
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000659685443748997,
      "loss": 0.0717,
      "theoretical_loss": 3.631577698055727,
      "tokens_seen": 1144782848
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006596051998074145,
      "loss": 0.0705,
      "theoretical_loss": 3.631501139553337,
      "tokens_seen": 1145044992
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006595249558658322,
      "loss": 0.0738,
      "theoretical_loss": 3.6314246034823867,
      "tokens_seen": 1145307136
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006594447119242497,
      "loss": 0.075,
      "theoretical_loss": 3.631348089831171,
      "tokens_seen": 1145569280
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006593644679826673,
      "loss": 0.0737,
      "theoretical_loss": 3.631271598587995,
      "tokens_seen": 1145831424
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006592842240410849,
      "loss": 0.0739,
      "theoretical_loss": 3.631195129741172,
      "tokens_seen": 1146093568
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006592039800995026,
      "loss": 0.0747,
      "theoretical_loss": 3.631118683279024,
      "tokens_seen": 1146355712
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006591237361579201,
      "loss": 0.0734,
      "theoretical_loss": 3.6310422591898814,
      "tokens_seen": 1146617856
    },
    {
      "epoch": 0.35,
      "objective/train/advantage_avg": 0.0006740565877407789,
      "objective/train/docs_used": 419954,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3567689657211304,
      "objective/train/original_loss": 1.3567688465118408,
      "objective/train/theoretical_loss": 3.630965857462084,
      "objective/train/tokens_used": 1167340000,
      "objective/train/value_avg": -0.00856781005859375,
      "objective/train/value_loss": 0.0002693459973670542,
      "objective/train/value_max": -5.9664249420166016e-05,
      "objective/train/value_min": -0.62451171875,
      "objective/train/value_reward_corr": 0.6644929738275986,
      "objective/train/value_std": 0.01461029052734375,
      "objective/train/weight_avg": 1.0007959604263306,
      "objective/train/weighted_lm_loss": 1.357459545135498,
      "objective/train/weights_max": 1.5120787620544434,
      "objective/train/weights_min": 0.3692566156387329,
      "theoretical_loss": 3.630965857462084,
      "tokens_seen": 1146880000
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006590434922163377,
      "loss": 0.0715,
      "theoretical_loss": 3.630965857462084,
      "tokens_seen": 1146880000
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006589632482747553,
      "loss": 0.0742,
      "theoretical_loss": 3.6308894780839798,
      "tokens_seen": 1147142144
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006588830043331728,
      "loss": 0.0715,
      "theoretical_loss": 3.630813121043926,
      "tokens_seen": 1147404288
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006588027603915905,
      "loss": 0.0738,
      "theoretical_loss": 3.630736786330288,
      "tokens_seen": 1147666432
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000658722516450008,
      "loss": 0.0747,
      "theoretical_loss": 3.630660473931441,
      "tokens_seen": 1147928576
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006586422725084256,
      "loss": 0.0754,
      "theoretical_loss": 3.6305841838357673,
      "tokens_seen": 1148190720
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006585620285668432,
      "loss": 0.0704,
      "theoretical_loss": 3.630507916031659,
      "tokens_seen": 1148452864
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006584817846252607,
      "loss": 0.0744,
      "theoretical_loss": 3.630431670507517,
      "tokens_seen": 1148715008
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006584015406836785,
      "loss": 0.0741,
      "theoretical_loss": 3.6303554472517496,
      "tokens_seen": 1148977152
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000658321296742096,
      "loss": 0.0715,
      "theoretical_loss": 3.6302792462527758,
      "tokens_seen": 1149239296
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006582410528005136,
      "loss": 0.071,
      "theoretical_loss": 3.6302030674990213,
      "tokens_seen": 1149501440
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006581608088589312,
      "loss": 0.075,
      "theoretical_loss": 3.6301269109789214,
      "tokens_seen": 1149763584
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006580805649173488,
      "loss": 0.0717,
      "theoretical_loss": 3.63005077668092,
      "tokens_seen": 1150025728
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006580003209757663,
      "loss": 0.0723,
      "theoretical_loss": 3.62997466459347,
      "tokens_seen": 1150287872
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006579200770341839,
      "loss": 0.0704,
      "theoretical_loss": 3.629898574705031,
      "tokens_seen": 1150550016
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006578398330926015,
      "loss": 0.0734,
      "theoretical_loss": 3.629822507004075,
      "tokens_seen": 1150812160
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000657759589151019,
      "loss": 0.0721,
      "theoretical_loss": 3.629746461479079,
      "tokens_seen": 1151074304
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006576793452094368,
      "loss": 0.0714,
      "theoretical_loss": 3.62967043811853,
      "tokens_seen": 1151336448
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006575991012678543,
      "loss": 0.0736,
      "theoretical_loss": 3.629594436910924,
      "tokens_seen": 1151598592
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006575188573262719,
      "loss": 0.0727,
      "theoretical_loss": 3.6295184578447643,
      "tokens_seen": 1151860736
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006574386133846895,
      "loss": 0.0698,
      "theoretical_loss": 3.6294425009085645,
      "tokens_seen": 1152122880
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000657358369443107,
      "loss": 0.073,
      "theoretical_loss": 3.6293665660908454,
      "tokens_seen": 1152385024
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006572781255015246,
      "loss": 0.0718,
      "theoretical_loss": 3.6292906533801372,
      "tokens_seen": 1152647168
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006571978815599422,
      "loss": 0.074,
      "theoretical_loss": 3.6292147627649776,
      "tokens_seen": 1152909312
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006571176376183598,
      "loss": 0.0751,
      "theoretical_loss": 3.6291388942339147,
      "tokens_seen": 1153171456
    },
    {
      "epoch": 0.35,
      "objective/train/advantage_avg": 0.0011869670124724507,
      "objective/train/docs_used": 422447,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4943604469299316,
      "objective/train/original_loss": 1.4943604469299316,
      "objective/train/theoretical_loss": 3.6290630477755026,
      "objective/train/tokens_used": 1173893600,
      "objective/train/value_avg": -0.0102081298828125,
      "objective/train/value_loss": 0.00040764882578514516,
      "objective/train/value_max": -7.31348991394043e-05,
      "objective/train/value_min": -0.70703125,
      "objective/train/value_reward_corr": 0.6485598952753047,
      "objective/train/value_std": 0.0184478759765625,
      "objective/train/weight_avg": 1.0013765096664429,
      "objective/train/weighted_lm_loss": 1.496140718460083,
      "objective/train/weights_max": 1.7507752180099487,
      "objective/train/weights_min": 0.3769771456718445,
      "theoretical_loss": 3.6290630477755026,
      "tokens_seen": 1153433600
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006570373936767774,
      "loss": 0.0714,
      "theoretical_loss": 3.6290630477755026,
      "tokens_seen": 1153433600
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006569571497351951,
      "loss": 0.0716,
      "theoretical_loss": 3.6289872233783065,
      "tokens_seen": 1153695744
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006568769057936126,
      "loss": 0.0704,
      "theoretical_loss": 3.6289114210308977,
      "tokens_seen": 1153957888
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006567966618520302,
      "loss": 0.0727,
      "theoretical_loss": 3.628835640721859,
      "tokens_seen": 1154220032
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006567164179104478,
      "loss": 0.0725,
      "theoretical_loss": 3.6287598824397787,
      "tokens_seen": 1154482176
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006566361739688653,
      "loss": 0.0745,
      "theoretical_loss": 3.6286841461732546,
      "tokens_seen": 1154744320
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000656555930027283,
      "loss": 0.0711,
      "theoretical_loss": 3.6286084319108944,
      "tokens_seen": 1155006464
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006564756860857005,
      "loss": 0.07,
      "theoretical_loss": 3.628532739641312,
      "tokens_seen": 1155268608
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006563954421441181,
      "loss": 0.0699,
      "theoretical_loss": 3.6284570693531317,
      "tokens_seen": 1155530752
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006563151982025357,
      "loss": 0.0715,
      "theoretical_loss": 3.6283814210349847,
      "tokens_seen": 1155792896
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006562349542609534,
      "loss": 0.0707,
      "theoretical_loss": 3.628305794675512,
      "tokens_seen": 1156055040
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006561547103193709,
      "loss": 0.0711,
      "theoretical_loss": 3.6282301902633627,
      "tokens_seen": 1156317184
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006560744663777885,
      "loss": 0.069,
      "theoretical_loss": 3.628154607787194,
      "tokens_seen": 1156579328
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006559942224362061,
      "loss": 0.0703,
      "theoretical_loss": 3.6280790472356705,
      "tokens_seen": 1156841472
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006559139784946236,
      "loss": 0.0698,
      "theoretical_loss": 3.628003508597468,
      "tokens_seen": 1157103616
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006558337345530413,
      "loss": 0.0734,
      "theoretical_loss": 3.6279279918612675,
      "tokens_seen": 1157365760
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006557534906114588,
      "loss": 0.0704,
      "theoretical_loss": 3.6278524970157613,
      "tokens_seen": 1157627904
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006556732466698765,
      "loss": 0.0729,
      "theoretical_loss": 3.6277770240496476,
      "tokens_seen": 1157890048
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000655593002728294,
      "loss": 0.0693,
      "theoretical_loss": 3.6277015729516355,
      "tokens_seen": 1158152192
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006555127587867115,
      "loss": 0.0708,
      "theoretical_loss": 3.6276261437104402,
      "tokens_seen": 1158414336
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006554325148451293,
      "loss": 0.0739,
      "theoretical_loss": 3.6275507363147868,
      "tokens_seen": 1158676480
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006553522709035468,
      "loss": 0.0713,
      "theoretical_loss": 3.6274753507534077,
      "tokens_seen": 1158938624
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006552720269619644,
      "loss": 0.0745,
      "theoretical_loss": 3.6273999870150444,
      "tokens_seen": 1159200768
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000655191783020382,
      "loss": 0.0682,
      "theoretical_loss": 3.627324645088446,
      "tokens_seen": 1159462912
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006551115390787996,
      "loss": 0.0721,
      "theoretical_loss": 3.627249324962371,
      "tokens_seen": 1159725056
    },
    {
      "epoch": 0.35,
      "objective/train/advantage_avg": 0.0011229667579755187,
      "objective/train/docs_used": 424758,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5378754138946533,
      "objective/train/original_loss": 1.5378755331039429,
      "objective/train/theoretical_loss": 3.6271740266255854,
      "objective/train/tokens_used": 1180447200,
      "objective/train/value_avg": -0.0070648193359375,
      "objective/train/value_loss": 0.00018219789490103722,
      "objective/train/value_max": -0.00013446807861328125,
      "objective/train/value_min": -0.4599609375,
      "objective/train/value_reward_corr": 0.6781258515515318,
      "objective/train/value_std": 0.01233673095703125,
      "objective/train/weight_avg": 1.0012083053588867,
      "objective/train/weighted_lm_loss": 1.5396517515182495,
      "objective/train/weights_max": 1.1805914640426636,
      "objective/train/weights_min": 0.37442609667778015,
      "theoretical_loss": 3.6271740266255854,
      "tokens_seen": 1159987200
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006550312951372171,
      "loss": 0.0719,
      "theoretical_loss": 3.6271740266255854,
      "tokens_seen": 1159987200
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006549510511956347,
      "loss": 0.0728,
      "theoretical_loss": 3.6270987500668648,
      "tokens_seen": 1160249344
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006548708072540523,
      "loss": 0.0726,
      "theoretical_loss": 3.6270234952749902,
      "tokens_seen": 1160511488
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006547905633124698,
      "loss": 0.0711,
      "theoretical_loss": 3.6269482622387548,
      "tokens_seen": 1160773632
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006547103193708876,
      "loss": 0.0723,
      "theoretical_loss": 3.6268730509469567,
      "tokens_seen": 1161035776
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006546300754293051,
      "loss": 0.0698,
      "theoretical_loss": 3.626797861388404,
      "tokens_seen": 1161297920
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006545498314877228,
      "loss": 0.073,
      "theoretical_loss": 3.6267226935519132,
      "tokens_seen": 1161560064
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006544695875461403,
      "loss": 0.0723,
      "theoretical_loss": 3.626647547426309,
      "tokens_seen": 1161822208
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006543893436045578,
      "loss": 0.0729,
      "theoretical_loss": 3.6265724230004226,
      "tokens_seen": 1162084352
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006543090996629755,
      "loss": 0.0711,
      "theoretical_loss": 3.6264973202630966,
      "tokens_seen": 1162346496
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000654228855721393,
      "loss": 0.0696,
      "theoretical_loss": 3.6264222392031797,
      "tokens_seen": 1162608640
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006541486117798106,
      "loss": 0.0696,
      "theoretical_loss": 3.6263471798095286,
      "tokens_seen": 1162870784
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006540683678382282,
      "loss": 0.0749,
      "theoretical_loss": 3.6262721420710093,
      "tokens_seen": 1163132928
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006539881238966459,
      "loss": 0.0707,
      "theoretical_loss": 3.6261971259764962,
      "tokens_seen": 1163395072
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006539078799550634,
      "loss": 0.0694,
      "theoretical_loss": 3.626122131514871,
      "tokens_seen": 1163657216
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000653827636013481,
      "loss": 0.0724,
      "theoretical_loss": 3.626047158675024,
      "tokens_seen": 1163919360
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006537473920718986,
      "loss": 0.0701,
      "theoretical_loss": 3.625972207445854,
      "tokens_seen": 1164181504
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006536671481303161,
      "loss": 0.0686,
      "theoretical_loss": 3.625897277816267,
      "tokens_seen": 1164443648
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006535869041887338,
      "loss": 0.0709,
      "theoretical_loss": 3.625822369775179,
      "tokens_seen": 1164705792
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006535066602471513,
      "loss": 0.0731,
      "theoretical_loss": 3.6257474833115113,
      "tokens_seen": 1164967936
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006534264163055689,
      "loss": 0.0726,
      "theoretical_loss": 3.625672618414198,
      "tokens_seen": 1165230080
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006533461723639865,
      "loss": 0.0714,
      "theoretical_loss": 3.6255977750721753,
      "tokens_seen": 1165492224
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006532659284224042,
      "loss": 0.0747,
      "theoretical_loss": 3.6255229532743933,
      "tokens_seen": 1165754368
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006531856844808218,
      "loss": 0.0712,
      "theoretical_loss": 3.625448153009807,
      "tokens_seen": 1166016512
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006531054405392393,
      "loss": 0.0693,
      "theoretical_loss": 3.6253733742673795,
      "tokens_seen": 1166278656
    },
    {
      "epoch": 0.35,
      "objective/train/advantage_avg": 0.002040127059444785,
      "objective/train/docs_used": 427016,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.472293734550476,
      "objective/train/original_loss": 1.4722938537597656,
      "objective/train/theoretical_loss": 3.625298617036084,
      "objective/train/tokens_used": 1187000800,
      "objective/train/value_avg": -0.01036834716796875,
      "objective/train/value_loss": 0.0002555120154283941,
      "objective/train/value_max": -0.00013339519500732422,
      "objective/train/value_min": -0.6376953125,
      "objective/train/value_reward_corr": 0.7320040769195669,
      "objective/train/value_std": 0.0182952880859375,
      "objective/train/weight_avg": 1.0021584033966064,
      "objective/train/weighted_lm_loss": 1.475948452949524,
      "objective/train/weights_max": 1.3658320903778076,
      "objective/train/weights_min": 0.3710109293460846,
      "theoretical_loss": 3.625298617036084,
      "tokens_seen": 1166540800
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006530251965976569,
      "loss": 0.0744,
      "theoretical_loss": 3.625298617036084,
      "tokens_seen": 1166540800
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006529449526560745,
      "loss": 0.0727,
      "theoretical_loss": 3.6252238813049,
      "tokens_seen": 1166802944
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006528647087144921,
      "loss": 0.073,
      "theoretical_loss": 3.6251491670628155,
      "tokens_seen": 1167065088
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006527844647729096,
      "loss": 0.071,
      "theoretical_loss": 3.6250744742988275,
      "tokens_seen": 1167327232
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006527042208313273,
      "loss": 0.0727,
      "theoretical_loss": 3.6249998030019404,
      "tokens_seen": 1167589376
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006526239768897448,
      "loss": 0.0707,
      "theoretical_loss": 3.6249251531611666,
      "tokens_seen": 1167851520
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006525437329481623,
      "loss": 0.0715,
      "theoretical_loss": 3.6248505247655265,
      "tokens_seen": 1168113664
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006524634890065801,
      "loss": 0.0731,
      "theoretical_loss": 3.6247759178040493,
      "tokens_seen": 1168375808
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006523832450649976,
      "loss": 0.0703,
      "theoretical_loss": 3.624701332265772,
      "tokens_seen": 1168637952
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006523030011234152,
      "loss": 0.0703,
      "theoretical_loss": 3.6246267681397386,
      "tokens_seen": 1168900096
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006522227571818328,
      "loss": 0.0715,
      "theoretical_loss": 3.624552225415003,
      "tokens_seen": 1169162240
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006521425132402504,
      "loss": 0.0691,
      "theoretical_loss": 3.6244777040806255,
      "tokens_seen": 1169424384
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006520622692986679,
      "loss": 0.0722,
      "theoretical_loss": 3.624403204125676,
      "tokens_seen": 1169686528
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006519820253570855,
      "loss": 0.0748,
      "theoretical_loss": 3.6243287255392307,
      "tokens_seen": 1169948672
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006519017814155031,
      "loss": 0.0722,
      "theoretical_loss": 3.624254268310375,
      "tokens_seen": 1170210816
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006518215374739207,
      "loss": 0.0697,
      "theoretical_loss": 3.6241798324282017,
      "tokens_seen": 1170472960
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006517412935323384,
      "loss": 0.0733,
      "theoretical_loss": 3.624105417881813,
      "tokens_seen": 1170735104
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006516610495907559,
      "loss": 0.0729,
      "theoretical_loss": 3.624031024660317,
      "tokens_seen": 1170997248
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0006515808056491736,
      "loss": 0.0734,
      "theoretical_loss": 3.6239566527528306,
      "tokens_seen": 1171259392
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006515005617075911,
      "loss": 0.0714,
      "theoretical_loss": 3.6238823021484796,
      "tokens_seen": 1171521536
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006514203177660086,
      "loss": 0.0728,
      "theoretical_loss": 3.6238079728363974,
      "tokens_seen": 1171783680
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006513400738244263,
      "loss": 0.0711,
      "theoretical_loss": 3.6237336648057243,
      "tokens_seen": 1172045824
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006512598298828438,
      "loss": 0.0691,
      "theoretical_loss": 3.62365937804561,
      "tokens_seen": 1172307968
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006511795859412614,
      "loss": 0.0682,
      "theoretical_loss": 3.6235851125452108,
      "tokens_seen": 1172570112
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000651099341999679,
      "loss": 0.072,
      "theoretical_loss": 3.6235108682936916,
      "tokens_seen": 1172832256
    },
    {
      "epoch": 0.36,
      "objective/train/advantage_avg": 0.0011571807553991675,
      "objective/train/docs_used": 429267,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3518650531768799,
      "objective/train/original_loss": 1.3518651723861694,
      "objective/train/theoretical_loss": 3.623436645280226,
      "objective/train/tokens_used": 1193554400,
      "objective/train/value_avg": -0.00867462158203125,
      "objective/train/value_loss": 0.00022238181554712355,
      "objective/train/value_max": -8.612871170043945e-05,
      "objective/train/value_min": -0.67236328125,
      "objective/train/value_reward_corr": 0.6452302653535441,
      "objective/train/value_std": 0.014129638671875,
      "objective/train/weight_avg": 1.0012578964233398,
      "objective/train/weighted_lm_loss": 1.3529399633407593,
      "objective/train/weights_max": 1.2395051717758179,
      "objective/train/weights_min": 0.37000179290771484,
      "theoretical_loss": 3.623436645280226,
      "tokens_seen": 1173094400
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006510190980580967,
      "loss": 0.0723,
      "theoretical_loss": 3.623436645280226,
      "tokens_seen": 1173094400
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006509388541165142,
      "loss": 0.0724,
      "theoretical_loss": 3.6233624434939946,
      "tokens_seen": 1173356544
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006508586101749318,
      "loss": 0.07,
      "theoretical_loss": 3.623288262924186,
      "tokens_seen": 1173618688
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006507783662333494,
      "loss": 0.0738,
      "theoretical_loss": 3.623214103559997,
      "tokens_seen": 1173880832
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000650698122291767,
      "loss": 0.0719,
      "theoretical_loss": 3.6231399653906315,
      "tokens_seen": 1174142976
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006506178783501846,
      "loss": 0.0686,
      "theoretical_loss": 3.623065848405303,
      "tokens_seen": 1174405120
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006505376344086021,
      "loss": 0.0717,
      "theoretical_loss": 3.622991752593231,
      "tokens_seen": 1174667264
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006504573904670198,
      "loss": 0.0724,
      "theoretical_loss": 3.6229176779436445,
      "tokens_seen": 1174929408
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006503771465254373,
      "loss": 0.0705,
      "theoretical_loss": 3.6228436244457796,
      "tokens_seen": 1175191552
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000650296902583855,
      "loss": 0.0719,
      "theoretical_loss": 3.6227695920888796,
      "tokens_seen": 1175453696
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006502166586422726,
      "loss": 0.0703,
      "theoretical_loss": 3.622695580862197,
      "tokens_seen": 1175715840
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006501364147006901,
      "loss": 0.0703,
      "theoretical_loss": 3.6226215907549912,
      "tokens_seen": 1175977984
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006500561707591077,
      "loss": 0.0717,
      "theoretical_loss": 3.6225476217565307,
      "tokens_seen": 1176240128
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006499759268175253,
      "loss": 0.0711,
      "theoretical_loss": 3.6224736738560894,
      "tokens_seen": 1176502272
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006498956828759429,
      "loss": 0.0754,
      "theoretical_loss": 3.6223997470429516,
      "tokens_seen": 1176764416
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006498154389343604,
      "loss": 0.0713,
      "theoretical_loss": 3.6223258413064086,
      "tokens_seen": 1177026560
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006497351949927781,
      "loss": 0.0733,
      "theoretical_loss": 3.6222519566357585,
      "tokens_seen": 1177288704
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006496549510511956,
      "loss": 0.071,
      "theoretical_loss": 3.6221780930203096,
      "tokens_seen": 1177550848
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006495747071096131,
      "loss": 0.073,
      "theoretical_loss": 3.6221042504493743,
      "tokens_seen": 1177812992
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006494944631680309,
      "loss": 0.0736,
      "theoretical_loss": 3.622030428912276,
      "tokens_seen": 1178075136
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006494142192264484,
      "loss": 0.0699,
      "theoretical_loss": 3.6219566283983458,
      "tokens_seen": 1178337280
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006493339752848661,
      "loss": 0.0747,
      "theoretical_loss": 3.6218828488969197,
      "tokens_seen": 1178599424
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006492537313432836,
      "loss": 0.0707,
      "theoretical_loss": 3.6218090903973446,
      "tokens_seen": 1178861568
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006491734874017012,
      "loss": 0.0765,
      "theoretical_loss": 3.621735352888974,
      "tokens_seen": 1179123712
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006490932434601188,
      "loss": 0.0787,
      "theoretical_loss": 3.621661636361169,
      "tokens_seen": 1179385856
    },
    {
      "epoch": 0.36,
      "objective/train/advantage_avg": -0.000662869424559176,
      "objective/train/docs_used": 431781,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5813153982162476,
      "objective/train/original_loss": 1.5813153982162476,
      "objective/train/theoretical_loss": 3.6215879408032987,
      "objective/train/tokens_used": 1200108000,
      "objective/train/value_avg": -0.00632476806640625,
      "objective/train/value_loss": 0.0001973478210857138,
      "objective/train/value_max": -7.545948028564453e-05,
      "objective/train/value_min": -0.392822265625,
      "objective/train/value_reward_corr": 0.6522775297311824,
      "objective/train/value_std": 0.01153564453125,
      "objective/train/weight_avg": 0.9994257688522339,
      "objective/train/weighted_lm_loss": 1.5801219940185547,
      "objective/train/weights_max": 1.2123867273330688,
      "objective/train/weights_min": 0.3755246698856354,
      "theoretical_loss": 3.6215879408032987,
      "tokens_seen": 1179648000
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006490129995185363,
      "loss": 0.0741,
      "theoretical_loss": 3.6215879408032987,
      "tokens_seen": 1179648000
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006489327555769539,
      "loss": 0.0739,
      "theoretical_loss": 3.6215142662047395,
      "tokens_seen": 1179910144
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006488525116353715,
      "loss": 0.074,
      "theoretical_loss": 3.621440612554876,
      "tokens_seen": 1180172288
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006487722676937892,
      "loss": 0.0739,
      "theoretical_loss": 3.6213669798431005,
      "tokens_seen": 1180434432
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006486920237522067,
      "loss": 0.0723,
      "theoretical_loss": 3.621293368058813,
      "tokens_seen": 1180696576
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006486117798106244,
      "loss": 0.074,
      "theoretical_loss": 3.621219777191421,
      "tokens_seen": 1180958720
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006485315358690419,
      "loss": 0.0744,
      "theoretical_loss": 3.62114620723034,
      "tokens_seen": 1181220864
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006484512919274594,
      "loss": 0.0737,
      "theoretical_loss": 3.621072658164993,
      "tokens_seen": 1181483008
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006483710479858771,
      "loss": 0.0735,
      "theoretical_loss": 3.6209991299848108,
      "tokens_seen": 1181745152
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006482908040442946,
      "loss": 0.0752,
      "theoretical_loss": 3.620925622679232,
      "tokens_seen": 1182007296
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006482105601027123,
      "loss": 0.074,
      "theoretical_loss": 3.620852136237702,
      "tokens_seen": 1182269440
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006481303161611298,
      "loss": 0.074,
      "theoretical_loss": 3.620778670649676,
      "tokens_seen": 1182531584
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006480500722195475,
      "loss": 0.0709,
      "theoretical_loss": 3.620705225904614,
      "tokens_seen": 1182793728
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006479698282779651,
      "loss": 0.0733,
      "theoretical_loss": 3.620631801991987,
      "tokens_seen": 1183055872
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006478895843363826,
      "loss": 0.075,
      "theoretical_loss": 3.6205583989012697,
      "tokens_seen": 1183318016
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006478093403948002,
      "loss": 0.0748,
      "theoretical_loss": 3.6204850166219478,
      "tokens_seen": 1183580160
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006477290964532178,
      "loss": 0.0699,
      "theoretical_loss": 3.6204116551435126,
      "tokens_seen": 1183842304
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006476488525116354,
      "loss": 0.0745,
      "theoretical_loss": 3.620338314455465,
      "tokens_seen": 1184104448
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006475686085700529,
      "loss": 0.0742,
      "theoretical_loss": 3.6202649945473113,
      "tokens_seen": 1184366592
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006474883646284706,
      "loss": 0.0724,
      "theoretical_loss": 3.6201916954085664,
      "tokens_seen": 1184628736
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006474081206868881,
      "loss": 0.0725,
      "theoretical_loss": 3.620118417028754,
      "tokens_seen": 1184890880
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006473278767453056,
      "loss": 0.0735,
      "theoretical_loss": 3.620045159397403,
      "tokens_seen": 1185153024
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006472476328037234,
      "loss": 0.0741,
      "theoretical_loss": 3.619971922504052,
      "tokens_seen": 1185415168
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006471673888621409,
      "loss": 0.0711,
      "theoretical_loss": 3.6198987063382457,
      "tokens_seen": 1185677312
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006470871449205585,
      "loss": 0.0757,
      "theoretical_loss": 3.6198255108895374,
      "tokens_seen": 1185939456
    },
    {
      "epoch": 0.36,
      "objective/train/advantage_avg": 0.0009745230199769139,
      "objective/train/docs_used": 434243,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5072544813156128,
      "objective/train/original_loss": 1.5072546005249023,
      "objective/train/theoretical_loss": 3.6197523361474877,
      "objective/train/tokens_used": 1206661600,
      "objective/train/value_avg": -0.00830841064453125,
      "objective/train/value_loss": 0.00030490392236970365,
      "objective/train/value_max": -9.763240814208984e-05,
      "objective/train/value_min": -0.658203125,
      "objective/train/value_reward_corr": 0.6350907920593141,
      "objective/train/value_std": 0.0146636962890625,
      "objective/train/weight_avg": 1.0011084079742432,
      "objective/train/weighted_lm_loss": 1.509034514427185,
      "objective/train/weights_max": 1.5940988063812256,
      "objective/train/weights_min": 0.37102511525154114,
      "theoretical_loss": 3.6197523361474877,
      "tokens_seen": 1186201600
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006470069009789761,
      "loss": 0.0753,
      "theoretical_loss": 3.6197523361474877,
      "tokens_seen": 1186201600
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006469266570373937,
      "loss": 0.0723,
      "theoretical_loss": 3.619679182101664,
      "tokens_seen": 1186463744
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006468464130958113,
      "loss": 0.0766,
      "theoretical_loss": 3.619606048741643,
      "tokens_seen": 1186725888
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006467661691542289,
      "loss": 0.0751,
      "theoretical_loss": 3.6195329360570065,
      "tokens_seen": 1186988032
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006466859252126464,
      "loss": 0.0743,
      "theoretical_loss": 3.6194598440373467,
      "tokens_seen": 1187250176
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000646605681271064,
      "loss": 0.072,
      "theoretical_loss": 3.619386772672261,
      "tokens_seen": 1187512320
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006465254373294817,
      "loss": 0.0766,
      "theoretical_loss": 3.6193137219513556,
      "tokens_seen": 1187774464
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006464451933878992,
      "loss": 0.0759,
      "theoretical_loss": 3.619240691864243,
      "tokens_seen": 1188036608
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006463649494463169,
      "loss": 0.0709,
      "theoretical_loss": 3.619167682400545,
      "tokens_seen": 1188298752
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006462847055047344,
      "loss": 0.0769,
      "theoretical_loss": 3.619094693549889,
      "tokens_seen": 1188560896
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000646204461563152,
      "loss": 0.0729,
      "theoretical_loss": 3.6190217253019124,
      "tokens_seen": 1188823040
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006461242176215696,
      "loss": 0.0747,
      "theoretical_loss": 3.6189487776462568,
      "tokens_seen": 1189085184
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006460439736799871,
      "loss": 0.0706,
      "theoretical_loss": 3.6188758505725738,
      "tokens_seen": 1189347328
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006459637297384047,
      "loss": 0.0767,
      "theoretical_loss": 3.618802944070522,
      "tokens_seen": 1189609472
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006458834857968223,
      "loss": 0.0741,
      "theoretical_loss": 3.618730058129766,
      "tokens_seen": 1189871616
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00064580324185524,
      "loss": 0.0745,
      "theoretical_loss": 3.61865719273998,
      "tokens_seen": 1190133760
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006457229979136575,
      "loss": 0.0736,
      "theoretical_loss": 3.6185843478908453,
      "tokens_seen": 1190395904
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006456427539720752,
      "loss": 0.0736,
      "theoretical_loss": 3.618511523572049,
      "tokens_seen": 1190658048
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006455625100304927,
      "loss": 0.0728,
      "theoretical_loss": 3.6184387197732875,
      "tokens_seen": 1190920192
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006454822660889103,
      "loss": 0.0775,
      "theoretical_loss": 3.6183659364842624,
      "tokens_seen": 1191182336
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006454020221473279,
      "loss": 0.0742,
      "theoretical_loss": 3.6182931736946857,
      "tokens_seen": 1191444480
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006453217782057454,
      "loss": 0.0723,
      "theoretical_loss": 3.618220431394274,
      "tokens_seen": 1191706624
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006452415342641631,
      "loss": 0.0704,
      "theoretical_loss": 3.618147709572754,
      "tokens_seen": 1191968768
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006451612903225806,
      "loss": 0.0738,
      "theoretical_loss": 3.618075008219858,
      "tokens_seen": 1192230912
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006450810463809983,
      "loss": 0.0715,
      "theoretical_loss": 3.6180023273253252,
      "tokens_seen": 1192493056
    },
    {
      "epoch": 0.36,
      "objective/train/advantage_avg": 0.001384525210596621,
      "objective/train/docs_used": 436657,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4595648050308228,
      "objective/train/original_loss": 1.4595648050308228,
      "objective/train/theoretical_loss": 3.6179296668789043,
      "objective/train/tokens_used": 1213215200,
      "objective/train/value_avg": -0.01062774658203125,
      "objective/train/value_loss": 0.00031874870182946324,
      "objective/train/value_max": -0.00014317035675048828,
      "objective/train/value_min": -0.7646484375,
      "objective/train/value_reward_corr": 0.6390448707495002,
      "objective/train/value_std": 0.0160675048828125,
      "objective/train/weight_avg": 1.0015238523483276,
      "objective/train/weighted_lm_loss": 1.4609448909759521,
      "objective/train/weights_max": 1.3013832569122314,
      "objective/train/weights_min": 0.3687829375267029,
      "theoretical_loss": 3.6179296668789043,
      "tokens_seen": 1192755200
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006450008024394159,
      "loss": 0.0753,
      "theoretical_loss": 3.6179296668789043,
      "tokens_seen": 1192755200
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006449205584978334,
      "loss": 0.0736,
      "theoretical_loss": 3.6178570268703494,
      "tokens_seen": 1193017344
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000644840314556251,
      "loss": 0.0733,
      "theoretical_loss": 3.617784407289424,
      "tokens_seen": 1193279488
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006447600706146686,
      "loss": 0.0747,
      "theoretical_loss": 3.617711808125896,
      "tokens_seen": 1193541632
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006446798266730862,
      "loss": 0.0751,
      "theoretical_loss": 3.6176392293695434,
      "tokens_seen": 1193803776
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006445995827315037,
      "loss": 0.0753,
      "theoretical_loss": 3.6175666710101506,
      "tokens_seen": 1194065920
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006445193387899214,
      "loss": 0.0733,
      "theoretical_loss": 3.6174941330375097,
      "tokens_seen": 1194328064
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006444390948483389,
      "loss": 0.0753,
      "theoretical_loss": 3.617421615441419,
      "tokens_seen": 1194590208
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006443588509067564,
      "loss": 0.0777,
      "theoretical_loss": 3.617349118211685,
      "tokens_seen": 1194852352
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006442786069651742,
      "loss": 0.0731,
      "theoretical_loss": 3.6172766413381225,
      "tokens_seen": 1195114496
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006441983630235917,
      "loss": 0.0704,
      "theoretical_loss": 3.617204184810552,
      "tokens_seen": 1195376640
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006441181190820094,
      "loss": 0.074,
      "theoretical_loss": 3.6171317486188013,
      "tokens_seen": 1195638784
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006440378751404269,
      "loss": 0.0744,
      "theoretical_loss": 3.6170593327527074,
      "tokens_seen": 1195900928
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006439576311988445,
      "loss": 0.0725,
      "theoretical_loss": 3.616986937202112,
      "tokens_seen": 1196163072
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006438773872572621,
      "loss": 0.0745,
      "theoretical_loss": 3.616914561956867,
      "tokens_seen": 1196425216
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006437971433156797,
      "loss": 0.0744,
      "theoretical_loss": 3.6168422070068287,
      "tokens_seen": 1196687360
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006437168993740972,
      "loss": 0.0729,
      "theoretical_loss": 3.6167698723418624,
      "tokens_seen": 1196949504
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006436366554325149,
      "loss": 0.076,
      "theoretical_loss": 3.6166975579518406,
      "tokens_seen": 1197211648
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006435564114909325,
      "loss": 0.0729,
      "theoretical_loss": 3.6166252638266423,
      "tokens_seen": 1197473792
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00064347616754935,
      "loss": 0.0746,
      "theoretical_loss": 3.616552989956155,
      "tokens_seen": 1197735936
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006433959236077677,
      "loss": 0.0745,
      "theoretical_loss": 3.616480736330272,
      "tokens_seen": 1197998080
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006433156796661852,
      "loss": 0.0709,
      "theoretical_loss": 3.616408502938895,
      "tokens_seen": 1198260224
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006432354357246028,
      "loss": 0.07,
      "theoretical_loss": 3.616336289771932,
      "tokens_seen": 1198522368
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006431551917830204,
      "loss": 0.0712,
      "theoretical_loss": 3.6162640968192994,
      "tokens_seen": 1198784512
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006430749478414379,
      "loss": 0.0726,
      "theoretical_loss": 3.61619192407092,
      "tokens_seen": 1199046656
    },
    {
      "epoch": 0.36,
      "objective/train/advantage_avg": 0.0016890951665118337,
      "objective/train/docs_used": 438488,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4938664436340332,
      "objective/train/original_loss": 1.493866205215454,
      "objective/train/theoretical_loss": 3.6161197715167237,
      "objective/train/tokens_used": 1219768800,
      "objective/train/value_avg": -0.00862884521484375,
      "objective/train/value_loss": 0.0003480366722214967,
      "objective/train/value_max": -0.00011146068572998047,
      "objective/train/value_min": -0.95068359375,
      "objective/train/value_reward_corr": 0.7120363623901087,
      "objective/train/value_std": 0.0188140869140625,
      "objective/train/weight_avg": 1.0018378496170044,
      "objective/train/weighted_lm_loss": 1.4969854354858398,
      "objective/train/weights_max": 1.3778945207595825,
      "objective/train/weights_min": 0.22749631106853485,
      "theoretical_loss": 3.6161197715167237,
      "tokens_seen": 1199308800
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006429947038998556,
      "loss": 0.072,
      "theoretical_loss": 3.6161197715167237,
      "tokens_seen": 1199308800
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006429144599582731,
      "loss": 0.0718,
      "theoretical_loss": 3.616047639146648,
      "tokens_seen": 1199570944
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006428342160166908,
      "loss": 0.0716,
      "theoretical_loss": 3.6159755269506375,
      "tokens_seen": 1199833088
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006427539720751084,
      "loss": 0.0734,
      "theoretical_loss": 3.6159034349186445,
      "tokens_seen": 1200095232
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000642673728133526,
      "loss": 0.0759,
      "theoretical_loss": 3.615831363040628,
      "tokens_seen": 1200357376
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006425934841919435,
      "loss": 0.0709,
      "theoretical_loss": 3.615759311306553,
      "tokens_seen": 1200619520
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006425132402503611,
      "loss": 0.0715,
      "theoretical_loss": 3.6156872797063944,
      "tokens_seen": 1200881664
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006424329963087787,
      "loss": 0.0745,
      "theoretical_loss": 3.6156152682301324,
      "tokens_seen": 1201143808
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006423527523671962,
      "loss": 0.0752,
      "theoretical_loss": 3.6155432768677542,
      "tokens_seen": 1201405952
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006422725084256139,
      "loss": 0.0726,
      "theoretical_loss": 3.615471305609255,
      "tokens_seen": 1201668096
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006421922644840314,
      "loss": 0.0736,
      "theoretical_loss": 3.6153993544446372,
      "tokens_seen": 1201930240
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006421120205424491,
      "loss": 0.073,
      "theoretical_loss": 3.61532742336391,
      "tokens_seen": 1202192384
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006420317766008667,
      "loss": 0.0709,
      "theoretical_loss": 3.6152555123570895,
      "tokens_seen": 1202454528
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006419515326592842,
      "loss": 0.0685,
      "theoretical_loss": 3.6151836214141992,
      "tokens_seen": 1202716672
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006418712887177018,
      "loss": 0.0737,
      "theoretical_loss": 3.61511175052527,
      "tokens_seen": 1202978816
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006417910447761194,
      "loss": 0.0711,
      "theoretical_loss": 3.6150398996803395,
      "tokens_seen": 1203240960
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000641710800834537,
      "loss": 0.0732,
      "theoretical_loss": 3.6149680688694525,
      "tokens_seen": 1203503104
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006416305568929546,
      "loss": 0.0737,
      "theoretical_loss": 3.614896258082661,
      "tokens_seen": 1203765248
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006415503129513722,
      "loss": 0.071,
      "theoretical_loss": 3.614824467310025,
      "tokens_seen": 1204027392
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0006414700690097897,
      "loss": 0.0743,
      "theoretical_loss": 3.614752696541609,
      "tokens_seen": 1204289536
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006413898250682074,
      "loss": 0.0732,
      "theoretical_loss": 3.614680945767488,
      "tokens_seen": 1204551680
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000641309581126625,
      "loss": 0.0728,
      "theoretical_loss": 3.614609214977741,
      "tokens_seen": 1204813824
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006412293371850425,
      "loss": 0.0751,
      "theoretical_loss": 3.614537504162457,
      "tokens_seen": 1205075968
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006411490932434602,
      "loss": 0.0736,
      "theoretical_loss": 3.614465813311729,
      "tokens_seen": 1205338112
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006410688493018777,
      "loss": 0.0736,
      "theoretical_loss": 3.6143941424156596,
      "tokens_seen": 1205600256
    },
    {
      "epoch": 0.37,
      "objective/train/advantage_avg": 0.0002445022400934249,
      "objective/train/docs_used": 441116,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4677330255508423,
      "objective/train/original_loss": 1.4677330255508423,
      "objective/train/theoretical_loss": 3.614322491464357,
      "objective/train/tokens_used": 1226322400,
      "objective/train/value_avg": -0.01202392578125,
      "objective/train/value_loss": 0.0005131938378326595,
      "objective/train/value_max": -9.608268737792969e-05,
      "objective/train/value_min": -0.91015625,
      "objective/train/value_reward_corr": 0.7615977591679253,
      "objective/train/value_std": 0.025634765625,
      "objective/train/weight_avg": 1.0004706382751465,
      "objective/train/weighted_lm_loss": 1.4688076972961426,
      "objective/train/weights_max": 1.7541979551315308,
      "objective/train/weights_min": 0.3682364523410797,
      "theoretical_loss": 3.614322491464357,
      "tokens_seen": 1205862400
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006409886053602953,
      "loss": 0.0725,
      "theoretical_loss": 3.614322491464357,
      "tokens_seen": 1205862400
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006409083614187129,
      "loss": 0.0761,
      "theoretical_loss": 3.614250860447936,
      "tokens_seen": 1206124544
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006408281174771304,
      "loss": 0.0747,
      "theoretical_loss": 3.6141792493565212,
      "tokens_seen": 1206386688
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000640747873535548,
      "loss": 0.071,
      "theoretical_loss": 3.6141076581802416,
      "tokens_seen": 1206648832
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006406676295939656,
      "loss": 0.0718,
      "theoretical_loss": 3.614036086909234,
      "tokens_seen": 1206910976
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006405873856523833,
      "loss": 0.0748,
      "theoretical_loss": 3.6139645355336425,
      "tokens_seen": 1207173120
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006405071417108009,
      "loss": 0.076,
      "theoretical_loss": 3.613893004043617,
      "tokens_seen": 1207435264
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006404268977692185,
      "loss": 0.0723,
      "theoretical_loss": 3.6138214924293166,
      "tokens_seen": 1207697408
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000640346653827636,
      "loss": 0.0771,
      "theoretical_loss": 3.6137500006809056,
      "tokens_seen": 1207959552
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006402664098860537,
      "loss": 0.074,
      "theoretical_loss": 3.6136785287885553,
      "tokens_seen": 1208221696
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006401861659444712,
      "loss": 0.0736,
      "theoretical_loss": 3.6136070767424457,
      "tokens_seen": 1208483840
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006401059220028887,
      "loss": 0.0746,
      "theoretical_loss": 3.6135356445327624,
      "tokens_seen": 1208745984
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006400256780613064,
      "loss": 0.0692,
      "theoretical_loss": 3.6134642321496977,
      "tokens_seen": 1209008128
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000639945434119724,
      "loss": 0.0706,
      "theoretical_loss": 3.613392839583452,
      "tokens_seen": 1209270272
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006398651901781416,
      "loss": 0.0741,
      "theoretical_loss": 3.6133214668242317,
      "tokens_seen": 1209532416
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006397849462365592,
      "loss": 0.0709,
      "theoretical_loss": 3.613250113862251,
      "tokens_seen": 1209794560
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006397047022949768,
      "loss": 0.0724,
      "theoretical_loss": 3.61317878068773,
      "tokens_seen": 1210056704
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006396244583533943,
      "loss": 0.0715,
      "theoretical_loss": 3.6131074672908965,
      "tokens_seen": 1210318848
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006395442144118119,
      "loss": 0.0715,
      "theoretical_loss": 3.6130361736619854,
      "tokens_seen": 1210580992
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006394639704702295,
      "loss": 0.0739,
      "theoretical_loss": 3.6129648997912382,
      "tokens_seen": 1210843136
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000639383726528647,
      "loss": 0.0723,
      "theoretical_loss": 3.6128936456689034,
      "tokens_seen": 1211105280
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006393034825870647,
      "loss": 0.0729,
      "theoretical_loss": 3.612822411285236,
      "tokens_seen": 1211367424
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006392232386454822,
      "loss": 0.0755,
      "theoretical_loss": 3.612751196630499,
      "tokens_seen": 1211629568
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006391429947039,
      "loss": 0.0725,
      "theoretical_loss": 3.61268000169496,
      "tokens_seen": 1211891712
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006390627507623175,
      "loss": 0.0742,
      "theoretical_loss": 3.612608826468897,
      "tokens_seen": 1212153856
    },
    {
      "epoch": 0.37,
      "objective/train/advantage_avg": -0.0005174296093173325,
      "objective/train/docs_used": 443456,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.490218162536621,
      "objective/train/original_loss": 1.490218162536621,
      "objective/train/theoretical_loss": 3.6125376709425923,
      "objective/train/tokens_used": 1232876000,
      "objective/train/value_avg": -0.008087158203125,
      "objective/train/value_loss": 0.00029237871058285236,
      "objective/train/value_max": -7.599592208862305e-05,
      "objective/train/value_min": -0.3349609375,
      "objective/train/value_reward_corr": 0.6280084039572571,
      "objective/train/value_std": 0.01226806640625,
      "objective/train/weight_avg": 0.999613881111145,
      "objective/train/weighted_lm_loss": 1.4892261028289795,
      "objective/train/weights_max": 1.3555288314819336,
      "objective/train/weights_min": 0.3726279139518738,
      "theoretical_loss": 3.6125376709425923,
      "tokens_seen": 1212416000
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000638982506820735,
      "loss": 0.0738,
      "theoretical_loss": 3.6125376709425923,
      "tokens_seen": 1212416000
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006389022628791527,
      "loss": 0.0683,
      "theoretical_loss": 3.6124665351063356,
      "tokens_seen": 1212678144
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006388220189375702,
      "loss": 0.0737,
      "theoretical_loss": 3.612395418950424,
      "tokens_seen": 1212940288
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006387417749959878,
      "loss": 0.0741,
      "theoretical_loss": 3.6123243224651604,
      "tokens_seen": 1213202432
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006386615310544054,
      "loss": 0.0719,
      "theoretical_loss": 3.612253245640856,
      "tokens_seen": 1213464576
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000638581287112823,
      "loss": 0.0716,
      "theoretical_loss": 3.6121821884678287,
      "tokens_seen": 1213726720
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006385010431712405,
      "loss": 0.0742,
      "theoretical_loss": 3.6121111509364017,
      "tokens_seen": 1213988864
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006384207992296582,
      "loss": 0.0751,
      "theoretical_loss": 3.6120401330369067,
      "tokens_seen": 1214251008
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006383405552880758,
      "loss": 0.0718,
      "theoretical_loss": 3.6119691347596814,
      "tokens_seen": 1214513152
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006382603113464933,
      "loss": 0.0739,
      "theoretical_loss": 3.611898156095071,
      "tokens_seen": 1214775296
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000638180067404911,
      "loss": 0.0736,
      "theoretical_loss": 3.6118271970334264,
      "tokens_seen": 1215037440
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006380998234633285,
      "loss": 0.0722,
      "theoretical_loss": 3.6117562575651068,
      "tokens_seen": 1215299584
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006380195795217462,
      "loss": 0.0728,
      "theoretical_loss": 3.611685337680477,
      "tokens_seen": 1215561728
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006379393355801637,
      "loss": 0.072,
      "theoretical_loss": 3.6116144373699086,
      "tokens_seen": 1215823872
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006378590916385812,
      "loss": 0.0729,
      "theoretical_loss": 3.611543556623782,
      "tokens_seen": 1216086016
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006377788476969989,
      "loss": 0.0728,
      "theoretical_loss": 3.6114726954324814,
      "tokens_seen": 1216348160
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006376986037554164,
      "loss": 0.0723,
      "theoretical_loss": 3.6114018537864,
      "tokens_seen": 1216610304
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006376183598138341,
      "loss": 0.0724,
      "theoretical_loss": 3.6113310316759373,
      "tokens_seen": 1216872448
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006375381158722517,
      "loss": 0.0691,
      "theoretical_loss": 3.6112602290914984,
      "tokens_seen": 1217134592
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006374578719306693,
      "loss": 0.0716,
      "theoretical_loss": 3.611189446023497,
      "tokens_seen": 1217396736
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006373776279890868,
      "loss": 0.0725,
      "theoretical_loss": 3.6111186824623527,
      "tokens_seen": 1217658880
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006372973840475045,
      "loss": 0.0737,
      "theoretical_loss": 3.6110479383984915,
      "tokens_seen": 1217921024
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000637217140105922,
      "loss": 0.0729,
      "theoretical_loss": 3.6109772138223466,
      "tokens_seen": 1218183168
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006371368961643395,
      "loss": 0.0735,
      "theoretical_loss": 3.6109065087243577,
      "tokens_seen": 1218445312
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006370566522227572,
      "loss": 0.0715,
      "theoretical_loss": 3.610835823094972,
      "tokens_seen": 1218707456
    },
    {
      "epoch": 0.37,
      "objective/train/advantage_avg": 0.0005563150043599308,
      "objective/train/docs_used": 445814,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4974902868270874,
      "objective/train/original_loss": 1.4974902868270874,
      "objective/train/theoretical_loss": 3.6107651569246424,
      "objective/train/tokens_used": 1239429600,
      "objective/train/value_avg": -0.0152435302734375,
      "objective/train/value_loss": 0.0005672244587913156,
      "objective/train/value_max": -0.0001055598258972168,
      "objective/train/value_min": -0.96044921875,
      "objective/train/value_reward_corr": 0.7760111785211823,
      "objective/train/value_std": 0.0281982421875,
      "objective/train/weight_avg": 1.0008221864700317,
      "objective/train/weighted_lm_loss": 1.4987361431121826,
      "objective/train/weights_max": 1.8249680995941162,
      "objective/train/weights_min": 0.3728724420070648,
      "theoretical_loss": 3.6107651569246424,
      "tokens_seen": 1218969600
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006369764082811747,
      "loss": 0.073,
      "theoretical_loss": 3.6107651569246424,
      "tokens_seen": 1218969600
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006368961643395924,
      "loss": 0.0712,
      "theoretical_loss": 3.610694510203829,
      "tokens_seen": 1219231744
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00063681592039801,
      "loss": 0.0716,
      "theoretical_loss": 3.610623882922999,
      "tokens_seen": 1219493888
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006367356764564276,
      "loss": 0.0716,
      "theoretical_loss": 3.6105532750726255,
      "tokens_seen": 1219756032
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006366554325148452,
      "loss": 0.0717,
      "theoretical_loss": 3.6104826866431887,
      "tokens_seen": 1220018176
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006365751885732627,
      "loss": 0.0717,
      "theoretical_loss": 3.6104121176251764,
      "tokens_seen": 1220280320
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006364949446316803,
      "loss": 0.0723,
      "theoretical_loss": 3.6103415680090816,
      "tokens_seen": 1220542464
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006364147006900979,
      "loss": 0.0739,
      "theoretical_loss": 3.6102710377854046,
      "tokens_seen": 1220804608
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006363344567485155,
      "loss": 0.0704,
      "theoretical_loss": 3.610200526944652,
      "tokens_seen": 1221066752
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000636254212806933,
      "loss": 0.0721,
      "theoretical_loss": 3.610130035477339,
      "tokens_seen": 1221328896
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006361739688653508,
      "loss": 0.0752,
      "theoretical_loss": 3.6100595633739854,
      "tokens_seen": 1221591040
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006360937249237683,
      "loss": 0.0727,
      "theoretical_loss": 3.6099891106251176,
      "tokens_seen": 1221853184
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006360134809821858,
      "loss": 0.0746,
      "theoretical_loss": 3.6099186772212697,
      "tokens_seen": 1222115328
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006359332370406035,
      "loss": 0.0709,
      "theoretical_loss": 3.6098482631529825,
      "tokens_seen": 1222377472
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000635852993099021,
      "loss": 0.0727,
      "theoretical_loss": 3.6097778684108026,
      "tokens_seen": 1222639616
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006357727491574386,
      "loss": 0.0716,
      "theoretical_loss": 3.609707492985284,
      "tokens_seen": 1222901760
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006356925052158562,
      "loss": 0.0714,
      "theoretical_loss": 3.609637136866987,
      "tokens_seen": 1223163904
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006356122612742738,
      "loss": 0.0697,
      "theoretical_loss": 3.609566800046478,
      "tokens_seen": 1223426048
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006355320173326913,
      "loss": 0.0711,
      "theoretical_loss": 3.609496482514332,
      "tokens_seen": 1223688192
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000635451773391109,
      "loss": 0.0732,
      "theoretical_loss": 3.609426184261128,
      "tokens_seen": 1223950336
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006353715294495266,
      "loss": 0.0735,
      "theoretical_loss": 3.6093559052774538,
      "tokens_seen": 1224212480
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006352912855079442,
      "loss": 0.0715,
      "theoretical_loss": 3.6092856455539017,
      "tokens_seen": 1224474624
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006352110415663618,
      "loss": 0.0745,
      "theoretical_loss": 3.6092154050810725,
      "tokens_seen": 1224736768
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006351307976247793,
      "loss": 0.073,
      "theoretical_loss": 3.6091451838495727,
      "tokens_seen": 1224998912
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000635050553683197,
      "loss": 0.0714,
      "theoretical_loss": 3.609074981850016,
      "tokens_seen": 1225261056
    },
    {
      "epoch": 0.37,
      "objective/train/advantage_avg": 0.000483504292787984,
      "objective/train/docs_used": 448161,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4086085557937622,
      "objective/train/original_loss": 1.4086084365844727,
      "objective/train/theoretical_loss": 3.6090047990730216,
      "objective/train/tokens_used": 1245983200,
      "objective/train/value_avg": -0.007053375244140625,
      "objective/train/value_loss": 0.0002645938075147569,
      "objective/train/value_max": -8.285045623779297e-05,
      "objective/train/value_min": -0.9697265625,
      "objective/train/value_reward_corr": 0.7658859999331806,
      "objective/train/value_std": 0.0177764892578125,
      "objective/train/weight_avg": 1.0006027221679688,
      "objective/train/weighted_lm_loss": 1.4089957475662231,
      "objective/train/weights_max": 1.5008656978607178,
      "objective/train/weights_min": 0.3695588707923889,
      "theoretical_loss": 3.6090047990730216,
      "tokens_seen": 1225523200
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006349703097416145,
      "loss": 0.0747,
      "theoretical_loss": 3.6090047990730216,
      "tokens_seen": 1225523200
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000634890065800032,
      "loss": 0.0711,
      "theoretical_loss": 3.6089346355092164,
      "tokens_seen": 1225785344
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006348098218584497,
      "loss": 0.0737,
      "theoretical_loss": 3.6088644911492334,
      "tokens_seen": 1226047488
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006347295779168672,
      "loss": 0.0731,
      "theoretical_loss": 3.6087943659837114,
      "tokens_seen": 1226309632
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006346493339752849,
      "loss": 0.0708,
      "theoretical_loss": 3.6087242600032976,
      "tokens_seen": 1226571776
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006345690900337025,
      "loss": 0.0721,
      "theoretical_loss": 3.6086541731986443,
      "tokens_seen": 1226833920
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006344888460921201,
      "loss": 0.0685,
      "theoretical_loss": 3.6085841055604106,
      "tokens_seen": 1227096064
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006344086021505376,
      "loss": 0.073,
      "theoretical_loss": 3.608514057079262,
      "tokens_seen": 1227358208
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006343283582089553,
      "loss": 0.0712,
      "theoretical_loss": 3.6084440277458714,
      "tokens_seen": 1227620352
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006342481142673728,
      "loss": 0.0702,
      "theoretical_loss": 3.6083740175509176,
      "tokens_seen": 1227882496
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006341678703257904,
      "loss": 0.0749,
      "theoretical_loss": 3.6083040264850856,
      "tokens_seen": 1228144640
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000634087626384208,
      "loss": 0.0711,
      "theoretical_loss": 3.6082340545390674,
      "tokens_seen": 1228406784
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006340073824426255,
      "loss": 0.0732,
      "theoretical_loss": 3.6081641017035615,
      "tokens_seen": 1228668928
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006339271385010433,
      "loss": 0.0748,
      "theoretical_loss": 3.608094167969273,
      "tokens_seen": 1228931072
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006338468945594608,
      "loss": 0.0717,
      "theoretical_loss": 3.6080242533269136,
      "tokens_seen": 1229193216
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006337666506178784,
      "loss": 0.0701,
      "theoretical_loss": 3.6079543577672,
      "tokens_seen": 1229455360
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000633686406676296,
      "loss": 0.074,
      "theoretical_loss": 3.6078844812808577,
      "tokens_seen": 1229717504
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006336061627347135,
      "loss": 0.0758,
      "theoretical_loss": 3.6078146238586175,
      "tokens_seen": 1229979648
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006335259187931311,
      "loss": 0.0737,
      "theoretical_loss": 3.6077447854912164,
      "tokens_seen": 1230241792
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006334456748515487,
      "loss": 0.0721,
      "theoretical_loss": 3.6076749661693985,
      "tokens_seen": 1230503936
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006333654309099663,
      "loss": 0.0742,
      "theoretical_loss": 3.607605165883914,
      "tokens_seen": 1230766080
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006332851869683838,
      "loss": 0.0728,
      "theoretical_loss": 3.60753538462552,
      "tokens_seen": 1231028224
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006332049430268016,
      "loss": 0.0718,
      "theoretical_loss": 3.607465622384979,
      "tokens_seen": 1231290368
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006331246990852191,
      "loss": 0.0731,
      "theoretical_loss": 3.6073958791530614,
      "tokens_seen": 1231552512
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006330444551436366,
      "loss": 0.0727,
      "theoretical_loss": 3.607326154920543,
      "tokens_seen": 1231814656
    },
    {
      "epoch": 0.37,
      "objective/train/advantage_avg": 0.0014651658711954951,
      "objective/train/docs_used": 450737,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4861600399017334,
      "objective/train/original_loss": 1.4861600399017334,
      "objective/train/theoretical_loss": 3.6072564496782062,
      "objective/train/tokens_used": 1252536800,
      "objective/train/value_avg": -0.01021575927734375,
      "objective/train/value_loss": 0.0003100598114542663,
      "objective/train/value_max": -0.0001055598258972168,
      "objective/train/value_min": -0.75244140625,
      "objective/train/value_reward_corr": 0.6669433027871567,
      "objective/train/value_std": 0.0163116455078125,
      "objective/train/weight_avg": 1.001604437828064,
      "objective/train/weighted_lm_loss": 1.4885464906692505,
      "objective/train/weights_max": 1.9935988187789917,
      "objective/train/weights_min": 0.36889123916625977,
      "theoretical_loss": 3.6072564496782062,
      "tokens_seen": 1232076800
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006329642112020543,
      "loss": 0.0716,
      "theoretical_loss": 3.6072564496782062,
      "tokens_seen": 1232076800
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006328839672604718,
      "loss": 0.0703,
      "theoretical_loss": 3.607186763416841,
      "tokens_seen": 1232338944
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006328037233188895,
      "loss": 0.0728,
      "theoretical_loss": 3.607117096127242,
      "tokens_seen": 1232601088
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000632723479377307,
      "loss": 0.0744,
      "theoretical_loss": 3.60704744780021,
      "tokens_seen": 1232863232
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006326432354357246,
      "loss": 0.0721,
      "theoretical_loss": 3.606977818426555,
      "tokens_seen": 1233125376
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006325629914941422,
      "loss": 0.0745,
      "theoretical_loss": 3.6069082079970913,
      "tokens_seen": 1233387520
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006324827475525598,
      "loss": 0.0739,
      "theoretical_loss": 3.60683861650264,
      "tokens_seen": 1233649664
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006324025036109774,
      "loss": 0.0708,
      "theoretical_loss": 3.606769043934027,
      "tokens_seen": 1233911808
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000632322259669395,
      "loss": 0.0736,
      "theoretical_loss": 3.606699490282088,
      "tokens_seen": 1234173952
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006322420157278126,
      "loss": 0.0699,
      "theoretical_loss": 3.606629955537663,
      "tokens_seen": 1234436096
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006321617717862301,
      "loss": 0.0711,
      "theoretical_loss": 3.606560439691598,
      "tokens_seen": 1234698240
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006320815278446478,
      "loss": 0.0722,
      "theoretical_loss": 3.6064909427347462,
      "tokens_seen": 1234960384
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006320012839030653,
      "loss": 0.0733,
      "theoretical_loss": 3.6064214646579673,
      "tokens_seen": 1235222528
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006319210399614828,
      "loss": 0.0743,
      "theoretical_loss": 3.606352005452126,
      "tokens_seen": 1235484672
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006318407960199005,
      "loss": 0.0728,
      "theoretical_loss": 3.6062825651080956,
      "tokens_seen": 1235746816
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000631760552078318,
      "loss": 0.0736,
      "theoretical_loss": 3.6062131436167544,
      "tokens_seen": 1236008960
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006316803081367358,
      "loss": 0.0717,
      "theoretical_loss": 3.606143740968986,
      "tokens_seen": 1236271104
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006316000641951533,
      "loss": 0.0749,
      "theoretical_loss": 3.6060743571556833,
      "tokens_seen": 1236533248
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006315198202535709,
      "loss": 0.0741,
      "theoretical_loss": 3.606004992167742,
      "tokens_seen": 1236795392
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0006314395763119885,
      "loss": 0.0716,
      "theoretical_loss": 3.605935645996067,
      "tokens_seen": 1237057536
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000631359332370406,
      "loss": 0.0698,
      "theoretical_loss": 3.605866318631568,
      "tokens_seen": 1237319680
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006312790884288236,
      "loss": 0.0712,
      "theoretical_loss": 3.605797010065161,
      "tokens_seen": 1237581824
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006311988444872412,
      "loss": 0.0714,
      "theoretical_loss": 3.6057277202877698,
      "tokens_seen": 1237843968
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006311186005456588,
      "loss": 0.0719,
      "theoretical_loss": 3.6056584492903223,
      "tokens_seen": 1238106112
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006310383566040763,
      "loss": 0.0714,
      "theoretical_loss": 3.6055891970637544,
      "tokens_seen": 1238368256
    },
    {
      "epoch": 0.38,
      "objective/train/advantage_avg": 0.0007562974351458251,
      "objective/train/docs_used": 453261,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5378789901733398,
      "objective/train/original_loss": 1.5378789901733398,
      "objective/train/theoretical_loss": 3.605519963599008,
      "objective/train/tokens_used": 1259090400,
      "objective/train/value_avg": -0.009429931640625,
      "objective/train/value_loss": 0.0003417150001041591,
      "objective/train/value_max": -0.00011593103408813477,
      "objective/train/value_min": -0.72119140625,
      "objective/train/value_reward_corr": 0.7248880939979041,
      "objective/train/value_std": 0.0178985595703125,
      "objective/train/weight_avg": 1.0009098052978516,
      "objective/train/weighted_lm_loss": 1.539913296699524,
      "objective/train/weights_max": 1.2381511926651,
      "objective/train/weights_min": 0.3683769106864929,
      "theoretical_loss": 3.605519963599008,
      "tokens_seen": 1238630400
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006309581126624941,
      "loss": 0.0748,
      "theoretical_loss": 3.605519963599008,
      "tokens_seen": 1238630400
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006308778687209116,
      "loss": 0.0723,
      "theoretical_loss": 3.60545074888703,
      "tokens_seen": 1238892544
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006307976247793292,
      "loss": 0.0714,
      "theoretical_loss": 3.6053815529187756,
      "tokens_seen": 1239154688
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006307173808377468,
      "loss": 0.0739,
      "theoretical_loss": 3.6053123756852052,
      "tokens_seen": 1239416832
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006306371368961643,
      "loss": 0.0709,
      "theoretical_loss": 3.605243217177285,
      "tokens_seen": 1239678976
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006305568929545819,
      "loss": 0.0711,
      "theoretical_loss": 3.6051740773859877,
      "tokens_seen": 1239941120
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006304766490129995,
      "loss": 0.0729,
      "theoretical_loss": 3.605104956302293,
      "tokens_seen": 1240203264
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006303964050714171,
      "loss": 0.0728,
      "theoretical_loss": 3.605035853917187,
      "tokens_seen": 1240465408
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006303161611298347,
      "loss": 0.0748,
      "theoretical_loss": 3.604966770221661,
      "tokens_seen": 1240727552
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006302359171882524,
      "loss": 0.0696,
      "theoretical_loss": 3.604897705206713,
      "tokens_seen": 1240989696
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006301556732466699,
      "loss": 0.0723,
      "theoretical_loss": 3.6048286588633465,
      "tokens_seen": 1241251840
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006300754293050875,
      "loss": 0.0747,
      "theoretical_loss": 3.6047596311825725,
      "tokens_seen": 1241513984
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006299951853635051,
      "loss": 0.0746,
      "theoretical_loss": 3.6046906221554087,
      "tokens_seen": 1241776128
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006299149414219226,
      "loss": 0.0737,
      "theoretical_loss": 3.6046216317728765,
      "tokens_seen": 1242038272
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006298346974803403,
      "loss": 0.0722,
      "theoretical_loss": 3.6045526600260054,
      "tokens_seen": 1242300416
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006297544535387578,
      "loss": 0.0716,
      "theoretical_loss": 3.6044837069058318,
      "tokens_seen": 1242562560
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006296742095971754,
      "loss": 0.0751,
      "theoretical_loss": 3.6044147724033957,
      "tokens_seen": 1242824704
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.000629593965655593,
      "loss": 0.0752,
      "theoretical_loss": 3.6043458565097453,
      "tokens_seen": 1243086848
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006295137217140106,
      "loss": 0.07,
      "theoretical_loss": 3.6042769592159356,
      "tokens_seen": 1243348992
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006294334777724282,
      "loss": 0.0724,
      "theoretical_loss": 3.6042080805130254,
      "tokens_seen": 1243611136
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006293532338308458,
      "loss": 0.0749,
      "theoretical_loss": 3.6041392203920815,
      "tokens_seen": 1243873280
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006292729898892634,
      "loss": 0.0747,
      "theoretical_loss": 3.6040703788441757,
      "tokens_seen": 1244135424
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006291927459476809,
      "loss": 0.0744,
      "theoretical_loss": 3.6040015558603877,
      "tokens_seen": 1244397568
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006291125020060986,
      "loss": 0.073,
      "theoretical_loss": 3.603932751431802,
      "tokens_seen": 1244659712
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006290322580645161,
      "loss": 0.072,
      "theoretical_loss": 3.6038639655495093,
      "tokens_seen": 1244921856
    },
    {
      "epoch": 0.38,
      "objective/train/advantage_avg": -0.000271944678388536,
      "objective/train/docs_used": 455531,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5588250160217285,
      "objective/train/original_loss": 1.558824896812439,
      "objective/train/theoretical_loss": 3.6037951982046064,
      "objective/train/tokens_used": 1265644000,
      "objective/train/value_avg": -0.009490966796875,
      "objective/train/value_loss": 0.000395492505049333,
      "objective/train/value_max": -8.153915405273438e-05,
      "objective/train/value_min": -0.5654296875,
      "objective/train/value_reward_corr": 0.6741612325784748,
      "objective/train/value_std": 0.01556396484375,
      "objective/train/weight_avg": 0.9999120831489563,
      "objective/train/weighted_lm_loss": 1.55759859085083,
      "objective/train/weights_max": 1.5399620532989502,
      "objective/train/weights_min": 0.3693220913410187,
      "theoretical_loss": 3.6037951982046064,
      "tokens_seen": 1245184000
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006289520141229337,
      "loss": 0.0727,
      "theoretical_loss": 3.6037951982046064,
      "tokens_seen": 1245184000
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006288717701813513,
      "loss": 0.0728,
      "theoretical_loss": 3.6037264493881973,
      "tokens_seen": 1245446144
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006287915262397688,
      "loss": 0.075,
      "theoretical_loss": 3.603657719091391,
      "tokens_seen": 1245708288
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006287112822981866,
      "loss": 0.0731,
      "theoretical_loss": 3.603589007305303,
      "tokens_seen": 1245970432
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006286310383566041,
      "loss": 0.075,
      "theoretical_loss": 3.603520314021055,
      "tokens_seen": 1246232576
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006285507944150217,
      "loss": 0.0718,
      "theoretical_loss": 3.6034516392297746,
      "tokens_seen": 1246494720
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006284705504734393,
      "loss": 0.076,
      "theoretical_loss": 3.6033829829225965,
      "tokens_seen": 1246756864
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006283903065318568,
      "loss": 0.075,
      "theoretical_loss": 3.6033143450906593,
      "tokens_seen": 1247019008
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006283100625902744,
      "loss": 0.0732,
      "theoretical_loss": 3.6032457257251105,
      "tokens_seen": 1247281152
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.000628229818648692,
      "loss": 0.0701,
      "theoretical_loss": 3.603177124817101,
      "tokens_seen": 1247543296
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006281495747071096,
      "loss": 0.0724,
      "theoretical_loss": 3.603108542357791,
      "tokens_seen": 1247805440
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006280693307655271,
      "loss": 0.0741,
      "theoretical_loss": 3.603039978338343,
      "tokens_seen": 1248067584
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006279890868239449,
      "loss": 0.0714,
      "theoretical_loss": 3.6029714327499285,
      "tokens_seen": 1248329728
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006279088428823624,
      "loss": 0.0752,
      "theoretical_loss": 3.602902905583724,
      "tokens_seen": 1248591872
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006278285989407801,
      "loss": 0.0717,
      "theoretical_loss": 3.602834396830912,
      "tokens_seen": 1248854016
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006277483549991976,
      "loss": 0.0725,
      "theoretical_loss": 3.6027659064826816,
      "tokens_seen": 1249116160
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006276681110576151,
      "loss": 0.0745,
      "theoretical_loss": 3.602697434530227,
      "tokens_seen": 1249378304
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006275878671160328,
      "loss": 0.0745,
      "theoretical_loss": 3.6026289809647487,
      "tokens_seen": 1249640448
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006275076231744503,
      "loss": 0.0741,
      "theoretical_loss": 3.6025605457774548,
      "tokens_seen": 1249902592
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006274273792328679,
      "loss": 0.0722,
      "theoretical_loss": 3.602492128959558,
      "tokens_seen": 1250164736
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006273471352912855,
      "loss": 0.073,
      "theoretical_loss": 3.6024237305022764,
      "tokens_seen": 1250426880
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006272668913497032,
      "loss": 0.0741,
      "theoretical_loss": 3.602355350396836,
      "tokens_seen": 1250689024
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006271866474081207,
      "loss": 0.072,
      "theoretical_loss": 3.602286988634467,
      "tokens_seen": 1250951168
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006271064034665383,
      "loss": 0.0719,
      "theoretical_loss": 3.6022186452064076,
      "tokens_seen": 1251213312
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006270261595249559,
      "loss": 0.0717,
      "theoretical_loss": 3.6021503201039,
      "tokens_seen": 1251475456
    },
    {
      "epoch": 0.38,
      "objective/train/advantage_avg": 0.0007342287572100759,
      "objective/train/docs_used": 458046,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4588574171066284,
      "objective/train/original_loss": 1.4588572978973389,
      "objective/train/theoretical_loss": 3.6020820133181934,
      "objective/train/tokens_used": 1272197600,
      "objective/train/value_avg": -0.006072998046875,
      "objective/train/value_loss": 0.00018203555373474956,
      "objective/train/value_max": -6.157159805297852e-05,
      "objective/train/value_min": -0.29638671875,
      "objective/train/value_reward_corr": 0.6572453525608541,
      "objective/train/value_std": 0.0111083984375,
      "objective/train/weight_avg": 1.0008169412612915,
      "objective/train/weighted_lm_loss": 1.460231065750122,
      "objective/train/weights_max": 1.3257553577423096,
      "objective/train/weights_min": 0.3682669997215271,
      "theoretical_loss": 3.6020820133181934,
      "tokens_seen": 1251737600
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006269459155833734,
      "loss": 0.0709,
      "theoretical_loss": 3.6020820133181934,
      "tokens_seen": 1251737600
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006268656716417911,
      "loss": 0.0742,
      "theoretical_loss": 3.6020137248405435,
      "tokens_seen": 1251999744
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006267854277002086,
      "loss": 0.0733,
      "theoretical_loss": 3.6019454546622107,
      "tokens_seen": 1252261888
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006267051837586262,
      "loss": 0.0712,
      "theoretical_loss": 3.601877202774463,
      "tokens_seen": 1252524032
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006266249398170438,
      "loss": 0.0727,
      "theoretical_loss": 3.601808969168573,
      "tokens_seen": 1252786176
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006265446958754613,
      "loss": 0.0721,
      "theoretical_loss": 3.60174075383582,
      "tokens_seen": 1253048320
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006264644519338791,
      "loss": 0.0753,
      "theoretical_loss": 3.6016725567674883,
      "tokens_seen": 1253310464
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006263842079922966,
      "loss": 0.0728,
      "theoretical_loss": 3.60160437795487,
      "tokens_seen": 1253572608
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006263039640507142,
      "loss": 0.073,
      "theoretical_loss": 3.6015362173892624,
      "tokens_seen": 1253834752
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006262237201091318,
      "loss": 0.0745,
      "theoretical_loss": 3.6014680750619674,
      "tokens_seen": 1254096896
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006261434761675494,
      "loss": 0.0726,
      "theoretical_loss": 3.6013999509642947,
      "tokens_seen": 1254359040
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006260632322259669,
      "loss": 0.0723,
      "theoretical_loss": 3.601331845087559,
      "tokens_seen": 1254621184
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006259829882843845,
      "loss": 0.073,
      "theoretical_loss": 3.601263757423082,
      "tokens_seen": 1254883328
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006259027443428021,
      "loss": 0.073,
      "theoretical_loss": 3.601195687962189,
      "tokens_seen": 1255145472
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006258225004012196,
      "loss": 0.0718,
      "theoretical_loss": 3.601127636696214,
      "tokens_seen": 1255407616
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006257422564596374,
      "loss": 0.0725,
      "theoretical_loss": 3.6010596036164957,
      "tokens_seen": 1255669760
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006256620125180549,
      "loss": 0.0724,
      "theoretical_loss": 3.600991588714378,
      "tokens_seen": 1255931904
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006255817685764725,
      "loss": 0.0705,
      "theoretical_loss": 3.6009235919812124,
      "tokens_seen": 1256194048
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006255015246348901,
      "loss": 0.0717,
      "theoretical_loss": 3.6008556134083545,
      "tokens_seen": 1256456192
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006254212806933076,
      "loss": 0.0733,
      "theoretical_loss": 3.600787652987167,
      "tokens_seen": 1256718336
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006253410367517252,
      "loss": 0.0717,
      "theoretical_loss": 3.6007197107090185,
      "tokens_seen": 1256980480
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006252607928101428,
      "loss": 0.0736,
      "theoretical_loss": 3.600651786565283,
      "tokens_seen": 1257242624
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006251805488685604,
      "loss": 0.0718,
      "theoretical_loss": 3.6005838805473402,
      "tokens_seen": 1257504768
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.000625100304926978,
      "loss": 0.0717,
      "theoretical_loss": 3.600515992646577,
      "tokens_seen": 1257766912
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006250200609853957,
      "loss": 0.0719,
      "theoretical_loss": 3.6004481228543854,
      "tokens_seen": 1258029056
    },
    {
      "epoch": 0.38,
      "objective/train/advantage_avg": 0.0012113068951293826,
      "objective/train/docs_used": 460354,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4824901819229126,
      "objective/train/original_loss": 1.482490062713623,
      "objective/train/theoretical_loss": 3.600380271162162,
      "objective/train/tokens_used": 1278751200,
      "objective/train/value_avg": -0.01016998291015625,
      "objective/train/value_loss": 0.00046331010526046157,
      "objective/train/value_max": -0.00010889768600463867,
      "objective/train/value_min": -0.97119140625,
      "objective/train/value_reward_corr": 0.7243134544742209,
      "objective/train/value_std": 0.0211181640625,
      "objective/train/weight_avg": 1.0014119148254395,
      "objective/train/weighted_lm_loss": 1.484576940536499,
      "objective/train/weights_max": 1.8089985847473145,
      "objective/train/weights_min": 0.36829686164855957,
      "theoretical_loss": 3.600380271162162,
      "tokens_seen": 1258291200
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006249398170438132,
      "loss": 0.0709,
      "theoretical_loss": 3.600380271162162,
      "tokens_seen": 1258291200
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006248595731022309,
      "loss": 0.0693,
      "theoretical_loss": 3.6003124375613114,
      "tokens_seen": 1258553344
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006247793291606484,
      "loss": 0.0746,
      "theoretical_loss": 3.6002446220432427,
      "tokens_seen": 1258815488
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006246990852190659,
      "loss": 0.074,
      "theoretical_loss": 3.6001768245993713,
      "tokens_seen": 1259077632
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006246188412774836,
      "loss": 0.0731,
      "theoretical_loss": 3.600109045221119,
      "tokens_seen": 1259339776
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006245385973359011,
      "loss": 0.0749,
      "theoretical_loss": 3.600041283899912,
      "tokens_seen": 1259601920
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006244583533943187,
      "loss": 0.0713,
      "theoretical_loss": 3.599973540627184,
      "tokens_seen": 1259864064
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006243781094527363,
      "loss": 0.0742,
      "theoretical_loss": 3.599905815394374,
      "tokens_seen": 1260126208
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.000624297865511154,
      "loss": 0.0736,
      "theoretical_loss": 3.599838108192926,
      "tokens_seen": 1260388352
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006242176215695715,
      "loss": 0.0724,
      "theoretical_loss": 3.5997704190142907,
      "tokens_seen": 1260650496
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006241373776279891,
      "loss": 0.0741,
      "theoretical_loss": 3.599702747849925,
      "tokens_seen": 1260912640
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006240571336864067,
      "loss": 0.073,
      "theoretical_loss": 3.5996350946912896,
      "tokens_seen": 1261174784
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006239768897448243,
      "loss": 0.0715,
      "theoretical_loss": 3.5995674595298537,
      "tokens_seen": 1261436928
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006238966458032419,
      "loss": 0.072,
      "theoretical_loss": 3.5994998423570914,
      "tokens_seen": 1261699072
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006238164018616594,
      "loss": 0.0714,
      "theoretical_loss": 3.599432243164481,
      "tokens_seen": 1261961216
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006237361579200771,
      "loss": 0.0713,
      "theoretical_loss": 3.599364661943509,
      "tokens_seen": 1262223360
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006236559139784946,
      "loss": 0.0726,
      "theoretical_loss": 3.599297098685666,
      "tokens_seen": 1262485504
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006235756700369121,
      "loss": 0.071,
      "theoretical_loss": 3.599229553382449,
      "tokens_seen": 1262747648
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006234954260953299,
      "loss": 0.0725,
      "theoretical_loss": 3.599162026025361,
      "tokens_seen": 1263009792
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006234151821537474,
      "loss": 0.0705,
      "theoretical_loss": 3.599094516605911,
      "tokens_seen": 1263271936
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.000623334938212165,
      "loss": 0.0738,
      "theoretical_loss": 3.5990270251156122,
      "tokens_seen": 1263534080
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006232546942705826,
      "loss": 0.0746,
      "theoretical_loss": 3.5989595515459856,
      "tokens_seen": 1263796224
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006231744503290002,
      "loss": 0.0743,
      "theoretical_loss": 3.598892095888557,
      "tokens_seen": 1264058368
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006230942063874177,
      "loss": 0.0716,
      "theoretical_loss": 3.5988246581348573,
      "tokens_seen": 1264320512
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006230139624458353,
      "loss": 0.073,
      "theoretical_loss": 3.5987572382764252,
      "tokens_seen": 1264582656
    },
    {
      "epoch": 0.38,
      "objective/train/advantage_avg": 0.0005436477949842811,
      "objective/train/docs_used": 462829,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.450318455696106,
      "objective/train/original_loss": 1.4503185749053955,
      "objective/train/theoretical_loss": 3.5986898363048025,
      "objective/train/tokens_used": 1285304800,
      "objective/train/value_avg": -0.007289886474609375,
      "objective/train/value_loss": 0.00032011562143452466,
      "objective/train/value_max": -8.285045623779297e-05,
      "objective/train/value_min": -0.841796875,
      "objective/train/value_reward_corr": 0.5236847438293369,
      "objective/train/value_std": 0.011871337890625,
      "objective/train/weight_avg": 1.0006617307662964,
      "objective/train/weighted_lm_loss": 1.4513938426971436,
      "objective/train/weights_max": 1.2407280206680298,
      "objective/train/weights_min": 0.05488771200180054,
      "theoretical_loss": 3.5986898363048025,
      "tokens_seen": 1264844800
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006229337185042529,
      "loss": 0.0724,
      "theoretical_loss": 3.5986898363048025,
      "tokens_seen": 1264844800
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006228534745626704,
      "loss": 0.0713,
      "theoretical_loss": 3.5986224522115395,
      "tokens_seen": 1265106944
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006227732306210882,
      "loss": 0.0704,
      "theoretical_loss": 3.59855508598819,
      "tokens_seen": 1265369088
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006226929866795057,
      "loss": 0.0736,
      "theoretical_loss": 3.5984877376263142,
      "tokens_seen": 1265631232
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006226127427379234,
      "loss": 0.0699,
      "theoretical_loss": 3.5984204071174783,
      "tokens_seen": 1265893376
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006225324987963409,
      "loss": 0.0763,
      "theoretical_loss": 3.598353094453255,
      "tokens_seen": 1266155520
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006224522548547584,
      "loss": 0.0724,
      "theoretical_loss": 3.598285799625221,
      "tokens_seen": 1266417664
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006223720109131761,
      "loss": 0.072,
      "theoretical_loss": 3.5982185226249594,
      "tokens_seen": 1266679808
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006222917669715936,
      "loss": 0.0718,
      "theoretical_loss": 3.5981512634440604,
      "tokens_seen": 1266941952
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006222115230300112,
      "loss": 0.0725,
      "theoretical_loss": 3.5980840220741177,
      "tokens_seen": 1267204096
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006221312790884289,
      "loss": 0.0748,
      "theoretical_loss": 3.5980167985067317,
      "tokens_seen": 1267466240
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006220510351468465,
      "loss": 0.074,
      "theoretical_loss": 3.5979495927335092,
      "tokens_seen": 1267728384
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.000621970791205264,
      "loss": 0.0727,
      "theoretical_loss": 3.5978824047460614,
      "tokens_seen": 1267990528
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006218905472636816,
      "loss": 0.0732,
      "theoretical_loss": 3.5978152345360055,
      "tokens_seen": 1268252672
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006218103033220992,
      "loss": 0.0723,
      "theoretical_loss": 3.5977480820949657,
      "tokens_seen": 1268514816
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006217300593805167,
      "loss": 0.0741,
      "theoretical_loss": 3.5976809474145703,
      "tokens_seen": 1268776960
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006216498154389344,
      "loss": 0.073,
      "theoretical_loss": 3.597613830486454,
      "tokens_seen": 1269039104
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006215695714973519,
      "loss": 0.0717,
      "theoretical_loss": 3.5975467313022564,
      "tokens_seen": 1269301248
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006214893275557696,
      "loss": 0.0734,
      "theoretical_loss": 3.5974796498536237,
      "tokens_seen": 1269563392
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006214090836141871,
      "loss": 0.0708,
      "theoretical_loss": 3.597412586132208,
      "tokens_seen": 1269825536
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006213288396726048,
      "loss": 0.0728,
      "theoretical_loss": 3.5973455401296652,
      "tokens_seen": 1270087680
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0006212485957310224,
      "loss": 0.0698,
      "theoretical_loss": 3.597278511837659,
      "tokens_seen": 1270349824
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006211683517894399,
      "loss": 0.0713,
      "theoretical_loss": 3.597211501247858,
      "tokens_seen": 1270611968
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006210881078478575,
      "loss": 0.07,
      "theoretical_loss": 3.5971445083519358,
      "tokens_seen": 1270874112
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006210078639062751,
      "loss": 0.0732,
      "theoretical_loss": 3.5970775331415723,
      "tokens_seen": 1271136256
    },
    {
      "epoch": 0.39,
      "objective/train/advantage_avg": 0.0007621328113600612,
      "objective/train/docs_used": 465264,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4400675296783447,
      "objective/train/original_loss": 1.4400676488876343,
      "objective/train/theoretical_loss": 3.5970105756084525,
      "objective/train/tokens_used": 1291858400,
      "objective/train/value_avg": -0.007965087890625,
      "objective/train/value_loss": 0.00021634330914821476,
      "objective/train/value_max": -9.101629257202148e-05,
      "objective/train/value_min": -0.335205078125,
      "objective/train/value_reward_corr": 0.6737940618129363,
      "objective/train/value_std": 0.0167388916015625,
      "objective/train/weight_avg": 1.0008677244186401,
      "objective/train/weighted_lm_loss": 1.4404685497283936,
      "objective/train/weights_max": 1.2927756309509277,
      "objective/train/weights_min": 0.37808898091316223,
      "theoretical_loss": 3.5970105756084525,
      "tokens_seen": 1271398400
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006209276199646927,
      "loss": 0.0703,
      "theoretical_loss": 3.5970105756084525,
      "tokens_seen": 1271398400
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006208473760231102,
      "loss": 0.0727,
      "theoretical_loss": 3.5969436357442683,
      "tokens_seen": 1271660544
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006207671320815279,
      "loss": 0.0719,
      "theoretical_loss": 3.5968767135407154,
      "tokens_seen": 1271922688
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006206868881399454,
      "loss": 0.0712,
      "theoretical_loss": 3.596809808989496,
      "tokens_seen": 1272184832
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000620606644198363,
      "loss": 0.0727,
      "theoretical_loss": 3.596742922082319,
      "tokens_seen": 1272446976
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006205264002567807,
      "loss": 0.0704,
      "theoretical_loss": 3.5966760528108965,
      "tokens_seen": 1272709120
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006204461563151982,
      "loss": 0.0725,
      "theoretical_loss": 3.596609201166948,
      "tokens_seen": 1272971264
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006203659123736158,
      "loss": 0.0726,
      "theoretical_loss": 3.596542367142198,
      "tokens_seen": 1273233408
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006202856684320334,
      "loss": 0.0725,
      "theoretical_loss": 3.5964755507283774,
      "tokens_seen": 1273495552
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000620205424490451,
      "loss": 0.0737,
      "theoretical_loss": 3.5964087519172203,
      "tokens_seen": 1273757696
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006201251805488686,
      "loss": 0.0744,
      "theoretical_loss": 3.59634197070047,
      "tokens_seen": 1274019840
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006200449366072861,
      "loss": 0.075,
      "theoretical_loss": 3.5962752070698727,
      "tokens_seen": 1274281984
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006199646926657037,
      "loss": 0.0693,
      "theoretical_loss": 3.59620846101718,
      "tokens_seen": 1274544128
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006198844487241214,
      "loss": 0.0726,
      "theoretical_loss": 3.596141732534151,
      "tokens_seen": 1274806272
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000619804204782539,
      "loss": 0.0749,
      "theoretical_loss": 3.596075021612549,
      "tokens_seen": 1275068416
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006197239608409565,
      "loss": 0.0739,
      "theoretical_loss": 3.5960083282441433,
      "tokens_seen": 1275330560
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006196437168993742,
      "loss": 0.0741,
      "theoretical_loss": 3.5959416524207084,
      "tokens_seen": 1275592704
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006195634729577917,
      "loss": 0.0732,
      "theoretical_loss": 3.5958749941340242,
      "tokens_seen": 1275854848
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006194832290162092,
      "loss": 0.0735,
      "theoretical_loss": 3.595808353375877,
      "tokens_seen": 1276116992
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006194029850746269,
      "loss": 0.0723,
      "theoretical_loss": 3.5957417301380588,
      "tokens_seen": 1276379136
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006193227411330444,
      "loss": 0.0735,
      "theoretical_loss": 3.595675124412365,
      "tokens_seen": 1276641280
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000619242497191462,
      "loss": 0.0762,
      "theoretical_loss": 3.595608536190599,
      "tokens_seen": 1276903424
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006191622532498797,
      "loss": 0.0724,
      "theoretical_loss": 3.595541965464568,
      "tokens_seen": 1277165568
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006190820093082973,
      "loss": 0.0702,
      "theoretical_loss": 3.5954754122260866,
      "tokens_seen": 1277427712
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006190017653667148,
      "loss": 0.0739,
      "theoretical_loss": 3.595408876466972,
      "tokens_seen": 1277689856
    },
    {
      "epoch": 0.39,
      "objective/train/advantage_avg": -0.0006468091160058975,
      "objective/train/docs_used": 467639,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4236520528793335,
      "objective/train/original_loss": 1.423651933670044,
      "objective/train/theoretical_loss": 3.59534235817905,
      "objective/train/tokens_used": 1298412000,
      "objective/train/value_avg": -0.0084686279296875,
      "objective/train/value_loss": 0.0006970210233703256,
      "objective/train/value_max": -8.547306060791016e-05,
      "objective/train/value_min": -0.8544921875,
      "objective/train/value_reward_corr": 0.7078890831361996,
      "objective/train/value_std": 0.022552490234375,
      "objective/train/weight_avg": 0.9996680021286011,
      "objective/train/weighted_lm_loss": 1.4211854934692383,
      "objective/train/weights_max": 2.232908010482788,
      "objective/train/weights_min": 0.36951375007629395,
      "theoretical_loss": 3.59534235817905,
      "tokens_seen": 1277952000
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006189215214251324,
      "loss": 0.0727,
      "theoretical_loss": 3.59534235817905,
      "tokens_seen": 1277952000
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00061884127748355,
      "loss": 0.0732,
      "theoretical_loss": 3.5952758573541503,
      "tokens_seen": 1278214144
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006187610335419676,
      "loss": 0.0722,
      "theoretical_loss": 3.5952093739841073,
      "tokens_seen": 1278476288
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006186807896003852,
      "loss": 0.0743,
      "theoretical_loss": 3.5951429080607635,
      "tokens_seen": 1278738432
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006186005456588027,
      "loss": 0.0717,
      "theoretical_loss": 3.5950764595759646,
      "tokens_seen": 1279000576
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006185203017172204,
      "loss": 0.0725,
      "theoretical_loss": 3.5950100285215623,
      "tokens_seen": 1279262720
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000618440057775638,
      "loss": 0.0704,
      "theoretical_loss": 3.594943614889414,
      "tokens_seen": 1279524864
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006183598138340556,
      "loss": 0.0745,
      "theoretical_loss": 3.5948772186713827,
      "tokens_seen": 1279787008
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006182795698924732,
      "loss": 0.0727,
      "theoretical_loss": 3.5948108398593357,
      "tokens_seen": 1280049152
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006181993259508907,
      "loss": 0.0735,
      "theoretical_loss": 3.5947444784451488,
      "tokens_seen": 1280311296
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006181190820093083,
      "loss": 0.0715,
      "theoretical_loss": 3.5946781344206995,
      "tokens_seen": 1280573440
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006180388380677259,
      "loss": 0.0745,
      "theoretical_loss": 3.5946118077778735,
      "tokens_seen": 1280835584
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006179585941261435,
      "loss": 0.0758,
      "theoretical_loss": 3.59454549850856,
      "tokens_seen": 1281097728
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000617878350184561,
      "loss": 0.0736,
      "theoretical_loss": 3.594479206604655,
      "tokens_seen": 1281359872
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006177981062429787,
      "loss": 0.0746,
      "theoretical_loss": 3.59441293205806,
      "tokens_seen": 1281622016
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006177178623013962,
      "loss": 0.0724,
      "theoretical_loss": 3.59434667486068,
      "tokens_seen": 1281884160
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006176376183598139,
      "loss": 0.0726,
      "theoretical_loss": 3.5942804350044284,
      "tokens_seen": 1282146304
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006175573744182315,
      "loss": 0.0723,
      "theoretical_loss": 3.594214212481222,
      "tokens_seen": 1282408448
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000617477130476649,
      "loss": 0.07,
      "theoretical_loss": 3.594148007282983,
      "tokens_seen": 1282670592
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006173968865350667,
      "loss": 0.0673,
      "theoretical_loss": 3.59408181940164,
      "tokens_seen": 1282932736
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006173166425934842,
      "loss": 0.0711,
      "theoretical_loss": 3.5940156488291266,
      "tokens_seen": 1283194880
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006172363986519018,
      "loss": 0.0704,
      "theoretical_loss": 3.593949495557381,
      "tokens_seen": 1283457024
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006171561547103194,
      "loss": 0.0721,
      "theoretical_loss": 3.5938833595783484,
      "tokens_seen": 1283719168
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006170759107687369,
      "loss": 0.0723,
      "theoretical_loss": 3.593817240883978,
      "tokens_seen": 1283981312
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006169956668271545,
      "loss": 0.072,
      "theoretical_loss": 3.5937511394662254,
      "tokens_seen": 1284243456
    },
    {
      "epoch": 0.39,
      "objective/train/advantage_avg": 0.0013448818353936076,
      "objective/train/docs_used": 470107,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3530259132385254,
      "objective/train/original_loss": 1.3530259132385254,
      "objective/train/theoretical_loss": 3.5936850553170503,
      "objective/train/tokens_used": 1304965600,
      "objective/train/value_avg": -0.0072174072265625,
      "objective/train/value_loss": 0.00014462636318057775,
      "objective/train/value_max": -8.887052536010742e-05,
      "objective/train/value_min": -0.44189453125,
      "objective/train/value_reward_corr": 0.6305170353978925,
      "objective/train/value_std": 0.011138916015625,
      "objective/train/weight_avg": 1.0014116764068604,
      "objective/train/weighted_lm_loss": 1.3553401231765747,
      "objective/train/weights_max": 1.2892757654190063,
      "objective/train/weights_min": 0.39345037937164307,
      "theoretical_loss": 3.5936850553170503,
      "tokens_seen": 1284505600
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006169154228855722,
      "loss": 0.0713,
      "theoretical_loss": 3.5936850553170503,
      "tokens_seen": 1284505600
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006168351789439898,
      "loss": 0.0734,
      "theoretical_loss": 3.593618988428419,
      "tokens_seen": 1284767744
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006167549350024073,
      "loss": 0.07,
      "theoretical_loss": 3.593552938792303,
      "tokens_seen": 1285029888
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000616674691060825,
      "loss": 0.0716,
      "theoretical_loss": 3.5934869064006785,
      "tokens_seen": 1285292032
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006165944471192425,
      "loss": 0.0734,
      "theoretical_loss": 3.593420891245527,
      "tokens_seen": 1285554176
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00061651420317766,
      "loss": 0.0748,
      "theoretical_loss": 3.5933548933188377,
      "tokens_seen": 1285816320
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006164339592360777,
      "loss": 0.0713,
      "theoretical_loss": 3.5932889126126017,
      "tokens_seen": 1286078464
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006163537152944952,
      "loss": 0.0738,
      "theoretical_loss": 3.5932229491188172,
      "tokens_seen": 1286340608
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006162734713529129,
      "loss": 0.0708,
      "theoretical_loss": 3.5931570028294884,
      "tokens_seen": 1286602752
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006161932274113304,
      "loss": 0.0719,
      "theoretical_loss": 3.5930910737366233,
      "tokens_seen": 1286864896
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006161129834697481,
      "loss": 0.0731,
      "theoretical_loss": 3.5930251618322364,
      "tokens_seen": 1287127040
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006160327395281657,
      "loss": 0.072,
      "theoretical_loss": 3.5929592671083466,
      "tokens_seen": 1287389184
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006159524955865832,
      "loss": 0.0716,
      "theoretical_loss": 3.5928933895569792,
      "tokens_seen": 1287651328
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006158722516450008,
      "loss": 0.07,
      "theoretical_loss": 3.5928275291701643,
      "tokens_seen": 1287913472
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006157920077034184,
      "loss": 0.071,
      "theoretical_loss": 3.592761685939937,
      "tokens_seen": 1288175616
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000615711763761836,
      "loss": 0.0717,
      "theoretical_loss": 3.5926958598583383,
      "tokens_seen": 1288437760
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006156315198202535,
      "loss": 0.0692,
      "theoretical_loss": 3.5926300509174136,
      "tokens_seen": 1288699904
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006155512758786712,
      "loss": 0.0711,
      "theoretical_loss": 3.5925642591092153,
      "tokens_seen": 1288962048
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006154710319370887,
      "loss": 0.0694,
      "theoretical_loss": 3.592498484425799,
      "tokens_seen": 1289224192
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006153907879955064,
      "loss": 0.0737,
      "theoretical_loss": 3.5924327268592267,
      "tokens_seen": 1289486336
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000615310544053924,
      "loss": 0.0696,
      "theoretical_loss": 3.5923669864015664,
      "tokens_seen": 1289748480
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006152303001123415,
      "loss": 0.0699,
      "theoretical_loss": 3.59230126304489,
      "tokens_seen": 1290010624
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006151500561707592,
      "loss": 0.0732,
      "theoretical_loss": 3.592235556781276,
      "tokens_seen": 1290272768
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006150698122291767,
      "loss": 0.0705,
      "theoretical_loss": 3.5921698676028058,
      "tokens_seen": 1290534912
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006149895682875943,
      "loss": 0.0684,
      "theoretical_loss": 3.59210419550157,
      "tokens_seen": 1290797056
    },
    {
      "epoch": 0.39,
      "objective/train/advantage_avg": 0.0011514985235407948,
      "objective/train/docs_used": 472355,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4526315927505493,
      "objective/train/original_loss": 1.4526317119598389,
      "objective/train/theoretical_loss": 3.5920385404696606,
      "objective/train/tokens_used": 1311519200,
      "objective/train/value_avg": -0.007213592529296875,
      "objective/train/value_loss": 0.00013709472841583192,
      "objective/train/value_max": -7.545948028564453e-05,
      "objective/train/value_min": -0.9716796875,
      "objective/train/value_reward_corr": 0.8126945195500097,
      "objective/train/value_std": 0.01486968994140625,
      "objective/train/weight_avg": 1.0012160539627075,
      "objective/train/weighted_lm_loss": 1.4547836780548096,
      "objective/train/weights_max": 1.2823024988174438,
      "objective/train/weights_min": 0.5326542258262634,
      "theoretical_loss": 3.5920385404696606,
      "tokens_seen": 1291059200
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006149093243460119,
      "loss": 0.0722,
      "theoretical_loss": 3.5920385404696606,
      "tokens_seen": 1291059200
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006148290804044295,
      "loss": 0.0715,
      "theoretical_loss": 3.591972902499177,
      "tokens_seen": 1291321344
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000614748836462847,
      "loss": 0.0714,
      "theoretical_loss": 3.5919072815822233,
      "tokens_seen": 1291583488
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006146685925212647,
      "loss": 0.0713,
      "theoretical_loss": 3.5918416777109092,
      "tokens_seen": 1291845632
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006145883485796823,
      "loss": 0.0698,
      "theoretical_loss": 3.591776090877349,
      "tokens_seen": 1292107776
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006145081046380998,
      "loss": 0.0753,
      "theoretical_loss": 3.5917105210736624,
      "tokens_seen": 1292369920
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006144278606965175,
      "loss": 0.0718,
      "theoretical_loss": 3.591644968291975,
      "tokens_seen": 1292632064
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000614347616754935,
      "loss": 0.0707,
      "theoretical_loss": 3.5915794325244175,
      "tokens_seen": 1292894208
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006142673728133526,
      "loss": 0.0722,
      "theoretical_loss": 3.5915139137631242,
      "tokens_seen": 1293156352
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006141871288717702,
      "loss": 0.07,
      "theoretical_loss": 3.5914484120002372,
      "tokens_seen": 1293418496
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006141068849301877,
      "loss": 0.0718,
      "theoretical_loss": 3.591382927227902,
      "tokens_seen": 1293680640
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006140266409886053,
      "loss": 0.0747,
      "theoretical_loss": 3.5913174594382693,
      "tokens_seen": 1293942784
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000613946397047023,
      "loss": 0.0725,
      "theoretical_loss": 3.5912520086234974,
      "tokens_seen": 1294204928
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006138661531054406,
      "loss": 0.0733,
      "theoretical_loss": 3.5911865747757457,
      "tokens_seen": 1294467072
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006137859091638582,
      "loss": 0.0715,
      "theoretical_loss": 3.5911211578871827,
      "tokens_seen": 1294729216
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006137056652222758,
      "loss": 0.0701,
      "theoretical_loss": 3.591055757949981,
      "tokens_seen": 1294991360
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006136254212806933,
      "loss": 0.0739,
      "theoretical_loss": 3.590990374956316,
      "tokens_seen": 1295253504
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006135451773391109,
      "loss": 0.0739,
      "theoretical_loss": 3.5909250088983713,
      "tokens_seen": 1295515648
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006134649333975285,
      "loss": 0.074,
      "theoretical_loss": 3.5908596597683347,
      "tokens_seen": 1295777792
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000613384689455946,
      "loss": 0.0713,
      "theoretical_loss": 3.590794327558399,
      "tokens_seen": 1296039936
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006133044455143637,
      "loss": 0.0697,
      "theoretical_loss": 3.590729012260762,
      "tokens_seen": 1296302080
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006132242015727812,
      "loss": 0.0739,
      "theoretical_loss": 3.5906637138676265,
      "tokens_seen": 1296564224
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006131439576311989,
      "loss": 0.0708,
      "theoretical_loss": 3.590598432371202,
      "tokens_seen": 1296826368
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006130637136896165,
      "loss": 0.0705,
      "theoretical_loss": 3.5905331677637013,
      "tokens_seen": 1297088512
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000612983469748034,
      "loss": 0.0712,
      "theoretical_loss": 3.5904679200373435,
      "tokens_seen": 1297350656
    },
    {
      "epoch": 0.39,
      "objective/train/advantage_avg": 0.0005244042840786278,
      "objective/train/docs_used": 474727,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4231724739074707,
      "objective/train/original_loss": 1.4231724739074707,
      "objective/train/theoretical_loss": 3.5904026891843523,
      "objective/train/tokens_used": 1318072800,
      "objective/train/value_avg": -0.006511688232421875,
      "objective/train/value_loss": 0.00018917610577773303,
      "objective/train/value_max": -5.1856040954589844e-05,
      "objective/train/value_min": -0.394287109375,
      "objective/train/value_reward_corr": 0.6175634721862767,
      "objective/train/value_std": 0.0108795166015625,
      "objective/train/weight_avg": 1.000605583190918,
      "objective/train/weighted_lm_loss": 1.4241219758987427,
      "objective/train/weights_max": 1.4251643419265747,
      "objective/train/weights_min": 0.23235876858234406,
      "theoretical_loss": 3.5904026891843523,
      "tokens_seen": 1297612800
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006129032258064516,
      "loss": 0.0701,
      "theoretical_loss": 3.5904026891843523,
      "tokens_seen": 1297612800
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006128229818648692,
      "loss": 0.0683,
      "theoretical_loss": 3.5903374751969563,
      "tokens_seen": 1297874944
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006127427379232868,
      "loss": 0.0687,
      "theoretical_loss": 3.59027227806739,
      "tokens_seen": 1298137088
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006126624939817043,
      "loss": 0.0709,
      "theoretical_loss": 3.5902070977878937,
      "tokens_seen": 1298399232
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000612582250040122,
      "loss": 0.07,
      "theoretical_loss": 3.5901419343507106,
      "tokens_seen": 1298661376
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006125020060985395,
      "loss": 0.0704,
      "theoretical_loss": 3.5900767877480906,
      "tokens_seen": 1298923520
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006124217621569572,
      "loss": 0.0695,
      "theoretical_loss": 3.5900116579722883,
      "tokens_seen": 1299185664
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006123415182153748,
      "loss": 0.0715,
      "theoretical_loss": 3.5899465450155637,
      "tokens_seen": 1299447808
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006122612742737923,
      "loss": 0.0724,
      "theoretical_loss": 3.589881448870182,
      "tokens_seen": 1299709952
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00061218103033221,
      "loss": 0.0722,
      "theoretical_loss": 3.589816369528413,
      "tokens_seen": 1299972096
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006121007863906275,
      "loss": 0.0702,
      "theoretical_loss": 3.5897513069825324,
      "tokens_seen": 1300234240
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006120205424490451,
      "loss": 0.0719,
      "theoretical_loss": 3.589686261224819,
      "tokens_seen": 1300496384
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006119402985074627,
      "loss": 0.0717,
      "theoretical_loss": 3.5896212322475605,
      "tokens_seen": 1300758528
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006118600545658803,
      "loss": 0.0712,
      "theoretical_loss": 3.589556220043046,
      "tokens_seen": 1301020672
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006117798106242978,
      "loss": 0.0715,
      "theoretical_loss": 3.589491224603571,
      "tokens_seen": 1301282816
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006116995666827155,
      "loss": 0.0714,
      "theoretical_loss": 3.5894262459214366,
      "tokens_seen": 1301544960
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006116193227411331,
      "loss": 0.0695,
      "theoretical_loss": 3.589361283988948,
      "tokens_seen": 1301807104
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006115390787995506,
      "loss": 0.0711,
      "theoretical_loss": 3.589296338798418,
      "tokens_seen": 1302069248
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006114588348579683,
      "loss": 0.0724,
      "theoretical_loss": 3.5892314103421596,
      "tokens_seen": 1302331392
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006113785909163858,
      "loss": 0.0712,
      "theoretical_loss": 3.589166498612496,
      "tokens_seen": 1302593536
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006112983469748035,
      "loss": 0.0718,
      "theoretical_loss": 3.589101603601752,
      "tokens_seen": 1302855680
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000611218103033221,
      "loss": 0.0731,
      "theoretical_loss": 3.58903672530226,
      "tokens_seen": 1303117824
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0006111378590916385,
      "loss": 0.0719,
      "theoretical_loss": 3.5889718637063552,
      "tokens_seen": 1303379968
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006110576151500562,
      "loss": 0.0701,
      "theoretical_loss": 3.5889070188063794,
      "tokens_seen": 1303642112
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006109773712084738,
      "loss": 0.0733,
      "theoretical_loss": 3.5888421905946783,
      "tokens_seen": 1303904256
    },
    {
      "epoch": 0.4,
      "objective/train/advantage_avg": 0.0004786965437233448,
      "objective/train/docs_used": 476908,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.320482850074768,
      "objective/train/original_loss": 1.3204827308654785,
      "objective/train/theoretical_loss": 3.588777379063604,
      "objective/train/tokens_used": 1324626400,
      "objective/train/value_avg": -0.007488250732421875,
      "objective/train/value_loss": 0.0003243642277084291,
      "objective/train/value_max": -8.821487426757812e-05,
      "objective/train/value_min": -0.79052734375,
      "objective/train/value_reward_corr": 0.5708697386938653,
      "objective/train/value_std": 0.01364898681640625,
      "objective/train/weight_avg": 1.0006197690963745,
      "objective/train/weighted_lm_loss": 1.3216181993484497,
      "objective/train/weights_max": 1.9110832214355469,
      "objective/train/weights_min": 0.368977814912796,
      "theoretical_loss": 3.588777379063604,
      "tokens_seen": 1304166400
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006108971272668914,
      "loss": 0.0702,
      "theoretical_loss": 3.588777379063604,
      "tokens_seen": 1304166400
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000610816883325309,
      "loss": 0.074,
      "theoretical_loss": 3.5887125842055116,
      "tokens_seen": 1304428544
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006107366393837266,
      "loss": 0.07,
      "theoretical_loss": 3.588647806012765,
      "tokens_seen": 1304690688
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006106563954421441,
      "loss": 0.0711,
      "theoretical_loss": 3.588583044477728,
      "tokens_seen": 1304952832
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006105761515005617,
      "loss": 0.0741,
      "theoretical_loss": 3.5885182995927734,
      "tokens_seen": 1305214976
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006104959075589793,
      "loss": 0.071,
      "theoretical_loss": 3.5884535713502776,
      "tokens_seen": 1305477120
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006104156636173968,
      "loss": 0.07,
      "theoretical_loss": 3.588388859742622,
      "tokens_seen": 1305739264
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006103354196758145,
      "loss": 0.0677,
      "theoretical_loss": 3.5883241647621933,
      "tokens_seen": 1306001408
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000610255175734232,
      "loss": 0.0699,
      "theoretical_loss": 3.588259486401383,
      "tokens_seen": 1306263552
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006101749317926497,
      "loss": 0.0711,
      "theoretical_loss": 3.5881948246525877,
      "tokens_seen": 1306525696
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006100946878510673,
      "loss": 0.071,
      "theoretical_loss": 3.588130179508209,
      "tokens_seen": 1306787840
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006100144439094848,
      "loss": 0.0716,
      "theoretical_loss": 3.5880655509606534,
      "tokens_seen": 1307049984
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006099341999679025,
      "loss": 0.0734,
      "theoretical_loss": 3.5880009390023324,
      "tokens_seen": 1307312128
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00060985395602632,
      "loss": 0.0694,
      "theoretical_loss": 3.5879363436256626,
      "tokens_seen": 1307574272
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006097737120847376,
      "loss": 0.0722,
      "theoretical_loss": 3.587871764823066,
      "tokens_seen": 1307836416
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006096934681431552,
      "loss": 0.073,
      "theoretical_loss": 3.5878072025869683,
      "tokens_seen": 1308098560
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006096132242015728,
      "loss": 0.0697,
      "theoretical_loss": 3.5877426569098017,
      "tokens_seen": 1308360704
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006095329802599903,
      "loss": 0.069,
      "theoretical_loss": 3.5876781277840024,
      "tokens_seen": 1308622848
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000609452736318408,
      "loss": 0.0712,
      "theoretical_loss": 3.5876136152020117,
      "tokens_seen": 1308884992
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006093724923768256,
      "loss": 0.0715,
      "theoretical_loss": 3.587549119156276,
      "tokens_seen": 1309147136
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006092922484352431,
      "loss": 0.0747,
      "theoretical_loss": 3.5874846396392472,
      "tokens_seen": 1309409280
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006092120044936608,
      "loss": 0.0718,
      "theoretical_loss": 3.5874201766433815,
      "tokens_seen": 1309671424
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006091317605520783,
      "loss": 0.0729,
      "theoretical_loss": 3.58735573016114,
      "tokens_seen": 1309933568
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006090515166104959,
      "loss": 0.0722,
      "theoretical_loss": 3.5872913001849884,
      "tokens_seen": 1310195712
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006089712726689135,
      "loss": 0.0694,
      "theoretical_loss": 3.5872268867073993,
      "tokens_seen": 1310457856
    },
    {
      "debugging/Compilability": 0.9473684210526315,
      "debugging/distinct-1-grams": 0.7438439469491407,
      "debugging/entropy-1-grams": 5.528104507223894,
      "debugging/length": 493.7894736842105,
      "debugging/num_segments": 19,
      "debugging/raw_token_scores_avg": 0.008907916024327278,
      "debugging/raw_token_scores_std": 0.03160863742232323,
      "debugging/score": 0.005095041850402146,
      "debugging/score_std": 0.005375833718357644,
      "epoch": 0.4,
      "objective/train/advantage_avg": 0.0013337053824216127,
      "objective/train/docs_used": 479375,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4477362632751465,
      "objective/train/original_loss": 1.4477365016937256,
      "objective/train/theoretical_loss": 3.587162489720847,
      "objective/train/tokens_used": 1331180000,
      "objective/train/value_avg": -0.0102386474609375,
      "objective/train/value_loss": 0.00042610923992469907,
      "objective/train/value_max": -0.00011235475540161133,
      "objective/train/value_min": -0.8984375,
      "objective/train/value_reward_corr": 0.7585738573996782,
      "objective/train/value_std": 0.0240631103515625,
      "objective/train/weight_avg": 1.0015311241149902,
      "objective/train/weighted_lm_loss": 1.4506300687789917,
      "objective/train/weights_max": 1.8746426105499268,
      "objective/train/weights_min": 0.3877526521682739,
      "theoretical_loss": 3.587162489720847,
      "tokens_seen": 1310720000
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006088910287273311,
      "loss": 0.0695,
      "theoretical_loss": 3.587162489720847,
      "tokens_seen": 1310720000
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006088107847857486,
      "loss": 0.0699,
      "theoretical_loss": 3.5870981092178136,
      "tokens_seen": 1310982144
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006087305408441663,
      "loss": 0.0719,
      "theoretical_loss": 3.5870337451907854,
      "tokens_seen": 1311244288
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006086502969025839,
      "loss": 0.0723,
      "theoretical_loss": 3.5869693976322523,
      "tokens_seen": 1311506432
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006085700529610015,
      "loss": 0.0676,
      "theoretical_loss": 3.586905066534711,
      "tokens_seen": 1311768576
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006084898090194191,
      "loss": 0.0664,
      "theoretical_loss": 3.5868407518906618,
      "tokens_seen": 1312030720
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006084095650778366,
      "loss": 0.0729,
      "theoretical_loss": 3.586776453692611,
      "tokens_seen": 1312292864
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006083293211362543,
      "loss": 0.0704,
      "theoretical_loss": 3.5867121719330677,
      "tokens_seen": 1312555008
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006082490771946718,
      "loss": 0.0711,
      "theoretical_loss": 3.586647906604549,
      "tokens_seen": 1312817152
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006081688332530893,
      "loss": 0.0709,
      "theoretical_loss": 3.5865836576995744,
      "tokens_seen": 1313079296
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000608088589311507,
      "loss": 0.0707,
      "theoretical_loss": 3.5865194252106694,
      "tokens_seen": 1313341440
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006080083453699246,
      "loss": 0.0727,
      "theoretical_loss": 3.586455209130364,
      "tokens_seen": 1313603584
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006079281014283422,
      "loss": 0.0739,
      "theoretical_loss": 3.5863910094511935,
      "tokens_seen": 1313865728
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006078478574867598,
      "loss": 0.0712,
      "theoretical_loss": 3.586326826165698,
      "tokens_seen": 1314127872
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006077676135451774,
      "loss": 0.0713,
      "theoretical_loss": 3.5862626592664215,
      "tokens_seen": 1314390016
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006076873696035949,
      "loss": 0.0703,
      "theoretical_loss": 3.586198508745915,
      "tokens_seen": 1314652160
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006076071256620125,
      "loss": 0.0705,
      "theoretical_loss": 3.586134374596732,
      "tokens_seen": 1314914304
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006075268817204301,
      "loss": 0.0721,
      "theoretical_loss": 3.586070256811432,
      "tokens_seen": 1315176448
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006074466377788477,
      "loss": 0.069,
      "theoretical_loss": 3.58600615538258,
      "tokens_seen": 1315438592
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006073663938372653,
      "loss": 0.0738,
      "theoretical_loss": 3.5859420703027447,
      "tokens_seen": 1315700736
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006072861498956828,
      "loss": 0.0701,
      "theoretical_loss": 3.5858780015644998,
      "tokens_seen": 1315962880
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006072059059541006,
      "loss": 0.0751,
      "theoretical_loss": 3.585813949160425,
      "tokens_seen": 1316225024
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006071256620125181,
      "loss": 0.0699,
      "theoretical_loss": 3.585749913083103,
      "tokens_seen": 1316487168
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006070454180709356,
      "loss": 0.0754,
      "theoretical_loss": 3.5856858933251234,
      "tokens_seen": 1316749312
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006069651741293533,
      "loss": 0.0727,
      "theoretical_loss": 3.5856218898790786,
      "tokens_seen": 1317011456
    },
    {
      "epoch": 0.4,
      "objective/train/advantage_avg": -0.0005380894290283322,
      "objective/train/docs_used": 481887,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4740736484527588,
      "objective/train/original_loss": 1.4740736484527588,
      "objective/train/theoretical_loss": 3.585557902737568,
      "objective/train/tokens_used": 1337733600,
      "objective/train/value_avg": -0.009063720703125,
      "objective/train/value_loss": 0.00029063946567475796,
      "objective/train/value_max": -6.014108657836914e-05,
      "objective/train/value_min": -0.982421875,
      "objective/train/value_reward_corr": 0.906233165609036,
      "objective/train/value_std": 0.030059814453125,
      "objective/train/weight_avg": 0.9995934963226318,
      "objective/train/weighted_lm_loss": 1.4738991260528564,
      "objective/train/weights_max": 1.3234916925430298,
      "objective/train/weights_min": 0.37601205706596375,
      "theoretical_loss": 3.585557902737568,
      "tokens_seen": 1317273600
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006068849301877708,
      "loss": 0.0742,
      "theoretical_loss": 3.585557902737568,
      "tokens_seen": 1317273600
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006068046862461884,
      "loss": 0.077,
      "theoretical_loss": 3.5854939318931933,
      "tokens_seen": 1317535744
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000606724442304606,
      "loss": 0.0721,
      "theoretical_loss": 3.5854299773385634,
      "tokens_seen": 1317797888
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006066441983630236,
      "loss": 0.0726,
      "theoretical_loss": 3.585366039066291,
      "tokens_seen": 1318060032
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006065639544214411,
      "loss": 0.0716,
      "theoretical_loss": 3.585302117068993,
      "tokens_seen": 1318322176
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006064837104798588,
      "loss": 0.072,
      "theoretical_loss": 3.585238211339292,
      "tokens_seen": 1318584320
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006064034665382764,
      "loss": 0.0703,
      "theoretical_loss": 3.5851743218698156,
      "tokens_seen": 1318846464
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006063232225966939,
      "loss": 0.0713,
      "theoretical_loss": 3.585110448653195,
      "tokens_seen": 1319108608
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006062429786551116,
      "loss": 0.0699,
      "theoretical_loss": 3.585046591682068,
      "tokens_seen": 1319370752
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006061627347135291,
      "loss": 0.0718,
      "theoretical_loss": 3.5849827509490746,
      "tokens_seen": 1319632896
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006060824907719468,
      "loss": 0.0718,
      "theoretical_loss": 3.584918926446863,
      "tokens_seen": 1319895040
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006060022468303643,
      "loss": 0.0706,
      "theoretical_loss": 3.5848551181680826,
      "tokens_seen": 1320157184
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006059220028887819,
      "loss": 0.0678,
      "theoretical_loss": 3.5847913261053908,
      "tokens_seen": 1320419328
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006058417589471995,
      "loss": 0.0719,
      "theoretical_loss": 3.584727550251447,
      "tokens_seen": 1320681472
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000605761515005617,
      "loss": 0.0733,
      "theoretical_loss": 3.5846637905989183,
      "tokens_seen": 1320943616
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006056812710640347,
      "loss": 0.0709,
      "theoretical_loss": 3.5846000471404738,
      "tokens_seen": 1321205760
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006056010271224523,
      "loss": 0.0763,
      "theoretical_loss": 3.5845363198687883,
      "tokens_seen": 1321467904
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006055207831808699,
      "loss": 0.0726,
      "theoretical_loss": 3.584472608776542,
      "tokens_seen": 1321730048
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006054405392392874,
      "loss": 0.0732,
      "theoretical_loss": 3.5844089138564197,
      "tokens_seen": 1321992192
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006053602952977051,
      "loss": 0.0693,
      "theoretical_loss": 3.584345235101111,
      "tokens_seen": 1322254336
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006052800513561226,
      "loss": 0.0709,
      "theoretical_loss": 3.584281572503309,
      "tokens_seen": 1322516480
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006051998074145401,
      "loss": 0.0728,
      "theoretical_loss": 3.584217926055713,
      "tokens_seen": 1322778624
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006051195634729578,
      "loss": 0.0734,
      "theoretical_loss": 3.584154295751027,
      "tokens_seen": 1323040768
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006050393195313753,
      "loss": 0.0718,
      "theoretical_loss": 3.5840906815819586,
      "tokens_seen": 1323302912
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006049590755897931,
      "loss": 0.0726,
      "theoretical_loss": 3.584027083541222,
      "tokens_seen": 1323565056
    },
    {
      "epoch": 0.4,
      "objective/train/advantage_avg": 0.00013670318003278226,
      "objective/train/docs_used": 484332,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4303979873657227,
      "objective/train/original_loss": 1.4303977489471436,
      "objective/train/theoretical_loss": 3.583963501621533,
      "objective/train/tokens_used": 1344287200,
      "objective/train/value_avg": -0.00768280029296875,
      "objective/train/value_loss": 0.0003027576894965023,
      "objective/train/value_max": -9.244680404663086e-05,
      "objective/train/value_min": -0.53466796875,
      "objective/train/value_reward_corr": 0.6955805663532895,
      "objective/train/value_std": 0.0156707763671875,
      "objective/train/weight_avg": 1.0002702474594116,
      "objective/train/weighted_lm_loss": 1.4310005903244019,
      "objective/train/weights_max": 1.3974761962890625,
      "objective/train/weights_min": 0.3838087022304535,
      "theoretical_loss": 3.583963501621533,
      "tokens_seen": 1323827200
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006048788316482106,
      "loss": 0.0721,
      "theoretical_loss": 3.583963501621533,
      "tokens_seen": 1323827200
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006047985877066282,
      "loss": 0.0715,
      "theoretical_loss": 3.583899935815616,
      "tokens_seen": 1324089344
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006047183437650458,
      "loss": 0.0715,
      "theoretical_loss": 3.583836386116197,
      "tokens_seen": 1324351488
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006046380998234633,
      "loss": 0.072,
      "theoretical_loss": 3.5837728525160086,
      "tokens_seen": 1324613632
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006045578558818809,
      "loss": 0.0765,
      "theoretical_loss": 3.5837093350077875,
      "tokens_seen": 1324875776
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006044776119402985,
      "loss": 0.0711,
      "theoretical_loss": 3.5836458335842747,
      "tokens_seen": 1325137920
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006043973679987161,
      "loss": 0.0728,
      "theoretical_loss": 3.5835823482382163,
      "tokens_seen": 1325400064
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006043171240571336,
      "loss": 0.0712,
      "theoretical_loss": 3.583518878962364,
      "tokens_seen": 1325662208
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006042368801155514,
      "loss": 0.0708,
      "theoretical_loss": 3.583455425749472,
      "tokens_seen": 1325924352
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006041566361739689,
      "loss": 0.0725,
      "theoretical_loss": 3.583391988592301,
      "tokens_seen": 1326186496
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006040763922323864,
      "loss": 0.0747,
      "theoretical_loss": 3.5833285674836164,
      "tokens_seen": 1326448640
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006039961482908041,
      "loss": 0.0724,
      "theoretical_loss": 3.583265162416187,
      "tokens_seen": 1326710784
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006039159043492216,
      "loss": 0.0722,
      "theoretical_loss": 3.583201773382788,
      "tokens_seen": 1326972928
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006038356604076392,
      "loss": 0.0731,
      "theoretical_loss": 3.583138400376197,
      "tokens_seen": 1327235072
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006037554164660568,
      "loss": 0.071,
      "theoretical_loss": 3.583075043389199,
      "tokens_seen": 1327497216
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006036751725244744,
      "loss": 0.0729,
      "theoretical_loss": 3.583011702414581,
      "tokens_seen": 1327759360
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000603594928582892,
      "loss": 0.0736,
      "theoretical_loss": 3.5829483774451374,
      "tokens_seen": 1328021504
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006035146846413096,
      "loss": 0.0711,
      "theoretical_loss": 3.582885068473665,
      "tokens_seen": 1328283648
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006034344406997272,
      "loss": 0.0724,
      "theoretical_loss": 3.582821775492966,
      "tokens_seen": 1328545792
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006033541967581448,
      "loss": 0.0712,
      "theoretical_loss": 3.5827584984958474,
      "tokens_seen": 1328807936
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006032739528165624,
      "loss": 0.0733,
      "theoretical_loss": 3.582695237475121,
      "tokens_seen": 1329070080
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006031937088749799,
      "loss": 0.0711,
      "theoretical_loss": 3.582631992423603,
      "tokens_seen": 1329332224
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006031134649333976,
      "loss": 0.0709,
      "theoretical_loss": 3.582568763334115,
      "tokens_seen": 1329594368
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006030332209918151,
      "loss": 0.0682,
      "theoretical_loss": 3.582505550199481,
      "tokens_seen": 1329856512
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006029529770502326,
      "loss": 0.0706,
      "theoretical_loss": 3.5824423530125324,
      "tokens_seen": 1330118656
    },
    {
      "epoch": 0.4,
      "objective/train/advantage_avg": 0.0006204114179126918,
      "objective/train/docs_used": 486514,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3864256143569946,
      "objective/train/original_loss": 1.386425495147705,
      "objective/train/theoretical_loss": 3.582379171766104,
      "objective/train/tokens_used": 1350840800,
      "objective/train/value_avg": -0.00786590576171875,
      "objective/train/value_loss": 0.00019436066213529557,
      "objective/train/value_max": -8.219480514526367e-05,
      "objective/train/value_min": -0.68359375,
      "objective/train/value_reward_corr": 0.6674128947141043,
      "objective/train/value_std": 0.0129241943359375,
      "objective/train/weight_avg": 1.0007107257843018,
      "objective/train/weighted_lm_loss": 1.3870501518249512,
      "objective/train/weights_max": 1.2174729108810425,
      "objective/train/weights_min": 0.3684675693511963,
      "theoretical_loss": 3.582379171766104,
      "tokens_seen": 1330380800
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006028727331086503,
      "loss": 0.0716,
      "theoretical_loss": 3.582379171766104,
      "tokens_seen": 1330380800
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006027924891670679,
      "loss": 0.0721,
      "theoretical_loss": 3.582316006453034,
      "tokens_seen": 1330642944
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006027122452254855,
      "loss": 0.072,
      "theoretical_loss": 3.5822528570661683,
      "tokens_seen": 1330905088
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006026320012839031,
      "loss": 0.073,
      "theoretical_loss": 3.582189723598354,
      "tokens_seen": 1331167232
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006025517573423207,
      "loss": 0.0721,
      "theoretical_loss": 3.582126606042446,
      "tokens_seen": 1331429376
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006024715134007382,
      "loss": 0.0719,
      "theoretical_loss": 3.5820635043913005,
      "tokens_seen": 1331691520
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006023912694591559,
      "loss": 0.073,
      "theoretical_loss": 3.582000418637781,
      "tokens_seen": 1331953664
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006023110255175734,
      "loss": 0.0686,
      "theoretical_loss": 3.581937348774755,
      "tokens_seen": 1332215808
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000602230781575991,
      "loss": 0.0723,
      "theoretical_loss": 3.5818742947950932,
      "tokens_seen": 1332477952
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006021505376344086,
      "loss": 0.0726,
      "theoretical_loss": 3.5818112566916724,
      "tokens_seen": 1332740096
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006020702936928261,
      "loss": 0.0708,
      "theoretical_loss": 3.5817482344573746,
      "tokens_seen": 1333002240
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006019900497512439,
      "loss": 0.0727,
      "theoretical_loss": 3.5816852280850835,
      "tokens_seen": 1333264384
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006019098058096614,
      "loss": 0.0704,
      "theoretical_loss": 3.5816222375676903,
      "tokens_seen": 1333526528
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000601829561868079,
      "loss": 0.0721,
      "theoretical_loss": 3.58155926289809,
      "tokens_seen": 1333788672
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006017493179264966,
      "loss": 0.0687,
      "theoretical_loss": 3.581496304069181,
      "tokens_seen": 1334050816
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006016690739849141,
      "loss": 0.0735,
      "theoretical_loss": 3.5814333610738673,
      "tokens_seen": 1334312960
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006015888300433317,
      "loss": 0.0718,
      "theoretical_loss": 3.5813704339050583,
      "tokens_seen": 1334575104
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006015085861017493,
      "loss": 0.0717,
      "theoretical_loss": 3.581307522555666,
      "tokens_seen": 1334837248
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006014283421601669,
      "loss": 0.0693,
      "theoretical_loss": 3.5812446270186085,
      "tokens_seen": 1335099392
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006013480982185844,
      "loss": 0.0709,
      "theoretical_loss": 3.5811817472868075,
      "tokens_seen": 1335361536
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006012678542770022,
      "loss": 0.0703,
      "theoretical_loss": 3.5811188833531897,
      "tokens_seen": 1335623680
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006011876103354197,
      "loss": 0.0681,
      "theoretical_loss": 3.5810560352106866,
      "tokens_seen": 1335885824
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006011073663938372,
      "loss": 0.0685,
      "theoretical_loss": 3.580993202852234,
      "tokens_seen": 1336147968
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0006010271224522549,
      "loss": 0.0675,
      "theoretical_loss": 3.580930386270772,
      "tokens_seen": 1336410112
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0006009468785106724,
      "loss": 0.0716,
      "theoretical_loss": 3.5808675854592464,
      "tokens_seen": 1336672256
    },
    {
      "epoch": 0.41,
      "objective/train/advantage_avg": -0.00022589701984543353,
      "objective/train/docs_used": 488954,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3641271591186523,
      "objective/train/original_loss": 1.3641269207000732,
      "objective/train/theoretical_loss": 3.5808048004106054,
      "objective/train/tokens_used": 1357394400,
      "objective/train/value_avg": -0.006866455078125,
      "objective/train/value_loss": 0.00021092750830575824,
      "objective/train/value_max": -9.608268737792969e-05,
      "objective/train/value_min": -0.333984375,
      "objective/train/value_reward_corr": 0.6062614211122952,
      "objective/train/value_std": 0.0108489990234375,
      "objective/train/weight_avg": 0.9998708367347717,
      "objective/train/weighted_lm_loss": 1.3648322820663452,
      "objective/train/weights_max": 1.2626718282699585,
      "objective/train/weights_min": 0.3777256906032562,
      "theoretical_loss": 3.5808048004106054,
      "tokens_seen": 1336934400
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0006008666345690901,
      "loss": 0.0717,
      "theoretical_loss": 3.5808048004106054,
      "tokens_seen": 1336934400
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0006007863906275076,
      "loss": 0.071,
      "theoretical_loss": 3.5807420311178033,
      "tokens_seen": 1337196544
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0006007061466859252,
      "loss": 0.071,
      "theoretical_loss": 3.580679277573799,
      "tokens_seen": 1337458688
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0006006259027443429,
      "loss": 0.0713,
      "theoretical_loss": 3.5806165397715546,
      "tokens_seen": 1337720832
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0006005456588027604,
      "loss": 0.0719,
      "theoretical_loss": 3.580553817704039,
      "tokens_seen": 1337982976
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000600465414861178,
      "loss": 0.0697,
      "theoretical_loss": 3.580491111364223,
      "tokens_seen": 1338245120
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0006003851709195956,
      "loss": 0.0693,
      "theoretical_loss": 3.5804284207450836,
      "tokens_seen": 1338507264
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0006003049269780132,
      "loss": 0.0682,
      "theoretical_loss": 3.580365745839602,
      "tokens_seen": 1338769408
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0006002246830364307,
      "loss": 0.0719,
      "theoretical_loss": 3.5803030866407637,
      "tokens_seen": 1339031552
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0006001444390948484,
      "loss": 0.0701,
      "theoretical_loss": 3.580240443141559,
      "tokens_seen": 1339293696
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0006000641951532659,
      "loss": 0.0716,
      "theoretical_loss": 3.5801778153349817,
      "tokens_seen": 1339555840
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005999839512116834,
      "loss": 0.0722,
      "theoretical_loss": 3.5801152032140315,
      "tokens_seen": 1339817984
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005999037072701011,
      "loss": 0.0698,
      "theoretical_loss": 3.580052606771712,
      "tokens_seen": 1340080128
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005998234633285187,
      "loss": 0.0697,
      "theoretical_loss": 3.579990026001031,
      "tokens_seen": 1340342272
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005997432193869364,
      "loss": 0.072,
      "theoretical_loss": 3.579927460895002,
      "tokens_seen": 1340604416
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005996629754453539,
      "loss": 0.0739,
      "theoretical_loss": 3.5798649114466405,
      "tokens_seen": 1340866560
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005995827315037715,
      "loss": 0.0703,
      "theoretical_loss": 3.579802377648969,
      "tokens_seen": 1341128704
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005995024875621891,
      "loss": 0.0724,
      "theoretical_loss": 3.579739859495013,
      "tokens_seen": 1341390848
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005994222436206067,
      "loss": 0.072,
      "theoretical_loss": 3.5796773569778026,
      "tokens_seen": 1341652992
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005993419996790242,
      "loss": 0.0713,
      "theoretical_loss": 3.579614870090374,
      "tokens_seen": 1341915136
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005992617557374418,
      "loss": 0.0749,
      "theoretical_loss": 3.5795523988257654,
      "tokens_seen": 1342177280
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005991815117958594,
      "loss": 0.072,
      "theoretical_loss": 3.5794899431770215,
      "tokens_seen": 1342439424
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000599101267854277,
      "loss": 0.0733,
      "theoretical_loss": 3.5794275031371896,
      "tokens_seen": 1342701568
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005990210239126947,
      "loss": 0.0694,
      "theoretical_loss": 3.579365078699323,
      "tokens_seen": 1342963712
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005989407799711122,
      "loss": 0.072,
      "theoretical_loss": 3.579302669856479,
      "tokens_seen": 1343225856
    },
    {
      "epoch": 0.41,
      "objective/train/advantage_avg": -2.97992642117606e-06,
      "objective/train/docs_used": 490931,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4495521783828735,
      "objective/train/original_loss": 1.449552059173584,
      "objective/train/theoretical_loss": 3.5792402766017197,
      "objective/train/tokens_used": 1363948000,
      "objective/train/value_avg": -0.00823211669921875,
      "objective/train/value_loss": 0.0004934009630233049,
      "objective/train/value_max": -8.547306060791016e-05,
      "objective/train/value_min": -0.64599609375,
      "objective/train/value_reward_corr": 0.603769803863347,
      "objective/train/value_std": 0.01506805419921875,
      "objective/train/weight_avg": 1.000206470489502,
      "objective/train/weighted_lm_loss": 1.4500666856765747,
      "objective/train/weights_max": 1.8629971742630005,
      "objective/train/weights_min": 0.3684026002883911,
      "theoretical_loss": 3.5792402766017197,
      "tokens_seen": 1343488000
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005988605360295298,
      "loss": 0.0703,
      "theoretical_loss": 3.5792402766017197,
      "tokens_seen": 1343488000
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005987802920879474,
      "loss": 0.0692,
      "theoretical_loss": 3.57917789892811,
      "tokens_seen": 1343750144
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005987000481463649,
      "loss": 0.0691,
      "theoretical_loss": 3.579115536828721,
      "tokens_seen": 1344012288
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005986198042047825,
      "loss": 0.07,
      "theoretical_loss": 3.5790531902966274,
      "tokens_seen": 1344274432
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005985395602632001,
      "loss": 0.0698,
      "theoretical_loss": 3.578990859324909,
      "tokens_seen": 1344536576
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005984593163216177,
      "loss": 0.0697,
      "theoretical_loss": 3.5789285439066494,
      "tokens_seen": 1344798720
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005983790723800354,
      "loss": 0.0701,
      "theoretical_loss": 3.578866244034937,
      "tokens_seen": 1345060864
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000598298828438453,
      "loss": 0.0733,
      "theoretical_loss": 3.5788039597028636,
      "tokens_seen": 1345323008
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005982185844968705,
      "loss": 0.0677,
      "theoretical_loss": 3.5787416909035272,
      "tokens_seen": 1345585152
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005981383405552881,
      "loss": 0.0707,
      "theoretical_loss": 3.578679437630029,
      "tokens_seen": 1345847296
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005980580966137057,
      "loss": 0.072,
      "theoretical_loss": 3.5786171998754748,
      "tokens_seen": 1346109440
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005979778526721232,
      "loss": 0.0697,
      "theoretical_loss": 3.5785549776329746,
      "tokens_seen": 1346371584
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005978976087305409,
      "loss": 0.0695,
      "theoretical_loss": 3.578492770895643,
      "tokens_seen": 1346633728
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005978173647889584,
      "loss": 0.0674,
      "theoretical_loss": 3.5784305796566,
      "tokens_seen": 1346895872
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000597737120847376,
      "loss": 0.0706,
      "theoretical_loss": 3.5783684039089687,
      "tokens_seen": 1347158016
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005976568769057937,
      "loss": 0.0689,
      "theoretical_loss": 3.578306243645876,
      "tokens_seen": 1347420160
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005975766329642112,
      "loss": 0.0684,
      "theoretical_loss": 3.5782440988604547,
      "tokens_seen": 1347682304
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005974963890226288,
      "loss": 0.0699,
      "theoretical_loss": 3.5781819695458417,
      "tokens_seen": 1347944448
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005974161450810464,
      "loss": 0.0717,
      "theoretical_loss": 3.578119855695178,
      "tokens_seen": 1348206592
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000597335901139464,
      "loss": 0.0701,
      "theoretical_loss": 3.5780577573016084,
      "tokens_seen": 1348468736
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005972556571978816,
      "loss": 0.0726,
      "theoretical_loss": 3.5779956743582835,
      "tokens_seen": 1348730880
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005971754132562992,
      "loss": 0.0705,
      "theoretical_loss": 3.5779336068583563,
      "tokens_seen": 1348993024
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005970951693147167,
      "loss": 0.0716,
      "theoretical_loss": 3.577871554794986,
      "tokens_seen": 1349255168
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005970149253731343,
      "loss": 0.0678,
      "theoretical_loss": 3.5778095181613354,
      "tokens_seen": 1349517312
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000596934681431552,
      "loss": 0.0713,
      "theoretical_loss": 3.577747496950572,
      "tokens_seen": 1349779456
    },
    {
      "epoch": 0.41,
      "objective/train/advantage_avg": 0.0013337216805666685,
      "objective/train/docs_used": 493399,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.339576244354248,
      "objective/train/original_loss": 1.3395761251449585,
      "objective/train/theoretical_loss": 3.577685491155866,
      "objective/train/tokens_used": 1370501600,
      "objective/train/value_avg": -0.00826263427734375,
      "objective/train/value_loss": 0.00026566104497760534,
      "objective/train/value_max": -0.0001252889633178711,
      "objective/train/value_min": -0.6865234375,
      "objective/train/value_reward_corr": 0.5799900901845774,
      "objective/train/value_std": 0.01233673095703125,
      "objective/train/weight_avg": 1.0014503002166748,
      "objective/train/weighted_lm_loss": 1.341780185699463,
      "objective/train/weights_max": 1.6780211925506592,
      "objective/train/weights_min": 0.3697971701622009,
      "theoretical_loss": 3.577685491155866,
      "tokens_seen": 1350041600
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005968544374899695,
      "loss": 0.0717,
      "theoretical_loss": 3.577685491155866,
      "tokens_seen": 1350041600
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005967741935483872,
      "loss": 0.0714,
      "theoretical_loss": 3.577623500770394,
      "tokens_seen": 1350303744
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005966939496068047,
      "loss": 0.0704,
      "theoretical_loss": 3.577561525787337,
      "tokens_seen": 1350565888
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005966137056652223,
      "loss": 0.0696,
      "theoretical_loss": 3.5774995661998785,
      "tokens_seen": 1350828032
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005965334617236399,
      "loss": 0.0701,
      "theoretical_loss": 3.5774376220012085,
      "tokens_seen": 1351090176
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005964532177820575,
      "loss": 0.0732,
      "theoretical_loss": 3.5773756931845186,
      "tokens_seen": 1351352320
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000596372973840475,
      "loss": 0.071,
      "theoretical_loss": 3.5773137797430077,
      "tokens_seen": 1351614464
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005962927298988926,
      "loss": 0.0719,
      "theoretical_loss": 3.577251881669877,
      "tokens_seen": 1351876608
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005962124859573102,
      "loss": 0.0755,
      "theoretical_loss": 3.5771899989583336,
      "tokens_seen": 1352138752
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005961322420157277,
      "loss": 0.0694,
      "theoretical_loss": 3.577128131601587,
      "tokens_seen": 1352400896
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005960519980741455,
      "loss": 0.0678,
      "theoretical_loss": 3.5770662795928527,
      "tokens_seen": 1352663040
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000595971754132563,
      "loss": 0.0681,
      "theoretical_loss": 3.5770044429253494,
      "tokens_seen": 1352925184
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005958915101909807,
      "loss": 0.0701,
      "theoretical_loss": 3.576942621592301,
      "tokens_seen": 1353187328
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005958112662493982,
      "loss": 0.0709,
      "theoretical_loss": 3.576880815586935,
      "tokens_seen": 1353449472
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005957310223078157,
      "loss": 0.0713,
      "theoretical_loss": 3.576819024902483,
      "tokens_seen": 1353711616
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005956507783662334,
      "loss": 0.0739,
      "theoretical_loss": 3.576757249532183,
      "tokens_seen": 1353973760
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005955705344246509,
      "loss": 0.0697,
      "theoretical_loss": 3.576695489469274,
      "tokens_seen": 1354235904
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005954902904830685,
      "loss": 0.0713,
      "theoretical_loss": 3.5766337447070016,
      "tokens_seen": 1354498048
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005954100465414862,
      "loss": 0.0706,
      "theoretical_loss": 3.5765720152386153,
      "tokens_seen": 1354760192
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005953298025999038,
      "loss": 0.0706,
      "theoretical_loss": 3.5765103010573682,
      "tokens_seen": 1355022336
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005952495586583213,
      "loss": 0.071,
      "theoretical_loss": 3.576448602156518,
      "tokens_seen": 1355284480
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005951693147167389,
      "loss": 0.0693,
      "theoretical_loss": 3.5763869185293276,
      "tokens_seen": 1355546624
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005950890707751565,
      "loss": 0.0729,
      "theoretical_loss": 3.576325250169062,
      "tokens_seen": 1355808768
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000595008826833574,
      "loss": 0.0696,
      "theoretical_loss": 3.5762635970689933,
      "tokens_seen": 1356070912
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005949285828919917,
      "loss": 0.0689,
      "theoretical_loss": 3.576201959222396,
      "tokens_seen": 1356333056
    },
    {
      "epoch": 0.41,
      "objective/train/advantage_avg": 0.0003139876062050462,
      "objective/train/docs_used": 495730,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4432828426361084,
      "objective/train/original_loss": 1.443282961845398,
      "objective/train/theoretical_loss": 3.576140336622548,
      "objective/train/tokens_used": 1377055200,
      "objective/train/value_avg": -0.0080413818359375,
      "objective/train/value_loss": 0.0003055589913856238,
      "objective/train/value_max": -7.545948028564453e-05,
      "objective/train/value_min": -0.8916015625,
      "objective/train/value_reward_corr": 0.8504052999798968,
      "objective/train/value_std": 0.0272674560546875,
      "objective/train/weight_avg": 1.0004558563232422,
      "objective/train/weighted_lm_loss": 1.4439336061477661,
      "objective/train/weights_max": 1.514553189277649,
      "objective/train/weights_min": 0.37005260586738586,
      "theoretical_loss": 3.576140336622548,
      "tokens_seen": 1356595200
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005948483389504092,
      "loss": 0.0699,
      "theoretical_loss": 3.576140336622548,
      "tokens_seen": 1356595200
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005947680950088269,
      "loss": 0.0727,
      "theoretical_loss": 3.5760787292627345,
      "tokens_seen": 1356857344
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005946878510672444,
      "loss": 0.0728,
      "theoretical_loss": 3.576017137136242,
      "tokens_seen": 1357119488
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000594607607125662,
      "loss": 0.0709,
      "theoretical_loss": 3.5759555602363635,
      "tokens_seen": 1357381632
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005945273631840797,
      "loss": 0.068,
      "theoretical_loss": 3.5758939985563942,
      "tokens_seen": 1357643776
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005944471192424972,
      "loss": 0.0709,
      "theoretical_loss": 3.5758324520896347,
      "tokens_seen": 1357905920
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005943668753009148,
      "loss": 0.0719,
      "theoretical_loss": 3.57577092082939,
      "tokens_seen": 1358168064
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005942866313593324,
      "loss": 0.0709,
      "theoretical_loss": 3.5757094047689684,
      "tokens_seen": 1358430208
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00059420638741775,
      "loss": 0.0696,
      "theoretical_loss": 3.575647903901684,
      "tokens_seen": 1358692352
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005941261434761675,
      "loss": 0.0713,
      "theoretical_loss": 3.575586418220853,
      "tokens_seen": 1358954496
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005940458995345851,
      "loss": 0.0717,
      "theoretical_loss": 3.5755249477197983,
      "tokens_seen": 1359216640
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005939656555930027,
      "loss": 0.0695,
      "theoretical_loss": 3.5754634923918447,
      "tokens_seen": 1359478784
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005938854116514203,
      "loss": 0.0694,
      "theoretical_loss": 3.5754020522303227,
      "tokens_seen": 1359740928
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000593805167709838,
      "loss": 0.0697,
      "theoretical_loss": 3.575340627228566,
      "tokens_seen": 1360003072
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005937249237682555,
      "loss": 0.0711,
      "theoretical_loss": 3.575279217379914,
      "tokens_seen": 1360265216
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005936446798266731,
      "loss": 0.0737,
      "theoretical_loss": 3.575217822677709,
      "tokens_seen": 1360527360
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005935644358850907,
      "loss": 0.0698,
      "theoretical_loss": 3.575156443115297,
      "tokens_seen": 1360789504
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005934841919435082,
      "loss": 0.0699,
      "theoretical_loss": 3.5750950786860307,
      "tokens_seen": 1361051648
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005934039480019259,
      "loss": 0.0703,
      "theoretical_loss": 3.5750337293832644,
      "tokens_seen": 1361313792
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005933237040603434,
      "loss": 0.0682,
      "theoretical_loss": 3.5749723952003576,
      "tokens_seen": 1361575936
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000593243460118761,
      "loss": 0.0742,
      "theoretical_loss": 3.5749110761306744,
      "tokens_seen": 1361838080
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005931632161771787,
      "loss": 0.0693,
      "theoretical_loss": 3.5748497721675823,
      "tokens_seen": 1362100224
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005930829722355963,
      "loss": 0.0723,
      "theoretical_loss": 3.574788483304453,
      "tokens_seen": 1362362368
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005930027282940138,
      "loss": 0.07,
      "theoretical_loss": 3.5747272095346636,
      "tokens_seen": 1362624512
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005929224843524315,
      "loss": 0.0717,
      "theoretical_loss": 3.5746659508515943,
      "tokens_seen": 1362886656
    },
    {
      "epoch": 0.41,
      "objective/train/advantage_avg": 0.0005423167604021728,
      "objective/train/docs_used": 498147,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3880330324172974,
      "objective/train/original_loss": 1.3880329132080078,
      "objective/train/theoretical_loss": 3.5746047072486293,
      "objective/train/tokens_used": 1383608800,
      "objective/train/value_avg": -0.00732421875,
      "objective/train/value_loss": 0.0004709110071416944,
      "objective/train/value_max": -7.724761962890625e-05,
      "objective/train/value_min": -0.80078125,
      "objective/train/value_reward_corr": 0.6727173948210152,
      "objective/train/value_std": 0.0166168212890625,
      "objective/train/weight_avg": 1.0007424354553223,
      "objective/train/weighted_lm_loss": 1.3886373043060303,
      "objective/train/weights_max": 1.4593173265457153,
      "objective/train/weights_min": 0.3683730661869049,
      "theoretical_loss": 3.5746047072486293,
      "tokens_seen": 1363148800
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000592842240410849,
      "loss": 0.0711,
      "theoretical_loss": 3.5746047072486293,
      "tokens_seen": 1363148800
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005927619964692665,
      "loss": 0.0691,
      "theoretical_loss": 3.574543478719158,
      "tokens_seen": 1363410944
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005926817525276842,
      "loss": 0.07,
      "theoretical_loss": 3.5744822652565724,
      "tokens_seen": 1363673088
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005926015085861017,
      "loss": 0.0727,
      "theoretical_loss": 3.5744210668542706,
      "tokens_seen": 1363935232
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005925212646445193,
      "loss": 0.0721,
      "theoretical_loss": 3.574359883505653,
      "tokens_seen": 1364197376
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000592441020702937,
      "loss": 0.0725,
      "theoretical_loss": 3.5742987152041255,
      "tokens_seen": 1364459520
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005923607767613546,
      "loss": 0.0688,
      "theoretical_loss": 3.574237561943098,
      "tokens_seen": 1364721664
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005922805328197721,
      "loss": 0.0713,
      "theoretical_loss": 3.5741764237159837,
      "tokens_seen": 1364983808
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005922002888781897,
      "loss": 0.0721,
      "theoretical_loss": 3.5741153005162003,
      "tokens_seen": 1365245952
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005921200449366073,
      "loss": 0.0729,
      "theoretical_loss": 3.57405419233717,
      "tokens_seen": 1365508096
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005920398009950249,
      "loss": 0.071,
      "theoretical_loss": 3.5739930991723194,
      "tokens_seen": 1365770240
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005919595570534425,
      "loss": 0.0735,
      "theoretical_loss": 3.5739320210150787,
      "tokens_seen": 1366032384
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00059187931311186,
      "loss": 0.0714,
      "theoretical_loss": 3.5738709578588814,
      "tokens_seen": 1366294528
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005917990691702777,
      "loss": 0.0687,
      "theoretical_loss": 3.573809909697167,
      "tokens_seen": 1366556672
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005917188252286952,
      "loss": 0.07,
      "theoretical_loss": 3.573748876523379,
      "tokens_seen": 1366818816
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005916385812871128,
      "loss": 0.0717,
      "theoretical_loss": 3.5736878583309624,
      "tokens_seen": 1367080960
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005915583373455305,
      "loss": 0.0716,
      "theoretical_loss": 3.5736268551133694,
      "tokens_seen": 1367343104
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000591478093403948,
      "loss": 0.0725,
      "theoretical_loss": 3.5735658668640538,
      "tokens_seen": 1367605248
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005913978494623656,
      "loss": 0.0698,
      "theoretical_loss": 3.573504893576476,
      "tokens_seen": 1367867392
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005913176055207832,
      "loss": 0.0698,
      "theoretical_loss": 3.573443935244099,
      "tokens_seen": 1368129536
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005912373615792008,
      "loss": 0.0702,
      "theoretical_loss": 3.5733829918603903,
      "tokens_seen": 1368391680
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005911571176376183,
      "loss": 0.0728,
      "theoretical_loss": 3.573322063418821,
      "tokens_seen": 1368653824
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005910768736960359,
      "loss": 0.0712,
      "theoretical_loss": 3.5732611499128666,
      "tokens_seen": 1368915968
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005909966297544535,
      "loss": 0.07,
      "theoretical_loss": 3.5732002513360075,
      "tokens_seen": 1369178112
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0005909163858128712,
      "loss": 0.0698,
      "theoretical_loss": 3.5731393676817267,
      "tokens_seen": 1369440256
    },
    {
      "epoch": 0.41,
      "objective/train/advantage_avg": 0.0006691030575893819,
      "objective/train/docs_used": 500422,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.462990403175354,
      "objective/train/original_loss": 1.462990164756775,
      "objective/train/theoretical_loss": 3.573078498943513,
      "objective/train/tokens_used": 1390162400,
      "objective/train/value_avg": -0.01076507568359375,
      "objective/train/value_loss": 0.00031313643557950854,
      "objective/train/value_max": -6.812810897827148e-05,
      "objective/train/value_min": -0.68896484375,
      "objective/train/value_reward_corr": 0.7292926677177232,
      "objective/train/value_std": 0.0184326171875,
      "objective/train/weight_avg": 1.0008124113082886,
      "objective/train/weighted_lm_loss": 1.4639732837677002,
      "objective/train/weights_max": 1.8267512321472168,
      "objective/train/weights_min": 0.37275585532188416,
      "theoretical_loss": 3.573078498943513,
      "tokens_seen": 1369702400
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005908361418712888,
      "loss": 0.0711,
      "theoretical_loss": 3.573078498943513,
      "tokens_seen": 1369702400
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005907558979297063,
      "loss": 0.0716,
      "theoretical_loss": 3.5730176451148568,
      "tokens_seen": 1369964544
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000590675653988124,
      "loss": 0.0698,
      "theoretical_loss": 3.572956806189256,
      "tokens_seen": 1370226688
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005905954100465415,
      "loss": 0.0707,
      "theoretical_loss": 3.5728959821602095,
      "tokens_seen": 1370488832
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000590515166104959,
      "loss": 0.0699,
      "theoretical_loss": 3.5728351730212218,
      "tokens_seen": 1370750976
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005904349221633767,
      "loss": 0.0707,
      "theoretical_loss": 3.5727743787658017,
      "tokens_seen": 1371013120
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005903546782217942,
      "loss": 0.0747,
      "theoretical_loss": 3.572713599387461,
      "tokens_seen": 1371275264
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005902744342802118,
      "loss": 0.0695,
      "theoretical_loss": 3.572652834879716,
      "tokens_seen": 1371537408
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005901941903386295,
      "loss": 0.0691,
      "theoretical_loss": 3.5725920852360877,
      "tokens_seen": 1371799552
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005901139463970471,
      "loss": 0.0688,
      "theoretical_loss": 3.5725313504501006,
      "tokens_seen": 1372061696
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005900337024554646,
      "loss": 0.0725,
      "theoretical_loss": 3.5724706305152827,
      "tokens_seen": 1372323840
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005899534585138823,
      "loss": 0.0667,
      "theoretical_loss": 3.572409925425167,
      "tokens_seen": 1372585984
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005898732145722998,
      "loss": 0.0709,
      "theoretical_loss": 3.5723492351732906,
      "tokens_seen": 1372848128
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005897929706307173,
      "loss": 0.0702,
      "theoretical_loss": 3.572288559753194,
      "tokens_seen": 1373110272
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000589712726689135,
      "loss": 0.0747,
      "theoretical_loss": 3.5722278991584218,
      "tokens_seen": 1373372416
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005896324827475525,
      "loss": 0.071,
      "theoretical_loss": 3.572167253382523,
      "tokens_seen": 1373634560
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005895522388059702,
      "loss": 0.0739,
      "theoretical_loss": 3.5721066224190503,
      "tokens_seen": 1373896704
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005894719948643878,
      "loss": 0.0702,
      "theoretical_loss": 3.572046006261561,
      "tokens_seen": 1374158848
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005893917509228054,
      "loss": 0.0677,
      "theoretical_loss": 3.5719854049036153,
      "tokens_seen": 1374420992
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000589311506981223,
      "loss": 0.0692,
      "theoretical_loss": 3.571924818338779,
      "tokens_seen": 1374683136
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005892312630396405,
      "loss": 0.0685,
      "theoretical_loss": 3.5718642465606214,
      "tokens_seen": 1374945280
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005891510190980581,
      "loss": 0.0684,
      "theoretical_loss": 3.571803689562714,
      "tokens_seen": 1375207424
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005890707751564757,
      "loss": 0.0722,
      "theoretical_loss": 3.571743147338635,
      "tokens_seen": 1375469568
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005889905312148933,
      "loss": 0.0718,
      "theoretical_loss": 3.5716826198819653,
      "tokens_seen": 1375731712
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005889102872733108,
      "loss": 0.0732,
      "theoretical_loss": 3.57162210718629,
      "tokens_seen": 1375993856
    },
    {
      "epoch": 0.42,
      "objective/train/advantage_avg": 0.001140189473517239,
      "objective/train/docs_used": 502650,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.329345464706421,
      "objective/train/original_loss": 1.329345464706421,
      "objective/train/theoretical_loss": 3.5715616092451983,
      "objective/train/tokens_used": 1396716000,
      "objective/train/value_avg": -0.009979248046875,
      "objective/train/value_loss": 0.0002272567362524569,
      "objective/train/value_max": -6.35385513305664e-05,
      "objective/train/value_min": -0.2432861328125,
      "objective/train/value_reward_corr": 0.7523422487981606,
      "objective/train/value_std": 0.01526641845703125,
      "objective/train/weight_avg": 1.0012439489364624,
      "objective/train/weighted_lm_loss": 1.3301997184753418,
      "objective/train/weights_max": 1.2260957956314087,
      "objective/train/weights_min": 0.3730090260505676,
      "theoretical_loss": 3.5715616092451983,
      "tokens_seen": 1376256000
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005888300433317285,
      "loss": 0.0693,
      "theoretical_loss": 3.5715616092451983,
      "tokens_seen": 1376256000
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000588749799390146,
      "loss": 0.0676,
      "theoretical_loss": 3.5715011260522824,
      "tokens_seen": 1376518144
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005886695554485636,
      "loss": 0.0701,
      "theoretical_loss": 3.57144065760114,
      "tokens_seen": 1376780288
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005885893115069813,
      "loss": 0.071,
      "theoretical_loss": 3.5713802038853726,
      "tokens_seen": 1377042432
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005885090675653988,
      "loss": 0.0677,
      "theoretical_loss": 3.5713197648985844,
      "tokens_seen": 1377304576
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005884288236238165,
      "loss": 0.0722,
      "theoretical_loss": 3.571259340634385,
      "tokens_seen": 1377566720
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000588348579682234,
      "loss": 0.0681,
      "theoretical_loss": 3.5711989310863874,
      "tokens_seen": 1377828864
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005882683357406516,
      "loss": 0.0663,
      "theoretical_loss": 3.571138536248209,
      "tokens_seen": 1378091008
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005881880917990692,
      "loss": 0.0715,
      "theoretical_loss": 3.57107815611347,
      "tokens_seen": 1378353152
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005881078478574867,
      "loss": 0.0718,
      "theoretical_loss": 3.571017790675796,
      "tokens_seen": 1378615296
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005880276039159043,
      "loss": 0.0712,
      "theoretical_loss": 3.570957439928815,
      "tokens_seen": 1378877440
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000587947359974322,
      "loss": 0.0708,
      "theoretical_loss": 3.5708971038661614,
      "tokens_seen": 1379139584
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005878671160327396,
      "loss": 0.0693,
      "theoretical_loss": 3.5708367824814715,
      "tokens_seen": 1379401728
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005877868720911571,
      "loss": 0.0706,
      "theoretical_loss": 3.570776475768386,
      "tokens_seen": 1379663872
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005877066281495748,
      "loss": 0.0715,
      "theoretical_loss": 3.57071618372055,
      "tokens_seen": 1379926016
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005876263842079923,
      "loss": 0.0719,
      "theoretical_loss": 3.570655906331612,
      "tokens_seen": 1380188160
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005875461402664098,
      "loss": 0.0708,
      "theoretical_loss": 3.570595643595225,
      "tokens_seen": 1380450304
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005874658963248275,
      "loss": 0.0709,
      "theoretical_loss": 3.570535395505045,
      "tokens_seen": 1380712448
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000587385652383245,
      "loss": 0.0682,
      "theoretical_loss": 3.570475162054734,
      "tokens_seen": 1380974592
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005873054084416626,
      "loss": 0.0705,
      "theoretical_loss": 3.570414943237956,
      "tokens_seen": 1381236736
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005872251645000803,
      "loss": 0.0705,
      "theoretical_loss": 3.570354739048379,
      "tokens_seen": 1381498880
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005871449205584979,
      "loss": 0.0694,
      "theoretical_loss": 3.5702945494796765,
      "tokens_seen": 1381761024
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005870646766169155,
      "loss": 0.0708,
      "theoretical_loss": 3.5702343745255236,
      "tokens_seen": 1382023168
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005869844326753331,
      "loss": 0.0707,
      "theoretical_loss": 3.5701742141796022,
      "tokens_seen": 1382285312
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005869041887337506,
      "loss": 0.0684,
      "theoretical_loss": 3.570114068435595,
      "tokens_seen": 1382547456
    },
    {
      "epoch": 0.42,
      "objective/train/advantage_avg": 0.0004002986242994666,
      "objective/train/docs_used": 505055,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4260497093200684,
      "objective/train/original_loss": 1.4260497093200684,
      "objective/train/theoretical_loss": 3.570053937287192,
      "objective/train/tokens_used": 1403269600,
      "objective/train/value_avg": -0.01218414306640625,
      "objective/train/value_loss": 0.00020151071657892317,
      "objective/train/value_max": -9.244680404663086e-05,
      "objective/train/value_min": -0.576171875,
      "objective/train/value_reward_corr": 0.8434513792631848,
      "objective/train/value_std": 0.022979736328125,
      "objective/train/weight_avg": 1.0004942417144775,
      "objective/train/weighted_lm_loss": 1.4266375303268433,
      "objective/train/weights_max": 1.1901416778564453,
      "objective/train/weights_min": 0.3686957061290741,
      "theoretical_loss": 3.570053937287192,
      "tokens_seen": 1382809600
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005868239447921682,
      "loss": 0.0716,
      "theoretical_loss": 3.570053937287192,
      "tokens_seen": 1382809600
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005867437008505858,
      "loss": 0.07,
      "theoretical_loss": 3.569993820728084,
      "tokens_seen": 1383071744
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005866634569090033,
      "loss": 0.0686,
      "theoretical_loss": 3.569933718751967,
      "tokens_seen": 1383333888
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000586583212967421,
      "loss": 0.0708,
      "theoretical_loss": 3.569873631352542,
      "tokens_seen": 1383596032
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005865029690258386,
      "loss": 0.0705,
      "theoretical_loss": 3.5698135585235122,
      "tokens_seen": 1383858176
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005864227250842562,
      "loss": 0.0684,
      "theoretical_loss": 3.5697535002585856,
      "tokens_seen": 1384120320
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005863424811426738,
      "loss": 0.0703,
      "theoretical_loss": 3.569693456551474,
      "tokens_seen": 1384382464
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005862622372010913,
      "loss": 0.0721,
      "theoretical_loss": 3.5696334273958925,
      "tokens_seen": 1384644608
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005861819932595089,
      "loss": 0.0702,
      "theoretical_loss": 3.569573412785561,
      "tokens_seen": 1384906752
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005861017493179265,
      "loss": 0.0719,
      "theoretical_loss": 3.569513412714203,
      "tokens_seen": 1385168896
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005860215053763441,
      "loss": 0.073,
      "theoretical_loss": 3.569453427175546,
      "tokens_seen": 1385431040
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005859412614347616,
      "loss": 0.0689,
      "theoretical_loss": 3.5693934561633203,
      "tokens_seen": 1385693184
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005858610174931793,
      "loss": 0.0692,
      "theoretical_loss": 3.5693334996712625,
      "tokens_seen": 1385955328
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005857807735515968,
      "loss": 0.0708,
      "theoretical_loss": 3.5692735576931103,
      "tokens_seen": 1386217472
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005857005296100145,
      "loss": 0.0675,
      "theoretical_loss": 3.569213630222607,
      "tokens_seen": 1386479616
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005856202856684321,
      "loss": 0.07,
      "theoretical_loss": 3.5691537172535,
      "tokens_seen": 1386741760
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005855400417268496,
      "loss": 0.0676,
      "theoretical_loss": 3.569093818779539,
      "tokens_seen": 1387003904
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005854597977852673,
      "loss": 0.0737,
      "theoretical_loss": 3.5690339347944784,
      "tokens_seen": 1387266048
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005853795538436848,
      "loss": 0.0684,
      "theoretical_loss": 3.568974065292077,
      "tokens_seen": 1387528192
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005852993099021024,
      "loss": 0.0723,
      "theoretical_loss": 3.5689142102660973,
      "tokens_seen": 1387790336
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00058521906596052,
      "loss": 0.0679,
      "theoretical_loss": 3.568854369710305,
      "tokens_seen": 1388052480
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005851388220189375,
      "loss": 0.067,
      "theoretical_loss": 3.5687945436184703,
      "tokens_seen": 1388314624
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005850585780773551,
      "loss": 0.0702,
      "theoretical_loss": 3.5687347319843665,
      "tokens_seen": 1388576768
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005849783341357728,
      "loss": 0.0707,
      "theoretical_loss": 3.5686749348017726,
      "tokens_seen": 1388838912
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005848980901941904,
      "loss": 0.0717,
      "theoretical_loss": 3.5686151520644684,
      "tokens_seen": 1389101056
    },
    {
      "epoch": 0.42,
      "objective/train/advantage_avg": 0.0012033990351483226,
      "objective/train/docs_used": 507487,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3310184478759766,
      "objective/train/original_loss": 1.3310184478759766,
      "objective/train/theoretical_loss": 3.56855538376624,
      "objective/train/tokens_used": 1409823200,
      "objective/train/value_avg": -0.006999969482421875,
      "objective/train/value_loss": 9.323181438958272e-05,
      "objective/train/value_max": -7.599592208862305e-05,
      "objective/train/value_min": -0.252685546875,
      "objective/train/value_reward_corr": 0.81285075839567,
      "objective/train/value_std": 0.01432037353515625,
      "objective/train/weight_avg": 1.0012476444244385,
      "objective/train/weighted_lm_loss": 1.3323184251785278,
      "objective/train/weights_max": 1.1576061248779297,
      "objective/train/weights_min": 0.3713989555835724,
      "theoretical_loss": 3.56855538376624,
      "tokens_seen": 1389363200
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005848178462526079,
      "loss": 0.0704,
      "theoretical_loss": 3.56855538376624,
      "tokens_seen": 1389363200
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005847376023110256,
      "loss": 0.0683,
      "theoretical_loss": 3.568495629900877,
      "tokens_seen": 1389625344
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005846573583694431,
      "loss": 0.0729,
      "theoretical_loss": 3.5684358904621725,
      "tokens_seen": 1389887488
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005845771144278606,
      "loss": 0.0675,
      "theoretical_loss": 3.5683761654439223,
      "tokens_seen": 1390149632
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005844968704862783,
      "loss": 0.067,
      "theoretical_loss": 3.5683164548399287,
      "tokens_seen": 1390411776
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005844166265446958,
      "loss": 0.0721,
      "theoretical_loss": 3.568256758643995,
      "tokens_seen": 1390673920
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005843363826031135,
      "loss": 0.0669,
      "theoretical_loss": 3.5681970768499305,
      "tokens_seen": 1390936064
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000584256138661531,
      "loss": 0.0683,
      "theoretical_loss": 3.5681374094515466,
      "tokens_seen": 1391198208
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005841758947199487,
      "loss": 0.0698,
      "theoretical_loss": 3.5680777564426602,
      "tokens_seen": 1391460352
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005840956507783663,
      "loss": 0.0667,
      "theoretical_loss": 3.56801811781709,
      "tokens_seen": 1391722496
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005840154068367838,
      "loss": 0.0687,
      "theoretical_loss": 3.5679584935686615,
      "tokens_seen": 1391984640
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005839351628952014,
      "loss": 0.068,
      "theoretical_loss": 3.5678988836912007,
      "tokens_seen": 1392246784
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000583854918953619,
      "loss": 0.0691,
      "theoretical_loss": 3.567839288178539,
      "tokens_seen": 1392508928
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005837746750120366,
      "loss": 0.0703,
      "theoretical_loss": 3.5677797070245125,
      "tokens_seen": 1392771072
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005836944310704541,
      "loss": 0.0688,
      "theoretical_loss": 3.567720140222959,
      "tokens_seen": 1393033216
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005836141871288718,
      "loss": 0.0713,
      "theoretical_loss": 3.567660587767722,
      "tokens_seen": 1393295360
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005835339431872894,
      "loss": 0.0693,
      "theoretical_loss": 3.567601049652648,
      "tokens_seen": 1393557504
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000583453699245707,
      "loss": 0.0689,
      "theoretical_loss": 3.567541525871587,
      "tokens_seen": 1393819648
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005833734553041246,
      "loss": 0.0691,
      "theoretical_loss": 3.5674820164183934,
      "tokens_seen": 1394081792
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005832932113625421,
      "loss": 0.0717,
      "theoretical_loss": 3.567422521286925,
      "tokens_seen": 1394343936
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005832129674209598,
      "loss": 0.071,
      "theoretical_loss": 3.5673630404710432,
      "tokens_seen": 1394606080
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005831327234793773,
      "loss": 0.0704,
      "theoretical_loss": 3.567303573964614,
      "tokens_seen": 1394868224
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005830524795377949,
      "loss": 0.0704,
      "theoretical_loss": 3.5672441217615063,
      "tokens_seen": 1395130368
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005829722355962125,
      "loss": 0.0732,
      "theoretical_loss": 3.5671846838555936,
      "tokens_seen": 1395392512
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005828919916546301,
      "loss": 0.0706,
      "theoretical_loss": 3.567125260240752,
      "tokens_seen": 1395654656
    },
    {
      "epoch": 0.42,
      "objective/train/advantage_avg": -0.0006299561937339604,
      "objective/train/docs_used": 509897,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3458755016326904,
      "objective/train/original_loss": 1.3458755016326904,
      "objective/train/theoretical_loss": 3.5670658509108626,
      "objective/train/tokens_used": 1416376800,
      "objective/train/value_avg": -0.006847381591796875,
      "objective/train/value_loss": 0.0002548126794863492,
      "objective/train/value_max": -9.763240814208984e-05,
      "objective/train/value_min": -0.53857421875,
      "objective/train/value_reward_corr": 0.7935790469163092,
      "objective/train/value_std": 0.016387939453125,
      "objective/train/weight_avg": 0.9994886517524719,
      "objective/train/weighted_lm_loss": 1.345136046409607,
      "objective/train/weights_max": 1.3395164012908936,
      "objective/train/weights_min": 0.3683066666126251,
      "theoretical_loss": 3.5670658509108626,
      "tokens_seen": 1395916800
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005828117477130476,
      "loss": 0.0683,
      "theoretical_loss": 3.5670658509108626,
      "tokens_seen": 1395916800
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005827315037714653,
      "loss": 0.0726,
      "theoretical_loss": 3.5670064558598096,
      "tokens_seen": 1396178944
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005826512598298829,
      "loss": 0.0703,
      "theoretical_loss": 3.5669470750814813,
      "tokens_seen": 1396441088
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005825710158883004,
      "loss": 0.0712,
      "theoretical_loss": 3.5668877085697694,
      "tokens_seen": 1396703232
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005824907719467181,
      "loss": 0.073,
      "theoretical_loss": 3.5668283563185703,
      "tokens_seen": 1396965376
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005824105280051356,
      "loss": 0.0695,
      "theoretical_loss": 3.566769018321782,
      "tokens_seen": 1397227520
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005823302840635532,
      "loss": 0.0731,
      "theoretical_loss": 3.5667096945733086,
      "tokens_seen": 1397489664
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005822500401219708,
      "loss": 0.0695,
      "theoretical_loss": 3.566650385067057,
      "tokens_seen": 1397751808
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005821697961803883,
      "loss": 0.0699,
      "theoretical_loss": 3.5665910897969377,
      "tokens_seen": 1398013952
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005820895522388059,
      "loss": 0.0724,
      "theoretical_loss": 3.5665318087568645,
      "tokens_seen": 1398276096
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005820093082972236,
      "loss": 0.071,
      "theoretical_loss": 3.5664725419407564,
      "tokens_seen": 1398538240
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005819290643556412,
      "loss": 0.0704,
      "theoretical_loss": 3.566413289342535,
      "tokens_seen": 1398800384
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005818488204140588,
      "loss": 0.0702,
      "theoretical_loss": 3.566354050956126,
      "tokens_seen": 1399062528
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005817685764724764,
      "loss": 0.0721,
      "theoretical_loss": 3.566294826775459,
      "tokens_seen": 1399324672
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005816883325308939,
      "loss": 0.0683,
      "theoretical_loss": 3.566235616794466,
      "tokens_seen": 1399586816
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005816080885893115,
      "loss": 0.0697,
      "theoretical_loss": 3.566176421007085,
      "tokens_seen": 1399848960
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005815278446477291,
      "loss": 0.0691,
      "theoretical_loss": 3.566117239407256,
      "tokens_seen": 1400111104
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005814476007061466,
      "loss": 0.0718,
      "theoretical_loss": 3.5660580719889237,
      "tokens_seen": 1400373248
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005813673567645643,
      "loss": 0.0727,
      "theoretical_loss": 3.5659989187460353,
      "tokens_seen": 1400635392
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005812871128229819,
      "loss": 0.0741,
      "theoretical_loss": 3.5659397796725427,
      "tokens_seen": 1400897536
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005812068688813995,
      "loss": 0.067,
      "theoretical_loss": 3.565880654762402,
      "tokens_seen": 1401159680
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005811266249398171,
      "loss": 0.0684,
      "theoretical_loss": 3.5658215440095717,
      "tokens_seen": 1401421824
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005810463809982346,
      "loss": 0.0698,
      "theoretical_loss": 3.565762447408015,
      "tokens_seen": 1401683968
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005809661370566522,
      "loss": 0.0722,
      "theoretical_loss": 3.5657033649516974,
      "tokens_seen": 1401946112
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0005808858931150698,
      "loss": 0.0693,
      "theoretical_loss": 3.5656442966345905,
      "tokens_seen": 1402208256
    },
    {
      "epoch": 0.42,
      "objective/train/advantage_avg": 0.00029971322510391474,
      "objective/train/docs_used": 512300,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5333430767059326,
      "objective/train/original_loss": 1.5333430767059326,
      "objective/train/theoretical_loss": 3.565585242450667,
      "objective/train/tokens_used": 1422930400,
      "objective/train/value_avg": -0.00736236572265625,
      "objective/train/value_loss": 0.00021842159912921488,
      "objective/train/value_max": -7.033348083496094e-05,
      "objective/train/value_min": -0.68017578125,
      "objective/train/value_reward_corr": 0.717017729068119,
      "objective/train/value_std": 0.01406097412109375,
      "objective/train/weight_avg": 1.0004018545150757,
      "objective/train/weighted_lm_loss": 1.5342278480529785,
      "objective/train/weights_max": 1.4047281742095947,
      "objective/train/weights_min": 0.3692037761211395,
      "theoretical_loss": 3.565585242450667,
      "tokens_seen": 1402470400
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005808056491734874,
      "loss": 0.0744,
      "theoretical_loss": 3.565585242450667,
      "tokens_seen": 1402470400
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000580725405231905,
      "loss": 0.0715,
      "theoretical_loss": 3.5655262023939054,
      "tokens_seen": 1402732544
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005806451612903226,
      "loss": 0.0688,
      "theoretical_loss": 3.5654671764582866,
      "tokens_seen": 1402994688
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005805649173487401,
      "loss": 0.0703,
      "theoretical_loss": 3.5654081646377955,
      "tokens_seen": 1403256832
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005804846734071579,
      "loss": 0.0705,
      "theoretical_loss": 3.5653491669264215,
      "tokens_seen": 1403518976
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005804044294655754,
      "loss": 0.0709,
      "theoretical_loss": 3.565290183318156,
      "tokens_seen": 1403781120
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005803241855239929,
      "loss": 0.0706,
      "theoretical_loss": 3.565231213806995,
      "tokens_seen": 1404043264
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005802439415824106,
      "loss": 0.0692,
      "theoretical_loss": 3.5651722583869394,
      "tokens_seen": 1404305408
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005801636976408281,
      "loss": 0.0723,
      "theoretical_loss": 3.565113317051991,
      "tokens_seen": 1404567552
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005800834536992457,
      "loss": 0.07,
      "theoretical_loss": 3.5650543897961584,
      "tokens_seen": 1404829696
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005800032097576633,
      "loss": 0.0722,
      "theoretical_loss": 3.5649954766134515,
      "tokens_seen": 1405091840
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005799229658160809,
      "loss": 0.0725,
      "theoretical_loss": 3.5649365774978845,
      "tokens_seen": 1405353984
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005798427218744984,
      "loss": 0.0698,
      "theoretical_loss": 3.5648776924434755,
      "tokens_seen": 1405616128
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005797624779329161,
      "loss": 0.0725,
      "theoretical_loss": 3.5648188214442467,
      "tokens_seen": 1405878272
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005796822339913337,
      "loss": 0.0734,
      "theoretical_loss": 3.5647599644942227,
      "tokens_seen": 1406140416
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005796019900497512,
      "loss": 0.0713,
      "theoretical_loss": 3.564701121587434,
      "tokens_seen": 1406402560
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005795217461081689,
      "loss": 0.067,
      "theoretical_loss": 3.5646422927179113,
      "tokens_seen": 1406664704
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005794415021665864,
      "loss": 0.0693,
      "theoretical_loss": 3.564583477879692,
      "tokens_seen": 1406926848
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005793612582250041,
      "loss": 0.0707,
      "theoretical_loss": 3.5645246770668164,
      "tokens_seen": 1407188992
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005792810142834216,
      "loss": 0.0717,
      "theoretical_loss": 3.5644658902733273,
      "tokens_seen": 1407451136
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005792007703418391,
      "loss": 0.0714,
      "theoretical_loss": 3.564407117493272,
      "tokens_seen": 1407713280
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005791205264002569,
      "loss": 0.0696,
      "theoretical_loss": 3.564348358720702,
      "tokens_seen": 1407975424
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005790402824586744,
      "loss": 0.0734,
      "theoretical_loss": 3.564289613949671,
      "tokens_seen": 1408237568
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000578960038517092,
      "loss": 0.0715,
      "theoretical_loss": 3.5642308831742384,
      "tokens_seen": 1408499712
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005788797945755096,
      "loss": 0.0707,
      "theoretical_loss": 3.564172166388465,
      "tokens_seen": 1408761856
    },
    {
      "epoch": 0.43,
      "objective/train/advantage_avg": 0.0017313800053671002,
      "objective/train/docs_used": 514827,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5287766456604004,
      "objective/train/original_loss": 1.5287766456604004,
      "objective/train/theoretical_loss": 3.5641134635864153,
      "objective/train/tokens_used": 1429484000,
      "objective/train/value_avg": -0.0082550048828125,
      "objective/train/value_loss": 0.00014103070134297013,
      "objective/train/value_max": -6.157159805297852e-05,
      "objective/train/value_min": -0.66455078125,
      "objective/train/value_reward_corr": 0.7031854825152273,
      "objective/train/value_std": 0.01306915283203125,
      "objective/train/weight_avg": 1.0018001794815063,
      "objective/train/weighted_lm_loss": 1.531276822090149,
      "objective/train/weights_max": 1.2216413021087646,
      "objective/train/weights_min": 0.5578534603118896,
      "theoretical_loss": 3.5641134635864153,
      "tokens_seen": 1409024000
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005787995506339272,
      "loss": 0.0709,
      "theoretical_loss": 3.5641134635864153,
      "tokens_seen": 1409024000
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005787193066923447,
      "loss": 0.0692,
      "theoretical_loss": 3.56405477476216,
      "tokens_seen": 1409286144
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005786390627507623,
      "loss": 0.0711,
      "theoretical_loss": 3.563996099909771,
      "tokens_seen": 1409548288
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005785588188091799,
      "loss": 0.0728,
      "theoretical_loss": 3.5639374390233245,
      "tokens_seen": 1409810432
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005784785748675974,
      "loss": 0.0705,
      "theoretical_loss": 3.563878792096901,
      "tokens_seen": 1410072576
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005783983309260151,
      "loss": 0.0707,
      "theoretical_loss": 3.5638201591245826,
      "tokens_seen": 1410334720
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005783180869844327,
      "loss": 0.0696,
      "theoretical_loss": 3.563761540100457,
      "tokens_seen": 1410596864
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005782378430428504,
      "loss": 0.0723,
      "theoretical_loss": 3.5637029350186156,
      "tokens_seen": 1410859008
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005781575991012679,
      "loss": 0.0715,
      "theoretical_loss": 3.563644343873152,
      "tokens_seen": 1411121152
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005780773551596854,
      "loss": 0.0741,
      "theoretical_loss": 3.5635857666581643,
      "tokens_seen": 1411383296
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005779971112181031,
      "loss": 0.0722,
      "theoretical_loss": 3.5635272033677534,
      "tokens_seen": 1411645440
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005779168672765206,
      "loss": 0.0734,
      "theoretical_loss": 3.5634686539960247,
      "tokens_seen": 1411907584
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005778366233349382,
      "loss": 0.0723,
      "theoretical_loss": 3.5634101185370874,
      "tokens_seen": 1412169728
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005777563793933558,
      "loss": 0.0709,
      "theoretical_loss": 3.5633515969850533,
      "tokens_seen": 1412431872
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005776761354517734,
      "loss": 0.0711,
      "theoretical_loss": 3.5632930893340378,
      "tokens_seen": 1412694016
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000577595891510191,
      "loss": 0.0722,
      "theoretical_loss": 3.5632345955781606,
      "tokens_seen": 1412956160
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005775156475686087,
      "loss": 0.0697,
      "theoretical_loss": 3.5631761157115456,
      "tokens_seen": 1413218304
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005774354036270262,
      "loss": 0.0694,
      "theoretical_loss": 3.5631176497283175,
      "tokens_seen": 1413480448
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005773551596854437,
      "loss": 0.0722,
      "theoretical_loss": 3.563059197622608,
      "tokens_seen": 1413742592
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005772749157438614,
      "loss": 0.0722,
      "theoretical_loss": 3.56300075938855,
      "tokens_seen": 1414004736
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005771946718022789,
      "loss": 0.072,
      "theoretical_loss": 3.5629423350202813,
      "tokens_seen": 1414266880
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005771144278606965,
      "loss": 0.0708,
      "theoretical_loss": 3.5628839245119424,
      "tokens_seen": 1414529024
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005770341839191141,
      "loss": 0.0734,
      "theoretical_loss": 3.5628255278576777,
      "tokens_seen": 1414791168
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005769539399775317,
      "loss": 0.0721,
      "theoretical_loss": 3.5627671450516347,
      "tokens_seen": 1415053312
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005768736960359494,
      "loss": 0.0721,
      "theoretical_loss": 3.5627087760879657,
      "tokens_seen": 1415315456
    },
    {
      "epoch": 0.43,
      "objective/train/advantage_avg": 0.000985055579803884,
      "objective/train/docs_used": 517252,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4529213905334473,
      "objective/train/original_loss": 1.4529212713241577,
      "objective/train/theoretical_loss": 3.5626504209608254,
      "objective/train/tokens_used": 1436037600,
      "objective/train/value_avg": -0.006443023681640625,
      "objective/train/value_loss": 0.00012476177653297782,
      "objective/train/value_max": -7.033348083496094e-05,
      "objective/train/value_min": -0.61279296875,
      "objective/train/value_reward_corr": 0.6847537745656294,
      "objective/train/value_std": 0.01068115234375,
      "objective/train/weight_avg": 1.0010443925857544,
      "objective/train/weighted_lm_loss": 1.4547419548034668,
      "objective/train/weights_max": 1.409537672996521,
      "objective/train/weights_min": 0.3798121213912964,
      "theoretical_loss": 3.5626504209608254,
      "tokens_seen": 1415577600
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005767934520943669,
      "loss": 0.0717,
      "theoretical_loss": 3.5626504209608254,
      "tokens_seen": 1415577600
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005767132081527845,
      "loss": 0.0676,
      "theoretical_loss": 3.5625920796643724,
      "tokens_seen": 1415839744
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005766329642112021,
      "loss": 0.074,
      "theoretical_loss": 3.5625337521927687,
      "tokens_seen": 1416101888
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005765527202696197,
      "loss": 0.0724,
      "theoretical_loss": 3.5624754385401802,
      "tokens_seen": 1416364032
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005764724763280372,
      "loss": 0.0704,
      "theoretical_loss": 3.562417138700776,
      "tokens_seen": 1416626176
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005763922323864549,
      "loss": 0.0727,
      "theoretical_loss": 3.5623588526687295,
      "tokens_seen": 1416888320
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005763119884448724,
      "loss": 0.0682,
      "theoretical_loss": 3.562300580438216,
      "tokens_seen": 1417150464
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005762317445032899,
      "loss": 0.073,
      "theoretical_loss": 3.562242322003416,
      "tokens_seen": 1417412608
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005761515005617077,
      "loss": 0.0713,
      "theoretical_loss": 3.5621840773585127,
      "tokens_seen": 1417674752
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005760712566201252,
      "loss": 0.0694,
      "theoretical_loss": 3.5621258464976924,
      "tokens_seen": 1417936896
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005759910126785428,
      "loss": 0.0689,
      "theoretical_loss": 3.5620676294151465,
      "tokens_seen": 1418199040
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005759107687369604,
      "loss": 0.0696,
      "theoretical_loss": 3.562009426105069,
      "tokens_seen": 1418461184
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000575830524795378,
      "loss": 0.0714,
      "theoretical_loss": 3.561951236561656,
      "tokens_seen": 1418723328
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005757502808537955,
      "loss": 0.0727,
      "theoretical_loss": 3.56189306077911,
      "tokens_seen": 1418985472
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005756700369122131,
      "loss": 0.072,
      "theoretical_loss": 3.561834898751635,
      "tokens_seen": 1419247616
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005755897929706307,
      "loss": 0.0709,
      "theoretical_loss": 3.561776750473439,
      "tokens_seen": 1419509760
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005755095490290483,
      "loss": 0.0729,
      "theoretical_loss": 3.561718615938733,
      "tokens_seen": 1419771904
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000575429305087466,
      "loss": 0.0705,
      "theoretical_loss": 3.5616604951417328,
      "tokens_seen": 1420034048
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005753490611458835,
      "loss": 0.0707,
      "theoretical_loss": 3.561602388076656,
      "tokens_seen": 1420296192
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005752688172043012,
      "loss": 0.0697,
      "theoretical_loss": 3.5615442947377254,
      "tokens_seen": 1420558336
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005751885732627187,
      "loss": 0.0722,
      "theoretical_loss": 3.5614862151191664,
      "tokens_seen": 1420820480
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005751083293211362,
      "loss": 0.0719,
      "theoretical_loss": 3.561428149215208,
      "tokens_seen": 1421082624
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005750280853795539,
      "loss": 0.0695,
      "theoretical_loss": 3.561370097020083,
      "tokens_seen": 1421344768
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005749478414379714,
      "loss": 0.0712,
      "theoretical_loss": 3.561312058528026,
      "tokens_seen": 1421606912
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000574867597496389,
      "loss": 0.0708,
      "theoretical_loss": 3.5612540337332783,
      "tokens_seen": 1421869056
    },
    {
      "epoch": 0.43,
      "objective/train/advantage_avg": 0.0007025928935036063,
      "objective/train/docs_used": 519513,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4979246854782104,
      "objective/train/original_loss": 1.497924566268921,
      "objective/train/theoretical_loss": 3.5611960226300816,
      "objective/train/tokens_used": 1442591200,
      "objective/train/value_avg": -0.007648468017578125,
      "objective/train/value_loss": 0.00016468969988636672,
      "objective/train/value_max": -7.545948028564453e-05,
      "objective/train/value_min": -0.35791015625,
      "objective/train/value_reward_corr": 0.7218915626636218,
      "objective/train/value_std": 0.01317596435546875,
      "objective/train/weight_avg": 1.0007803440093994,
      "objective/train/weighted_lm_loss": 1.4997540712356567,
      "objective/train/weights_max": 1.2718398571014404,
      "objective/train/weights_min": 0.3820907473564148,
      "theoretical_loss": 3.5611960226300816,
      "tokens_seen": 1422131200
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005747873535548066,
      "loss": 0.0722,
      "theoretical_loss": 3.5611960226300816,
      "tokens_seen": 1422131200
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005747071096132242,
      "loss": 0.0712,
      "theoretical_loss": 3.561138025212683,
      "tokens_seen": 1422393344
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005746268656716417,
      "loss": 0.07,
      "theoretical_loss": 3.561080041475332,
      "tokens_seen": 1422655488
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005745466217300594,
      "loss": 0.0693,
      "theoretical_loss": 3.5610220714122827,
      "tokens_seen": 1422917632
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000574466377788477,
      "loss": 0.0698,
      "theoretical_loss": 3.560964115017791,
      "tokens_seen": 1423179776
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005743861338468946,
      "loss": 0.07,
      "theoretical_loss": 3.560906172286118,
      "tokens_seen": 1423441920
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005743058899053122,
      "loss": 0.0703,
      "theoretical_loss": 3.5608482432115265,
      "tokens_seen": 1423704064
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005742256459637297,
      "loss": 0.0699,
      "theoretical_loss": 3.5607903277882853,
      "tokens_seen": 1423966208
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005741454020221474,
      "loss": 0.0712,
      "theoretical_loss": 3.560732426010664,
      "tokens_seen": 1424228352
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005740651580805649,
      "loss": 0.0726,
      "theoretical_loss": 3.560674537872937,
      "tokens_seen": 1424490496
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005739849141389825,
      "loss": 0.0693,
      "theoretical_loss": 3.560616663369382,
      "tokens_seen": 1424752640
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005739046701974002,
      "loss": 0.0704,
      "theoretical_loss": 3.5605588024942803,
      "tokens_seen": 1425014784
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005738244262558177,
      "loss": 0.0706,
      "theoretical_loss": 3.560500955241916,
      "tokens_seen": 1425276928
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005737441823142353,
      "loss": 0.0714,
      "theoretical_loss": 3.5604431216065775,
      "tokens_seen": 1425539072
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005736639383726529,
      "loss": 0.0704,
      "theoretical_loss": 3.5603853015825564,
      "tokens_seen": 1425801216
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005735836944310705,
      "loss": 0.0693,
      "theoretical_loss": 3.560327495164147,
      "tokens_seen": 1426063360
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000573503450489488,
      "loss": 0.0681,
      "theoretical_loss": 3.5602697023456473,
      "tokens_seen": 1426325504
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005734232065479057,
      "loss": 0.0706,
      "theoretical_loss": 3.5602119231213605,
      "tokens_seen": 1426587648
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005733429626063232,
      "loss": 0.0703,
      "theoretical_loss": 3.5601541574855906,
      "tokens_seen": 1426849792
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005732627186647407,
      "loss": 0.071,
      "theoretical_loss": 3.5600964054326463,
      "tokens_seen": 1427111936
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005731824747231584,
      "loss": 0.0721,
      "theoretical_loss": 3.5600386669568405,
      "tokens_seen": 1427374080
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000573102230781576,
      "loss": 0.0753,
      "theoretical_loss": 3.559980942052488,
      "tokens_seen": 1427636224
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005730219868399937,
      "loss": 0.0698,
      "theoretical_loss": 3.559923230713907,
      "tokens_seen": 1427898368
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005729417428984112,
      "loss": 0.0702,
      "theoretical_loss": 3.5598655329354214,
      "tokens_seen": 1428160512
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005728614989568288,
      "loss": 0.0721,
      "theoretical_loss": 3.5598078487113556,
      "tokens_seen": 1428422656
    },
    {
      "epoch": 0.43,
      "objective/train/advantage_avg": 0.0015544474590569735,
      "objective/train/docs_used": 521864,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4286561012268066,
      "objective/train/original_loss": 1.4286558628082275,
      "objective/train/theoretical_loss": 3.55975017803604,
      "objective/train/tokens_used": 1449144800,
      "objective/train/value_avg": -0.00952911376953125,
      "objective/train/value_loss": 0.0003103648195974529,
      "objective/train/value_max": -4.83393669128418e-05,
      "objective/train/value_min": -0.6337890625,
      "objective/train/value_reward_corr": 0.6740575950147936,
      "objective/train/value_std": 0.0167236328125,
      "objective/train/weight_avg": 1.0016885995864868,
      "objective/train/weighted_lm_loss": 1.430799126625061,
      "objective/train/weights_max": 1.1823221445083618,
      "objective/train/weights_min": 0.3692566156387329,
      "theoretical_loss": 3.55975017803604,
      "tokens_seen": 1428684800
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005727812550152464,
      "loss": 0.0732,
      "theoretical_loss": 3.55975017803604,
      "tokens_seen": 1428684800
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005727010110736639,
      "loss": 0.0752,
      "theoretical_loss": 3.5596925209038055,
      "tokens_seen": 1428946944
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005726207671320815,
      "loss": 0.072,
      "theoretical_loss": 3.55963487730899,
      "tokens_seen": 1429209088
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005725405231904991,
      "loss": 0.0698,
      "theoretical_loss": 3.5595772472459313,
      "tokens_seen": 1429471232
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005724602792489167,
      "loss": 0.0707,
      "theoretical_loss": 3.559519630708973,
      "tokens_seen": 1429733376
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005723800353073343,
      "loss": 0.0711,
      "theoretical_loss": 3.5594620276924607,
      "tokens_seen": 1429995520
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000572299791365752,
      "loss": 0.0703,
      "theoretical_loss": 3.559404438190745,
      "tokens_seen": 1430257664
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005722195474241695,
      "loss": 0.0706,
      "theoretical_loss": 3.559346862198178,
      "tokens_seen": 1430519808
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000572139303482587,
      "loss": 0.0717,
      "theoretical_loss": 3.559289299709116,
      "tokens_seen": 1430781952
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005720590595410047,
      "loss": 0.0709,
      "theoretical_loss": 3.5592317507179194,
      "tokens_seen": 1431044096
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005719788155994222,
      "loss": 0.072,
      "theoretical_loss": 3.5591742152189507,
      "tokens_seen": 1431306240
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005718985716578399,
      "loss": 0.0694,
      "theoretical_loss": 3.559116693206577,
      "tokens_seen": 1431568384
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005718183277162574,
      "loss": 0.0735,
      "theoretical_loss": 3.5590591846751685,
      "tokens_seen": 1431830528
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000571738083774675,
      "loss": 0.07,
      "theoretical_loss": 3.5590016896190977,
      "tokens_seen": 1432092672
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005716578398330927,
      "loss": 0.07,
      "theoretical_loss": 3.5589442080327416,
      "tokens_seen": 1432354816
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005715775958915102,
      "loss": 0.071,
      "theoretical_loss": 3.5588867399104798,
      "tokens_seen": 1432616960
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005714973519499278,
      "loss": 0.0718,
      "theoretical_loss": 3.5588292852466967,
      "tokens_seen": 1432879104
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005714171080083454,
      "loss": 0.0671,
      "theoretical_loss": 3.558771844035779,
      "tokens_seen": 1433141248
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000571336864066763,
      "loss": 0.0681,
      "theoretical_loss": 3.5587144162721156,
      "tokens_seen": 1433403392
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005712566201251805,
      "loss": 0.0687,
      "theoretical_loss": 3.5586570019501016,
      "tokens_seen": 1433665536
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005711763761835982,
      "loss": 0.0737,
      "theoretical_loss": 3.558599601064133,
      "tokens_seen": 1433927680
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005710961322420157,
      "loss": 0.072,
      "theoretical_loss": 3.5585422136086104,
      "tokens_seen": 1434189824
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005710158883004333,
      "loss": 0.0719,
      "theoretical_loss": 3.558484839577937,
      "tokens_seen": 1434451968
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000570935644358851,
      "loss": 0.0711,
      "theoretical_loss": 3.55842747896652,
      "tokens_seen": 1434714112
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005708554004172685,
      "loss": 0.0678,
      "theoretical_loss": 3.55837013176877,
      "tokens_seen": 1434976256
    },
    {
      "epoch": 0.43,
      "objective/train/advantage_avg": 0.001079781330190599,
      "objective/train/docs_used": 524207,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4617466926574707,
      "objective/train/original_loss": 1.4617464542388916,
      "objective/train/theoretical_loss": 3.5583127979791005,
      "objective/train/tokens_used": 1455698400,
      "objective/train/value_avg": -0.0064697265625,
      "objective/train/value_loss": 0.00015081166930031031,
      "objective/train/value_max": -6.657838821411133e-05,
      "objective/train/value_min": -0.29443359375,
      "objective/train/value_reward_corr": 0.6869328360522499,
      "objective/train/value_std": 0.012359619140625,
      "objective/train/weight_avg": 1.0011464357376099,
      "objective/train/weighted_lm_loss": 1.463802695274353,
      "objective/train/weights_max": 1.2564263343811035,
      "objective/train/weights_min": 0.3687688410282135,
      "theoretical_loss": 3.5583127979791005,
      "tokens_seen": 1435238400
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0005707751564756861,
      "loss": 0.0698,
      "theoretical_loss": 3.5583127979791005,
      "tokens_seen": 1435238400
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005706949125341037,
      "loss": 0.0712,
      "theoretical_loss": 3.558255477591928,
      "tokens_seen": 1435500544
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005706146685925213,
      "loss": 0.0694,
      "theoretical_loss": 3.558198170601674,
      "tokens_seen": 1435762688
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005705344246509389,
      "loss": 0.0695,
      "theoretical_loss": 3.558140877002761,
      "tokens_seen": 1436024832
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005704541807093565,
      "loss": 0.0726,
      "theoretical_loss": 3.5580835967896167,
      "tokens_seen": 1436286976
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000570373936767774,
      "loss": 0.0696,
      "theoretical_loss": 3.5580263299566712,
      "tokens_seen": 1436549120
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005702936928261916,
      "loss": 0.0726,
      "theoretical_loss": 3.5579690764983587,
      "tokens_seen": 1436811264
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005702134488846092,
      "loss": 0.0669,
      "theoretical_loss": 3.557911836409115,
      "tokens_seen": 1437073408
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005701332049430268,
      "loss": 0.0695,
      "theoretical_loss": 3.557854609683382,
      "tokens_seen": 1437335552
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005700529610014445,
      "loss": 0.0694,
      "theoretical_loss": 3.5577973963156024,
      "tokens_seen": 1437597696
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000569972717059862,
      "loss": 0.0692,
      "theoretical_loss": 3.557740196300224,
      "tokens_seen": 1437859840
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005698924731182796,
      "loss": 0.0726,
      "theoretical_loss": 3.5576830096316963,
      "tokens_seen": 1438121984
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005698122291766972,
      "loss": 0.0714,
      "theoretical_loss": 3.5576258363044735,
      "tokens_seen": 1438384128
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005697319852351147,
      "loss": 0.073,
      "theoretical_loss": 3.5575686763130117,
      "tokens_seen": 1438646272
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005696517412935323,
      "loss": 0.0717,
      "theoretical_loss": 3.5575115296517725,
      "tokens_seen": 1438908416
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005695714973519499,
      "loss": 0.0707,
      "theoretical_loss": 3.5574543963152188,
      "tokens_seen": 1439170560
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005694912534103675,
      "loss": 0.0688,
      "theoretical_loss": 3.5573972762978174,
      "tokens_seen": 1439432704
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000569411009468785,
      "loss": 0.0742,
      "theoretical_loss": 3.5573401695940383,
      "tokens_seen": 1439694848
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005693307655272028,
      "loss": 0.0704,
      "theoretical_loss": 3.557283076198356,
      "tokens_seen": 1439956992
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005692505215856203,
      "loss": 0.0725,
      "theoretical_loss": 3.557225996105246,
      "tokens_seen": 1440219136
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005691702776440379,
      "loss": 0.0735,
      "theoretical_loss": 3.5571689293091895,
      "tokens_seen": 1440481280
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005690900337024555,
      "loss": 0.0693,
      "theoretical_loss": 3.5571118758046696,
      "tokens_seen": 1440743424
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000569009789760873,
      "loss": 0.0682,
      "theoretical_loss": 3.5570548355861726,
      "tokens_seen": 1441005568
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005689295458192907,
      "loss": 0.0733,
      "theoretical_loss": 3.5569978086481884,
      "tokens_seen": 1441267712
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005688493018777082,
      "loss": 0.0682,
      "theoretical_loss": 3.556940794985211,
      "tokens_seen": 1441529856
    },
    {
      "epoch": 0.44,
      "objective/train/advantage_avg": 0.001043870928697288,
      "objective/train/docs_used": 526671,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3674421310424805,
      "objective/train/original_loss": 1.3674421310424805,
      "objective/train/theoretical_loss": 3.5568837945917364,
      "objective/train/tokens_used": 1462252000,
      "objective/train/value_avg": -0.00707244873046875,
      "objective/train/value_loss": 0.00022587105922866613,
      "objective/train/value_max": -6.556510925292969e-05,
      "objective/train/value_min": -0.299560546875,
      "objective/train/value_reward_corr": 0.6144987543879608,
      "objective/train/value_std": 0.0108795166015625,
      "objective/train/weight_avg": 1.001141905784607,
      "objective/train/weighted_lm_loss": 1.3684946298599243,
      "objective/train/weights_max": 1.2918591499328613,
      "objective/train/weights_min": 0.37235790491104126,
      "theoretical_loss": 3.5568837945917364,
      "tokens_seen": 1441792000
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005687690579361258,
      "loss": 0.0703,
      "theoretical_loss": 3.5568837945917364,
      "tokens_seen": 1441792000
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005686888139945435,
      "loss": 0.0734,
      "theoretical_loss": 3.556826807462264,
      "tokens_seen": 1442054144
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000568608570052961,
      "loss": 0.0711,
      "theoretical_loss": 3.5567698335912983,
      "tokens_seen": 1442316288
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005685283261113786,
      "loss": 0.0725,
      "theoretical_loss": 3.5567128729733444,
      "tokens_seen": 1442578432
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005684480821697962,
      "loss": 0.0701,
      "theoretical_loss": 3.5566559256029118,
      "tokens_seen": 1442840576
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005683678382282138,
      "loss": 0.0688,
      "theoretical_loss": 3.556598991474515,
      "tokens_seen": 1443102720
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005682875942866313,
      "loss": 0.071,
      "theoretical_loss": 3.556542070582669,
      "tokens_seen": 1443364864
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000568207350345049,
      "loss": 0.0706,
      "theoretical_loss": 3.5564851629218928,
      "tokens_seen": 1443627008
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005681271064034665,
      "loss": 0.0688,
      "theoretical_loss": 3.55642826848671,
      "tokens_seen": 1443889152
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000568046862461884,
      "loss": 0.0681,
      "theoretical_loss": 3.5563713872716467,
      "tokens_seen": 1444151296
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005679666185203018,
      "loss": 0.0719,
      "theoretical_loss": 3.5563145192712318,
      "tokens_seen": 1444413440
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005678863745787193,
      "loss": 0.0722,
      "theoretical_loss": 3.5562576644799977,
      "tokens_seen": 1444675584
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000567806130637137,
      "loss": 0.0719,
      "theoretical_loss": 3.5562008228924804,
      "tokens_seen": 1444937728
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005677258866955545,
      "loss": 0.07,
      "theoretical_loss": 3.556143994503219,
      "tokens_seen": 1445199872
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005676456427539721,
      "loss": 0.0672,
      "theoretical_loss": 3.5560871793067554,
      "tokens_seen": 1445462016
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005675653988123897,
      "loss": 0.0695,
      "theoretical_loss": 3.556030377297635,
      "tokens_seen": 1445724160
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005674851548708073,
      "loss": 0.0715,
      "theoretical_loss": 3.5559735884704073,
      "tokens_seen": 1445986304
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005674049109292248,
      "loss": 0.0703,
      "theoretical_loss": 3.5559168128196235,
      "tokens_seen": 1446248448
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005673246669876424,
      "loss": 0.0698,
      "theoretical_loss": 3.555860050339839,
      "tokens_seen": 1446510592
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00056724442304606,
      "loss": 0.0705,
      "theoretical_loss": 3.555803301025613,
      "tokens_seen": 1446772736
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005671641791044776,
      "loss": 0.0694,
      "theoretical_loss": 3.555746564871506,
      "tokens_seen": 1447034880
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005670839351628953,
      "loss": 0.068,
      "theoretical_loss": 3.5556898418720837,
      "tokens_seen": 1447297024
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005670036912213128,
      "loss": 0.0674,
      "theoretical_loss": 3.555633132021914,
      "tokens_seen": 1447559168
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005669234472797304,
      "loss": 0.07,
      "theoretical_loss": 3.5555764353155688,
      "tokens_seen": 1447821312
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000566843203338148,
      "loss": 0.07,
      "theoretical_loss": 3.555519751747622,
      "tokens_seen": 1448083456
    },
    {
      "epoch": 0.44,
      "objective/train/advantage_avg": 0.0011815315810963511,
      "objective/train/docs_used": 529166,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4328693151474,
      "objective/train/original_loss": 1.4328691959381104,
      "objective/train/theoretical_loss": 3.5554630813126513,
      "objective/train/tokens_used": 1468805600,
      "objective/train/value_avg": -0.00795745849609375,
      "objective/train/value_loss": 0.0003213270101696253,
      "objective/train/value_max": -9.685754776000977e-05,
      "objective/train/value_min": -0.9267578125,
      "objective/train/value_reward_corr": 0.7253755597865478,
      "objective/train/value_std": 0.0184173583984375,
      "objective/train/weight_avg": 1.0013219118118286,
      "objective/train/weighted_lm_loss": 1.4348268508911133,
      "objective/train/weights_max": 1.3883624076843262,
      "objective/train/weights_min": 0.3690347969532013,
      "theoretical_loss": 3.5554630813126513,
      "tokens_seen": 1448345600
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005667629593965655,
      "loss": 0.0709,
      "theoretical_loss": 3.5554630813126513,
      "tokens_seen": 1448345600
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005666827154549832,
      "loss": 0.0718,
      "theoretical_loss": 3.5554064240052385,
      "tokens_seen": 1448607744
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005666024715134007,
      "loss": 0.0695,
      "theoretical_loss": 3.5553497798199674,
      "tokens_seen": 1448869888
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005665222275718183,
      "loss": 0.0723,
      "theoretical_loss": 3.555293148751426,
      "tokens_seen": 1449132032
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000566441983630236,
      "loss": 0.0721,
      "theoretical_loss": 3.555236530794204,
      "tokens_seen": 1449394176
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005663617396886536,
      "loss": 0.07,
      "theoretical_loss": 3.5551799259428964,
      "tokens_seen": 1449656320
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005662814957470711,
      "loss": 0.0696,
      "theoretical_loss": 3.5551233341920994,
      "tokens_seen": 1449918464
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005662012518054887,
      "loss": 0.0717,
      "theoretical_loss": 3.555066755536414,
      "tokens_seen": 1450180608
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005661210078639063,
      "loss": 0.0736,
      "theoretical_loss": 3.555010189970443,
      "tokens_seen": 1450442752
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005660407639223238,
      "loss": 0.0681,
      "theoretical_loss": 3.5549536374887936,
      "tokens_seen": 1450704896
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005659605199807415,
      "loss": 0.0708,
      "theoretical_loss": 3.5548970980860757,
      "tokens_seen": 1450967040
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000565880276039159,
      "loss": 0.0697,
      "theoretical_loss": 3.5548405717569023,
      "tokens_seen": 1451229184
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005658000320975766,
      "loss": 0.0719,
      "theoretical_loss": 3.5547840584958896,
      "tokens_seen": 1451491328
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005657197881559943,
      "loss": 0.0725,
      "theoretical_loss": 3.554727558297657,
      "tokens_seen": 1451753472
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005656395442144118,
      "loss": 0.0701,
      "theoretical_loss": 3.554671071156828,
      "tokens_seen": 1452015616
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005655593002728294,
      "loss": 0.0707,
      "theoretical_loss": 3.554614597068027,
      "tokens_seen": 1452277760
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000565479056331247,
      "loss": 0.0716,
      "theoretical_loss": 3.554558136025884,
      "tokens_seen": 1452539904
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005653988123896646,
      "loss": 0.072,
      "theoretical_loss": 3.554501688025031,
      "tokens_seen": 1452802048
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005653185684480822,
      "loss": 0.0703,
      "theoretical_loss": 3.554445253060103,
      "tokens_seen": 1453064192
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005652383245064998,
      "loss": 0.07,
      "theoretical_loss": 3.5543888311257397,
      "tokens_seen": 1453326336
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005651580805649173,
      "loss": 0.0716,
      "theoretical_loss": 3.5543324222165813,
      "tokens_seen": 1453588480
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005650778366233349,
      "loss": 0.0693,
      "theoretical_loss": 3.554276026327274,
      "tokens_seen": 1453850624
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005649975926817526,
      "loss": 0.0715,
      "theoretical_loss": 3.5542196434524653,
      "tokens_seen": 1454112768
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005649173487401701,
      "loss": 0.0721,
      "theoretical_loss": 3.554163273586806,
      "tokens_seen": 1454374912
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005648371047985878,
      "loss": 0.0696,
      "theoretical_loss": 3.554106916724951,
      "tokens_seen": 1454637056
    },
    {
      "epoch": 0.44,
      "objective/train/advantage_avg": 0.0007077401387505233,
      "objective/train/docs_used": 531642,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3581182956695557,
      "objective/train/original_loss": 1.3581182956695557,
      "objective/train/theoretical_loss": 3.5540505728615583,
      "objective/train/tokens_used": 1475359200,
      "objective/train/value_avg": -0.006244659423828125,
      "objective/train/value_loss": 0.0002848842123057693,
      "objective/train/value_max": -7.599592208862305e-05,
      "objective/train/value_min": -0.51806640625,
      "objective/train/value_reward_corr": 0.5166717049484271,
      "objective/train/value_std": 0.01073455810546875,
      "objective/train/weight_avg": 1.0008255243301392,
      "objective/train/weighted_lm_loss": 1.3589487075805664,
      "objective/train/weights_max": 1.625488519668579,
      "objective/train/weights_min": 0.22773942351341248,
      "theoretical_loss": 3.5540505728615583,
      "tokens_seen": 1454899200
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005647568608570053,
      "loss": 0.0699,
      "theoretical_loss": 3.5540505728615583,
      "tokens_seen": 1454899200
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005646766169154229,
      "loss": 0.0704,
      "theoretical_loss": 3.5539942419912878,
      "tokens_seen": 1455161344
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005645963729738405,
      "loss": 0.0703,
      "theoretical_loss": 3.553937924108804,
      "tokens_seen": 1455423488
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005645161290322581,
      "loss": 0.0687,
      "theoretical_loss": 3.5538816192087728,
      "tokens_seen": 1455685632
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005644358850906756,
      "loss": 0.0693,
      "theoretical_loss": 3.5538253272858658,
      "tokens_seen": 1455947776
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005643556411490932,
      "loss": 0.07,
      "theoretical_loss": 3.5537690483347557,
      "tokens_seen": 1456209920
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005642753972075108,
      "loss": 0.0681,
      "theoretical_loss": 3.5537127823501184,
      "tokens_seen": 1456472064
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005641951532659285,
      "loss": 0.0694,
      "theoretical_loss": 3.5536565293266342,
      "tokens_seen": 1456734208
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005641149093243461,
      "loss": 0.0754,
      "theoretical_loss": 3.553600289258986,
      "tokens_seen": 1456996352
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005640346653827636,
      "loss": 0.0712,
      "theoretical_loss": 3.553544062141859,
      "tokens_seen": 1457258496
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005639544214411813,
      "loss": 0.0704,
      "theoretical_loss": 3.5534878479699423,
      "tokens_seen": 1457520640
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005638741774995988,
      "loss": 0.073,
      "theoretical_loss": 3.5534316467379288,
      "tokens_seen": 1457782784
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005637939335580163,
      "loss": 0.0727,
      "theoretical_loss": 3.5533754584405126,
      "tokens_seen": 1458044928
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000563713689616434,
      "loss": 0.0697,
      "theoretical_loss": 3.553319283072393,
      "tokens_seen": 1458307072
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005636334456748515,
      "loss": 0.0725,
      "theoretical_loss": 3.553263120628271,
      "tokens_seen": 1458569216
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005635532017332691,
      "loss": 0.0706,
      "theoretical_loss": 3.553206971102852,
      "tokens_seen": 1458831360
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005634729577916868,
      "loss": 0.0699,
      "theoretical_loss": 3.5531508344908436,
      "tokens_seen": 1459093504
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005633927138501044,
      "loss": 0.0711,
      "theoretical_loss": 3.5530947107869557,
      "tokens_seen": 1459355648
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005633124699085219,
      "loss": 0.0722,
      "theoretical_loss": 3.5530385999859035,
      "tokens_seen": 1459617792
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005632322259669395,
      "loss": 0.0709,
      "theoretical_loss": 3.5529825020824033,
      "tokens_seen": 1459879936
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005631519820253571,
      "loss": 0.0713,
      "theoretical_loss": 3.5529264170711756,
      "tokens_seen": 1460142080
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005630717380837746,
      "loss": 0.0715,
      "theoretical_loss": 3.552870344946944,
      "tokens_seen": 1460404224
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005629914941421923,
      "loss": 0.0713,
      "theoretical_loss": 3.5528142857044345,
      "tokens_seen": 1460666368
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005629112502006098,
      "loss": 0.0707,
      "theoretical_loss": 3.5527582393383765,
      "tokens_seen": 1460928512
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005628310062590275,
      "loss": 0.0727,
      "theoretical_loss": 3.5527022058435036,
      "tokens_seen": 1461190656
    },
    {
      "epoch": 0.44,
      "objective/train/advantage_avg": 0.0008472163463011384,
      "objective/train/docs_used": 534190,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4311131238937378,
      "objective/train/original_loss": 1.4311130046844482,
      "objective/train/theoretical_loss": 3.552646185214551,
      "objective/train/tokens_used": 1481912800,
      "objective/train/value_avg": -0.00859832763671875,
      "objective/train/value_loss": 0.00034199925721623003,
      "objective/train/value_max": -5.692243576049805e-05,
      "objective/train/value_min": -0.50927734375,
      "objective/train/value_reward_corr": 0.6411651572047227,
      "objective/train/value_std": 0.01556396484375,
      "objective/train/weight_avg": 1.00099515914917,
      "objective/train/weighted_lm_loss": 1.4326586723327637,
      "objective/train/weights_max": 1.275532603263855,
      "objective/train/weights_min": 0.24629011750221252,
      "theoretical_loss": 3.552646185214551,
      "tokens_seen": 1461452800
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000562750762317445,
      "loss": 0.0702,
      "theoretical_loss": 3.552646185214551,
      "tokens_seen": 1461452800
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005626705183758626,
      "loss": 0.07,
      "theoretical_loss": 3.552590177446257,
      "tokens_seen": 1461714944
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005625902744342803,
      "loss": 0.0701,
      "theoretical_loss": 3.5525341825333645,
      "tokens_seen": 1461977088
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005625100304926978,
      "loss": 0.0727,
      "theoretical_loss": 3.552478200470618,
      "tokens_seen": 1462239232
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005624297865511154,
      "loss": 0.072,
      "theoretical_loss": 3.552422231252766,
      "tokens_seen": 1462501376
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000562349542609533,
      "loss": 0.0699,
      "theoretical_loss": 3.552366274874559,
      "tokens_seen": 1462763520
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005622692986679506,
      "loss": 0.0707,
      "theoretical_loss": 3.5523103313307516,
      "tokens_seen": 1463025664
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005621890547263681,
      "loss": 0.0729,
      "theoretical_loss": 3.5522544006161016,
      "tokens_seen": 1463287808
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005621088107847857,
      "loss": 0.0703,
      "theoretical_loss": 3.5521984827253688,
      "tokens_seen": 1463549952
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005620285668432034,
      "loss": 0.0686,
      "theoretical_loss": 3.5521425776533175,
      "tokens_seen": 1463812096
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005619483229016209,
      "loss": 0.0696,
      "theoretical_loss": 3.5520866853947135,
      "tokens_seen": 1464074240
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005618680789600386,
      "loss": 0.0688,
      "theoretical_loss": 3.5520308059443275,
      "tokens_seen": 1464336384
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005617878350184561,
      "loss": 0.0723,
      "theoretical_loss": 3.5519749392969313,
      "tokens_seen": 1464598528
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005617075910768738,
      "loss": 0.0714,
      "theoretical_loss": 3.5519190854473006,
      "tokens_seen": 1464860672
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005616273471352913,
      "loss": 0.0707,
      "theoretical_loss": 3.5518632443902156,
      "tokens_seen": 1465122816
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005615471031937089,
      "loss": 0.0703,
      "theoretical_loss": 3.5518074161204565,
      "tokens_seen": 1465384960
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005614668592521265,
      "loss": 0.0731,
      "theoretical_loss": 3.5517516006328096,
      "tokens_seen": 1465647104
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000561386615310544,
      "loss": 0.0709,
      "theoretical_loss": 3.5516957979220627,
      "tokens_seen": 1465909248
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005613063713689616,
      "loss": 0.071,
      "theoretical_loss": 3.551640007983007,
      "tokens_seen": 1466171392
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005612261274273793,
      "loss": 0.0683,
      "theoretical_loss": 3.551584230810436,
      "tokens_seen": 1466433536
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005611458834857969,
      "loss": 0.0708,
      "theoretical_loss": 3.551528466399148,
      "tokens_seen": 1466695680
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005610656395442144,
      "loss": 0.0711,
      "theoretical_loss": 3.551472714743942,
      "tokens_seen": 1466957824
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005609853956026321,
      "loss": 0.069,
      "theoretical_loss": 3.551416975839623,
      "tokens_seen": 1467219968
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005609051516610496,
      "loss": 0.0699,
      "theoretical_loss": 3.551361249680996,
      "tokens_seen": 1467482112
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005608249077194671,
      "loss": 0.069,
      "theoretical_loss": 3.5513055362628707,
      "tokens_seen": 1467744256
    },
    {
      "epoch": 0.44,
      "objective/train/advantage_avg": 0.0009279769728891551,
      "objective/train/docs_used": 535989,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4145874977111816,
      "objective/train/original_loss": 1.4145876169204712,
      "objective/train/theoretical_loss": 3.5512498355800597,
      "objective/train/tokens_used": 1488466400,
      "objective/train/value_avg": -0.00853729248046875,
      "objective/train/value_loss": 0.0003833299851976335,
      "objective/train/value_max": -5.519390106201172e-05,
      "objective/train/value_min": -0.701171875,
      "objective/train/value_reward_corr": 0.6353014795865455,
      "objective/train/value_std": 0.017333984375,
      "objective/train/weight_avg": 1.0010976791381836,
      "objective/train/weighted_lm_loss": 1.4164438247680664,
      "objective/train/weights_max": 1.698567271232605,
      "objective/train/weights_min": 0.36841312050819397,
      "theoretical_loss": 3.5512498355800597,
      "tokens_seen": 1468006400
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005607446637778848,
      "loss": 0.0726,
      "theoretical_loss": 3.5512498355800597,
      "tokens_seen": 1468006400
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0005606644198363023,
      "loss": 0.0696,
      "theoretical_loss": 3.5511941476273785,
      "tokens_seen": 1468268544
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005605841758947199,
      "loss": 0.0688,
      "theoretical_loss": 3.551138472399646,
      "tokens_seen": 1468530688
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005605039319531376,
      "loss": 0.0703,
      "theoretical_loss": 3.5510828098916836,
      "tokens_seen": 1468792832
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005604236880115552,
      "loss": 0.0705,
      "theoretical_loss": 3.5510271600983154,
      "tokens_seen": 1469054976
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005603434440699728,
      "loss": 0.0702,
      "theoretical_loss": 3.5509715230143692,
      "tokens_seen": 1469317120
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005602632001283903,
      "loss": 0.0702,
      "theoretical_loss": 3.550915898634676,
      "tokens_seen": 1469579264
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005601829561868079,
      "loss": 0.0678,
      "theoretical_loss": 3.550860286954069,
      "tokens_seen": 1469841408
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005601027122452255,
      "loss": 0.068,
      "theoretical_loss": 3.5508046879673856,
      "tokens_seen": 1470103552
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005600224683036431,
      "loss": 0.0721,
      "theoretical_loss": 3.550749101669465,
      "tokens_seen": 1470365696
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005599422243620606,
      "loss": 0.0678,
      "theoretical_loss": 3.5506935280551497,
      "tokens_seen": 1470627840
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005598619804204783,
      "loss": 0.0711,
      "theoretical_loss": 3.5506379671192865,
      "tokens_seen": 1470889984
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005597817364788959,
      "loss": 0.0711,
      "theoretical_loss": 3.550582418856723,
      "tokens_seen": 1471152128
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005597014925373134,
      "loss": 0.069,
      "theoretical_loss": 3.550526883262312,
      "tokens_seen": 1471414272
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005596212485957311,
      "loss": 0.0711,
      "theoretical_loss": 3.550471360330907,
      "tokens_seen": 1471676416
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005595410046541486,
      "loss": 0.069,
      "theoretical_loss": 3.550415850057367,
      "tokens_seen": 1471938560
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005594607607125662,
      "loss": 0.0695,
      "theoretical_loss": 3.5503603524365523,
      "tokens_seen": 1472200704
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005593805167709838,
      "loss": 0.0735,
      "theoretical_loss": 3.5503048674633266,
      "tokens_seen": 1472462848
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005593002728294014,
      "loss": 0.0694,
      "theoretical_loss": 3.5502493951325564,
      "tokens_seen": 1472724992
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005592200288878189,
      "loss": 0.0739,
      "theoretical_loss": 3.550193935439112,
      "tokens_seen": 1472987136
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005591397849462365,
      "loss": 0.0697,
      "theoretical_loss": 3.5501384883778666,
      "tokens_seen": 1473249280
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005590595410046541,
      "loss": 0.0729,
      "theoretical_loss": 3.5500830539436956,
      "tokens_seen": 1473511424
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005589792970630718,
      "loss": 0.0692,
      "theoretical_loss": 3.550027632131477,
      "tokens_seen": 1473773568
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005588990531214894,
      "loss": 0.0708,
      "theoretical_loss": 3.549972222936094,
      "tokens_seen": 1474035712
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005588188091799069,
      "loss": 0.0696,
      "theoretical_loss": 3.5499168263524297,
      "tokens_seen": 1474297856
    },
    {
      "epoch": 0.45,
      "objective/train/advantage_avg": 0.000858014915138483,
      "objective/train/docs_used": 538469,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.343380331993103,
      "objective/train/original_loss": 1.343380331993103,
      "objective/train/theoretical_loss": 3.549861442375373,
      "objective/train/tokens_used": 1495020000,
      "objective/train/value_avg": -0.0080413818359375,
      "objective/train/value_loss": 0.00021646420645993203,
      "objective/train/value_max": -5.9664249420166016e-05,
      "objective/train/value_min": -0.485107421875,
      "objective/train/value_reward_corr": 0.6710323577309563,
      "objective/train/value_std": 0.015411376953125,
      "objective/train/weight_avg": 1.0009558200836182,
      "objective/train/weighted_lm_loss": 1.3450497388839722,
      "objective/train/weights_max": 1.3448151350021362,
      "objective/train/weights_min": 0.36848655343055725,
      "theoretical_loss": 3.549861442375373,
      "tokens_seen": 1474560000
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005587385652383246,
      "loss": 0.0717,
      "theoretical_loss": 3.549861442375373,
      "tokens_seen": 1474560000
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005586583212967421,
      "loss": 0.0681,
      "theoretical_loss": 3.5498060709998143,
      "tokens_seen": 1474822144
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005585780773551596,
      "loss": 0.071,
      "theoretical_loss": 3.5497507122206473,
      "tokens_seen": 1475084288
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005584978334135773,
      "loss": 0.0705,
      "theoretical_loss": 3.5496953660327684,
      "tokens_seen": 1475346432
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005584175894719948,
      "loss": 0.0705,
      "theoretical_loss": 3.5496400324310775,
      "tokens_seen": 1475608576
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005583373455304124,
      "loss": 0.0668,
      "theoretical_loss": 3.5495847114104766,
      "tokens_seen": 1475870720
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005582571015888301,
      "loss": 0.0721,
      "theoretical_loss": 3.549529402965873,
      "tokens_seen": 1476132864
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005581768576472477,
      "loss": 0.0699,
      "theoretical_loss": 3.549474107092173,
      "tokens_seen": 1476395008
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005580966137056652,
      "loss": 0.0734,
      "theoretical_loss": 3.5494188237842894,
      "tokens_seen": 1476657152
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005580163697640829,
      "loss": 0.0716,
      "theoretical_loss": 3.5493635530371366,
      "tokens_seen": 1476919296
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005579361258225004,
      "loss": 0.0694,
      "theoretical_loss": 3.5493082948456314,
      "tokens_seen": 1477181440
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005578558818809179,
      "loss": 0.0722,
      "theoretical_loss": 3.549253049204695,
      "tokens_seen": 1477443584
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005577756379393356,
      "loss": 0.0679,
      "theoretical_loss": 3.549197816109251,
      "tokens_seen": 1477705728
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005576953939977531,
      "loss": 0.0693,
      "theoretical_loss": 3.549142595554224,
      "tokens_seen": 1477967872
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005576151500561709,
      "loss": 0.0721,
      "theoretical_loss": 3.5490873875345446,
      "tokens_seen": 1478230016
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005575349061145884,
      "loss": 0.0691,
      "theoretical_loss": 3.5490321920451446,
      "tokens_seen": 1478492160
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000557454662173006,
      "loss": 0.0703,
      "theoretical_loss": 3.54897700908096,
      "tokens_seen": 1478754304
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005573744182314236,
      "loss": 0.0725,
      "theoretical_loss": 3.548921838636927,
      "tokens_seen": 1479016448
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005572941742898411,
      "loss": 0.0726,
      "theoretical_loss": 3.5488666807079885,
      "tokens_seen": 1479278592
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005572139303482587,
      "loss": 0.0703,
      "theoretical_loss": 3.5488115352890874,
      "tokens_seen": 1479540736
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005571336864066763,
      "loss": 0.071,
      "theoretical_loss": 3.5487564023751714,
      "tokens_seen": 1479802880
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005570534424650939,
      "loss": 0.0715,
      "theoretical_loss": 3.5487012819611894,
      "tokens_seen": 1480065024
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005569731985235114,
      "loss": 0.0716,
      "theoretical_loss": 3.548646174042095,
      "tokens_seen": 1480327168
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005568929545819291,
      "loss": 0.0727,
      "theoretical_loss": 3.5485910786128434,
      "tokens_seen": 1480589312
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005568127106403467,
      "loss": 0.0721,
      "theoretical_loss": 3.5485359956683933,
      "tokens_seen": 1480851456
    },
    {
      "epoch": 0.45,
      "objective/train/advantage_avg": 0.0011171478545293212,
      "objective/train/docs_used": 540822,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4887363910675049,
      "objective/train/original_loss": 1.4887363910675049,
      "objective/train/theoretical_loss": 3.548480925203706,
      "objective/train/tokens_used": 1501573600,
      "objective/train/value_avg": -0.01031494140625,
      "objective/train/value_loss": 0.0003310543834231794,
      "objective/train/value_max": -8.749961853027344e-05,
      "objective/train/value_min": -0.446533203125,
      "objective/train/value_reward_corr": 0.6941375660042437,
      "objective/train/value_std": 0.017333984375,
      "objective/train/weight_avg": 1.0012686252593994,
      "objective/train/weighted_lm_loss": 1.490714430809021,
      "objective/train/weights_max": 1.4081617593765259,
      "objective/train/weights_min": 0.3702827990055084,
      "theoretical_loss": 3.548480925203706,
      "tokens_seen": 1481113600
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005567324666987642,
      "loss": 0.0724,
      "theoretical_loss": 3.548480925203706,
      "tokens_seen": 1481113600
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005566522227571819,
      "loss": 0.0711,
      "theoretical_loss": 3.548425867213747,
      "tokens_seen": 1481375744
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005565719788155994,
      "loss": 0.0707,
      "theoretical_loss": 3.5483708216934833,
      "tokens_seen": 1481637888
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005564917348740171,
      "loss": 0.0705,
      "theoretical_loss": 3.5483157886378844,
      "tokens_seen": 1481900032
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005564114909324346,
      "loss": 0.0725,
      "theoretical_loss": 3.5482607680419243,
      "tokens_seen": 1482162176
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005563312469908522,
      "loss": 0.0706,
      "theoretical_loss": 3.548205759900579,
      "tokens_seen": 1482424320
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005562510030492698,
      "loss": 0.0706,
      "theoretical_loss": 3.548150764208828,
      "tokens_seen": 1482686464
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005561707591076873,
      "loss": 0.0699,
      "theoretical_loss": 3.5480957809616527,
      "tokens_seen": 1482948608
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000556090515166105,
      "loss": 0.0725,
      "theoretical_loss": 3.548040810154038,
      "tokens_seen": 1483210752
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005560102712245226,
      "loss": 0.0741,
      "theoretical_loss": 3.5479858517809717,
      "tokens_seen": 1483472896
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005559300272829402,
      "loss": 0.0722,
      "theoretical_loss": 3.547930905837445,
      "tokens_seen": 1483735040
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005558497833413577,
      "loss": 0.0748,
      "theoretical_loss": 3.547875972318451,
      "tokens_seen": 1483997184
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005557695393997754,
      "loss": 0.0691,
      "theoretical_loss": 3.547821051218987,
      "tokens_seen": 1484259328
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005556892954581929,
      "loss": 0.071,
      "theoretical_loss": 3.5477661425340514,
      "tokens_seen": 1484521472
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005556090515166104,
      "loss": 0.0719,
      "theoretical_loss": 3.547711246258647,
      "tokens_seen": 1484783616
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005555288075750281,
      "loss": 0.0731,
      "theoretical_loss": 3.547656362387779,
      "tokens_seen": 1485045760
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005554485636334456,
      "loss": 0.0733,
      "theoretical_loss": 3.5476014909164553,
      "tokens_seen": 1485307904
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005553683196918632,
      "loss": 0.0697,
      "theoretical_loss": 3.547546631839687,
      "tokens_seen": 1485570048
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005552880757502809,
      "loss": 0.0705,
      "theoretical_loss": 3.547491785152488,
      "tokens_seen": 1485832192
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005552078318086985,
      "loss": 0.0716,
      "theoretical_loss": 3.5474369508498755,
      "tokens_seen": 1486094336
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005551275878671161,
      "loss": 0.0717,
      "theoretical_loss": 3.547382128926868,
      "tokens_seen": 1486356480
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005550473439255337,
      "loss": 0.0698,
      "theoretical_loss": 3.5473273193784896,
      "tokens_seen": 1486618624
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005549670999839512,
      "loss": 0.0696,
      "theoretical_loss": 3.547272522199764,
      "tokens_seen": 1486880768
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005548868560423688,
      "loss": 0.072,
      "theoretical_loss": 3.5472177373857208,
      "tokens_seen": 1487142912
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005548066121007864,
      "loss": 0.0682,
      "theoretical_loss": 3.5471629649313905,
      "tokens_seen": 1487405056
    },
    {
      "epoch": 0.45,
      "objective/train/advantage_avg": -3.483461114228703e-05,
      "objective/train/docs_used": 543335,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3849729299545288,
      "objective/train/original_loss": 1.3849728107452393,
      "objective/train/theoretical_loss": 3.547108204831807,
      "objective/train/tokens_used": 1508127200,
      "objective/train/value_avg": -0.007526397705078125,
      "objective/train/value_loss": 0.000498010718729347,
      "objective/train/value_max": -7.253885269165039e-05,
      "objective/train/value_min": -0.91748046875,
      "objective/train/value_reward_corr": 0.6365641012412713,
      "objective/train/value_std": 0.0164794921875,
      "objective/train/weight_avg": 1.0001769065856934,
      "objective/train/weighted_lm_loss": 1.3848627805709839,
      "objective/train/weights_max": 2.075793504714966,
      "objective/train/weights_min": 0.23170025646686554,
      "theoretical_loss": 3.547108204831807,
      "tokens_seen": 1487667200
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005547263681592039,
      "loss": 0.0689,
      "theoretical_loss": 3.547108204831807,
      "tokens_seen": 1487667200
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005546461242176217,
      "loss": 0.0726,
      "theoretical_loss": 3.547053457082008,
      "tokens_seen": 1487929344
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005545658802760392,
      "loss": 0.0718,
      "theoretical_loss": 3.5469987216770322,
      "tokens_seen": 1488191488
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005544856363344568,
      "loss": 0.0724,
      "theoretical_loss": 3.5469439986119227,
      "tokens_seen": 1488453632
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005544053923928744,
      "loss": 0.07,
      "theoretical_loss": 3.5468892878817253,
      "tokens_seen": 1488715776
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005543251484512919,
      "loss": 0.0714,
      "theoretical_loss": 3.546834589481488,
      "tokens_seen": 1488977920
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005542449045097095,
      "loss": 0.07,
      "theoretical_loss": 3.5467799034062617,
      "tokens_seen": 1489240064
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005541646605681271,
      "loss": 0.0688,
      "theoretical_loss": 3.546725229651101,
      "tokens_seen": 1489502208
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005540844166265447,
      "loss": 0.071,
      "theoretical_loss": 3.5466705682110633,
      "tokens_seen": 1489764352
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005540041726849623,
      "loss": 0.0715,
      "theoretical_loss": 3.546615919081207,
      "tokens_seen": 1490026496
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00055392392874338,
      "loss": 0.0679,
      "theoretical_loss": 3.546561282256596,
      "tokens_seen": 1490288640
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005538436848017975,
      "loss": 0.0677,
      "theoretical_loss": 3.546506657732295,
      "tokens_seen": 1490550784
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005537634408602151,
      "loss": 0.0712,
      "theoretical_loss": 3.546452045503372,
      "tokens_seen": 1490812928
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005536831969186327,
      "loss": 0.0699,
      "theoretical_loss": 3.5463974455648994,
      "tokens_seen": 1491075072
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005536029529770502,
      "loss": 0.0696,
      "theoretical_loss": 3.5463428579119505,
      "tokens_seen": 1491337216
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005535227090354679,
      "loss": 0.0693,
      "theoretical_loss": 3.546288282539602,
      "tokens_seen": 1491599360
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005534424650938854,
      "loss": 0.0701,
      "theoretical_loss": 3.5462337194429336,
      "tokens_seen": 1491861504
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000553362221152303,
      "loss": 0.0722,
      "theoretical_loss": 3.546179168617028,
      "tokens_seen": 1492123648
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005532819772107206,
      "loss": 0.0689,
      "theoretical_loss": 3.54612463005697,
      "tokens_seen": 1492385792
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005532017332691381,
      "loss": 0.0731,
      "theoretical_loss": 3.546070103757849,
      "tokens_seen": 1492647936
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005531214893275557,
      "loss": 0.0681,
      "theoretical_loss": 3.546015589714755,
      "tokens_seen": 1492910080
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005530412453859734,
      "loss": 0.0715,
      "theoretical_loss": 3.545961087922782,
      "tokens_seen": 1493172224
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000552961001444391,
      "loss": 0.0696,
      "theoretical_loss": 3.5459065983770266,
      "tokens_seen": 1493434368
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005528807575028085,
      "loss": 0.0669,
      "theoretical_loss": 3.5458521210725893,
      "tokens_seen": 1493696512
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005528005135612262,
      "loss": 0.069,
      "theoretical_loss": 3.545797656004571,
      "tokens_seen": 1493958656
    },
    {
      "epoch": 0.45,
      "objective/train/advantage_avg": 0.002028031274676323,
      "objective/train/docs_used": 545793,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.288960337638855,
      "objective/train/original_loss": 1.2889604568481445,
      "objective/train/theoretical_loss": 3.545743203168077,
      "objective/train/tokens_used": 1514680800,
      "objective/train/value_avg": -0.0094146728515625,
      "objective/train/value_loss": 0.00047856621677055955,
      "objective/train/value_max": -6.973743438720703e-05,
      "objective/train/value_min": -0.8828125,
      "objective/train/value_reward_corr": 0.7416871098120547,
      "objective/train/value_std": 0.0251922607421875,
      "objective/train/weight_avg": 1.0022516250610352,
      "objective/train/weighted_lm_loss": 1.2908267974853516,
      "objective/train/weights_max": 2.2083611488342285,
      "objective/train/weights_min": 0.36811721324920654,
      "theoretical_loss": 3.545743203168077,
      "tokens_seen": 1494220800
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005527202696196437,
      "loss": 0.0683,
      "theoretical_loss": 3.545743203168077,
      "tokens_seen": 1494220800
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005526400256780613,
      "loss": 0.0692,
      "theoretical_loss": 3.545688762558216,
      "tokens_seen": 1494482944
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005525597817364789,
      "loss": 0.0704,
      "theoretical_loss": 3.5456343341700984,
      "tokens_seen": 1494745088
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005524795377948964,
      "loss": 0.07,
      "theoretical_loss": 3.545579917998838,
      "tokens_seen": 1495007232
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005523992938533142,
      "loss": 0.0689,
      "theoretical_loss": 3.5455255140395505,
      "tokens_seen": 1495269376
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005523190499117317,
      "loss": 0.071,
      "theoretical_loss": 3.5454711222873554,
      "tokens_seen": 1495531520
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005522388059701493,
      "loss": 0.0712,
      "theoretical_loss": 3.545416742737375,
      "tokens_seen": 1495793664
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005521585620285669,
      "loss": 0.0714,
      "theoretical_loss": 3.5453623753847343,
      "tokens_seen": 1496055808
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005520783180869845,
      "loss": 0.0688,
      "theoretical_loss": 3.54530802022456,
      "tokens_seen": 1496317952
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000551998074145402,
      "loss": 0.0744,
      "theoretical_loss": 3.545253677251983,
      "tokens_seen": 1496580096
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005519178302038196,
      "loss": 0.0698,
      "theoretical_loss": 3.5451993464621365,
      "tokens_seen": 1496842240
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005518375862622372,
      "loss": 0.0688,
      "theoretical_loss": 3.5451450278501566,
      "tokens_seen": 1497104384
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005517573423206547,
      "loss": 0.0736,
      "theoretical_loss": 3.545090721411182,
      "tokens_seen": 1497366528
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005516770983790724,
      "loss": 0.0708,
      "theoretical_loss": 3.545036427140354,
      "tokens_seen": 1497628672
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00055159685443749,
      "loss": 0.0727,
      "theoretical_loss": 3.544982145032817,
      "tokens_seen": 1497890816
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005515166104959077,
      "loss": 0.0694,
      "theoretical_loss": 3.5449278750837188,
      "tokens_seen": 1498152960
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005514363665543252,
      "loss": 0.0734,
      "theoretical_loss": 3.5448736172882085,
      "tokens_seen": 1498415104
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005513561226127427,
      "loss": 0.069,
      "theoretical_loss": 3.5448193716414393,
      "tokens_seen": 1498677248
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005512758786711604,
      "loss": 0.071,
      "theoretical_loss": 3.5447651381385668,
      "tokens_seen": 1498939392
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005511956347295779,
      "loss": 0.0708,
      "theoretical_loss": 3.544710916774749,
      "tokens_seen": 1499201536
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005511153907879955,
      "loss": 0.0701,
      "theoretical_loss": 3.5446567075451463,
      "tokens_seen": 1499463680
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005510351468464131,
      "loss": 0.0713,
      "theoretical_loss": 3.544602510444924,
      "tokens_seen": 1499725824
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005509549029048307,
      "loss": 0.0701,
      "theoretical_loss": 3.544548325469247,
      "tokens_seen": 1499987968
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005508746589632483,
      "loss": 0.0736,
      "theoretical_loss": 3.5444941526132863,
      "tokens_seen": 1500250112
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005507944150216659,
      "loss": 0.0725,
      "theoretical_loss": 3.5444399918722134,
      "tokens_seen": 1500512256
    },
    {
      "epoch": 0.45,
      "objective/train/advantage_avg": 0.00016552148736082017,
      "objective/train/docs_used": 548178,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3277549743652344,
      "objective/train/original_loss": 1.327755093574524,
      "objective/train/theoretical_loss": 3.5443858432412028,
      "objective/train/tokens_used": 1521234400,
      "objective/train/value_avg": -0.0071868896484375,
      "objective/train/value_loss": 0.00020315258007030934,
      "objective/train/value_max": -4.988908767700195e-05,
      "objective/train/value_min": -0.98486328125,
      "objective/train/value_reward_corr": 0.6999380374007601,
      "objective/train/value_std": 0.01276397705078125,
      "objective/train/weight_avg": 1.0002577304840088,
      "objective/train/weighted_lm_loss": 1.3282614946365356,
      "objective/train/weights_max": 1.1451897621154785,
      "objective/train/weights_min": 0.23797065019607544,
      "theoretical_loss": 3.5443858432412028,
      "tokens_seen": 1500774400
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005507141710800835,
      "loss": 0.0685,
      "theoretical_loss": 3.5443858432412028,
      "tokens_seen": 1500774400
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000550633927138501,
      "loss": 0.0691,
      "theoretical_loss": 3.5443317067154325,
      "tokens_seen": 1501036544
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0005505536831969187,
      "loss": 0.0717,
      "theoretical_loss": 3.544277582290083,
      "tokens_seen": 1501298688
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005504734392553362,
      "loss": 0.0713,
      "theoretical_loss": 3.544223469960337,
      "tokens_seen": 1501560832
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005503931953137538,
      "loss": 0.0723,
      "theoretical_loss": 3.5441693697213816,
      "tokens_seen": 1501822976
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005503129513721714,
      "loss": 0.0712,
      "theoretical_loss": 3.5441152815684043,
      "tokens_seen": 1502085120
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005502327074305889,
      "loss": 0.0697,
      "theoretical_loss": 3.5440612054965968,
      "tokens_seen": 1502347264
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005501524634890067,
      "loss": 0.0722,
      "theoretical_loss": 3.544007141501154,
      "tokens_seen": 1502609408
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005500722195474242,
      "loss": 0.0697,
      "theoretical_loss": 3.543953089577272,
      "tokens_seen": 1502871552
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005499919756058418,
      "loss": 0.0717,
      "theoretical_loss": 3.5438990497201512,
      "tokens_seen": 1503133696
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005499117316642594,
      "loss": 0.0692,
      "theoretical_loss": 3.5438450219249935,
      "tokens_seen": 1503395840
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000549831487722677,
      "loss": 0.0688,
      "theoretical_loss": 3.5437910061870044,
      "tokens_seen": 1503657984
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005497512437810945,
      "loss": 0.071,
      "theoretical_loss": 3.543737002501392,
      "tokens_seen": 1503920128
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005496709998395121,
      "loss": 0.0702,
      "theoretical_loss": 3.5436830108633663,
      "tokens_seen": 1504182272
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005495907558979297,
      "loss": 0.0679,
      "theoretical_loss": 3.5436290312681415,
      "tokens_seen": 1504444416
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005495105119563472,
      "loss": 0.0685,
      "theoretical_loss": 3.543575063710933,
      "tokens_seen": 1504706560
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000549430268014765,
      "loss": 0.0702,
      "theoretical_loss": 3.543521108186961,
      "tokens_seen": 1504968704
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005493500240731825,
      "loss": 0.0731,
      "theoretical_loss": 3.543467164691445,
      "tokens_seen": 1505230848
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005492697801316001,
      "loss": 0.0678,
      "theoretical_loss": 3.5434132332196113,
      "tokens_seen": 1505492992
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005491895361900177,
      "loss": 0.0698,
      "theoretical_loss": 3.543359313766686,
      "tokens_seen": 1505755136
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005491092922484352,
      "loss": 0.0703,
      "theoretical_loss": 3.543305406327899,
      "tokens_seen": 1506017280
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005490290483068528,
      "loss": 0.0682,
      "theoretical_loss": 3.5432515108984832,
      "tokens_seen": 1506279424
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005489488043652704,
      "loss": 0.0672,
      "theoretical_loss": 3.543197627473673,
      "tokens_seen": 1506541568
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000548868560423688,
      "loss": 0.0713,
      "theoretical_loss": 3.543143756048708,
      "tokens_seen": 1506803712
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005487883164821056,
      "loss": 0.0692,
      "theoretical_loss": 3.5430898966188265,
      "tokens_seen": 1507065856
    },
    {
      "epoch": 0.46,
      "objective/train/advantage_avg": 0.00014458467194344848,
      "objective/train/docs_used": 550589,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3810093402862549,
      "objective/train/original_loss": 1.3810093402862549,
      "objective/train/theoretical_loss": 3.543036049179274,
      "objective/train/tokens_used": 1527788000,
      "objective/train/value_avg": -0.0096435546875,
      "objective/train/value_loss": 0.00022767337213736027,
      "objective/train/value_max": -8.547306060791016e-05,
      "objective/train/value_min": -0.396728515625,
      "objective/train/value_reward_corr": 0.7315385161777496,
      "objective/train/value_std": 0.01538848876953125,
      "objective/train/weight_avg": 1.0002518892288208,
      "objective/train/weighted_lm_loss": 1.3807989358901978,
      "objective/train/weights_max": 1.3733340501785278,
      "objective/train/weights_min": 0.3778582811355591,
      "theoretical_loss": 3.543036049179274,
      "tokens_seen": 1507328000
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005487080725405232,
      "loss": 0.0707,
      "theoretical_loss": 3.543036049179274,
      "tokens_seen": 1507328000
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005486278285989408,
      "loss": 0.073,
      "theoretical_loss": 3.5429822137252955,
      "tokens_seen": 1507590144
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005485475846573585,
      "loss": 0.0696,
      "theoretical_loss": 3.54292839025214,
      "tokens_seen": 1507852288
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000548467340715776,
      "loss": 0.0697,
      "theoretical_loss": 3.5428745787550593,
      "tokens_seen": 1508114432
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005483870967741935,
      "loss": 0.0715,
      "theoretical_loss": 3.5428207792293076,
      "tokens_seen": 1508376576
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005483068528326112,
      "loss": 0.0698,
      "theoretical_loss": 3.542766991670142,
      "tokens_seen": 1508638720
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005482266088910287,
      "loss": 0.0706,
      "theoretical_loss": 3.542713216072821,
      "tokens_seen": 1508900864
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005481463649494463,
      "loss": 0.0721,
      "theoretical_loss": 3.542659452432608,
      "tokens_seen": 1509163008
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005480661210078639,
      "loss": 0.0705,
      "theoretical_loss": 3.542605700744768,
      "tokens_seen": 1509425152
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005479858770662815,
      "loss": 0.0694,
      "theoretical_loss": 3.542551961004568,
      "tokens_seen": 1509687296
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000547905633124699,
      "loss": 0.0716,
      "theoretical_loss": 3.5424982332072794,
      "tokens_seen": 1509949440
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005478253891831167,
      "loss": 0.0721,
      "theoretical_loss": 3.5424445173481756,
      "tokens_seen": 1510211584
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005477451452415343,
      "loss": 0.0711,
      "theoretical_loss": 3.5423908134225304,
      "tokens_seen": 1510473728
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005476649012999519,
      "loss": 0.0708,
      "theoretical_loss": 3.5423371214256245,
      "tokens_seen": 1510735872
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005475846573583695,
      "loss": 0.0708,
      "theoretical_loss": 3.5422834413527378,
      "tokens_seen": 1510998016
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000547504413416787,
      "loss": 0.0716,
      "theoretical_loss": 3.5422297731991548,
      "tokens_seen": 1511260160
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005474241694752047,
      "loss": 0.0686,
      "theoretical_loss": 3.542176116960162,
      "tokens_seen": 1511522304
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005473439255336222,
      "loss": 0.0718,
      "theoretical_loss": 3.542122472631048,
      "tokens_seen": 1511784448
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005472636815920397,
      "loss": 0.072,
      "theoretical_loss": 3.542068840207105,
      "tokens_seen": 1512046592
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005471834376504575,
      "loss": 0.0696,
      "theoretical_loss": 3.5420152196836288,
      "tokens_seen": 1512308736
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000547103193708875,
      "loss": 0.0683,
      "theoretical_loss": 3.541961611055915,
      "tokens_seen": 1512570880
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005470229497672926,
      "loss": 0.0701,
      "theoretical_loss": 3.5419080143192643,
      "tokens_seen": 1512833024
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005469427058257102,
      "loss": 0.0704,
      "theoretical_loss": 3.541854429468979,
      "tokens_seen": 1513095168
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005468624618841278,
      "loss": 0.0708,
      "theoretical_loss": 3.5418008565003647,
      "tokens_seen": 1513357312
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005467822179425453,
      "loss": 0.0718,
      "theoretical_loss": 3.541747295408729,
      "tokens_seen": 1513619456
    },
    {
      "epoch": 0.46,
      "objective/train/advantage_avg": 0.0009963659103959799,
      "objective/train/docs_used": 552897,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3523280620574951,
      "objective/train/original_loss": 1.3523280620574951,
      "objective/train/theoretical_loss": 3.541693746189383,
      "objective/train/tokens_used": 1534341600,
      "objective/train/value_avg": -0.0100250244140625,
      "objective/train/value_loss": 0.00032382109202444553,
      "objective/train/value_max": -6.252527236938477e-05,
      "objective/train/value_min": -0.81396484375,
      "objective/train/value_reward_corr": 0.6972783877515103,
      "objective/train/value_std": 0.01727294921875,
      "objective/train/weight_avg": 1.0011416673660278,
      "objective/train/weighted_lm_loss": 1.3530480861663818,
      "objective/train/weights_max": 1.5500072240829468,
      "objective/train/weights_min": 0.3707548677921295,
      "theoretical_loss": 3.541693746189383,
      "tokens_seen": 1513881600
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005467019740009629,
      "loss": 0.0701,
      "theoretical_loss": 3.541693746189383,
      "tokens_seen": 1513881600
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005466217300593805,
      "loss": 0.0687,
      "theoretical_loss": 3.5416402088376397,
      "tokens_seen": 1514143744
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000546541486117798,
      "loss": 0.0701,
      "theoretical_loss": 3.5415866833488154,
      "tokens_seen": 1514405888
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005464612421762158,
      "loss": 0.0709,
      "theoretical_loss": 3.541533169718228,
      "tokens_seen": 1514668032
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005463809982346333,
      "loss": 0.0701,
      "theoretical_loss": 3.541479667941199,
      "tokens_seen": 1514930176
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000546300754293051,
      "loss": 0.0706,
      "theoretical_loss": 3.5414261780130527,
      "tokens_seen": 1515192320
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005462205103514685,
      "loss": 0.0698,
      "theoretical_loss": 3.5413726999291155,
      "tokens_seen": 1515454464
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000546140266409886,
      "loss": 0.0724,
      "theoretical_loss": 3.5413192336847166,
      "tokens_seen": 1515716608
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005460600224683037,
      "loss": 0.0686,
      "theoretical_loss": 3.5412657792751876,
      "tokens_seen": 1515978752
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005459797785267212,
      "loss": 0.0698,
      "theoretical_loss": 3.541212336695863,
      "tokens_seen": 1516240896
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005458995345851388,
      "loss": 0.0702,
      "theoretical_loss": 3.54115890594208,
      "tokens_seen": 1516503040
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005458192906435564,
      "loss": 0.0677,
      "theoretical_loss": 3.5411054870091787,
      "tokens_seen": 1516765184
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000545739046701974,
      "loss": 0.0701,
      "theoretical_loss": 3.541052079892502,
      "tokens_seen": 1517027328
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005456588027603916,
      "loss": 0.0709,
      "theoretical_loss": 3.540998684587394,
      "tokens_seen": 1517289472
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005455785588188093,
      "loss": 0.071,
      "theoretical_loss": 3.5409453010892022,
      "tokens_seen": 1517551616
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005454983148772268,
      "loss": 0.0696,
      "theoretical_loss": 3.540891929393278,
      "tokens_seen": 1517813760
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005454180709356443,
      "loss": 0.0704,
      "theoretical_loss": 3.540838569494974,
      "tokens_seen": 1518075904
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000545337826994062,
      "loss": 0.0676,
      "theoretical_loss": 3.540785221389646,
      "tokens_seen": 1518338048
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005452575830524795,
      "loss": 0.072,
      "theoretical_loss": 3.5407318850726517,
      "tokens_seen": 1518600192
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005451773391108972,
      "loss": 0.0713,
      "theoretical_loss": 3.5406785605393525,
      "tokens_seen": 1518862336
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005450970951693147,
      "loss": 0.0713,
      "theoretical_loss": 3.540625247785111,
      "tokens_seen": 1519124480
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005450168512277323,
      "loss": 0.0701,
      "theoretical_loss": 3.5405719468052945,
      "tokens_seen": 1519386624
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00054493660728615,
      "loss": 0.0682,
      "theoretical_loss": 3.5405186575952716,
      "tokens_seen": 1519648768
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005448563633445675,
      "loss": 0.0697,
      "theoretical_loss": 3.5404653801504127,
      "tokens_seen": 1519910912
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005447761194029851,
      "loss": 0.0713,
      "theoretical_loss": 3.540412114466093,
      "tokens_seen": 1520173056
    },
    {
      "epoch": 0.46,
      "objective/train/advantage_avg": 0.000899738457519561,
      "objective/train/docs_used": 555119,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4808071851730347,
      "objective/train/original_loss": 1.4808071851730347,
      "objective/train/theoretical_loss": 3.5403588605376886,
      "objective/train/tokens_used": 1540895200,
      "objective/train/value_avg": -0.00998687744140625,
      "objective/train/value_loss": 0.0002470078179612756,
      "objective/train/value_max": -8.153915405273438e-05,
      "objective/train/value_min": -0.38720703125,
      "objective/train/value_reward_corr": 0.7553053169241293,
      "objective/train/value_std": 0.01812744140625,
      "objective/train/weight_avg": 1.0010114908218384,
      "objective/train/weighted_lm_loss": 1.4823392629623413,
      "objective/train/weights_max": 1.1878193616867065,
      "objective/train/weights_min": 0.3683270514011383,
      "theoretical_loss": 3.5403588605376886,
      "tokens_seen": 1520435200
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005446958754614027,
      "loss": 0.0679,
      "theoretical_loss": 3.5403588605376886,
      "tokens_seen": 1520435200
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005446156315198203,
      "loss": 0.0681,
      "theoretical_loss": 3.540305618360578,
      "tokens_seen": 1520697344
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005445353875782378,
      "loss": 0.0732,
      "theoretical_loss": 3.540252387930144,
      "tokens_seen": 1520959488
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005444551436366555,
      "loss": 0.0688,
      "theoretical_loss": 3.540199169241771,
      "tokens_seen": 1521221632
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000544374899695073,
      "loss": 0.0711,
      "theoretical_loss": 3.540145962290845,
      "tokens_seen": 1521483776
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005442946557534905,
      "loss": 0.0694,
      "theoretical_loss": 3.5400927670727573,
      "tokens_seen": 1521745920
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005442144118119083,
      "loss": 0.0708,
      "theoretical_loss": 3.5400395835828986,
      "tokens_seen": 1522008064
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005441341678703258,
      "loss": 0.0726,
      "theoretical_loss": 3.539986411816665,
      "tokens_seen": 1522270208
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005440539239287434,
      "loss": 0.0723,
      "theoretical_loss": 3.5399332517694533,
      "tokens_seen": 1522532352
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000543973679987161,
      "loss": 0.0685,
      "theoretical_loss": 3.539880103436664,
      "tokens_seen": 1522794496
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005438934360455786,
      "loss": 0.0696,
      "theoretical_loss": 3.5398269668136986,
      "tokens_seen": 1523056640
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005438131921039962,
      "loss": 0.0677,
      "theoretical_loss": 3.539773841895964,
      "tokens_seen": 1523318784
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005437329481624137,
      "loss": 0.0707,
      "theoretical_loss": 3.5397207286788666,
      "tokens_seen": 1523580928
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005436527042208313,
      "loss": 0.0736,
      "theoretical_loss": 3.539667627157818,
      "tokens_seen": 1523843072
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005435724602792489,
      "loss": 0.0704,
      "theoretical_loss": 3.53961453732823,
      "tokens_seen": 1524105216
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005434922163376666,
      "loss": 0.0697,
      "theoretical_loss": 3.53956145918552,
      "tokens_seen": 1524367360
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005434119723960841,
      "loss": 0.0698,
      "theoretical_loss": 3.5395083927251045,
      "tokens_seen": 1524629504
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005433317284545018,
      "loss": 0.0737,
      "theoretical_loss": 3.539455337942405,
      "tokens_seen": 1524891648
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005432514845129193,
      "loss": 0.068,
      "theoretical_loss": 3.5394022948328447,
      "tokens_seen": 1525153792
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005431712405713368,
      "loss": 0.069,
      "theoretical_loss": 3.5393492633918497,
      "tokens_seen": 1525415936
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005430909966297545,
      "loss": 0.0703,
      "theoretical_loss": 3.5392962436148485,
      "tokens_seen": 1525678080
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000543010752688172,
      "loss": 0.0682,
      "theoretical_loss": 3.5392432354972723,
      "tokens_seen": 1525940224
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005429305087465896,
      "loss": 0.0719,
      "theoretical_loss": 3.5391902390345544,
      "tokens_seen": 1526202368
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005428502648050072,
      "loss": 0.0695,
      "theoretical_loss": 3.5391372542221315,
      "tokens_seen": 1526464512
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005427700208634248,
      "loss": 0.0692,
      "theoretical_loss": 3.539084281055443,
      "tokens_seen": 1526726656
    },
    {
      "epoch": 0.46,
      "objective/train/advantage_avg": 0.001032950938679278,
      "objective/train/docs_used": 557485,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3883706331253052,
      "objective/train/original_loss": 1.3883706331253052,
      "objective/train/theoretical_loss": 3.5390313195299283,
      "objective/train/tokens_used": 1547448800,
      "objective/train/value_avg": -0.00554656982421875,
      "objective/train/value_loss": 0.00018550669483374804,
      "objective/train/value_max": -2.9087066650390625e-05,
      "objective/train/value_min": -0.630859375,
      "objective/train/value_reward_corr": 0.6172755507616114,
      "objective/train/value_std": 0.00994110107421875,
      "objective/train/weight_avg": 1.0011130571365356,
      "objective/train/weighted_lm_loss": 1.39009690284729,
      "objective/train/weights_max": 1.3466330766677856,
      "objective/train/weights_min": 0.3682669997215271,
      "theoretical_loss": 3.5390313195299283,
      "tokens_seen": 1526988800
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005426897769218424,
      "loss": 0.0712,
      "theoretical_loss": 3.5390313195299283,
      "tokens_seen": 1526988800
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005426095329802601,
      "loss": 0.067,
      "theoretical_loss": 3.538978369641033,
      "tokens_seen": 1527250944
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005425292890386776,
      "loss": 0.0697,
      "theoretical_loss": 3.538925431384203,
      "tokens_seen": 1527513088
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005424490450970952,
      "loss": 0.0702,
      "theoretical_loss": 3.538872504754888,
      "tokens_seen": 1527775232
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005423688011555128,
      "loss": 0.0697,
      "theoretical_loss": 3.538819589748539,
      "tokens_seen": 1528037376
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005422885572139303,
      "loss": 0.0688,
      "theoretical_loss": 3.5387666863606104,
      "tokens_seen": 1528299520
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000542208313272348,
      "loss": 0.0718,
      "theoretical_loss": 3.5387137945865588,
      "tokens_seen": 1528561664
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005421280693307655,
      "loss": 0.0701,
      "theoretical_loss": 3.538660914421844,
      "tokens_seen": 1528823808
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005420478253891831,
      "loss": 0.0718,
      "theoretical_loss": 3.5386080458619276,
      "tokens_seen": 1529085952
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005419675814476008,
      "loss": 0.0709,
      "theoretical_loss": 3.538555188902274,
      "tokens_seen": 1529348096
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005418873375060183,
      "loss": 0.0701,
      "theoretical_loss": 3.53850234353835,
      "tokens_seen": 1529610240
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005418070935644359,
      "loss": 0.0691,
      "theoretical_loss": 3.5384495097656252,
      "tokens_seen": 1529872384
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005417268496228535,
      "loss": 0.0723,
      "theoretical_loss": 3.5383966875795716,
      "tokens_seen": 1530134528
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005416466056812711,
      "loss": 0.0719,
      "theoretical_loss": 3.538343876975664,
      "tokens_seen": 1530396672
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005415663617396886,
      "loss": 0.0714,
      "theoretical_loss": 3.5382910779493795,
      "tokens_seen": 1530658816
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005414861177981063,
      "loss": 0.07,
      "theoretical_loss": 3.538238290496198,
      "tokens_seen": 1530920960
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005414058738565238,
      "loss": 0.0688,
      "theoretical_loss": 3.5381855146116017,
      "tokens_seen": 1531183104
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005413256299149413,
      "loss": 0.0717,
      "theoretical_loss": 3.5381327502910747,
      "tokens_seen": 1531445248
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000541245385973359,
      "loss": 0.0708,
      "theoretical_loss": 3.538079997530105,
      "tokens_seen": 1531707392
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005411651420317766,
      "loss": 0.072,
      "theoretical_loss": 3.538027256324182,
      "tokens_seen": 1531969536
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005410848980901943,
      "loss": 0.0696,
      "theoretical_loss": 3.5379745266687985,
      "tokens_seen": 1532231680
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005410046541486118,
      "loss": 0.0676,
      "theoretical_loss": 3.537921808559449,
      "tokens_seen": 1532493824
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005409244102070294,
      "loss": 0.07,
      "theoretical_loss": 3.537869101991631,
      "tokens_seen": 1532755968
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000540844166265447,
      "loss": 0.0731,
      "theoretical_loss": 3.537816406960845,
      "tokens_seen": 1533018112
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005407639223238645,
      "loss": 0.0715,
      "theoretical_loss": 3.537763723462593,
      "tokens_seen": 1533280256
    },
    {
      "epoch": 0.46,
      "objective/train/advantage_avg": 0.0005670369719155133,
      "objective/train/docs_used": 559697,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5663200616836548,
      "objective/train/original_loss": 1.5663201808929443,
      "objective/train/theoretical_loss": 3.537711051492379,
      "objective/train/tokens_used": 1554002400,
      "objective/train/value_avg": -0.00865936279296875,
      "objective/train/value_loss": 0.000221301379497163,
      "objective/train/value_max": -4.267692565917969e-05,
      "objective/train/value_min": -0.316650390625,
      "objective/train/value_reward_corr": 0.6577696228437152,
      "objective/train/value_std": 0.0131683349609375,
      "objective/train/weight_avg": 1.0006672143936157,
      "objective/train/weighted_lm_loss": 1.5662798881530762,
      "objective/train/weights_max": 1.3565322160720825,
      "objective/train/weights_min": 0.37095150351524353,
      "theoretical_loss": 3.537711051492379,
      "tokens_seen": 1533542400
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005406836783822821,
      "loss": 0.0736,
      "theoretical_loss": 3.537711051492379,
      "tokens_seen": 1533542400
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005406034344406997,
      "loss": 0.0708,
      "theoretical_loss": 3.5376583910457127,
      "tokens_seen": 1533804544
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005405231904991174,
      "loss": 0.0711,
      "theoretical_loss": 3.537605742118102,
      "tokens_seen": 1534066688
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0005404429465575349,
      "loss": 0.0721,
      "theoretical_loss": 3.537553104705061,
      "tokens_seen": 1534328832
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005403627026159526,
      "loss": 0.0747,
      "theoretical_loss": 3.5375004788021043,
      "tokens_seen": 1534590976
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005402824586743701,
      "loss": 0.0725,
      "theoretical_loss": 3.537447864404749,
      "tokens_seen": 1534853120
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005402022147327876,
      "loss": 0.0714,
      "theoretical_loss": 3.5373952615085154,
      "tokens_seen": 1535115264
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005401219707912053,
      "loss": 0.0695,
      "theoretical_loss": 3.5373426701089263,
      "tokens_seen": 1535377408
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005400417268496228,
      "loss": 0.0728,
      "theoretical_loss": 3.5372900902015063,
      "tokens_seen": 1535639552
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005399614829080405,
      "loss": 0.0734,
      "theoretical_loss": 3.537237521781784,
      "tokens_seen": 1535901696
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.000539881238966458,
      "loss": 0.072,
      "theoretical_loss": 3.537184964845289,
      "tokens_seen": 1536163840
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005398009950248756,
      "loss": 0.0756,
      "theoretical_loss": 3.5371324193875533,
      "tokens_seen": 1536425984
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005397207510832933,
      "loss": 0.0702,
      "theoretical_loss": 3.537079885404113,
      "tokens_seen": 1536688128
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005396405071417108,
      "loss": 0.0732,
      "theoretical_loss": 3.5370273628905045,
      "tokens_seen": 1536950272
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005395602632001284,
      "loss": 0.0706,
      "theoretical_loss": 3.5369748518422695,
      "tokens_seen": 1537212416
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.000539480019258546,
      "loss": 0.0733,
      "theoretical_loss": 3.5369223522549493,
      "tokens_seen": 1537474560
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005393997753169636,
      "loss": 0.0715,
      "theoretical_loss": 3.53686986412409,
      "tokens_seen": 1537736704
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005393195313753811,
      "loss": 0.0711,
      "theoretical_loss": 3.5368173874452378,
      "tokens_seen": 1537998848
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005392392874337988,
      "loss": 0.0716,
      "theoretical_loss": 3.536764922213944,
      "tokens_seen": 1538260992
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005391590434922163,
      "loss": 0.071,
      "theoretical_loss": 3.536712468425761,
      "tokens_seen": 1538523136
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005390787995506339,
      "loss": 0.0705,
      "theoretical_loss": 3.5366600260762433,
      "tokens_seen": 1538785280
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005389985556090516,
      "loss": 0.069,
      "theoretical_loss": 3.5366075951609486,
      "tokens_seen": 1539047424
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005389183116674691,
      "loss": 0.0712,
      "theoretical_loss": 3.5365551756754376,
      "tokens_seen": 1539309568
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005388380677258867,
      "loss": 0.0711,
      "theoretical_loss": 3.5365027676152714,
      "tokens_seen": 1539571712
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005387578237843043,
      "loss": 0.0726,
      "theoretical_loss": 3.5364503709760164,
      "tokens_seen": 1539833856
    },
    {
      "epoch": 0.47,
      "objective/train/advantage_avg": 0.0006242617382667959,
      "objective/train/docs_used": 562258,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4021567106246948,
      "objective/train/original_loss": 1.4021568298339844,
      "objective/train/theoretical_loss": 3.536397985753239,
      "objective/train/tokens_used": 1560556000,
      "objective/train/value_avg": -0.007061004638671875,
      "objective/train/value_loss": 0.00021928877686150372,
      "objective/train/value_max": -5.561113357543945e-05,
      "objective/train/value_min": -0.7822265625,
      "objective/train/value_reward_corr": 0.7640836101638206,
      "objective/train/value_std": 0.0187225341796875,
      "objective/train/weight_avg": 1.0007275342941284,
      "objective/train/weighted_lm_loss": 1.4030952453613281,
      "objective/train/weights_max": 1.5728368759155273,
      "objective/train/weights_min": 0.5366626381874084,
      "theoretical_loss": 3.536397985753239,
      "tokens_seen": 1540096000
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005386775798427219,
      "loss": 0.0733,
      "theoretical_loss": 3.536397985753239,
      "tokens_seen": 1540096000
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005385973359011395,
      "loss": 0.0762,
      "theoretical_loss": 3.53634561194251,
      "tokens_seen": 1540358144
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005385170919595571,
      "loss": 0.0726,
      "theoretical_loss": 3.5362932495394013,
      "tokens_seen": 1540620288
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005384368480179746,
      "loss": 0.0696,
      "theoretical_loss": 3.5362408985394875,
      "tokens_seen": 1540882432
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005383566040763922,
      "loss": 0.0713,
      "theoretical_loss": 3.5361885589383464,
      "tokens_seen": 1541144576
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005382763601348099,
      "loss": 0.0721,
      "theoretical_loss": 3.5361362307315574,
      "tokens_seen": 1541406720
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005381961161932274,
      "loss": 0.0726,
      "theoretical_loss": 3.5360839139147036,
      "tokens_seen": 1541668864
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005381158722516451,
      "loss": 0.0716,
      "theoretical_loss": 3.5360316084833685,
      "tokens_seen": 1541931008
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005380356283100626,
      "loss": 0.0704,
      "theoretical_loss": 3.53597931443314,
      "tokens_seen": 1542193152
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005379553843684802,
      "loss": 0.0723,
      "theoretical_loss": 3.535927031759608,
      "tokens_seen": 1542455296
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005378751404268978,
      "loss": 0.072,
      "theoretical_loss": 3.5358747604583636,
      "tokens_seen": 1542717440
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005377948964853153,
      "loss": 0.0706,
      "theoretical_loss": 3.5358225005250024,
      "tokens_seen": 1542979584
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005377146525437329,
      "loss": 0.0735,
      "theoretical_loss": 3.535770251955121,
      "tokens_seen": 1543241728
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005376344086021505,
      "loss": 0.07,
      "theoretical_loss": 3.5357180147443197,
      "tokens_seen": 1543503872
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005375541646605681,
      "loss": 0.073,
      "theoretical_loss": 3.5356657888881986,
      "tokens_seen": 1543766016
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005374739207189858,
      "loss": 0.0704,
      "theoretical_loss": 3.5356135743823636,
      "tokens_seen": 1544028160
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005373936767774034,
      "loss": 0.0723,
      "theoretical_loss": 3.5355613712224203,
      "tokens_seen": 1544290304
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005373134328358209,
      "loss": 0.0744,
      "theoretical_loss": 3.53550917940398,
      "tokens_seen": 1544552448
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005372331888942385,
      "loss": 0.0691,
      "theoretical_loss": 3.535456998922652,
      "tokens_seen": 1544814592
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005371529449526561,
      "loss": 0.0711,
      "theoretical_loss": 3.535404829774052,
      "tokens_seen": 1545076736
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005370727010110736,
      "loss": 0.0699,
      "theoretical_loss": 3.535352671953796,
      "tokens_seen": 1545338880
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005369924570694913,
      "loss": 0.0724,
      "theoretical_loss": 3.5353005254575027,
      "tokens_seen": 1545601024
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005369122131279088,
      "loss": 0.0769,
      "theoretical_loss": 3.5352483902807945,
      "tokens_seen": 1545863168
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005368319691863264,
      "loss": 0.0718,
      "theoretical_loss": 3.535196266419295,
      "tokens_seen": 1546125312
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005367517252447441,
      "loss": 0.0721,
      "theoretical_loss": 3.5351441538686306,
      "tokens_seen": 1546387456
    },
    {
      "epoch": 0.47,
      "objective/train/advantage_avg": 0.0003114771971013397,
      "objective/train/docs_used": 564726,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4200135469436646,
      "objective/train/original_loss": 1.420013427734375,
      "objective/train/theoretical_loss": 3.535092052624429,
      "objective/train/tokens_used": 1567109600,
      "objective/train/value_avg": -0.00567626953125,
      "objective/train/value_loss": 0.0001583931443747133,
      "objective/train/value_max": -3.0934810638427734e-05,
      "objective/train/value_min": -0.67724609375,
      "objective/train/value_reward_corr": 0.6668066682699528,
      "objective/train/value_std": 0.0106658935546875,
      "objective/train/weight_avg": 1.0003831386566162,
      "objective/train/weighted_lm_loss": 1.4202409982681274,
      "objective/train/weights_max": 1.356201171875,
      "objective/train/weights_min": 0.3696674406528473,
      "theoretical_loss": 3.535092052624429,
      "tokens_seen": 1546649600
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005366714813031616,
      "loss": 0.0688,
      "theoretical_loss": 3.535092052624429,
      "tokens_seen": 1546649600
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005365912373615792,
      "loss": 0.0735,
      "theoretical_loss": 3.5350399626823226,
      "tokens_seen": 1546911744
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005365109934199968,
      "loss": 0.0729,
      "theoretical_loss": 3.534987884037945,
      "tokens_seen": 1547173888
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005364307494784144,
      "loss": 0.072,
      "theoretical_loss": 3.5349358166869314,
      "tokens_seen": 1547436032
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005363505055368319,
      "loss": 0.0719,
      "theoretical_loss": 3.534883760624921,
      "tokens_seen": 1547698176
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005362702615952496,
      "loss": 0.0765,
      "theoretical_loss": 3.534831715847555,
      "tokens_seen": 1547960320
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005361900176536671,
      "loss": 0.0733,
      "theoretical_loss": 3.534779682350475,
      "tokens_seen": 1548222464
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005361097737120849,
      "loss": 0.0748,
      "theoretical_loss": 3.534727660129329,
      "tokens_seen": 1548484608
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005360295297705024,
      "loss": 0.071,
      "theoretical_loss": 3.534675649179764,
      "tokens_seen": 1548746752
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005359492858289199,
      "loss": 0.0735,
      "theoretical_loss": 3.53462364949743,
      "tokens_seen": 1549008896
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005358690418873376,
      "loss": 0.0706,
      "theoretical_loss": 3.5345716610779814,
      "tokens_seen": 1549271040
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005357887979457551,
      "loss": 0.0721,
      "theoretical_loss": 3.5345196839170723,
      "tokens_seen": 1549533184
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005357085540041727,
      "loss": 0.0737,
      "theoretical_loss": 3.534467718010361,
      "tokens_seen": 1549795328
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005356283100625903,
      "loss": 0.0716,
      "theoretical_loss": 3.5344157633535085,
      "tokens_seen": 1550057472
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005355480661210079,
      "loss": 0.0723,
      "theoretical_loss": 3.5343638199421763,
      "tokens_seen": 1550319616
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005354678221794254,
      "loss": 0.0732,
      "theoretical_loss": 3.5343118877720294,
      "tokens_seen": 1550581760
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.000535387578237843,
      "loss": 0.0717,
      "theoretical_loss": 3.534259966838736,
      "tokens_seen": 1550843904
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005353073342962607,
      "loss": 0.0696,
      "theoretical_loss": 3.534208057137966,
      "tokens_seen": 1551106048
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005352270903546782,
      "loss": 0.0727,
      "theoretical_loss": 3.5341561586653905,
      "tokens_seen": 1551368192
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005351468464130959,
      "loss": 0.0693,
      "theoretical_loss": 3.5341042714166853,
      "tokens_seen": 1551630336
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005350666024715134,
      "loss": 0.0702,
      "theoretical_loss": 3.5340523953875267,
      "tokens_seen": 1551892480
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005349863585299311,
      "loss": 0.0706,
      "theoretical_loss": 3.5340005305735946,
      "tokens_seen": 1552154624
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005349061145883486,
      "loss": 0.0686,
      "theoretical_loss": 3.533948676970571,
      "tokens_seen": 1552416768
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005348258706467661,
      "loss": 0.072,
      "theoretical_loss": 3.533896834574139,
      "tokens_seen": 1552678912
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005347456267051838,
      "loss": 0.0722,
      "theoretical_loss": 3.5338450033799864,
      "tokens_seen": 1552941056
    },
    {
      "epoch": 0.47,
      "objective/train/advantage_avg": -0.0007033547153696418,
      "objective/train/docs_used": 567156,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3911396265029907,
      "objective/train/original_loss": 1.3911396265029907,
      "objective/train/theoretical_loss": 3.533793183383802,
      "objective/train/tokens_used": 1573663200,
      "objective/train/value_avg": -0.007610321044921875,
      "objective/train/value_loss": 0.00024034272064454854,
      "objective/train/value_max": -4.988908767700195e-05,
      "objective/train/value_min": -0.38916015625,
      "objective/train/value_reward_corr": 0.6741884057137439,
      "objective/train/value_std": 0.013458251953125,
      "objective/train/weight_avg": 0.9994078874588013,
      "objective/train/weighted_lm_loss": 1.3903698921203613,
      "objective/train/weights_max": 1.2013026475906372,
      "objective/train/weights_min": 0.37362140417099,
      "theoretical_loss": 3.533793183383802,
      "tokens_seen": 1553203200
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005346653827636013,
      "loss": 0.0693,
      "theoretical_loss": 3.533793183383802,
      "tokens_seen": 1553203200
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.000534585138822019,
      "loss": 0.0685,
      "theoretical_loss": 3.5337413745812767,
      "tokens_seen": 1553465344
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005345048948804366,
      "loss": 0.0718,
      "theoretical_loss": 3.5336895769681043,
      "tokens_seen": 1553727488
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005344246509388542,
      "loss": 0.0697,
      "theoretical_loss": 3.533637790539981,
      "tokens_seen": 1553989632
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005343444069972717,
      "loss": 0.0671,
      "theoretical_loss": 3.533586015292606,
      "tokens_seen": 1554251776
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005342641630556893,
      "loss": 0.0708,
      "theoretical_loss": 3.5335342512216794,
      "tokens_seen": 1554513920
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005341839191141069,
      "loss": 0.073,
      "theoretical_loss": 3.5334824983229045,
      "tokens_seen": 1554776064
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005341036751725244,
      "loss": 0.072,
      "theoretical_loss": 3.5334307565919874,
      "tokens_seen": 1555038208
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005340234312309421,
      "loss": 0.0695,
      "theoretical_loss": 3.533379026024636,
      "tokens_seen": 1555300352
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005339431872893596,
      "loss": 0.0719,
      "theoretical_loss": 3.53332730661656,
      "tokens_seen": 1555562496
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005338629433477772,
      "loss": 0.0716,
      "theoretical_loss": 3.533275598363473,
      "tokens_seen": 1555824640
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005337826994061949,
      "loss": 0.0676,
      "theoretical_loss": 3.53322390126109,
      "tokens_seen": 1556086784
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005337024554646124,
      "loss": 0.0731,
      "theoretical_loss": 3.533172215305129,
      "tokens_seen": 1556348928
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005336222115230301,
      "loss": 0.0681,
      "theoretical_loss": 3.533120540491309,
      "tokens_seen": 1556611072
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005335419675814476,
      "loss": 0.0728,
      "theoretical_loss": 3.533068876815352,
      "tokens_seen": 1556873216
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005334617236398652,
      "loss": 0.0687,
      "theoretical_loss": 3.5330172242729834,
      "tokens_seen": 1557135360
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005333814796982828,
      "loss": 0.072,
      "theoretical_loss": 3.53296558285993,
      "tokens_seen": 1557397504
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005333012357567004,
      "loss": 0.067,
      "theoretical_loss": 3.532913952571921,
      "tokens_seen": 1557659648
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005332209918151179,
      "loss": 0.0699,
      "theoretical_loss": 3.5328623334046885,
      "tokens_seen": 1557921792
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005331407478735357,
      "loss": 0.0697,
      "theoretical_loss": 3.532810725353966,
      "tokens_seen": 1558183936
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005330605039319532,
      "loss": 0.0721,
      "theoretical_loss": 3.5327591284154893,
      "tokens_seen": 1558446080
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005329802599903707,
      "loss": 0.0692,
      "theoretical_loss": 3.5327075425849985,
      "tokens_seen": 1558708224
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005329000160487884,
      "loss": 0.0704,
      "theoretical_loss": 3.5326559678582337,
      "tokens_seen": 1558970368
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005328197721072059,
      "loss": 0.0701,
      "theoretical_loss": 3.532604404230939,
      "tokens_seen": 1559232512
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005327395281656235,
      "loss": 0.0688,
      "theoretical_loss": 3.53255285169886,
      "tokens_seen": 1559494656
    },
    {
      "epoch": 0.47,
      "objective/train/advantage_avg": 0.0007141765672713518,
      "objective/train/docs_used": 569616,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4324969053268433,
      "objective/train/original_loss": 1.4324970245361328,
      "objective/train/theoretical_loss": 3.5325013102577443,
      "objective/train/tokens_used": 1580216800,
      "objective/train/value_avg": -0.00835418701171875,
      "objective/train/value_loss": 0.0002993447124026716,
      "objective/train/value_max": -0.00010073184967041016,
      "objective/train/value_min": -0.72412109375,
      "objective/train/value_reward_corr": 0.7243982875759303,
      "objective/train/value_std": 0.018646240234375,
      "objective/train/weight_avg": 1.0008517503738403,
      "objective/train/weighted_lm_loss": 1.433107614517212,
      "objective/train/weights_max": 1.4205234050750732,
      "objective/train/weights_min": 0.36909955739974976,
      "theoretical_loss": 3.5325013102577443,
      "tokens_seen": 1559756800
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005326592842240411,
      "loss": 0.0721,
      "theoretical_loss": 3.5325013102577443,
      "tokens_seen": 1559756800
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005325790402824587,
      "loss": 0.0729,
      "theoretical_loss": 3.532449779903343,
      "tokens_seen": 1560018944
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005324987963408762,
      "loss": 0.0693,
      "theoretical_loss": 3.5323982606314086,
      "tokens_seen": 1560281088
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005324185523992938,
      "loss": 0.0722,
      "theoretical_loss": 3.5323467524376966,
      "tokens_seen": 1560543232
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005323383084577115,
      "loss": 0.0715,
      "theoretical_loss": 3.532295255317964,
      "tokens_seen": 1560805376
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005322580645161291,
      "loss": 0.0673,
      "theoretical_loss": 3.532243769267971,
      "tokens_seen": 1561067520
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005321778205745467,
      "loss": 0.0701,
      "theoretical_loss": 3.5321922942834796,
      "tokens_seen": 1561329664
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005320975766329642,
      "loss": 0.0738,
      "theoretical_loss": 3.5321408303602544,
      "tokens_seen": 1561591808
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005320173326913819,
      "loss": 0.0701,
      "theoretical_loss": 3.532089377494062,
      "tokens_seen": 1561853952
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005319370887497994,
      "loss": 0.0753,
      "theoretical_loss": 3.532037935680672,
      "tokens_seen": 1562116096
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005318568448082169,
      "loss": 0.0714,
      "theoretical_loss": 3.5319865049158556,
      "tokens_seen": 1562378240
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005317766008666346,
      "loss": 0.0686,
      "theoretical_loss": 3.5319350851953866,
      "tokens_seen": 1562640384
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005316963569250521,
      "loss": 0.0703,
      "theoretical_loss": 3.531883676515041,
      "tokens_seen": 1562902528
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005316161129834697,
      "loss": 0.073,
      "theoretical_loss": 3.5318322788705974,
      "tokens_seen": 1563164672
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005315358690418874,
      "loss": 0.068,
      "theoretical_loss": 3.531780892257837,
      "tokens_seen": 1563426816
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.000531455625100305,
      "loss": 0.0691,
      "theoretical_loss": 3.5317295166725424,
      "tokens_seen": 1563688960
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005313753811587225,
      "loss": 0.0688,
      "theoretical_loss": 3.5316781521104996,
      "tokens_seen": 1563951104
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005312951372171401,
      "loss": 0.0687,
      "theoretical_loss": 3.5316267985674954,
      "tokens_seen": 1564213248
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005312148932755577,
      "loss": 0.0695,
      "theoretical_loss": 3.5315754560393207,
      "tokens_seen": 1564475392
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005311346493339753,
      "loss": 0.0688,
      "theoretical_loss": 3.5315241245217672,
      "tokens_seen": 1564737536
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005310544053923929,
      "loss": 0.0697,
      "theoretical_loss": 3.5314728040106305,
      "tokens_seen": 1564999680
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005309741614508104,
      "loss": 0.073,
      "theoretical_loss": 3.531421494501707,
      "tokens_seen": 1565261824
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005308939175092282,
      "loss": 0.0713,
      "theoretical_loss": 3.5313701959907955,
      "tokens_seen": 1565523968
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005308136735676457,
      "loss": 0.0715,
      "theoretical_loss": 3.5313189084736987,
      "tokens_seen": 1565786112
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005307334296260632,
      "loss": 0.0715,
      "theoretical_loss": 3.5312676319462195,
      "tokens_seen": 1566048256
    },
    {
      "epoch": 0.47,
      "objective/train/advantage_avg": -0.00031399555155076087,
      "objective/train/docs_used": 572070,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3572434186935425,
      "objective/train/original_loss": 1.3572434186935425,
      "objective/train/theoretical_loss": 3.531216366404165,
      "objective/train/tokens_used": 1586770400,
      "objective/train/value_avg": -0.009185791015625,
      "objective/train/value_loss": 0.00043531539267860353,
      "objective/train/value_max": -3.11732292175293e-05,
      "objective/train/value_min": -0.97998046875,
      "objective/train/value_reward_corr": 0.7681454448311413,
      "objective/train/value_std": 0.0237579345703125,
      "objective/train/weight_avg": 0.999882698059082,
      "objective/train/weighted_lm_loss": 1.3566851615905762,
      "objective/train/weights_max": 2.3744029998779297,
      "objective/train/weights_min": 0.3780485987663269,
      "theoretical_loss": 3.531216366404165,
      "tokens_seen": 1566310400
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005306531856844809,
      "loss": 0.0684,
      "theoretical_loss": 3.531216366404165,
      "tokens_seen": 1566310400
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005305729417428984,
      "loss": 0.0682,
      "theoretical_loss": 3.531165111843343,
      "tokens_seen": 1566572544
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.000530492697801316,
      "loss": 0.0709,
      "theoretical_loss": 3.531113868259565,
      "tokens_seen": 1566834688
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005304124538597336,
      "loss": 0.0684,
      "theoretical_loss": 3.5310626356486434,
      "tokens_seen": 1567096832
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0005303322099181512,
      "loss": 0.0689,
      "theoretical_loss": 3.5310114140063944,
      "tokens_seen": 1567358976
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005302519659765687,
      "loss": 0.0703,
      "theoretical_loss": 3.530960203328635,
      "tokens_seen": 1567621120
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005301717220349863,
      "loss": 0.0704,
      "theoretical_loss": 3.5309090036111854,
      "tokens_seen": 1567883264
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000530091478093404,
      "loss": 0.0704,
      "theoretical_loss": 3.5308578148498686,
      "tokens_seen": 1568145408
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005300112341518215,
      "loss": 0.0705,
      "theoretical_loss": 3.5308066370405076,
      "tokens_seen": 1568407552
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005299309902102392,
      "loss": 0.0695,
      "theoretical_loss": 3.530755470178931,
      "tokens_seen": 1568669696
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005298507462686567,
      "loss": 0.0697,
      "theoretical_loss": 3.5307043142609666,
      "tokens_seen": 1568931840
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005297705023270744,
      "loss": 0.0696,
      "theoretical_loss": 3.530653169282447,
      "tokens_seen": 1569193984
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005296902583854919,
      "loss": 0.0689,
      "theoretical_loss": 3.5306020352392053,
      "tokens_seen": 1569456128
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005296100144439095,
      "loss": 0.0699,
      "theoretical_loss": 3.5305509121270777,
      "tokens_seen": 1569718272
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005295297705023271,
      "loss": 0.0664,
      "theoretical_loss": 3.5304997999419028,
      "tokens_seen": 1569980416
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005294495265607446,
      "loss": 0.0693,
      "theoretical_loss": 3.5304486986795203,
      "tokens_seen": 1570242560
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005293692826191623,
      "loss": 0.0701,
      "theoretical_loss": 3.5303976083357735,
      "tokens_seen": 1570504704
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005292890386775799,
      "loss": 0.0648,
      "theoretical_loss": 3.5303465289065077,
      "tokens_seen": 1570766848
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005292087947359975,
      "loss": 0.0674,
      "theoretical_loss": 3.5302954603875696,
      "tokens_seen": 1571028992
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000529128550794415,
      "loss": 0.0695,
      "theoretical_loss": 3.5302444027748106,
      "tokens_seen": 1571291136
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005290483068528327,
      "loss": 0.0702,
      "theoretical_loss": 3.530193356064081,
      "tokens_seen": 1571553280
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005289680629112502,
      "loss": 0.0689,
      "theoretical_loss": 3.530142320251235,
      "tokens_seen": 1571815424
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005288878189696677,
      "loss": 0.0688,
      "theoretical_loss": 3.5300912953321304,
      "tokens_seen": 1572077568
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005288075750280854,
      "loss": 0.0685,
      "theoretical_loss": 3.530040281302625,
      "tokens_seen": 1572339712
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005287273310865029,
      "loss": 0.0736,
      "theoretical_loss": 3.5299892781585793,
      "tokens_seen": 1572601856
    },
    {
      "epoch": 0.48,
      "objective/train/advantage_avg": 0.0002667908265721053,
      "objective/train/docs_used": 574423,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4923175573349,
      "objective/train/original_loss": 1.492317795753479,
      "objective/train/theoretical_loss": 3.529938285895858,
      "objective/train/tokens_used": 1593324000,
      "objective/train/value_avg": -0.0081787109375,
      "objective/train/value_loss": 0.0002718984615057707,
      "objective/train/value_max": -8.481740951538086e-05,
      "objective/train/value_min": -0.552734375,
      "objective/train/value_reward_corr": 0.7223534480290491,
      "objective/train/value_std": 0.0150909423828125,
      "objective/train/weight_avg": 1.0003913640975952,
      "objective/train/weighted_lm_loss": 1.4929695129394531,
      "objective/train/weights_max": 1.278699278831482,
      "objective/train/weights_min": 0.3693629503250122,
      "theoretical_loss": 3.529938285895858,
      "tokens_seen": 1572864000
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005286470871449207,
      "loss": 0.0723,
      "theoretical_loss": 3.529938285895858,
      "tokens_seen": 1572864000
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005285668432033382,
      "loss": 0.0671,
      "theoretical_loss": 3.529887304510326,
      "tokens_seen": 1573126144
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005284865992617558,
      "loss": 0.0665,
      "theoretical_loss": 3.5298363339978502,
      "tokens_seen": 1573388288
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005284063553201734,
      "loss": 0.0673,
      "theoretical_loss": 3.529785374354302,
      "tokens_seen": 1573650432
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005283261113785909,
      "loss": 0.0687,
      "theoretical_loss": 3.529734425575553,
      "tokens_seen": 1573912576
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005282458674370085,
      "loss": 0.0688,
      "theoretical_loss": 3.5296834876574783,
      "tokens_seen": 1574174720
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005281656234954261,
      "loss": 0.0702,
      "theoretical_loss": 3.529632560595954,
      "tokens_seen": 1574436864
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005280853795538437,
      "loss": 0.0676,
      "theoretical_loss": 3.5295816443868593,
      "tokens_seen": 1574699008
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005280051356122612,
      "loss": 0.0643,
      "theoretical_loss": 3.529530739026076,
      "tokens_seen": 1574961152
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000527924891670679,
      "loss": 0.0678,
      "theoretical_loss": 3.5294798445094875,
      "tokens_seen": 1575223296
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005278446477290965,
      "loss": 0.0702,
      "theoretical_loss": 3.529428960832979,
      "tokens_seen": 1575485440
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000527764403787514,
      "loss": 0.0679,
      "theoretical_loss": 3.5293780879924395,
      "tokens_seen": 1575747584
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005276841598459317,
      "loss": 0.0664,
      "theoretical_loss": 3.529327225983759,
      "tokens_seen": 1576009728
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005276039159043492,
      "loss": 0.0683,
      "theoretical_loss": 3.5292763748028295,
      "tokens_seen": 1576271872
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005275236719627668,
      "loss": 0.0684,
      "theoretical_loss": 3.5292255344455468,
      "tokens_seen": 1576534016
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005274434280211844,
      "loss": 0.0677,
      "theoretical_loss": 3.529174704907807,
      "tokens_seen": 1576796160
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000527363184079602,
      "loss": 0.069,
      "theoretical_loss": 3.529123886185509,
      "tokens_seen": 1577058304
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005272829401380196,
      "loss": 0.0671,
      "theoretical_loss": 3.529073078274556,
      "tokens_seen": 1577320448
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005272026961964371,
      "loss": 0.0679,
      "theoretical_loss": 3.5290222811708505,
      "tokens_seen": 1577582592
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005271224522548548,
      "loss": 0.0703,
      "theoretical_loss": 3.528971494870299,
      "tokens_seen": 1577844736
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005270422083132724,
      "loss": 0.0663,
      "theoretical_loss": 3.5289207193688092,
      "tokens_seen": 1578106880
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00052696196437169,
      "loss": 0.0671,
      "theoretical_loss": 3.5288699546622913,
      "tokens_seen": 1578369024
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005268817204301075,
      "loss": 0.07,
      "theoretical_loss": 3.528819200746659,
      "tokens_seen": 1578631168
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005268014764885252,
      "loss": 0.0709,
      "theoretical_loss": 3.5287684576178258,
      "tokens_seen": 1578893312
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005267212325469427,
      "loss": 0.0711,
      "theoretical_loss": 3.5287177252717097,
      "tokens_seen": 1579155456
    },
    {
      "epoch": 0.48,
      "objective/train/advantage_avg": 0.0015256733167916536,
      "objective/train/docs_used": 576827,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.413283348083496,
      "objective/train/original_loss": 1.413283348083496,
      "objective/train/theoretical_loss": 3.5286670037042303,
      "objective/train/tokens_used": 1599877600,
      "objective/train/value_avg": -0.007694244384765625,
      "objective/train/value_loss": 0.00026877259369939566,
      "objective/train/value_max": -8.481740951538086e-05,
      "objective/train/value_min": -0.9794921875,
      "objective/train/value_reward_corr": 0.7313708202021553,
      "objective/train/value_std": 0.0199737548828125,
      "objective/train/weight_avg": 1.0016505718231201,
      "objective/train/weighted_lm_loss": 1.415053367614746,
      "objective/train/weights_max": 2.1128692626953125,
      "objective/train/weights_min": 0.3688335716724396,
      "theoretical_loss": 3.5286670037042303,
      "tokens_seen": 1579417600
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005266409886053603,
      "loss": 0.07,
      "theoretical_loss": 3.5286670037042303,
      "tokens_seen": 1579417600
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005265607446637779,
      "loss": 0.0683,
      "theoretical_loss": 3.528616292911309,
      "tokens_seen": 1579679744
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005264805007221954,
      "loss": 0.07,
      "theoretical_loss": 3.5285655928888686,
      "tokens_seen": 1579941888
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000526400256780613,
      "loss": 0.0688,
      "theoretical_loss": 3.5285149036328356,
      "tokens_seen": 1580204032
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005263200128390307,
      "loss": 0.0696,
      "theoretical_loss": 3.528464225139139,
      "tokens_seen": 1580466176
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005262397688974483,
      "loss": 0.0731,
      "theoretical_loss": 3.5284135574037085,
      "tokens_seen": 1580728320
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005261595249558658,
      "loss": 0.0699,
      "theoretical_loss": 3.5283629004224766,
      "tokens_seen": 1580990464
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005260792810142835,
      "loss": 0.0709,
      "theoretical_loss": 3.5283122541913787,
      "tokens_seen": 1581252608
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000525999037072701,
      "loss": 0.0687,
      "theoretical_loss": 3.5282616187063516,
      "tokens_seen": 1581514752
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005259187931311186,
      "loss": 0.0686,
      "theoretical_loss": 3.528210993963334,
      "tokens_seen": 1581776896
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005258385491895362,
      "loss": 0.0714,
      "theoretical_loss": 3.528160379958268,
      "tokens_seen": 1582039040
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005257583052479537,
      "loss": 0.0707,
      "theoretical_loss": 3.528109776687097,
      "tokens_seen": 1582301184
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005256780613063715,
      "loss": 0.0721,
      "theoretical_loss": 3.528059184145767,
      "tokens_seen": 1582563328
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000525597817364789,
      "loss": 0.0704,
      "theoretical_loss": 3.5280086023302264,
      "tokens_seen": 1582825472
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005255175734232066,
      "loss": 0.0687,
      "theoretical_loss": 3.527958031236425,
      "tokens_seen": 1583087616
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005254373294816242,
      "loss": 0.0693,
      "theoretical_loss": 3.527907470860315,
      "tokens_seen": 1583349760
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005253570855400417,
      "loss": 0.0709,
      "theoretical_loss": 3.527856921197852,
      "tokens_seen": 1583611904
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005252768415984593,
      "loss": 0.0709,
      "theoretical_loss": 3.5278063822449925,
      "tokens_seen": 1583874048
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005251965976568769,
      "loss": 0.0699,
      "theoretical_loss": 3.5277558539976956,
      "tokens_seen": 1584136192
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005251163537152945,
      "loss": 0.0702,
      "theoretical_loss": 3.5277053364519215,
      "tokens_seen": 1584398336
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000525036109773712,
      "loss": 0.0707,
      "theoretical_loss": 3.5276548296036356,
      "tokens_seen": 1584660480
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005249558658321298,
      "loss": 0.0684,
      "theoretical_loss": 3.527604333448802,
      "tokens_seen": 1584922624
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005248756218905473,
      "loss": 0.0687,
      "theoretical_loss": 3.527553847983389,
      "tokens_seen": 1585184768
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005247953779489648,
      "loss": 0.07,
      "theoretical_loss": 3.5275033732033667,
      "tokens_seen": 1585446912
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005247151340073825,
      "loss": 0.067,
      "theoretical_loss": 3.5274529091047073,
      "tokens_seen": 1585709056
    },
    {
      "epoch": 0.48,
      "objective/train/advantage_avg": 0.00030591359245590866,
      "objective/train/docs_used": 579278,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2970331907272339,
      "objective/train/original_loss": 1.2970331907272339,
      "objective/train/theoretical_loss": 3.527402455683385,
      "objective/train/tokens_used": 1606431200,
      "objective/train/value_avg": -0.00768280029296875,
      "objective/train/value_loss": 0.00016412846161983907,
      "objective/train/value_max": -5.608797073364258e-05,
      "objective/train/value_min": -0.393310546875,
      "objective/train/value_reward_corr": 0.7478747737276108,
      "objective/train/value_std": 0.013519287109375,
      "objective/train/weight_avg": 1.000381588935852,
      "objective/train/weighted_lm_loss": 1.2976466417312622,
      "objective/train/weights_max": 1.1971355676651,
      "objective/train/weights_min": 0.36882299184799194,
      "theoretical_loss": 3.527402455683385,
      "tokens_seen": 1585971200
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005246348900658,
      "loss": 0.0679,
      "theoretical_loss": 3.527402455683385,
      "tokens_seen": 1585971200
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005245546461242177,
      "loss": 0.0697,
      "theoretical_loss": 3.527352012935377,
      "tokens_seen": 1586233344
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005244744021826352,
      "loss": 0.0696,
      "theoretical_loss": 3.527301580856661,
      "tokens_seen": 1586495488
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005243941582410528,
      "loss": 0.0702,
      "theoretical_loss": 3.527251159443219,
      "tokens_seen": 1586757632
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005243139142994704,
      "loss": 0.0722,
      "theoretical_loss": 3.5272007486910333,
      "tokens_seen": 1587019776
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005242336703578879,
      "loss": 0.0701,
      "theoretical_loss": 3.5271503485960896,
      "tokens_seen": 1587281920
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005241534264163056,
      "loss": 0.0694,
      "theoretical_loss": 3.5270999591543752,
      "tokens_seen": 1587544064
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005240731824747232,
      "loss": 0.0685,
      "theoretical_loss": 3.5270495803618793,
      "tokens_seen": 1587806208
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005239929385331408,
      "loss": 0.0656,
      "theoretical_loss": 3.5269992122145952,
      "tokens_seen": 1588068352
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005239126945915583,
      "loss": 0.0678,
      "theoretical_loss": 3.526948854708515,
      "tokens_seen": 1588330496
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000523832450649976,
      "loss": 0.0688,
      "theoretical_loss": 3.526898507839636,
      "tokens_seen": 1588592640
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005237522067083935,
      "loss": 0.0674,
      "theoretical_loss": 3.526848171603956,
      "tokens_seen": 1588854784
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005236719627668111,
      "loss": 0.0706,
      "theoretical_loss": 3.526797845997476,
      "tokens_seen": 1589116928
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005235917188252287,
      "loss": 0.0669,
      "theoretical_loss": 3.5267475310161984,
      "tokens_seen": 1589379072
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005235114748836462,
      "loss": 0.0713,
      "theoretical_loss": 3.5266972266561276,
      "tokens_seen": 1589641216
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000523431230942064,
      "loss": 0.0673,
      "theoretical_loss": 3.526646932913271,
      "tokens_seen": 1589903360
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005233509870004815,
      "loss": 0.0668,
      "theoretical_loss": 3.5265966497836376,
      "tokens_seen": 1590165504
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005232707430588991,
      "loss": 0.0726,
      "theoretical_loss": 3.5265463772632386,
      "tokens_seen": 1590427648
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005231904991173167,
      "loss": 0.0681,
      "theoretical_loss": 3.5264961153480874,
      "tokens_seen": 1590689792
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005231102551757343,
      "loss": 0.0706,
      "theoretical_loss": 3.5264458640342,
      "tokens_seen": 1590951936
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005230300112341518,
      "loss": 0.0701,
      "theoretical_loss": 3.5263956233175935,
      "tokens_seen": 1591214080
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005229497672925694,
      "loss": 0.067,
      "theoretical_loss": 3.5263453931942883,
      "tokens_seen": 1591476224
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000522869523350987,
      "loss": 0.0674,
      "theoretical_loss": 3.5262951736603063,
      "tokens_seen": 1591738368
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005227892794094045,
      "loss": 0.069,
      "theoretical_loss": 3.5262449647116716,
      "tokens_seen": 1592000512
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005227090354678223,
      "loss": 0.0665,
      "theoretical_loss": 3.5261947663444104,
      "tokens_seen": 1592262656
    },
    {
      "epoch": 0.48,
      "objective/train/advantage_avg": -0.0003691379679366946,
      "objective/train/docs_used": 581545,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4430354833602905,
      "objective/train/original_loss": 1.44303560256958,
      "objective/train/theoretical_loss": 3.5261445785545513,
      "objective/train/tokens_used": 1612984800,
      "objective/train/value_avg": -0.006046295166015625,
      "objective/train/value_loss": 0.000189928337931633,
      "objective/train/value_max": -3.534555435180664e-05,
      "objective/train/value_min": -0.60888671875,
      "objective/train/value_reward_corr": 0.6734293856838923,
      "objective/train/value_std": 0.011077880859375,
      "objective/train/weight_avg": 0.9997175931930542,
      "objective/train/weighted_lm_loss": 1.442589521408081,
      "objective/train/weights_max": 1.4511600732803345,
      "objective/train/weights_min": 0.3682490885257721,
      "theoretical_loss": 3.5261445785545513,
      "tokens_seen": 1592524800
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005226287915262398,
      "loss": 0.0679,
      "theoretical_loss": 3.5261445785545513,
      "tokens_seen": 1592524800
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005225485475846574,
      "loss": 0.069,
      "theoretical_loss": 3.526094401338125,
      "tokens_seen": 1592786944
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000522468303643075,
      "loss": 0.0695,
      "theoretical_loss": 3.526044234691165,
      "tokens_seen": 1593049088
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005223880597014925,
      "loss": 0.0681,
      "theoretical_loss": 3.525994078609705,
      "tokens_seen": 1593311232
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005223078157599101,
      "loss": 0.0694,
      "theoretical_loss": 3.525943933089782,
      "tokens_seen": 1593573376
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005222275718183277,
      "loss": 0.0716,
      "theoretical_loss": 3.5258937981274365,
      "tokens_seen": 1593835520
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005221473278767453,
      "loss": 0.0689,
      "theoretical_loss": 3.5258436737187084,
      "tokens_seen": 1594097664
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005220670839351629,
      "loss": 0.0673,
      "theoretical_loss": 3.5257935598596424,
      "tokens_seen": 1594359808
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005219868399935806,
      "loss": 0.0676,
      "theoretical_loss": 3.5257434565462833,
      "tokens_seen": 1594621952
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005219065960519981,
      "loss": 0.0695,
      "theoretical_loss": 3.5256933637746792,
      "tokens_seen": 1594884096
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005218263521104157,
      "loss": 0.0652,
      "theoretical_loss": 3.5256432815408796,
      "tokens_seen": 1595146240
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005217461081688333,
      "loss": 0.0661,
      "theoretical_loss": 3.5255932098409364,
      "tokens_seen": 1595408384
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005216658642272508,
      "loss": 0.068,
      "theoretical_loss": 3.5255431486709043,
      "tokens_seen": 1595670528
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005215856202856685,
      "loss": 0.0651,
      "theoretical_loss": 3.525493098026839,
      "tokens_seen": 1595932672
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000521505376344086,
      "loss": 0.0693,
      "theoretical_loss": 3.5254430579047993,
      "tokens_seen": 1596194816
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005214251324025036,
      "loss": 0.0658,
      "theoretical_loss": 3.5253930283008454,
      "tokens_seen": 1596456960
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005213448884609212,
      "loss": 0.069,
      "theoretical_loss": 3.5253430092110403,
      "tokens_seen": 1596719104
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005212646445193387,
      "loss": 0.067,
      "theoretical_loss": 3.5252930006314482,
      "tokens_seen": 1596981248
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005211844005777564,
      "loss": 0.0678,
      "theoretical_loss": 3.5252430025581356,
      "tokens_seen": 1597243392
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000521104156636174,
      "loss": 0.0679,
      "theoretical_loss": 3.5251930149871726,
      "tokens_seen": 1597505536
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005210239126945916,
      "loss": 0.0675,
      "theoretical_loss": 3.5251430379146296,
      "tokens_seen": 1597767680
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005209436687530092,
      "loss": 0.0674,
      "theoretical_loss": 3.5250930713365802,
      "tokens_seen": 1598029824
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005208634248114268,
      "loss": 0.0656,
      "theoretical_loss": 3.525043115249099,
      "tokens_seen": 1598291968
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005207831808698443,
      "loss": 0.0699,
      "theoretical_loss": 3.5249931696482637,
      "tokens_seen": 1598554112
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005207029369282619,
      "loss": 0.0698,
      "theoretical_loss": 3.524943234530154,
      "tokens_seen": 1598816256
    },
    {
      "epoch": 0.48,
      "objective/train/advantage_avg": 0.0008163555758073926,
      "objective/train/docs_used": 583998,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3744968175888062,
      "objective/train/original_loss": 1.3744966983795166,
      "objective/train/theoretical_loss": 3.5248933098908517,
      "objective/train/tokens_used": 1619538400,
      "objective/train/value_avg": -0.00650787353515625,
      "objective/train/value_loss": 0.000182785777724348,
      "objective/train/value_max": -3.24249267578125e-05,
      "objective/train/value_min": -0.49658203125,
      "objective/train/value_reward_corr": 0.6320282058663177,
      "objective/train/value_std": 0.0113067626953125,
      "objective/train/weight_avg": 1.0009000301361084,
      "objective/train/weighted_lm_loss": 1.3754725456237793,
      "objective/train/weights_max": 1.3192977905273438,
      "objective/train/weights_min": 0.3684886395931244,
      "theoretical_loss": 3.5248933098908517,
      "tokens_seen": 1599078400
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005206226929866795,
      "loss": 0.0687,
      "theoretical_loss": 3.5248933098908517,
      "tokens_seen": 1599078400
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000520542449045097,
      "loss": 0.0682,
      "theoretical_loss": 3.5248433957264402,
      "tokens_seen": 1599340544
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005204622051035148,
      "loss": 0.0703,
      "theoretical_loss": 3.5247934920330053,
      "tokens_seen": 1599602688
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005203819611619323,
      "loss": 0.0702,
      "theoretical_loss": 3.5247435988066353,
      "tokens_seen": 1599864832
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005203017172203499,
      "loss": 0.0694,
      "theoretical_loss": 3.52469371604342,
      "tokens_seen": 1600126976
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0005202214732787675,
      "loss": 0.0692,
      "theoretical_loss": 3.524643843739452,
      "tokens_seen": 1600389120
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005201412293371851,
      "loss": 0.0727,
      "theoretical_loss": 3.524593981890825,
      "tokens_seen": 1600651264
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005200609853956026,
      "loss": 0.069,
      "theoretical_loss": 3.524544130493635,
      "tokens_seen": 1600913408
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005199807414540202,
      "loss": 0.0686,
      "theoretical_loss": 3.5244942895439815,
      "tokens_seen": 1601175552
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005199004975124378,
      "loss": 0.069,
      "theoretical_loss": 3.524444459037965,
      "tokens_seen": 1601437696
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005198202535708553,
      "loss": 0.0686,
      "theoretical_loss": 3.5243946389716867,
      "tokens_seen": 1601699840
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000519740009629273,
      "loss": 0.0704,
      "theoretical_loss": 3.5243448293412527,
      "tokens_seen": 1601961984
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005196597656876906,
      "loss": 0.0705,
      "theoretical_loss": 3.5242950301427696,
      "tokens_seen": 1602224128
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005195795217461083,
      "loss": 0.0675,
      "theoretical_loss": 3.5242452413723457,
      "tokens_seen": 1602486272
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005194992778045258,
      "loss": 0.069,
      "theoretical_loss": 3.5241954630260923,
      "tokens_seen": 1602748416
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005194190338629433,
      "loss": 0.0653,
      "theoretical_loss": 3.524145695100123,
      "tokens_seen": 1603010560
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000519338789921361,
      "loss": 0.0685,
      "theoretical_loss": 3.5240959375905527,
      "tokens_seen": 1603272704
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005192585459797785,
      "loss": 0.0689,
      "theoretical_loss": 3.5240461904934977,
      "tokens_seen": 1603534848
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005191783020381961,
      "loss": 0.0696,
      "theoretical_loss": 3.5239964538050788,
      "tokens_seen": 1603796992
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005190980580966137,
      "loss": 0.0691,
      "theoretical_loss": 3.5239467275214165,
      "tokens_seen": 1604059136
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005190178141550314,
      "loss": 0.0707,
      "theoretical_loss": 3.523897011638635,
      "tokens_seen": 1604321280
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005189375702134489,
      "loss": 0.069,
      "theoretical_loss": 3.5238473061528586,
      "tokens_seen": 1604583424
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005188573262718665,
      "loss": 0.066,
      "theoretical_loss": 3.5237976110602163,
      "tokens_seen": 1604845568
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005187770823302841,
      "loss": 0.0714,
      "theoretical_loss": 3.523747926356837,
      "tokens_seen": 1605107712
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005186968383887016,
      "loss": 0.0684,
      "theoretical_loss": 3.523698252038853,
      "tokens_seen": 1605369856
    },
    {
      "epoch": 0.49,
      "objective/train/advantage_avg": 0.0014611550141125917,
      "objective/train/docs_used": 586391,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3702470064163208,
      "objective/train/original_loss": 1.3702466487884521,
      "objective/train/theoretical_loss": 3.5236485881023976,
      "objective/train/tokens_used": 1626092000,
      "objective/train/value_avg": -0.006923675537109375,
      "objective/train/value_loss": 0.00013820224558003247,
      "objective/train/value_max": -2.6464462280273438e-05,
      "objective/train/value_min": -0.95068359375,
      "objective/train/value_reward_corr": 0.6985552108151838,
      "objective/train/value_std": 0.01245880126953125,
      "objective/train/weight_avg": 1.0015236139297485,
      "objective/train/weighted_lm_loss": 1.3721559047698975,
      "objective/train/weights_max": 1.1226495504379272,
      "objective/train/weights_min": 0.3682887554168701,
      "theoretical_loss": 3.5236485881023976,
      "tokens_seen": 1605632000
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005186165944471193,
      "loss": 0.0676,
      "theoretical_loss": 3.5236485881023976,
      "tokens_seen": 1605632000
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005185363505055368,
      "loss": 0.0671,
      "theoretical_loss": 3.5235989345436076,
      "tokens_seen": 1605894144
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005184561065639545,
      "loss": 0.0693,
      "theoretical_loss": 3.52354929135862,
      "tokens_seen": 1606156288
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000518375862622372,
      "loss": 0.0665,
      "theoretical_loss": 3.5234996585435754,
      "tokens_seen": 1606418432
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005182956186807895,
      "loss": 0.0686,
      "theoretical_loss": 3.5234500360946157,
      "tokens_seen": 1606680576
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005182153747392073,
      "loss": 0.0663,
      "theoretical_loss": 3.5234004240078853,
      "tokens_seen": 1606942720
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005181351307976248,
      "loss": 0.069,
      "theoretical_loss": 3.5233508222795304,
      "tokens_seen": 1607204864
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005180548868560424,
      "loss": 0.0673,
      "theoretical_loss": 3.5233012309057,
      "tokens_seen": 1607467008
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00051797464291446,
      "loss": 0.0685,
      "theoretical_loss": 3.5232516498825426,
      "tokens_seen": 1607729152
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005178943989728776,
      "loss": 0.0706,
      "theoretical_loss": 3.5232020792062126,
      "tokens_seen": 1607991296
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005178141550312951,
      "loss": 0.0671,
      "theoretical_loss": 3.523152518872864,
      "tokens_seen": 1608253440
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005177339110897127,
      "loss": 0.0694,
      "theoretical_loss": 3.5231029688786526,
      "tokens_seen": 1608515584
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005176536671481303,
      "loss": 0.0692,
      "theoretical_loss": 3.523053429219738,
      "tokens_seen": 1608777728
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005175734232065478,
      "loss": 0.0743,
      "theoretical_loss": 3.52300389989228,
      "tokens_seen": 1609039872
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005174931792649656,
      "loss": 0.0716,
      "theoretical_loss": 3.5229543808924415,
      "tokens_seen": 1609302016
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005174129353233831,
      "loss": 0.0684,
      "theoretical_loss": 3.522904872216388,
      "tokens_seen": 1609564160
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005173326913818007,
      "loss": 0.0689,
      "theoretical_loss": 3.5228553738602857,
      "tokens_seen": 1609826304
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005172524474402183,
      "loss": 0.0705,
      "theoretical_loss": 3.5228058858203033,
      "tokens_seen": 1610088448
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005171722034986359,
      "loss": 0.0704,
      "theoretical_loss": 3.5227564080926115,
      "tokens_seen": 1610350592
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005170919595570535,
      "loss": 0.0698,
      "theoretical_loss": 3.5227069406733844,
      "tokens_seen": 1610612736
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000517011715615471,
      "loss": 0.0708,
      "theoretical_loss": 3.5226574835587963,
      "tokens_seen": 1610874880
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005169314716738886,
      "loss": 0.0676,
      "theoretical_loss": 3.522608036745024,
      "tokens_seen": 1611137024
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005168512277323062,
      "loss": 0.0698,
      "theoretical_loss": 3.5225586002282467,
      "tokens_seen": 1611399168
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005167709837907239,
      "loss": 0.0702,
      "theoretical_loss": 3.5225091740046457,
      "tokens_seen": 1611661312
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005166907398491414,
      "loss": 0.072,
      "theoretical_loss": 3.5224597580704033,
      "tokens_seen": 1611923456
    },
    {
      "epoch": 0.49,
      "objective/train/advantage_avg": -0.0002596320991870016,
      "objective/train/docs_used": 588885,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3582322597503662,
      "objective/train/original_loss": 1.3582322597503662,
      "objective/train/theoretical_loss": 3.5224103524217067,
      "objective/train/tokens_used": 1632645600,
      "objective/train/value_avg": -0.008026123046875,
      "objective/train/value_loss": 0.0002943206636700779,
      "objective/train/value_max": -4.684925079345703e-05,
      "objective/train/value_min": -0.91357421875,
      "objective/train/value_reward_corr": 0.7926579179126808,
      "objective/train/value_std": 0.0210723876953125,
      "objective/train/weight_avg": 0.9998724460601807,
      "objective/train/weighted_lm_loss": 1.3586115837097168,
      "objective/train/weights_max": 1.6367690563201904,
      "objective/train/weights_min": 0.3682592809200287,
      "theoretical_loss": 3.5224103524217067,
      "tokens_seen": 1612185600
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005166104959075591,
      "loss": 0.0693,
      "theoretical_loss": 3.5224103524217067,
      "tokens_seen": 1612185600
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005165302519659766,
      "loss": 0.0705,
      "theoretical_loss": 3.522360957054741,
      "tokens_seen": 1612447744
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005164500080243941,
      "loss": 0.0669,
      "theoretical_loss": 3.5223115719656963,
      "tokens_seen": 1612709888
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005163697640828118,
      "loss": 0.0705,
      "theoretical_loss": 3.5222621971507646,
      "tokens_seen": 1612972032
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005162895201412293,
      "loss": 0.0695,
      "theoretical_loss": 3.522212832606138,
      "tokens_seen": 1613234176
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005162092761996469,
      "loss": 0.0714,
      "theoretical_loss": 3.5221634783280122,
      "tokens_seen": 1613496320
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005161290322580645,
      "loss": 0.0691,
      "theoretical_loss": 3.5221141343125852,
      "tokens_seen": 1613758464
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005160487883164821,
      "loss": 0.0713,
      "theoretical_loss": 3.5220648005560555,
      "tokens_seen": 1614020608
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005159685443748997,
      "loss": 0.0748,
      "theoretical_loss": 3.522015477054625,
      "tokens_seen": 1614282752
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005158883004333173,
      "loss": 0.0729,
      "theoretical_loss": 3.521966163804497,
      "tokens_seen": 1614544896
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005158080564917349,
      "loss": 0.0708,
      "theoretical_loss": 3.521916860801877,
      "tokens_seen": 1614807040
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005157278125501525,
      "loss": 0.0704,
      "theoretical_loss": 3.521867568042973,
      "tokens_seen": 1615069184
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005156475686085701,
      "loss": 0.07,
      "theoretical_loss": 3.5218182855239935,
      "tokens_seen": 1615331328
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005155673246669876,
      "loss": 0.0683,
      "theoretical_loss": 3.5217690132411508,
      "tokens_seen": 1615593472
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005154870807254053,
      "loss": 0.0708,
      "theoretical_loss": 3.521719751190658,
      "tokens_seen": 1615855616
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005154068367838228,
      "loss": 0.0696,
      "theoretical_loss": 3.5216704993687307,
      "tokens_seen": 1616117760
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005153265928422403,
      "loss": 0.0694,
      "theoretical_loss": 3.5216212577715873,
      "tokens_seen": 1616379904
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005152463489006581,
      "loss": 0.0711,
      "theoretical_loss": 3.5215720263954458,
      "tokens_seen": 1616642048
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005151661049590756,
      "loss": 0.0692,
      "theoretical_loss": 3.521522805236529,
      "tokens_seen": 1616904192
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005150858610174932,
      "loss": 0.0725,
      "theoretical_loss": 3.52147359429106,
      "tokens_seen": 1617166336
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005150056170759108,
      "loss": 0.0708,
      "theoretical_loss": 3.5214243935552654,
      "tokens_seen": 1617428480
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005149253731343284,
      "loss": 0.0664,
      "theoretical_loss": 3.5213752030253715,
      "tokens_seen": 1617690624
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005148451291927459,
      "loss": 0.07,
      "theoretical_loss": 3.5213260226976084,
      "tokens_seen": 1617952768
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005147648852511635,
      "loss": 0.0702,
      "theoretical_loss": 3.5212768525682074,
      "tokens_seen": 1618214912
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005146846413095811,
      "loss": 0.0714,
      "theoretical_loss": 3.5212276926334027,
      "tokens_seen": 1618477056
    },
    {
      "epoch": 0.49,
      "objective/train/advantage_avg": -8.634123514639214e-05,
      "objective/train/docs_used": 591377,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2014659643173218,
      "objective/train/original_loss": 1.2014658451080322,
      "objective/train/theoretical_loss": 3.5211785428894293,
      "objective/train/tokens_used": 1639199200,
      "objective/train/value_avg": -0.007335662841796875,
      "objective/train/value_loss": 0.00014820354408584535,
      "objective/train/value_max": -2.777576446533203e-05,
      "objective/train/value_min": -0.5302734375,
      "objective/train/value_reward_corr": 0.7609983102667354,
      "objective/train/value_std": 0.0133056640625,
      "objective/train/weight_avg": 0.9999821782112122,
      "objective/train/weighted_lm_loss": 1.2018018960952759,
      "objective/train/weights_max": 1.1847867965698242,
      "objective/train/weights_min": 0.3695588707923889,
      "theoretical_loss": 3.5211785428894293,
      "tokens_seen": 1618739200
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005146043973679986,
      "loss": 0.0672,
      "theoretical_loss": 3.5211785428894293,
      "tokens_seen": 1618739200
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005145241534264164,
      "loss": 0.0719,
      "theoretical_loss": 3.5211294033325258,
      "tokens_seen": 1619001344
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005144439094848339,
      "loss": 0.0687,
      "theoretical_loss": 3.5210802739589306,
      "tokens_seen": 1619263488
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005143636655432516,
      "loss": 0.0697,
      "theoretical_loss": 3.521031154764886,
      "tokens_seen": 1619525632
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005142834216016691,
      "loss": 0.0707,
      "theoretical_loss": 3.520982045746636,
      "tokens_seen": 1619787776
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005142031776600867,
      "loss": 0.0699,
      "theoretical_loss": 3.5209329469004254,
      "tokens_seen": 1620049920
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005141229337185043,
      "loss": 0.0715,
      "theoretical_loss": 3.5208838582225024,
      "tokens_seen": 1620312064
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005140426897769218,
      "loss": 0.0685,
      "theoretical_loss": 3.5208347797091157,
      "tokens_seen": 1620574208
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005139624458353394,
      "loss": 0.0681,
      "theoretical_loss": 3.5207857113565177,
      "tokens_seen": 1620836352
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000513882201893757,
      "loss": 0.0692,
      "theoretical_loss": 3.520736653160962,
      "tokens_seen": 1621098496
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005138019579521747,
      "loss": 0.0693,
      "theoretical_loss": 3.520687605118704,
      "tokens_seen": 1621360640
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005137217140105922,
      "loss": 0.072,
      "theoretical_loss": 3.520638567226001,
      "tokens_seen": 1621622784
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005136414700690099,
      "loss": 0.0697,
      "theoretical_loss": 3.5205895394791127,
      "tokens_seen": 1621884928
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005135612261274274,
      "loss": 0.0686,
      "theoretical_loss": 3.5205405218743007,
      "tokens_seen": 1622147072
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005134809821858449,
      "loss": 0.0702,
      "theoretical_loss": 3.520491514407828,
      "tokens_seen": 1622409216
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005134007382442626,
      "loss": 0.0702,
      "theoretical_loss": 3.520442517075961,
      "tokens_seen": 1622671360
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005133204943026801,
      "loss": 0.068,
      "theoretical_loss": 3.5203935298749656,
      "tokens_seen": 1622933504
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005132402503610978,
      "loss": 0.0689,
      "theoretical_loss": 3.520344552801113,
      "tokens_seen": 1623195648
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005131600064195153,
      "loss": 0.0704,
      "theoretical_loss": 3.5202955858506737,
      "tokens_seen": 1623457792
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000513079762477933,
      "loss": 0.0726,
      "theoretical_loss": 3.520246629019921,
      "tokens_seen": 1623719936
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005129995185363506,
      "loss": 0.0694,
      "theoretical_loss": 3.5201976823051306,
      "tokens_seen": 1623982080
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005129192745947681,
      "loss": 0.0711,
      "theoretical_loss": 3.5201487457025795,
      "tokens_seen": 1624244224
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005128390306531857,
      "loss": 0.0694,
      "theoretical_loss": 3.5200998192085473,
      "tokens_seen": 1624506368
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005127587867116033,
      "loss": 0.069,
      "theoretical_loss": 3.5200509028193148,
      "tokens_seen": 1624768512
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005126785427700209,
      "loss": 0.0723,
      "theoretical_loss": 3.5200019965311657,
      "tokens_seen": 1625030656
    },
    {
      "epoch": 0.49,
      "objective/train/advantage_avg": 0.0010418968740850687,
      "objective/train/docs_used": 593743,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.481257438659668,
      "objective/train/original_loss": 1.481257438659668,
      "objective/train/theoretical_loss": 3.519953100340385,
      "objective/train/tokens_used": 1645752800,
      "objective/train/value_avg": -0.0069732666015625,
      "objective/train/value_loss": 0.00027382862754166126,
      "objective/train/value_max": -6.502866744995117e-05,
      "objective/train/value_min": -0.8955078125,
      "objective/train/value_reward_corr": 0.6842025943888881,
      "objective/train/value_std": 0.01538848876953125,
      "objective/train/weight_avg": 1.0011622905731201,
      "objective/train/weighted_lm_loss": 1.4833272695541382,
      "objective/train/weights_max": 1.6334962844848633,
      "objective/train/weights_min": 0.36918336153030396,
      "theoretical_loss": 3.519953100340385,
      "tokens_seen": 1625292800
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005125982988284384,
      "loss": 0.0712,
      "theoretical_loss": 3.519953100340385,
      "tokens_seen": 1625292800
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005125180548868561,
      "loss": 0.073,
      "theoretical_loss": 3.51990421424326,
      "tokens_seen": 1625554944
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005124378109452736,
      "loss": 0.0703,
      "theoretical_loss": 3.519855338236079,
      "tokens_seen": 1625817088
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005123575670036911,
      "loss": 0.0687,
      "theoretical_loss": 3.5198064723151345,
      "tokens_seen": 1626079232
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005122773230621089,
      "loss": 0.0707,
      "theoretical_loss": 3.519757616476719,
      "tokens_seen": 1626341376
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005121970791205264,
      "loss": 0.0693,
      "theoretical_loss": 3.519708770717126,
      "tokens_seen": 1626603520
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000512116835178944,
      "loss": 0.0718,
      "theoretical_loss": 3.519659935032655,
      "tokens_seen": 1626865664
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005120365912373616,
      "loss": 0.0708,
      "theoretical_loss": 3.5196111094196034,
      "tokens_seen": 1627127808
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005119563472957792,
      "loss": 0.0684,
      "theoretical_loss": 3.5195622938742726,
      "tokens_seen": 1627389952
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005118761033541968,
      "loss": 0.0713,
      "theoretical_loss": 3.519513488392965,
      "tokens_seen": 1627652096
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005117958594126143,
      "loss": 0.0706,
      "theoretical_loss": 3.5194646929719853,
      "tokens_seen": 1627914240
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005117156154710319,
      "loss": 0.0706,
      "theoretical_loss": 3.519415907607641,
      "tokens_seen": 1628176384
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005116353715294495,
      "loss": 0.0682,
      "theoretical_loss": 3.5193671322962397,
      "tokens_seen": 1628438528
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005115551275878672,
      "loss": 0.0704,
      "theoretical_loss": 3.519318367034093,
      "tokens_seen": 1628700672
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005114748836462847,
      "loss": 0.0693,
      "theoretical_loss": 3.519269611817513,
      "tokens_seen": 1628962816
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005113946397047024,
      "loss": 0.0713,
      "theoretical_loss": 3.5192208666428146,
      "tokens_seen": 1629224960
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005113143957631199,
      "loss": 0.0703,
      "theoretical_loss": 3.5191721315063136,
      "tokens_seen": 1629487104
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005112341518215374,
      "loss": 0.0682,
      "theoretical_loss": 3.5191234064043293,
      "tokens_seen": 1629749248
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005111539078799551,
      "loss": 0.0679,
      "theoretical_loss": 3.5190746913331816,
      "tokens_seen": 1630011392
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005110736639383726,
      "loss": 0.0705,
      "theoretical_loss": 3.5190259862891926,
      "tokens_seen": 1630273536
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005109934199967902,
      "loss": 0.0683,
      "theoretical_loss": 3.518977291268686,
      "tokens_seen": 1630535680
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005109131760552078,
      "loss": 0.0708,
      "theoretical_loss": 3.5189286062679894,
      "tokens_seen": 1630797824
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005108329321136255,
      "loss": 0.0697,
      "theoretical_loss": 3.51887993128343,
      "tokens_seen": 1631059968
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005107526881720431,
      "loss": 0.0661,
      "theoretical_loss": 3.518831266311339,
      "tokens_seen": 1631322112
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005106724442304607,
      "loss": 0.0705,
      "theoretical_loss": 3.518782611348046,
      "tokens_seen": 1631584256
    },
    {
      "epoch": 0.49,
      "objective/train/advantage_avg": 0.0007258797995746136,
      "objective/train/docs_used": 596193,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4411818981170654,
      "objective/train/original_loss": 1.4411818981170654,
      "objective/train/theoretical_loss": 3.5187339663898873,
      "objective/train/tokens_used": 1652306400,
      "objective/train/value_avg": -0.005550384521484375,
      "objective/train/value_loss": 0.00012820863048546016,
      "objective/train/value_max": -5.9664249420166016e-05,
      "objective/train/value_min": -0.69775390625,
      "objective/train/value_reward_corr": 0.6792193647270048,
      "objective/train/value_std": 0.01087188720703125,
      "objective/train/weight_avg": 1.0007840394973755,
      "objective/train/weighted_lm_loss": 1.4419759511947632,
      "objective/train/weights_max": 1.2253751754760742,
      "objective/train/weights_min": 0.36889898777008057,
      "theoretical_loss": 3.5187339663898873,
      "tokens_seen": 1631846400
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005105922002888782,
      "loss": 0.0707,
      "theoretical_loss": 3.5187339663898873,
      "tokens_seen": 1631846400
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005105119563472958,
      "loss": 0.071,
      "theoretical_loss": 3.5186853314331974,
      "tokens_seen": 1632108544
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005104317124057134,
      "loss": 0.0703,
      "theoretical_loss": 3.5186367064743145,
      "tokens_seen": 1632370688
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005103514684641309,
      "loss": 0.073,
      "theoretical_loss": 3.518588091509578,
      "tokens_seen": 1632632832
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005102712245225486,
      "loss": 0.0695,
      "theoretical_loss": 3.51853948653533,
      "tokens_seen": 1632894976
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005101909805809661,
      "loss": 0.0688,
      "theoretical_loss": 3.5184908915479145,
      "tokens_seen": 1633157120
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0005101107366393837,
      "loss": 0.0696,
      "theoretical_loss": 3.5184423065436756,
      "tokens_seen": 1633419264
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005100304926978014,
      "loss": 0.0693,
      "theoretical_loss": 3.5183937315189615,
      "tokens_seen": 1633681408
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005099502487562189,
      "loss": 0.068,
      "theoretical_loss": 3.5183451664701217,
      "tokens_seen": 1633943552
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005098700048146365,
      "loss": 0.0693,
      "theoretical_loss": 3.5182966113935072,
      "tokens_seen": 1634205696
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005097897608730541,
      "loss": 0.0696,
      "theoretical_loss": 3.518248066285471,
      "tokens_seen": 1634467840
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005097095169314717,
      "loss": 0.0675,
      "theoretical_loss": 3.518199531142369,
      "tokens_seen": 1634729984
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005096292729898892,
      "loss": 0.0704,
      "theoretical_loss": 3.518151005960557,
      "tokens_seen": 1634992128
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005095490290483069,
      "loss": 0.0699,
      "theoretical_loss": 3.5181024907363945,
      "tokens_seen": 1635254272
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005094687851067244,
      "loss": 0.0691,
      "theoretical_loss": 3.518053985466243,
      "tokens_seen": 1635516416
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.000509388541165142,
      "loss": 0.0662,
      "theoretical_loss": 3.518005490146464,
      "tokens_seen": 1635778560
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005093082972235597,
      "loss": 0.0658,
      "theoretical_loss": 3.5179570047734225,
      "tokens_seen": 1636040704
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005092280532819772,
      "loss": 0.0732,
      "theoretical_loss": 3.5179085293434857,
      "tokens_seen": 1636302848
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005091478093403949,
      "loss": 0.0653,
      "theoretical_loss": 3.517860063853022,
      "tokens_seen": 1636564992
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005090675653988124,
      "loss": 0.0694,
      "theoretical_loss": 3.517811608298401,
      "tokens_seen": 1636827136
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00050898732145723,
      "loss": 0.0706,
      "theoretical_loss": 3.5177631626759958,
      "tokens_seen": 1637089280
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005089070775156476,
      "loss": 0.0665,
      "theoretical_loss": 3.5177147269821805,
      "tokens_seen": 1637351424
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005088268335740651,
      "loss": 0.0678,
      "theoretical_loss": 3.517666301213331,
      "tokens_seen": 1637613568
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005087465896324827,
      "loss": 0.0691,
      "theoretical_loss": 3.5176178853658246,
      "tokens_seen": 1637875712
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005086663456909003,
      "loss": 0.0713,
      "theoretical_loss": 3.5175694794360424,
      "tokens_seen": 1638137856
    },
    {
      "epoch": 0.5,
      "objective/train/advantage_avg": 0.0005885313730686903,
      "objective/train/docs_used": 598618,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3161839246749878,
      "objective/train/original_loss": 1.3161839246749878,
      "objective/train/theoretical_loss": 3.5175210834203656,
      "objective/train/tokens_used": 1658860000,
      "objective/train/value_avg": -0.00928497314453125,
      "objective/train/value_loss": 0.0003415115352254361,
      "objective/train/value_max": -7.140636444091797e-05,
      "objective/train/value_min": -0.94189453125,
      "objective/train/value_reward_corr": 0.7613534650600375,
      "objective/train/value_std": 0.0201263427734375,
      "objective/train/weight_avg": 1.0007390975952148,
      "objective/train/weighted_lm_loss": 1.3172998428344727,
      "objective/train/weights_max": 1.4789870977401733,
      "objective/train/weights_min": 0.3708581030368805,
      "theoretical_loss": 3.5175210834203656,
      "tokens_seen": 1638400000
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.000508586101749318,
      "loss": 0.0704,
      "theoretical_loss": 3.5175210834203656,
      "tokens_seen": 1638400000
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005085058578077355,
      "loss": 0.0736,
      "theoretical_loss": 3.5174726973151778,
      "tokens_seen": 1638662144
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005084256138661532,
      "loss": 0.0679,
      "theoretical_loss": 3.517424321116865,
      "tokens_seen": 1638924288
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005083453699245707,
      "loss": 0.071,
      "theoretical_loss": 3.517375954821815,
      "tokens_seen": 1639186432
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005082651259829882,
      "loss": 0.0691,
      "theoretical_loss": 3.517327598426416,
      "tokens_seen": 1639448576
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005081848820414059,
      "loss": 0.0685,
      "theoretical_loss": 3.5172792519270604,
      "tokens_seen": 1639710720
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005081046380998234,
      "loss": 0.0716,
      "theoretical_loss": 3.517230915320141,
      "tokens_seen": 1639972864
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005080243941582411,
      "loss": 0.07,
      "theoretical_loss": 3.5171825886020525,
      "tokens_seen": 1640235008
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005079441502166586,
      "loss": 0.0673,
      "theoretical_loss": 3.5171342717691925,
      "tokens_seen": 1640497152
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005078639062750763,
      "loss": 0.0675,
      "theoretical_loss": 3.517085964817959,
      "tokens_seen": 1640759296
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005077836623334939,
      "loss": 0.0694,
      "theoretical_loss": 3.517037667744754,
      "tokens_seen": 1641021440
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005077034183919115,
      "loss": 0.0687,
      "theoretical_loss": 3.516989380545979,
      "tokens_seen": 1641283584
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.000507623174450329,
      "loss": 0.0708,
      "theoretical_loss": 3.5169411032180387,
      "tokens_seen": 1641545728
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005075429305087466,
      "loss": 0.0713,
      "theoretical_loss": 3.51689283575734,
      "tokens_seen": 1641807872
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005074626865671642,
      "loss": 0.0724,
      "theoretical_loss": 3.516844578160291,
      "tokens_seen": 1642070016
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005073824426255817,
      "loss": 0.0684,
      "theoretical_loss": 3.5167963304233014,
      "tokens_seen": 1642332160
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005073021986839994,
      "loss": 0.0702,
      "theoretical_loss": 3.516748092542784,
      "tokens_seen": 1642594304
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005072219547424169,
      "loss": 0.0718,
      "theoretical_loss": 3.5166998645151515,
      "tokens_seen": 1642856448
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005071417108008345,
      "loss": 0.0721,
      "theoretical_loss": 3.516651646336821,
      "tokens_seen": 1643118592
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005070614668592522,
      "loss": 0.0703,
      "theoretical_loss": 3.5166034380042093,
      "tokens_seen": 1643380736
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005069812229176697,
      "loss": 0.0701,
      "theoretical_loss": 3.5165552395137363,
      "tokens_seen": 1643642880
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005069009789760874,
      "loss": 0.0681,
      "theoretical_loss": 3.516507050861823,
      "tokens_seen": 1643905024
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005068207350345049,
      "loss": 0.0671,
      "theoretical_loss": 3.5164588720448937,
      "tokens_seen": 1644167168
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005067404910929225,
      "loss": 0.0699,
      "theoretical_loss": 3.5164107030593725,
      "tokens_seen": 1644429312
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005066602471513401,
      "loss": 0.0707,
      "theoretical_loss": 3.516362543901687,
      "tokens_seen": 1644691456
    },
    {
      "epoch": 0.5,
      "objective/train/advantage_avg": 0.001454401994124055,
      "objective/train/docs_used": 601010,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3577359914779663,
      "objective/train/original_loss": 1.3577358722686768,
      "objective/train/theoretical_loss": 3.5163143945682656,
      "objective/train/tokens_used": 1665413600,
      "objective/train/value_avg": -0.007610321044921875,
      "objective/train/value_loss": 0.00015290330338757485,
      "objective/train/value_max": -6.657838821411133e-05,
      "objective/train/value_min": -0.483642578125,
      "objective/train/value_reward_corr": 0.7295783796450883,
      "objective/train/value_std": 0.01434326171875,
      "objective/train/weight_avg": 1.0015279054641724,
      "objective/train/weighted_lm_loss": 1.3596255779266357,
      "objective/train/weights_max": 1.3309441804885864,
      "objective/train/weights_min": 0.369488388299942,
      "theoretical_loss": 3.5163143945682656,
      "tokens_seen": 1644953600
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005065800032097577,
      "loss": 0.0724,
      "theoretical_loss": 3.5163143945682656,
      "tokens_seen": 1644953600
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005064997592681752,
      "loss": 0.0716,
      "theoretical_loss": 3.5162662550555392,
      "tokens_seen": 1645215744
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005064195153265928,
      "loss": 0.0721,
      "theoretical_loss": 3.5162181253599405,
      "tokens_seen": 1645477888
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005063392713850105,
      "loss": 0.0704,
      "theoretical_loss": 3.516170005477904,
      "tokens_seen": 1645740032
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.000506259027443428,
      "loss": 0.0698,
      "theoretical_loss": 3.516121895405866,
      "tokens_seen": 1646002176
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005061787835018457,
      "loss": 0.0733,
      "theoretical_loss": 3.516073795140265,
      "tokens_seen": 1646264320
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005060985395602632,
      "loss": 0.0693,
      "theoretical_loss": 3.5160257046775407,
      "tokens_seen": 1646526464
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005060182956186808,
      "loss": 0.0695,
      "theoretical_loss": 3.515977624014135,
      "tokens_seen": 1646788608
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005059380516770984,
      "loss": 0.0682,
      "theoretical_loss": 3.515929553146492,
      "tokens_seen": 1647050752
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005058578077355159,
      "loss": 0.0681,
      "theoretical_loss": 3.515881492071057,
      "tokens_seen": 1647312896
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005057775637939335,
      "loss": 0.0705,
      "theoretical_loss": 3.515833440784278,
      "tokens_seen": 1647575040
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005056973198523511,
      "loss": 0.0688,
      "theoretical_loss": 3.5157853992826036,
      "tokens_seen": 1647837184
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005056170759107688,
      "loss": 0.0697,
      "theoretical_loss": 3.5157373675624854,
      "tokens_seen": 1648099328
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005055368319691864,
      "loss": 0.0712,
      "theoretical_loss": 3.515689345620377,
      "tokens_seen": 1648361472
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.000505456588027604,
      "loss": 0.071,
      "theoretical_loss": 3.5156413334527317,
      "tokens_seen": 1648623616
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005053763440860215,
      "loss": 0.0698,
      "theoretical_loss": 3.515593331056008,
      "tokens_seen": 1648885760
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005052961001444391,
      "loss": 0.068,
      "theoretical_loss": 3.5155453384266635,
      "tokens_seen": 1649147904
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005052158562028567,
      "loss": 0.068,
      "theoretical_loss": 3.515497355561159,
      "tokens_seen": 1649410048
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005051356122612742,
      "loss": 0.0712,
      "theoretical_loss": 3.515449382455957,
      "tokens_seen": 1649672192
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005050553683196919,
      "loss": 0.0699,
      "theoretical_loss": 3.515401419107521,
      "tokens_seen": 1649934336
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005049751243781094,
      "loss": 0.0683,
      "theoretical_loss": 3.515353465512317,
      "tokens_seen": 1650196480
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.000504894880436527,
      "loss": 0.0721,
      "theoretical_loss": 3.5153055216668134,
      "tokens_seen": 1650458624
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005048146364949447,
      "loss": 0.0704,
      "theoretical_loss": 3.51525758756748,
      "tokens_seen": 1650720768
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005047343925533623,
      "loss": 0.0692,
      "theoretical_loss": 3.5152096632107876,
      "tokens_seen": 1650982912
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005046541486117798,
      "loss": 0.0704,
      "theoretical_loss": 3.5151617485932096,
      "tokens_seen": 1651245056
    },
    {
      "epoch": 0.5,
      "objective/train/advantage_avg": 0.0005821323138661683,
      "objective/train/docs_used": 603335,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3430393934249878,
      "objective/train/original_loss": 1.3430393934249878,
      "objective/train/theoretical_loss": 3.5151138437112213,
      "objective/train/tokens_used": 1671967200,
      "objective/train/value_avg": -0.00864410400390625,
      "objective/train/value_loss": 0.00044420172343961895,
      "objective/train/value_max": -5.1856040954589844e-05,
      "objective/train/value_min": -0.92138671875,
      "objective/train/value_reward_corr": 0.6653913925705341,
      "objective/train/value_std": 0.01959228515625,
      "objective/train/weight_avg": 1.0007838010787964,
      "objective/train/weighted_lm_loss": 1.3428455591201782,
      "objective/train/weights_max": 2.4011125564575195,
      "objective/train/weights_min": 0.3695729672908783,
      "theoretical_loss": 3.5151138437112213,
      "tokens_seen": 1651507200
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005045739046701974,
      "loss": 0.0665,
      "theoretical_loss": 3.5151138437112213,
      "tokens_seen": 1651507200
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.000504493660728615,
      "loss": 0.0712,
      "theoretical_loss": 3.5150659485613005,
      "tokens_seen": 1651769344
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005044134167870326,
      "loss": 0.0693,
      "theoretical_loss": 3.5150180631399244,
      "tokens_seen": 1652031488
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005043331728454502,
      "loss": 0.0689,
      "theoretical_loss": 3.5149701874435753,
      "tokens_seen": 1652293632
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005042529289038677,
      "loss": 0.0741,
      "theoretical_loss": 3.514922321468734,
      "tokens_seen": 1652555776
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005041726849622855,
      "loss": 0.071,
      "theoretical_loss": 3.5148744652118866,
      "tokens_seen": 1652817920
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.000504092441020703,
      "loss": 0.071,
      "theoretical_loss": 3.5148266186695185,
      "tokens_seen": 1653080064
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005040121970791205,
      "loss": 0.0687,
      "theoretical_loss": 3.5147787818381175,
      "tokens_seen": 1653342208
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005039319531375382,
      "loss": 0.0696,
      "theoretical_loss": 3.514730954714173,
      "tokens_seen": 1653604352
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005038517091959557,
      "loss": 0.0693,
      "theoretical_loss": 3.5146831372941776,
      "tokens_seen": 1653866496
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005037714652543733,
      "loss": 0.0705,
      "theoretical_loss": 3.5146353295746247,
      "tokens_seen": 1654128640
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005036912213127909,
      "loss": 0.0686,
      "theoretical_loss": 3.514587531552009,
      "tokens_seen": 1654390784
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005036109773712085,
      "loss": 0.0681,
      "theoretical_loss": 3.5145397432228274,
      "tokens_seen": 1654652928
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.000503530733429626,
      "loss": 0.0723,
      "theoretical_loss": 3.5144919645835797,
      "tokens_seen": 1654915072
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005034504894880436,
      "loss": 0.0681,
      "theoretical_loss": 3.514444195630766,
      "tokens_seen": 1655177216
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005033702455464613,
      "loss": 0.0683,
      "theoretical_loss": 3.5143964363608893,
      "tokens_seen": 1655439360
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005032900016048788,
      "loss": 0.0718,
      "theoretical_loss": 3.514348686770454,
      "tokens_seen": 1655701504
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005032097576632965,
      "loss": 0.0683,
      "theoretical_loss": 3.5143009468559656,
      "tokens_seen": 1655963648
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.000503129513721714,
      "loss": 0.0684,
      "theoretical_loss": 3.514253216613932,
      "tokens_seen": 1656225792
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005030492697801317,
      "loss": 0.0711,
      "theoretical_loss": 3.514205496040865,
      "tokens_seen": 1656487936
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005029690258385492,
      "loss": 0.0692,
      "theoretical_loss": 3.5141577851332735,
      "tokens_seen": 1656750080
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005028887818969667,
      "loss": 0.0685,
      "theoretical_loss": 3.5141100838876724,
      "tokens_seen": 1657012224
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005028085379553844,
      "loss": 0.0676,
      "theoretical_loss": 3.5140623923005774,
      "tokens_seen": 1657274368
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005027282940138019,
      "loss": 0.0675,
      "theoretical_loss": 3.514014710368505,
      "tokens_seen": 1657536512
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005026480500722196,
      "loss": 0.0673,
      "theoretical_loss": 3.513967038087973,
      "tokens_seen": 1657798656
    },
    {
      "epoch": 0.5,
      "objective/train/advantage_avg": 0.00016826541104819626,
      "objective/train/docs_used": 605625,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3199695348739624,
      "objective/train/original_loss": 1.3199697732925415,
      "objective/train/theoretical_loss": 3.5139193754555036,
      "objective/train/tokens_used": 1678520800,
      "objective/train/value_avg": -0.00653076171875,
      "objective/train/value_loss": 0.0002841904351953417,
      "objective/train/value_max": -0.00011771917343139648,
      "objective/train/value_min": -0.70263671875,
      "objective/train/value_reward_corr": 0.6535262240804112,
      "objective/train/value_std": 0.01262664794921875,
      "objective/train/weight_avg": 1.0002937316894531,
      "objective/train/weighted_lm_loss": 1.3204420804977417,
      "objective/train/weights_max": 1.227171540260315,
      "objective/train/weights_min": 0.3689693808555603,
      "theoretical_loss": 3.5139193754555036,
      "tokens_seen": 1658060800
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005025678061306372,
      "loss": 0.0688,
      "theoretical_loss": 3.5139193754555036,
      "tokens_seen": 1658060800
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005024875621890548,
      "loss": 0.0673,
      "theoretical_loss": 3.513871722467619,
      "tokens_seen": 1658322944
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005024073182474723,
      "loss": 0.0701,
      "theoretical_loss": 3.513824079120843,
      "tokens_seen": 1658585088
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005023270743058899,
      "loss": 0.0696,
      "theoretical_loss": 3.513776445411702,
      "tokens_seen": 1658847232
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005022468303643075,
      "loss": 0.067,
      "theoretical_loss": 3.5137288213367235,
      "tokens_seen": 1659109376
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.000502166586422725,
      "loss": 0.0698,
      "theoretical_loss": 3.5136812068924375,
      "tokens_seen": 1659371520
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005020863424811427,
      "loss": 0.069,
      "theoretical_loss": 3.513633602075376,
      "tokens_seen": 1659633664
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005020060985395602,
      "loss": 0.0685,
      "theoretical_loss": 3.513586006882071,
      "tokens_seen": 1659895808
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.000501925854597978,
      "loss": 0.0691,
      "theoretical_loss": 3.513538421309059,
      "tokens_seen": 1660157952
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005018456106563955,
      "loss": 0.0697,
      "theoretical_loss": 3.5134908453528757,
      "tokens_seen": 1660420096
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.000501765366714813,
      "loss": 0.0685,
      "theoretical_loss": 3.51344327901006,
      "tokens_seen": 1660682240
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005016851227732307,
      "loss": 0.0699,
      "theoretical_loss": 3.513395722277153,
      "tokens_seen": 1660944384
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005016048788316482,
      "loss": 0.0706,
      "theoretical_loss": 3.513348175150696,
      "tokens_seen": 1661206528
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005015246348900658,
      "loss": 0.0658,
      "theoretical_loss": 3.5133006376272338,
      "tokens_seen": 1661468672
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005014443909484834,
      "loss": 0.0685,
      "theoretical_loss": 3.5132531097033115,
      "tokens_seen": 1661730816
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.000501364147006901,
      "loss": 0.0655,
      "theoretical_loss": 3.5132055913754776,
      "tokens_seen": 1661992960
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005012839030653185,
      "loss": 0.0694,
      "theoretical_loss": 3.5131580826402806,
      "tokens_seen": 1662255104
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005012036591237363,
      "loss": 0.0703,
      "theoretical_loss": 3.5131105834942726,
      "tokens_seen": 1662517248
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005011234151821538,
      "loss": 0.0698,
      "theoretical_loss": 3.5130630939340053,
      "tokens_seen": 1662779392
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005010431712405713,
      "loss": 0.069,
      "theoretical_loss": 3.5130156139560347,
      "tokens_seen": 1663041536
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.000500962927298989,
      "loss": 0.0703,
      "theoretical_loss": 3.512968143556917,
      "tokens_seen": 1663303680
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005008826833574065,
      "loss": 0.0693,
      "theoretical_loss": 3.51292068273321,
      "tokens_seen": 1663565824
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005008024394158241,
      "loss": 0.0682,
      "theoretical_loss": 3.512873231481474,
      "tokens_seen": 1663827968
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005007221954742417,
      "loss": 0.0681,
      "theoretical_loss": 3.512825789798271,
      "tokens_seen": 1664090112
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005006419515326593,
      "loss": 0.0683,
      "theoretical_loss": 3.5127783576801646,
      "tokens_seen": 1664352256
    },
    {
      "epoch": 0.5,
      "objective/train/advantage_avg": 0.00016097695333883166,
      "objective/train/docs_used": 608106,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3378410339355469,
      "objective/train/original_loss": 1.3378409147262573,
      "objective/train/theoretical_loss": 3.51273093512372,
      "objective/train/tokens_used": 1685074400,
      "objective/train/value_avg": -0.00897979736328125,
      "objective/train/value_loss": 0.00020667578792199492,
      "objective/train/value_max": -4.6133995056152344e-05,
      "objective/train/value_min": -0.70166015625,
      "objective/train/value_reward_corr": 0.7562442994128528,
      "objective/train/value_std": 0.0157470703125,
      "objective/train/weight_avg": 1.000257968902588,
      "objective/train/weighted_lm_loss": 1.3374638557434082,
      "objective/train/weights_max": 1.2285205125808716,
      "objective/train/weights_min": 0.39345037937164307,
      "theoretical_loss": 3.51273093512372,
      "tokens_seen": 1664614400
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005005617075910769,
      "loss": 0.0709,
      "theoretical_loss": 3.51273093512372,
      "tokens_seen": 1664614400
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005004814636494944,
      "loss": 0.0681,
      "theoretical_loss": 3.512683522125505,
      "tokens_seen": 1664876544
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005004012197079121,
      "loss": 0.0683,
      "theoretical_loss": 3.512636118682088,
      "tokens_seen": 1665138688
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005003209757663297,
      "loss": 0.0687,
      "theoretical_loss": 3.5125887247900396,
      "tokens_seen": 1665400832
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005002407318247473,
      "loss": 0.0711,
      "theoretical_loss": 3.512541340445933,
      "tokens_seen": 1665662976
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005001604878831648,
      "loss": 0.0695,
      "theoretical_loss": 3.5124939656463416,
      "tokens_seen": 1665925120
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0005000802439415825,
      "loss": 0.0678,
      "theoretical_loss": 3.5124466003878423,
      "tokens_seen": 1666187264
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0005,
      "loss": 0.0694,
      "theoretical_loss": 3.512399244667012,
      "tokens_seen": 1666449408
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004999197560584176,
      "loss": 0.0711,
      "theoretical_loss": 3.5123518984804303,
      "tokens_seen": 1666711552
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004998395121168352,
      "loss": 0.0727,
      "theoretical_loss": 3.5123045618246795,
      "tokens_seen": 1666973696
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004997592681752528,
      "loss": 0.0685,
      "theoretical_loss": 3.5122572346963423,
      "tokens_seen": 1667235840
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004996790242336704,
      "loss": 0.0683,
      "theoretical_loss": 3.512209917092003,
      "tokens_seen": 1667497984
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.000499598780292088,
      "loss": 0.0699,
      "theoretical_loss": 3.5121626090082487,
      "tokens_seen": 1667760128
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004995185363505056,
      "loss": 0.0694,
      "theoretical_loss": 3.5121153104416676,
      "tokens_seen": 1668022272
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004994382924089231,
      "loss": 0.0653,
      "theoretical_loss": 3.5120680213888504,
      "tokens_seen": 1668284416
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004993580484673407,
      "loss": 0.0686,
      "theoretical_loss": 3.512020741846388,
      "tokens_seen": 1668546560
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004992778045257583,
      "loss": 0.0676,
      "theoretical_loss": 3.511973471810875,
      "tokens_seen": 1668808704
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004991975605841759,
      "loss": 0.069,
      "theoretical_loss": 3.5119262112789063,
      "tokens_seen": 1669070848
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004991173166425935,
      "loss": 0.0679,
      "theoretical_loss": 3.5118789602470786,
      "tokens_seen": 1669332992
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.000499037072701011,
      "loss": 0.0727,
      "theoretical_loss": 3.5118317187119916,
      "tokens_seen": 1669595136
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004989568287594286,
      "loss": 0.07,
      "theoretical_loss": 3.511784486670246,
      "tokens_seen": 1669857280
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004988765848178463,
      "loss": 0.0673,
      "theoretical_loss": 3.5117372641184432,
      "tokens_seen": 1670119424
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004987963408762639,
      "loss": 0.0707,
      "theoretical_loss": 3.5116900510531885,
      "tokens_seen": 1670381568
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004987160969346815,
      "loss": 0.069,
      "theoretical_loss": 3.5116428474710872,
      "tokens_seen": 1670643712
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004986358529930991,
      "loss": 0.0706,
      "theoretical_loss": 3.5115956533687473,
      "tokens_seen": 1670905856
    },
    {
      "epoch": 0.51,
      "objective/train/advantage_avg": 0.0005539392586797476,
      "objective/train/docs_used": 610761,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.340118169784546,
      "objective/train/original_loss": 1.3401179313659668,
      "objective/train/theoretical_loss": 3.5115484687427783,
      "objective/train/tokens_used": 1691628000,
      "objective/train/value_avg": -0.0100250244140625,
      "objective/train/value_loss": 0.00027908929041586816,
      "objective/train/value_max": -8.028745651245117e-05,
      "objective/train/value_min": -0.5888671875,
      "objective/train/value_reward_corr": 0.7176989029137323,
      "objective/train/value_std": 0.01812744140625,
      "objective/train/weight_avg": 1.0006839036941528,
      "objective/train/weighted_lm_loss": 1.3407360315322876,
      "objective/train/weights_max": 1.7151225805282593,
      "objective/train/weights_min": 0.3900670111179352,
      "theoretical_loss": 3.5115484687427783,
      "tokens_seen": 1671168000
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004985556090515166,
      "loss": 0.0734,
      "theoretical_loss": 3.5115484687427783,
      "tokens_seen": 1671168000
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004984753651099342,
      "loss": 0.0652,
      "theoretical_loss": 3.5115012935897907,
      "tokens_seen": 1671430144
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004983951211683518,
      "loss": 0.0676,
      "theoretical_loss": 3.5114541279063975,
      "tokens_seen": 1671692288
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004983148772267693,
      "loss": 0.0692,
      "theoretical_loss": 3.511406971689214,
      "tokens_seen": 1671954432
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004982346332851869,
      "loss": 0.0673,
      "theoretical_loss": 3.511359824934856,
      "tokens_seen": 1672216576
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004981543893436046,
      "loss": 0.0666,
      "theoretical_loss": 3.5113126876399416,
      "tokens_seen": 1672478720
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004980741454020222,
      "loss": 0.0677,
      "theoretical_loss": 3.511265559801091,
      "tokens_seen": 1672740864
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004979939014604398,
      "loss": 0.069,
      "theoretical_loss": 3.5112184414149255,
      "tokens_seen": 1673003008
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004979136575188573,
      "loss": 0.0681,
      "theoretical_loss": 3.5111713324780687,
      "tokens_seen": 1673265152
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004978334135772749,
      "loss": 0.0702,
      "theoretical_loss": 3.5111242329871457,
      "tokens_seen": 1673527296
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004977531696356925,
      "loss": 0.0689,
      "theoretical_loss": 3.5110771429387824,
      "tokens_seen": 1673789440
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004976729256941101,
      "loss": 0.0688,
      "theoretical_loss": 3.511030062329608,
      "tokens_seen": 1674051584
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004975926817525277,
      "loss": 0.0691,
      "theoretical_loss": 3.5109829911562533,
      "tokens_seen": 1674313728
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004975124378109452,
      "loss": 0.0661,
      "theoretical_loss": 3.5109359294153495,
      "tokens_seen": 1674575872
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004974321938693629,
      "loss": 0.0678,
      "theoretical_loss": 3.5108888771035307,
      "tokens_seen": 1674838016
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004973519499277805,
      "loss": 0.0681,
      "theoretical_loss": 3.5108418342174317,
      "tokens_seen": 1675100160
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004972717059861981,
      "loss": 0.0678,
      "theoretical_loss": 3.5107948007536907,
      "tokens_seen": 1675362304
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004971914620446156,
      "loss": 0.0673,
      "theoretical_loss": 3.5107477767089454,
      "tokens_seen": 1675624448
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004971112181030332,
      "loss": 0.0663,
      "theoretical_loss": 3.5107007620798374,
      "tokens_seen": 1675886592
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004970309741614508,
      "loss": 0.0682,
      "theoretical_loss": 3.510653756863009,
      "tokens_seen": 1676148736
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004969507302198684,
      "loss": 0.0676,
      "theoretical_loss": 3.5106067610551035,
      "tokens_seen": 1676410880
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.000496870486278286,
      "loss": 0.071,
      "theoretical_loss": 3.5105597746527675,
      "tokens_seen": 1676673024
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004967902423367036,
      "loss": 0.0657,
      "theoretical_loss": 3.5105127976526482,
      "tokens_seen": 1676935168
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004967099983951212,
      "loss": 0.0679,
      "theoretical_loss": 3.5104658300513942,
      "tokens_seen": 1677197312
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004966297544535388,
      "loss": 0.0655,
      "theoretical_loss": 3.5104188718456575,
      "tokens_seen": 1677459456
    },
    {
      "epoch": 0.51,
      "objective/train/advantage_avg": 0.00019064874504692852,
      "objective/train/docs_used": 612648,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3213294744491577,
      "objective/train/original_loss": 1.3213295936584473,
      "objective/train/theoretical_loss": 3.5103719230320904,
      "objective/train/tokens_used": 1698181600,
      "objective/train/value_avg": -0.0089111328125,
      "objective/train/value_loss": 0.000392118381569162,
      "objective/train/value_max": -7.367134094238281e-05,
      "objective/train/value_min": -0.72900390625,
      "objective/train/value_reward_corr": 0.6955816812324872,
      "objective/train/value_std": 0.0175933837890625,
      "objective/train/weight_avg": 1.0003646612167358,
      "objective/train/weighted_lm_loss": 1.3213082551956177,
      "objective/train/weights_max": 1.415054440498352,
      "objective/train/weights_min": 0.3721874952316284,
      "theoretical_loss": 3.5103719230320904,
      "tokens_seen": 1677721600
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004965495105119564,
      "loss": 0.0681,
      "theoretical_loss": 3.5103719230320904,
      "tokens_seen": 1677721600
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.000496469266570374,
      "loss": 0.069,
      "theoretical_loss": 3.510324983607347,
      "tokens_seen": 1677983744
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004963890226287915,
      "loss": 0.0682,
      "theoretical_loss": 3.5102780535680838,
      "tokens_seen": 1678245888
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004963087786872091,
      "loss": 0.0701,
      "theoretical_loss": 3.510231132910958,
      "tokens_seen": 1678508032
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004962285347456267,
      "loss": 0.0679,
      "theoretical_loss": 3.5101842216326293,
      "tokens_seen": 1678770176
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004961482908040443,
      "loss": 0.0674,
      "theoretical_loss": 3.5101373197297594,
      "tokens_seen": 1679032320
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004960680468624618,
      "loss": 0.0672,
      "theoretical_loss": 3.510090427199011,
      "tokens_seen": 1679294464
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004959878029208794,
      "loss": 0.0689,
      "theoretical_loss": 3.5100435440370483,
      "tokens_seen": 1679556608
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004959075589792971,
      "loss": 0.0684,
      "theoretical_loss": 3.5099966702405383,
      "tokens_seen": 1679818752
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004958273150377147,
      "loss": 0.0712,
      "theoretical_loss": 3.5099498058061487,
      "tokens_seen": 1680080896
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004957470710961323,
      "loss": 0.068,
      "theoretical_loss": 3.5099029507305497,
      "tokens_seen": 1680343040
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004956668271545499,
      "loss": 0.0703,
      "theoretical_loss": 3.509856105010412,
      "tokens_seen": 1680605184
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004955865832129674,
      "loss": 0.073,
      "theoretical_loss": 3.509809268642409,
      "tokens_seen": 1680867328
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.000495506339271385,
      "loss": 0.071,
      "theoretical_loss": 3.5097624416232156,
      "tokens_seen": 1681129472
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004954260953298026,
      "loss": 0.0702,
      "theoretical_loss": 3.509715623949509,
      "tokens_seen": 1681391616
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004953458513882202,
      "loss": 0.0712,
      "theoretical_loss": 3.509668815617967,
      "tokens_seen": 1681653760
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004952656074466377,
      "loss": 0.0692,
      "theoretical_loss": 3.5096220166252694,
      "tokens_seen": 1681915904
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004951853635050554,
      "loss": 0.0683,
      "theoretical_loss": 3.509575226968098,
      "tokens_seen": 1682178048
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.000495105119563473,
      "loss": 0.0691,
      "theoretical_loss": 3.5095284466431362,
      "tokens_seen": 1682440192
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004950248756218906,
      "loss": 0.068,
      "theoretical_loss": 3.509481675647069,
      "tokens_seen": 1682702336
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004949446316803081,
      "loss": 0.0684,
      "theoretical_loss": 3.509434913976583,
      "tokens_seen": 1682964480
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004948643877387257,
      "loss": 0.0694,
      "theoretical_loss": 3.509388161628367,
      "tokens_seen": 1683226624
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004947841437971433,
      "loss": 0.0698,
      "theoretical_loss": 3.509341418599111,
      "tokens_seen": 1683488768
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004947038998555609,
      "loss": 0.0695,
      "theoretical_loss": 3.509294684885506,
      "tokens_seen": 1683750912
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004946236559139785,
      "loss": 0.0712,
      "theoretical_loss": 3.509247960484247,
      "tokens_seen": 1684013056
    },
    {
      "epoch": 0.51,
      "objective/train/advantage_avg": 0.000966555206105113,
      "objective/train/docs_used": 615159,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3198797702789307,
      "objective/train/original_loss": 1.3198797702789307,
      "objective/train/theoretical_loss": 3.5092012453920285,
      "objective/train/tokens_used": 1704735200,
      "objective/train/value_avg": -0.00679779052734375,
      "objective/train/value_loss": 0.00012040092406095937,
      "objective/train/value_max": -6.300210952758789e-05,
      "objective/train/value_min": -0.474609375,
      "objective/train/value_reward_corr": 0.700724774795992,
      "objective/train/value_std": 0.0108489990234375,
      "objective/train/weight_avg": 1.0010249614715576,
      "objective/train/weighted_lm_loss": 1.3210926055908203,
      "objective/train/weights_max": 1.2091786861419678,
      "objective/train/weights_min": 0.6098971962928772,
      "theoretical_loss": 3.5092012453920285,
      "tokens_seen": 1684275200
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004945434119723962,
      "loss": 0.0686,
      "theoretical_loss": 3.5092012453920285,
      "tokens_seen": 1684275200
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004944631680308137,
      "loss": 0.0685,
      "theoretical_loss": 3.5091545396055475,
      "tokens_seen": 1684537344
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004943829240892313,
      "loss": 0.0674,
      "theoretical_loss": 3.5091078431215017,
      "tokens_seen": 1684799488
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004943026801476489,
      "loss": 0.0674,
      "theoretical_loss": 3.5090611559365925,
      "tokens_seen": 1685061632
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004942224362060665,
      "loss": 0.0685,
      "theoretical_loss": 3.509014478047522,
      "tokens_seen": 1685323776
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.000494142192264484,
      "loss": 0.0709,
      "theoretical_loss": 3.5089678094509926,
      "tokens_seen": 1685585920
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004940619483229016,
      "loss": 0.0691,
      "theoretical_loss": 3.5089211501437103,
      "tokens_seen": 1685848064
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004939817043813192,
      "loss": 0.0684,
      "theoretical_loss": 3.5088745001223822,
      "tokens_seen": 1686110208
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004939014604397368,
      "loss": 0.0698,
      "theoretical_loss": 3.508827859383717,
      "tokens_seen": 1686372352
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004938212164981544,
      "loss": 0.0674,
      "theoretical_loss": 3.5087812279244246,
      "tokens_seen": 1686634496
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.000493740972556572,
      "loss": 0.0691,
      "theoretical_loss": 3.5087346057412176,
      "tokens_seen": 1686896640
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004936607286149896,
      "loss": 0.0734,
      "theoretical_loss": 3.508687992830809,
      "tokens_seen": 1687158784
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004935804846734072,
      "loss": 0.0731,
      "theoretical_loss": 3.5086413891899144,
      "tokens_seen": 1687420928
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004935002407318248,
      "loss": 0.0714,
      "theoretical_loss": 3.5085947948152514,
      "tokens_seen": 1687683072
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004934199967902424,
      "loss": 0.0688,
      "theoretical_loss": 3.5085482097035383,
      "tokens_seen": 1687945216
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004933397528486599,
      "loss": 0.0693,
      "theoretical_loss": 3.508501633851495,
      "tokens_seen": 1688207360
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004932595089070775,
      "loss": 0.0672,
      "theoretical_loss": 3.5084550672558446,
      "tokens_seen": 1688469504
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004931792649654951,
      "loss": 0.0694,
      "theoretical_loss": 3.50840850991331,
      "tokens_seen": 1688731648
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004930990210239126,
      "loss": 0.0655,
      "theoretical_loss": 3.5083619618206168,
      "tokens_seen": 1688993792
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004930187770823302,
      "loss": 0.0664,
      "theoretical_loss": 3.5083154229744924,
      "tokens_seen": 1689255936
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004929385331407479,
      "loss": 0.0735,
      "theoretical_loss": 3.5082688933716653,
      "tokens_seen": 1689518080
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004928582891991655,
      "loss": 0.0701,
      "theoretical_loss": 3.5082223730088655,
      "tokens_seen": 1689780224
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004927780452575831,
      "loss": 0.0695,
      "theoretical_loss": 3.5081758618828256,
      "tokens_seen": 1690042368
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004926978013160007,
      "loss": 0.0682,
      "theoretical_loss": 3.5081293599902788,
      "tokens_seen": 1690304512
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004926175573744183,
      "loss": 0.0693,
      "theoretical_loss": 3.5080828673279614,
      "tokens_seen": 1690566656
    },
    {
      "epoch": 0.51,
      "objective/train/advantage_avg": 0.000561828725039959,
      "objective/train/docs_used": 617433,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3782442808151245,
      "objective/train/original_loss": 1.3782442808151245,
      "objective/train/theoretical_loss": 3.5080363838926094,
      "objective/train/tokens_used": 1711288800,
      "objective/train/value_avg": -0.006938934326171875,
      "objective/train/value_loss": 0.00027447956381365657,
      "objective/train/value_max": -5.519390106201172e-05,
      "objective/train/value_min": -0.72607421875,
      "objective/train/value_reward_corr": 0.6786282922023184,
      "objective/train/value_std": 0.016754150390625,
      "objective/train/weight_avg": 1.0006831884384155,
      "objective/train/weighted_lm_loss": 1.3784576654434204,
      "objective/train/weights_max": 1.5841267108917236,
      "objective/train/weights_min": 0.3688187897205353,
      "theoretical_loss": 3.5080363838926094,
      "tokens_seen": 1690828800
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004925373134328358,
      "loss": 0.0686,
      "theoretical_loss": 3.5080363838926094,
      "tokens_seen": 1690828800
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004924570694912534,
      "loss": 0.0688,
      "theoretical_loss": 3.5079899096809624,
      "tokens_seen": 1691090944
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.000492376825549671,
      "loss": 0.0691,
      "theoretical_loss": 3.5079434446897597,
      "tokens_seen": 1691353088
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004922965816080887,
      "loss": 0.0695,
      "theoretical_loss": 3.5078969889157445,
      "tokens_seen": 1691615232
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004922163376665062,
      "loss": 0.0685,
      "theoretical_loss": 3.5078505423556594,
      "tokens_seen": 1691877376
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004921360937249238,
      "loss": 0.0708,
      "theoretical_loss": 3.50780410500625,
      "tokens_seen": 1692139520
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004920558497833414,
      "loss": 0.0707,
      "theoretical_loss": 3.507757676864264,
      "tokens_seen": 1692401664
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004919756058417589,
      "loss": 0.067,
      "theoretical_loss": 3.5077112579264496,
      "tokens_seen": 1692663808
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004918953619001765,
      "loss": 0.0693,
      "theoretical_loss": 3.507664848189557,
      "tokens_seen": 1692925952
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004918151179585941,
      "loss": 0.0715,
      "theoretical_loss": 3.507618447650337,
      "tokens_seen": 1693188096
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004917348740170117,
      "loss": 0.0696,
      "theoretical_loss": 3.5075720563055457,
      "tokens_seen": 1693450240
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004916546300754293,
      "loss": 0.0667,
      "theoretical_loss": 3.5075256741519363,
      "tokens_seen": 1693712384
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.000491574386133847,
      "loss": 0.0667,
      "theoretical_loss": 3.507479301186266,
      "tokens_seen": 1693974528
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004914941421922646,
      "loss": 0.0708,
      "theoretical_loss": 3.5074329374052944,
      "tokens_seen": 1694236672
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004914138982506821,
      "loss": 0.0662,
      "theoretical_loss": 3.507386582805781,
      "tokens_seen": 1694498816
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004913336543090997,
      "loss": 0.0681,
      "theoretical_loss": 3.5073402373844864,
      "tokens_seen": 1694760960
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004912534103675173,
      "loss": 0.0695,
      "theoretical_loss": 3.5072939011381763,
      "tokens_seen": 1695023104
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004911731664259348,
      "loss": 0.0672,
      "theoretical_loss": 3.507247574063614,
      "tokens_seen": 1695285248
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004910929224843524,
      "loss": 0.0664,
      "theoretical_loss": 3.5072012561575674,
      "tokens_seen": 1695547392
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.00049101267854277,
      "loss": 0.0675,
      "theoretical_loss": 3.5071549474168036,
      "tokens_seen": 1695809536
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004909324346011876,
      "loss": 0.0699,
      "theoretical_loss": 3.507108647838094,
      "tokens_seen": 1696071680
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004908521906596052,
      "loss": 0.0671,
      "theoretical_loss": 3.507062357418209,
      "tokens_seen": 1696333824
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004907719467180228,
      "loss": 0.0681,
      "theoretical_loss": 3.5070160761539233,
      "tokens_seen": 1696595968
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004906917027764404,
      "loss": 0.0682,
      "theoretical_loss": 3.506969804042011,
      "tokens_seen": 1696858112
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.000490611458834858,
      "loss": 0.0707,
      "theoretical_loss": 3.5069235410792485,
      "tokens_seen": 1697120256
    },
    {
      "epoch": 0.51,
      "objective/train/advantage_avg": 0.00037086568772792816,
      "objective/train/docs_used": 619912,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4011075496673584,
      "objective/train/original_loss": 1.4011075496673584,
      "objective/train/theoretical_loss": 3.506877287262414,
      "objective/train/tokens_used": 1717842400,
      "objective/train/value_avg": -0.00641632080078125,
      "objective/train/value_loss": 0.00015510381490457803,
      "objective/train/value_max": -5.7816505432128906e-05,
      "objective/train/value_min": -0.32763671875,
      "objective/train/value_reward_corr": 0.6610992539997755,
      "objective/train/value_std": 0.01039886474609375,
      "objective/train/weight_avg": 1.000441551208496,
      "objective/train/weighted_lm_loss": 1.4019138813018799,
      "objective/train/weights_max": 1.1068991422653198,
      "objective/train/weights_min": 0.36844298243522644,
      "theoretical_loss": 3.506877287262414,
      "tokens_seen": 1697382400
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004905312148932756,
      "loss": 0.0692,
      "theoretical_loss": 3.506877287262414,
      "tokens_seen": 1697382400
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004904509709516932,
      "loss": 0.0671,
      "theoretical_loss": 3.5068310425882876,
      "tokens_seen": 1697644544
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004903707270101108,
      "loss": 0.0696,
      "theoretical_loss": 3.506784807053651,
      "tokens_seen": 1697906688
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004902904830685283,
      "loss": 0.0704,
      "theoretical_loss": 3.506738580655287,
      "tokens_seen": 1698168832
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004902102391269459,
      "loss": 0.0673,
      "theoretical_loss": 3.5066923633899796,
      "tokens_seen": 1698430976
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004901299951853635,
      "loss": 0.0683,
      "theoretical_loss": 3.506646155254516,
      "tokens_seen": 1698693120
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.000490049751243781,
      "loss": 0.0686,
      "theoretical_loss": 3.506599956245684,
      "tokens_seen": 1698955264
    },
    {
      "epoch": 0.51,
      "learning_rate": 0.0004899695073021987,
      "loss": 0.0689,
      "theoretical_loss": 3.5065537663602737,
      "tokens_seen": 1699217408
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004898892633606163,
      "loss": 0.0678,
      "theoretical_loss": 3.506507585595075,
      "tokens_seen": 1699479552
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004898090194190339,
      "loss": 0.0681,
      "theoretical_loss": 3.506461413946882,
      "tokens_seen": 1699741696
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004897287754774515,
      "loss": 0.0653,
      "theoretical_loss": 3.5064152514124887,
      "tokens_seen": 1700003840
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004896485315358691,
      "loss": 0.0701,
      "theoretical_loss": 3.506369097988691,
      "tokens_seen": 1700265984
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004895682875942866,
      "loss": 0.0678,
      "theoretical_loss": 3.5063229536722864,
      "tokens_seen": 1700528128
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004894880436527042,
      "loss": 0.0688,
      "theoretical_loss": 3.5062768184600754,
      "tokens_seen": 1700790272
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004894077997111218,
      "loss": 0.0674,
      "theoretical_loss": 3.5062306923488573,
      "tokens_seen": 1701052416
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004893275557695395,
      "loss": 0.0697,
      "theoretical_loss": 3.5061845753354355,
      "tokens_seen": 1701314560
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.000489247311827957,
      "loss": 0.0673,
      "theoretical_loss": 3.5061384674166147,
      "tokens_seen": 1701576704
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004891670678863746,
      "loss": 0.0663,
      "theoretical_loss": 3.5060923685892,
      "tokens_seen": 1701838848
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004890868239447922,
      "loss": 0.0714,
      "theoretical_loss": 3.5060462788499986,
      "tokens_seen": 1702100992
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004890065800032098,
      "loss": 0.0684,
      "theoretical_loss": 3.50600019819582,
      "tokens_seen": 1702363136
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004889263360616273,
      "loss": 0.0679,
      "theoretical_loss": 3.5059541266234744,
      "tokens_seen": 1702625280
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004888460921200449,
      "loss": 0.0688,
      "theoretical_loss": 3.505908064129775,
      "tokens_seen": 1702887424
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004887658481784625,
      "loss": 0.0655,
      "theoretical_loss": 3.5058620107115344,
      "tokens_seen": 1703149568
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004886856042368801,
      "loss": 0.068,
      "theoretical_loss": 3.505815966365568,
      "tokens_seen": 1703411712
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004886053602952977,
      "loss": 0.0701,
      "theoretical_loss": 3.5057699310886945,
      "tokens_seen": 1703673856
    },
    {
      "epoch": 0.52,
      "objective/train/advantage_avg": 0.0003058328293263912,
      "objective/train/docs_used": 622235,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3641868829727173,
      "objective/train/original_loss": 1.3641867637634277,
      "objective/train/theoretical_loss": 3.505723904877731,
      "objective/train/tokens_used": 1724396000,
      "objective/train/value_avg": -0.00513458251953125,
      "objective/train/value_loss": 0.00021216008462943137,
      "objective/train/value_max": -5.066394805908203e-05,
      "objective/train/value_min": -0.95166015625,
      "objective/train/value_reward_corr": 0.6547522900747396,
      "objective/train/value_std": 0.0124664306640625,
      "objective/train/weight_avg": 1.0004013776779175,
      "objective/train/weighted_lm_loss": 1.3642295598983765,
      "objective/train/weights_max": 2.1640307903289795,
      "objective/train/weights_min": 0.3709288537502289,
      "theoretical_loss": 3.505723904877731,
      "tokens_seen": 1703936000
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004885251163537154,
      "loss": 0.0687,
      "theoretical_loss": 3.505723904877731,
      "tokens_seen": 1703936000
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004884448724121329,
      "loss": 0.0668,
      "theoretical_loss": 3.5056778877294983,
      "tokens_seen": 1704198144
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004883646284705505,
      "loss": 0.0681,
      "theoretical_loss": 3.5056318796408186,
      "tokens_seen": 1704460288
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004882843845289681,
      "loss": 0.0684,
      "theoretical_loss": 3.505585880608515,
      "tokens_seen": 1704722432
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048820414058738565,
      "loss": 0.0682,
      "theoretical_loss": 3.505539890629412,
      "tokens_seen": 1704984576
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048812389664580326,
      "loss": 0.0683,
      "theoretical_loss": 3.5054939097003377,
      "tokens_seen": 1705246720
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004880436527042209,
      "loss": 0.0694,
      "theoretical_loss": 3.505447937818119,
      "tokens_seen": 1705508864
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048796340876263843,
      "loss": 0.0686,
      "theoretical_loss": 3.5054019749795864,
      "tokens_seen": 1705771008
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048788316482105605,
      "loss": 0.071,
      "theoretical_loss": 3.5053560211815715,
      "tokens_seen": 1706033152
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004878029208794736,
      "loss": 0.0683,
      "theoretical_loss": 3.505310076420907,
      "tokens_seen": 1706295296
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048772267693789116,
      "loss": 0.0672,
      "theoretical_loss": 3.505264140694428,
      "tokens_seen": 1706557440
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004876424329963088,
      "loss": 0.0696,
      "theoretical_loss": 3.5052182139989707,
      "tokens_seen": 1706819584
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004875621890547264,
      "loss": 0.0698,
      "theoretical_loss": 3.5051722963313723,
      "tokens_seen": 1707081728
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.000487481945113144,
      "loss": 0.0671,
      "theoretical_loss": 3.505126387688473,
      "tokens_seen": 1707343872
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048740170117156156,
      "loss": 0.0654,
      "theoretical_loss": 3.5050804880671134,
      "tokens_seen": 1707606016
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048732145722997917,
      "loss": 0.0677,
      "theoretical_loss": 3.505034597464137,
      "tokens_seen": 1707868160
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048724121328839673,
      "loss": 0.0691,
      "theoretical_loss": 3.5049887158763866,
      "tokens_seen": 1708130304
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004871609693468143,
      "loss": 0.0658,
      "theoretical_loss": 3.504942843300709,
      "tokens_seen": 1708392448
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004870807254052319,
      "loss": 0.0683,
      "theoretical_loss": 3.5048969797339513,
      "tokens_seen": 1708654592
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004870004814636495,
      "loss": 0.0703,
      "theoretical_loss": 3.5048511251729626,
      "tokens_seen": 1708916736
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004869202375220671,
      "loss": 0.068,
      "theoretical_loss": 3.504805279614594,
      "tokens_seen": 1709178880
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004868399935804847,
      "loss": 0.0704,
      "theoretical_loss": 3.504759443055696,
      "tokens_seen": 1709441024
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004867597496389023,
      "loss": 0.0679,
      "theoretical_loss": 3.5047136154931238,
      "tokens_seen": 1709703168
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048667950569731986,
      "loss": 0.0696,
      "theoretical_loss": 3.5046677969237328,
      "tokens_seen": 1709965312
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004865992617557374,
      "loss": 0.0679,
      "theoretical_loss": 3.504621987344379,
      "tokens_seen": 1710227456
    },
    {
      "epoch": 0.52,
      "objective/train/advantage_avg": 0.0006008681957609951,
      "objective/train/docs_used": 624643,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3755950927734375,
      "objective/train/original_loss": 1.3755950927734375,
      "objective/train/theoretical_loss": 3.504576186751921,
      "objective/train/tokens_used": 1730949600,
      "objective/train/value_avg": -0.008392333984375,
      "objective/train/value_loss": 0.00020491065515670925,
      "objective/train/value_max": -4.267692565917969e-05,
      "objective/train/value_min": -0.37646484375,
      "objective/train/value_reward_corr": 0.7498691017721519,
      "objective/train/value_std": 0.0162200927734375,
      "objective/train/weight_avg": 1.0006961822509766,
      "objective/train/weighted_lm_loss": 1.3765302896499634,
      "objective/train/weights_max": 1.2192577123641968,
      "objective/train/weights_min": 0.37706345319747925,
      "theoretical_loss": 3.504576186751921,
      "tokens_seen": 1710489600
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.000486519017814155,
      "loss": 0.0693,
      "theoretical_loss": 3.504576186751921,
      "tokens_seen": 1710489600
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048643877387257264,
      "loss": 0.0698,
      "theoretical_loss": 3.5045303951432194,
      "tokens_seen": 1710751744
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004863585299309902,
      "loss": 0.068,
      "theoretical_loss": 3.504484612515136,
      "tokens_seen": 1711013888
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004862782859894078,
      "loss": 0.07,
      "theoretical_loss": 3.5044388388645333,
      "tokens_seen": 1711276032
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004861980420478254,
      "loss": 0.0651,
      "theoretical_loss": 3.5043930741882763,
      "tokens_seen": 1711538176
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048611779810624304,
      "loss": 0.0695,
      "theoretical_loss": 3.504347318483232,
      "tokens_seen": 1711800320
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048603755416466054,
      "loss": 0.071,
      "theoretical_loss": 3.504301571746267,
      "tokens_seen": 1712062464
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048595731022307815,
      "loss": 0.0694,
      "theoretical_loss": 3.504255833974252,
      "tokens_seen": 1712324608
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048587706628149576,
      "loss": 0.0683,
      "theoretical_loss": 3.5042101051640575,
      "tokens_seen": 1712586752
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004857968223399133,
      "loss": 0.0707,
      "theoretical_loss": 3.5041643853125564,
      "tokens_seen": 1712848896
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048571657839833094,
      "loss": 0.0698,
      "theoretical_loss": 3.504118674416623,
      "tokens_seen": 1713111040
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048563633445674855,
      "loss": 0.0706,
      "theoretical_loss": 3.504072972473133,
      "tokens_seen": 1713373184
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048555609051516616,
      "loss": 0.0671,
      "theoretical_loss": 3.504027279478964,
      "tokens_seen": 1713635328
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048547584657358366,
      "loss": 0.0689,
      "theoretical_loss": 3.5039815954309943,
      "tokens_seen": 1713897472
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004853956026320013,
      "loss": 0.0678,
      "theoretical_loss": 3.5039359203261045,
      "tokens_seen": 1714159616
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004853153586904189,
      "loss": 0.0678,
      "theoretical_loss": 3.5038902541611776,
      "tokens_seen": 1714421760
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048523511474883645,
      "loss": 0.0685,
      "theoretical_loss": 3.503844596933096,
      "tokens_seen": 1714683904
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048515487080725406,
      "loss": 0.0709,
      "theoretical_loss": 3.503798948638746,
      "tokens_seen": 1714946048
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004850746268656717,
      "loss": 0.0715,
      "theoretical_loss": 3.503753309275013,
      "tokens_seen": 1715208192
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004849943829240893,
      "loss": 0.072,
      "theoretical_loss": 3.5037076788387864,
      "tokens_seen": 1715470336
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048491413898250684,
      "loss": 0.0689,
      "theoretical_loss": 3.503662057326956,
      "tokens_seen": 1715732480
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004848338950409244,
      "loss": 0.0688,
      "theoretical_loss": 3.503616444736412,
      "tokens_seen": 1715994624
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.000484753651099342,
      "loss": 0.0675,
      "theoretical_loss": 3.5035708410640494,
      "tokens_seen": 1716256768
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004846734071577596,
      "loss": 0.0687,
      "theoretical_loss": 3.5035252463067614,
      "tokens_seen": 1716518912
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004845931632161772,
      "loss": 0.0686,
      "theoretical_loss": 3.503479660461444,
      "tokens_seen": 1716781056
    },
    {
      "epoch": 0.52,
      "objective/train/advantage_avg": 0.0005554624949581921,
      "objective/train/docs_used": 627098,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2948997020721436,
      "objective/train/original_loss": 1.2948997020721436,
      "objective/train/theoretical_loss": 3.5034340835249953,
      "objective/train/tokens_used": 1737503200,
      "objective/train/value_avg": -0.008697509765625,
      "objective/train/value_loss": 0.0003120089240837842,
      "objective/train/value_max": -3.2961368560791016e-05,
      "objective/train/value_min": -0.9716796875,
      "objective/train/value_reward_corr": 0.7148762752931084,
      "objective/train/value_std": 0.0185394287109375,
      "objective/train/weight_avg": 1.0006957054138184,
      "objective/train/weighted_lm_loss": 1.2951641082763672,
      "objective/train/weights_max": 2.2991042137145996,
      "objective/train/weights_min": 0.3745975196361542,
      "theoretical_loss": 3.5034340835249953,
      "tokens_seen": 1717043200
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004845129192745948,
      "loss": 0.0663,
      "theoretical_loss": 3.5034340835249953,
      "tokens_seen": 1717043200
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048443267533301236,
      "loss": 0.0684,
      "theoretical_loss": 3.503388515494315,
      "tokens_seen": 1717305344
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048435243139142997,
      "loss": 0.0681,
      "theoretical_loss": 3.5033429563663026,
      "tokens_seen": 1717567488
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048427218744984753,
      "loss": 0.0709,
      "theoretical_loss": 3.503297406137861,
      "tokens_seen": 1717829632
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048419194350826514,
      "loss": 0.0691,
      "theoretical_loss": 3.5032518648058946,
      "tokens_seen": 1718091776
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004841116995666827,
      "loss": 0.0664,
      "theoretical_loss": 3.503206332367308,
      "tokens_seen": 1718353920
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004840314556251003,
      "loss": 0.0682,
      "theoretical_loss": 3.5031608088190085,
      "tokens_seen": 1718616064
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004839512116835179,
      "loss": 0.0674,
      "theoretical_loss": 3.5031152941579045,
      "tokens_seen": 1718878208
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004838709677419355,
      "loss": 0.0674,
      "theoretical_loss": 3.5030697883809063,
      "tokens_seen": 1719140352
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004837907238003531,
      "loss": 0.0697,
      "theoretical_loss": 3.5030242914849254,
      "tokens_seen": 1719402496
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048371047985877065,
      "loss": 0.0709,
      "theoretical_loss": 3.502978803466875,
      "tokens_seen": 1719664640
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048363023591718827,
      "loss": 0.0704,
      "theoretical_loss": 3.502933324323669,
      "tokens_seen": 1719926784
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004835499919756058,
      "loss": 0.0706,
      "theoretical_loss": 3.502887854052225,
      "tokens_seen": 1720188928
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048346974803402344,
      "loss": 0.0689,
      "theoretical_loss": 3.5028423926494594,
      "tokens_seen": 1720451072
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048338950409244105,
      "loss": 0.0686,
      "theoretical_loss": 3.5027969401122925,
      "tokens_seen": 1720713216
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004833092601508586,
      "loss": 0.0688,
      "theoretical_loss": 3.5027514964376447,
      "tokens_seen": 1720975360
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004832290162092762,
      "loss": 0.0703,
      "theoretical_loss": 3.502706061622438,
      "tokens_seen": 1721237504
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048314877226769383,
      "loss": 0.0684,
      "theoretical_loss": 3.502660635663597,
      "tokens_seen": 1721499648
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004830685283261114,
      "loss": 0.0699,
      "theoretical_loss": 3.5026152185580472,
      "tokens_seen": 1721761792
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048298828438452895,
      "loss": 0.0686,
      "theoretical_loss": 3.502569810302715,
      "tokens_seen": 1722023936
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048290804044294656,
      "loss": 0.0683,
      "theoretical_loss": 3.5025244108945293,
      "tokens_seen": 1722286080
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004828277965013642,
      "loss": 0.0681,
      "theoretical_loss": 3.50247902033042,
      "tokens_seen": 1722548224
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048274755255978173,
      "loss": 0.0704,
      "theoretical_loss": 3.502433638607319,
      "tokens_seen": 1722810368
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048266730861819935,
      "loss": 0.0693,
      "theoretical_loss": 3.502388265722159,
      "tokens_seen": 1723072512
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048258706467661696,
      "loss": 0.0699,
      "theoretical_loss": 3.5023429016718755,
      "tokens_seen": 1723334656
    },
    {
      "epoch": 0.52,
      "objective/train/advantage_avg": -2.2113810700830072e-05,
      "objective/train/docs_used": 629447,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4110980033874512,
      "objective/train/original_loss": 1.4110978841781616,
      "objective/train/theoretical_loss": 3.502297546453403,
      "objective/train/tokens_used": 1744056800,
      "objective/train/value_avg": -0.005741119384765625,
      "objective/train/value_loss": 0.00019440604955889285,
      "objective/train/value_max": -4.83393669128418e-05,
      "objective/train/value_min": -0.241455078125,
      "objective/train/value_reward_corr": 0.6586684600559165,
      "objective/train/value_std": 0.01081085205078125,
      "objective/train/weight_avg": 1.0000641345977783,
      "objective/train/weighted_lm_loss": 1.411281704902649,
      "objective/train/weights_max": 1.1972782611846924,
      "objective/train/weights_min": 0.3683987259864807,
      "theoretical_loss": 3.502297546453403,
      "tokens_seen": 1723596800
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048250682073503446,
      "loss": 0.0704,
      "theoretical_loss": 3.502297546453403,
      "tokens_seen": 1723596800
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004824265767934521,
      "loss": 0.0682,
      "theoretical_loss": 3.5022522000636807,
      "tokens_seen": 1723858944
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004823463328518697,
      "loss": 0.0701,
      "theoretical_loss": 3.502206862499647,
      "tokens_seen": 1724121088
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004822660889102873,
      "loss": 0.0683,
      "theoretical_loss": 3.5021615337582435,
      "tokens_seen": 1724383232
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048218584496870486,
      "loss": 0.0686,
      "theoretical_loss": 3.5021162138364117,
      "tokens_seen": 1724645376
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004821056010271225,
      "loss": 0.0692,
      "theoretical_loss": 3.502070902731096,
      "tokens_seen": 1724907520
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004820253570855401,
      "loss": 0.0715,
      "theoretical_loss": 3.502025600439241,
      "tokens_seen": 1725169664
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004819451131439576,
      "loss": 0.0717,
      "theoretical_loss": 3.5019803069577944,
      "tokens_seen": 1725431808
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004818648692023752,
      "loss": 0.0693,
      "theoretical_loss": 3.5019350222837042,
      "tokens_seen": 1725693952
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004817846252607928,
      "loss": 0.07,
      "theoretical_loss": 3.5018897464139203,
      "tokens_seen": 1725956096
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048170438131921043,
      "loss": 0.0717,
      "theoretical_loss": 3.501844479345394,
      "tokens_seen": 1726218240
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.000481624137377628,
      "loss": 0.069,
      "theoretical_loss": 3.5017992210750783,
      "tokens_seen": 1726480384
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004815438934360456,
      "loss": 0.0716,
      "theoretical_loss": 3.5017539715999275,
      "tokens_seen": 1726742528
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004814636494944632,
      "loss": 0.0697,
      "theoretical_loss": 3.5017087309168975,
      "tokens_seen": 1727004672
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048138340555288077,
      "loss": 0.0706,
      "theoretical_loss": 3.501663499022947,
      "tokens_seen": 1727266816
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048130316161129833,
      "loss": 0.068,
      "theoretical_loss": 3.5016182759150336,
      "tokens_seen": 1727528960
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048122291766971594,
      "loss": 0.0687,
      "theoretical_loss": 3.501573061590118,
      "tokens_seen": 1727791104
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048114267372813355,
      "loss": 0.0678,
      "theoretical_loss": 3.5015278560451626,
      "tokens_seen": 1728053248
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004810624297865511,
      "loss": 0.069,
      "theoretical_loss": 3.501482659277131,
      "tokens_seen": 1728315392
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004809821858449687,
      "loss": 0.0698,
      "theoretical_loss": 3.501437471282988,
      "tokens_seen": 1728577536
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048090194190338634,
      "loss": 0.067,
      "theoretical_loss": 3.5013922920597,
      "tokens_seen": 1728839680
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004808216979618039,
      "loss": 0.0687,
      "theoretical_loss": 3.5013471216042356,
      "tokens_seen": 1729101824
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048074145402022145,
      "loss": 0.0665,
      "theoretical_loss": 3.5013019599135635,
      "tokens_seen": 1729363968
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048066121007863907,
      "loss": 0.0698,
      "theoretical_loss": 3.501256806984656,
      "tokens_seen": 1729626112
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004805809661370567,
      "loss": 0.0698,
      "theoretical_loss": 3.5012116628144847,
      "tokens_seen": 1729888256
    },
    {
      "epoch": 0.52,
      "objective/train/advantage_avg": -0.00018954773258883506,
      "objective/train/docs_used": 631803,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4000455141067505,
      "objective/train/original_loss": 1.4000455141067505,
      "objective/train/theoretical_loss": 3.5011665274000245,
      "objective/train/tokens_used": 1750610400,
      "objective/train/value_avg": -0.00847625732421875,
      "objective/train/value_loss": 0.0005068992613814771,
      "objective/train/value_max": -4.792213439941406e-05,
      "objective/train/value_min": -0.79736328125,
      "objective/train/value_reward_corr": 0.705849457365212,
      "objective/train/value_std": 0.02032470703125,
      "objective/train/weight_avg": 1.0000356435775757,
      "objective/train/weighted_lm_loss": 1.4004305601119995,
      "objective/train/weights_max": 1.8320720195770264,
      "objective/train/weights_min": 0.37181854248046875,
      "theoretical_loss": 3.5011665274000245,
      "tokens_seen": 1730150400
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048050072219547424,
      "loss": 0.0667,
      "theoretical_loss": 3.5011665274000245,
      "tokens_seen": 1730150400
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048042047825389185,
      "loss": 0.0721,
      "theoretical_loss": 3.50112140073825,
      "tokens_seen": 1730412544
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048034023431230946,
      "loss": 0.0692,
      "theoretical_loss": 3.501076282826139,
      "tokens_seen": 1730674688
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.000480259990370727,
      "loss": 0.0694,
      "theoretical_loss": 3.5010311736606696,
      "tokens_seen": 1730936832
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048017974642914463,
      "loss": 0.0709,
      "theoretical_loss": 3.5009860732388223,
      "tokens_seen": 1731198976
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.0004800995024875622,
      "loss": 0.0705,
      "theoretical_loss": 3.5009409815575787,
      "tokens_seen": 1731461120
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00048001925854597975,
      "loss": 0.0676,
      "theoretical_loss": 3.500895898613922,
      "tokens_seen": 1731723264
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.00047993901460439736,
      "loss": 0.0709,
      "theoretical_loss": 3.5008508244048366,
      "tokens_seen": 1731985408
    },
    {
      "epoch": 0.52,
      "learning_rate": 0.000479858770662815,
      "loss": 0.0688,
      "theoretical_loss": 3.5008057589273083,
      "tokens_seen": 1732247552
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004797785267212326,
      "loss": 0.0674,
      "theoretical_loss": 3.500760702178325,
      "tokens_seen": 1732509696
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047969828277965015,
      "loss": 0.0655,
      "theoretical_loss": 3.5007156541548756,
      "tokens_seen": 1732771840
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047961803883806776,
      "loss": 0.0708,
      "theoretical_loss": 3.5006706148539513,
      "tokens_seen": 1733033984
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004795377948964853,
      "loss": 0.0717,
      "theoretical_loss": 3.500625584272543,
      "tokens_seen": 1733296128
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004794575509549029,
      "loss": 0.0682,
      "theoretical_loss": 3.5005805624076456,
      "tokens_seen": 1733558272
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004793773070133205,
      "loss": 0.0688,
      "theoretical_loss": 3.500535549256253,
      "tokens_seen": 1733820416
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004792970630717381,
      "loss": 0.0672,
      "theoretical_loss": 3.5004905448153627,
      "tokens_seen": 1734082560
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004792168191301557,
      "loss": 0.0705,
      "theoretical_loss": 3.500445549081972,
      "tokens_seen": 1734344704
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047913657518857327,
      "loss": 0.0691,
      "theoretical_loss": 3.500400562053081,
      "tokens_seen": 1734606848
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004790563312469909,
      "loss": 0.0699,
      "theoretical_loss": 3.50035558372569,
      "tokens_seen": 1734868992
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047897608730540844,
      "loss": 0.071,
      "theoretical_loss": 3.5003106140968026,
      "tokens_seen": 1735131136
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.000478895843363826,
      "loss": 0.0701,
      "theoretical_loss": 3.5002656531634213,
      "tokens_seen": 1735393280
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004788155994222436,
      "loss": 0.0693,
      "theoretical_loss": 3.5002207009225526,
      "tokens_seen": 1735655424
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004787353554806612,
      "loss": 0.0684,
      "theoretical_loss": 3.5001757573712036,
      "tokens_seen": 1735917568
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047865511153907884,
      "loss": 0.0701,
      "theoretical_loss": 3.5001308225063816,
      "tokens_seen": 1736179712
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004785748675974964,
      "loss": 0.0689,
      "theoretical_loss": 3.500085896325098,
      "tokens_seen": 1736441856
    },
    {
      "epoch": 0.53,
      "objective/train/advantage_avg": 0.0006846745382063091,
      "objective/train/docs_used": 634251,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4682844877243042,
      "objective/train/original_loss": 1.4682844877243042,
      "objective/train/theoretical_loss": 3.5000409788243623,
      "objective/train/tokens_used": 1757164000,
      "objective/train/value_avg": -0.00629425048828125,
      "objective/train/value_loss": 0.00014210244989953935,
      "objective/train/value_max": -5.9664249420166016e-05,
      "objective/train/value_min": -0.6083984375,
      "objective/train/value_reward_corr": 0.7120896194157069,
      "objective/train/value_std": 0.0111541748046875,
      "objective/train/weight_avg": 1.0007526874542236,
      "objective/train/weighted_lm_loss": 1.4696582555770874,
      "objective/train/weights_max": 1.7635035514831543,
      "objective/train/weights_min": 0.37317126989364624,
      "theoretical_loss": 3.5000409788243623,
      "tokens_seen": 1736704000
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.000478494623655914,
      "loss": 0.0691,
      "theoretical_loss": 3.5000409788243623,
      "tokens_seen": 1736704000
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004784143797143316,
      "loss": 0.0707,
      "theoretical_loss": 3.4999960700011896,
      "tokens_seen": 1736966144
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047833413577274913,
      "loss": 0.0697,
      "theoretical_loss": 3.4999511698525927,
      "tokens_seen": 1737228288
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047825389183116674,
      "loss": 0.0692,
      "theoretical_loss": 3.4999062783755877,
      "tokens_seen": 1737490432
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047817364788958435,
      "loss": 0.0675,
      "theoretical_loss": 3.4998613955671924,
      "tokens_seen": 1737752576
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004780934039480019,
      "loss": 0.0682,
      "theoretical_loss": 3.4998165214244246,
      "tokens_seen": 1738014720
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004780131600064195,
      "loss": 0.0694,
      "theoretical_loss": 3.499771655944306,
      "tokens_seen": 1738276864
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047793291606483714,
      "loss": 0.0682,
      "theoretical_loss": 3.4997267991238568,
      "tokens_seen": 1738539008
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047785267212325475,
      "loss": 0.0669,
      "theoretical_loss": 3.4996819509601016,
      "tokens_seen": 1738801152
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047777242818167225,
      "loss": 0.0672,
      "theoretical_loss": 3.499637111450064,
      "tokens_seen": 1739063296
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047769218424008987,
      "loss": 0.0707,
      "theoretical_loss": 3.499592280590771,
      "tokens_seen": 1739325440
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004776119402985075,
      "loss": 0.071,
      "theoretical_loss": 3.499547458379249,
      "tokens_seen": 1739587584
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047753169635692504,
      "loss": 0.0693,
      "theoretical_loss": 3.4995026448125284,
      "tokens_seen": 1739849728
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047745145241534265,
      "loss": 0.0669,
      "theoretical_loss": 3.499457839887639,
      "tokens_seen": 1740111872
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047737120847376026,
      "loss": 0.0674,
      "theoretical_loss": 3.4994130436016126,
      "tokens_seen": 1740374016
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004772909645321779,
      "loss": 0.0694,
      "theoretical_loss": 3.4993682559514836,
      "tokens_seen": 1740636160
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004772107205905954,
      "loss": 0.0689,
      "theoretical_loss": 3.4993234769342862,
      "tokens_seen": 1740898304
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.000477130476649013,
      "loss": 0.0674,
      "theoretical_loss": 3.4992787065470567,
      "tokens_seen": 1741160448
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004770502327074306,
      "loss": 0.0674,
      "theoretical_loss": 3.499233944786834,
      "tokens_seen": 1741422592
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047696998876584816,
      "loss": 0.068,
      "theoretical_loss": 3.499189191650656,
      "tokens_seen": 1741684736
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004768897448242658,
      "loss": 0.0684,
      "theoretical_loss": 3.499144447135565,
      "tokens_seen": 1741946880
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004768095008826834,
      "loss": 0.0716,
      "theoretical_loss": 3.499099711238602,
      "tokens_seen": 1742209024
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.000476729256941101,
      "loss": 0.0676,
      "theoretical_loss": 3.4990549839568112,
      "tokens_seen": 1742471168
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047664901299951856,
      "loss": 0.0704,
      "theoretical_loss": 3.4990102652872377,
      "tokens_seen": 1742733312
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004765687690579361,
      "loss": 0.0712,
      "theoretical_loss": 3.4989655552269285,
      "tokens_seen": 1742995456
    },
    {
      "epoch": 0.53,
      "objective/train/advantage_avg": 0.0005523924482986331,
      "objective/train/docs_used": 636817,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.377171277999878,
      "objective/train/original_loss": 1.377171277999878,
      "objective/train/theoretical_loss": 3.4989208537729315,
      "objective/train/tokens_used": 1763717600,
      "objective/train/value_avg": -0.00604248046875,
      "objective/train/value_loss": 0.00014253681001719087,
      "objective/train/value_max": -4.57763671875e-05,
      "objective/train/value_min": -0.94677734375,
      "objective/train/value_reward_corr": 0.7605732795954262,
      "objective/train/value_std": 0.01490020751953125,
      "objective/train/weight_avg": 1.0006181001663208,
      "objective/train/weighted_lm_loss": 1.378185510635376,
      "objective/train/weights_max": 1.3232762813568115,
      "objective/train/weights_min": 0.37375253438949585,
      "theoretical_loss": 3.4989208537729315,
      "tokens_seen": 1743257600
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047648852511635373,
      "loss": 0.0709,
      "theoretical_loss": 3.4989208537729315,
      "tokens_seen": 1743257600
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004764082811747713,
      "loss": 0.0703,
      "theoretical_loss": 3.498876160922296,
      "tokens_seen": 1743519744
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004763280372331889,
      "loss": 0.0697,
      "theoretical_loss": 3.4988314766720734,
      "tokens_seen": 1743781888
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004762477932916065,
      "loss": 0.0704,
      "theoretical_loss": 3.4987868010193157,
      "tokens_seen": 1744044032
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047616754935002407,
      "loss": 0.0647,
      "theoretical_loss": 3.4987421339610774,
      "tokens_seen": 1744306176
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004760873054084417,
      "loss": 0.0682,
      "theoretical_loss": 3.498697475494413,
      "tokens_seen": 1744568320
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047600706146685924,
      "loss": 0.0694,
      "theoretical_loss": 3.49865282561638,
      "tokens_seen": 1744830464
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047592681752527686,
      "loss": 0.07,
      "theoretical_loss": 3.498608184324037,
      "tokens_seen": 1745092608
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004758465735836944,
      "loss": 0.0699,
      "theoretical_loss": 3.498563551614443,
      "tokens_seen": 1745354752
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.000475766329642112,
      "loss": 0.0685,
      "theoretical_loss": 3.4985189274846586,
      "tokens_seen": 1745616896
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047568608570052964,
      "loss": 0.0711,
      "theoretical_loss": 3.498474311931748,
      "tokens_seen": 1745879040
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004756058417589472,
      "loss": 0.071,
      "theoretical_loss": 3.4984297049527737,
      "tokens_seen": 1746141184
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004755255978173648,
      "loss": 0.0696,
      "theoretical_loss": 3.498385106544802,
      "tokens_seen": 1746403328
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004754453538757824,
      "loss": 0.0667,
      "theoretical_loss": 3.4983405167049004,
      "tokens_seen": 1746665472
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004753651099342,
      "loss": 0.0692,
      "theoretical_loss": 3.4982959354301366,
      "tokens_seen": 1746927616
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047528486599261754,
      "loss": 0.0651,
      "theoretical_loss": 3.49825136271758,
      "tokens_seen": 1747189760
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047520462205103515,
      "loss": 0.0704,
      "theoretical_loss": 3.498206798564303,
      "tokens_seen": 1747451904
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047512437810945276,
      "loss": 0.0688,
      "theoretical_loss": 3.498162242967377,
      "tokens_seen": 1747714048
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004750441341678703,
      "loss": 0.0699,
      "theoretical_loss": 3.4981176959238773,
      "tokens_seen": 1747976192
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047496389022628794,
      "loss": 0.0676,
      "theoretical_loss": 3.4980731574308788,
      "tokens_seen": 1748238336
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047488364628470555,
      "loss": 0.0657,
      "theoretical_loss": 3.498028627485459,
      "tokens_seen": 1748500480
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004748034023431231,
      "loss": 0.0687,
      "theoretical_loss": 3.497984106084696,
      "tokens_seen": 1748762624
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047472315840154066,
      "loss": 0.0716,
      "theoretical_loss": 3.4979395932256696,
      "tokens_seen": 1749024768
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004746429144599583,
      "loss": 0.0677,
      "theoretical_loss": 3.4978950889054614,
      "tokens_seen": 1749286912
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004745626705183759,
      "loss": 0.0684,
      "theoretical_loss": 3.4978505931211545,
      "tokens_seen": 1749549056
    },
    {
      "epoch": 0.53,
      "objective/train/advantage_avg": 4.875126978731714e-05,
      "objective/train/docs_used": 639149,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4085463285446167,
      "objective/train/original_loss": 1.4085463285446167,
      "objective/train/theoretical_loss": 3.4978061058698326,
      "objective/train/tokens_used": 1770271200,
      "objective/train/value_avg": -0.00791168212890625,
      "objective/train/value_loss": 0.00021851272322237492,
      "objective/train/value_max": -4.6133995056152344e-05,
      "objective/train/value_min": -0.2459716796875,
      "objective/train/value_reward_corr": 0.7395755158288367,
      "objective/train/value_std": 0.015655517578125,
      "objective/train/weight_avg": 1.0001466274261475,
      "objective/train/weighted_lm_loss": 1.4086047410964966,
      "objective/train/weights_max": 1.2113949060440063,
      "objective/train/weights_min": 0.36827927827835083,
      "theoretical_loss": 3.4978061058698326,
      "tokens_seen": 1749811200
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047448242657679345,
      "loss": 0.0693,
      "theoretical_loss": 3.4978061058698326,
      "tokens_seen": 1749811200
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047440218263521106,
      "loss": 0.072,
      "theoretical_loss": 3.4977616271485816,
      "tokens_seen": 1750073344
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004743219386936287,
      "loss": 0.0688,
      "theoretical_loss": 3.4977171569544883,
      "tokens_seen": 1750335488
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004742416947520462,
      "loss": 0.0672,
      "theoretical_loss": 3.497672695284641,
      "tokens_seen": 1750597632
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004741614508104638,
      "loss": 0.0684,
      "theoretical_loss": 3.497628242136131,
      "tokens_seen": 1750859776
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004740812068688814,
      "loss": 0.0676,
      "theoretical_loss": 3.497583797506048,
      "tokens_seen": 1751121920
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.000474000962927299,
      "loss": 0.0701,
      "theoretical_loss": 3.4975393613914854,
      "tokens_seen": 1751384064
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004739207189857166,
      "loss": 0.0691,
      "theoretical_loss": 3.497494933789538,
      "tokens_seen": 1751646208
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004738404750441342,
      "loss": 0.071,
      "theoretical_loss": 3.4974505146973005,
      "tokens_seen": 1751908352
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004737602311025518,
      "loss": 0.0722,
      "theoretical_loss": 3.497406104111871,
      "tokens_seen": 1752170496
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047367998716096936,
      "loss": 0.0709,
      "theoretical_loss": 3.4973617020303465,
      "tokens_seen": 1752432640
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004735997432193869,
      "loss": 0.0704,
      "theoretical_loss": 3.497317308449828,
      "tokens_seen": 1752694784
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047351949927780453,
      "loss": 0.0659,
      "theoretical_loss": 3.4972729233674174,
      "tokens_seen": 1752956928
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047343925533622214,
      "loss": 0.0681,
      "theoretical_loss": 3.4972285467802164,
      "tokens_seen": 1753219072
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004733590113946397,
      "loss": 0.0708,
      "theoretical_loss": 3.4971841786853295,
      "tokens_seen": 1753481216
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004732787674530573,
      "loss": 0.0707,
      "theoretical_loss": 3.4971398190798615,
      "tokens_seen": 1753743360
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004731985235114749,
      "loss": 0.0687,
      "theoretical_loss": 3.4970954679609214,
      "tokens_seen": 1754005504
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004731182795698925,
      "loss": 0.0725,
      "theoretical_loss": 3.4970511253256156,
      "tokens_seen": 1754267648
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047303803562831004,
      "loss": 0.0718,
      "theoretical_loss": 3.4970067911710556,
      "tokens_seen": 1754529792
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047295779168672765,
      "loss": 0.0726,
      "theoretical_loss": 3.4969624654943514,
      "tokens_seen": 1754791936
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047287754774514527,
      "loss": 0.0696,
      "theoretical_loss": 3.496918148292616,
      "tokens_seen": 1755054080
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004727973038035628,
      "loss": 0.0701,
      "theoretical_loss": 3.496873839562964,
      "tokens_seen": 1755316224
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047271705986198044,
      "loss": 0.0692,
      "theoretical_loss": 3.49682953930251,
      "tokens_seen": 1755578368
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047263681592039805,
      "loss": 0.0686,
      "theoretical_loss": 3.4967852475083725,
      "tokens_seen": 1755840512
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004725565719788156,
      "loss": 0.07,
      "theoretical_loss": 3.4967409641776683,
      "tokens_seen": 1756102656
    },
    {
      "epoch": 0.53,
      "objective/train/advantage_avg": 0.000851556018460542,
      "objective/train/docs_used": 641471,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.289823293685913,
      "objective/train/original_loss": 1.289823293685913,
      "objective/train/theoretical_loss": 3.4966966893075178,
      "objective/train/tokens_used": 1776824800,
      "objective/train/value_avg": -0.00839996337890625,
      "objective/train/value_loss": 0.00039937844849191606,
      "objective/train/value_max": -6.014108657836914e-05,
      "objective/train/value_min": -0.80615234375,
      "objective/train/value_reward_corr": 0.7816001366928128,
      "objective/train/value_std": 0.0249481201171875,
      "objective/train/weight_avg": 1.0010340213775635,
      "objective/train/weighted_lm_loss": 1.2905949354171753,
      "objective/train/weights_max": 1.8249680995941162,
      "objective/train/weights_min": 0.36997780203819275,
      "theoretical_loss": 3.4966966893075178,
      "tokens_seen": 1756364800
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047247632803723317,
      "loss": 0.0674,
      "theoretical_loss": 3.4966966893075178,
      "tokens_seen": 1756364800
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004723960840956508,
      "loss": 0.0704,
      "theoretical_loss": 3.496652422895042,
      "tokens_seen": 1756626944
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004723158401540684,
      "loss": 0.0691,
      "theoretical_loss": 3.496608164937364,
      "tokens_seen": 1756889088
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047223559621248595,
      "loss": 0.0706,
      "theoretical_loss": 3.496563915431607,
      "tokens_seen": 1757151232
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047215535227090356,
      "loss": 0.0707,
      "theoretical_loss": 3.4965196743748965,
      "tokens_seen": 1757413376
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004720751083293212,
      "loss": 0.0709,
      "theoretical_loss": 3.49647544176436,
      "tokens_seen": 1757675520
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047199486438773873,
      "loss": 0.0731,
      "theoretical_loss": 3.4964312175971246,
      "tokens_seen": 1757937664
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047191462044615635,
      "loss": 0.0721,
      "theoretical_loss": 3.496387001870321,
      "tokens_seen": 1758199808
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004718343765045739,
      "loss": 0.0707,
      "theoretical_loss": 3.49634279458108,
      "tokens_seen": 1758461952
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047175413256299146,
      "loss": 0.0684,
      "theoretical_loss": 3.4962985957265333,
      "tokens_seen": 1758724096
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004716738886214091,
      "loss": 0.0681,
      "theoretical_loss": 3.496254405303815,
      "tokens_seen": 1758986240
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004715936446798267,
      "loss": 0.0704,
      "theoretical_loss": 3.4962102233100607,
      "tokens_seen": 1759248384
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004715134007382443,
      "loss": 0.0728,
      "theoretical_loss": 3.4961660497424063,
      "tokens_seen": 1759510528
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047143315679666186,
      "loss": 0.0702,
      "theoretical_loss": 3.496121884597991,
      "tokens_seen": 1759772672
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004713529128550795,
      "loss": 0.0698,
      "theoretical_loss": 3.4960777278739528,
      "tokens_seen": 1760034816
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047127266891349703,
      "loss": 0.0722,
      "theoretical_loss": 3.4960335795674338,
      "tokens_seen": 1760296960
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004711924249719146,
      "loss": 0.0703,
      "theoretical_loss": 3.495989439675575,
      "tokens_seen": 1760559104
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004711121810303322,
      "loss": 0.07,
      "theoretical_loss": 3.4959453081955205,
      "tokens_seen": 1760821248
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004710319370887498,
      "loss": 0.0687,
      "theoretical_loss": 3.495901185124416,
      "tokens_seen": 1761083392
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047095169314716743,
      "loss": 0.0692,
      "theoretical_loss": 3.4958570704594067,
      "tokens_seen": 1761345536
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.000470871449205585,
      "loss": 0.0685,
      "theoretical_loss": 3.495812964197641,
      "tokens_seen": 1761607680
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004707912052640026,
      "loss": 0.0701,
      "theoretical_loss": 3.4957688663362685,
      "tokens_seen": 1761869824
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004707109613224202,
      "loss": 0.0717,
      "theoretical_loss": 3.495724776872439,
      "tokens_seen": 1762131968
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004706307173808377,
      "loss": 0.071,
      "theoretical_loss": 3.4956806958033044,
      "tokens_seen": 1762394112
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047055047343925533,
      "loss": 0.0677,
      "theoretical_loss": 3.4956366231260185,
      "tokens_seen": 1762656256
    },
    {
      "epoch": 0.53,
      "objective/train/advantage_avg": 0.0005501354462467134,
      "objective/train/docs_used": 643830,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2951899766921997,
      "objective/train/original_loss": 1.2951900959014893,
      "objective/train/theoretical_loss": 3.4955925588377363,
      "objective/train/tokens_used": 1783378400,
      "objective/train/value_avg": -0.005947113037109375,
      "objective/train/value_loss": 0.0003123681526631117,
      "objective/train/value_max": -4.57763671875e-05,
      "objective/train/value_min": -0.490478515625,
      "objective/train/value_reward_corr": 0.5554375514170562,
      "objective/train/value_std": 0.01132965087890625,
      "objective/train/weight_avg": 1.00067937374115,
      "objective/train/weighted_lm_loss": 1.296012282371521,
      "objective/train/weights_max": 1.4989490509033203,
      "objective/train/weights_min": 0.1461181342601776,
      "theoretical_loss": 3.4955925588377363,
      "tokens_seen": 1762918400
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047047022949767294,
      "loss": 0.0682,
      "theoretical_loss": 3.4955925588377363,
      "tokens_seen": 1762918400
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047038998555609055,
      "loss": 0.0729,
      "theoretical_loss": 3.4955485029356135,
      "tokens_seen": 1763180544
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004703097416145081,
      "loss": 0.0694,
      "theoretical_loss": 3.495504455416807,
      "tokens_seen": 1763442688
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004702294976729257,
      "loss": 0.0674,
      "theoretical_loss": 3.495460416278477,
      "tokens_seen": 1763704832
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047014925373134334,
      "loss": 0.0691,
      "theoretical_loss": 3.4954163855177827,
      "tokens_seen": 1763966976
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00047006900978976084,
      "loss": 0.0702,
      "theoretical_loss": 3.495372363131886,
      "tokens_seen": 1764229120
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00046998876584817845,
      "loss": 0.0665,
      "theoretical_loss": 3.4953283491179503,
      "tokens_seen": 1764491264
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00046990852190659607,
      "loss": 0.0695,
      "theoretical_loss": 3.4952843434731395,
      "tokens_seen": 1764753408
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.0004698282779650136,
      "loss": 0.0698,
      "theoretical_loss": 3.49524034619462,
      "tokens_seen": 1765015552
    },
    {
      "epoch": 0.53,
      "learning_rate": 0.00046974803402343124,
      "loss": 0.0717,
      "theoretical_loss": 3.495196357279559,
      "tokens_seen": 1765277696
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046966779008184885,
      "loss": 0.0716,
      "theoretical_loss": 3.495152376725124,
      "tokens_seen": 1765539840
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046958754614026646,
      "loss": 0.0662,
      "theoretical_loss": 3.4951084045284864,
      "tokens_seen": 1765801984
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046950730219868397,
      "loss": 0.0698,
      "theoretical_loss": 3.495064440686816,
      "tokens_seen": 1766064128
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004694270582571016,
      "loss": 0.0699,
      "theoretical_loss": 3.495020485197287,
      "tokens_seen": 1766326272
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004693468143155192,
      "loss": 0.0703,
      "theoretical_loss": 3.494976538057073,
      "tokens_seen": 1766588416
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046926657037393675,
      "loss": 0.0709,
      "theoretical_loss": 3.4949325992633486,
      "tokens_seen": 1766850560
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046918632643235436,
      "loss": 0.0706,
      "theoretical_loss": 3.4948886688132923,
      "tokens_seen": 1767112704
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.000469106082490772,
      "loss": 0.0701,
      "theoretical_loss": 3.4948447467040804,
      "tokens_seen": 1767374848
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004690258385491896,
      "loss": 0.0694,
      "theoretical_loss": 3.494800832932894,
      "tokens_seen": 1767636992
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046894559460760715,
      "loss": 0.0662,
      "theoretical_loss": 3.494756927496913,
      "tokens_seen": 1767899136
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004688653506660247,
      "loss": 0.0684,
      "theoretical_loss": 3.49471303039332,
      "tokens_seen": 1768161280
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004687851067244423,
      "loss": 0.0707,
      "theoretical_loss": 3.4946691416192985,
      "tokens_seen": 1768423424
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004687048627828599,
      "loss": 0.069,
      "theoretical_loss": 3.4946252611720348,
      "tokens_seen": 1768685568
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004686246188412775,
      "loss": 0.0706,
      "theoretical_loss": 3.494581389048714,
      "tokens_seen": 1768947712
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004685443748996951,
      "loss": 0.0695,
      "theoretical_loss": 3.494537525246524,
      "tokens_seen": 1769209856
    },
    {
      "epoch": 0.54,
      "objective/train/advantage_avg": 0.0003884284815285355,
      "objective/train/docs_used": 646001,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2914732694625854,
      "objective/train/original_loss": 1.2914732694625854,
      "objective/train/theoretical_loss": 3.4944936697626545,
      "objective/train/tokens_used": 1789932000,
      "objective/train/value_avg": -0.00746917724609375,
      "objective/train/value_loss": 0.0005339648341760039,
      "objective/train/value_max": -4.035234451293945e-05,
      "objective/train/value_min": -0.84521484375,
      "objective/train/value_reward_corr": 0.646307946059838,
      "objective/train/value_std": 0.016021728515625,
      "objective/train/weight_avg": 1.0005794763565063,
      "objective/train/weighted_lm_loss": 1.2920138835906982,
      "objective/train/weights_max": 2.0619101524353027,
      "objective/train/weights_min": 0.03041936829686165,
      "theoretical_loss": 3.4944936697626545,
      "tokens_seen": 1769472000
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004684641309581127,
      "loss": 0.0655,
      "theoretical_loss": 3.4944936697626545,
      "tokens_seen": 1769472000
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046838388701653027,
      "loss": 0.068,
      "theoretical_loss": 3.4944498225942953,
      "tokens_seen": 1769734144
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046830364307494783,
      "loss": 0.0698,
      "theoretical_loss": 3.4944059837386394,
      "tokens_seen": 1769996288
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046822339913336544,
      "loss": 0.0712,
      "theoretical_loss": 3.494362153192879,
      "tokens_seen": 1770258432
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.000468143155191783,
      "loss": 0.0698,
      "theoretical_loss": 3.494318330954209,
      "tokens_seen": 1770520576
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004680629112502006,
      "loss": 0.069,
      "theoretical_loss": 3.494274517019826,
      "tokens_seen": 1770782720
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004679826673086182,
      "loss": 0.0701,
      "theoretical_loss": 3.494230711386926,
      "tokens_seen": 1771044864
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004679024233670358,
      "loss": 0.0682,
      "theoretical_loss": 3.4941869140527095,
      "tokens_seen": 1771307008
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004678221794254534,
      "loss": 0.0688,
      "theoretical_loss": 3.494143125014375,
      "tokens_seen": 1771569152
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046774193548387096,
      "loss": 0.0715,
      "theoretical_loss": 3.4940993442691246,
      "tokens_seen": 1771831296
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046766169154228857,
      "loss": 0.07,
      "theoretical_loss": 3.4940555718141613,
      "tokens_seen": 1772093440
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046758144760070613,
      "loss": 0.0692,
      "theoretical_loss": 3.4940118076466886,
      "tokens_seen": 1772355584
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046750120365912374,
      "loss": 0.0689,
      "theoretical_loss": 3.493968051763912,
      "tokens_seen": 1772617728
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046742095971754135,
      "loss": 0.0693,
      "theoretical_loss": 3.4939243041630395,
      "tokens_seen": 1772879872
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004673407157759589,
      "loss": 0.0699,
      "theoretical_loss": 3.4938805648412776,
      "tokens_seen": 1773142016
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004672604718343765,
      "loss": 0.07,
      "theoretical_loss": 3.4938368337958368,
      "tokens_seen": 1773404160
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046718022789279414,
      "loss": 0.0677,
      "theoretical_loss": 3.493793111023928,
      "tokens_seen": 1773666304
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004670999839512117,
      "loss": 0.0677,
      "theoretical_loss": 3.4937493965227633,
      "tokens_seen": 1773928448
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046701974000962925,
      "loss": 0.0666,
      "theoretical_loss": 3.4937056902895565,
      "tokens_seen": 1774190592
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046693949606804687,
      "loss": 0.0696,
      "theoretical_loss": 3.4936619923215226,
      "tokens_seen": 1774452736
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004668592521264645,
      "loss": 0.0661,
      "theoretical_loss": 3.493618302615878,
      "tokens_seen": 1774714880
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046677900818488204,
      "loss": 0.0687,
      "theoretical_loss": 3.4935746211698393,
      "tokens_seen": 1774977024
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046669876424329965,
      "loss": 0.0683,
      "theoretical_loss": 3.4935309479806262,
      "tokens_seen": 1775239168
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046661852030171726,
      "loss": 0.0716,
      "theoretical_loss": 3.49348728304546,
      "tokens_seen": 1775501312
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004665382763601348,
      "loss": 0.0663,
      "theoretical_loss": 3.493443626361561,
      "tokens_seen": 1775763456
    },
    {
      "epoch": 0.54,
      "objective/train/advantage_avg": 0.0006577958702109754,
      "objective/train/docs_used": 648417,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4378572702407837,
      "objective/train/original_loss": 1.4378571510314941,
      "objective/train/theoretical_loss": 3.4933999779261526,
      "objective/train/tokens_used": 1796485600,
      "objective/train/value_avg": -0.005893707275390625,
      "objective/train/value_loss": 0.00018775436910800636,
      "objective/train/value_max": -4.792213439941406e-05,
      "objective/train/value_min": -0.96630859375,
      "objective/train/value_reward_corr": 0.6805375710021935,
      "objective/train/value_std": 0.01258087158203125,
      "objective/train/weight_avg": 1.000742793083191,
      "objective/train/weighted_lm_loss": 1.4388798475265503,
      "objective/train/weights_max": 1.6557923555374146,
      "objective/train/weights_min": 0.368775874376297,
      "theoretical_loss": 3.4933999779261526,
      "tokens_seen": 1776025600
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004664580324185524,
      "loss": 0.0693,
      "theoretical_loss": 3.4933999779261526,
      "tokens_seen": 1776025600
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046637778847697,
      "loss": 0.0691,
      "theoretical_loss": 3.4933563377364596,
      "tokens_seen": 1776287744
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004662975445353876,
      "loss": 0.068,
      "theoretical_loss": 3.493312705789708,
      "tokens_seen": 1776549888
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046621730059380516,
      "loss": 0.07,
      "theoretical_loss": 3.493269082083123,
      "tokens_seen": 1776812032
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004661370566522228,
      "loss": 0.0708,
      "theoretical_loss": 3.4932254666139357,
      "tokens_seen": 1777074176
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004660568127106404,
      "loss": 0.0648,
      "theoretical_loss": 3.493181859379374,
      "tokens_seen": 1777336320
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.000465976568769058,
      "loss": 0.0684,
      "theoretical_loss": 3.4931382603766696,
      "tokens_seen": 1777598464
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004658963248274755,
      "loss": 0.0661,
      "theoretical_loss": 3.493094669603055,
      "tokens_seen": 1777860608
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004658160808858931,
      "loss": 0.0686,
      "theoretical_loss": 3.493051087055764,
      "tokens_seen": 1778122752
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046573583694431073,
      "loss": 0.071,
      "theoretical_loss": 3.493007512732031,
      "tokens_seen": 1778384896
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004656555930027283,
      "loss": 0.0697,
      "theoretical_loss": 3.4929639466290934,
      "tokens_seen": 1778647040
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004655753490611459,
      "loss": 0.0681,
      "theoretical_loss": 3.492920388744188,
      "tokens_seen": 1778909184
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004654951051195635,
      "loss": 0.0694,
      "theoretical_loss": 3.4928768390745555,
      "tokens_seen": 1779171328
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046541486117798107,
      "loss": 0.0688,
      "theoretical_loss": 3.4928332976174348,
      "tokens_seen": 1779433472
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046533461723639863,
      "loss": 0.068,
      "theoretical_loss": 3.492789764370068,
      "tokens_seen": 1779695616
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046525437329481624,
      "loss": 0.0694,
      "theoretical_loss": 3.4927462393296986,
      "tokens_seen": 1779957760
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046517412935323386,
      "loss": 0.0721,
      "theoretical_loss": 3.492702722493571,
      "tokens_seen": 1780219904
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004650938854116514,
      "loss": 0.0672,
      "theoretical_loss": 3.4926592138589307,
      "tokens_seen": 1780482048
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.000465013641470069,
      "loss": 0.0673,
      "theoretical_loss": 3.4926157134230253,
      "tokens_seen": 1780744192
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046493339752848664,
      "loss": 0.0699,
      "theoretical_loss": 3.4925722211831025,
      "tokens_seen": 1781006336
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004648531535869042,
      "loss": 0.0688,
      "theoretical_loss": 3.4925287371364124,
      "tokens_seen": 1781268480
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046477290964532176,
      "loss": 0.0661,
      "theoretical_loss": 3.4924852612802066,
      "tokens_seen": 1781530624
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046469266570373937,
      "loss": 0.0689,
      "theoretical_loss": 3.4924417936117376,
      "tokens_seen": 1781792768
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.000464612421762157,
      "loss": 0.0705,
      "theoretical_loss": 3.492398334128258,
      "tokens_seen": 1782054912
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046453217782057454,
      "loss": 0.0664,
      "theoretical_loss": 3.492354882827023,
      "tokens_seen": 1782317056
    },
    {
      "epoch": 0.54,
      "objective/train/advantage_avg": 0.0009097594302147627,
      "objective/train/docs_used": 650724,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4593218564987183,
      "objective/train/original_loss": 1.4593218564987183,
      "objective/train/theoretical_loss": 3.4923114397052903,
      "objective/train/tokens_used": 1803039200,
      "objective/train/value_avg": -0.0082244873046875,
      "objective/train/value_loss": 0.00018493611423764378,
      "objective/train/value_max": -8.350610733032227e-05,
      "objective/train/value_min": -0.359375,
      "objective/train/value_reward_corr": 0.6688092284650106,
      "objective/train/value_std": 0.01255035400390625,
      "objective/train/weight_avg": 1.0009934902191162,
      "objective/train/weighted_lm_loss": 1.4602468013763428,
      "objective/train/weights_max": 1.2040235996246338,
      "objective/train/weights_min": 0.3700779974460602,
      "theoretical_loss": 3.4923114397052903,
      "tokens_seen": 1782579200
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046445193387899215,
      "loss": 0.0709,
      "theoretical_loss": 3.4923114397052903,
      "tokens_seen": 1782579200
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046437168993740976,
      "loss": 0.0694,
      "theoretical_loss": 3.4922680047603167,
      "tokens_seen": 1782841344
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004642914459958273,
      "loss": 0.0719,
      "theoretical_loss": 3.4922245779893615,
      "tokens_seen": 1783103488
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046421120205424494,
      "loss": 0.0726,
      "theoretical_loss": 3.492181159389685,
      "tokens_seen": 1783365632
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004641309581126625,
      "loss": 0.0702,
      "theoretical_loss": 3.4921377489585486,
      "tokens_seen": 1783627776
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046405071417108005,
      "loss": 0.0692,
      "theoretical_loss": 3.4920943466932153,
      "tokens_seen": 1783889920
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046397047022949767,
      "loss": 0.068,
      "theoretical_loss": 3.49205095259095,
      "tokens_seen": 1784152064
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004638902262879153,
      "loss": 0.0685,
      "theoretical_loss": 3.492007566649018,
      "tokens_seen": 1784414208
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004638099823463329,
      "loss": 0.0681,
      "theoretical_loss": 3.491964188864686,
      "tokens_seen": 1784676352
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046372973840475045,
      "loss": 0.0728,
      "theoretical_loss": 3.491920819235223,
      "tokens_seen": 1784938496
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046364949446316806,
      "loss": 0.0719,
      "theoretical_loss": 3.491877457757898,
      "tokens_seen": 1785200640
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004635692505215856,
      "loss": 0.0681,
      "theoretical_loss": 3.491834104429982,
      "tokens_seen": 1785462784
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004634890065800032,
      "loss": 0.0672,
      "theoretical_loss": 3.491790759248747,
      "tokens_seen": 1785724928
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004634087626384208,
      "loss": 0.0707,
      "theoretical_loss": 3.491747422211467,
      "tokens_seen": 1785987072
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004633285186968384,
      "loss": 0.068,
      "theoretical_loss": 3.491704093315416,
      "tokens_seen": 1786249216
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.000463248274755256,
      "loss": 0.0692,
      "theoretical_loss": 3.4916607725578714,
      "tokens_seen": 1786511360
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004631680308136736,
      "loss": 0.0688,
      "theoretical_loss": 3.4916174599361103,
      "tokens_seen": 1786773504
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004630877868720912,
      "loss": 0.0724,
      "theoretical_loss": 3.491574155447411,
      "tokens_seen": 1787035648
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046300754293050875,
      "loss": 0.0685,
      "theoretical_loss": 3.491530859089054,
      "tokens_seen": 1787297792
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004629272989889263,
      "loss": 0.0669,
      "theoretical_loss": 3.49148757085832,
      "tokens_seen": 1787559936
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004628470550473439,
      "loss": 0.0723,
      "theoretical_loss": 3.4914442907524927,
      "tokens_seen": 1787822080
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046276681110576153,
      "loss": 0.0673,
      "theoretical_loss": 3.4914010187688556,
      "tokens_seen": 1788084224
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046268656716417914,
      "loss": 0.0692,
      "theoretical_loss": 3.4913577549046937,
      "tokens_seen": 1788346368
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004626063232225967,
      "loss": 0.0672,
      "theoretical_loss": 3.491314499157294,
      "tokens_seen": 1788608512
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004625260792810143,
      "loss": 0.0682,
      "theoretical_loss": 3.491271251523945,
      "tokens_seen": 1788870656
    },
    {
      "epoch": 0.54,
      "objective/train/advantage_avg": 0.0006975163705646992,
      "objective/train/docs_used": 653179,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.354465126991272,
      "objective/train/original_loss": 1.3544648885726929,
      "objective/train/theoretical_loss": 3.491228012001935,
      "objective/train/tokens_used": 1809592800,
      "objective/train/value_avg": -0.007549285888671875,
      "objective/train/value_loss": 0.00028828682843595743,
      "objective/train/value_max": -5.4776668548583984e-05,
      "objective/train/value_min": -0.7099609375,
      "objective/train/value_reward_corr": 0.613868199622637,
      "objective/train/value_std": 0.012359619140625,
      "objective/train/weight_avg": 1.0008219480514526,
      "objective/train/weighted_lm_loss": 1.3553131818771362,
      "objective/train/weights_max": 1.442611575126648,
      "objective/train/weights_min": 0.36832353472709656,
      "theoretical_loss": 3.491228012001935,
      "tokens_seen": 1789132800
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004624458353394319,
      "loss": 0.0687,
      "theoretical_loss": 3.491228012001935,
      "tokens_seen": 1789132800
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046236559139784943,
      "loss": 0.0684,
      "theoretical_loss": 3.4911847805885547,
      "tokens_seen": 1789394944
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046228534745626704,
      "loss": 0.0684,
      "theoretical_loss": 3.491141557281096,
      "tokens_seen": 1789657088
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046220510351468465,
      "loss": 0.0728,
      "theoretical_loss": 3.491098342076852,
      "tokens_seen": 1789919232
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046212485957310227,
      "loss": 0.0681,
      "theoretical_loss": 3.4910551349731183,
      "tokens_seen": 1790181376
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004620446156315198,
      "loss": 0.0677,
      "theoretical_loss": 3.4910119359671885,
      "tokens_seen": 1790443520
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046196437168993744,
      "loss": 0.0655,
      "theoretical_loss": 3.490968745056361,
      "tokens_seen": 1790705664
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046188412774835505,
      "loss": 0.0686,
      "theoretical_loss": 3.4909255622379343,
      "tokens_seen": 1790967808
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046180388380677256,
      "loss": 0.0686,
      "theoretical_loss": 3.490882387509207,
      "tokens_seen": 1791229952
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046172363986519017,
      "loss": 0.0702,
      "theoretical_loss": 3.490839220867481,
      "tokens_seen": 1791492096
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004616433959236078,
      "loss": 0.0659,
      "theoretical_loss": 3.490796062310058,
      "tokens_seen": 1791754240
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046156315198202534,
      "loss": 0.0683,
      "theoretical_loss": 3.4907529118342415,
      "tokens_seen": 1792016384
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046148290804044295,
      "loss": 0.0677,
      "theoretical_loss": 3.490709769437337,
      "tokens_seen": 1792278528
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046140266409886056,
      "loss": 0.0654,
      "theoretical_loss": 3.49066663511665,
      "tokens_seen": 1792540672
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004613224201572782,
      "loss": 0.0686,
      "theoretical_loss": 3.4906235088694872,
      "tokens_seen": 1792802816
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004612421762156957,
      "loss": 0.0678,
      "theoretical_loss": 3.4905803906931587,
      "tokens_seen": 1793064960
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004611619322741133,
      "loss": 0.0653,
      "theoretical_loss": 3.4905372805849737,
      "tokens_seen": 1793327104
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004610816883325309,
      "loss": 0.0658,
      "theoretical_loss": 3.490494178542243,
      "tokens_seen": 1793589248
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046100144439094846,
      "loss": 0.0662,
      "theoretical_loss": 3.4904510845622805,
      "tokens_seen": 1793851392
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004609212004493661,
      "loss": 0.0691,
      "theoretical_loss": 3.490407998642399,
      "tokens_seen": 1794113536
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004608409565077837,
      "loss": 0.0693,
      "theoretical_loss": 3.4903649207799137,
      "tokens_seen": 1794375680
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004607607125662013,
      "loss": 0.0691,
      "theoretical_loss": 3.4903218509721414,
      "tokens_seen": 1794637824
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046068046862461886,
      "loss": 0.0675,
      "theoretical_loss": 3.4902787892163993,
      "tokens_seen": 1794899968
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004606002246830364,
      "loss": 0.0683,
      "theoretical_loss": 3.490235735510007,
      "tokens_seen": 1795162112
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046051998074145403,
      "loss": 0.0701,
      "theoretical_loss": 3.490192689850284,
      "tokens_seen": 1795424256
    },
    {
      "epoch": 0.54,
      "objective/train/advantage_avg": -0.0002183630713261664,
      "objective/train/docs_used": 655419,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.333343505859375,
      "objective/train/original_loss": 1.333343505859375,
      "objective/train/theoretical_loss": 3.4901496522345528,
      "objective/train/tokens_used": 1816146400,
      "objective/train/value_avg": -0.00982666015625,
      "objective/train/value_loss": 0.0002944105945061892,
      "objective/train/value_max": -4.1961669921875e-05,
      "objective/train/value_min": -0.771484375,
      "objective/train/value_reward_corr": 0.7564874172701911,
      "objective/train/value_std": 0.0181732177734375,
      "objective/train/weight_avg": 0.9999178647994995,
      "objective/train/weighted_lm_loss": 1.3324216604232788,
      "objective/train/weights_max": 1.3296234607696533,
      "objective/train/weights_min": 0.3700949549674988,
      "theoretical_loss": 3.4901496522345528,
      "tokens_seen": 1795686400
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004604397367998716,
      "loss": 0.0698,
      "theoretical_loss": 3.4901496522345528,
      "tokens_seen": 1795686400
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004603594928582892,
      "loss": 0.0687,
      "theoretical_loss": 3.490106622660136,
      "tokens_seen": 1795948544
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004602792489167068,
      "loss": 0.0695,
      "theoretical_loss": 3.4900636011243567,
      "tokens_seen": 1796210688
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046019900497512443,
      "loss": 0.0704,
      "theoretical_loss": 3.4900205876245414,
      "tokens_seen": 1796472832
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.000460118761033542,
      "loss": 0.0706,
      "theoretical_loss": 3.4899775821580166,
      "tokens_seen": 1796734976
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00046003851709195954,
      "loss": 0.0689,
      "theoretical_loss": 3.4899345847221097,
      "tokens_seen": 1796997120
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00045995827315037716,
      "loss": 0.0665,
      "theoretical_loss": 3.4898915953141505,
      "tokens_seen": 1797259264
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004598780292087947,
      "loss": 0.0678,
      "theoretical_loss": 3.4898486139314695,
      "tokens_seen": 1797521408
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00045979778526721233,
      "loss": 0.0699,
      "theoretical_loss": 3.489805640571398,
      "tokens_seen": 1797783552
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.00045971754132562994,
      "loss": 0.0664,
      "theoretical_loss": 3.48976267523127,
      "tokens_seen": 1798045696
    },
    {
      "epoch": 0.54,
      "learning_rate": 0.0004596372973840475,
      "loss": 0.0679,
      "theoretical_loss": 3.4897197179084185,
      "tokens_seen": 1798307840
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004595570534424651,
      "loss": 0.07,
      "theoretical_loss": 3.48967676860018,
      "tokens_seen": 1798569984
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004594768095008827,
      "loss": 0.0704,
      "theoretical_loss": 3.4896338273038916,
      "tokens_seen": 1798832128
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004593965655593003,
      "loss": 0.0665,
      "theoretical_loss": 3.4895908940168905,
      "tokens_seen": 1799094272
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045931632161771784,
      "loss": 0.07,
      "theoretical_loss": 3.489547968736517,
      "tokens_seen": 1799356416
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045923607767613545,
      "loss": 0.0702,
      "theoretical_loss": 3.4895050514601116,
      "tokens_seen": 1799618560
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045915583373455307,
      "loss": 0.0686,
      "theoretical_loss": 3.4894621421850163,
      "tokens_seen": 1799880704
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004590755897929706,
      "loss": 0.069,
      "theoretical_loss": 3.489419240908574,
      "tokens_seen": 1800142848
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045899534585138824,
      "loss": 0.0706,
      "theoretical_loss": 3.48937634762813,
      "tokens_seen": 1800404992
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045891510190980585,
      "loss": 0.0673,
      "theoretical_loss": 3.489333462341029,
      "tokens_seen": 1800667136
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004588348579682234,
      "loss": 0.0686,
      "theoretical_loss": 3.4892905850446185,
      "tokens_seen": 1800929280
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045875461402664097,
      "loss": 0.0674,
      "theoretical_loss": 3.489247715736247,
      "tokens_seen": 1801191424
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004586743700850586,
      "loss": 0.0695,
      "theoretical_loss": 3.489204854413264,
      "tokens_seen": 1801453568
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004585941261434762,
      "loss": 0.0662,
      "theoretical_loss": 3.48916200107302,
      "tokens_seen": 1801715712
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045851388220189375,
      "loss": 0.0668,
      "theoretical_loss": 3.489119155712868,
      "tokens_seen": 1801977856
    },
    {
      "epoch": 0.55,
      "objective/train/advantage_avg": 0.0006118986057117581,
      "objective/train/docs_used": 657910,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.438297986984253,
      "objective/train/original_loss": 1.4382978677749634,
      "objective/train/theoretical_loss": 3.4890763183301607,
      "objective/train/tokens_used": 1822700000,
      "objective/train/value_avg": -0.00920867919921875,
      "objective/train/value_loss": 0.00051107257604599,
      "objective/train/value_max": -4.9114227294921875e-05,
      "objective/train/value_min": -0.7216796875,
      "objective/train/value_reward_corr": 0.6640711326255471,
      "objective/train/value_std": 0.019927978515625,
      "objective/train/weight_avg": 1.0008348226547241,
      "objective/train/weighted_lm_loss": 1.4394363164901733,
      "objective/train/weights_max": 2.024425983428955,
      "objective/train/weights_min": 0.37084680795669556,
      "theoretical_loss": 3.4890763183301607,
      "tokens_seen": 1802240000
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045843363826031136,
      "loss": 0.0682,
      "theoretical_loss": 3.4890763183301607,
      "tokens_seen": 1802240000
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.000458353394318729,
      "loss": 0.0694,
      "theoretical_loss": 3.489033488922253,
      "tokens_seen": 1802502144
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045827315037714653,
      "loss": 0.0672,
      "theoretical_loss": 3.4889906674865,
      "tokens_seen": 1802764288
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004581929064355641,
      "loss": 0.0719,
      "theoretical_loss": 3.48894785402026,
      "tokens_seen": 1803026432
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004581126624939817,
      "loss": 0.0676,
      "theoretical_loss": 3.4889050485208912,
      "tokens_seen": 1803288576
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004580324185523993,
      "loss": 0.0682,
      "theoretical_loss": 3.4888622509857523,
      "tokens_seen": 1803550720
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004579521746108169,
      "loss": 0.0653,
      "theoretical_loss": 3.488819461412205,
      "tokens_seen": 1803812864
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004578719306692345,
      "loss": 0.0706,
      "theoretical_loss": 3.4887766797976116,
      "tokens_seen": 1804075008
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004577916867276521,
      "loss": 0.0705,
      "theoretical_loss": 3.488733906139336,
      "tokens_seen": 1804337152
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004577114427860697,
      "loss": 0.0709,
      "theoretical_loss": 3.4886911404347414,
      "tokens_seen": 1804599296
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004576311988444872,
      "loss": 0.0675,
      "theoretical_loss": 3.4886483826811947,
      "tokens_seen": 1804861440
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045755095490290483,
      "loss": 0.0675,
      "theoretical_loss": 3.4886056328760633,
      "tokens_seen": 1805123584
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045747071096132244,
      "loss": 0.0697,
      "theoretical_loss": 3.4885628910167155,
      "tokens_seen": 1805385728
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045739046701974,
      "loss": 0.0703,
      "theoretical_loss": 3.488520157100521,
      "tokens_seen": 1805647872
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004573102230781576,
      "loss": 0.0703,
      "theoretical_loss": 3.4884774311248505,
      "tokens_seen": 1805910016
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045722997913657523,
      "loss": 0.0676,
      "theoretical_loss": 3.4884347130870768,
      "tokens_seen": 1806172160
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004571497351949928,
      "loss": 0.0662,
      "theoretical_loss": 3.4883920029845727,
      "tokens_seen": 1806434304
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045706949125341034,
      "loss": 0.0691,
      "theoretical_loss": 3.488349300814713,
      "tokens_seen": 1806696448
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045698924731182796,
      "loss": 0.0691,
      "theoretical_loss": 3.4883066065748745,
      "tokens_seen": 1806958592
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045690900337024557,
      "loss": 0.0706,
      "theoretical_loss": 3.488263920262434,
      "tokens_seen": 1807220736
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045682875942866313,
      "loss": 0.0687,
      "theoretical_loss": 3.4882212418747693,
      "tokens_seen": 1807482880
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045674851548708074,
      "loss": 0.0686,
      "theoretical_loss": 3.4881785714092617,
      "tokens_seen": 1807745024
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045666827154549835,
      "loss": 0.0704,
      "theoretical_loss": 3.48813590886329,
      "tokens_seen": 1808007168
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004565880276039159,
      "loss": 0.0687,
      "theoretical_loss": 3.488093254234238,
      "tokens_seen": 1808269312
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045650778366233347,
      "loss": 0.0682,
      "theoretical_loss": 3.488050607519489,
      "tokens_seen": 1808531456
    },
    {
      "epoch": 0.55,
      "objective/train/advantage_avg": 0.0004177616792730987,
      "objective/train/docs_used": 660069,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3710644245147705,
      "objective/train/original_loss": 1.37106454372406,
      "objective/train/theoretical_loss": 3.4880079687164267,
      "objective/train/tokens_used": 1829253600,
      "objective/train/value_avg": -0.0078277587890625,
      "objective/train/value_loss": 0.0003445453767199069,
      "objective/train/value_max": -6.920099258422852e-05,
      "objective/train/value_min": -0.86474609375,
      "objective/train/value_reward_corr": 0.692850757955269,
      "objective/train/value_std": 0.01751708984375,
      "objective/train/weight_avg": 1.000558853149414,
      "objective/train/weighted_lm_loss": 1.3716151714324951,
      "objective/train/weights_max": 1.3765496015548706,
      "objective/train/weights_min": 0.2556014955043793,
      "theoretical_loss": 3.4880079687164267,
      "tokens_seen": 1808793600
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004564275397207511,
      "loss": 0.07,
      "theoretical_loss": 3.4880079687164267,
      "tokens_seen": 1808793600
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004563472957791687,
      "loss": 0.0677,
      "theoretical_loss": 3.4879653378224384,
      "tokens_seen": 1809055744
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045626705183758625,
      "loss": 0.0696,
      "theoretical_loss": 3.4879227148349106,
      "tokens_seen": 1809317888
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045618680789600387,
      "loss": 0.0664,
      "theoretical_loss": 3.487880099751232,
      "tokens_seen": 1809580032
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004561065639544215,
      "loss": 0.0645,
      "theoretical_loss": 3.487837492568792,
      "tokens_seen": 1809842176
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045602632001283904,
      "loss": 0.0697,
      "theoretical_loss": 3.487794893284981,
      "tokens_seen": 1810104320
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045594607607125665,
      "loss": 0.0684,
      "theoretical_loss": 3.487752301897192,
      "tokens_seen": 1810366464
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004558658321296742,
      "loss": 0.0701,
      "theoretical_loss": 3.487709718402818,
      "tokens_seen": 1810628608
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045578558818809177,
      "loss": 0.0676,
      "theoretical_loss": 3.4876671427992543,
      "tokens_seen": 1810890752
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004557053442465094,
      "loss": 0.0671,
      "theoretical_loss": 3.4876245750838955,
      "tokens_seen": 1811152896
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.000455625100304927,
      "loss": 0.0693,
      "theoretical_loss": 3.48758201525414,
      "tokens_seen": 1811415040
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004555448563633446,
      "loss": 0.0669,
      "theoretical_loss": 3.487539463307385,
      "tokens_seen": 1811677184
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045546461242176216,
      "loss": 0.0679,
      "theoretical_loss": 3.4874969192410306,
      "tokens_seen": 1811939328
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004553843684801798,
      "loss": 0.0674,
      "theoretical_loss": 3.4874543830524782,
      "tokens_seen": 1812201472
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045530412453859733,
      "loss": 0.0693,
      "theoretical_loss": 3.487411854739128,
      "tokens_seen": 1812463616
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004552238805970149,
      "loss": 0.0691,
      "theoretical_loss": 3.487369334298386,
      "tokens_seen": 1812725760
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004551436366554325,
      "loss": 0.0673,
      "theoretical_loss": 3.4873268217276543,
      "tokens_seen": 1812987904
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004550633927138501,
      "loss": 0.0696,
      "theoretical_loss": 3.48728431702434,
      "tokens_seen": 1813250048
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045498314877226773,
      "loss": 0.0672,
      "theoretical_loss": 3.4872418201858495,
      "tokens_seen": 1813512192
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004549029048306853,
      "loss": 0.069,
      "theoretical_loss": 3.487199331209591,
      "tokens_seen": 1813774336
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004548226608891029,
      "loss": 0.0707,
      "theoretical_loss": 3.487156850092974,
      "tokens_seen": 1814036480
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004547424169475205,
      "loss": 0.0717,
      "theoretical_loss": 3.487114376833409,
      "tokens_seen": 1814298624
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.000454662173005938,
      "loss": 0.0678,
      "theoretical_loss": 3.487071911428308,
      "tokens_seen": 1814560768
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045458192906435563,
      "loss": 0.0659,
      "theoretical_loss": 3.487029453875085,
      "tokens_seen": 1814822912
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045450168512277324,
      "loss": 0.0706,
      "theoretical_loss": 3.4869870041711524,
      "tokens_seen": 1815085056
    },
    {
      "epoch": 0.55,
      "objective/train/advantage_avg": 0.00177646370138973,
      "objective/train/docs_used": 662349,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.329512119293213,
      "objective/train/original_loss": 1.329512119293213,
      "objective/train/theoretical_loss": 3.4869445623139272,
      "objective/train/tokens_used": 1835807200,
      "objective/train/value_avg": -0.00986480712890625,
      "objective/train/value_loss": 0.0003300151147413999,
      "objective/train/value_max": -3.0934810638427734e-05,
      "objective/train/value_min": -0.7333984375,
      "objective/train/value_reward_corr": 0.7485666886816452,
      "objective/train/value_std": 0.021759033203125,
      "objective/train/weight_avg": 1.0019299983978271,
      "objective/train/weighted_lm_loss": 1.3307819366455078,
      "objective/train/weights_max": 1.4997124671936035,
      "objective/train/weights_min": 0.38332873582839966,
      "theoretical_loss": 3.4869445623139272,
      "tokens_seen": 1815347200
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045442144118119086,
      "loss": 0.0685,
      "theoretical_loss": 3.4869445623139272,
      "tokens_seen": 1815347200
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004543411972396084,
      "loss": 0.0717,
      "theoretical_loss": 3.4869021283008257,
      "tokens_seen": 1815609344
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.000454260953298026,
      "loss": 0.067,
      "theoretical_loss": 3.4868597021292658,
      "tokens_seen": 1815871488
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045418070935644364,
      "loss": 0.0711,
      "theoretical_loss": 3.4868172837966673,
      "tokens_seen": 1816133632
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045410046541486114,
      "loss": 0.0664,
      "theoretical_loss": 3.48677487330045,
      "tokens_seen": 1816395776
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045402022147327876,
      "loss": 0.0683,
      "theoretical_loss": 3.486732470638036,
      "tokens_seen": 1816657920
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045393997753169637,
      "loss": 0.0686,
      "theoretical_loss": 3.4866900758068478,
      "tokens_seen": 1816920064
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.000453859733590114,
      "loss": 0.0673,
      "theoretical_loss": 3.4866476888043096,
      "tokens_seen": 1817182208
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045377948964853154,
      "loss": 0.0682,
      "theoretical_loss": 3.486605309627847,
      "tokens_seen": 1817444352
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045369924570694915,
      "loss": 0.0693,
      "theoretical_loss": 3.4865629382748864,
      "tokens_seen": 1817706496
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045361900176536676,
      "loss": 0.07,
      "theoretical_loss": 3.486520574742855,
      "tokens_seen": 1817968640
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045353875782378427,
      "loss": 0.0667,
      "theoretical_loss": 3.486478219029183,
      "tokens_seen": 1818230784
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004534585138822019,
      "loss": 0.0698,
      "theoretical_loss": 3.4864358711312997,
      "tokens_seen": 1818492928
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004533782699406195,
      "loss": 0.068,
      "theoretical_loss": 3.4863935310466365,
      "tokens_seen": 1818755072
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045329802599903705,
      "loss": 0.0688,
      "theoretical_loss": 3.486351198772626,
      "tokens_seen": 1819017216
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045321778205745467,
      "loss": 0.068,
      "theoretical_loss": 3.4863088743067023,
      "tokens_seen": 1819279360
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004531375381158723,
      "loss": 0.0707,
      "theoretical_loss": 3.4862665576463003,
      "tokens_seen": 1819541504
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004530572941742899,
      "loss": 0.0678,
      "theoretical_loss": 3.4862242487888566,
      "tokens_seen": 1819803648
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045297705023270745,
      "loss": 0.0679,
      "theoretical_loss": 3.486181947731808,
      "tokens_seen": 1820065792
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.000452896806291125,
      "loss": 0.0687,
      "theoretical_loss": 3.486139654472594,
      "tokens_seen": 1820327936
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004528165623495426,
      "loss": 0.0681,
      "theoretical_loss": 3.486097369008654,
      "tokens_seen": 1820590080
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004527363184079602,
      "loss": 0.0704,
      "theoretical_loss": 3.4860550913374286,
      "tokens_seen": 1820852224
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004526560744663778,
      "loss": 0.0709,
      "theoretical_loss": 3.4860128214563613,
      "tokens_seen": 1821114368
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004525758305247954,
      "loss": 0.0709,
      "theoretical_loss": 3.4859705593628947,
      "tokens_seen": 1821376512
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.000452495586583213,
      "loss": 0.0679,
      "theoretical_loss": 3.4859283050544736,
      "tokens_seen": 1821638656
    },
    {
      "epoch": 0.55,
      "objective/train/advantage_avg": 2.5156406991300173e-05,
      "objective/train/docs_used": 664649,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.30966317653656,
      "objective/train/original_loss": 1.3096632957458496,
      "objective/train/theoretical_loss": 3.4858860585285445,
      "objective/train/tokens_used": 1842360800,
      "objective/train/value_avg": -0.008636474609375,
      "objective/train/value_loss": 0.00020235766714904457,
      "objective/train/value_max": -5.269050598144531e-05,
      "objective/train/value_min": -0.487548828125,
      "objective/train/value_reward_corr": 0.7453598247617829,
      "objective/train/value_std": 0.0155029296875,
      "objective/train/weight_avg": 1.0001190900802612,
      "objective/train/weighted_lm_loss": 1.3092310428619385,
      "objective/train/weights_max": 1.396862268447876,
      "objective/train/weights_min": 0.37186679244041443,
      "theoretical_loss": 3.4858860585285445,
      "tokens_seen": 1821900800
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004524153426416306,
      "loss": 0.0679,
      "theoretical_loss": 3.4858860585285445,
      "tokens_seen": 1821900800
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045233509870004813,
      "loss": 0.0697,
      "theoretical_loss": 3.485843819782554,
      "tokens_seen": 1822162944
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045225485475846575,
      "loss": 0.0718,
      "theoretical_loss": 3.48580158881395,
      "tokens_seen": 1822425088
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004521746108168833,
      "loss": 0.0691,
      "theoretical_loss": 3.4857593656201833,
      "tokens_seen": 1822687232
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004520943668753009,
      "loss": 0.0702,
      "theoretical_loss": 3.4857171501987034,
      "tokens_seen": 1822949376
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045201412293371853,
      "loss": 0.067,
      "theoretical_loss": 3.4856749425469635,
      "tokens_seen": 1823211520
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045193387899213614,
      "loss": 0.0737,
      "theoretical_loss": 3.485632742662416,
      "tokens_seen": 1823473664
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004518536350505537,
      "loss": 0.0698,
      "theoretical_loss": 3.4855905505425144,
      "tokens_seen": 1823735808
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045177339110897126,
      "loss": 0.069,
      "theoretical_loss": 3.485548366184716,
      "tokens_seen": 1823997952
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045169314716738887,
      "loss": 0.0711,
      "theoretical_loss": 3.4855061895864763,
      "tokens_seen": 1824260096
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045161290322580643,
      "loss": 0.0691,
      "theoretical_loss": 3.485464020745253,
      "tokens_seen": 1824522240
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045153265928422404,
      "loss": 0.0668,
      "theoretical_loss": 3.4854218596585067,
      "tokens_seen": 1824784384
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045145241534264165,
      "loss": 0.0675,
      "theoretical_loss": 3.485379706323697,
      "tokens_seen": 1825046528
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004513721714010592,
      "loss": 0.0691,
      "theoretical_loss": 3.4853375607382846,
      "tokens_seen": 1825308672
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004512919274594768,
      "loss": 0.0689,
      "theoretical_loss": 3.4852954228997337,
      "tokens_seen": 1825570816
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045121168351789444,
      "loss": 0.069,
      "theoretical_loss": 3.485253292805507,
      "tokens_seen": 1825832960
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.000451131439576312,
      "loss": 0.0703,
      "theoretical_loss": 3.48521117045307,
      "tokens_seen": 1826095104
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045105119563472956,
      "loss": 0.0706,
      "theoretical_loss": 3.4851690558398896,
      "tokens_seen": 1826357248
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045097095169314717,
      "loss": 0.0669,
      "theoretical_loss": 3.4851269489634324,
      "tokens_seen": 1826619392
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004508907077515648,
      "loss": 0.0662,
      "theoretical_loss": 3.4850848498211677,
      "tokens_seen": 1826881536
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045081046380998234,
      "loss": 0.0668,
      "theoretical_loss": 3.4850427584105654,
      "tokens_seen": 1827143680
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045073021986839995,
      "loss": 0.0679,
      "theoretical_loss": 3.4850006747290965,
      "tokens_seen": 1827405824
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045064997592681756,
      "loss": 0.0708,
      "theoretical_loss": 3.4849585987742326,
      "tokens_seen": 1827667968
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004505697319852351,
      "loss": 0.0712,
      "theoretical_loss": 3.4849165305434484,
      "tokens_seen": 1827930112
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004504894880436527,
      "loss": 0.0668,
      "theoretical_loss": 3.4848744700342174,
      "tokens_seen": 1828192256
    },
    {
      "epoch": 0.55,
      "objective/train/advantage_avg": -0.00015672965673729777,
      "objective/train/docs_used": 666768,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3809049129486084,
      "objective/train/original_loss": 1.3809049129486084,
      "objective/train/theoretical_loss": 3.484832417244016,
      "objective/train/tokens_used": 1848914400,
      "objective/train/value_avg": -0.0067596435546875,
      "objective/train/value_loss": 0.0001564290578244254,
      "objective/train/value_max": -6.502866744995117e-05,
      "objective/train/value_min": -0.6142578125,
      "objective/train/value_reward_corr": 0.7000283534341674,
      "objective/train/value_std": 0.012420654296875,
      "objective/train/weight_avg": 0.9999170303344727,
      "objective/train/weighted_lm_loss": 1.3802940845489502,
      "objective/train/weights_max": 1.7114393711090088,
      "objective/train/weights_min": 0.36869922280311584,
      "theoretical_loss": 3.484832417244016,
      "tokens_seen": 1828454400
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004504092441020703,
      "loss": 0.0698,
      "theoretical_loss": 3.484832417244016,
      "tokens_seen": 1828454400
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004503290001604879,
      "loss": 0.0681,
      "theoretical_loss": 3.484790372170321,
      "tokens_seen": 1828716544
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00045024875621890546,
      "loss": 0.067,
      "theoretical_loss": 3.484748334810611,
      "tokens_seen": 1828978688
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004501685122773231,
      "loss": 0.0678,
      "theoretical_loss": 3.484706305162365,
      "tokens_seen": 1829240832
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004500882683357407,
      "loss": 0.0672,
      "theoretical_loss": 3.484664283223064,
      "tokens_seen": 1829502976
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004500080243941583,
      "loss": 0.0691,
      "theoretical_loss": 3.484622268990189,
      "tokens_seen": 1829765120
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004499277804525758,
      "loss": 0.0691,
      "theoretical_loss": 3.4845802624612237,
      "tokens_seen": 1830027264
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004498475365109934,
      "loss": 0.0677,
      "theoretical_loss": 3.484538263633652,
      "tokens_seen": 1830289408
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.00044976729256941103,
      "loss": 0.0694,
      "theoretical_loss": 3.484496272504959,
      "tokens_seen": 1830551552
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004496870486278286,
      "loss": 0.0676,
      "theoretical_loss": 3.484454289072631,
      "tokens_seen": 1830813696
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004496068046862462,
      "loss": 0.0693,
      "theoretical_loss": 3.4844123133341567,
      "tokens_seen": 1831075840
    },
    {
      "epoch": 0.55,
      "learning_rate": 0.0004495265607446638,
      "loss": 0.0672,
      "theoretical_loss": 3.484370345287024,
      "tokens_seen": 1831337984
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044944631680308143,
      "loss": 0.0665,
      "theoretical_loss": 3.484328384928723,
      "tokens_seen": 1831600128
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044936607286149893,
      "loss": 0.0675,
      "theoretical_loss": 3.484286432256745,
      "tokens_seen": 1831862272
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044928582891991654,
      "loss": 0.0686,
      "theoretical_loss": 3.484244487268583,
      "tokens_seen": 1832124416
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044920558497833416,
      "loss": 0.0688,
      "theoretical_loss": 3.4842025499617297,
      "tokens_seen": 1832386560
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004491253410367517,
      "loss": 0.0717,
      "theoretical_loss": 3.4841606203336806,
      "tokens_seen": 1832648704
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044904509709516933,
      "loss": 0.0664,
      "theoretical_loss": 3.4841186983819306,
      "tokens_seen": 1832910848
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044896485315358694,
      "loss": 0.0681,
      "theoretical_loss": 3.4840767841039777,
      "tokens_seen": 1833172992
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004488846092120045,
      "loss": 0.0697,
      "theoretical_loss": 3.48403487749732,
      "tokens_seen": 1833435136
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044880436527042206,
      "loss": 0.0707,
      "theoretical_loss": 3.4839929785594563,
      "tokens_seen": 1833697280
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044872412132883967,
      "loss": 0.0679,
      "theoretical_loss": 3.4839510872878883,
      "tokens_seen": 1833959424
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004486438773872573,
      "loss": 0.0674,
      "theoretical_loss": 3.483909203680117,
      "tokens_seen": 1834221568
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044856363344567484,
      "loss": 0.0695,
      "theoretical_loss": 3.483867327733645,
      "tokens_seen": 1834483712
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044848338950409245,
      "loss": 0.0676,
      "theoretical_loss": 3.4838254594459777,
      "tokens_seen": 1834745856
    },
    {
      "epoch": 0.56,
      "objective/train/advantage_avg": -0.00017360829224344343,
      "objective/train/docs_used": 669245,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2899640798568726,
      "objective/train/original_loss": 1.289963960647583,
      "objective/train/theoretical_loss": 3.483783598814619,
      "objective/train/tokens_used": 1855468000,
      "objective/train/value_avg": -0.0079498291015625,
      "objective/train/value_loss": 0.0003152432036586106,
      "objective/train/value_max": -5.561113357543945e-05,
      "objective/train/value_min": -0.9013671875,
      "objective/train/value_reward_corr": 0.652301360066071,
      "objective/train/value_std": 0.01739501953125,
      "objective/train/weight_avg": 0.9999733567237854,
      "objective/train/weighted_lm_loss": 1.290113091468811,
      "objective/train/weights_max": 2.32053279876709,
      "objective/train/weights_min": 0.37400925159454346,
      "theoretical_loss": 3.483783598814619,
      "tokens_seen": 1835008000
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044840314556251007,
      "loss": 0.0685,
      "theoretical_loss": 3.483783598814619,
      "tokens_seen": 1835008000
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004483229016209276,
      "loss": 0.069,
      "theoretical_loss": 3.4837417458370767,
      "tokens_seen": 1835270144
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044824265767934524,
      "loss": 0.0672,
      "theoretical_loss": 3.483699900510857,
      "tokens_seen": 1835532288
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004481624137377628,
      "loss": 0.069,
      "theoretical_loss": 3.4836580628334697,
      "tokens_seen": 1835794432
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004480821697961804,
      "loss": 0.0719,
      "theoretical_loss": 3.4836162328024245,
      "tokens_seen": 1836056576
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044800192585459797,
      "loss": 0.0683,
      "theoretical_loss": 3.4835744104152324,
      "tokens_seen": 1836318720
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004479216819130156,
      "loss": 0.0698,
      "theoretical_loss": 3.483532595669406,
      "tokens_seen": 1836580864
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004478414379714332,
      "loss": 0.0654,
      "theoretical_loss": 3.4834907885624586,
      "tokens_seen": 1836843008
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044776119402985075,
      "loss": 0.0658,
      "theoretical_loss": 3.4834489890919045,
      "tokens_seen": 1837105152
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044768095008826836,
      "loss": 0.0674,
      "theoretical_loss": 3.48340719725526,
      "tokens_seen": 1837367296
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004476007061466859,
      "loss": 0.0697,
      "theoretical_loss": 3.4833654130500413,
      "tokens_seen": 1837629440
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004475204622051035,
      "loss": 0.0665,
      "theoretical_loss": 3.4833236364737674,
      "tokens_seen": 1837891584
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004474402182635211,
      "loss": 0.0677,
      "theoretical_loss": 3.4832818675239574,
      "tokens_seen": 1838153728
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004473599743219387,
      "loss": 0.0687,
      "theoretical_loss": 3.483240106198131,
      "tokens_seen": 1838415872
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004472797303803563,
      "loss": 0.0709,
      "theoretical_loss": 3.483198352493811,
      "tokens_seen": 1838678016
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004471994864387739,
      "loss": 0.0719,
      "theoretical_loss": 3.4831566064085187,
      "tokens_seen": 1838940160
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004471192424971915,
      "loss": 0.068,
      "theoretical_loss": 3.483114867939779,
      "tokens_seen": 1839202304
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044703899855560905,
      "loss": 0.0676,
      "theoretical_loss": 3.4830731370851167,
      "tokens_seen": 1839464448
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004469587546140266,
      "loss": 0.0674,
      "theoretical_loss": 3.483031413842058,
      "tokens_seen": 1839726592
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004468785106724442,
      "loss": 0.07,
      "theoretical_loss": 3.4829896982081303,
      "tokens_seen": 1839988736
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044679826673086183,
      "loss": 0.0701,
      "theoretical_loss": 3.4829479901808624,
      "tokens_seen": 1840250880
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044671802278927944,
      "loss": 0.0668,
      "theoretical_loss": 3.482906289757784,
      "tokens_seen": 1840513024
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.000446637778847697,
      "loss": 0.0699,
      "theoretical_loss": 3.482864596936425,
      "tokens_seen": 1840775168
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004465575349061146,
      "loss": 0.0684,
      "theoretical_loss": 3.482822911714318,
      "tokens_seen": 1841037312
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044647729096453223,
      "loss": 0.0688,
      "theoretical_loss": 3.4827812340889963,
      "tokens_seen": 1841299456
    },
    {
      "epoch": 0.56,
      "objective/train/advantage_avg": -8.307833195431158e-05,
      "objective/train/docs_used": 671743,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3671631813049316,
      "objective/train/original_loss": 1.367163062095642,
      "objective/train/theoretical_loss": 3.4827395640579946,
      "objective/train/tokens_used": 1862021600,
      "objective/train/value_avg": -0.0087738037109375,
      "objective/train/value_loss": 0.0003252418537158519,
      "objective/train/value_max": -2.110004425048828e-05,
      "objective/train/value_min": -0.84423828125,
      "objective/train/value_reward_corr": 0.8539613338364523,
      "objective/train/value_std": 0.0264739990234375,
      "objective/train/weight_avg": 1.0000675916671753,
      "objective/train/weighted_lm_loss": 1.3670684099197388,
      "objective/train/weights_max": 1.6681559085845947,
      "objective/train/weights_min": 0.3783082664012909,
      "theoretical_loss": 3.4827395640579946,
      "tokens_seen": 1841561600
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044639704702294973,
      "loss": 0.069,
      "theoretical_loss": 3.4827395640579946,
      "tokens_seen": 1841561600
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044631680308136734,
      "loss": 0.0656,
      "theoretical_loss": 3.4826979016188475,
      "tokens_seen": 1841823744
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044623655913978496,
      "loss": 0.069,
      "theoretical_loss": 3.482656246769092,
      "tokens_seen": 1842085888
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044615631519820257,
      "loss": 0.0682,
      "theoretical_loss": 3.482614599506266,
      "tokens_seen": 1842348032
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044607607125662013,
      "loss": 0.0695,
      "theoretical_loss": 3.482572959827908,
      "tokens_seen": 1842610176
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044599582731503774,
      "loss": 0.07,
      "theoretical_loss": 3.482531327731558,
      "tokens_seen": 1842872320
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044591558337345535,
      "loss": 0.0688,
      "theoretical_loss": 3.4824897032147577,
      "tokens_seen": 1843134464
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044583533943187286,
      "loss": 0.0674,
      "theoretical_loss": 3.482448086275049,
      "tokens_seen": 1843396608
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044575509549029047,
      "loss": 0.0658,
      "theoretical_loss": 3.4824064769099756,
      "tokens_seen": 1843658752
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004456748515487081,
      "loss": 0.0666,
      "theoretical_loss": 3.4823648751170824,
      "tokens_seen": 1843920896
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004455946076071257,
      "loss": 0.0667,
      "theoretical_loss": 3.482323280893915,
      "tokens_seen": 1844183040
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044551436366554325,
      "loss": 0.0669,
      "theoretical_loss": 3.4822816942380195,
      "tokens_seen": 1844445184
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044543411972396087,
      "loss": 0.0671,
      "theoretical_loss": 3.4822401151469453,
      "tokens_seen": 1844707328
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004453538757823785,
      "loss": 0.0703,
      "theoretical_loss": 3.4821985436182405,
      "tokens_seen": 1844969472
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044527363184079604,
      "loss": 0.0655,
      "theoretical_loss": 3.4821569796494565,
      "tokens_seen": 1845231616
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004451933878992136,
      "loss": 0.0659,
      "theoretical_loss": 3.482115423238144,
      "tokens_seen": 1845493760
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004451131439576312,
      "loss": 0.0711,
      "theoretical_loss": 3.4820738743818556,
      "tokens_seen": 1845755904
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044503290001604877,
      "loss": 0.0696,
      "theoretical_loss": 3.4820323330781457,
      "tokens_seen": 1846018048
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004449526560744664,
      "loss": 0.0677,
      "theoretical_loss": 3.481990799324568,
      "tokens_seen": 1846280192
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.000444872412132884,
      "loss": 0.0686,
      "theoretical_loss": 3.4819492731186807,
      "tokens_seen": 1846542336
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004447921681913016,
      "loss": 0.0667,
      "theoretical_loss": 3.481907754458039,
      "tokens_seen": 1846804480
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044471192424971916,
      "loss": 0.0685,
      "theoretical_loss": 3.4818662433402014,
      "tokens_seen": 1847066624
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004446316803081367,
      "loss": 0.0686,
      "theoretical_loss": 3.4818247397627284,
      "tokens_seen": 1847328768
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044455143636655433,
      "loss": 0.0669,
      "theoretical_loss": 3.48178324372318,
      "tokens_seen": 1847590912
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004444711924249719,
      "loss": 0.0649,
      "theoretical_loss": 3.481741755219118,
      "tokens_seen": 1847853056
    },
    {
      "epoch": 0.56,
      "objective/train/advantage_avg": 0.00012247166887391359,
      "objective/train/docs_used": 673981,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3901854753494263,
      "objective/train/original_loss": 1.3901854753494263,
      "objective/train/theoretical_loss": 3.481700274248105,
      "objective/train/tokens_used": 1868575200,
      "objective/train/value_avg": -0.007476806640625,
      "objective/train/value_loss": 0.00014100054977461696,
      "objective/train/value_max": -5.346536636352539e-05,
      "objective/train/value_min": -0.218505859375,
      "objective/train/value_reward_corr": 0.7216289321788012,
      "objective/train/value_std": 0.01219940185546875,
      "objective/train/weight_avg": 1.0001883506774902,
      "objective/train/weighted_lm_loss": 1.3908172845840454,
      "objective/train/weights_max": 1.1437007188796997,
      "objective/train/weights_min": 0.37020933628082275,
      "theoretical_loss": 3.481700274248105,
      "tokens_seen": 1848115200
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004443909484833895,
      "loss": 0.0708,
      "theoretical_loss": 3.481700274248105,
      "tokens_seen": 1848115200
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004443107045418071,
      "loss": 0.0708,
      "theoretical_loss": 3.481658800807706,
      "tokens_seen": 1848377344
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044423046060022473,
      "loss": 0.0704,
      "theoretical_loss": 3.4816173348954846,
      "tokens_seen": 1848639488
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004441502166586423,
      "loss": 0.0696,
      "theoretical_loss": 3.481575876509008,
      "tokens_seen": 1848901632
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044406997271705985,
      "loss": 0.0726,
      "theoretical_loss": 3.4815344256458434,
      "tokens_seen": 1849163776
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044398972877547746,
      "loss": 0.0702,
      "theoretical_loss": 3.4814929823035596,
      "tokens_seen": 1849425920
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.000443909484833895,
      "loss": 0.0708,
      "theoretical_loss": 3.481451546479726,
      "tokens_seen": 1849688064
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044382924089231263,
      "loss": 0.0685,
      "theoretical_loss": 3.4814101181719135,
      "tokens_seen": 1849950208
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044374899695073024,
      "loss": 0.0693,
      "theoretical_loss": 3.4813686973776936,
      "tokens_seen": 1850212352
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044366875300914786,
      "loss": 0.067,
      "theoretical_loss": 3.48132728409464,
      "tokens_seen": 1850474496
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004435885090675654,
      "loss": 0.0712,
      "theoretical_loss": 3.4812858783203264,
      "tokens_seen": 1850736640
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.000443508265125983,
      "loss": 0.0718,
      "theoretical_loss": 3.481244480052329,
      "tokens_seen": 1850998784
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004434280211844006,
      "loss": 0.0674,
      "theoretical_loss": 3.4812030892882224,
      "tokens_seen": 1851260928
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044334777724281814,
      "loss": 0.0671,
      "theoretical_loss": 3.4811617060255857,
      "tokens_seen": 1851523072
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044326753330123576,
      "loss": 0.0692,
      "theoretical_loss": 3.481120330261997,
      "tokens_seen": 1851785216
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044318728935965337,
      "loss": 0.0677,
      "theoretical_loss": 3.4810789619950366,
      "tokens_seen": 1852047360
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004431070454180709,
      "loss": 0.0679,
      "theoretical_loss": 3.481037601222285,
      "tokens_seen": 1852309504
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044302680147648854,
      "loss": 0.069,
      "theoretical_loss": 3.480996247941324,
      "tokens_seen": 1852571648
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044294655753490615,
      "loss": 0.0682,
      "theoretical_loss": 3.4809549021497372,
      "tokens_seen": 1852833792
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004428663135933237,
      "loss": 0.0691,
      "theoretical_loss": 3.480913563845109,
      "tokens_seen": 1853095936
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044278606965174127,
      "loss": 0.066,
      "theoretical_loss": 3.480872233025024,
      "tokens_seen": 1853358080
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004427058257101589,
      "loss": 0.0694,
      "theoretical_loss": 3.4808309096870698,
      "tokens_seen": 1853620224
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004426255817685765,
      "loss": 0.0667,
      "theoretical_loss": 3.480789593828834,
      "tokens_seen": 1853882368
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044254533782699405,
      "loss": 0.0695,
      "theoretical_loss": 3.4807482854479037,
      "tokens_seen": 1854144512
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044246509388541167,
      "loss": 0.0692,
      "theoretical_loss": 3.4807069845418708,
      "tokens_seen": 1854406656
    },
    {
      "epoch": 0.56,
      "objective/train/advantage_avg": 0.0009944553021341562,
      "objective/train/docs_used": 676483,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.5289674997329712,
      "objective/train/original_loss": 1.5289674997329712,
      "objective/train/theoretical_loss": 3.4806656911083254,
      "objective/train/tokens_used": 1875128800,
      "objective/train/value_avg": -0.00623321533203125,
      "objective/train/value_loss": 0.00023596876417286694,
      "objective/train/value_max": -6.35385513305664e-05,
      "objective/train/value_min": -0.64013671875,
      "objective/train/value_reward_corr": 0.6387599212671781,
      "objective/train/value_std": 0.01309967041015625,
      "objective/train/weight_avg": 1.0010981559753418,
      "objective/train/weighted_lm_loss": 1.5305681228637695,
      "objective/train/weights_max": 1.3230394124984741,
      "objective/train/weights_min": 0.37203988432884216,
      "theoretical_loss": 3.4806656911083254,
      "tokens_seen": 1854668800
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004423848499438293,
      "loss": 0.0721,
      "theoretical_loss": 3.4806656911083254,
      "tokens_seen": 1854668800
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044230460600224684,
      "loss": 0.069,
      "theoretical_loss": 3.48062440514486,
      "tokens_seen": 1854930944
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004422243620606644,
      "loss": 0.0693,
      "theoretical_loss": 3.4805831266490674,
      "tokens_seen": 1855193088
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.000442144118119082,
      "loss": 0.067,
      "theoretical_loss": 3.480541855618542,
      "tokens_seen": 1855455232
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004420638741774996,
      "loss": 0.0668,
      "theoretical_loss": 3.4805005920508796,
      "tokens_seen": 1855717376
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004419836302359172,
      "loss": 0.0689,
      "theoretical_loss": 3.480459335943676,
      "tokens_seen": 1855979520
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004419033862943348,
      "loss": 0.0684,
      "theoretical_loss": 3.4804180872945305,
      "tokens_seen": 1856241664
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004418231423527524,
      "loss": 0.0653,
      "theoretical_loss": 3.4803768461010405,
      "tokens_seen": 1856503808
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044174289841117,
      "loss": 0.0707,
      "theoretical_loss": 3.4803356123608062,
      "tokens_seen": 1856765952
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004416626544695875,
      "loss": 0.0673,
      "theoretical_loss": 3.480294386071429,
      "tokens_seen": 1857028096
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044158241052800513,
      "loss": 0.068,
      "theoretical_loss": 3.4802531672305106,
      "tokens_seen": 1857290240
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044150216658642275,
      "loss": 0.0702,
      "theoretical_loss": 3.480211955835654,
      "tokens_seen": 1857552384
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004414219226448403,
      "loss": 0.0687,
      "theoretical_loss": 3.4801707518844647,
      "tokens_seen": 1857814528
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004413416787032579,
      "loss": 0.0655,
      "theoretical_loss": 3.480129555374547,
      "tokens_seen": 1858076672
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044126143476167553,
      "loss": 0.0692,
      "theoretical_loss": 3.4800883663035083,
      "tokens_seen": 1858338816
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004411811908200931,
      "loss": 0.0654,
      "theoretical_loss": 3.4800471846689556,
      "tokens_seen": 1858600960
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044110094687851065,
      "loss": 0.0707,
      "theoretical_loss": 3.4800060104684984,
      "tokens_seen": 1858863104
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044102070293692826,
      "loss": 0.0642,
      "theoretical_loss": 3.4799648436997463,
      "tokens_seen": 1859125248
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044094045899534587,
      "loss": 0.0671,
      "theoretical_loss": 3.47992368436031,
      "tokens_seen": 1859387392
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044086021505376343,
      "loss": 0.0661,
      "theoretical_loss": 3.4798825324478018,
      "tokens_seen": 1859649536
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044077997111218104,
      "loss": 0.07,
      "theoretical_loss": 3.4798413879598353,
      "tokens_seen": 1859911680
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044069972717059865,
      "loss": 0.0657,
      "theoretical_loss": 3.4798002508940242,
      "tokens_seen": 1860173824
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004406194832290162,
      "loss": 0.0696,
      "theoretical_loss": 3.479759121247984,
      "tokens_seen": 1860435968
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004405392392874338,
      "loss": 0.0692,
      "theoretical_loss": 3.479717999019332,
      "tokens_seen": 1860698112
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004404589953458514,
      "loss": 0.067,
      "theoretical_loss": 3.4796768842056847,
      "tokens_seen": 1860960256
    },
    {
      "epoch": 0.56,
      "objective/train/advantage_avg": 0.00022138823987916112,
      "objective/train/docs_used": 678968,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3515822887420654,
      "objective/train/original_loss": 1.3515822887420654,
      "objective/train/theoretical_loss": 3.4796357768046615,
      "objective/train/tokens_used": 1881682400,
      "objective/train/value_avg": -0.008880615234375,
      "objective/train/value_loss": 0.0006506952340714633,
      "objective/train/value_max": -6.604194641113281e-05,
      "objective/train/value_min": -0.92333984375,
      "objective/train/value_reward_corr": 0.6781693074134842,
      "objective/train/value_std": 0.0211181640625,
      "objective/train/weight_avg": 1.0004839897155762,
      "objective/train/weighted_lm_loss": 1.3513437509536743,
      "objective/train/weights_max": 2.46537446975708,
      "objective/train/weights_min": 0.0841136947274208,
      "theoretical_loss": 3.4796357768046615,
      "tokens_seen": 1861222400
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.000440378751404269,
      "loss": 0.07,
      "theoretical_loss": 3.4796357768046615,
      "tokens_seen": 1861222400
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044029850746268656,
      "loss": 0.0674,
      "theoretical_loss": 3.4795946768138823,
      "tokens_seen": 1861484544
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044021826352110417,
      "loss": 0.0688,
      "theoretical_loss": 3.4795535842309677,
      "tokens_seen": 1861746688
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004401380195795218,
      "loss": 0.0714,
      "theoretical_loss": 3.4795124990535395,
      "tokens_seen": 1862008832
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00044005777563793934,
      "loss": 0.0658,
      "theoretical_loss": 3.479471421279222,
      "tokens_seen": 1862270976
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00043997753169635695,
      "loss": 0.0695,
      "theoretical_loss": 3.4794303509056377,
      "tokens_seen": 1862533120
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004398972877547745,
      "loss": 0.0679,
      "theoretical_loss": 3.479389287930413,
      "tokens_seen": 1862795264
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004398170438131921,
      "loss": 0.0683,
      "theoretical_loss": 3.4793482323511746,
      "tokens_seen": 1863057408
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004397367998716097,
      "loss": 0.071,
      "theoretical_loss": 3.479307184165549,
      "tokens_seen": 1863319552
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004396565559300273,
      "loss": 0.0677,
      "theoretical_loss": 3.4792661433711656,
      "tokens_seen": 1863581696
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004395763119884449,
      "loss": 0.0688,
      "theoretical_loss": 3.479225109965653,
      "tokens_seen": 1863843840
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.00043949606804686246,
      "loss": 0.0706,
      "theoretical_loss": 3.4791840839466435,
      "tokens_seen": 1864105984
    },
    {
      "epoch": 0.56,
      "learning_rate": 0.0004394158241052801,
      "loss": 0.0689,
      "theoretical_loss": 3.479143065311768,
      "tokens_seen": 1864368128
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043933558016369764,
      "loss": 0.0668,
      "theoretical_loss": 3.47910205405866,
      "tokens_seen": 1864630272
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004392553362221152,
      "loss": 0.0649,
      "theoretical_loss": 3.479061050184953,
      "tokens_seen": 1864892416
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004391750922805328,
      "loss": 0.0689,
      "theoretical_loss": 3.4790200536882825,
      "tokens_seen": 1865154560
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004390948483389504,
      "loss": 0.0687,
      "theoretical_loss": 3.4789790645662846,
      "tokens_seen": 1865416704
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043901460439736803,
      "loss": 0.07,
      "theoretical_loss": 3.478938082816597,
      "tokens_seen": 1865678848
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004389343604557856,
      "loss": 0.068,
      "theoretical_loss": 3.4788971084368576,
      "tokens_seen": 1865940992
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004388541165142032,
      "loss": 0.0706,
      "theoretical_loss": 3.478856141424706,
      "tokens_seen": 1866203136
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004387738725726208,
      "loss": 0.0697,
      "theoretical_loss": 3.478815181777783,
      "tokens_seen": 1866465280
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004386936286310383,
      "loss": 0.0665,
      "theoretical_loss": 3.4787742294937303,
      "tokens_seen": 1866727424
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043861338468945593,
      "loss": 0.0688,
      "theoretical_loss": 3.4787332845701906,
      "tokens_seen": 1866989568
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043853314074787354,
      "loss": 0.0672,
      "theoretical_loss": 3.4786923470048077,
      "tokens_seen": 1867251712
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043845289680629116,
      "loss": 0.0699,
      "theoretical_loss": 3.478651416795227,
      "tokens_seen": 1867513856
    },
    {
      "epoch": 0.57,
      "objective/train/advantage_avg": -3.919263690477237e-05,
      "objective/train/docs_used": 681429,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3138712644577026,
      "objective/train/original_loss": 1.313871145248413,
      "objective/train/theoretical_loss": 3.478610493939094,
      "objective/train/tokens_used": 1888236000,
      "objective/train/value_avg": -0.005817413330078125,
      "objective/train/value_loss": 0.0002530510537326336,
      "objective/train/value_max": -4.792213439941406e-05,
      "objective/train/value_min": -0.5234375,
      "objective/train/value_reward_corr": 0.7415155710875803,
      "objective/train/value_std": 0.0134124755859375,
      "objective/train/weight_avg": 1.0000742673873901,
      "objective/train/weighted_lm_loss": 1.3138768672943115,
      "objective/train/weights_max": 1.5628552436828613,
      "objective/train/weights_min": 0.36993688344955444,
      "theoretical_loss": 3.478610493939094,
      "tokens_seen": 1867776000
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004383726528647087,
      "loss": 0.0671,
      "theoretical_loss": 3.478610493939094,
      "tokens_seen": 1867776000
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043829240892312633,
      "loss": 0.0688,
      "theoretical_loss": 3.4785695784340556,
      "tokens_seen": 1868038144
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043821216498154394,
      "loss": 0.0668,
      "theoretical_loss": 3.4785286702777602,
      "tokens_seen": 1868300288
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043813192103996145,
      "loss": 0.0661,
      "theoretical_loss": 3.4784877694678573,
      "tokens_seen": 1868562432
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043805167709837906,
      "loss": 0.0696,
      "theoretical_loss": 3.478446876001997,
      "tokens_seen": 1868824576
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043797143315679667,
      "loss": 0.0703,
      "theoretical_loss": 3.4784059898778312,
      "tokens_seen": 1869086720
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004378911892152143,
      "loss": 0.0695,
      "theoretical_loss": 3.4783651110930123,
      "tokens_seen": 1869348864
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043781094527363184,
      "loss": 0.0697,
      "theoretical_loss": 3.478324239645193,
      "tokens_seen": 1869611008
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043773070133204945,
      "loss": 0.0714,
      "theoretical_loss": 3.478283375532029,
      "tokens_seen": 1869873152
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043765045739046707,
      "loss": 0.07,
      "theoretical_loss": 3.4782425187511756,
      "tokens_seen": 1870135296
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043757021344888457,
      "loss": 0.0687,
      "theoretical_loss": 3.47820166930029,
      "tokens_seen": 1870397440
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004374899695073022,
      "loss": 0.0675,
      "theoretical_loss": 3.47816082717703,
      "tokens_seen": 1870659584
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004374097255657198,
      "loss": 0.0664,
      "theoretical_loss": 3.478119992379054,
      "tokens_seen": 1870921728
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004373294816241374,
      "loss": 0.0721,
      "theoretical_loss": 3.478079164904022,
      "tokens_seen": 1871183872
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043724923768255497,
      "loss": 0.071,
      "theoretical_loss": 3.4780383447495966,
      "tokens_seen": 1871446016
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004371689937409726,
      "loss": 0.0667,
      "theoretical_loss": 3.477997531913439,
      "tokens_seen": 1871708160
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004370887497993902,
      "loss": 0.0716,
      "theoretical_loss": 3.477956726393212,
      "tokens_seen": 1871970304
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043700850585780775,
      "loss": 0.0702,
      "theoretical_loss": 3.477915928186581,
      "tokens_seen": 1872232448
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004369282619162253,
      "loss": 0.0696,
      "theoretical_loss": 3.4778751372912105,
      "tokens_seen": 1872494592
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004368480179746429,
      "loss": 0.0682,
      "theoretical_loss": 3.4778343537047673,
      "tokens_seen": 1872756736
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004367677740330605,
      "loss": 0.0688,
      "theoretical_loss": 3.4777935774249196,
      "tokens_seen": 1873018880
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004366875300914781,
      "loss": 0.0706,
      "theoretical_loss": 3.4777528084493348,
      "tokens_seen": 1873281024
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004366072861498957,
      "loss": 0.0682,
      "theoretical_loss": 3.477712046775684,
      "tokens_seen": 1873543168
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004365270422083133,
      "loss": 0.0684,
      "theoretical_loss": 3.477671292401637,
      "tokens_seen": 1873805312
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004364467982667309,
      "loss": 0.068,
      "theoretical_loss": 3.477630545324866,
      "tokens_seen": 1874067456
    },
    {
      "epoch": 0.57,
      "objective/train/advantage_avg": -0.00036099160206504166,
      "objective/train/docs_used": 683764,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4436589479446411,
      "objective/train/original_loss": 1.4436590671539307,
      "objective/train/theoretical_loss": 3.477589805543044,
      "objective/train/tokens_used": 1894789600,
      "objective/train/value_avg": -0.00724029541015625,
      "objective/train/value_loss": 0.00032180329435504973,
      "objective/train/value_max": -3.8504600524902344e-05,
      "objective/train/value_min": -0.9072265625,
      "objective/train/value_reward_corr": 0.6553256771569635,
      "objective/train/value_std": 0.01497650146484375,
      "objective/train/weight_avg": 0.9997800588607788,
      "objective/train/weighted_lm_loss": 1.4435096979141235,
      "objective/train/weights_max": 1.7920018434524536,
      "objective/train/weights_min": 0.3784064054489136,
      "theoretical_loss": 3.477589805543044,
      "tokens_seen": 1874329600
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043636655432514843,
      "loss": 0.0707,
      "theoretical_loss": 3.477589805543044,
      "tokens_seen": 1874329600
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043628631038356605,
      "loss": 0.072,
      "theoretical_loss": 3.477549073053845,
      "tokens_seen": 1874591744
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004362060664419836,
      "loss": 0.0702,
      "theoretical_loss": 3.477508347854944,
      "tokens_seen": 1874853888
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004361258225004012,
      "loss": 0.0688,
      "theoretical_loss": 3.477467629944017,
      "tokens_seen": 1875116032
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043604557855881883,
      "loss": 0.0667,
      "theoretical_loss": 3.4774269193187406,
      "tokens_seen": 1875378176
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043596533461723644,
      "loss": 0.0697,
      "theoretical_loss": 3.4773862159767943,
      "tokens_seen": 1875640320
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.000435885090675654,
      "loss": 0.0696,
      "theoretical_loss": 3.4773455199158567,
      "tokens_seen": 1875902464
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004358048467340716,
      "loss": 0.0718,
      "theoretical_loss": 3.4773048311336083,
      "tokens_seen": 1876164608
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004357246027924892,
      "loss": 0.0682,
      "theoretical_loss": 3.4772641496277306,
      "tokens_seen": 1876426752
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043564435885090673,
      "loss": 0.0674,
      "theoretical_loss": 3.4772234753959057,
      "tokens_seen": 1876688896
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043556411490932434,
      "loss": 0.0691,
      "theoretical_loss": 3.477182808435818,
      "tokens_seen": 1876951040
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043548387096774196,
      "loss": 0.0676,
      "theoretical_loss": 3.477142148745151,
      "tokens_seen": 1877213184
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043540362702615957,
      "loss": 0.07,
      "theoretical_loss": 3.477101496321591,
      "tokens_seen": 1877475328
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043532338308457713,
      "loss": 0.0734,
      "theoretical_loss": 3.4770608511628254,
      "tokens_seen": 1877737472
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043524313914299474,
      "loss": 0.0724,
      "theoretical_loss": 3.477020213266541,
      "tokens_seen": 1877999616
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004351628952014123,
      "loss": 0.0716,
      "theoretical_loss": 3.476979582630427,
      "tokens_seen": 1878261760
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043508265125982986,
      "loss": 0.0701,
      "theoretical_loss": 3.4769389592521733,
      "tokens_seen": 1878523904
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043500240731824747,
      "loss": 0.0721,
      "theoretical_loss": 3.476898343129471,
      "tokens_seen": 1878786048
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004349221633766651,
      "loss": 0.0659,
      "theoretical_loss": 3.476857734260012,
      "tokens_seen": 1879048192
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043484191943508264,
      "loss": 0.0692,
      "theoretical_loss": 3.4768171326414894,
      "tokens_seen": 1879310336
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043476167549350025,
      "loss": 0.0677,
      "theoretical_loss": 3.4767765382715976,
      "tokens_seen": 1879572480
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043468143155191787,
      "loss": 0.0697,
      "theoretical_loss": 3.4767359511480316,
      "tokens_seen": 1879834624
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004346011876103354,
      "loss": 0.0692,
      "theoretical_loss": 3.4766953712684874,
      "tokens_seen": 1880096768
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.000434520943668753,
      "loss": 0.0678,
      "theoretical_loss": 3.476654798630663,
      "tokens_seen": 1880358912
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004344406997271706,
      "loss": 0.0683,
      "theoretical_loss": 3.476614233232256,
      "tokens_seen": 1880621056
    },
    {
      "epoch": 0.57,
      "objective/train/advantage_avg": -0.0004898487823083997,
      "objective/train/docs_used": 686045,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3733599185943604,
      "objective/train/original_loss": 1.3733599185943604,
      "objective/train/theoretical_loss": 3.476573675070966,
      "objective/train/tokens_used": 1901343200,
      "objective/train/value_avg": -0.0057830810546875,
      "objective/train/value_loss": 0.00014441322127822787,
      "objective/train/value_max": -3.510713577270508e-05,
      "objective/train/value_min": -0.3486328125,
      "objective/train/value_reward_corr": 0.7300531562767005,
      "objective/train/value_std": 0.010986328125,
      "objective/train/weight_avg": 0.9995789527893066,
      "objective/train/weighted_lm_loss": 1.3724981546401978,
      "objective/train/weights_max": 1.1518362760543823,
      "objective/train/weights_min": 0.36823925375938416,
      "theoretical_loss": 3.476573675070966,
      "tokens_seen": 1880883200
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004343604557855882,
      "loss": 0.0674,
      "theoretical_loss": 3.476573675070966,
      "tokens_seen": 1880883200
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043428021184400577,
      "loss": 0.0704,
      "theoretical_loss": 3.4765331241444937,
      "tokens_seen": 1881145344
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004341999679024234,
      "loss": 0.0665,
      "theoretical_loss": 3.4764925804505404,
      "tokens_seen": 1881407488
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.000434119723960841,
      "loss": 0.067,
      "theoretical_loss": 3.476452043986809,
      "tokens_seen": 1881669632
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004340394800192586,
      "loss": 0.067,
      "theoretical_loss": 3.4764115147510033,
      "tokens_seen": 1881931776
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004339592360776761,
      "loss": 0.0705,
      "theoretical_loss": 3.4763709927408266,
      "tokens_seen": 1882193920
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004338789921360937,
      "loss": 0.0678,
      "theoretical_loss": 3.4763304779539865,
      "tokens_seen": 1882456064
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043379874819451133,
      "loss": 0.0713,
      "theoretical_loss": 3.476289970388188,
      "tokens_seen": 1882718208
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004337185042529289,
      "loss": 0.0694,
      "theoretical_loss": 3.4762494700411404,
      "tokens_seen": 1882980352
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004336382603113465,
      "loss": 0.0681,
      "theoretical_loss": 3.476208976910552,
      "tokens_seen": 1883242496
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004335580163697641,
      "loss": 0.0688,
      "theoretical_loss": 3.476168490994132,
      "tokens_seen": 1883504640
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043347777242818173,
      "loss": 0.0692,
      "theoretical_loss": 3.4761280122895926,
      "tokens_seen": 1883766784
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043339752848659923,
      "loss": 0.0704,
      "theoretical_loss": 3.476087540794645,
      "tokens_seen": 1884028928
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043331728454501685,
      "loss": 0.0708,
      "theoretical_loss": 3.476047076507002,
      "tokens_seen": 1884291072
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043323704060343446,
      "loss": 0.0687,
      "theoretical_loss": 3.4760066194243784,
      "tokens_seen": 1884553216
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.000433156796661852,
      "loss": 0.0729,
      "theoretical_loss": 3.4759661695444892,
      "tokens_seen": 1884815360
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043307655272026963,
      "loss": 0.071,
      "theoretical_loss": 3.47592572686505,
      "tokens_seen": 1885077504
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043299630877868724,
      "loss": 0.071,
      "theoretical_loss": 3.4758852913837783,
      "tokens_seen": 1885339648
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004329160648371048,
      "loss": 0.0671,
      "theoretical_loss": 3.475844863098393,
      "tokens_seen": 1885601792
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043283582089552236,
      "loss": 0.0684,
      "theoretical_loss": 3.475804442006612,
      "tokens_seen": 1885863936
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043275557695393997,
      "loss": 0.0718,
      "theoretical_loss": 3.4757640281061573,
      "tokens_seen": 1886126080
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004326753330123576,
      "loss": 0.0693,
      "theoretical_loss": 3.4757236213947484,
      "tokens_seen": 1886388224
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043259508907077514,
      "loss": 0.0689,
      "theoretical_loss": 3.4756832218701095,
      "tokens_seen": 1886650368
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043251484512919276,
      "loss": 0.0684,
      "theoretical_loss": 3.475642829529963,
      "tokens_seen": 1886912512
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043243460118761037,
      "loss": 0.0686,
      "theoretical_loss": 3.4756024443720337,
      "tokens_seen": 1887174656
    },
    {
      "epoch": 0.57,
      "objective/train/advantage_avg": -0.0003805331070907414,
      "objective/train/docs_used": 688313,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3872874975204468,
      "objective/train/original_loss": 1.3872874975204468,
      "objective/train/theoretical_loss": 3.4755620663940467,
      "objective/train/tokens_used": 1907896800,
      "objective/train/value_avg": -0.00792694091796875,
      "objective/train/value_loss": 0.0004488099948503077,
      "objective/train/value_max": -4.988908767700195e-05,
      "objective/train/value_min": -0.88671875,
      "objective/train/value_reward_corr": 0.72679213095778,
      "objective/train/value_std": 0.021087646484375,
      "objective/train/weight_avg": 0.9998147487640381,
      "objective/train/weighted_lm_loss": 1.3864668607711792,
      "objective/train/weights_max": 1.4310088157653809,
      "objective/train/weights_min": 0.3778928816318512,
      "theoretical_loss": 3.4755620663940467,
      "tokens_seen": 1887436800
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043235435724602793,
      "loss": 0.0692,
      "theoretical_loss": 3.4755620663940467,
      "tokens_seen": 1887436800
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043227411330444554,
      "loss": 0.0712,
      "theoretical_loss": 3.4755216955937294,
      "tokens_seen": 1887698944
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004321938693628631,
      "loss": 0.0672,
      "theoretical_loss": 3.4754813319688087,
      "tokens_seen": 1887961088
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004321136254212807,
      "loss": 0.0683,
      "theoretical_loss": 3.4754409755170133,
      "tokens_seen": 1888223232
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043203338147969827,
      "loss": 0.0685,
      "theoretical_loss": 3.4754006262360733,
      "tokens_seen": 1888485376
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004319531375381159,
      "loss": 0.0638,
      "theoretical_loss": 3.4753602841237186,
      "tokens_seen": 1888747520
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004318728935965335,
      "loss": 0.0671,
      "theoretical_loss": 3.4753199491776816,
      "tokens_seen": 1889009664
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043179264965495105,
      "loss": 0.0673,
      "theoretical_loss": 3.475279621395695,
      "tokens_seen": 1889271808
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043171240571336867,
      "loss": 0.068,
      "theoretical_loss": 3.4752393007754923,
      "tokens_seen": 1889533952
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004316321617717862,
      "loss": 0.0694,
      "theoretical_loss": 3.4751989873148084,
      "tokens_seen": 1889796096
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043155191783020384,
      "loss": 0.0716,
      "theoretical_loss": 3.4751586810113793,
      "tokens_seen": 1890058240
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004314716738886214,
      "loss": 0.0668,
      "theoretical_loss": 3.4751183818629414,
      "tokens_seen": 1890320384
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.000431391429947039,
      "loss": 0.069,
      "theoretical_loss": 3.4750780898672335,
      "tokens_seen": 1890582528
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004313111860054566,
      "loss": 0.0654,
      "theoretical_loss": 3.475037805021994,
      "tokens_seen": 1890844672
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004312309420638742,
      "loss": 0.0672,
      "theoretical_loss": 3.4749975273249625,
      "tokens_seen": 1891106816
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004311506981222918,
      "loss": 0.0695,
      "theoretical_loss": 3.4749572567738807,
      "tokens_seen": 1891368960
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004310704541807094,
      "loss": 0.0693,
      "theoretical_loss": 3.4749169933664903,
      "tokens_seen": 1891631104
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004309902102391269,
      "loss": 0.0671,
      "theoretical_loss": 3.4748767371005345,
      "tokens_seen": 1891893248
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004309099662975445,
      "loss": 0.0684,
      "theoretical_loss": 3.474836487973757,
      "tokens_seen": 1892155392
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043082972235596213,
      "loss": 0.0699,
      "theoretical_loss": 3.474796245983903,
      "tokens_seen": 1892417536
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043074947841437975,
      "loss": 0.0703,
      "theoretical_loss": 3.4747560111287195,
      "tokens_seen": 1892679680
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004306692344727973,
      "loss": 0.0673,
      "theoretical_loss": 3.4747157834059523,
      "tokens_seen": 1892941824
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004305889905312149,
      "loss": 0.0679,
      "theoretical_loss": 3.47467556281335,
      "tokens_seen": 1893203968
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043050874658963253,
      "loss": 0.0699,
      "theoretical_loss": 3.474635349348662,
      "tokens_seen": 1893466112
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043042850264805003,
      "loss": 0.0683,
      "theoretical_loss": 3.4745951430096387,
      "tokens_seen": 1893728256
    },
    {
      "epoch": 0.57,
      "objective/train/advantage_avg": 0.0003969900426454842,
      "objective/train/docs_used": 690600,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.344144344329834,
      "objective/train/original_loss": 1.344144344329834,
      "objective/train/theoretical_loss": 3.474554943794031,
      "objective/train/tokens_used": 1914450400,
      "objective/train/value_avg": -0.005825042724609375,
      "objective/train/value_loss": 0.00022731033095624298,
      "objective/train/value_max": -3.7610530853271484e-05,
      "objective/train/value_min": -0.96142578125,
      "objective/train/value_reward_corr": 0.781940150301403,
      "objective/train/value_std": 0.0180816650390625,
      "objective/train/weight_avg": 1.0004974603652954,
      "objective/train/weighted_lm_loss": 1.3450180292129517,
      "objective/train/weights_max": 1.4332808256149292,
      "objective/train/weights_min": 0.3768678903579712,
      "theoretical_loss": 3.474554943794031,
      "tokens_seen": 1893990400
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043034825870646765,
      "loss": 0.0688,
      "theoretical_loss": 3.474554943794031,
      "tokens_seen": 1893990400
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043026801476488526,
      "loss": 0.0697,
      "theoretical_loss": 3.4745147516995916,
      "tokens_seen": 1894252544
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043018777082330287,
      "loss": 0.0688,
      "theoretical_loss": 3.474474566724073,
      "tokens_seen": 1894514688
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043010752688172043,
      "loss": 0.0692,
      "theoretical_loss": 3.4744343888652303,
      "tokens_seen": 1894776832
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00043002728294013804,
      "loss": 0.0664,
      "theoretical_loss": 3.474394218120818,
      "tokens_seen": 1895038976
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00042994703899855565,
      "loss": 0.0689,
      "theoretical_loss": 3.474354054488593,
      "tokens_seen": 1895301120
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00042986679505697316,
      "loss": 0.0679,
      "theoretical_loss": 3.4743138979663133,
      "tokens_seen": 1895563264
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00042978655111539077,
      "loss": 0.072,
      "theoretical_loss": 3.4742737485517354,
      "tokens_seen": 1895825408
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004297063071738084,
      "loss": 0.0683,
      "theoretical_loss": 3.4742336062426205,
      "tokens_seen": 1896087552
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.000429626063232226,
      "loss": 0.0682,
      "theoretical_loss": 3.4741934710367284,
      "tokens_seen": 1896349696
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00042954581929064356,
      "loss": 0.0695,
      "theoretical_loss": 3.47415334293182,
      "tokens_seen": 1896611840
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00042946557534906117,
      "loss": 0.0684,
      "theoretical_loss": 3.4741132219256583,
      "tokens_seen": 1896873984
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.0004293853314074788,
      "loss": 0.0688,
      "theoretical_loss": 3.4740731080160066,
      "tokens_seen": 1897136128
    },
    {
      "epoch": 0.57,
      "learning_rate": 0.00042930508746589634,
      "loss": 0.0718,
      "theoretical_loss": 3.4740330012006293,
      "tokens_seen": 1897398272
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004292248435243139,
      "loss": 0.0682,
      "theoretical_loss": 3.473992901477292,
      "tokens_seen": 1897660416
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004291445995827315,
      "loss": 0.0673,
      "theoretical_loss": 3.4739528088437606,
      "tokens_seen": 1897922560
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004290643556411491,
      "loss": 0.0673,
      "theoretical_loss": 3.4739127232978033,
      "tokens_seen": 1898184704
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004289841116995667,
      "loss": 0.0657,
      "theoretical_loss": 3.473872644837189,
      "tokens_seen": 1898446848
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004289038677579843,
      "loss": 0.0703,
      "theoretical_loss": 3.4738325734596858,
      "tokens_seen": 1898708992
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004288236238164019,
      "loss": 0.0652,
      "theoretical_loss": 3.473792509163066,
      "tokens_seen": 1898971136
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042874337987481946,
      "loss": 0.068,
      "theoretical_loss": 3.4737524519450993,
      "tokens_seen": 1899233280
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.000428663135933237,
      "loss": 0.068,
      "theoretical_loss": 3.4737124018035597,
      "tokens_seen": 1899495424
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042858289199165464,
      "loss": 0.0677,
      "theoretical_loss": 3.47367235873622,
      "tokens_seen": 1899757568
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004285026480500722,
      "loss": 0.0677,
      "theoretical_loss": 3.4736323227408548,
      "tokens_seen": 1900019712
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004284224041084898,
      "loss": 0.0687,
      "theoretical_loss": 3.4735922938152397,
      "tokens_seen": 1900281856
    },
    {
      "epoch": 0.58,
      "objective/train/advantage_avg": 0.000148945051478222,
      "objective/train/docs_used": 693007,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3303676843643188,
      "objective/train/original_loss": 1.3303678035736084,
      "objective/train/theoretical_loss": 3.4735522719571517,
      "objective/train/tokens_used": 1921004000,
      "objective/train/value_avg": -0.00881195068359375,
      "objective/train/value_loss": 0.0003545716463122517,
      "objective/train/value_max": -6.109476089477539e-05,
      "objective/train/value_min": -0.9736328125,
      "objective/train/value_reward_corr": 0.7112460178861546,
      "objective/train/value_std": 0.018310546875,
      "objective/train/weight_avg": 1.0003077983856201,
      "objective/train/weighted_lm_loss": 1.331317663192749,
      "objective/train/weights_max": 1.3738635778427124,
      "objective/train/weights_min": 0.3704579770565033,
      "theoretical_loss": 3.4735522719571517,
      "tokens_seen": 1900544000
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004283421601669074,
      "loss": 0.068,
      "theoretical_loss": 3.4735522719571517,
      "tokens_seen": 1900544000
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042826191622532503,
      "loss": 0.0671,
      "theoretical_loss": 3.473512257164368,
      "tokens_seen": 1900806144
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004281816722837426,
      "loss": 0.068,
      "theoretical_loss": 3.4734722494346673,
      "tokens_seen": 1901068288
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042810142834216015,
      "loss": 0.0685,
      "theoretical_loss": 3.473432248765829,
      "tokens_seen": 1901330432
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042802118440057776,
      "loss": 0.0694,
      "theoretical_loss": 3.473392255155634,
      "tokens_seen": 1901592576
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004279409404589953,
      "loss": 0.0694,
      "theoretical_loss": 3.4733522686018636,
      "tokens_seen": 1901854720
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042786069651741293,
      "loss": 0.0696,
      "theoretical_loss": 3.4733122891023007,
      "tokens_seen": 1902116864
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042778045257583054,
      "loss": 0.0681,
      "theoretical_loss": 3.473272316654729,
      "tokens_seen": 1902379008
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042770020863424816,
      "loss": 0.0712,
      "theoretical_loss": 3.4732323512569323,
      "tokens_seen": 1902641152
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004276199646926657,
      "loss": 0.068,
      "theoretical_loss": 3.473192392906697,
      "tokens_seen": 1902903296
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042753972075108333,
      "loss": 0.0689,
      "theoretical_loss": 3.473152441601809,
      "tokens_seen": 1903165440
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004274594768095009,
      "loss": 0.0723,
      "theoretical_loss": 3.473112497340057,
      "tokens_seen": 1903427584
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042737923286791845,
      "loss": 0.0686,
      "theoretical_loss": 3.473072560119229,
      "tokens_seen": 1903689728
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042729898892633606,
      "loss": 0.0701,
      "theoretical_loss": 3.473032629937114,
      "tokens_seen": 1903951872
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042721874498475367,
      "loss": 0.068,
      "theoretical_loss": 3.472992706791504,
      "tokens_seen": 1904214016
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004271385010431713,
      "loss": 0.0688,
      "theoretical_loss": 3.472952790680189,
      "tokens_seen": 1904476160
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042705825710158884,
      "loss": 0.0689,
      "theoretical_loss": 3.472912881600963,
      "tokens_seen": 1904738304
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042697801316000645,
      "loss": 0.0688,
      "theoretical_loss": 3.4728729795516182,
      "tokens_seen": 1905000448
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.000426897769218424,
      "loss": 0.0697,
      "theoretical_loss": 3.4728330845299507,
      "tokens_seen": 1905262592
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042681752527684157,
      "loss": 0.0694,
      "theoretical_loss": 3.472793196533755,
      "tokens_seen": 1905524736
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004267372813352592,
      "loss": 0.0704,
      "theoretical_loss": 3.472753315560828,
      "tokens_seen": 1905786880
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004266570373936768,
      "loss": 0.0673,
      "theoretical_loss": 3.4727134416089678,
      "tokens_seen": 1906049024
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042657679345209435,
      "loss": 0.0713,
      "theoretical_loss": 3.472673574675972,
      "tokens_seen": 1906311168
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042649654951051197,
      "loss": 0.0667,
      "theoretical_loss": 3.472633714759641,
      "tokens_seen": 1906573312
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004264163055689296,
      "loss": 0.0695,
      "theoretical_loss": 3.4725938618577743,
      "tokens_seen": 1906835456
    },
    {
      "epoch": 0.58,
      "objective/train/advantage_avg": 0.0010231251362711191,
      "objective/train/docs_used": 695522,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.331170678138733,
      "objective/train/original_loss": 1.3311705589294434,
      "objective/train/theoretical_loss": 3.4725540159681745,
      "objective/train/tokens_used": 1927557600,
      "objective/train/value_avg": -0.0065460205078125,
      "objective/train/value_loss": 0.0002442924596834928,
      "objective/train/value_max": -6.35385513305664e-05,
      "objective/train/value_min": -0.9287109375,
      "objective/train/value_reward_corr": 0.7102165433933779,
      "objective/train/value_std": 0.016693115234375,
      "objective/train/weight_avg": 1.0011364221572876,
      "objective/train/weighted_lm_loss": 1.3334146738052368,
      "objective/train/weights_max": 1.9407720565795898,
      "objective/train/weights_min": 0.368646502494812,
      "theoretical_loss": 3.4725540159681745,
      "tokens_seen": 1907097600
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004263360616273472,
      "loss": 0.069,
      "theoretical_loss": 3.4725540159681745,
      "tokens_seen": 1907097600
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004262558176857647,
      "loss": 0.0669,
      "theoretical_loss": 3.4725141770886436,
      "tokens_seen": 1907359744
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004261755737441823,
      "loss": 0.0666,
      "theoretical_loss": 3.4724743452169857,
      "tokens_seen": 1907621888
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004260953298025999,
      "loss": 0.0678,
      "theoretical_loss": 3.4724345203510047,
      "tokens_seen": 1907884032
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004260150858610175,
      "loss": 0.0679,
      "theoretical_loss": 3.472394702488506,
      "tokens_seen": 1908146176
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004259348419194351,
      "loss": 0.0672,
      "theoretical_loss": 3.4723548916272966,
      "tokens_seen": 1908408320
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004258545979778527,
      "loss": 0.0693,
      "theoretical_loss": 3.4723150877651836,
      "tokens_seen": 1908670464
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004257743540362703,
      "loss": 0.0681,
      "theoretical_loss": 3.472275290899976,
      "tokens_seen": 1908932608
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004256941100946878,
      "loss": 0.0689,
      "theoretical_loss": 3.472235501029483,
      "tokens_seen": 1909194752
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042561386615310543,
      "loss": 0.0699,
      "theoretical_loss": 3.4721957181515144,
      "tokens_seen": 1909456896
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042553362221152305,
      "loss": 0.0694,
      "theoretical_loss": 3.472155942263883,
      "tokens_seen": 1909719040
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004254533782699406,
      "loss": 0.07,
      "theoretical_loss": 3.4721161733643995,
      "tokens_seen": 1909981184
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004253731343283582,
      "loss": 0.0675,
      "theoretical_loss": 3.472076411450878,
      "tokens_seen": 1910243328
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042529289038677583,
      "loss": 0.0679,
      "theoretical_loss": 3.472036656521134,
      "tokens_seen": 1910505472
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042521264644519344,
      "loss": 0.0662,
      "theoretical_loss": 3.4719969085729816,
      "tokens_seen": 1910767616
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042513240250361095,
      "loss": 0.0717,
      "theoretical_loss": 3.4719571676042373,
      "tokens_seen": 1911029760
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042505215856202856,
      "loss": 0.0708,
      "theoretical_loss": 3.4719174336127185,
      "tokens_seen": 1911291904
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004249719146204462,
      "loss": 0.0691,
      "theoretical_loss": 3.471877706596244,
      "tokens_seen": 1911554048
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042489167067886373,
      "loss": 0.0669,
      "theoretical_loss": 3.4718379865526323,
      "tokens_seen": 1911816192
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042481142673728134,
      "loss": 0.0705,
      "theoretical_loss": 3.4717982734797044,
      "tokens_seen": 1912078336
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042473118279569896,
      "loss": 0.0695,
      "theoretical_loss": 3.4717585673752813,
      "tokens_seen": 1912340480
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004246509388541165,
      "loss": 0.0692,
      "theoretical_loss": 3.4717188682371853,
      "tokens_seen": 1912602624
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042457069491253413,
      "loss": 0.0678,
      "theoretical_loss": 3.471679176063239,
      "tokens_seen": 1912864768
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004244904509709517,
      "loss": 0.0682,
      "theoretical_loss": 3.4716394908512678,
      "tokens_seen": 1913126912
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004244102070293693,
      "loss": 0.0661,
      "theoretical_loss": 3.471599812599095,
      "tokens_seen": 1913389056
    },
    {
      "epoch": 0.58,
      "objective/train/advantage_avg": 0.0013305656611919403,
      "objective/train/docs_used": 697951,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4070523977279663,
      "objective/train/original_loss": 1.4070522785186768,
      "objective/train/theoretical_loss": 3.4715601413045487,
      "objective/train/tokens_used": 1934111200,
      "objective/train/value_avg": -0.01123046875,
      "objective/train/value_loss": 0.0003065956407226622,
      "objective/train/value_max": -2.485513687133789e-05,
      "objective/train/value_min": -0.900390625,
      "objective/train/value_reward_corr": 0.7024227752945549,
      "objective/train/value_std": 0.018707275390625,
      "objective/train/weight_avg": 1.0014679431915283,
      "objective/train/weighted_lm_loss": 1.4086270332336426,
      "objective/train/weights_max": 2.3405609130859375,
      "objective/train/weights_min": 0.2271009087562561,
      "theoretical_loss": 3.4715601413045487,
      "tokens_seen": 1913651200
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042432996308778686,
      "loss": 0.0712,
      "theoretical_loss": 3.4715601413045487,
      "tokens_seen": 1913651200
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042424971914620447,
      "loss": 0.0687,
      "theoretical_loss": 3.4715204769654555,
      "tokens_seen": 1913913344
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004241694752046221,
      "loss": 0.0694,
      "theoretical_loss": 3.471480819579643,
      "tokens_seen": 1914175488
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042408923126303964,
      "loss": 0.0684,
      "theoretical_loss": 3.4714411691449403,
      "tokens_seen": 1914437632
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042400898732145725,
      "loss": 0.0674,
      "theoretical_loss": 3.4714015256591777,
      "tokens_seen": 1914699776
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004239287433798748,
      "loss": 0.0716,
      "theoretical_loss": 3.4713618891201863,
      "tokens_seen": 1914961920
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004238484994382924,
      "loss": 0.0694,
      "theoretical_loss": 3.471322259525798,
      "tokens_seen": 1915224064
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042376825549671,
      "loss": 0.0696,
      "theoretical_loss": 3.4712826368738456,
      "tokens_seen": 1915486208
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004236880115551276,
      "loss": 0.0684,
      "theoretical_loss": 3.471243021162163,
      "tokens_seen": 1915748352
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004236077676135452,
      "loss": 0.0675,
      "theoretical_loss": 3.4712034123885855,
      "tokens_seen": 1916010496
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042352752367196277,
      "loss": 0.0661,
      "theoretical_loss": 3.471163810550949,
      "tokens_seen": 1916272640
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004234472797303804,
      "loss": 0.0696,
      "theoretical_loss": 3.47112421564709,
      "tokens_seen": 1916534784
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042336703578879794,
      "loss": 0.0673,
      "theoretical_loss": 3.4710846276748466,
      "tokens_seen": 1916796928
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042328679184721555,
      "loss": 0.0682,
      "theoretical_loss": 3.4710450466320575,
      "tokens_seen": 1917059072
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004232065479056331,
      "loss": 0.0677,
      "theoretical_loss": 3.471005472516562,
      "tokens_seen": 1917321216
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004231263039640507,
      "loss": 0.0687,
      "theoretical_loss": 3.4709659053262016,
      "tokens_seen": 1917583360
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042304606002246833,
      "loss": 0.0713,
      "theoretical_loss": 3.470926345058818,
      "tokens_seen": 1917845504
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004229658160808859,
      "loss": 0.0666,
      "theoretical_loss": 3.4708867917122532,
      "tokens_seen": 1918107648
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004228855721393035,
      "loss": 0.0696,
      "theoretical_loss": 3.470847245284351,
      "tokens_seen": 1918369792
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004228053281977211,
      "loss": 0.067,
      "theoretical_loss": 3.4708077057729567,
      "tokens_seen": 1918631936
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004227250842561386,
      "loss": 0.069,
      "theoretical_loss": 3.4707681731759155,
      "tokens_seen": 1918894080
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042264484031455623,
      "loss": 0.0679,
      "theoretical_loss": 3.4707286474910735,
      "tokens_seen": 1919156224
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042256459637297385,
      "loss": 0.0665,
      "theoretical_loss": 3.4706891287162787,
      "tokens_seen": 1919418368
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042248435243139146,
      "loss": 0.0691,
      "theoretical_loss": 3.470649616849379,
      "tokens_seen": 1919680512
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.000422404108489809,
      "loss": 0.0683,
      "theoretical_loss": 3.4706101118882247,
      "tokens_seen": 1919942656
    },
    {
      "epoch": 0.58,
      "objective/train/advantage_avg": 0.0006141560152173042,
      "objective/train/docs_used": 700307,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4711885452270508,
      "objective/train/original_loss": 1.4711885452270508,
      "objective/train/theoretical_loss": 3.4705706138306653,
      "objective/train/tokens_used": 1940664800,
      "objective/train/value_avg": -0.005828857421875,
      "objective/train/value_loss": 0.0001746070192893967,
      "objective/train/value_max": -5.561113357543945e-05,
      "objective/train/value_min": -0.26611328125,
      "objective/train/value_reward_corr": 0.5963098498497844,
      "objective/train/value_std": 0.009674072265625,
      "objective/train/weight_avg": 1.0006877183914185,
      "objective/train/weighted_lm_loss": 1.4726924896240234,
      "objective/train/weights_max": 1.165907382965088,
      "objective/train/weights_min": 0.2251291126012802,
      "theoretical_loss": 3.4705706138306653,
      "tokens_seen": 1920204800
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042232386454822663,
      "loss": 0.0705,
      "theoretical_loss": 3.4705706138306653,
      "tokens_seen": 1920204800
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042224362060664424,
      "loss": 0.0686,
      "theoretical_loss": 3.4705311226745525,
      "tokens_seen": 1920466944
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042216337666506175,
      "loss": 0.0679,
      "theoretical_loss": 3.470491638417739,
      "tokens_seen": 1920729088
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042208313272347936,
      "loss": 0.0665,
      "theoretical_loss": 3.470452161058078,
      "tokens_seen": 1920991232
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042200288878189697,
      "loss": 0.0694,
      "theoretical_loss": 3.470412690593423,
      "tokens_seen": 1921253376
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004219226448403146,
      "loss": 0.0665,
      "theoretical_loss": 3.4703732270216303,
      "tokens_seen": 1921515520
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042184240089873214,
      "loss": 0.0671,
      "theoretical_loss": 3.470333770340555,
      "tokens_seen": 1921777664
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042176215695714976,
      "loss": 0.0682,
      "theoretical_loss": 3.4702943205480548,
      "tokens_seen": 1922039808
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042168191301556737,
      "loss": 0.0678,
      "theoretical_loss": 3.470254877641988,
      "tokens_seen": 1922301952
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042160166907398493,
      "loss": 0.0687,
      "theoretical_loss": 3.470215441620213,
      "tokens_seen": 1922564096
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004215214251324025,
      "loss": 0.0715,
      "theoretical_loss": 3.47017601248059,
      "tokens_seen": 1922826240
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004214411811908201,
      "loss": 0.0679,
      "theoretical_loss": 3.4701365902209798,
      "tokens_seen": 1923088384
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004213609372492377,
      "loss": 0.0708,
      "theoretical_loss": 3.4700971748392453,
      "tokens_seen": 1923350528
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042128069330765527,
      "loss": 0.0678,
      "theoretical_loss": 3.470057766333248,
      "tokens_seen": 1923612672
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004212004493660729,
      "loss": 0.0706,
      "theoretical_loss": 3.4700183647008522,
      "tokens_seen": 1923874816
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004211202054244905,
      "loss": 0.0676,
      "theoretical_loss": 3.4699789699399233,
      "tokens_seen": 1924136960
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042103996148290805,
      "loss": 0.0661,
      "theoretical_loss": 3.469939582048326,
      "tokens_seen": 1924399104
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004209597175413256,
      "loss": 0.0676,
      "theoretical_loss": 3.469900201023928,
      "tokens_seen": 1924661248
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004208794735997432,
      "loss": 0.0651,
      "theoretical_loss": 3.469860826864596,
      "tokens_seen": 1924923392
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004207992296581608,
      "loss": 0.0689,
      "theoretical_loss": 3.4698214595681995,
      "tokens_seen": 1925185536
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004207189857165784,
      "loss": 0.0677,
      "theoretical_loss": 3.4697820991326074,
      "tokens_seen": 1925447680
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.000420638741774996,
      "loss": 0.0658,
      "theoretical_loss": 3.46974274555569,
      "tokens_seen": 1925709824
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004205584978334136,
      "loss": 0.0701,
      "theoretical_loss": 3.4697033988353194,
      "tokens_seen": 1925971968
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004204782538918312,
      "loss": 0.0666,
      "theoretical_loss": 3.4696640589693675,
      "tokens_seen": 1926234112
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042039800995024874,
      "loss": 0.0681,
      "theoretical_loss": 3.4696247259557076,
      "tokens_seen": 1926496256
    },
    {
      "epoch": 0.58,
      "objective/train/advantage_avg": 0.000970550230704248,
      "objective/train/docs_used": 702709,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4049184322357178,
      "objective/train/original_loss": 1.4049184322357178,
      "objective/train/theoretical_loss": 3.469585399792215,
      "objective/train/tokens_used": 1947218400,
      "objective/train/value_avg": -0.0082550048828125,
      "objective/train/value_loss": 0.00031141238287091255,
      "objective/train/value_max": -4.2319297790527344e-05,
      "objective/train/value_min": -0.669921875,
      "objective/train/value_reward_corr": 0.6947720833693894,
      "objective/train/value_std": 0.017974853515625,
      "objective/train/weight_avg": 1.0011149644851685,
      "objective/train/weighted_lm_loss": 1.406195044517517,
      "objective/train/weights_max": 1.8169658184051514,
      "objective/train/weights_min": 0.3840899169445038,
      "theoretical_loss": 3.469585399792215,
      "tokens_seen": 1926758400
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042031776600866635,
      "loss": 0.0706,
      "theoretical_loss": 3.469585399792215,
      "tokens_seen": 1926758400
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004202375220670839,
      "loss": 0.0666,
      "theoretical_loss": 3.4695460804767633,
      "tokens_seen": 1927020544
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004201572781255015,
      "loss": 0.0694,
      "theoretical_loss": 3.4695067680072293,
      "tokens_seen": 1927282688
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00042007703418391913,
      "loss": 0.0709,
      "theoretical_loss": 3.469467462381491,
      "tokens_seen": 1927544832
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00041999679024233675,
      "loss": 0.0682,
      "theoretical_loss": 3.469428163597426,
      "tokens_seen": 1927806976
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004199165463007543,
      "loss": 0.0676,
      "theoretical_loss": 3.4693888716529124,
      "tokens_seen": 1928069120
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004198363023591719,
      "loss": 0.0706,
      "theoretical_loss": 3.4693495865458313,
      "tokens_seen": 1928331264
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.0004197560584175895,
      "loss": 0.068,
      "theoretical_loss": 3.4693103082740633,
      "tokens_seen": 1928593408
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00041967581447600703,
      "loss": 0.0717,
      "theoretical_loss": 3.4692710368354898,
      "tokens_seen": 1928855552
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00041959557053442465,
      "loss": 0.0682,
      "theoretical_loss": 3.469231772227994,
      "tokens_seen": 1929117696
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00041951532659284226,
      "loss": 0.0667,
      "theoretical_loss": 3.4691925144494604,
      "tokens_seen": 1929379840
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00041943508265125987,
      "loss": 0.0715,
      "theoretical_loss": 3.4691532634977724,
      "tokens_seen": 1929641984
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00041935483870967743,
      "loss": 0.0689,
      "theoretical_loss": 3.469114019370816,
      "tokens_seen": 1929904128
    },
    {
      "epoch": 0.58,
      "learning_rate": 0.00041927459476809504,
      "loss": 0.0686,
      "theoretical_loss": 3.4690747820664782,
      "tokens_seen": 1930166272
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004191943508265126,
      "loss": 0.0673,
      "theoretical_loss": 3.469035551582646,
      "tokens_seen": 1930428416
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041911410688493016,
      "loss": 0.068,
      "theoretical_loss": 3.468996327917208,
      "tokens_seen": 1930690560
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041903386294334777,
      "loss": 0.0695,
      "theoretical_loss": 3.468957111068054,
      "tokens_seen": 1930952704
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004189536190017654,
      "loss": 0.067,
      "theoretical_loss": 3.468917901033074,
      "tokens_seen": 1931214848
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.000418873375060183,
      "loss": 0.0694,
      "theoretical_loss": 3.468878697810159,
      "tokens_seen": 1931476992
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041879313111860056,
      "loss": 0.0699,
      "theoretical_loss": 3.468839501397202,
      "tokens_seen": 1931739136
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041871288717701817,
      "loss": 0.0701,
      "theoretical_loss": 3.4688003117920956,
      "tokens_seen": 1932001280
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004186326432354357,
      "loss": 0.0673,
      "theoretical_loss": 3.4687611289927336,
      "tokens_seen": 1932263424
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004185523992938533,
      "loss": 0.0684,
      "theoretical_loss": 3.468721952997012,
      "tokens_seen": 1932525568
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004184721553522709,
      "loss": 0.0683,
      "theoretical_loss": 3.4686827838028256,
      "tokens_seen": 1932787712
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004183919114106885,
      "loss": 0.0693,
      "theoretical_loss": 3.4686436214080727,
      "tokens_seen": 1933049856
    },
    {
      "epoch": 0.59,
      "objective/train/advantage_avg": 0.0004142014076933265,
      "objective/train/docs_used": 705162,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3592551946640015,
      "objective/train/original_loss": 1.3592551946640015,
      "objective/train/theoretical_loss": 3.4686044658106496,
      "objective/train/tokens_used": 1953772000,
      "objective/train/value_avg": -0.00897216796875,
      "objective/train/value_loss": 0.00020762467465829104,
      "objective/train/value_max": -5.227327346801758e-05,
      "objective/train/value_min": -0.3525390625,
      "objective/train/value_reward_corr": 0.7716059997343598,
      "objective/train/value_std": 0.016326904296875,
      "objective/train/weight_avg": 1.0005117654800415,
      "objective/train/weighted_lm_loss": 1.3597215414047241,
      "objective/train/weights_max": 1.2099170684814453,
      "objective/train/weights_min": 0.36951375007629395,
      "theoretical_loss": 3.4686044658106496,
      "tokens_seen": 1933312000
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041831166746910607,
      "loss": 0.0663,
      "theoretical_loss": 3.4686044658106496,
      "tokens_seen": 1933312000
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004182314235275237,
      "loss": 0.0669,
      "theoretical_loss": 3.468565317008456,
      "tokens_seen": 1933574144
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004181511795859413,
      "loss": 0.0717,
      "theoretical_loss": 3.4685261749993908,
      "tokens_seen": 1933836288
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004180709356443589,
      "loss": 0.0658,
      "theoretical_loss": 3.4684870397813556,
      "tokens_seen": 1934098432
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004179906917027764,
      "loss": 0.0698,
      "theoretical_loss": 3.4684479113522517,
      "tokens_seen": 1934360576
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.000417910447761194,
      "loss": 0.0688,
      "theoretical_loss": 3.4684087897099816,
      "tokens_seen": 1934622720
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041783020381961164,
      "loss": 0.0683,
      "theoretical_loss": 3.4683696748524486,
      "tokens_seen": 1934884864
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004177499598780292,
      "loss": 0.0709,
      "theoretical_loss": 3.468330566777557,
      "tokens_seen": 1935147008
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004176697159364468,
      "loss": 0.0676,
      "theoretical_loss": 3.4682914654832118,
      "tokens_seen": 1935409152
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004175894719948644,
      "loss": 0.0651,
      "theoretical_loss": 3.46825237096732,
      "tokens_seen": 1935671296
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041750922805328203,
      "loss": 0.0671,
      "theoretical_loss": 3.4682132832277883,
      "tokens_seen": 1935933440
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041742898411169954,
      "loss": 0.0691,
      "theoretical_loss": 3.4681742022625253,
      "tokens_seen": 1936195584
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041734874017011715,
      "loss": 0.0735,
      "theoretical_loss": 3.468135128069439,
      "tokens_seen": 1936457728
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041726849622853476,
      "loss": 0.0689,
      "theoretical_loss": 3.4680960606464404,
      "tokens_seen": 1936719872
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004171882522869523,
      "loss": 0.0692,
      "theoretical_loss": 3.4680569999914397,
      "tokens_seen": 1936982016
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041710800834536993,
      "loss": 0.0677,
      "theoretical_loss": 3.468017946102349,
      "tokens_seen": 1937244160
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041702776440378755,
      "loss": 0.0692,
      "theoretical_loss": 3.467978898977081,
      "tokens_seen": 1937506304
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041694752046220516,
      "loss": 0.0699,
      "theoretical_loss": 3.4679398586135486,
      "tokens_seen": 1937768448
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004168672765206227,
      "loss": 0.0679,
      "theoretical_loss": 3.467900825009668,
      "tokens_seen": 1938030592
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004167870325790403,
      "loss": 0.0695,
      "theoretical_loss": 3.4678617981633533,
      "tokens_seen": 1938292736
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004167067886374579,
      "loss": 0.0683,
      "theoretical_loss": 3.467822778072521,
      "tokens_seen": 1938554880
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041662654469587545,
      "loss": 0.0677,
      "theoretical_loss": 3.4677837647350898,
      "tokens_seen": 1938817024
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041654630075429306,
      "loss": 0.068,
      "theoretical_loss": 3.467744758148976,
      "tokens_seen": 1939079168
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041646605681271067,
      "loss": 0.0686,
      "theoretical_loss": 3.4677057583121007,
      "tokens_seen": 1939341312
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041638581287112823,
      "loss": 0.0688,
      "theoretical_loss": 3.467666765222383,
      "tokens_seen": 1939603456
    },
    {
      "epoch": 0.59,
      "objective/train/advantage_avg": 0.00022620317758992314,
      "objective/train/docs_used": 707259,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3885008096694946,
      "objective/train/original_loss": 1.388500452041626,
      "objective/train/theoretical_loss": 3.4676277788777443,
      "objective/train/tokens_used": 1960325600,
      "objective/train/value_avg": -0.0106658935546875,
      "objective/train/value_loss": 0.0003325411817058921,
      "objective/train/value_max": -3.045797348022461e-05,
      "objective/train/value_min": -0.95068359375,
      "objective/train/value_reward_corr": 0.6915883477212278,
      "objective/train/value_std": 0.0175933837890625,
      "objective/train/weight_avg": 1.0003694295883179,
      "objective/train/weighted_lm_loss": 1.3884354829788208,
      "objective/train/weights_max": 1.4233535528182983,
      "objective/train/weights_min": 0.23048020899295807,
      "theoretical_loss": 3.4676277788777443,
      "tokens_seen": 1939865600
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041630556892954584,
      "loss": 0.0682,
      "theoretical_loss": 3.4676277788777443,
      "tokens_seen": 1939865600
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004162253249879634,
      "loss": 0.0683,
      "theoretical_loss": 3.467588799276106,
      "tokens_seen": 1940127744
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.000416145081046381,
      "loss": 0.0681,
      "theoretical_loss": 3.4675498264153912,
      "tokens_seen": 1940389888
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041606483710479857,
      "loss": 0.0693,
      "theoretical_loss": 3.4675108602935243,
      "tokens_seen": 1940652032
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004159845931632162,
      "loss": 0.0662,
      "theoretical_loss": 3.4674719009084294,
      "tokens_seen": 1940914176
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004159043492216338,
      "loss": 0.0691,
      "theoretical_loss": 3.4674329482580326,
      "tokens_seen": 1941176320
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041582410528005135,
      "loss": 0.0666,
      "theoretical_loss": 3.46739400234026,
      "tokens_seen": 1941438464
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041574386133846897,
      "loss": 0.0706,
      "theoretical_loss": 3.467355063153039,
      "tokens_seen": 1941700608
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004156636173968865,
      "loss": 0.0647,
      "theoretical_loss": 3.467316130694299,
      "tokens_seen": 1941962752
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041558337345530414,
      "loss": 0.0663,
      "theoretical_loss": 3.467277204961968,
      "tokens_seen": 1942224896
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004155031295137217,
      "loss": 0.0671,
      "theoretical_loss": 3.467238285953977,
      "tokens_seen": 1942487040
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004154228855721393,
      "loss": 0.0696,
      "theoretical_loss": 3.467199373668257,
      "tokens_seen": 1942749184
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004153426416305569,
      "loss": 0.067,
      "theoretical_loss": 3.4671604681027404,
      "tokens_seen": 1943011328
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004152623976889745,
      "loss": 0.0704,
      "theoretical_loss": 3.467121569255359,
      "tokens_seen": 1943273472
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004151821537473921,
      "loss": 0.0679,
      "theoretical_loss": 3.4670826771240484,
      "tokens_seen": 1943535616
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004151019098058097,
      "loss": 0.0711,
      "theoretical_loss": 3.4670437917067423,
      "tokens_seen": 1943797760
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041502166586422726,
      "loss": 0.0687,
      "theoretical_loss": 3.4670049130013765,
      "tokens_seen": 1944059904
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004149414219226448,
      "loss": 0.0657,
      "theoretical_loss": 3.466966041005888,
      "tokens_seen": 1944322048
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041486117798106244,
      "loss": 0.0695,
      "theoretical_loss": 3.4669271757182143,
      "tokens_seen": 1944584192
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041478093403948005,
      "loss": 0.0689,
      "theoretical_loss": 3.466888317136293,
      "tokens_seen": 1944846336
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004147006900978976,
      "loss": 0.0678,
      "theoretical_loss": 3.466849465258065,
      "tokens_seen": 1945108480
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004146204461563152,
      "loss": 0.068,
      "theoretical_loss": 3.4668106200814695,
      "tokens_seen": 1945370624
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041454020221473283,
      "loss": 0.0677,
      "theoretical_loss": 3.466771781604448,
      "tokens_seen": 1945632768
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041445995827315034,
      "loss": 0.0662,
      "theoretical_loss": 3.4667329498249426,
      "tokens_seen": 1945894912
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041437971433156795,
      "loss": 0.0675,
      "theoretical_loss": 3.466694124740896,
      "tokens_seen": 1946157056
    },
    {
      "epoch": 0.59,
      "objective/train/advantage_avg": 0.0005765951937064528,
      "objective/train/docs_used": 709628,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3162634372711182,
      "objective/train/original_loss": 1.316263198852539,
      "objective/train/theoretical_loss": 3.466655306350253,
      "objective/train/tokens_used": 1966879200,
      "objective/train/value_avg": -0.00644683837890625,
      "objective/train/value_loss": 0.0001964610710274428,
      "objective/train/value_max": -6.502866744995117e-05,
      "objective/train/value_min": -0.73486328125,
      "objective/train/value_reward_corr": 0.7131132573437041,
      "objective/train/value_std": 0.01448822021484375,
      "objective/train/weight_avg": 1.0006669759750366,
      "objective/train/weighted_lm_loss": 1.3172144889831543,
      "objective/train/weights_max": 1.8098821640014648,
      "objective/train/weights_min": 0.3702559173107147,
      "theoretical_loss": 3.466655306350253,
      "tokens_seen": 1946419200
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041429947038998556,
      "loss": 0.0682,
      "theoretical_loss": 3.466655306350253,
      "tokens_seen": 1946419200
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004142192264484032,
      "loss": 0.0679,
      "theoretical_loss": 3.4666164946509572,
      "tokens_seen": 1946681344
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041413898250682073,
      "loss": 0.0697,
      "theoretical_loss": 3.466577689640955,
      "tokens_seen": 1946943488
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041405873856523834,
      "loss": 0.0695,
      "theoretical_loss": 3.4665388913181934,
      "tokens_seen": 1947205632
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041397849462365596,
      "loss": 0.0666,
      "theoretical_loss": 3.4665000996806192,
      "tokens_seen": 1947467776
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041389825068207346,
      "loss": 0.0686,
      "theoretical_loss": 3.466461314726182,
      "tokens_seen": 1947729920
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004138180067404911,
      "loss": 0.0672,
      "theoretical_loss": 3.4664225364528294,
      "tokens_seen": 1947992064
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004137377627989087,
      "loss": 0.0697,
      "theoretical_loss": 3.466383764858513,
      "tokens_seen": 1948254208
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004136575188573263,
      "loss": 0.0689,
      "theoretical_loss": 3.466344999941184,
      "tokens_seen": 1948516352
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041357727491574386,
      "loss": 0.0683,
      "theoretical_loss": 3.4663062416987938,
      "tokens_seen": 1948778496
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041349703097416147,
      "loss": 0.0697,
      "theoretical_loss": 3.4662674901292956,
      "tokens_seen": 1949040640
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004134167870325791,
      "loss": 0.0638,
      "theoretical_loss": 3.4662287452306435,
      "tokens_seen": 1949302784
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041333654309099664,
      "loss": 0.0683,
      "theoretical_loss": 3.466190007000792,
      "tokens_seen": 1949564928
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004132562991494142,
      "loss": 0.0709,
      "theoretical_loss": 3.466151275437697,
      "tokens_seen": 1949827072
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004131760552078318,
      "loss": 0.0659,
      "theoretical_loss": 3.4661125505393153,
      "tokens_seen": 1950089216
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004130958112662494,
      "loss": 0.0688,
      "theoretical_loss": 3.4660738323036036,
      "tokens_seen": 1950351360
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.000413015567324667,
      "loss": 0.0705,
      "theoretical_loss": 3.466035120728521,
      "tokens_seen": 1950613504
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004129353233830846,
      "loss": 0.0669,
      "theoretical_loss": 3.465996415812027,
      "tokens_seen": 1950875648
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004128550794415022,
      "loss": 0.0699,
      "theoretical_loss": 3.4659577175520813,
      "tokens_seen": 1951137792
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041277483549991977,
      "loss": 0.0668,
      "theoretical_loss": 3.4659190259466444,
      "tokens_seen": 1951399936
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004126945915583373,
      "loss": 0.0673,
      "theoretical_loss": 3.46588034099368,
      "tokens_seen": 1951662080
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041261434761675494,
      "loss": 0.0689,
      "theoretical_loss": 3.465841662691149,
      "tokens_seen": 1951924224
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004125341036751725,
      "loss": 0.0672,
      "theoretical_loss": 3.465802991037016,
      "tokens_seen": 1952186368
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004124538597335901,
      "loss": 0.0702,
      "theoretical_loss": 3.4657643260292463,
      "tokens_seen": 1952448512
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004123736157920077,
      "loss": 0.0689,
      "theoretical_loss": 3.4657256676658053,
      "tokens_seen": 1952710656
    },
    {
      "epoch": 0.59,
      "objective/train/advantage_avg": -0.0003757055092137307,
      "objective/train/docs_used": 712021,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3009216785430908,
      "objective/train/original_loss": 1.3009214401245117,
      "objective/train/theoretical_loss": 3.4656870159446584,
      "objective/train/tokens_used": 1973432800,
      "objective/train/value_avg": -0.01009368896484375,
      "objective/train/value_loss": 0.0003471270902082324,
      "objective/train/value_max": -5.608797073364258e-05,
      "objective/train/value_min": -0.74755859375,
      "objective/train/value_reward_corr": 0.8502257457911682,
      "objective/train/value_std": 0.0214691162109375,
      "objective/train/weight_avg": 0.9997867345809937,
      "objective/train/weighted_lm_loss": 1.2977733612060547,
      "objective/train/weights_max": 1.6069636344909668,
      "objective/train/weights_min": 0.40155351161956787,
      "theoretical_loss": 3.4656870159446584,
      "tokens_seen": 1952972800
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041229337185042533,
      "loss": 0.0688,
      "theoretical_loss": 3.4656870159446584,
      "tokens_seen": 1952972800
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004122131279088429,
      "loss": 0.0676,
      "theoretical_loss": 3.465648370863774,
      "tokens_seen": 1953234944
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004121328839672605,
      "loss": 0.0664,
      "theoretical_loss": 3.46560973242112,
      "tokens_seen": 1953497088
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041205264002567806,
      "loss": 0.0695,
      "theoretical_loss": 3.4655711006146657,
      "tokens_seen": 1953759232
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004119723960840956,
      "loss": 0.0701,
      "theoretical_loss": 3.465532475442381,
      "tokens_seen": 1954021376
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041189215214251323,
      "loss": 0.0685,
      "theoretical_loss": 3.4654938569022375,
      "tokens_seen": 1954283520
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041181190820093085,
      "loss": 0.0684,
      "theoretical_loss": 3.4654552449922056,
      "tokens_seen": 1954545664
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041173166425934846,
      "loss": 0.0698,
      "theoretical_loss": 3.4654166397102593,
      "tokens_seen": 1954807808
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.000411651420317766,
      "loss": 0.0693,
      "theoretical_loss": 3.4653780410543717,
      "tokens_seen": 1955069952
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041157117637618363,
      "loss": 0.0687,
      "theoretical_loss": 3.465339449022517,
      "tokens_seen": 1955332096
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004114909324346012,
      "loss": 0.0702,
      "theoretical_loss": 3.4653008636126716,
      "tokens_seen": 1955594240
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041141068849301875,
      "loss": 0.0684,
      "theoretical_loss": 3.4652622848228107,
      "tokens_seen": 1955856384
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041133044455143636,
      "loss": 0.069,
      "theoretical_loss": 3.4652237126509124,
      "tokens_seen": 1956118528
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041125020060985397,
      "loss": 0.0693,
      "theoretical_loss": 3.465185147094954,
      "tokens_seen": 1956380672
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004111699566682716,
      "loss": 0.0695,
      "theoretical_loss": 3.465146588152915,
      "tokens_seen": 1956642816
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041108971272668914,
      "loss": 0.0681,
      "theoretical_loss": 3.465108035822775,
      "tokens_seen": 1956904960
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041100946878510676,
      "loss": 0.0702,
      "theoretical_loss": 3.4650694901025147,
      "tokens_seen": 1957167104
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004109292248435243,
      "loss": 0.0669,
      "theoretical_loss": 3.4650309509901156,
      "tokens_seen": 1957429248
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004108489809019419,
      "loss": 0.0722,
      "theoretical_loss": 3.4649924184835603,
      "tokens_seen": 1957691392
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004107687369603595,
      "loss": 0.0703,
      "theoretical_loss": 3.4649538925808328,
      "tokens_seen": 1957953536
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004106884930187771,
      "loss": 0.0689,
      "theoretical_loss": 3.4649153732799167,
      "tokens_seen": 1958215680
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004106082490771947,
      "loss": 0.0673,
      "theoretical_loss": 3.464876860578797,
      "tokens_seen": 1958477824
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00041052800513561227,
      "loss": 0.0702,
      "theoretical_loss": 3.46483835447546,
      "tokens_seen": 1958739968
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004104477611940299,
      "loss": 0.0675,
      "theoretical_loss": 3.464799854967893,
      "tokens_seen": 1959002112
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004103675172524475,
      "loss": 0.0697,
      "theoretical_loss": 3.464761362054084,
      "tokens_seen": 1959264256
    },
    {
      "epoch": 0.59,
      "objective/train/advantage_avg": 0.00013065687380731106,
      "objective/train/docs_used": 714425,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4167454242706299,
      "objective/train/original_loss": 1.4167454242706299,
      "objective/train/theoretical_loss": 3.4647228757320203,
      "objective/train/tokens_used": 1979986400,
      "objective/train/value_avg": -0.01082611083984375,
      "objective/train/value_loss": 0.0005280160694383085,
      "objective/train/value_max": -7.033348083496094e-05,
      "objective/train/value_min": -0.958984375,
      "objective/train/value_reward_corr": 0.7742440701143263,
      "objective/train/value_std": 0.0279388427734375,
      "objective/train/weight_avg": 1.0003762245178223,
      "objective/train/weighted_lm_loss": 1.4171031713485718,
      "objective/train/weights_max": 2.2862038612365723,
      "objective/train/weights_min": 0.36873859167099,
      "theoretical_loss": 3.4647228757320203,
      "tokens_seen": 1959526400
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.000410287273310865,
      "loss": 0.0681,
      "theoretical_loss": 3.4647228757320203,
      "tokens_seen": 1959526400
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004102070293692826,
      "loss": 0.0697,
      "theoretical_loss": 3.4646843959996927,
      "tokens_seen": 1959788544
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004101267854277002,
      "loss": 0.0676,
      "theoretical_loss": 3.464645922855091,
      "tokens_seen": 1960050688
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004100465414861178,
      "loss": 0.07,
      "theoretical_loss": 3.464607456296207,
      "tokens_seen": 1960312832
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004099662975445354,
      "loss": 0.0708,
      "theoretical_loss": 3.464568996321033,
      "tokens_seen": 1960574976
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.000409886053602953,
      "loss": 0.0704,
      "theoretical_loss": 3.4645305429275624,
      "tokens_seen": 1960837120
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004098058096613706,
      "loss": 0.0695,
      "theoretical_loss": 3.464492096113788,
      "tokens_seen": 1961099264
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004097255657197881,
      "loss": 0.0694,
      "theoretical_loss": 3.4644536558777057,
      "tokens_seen": 1961361408
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00040964532177820574,
      "loss": 0.07,
      "theoretical_loss": 3.4644152222173106,
      "tokens_seen": 1961623552
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00040956507783662335,
      "loss": 0.0688,
      "theoretical_loss": 3.4643767951305997,
      "tokens_seen": 1961885696
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004094848338950409,
      "loss": 0.0666,
      "theoretical_loss": 3.4643383746155703,
      "tokens_seen": 1962147840
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.0004094045899534585,
      "loss": 0.0677,
      "theoretical_loss": 3.464299960670221,
      "tokens_seen": 1962409984
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00040932434601187613,
      "loss": 0.0693,
      "theoretical_loss": 3.464261553292551,
      "tokens_seen": 1962672128
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00040924410207029375,
      "loss": 0.0697,
      "theoretical_loss": 3.4642231524805607,
      "tokens_seen": 1962934272
    },
    {
      "epoch": 0.59,
      "learning_rate": 0.00040916385812871125,
      "loss": 0.0703,
      "theoretical_loss": 3.4641847582322507,
      "tokens_seen": 1963196416
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040908361418712886,
      "loss": 0.0711,
      "theoretical_loss": 3.4641463705456226,
      "tokens_seen": 1963458560
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004090033702455465,
      "loss": 0.0704,
      "theoretical_loss": 3.4641079894186797,
      "tokens_seen": 1963720704
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040892312630396403,
      "loss": 0.0674,
      "theoretical_loss": 3.4640696148494254,
      "tokens_seen": 1963982848
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040884288236238165,
      "loss": 0.068,
      "theoretical_loss": 3.464031246835864,
      "tokens_seen": 1964244992
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040876263842079926,
      "loss": 0.0684,
      "theoretical_loss": 3.4639928853760016,
      "tokens_seen": 1964507136
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040868239447921687,
      "loss": 0.0688,
      "theoretical_loss": 3.463954530467844,
      "tokens_seen": 1964769280
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040860215053763443,
      "loss": 0.0672,
      "theoretical_loss": 3.463916182109398,
      "tokens_seen": 1965031424
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.000408521906596052,
      "loss": 0.0686,
      "theoretical_loss": 3.463877840298672,
      "tokens_seen": 1965293568
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004084416626544696,
      "loss": 0.0683,
      "theoretical_loss": 3.4638395050336745,
      "tokens_seen": 1965555712
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040836141871288716,
      "loss": 0.0696,
      "theoretical_loss": 3.4638011763124164,
      "tokens_seen": 1965817856
    },
    {
      "debugging/Compilability": 1.0,
      "debugging/distinct-1-grams": 0.7414976990322337,
      "debugging/entropy-1-grams": 5.624956263227352,
      "debugging/length": 461.2173913043478,
      "debugging/num_segments": 23,
      "debugging/raw_token_scores_avg": 0.006980398204177618,
      "debugging/raw_token_scores_std": 0.02224303036928177,
      "debugging/score": 0.007207965953219283,
      "debugging/score_std": 0.01109644612420906,
      "epoch": 0.6,
      "objective/train/advantage_avg": 0.00012892790255136788,
      "objective/train/docs_used": 716760,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4562304019927979,
      "objective/train/original_loss": 1.456230640411377,
      "objective/train/theoretical_loss": 3.4637628541329066,
      "objective/train/tokens_used": 1986540000,
      "objective/train/value_avg": -0.007106781005859375,
      "objective/train/value_loss": 0.00028058464522473514,
      "objective/train/value_max": -5.269050598144531e-05,
      "objective/train/value_min": -0.841796875,
      "objective/train/value_reward_corr": 0.6591656050447988,
      "objective/train/value_std": 0.01369476318359375,
      "objective/train/weight_avg": 1.0002350807189941,
      "objective/train/weighted_lm_loss": 1.456529974937439,
      "objective/train/weights_max": 2.0581281185150146,
      "objective/train/weights_min": 0.06374084949493408,
      "theoretical_loss": 3.4637628541329066,
      "tokens_seen": 1966080000
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040828117477130477,
      "loss": 0.0681,
      "theoretical_loss": 3.4637628541329066,
      "tokens_seen": 1966080000
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004082009308297224,
      "loss": 0.0666,
      "theoretical_loss": 3.4637245384931576,
      "tokens_seen": 1966342144
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040812068688813994,
      "loss": 0.0694,
      "theoretical_loss": 3.4636862293911816,
      "tokens_seen": 1966604288
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040804044294655756,
      "loss": 0.0677,
      "theoretical_loss": 3.463647926824992,
      "tokens_seen": 1966866432
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004079601990049751,
      "loss": 0.0693,
      "theoretical_loss": 3.4636096307926016,
      "tokens_seen": 1967128576
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004078799550633927,
      "loss": 0.0695,
      "theoretical_loss": 3.4635713412920275,
      "tokens_seen": 1967390720
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004077997111218103,
      "loss": 0.0704,
      "theoretical_loss": 3.4635330583212838,
      "tokens_seen": 1967652864
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004077194671802279,
      "loss": 0.0694,
      "theoretical_loss": 3.463494781878388,
      "tokens_seen": 1967915008
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004076392232386455,
      "loss": 0.0715,
      "theoretical_loss": 3.4634565119613576,
      "tokens_seen": 1968177152
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040755897929706307,
      "loss": 0.0699,
      "theoretical_loss": 3.463418248568211,
      "tokens_seen": 1968439296
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004074787353554807,
      "loss": 0.0673,
      "theoretical_loss": 3.463379991696967,
      "tokens_seen": 1968701440
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004073984914138983,
      "loss": 0.0681,
      "theoretical_loss": 3.463341741345646,
      "tokens_seen": 1968963584
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040731824747231585,
      "loss": 0.0663,
      "theoretical_loss": 3.4633034975122694,
      "tokens_seen": 1969225728
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004072380035307334,
      "loss": 0.0693,
      "theoretical_loss": 3.4632652601948593,
      "tokens_seen": 1969487872
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.000407157759589151,
      "loss": 0.0685,
      "theoretical_loss": 3.463227029391437,
      "tokens_seen": 1969750016
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040707751564756864,
      "loss": 0.0695,
      "theoretical_loss": 3.4631888051000272,
      "tokens_seen": 1970012160
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004069972717059862,
      "loss": 0.0669,
      "theoretical_loss": 3.4631505873186548,
      "tokens_seen": 1970274304
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004069170277644038,
      "loss": 0.067,
      "theoretical_loss": 3.463112376045344,
      "tokens_seen": 1970536448
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004068367838228214,
      "loss": 0.0679,
      "theoretical_loss": 3.463074171278122,
      "tokens_seen": 1970798592
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.000406756539881239,
      "loss": 0.0679,
      "theoretical_loss": 3.4630359730150153,
      "tokens_seen": 1971060736
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040667629593965654,
      "loss": 0.0694,
      "theoretical_loss": 3.4629977812540518,
      "tokens_seen": 1971322880
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040659605199807415,
      "loss": 0.0709,
      "theoretical_loss": 3.46295959599326,
      "tokens_seen": 1971585024
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040651580805649176,
      "loss": 0.0695,
      "theoretical_loss": 3.4629214172306706,
      "tokens_seen": 1971847168
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004064355641149093,
      "loss": 0.0704,
      "theoretical_loss": 3.462883244964313,
      "tokens_seen": 1972109312
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040635532017332693,
      "loss": 0.0663,
      "theoretical_loss": 3.4628450791922187,
      "tokens_seen": 1972371456
    },
    {
      "epoch": 0.6,
      "objective/train/advantage_avg": 0.00040722067933529615,
      "objective/train/docs_used": 719209,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3216396570205688,
      "objective/train/original_loss": 1.3216397762298584,
      "objective/train/theoretical_loss": 3.4628069199124205,
      "objective/train/tokens_used": 1993093600,
      "objective/train/value_avg": -0.00849151611328125,
      "objective/train/value_loss": 0.00020072172628715634,
      "objective/train/value_max": -6.657838821411133e-05,
      "objective/train/value_min": -0.3955078125,
      "objective/train/value_reward_corr": 0.7492742374040502,
      "objective/train/value_std": 0.01561737060546875,
      "objective/train/weight_avg": 1.000498652458191,
      "objective/train/weighted_lm_loss": 1.3224918842315674,
      "objective/train/weights_max": 1.2429403066635132,
      "objective/train/weights_min": 0.36926716566085815,
      "theoretical_loss": 3.4628069199124205,
      "tokens_seen": 1972633600
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040627507623174455,
      "loss": 0.066,
      "theoretical_loss": 3.4628069199124205,
      "tokens_seen": 1972633600
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040619483229016205,
      "loss": 0.0675,
      "theoretical_loss": 3.4627687671229515,
      "tokens_seen": 1972895744
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040611458834857966,
      "loss": 0.0687,
      "theoretical_loss": 3.462730620821845,
      "tokens_seen": 1973157888
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004060343444069973,
      "loss": 0.0695,
      "theoretical_loss": 3.462692481007136,
      "tokens_seen": 1973420032
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004059541004654149,
      "loss": 0.0678,
      "theoretical_loss": 3.46265434767686,
      "tokens_seen": 1973682176
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040587385652383245,
      "loss": 0.068,
      "theoretical_loss": 3.4626162208290534,
      "tokens_seen": 1973944320
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040579361258225006,
      "loss": 0.0691,
      "theoretical_loss": 3.462578100461754,
      "tokens_seen": 1974206464
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040571336864066767,
      "loss": 0.069,
      "theoretical_loss": 3.4625399865730007,
      "tokens_seen": 1974468608
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040563312469908523,
      "loss": 0.067,
      "theoretical_loss": 3.462501879160831,
      "tokens_seen": 1974730752
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004055528807575028,
      "loss": 0.0702,
      "theoretical_loss": 3.462463778223285,
      "tokens_seen": 1974992896
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004054726368159204,
      "loss": 0.0682,
      "theoretical_loss": 3.462425683758404,
      "tokens_seen": 1975255040
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.000405392392874338,
      "loss": 0.0721,
      "theoretical_loss": 3.4623875957642305,
      "tokens_seen": 1975517184
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040531214893275557,
      "loss": 0.0698,
      "theoretical_loss": 3.462349514238805,
      "tokens_seen": 1975779328
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004052319049911732,
      "loss": 0.0705,
      "theoretical_loss": 3.462311439180173,
      "tokens_seen": 1976041472
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004051516610495908,
      "loss": 0.0677,
      "theoretical_loss": 3.4622733705863764,
      "tokens_seen": 1976303616
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040507141710800835,
      "loss": 0.0654,
      "theoretical_loss": 3.462235308455462,
      "tokens_seen": 1976565760
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004049911731664259,
      "loss": 0.0687,
      "theoretical_loss": 3.4621972527854745,
      "tokens_seen": 1976827904
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004049109292248435,
      "loss": 0.0705,
      "theoretical_loss": 3.462159203574461,
      "tokens_seen": 1977090048
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040483068528326114,
      "loss": 0.0686,
      "theoretical_loss": 3.46212116082047,
      "tokens_seen": 1977352192
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004047504413416787,
      "loss": 0.0699,
      "theoretical_loss": 3.4620831245215484,
      "tokens_seen": 1977614336
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004046701974000963,
      "loss": 0.0692,
      "theoretical_loss": 3.462045094675747,
      "tokens_seen": 1977876480
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004045899534585139,
      "loss": 0.0683,
      "theoretical_loss": 3.462007071281114,
      "tokens_seen": 1978138624
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004045097095169315,
      "loss": 0.0666,
      "theoretical_loss": 3.461969054335703,
      "tokens_seen": 1978400768
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040442946557534904,
      "loss": 0.0687,
      "theoretical_loss": 3.461931043837563,
      "tokens_seen": 1978662912
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040434922163376665,
      "loss": 0.0681,
      "theoretical_loss": 3.4618930397847487,
      "tokens_seen": 1978925056
    },
    {
      "epoch": 0.6,
      "objective/train/advantage_avg": 0.0015131087275221944,
      "objective/train/docs_used": 721537,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4136056900024414,
      "objective/train/original_loss": 1.4136054515838623,
      "objective/train/theoretical_loss": 3.461855042175312,
      "objective/train/tokens_used": 1999647200,
      "objective/train/value_avg": -0.007205963134765625,
      "objective/train/value_loss": 0.00017736262816470116,
      "objective/train/value_max": -6.204843521118164e-05,
      "objective/train/value_min": -0.340087890625,
      "objective/train/value_reward_corr": 0.6555204938605945,
      "objective/train/value_std": 0.0123291015625,
      "objective/train/weight_avg": 1.001592993736267,
      "objective/train/weighted_lm_loss": 1.4155278205871582,
      "objective/train/weights_max": 1.4050710201263428,
      "objective/train/weights_min": 0.3719121813774109,
      "theoretical_loss": 3.461855042175312,
      "tokens_seen": 1979187200
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004042689776921842,
      "loss": 0.0679,
      "theoretical_loss": 3.461855042175312,
      "tokens_seen": 1979187200
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004041887337506018,
      "loss": 0.0687,
      "theoretical_loss": 3.4618170510073085,
      "tokens_seen": 1979449344
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040410848980901944,
      "loss": 0.0696,
      "theoretical_loss": 3.4617790662787935,
      "tokens_seen": 1979711488
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040402824586743705,
      "loss": 0.0724,
      "theoretical_loss": 3.4617410879878223,
      "tokens_seen": 1979973632
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004039480019258546,
      "loss": 0.0693,
      "theoretical_loss": 3.461703116132452,
      "tokens_seen": 1980235776
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004038677579842722,
      "loss": 0.0695,
      "theoretical_loss": 3.46166515071074,
      "tokens_seen": 1980497920
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004037875140426898,
      "loss": 0.0684,
      "theoretical_loss": 3.461627191720745,
      "tokens_seen": 1980760064
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040370727010110734,
      "loss": 0.0695,
      "theoretical_loss": 3.461589239160528,
      "tokens_seen": 1981022208
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040362702615952495,
      "loss": 0.0682,
      "theoretical_loss": 3.4615512930281467,
      "tokens_seen": 1981284352
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040354678221794256,
      "loss": 0.0676,
      "theoretical_loss": 3.4615133533216635,
      "tokens_seen": 1981546496
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004034665382763602,
      "loss": 0.068,
      "theoretical_loss": 3.4614754200391404,
      "tokens_seen": 1981808640
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040338629433477773,
      "loss": 0.0658,
      "theoretical_loss": 3.4614374931786402,
      "tokens_seen": 1982070784
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040330605039319534,
      "loss": 0.0669,
      "theoretical_loss": 3.461399572738226,
      "tokens_seen": 1982332928
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004032258064516129,
      "loss": 0.0687,
      "theoretical_loss": 3.461361658715963,
      "tokens_seen": 1982595072
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040314556251003046,
      "loss": 0.0695,
      "theoretical_loss": 3.4613237511099157,
      "tokens_seen": 1982857216
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004030653185684481,
      "loss": 0.0684,
      "theoretical_loss": 3.4612858499181502,
      "tokens_seen": 1983119360
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004029850746268657,
      "loss": 0.07,
      "theoretical_loss": 3.4612479551387345,
      "tokens_seen": 1983381504
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004029048306852833,
      "loss": 0.0689,
      "theoretical_loss": 3.461210066769736,
      "tokens_seen": 1983643648
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040282458674370086,
      "loss": 0.0688,
      "theoretical_loss": 3.4611721848092225,
      "tokens_seen": 1983905792
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040274434280211847,
      "loss": 0.0692,
      "theoretical_loss": 3.461134309255265,
      "tokens_seen": 1984167936
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004026640988605361,
      "loss": 0.068,
      "theoretical_loss": 3.4610964401059325,
      "tokens_seen": 1984430080
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004025838549189536,
      "loss": 0.0691,
      "theoretical_loss": 3.461058577359297,
      "tokens_seen": 1984692224
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004025036109773712,
      "loss": 0.0684,
      "theoretical_loss": 3.4610207210134294,
      "tokens_seen": 1984954368
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004024233670357888,
      "loss": 0.0702,
      "theoretical_loss": 3.4609828710664035,
      "tokens_seen": 1985216512
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004023431230942064,
      "loss": 0.0673,
      "theoretical_loss": 3.460945027516293,
      "tokens_seen": 1985478656
    },
    {
      "epoch": 0.6,
      "objective/train/advantage_avg": 0.0006132093258202076,
      "objective/train/docs_used": 723995,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3630348443984985,
      "objective/train/original_loss": 1.3630346059799194,
      "objective/train/theoretical_loss": 3.460907190361172,
      "objective/train/tokens_used": 2006200800,
      "objective/train/value_avg": -0.00844573974609375,
      "objective/train/value_loss": 0.00041758434963412583,
      "objective/train/value_max": -5.143880844116211e-05,
      "objective/train/value_min": -0.984375,
      "objective/train/value_reward_corr": 0.6131844355731204,
      "objective/train/value_std": 0.01934814453125,
      "objective/train/weight_avg": 1.0008108615875244,
      "objective/train/weighted_lm_loss": 1.3635249137878418,
      "objective/train/weights_max": 2.572547435760498,
      "objective/train/weights_min": 0.3744775056838989,
      "theoretical_loss": 3.460907190361172,
      "tokens_seen": 1985740800
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.000402262879152624,
      "loss": 0.0675,
      "theoretical_loss": 3.460907190361172,
      "tokens_seen": 1985740800
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004021826352110416,
      "loss": 0.0679,
      "theoretical_loss": 3.460869359599116,
      "tokens_seen": 1986002944
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004021023912694592,
      "loss": 0.0682,
      "theoretical_loss": 3.460831535228201,
      "tokens_seen": 1986265088
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004020221473278767,
      "loss": 0.0695,
      "theoretical_loss": 3.4607937172465046,
      "tokens_seen": 1986527232
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004019419033862943,
      "loss": 0.0671,
      "theoretical_loss": 3.4607559056521033,
      "tokens_seen": 1986789376
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040186165944471194,
      "loss": 0.069,
      "theoretical_loss": 3.4607181004430774,
      "tokens_seen": 1987051520
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004017814155031295,
      "loss": 0.0683,
      "theoretical_loss": 3.460680301617505,
      "tokens_seen": 1987313664
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004017011715615471,
      "loss": 0.069,
      "theoretical_loss": 3.460642509173468,
      "tokens_seen": 1987575808
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004016209276199647,
      "loss": 0.0719,
      "theoretical_loss": 3.4606047231090455,
      "tokens_seen": 1987837952
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040154068367838233,
      "loss": 0.0658,
      "theoretical_loss": 3.4605669434223216,
      "tokens_seen": 1988100096
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040146043973679984,
      "loss": 0.0671,
      "theoretical_loss": 3.4605291701113776,
      "tokens_seen": 1988362240
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040138019579521745,
      "loss": 0.0663,
      "theoretical_loss": 3.4604914031742977,
      "tokens_seen": 1988624384
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040129995185363506,
      "loss": 0.0673,
      "theoretical_loss": 3.460453642609166,
      "tokens_seen": 1988886528
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004012197079120526,
      "loss": 0.0697,
      "theoretical_loss": 3.4604158884140683,
      "tokens_seen": 1989148672
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040113946397047023,
      "loss": 0.0686,
      "theoretical_loss": 3.460378140587091,
      "tokens_seen": 1989410816
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040105922002888785,
      "loss": 0.0693,
      "theoretical_loss": 3.46034039912632,
      "tokens_seen": 1989672960
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040097897608730546,
      "loss": 0.0662,
      "theoretical_loss": 3.460302664029844,
      "tokens_seen": 1989935104
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.000400898732145723,
      "loss": 0.0708,
      "theoretical_loss": 3.4602649352957515,
      "tokens_seen": 1990197248
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004008184882041406,
      "loss": 0.0675,
      "theoretical_loss": 3.460227212922131,
      "tokens_seen": 1990459392
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004007382442625582,
      "loss": 0.0693,
      "theoretical_loss": 3.4601894969070743,
      "tokens_seen": 1990721536
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040065800032097575,
      "loss": 0.067,
      "theoretical_loss": 3.460151787248672,
      "tokens_seen": 1990983680
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040057775637939336,
      "loss": 0.0682,
      "theoretical_loss": 3.460114083945015,
      "tokens_seen": 1991245824
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040049751243781097,
      "loss": 0.0689,
      "theoretical_loss": 3.4600763869941966,
      "tokens_seen": 1991507968
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004004172684962286,
      "loss": 0.0687,
      "theoretical_loss": 3.460038696394311,
      "tokens_seen": 1991770112
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00040033702455464614,
      "loss": 0.0706,
      "theoretical_loss": 3.4600010121434517,
      "tokens_seen": 1992032256
    },
    {
      "epoch": 0.6,
      "objective/train/advantage_avg": 0.0002454657224006951,
      "objective/train/docs_used": 726441,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3276835680007935,
      "objective/train/original_loss": 1.327683448791504,
      "objective/train/theoretical_loss": 3.4599633342397142,
      "objective/train/tokens_used": 2012754400,
      "objective/train/value_avg": -0.0082244873046875,
      "objective/train/value_loss": 0.00023926161520648748,
      "objective/train/value_max": -5.066394805908203e-05,
      "objective/train/value_min": -0.59423828125,
      "objective/train/value_reward_corr": 0.7205625276168954,
      "objective/train/value_std": 0.0155487060546875,
      "objective/train/weight_avg": 1.000353455543518,
      "objective/train/weighted_lm_loss": 1.3281760215759277,
      "objective/train/weights_max": 1.299572229385376,
      "objective/train/weights_min": 0.36829331517219543,
      "theoretical_loss": 3.4599633342397142,
      "tokens_seen": 1992294400
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004002567806130637,
      "loss": 0.068,
      "theoretical_loss": 3.4599633342397142,
      "tokens_seen": 1992294400
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004001765366714813,
      "loss": 0.0682,
      "theoretical_loss": 3.4599256626811945,
      "tokens_seen": 1992556544
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004000962927298989,
      "loss": 0.0682,
      "theoretical_loss": 3.45988799746599,
      "tokens_seen": 1992818688
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0004000160487883165,
      "loss": 0.0677,
      "theoretical_loss": 3.4598503385921977,
      "tokens_seen": 1993080832
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0003999358048467341,
      "loss": 0.0677,
      "theoretical_loss": 3.4598126860579166,
      "tokens_seen": 1993342976
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00039985556090515166,
      "loss": 0.0673,
      "theoretical_loss": 3.4597750398612455,
      "tokens_seen": 1993605120
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00039977531696356927,
      "loss": 0.0702,
      "theoretical_loss": 3.459737400000284,
      "tokens_seen": 1993867264
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00039969507302198683,
      "loss": 0.0675,
      "theoretical_loss": 3.4596997664731344,
      "tokens_seen": 1994129408
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00039961482908040444,
      "loss": 0.0685,
      "theoretical_loss": 3.4596621392778983,
      "tokens_seen": 1994391552
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.000399534585138822,
      "loss": 0.0685,
      "theoretical_loss": 3.459624518412677,
      "tokens_seen": 1994653696
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0003994543411972396,
      "loss": 0.0668,
      "theoretical_loss": 3.459586903875575,
      "tokens_seen": 1994915840
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0003993740972556572,
      "loss": 0.0692,
      "theoretical_loss": 3.4595492956646963,
      "tokens_seen": 1995177984
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0003992938533140748,
      "loss": 0.0675,
      "theoretical_loss": 3.459511693778146,
      "tokens_seen": 1995440128
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.0003992136093724924,
      "loss": 0.0679,
      "theoretical_loss": 3.4594740982140295,
      "tokens_seen": 1995702272
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00039913336543091,
      "loss": 0.0699,
      "theoretical_loss": 3.459436508970454,
      "tokens_seen": 1995964416
    },
    {
      "epoch": 0.6,
      "learning_rate": 0.00039905312148932757,
      "loss": 0.0689,
      "theoretical_loss": 3.4593989260455267,
      "tokens_seen": 1996226560
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003989728775477451,
      "loss": 0.0677,
      "theoretical_loss": 3.459361349437356,
      "tokens_seen": 1996488704
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039889263360616274,
      "loss": 0.0695,
      "theoretical_loss": 3.459323779144051,
      "tokens_seen": 1996750848
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039881238966458035,
      "loss": 0.0677,
      "theoretical_loss": 3.4592862151637216,
      "tokens_seen": 1997012992
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003987321457229979,
      "loss": 0.0673,
      "theoretical_loss": 3.4592486574944785,
      "tokens_seen": 1997275136
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003986519017814155,
      "loss": 0.0693,
      "theoretical_loss": 3.4592111061344335,
      "tokens_seen": 1997537280
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039857165783983313,
      "loss": 0.07,
      "theoretical_loss": 3.4591735610816983,
      "tokens_seen": 1997799424
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003984914138982507,
      "loss": 0.0689,
      "theoretical_loss": 3.459136022334387,
      "tokens_seen": 1998061568
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039841116995666825,
      "loss": 0.0715,
      "theoretical_loss": 3.4590984898906134,
      "tokens_seen": 1998323712
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039833092601508586,
      "loss": 0.0672,
      "theoretical_loss": 3.4590609637484913,
      "tokens_seen": 1998585856
    },
    {
      "epoch": 0.61,
      "objective/train/advantage_avg": 1.4012218343850691e-05,
      "objective/train/docs_used": 728818,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3711408376693726,
      "objective/train/original_loss": 1.371140718460083,
      "objective/train/theoretical_loss": 3.459023443906138,
      "objective/train/tokens_used": 2019308000,
      "objective/train/value_avg": -0.0090179443359375,
      "objective/train/value_loss": 0.0004112530150450766,
      "objective/train/value_max": -6.556510925292969e-05,
      "objective/train/value_min": -0.98876953125,
      "objective/train/value_reward_corr": 0.7207536114861102,
      "objective/train/value_std": 0.0240020751953125,
      "objective/train/weight_avg": 1.0002104043960571,
      "objective/train/weighted_lm_loss": 1.3707592487335205,
      "objective/train/weights_max": 2.44260835647583,
      "objective/train/weights_min": 0.229869082570076,
      "theoretical_loss": 3.459023443906138,
      "tokens_seen": 1998848000
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003982506820735035,
      "loss": 0.0686,
      "theoretical_loss": 3.459023443906138,
      "tokens_seen": 1998848000
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039817043813192103,
      "loss": 0.0677,
      "theoretical_loss": 3.458985930361668,
      "tokens_seen": 1999110144
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039809019419033865,
      "loss": 0.0679,
      "theoretical_loss": 3.4589484231132,
      "tokens_seen": 1999372288
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039800995024875626,
      "loss": 0.0668,
      "theoretical_loss": 3.4589109221588514,
      "tokens_seen": 1999634432
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003979297063071738,
      "loss": 0.072,
      "theoretical_loss": 3.4588734274967416,
      "tokens_seen": 1999896576
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003978494623655914,
      "loss": 0.0699,
      "theoretical_loss": 3.4588359391249894,
      "tokens_seen": 2000158720
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.000397769218424009,
      "loss": 0.068,
      "theoretical_loss": 3.458798457041716,
      "tokens_seen": 2000420864
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003976889744824266,
      "loss": 0.0683,
      "theoretical_loss": 3.4587609812450424,
      "tokens_seen": 2000683008
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039760873054084416,
      "loss": 0.0691,
      "theoretical_loss": 3.4587235117330906,
      "tokens_seen": 2000945152
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039752848659926177,
      "loss": 0.0663,
      "theoretical_loss": 3.4586860485039836,
      "tokens_seen": 2001207296
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003974482426576794,
      "loss": 0.0684,
      "theoretical_loss": 3.4586485915558454,
      "tokens_seen": 2001469440
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039736799871609694,
      "loss": 0.0697,
      "theoretical_loss": 3.4586111408868,
      "tokens_seen": 2001731584
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003972877547745145,
      "loss": 0.0687,
      "theoretical_loss": 3.4585736964949727,
      "tokens_seen": 2001993728
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003972075108329321,
      "loss": 0.0665,
      "theoretical_loss": 3.45853625837849,
      "tokens_seen": 2002255872
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003971272668913497,
      "loss": 0.0666,
      "theoretical_loss": 3.458498826535479,
      "tokens_seen": 2002518016
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003970470229497673,
      "loss": 0.0699,
      "theoretical_loss": 3.4584614009640666,
      "tokens_seen": 2002780160
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003969667790081849,
      "loss": 0.0666,
      "theoretical_loss": 3.4584239816623823,
      "tokens_seen": 2003042304
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003968865350666025,
      "loss": 0.0688,
      "theoretical_loss": 3.4583865686285544,
      "tokens_seen": 2003304448
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039680629112502007,
      "loss": 0.0705,
      "theoretical_loss": 3.4583491618607134,
      "tokens_seen": 2003566592
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039672604718343763,
      "loss": 0.0687,
      "theoretical_loss": 3.458311761356991,
      "tokens_seen": 2003828736
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039664580324185524,
      "loss": 0.0695,
      "theoretical_loss": 3.4582743671155183,
      "tokens_seen": 2004090880
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039656555930027285,
      "loss": 0.0692,
      "theoretical_loss": 3.458236979134428,
      "tokens_seen": 2004353024
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003964853153586904,
      "loss": 0.0669,
      "theoretical_loss": 3.458199597411853,
      "tokens_seen": 2004615168
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.000396405071417108,
      "loss": 0.069,
      "theoretical_loss": 3.4581622219459276,
      "tokens_seen": 2004877312
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039632482747552564,
      "loss": 0.0657,
      "theoretical_loss": 3.4581248527347874,
      "tokens_seen": 2005139456
    },
    {
      "epoch": 0.61,
      "objective/train/advantage_avg": 0.0006999548641033471,
      "objective/train/docs_used": 731237,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3625905513763428,
      "objective/train/original_loss": 1.3625905513763428,
      "objective/train/theoretical_loss": 3.458087489776567,
      "objective/train/tokens_used": 2025861600,
      "objective/train/value_avg": -0.0076751708984375,
      "objective/train/value_loss": 0.0002253977581858635,
      "objective/train/value_max": -2.9981136322021484e-05,
      "objective/train/value_min": -0.70458984375,
      "objective/train/value_reward_corr": 0.7154743835992182,
      "objective/train/value_std": 0.016448974609375,
      "objective/train/weight_avg": 1.0008041858673096,
      "objective/train/weighted_lm_loss": 1.363987684249878,
      "objective/train/weights_max": 1.384639024734497,
      "objective/train/weights_min": 0.3804037272930145,
      "theoretical_loss": 3.458087489776567,
      "tokens_seen": 2005401600
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003962445835339432,
      "loss": 0.0674,
      "theoretical_loss": 3.458087489776567,
      "tokens_seen": 2005401600
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003961643395923608,
      "loss": 0.0677,
      "theoretical_loss": 3.458050133069404,
      "tokens_seen": 2005663744
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039608409565077837,
      "loss": 0.0685,
      "theoretical_loss": 3.4580127826114353,
      "tokens_seen": 2005925888
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003960038517091959,
      "loss": 0.0672,
      "theoretical_loss": 3.457975438400799,
      "tokens_seen": 2006188032
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039592360776761354,
      "loss": 0.0655,
      "theoretical_loss": 3.4579381004356344,
      "tokens_seen": 2006450176
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039584336382603115,
      "loss": 0.0684,
      "theoretical_loss": 3.4579007687140804,
      "tokens_seen": 2006712320
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039576311988444876,
      "loss": 0.0702,
      "theoretical_loss": 3.4578634432342783,
      "tokens_seen": 2006974464
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003956828759428663,
      "loss": 0.0702,
      "theoretical_loss": 3.4578261239943693,
      "tokens_seen": 2007236608
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039560263200128393,
      "loss": 0.0704,
      "theoretical_loss": 3.4577888109924952,
      "tokens_seen": 2007498752
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003955223880597015,
      "loss": 0.0688,
      "theoretical_loss": 3.4577515042267994,
      "tokens_seen": 2007760896
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039544214411811905,
      "loss": 0.0655,
      "theoretical_loss": 3.457714203695425,
      "tokens_seen": 2008023040
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039536190017653666,
      "loss": 0.067,
      "theoretical_loss": 3.4576769093965174,
      "tokens_seen": 2008285184
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003952816562349543,
      "loss": 0.0681,
      "theoretical_loss": 3.4576396213282212,
      "tokens_seen": 2008547328
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003952014122933719,
      "loss": 0.0703,
      "theoretical_loss": 3.457602339488682,
      "tokens_seen": 2008809472
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039512116835178945,
      "loss": 0.0683,
      "theoretical_loss": 3.4575650638760482,
      "tokens_seen": 2009071616
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039504092441020706,
      "loss": 0.0687,
      "theoretical_loss": 3.457527794488466,
      "tokens_seen": 2009333760
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003949606804686246,
      "loss": 0.0685,
      "theoretical_loss": 3.457490531324085,
      "tokens_seen": 2009595904
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003948804365270422,
      "loss": 0.0662,
      "theoretical_loss": 3.4574532743810535,
      "tokens_seen": 2009858048
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003948001925854598,
      "loss": 0.0674,
      "theoretical_loss": 3.4574160236575224,
      "tokens_seen": 2010120192
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003947199486438774,
      "loss": 0.0686,
      "theoretical_loss": 3.457378779151642,
      "tokens_seen": 2010382336
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.000394639704702295,
      "loss": 0.0674,
      "theoretical_loss": 3.457341540861564,
      "tokens_seen": 2010644480
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039455946076071257,
      "loss": 0.0667,
      "theoretical_loss": 3.4573043087854414,
      "tokens_seen": 2010906624
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003944792168191302,
      "loss": 0.0687,
      "theoretical_loss": 3.4572670829214265,
      "tokens_seen": 2011168768
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003943989728775478,
      "loss": 0.0683,
      "theoretical_loss": 3.457229863267674,
      "tokens_seen": 2011430912
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003943187289359653,
      "loss": 0.0678,
      "theoretical_loss": 3.457192649822338,
      "tokens_seen": 2011693056
    },
    {
      "epoch": 0.61,
      "objective/train/advantage_avg": 0.0003829321067314595,
      "objective/train/docs_used": 733698,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4594643115997314,
      "objective/train/original_loss": 1.4594643115997314,
      "objective/train/theoretical_loss": 3.4571554425835753,
      "objective/train/tokens_used": 2032415200,
      "objective/train/value_avg": -0.00807952880859375,
      "objective/train/value_loss": 0.0001753742399159819,
      "objective/train/value_max": -2.6285648345947266e-05,
      "objective/train/value_min": -0.81005859375,
      "objective/train/value_reward_corr": 0.7278631297172813,
      "objective/train/value_std": 0.0141754150390625,
      "objective/train/weight_avg": 1.0004661083221436,
      "objective/train/weighted_lm_loss": 1.4600026607513428,
      "objective/train/weights_max": 1.5876256227493286,
      "objective/train/weights_min": 0.3781697452068329,
      "theoretical_loss": 3.4571554425835753,
      "tokens_seen": 2011955200
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003942384849943829,
      "loss": 0.0698,
      "theoretical_loss": 3.4571554425835753,
      "tokens_seen": 2011955200
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003941582410528005,
      "loss": 0.0688,
      "theoretical_loss": 3.457118241549541,
      "tokens_seen": 2012217344
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039407799711121814,
      "loss": 0.0698,
      "theoretical_loss": 3.4570810467183932,
      "tokens_seen": 2012479488
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003939977531696357,
      "loss": 0.0696,
      "theoretical_loss": 3.45704385808829,
      "tokens_seen": 2012741632
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003939175092280533,
      "loss": 0.0704,
      "theoretical_loss": 3.4570066756573885,
      "tokens_seen": 2013003776
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003938372652864709,
      "loss": 0.0675,
      "theoretical_loss": 3.45696949942385,
      "tokens_seen": 2013265920
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003937570213448884,
      "loss": 0.0683,
      "theoretical_loss": 3.4569323293858334,
      "tokens_seen": 2013528064
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039367677740330604,
      "loss": 0.0678,
      "theoretical_loss": 3.4568951655415017,
      "tokens_seen": 2013790208
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039359653346172365,
      "loss": 0.0664,
      "theoretical_loss": 3.4568580078890143,
      "tokens_seen": 2014052352
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003935162895201412,
      "loss": 0.0694,
      "theoretical_loss": 3.4568208564265364,
      "tokens_seen": 2014314496
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003934360455785588,
      "loss": 0.0677,
      "theoretical_loss": 3.4567837111522293,
      "tokens_seen": 2014576640
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039335580163697644,
      "loss": 0.0675,
      "theoretical_loss": 3.456746572064259,
      "tokens_seen": 2014838784
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039327555769539405,
      "loss": 0.0676,
      "theoretical_loss": 3.456709439160789,
      "tokens_seen": 2015100928
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003931953137538116,
      "loss": 0.0678,
      "theoretical_loss": 3.456672312439986,
      "tokens_seen": 2015363072
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039311506981222916,
      "loss": 0.0679,
      "theoretical_loss": 3.4566351919000167,
      "tokens_seen": 2015625216
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003930348258706468,
      "loss": 0.0649,
      "theoretical_loss": 3.4565980775390477,
      "tokens_seen": 2015887360
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039295458192906434,
      "loss": 0.0688,
      "theoretical_loss": 3.456560969355248,
      "tokens_seen": 2016149504
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039287433798748195,
      "loss": 0.0673,
      "theoretical_loss": 3.456523867346786,
      "tokens_seen": 2016411648
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039279409404589956,
      "loss": 0.0654,
      "theoretical_loss": 3.4564867715118313,
      "tokens_seen": 2016673792
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003927138501043172,
      "loss": 0.0686,
      "theoretical_loss": 3.4564496818485546,
      "tokens_seen": 2016935936
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039263360616273473,
      "loss": 0.0693,
      "theoretical_loss": 3.4564125983551275,
      "tokens_seen": 2017198080
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003925533622211523,
      "loss": 0.0696,
      "theoretical_loss": 3.4563755210297216,
      "tokens_seen": 2017460224
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003924731182795699,
      "loss": 0.0685,
      "theoretical_loss": 3.4563384498705094,
      "tokens_seen": 2017722368
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039239287433798746,
      "loss": 0.0679,
      "theoretical_loss": 3.456301384875666,
      "tokens_seen": 2017984512
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003923126303964051,
      "loss": 0.0666,
      "theoretical_loss": 3.456264326043364,
      "tokens_seen": 2018246656
    },
    {
      "epoch": 0.61,
      "objective/train/advantage_avg": 0.00186018249951303,
      "objective/train/docs_used": 736110,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.349985957145691,
      "objective/train/original_loss": 1.3499860763549805,
      "objective/train/theoretical_loss": 3.4562272733717796,
      "objective/train/tokens_used": 2038968800,
      "objective/train/value_avg": -0.01027679443359375,
      "objective/train/value_loss": 0.00021227422985248268,
      "objective/train/value_max": -5.4776668548583984e-05,
      "objective/train/value_min": -0.449462890625,
      "objective/train/value_reward_corr": 0.785820240253668,
      "objective/train/value_std": 0.0207061767578125,
      "objective/train/weight_avg": 1.0019619464874268,
      "objective/train/weighted_lm_loss": 1.3521424531936646,
      "objective/train/weights_max": 1.5077892541885376,
      "objective/train/weights_min": 0.3687315881252289,
      "theoretical_loss": 3.4562272733717796,
      "tokens_seen": 2018508800
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003922323864548227,
      "loss": 0.0684,
      "theoretical_loss": 3.4562272733717796,
      "tokens_seen": 2018508800
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003921521425132403,
      "loss": 0.068,
      "theoretical_loss": 3.4561902268590883,
      "tokens_seen": 2018770944
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039207189857165786,
      "loss": 0.0674,
      "theoretical_loss": 3.4561531865034665,
      "tokens_seen": 2019033088
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003919916546300754,
      "loss": 0.0654,
      "theoretical_loss": 3.4561161523030925,
      "tokens_seen": 2019295232
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039191141068849303,
      "loss": 0.0672,
      "theoretical_loss": 3.456079124256145,
      "tokens_seen": 2019557376
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003918311667469106,
      "loss": 0.0706,
      "theoretical_loss": 3.4560421023608012,
      "tokens_seen": 2019819520
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003917509228053282,
      "loss": 0.0705,
      "theoretical_loss": 3.4560050866152423,
      "tokens_seen": 2020081664
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003916706788637458,
      "loss": 0.0665,
      "theoretical_loss": 3.455968077017649,
      "tokens_seen": 2020343808
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039159043492216337,
      "loss": 0.0675,
      "theoretical_loss": 3.455931073566202,
      "tokens_seen": 2020605952
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.000391510190980581,
      "loss": 0.0674,
      "theoretical_loss": 3.4558940762590837,
      "tokens_seen": 2020868096
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003914299470389986,
      "loss": 0.0697,
      "theoretical_loss": 3.455857085094477,
      "tokens_seen": 2021130240
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039134970309741615,
      "loss": 0.066,
      "theoretical_loss": 3.4558201000705653,
      "tokens_seen": 2021392384
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003912694591558337,
      "loss": 0.0701,
      "theoretical_loss": 3.455783121185534,
      "tokens_seen": 2021654528
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003911892152142513,
      "loss": 0.0651,
      "theoretical_loss": 3.4557461484375676,
      "tokens_seen": 2021916672
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039110897127266894,
      "loss": 0.0692,
      "theoretical_loss": 3.4557091818248518,
      "tokens_seen": 2022178816
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003910287273310865,
      "loss": 0.0682,
      "theoretical_loss": 3.4556722213455737,
      "tokens_seen": 2022440960
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003909484833895041,
      "loss": 0.0674,
      "theoretical_loss": 3.455635266997921,
      "tokens_seen": 2022703104
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003908682394479217,
      "loss": 0.0664,
      "theoretical_loss": 3.4555983187800825,
      "tokens_seen": 2022965248
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003907879955063393,
      "loss": 0.0682,
      "theoretical_loss": 3.455561376690246,
      "tokens_seen": 2023227392
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039070775156475684,
      "loss": 0.0678,
      "theoretical_loss": 3.4555244407266024,
      "tokens_seen": 2023489536
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039062750762317445,
      "loss": 0.0688,
      "theoretical_loss": 3.455487510887342,
      "tokens_seen": 2023751680
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039054726368159206,
      "loss": 0.0689,
      "theoretical_loss": 3.455450587170656,
      "tokens_seen": 2024013824
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003904670197400096,
      "loss": 0.0705,
      "theoretical_loss": 3.455413669574737,
      "tokens_seen": 2024275968
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039038677579842723,
      "loss": 0.068,
      "theoretical_loss": 3.4553767580977777,
      "tokens_seen": 2024538112
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039030653185684485,
      "loss": 0.0681,
      "theoretical_loss": 3.4553398527379717,
      "tokens_seen": 2024800256
    },
    {
      "epoch": 0.61,
      "objective/train/advantage_avg": 0.0008612438105046749,
      "objective/train/docs_used": 738608,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2844445705413818,
      "objective/train/original_loss": 1.2844444513320923,
      "objective/train/theoretical_loss": 3.4553029534935136,
      "objective/train/tokens_used": 2045522400,
      "objective/train/value_avg": -0.00919342041015625,
      "objective/train/value_loss": 0.00018717643979471177,
      "objective/train/value_max": -5.7816505432128906e-05,
      "objective/train/value_min": -0.8095703125,
      "objective/train/value_reward_corr": 0.85092619597115,
      "objective/train/value_std": 0.0222930908203125,
      "objective/train/weight_avg": 1.0009496212005615,
      "objective/train/weighted_lm_loss": 1.2855502367019653,
      "objective/train/weights_max": 1.3055201768875122,
      "objective/train/weights_min": 0.3694390654563904,
      "theoretical_loss": 3.4553029534935136,
      "tokens_seen": 2025062400
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003902262879152624,
      "loss": 0.0669,
      "theoretical_loss": 3.4553029534935136,
      "tokens_seen": 2025062400
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00039014604397367996,
      "loss": 0.0669,
      "theoretical_loss": 3.4552660603625984,
      "tokens_seen": 2025324544
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003900658000320976,
      "loss": 0.067,
      "theoretical_loss": 3.455229173343423,
      "tokens_seen": 2025586688
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003899855560905152,
      "loss": 0.068,
      "theoretical_loss": 3.455192292434183,
      "tokens_seen": 2025848832
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00038990531214893275,
      "loss": 0.0657,
      "theoretical_loss": 3.455155417633076,
      "tokens_seen": 2026110976
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00038982506820735036,
      "loss": 0.0652,
      "theoretical_loss": 3.4551185489383007,
      "tokens_seen": 2026373120
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00038974482426576797,
      "loss": 0.0675,
      "theoretical_loss": 3.4550816863480565,
      "tokens_seen": 2026635264
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00038966458032418553,
      "loss": 0.0678,
      "theoretical_loss": 3.455044829860543,
      "tokens_seen": 2026897408
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003895843363826031,
      "loss": 0.0691,
      "theoretical_loss": 3.45500797947396,
      "tokens_seen": 2027159552
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003895040924410207,
      "loss": 0.0679,
      "theoretical_loss": 3.45497113518651,
      "tokens_seen": 2027421696
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003894238484994383,
      "loss": 0.0699,
      "theoretical_loss": 3.4549342969963943,
      "tokens_seen": 2027683840
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003893436045578559,
      "loss": 0.0678,
      "theoretical_loss": 3.4548974649018165,
      "tokens_seen": 2027945984
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003892633606162735,
      "loss": 0.0684,
      "theoretical_loss": 3.4548606389009793,
      "tokens_seen": 2028208128
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003891831166746911,
      "loss": 0.0654,
      "theoretical_loss": 3.4548238189920877,
      "tokens_seen": 2028470272
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00038910287273310866,
      "loss": 0.0661,
      "theoretical_loss": 3.4547870051733467,
      "tokens_seen": 2028732416
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.0003890226287915262,
      "loss": 0.0674,
      "theoretical_loss": 3.4547501974429626,
      "tokens_seen": 2028994560
    },
    {
      "epoch": 0.61,
      "learning_rate": 0.00038894238484994383,
      "loss": 0.066,
      "theoretical_loss": 3.454713395799142,
      "tokens_seen": 2029256704
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038886214090836144,
      "loss": 0.068,
      "theoretical_loss": 3.4546766002400915,
      "tokens_seen": 2029518848
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.000388781896966779,
      "loss": 0.0674,
      "theoretical_loss": 3.4546398107640197,
      "tokens_seen": 2029780992
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003887016530251966,
      "loss": 0.0701,
      "theoretical_loss": 3.4546030273691364,
      "tokens_seen": 2030043136
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003886214090836142,
      "loss": 0.0659,
      "theoretical_loss": 3.4545662500536505,
      "tokens_seen": 2030305280
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003885411651420318,
      "loss": 0.071,
      "theoretical_loss": 3.4545294788157728,
      "tokens_seen": 2030567424
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038846092120044934,
      "loss": 0.0687,
      "theoretical_loss": 3.454492713653714,
      "tokens_seen": 2030829568
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038838067725886695,
      "loss": 0.0693,
      "theoretical_loss": 3.4544559545656863,
      "tokens_seen": 2031091712
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038830043331728457,
      "loss": 0.0663,
      "theoretical_loss": 3.4544192015499027,
      "tokens_seen": 2031353856
    },
    {
      "epoch": 0.62,
      "objective/train/advantage_avg": 0.0007523357635363936,
      "objective/train/docs_used": 741020,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3411290645599365,
      "objective/train/original_loss": 1.341128945350647,
      "objective/train/theoretical_loss": 3.454382454604577,
      "objective/train/tokens_used": 2052076000,
      "objective/train/value_avg": -0.007038116455078125,
      "objective/train/value_loss": 0.00034059895551763475,
      "objective/train/value_max": -5.066394805908203e-05,
      "objective/train/value_min": -0.73388671875,
      "objective/train/value_reward_corr": 0.7047397016521502,
      "objective/train/value_std": 0.0164031982421875,
      "objective/train/weight_avg": 1.0008991956710815,
      "objective/train/weighted_lm_loss": 1.3422584533691406,
      "objective/train/weights_max": 1.5874963998794556,
      "objective/train/weights_min": 0.36873507499694824,
      "theoretical_loss": 3.454382454604577,
      "tokens_seen": 2031616000
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003882201893757021,
      "loss": 0.0668,
      "theoretical_loss": 3.454382454604577,
      "tokens_seen": 2031616000
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038813994543411974,
      "loss": 0.0666,
      "theoretical_loss": 3.454345713727923,
      "tokens_seen": 2031878144
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038805970149253735,
      "loss": 0.0685,
      "theoretical_loss": 3.4543089789181556,
      "tokens_seen": 2032140288
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003879794575509549,
      "loss": 0.0661,
      "theoretical_loss": 3.4542722501734904,
      "tokens_seen": 2032402432
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003878992136093725,
      "loss": 0.0681,
      "theoretical_loss": 3.454235527492145,
      "tokens_seen": 2032664576
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003878189696677901,
      "loss": 0.0645,
      "theoretical_loss": 3.4541988108723354,
      "tokens_seen": 2032926720
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038773872572620764,
      "loss": 0.0698,
      "theoretical_loss": 3.4541621003122804,
      "tokens_seen": 2033188864
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038765848178462525,
      "loss": 0.0707,
      "theoretical_loss": 3.4541253958101983,
      "tokens_seen": 2033451008
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038757823784304286,
      "loss": 0.0679,
      "theoretical_loss": 3.454088697364309,
      "tokens_seen": 2033713152
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003874979939014605,
      "loss": 0.0651,
      "theoretical_loss": 3.454052004972833,
      "tokens_seen": 2033975296
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038741774995987803,
      "loss": 0.068,
      "theoretical_loss": 3.4540153186339912,
      "tokens_seen": 2034237440
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038733750601829565,
      "loss": 0.0685,
      "theoretical_loss": 3.4539786383460047,
      "tokens_seen": 2034499584
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003872572620767132,
      "loss": 0.0667,
      "theoretical_loss": 3.4539419641070968,
      "tokens_seen": 2034761728
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038717701813513076,
      "loss": 0.0676,
      "theoretical_loss": 3.4539052959154906,
      "tokens_seen": 2035023872
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003870967741935484,
      "loss": 0.0689,
      "theoretical_loss": 3.4538686337694102,
      "tokens_seen": 2035286016
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.000387016530251966,
      "loss": 0.0687,
      "theoretical_loss": 3.45383197766708,
      "tokens_seen": 2035548160
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003869362863103836,
      "loss": 0.0644,
      "theoretical_loss": 3.453795327606726,
      "tokens_seen": 2035810304
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038685604236880116,
      "loss": 0.0687,
      "theoretical_loss": 3.4537586835865746,
      "tokens_seen": 2036072448
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038677579842721877,
      "loss": 0.0698,
      "theoretical_loss": 3.4537220456048523,
      "tokens_seen": 2036334592
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003866955544856364,
      "loss": 0.0683,
      "theoretical_loss": 3.453685413659788,
      "tokens_seen": 2036596736
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003866153105440539,
      "loss": 0.0682,
      "theoretical_loss": 3.4536487877496085,
      "tokens_seen": 2036858880
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003865350666024715,
      "loss": 0.0695,
      "theoretical_loss": 3.4536121678725444,
      "tokens_seen": 2037121024
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003864548226608891,
      "loss": 0.0677,
      "theoretical_loss": 3.453575554026825,
      "tokens_seen": 2037383168
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003863745787193067,
      "loss": 0.0652,
      "theoretical_loss": 3.4535389462106822,
      "tokens_seen": 2037645312
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003862943347777243,
      "loss": 0.0686,
      "theoretical_loss": 3.4535023444223465,
      "tokens_seen": 2037907456
    },
    {
      "epoch": 0.62,
      "objective/train/advantage_avg": -0.00029666503542102873,
      "objective/train/docs_used": 743253,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3327980041503906,
      "objective/train/original_loss": 1.3327980041503906,
      "objective/train/theoretical_loss": 3.453465748660051,
      "objective/train/tokens_used": 2058629600,
      "objective/train/value_avg": -0.007541656494140625,
      "objective/train/value_loss": 0.0001606981095392257,
      "objective/train/value_max": -4.5418739318847656e-05,
      "objective/train/value_min": -0.45166015625,
      "objective/train/value_reward_corr": 0.7653297176662357,
      "objective/train/value_std": 0.01525115966796875,
      "objective/train/weight_avg": 0.9997789859771729,
      "objective/train/weighted_lm_loss": 1.3319969177246094,
      "objective/train/weights_max": 1.4667052030563354,
      "objective/train/weights_min": 0.3728525638580322,
      "theoretical_loss": 3.453465748660051,
      "tokens_seen": 2038169600
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003862140908361419,
      "loss": 0.0676,
      "theoretical_loss": 3.453465748660051,
      "tokens_seen": 2038169600
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003861338468945595,
      "loss": 0.0671,
      "theoretical_loss": 3.4534291589220274,
      "tokens_seen": 2038431744
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.000386053602952977,
      "loss": 0.0685,
      "theoretical_loss": 3.4533925752065104,
      "tokens_seen": 2038693888
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038597335901139463,
      "loss": 0.0684,
      "theoretical_loss": 3.4533559975117347,
      "tokens_seen": 2038956032
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038589311506981224,
      "loss": 0.0697,
      "theoretical_loss": 3.453319425835935,
      "tokens_seen": 2039218176
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038581287112822985,
      "loss": 0.0667,
      "theoretical_loss": 3.4532828601773478,
      "tokens_seen": 2039480320
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003857326271866474,
      "loss": 0.0664,
      "theoretical_loss": 3.453246300534209,
      "tokens_seen": 2039742464
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.000385652383245065,
      "loss": 0.0695,
      "theoretical_loss": 3.4532097469047573,
      "tokens_seen": 2040004608
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038557213930348264,
      "loss": 0.0683,
      "theoretical_loss": 3.4531731992872303,
      "tokens_seen": 2040266752
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038549189536190014,
      "loss": 0.0668,
      "theoretical_loss": 3.4531366576798668,
      "tokens_seen": 2040528896
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038541165142031775,
      "loss": 0.0699,
      "theoretical_loss": 3.4531001220809068,
      "tokens_seen": 2040791040
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038533140747873537,
      "loss": 0.0684,
      "theoretical_loss": 3.453063592488591,
      "tokens_seen": 2041053184
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003852511635371529,
      "loss": 0.0663,
      "theoretical_loss": 3.4530270689011595,
      "tokens_seen": 2041315328
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038517091959557054,
      "loss": 0.0692,
      "theoretical_loss": 3.4529905513168555,
      "tokens_seen": 2041577472
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038509067565398815,
      "loss": 0.0675,
      "theoretical_loss": 3.452954039733921,
      "tokens_seen": 2041839616
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038501043171240576,
      "loss": 0.0653,
      "theoretical_loss": 3.4529175341505995,
      "tokens_seen": 2042101760
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003849301877708233,
      "loss": 0.0675,
      "theoretical_loss": 3.4528810345651357,
      "tokens_seen": 2042363904
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003848499438292409,
      "loss": 0.0677,
      "theoretical_loss": 3.4528445409757738,
      "tokens_seen": 2042626048
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003847696998876585,
      "loss": 0.0692,
      "theoretical_loss": 3.45280805338076,
      "tokens_seen": 2042888192
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038468945594607605,
      "loss": 0.0658,
      "theoretical_loss": 3.45277157177834,
      "tokens_seen": 2043150336
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038460921200449366,
      "loss": 0.0664,
      "theoretical_loss": 3.4527350961667613,
      "tokens_seen": 2043412480
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003845289680629113,
      "loss": 0.0669,
      "theoretical_loss": 3.452698626544272,
      "tokens_seen": 2043674624
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003844487241213289,
      "loss": 0.0674,
      "theoretical_loss": 3.45266216290912,
      "tokens_seen": 2043936768
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038436848017974645,
      "loss": 0.0681,
      "theoretical_loss": 3.452625705259556,
      "tokens_seen": 2044198912
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.000384288236238164,
      "loss": 0.0686,
      "theoretical_loss": 3.4525892535938283,
      "tokens_seen": 2044461056
    },
    {
      "epoch": 0.62,
      "objective/train/advantage_avg": 4.471498687053099e-05,
      "objective/train/docs_used": 745676,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2721974849700928,
      "objective/train/original_loss": 1.2721973657608032,
      "objective/train/theoretical_loss": 3.4525528079101884,
      "objective/train/tokens_used": 2065183200,
      "objective/train/value_avg": -0.006793975830078125,
      "objective/train/value_loss": 0.00025227837613783777,
      "objective/train/value_max": -4.398822784423828e-05,
      "objective/train/value_min": -0.92236328125,
      "objective/train/value_reward_corr": 0.6778463347287009,
      "objective/train/value_std": 0.0141448974609375,
      "objective/train/weight_avg": 1.0001567602157593,
      "objective/train/weighted_lm_loss": 1.272079348564148,
      "objective/train/weights_max": 1.4578360319137573,
      "objective/train/weights_min": 0.3682824373245239,
      "theoretical_loss": 3.4525528079101884,
      "tokens_seen": 2044723200
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003842079922965816,
      "loss": 0.068,
      "theoretical_loss": 3.4525528079101884,
      "tokens_seen": 2044723200
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003841277483549992,
      "loss": 0.0681,
      "theoretical_loss": 3.4525163682068882,
      "tokens_seen": 2044985344
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003840475044134168,
      "loss": 0.0706,
      "theoretical_loss": 3.45247993448218,
      "tokens_seen": 2045247488
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003839672604718344,
      "loss": 0.067,
      "theoretical_loss": 3.4524435067343164,
      "tokens_seen": 2045509632
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.000383887016530252,
      "loss": 0.0643,
      "theoretical_loss": 3.452407084961551,
      "tokens_seen": 2045771776
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038380677258866957,
      "loss": 0.0643,
      "theoretical_loss": 3.452370669162139,
      "tokens_seen": 2046033920
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038372652864708713,
      "loss": 0.0678,
      "theoretical_loss": 3.452334259334335,
      "tokens_seen": 2046296064
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038364628470550474,
      "loss": 0.0676,
      "theoretical_loss": 3.452297855476395,
      "tokens_seen": 2046558208
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003835660407639223,
      "loss": 0.0668,
      "theoretical_loss": 3.4522614575865753,
      "tokens_seen": 2046820352
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003834857968223399,
      "loss": 0.0646,
      "theoretical_loss": 3.452225065663134,
      "tokens_seen": 2047082496
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003834055528807575,
      "loss": 0.0687,
      "theoretical_loss": 3.4521886797043293,
      "tokens_seen": 2047344640
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003833253089391751,
      "loss": 0.0661,
      "theoretical_loss": 3.4521522997084197,
      "tokens_seen": 2047606784
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003832450649975927,
      "loss": 0.0683,
      "theoretical_loss": 3.452115925673665,
      "tokens_seen": 2047868928
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003831648210560103,
      "loss": 0.063,
      "theoretical_loss": 3.4520795575983247,
      "tokens_seen": 2048131072
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038308457711442787,
      "loss": 0.071,
      "theoretical_loss": 3.4520431954806607,
      "tokens_seen": 2048393216
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003830043331728454,
      "loss": 0.068,
      "theoretical_loss": 3.452006839318935,
      "tokens_seen": 2048655360
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038292408923126304,
      "loss": 0.0687,
      "theoretical_loss": 3.45197048911141,
      "tokens_seen": 2048917504
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038284384528968065,
      "loss": 0.0693,
      "theoretical_loss": 3.451934144856348,
      "tokens_seen": 2049179648
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003827636013480982,
      "loss": 0.0681,
      "theoretical_loss": 3.451897806552014,
      "tokens_seen": 2049441792
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003826833574065158,
      "loss": 0.0692,
      "theoretical_loss": 3.451861474196672,
      "tokens_seen": 2049703936
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038260311346493344,
      "loss": 0.0686,
      "theoretical_loss": 3.4518251477885884,
      "tokens_seen": 2049966080
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.000382522869523351,
      "loss": 0.0713,
      "theoretical_loss": 3.4517888273260287,
      "tokens_seen": 2050228224
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038244262558176855,
      "loss": 0.0677,
      "theoretical_loss": 3.4517525128072593,
      "tokens_seen": 2050490368
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038236238164018616,
      "loss": 0.0669,
      "theoretical_loss": 3.4517162042305483,
      "tokens_seen": 2050752512
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003822821376986038,
      "loss": 0.0671,
      "theoretical_loss": 3.4516799015941646,
      "tokens_seen": 2051014656
    },
    {
      "epoch": 0.62,
      "objective/train/advantage_avg": 0.0005698420573025942,
      "objective/train/docs_used": 748156,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3751286268234253,
      "objective/train/original_loss": 1.3751286268234253,
      "objective/train/theoretical_loss": 3.451643604896377,
      "objective/train/tokens_used": 2071736800,
      "objective/train/value_avg": -0.010772705078125,
      "objective/train/value_loss": 0.00031448592199012637,
      "objective/train/value_max": -7.545948028564453e-05,
      "objective/train/value_min": -0.966796875,
      "objective/train/value_reward_corr": 0.7301633754645884,
      "objective/train/value_std": 0.0200653076171875,
      "objective/train/weight_avg": 1.0007132291793823,
      "objective/train/weighted_lm_loss": 1.3746947050094604,
      "objective/train/weights_max": 1.821274757385254,
      "objective/train/weights_min": 0.370111882686615,
      "theoretical_loss": 3.451643604896377,
      "tokens_seen": 2051276800
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038220189375702134,
      "loss": 0.0682,
      "theoretical_loss": 3.451643604896377,
      "tokens_seen": 2051276800
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038212164981543895,
      "loss": 0.0699,
      "theoretical_loss": 3.4516073141354546,
      "tokens_seen": 2051538944
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038204140587385656,
      "loss": 0.0691,
      "theoretical_loss": 3.451571029309668,
      "tokens_seen": 2051801088
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003819611619322742,
      "loss": 0.0666,
      "theoretical_loss": 3.4515347504172893,
      "tokens_seen": 2052063232
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003818809179906917,
      "loss": 0.0686,
      "theoretical_loss": 3.4514984774565898,
      "tokens_seen": 2052325376
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003818006740491093,
      "loss": 0.0676,
      "theoretical_loss": 3.4514622104258423,
      "tokens_seen": 2052587520
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003817204301075269,
      "loss": 0.0655,
      "theoretical_loss": 3.451425949323321,
      "tokens_seen": 2052849664
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038164018616594446,
      "loss": 0.0668,
      "theoretical_loss": 3.451389694147298,
      "tokens_seen": 2053111808
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003815599422243621,
      "loss": 0.068,
      "theoretical_loss": 3.45135344489605,
      "tokens_seen": 2053373952
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003814796982827797,
      "loss": 0.0691,
      "theoretical_loss": 3.4513172015678526,
      "tokens_seen": 2053636096
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038139945434119725,
      "loss": 0.0651,
      "theoretical_loss": 3.451280964160981,
      "tokens_seen": 2053898240
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003813192103996148,
      "loss": 0.0677,
      "theoretical_loss": 3.451244732673713,
      "tokens_seen": 2054160384
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003812389664580324,
      "loss": 0.0657,
      "theoretical_loss": 3.451208507104326,
      "tokens_seen": 2054422528
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038115872251645003,
      "loss": 0.0675,
      "theoretical_loss": 3.451172287451098,
      "tokens_seen": 2054684672
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003810784785748676,
      "loss": 0.0697,
      "theoretical_loss": 3.45113607371231,
      "tokens_seen": 2054946816
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003809982346332852,
      "loss": 0.0673,
      "theoretical_loss": 3.4510998658862397,
      "tokens_seen": 2055208960
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003809179906917028,
      "loss": 0.0673,
      "theoretical_loss": 3.451063663971169,
      "tokens_seen": 2055471104
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038083774675012037,
      "loss": 0.0684,
      "theoretical_loss": 3.4510274679653787,
      "tokens_seen": 2055733248
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038075750280853793,
      "loss": 0.0686,
      "theoretical_loss": 3.4509912778671517,
      "tokens_seen": 2055995392
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038067725886695554,
      "loss": 0.0665,
      "theoretical_loss": 3.4509550936747697,
      "tokens_seen": 2056257536
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038059701492537315,
      "loss": 0.0697,
      "theoretical_loss": 3.4509189153865165,
      "tokens_seen": 2056519680
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003805167709837907,
      "loss": 0.0683,
      "theoretical_loss": 3.450882743000677,
      "tokens_seen": 2056781824
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003804365270422083,
      "loss": 0.0706,
      "theoretical_loss": 3.450846576515535,
      "tokens_seen": 2057043968
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038035628310062594,
      "loss": 0.0682,
      "theoretical_loss": 3.4508104159293773,
      "tokens_seen": 2057306112
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003802760391590435,
      "loss": 0.0696,
      "theoretical_loss": 3.4507742612404897,
      "tokens_seen": 2057568256
    },
    {
      "epoch": 0.62,
      "objective/train/advantage_avg": 0.0001572015753481537,
      "objective/train/docs_used": 750571,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4363138675689697,
      "objective/train/original_loss": 1.4363138675689697,
      "objective/train/theoretical_loss": 3.4507381124471594,
      "objective/train/tokens_used": 2078290400,
      "objective/train/value_avg": -0.00695037841796875,
      "objective/train/value_loss": 0.00017691683024168015,
      "objective/train/value_max": -3.427267074584961e-05,
      "objective/train/value_min": -0.2410888671875,
      "objective/train/value_reward_corr": 0.6409099064873156,
      "objective/train/value_std": 0.01142120361328125,
      "objective/train/weight_avg": 1.000238060951233,
      "objective/train/weighted_lm_loss": 1.4368598461151123,
      "objective/train/weights_max": 1.2355972528457642,
      "objective/train/weights_min": 0.37141308188438416,
      "theoretical_loss": 3.4507381124471594,
      "tokens_seen": 2057830400
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003801957952174611,
      "loss": 0.0687,
      "theoretical_loss": 3.4507381124471594,
      "tokens_seen": 2057830400
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00038011555127587867,
      "loss": 0.0676,
      "theoretical_loss": 3.4507019695476737,
      "tokens_seen": 2058092544
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003800353073342963,
      "loss": 0.0685,
      "theoretical_loss": 3.450665832540322,
      "tokens_seen": 2058354688
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00037995506339271384,
      "loss": 0.0672,
      "theoretical_loss": 3.450629701423393,
      "tokens_seen": 2058616832
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00037987481945113145,
      "loss": 0.0705,
      "theoretical_loss": 3.4505935761951765,
      "tokens_seen": 2058878976
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00037979457550954906,
      "loss": 0.0681,
      "theoretical_loss": 3.4505574568539634,
      "tokens_seen": 2059141120
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003797143315679666,
      "loss": 0.0665,
      "theoretical_loss": 3.4505213433980453,
      "tokens_seen": 2059403264
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00037963408762638423,
      "loss": 0.0699,
      "theoretical_loss": 3.450485235825714,
      "tokens_seen": 2059665408
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003795538436848018,
      "loss": 0.067,
      "theoretical_loss": 3.450449134135262,
      "tokens_seen": 2059927552
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00037947359974321935,
      "loss": 0.0673,
      "theoretical_loss": 3.4504130383249834,
      "tokens_seen": 2060189696
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00037939335580163696,
      "loss": 0.0676,
      "theoretical_loss": 3.4503769483931723,
      "tokens_seen": 2060451840
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003793131118600546,
      "loss": 0.0667,
      "theoretical_loss": 3.4503408643381235,
      "tokens_seen": 2060713984
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003792328679184722,
      "loss": 0.066,
      "theoretical_loss": 3.4503047861581324,
      "tokens_seen": 2060976128
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00037915262397688975,
      "loss": 0.069,
      "theoretical_loss": 3.4502687138514956,
      "tokens_seen": 2061238272
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.00037907238003530736,
      "loss": 0.0687,
      "theoretical_loss": 3.4502326474165104,
      "tokens_seen": 2061500416
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003789921360937249,
      "loss": 0.0693,
      "theoretical_loss": 3.450196586851474,
      "tokens_seen": 2061762560
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003789118921521425,
      "loss": 0.069,
      "theoretical_loss": 3.450160532154685,
      "tokens_seen": 2062024704
    },
    {
      "epoch": 0.62,
      "learning_rate": 0.0003788316482105601,
      "loss": 0.0675,
      "theoretical_loss": 3.4501244833244438,
      "tokens_seen": 2062286848
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003787514042689777,
      "loss": 0.0687,
      "theoretical_loss": 3.450088440359049,
      "tokens_seen": 2062548992
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003786711603273953,
      "loss": 0.0666,
      "theoretical_loss": 3.450052403256801,
      "tokens_seen": 2062811136
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003785909163858129,
      "loss": 0.0673,
      "theoretical_loss": 3.450016372016002,
      "tokens_seen": 2063073280
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003785106724442305,
      "loss": 0.0694,
      "theoretical_loss": 3.4499803466349537,
      "tokens_seen": 2063335424
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003784304285026481,
      "loss": 0.0695,
      "theoretical_loss": 3.449944327111959,
      "tokens_seen": 2063597568
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003783501845610656,
      "loss": 0.0678,
      "theoretical_loss": 3.449908313445321,
      "tokens_seen": 2063859712
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003782699406194832,
      "loss": 0.0677,
      "theoretical_loss": 3.4498723056333445,
      "tokens_seen": 2064121856
    },
    {
      "epoch": 0.63,
      "objective/train/advantage_avg": 0.0011079288087785244,
      "objective/train/docs_used": 752866,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4321738481521606,
      "objective/train/original_loss": 1.4321739673614502,
      "objective/train/theoretical_loss": 3.4498363036743336,
      "objective/train/tokens_used": 2084844000,
      "objective/train/value_avg": -0.00701141357421875,
      "objective/train/value_loss": 0.0003250784066040069,
      "objective/train/value_max": -4.9114227294921875e-05,
      "objective/train/value_min": -0.75634765625,
      "objective/train/value_reward_corr": 0.65196970836296,
      "objective/train/value_std": 0.01462554931640625,
      "objective/train/weight_avg": 1.0012527704238892,
      "objective/train/weighted_lm_loss": 1.43328058719635,
      "objective/train/weights_max": 1.950271725654602,
      "objective/train/weights_min": 0.3689524829387665,
      "theoretical_loss": 3.4498363036743336,
      "tokens_seen": 2064384000
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037818969667790083,
      "loss": 0.0716,
      "theoretical_loss": 3.4498363036743336,
      "tokens_seen": 2064384000
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037810945273631844,
      "loss": 0.0667,
      "theoretical_loss": 3.449800307566594,
      "tokens_seen": 2064646144
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.000378029208794736,
      "loss": 0.0681,
      "theoretical_loss": 3.449764317308432,
      "tokens_seen": 2064908288
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003779489648531536,
      "loss": 0.0703,
      "theoretical_loss": 3.449728332898155,
      "tokens_seen": 2065170432
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003778687209115712,
      "loss": 0.0662,
      "theoretical_loss": 3.4496923543340703,
      "tokens_seen": 2065432576
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037778847696998873,
      "loss": 0.0692,
      "theoretical_loss": 3.4496563816144867,
      "tokens_seen": 2065694720
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037770823302840634,
      "loss": 0.0676,
      "theoretical_loss": 3.449620414737713,
      "tokens_seen": 2065956864
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037762798908682395,
      "loss": 0.0661,
      "theoretical_loss": 3.449584453702059,
      "tokens_seen": 2066219008
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003775477451452415,
      "loss": 0.0675,
      "theoretical_loss": 3.449548498505834,
      "tokens_seen": 2066481152
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003774675012036591,
      "loss": 0.0714,
      "theoretical_loss": 3.4495125491473515,
      "tokens_seen": 2066743296
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037738725726207674,
      "loss": 0.0674,
      "theoretical_loss": 3.449476605624922,
      "tokens_seen": 2067005440
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037730701332049435,
      "loss": 0.0682,
      "theoretical_loss": 3.4494406679368583,
      "tokens_seen": 2067267584
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003772267693789119,
      "loss": 0.0696,
      "theoretical_loss": 3.449404736081474,
      "tokens_seen": 2067529728
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037714652543732947,
      "loss": 0.0713,
      "theoretical_loss": 3.4493688100570825,
      "tokens_seen": 2067791872
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003770662814957471,
      "loss": 0.0702,
      "theoretical_loss": 3.4493328898619993,
      "tokens_seen": 2068054016
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037698603755416464,
      "loss": 0.0676,
      "theoretical_loss": 3.449296975494539,
      "tokens_seen": 2068316160
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037690579361258225,
      "loss": 0.0667,
      "theoretical_loss": 3.449261066953018,
      "tokens_seen": 2068578304
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037682554967099986,
      "loss": 0.0687,
      "theoretical_loss": 3.4492251642357536,
      "tokens_seen": 2068840448
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003767453057294175,
      "loss": 0.0695,
      "theoretical_loss": 3.4491892673410627,
      "tokens_seen": 2069102592
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037666506178783503,
      "loss": 0.069,
      "theoretical_loss": 3.449153376267264,
      "tokens_seen": 2069364736
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003765848178462526,
      "loss": 0.0681,
      "theoretical_loss": 3.449117491012676,
      "tokens_seen": 2069626880
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003765045739046702,
      "loss": 0.0674,
      "theoretical_loss": 3.449081611575618,
      "tokens_seen": 2069889024
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037642432996308776,
      "loss": 0.0675,
      "theoretical_loss": 3.4490457379544113,
      "tokens_seen": 2070151168
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003763440860215054,
      "loss": 0.0706,
      "theoretical_loss": 3.4490098701473757,
      "tokens_seen": 2070413312
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.000376263842079923,
      "loss": 0.0673,
      "theoretical_loss": 3.448974008152834,
      "tokens_seen": 2070675456
    },
    {
      "epoch": 0.63,
      "objective/train/advantage_avg": 0.0006079924060031772,
      "objective/train/docs_used": 755047,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3184467554092407,
      "objective/train/original_loss": 1.3184466361999512,
      "objective/train/theoretical_loss": 3.4489381519691085,
      "objective/train/tokens_used": 2091397600,
      "objective/train/value_avg": -0.00807952880859375,
      "objective/train/value_loss": 0.0003275515919085592,
      "objective/train/value_max": -5.7816505432128906e-05,
      "objective/train/value_min": -0.6318359375,
      "objective/train/value_reward_corr": 0.6679204773929055,
      "objective/train/value_std": 0.0141143798828125,
      "objective/train/weight_avg": 1.0007538795471191,
      "objective/train/weighted_lm_loss": 1.3194092512130737,
      "objective/train/weights_max": 1.8273200988769531,
      "objective/train/weights_min": 0.3784526288509369,
      "theoretical_loss": 3.4489381519691085,
      "tokens_seen": 2070937600
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003761835981383406,
      "loss": 0.0685,
      "theoretical_loss": 3.4489381519691085,
      "tokens_seen": 2070937600
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037610335419675816,
      "loss": 0.0707,
      "theoretical_loss": 3.4489023015945213,
      "tokens_seen": 2071199744
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003760231102551757,
      "loss": 0.0684,
      "theoretical_loss": 3.448866457027397,
      "tokens_seen": 2071461888
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037594286631359333,
      "loss": 0.0686,
      "theoretical_loss": 3.4488306182660597,
      "tokens_seen": 2071724032
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003758626223720109,
      "loss": 0.0713,
      "theoretical_loss": 3.448794785308835,
      "tokens_seen": 2071986176
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003757823784304285,
      "loss": 0.0691,
      "theoretical_loss": 3.4487589581540483,
      "tokens_seen": 2072248320
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003757021344888461,
      "loss": 0.0677,
      "theoretical_loss": 3.4487231368000266,
      "tokens_seen": 2072510464
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003756218905472637,
      "loss": 0.0676,
      "theoretical_loss": 3.448687321245097,
      "tokens_seen": 2072772608
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003755416466056813,
      "loss": 0.066,
      "theoretical_loss": 3.4486515114875873,
      "tokens_seen": 2073034752
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003754614026640989,
      "loss": 0.0704,
      "theoretical_loss": 3.448615707525826,
      "tokens_seen": 2073296896
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037538115872251646,
      "loss": 0.0687,
      "theoretical_loss": 3.448579909358143,
      "tokens_seen": 2073559040
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.000375300914780934,
      "loss": 0.0687,
      "theoretical_loss": 3.448544116982868,
      "tokens_seen": 2073821184
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037522067083935163,
      "loss": 0.0696,
      "theoretical_loss": 3.448508330398332,
      "tokens_seen": 2074083328
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037514042689776924,
      "loss": 0.0672,
      "theoretical_loss": 3.448472549602866,
      "tokens_seen": 2074345472
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003750601829561868,
      "loss": 0.0651,
      "theoretical_loss": 3.4484367745948026,
      "tokens_seen": 2074607616
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003749799390146044,
      "loss": 0.0698,
      "theoretical_loss": 3.4484010053724736,
      "tokens_seen": 2074869760
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.000374899695073022,
      "loss": 0.0687,
      "theoretical_loss": 3.448365241934214,
      "tokens_seen": 2075131904
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003748194511314396,
      "loss": 0.0674,
      "theoretical_loss": 3.4483294842783563,
      "tokens_seen": 2075394048
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037473920718985714,
      "loss": 0.0695,
      "theoretical_loss": 3.4482937324032368,
      "tokens_seen": 2075656192
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037465896324827475,
      "loss": 0.0709,
      "theoretical_loss": 3.4482579863071905,
      "tokens_seen": 2075918336
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037457871930669237,
      "loss": 0.0681,
      "theoretical_loss": 3.4482222459885534,
      "tokens_seen": 2076180480
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003744984753651099,
      "loss": 0.0678,
      "theoretical_loss": 3.4481865114456625,
      "tokens_seen": 2076442624
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037441823142352754,
      "loss": 0.0671,
      "theoretical_loss": 3.448150782676856,
      "tokens_seen": 2076704768
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037433798748194515,
      "loss": 0.0642,
      "theoretical_loss": 3.4481150596804717,
      "tokens_seen": 2076966912
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003742577435403627,
      "loss": 0.0668,
      "theoretical_loss": 3.4480793424548493,
      "tokens_seen": 2077229056
    },
    {
      "epoch": 0.63,
      "objective/train/advantage_avg": 0.0019641020335257053,
      "objective/train/docs_used": 757304,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3592127561569214,
      "objective/train/original_loss": 1.3592126369476318,
      "objective/train/theoretical_loss": 3.4480436309983276,
      "objective/train/tokens_used": 2097951200,
      "objective/train/value_avg": -0.0092926025390625,
      "objective/train/value_loss": 0.0001754560216795653,
      "objective/train/value_max": -6.657838821411133e-05,
      "objective/train/value_min": -0.55322265625,
      "objective/train/value_reward_corr": 0.7035732028469984,
      "objective/train/value_std": 0.01471710205078125,
      "objective/train/weight_avg": 1.0020464658737183,
      "objective/train/weighted_lm_loss": 1.3608214855194092,
      "objective/train/weights_max": 1.3589965105056763,
      "objective/train/weights_min": 0.36952081322669983,
      "theoretical_loss": 3.4480436309983276,
      "tokens_seen": 2077491200
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037417749959878027,
      "loss": 0.0667,
      "theoretical_loss": 3.4480436309983276,
      "tokens_seen": 2077491200
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003740972556571979,
      "loss": 0.0689,
      "theoretical_loss": 3.448007925309247,
      "tokens_seen": 2077753344
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003740170117156155,
      "loss": 0.0695,
      "theoretical_loss": 3.447972225385949,
      "tokens_seen": 2078015488
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037393676777403305,
      "loss": 0.0651,
      "theoretical_loss": 3.447936531226776,
      "tokens_seen": 2078277632
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037385652383245066,
      "loss": 0.0652,
      "theoretical_loss": 3.4479008428300686,
      "tokens_seen": 2078539776
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003737762798908683,
      "loss": 0.0678,
      "theoretical_loss": 3.447865160194171,
      "tokens_seen": 2078801920
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003736960359492859,
      "loss": 0.0688,
      "theoretical_loss": 3.447829483317428,
      "tokens_seen": 2079064064
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003736157920077034,
      "loss": 0.0656,
      "theoretical_loss": 3.4477938121981824,
      "tokens_seen": 2079326208
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.000373535548066121,
      "loss": 0.0645,
      "theoretical_loss": 3.44775814683478,
      "tokens_seen": 2079588352
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003734553041245386,
      "loss": 0.0665,
      "theoretical_loss": 3.447722487225567,
      "tokens_seen": 2079850496
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003733750601829562,
      "loss": 0.064,
      "theoretical_loss": 3.4476868333688904,
      "tokens_seen": 2080112640
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003732948162413738,
      "loss": 0.0684,
      "theoretical_loss": 3.447651185263096,
      "tokens_seen": 2080374784
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003732145722997914,
      "loss": 0.0677,
      "theoretical_loss": 3.447615542906532,
      "tokens_seen": 2080636928
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037313432835820896,
      "loss": 0.0693,
      "theoretical_loss": 3.4475799062975483,
      "tokens_seen": 2080899072
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003730540844166265,
      "loss": 0.0679,
      "theoretical_loss": 3.4475442754344927,
      "tokens_seen": 2081161216
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037297384047504413,
      "loss": 0.0641,
      "theoretical_loss": 3.447508650315716,
      "tokens_seen": 2081423360
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037289359653346174,
      "loss": 0.0664,
      "theoretical_loss": 3.4474730309395687,
      "tokens_seen": 2081685504
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003728133525918793,
      "loss": 0.0708,
      "theoretical_loss": 3.4474374173044025,
      "tokens_seen": 2081947648
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003727331086502969,
      "loss": 0.0684,
      "theoretical_loss": 3.4474018094085683,
      "tokens_seen": 2082209792
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003726528647087145,
      "loss": 0.0689,
      "theoretical_loss": 3.4473662072504196,
      "tokens_seen": 2082471936
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003725726207671321,
      "loss": 0.0676,
      "theoretical_loss": 3.4473306108283097,
      "tokens_seen": 2082734080
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003724923768255497,
      "loss": 0.0677,
      "theoretical_loss": 3.4472950201405923,
      "tokens_seen": 2082996224
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037241213288396726,
      "loss": 0.0656,
      "theoretical_loss": 3.4472594351856225,
      "tokens_seen": 2083258368
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037233188894238487,
      "loss": 0.0687,
      "theoretical_loss": 3.447223855961756,
      "tokens_seen": 2083520512
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003722516450008024,
      "loss": 0.0677,
      "theoretical_loss": 3.447188282467348,
      "tokens_seen": 2083782656
    },
    {
      "epoch": 0.63,
      "objective/train/advantage_avg": 0.001002849661745131,
      "objective/train/docs_used": 759560,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4999350309371948,
      "objective/train/original_loss": 1.4999351501464844,
      "objective/train/theoretical_loss": 3.4471527147007555,
      "objective/train/tokens_used": 2104504800,
      "objective/train/value_avg": -0.0079498291015625,
      "objective/train/value_loss": 0.00027177634183317423,
      "objective/train/value_max": -4.7206878662109375e-05,
      "objective/train/value_min": -0.74853515625,
      "objective/train/value_reward_corr": 0.6737963482430739,
      "objective/train/value_std": 0.016357421875,
      "objective/train/weight_avg": 1.0011271238327026,
      "objective/train/weighted_lm_loss": 1.5015414953231812,
      "objective/train/weights_max": 1.7329145669937134,
      "objective/train/weights_min": 0.38661831617355347,
      "theoretical_loss": 3.4471527147007555,
      "tokens_seen": 2084044800
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037217140105922004,
      "loss": 0.0712,
      "theoretical_loss": 3.4471527147007555,
      "tokens_seen": 2084044800
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037209115711763765,
      "loss": 0.0672,
      "theoretical_loss": 3.447117152660337,
      "tokens_seen": 2084306944
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003720109131760552,
      "loss": 0.0686,
      "theoretical_loss": 3.447081596344449,
      "tokens_seen": 2084569088
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003719306692344728,
      "loss": 0.0678,
      "theoretical_loss": 3.447046045751451,
      "tokens_seen": 2084831232
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003718504252928904,
      "loss": 0.066,
      "theoretical_loss": 3.447010500879703,
      "tokens_seen": 2085093376
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.000371770181351308,
      "loss": 0.0648,
      "theoretical_loss": 3.4469749617275642,
      "tokens_seen": 2085355520
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037168993740972555,
      "loss": 0.066,
      "theoretical_loss": 3.4469394282933967,
      "tokens_seen": 2085617664
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037160969346814316,
      "loss": 0.0699,
      "theoretical_loss": 3.4469039005755606,
      "tokens_seen": 2085879808
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003715294495265608,
      "loss": 0.069,
      "theoretical_loss": 3.446868378572419,
      "tokens_seen": 2086141952
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037144920558497834,
      "loss": 0.0693,
      "theoretical_loss": 3.446832862282334,
      "tokens_seen": 2086404096
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037136896164339595,
      "loss": 0.0656,
      "theoretical_loss": 3.44679735170367,
      "tokens_seen": 2086666240
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003712887177018135,
      "loss": 0.0656,
      "theoretical_loss": 3.4467618468347903,
      "tokens_seen": 2086928384
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037120847376023107,
      "loss": 0.0667,
      "theoretical_loss": 3.44672634767406,
      "tokens_seen": 2087190528
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003711282298186487,
      "loss": 0.0668,
      "theoretical_loss": 3.4466908542198453,
      "tokens_seen": 2087452672
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003710479858770663,
      "loss": 0.0679,
      "theoretical_loss": 3.4466553664705124,
      "tokens_seen": 2087714816
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003709677419354839,
      "loss": 0.0669,
      "theoretical_loss": 3.446619884424427,
      "tokens_seen": 2087976960
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037088749799390146,
      "loss": 0.0678,
      "theoretical_loss": 3.446584408079958,
      "tokens_seen": 2088239104
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003708072540523191,
      "loss": 0.0692,
      "theoretical_loss": 3.4465489374354727,
      "tokens_seen": 2088501248
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003707270101107367,
      "loss": 0.0687,
      "theoretical_loss": 3.4465134724893405,
      "tokens_seen": 2088763392
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003706467661691542,
      "loss": 0.0681,
      "theoretical_loss": 3.446478013239931,
      "tokens_seen": 2089025536
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003705665222275718,
      "loss": 0.0694,
      "theoretical_loss": 3.4464425596856136,
      "tokens_seen": 2089287680
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003704862782859894,
      "loss": 0.0691,
      "theoretical_loss": 3.446407111824761,
      "tokens_seen": 2089549824
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037040603434440703,
      "loss": 0.0684,
      "theoretical_loss": 3.4463716696557425,
      "tokens_seen": 2089811968
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003703257904028246,
      "loss": 0.0675,
      "theoretical_loss": 3.4463362331769325,
      "tokens_seen": 2090074112
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003702455464612422,
      "loss": 0.0671,
      "theoretical_loss": 3.4463008023867028,
      "tokens_seen": 2090336256
    },
    {
      "epoch": 0.63,
      "objective/train/advantage_avg": -0.00044457524199970067,
      "objective/train/docs_used": 762048,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4181301593780518,
      "objective/train/original_loss": 1.4181302785873413,
      "objective/train/theoretical_loss": 3.4462653772834266,
      "objective/train/tokens_used": 2111058400,
      "objective/train/value_avg": -0.00865936279296875,
      "objective/train/value_loss": 0.0003374247462488711,
      "objective/train/value_max": -2.9087066650390625e-05,
      "objective/train/value_min": -0.91943359375,
      "objective/train/value_reward_corr": 0.747619425608935,
      "objective/train/value_std": 0.0189666748046875,
      "objective/train/weight_avg": 0.9997140765190125,
      "objective/train/weighted_lm_loss": 1.4173901081085205,
      "objective/train/weights_max": 1.5935001373291016,
      "objective/train/weights_min": 0.3686169385910034,
      "theoretical_loss": 3.4462653772834266,
      "tokens_seen": 2090598400
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003701653025196598,
      "loss": 0.0694,
      "theoretical_loss": 3.4462653772834266,
      "tokens_seen": 2090598400
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003700850585780773,
      "loss": 0.0679,
      "theoretical_loss": 3.446229957865479,
      "tokens_seen": 2090860544
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00037000481463649493,
      "loss": 0.069,
      "theoretical_loss": 3.4461945441312354,
      "tokens_seen": 2091122688
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00036992457069491254,
      "loss": 0.067,
      "theoretical_loss": 3.4461591360790704,
      "tokens_seen": 2091384832
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00036984432675333015,
      "loss": 0.0657,
      "theoretical_loss": 3.44612373370736,
      "tokens_seen": 2091646976
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003697640828117477,
      "loss": 0.0687,
      "theoretical_loss": 3.446088337014482,
      "tokens_seen": 2091909120
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003696838388701653,
      "loss": 0.0684,
      "theoretical_loss": 3.446052945998814,
      "tokens_seen": 2092171264
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00036960359492858294,
      "loss": 0.068,
      "theoretical_loss": 3.446017560658734,
      "tokens_seen": 2092433408
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00036952335098700044,
      "loss": 0.0673,
      "theoretical_loss": 3.4459821809926208,
      "tokens_seen": 2092695552
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00036944310704541805,
      "loss": 0.0667,
      "theoretical_loss": 3.4459468069988537,
      "tokens_seen": 2092957696
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00036936286310383567,
      "loss": 0.0666,
      "theoretical_loss": 3.445911438675814,
      "tokens_seen": 2093219840
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003692826191622532,
      "loss": 0.0638,
      "theoretical_loss": 3.445876076021882,
      "tokens_seen": 2093481984
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00036920237522067084,
      "loss": 0.0683,
      "theoretical_loss": 3.4458407190354388,
      "tokens_seen": 2093744128
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00036912213127908845,
      "loss": 0.0666,
      "theoretical_loss": 3.445805367714868,
      "tokens_seen": 2094006272
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00036904188733750606,
      "loss": 0.0662,
      "theoretical_loss": 3.445770022058551,
      "tokens_seen": 2094268416
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003689616433959236,
      "loss": 0.0678,
      "theoretical_loss": 3.445734682064873,
      "tokens_seen": 2094530560
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003688813994543412,
      "loss": 0.0664,
      "theoretical_loss": 3.4456993477322166,
      "tokens_seen": 2094792704
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.0003688011555127588,
      "loss": 0.0667,
      "theoretical_loss": 3.4456640190589676,
      "tokens_seen": 2095054848
    },
    {
      "epoch": 0.63,
      "learning_rate": 0.00036872091157117635,
      "loss": 0.0685,
      "theoretical_loss": 3.445628696043512,
      "tokens_seen": 2095316992
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036864066762959396,
      "loss": 0.0664,
      "theoretical_loss": 3.4455933786842348,
      "tokens_seen": 2095579136
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003685604236880116,
      "loss": 0.0699,
      "theoretical_loss": 3.4455580669795243,
      "tokens_seen": 2095841280
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003684801797464292,
      "loss": 0.0666,
      "theoretical_loss": 3.4455227609277674,
      "tokens_seen": 2096103424
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036839993580484675,
      "loss": 0.0709,
      "theoretical_loss": 3.445487460527352,
      "tokens_seen": 2096365568
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003683196918632643,
      "loss": 0.0693,
      "theoretical_loss": 3.4454521657766675,
      "tokens_seen": 2096627712
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003682394479216819,
      "loss": 0.0698,
      "theoretical_loss": 3.445416876674104,
      "tokens_seen": 2096889856
    },
    {
      "epoch": 0.64,
      "objective/train/advantage_avg": 0.0006668668356724083,
      "objective/train/docs_used": 764413,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4101420640945435,
      "objective/train/original_loss": 1.4101420640945435,
      "objective/train/theoretical_loss": 3.44538159321805,
      "objective/train/tokens_used": 2117612000,
      "objective/train/value_avg": -0.00644683837890625,
      "objective/train/value_loss": 0.0001926971017383039,
      "objective/train/value_max": -4.2319297790527344e-05,
      "objective/train/value_min": -0.8056640625,
      "objective/train/value_reward_corr": 0.6798013959111557,
      "objective/train/value_std": 0.0125732421875,
      "objective/train/weight_avg": 1.00075101852417,
      "objective/train/weighted_lm_loss": 1.4113109111785889,
      "objective/train/weights_max": 1.2963095903396606,
      "objective/train/weights_min": 0.22838670015335083,
      "theoretical_loss": 3.44538159321805,
      "tokens_seen": 2097152000
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003681592039800995,
      "loss": 0.0677,
      "theoretical_loss": 3.44538159321805,
      "tokens_seen": 2097152000
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003680789600385171,
      "loss": 0.0676,
      "theoretical_loss": 3.4453463154068977,
      "tokens_seen": 2097414144
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003679987160969347,
      "loss": 0.0691,
      "theoretical_loss": 3.4453110432390384,
      "tokens_seen": 2097676288
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003679184721553523,
      "loss": 0.0683,
      "theoretical_loss": 3.4452757767128643,
      "tokens_seen": 2097938432
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003678382282137699,
      "loss": 0.0644,
      "theoretical_loss": 3.445240515826768,
      "tokens_seen": 2098200576
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003677579842721875,
      "loss": 0.0694,
      "theoretical_loss": 3.4452052605791432,
      "tokens_seen": 2098462720
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036767774033060504,
      "loss": 0.0672,
      "theoretical_loss": 3.4451700109683836,
      "tokens_seen": 2098724864
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003675974963890226,
      "loss": 0.0685,
      "theoretical_loss": 3.4451347669928856,
      "tokens_seen": 2098987008
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003675172524474402,
      "loss": 0.0691,
      "theoretical_loss": 3.4450995286510424,
      "tokens_seen": 2099249152
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036743700850585783,
      "loss": 0.0689,
      "theoretical_loss": 3.445064295941252,
      "tokens_seen": 2099511296
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036735676456427544,
      "loss": 0.0667,
      "theoretical_loss": 3.4450290688619103,
      "tokens_seen": 2099773440
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.000367276520622693,
      "loss": 0.0637,
      "theoretical_loss": 3.444993847411415,
      "tokens_seen": 2100035584
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003671962766811106,
      "loss": 0.0689,
      "theoretical_loss": 3.4449586315881637,
      "tokens_seen": 2100297728
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036711603273952817,
      "loss": 0.065,
      "theoretical_loss": 3.4449234213905564,
      "tokens_seen": 2100559872
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036703578879794573,
      "loss": 0.0652,
      "theoretical_loss": 3.4448882168169908,
      "tokens_seen": 2100822016
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036695554485636334,
      "loss": 0.0654,
      "theoretical_loss": 3.444853017865869,
      "tokens_seen": 2101084160
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036687530091478095,
      "loss": 0.0646,
      "theoretical_loss": 3.4448178245355896,
      "tokens_seen": 2101346304
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003667950569731985,
      "loss": 0.0657,
      "theoretical_loss": 3.4447826368245558,
      "tokens_seen": 2101608448
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003667148130316161,
      "loss": 0.0659,
      "theoretical_loss": 3.4447474547311683,
      "tokens_seen": 2101870592
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036663456909003374,
      "loss": 0.0669,
      "theoretical_loss": 3.4447122782538306,
      "tokens_seen": 2102132736
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003665543251484513,
      "loss": 0.0655,
      "theoretical_loss": 3.444677107390946,
      "tokens_seen": 2102394880
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036647408120686885,
      "loss": 0.0644,
      "theoretical_loss": 3.444641942140918,
      "tokens_seen": 2102657024
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036639383726528647,
      "loss": 0.0662,
      "theoretical_loss": 3.4446067825021514,
      "tokens_seen": 2102919168
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003663135933237041,
      "loss": 0.0657,
      "theoretical_loss": 3.444571628473052,
      "tokens_seen": 2103181312
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036623334938212164,
      "loss": 0.0662,
      "theoretical_loss": 3.4445364800520255,
      "tokens_seen": 2103443456
    },
    {
      "epoch": 0.64,
      "objective/train/advantage_avg": -0.0007185638532973826,
      "objective/train/docs_used": 766743,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4546356201171875,
      "objective/train/original_loss": 1.4546356201171875,
      "objective/train/theoretical_loss": 3.444501337237478,
      "objective/train/tokens_used": 2124165600,
      "objective/train/value_avg": -0.0100555419921875,
      "objective/train/value_loss": 0.00047717380221001804,
      "objective/train/value_max": -3.218650817871094e-05,
      "objective/train/value_min": -0.8232421875,
      "objective/train/value_reward_corr": 0.7003188218258403,
      "objective/train/value_std": 0.0203704833984375,
      "objective/train/weight_avg": 0.999488353729248,
      "objective/train/weighted_lm_loss": 1.453125238418579,
      "objective/train/weights_max": 1.6461470127105713,
      "objective/train/weights_min": 0.23570120334625244,
      "theoretical_loss": 3.444501337237478,
      "tokens_seen": 2103705600
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036615310544053925,
      "loss": 0.0694,
      "theoretical_loss": 3.444501337237478,
      "tokens_seen": 2103705600
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036607286149895686,
      "loss": 0.0668,
      "theoretical_loss": 3.4444662000278177,
      "tokens_seen": 2103967744
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003659926175573745,
      "loss": 0.0627,
      "theoretical_loss": 3.4444310684214514,
      "tokens_seen": 2104229888
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.000365912373615792,
      "loss": 0.064,
      "theoretical_loss": 3.444395942416789,
      "tokens_seen": 2104492032
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003658321296742096,
      "loss": 0.0654,
      "theoretical_loss": 3.4443608220122384,
      "tokens_seen": 2104754176
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003657518857326272,
      "loss": 0.0668,
      "theoretical_loss": 3.44432570720621,
      "tokens_seen": 2105016320
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036567164179104476,
      "loss": 0.0623,
      "theoretical_loss": 3.4442905979971146,
      "tokens_seen": 2105278464
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003655913978494624,
      "loss": 0.0672,
      "theoretical_loss": 3.4442554943833628,
      "tokens_seen": 2105540608
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036551115390788,
      "loss": 0.0673,
      "theoretical_loss": 3.444220396363367,
      "tokens_seen": 2105802752
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003654309099662976,
      "loss": 0.0642,
      "theoretical_loss": 3.444185303935539,
      "tokens_seen": 2106064896
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003653506660247151,
      "loss": 0.0659,
      "theoretical_loss": 3.4441502170982927,
      "tokens_seen": 2106327040
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003652704220831327,
      "loss": 0.0689,
      "theoretical_loss": 3.444115135850041,
      "tokens_seen": 2106589184
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036519017814155033,
      "loss": 0.0686,
      "theoretical_loss": 3.444080060189199,
      "tokens_seen": 2106851328
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003651099341999679,
      "loss": 0.0692,
      "theoretical_loss": 3.444044990114181,
      "tokens_seen": 2107113472
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003650296902583855,
      "loss": 0.0677,
      "theoretical_loss": 3.4440099256234036,
      "tokens_seen": 2107375616
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003649494463168031,
      "loss": 0.0689,
      "theoretical_loss": 3.4439748667152825,
      "tokens_seen": 2107637760
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036486920237522067,
      "loss": 0.0667,
      "theoretical_loss": 3.443939813388235,
      "tokens_seen": 2107899904
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036478895843363823,
      "loss": 0.0659,
      "theoretical_loss": 3.4439047656406783,
      "tokens_seen": 2108162048
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036470871449205584,
      "loss": 0.0693,
      "theoretical_loss": 3.443869723471031,
      "tokens_seen": 2108424192
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036462847055047346,
      "loss": 0.067,
      "theoretical_loss": 3.443834686877712,
      "tokens_seen": 2108686336
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.000364548226608891,
      "loss": 0.0675,
      "theoretical_loss": 3.443799655859141,
      "tokens_seen": 2108948480
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036446798266730863,
      "loss": 0.0675,
      "theoretical_loss": 3.443764630413738,
      "tokens_seen": 2109210624
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036438773872572624,
      "loss": 0.0656,
      "theoretical_loss": 3.4437296105399238,
      "tokens_seen": 2109472768
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003643074947841438,
      "loss": 0.0693,
      "theoretical_loss": 3.4436945962361203,
      "tokens_seen": 2109734912
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003642272508425614,
      "loss": 0.0665,
      "theoretical_loss": 3.4436595875007487,
      "tokens_seen": 2109997056
    },
    {
      "epoch": 0.64,
      "objective/train/advantage_avg": -6.150845001684502e-05,
      "objective/train/docs_used": 769058,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.341748595237732,
      "objective/train/original_loss": 1.341748595237732,
      "objective/train/theoretical_loss": 3.443624584332233,
      "objective/train/tokens_used": 2130719200,
      "objective/train/value_avg": -0.0086822509765625,
      "objective/train/value_loss": 0.00023672039969824255,
      "objective/train/value_max": -4.398822784423828e-05,
      "objective/train/value_min": -0.70703125,
      "objective/train/value_reward_corr": 0.7215973019732559,
      "objective/train/value_std": 0.0160980224609375,
      "objective/train/weight_avg": 1.0000429153442383,
      "objective/train/weighted_lm_loss": 1.3412035703659058,
      "objective/train/weights_max": 1.2542415857315063,
      "objective/train/weights_min": 0.3687976598739624,
      "theoretical_loss": 3.443624584332233,
      "tokens_seen": 2110259200
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036414700690097897,
      "loss": 0.0673,
      "theoretical_loss": 3.443624584332233,
      "tokens_seen": 2110259200
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003640667629593966,
      "loss": 0.0678,
      "theoretical_loss": 3.4435895867289963,
      "tokens_seen": 2110521344
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036398651901781414,
      "loss": 0.0668,
      "theoretical_loss": 3.443554594689462,
      "tokens_seen": 2110783488
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036390627507623175,
      "loss": 0.0638,
      "theoretical_loss": 3.443519608212055,
      "tokens_seen": 2111045632
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036382603113464937,
      "loss": 0.0638,
      "theoretical_loss": 3.4434846272952013,
      "tokens_seen": 2111307776
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003637457871930669,
      "loss": 0.0669,
      "theoretical_loss": 3.4434496519373265,
      "tokens_seen": 2111569920
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036366554325148454,
      "loss": 0.0684,
      "theoretical_loss": 3.443414682136857,
      "tokens_seen": 2111832064
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003635852993099021,
      "loss": 0.0664,
      "theoretical_loss": 3.4433797178922205,
      "tokens_seen": 2112094208
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003635050553683197,
      "loss": 0.0646,
      "theoretical_loss": 3.4433447592018447,
      "tokens_seen": 2112356352
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036342481142673727,
      "loss": 0.0673,
      "theoretical_loss": 3.443309806064158,
      "tokens_seen": 2112618496
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003633445674851549,
      "loss": 0.0663,
      "theoretical_loss": 3.44327485847759,
      "tokens_seen": 2112880640
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003632643235435725,
      "loss": 0.0669,
      "theoretical_loss": 3.44323991644057,
      "tokens_seen": 2113142784
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036318407960199005,
      "loss": 0.065,
      "theoretical_loss": 3.443204979951529,
      "tokens_seen": 2113404928
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036310383566040766,
      "loss": 0.0642,
      "theoretical_loss": 3.4431700490088977,
      "tokens_seen": 2113667072
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003630235917188253,
      "loss": 0.0675,
      "theoretical_loss": 3.443135123611108,
      "tokens_seen": 2113929216
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003629433477772428,
      "loss": 0.0669,
      "theoretical_loss": 3.4431002037565923,
      "tokens_seen": 2114191360
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003628631038356604,
      "loss": 0.0657,
      "theoretical_loss": 3.4430652894437834,
      "tokens_seen": 2114453504
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.000362782859894078,
      "loss": 0.0698,
      "theoretical_loss": 3.4430303806711158,
      "tokens_seen": 2114715648
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003627026159524956,
      "loss": 0.0678,
      "theoretical_loss": 3.442995477437023,
      "tokens_seen": 2114977792
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003626223720109132,
      "loss": 0.066,
      "theoretical_loss": 3.44296057973994,
      "tokens_seen": 2115239936
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003625421280693308,
      "loss": 0.067,
      "theoretical_loss": 3.442925687578302,
      "tokens_seen": 2115502080
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003624618841277484,
      "loss": 0.0674,
      "theoretical_loss": 3.442890800950546,
      "tokens_seen": 2115764224
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003623816401861659,
      "loss": 0.0637,
      "theoretical_loss": 3.442855919855109,
      "tokens_seen": 2116026368
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003623013962445835,
      "loss": 0.0638,
      "theoretical_loss": 3.4428210442904277,
      "tokens_seen": 2116288512
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036222115230300113,
      "loss": 0.0654,
      "theoretical_loss": 3.4427861742549406,
      "tokens_seen": 2116550656
    },
    {
      "epoch": 0.64,
      "objective/train/advantage_avg": 0.0007084216340444982,
      "objective/train/docs_used": 771474,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3125371932983398,
      "objective/train/original_loss": 1.3125371932983398,
      "objective/train/theoretical_loss": 3.442751309747086,
      "objective/train/tokens_used": 2137272800,
      "objective/train/value_avg": -0.007656097412109375,
      "objective/train/value_loss": 0.0002038901875494048,
      "objective/train/value_max": -5.4776668548583984e-05,
      "objective/train/value_min": -0.394287109375,
      "objective/train/value_reward_corr": 0.7548622300814178,
      "objective/train/value_std": 0.0170745849609375,
      "objective/train/weight_avg": 1.000800609588623,
      "objective/train/weighted_lm_loss": 1.3134653568267822,
      "objective/train/weights_max": 1.2419233322143555,
      "objective/train/weights_min": 0.38047337532043457,
      "theoretical_loss": 3.442751309747086,
      "tokens_seen": 2116812800
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036214090836141874,
      "loss": 0.0648,
      "theoretical_loss": 3.442751309747086,
      "tokens_seen": 2116812800
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003620606644198363,
      "loss": 0.0664,
      "theoretical_loss": 3.442716450765304,
      "tokens_seen": 2117074944
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003619804204782539,
      "loss": 0.0669,
      "theoretical_loss": 3.4426815973080345,
      "tokens_seen": 2117337088
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003619001765366715,
      "loss": 0.0686,
      "theoretical_loss": 3.4426467493737176,
      "tokens_seen": 2117599232
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036181993259508903,
      "loss": 0.0664,
      "theoretical_loss": 3.4426119069607948,
      "tokens_seen": 2117861376
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036173968865350664,
      "loss": 0.0676,
      "theoretical_loss": 3.4425770700677085,
      "tokens_seen": 2118123520
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036165944471192426,
      "loss": 0.0644,
      "theoretical_loss": 3.4425422386929005,
      "tokens_seen": 2118385664
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036157920077034187,
      "loss": 0.0656,
      "theoretical_loss": 3.4425074128348148,
      "tokens_seen": 2118647808
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003614989568287594,
      "loss": 0.0665,
      "theoretical_loss": 3.442472592491894,
      "tokens_seen": 2118909952
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036141871288717704,
      "loss": 0.0671,
      "theoretical_loss": 3.442437777662584,
      "tokens_seen": 2119172096
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036133846894559465,
      "loss": 0.0679,
      "theoretical_loss": 3.4424029683453288,
      "tokens_seen": 2119434240
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003612582250040122,
      "loss": 0.0683,
      "theoretical_loss": 3.442368164538575,
      "tokens_seen": 2119696384
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036117798106242977,
      "loss": 0.0655,
      "theoretical_loss": 3.4423333662407676,
      "tokens_seen": 2119958528
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003610977371208474,
      "loss": 0.0673,
      "theoretical_loss": 3.442298573450355,
      "tokens_seen": 2120220672
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036101749317926494,
      "loss": 0.0661,
      "theoretical_loss": 3.4422637861657837,
      "tokens_seen": 2120482816
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036093724923768255,
      "loss": 0.066,
      "theoretical_loss": 3.442229004385502,
      "tokens_seen": 2120744960
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036085700529610017,
      "loss": 0.0661,
      "theoretical_loss": 3.44219422810796,
      "tokens_seen": 2121007104
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003607767613545178,
      "loss": 0.0646,
      "theoretical_loss": 3.4421594573316057,
      "tokens_seen": 2121269248
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036069651741293534,
      "loss": 0.0672,
      "theoretical_loss": 3.44212469205489,
      "tokens_seen": 2121531392
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003606162734713529,
      "loss": 0.0693,
      "theoretical_loss": 3.4420899322762635,
      "tokens_seen": 2121793536
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003605360295297705,
      "loss": 0.0682,
      "theoretical_loss": 3.4420551779941775,
      "tokens_seen": 2122055680
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036045578558818807,
      "loss": 0.0676,
      "theoretical_loss": 3.4420204292070844,
      "tokens_seen": 2122317824
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003603755416466057,
      "loss": 0.066,
      "theoretical_loss": 3.4419856859134357,
      "tokens_seen": 2122579968
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003602952977050233,
      "loss": 0.0664,
      "theoretical_loss": 3.4419509481116854,
      "tokens_seen": 2122842112
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003602150537634409,
      "loss": 0.0691,
      "theoretical_loss": 3.441916215800288,
      "tokens_seen": 2123104256
    },
    {
      "epoch": 0.64,
      "objective/train/advantage_avg": -0.0004949113936163485,
      "objective/train/docs_used": 773937,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2716563940048218,
      "objective/train/original_loss": 1.2716562747955322,
      "objective/train/theoretical_loss": 3.441881488977697,
      "objective/train/tokens_used": 2143826400,
      "objective/train/value_avg": -0.01212310791015625,
      "objective/train/value_loss": 0.0005141652654856443,
      "objective/train/value_max": -2.1278858184814453e-05,
      "objective/train/value_min": -0.85888671875,
      "objective/train/value_reward_corr": 0.7393573499560793,
      "objective/train/value_std": 0.0230712890625,
      "objective/train/weight_avg": 0.9997339248657227,
      "objective/train/weighted_lm_loss": 1.2710819244384766,
      "objective/train/weights_max": 1.5846818685531616,
      "objective/train/weights_min": 0.3693798780441284,
      "theoretical_loss": 3.441881488977697,
      "tokens_seen": 2123366400
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00036013480982185846,
      "loss": 0.0636,
      "theoretical_loss": 3.441881488977697,
      "tokens_seen": 2123366400
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.000360054565880276,
      "loss": 0.0633,
      "theoretical_loss": 3.441846767642368,
      "tokens_seen": 2123628544
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00035997432193869363,
      "loss": 0.0662,
      "theoretical_loss": 3.4418120517927564,
      "tokens_seen": 2123890688
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003598940779971112,
      "loss": 0.0687,
      "theoretical_loss": 3.441777341427319,
      "tokens_seen": 2124152832
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003598138340555288,
      "loss": 0.0703,
      "theoretical_loss": 3.4417426365445127,
      "tokens_seen": 2124414976
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003597335901139464,
      "loss": 0.0644,
      "theoretical_loss": 3.4417079371427945,
      "tokens_seen": 2124677120
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00035965334617236403,
      "loss": 0.0666,
      "theoretical_loss": 3.441673243220624,
      "tokens_seen": 2124939264
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003595731022307816,
      "loss": 0.0655,
      "theoretical_loss": 3.4416385547764583,
      "tokens_seen": 2125201408
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003594928582891992,
      "loss": 0.0699,
      "theoretical_loss": 3.4416038718087583,
      "tokens_seen": 2125463552
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00035941261434761676,
      "loss": 0.0671,
      "theoretical_loss": 3.4415691943159836,
      "tokens_seen": 2125725696
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003593323704060343,
      "loss": 0.0672,
      "theoretical_loss": 3.441534522296595,
      "tokens_seen": 2125987840
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00035925212646445193,
      "loss": 0.0643,
      "theoretical_loss": 3.441499855749054,
      "tokens_seen": 2126249984
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00035917188252286954,
      "loss": 0.0649,
      "theoretical_loss": 3.441465194671822,
      "tokens_seen": 2126512128
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00035909163858128715,
      "loss": 0.0665,
      "theoretical_loss": 3.4414305390633624,
      "tokens_seen": 2126774272
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003590113946397047,
      "loss": 0.0649,
      "theoretical_loss": 3.4413958889221385,
      "tokens_seen": 2127036416
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003589311506981223,
      "loss": 0.0653,
      "theoretical_loss": 3.4413612442466133,
      "tokens_seen": 2127298560
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.0003588509067565399,
      "loss": 0.0668,
      "theoretical_loss": 3.441326605035252,
      "tokens_seen": 2127560704
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00035877066281495744,
      "loss": 0.0671,
      "theoretical_loss": 3.441291971286519,
      "tokens_seen": 2127822848
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00035869041887337506,
      "loss": 0.0643,
      "theoretical_loss": 3.44125734299888,
      "tokens_seen": 2128084992
    },
    {
      "epoch": 0.64,
      "learning_rate": 0.00035861017493179267,
      "loss": 0.068,
      "theoretical_loss": 3.4412227201708028,
      "tokens_seen": 2128347136
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003585299309902102,
      "loss": 0.0681,
      "theoretical_loss": 3.4411881028007527,
      "tokens_seen": 2128609280
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035844968704862784,
      "loss": 0.0647,
      "theoretical_loss": 3.4411534908871984,
      "tokens_seen": 2128871424
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035836944310704545,
      "loss": 0.068,
      "theoretical_loss": 3.4411188844286067,
      "tokens_seen": 2129133568
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035828919916546306,
      "loss": 0.0677,
      "theoretical_loss": 3.441084283423448,
      "tokens_seen": 2129395712
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035820895522388057,
      "loss": 0.0692,
      "theoretical_loss": 3.4410496878701906,
      "tokens_seen": 2129657856
    },
    {
      "epoch": 0.65,
      "objective/train/advantage_avg": 0.00014313262363430113,
      "objective/train/docs_used": 775953,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3686505556106567,
      "objective/train/original_loss": 1.3686505556106567,
      "objective/train/theoretical_loss": 3.441015097767305,
      "objective/train/tokens_used": 2150380000,
      "objective/train/value_avg": -0.00724029541015625,
      "objective/train/value_loss": 0.00016575136396568269,
      "objective/train/value_max": -6.866455078125e-05,
      "objective/train/value_min": -0.2313232421875,
      "objective/train/value_reward_corr": 0.7221669709099741,
      "objective/train/value_std": 0.01239776611328125,
      "objective/train/weight_avg": 1.0002208948135376,
      "objective/train/weighted_lm_loss": 1.3682907819747925,
      "objective/train/weights_max": 1.2552000284194946,
      "objective/train/weights_min": 0.3698423504829407,
      "theoretical_loss": 3.441015097767305,
      "tokens_seen": 2129920000
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003581287112822982,
      "loss": 0.0652,
      "theoretical_loss": 3.441015097767305,
      "tokens_seen": 2129920000
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003580484673407158,
      "loss": 0.069,
      "theoretical_loss": 3.440980513113262,
      "tokens_seen": 2130182144
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035796822339913335,
      "loss": 0.0679,
      "theoretical_loss": 3.4409459339065327,
      "tokens_seen": 2130444288
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035788797945755096,
      "loss": 0.0705,
      "theoretical_loss": 3.4409113601455887,
      "tokens_seen": 2130706432
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003578077355159686,
      "loss": 0.0665,
      "theoretical_loss": 3.4408767918289027,
      "tokens_seen": 2130968576
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003577274915743862,
      "loss": 0.0666,
      "theoretical_loss": 3.4408422289549483,
      "tokens_seen": 2131230720
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003576472476328037,
      "loss": 0.0657,
      "theoretical_loss": 3.4408076715221982,
      "tokens_seen": 2131492864
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003575670036912213,
      "loss": 0.0659,
      "theoretical_loss": 3.440773119529128,
      "tokens_seen": 2131755008
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003574867597496389,
      "loss": 0.0678,
      "theoretical_loss": 3.440738572974212,
      "tokens_seen": 2132017152
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003574065158080565,
      "loss": 0.0691,
      "theoretical_loss": 3.440704031855926,
      "tokens_seen": 2132279296
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003573262718664741,
      "loss": 0.0695,
      "theoretical_loss": 3.440669496172746,
      "tokens_seen": 2132541440
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003572460279248917,
      "loss": 0.0684,
      "theoretical_loss": 3.4406349659231488,
      "tokens_seen": 2132803584
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003571657839833093,
      "loss": 0.064,
      "theoretical_loss": 3.4406004411056124,
      "tokens_seen": 2133065728
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003570855400417268,
      "loss": 0.0643,
      "theoretical_loss": 3.4405659217186138,
      "tokens_seen": 2133327872
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035700529610014443,
      "loss": 0.0643,
      "theoretical_loss": 3.440531407760633,
      "tokens_seen": 2133590016
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035692505215856204,
      "loss": 0.0674,
      "theoretical_loss": 3.4404968992301477,
      "tokens_seen": 2133852160
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003568448082169796,
      "loss": 0.0641,
      "theoretical_loss": 3.440462396125639,
      "tokens_seen": 2134114304
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003567645642753972,
      "loss": 0.068,
      "theoretical_loss": 3.4404278984455874,
      "tokens_seen": 2134376448
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035668432033381483,
      "loss": 0.069,
      "theoretical_loss": 3.4403934061884733,
      "tokens_seen": 2134638592
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003566040763922324,
      "loss": 0.068,
      "theoretical_loss": 3.4403589193527786,
      "tokens_seen": 2134900736
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035652383245065,
      "loss": 0.0677,
      "theoretical_loss": 3.440324437936986,
      "tokens_seen": 2135162880
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035644358850906756,
      "loss": 0.0662,
      "theoretical_loss": 3.4402899619395786,
      "tokens_seen": 2135425024
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035636334456748517,
      "loss": 0.0632,
      "theoretical_loss": 3.440255491359039,
      "tokens_seen": 2135687168
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035628310062590273,
      "loss": 0.0672,
      "theoretical_loss": 3.440221026193852,
      "tokens_seen": 2135949312
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035620285668432034,
      "loss": 0.0645,
      "theoretical_loss": 3.4401865664425024,
      "tokens_seen": 2136211456
    },
    {
      "epoch": 0.65,
      "objective/train/advantage_avg": 0.0002571629884187132,
      "objective/train/docs_used": 778288,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3589222431182861,
      "objective/train/original_loss": 1.3589223623275757,
      "objective/train/theoretical_loss": 3.440152112103476,
      "objective/train/tokens_used": 2156933600,
      "objective/train/value_avg": -0.00910186767578125,
      "objective/train/value_loss": 0.00038075895281508565,
      "objective/train/value_max": -4.363059997558594e-05,
      "objective/train/value_min": -0.90185546875,
      "objective/train/value_reward_corr": 0.6478648047552944,
      "objective/train/value_std": 0.016845703125,
      "objective/train/weight_avg": 1.0004286766052246,
      "objective/train/weighted_lm_loss": 1.358363389968872,
      "objective/train/weights_max": 1.8600143194198608,
      "objective/train/weights_min": 0.3690122961997986,
      "theoretical_loss": 3.440152112103476,
      "tokens_seen": 2136473600
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035612261274273795,
      "loss": 0.0661,
      "theoretical_loss": 3.440152112103476,
      "tokens_seen": 2136473600
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003560423688011555,
      "loss": 0.0669,
      "theoretical_loss": 3.4401176631752577,
      "tokens_seen": 2136735744
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003559621248595731,
      "loss": 0.0658,
      "theoretical_loss": 3.4400832196563353,
      "tokens_seen": 2136997888
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003558818809179907,
      "loss": 0.0669,
      "theoretical_loss": 3.440048781545195,
      "tokens_seen": 2137260032
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003558016369764083,
      "loss": 0.0691,
      "theoretical_loss": 3.440014348840325,
      "tokens_seen": 2137522176
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035572139303482585,
      "loss": 0.0674,
      "theoretical_loss": 3.439979921540214,
      "tokens_seen": 2137784320
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035564114909324347,
      "loss": 0.0666,
      "theoretical_loss": 3.4399454996433514,
      "tokens_seen": 2138046464
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003555609051516611,
      "loss": 0.0686,
      "theoretical_loss": 3.4399110831482256,
      "tokens_seen": 2138308608
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035548066121007864,
      "loss": 0.0673,
      "theoretical_loss": 3.4398766720533276,
      "tokens_seen": 2138570752
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035540041726849625,
      "loss": 0.0692,
      "theoretical_loss": 3.4398422663571484,
      "tokens_seen": 2138832896
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003553201733269138,
      "loss": 0.0663,
      "theoretical_loss": 3.439807866058179,
      "tokens_seen": 2139095040
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003552399293853314,
      "loss": 0.0697,
      "theoretical_loss": 3.439773471154912,
      "tokens_seen": 2139357184
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.000355159685443749,
      "loss": 0.0675,
      "theoretical_loss": 3.4397390816458397,
      "tokens_seen": 2139619328
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003550794415021666,
      "loss": 0.0674,
      "theoretical_loss": 3.4397046975294554,
      "tokens_seen": 2139881472
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003549991975605842,
      "loss": 0.069,
      "theoretical_loss": 3.4396703188042537,
      "tokens_seen": 2140143616
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035491895361900176,
      "loss": 0.0653,
      "theoretical_loss": 3.4396359454687278,
      "tokens_seen": 2140405760
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003548387096774194,
      "loss": 0.0649,
      "theoretical_loss": 3.4396015775213735,
      "tokens_seen": 2140667904
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.000354758465735837,
      "loss": 0.0664,
      "theoretical_loss": 3.439567214960687,
      "tokens_seen": 2140930048
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003546782217942545,
      "loss": 0.0648,
      "theoretical_loss": 3.439532857785164,
      "tokens_seen": 2141192192
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003545979778526721,
      "loss": 0.0658,
      "theoretical_loss": 3.4394985059933014,
      "tokens_seen": 2141454336
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003545177339110897,
      "loss": 0.0667,
      "theoretical_loss": 3.439464159583597,
      "tokens_seen": 2141716480
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035443748996950733,
      "loss": 0.0654,
      "theoretical_loss": 3.4394298185545487,
      "tokens_seen": 2141978624
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003543572460279249,
      "loss": 0.0683,
      "theoretical_loss": 3.439395482904655,
      "tokens_seen": 2142240768
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003542770020863425,
      "loss": 0.0649,
      "theoretical_loss": 3.439361152632416,
      "tokens_seen": 2142502912
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003541967581447601,
      "loss": 0.0662,
      "theoretical_loss": 3.4393268277363305,
      "tokens_seen": 2142765056
    },
    {
      "epoch": 0.65,
      "objective/train/advantage_avg": 0.00045191447134129703,
      "objective/train/docs_used": 780500,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3032023906707764,
      "objective/train/original_loss": 1.3032021522521973,
      "objective/train/theoretical_loss": 3.4392925082149004,
      "objective/train/tokens_used": 2163487200,
      "objective/train/value_avg": -0.006439208984375,
      "objective/train/value_loss": 0.00019448416423983872,
      "objective/train/value_max": -4.363059997558594e-05,
      "objective/train/value_min": -0.46142578125,
      "objective/train/value_reward_corr": 0.6783022603004962,
      "objective/train/value_std": 0.012542724609375,
      "objective/train/weight_avg": 1.0005395412445068,
      "objective/train/weighted_lm_loss": 1.3039817810058594,
      "objective/train/weights_max": 1.1519769430160522,
      "objective/train/weights_min": 0.3689151704311371,
      "theoretical_loss": 3.4392925082149004,
      "tokens_seen": 2143027200
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003541165142031776,
      "loss": 0.0683,
      "theoretical_loss": 3.4392925082149004,
      "tokens_seen": 2143027200
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035403627026159523,
      "loss": 0.0679,
      "theoretical_loss": 3.4392581940666256,
      "tokens_seen": 2143289344
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035395602632001284,
      "loss": 0.0647,
      "theoretical_loss": 3.4392238852900086,
      "tokens_seen": 2143551488
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035387578237843046,
      "loss": 0.0665,
      "theoretical_loss": 3.4391895818835514,
      "tokens_seen": 2143813632
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.000353795538436848,
      "loss": 0.0639,
      "theoretical_loss": 3.439155283845757,
      "tokens_seen": 2144075776
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035371529449526563,
      "loss": 0.0688,
      "theoretical_loss": 3.4391209911751286,
      "tokens_seen": 2144337920
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035363505055368324,
      "loss": 0.0654,
      "theoretical_loss": 3.4390867038701716,
      "tokens_seen": 2144600064
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003535548066121008,
      "loss": 0.0669,
      "theoretical_loss": 3.439052421929389,
      "tokens_seen": 2144862208
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035347456267051836,
      "loss": 0.0646,
      "theoretical_loss": 3.439018145351287,
      "tokens_seen": 2145124352
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035339431872893597,
      "loss": 0.0681,
      "theoretical_loss": 3.4389838741343715,
      "tokens_seen": 2145386496
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003533140747873536,
      "loss": 0.0661,
      "theoretical_loss": 3.438949608277149,
      "tokens_seen": 2145648640
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035323383084577114,
      "loss": 0.065,
      "theoretical_loss": 3.438915347778127,
      "tokens_seen": 2145910784
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035315358690418875,
      "loss": 0.0686,
      "theoretical_loss": 3.4388810926358127,
      "tokens_seen": 2146172928
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035307334296260637,
      "loss": 0.0652,
      "theoretical_loss": 3.4388468428487142,
      "tokens_seen": 2146435072
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003529930990210239,
      "loss": 0.0658,
      "theoretical_loss": 3.4388125984153413,
      "tokens_seen": 2146697216
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003529128550794415,
      "loss": 0.0666,
      "theoretical_loss": 3.4387783593342025,
      "tokens_seen": 2146959360
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003528326111378591,
      "loss": 0.07,
      "theoretical_loss": 3.438744125603809,
      "tokens_seen": 2147221504
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035275236719627665,
      "loss": 0.0652,
      "theoretical_loss": 3.4387098972226706,
      "tokens_seen": 2147483648
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035267212325469427,
      "loss": 0.0623,
      "theoretical_loss": 3.4386756741892994,
      "tokens_seen": 2147745792
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003525918793131119,
      "loss": 0.0694,
      "theoretical_loss": 3.4386414565022063,
      "tokens_seen": 2148007936
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003525116353715295,
      "loss": 0.0669,
      "theoretical_loss": 3.438607244159905,
      "tokens_seen": 2148270080
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035243139142994705,
      "loss": 0.0656,
      "theoretical_loss": 3.4385730371609076,
      "tokens_seen": 2148532224
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003523511474883646,
      "loss": 0.0672,
      "theoretical_loss": 3.438538835503728,
      "tokens_seen": 2148794368
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003522709035467822,
      "loss": 0.0679,
      "theoretical_loss": 3.438504639186881,
      "tokens_seen": 2149056512
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003521906596051998,
      "loss": 0.0654,
      "theoretical_loss": 3.4384704482088813,
      "tokens_seen": 2149318656
    },
    {
      "epoch": 0.65,
      "objective/train/advantage_avg": 0.00025573305902071297,
      "objective/train/docs_used": 782928,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.186272382736206,
      "objective/train/original_loss": 1.186272144317627,
      "objective/train/theoretical_loss": 3.4384362625682434,
      "objective/train/tokens_used": 2170040800,
      "objective/train/value_avg": -0.006961822509765625,
      "objective/train/value_loss": 0.00010668325558071956,
      "objective/train/value_max": -3.427267074584961e-05,
      "objective/train/value_min": -0.501953125,
      "objective/train/value_reward_corr": 0.7916194071649223,
      "objective/train/value_std": 0.012664794921875,
      "objective/train/weight_avg": 1.0003083944320679,
      "objective/train/weighted_lm_loss": 1.1869118213653564,
      "objective/train/weights_max": 1.146925687789917,
      "objective/train/weights_min": 0.8217722177505493,
      "theoretical_loss": 3.4384362625682434,
      "tokens_seen": 2149580800
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003521104156636174,
      "loss": 0.0637,
      "theoretical_loss": 3.4384362625682434,
      "tokens_seen": 2149580800
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.000352030171722035,
      "loss": 0.0656,
      "theoretical_loss": 3.438402082263485,
      "tokens_seen": 2149842944
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003519499277804526,
      "loss": 0.0681,
      "theoretical_loss": 3.4383679072931215,
      "tokens_seen": 2150105088
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003518696838388702,
      "loss": 0.0655,
      "theoretical_loss": 3.438333737655671,
      "tokens_seen": 2150367232
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003517894398972878,
      "loss": 0.0688,
      "theoretical_loss": 3.43829957334965,
      "tokens_seen": 2150629376
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035170919595570535,
      "loss": 0.0655,
      "theoretical_loss": 3.4382654143735785,
      "tokens_seen": 2150891520
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003516289520141229,
      "loss": 0.0679,
      "theoretical_loss": 3.438231260725975,
      "tokens_seen": 2151153664
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003515487080725405,
      "loss": 0.0662,
      "theoretical_loss": 3.4381971124053594,
      "tokens_seen": 2151415808
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035146846413095813,
      "loss": 0.0683,
      "theoretical_loss": 3.438162969410251,
      "tokens_seen": 2151677952
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035138822018937574,
      "loss": 0.0673,
      "theoretical_loss": 3.438128831739171,
      "tokens_seen": 2151940096
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003513079762477933,
      "loss": 0.0687,
      "theoretical_loss": 3.4380946993906414,
      "tokens_seen": 2152202240
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003512277323062109,
      "loss": 0.0664,
      "theoretical_loss": 3.4380605723631836,
      "tokens_seen": 2152464384
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035114748836462847,
      "loss": 0.0676,
      "theoretical_loss": 3.4380264506553204,
      "tokens_seen": 2152726528
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035106724442304603,
      "loss": 0.0669,
      "theoretical_loss": 3.4379923342655747,
      "tokens_seen": 2152988672
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035098700048146364,
      "loss": 0.0671,
      "theoretical_loss": 3.437958223192471,
      "tokens_seen": 2153250816
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035090675653988126,
      "loss": 0.0693,
      "theoretical_loss": 3.4379241174345325,
      "tokens_seen": 2153512960
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035082651259829887,
      "loss": 0.0635,
      "theoretical_loss": 3.437890016990285,
      "tokens_seen": 2153775104
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003507462686567164,
      "loss": 0.0669,
      "theoretical_loss": 3.4378559218582536,
      "tokens_seen": 2154037248
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035066602471513404,
      "loss": 0.0651,
      "theoretical_loss": 3.4378218320369647,
      "tokens_seen": 2154299392
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003505857807735516,
      "loss": 0.0652,
      "theoretical_loss": 3.437787747524945,
      "tokens_seen": 2154561536
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035050553683196916,
      "loss": 0.0665,
      "theoretical_loss": 3.4377536683207217,
      "tokens_seen": 2154823680
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035042529289038677,
      "loss": 0.0672,
      "theoretical_loss": 3.4377195944228225,
      "tokens_seen": 2155085824
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003503450489488044,
      "loss": 0.0654,
      "theoretical_loss": 3.4376855258297763,
      "tokens_seen": 2155347968
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035026480500722194,
      "loss": 0.0664,
      "theoretical_loss": 3.4376514625401113,
      "tokens_seen": 2155610112
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035018456106563955,
      "loss": 0.0665,
      "theoretical_loss": 3.437617404552358,
      "tokens_seen": 2155872256
    },
    {
      "epoch": 0.65,
      "objective/train/advantage_avg": 0.00036321027437224984,
      "objective/train/docs_used": 785270,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3421927690505981,
      "objective/train/original_loss": 1.3421927690505981,
      "objective/train/theoretical_loss": 3.4375833518650465,
      "objective/train/tokens_used": 2176594400,
      "objective/train/value_avg": -0.0075225830078125,
      "objective/train/value_loss": 0.00023389572743326426,
      "objective/train/value_max": -5.346536636352539e-05,
      "objective/train/value_min": -0.8916015625,
      "objective/train/value_reward_corr": 0.7332820517109091,
      "objective/train/value_std": 0.0157470703125,
      "objective/train/weight_avg": 1.000470519065857,
      "objective/train/weighted_lm_loss": 1.3425078392028809,
      "objective/train/weights_max": 2.007188320159912,
      "objective/train/weights_min": 0.24055545032024384,
      "theoretical_loss": 3.4375833518650465,
      "tokens_seen": 2156134400
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00035010431712405717,
      "loss": 0.0679,
      "theoretical_loss": 3.4375833518650465,
      "tokens_seen": 2156134400
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003500240731824748,
      "loss": 0.0662,
      "theoretical_loss": 3.4375493044767076,
      "tokens_seen": 2156396544
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003499438292408923,
      "loss": 0.065,
      "theoretical_loss": 3.4375152623858725,
      "tokens_seen": 2156658688
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003498635852993099,
      "loss": 0.0664,
      "theoretical_loss": 3.437481225591073,
      "tokens_seen": 2156920832
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003497833413577275,
      "loss": 0.0664,
      "theoretical_loss": 3.4374471940908418,
      "tokens_seen": 2157182976
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00034970309741614507,
      "loss": 0.0648,
      "theoretical_loss": 3.4374131678837125,
      "tokens_seen": 2157445120
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003496228534745627,
      "loss": 0.0661,
      "theoretical_loss": 3.4373791469682184,
      "tokens_seen": 2157707264
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003495426095329803,
      "loss": 0.0662,
      "theoretical_loss": 3.437345131342894,
      "tokens_seen": 2157969408
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003494623655913979,
      "loss": 0.0665,
      "theoretical_loss": 3.437311121006274,
      "tokens_seen": 2158231552
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003493821216498154,
      "loss": 0.0669,
      "theoretical_loss": 3.4372771159568942,
      "tokens_seen": 2158493696
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.000349301877708233,
      "loss": 0.0678,
      "theoretical_loss": 3.43724311619329,
      "tokens_seen": 2158755840
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00034922163376665063,
      "loss": 0.0662,
      "theoretical_loss": 3.437209121713999,
      "tokens_seen": 2159017984
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003491413898250682,
      "loss": 0.0669,
      "theoretical_loss": 3.4371751325175586,
      "tokens_seen": 2159280128
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003490611458834858,
      "loss": 0.0656,
      "theoretical_loss": 3.437141148602505,
      "tokens_seen": 2159542272
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003489809019419034,
      "loss": 0.0645,
      "theoretical_loss": 3.437107169967378,
      "tokens_seen": 2159804416
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00034890065800032103,
      "loss": 0.0672,
      "theoretical_loss": 3.437073196610716,
      "tokens_seen": 2160066560
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003488204140587386,
      "loss": 0.0643,
      "theoretical_loss": 3.4370392285310594,
      "tokens_seen": 2160328704
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00034874017011715615,
      "loss": 0.0642,
      "theoretical_loss": 3.437005265726947,
      "tokens_seen": 2160590848
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00034865992617557376,
      "loss": 0.0654,
      "theoretical_loss": 3.4369713081969206,
      "tokens_seen": 2160852992
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.0003485796822339913,
      "loss": 0.068,
      "theoretical_loss": 3.436937355939521,
      "tokens_seen": 2161115136
    },
    {
      "epoch": 0.65,
      "learning_rate": 0.00034849943829240893,
      "loss": 0.0683,
      "theoretical_loss": 3.4369034089532904,
      "tokens_seen": 2161377280
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034841919435082654,
      "loss": 0.0655,
      "theoretical_loss": 3.436869467236771,
      "tokens_seen": 2161639424
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003483389504092441,
      "loss": 0.0644,
      "theoretical_loss": 3.4368355307885063,
      "tokens_seen": 2161901568
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003482587064676617,
      "loss": 0.0667,
      "theoretical_loss": 3.436801599607039,
      "tokens_seen": 2162163712
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034817846252607927,
      "loss": 0.0681,
      "theoretical_loss": 3.4367676736909143,
      "tokens_seen": 2162425856
    },
    {
      "epoch": 0.66,
      "objective/train/advantage_avg": 0.0004280103603377938,
      "objective/train/docs_used": 787745,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.441954493522644,
      "objective/train/original_loss": 1.4419546127319336,
      "objective/train/theoretical_loss": 3.4367337530386766,
      "objective/train/tokens_used": 2183148000,
      "objective/train/value_avg": -0.00835418701171875,
      "objective/train/value_loss": 0.00033444137079641223,
      "objective/train/value_max": -5.829334259033203e-05,
      "objective/train/value_min": -0.71337890625,
      "objective/train/value_reward_corr": 0.7118774960473635,
      "objective/train/value_std": 0.018341064453125,
      "objective/train/weight_avg": 1.0005724430084229,
      "objective/train/weighted_lm_loss": 1.4417777061462402,
      "objective/train/weights_max": 1.43697190284729,
      "objective/train/weights_min": 0.22480729222297668,
      "theoretical_loss": 3.4367337530386766,
      "tokens_seen": 2162688000
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003480982185844969,
      "loss": 0.067,
      "theoretical_loss": 3.4367337530386766,
      "tokens_seen": 2162688000
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034801797464291444,
      "loss": 0.0679,
      "theoretical_loss": 3.436699837648871,
      "tokens_seen": 2162950144
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034793773070133206,
      "loss": 0.0692,
      "theoretical_loss": 3.4366659275200444,
      "tokens_seen": 2163212288
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034785748675974967,
      "loss": 0.0677,
      "theoretical_loss": 3.436632022650742,
      "tokens_seen": 2163474432
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003477772428181672,
      "loss": 0.0683,
      "theoretical_loss": 3.4365981230395115,
      "tokens_seen": 2163736576
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034769699887658484,
      "loss": 0.0653,
      "theoretical_loss": 3.436564228684901,
      "tokens_seen": 2163998720
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003476167549350024,
      "loss": 0.0661,
      "theoretical_loss": 3.436530339585458,
      "tokens_seen": 2164260864
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034753651099342,
      "loss": 0.0668,
      "theoretical_loss": 3.4364964557397317,
      "tokens_seen": 2164523008
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034745626705183757,
      "loss": 0.0663,
      "theoretical_loss": 3.436462577146272,
      "tokens_seen": 2164785152
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003473760231102552,
      "loss": 0.0661,
      "theoretical_loss": 3.4364287038036276,
      "tokens_seen": 2165047296
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003472957791686728,
      "loss": 0.067,
      "theoretical_loss": 3.4363948357103506,
      "tokens_seen": 2165309440
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034721553522709035,
      "loss": 0.0668,
      "theoretical_loss": 3.436360972864991,
      "tokens_seen": 2165571584
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034713529128550796,
      "loss": 0.0668,
      "theoretical_loss": 3.4363271152661006,
      "tokens_seen": 2165833728
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003470550473439256,
      "loss": 0.0682,
      "theoretical_loss": 3.4362932629122325,
      "tokens_seen": 2166095872
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034697480340234314,
      "loss": 0.0729,
      "theoretical_loss": 3.436259415801939,
      "tokens_seen": 2166358016
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003468945594607607,
      "loss": 0.0654,
      "theoretical_loss": 3.436225573933773,
      "tokens_seen": 2166620160
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003468143155191783,
      "loss": 0.0669,
      "theoretical_loss": 3.43619173730629,
      "tokens_seen": 2166882304
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003467340715775959,
      "loss": 0.0677,
      "theoretical_loss": 3.4361579059180425,
      "tokens_seen": 2167144448
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003466538276360135,
      "loss": 0.0677,
      "theoretical_loss": 3.4361240797675876,
      "tokens_seen": 2167406592
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003465735836944311,
      "loss": 0.0691,
      "theoretical_loss": 3.43609025885348,
      "tokens_seen": 2167668736
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003464933397528487,
      "loss": 0.0662,
      "theoretical_loss": 3.436056443174276,
      "tokens_seen": 2167930880
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003464130958112662,
      "loss": 0.0686,
      "theoretical_loss": 3.436022632728533,
      "tokens_seen": 2168193024
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003463328518696838,
      "loss": 0.07,
      "theoretical_loss": 3.4359888275148083,
      "tokens_seen": 2168455168
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034625260792810143,
      "loss": 0.0653,
      "theoretical_loss": 3.4359550275316595,
      "tokens_seen": 2168717312
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034617236398651904,
      "loss": 0.0662,
      "theoretical_loss": 3.4359212327776456,
      "tokens_seen": 2168979456
    },
    {
      "epoch": 0.66,
      "objective/train/advantage_avg": -0.00015192697173915803,
      "objective/train/docs_used": 790125,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4038983583450317,
      "objective/train/original_loss": 1.4038983583450317,
      "objective/train/theoretical_loss": 3.4358874432513256,
      "objective/train/tokens_used": 2189701600,
      "objective/train/value_avg": -0.00885772705078125,
      "objective/train/value_loss": 0.00030142103787511587,
      "objective/train/value_max": -1.5437602996826172e-05,
      "objective/train/value_min": -0.87890625,
      "objective/train/value_reward_corr": 0.8012933605623311,
      "objective/train/value_std": 0.0206451416015625,
      "objective/train/weight_avg": 0.9999855756759644,
      "objective/train/weighted_lm_loss": 1.4039148092269897,
      "objective/train/weights_max": 1.7211087942123413,
      "objective/train/weights_min": 0.3775528371334076,
      "theoretical_loss": 3.4358874432513256,
      "tokens_seen": 2169241600
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003460921200449366,
      "loss": 0.067,
      "theoretical_loss": 3.4358874432513256,
      "tokens_seen": 2169241600
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003460118761033542,
      "loss": 0.0659,
      "theoretical_loss": 3.435853658951259,
      "tokens_seen": 2169503744
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034593163216177183,
      "loss": 0.064,
      "theoretical_loss": 3.4358198798760067,
      "tokens_seen": 2169765888
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034585138822018933,
      "loss": 0.066,
      "theoretical_loss": 3.4357861060241293,
      "tokens_seen": 2170028032
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034577114427860695,
      "loss": 0.0669,
      "theoretical_loss": 3.4357523373941876,
      "tokens_seen": 2170290176
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034569090033702456,
      "loss": 0.0676,
      "theoretical_loss": 3.4357185739847447,
      "tokens_seen": 2170552320
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034561065639544217,
      "loss": 0.0659,
      "theoretical_loss": 3.4356848157943625,
      "tokens_seen": 2170814464
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034553041245385973,
      "loss": 0.0686,
      "theoretical_loss": 3.4356510628216044,
      "tokens_seen": 2171076608
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034545016851227734,
      "loss": 0.0696,
      "theoretical_loss": 3.435617315065034,
      "tokens_seen": 2171338752
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034536992457069495,
      "loss": 0.0679,
      "theoretical_loss": 3.435583572523216,
      "tokens_seen": 2171600896
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003452896806291125,
      "loss": 0.0673,
      "theoretical_loss": 3.4355498351947142,
      "tokens_seen": 2171863040
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034520943668753007,
      "loss": 0.0668,
      "theoretical_loss": 3.435516103078095,
      "tokens_seen": 2172125184
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003451291927459477,
      "loss": 0.0689,
      "theoretical_loss": 3.435482376171924,
      "tokens_seen": 2172387328
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003450489488043653,
      "loss": 0.0654,
      "theoretical_loss": 3.435448654474768,
      "tokens_seen": 2172649472
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034496870486278285,
      "loss": 0.0674,
      "theoretical_loss": 3.435414937985194,
      "tokens_seen": 2172911616
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034488846092120047,
      "loss": 0.0689,
      "theoretical_loss": 3.43538122670177,
      "tokens_seen": 2173173760
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003448082169796181,
      "loss": 0.0689,
      "theoretical_loss": 3.435347520623063,
      "tokens_seen": 2173435904
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034472797303803564,
      "loss": 0.0677,
      "theoretical_loss": 3.435313819747644,
      "tokens_seen": 2173698048
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003446477290964532,
      "loss": 0.0664,
      "theoretical_loss": 3.4352801240740805,
      "tokens_seen": 2173960192
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003445674851548708,
      "loss": 0.0643,
      "theoretical_loss": 3.435246433600943,
      "tokens_seen": 2174222336
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034448724121328837,
      "loss": 0.0647,
      "theoretical_loss": 3.4352127483268022,
      "tokens_seen": 2174484480
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.000344406997271706,
      "loss": 0.0644,
      "theoretical_loss": 3.4351790682502297,
      "tokens_seen": 2174746624
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003443267533301236,
      "loss": 0.0674,
      "theoretical_loss": 3.435145393369796,
      "tokens_seen": 2175008768
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003442465093885412,
      "loss": 0.067,
      "theoretical_loss": 3.4351117236840745,
      "tokens_seen": 2175270912
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034416626544695876,
      "loss": 0.0643,
      "theoretical_loss": 3.4350780591916372,
      "tokens_seen": 2175533056
    },
    {
      "epoch": 0.66,
      "objective/train/advantage_avg": 7.212981290649623e-05,
      "objective/train/docs_used": 792571,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3175489902496338,
      "objective/train/original_loss": 1.3175489902496338,
      "objective/train/theoretical_loss": 3.4350443998910576,
      "objective/train/tokens_used": 2196255200,
      "objective/train/value_avg": -0.01052093505859375,
      "objective/train/value_loss": 0.0004599344392772764,
      "objective/train/value_max": -4.7206878662109375e-05,
      "objective/train/value_min": -0.9482421875,
      "objective/train/value_reward_corr": 0.6958288722820336,
      "objective/train/value_std": 0.02197265625,
      "objective/train/weight_avg": 1.0002882480621338,
      "objective/train/weighted_lm_loss": 1.3171955347061157,
      "objective/train/weights_max": 2.5474281311035156,
      "objective/train/weights_min": 0.3810078501701355,
      "theoretical_loss": 3.4350443998910576,
      "tokens_seen": 2175795200
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003440860215053764,
      "loss": 0.065,
      "theoretical_loss": 3.4350443998910576,
      "tokens_seen": 2175795200
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034400577756379393,
      "loss": 0.0674,
      "theoretical_loss": 3.435010745780909,
      "tokens_seen": 2176057344
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003439255336222115,
      "loss": 0.0651,
      "theoretical_loss": 3.4349770968597677,
      "tokens_seen": 2176319488
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003438452896806291,
      "loss": 0.0652,
      "theoretical_loss": 3.4349434531262073,
      "tokens_seen": 2176581632
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003437650457390467,
      "loss": 0.0666,
      "theoretical_loss": 3.4349098145788033,
      "tokens_seen": 2176843776
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034368480179746433,
      "loss": 0.0672,
      "theoretical_loss": 3.434876181216133,
      "tokens_seen": 2177105920
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003436045578558819,
      "loss": 0.0648,
      "theoretical_loss": 3.4348425530367717,
      "tokens_seen": 2177368064
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003435243139142995,
      "loss": 0.0672,
      "theoretical_loss": 3.4348089300392974,
      "tokens_seen": 2177630208
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034344406997271706,
      "loss": 0.0661,
      "theoretical_loss": 3.434775312222288,
      "tokens_seen": 2177892352
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003433638260311346,
      "loss": 0.0677,
      "theoretical_loss": 3.434741699584322,
      "tokens_seen": 2178154496
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034328358208955223,
      "loss": 0.0678,
      "theoretical_loss": 3.434708092123978,
      "tokens_seen": 2178416640
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034320333814796984,
      "loss": 0.0675,
      "theoretical_loss": 3.434674489839836,
      "tokens_seen": 2178678784
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034312309420638746,
      "loss": 0.0649,
      "theoretical_loss": 3.4346408927304757,
      "tokens_seen": 2178940928
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.000343042850264805,
      "loss": 0.067,
      "theoretical_loss": 3.434607300794478,
      "tokens_seen": 2179203072
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034296260632322263,
      "loss": 0.0645,
      "theoretical_loss": 3.434573714030424,
      "tokens_seen": 2179465216
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003428823623816402,
      "loss": 0.0647,
      "theoretical_loss": 3.4345401324368954,
      "tokens_seen": 2179727360
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034280211844005774,
      "loss": 0.067,
      "theoretical_loss": 3.4345065560124746,
      "tokens_seen": 2179989504
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034272187449847536,
      "loss": 0.0659,
      "theoretical_loss": 3.434472984755745,
      "tokens_seen": 2180251648
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034264163055689297,
      "loss": 0.0675,
      "theoretical_loss": 3.434439418665289,
      "tokens_seen": 2180513792
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034256138661531053,
      "loss": 0.0639,
      "theoretical_loss": 3.434405857739691,
      "tokens_seen": 2180775936
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034248114267372814,
      "loss": 0.0669,
      "theoretical_loss": 3.434372301977536,
      "tokens_seen": 2181038080
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034240089873214575,
      "loss": 0.0671,
      "theoretical_loss": 3.4343387513774095,
      "tokens_seen": 2181300224
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034232065479056337,
      "loss": 0.0697,
      "theoretical_loss": 3.434305205937896,
      "tokens_seen": 2181562368
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034224041084898087,
      "loss": 0.0679,
      "theoretical_loss": 3.4342716656575822,
      "tokens_seen": 2181824512
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003421601669073985,
      "loss": 0.0671,
      "theoretical_loss": 3.4342381305350553,
      "tokens_seen": 2182086656
    },
    {
      "epoch": 0.66,
      "objective/train/advantage_avg": -2.115079951181542e-05,
      "objective/train/docs_used": 794520,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.430501103401184,
      "objective/train/original_loss": 1.4305012226104736,
      "objective/train/theoretical_loss": 3.434204600568902,
      "objective/train/tokens_used": 2202808800,
      "objective/train/value_avg": -0.006259918212890625,
      "objective/train/value_loss": 0.0002908243041019887,
      "objective/train/value_max": -5.650520324707031e-05,
      "objective/train/value_min": -0.7333984375,
      "objective/train/value_reward_corr": 0.6777902999039132,
      "objective/train/value_std": 0.01520538330078125,
      "objective/train/weight_avg": 1.0001102685928345,
      "objective/train/weighted_lm_loss": 1.4302117824554443,
      "objective/train/weights_max": 1.45076322555542,
      "objective/train/weights_min": 0.3727388083934784,
      "theoretical_loss": 3.434204600568902,
      "tokens_seen": 2182348800
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003420799229658161,
      "loss": 0.0696,
      "theoretical_loss": 3.434204600568902,
      "tokens_seen": 2182348800
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034199967902423365,
      "loss": 0.07,
      "theoretical_loss": 3.434171075757711,
      "tokens_seen": 2182610944
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034191943508265127,
      "loss": 0.0654,
      "theoretical_loss": 3.4341375561000698,
      "tokens_seen": 2182873088
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003418391911410689,
      "loss": 0.0656,
      "theoretical_loss": 3.4341040415945683,
      "tokens_seen": 2183135232
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003417589471994865,
      "loss": 0.0673,
      "theoretical_loss": 3.434070532239796,
      "tokens_seen": 2183397376
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.000341678703257904,
      "loss": 0.066,
      "theoretical_loss": 3.4340370280343424,
      "tokens_seen": 2183659520
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003415984593163216,
      "loss": 0.065,
      "theoretical_loss": 3.434003528976799,
      "tokens_seen": 2183921664
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003415182153747392,
      "loss": 0.0663,
      "theoretical_loss": 3.433970035065756,
      "tokens_seen": 2184183808
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003414379714331568,
      "loss": 0.0674,
      "theoretical_loss": 3.4339365462998064,
      "tokens_seen": 2184445952
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003413577274915744,
      "loss": 0.065,
      "theoretical_loss": 3.4339030626775413,
      "tokens_seen": 2184708096
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.000341277483549992,
      "loss": 0.0653,
      "theoretical_loss": 3.433869584197555,
      "tokens_seen": 2184970240
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003411972396084096,
      "loss": 0.0665,
      "theoretical_loss": 3.43383611085844,
      "tokens_seen": 2185232384
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003411169956668271,
      "loss": 0.0661,
      "theoretical_loss": 3.43380264265879,
      "tokens_seen": 2185494528
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034103675172524473,
      "loss": 0.0652,
      "theoretical_loss": 3.433769179597201,
      "tokens_seen": 2185756672
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034095650778366235,
      "loss": 0.0669,
      "theoretical_loss": 3.433735721672267,
      "tokens_seen": 2186018816
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003408762638420799,
      "loss": 0.0664,
      "theoretical_loss": 3.4337022688825836,
      "tokens_seen": 2186280960
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003407960199004975,
      "loss": 0.0687,
      "theoretical_loss": 3.4336688212267474,
      "tokens_seen": 2186543104
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034071577595891513,
      "loss": 0.0695,
      "theoretical_loss": 3.4336353787033556,
      "tokens_seen": 2186805248
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034063553201733274,
      "loss": 0.066,
      "theoretical_loss": 3.4336019413110046,
      "tokens_seen": 2187067392
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003405552880757503,
      "loss": 0.0653,
      "theoretical_loss": 3.4335685090482926,
      "tokens_seen": 2187329536
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034047504413416786,
      "loss": 0.0701,
      "theoretical_loss": 3.4335350819138184,
      "tokens_seen": 2187591680
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034039480019258547,
      "loss": 0.0664,
      "theoretical_loss": 3.433501659906181,
      "tokens_seen": 2187853824
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034031455625100303,
      "loss": 0.0656,
      "theoretical_loss": 3.433468243023979,
      "tokens_seen": 2188115968
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034023431230942064,
      "loss": 0.0658,
      "theoretical_loss": 3.433434831265814,
      "tokens_seen": 2188378112
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00034015406836783826,
      "loss": 0.0675,
      "theoretical_loss": 3.4334014246302855,
      "tokens_seen": 2188640256
    },
    {
      "epoch": 0.66,
      "objective/train/advantage_avg": 4.372426701593213e-05,
      "objective/train/docs_used": 796788,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2543916702270508,
      "objective/train/original_loss": 1.2543916702270508,
      "objective/train/theoretical_loss": 3.4333680231159946,
      "objective/train/tokens_used": 2209362400,
      "objective/train/value_avg": -0.0097503662109375,
      "objective/train/value_loss": 0.00034007534850388765,
      "objective/train/value_max": -2.7298927307128906e-05,
      "objective/train/value_min": -0.71875,
      "objective/train/value_reward_corr": 0.7060483238102327,
      "objective/train/value_std": 0.0169219970703125,
      "objective/train/weight_avg": 1.0001929998397827,
      "objective/train/weighted_lm_loss": 1.2542983293533325,
      "objective/train/weights_max": 1.3613519668579102,
      "objective/train/weights_min": 0.3707548677921295,
      "theoretical_loss": 3.4333680231159946,
      "tokens_seen": 2188902400
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003400738244262558,
      "loss": 0.0668,
      "theoretical_loss": 3.4333680231159946,
      "tokens_seen": 2188902400
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003399935804846734,
      "loss": 0.066,
      "theoretical_loss": 3.433334626721544,
      "tokens_seen": 2189164544
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.000339913336543091,
      "loss": 0.0669,
      "theoretical_loss": 3.4333012354455352,
      "tokens_seen": 2189426688
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003398330926015086,
      "loss": 0.0669,
      "theoretical_loss": 3.433267849286571,
      "tokens_seen": 2189688832
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00033975284865992616,
      "loss": 0.072,
      "theoretical_loss": 3.4332344682432554,
      "tokens_seen": 2189950976
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00033967260471834377,
      "loss": 0.0674,
      "theoretical_loss": 3.4332010923141913,
      "tokens_seen": 2190213120
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003395923607767614,
      "loss": 0.0646,
      "theoretical_loss": 3.4331677214979845,
      "tokens_seen": 2190475264
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00033951211683517894,
      "loss": 0.069,
      "theoretical_loss": 3.4331343557932392,
      "tokens_seen": 2190737408
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00033943187289359655,
      "loss": 0.0655,
      "theoretical_loss": 3.433100995198561,
      "tokens_seen": 2190999552
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00033935162895201417,
      "loss": 0.0672,
      "theoretical_loss": 3.433067639712556,
      "tokens_seen": 2191261696
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003392713850104317,
      "loss": 0.0652,
      "theoretical_loss": 3.4330342893338313,
      "tokens_seen": 2191523840
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003391911410688493,
      "loss": 0.0638,
      "theoretical_loss": 3.4330009440609937,
      "tokens_seen": 2191785984
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003391108971272669,
      "loss": 0.0661,
      "theoretical_loss": 3.432967603892651,
      "tokens_seen": 2192048128
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003390306531856845,
      "loss": 0.0696,
      "theoretical_loss": 3.432934268827412,
      "tokens_seen": 2192310272
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00033895040924410207,
      "loss": 0.0664,
      "theoretical_loss": 3.4329009388638845,
      "tokens_seen": 2192572416
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003388701653025197,
      "loss": 0.068,
      "theoretical_loss": 3.4328676140006786,
      "tokens_seen": 2192834560
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003387899213609373,
      "loss": 0.0681,
      "theoretical_loss": 3.4328342942364043,
      "tokens_seen": 2193096704
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00033870967741935485,
      "loss": 0.0648,
      "theoretical_loss": 3.432800979569672,
      "tokens_seen": 2193358848
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003386294334777724,
      "loss": 0.0667,
      "theoretical_loss": 3.4327676699990928,
      "tokens_seen": 2193620992
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00033854918953619,
      "loss": 0.0674,
      "theoretical_loss": 3.432734365523278,
      "tokens_seen": 2193883136
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.00033846894559460763,
      "loss": 0.0694,
      "theoretical_loss": 3.4327010661408397,
      "tokens_seen": 2194145280
    },
    {
      "epoch": 0.66,
      "learning_rate": 0.0003383887016530252,
      "loss": 0.067,
      "theoretical_loss": 3.432667771850391,
      "tokens_seen": 2194407424
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003383084577114428,
      "loss": 0.0668,
      "theoretical_loss": 3.4326344826505446,
      "tokens_seen": 2194669568
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003382282137698604,
      "loss": 0.0669,
      "theoretical_loss": 3.4326011985399147,
      "tokens_seen": 2194931712
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003381479698282779,
      "loss": 0.068,
      "theoretical_loss": 3.4325679195171155,
      "tokens_seen": 2195193856
    },
    {
      "epoch": 0.67,
      "objective/train/advantage_avg": 0.00028255730285309255,
      "objective/train/docs_used": 799213,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3589651584625244,
      "objective/train/original_loss": 1.3589650392532349,
      "objective/train/theoretical_loss": 3.4325346455807613,
      "objective/train/tokens_used": 2215916000,
      "objective/train/value_avg": -0.00843048095703125,
      "objective/train/value_loss": 0.00023773591965436935,
      "objective/train/value_max": -5.519390106201172e-05,
      "objective/train/value_min": -0.79150390625,
      "objective/train/value_reward_corr": 0.7414294469085657,
      "objective/train/value_std": 0.0166168212890625,
      "objective/train/weight_avg": 1.0003902912139893,
      "objective/train/weighted_lm_loss": 1.3589110374450684,
      "objective/train/weights_max": 1.6519445180892944,
      "objective/train/weights_min": 0.23257870972156525,
      "theoretical_loss": 3.4325346455807613,
      "tokens_seen": 2195456000
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033806772588669553,
      "loss": 0.0643,
      "theoretical_loss": 3.4325346455807613,
      "tokens_seen": 2195456000
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033798748194511315,
      "loss": 0.0677,
      "theoretical_loss": 3.4325013767294683,
      "tokens_seen": 2195718144
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033790723800353076,
      "loss": 0.0686,
      "theoretical_loss": 3.4324681129618515,
      "tokens_seen": 2195980288
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003378269940619483,
      "loss": 0.0668,
      "theoretical_loss": 3.4324348542765284,
      "tokens_seen": 2196242432
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033774675012036593,
      "loss": 0.0699,
      "theoretical_loss": 3.432401600672115,
      "tokens_seen": 2196504576
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033766650617878354,
      "loss": 0.0671,
      "theoretical_loss": 3.43236835214723,
      "tokens_seen": 2196766720
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003375862622372011,
      "loss": 0.0664,
      "theoretical_loss": 3.4323351087004905,
      "tokens_seen": 2197028864
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033750601829561866,
      "loss": 0.0645,
      "theoretical_loss": 3.4323018703305155,
      "tokens_seen": 2197291008
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033742577435403627,
      "loss": 0.0656,
      "theoretical_loss": 3.432268637035924,
      "tokens_seen": 2197553152
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003373455304124539,
      "loss": 0.0684,
      "theoretical_loss": 3.4322354088153357,
      "tokens_seen": 2197815296
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033726528647087144,
      "loss": 0.0679,
      "theoretical_loss": 3.432202185667371,
      "tokens_seen": 2198077440
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033718504252928906,
      "loss": 0.0668,
      "theoretical_loss": 3.432168967590651,
      "tokens_seen": 2198339584
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033710479858770667,
      "loss": 0.0674,
      "theoretical_loss": 3.432135754583796,
      "tokens_seen": 2198601728
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003370245546461242,
      "loss": 0.0672,
      "theoretical_loss": 3.432102546645429,
      "tokens_seen": 2198863872
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003369443107045418,
      "loss": 0.0678,
      "theoretical_loss": 3.432069343774172,
      "tokens_seen": 2199126016
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003368640667629594,
      "loss": 0.068,
      "theoretical_loss": 3.432036145968648,
      "tokens_seen": 2199388160
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.000336783822821377,
      "loss": 0.0682,
      "theoretical_loss": 3.43200295322748,
      "tokens_seen": 2199650304
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033670357887979457,
      "loss": 0.0646,
      "theoretical_loss": 3.431969765549293,
      "tokens_seen": 2199912448
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003366233349382122,
      "loss": 0.0661,
      "theoretical_loss": 3.43193658293271,
      "tokens_seen": 2200174592
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003365430909966298,
      "loss": 0.068,
      "theoretical_loss": 3.4319034053763575,
      "tokens_seen": 2200436736
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033646284705504735,
      "loss": 0.0656,
      "theoretical_loss": 3.431870232878861,
      "tokens_seen": 2200698880
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003363826031134649,
      "loss": 0.0671,
      "theoretical_loss": 3.431837065438846,
      "tokens_seen": 2200961024
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003363023591718825,
      "loss": 0.0682,
      "theoretical_loss": 3.4318039030549397,
      "tokens_seen": 2201223168
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003362221152303001,
      "loss": 0.0663,
      "theoretical_loss": 3.431770745725769,
      "tokens_seen": 2201485312
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003361418712887177,
      "loss": 0.0642,
      "theoretical_loss": 3.431737593449962,
      "tokens_seen": 2201747456
    },
    {
      "epoch": 0.67,
      "objective/train/advantage_avg": 0.000496376771479845,
      "objective/train/docs_used": 801578,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3730624914169312,
      "objective/train/original_loss": 1.3730623722076416,
      "objective/train/theoretical_loss": 3.4317044462261466,
      "objective/train/tokens_used": 2222469600,
      "objective/train/value_avg": -0.00689697265625,
      "objective/train/value_loss": 0.00029629369964823127,
      "objective/train/value_max": -3.3974647521972656e-05,
      "objective/train/value_min": -0.7666015625,
      "objective/train/value_reward_corr": 0.737004677943326,
      "objective/train/value_std": 0.0198822021484375,
      "objective/train/weight_avg": 1.0006368160247803,
      "objective/train/weighted_lm_loss": 1.373940110206604,
      "objective/train/weights_max": 2.072002649307251,
      "objective/train/weights_min": 0.3690805733203888,
      "theoretical_loss": 3.4317044462261466,
      "tokens_seen": 2202009600
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003360616273471353,
      "loss": 0.0654,
      "theoretical_loss": 3.4317044462261466,
      "tokens_seen": 2202009600
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003359813834055529,
      "loss": 0.0665,
      "theoretical_loss": 3.4316713040529523,
      "tokens_seen": 2202271744
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003359011394639705,
      "loss": 0.0672,
      "theoretical_loss": 3.4316381669290084,
      "tokens_seen": 2202533888
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003358208955223881,
      "loss": 0.0655,
      "theoretical_loss": 3.431605034852944,
      "tokens_seen": 2202796032
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033574065158080565,
      "loss": 0.0703,
      "theoretical_loss": 3.43157190782339,
      "tokens_seen": 2203058176
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003356604076392232,
      "loss": 0.065,
      "theoretical_loss": 3.431538785838978,
      "tokens_seen": 2203320320
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003355801636976408,
      "loss": 0.0675,
      "theoretical_loss": 3.4315056688983385,
      "tokens_seen": 2203582464
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033549991975605843,
      "loss": 0.0687,
      "theoretical_loss": 3.4314725570001046,
      "tokens_seen": 2203844608
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033541967581447604,
      "loss": 0.066,
      "theoretical_loss": 3.431439450142908,
      "tokens_seen": 2204106752
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003353394318728936,
      "loss": 0.0692,
      "theoretical_loss": 3.4314063483253823,
      "tokens_seen": 2204368896
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003352591879313112,
      "loss": 0.0656,
      "theoretical_loss": 3.431373251546161,
      "tokens_seen": 2204631040
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003351789439897288,
      "loss": 0.069,
      "theoretical_loss": 3.431340159803878,
      "tokens_seen": 2204893184
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033509870004814633,
      "loss": 0.0666,
      "theoretical_loss": 3.4313070730971686,
      "tokens_seen": 2205155328
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033501845610656395,
      "loss": 0.0691,
      "theoretical_loss": 3.431273991424668,
      "tokens_seen": 2205417472
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033493821216498156,
      "loss": 0.0681,
      "theoretical_loss": 3.431240914785012,
      "tokens_seen": 2205679616
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033485796822339917,
      "loss": 0.0654,
      "theoretical_loss": 3.431207843176836,
      "tokens_seen": 2205941760
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033477772428181673,
      "loss": 0.0678,
      "theoretical_loss": 3.431174776598778,
      "tokens_seen": 2206203904
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033469748034023434,
      "loss": 0.0675,
      "theoretical_loss": 3.431141715049475,
      "tokens_seen": 2206466048
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033461723639865195,
      "loss": 0.0672,
      "theoretical_loss": 3.4311086585275645,
      "tokens_seen": 2206728192
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033453699245706946,
      "loss": 0.0656,
      "theoretical_loss": 3.4310756070316857,
      "tokens_seen": 2206990336
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033445674851548707,
      "loss": 0.0673,
      "theoretical_loss": 3.431042560560477,
      "tokens_seen": 2207252480
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003343765045739047,
      "loss": 0.0689,
      "theoretical_loss": 3.431009519112578,
      "tokens_seen": 2207514624
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033429626063232224,
      "loss": 0.0684,
      "theoretical_loss": 3.4309764826866287,
      "tokens_seen": 2207776768
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033421601669073985,
      "loss": 0.0675,
      "theoretical_loss": 3.43094345128127,
      "tokens_seen": 2208038912
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033413577274915747,
      "loss": 0.0699,
      "theoretical_loss": 3.430910424895143,
      "tokens_seen": 2208301056
    },
    {
      "epoch": 0.67,
      "objective/train/advantage_avg": 0.0003951071994379163,
      "objective/train/docs_used": 804023,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3737064599990845,
      "objective/train/original_loss": 1.373706579208374,
      "objective/train/theoretical_loss": 3.4308774035268885,
      "objective/train/tokens_used": 2229023200,
      "objective/train/value_avg": -0.00640869140625,
      "objective/train/value_loss": 0.00019101255747955292,
      "objective/train/value_max": -4.9114227294921875e-05,
      "objective/train/value_min": -0.58740234375,
      "objective/train/value_reward_corr": 0.6632108587219194,
      "objective/train/value_std": 0.012176513671875,
      "objective/train/weight_avg": 1.0004802942276,
      "objective/train/weighted_lm_loss": 1.3744759559631348,
      "objective/train/weights_max": 1.470705509185791,
      "objective/train/weights_min": 0.36826980113983154,
      "theoretical_loss": 3.4308774035268885,
      "tokens_seen": 2208563200
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003340555288075751,
      "loss": 0.0686,
      "theoretical_loss": 3.4308774035268885,
      "tokens_seen": 2208563200
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003339752848659926,
      "loss": 0.0676,
      "theoretical_loss": 3.4308443871751497,
      "tokens_seen": 2208825344
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003338950409244102,
      "loss": 0.0644,
      "theoretical_loss": 3.4308113758385685,
      "tokens_seen": 2209087488
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003338147969828278,
      "loss": 0.0685,
      "theoretical_loss": 3.4307783695157887,
      "tokens_seen": 2209349632
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033373455304124537,
      "loss": 0.0675,
      "theoretical_loss": 3.4307453682054536,
      "tokens_seen": 2209611776
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.000333654309099663,
      "loss": 0.0674,
      "theoretical_loss": 3.430712371906208,
      "tokens_seen": 2209873920
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003335740651580806,
      "loss": 0.0674,
      "theoretical_loss": 3.4306793806166955,
      "tokens_seen": 2210136064
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003334938212164982,
      "loss": 0.0662,
      "theoretical_loss": 3.4306463943355627,
      "tokens_seen": 2210398208
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003334135772749157,
      "loss": 0.0672,
      "theoretical_loss": 3.430613413061455,
      "tokens_seen": 2210660352
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003333333333333333,
      "loss": 0.0673,
      "theoretical_loss": 3.4305804367930186,
      "tokens_seen": 2210922496
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033325308939175093,
      "loss": 0.0639,
      "theoretical_loss": 3.4305474655289006,
      "tokens_seen": 2211184640
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003331728454501685,
      "loss": 0.0678,
      "theoretical_loss": 3.4305144992677485,
      "tokens_seen": 2211446784
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003330926015085861,
      "loss": 0.066,
      "theoretical_loss": 3.4304815380082103,
      "tokens_seen": 2211708928
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003330123575670037,
      "loss": 0.065,
      "theoretical_loss": 3.430448581748934,
      "tokens_seen": 2211971072
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033293211362542133,
      "loss": 0.0692,
      "theoretical_loss": 3.430415630488569,
      "tokens_seen": 2212233216
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003328518696838389,
      "loss": 0.0686,
      "theoretical_loss": 3.4303826842257648,
      "tokens_seen": 2212495360
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033277162574225645,
      "loss": 0.0676,
      "theoretical_loss": 3.430349742959171,
      "tokens_seen": 2212757504
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033269138180067406,
      "loss": 0.0657,
      "theoretical_loss": 3.4303168066874385,
      "tokens_seen": 2213019648
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003326111378590916,
      "loss": 0.067,
      "theoretical_loss": 3.430283875409219,
      "tokens_seen": 2213281792
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033253089391750923,
      "loss": 0.065,
      "theoretical_loss": 3.430250949123163,
      "tokens_seen": 2213543936
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033245064997592684,
      "loss": 0.0669,
      "theoretical_loss": 3.430218027827924,
      "tokens_seen": 2213806080
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033237040603434446,
      "loss": 0.068,
      "theoretical_loss": 3.430185111522153,
      "tokens_seen": 2214068224
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.000332290162092762,
      "loss": 0.0642,
      "theoretical_loss": 3.4301522002045046,
      "tokens_seen": 2214330368
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003322099181511796,
      "loss": 0.0692,
      "theoretical_loss": 3.430119293873632,
      "tokens_seen": 2214592512
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003321296742095972,
      "loss": 0.067,
      "theoretical_loss": 3.4300863925281893,
      "tokens_seen": 2214854656
    },
    {
      "epoch": 0.67,
      "objective/train/advantage_avg": 0.000551328994333744,
      "objective/train/docs_used": 806281,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2843917608261108,
      "objective/train/original_loss": 1.2843916416168213,
      "objective/train/theoretical_loss": 3.430053496166831,
      "objective/train/tokens_used": 2235576800,
      "objective/train/value_avg": -0.00879669189453125,
      "objective/train/value_loss": 0.0003198671620339155,
      "objective/train/value_max": -2.372264862060547e-05,
      "objective/train/value_min": -0.97216796875,
      "objective/train/value_reward_corr": 0.741845592361158,
      "objective/train/value_std": 0.0195159912109375,
      "objective/train/weight_avg": 1.0007034540176392,
      "objective/train/weighted_lm_loss": 1.2850383520126343,
      "objective/train/weights_max": 1.8766188621520996,
      "objective/train/weights_min": 0.38213738799095154,
      "theoretical_loss": 3.430053496166831,
      "tokens_seen": 2215116800
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033204943026801474,
      "loss": 0.0662,
      "theoretical_loss": 3.430053496166831,
      "tokens_seen": 2215116800
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033196918632643236,
      "loss": 0.0668,
      "theoretical_loss": 3.4300206047882136,
      "tokens_seen": 2215378944
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033188894238484997,
      "loss": 0.0678,
      "theoretical_loss": 3.4299877183909917,
      "tokens_seen": 2215641088
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033180869844326753,
      "loss": 0.0665,
      "theoretical_loss": 3.4299548369738218,
      "tokens_seen": 2215903232
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033172845450168514,
      "loss": 0.0648,
      "theoretical_loss": 3.429921960535361,
      "tokens_seen": 2216165376
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003316482105601027,
      "loss": 0.0667,
      "theoretical_loss": 3.429889089074267,
      "tokens_seen": 2216427520
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003315679666185203,
      "loss": 0.0636,
      "theoretical_loss": 3.429856222589197,
      "tokens_seen": 2216689664
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033148772267693787,
      "loss": 0.0673,
      "theoretical_loss": 3.4298233610788094,
      "tokens_seen": 2216951808
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003314074787353555,
      "loss": 0.067,
      "theoretical_loss": 3.429790504541764,
      "tokens_seen": 2217213952
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003313272347937731,
      "loss": 0.066,
      "theoretical_loss": 3.4297576529767193,
      "tokens_seen": 2217476096
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033124699085219065,
      "loss": 0.0671,
      "theoretical_loss": 3.4297248063823362,
      "tokens_seen": 2217738240
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033116674691060827,
      "loss": 0.066,
      "theoretical_loss": 3.4296919647572746,
      "tokens_seen": 2218000384
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003310865029690259,
      "loss": 0.067,
      "theoretical_loss": 3.4296591281001954,
      "tokens_seen": 2218262528
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033100625902744344,
      "loss": 0.0643,
      "theoretical_loss": 3.42962629640976,
      "tokens_seen": 2218524672
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.000330926015085861,
      "loss": 0.0682,
      "theoretical_loss": 3.4295934696846313,
      "tokens_seen": 2218786816
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003308457711442786,
      "loss": 0.0642,
      "theoretical_loss": 3.4295606479234713,
      "tokens_seen": 2219048960
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003307655272026962,
      "loss": 0.0661,
      "theoretical_loss": 3.429527831124943,
      "tokens_seen": 2219311104
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003306852832611138,
      "loss": 0.0673,
      "theoretical_loss": 3.42949501928771,
      "tokens_seen": 2219573248
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003306050393195314,
      "loss": 0.0649,
      "theoretical_loss": 3.4294622124104364,
      "tokens_seen": 2219835392
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.000330524795377949,
      "loss": 0.064,
      "theoretical_loss": 3.429429410491787,
      "tokens_seen": 2220097536
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033044455143636656,
      "loss": 0.066,
      "theoretical_loss": 3.429396613530427,
      "tokens_seen": 2220359680
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003303643074947841,
      "loss": 0.0668,
      "theoretical_loss": 3.429363821525022,
      "tokens_seen": 2220621824
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033028406355320173,
      "loss": 0.0667,
      "theoretical_loss": 3.4293310344742385,
      "tokens_seen": 2220883968
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00033020381961161935,
      "loss": 0.065,
      "theoretical_loss": 3.429298252376743,
      "tokens_seen": 2221146112
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003301235756700369,
      "loss": 0.0638,
      "theoretical_loss": 3.429265475231202,
      "tokens_seen": 2221408256
    },
    {
      "epoch": 0.67,
      "objective/train/advantage_avg": 0.0007514514145441353,
      "objective/train/docs_used": 808517,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2400312423706055,
      "objective/train/original_loss": 1.2400312423706055,
      "objective/train/theoretical_loss": 3.429232703036284,
      "objective/train/tokens_used": 2242130400,
      "objective/train/value_avg": -0.00983428955078125,
      "objective/train/value_loss": 0.0003172192955389619,
      "objective/train/value_max": -7.724761962890625e-05,
      "objective/train/value_min": -0.80078125,
      "objective/train/value_reward_corr": 0.7819778708574383,
      "objective/train/value_std": 0.0235595703125,
      "objective/train/weight_avg": 1.000901460647583,
      "objective/train/weighted_lm_loss": 1.2409225702285767,
      "objective/train/weights_max": 1.667690396308899,
      "objective/train/weights_min": 0.41253384947776794,
      "theoretical_loss": 3.429232703036284,
      "tokens_seen": 2221670400
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003300433317284545,
      "loss": 0.0651,
      "theoretical_loss": 3.429232703036284,
      "tokens_seen": 2221670400
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00032996308778687213,
      "loss": 0.067,
      "theoretical_loss": 3.429199935790657,
      "tokens_seen": 2221932544
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003298828438452897,
      "loss": 0.0689,
      "theoretical_loss": 3.4291671734929903,
      "tokens_seen": 2222194688
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00032980259990370725,
      "loss": 0.0683,
      "theoretical_loss": 3.4291344161419524,
      "tokens_seen": 2222456832
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00032972235596212486,
      "loss": 0.066,
      "theoretical_loss": 3.4291016637362137,
      "tokens_seen": 2222718976
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00032964211202054247,
      "loss": 0.0657,
      "theoretical_loss": 3.429068916274444,
      "tokens_seen": 2222981120
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00032956186807896003,
      "loss": 0.0654,
      "theoretical_loss": 3.429036173755314,
      "tokens_seen": 2223243264
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00032948162413737764,
      "loss": 0.0693,
      "theoretical_loss": 3.429003436177496,
      "tokens_seen": 2223505408
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00032940138019579526,
      "loss": 0.067,
      "theoretical_loss": 3.4289707035396613,
      "tokens_seen": 2223767552
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003293211362542128,
      "loss": 0.0697,
      "theoretical_loss": 3.428937975840482,
      "tokens_seen": 2224029696
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00032924089231263037,
      "loss": 0.0686,
      "theoretical_loss": 3.4289052530786313,
      "tokens_seen": 2224291840
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.000329160648371048,
      "loss": 0.068,
      "theoretical_loss": 3.4288725352527827,
      "tokens_seen": 2224553984
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003290804044294656,
      "loss": 0.0661,
      "theoretical_loss": 3.42883982236161,
      "tokens_seen": 2224816128
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00032900016048788316,
      "loss": 0.0649,
      "theoretical_loss": 3.428807114403787,
      "tokens_seen": 2225078272
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00032891991654630077,
      "loss": 0.0627,
      "theoretical_loss": 3.428774411377989,
      "tokens_seen": 2225340416
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003288396726047184,
      "loss": 0.0677,
      "theoretical_loss": 3.4287417132828923,
      "tokens_seen": 2225602560
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.00032875942866313594,
      "loss": 0.0656,
      "theoretical_loss": 3.428709020117172,
      "tokens_seen": 2225864704
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003286791847215535,
      "loss": 0.0646,
      "theoretical_loss": 3.4286763318795046,
      "tokens_seen": 2226126848
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003285989407799711,
      "loss": 0.064,
      "theoretical_loss": 3.428643648568567,
      "tokens_seen": 2226388992
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003285186968383887,
      "loss": 0.0684,
      "theoretical_loss": 3.4286109701830374,
      "tokens_seen": 2226651136
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003284384528968063,
      "loss": 0.0678,
      "theoretical_loss": 3.428578296721593,
      "tokens_seen": 2226913280
    },
    {
      "epoch": 0.67,
      "learning_rate": 0.0003283582089552239,
      "loss": 0.0669,
      "theoretical_loss": 3.4285456281829125,
      "tokens_seen": 2227175424
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003282779650136415,
      "loss": 0.0694,
      "theoretical_loss": 3.428512964565675,
      "tokens_seen": 2227437568
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032819772107205907,
      "loss": 0.0664,
      "theoretical_loss": 3.4284803058685602,
      "tokens_seen": 2227699712
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003281174771304767,
      "loss": 0.0659,
      "theoretical_loss": 3.428447652090248,
      "tokens_seen": 2227961856
    },
    {
      "epoch": 0.68,
      "objective/train/advantage_avg": -0.0001387705997331068,
      "objective/train/docs_used": 811140,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2117676734924316,
      "objective/train/original_loss": 1.2117676734924316,
      "objective/train/theoretical_loss": 3.4284150032294187,
      "objective/train/tokens_used": 2248684000,
      "objective/train/value_avg": -0.009246826171875,
      "objective/train/value_loss": 0.0001913748710649088,
      "objective/train/value_max": -3.5643577575683594e-05,
      "objective/train/value_min": -0.2802734375,
      "objective/train/value_reward_corr": 0.834702300274055,
      "objective/train/value_std": 0.0182647705078125,
      "objective/train/weight_avg": 0.999951958656311,
      "objective/train/weighted_lm_loss": 1.212104320526123,
      "objective/train/weights_max": 1.2279880046844482,
      "objective/train/weights_min": 0.372374951839447,
      "theoretical_loss": 3.4284150032294187,
      "tokens_seen": 2228224000
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032803723318889424,
      "loss": 0.0672,
      "theoretical_loss": 3.4284150032294187,
      "tokens_seen": 2228224000
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003279569892473118,
      "loss": 0.0666,
      "theoretical_loss": 3.428382359284754,
      "tokens_seen": 2228486144
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003278767453057294,
      "loss": 0.0673,
      "theoretical_loss": 3.4283497202549347,
      "tokens_seen": 2228748288
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.000327796501364147,
      "loss": 0.0689,
      "theoretical_loss": 3.4283170861386436,
      "tokens_seen": 2229010432
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032771625742256463,
      "loss": 0.0684,
      "theoretical_loss": 3.4282844569345623,
      "tokens_seen": 2229272576
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003276360134809822,
      "loss": 0.0675,
      "theoretical_loss": 3.428251832641375,
      "tokens_seen": 2229534720
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003275557695393998,
      "loss": 0.0685,
      "theoretical_loss": 3.4282192132577647,
      "tokens_seen": 2229796864
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032747552559781736,
      "loss": 0.0666,
      "theoretical_loss": 3.4281865987824154,
      "tokens_seen": 2230059008
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003273952816562349,
      "loss": 0.0661,
      "theoretical_loss": 3.428153989214012,
      "tokens_seen": 2230321152
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032731503771465253,
      "loss": 0.0655,
      "theoretical_loss": 3.4281213845512397,
      "tokens_seen": 2230583296
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032723479377307015,
      "loss": 0.0643,
      "theoretical_loss": 3.428088784792784,
      "tokens_seen": 2230845440
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032715454983148776,
      "loss": 0.0696,
      "theoretical_loss": 3.4280561899373305,
      "tokens_seen": 2231107584
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003270743058899053,
      "loss": 0.0672,
      "theoretical_loss": 3.428023599983567,
      "tokens_seen": 2231369728
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032699406194832293,
      "loss": 0.0675,
      "theoretical_loss": 3.4279910149301798,
      "tokens_seen": 2231631872
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003269138180067405,
      "loss": 0.0674,
      "theoretical_loss": 3.4279584347758565,
      "tokens_seen": 2231894016
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032683357406515805,
      "loss": 0.065,
      "theoretical_loss": 3.4279258595192856,
      "tokens_seen": 2232156160
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032675333012357566,
      "loss": 0.0649,
      "theoretical_loss": 3.427893289159156,
      "tokens_seen": 2232418304
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032667308618199327,
      "loss": 0.0672,
      "theoretical_loss": 3.427860723694156,
      "tokens_seen": 2232680448
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003265928422404109,
      "loss": 0.0677,
      "theoretical_loss": 3.427828163122976,
      "tokens_seen": 2232942592
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032651259829882844,
      "loss": 0.067,
      "theoretical_loss": 3.427795607444306,
      "tokens_seen": 2233204736
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032643235435724606,
      "loss": 0.0689,
      "theoretical_loss": 3.4277630566568367,
      "tokens_seen": 2233466880
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032635211041566367,
      "loss": 0.0658,
      "theoretical_loss": 3.4277305107592593,
      "tokens_seen": 2233729024
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032627186647408117,
      "loss": 0.0682,
      "theoretical_loss": 3.427697969750265,
      "tokens_seen": 2233991168
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003261916225324988,
      "loss": 0.0684,
      "theoretical_loss": 3.427665433628547,
      "tokens_seen": 2234253312
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003261113785909164,
      "loss": 0.0679,
      "theoretical_loss": 3.427632902392797,
      "tokens_seen": 2234515456
    },
    {
      "epoch": 0.68,
      "objective/train/advantage_avg": 0.0003541047917678952,
      "objective/train/docs_used": 813661,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.345414400100708,
      "objective/train/original_loss": 1.345414400100708,
      "objective/train/theoretical_loss": 3.427600376041709,
      "objective/train/tokens_used": 2255237600,
      "objective/train/value_avg": -0.006504058837890625,
      "objective/train/value_loss": 0.00015749734302517027,
      "objective/train/value_max": -1.8656253814697266e-05,
      "objective/train/value_min": -0.7744140625,
      "objective/train/value_reward_corr": 0.7244407312411731,
      "objective/train/value_std": 0.0146942138671875,
      "objective/train/weight_avg": 1.000430703163147,
      "objective/train/weighted_lm_loss": 1.3458590507507324,
      "objective/train/weights_max": 1.8365892171859741,
      "objective/train/weights_min": 0.39913442730903625,
      "theoretical_loss": 3.427600376041709,
      "tokens_seen": 2234777600
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032603113464933396,
      "loss": 0.0672,
      "theoretical_loss": 3.427600376041709,
      "tokens_seen": 2234777600
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032595089070775157,
      "loss": 0.0647,
      "theoretical_loss": 3.4275678545739763,
      "tokens_seen": 2235039744
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003258706467661692,
      "loss": 0.067,
      "theoretical_loss": 3.4275353379882927,
      "tokens_seen": 2235301888
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003257904028245868,
      "loss": 0.0669,
      "theoretical_loss": 3.427502826283354,
      "tokens_seen": 2235564032
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003257101588830043,
      "loss": 0.0668,
      "theoretical_loss": 3.427470319457854,
      "tokens_seen": 2235826176
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003256299149414219,
      "loss": 0.0685,
      "theoretical_loss": 3.4274378175104894,
      "tokens_seen": 2236088320
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003255496709998395,
      "loss": 0.0658,
      "theoretical_loss": 3.427405320439956,
      "tokens_seen": 2236350464
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003254694270582571,
      "loss": 0.0656,
      "theoretical_loss": 3.4273728282449514,
      "tokens_seen": 2236612608
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003253891831166747,
      "loss": 0.066,
      "theoretical_loss": 3.4273403409241716,
      "tokens_seen": 2236874752
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003253089391750923,
      "loss": 0.0667,
      "theoretical_loss": 3.427307858476315,
      "tokens_seen": 2237136896
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003252286952335099,
      "loss": 0.0689,
      "theoretical_loss": 3.4272753809000793,
      "tokens_seen": 2237399040
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003251484512919275,
      "loss": 0.0675,
      "theoretical_loss": 3.4272429081941636,
      "tokens_seen": 2237661184
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032506820735034504,
      "loss": 0.0693,
      "theoretical_loss": 3.4272104403572667,
      "tokens_seen": 2237923328
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032498796340876265,
      "loss": 0.0667,
      "theoretical_loss": 3.4271779773880895,
      "tokens_seen": 2238185472
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003249077194671802,
      "loss": 0.0657,
      "theoretical_loss": 3.427145519285331,
      "tokens_seen": 2238447616
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003248274755255978,
      "loss": 0.0693,
      "theoretical_loss": 3.427113066047692,
      "tokens_seen": 2238709760
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032474723158401543,
      "loss": 0.0687,
      "theoretical_loss": 3.4270806176738744,
      "tokens_seen": 2238971904
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032466698764243304,
      "loss": 0.067,
      "theoretical_loss": 3.4270481741625796,
      "tokens_seen": 2239234048
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003245867437008506,
      "loss": 0.0677,
      "theoretical_loss": 3.4270157355125095,
      "tokens_seen": 2239496192
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032450649975926816,
      "loss": 0.0687,
      "theoretical_loss": 3.426983301722367,
      "tokens_seen": 2239758336
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003244262558176858,
      "loss": 0.0668,
      "theoretical_loss": 3.4269508727908553,
      "tokens_seen": 2240020480
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032434601187610333,
      "loss": 0.0667,
      "theoretical_loss": 3.426918448716678,
      "tokens_seen": 2240282624
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032426576793452095,
      "loss": 0.068,
      "theoretical_loss": 3.42688602949854,
      "tokens_seen": 2240544768
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032418552399293856,
      "loss": 0.064,
      "theoretical_loss": 3.426853615135145,
      "tokens_seen": 2240806912
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032410528005135617,
      "loss": 0.0663,
      "theoretical_loss": 3.426821205625199,
      "tokens_seen": 2241069056
    },
    {
      "epoch": 0.68,
      "objective/train/advantage_avg": 0.0005054355133324862,
      "objective/train/docs_used": 816099,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2898792028427124,
      "objective/train/original_loss": 1.2898790836334229,
      "objective/train/theoretical_loss": 3.4267888009674072,
      "objective/train/tokens_used": 2261791200,
      "objective/train/value_avg": -0.00934600830078125,
      "objective/train/value_loss": 0.000255974184256047,
      "objective/train/value_max": -1.0728836059570312e-05,
      "objective/train/value_min": -0.72900390625,
      "objective/train/value_reward_corr": 0.7769306544821992,
      "objective/train/value_std": 0.0211029052734375,
      "objective/train/weight_avg": 1.0006273984909058,
      "objective/train/weighted_lm_loss": 1.2898268699645996,
      "objective/train/weights_max": 1.6812392473220825,
      "objective/train/weights_min": 0.3951832354068756,
      "theoretical_loss": 3.4267888009674072,
      "tokens_seen": 2241331200
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032402503610977373,
      "loss": 0.0653,
      "theoretical_loss": 3.4267888009674072,
      "tokens_seen": 2241331200
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003239447921681913,
      "loss": 0.0664,
      "theoretical_loss": 3.4267564011604756,
      "tokens_seen": 2241593344
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003238645482266089,
      "loss": 0.0651,
      "theoretical_loss": 3.426724006203112,
      "tokens_seen": 2241855488
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032378430428502646,
      "loss": 0.0647,
      "theoretical_loss": 3.426691616094022,
      "tokens_seen": 2242117632
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032370406034344407,
      "loss": 0.0679,
      "theoretical_loss": 3.4266592308319144,
      "tokens_seen": 2242379776
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003236238164018617,
      "loss": 0.067,
      "theoretical_loss": 3.426626850415497,
      "tokens_seen": 2242641920
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032354357246027924,
      "loss": 0.0662,
      "theoretical_loss": 3.4265944748434785,
      "tokens_seen": 2242904064
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032346332851869685,
      "loss": 0.0662,
      "theoretical_loss": 3.4265621041145677,
      "tokens_seen": 2243166208
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032338308457711447,
      "loss": 0.0642,
      "theoretical_loss": 3.426529738227475,
      "tokens_seen": 2243428352
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.000323302840635532,
      "loss": 0.0643,
      "theoretical_loss": 3.4264973771809104,
      "tokens_seen": 2243690496
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003232225966939496,
      "loss": 0.0666,
      "theoretical_loss": 3.426465020973584,
      "tokens_seen": 2243952640
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003231423527523672,
      "loss": 0.0676,
      "theoretical_loss": 3.4264326696042073,
      "tokens_seen": 2244214784
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003230621088107848,
      "loss": 0.069,
      "theoretical_loss": 3.4264003230714923,
      "tokens_seen": 2244476928
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032298186486920237,
      "loss": 0.0651,
      "theoretical_loss": 3.4263679813741503,
      "tokens_seen": 2244739072
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032290162092762,
      "loss": 0.068,
      "theoretical_loss": 3.4263356445108943,
      "tokens_seen": 2245001216
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003228213769860376,
      "loss": 0.0648,
      "theoretical_loss": 3.426303312480438,
      "tokens_seen": 2245263360
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032274113304445515,
      "loss": 0.0688,
      "theoretical_loss": 3.426270985281494,
      "tokens_seen": 2245525504
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003226608891028727,
      "loss": 0.0667,
      "theoretical_loss": 3.426238662912777,
      "tokens_seen": 2245787648
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003225806451612903,
      "loss": 0.0678,
      "theoretical_loss": 3.4262063453730014,
      "tokens_seen": 2246049792
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032250040121970793,
      "loss": 0.0678,
      "theoretical_loss": 3.4261740326608825,
      "tokens_seen": 2246311936
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003224201572781255,
      "loss": 0.0646,
      "theoretical_loss": 3.4261417247751353,
      "tokens_seen": 2246574080
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003223399133365431,
      "loss": 0.0663,
      "theoretical_loss": 3.4261094217144765,
      "tokens_seen": 2246836224
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003222596693949607,
      "loss": 0.0681,
      "theoretical_loss": 3.4260771234776226,
      "tokens_seen": 2247098368
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003221794254533782,
      "loss": 0.0678,
      "theoretical_loss": 3.4260448300632906,
      "tokens_seen": 2247360512
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032209918151179584,
      "loss": 0.0679,
      "theoretical_loss": 3.4260125414701976,
      "tokens_seen": 2247622656
    },
    {
      "epoch": 0.68,
      "objective/train/advantage_avg": 0.0004595842328853905,
      "objective/train/docs_used": 818241,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.241592288017273,
      "objective/train/original_loss": 1.241592288017273,
      "objective/train/theoretical_loss": 3.4259802576970624,
      "objective/train/tokens_used": 2268344800,
      "objective/train/value_avg": -0.004749298095703125,
      "objective/train/value_loss": 0.00011237629951210693,
      "objective/train/value_max": -1.3947486877441406e-05,
      "objective/train/value_min": -0.71484375,
      "objective/train/value_reward_corr": 0.6671147463286998,
      "objective/train/value_std": 0.00948333740234375,
      "objective/train/weight_avg": 1.0005121231079102,
      "objective/train/weighted_lm_loss": 1.2421815395355225,
      "objective/train/weights_max": 1.2712233066558838,
      "objective/train/weights_min": 0.37275585532188416,
      "theoretical_loss": 3.4259802576970624,
      "tokens_seen": 2247884800
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032201893757021345,
      "loss": 0.0648,
      "theoretical_loss": 3.4259802576970624,
      "tokens_seen": 2247884800
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032193869362863106,
      "loss": 0.0671,
      "theoretical_loss": 3.425947978742603,
      "tokens_seen": 2248146944
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003218584496870486,
      "loss": 0.0666,
      "theoretical_loss": 3.425915704605538,
      "tokens_seen": 2248409088
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032177820574546623,
      "loss": 0.0664,
      "theoretical_loss": 3.4258834352845877,
      "tokens_seen": 2248671232
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032169796180388384,
      "loss": 0.0666,
      "theoretical_loss": 3.425851170778472,
      "tokens_seen": 2248933376
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003216177178623014,
      "loss": 0.0668,
      "theoretical_loss": 3.4258189110859107,
      "tokens_seen": 2249195520
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032153747392071896,
      "loss": 0.0657,
      "theoretical_loss": 3.425786656205626,
      "tokens_seen": 2249457664
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003214572299791366,
      "loss": 0.0671,
      "theoretical_loss": 3.425754406136338,
      "tokens_seen": 2249719808
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003213769860375542,
      "loss": 0.0649,
      "theoretical_loss": 3.425722160876769,
      "tokens_seen": 2249981952
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032129674209597174,
      "loss": 0.069,
      "theoretical_loss": 3.425689920425642,
      "tokens_seen": 2250244096
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032121649815438936,
      "loss": 0.0667,
      "theoretical_loss": 3.42565768478168,
      "tokens_seen": 2250506240
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032113625421280697,
      "loss": 0.0678,
      "theoretical_loss": 3.4256254539436055,
      "tokens_seen": 2250768384
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032105601027122453,
      "loss": 0.0669,
      "theoretical_loss": 3.4255932279101433,
      "tokens_seen": 2251030528
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003209757663296421,
      "loss": 0.0679,
      "theoretical_loss": 3.4255610066800166,
      "tokens_seen": 2251292672
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003208955223880597,
      "loss": 0.0672,
      "theoretical_loss": 3.425528790251952,
      "tokens_seen": 2251554816
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003208152784464773,
      "loss": 0.0673,
      "theoretical_loss": 3.4254965786246734,
      "tokens_seen": 2251816960
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032073503450489487,
      "loss": 0.0677,
      "theoretical_loss": 3.4254643717969073,
      "tokens_seen": 2252079104
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003206547905633125,
      "loss": 0.066,
      "theoretical_loss": 3.42543216976738,
      "tokens_seen": 2252341248
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003205745466217301,
      "loss": 0.0645,
      "theoretical_loss": 3.425399972534818,
      "tokens_seen": 2252603392
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032049430268014765,
      "loss": 0.0643,
      "theoretical_loss": 3.4253677800979494,
      "tokens_seen": 2252865536
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003204140587385652,
      "loss": 0.0677,
      "theoretical_loss": 3.425335592455501,
      "tokens_seen": 2253127680
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003203338147969828,
      "loss": 0.0668,
      "theoretical_loss": 3.4253034096062014,
      "tokens_seen": 2253389824
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00032025357085540044,
      "loss": 0.0706,
      "theoretical_loss": 3.4252712315487797,
      "tokens_seen": 2253651968
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.000320173326913818,
      "loss": 0.0691,
      "theoretical_loss": 3.4252390582819645,
      "tokens_seen": 2253914112
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003200930829722356,
      "loss": 0.0668,
      "theoretical_loss": 3.4252068898044863,
      "tokens_seen": 2254176256
    },
    {
      "epoch": 0.68,
      "objective/train/advantage_avg": -4.281154542695731e-05,
      "objective/train/docs_used": 820753,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3093576431274414,
      "objective/train/original_loss": 1.309357762336731,
      "objective/train/theoretical_loss": 3.425174726115075,
      "objective/train/tokens_used": 2274898400,
      "objective/train/value_avg": -0.0074005126953125,
      "objective/train/value_loss": 0.00020979443797841668,
      "objective/train/value_max": -3.045797348022461e-05,
      "objective/train/value_min": -0.39990234375,
      "objective/train/value_reward_corr": 0.6900546208748884,
      "objective/train/value_std": 0.013519287109375,
      "objective/train/weight_avg": 1.000051498413086,
      "objective/train/weighted_lm_loss": 1.309179425239563,
      "objective/train/weights_max": 1.2513753175735474,
      "objective/train/weights_min": 0.36851292848587036,
      "theoretical_loss": 3.425174726115075,
      "tokens_seen": 2254438400
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003200128390306532,
      "loss": 0.0646,
      "theoretical_loss": 3.425174726115075,
      "tokens_seen": 2254438400
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003199325950890708,
      "loss": 0.0659,
      "theoretical_loss": 3.425142567212461,
      "tokens_seen": 2254700544
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003198523511474884,
      "loss": 0.0655,
      "theoretical_loss": 3.4251104130953762,
      "tokens_seen": 2254962688
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00031977210720590595,
      "loss": 0.0692,
      "theoretical_loss": 3.4250782637625514,
      "tokens_seen": 2255224832
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003196918632643235,
      "loss": 0.0674,
      "theoretical_loss": 3.4250461192127193,
      "tokens_seen": 2255486976
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003196116193227411,
      "loss": 0.0669,
      "theoretical_loss": 3.4250139794446124,
      "tokens_seen": 2255749120
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00031953137538115873,
      "loss": 0.0636,
      "theoretical_loss": 3.4249818444569637,
      "tokens_seen": 2256011264
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00031945113143957635,
      "loss": 0.0649,
      "theoretical_loss": 3.424949714248507,
      "tokens_seen": 2256273408
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003193708874979939,
      "loss": 0.0664,
      "theoretical_loss": 3.4249175888179764,
      "tokens_seen": 2256535552
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003192906435564115,
      "loss": 0.0646,
      "theoretical_loss": 3.424885468164106,
      "tokens_seen": 2256797696
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003192103996148291,
      "loss": 0.0675,
      "theoretical_loss": 3.4248533522856315,
      "tokens_seen": 2257059840
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00031913015567324663,
      "loss": 0.0703,
      "theoretical_loss": 3.424821241181288,
      "tokens_seen": 2257321984
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00031904991173166425,
      "loss": 0.0676,
      "theoretical_loss": 3.4247891348498114,
      "tokens_seen": 2257584128
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00031896966779008186,
      "loss": 0.0651,
      "theoretical_loss": 3.424757033289939,
      "tokens_seen": 2257846272
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00031888942384849947,
      "loss": 0.0677,
      "theoretical_loss": 3.424724936500407,
      "tokens_seen": 2258108416
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00031880917990691703,
      "loss": 0.0658,
      "theoretical_loss": 3.424692844479953,
      "tokens_seen": 2258370560
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00031872893596533464,
      "loss": 0.0689,
      "theoretical_loss": 3.424660757227315,
      "tokens_seen": 2258632704
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00031864869202375226,
      "loss": 0.0696,
      "theoretical_loss": 3.4246286747412316,
      "tokens_seen": 2258894848
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00031856844808216976,
      "loss": 0.0691,
      "theoretical_loss": 3.4245965970204413,
      "tokens_seen": 2259156992
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003184882041405874,
      "loss": 0.0661,
      "theoretical_loss": 3.4245645240636833,
      "tokens_seen": 2259419136
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.000318407960199005,
      "loss": 0.0695,
      "theoretical_loss": 3.4245324558696986,
      "tokens_seen": 2259681280
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.0003183277162574226,
      "loss": 0.0673,
      "theoretical_loss": 3.4245003924372264,
      "tokens_seen": 2259943424
    },
    {
      "epoch": 0.68,
      "learning_rate": 0.00031824747231584016,
      "loss": 0.0637,
      "theoretical_loss": 3.424468333765008,
      "tokens_seen": 2260205568
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031816722837425777,
      "loss": 0.0672,
      "theoretical_loss": 3.4244362798517844,
      "tokens_seen": 2260467712
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003180869844326754,
      "loss": 0.065,
      "theoretical_loss": 3.4244042306962976,
      "tokens_seen": 2260729856
    },
    {
      "epoch": 0.69,
      "objective/train/advantage_avg": 0.0001269732601940632,
      "objective/train/docs_used": 823330,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3321760892868042,
      "objective/train/original_loss": 1.3321759700775146,
      "objective/train/theoretical_loss": 3.42437218629729,
      "objective/train/tokens_used": 2281452000,
      "objective/train/value_avg": -0.01062774658203125,
      "objective/train/value_loss": 0.00037752685602754354,
      "objective/train/value_max": -3.451108932495117e-05,
      "objective/train/value_min": -0.71044921875,
      "objective/train/value_reward_corr": 0.7460287555104631,
      "objective/train/value_std": 0.02197265625,
      "objective/train/weight_avg": 1.000304102897644,
      "objective/train/weighted_lm_loss": 1.3322685956954956,
      "objective/train/weights_max": 1.8926640748977661,
      "objective/train/weights_min": 0.38069406151771545,
      "theoretical_loss": 3.42437218629729,
      "tokens_seen": 2260992000
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003180067404910929,
      "loss": 0.0677,
      "theoretical_loss": 3.42437218629729,
      "tokens_seen": 2260992000
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003179264965495105,
      "loss": 0.0685,
      "theoretical_loss": 3.4243401466535044,
      "tokens_seen": 2261254144
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003178462526079281,
      "loss": 0.0697,
      "theoretical_loss": 3.424308111763683,
      "tokens_seen": 2261516288
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031776600866634567,
      "loss": 0.0666,
      "theoretical_loss": 3.4242760816265707,
      "tokens_seen": 2261778432
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003176857647247633,
      "loss": 0.0699,
      "theoretical_loss": 3.4242440562409113,
      "tokens_seen": 2262040576
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003176055207831809,
      "loss": 0.0685,
      "theoretical_loss": 3.424212035605449,
      "tokens_seen": 2262302720
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003175252768415985,
      "loss": 0.0694,
      "theoretical_loss": 3.4241800197189294,
      "tokens_seen": 2262564864
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.000317445032900016,
      "loss": 0.0672,
      "theoretical_loss": 3.4241480085800977,
      "tokens_seen": 2262827008
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003173647889584336,
      "loss": 0.0685,
      "theoretical_loss": 3.4241160021877004,
      "tokens_seen": 2263089152
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031728454501685124,
      "loss": 0.0695,
      "theoretical_loss": 3.424084000540484,
      "tokens_seen": 2263351296
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003172043010752688,
      "loss": 0.0662,
      "theoretical_loss": 3.424052003637195,
      "tokens_seen": 2263613440
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003171240571336864,
      "loss": 0.0682,
      "theoretical_loss": 3.4240200114765815,
      "tokens_seen": 2263875584
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.000317043813192104,
      "loss": 0.0653,
      "theoretical_loss": 3.423988024057391,
      "tokens_seen": 2264137728
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031696356925052163,
      "loss": 0.0692,
      "theoretical_loss": 3.423956041378373,
      "tokens_seen": 2264399872
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003168833253089392,
      "loss": 0.066,
      "theoretical_loss": 3.423924063438275,
      "tokens_seen": 2264662016
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031680308136735675,
      "loss": 0.0656,
      "theoretical_loss": 3.4238920902358467,
      "tokens_seen": 2264924160
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031672283742577436,
      "loss": 0.0665,
      "theoretical_loss": 3.423860121769839,
      "tokens_seen": 2265186304
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003166425934841919,
      "loss": 0.0661,
      "theoretical_loss": 3.423828158039001,
      "tokens_seen": 2265448448
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031656234954260953,
      "loss": 0.0695,
      "theoretical_loss": 3.4237961990420844,
      "tokens_seen": 2265710592
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031648210560102715,
      "loss": 0.0668,
      "theoretical_loss": 3.42376424477784,
      "tokens_seen": 2265972736
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031640186165944476,
      "loss": 0.0675,
      "theoretical_loss": 3.4237322952450198,
      "tokens_seen": 2266234880
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003163216177178623,
      "loss": 0.0677,
      "theoretical_loss": 3.423700350442376,
      "tokens_seen": 2266497024
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003162413737762799,
      "loss": 0.0659,
      "theoretical_loss": 3.4236684103686614,
      "tokens_seen": 2266759168
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003161611298346975,
      "loss": 0.068,
      "theoretical_loss": 3.423636475022629,
      "tokens_seen": 2267021312
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031608088589311505,
      "loss": 0.067,
      "theoretical_loss": 3.423604544403032,
      "tokens_seen": 2267283456
    },
    {
      "epoch": 0.69,
      "objective/train/advantage_avg": 0.00016212603077292442,
      "objective/train/docs_used": 825734,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2646903991699219,
      "objective/train/original_loss": 1.2646903991699219,
      "objective/train/theoretical_loss": 3.4235726185086257,
      "objective/train/tokens_used": 2288005600,
      "objective/train/value_avg": -0.0071868896484375,
      "objective/train/value_loss": 0.0001659700064919889,
      "objective/train/value_max": -3.9458274841308594e-05,
      "objective/train/value_min": -0.3583984375,
      "objective/train/value_reward_corr": 0.6855385326256844,
      "objective/train/value_std": 0.01271820068359375,
      "objective/train/weight_avg": 1.0002398490905762,
      "objective/train/weighted_lm_loss": 1.2645337581634521,
      "objective/train/weights_max": 1.3612432479858398,
      "objective/train/weights_min": 0.37317126989364624,
      "theoretical_loss": 3.4235726185086257,
      "tokens_seen": 2267545600
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031600064195153266,
      "loss": 0.0663,
      "theoretical_loss": 3.4235726185086257,
      "tokens_seen": 2267545600
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031592039800995027,
      "loss": 0.0663,
      "theoretical_loss": 3.423540697338164,
      "tokens_seen": 2267807744
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003158401540683679,
      "loss": 0.0673,
      "theoretical_loss": 3.423508780890402,
      "tokens_seen": 2268069888
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031575991012678544,
      "loss": 0.0678,
      "theoretical_loss": 3.4234768691640953,
      "tokens_seen": 2268332032
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.000315679666185203,
      "loss": 0.0661,
      "theoretical_loss": 3.4234449621580003,
      "tokens_seen": 2268594176
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003155994222436206,
      "loss": 0.0669,
      "theoretical_loss": 3.4234130598708727,
      "tokens_seen": 2268856320
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031551917830203817,
      "loss": 0.0667,
      "theoretical_loss": 3.42338116230147,
      "tokens_seen": 2269118464
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003154389343604558,
      "loss": 0.0662,
      "theoretical_loss": 3.4233492694485497,
      "tokens_seen": 2269380608
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003153586904188734,
      "loss": 0.0641,
      "theoretical_loss": 3.4233173813108695,
      "tokens_seen": 2269642752
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031527844647729096,
      "loss": 0.0662,
      "theoretical_loss": 3.423285497887188,
      "tokens_seen": 2269904896
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031519820253570857,
      "loss": 0.0672,
      "theoretical_loss": 3.4232536191762635,
      "tokens_seen": 2270167040
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003151179585941262,
      "loss": 0.0683,
      "theoretical_loss": 3.423221745176856,
      "tokens_seen": 2270429184
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031503771465254374,
      "loss": 0.0648,
      "theoretical_loss": 3.423189875887725,
      "tokens_seen": 2270691328
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003149574707109613,
      "loss": 0.0648,
      "theoretical_loss": 3.4231580113076303,
      "tokens_seen": 2270953472
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003148772267693789,
      "loss": 0.0669,
      "theoretical_loss": 3.4231261514353335,
      "tokens_seen": 2271215616
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003147969828277965,
      "loss": 0.0649,
      "theoretical_loss": 3.4230942962695954,
      "tokens_seen": 2271477760
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003147167388862141,
      "loss": 0.0707,
      "theoretical_loss": 3.4230624458091774,
      "tokens_seen": 2271739904
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003146364949446317,
      "loss": 0.0684,
      "theoretical_loss": 3.423030600052842,
      "tokens_seen": 2272002048
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003145562510030493,
      "loss": 0.0646,
      "theoretical_loss": 3.422998758999352,
      "tokens_seen": 2272264192
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031447600706146687,
      "loss": 0.0629,
      "theoretical_loss": 3.4229669226474697,
      "tokens_seen": 2272526336
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003143957631198844,
      "loss": 0.0669,
      "theoretical_loss": 3.422935090995959,
      "tokens_seen": 2272788480
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031431551917830204,
      "loss": 0.0675,
      "theoretical_loss": 3.4229032640435846,
      "tokens_seen": 2273050624
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031423527523671965,
      "loss": 0.0655,
      "theoretical_loss": 3.42287144178911,
      "tokens_seen": 2273312768
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003141550312951372,
      "loss": 0.0715,
      "theoretical_loss": 3.4228396242313006,
      "tokens_seen": 2273574912
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003140747873535548,
      "loss": 0.0647,
      "theoretical_loss": 3.422807811368922,
      "tokens_seen": 2273837056
    },
    {
      "epoch": 0.69,
      "objective/train/advantage_avg": 0.001269850879907608,
      "objective/train/docs_used": 828122,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3244622945785522,
      "objective/train/original_loss": 1.3244624137878418,
      "objective/train/theoretical_loss": 3.422776003200739,
      "objective/train/tokens_used": 2294559200,
      "objective/train/value_avg": -0.007564544677734375,
      "objective/train/value_loss": 0.0001630596088944003,
      "objective/train/value_max": -2.0623207092285156e-05,
      "objective/train/value_min": -0.60986328125,
      "objective/train/value_reward_corr": 0.7459873437025508,
      "objective/train/value_std": 0.01558685302734375,
      "objective/train/weight_avg": 1.0013480186462402,
      "objective/train/weighted_lm_loss": 1.326235294342041,
      "objective/train/weights_max": 1.236796259880066,
      "objective/train/weights_min": 0.39109206199645996,
      "theoretical_loss": 3.422776003200739,
      "tokens_seen": 2274099200
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031399454341197243,
      "loss": 0.0658,
      "theoretical_loss": 3.422776003200739,
      "tokens_seen": 2274099200
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031391429947039004,
      "loss": 0.0709,
      "theoretical_loss": 3.4227441997255195,
      "tokens_seen": 2274361344
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031383405552880755,
      "loss": 0.0695,
      "theoretical_loss": 3.4227124009420296,
      "tokens_seen": 2274623488
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031375381158722516,
      "loss": 0.0655,
      "theoretical_loss": 3.422680606849036,
      "tokens_seen": 2274885632
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003136735676456428,
      "loss": 0.0658,
      "theoretical_loss": 3.4226488174453076,
      "tokens_seen": 2275147776
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031359332370406033,
      "loss": 0.0678,
      "theoretical_loss": 3.4226170327296113,
      "tokens_seen": 2275409920
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031351307976247795,
      "loss": 0.07,
      "theoretical_loss": 3.422585252700717,
      "tokens_seen": 2275672064
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031343283582089556,
      "loss": 0.0678,
      "theoretical_loss": 3.422553477357393,
      "tokens_seen": 2275934208
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003133525918793131,
      "loss": 0.0669,
      "theoretical_loss": 3.4225217066984093,
      "tokens_seen": 2276196352
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003132723479377307,
      "loss": 0.0645,
      "theoretical_loss": 3.422489940722536,
      "tokens_seen": 2276458496
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003131921039961483,
      "loss": 0.0664,
      "theoretical_loss": 3.422458179428543,
      "tokens_seen": 2276720640
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003131118600545659,
      "loss": 0.0651,
      "theoretical_loss": 3.422426422815202,
      "tokens_seen": 2276982784
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031303161611298346,
      "loss": 0.0669,
      "theoretical_loss": 3.4223946708812845,
      "tokens_seen": 2277244928
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031295137217140107,
      "loss": 0.0681,
      "theoretical_loss": 3.422362923625562,
      "tokens_seen": 2277507072
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003128711282298187,
      "loss": 0.067,
      "theoretical_loss": 3.422331181046807,
      "tokens_seen": 2277769216
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031279088428823624,
      "loss": 0.0666,
      "theoretical_loss": 3.4222994431437925,
      "tokens_seen": 2278031360
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003127106403466538,
      "loss": 0.0658,
      "theoretical_loss": 3.4222677099152916,
      "tokens_seen": 2278293504
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003126303964050714,
      "loss": 0.0666,
      "theoretical_loss": 3.4222359813600782,
      "tokens_seen": 2278555648
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.000312550152463489,
      "loss": 0.0646,
      "theoretical_loss": 3.422204257476927,
      "tokens_seen": 2278817792
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003124699085219066,
      "loss": 0.0645,
      "theoretical_loss": 3.4221725382646118,
      "tokens_seen": 2279079936
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003123896645803242,
      "loss": 0.0663,
      "theoretical_loss": 3.422140823721908,
      "tokens_seen": 2279342080
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003123094206387418,
      "loss": 0.0684,
      "theoretical_loss": 3.422109113847592,
      "tokens_seen": 2279604224
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031222917669715937,
      "loss": 0.0644,
      "theoretical_loss": 3.422077408640439,
      "tokens_seen": 2279866368
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.000312148932755577,
      "loss": 0.0671,
      "theoretical_loss": 3.422045708099226,
      "tokens_seen": 2280128512
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031206868881399454,
      "loss": 0.0687,
      "theoretical_loss": 3.42201401222273,
      "tokens_seen": 2280390656
    },
    {
      "epoch": 0.69,
      "objective/train/advantage_avg": 0.0003086750512011349,
      "objective/train/docs_used": 830439,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2479599714279175,
      "objective/train/original_loss": 1.2479599714279175,
      "objective/train/theoretical_loss": 3.421982321009728,
      "objective/train/tokens_used": 2301112800,
      "objective/train/value_avg": -0.00862884521484375,
      "objective/train/value_loss": 0.0004476759640965611,
      "objective/train/value_max": -4.5418739318847656e-05,
      "objective/train/value_min": -0.66748046875,
      "objective/train/value_reward_corr": 0.6656916804020896,
      "objective/train/value_std": 0.0174102783203125,
      "objective/train/weight_avg": 1.0004971027374268,
      "objective/train/weighted_lm_loss": 1.2481622695922852,
      "objective/train/weights_max": 1.3128468990325928,
      "objective/train/weights_min": 0.22669236361980438,
      "theoretical_loss": 3.421982321009728,
      "tokens_seen": 2280652800
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031198844487241215,
      "loss": 0.0665,
      "theoretical_loss": 3.421982321009728,
      "tokens_seen": 2280652800
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003119082009308297,
      "loss": 0.0675,
      "theoretical_loss": 3.4219506344589985,
      "tokens_seen": 2280914944
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003118279569892473,
      "loss": 0.0659,
      "theoretical_loss": 3.421918952569319,
      "tokens_seen": 2281177088
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031174771304766494,
      "loss": 0.0643,
      "theoretical_loss": 3.4218872753394702,
      "tokens_seen": 2281439232
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003116674691060825,
      "loss": 0.0663,
      "theoretical_loss": 3.4218556027682294,
      "tokens_seen": 2281701376
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003115872251645001,
      "loss": 0.0645,
      "theoretical_loss": 3.4218239348543777,
      "tokens_seen": 2281963520
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031150698122291766,
      "loss": 0.0677,
      "theoretical_loss": 3.4217922715966944,
      "tokens_seen": 2282225664
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003114267372813352,
      "loss": 0.0662,
      "theoretical_loss": 3.421760612993961,
      "tokens_seen": 2282487808
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031134649333975284,
      "loss": 0.0663,
      "theoretical_loss": 3.421728959044958,
      "tokens_seen": 2282749952
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031126624939817045,
      "loss": 0.066,
      "theoretical_loss": 3.4216973097484678,
      "tokens_seen": 2283012096
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031118600545658806,
      "loss": 0.0667,
      "theoretical_loss": 3.4216656651032715,
      "tokens_seen": 2283274240
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003111057615150056,
      "loss": 0.0654,
      "theoretical_loss": 3.421634025108152,
      "tokens_seen": 2283536384
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031102551757342323,
      "loss": 0.0659,
      "theoretical_loss": 3.421602389761893,
      "tokens_seen": 2283798528
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003109452736318408,
      "loss": 0.0664,
      "theoretical_loss": 3.421570759063277,
      "tokens_seen": 2284060672
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031086502969025835,
      "loss": 0.0668,
      "theoretical_loss": 3.4215391330110885,
      "tokens_seen": 2284322816
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031078478574867596,
      "loss": 0.0677,
      "theoretical_loss": 3.421507511604111,
      "tokens_seen": 2284584960
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003107045418070936,
      "loss": 0.065,
      "theoretical_loss": 3.42147589484113,
      "tokens_seen": 2284847104
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003106242978655112,
      "loss": 0.0681,
      "theoretical_loss": 3.4214442827209313,
      "tokens_seen": 2285109248
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031054405392392874,
      "loss": 0.0632,
      "theoretical_loss": 3.4214126752423,
      "tokens_seen": 2285371392
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031046380998234636,
      "loss": 0.0653,
      "theoretical_loss": 3.4213810724040217,
      "tokens_seen": 2285633536
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00031038356604076397,
      "loss": 0.065,
      "theoretical_loss": 3.421349474204884,
      "tokens_seen": 2285895680
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003103033220991815,
      "loss": 0.0661,
      "theoretical_loss": 3.4213178806436737,
      "tokens_seen": 2286157824
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003102230781575991,
      "loss": 0.0637,
      "theoretical_loss": 3.4212862917191784,
      "tokens_seen": 2286419968
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003101428342160167,
      "loss": 0.0659,
      "theoretical_loss": 3.4212547074301862,
      "tokens_seen": 2286682112
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003100625902744343,
      "loss": 0.0645,
      "theoretical_loss": 3.4212231277754848,
      "tokens_seen": 2286944256
    },
    {
      "epoch": 0.69,
      "objective/train/advantage_avg": -0.0003998603788204491,
      "objective/train/docs_used": 832763,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.245110273361206,
      "objective/train/original_loss": 1.245110273361206,
      "objective/train/theoretical_loss": 3.4211915527538643,
      "objective/train/tokens_used": 2307666400,
      "objective/train/value_avg": -0.00817108154296875,
      "objective/train/value_loss": 0.00033646036172285676,
      "objective/train/value_max": -3.349781036376953e-05,
      "objective/train/value_min": -0.95166015625,
      "objective/train/value_reward_corr": 0.7729082380774197,
      "objective/train/value_std": 0.0215911865234375,
      "objective/train/weight_avg": 0.9997501373291016,
      "objective/train/weighted_lm_loss": 1.2448906898498535,
      "objective/train/weights_max": 1.6519445180892944,
      "objective/train/weights_min": 0.36829686164855957,
      "theoretical_loss": 3.4211915527538643,
      "tokens_seen": 2287206400
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00030998234633285187,
      "loss": 0.0652,
      "theoretical_loss": 3.4211915527538643,
      "tokens_seen": 2287206400
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003099021023912695,
      "loss": 0.0649,
      "theoretical_loss": 3.4211599823641134,
      "tokens_seen": 2287468544
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003098218584496871,
      "loss": 0.0668,
      "theoretical_loss": 3.421128416605022,
      "tokens_seen": 2287730688
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003097416145081046,
      "loss": 0.0662,
      "theoretical_loss": 3.4210968554753807,
      "tokens_seen": 2287992832
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003096613705665222,
      "loss": 0.0651,
      "theoretical_loss": 3.42106529897398,
      "tokens_seen": 2288254976
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003095811266249398,
      "loss": 0.0669,
      "theoretical_loss": 3.4210337470996106,
      "tokens_seen": 2288517120
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003095008826833574,
      "loss": 0.0664,
      "theoretical_loss": 3.4210021998510647,
      "tokens_seen": 2288779264
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.000309420638741775,
      "loss": 0.0663,
      "theoretical_loss": 3.4209706572271346,
      "tokens_seen": 2289041408
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003093403948001926,
      "loss": 0.0668,
      "theoretical_loss": 3.4209391192266128,
      "tokens_seen": 2289303552
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003092601508586102,
      "loss": 0.0672,
      "theoretical_loss": 3.420907585848292,
      "tokens_seen": 2289565696
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003091799069170278,
      "loss": 0.0669,
      "theoretical_loss": 3.4208760570909655,
      "tokens_seen": 2289827840
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00030909966297544534,
      "loss": 0.0649,
      "theoretical_loss": 3.420844532953428,
      "tokens_seen": 2290089984
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00030901941903386295,
      "loss": 0.0677,
      "theoretical_loss": 3.420813013434473,
      "tokens_seen": 2290352128
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003089391750922805,
      "loss": 0.0655,
      "theoretical_loss": 3.4207814985328957,
      "tokens_seen": 2290614272
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003088589311506981,
      "loss": 0.0655,
      "theoretical_loss": 3.4207499882474917,
      "tokens_seen": 2290876416
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00030877868720911573,
      "loss": 0.0632,
      "theoretical_loss": 3.4207184825770565,
      "tokens_seen": 2291138560
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00030869844326753335,
      "loss": 0.0643,
      "theoretical_loss": 3.420686981520386,
      "tokens_seen": 2291400704
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003086181993259509,
      "loss": 0.0665,
      "theoretical_loss": 3.420655485076277,
      "tokens_seen": 2291662848
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00030853795538436846,
      "loss": 0.0652,
      "theoretical_loss": 3.420623993243527,
      "tokens_seen": 2291924992
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.0003084577114427861,
      "loss": 0.0661,
      "theoretical_loss": 3.420592506020933,
      "tokens_seen": 2292187136
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00030837746750120363,
      "loss": 0.0646,
      "theoretical_loss": 3.420561023407293,
      "tokens_seen": 2292449280
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00030829722355962125,
      "loss": 0.0675,
      "theoretical_loss": 3.420529545401406,
      "tokens_seen": 2292711424
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00030821697961803886,
      "loss": 0.067,
      "theoretical_loss": 3.42049807200207,
      "tokens_seen": 2292973568
    },
    {
      "epoch": 0.69,
      "learning_rate": 0.00030813673567645647,
      "loss": 0.0659,
      "theoretical_loss": 3.420466603208085,
      "tokens_seen": 2293235712
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030805649173487403,
      "loss": 0.0634,
      "theoretical_loss": 3.4204351390182506,
      "tokens_seen": 2293497856
    },
    {
      "epoch": 0.7,
      "objective/train/advantage_avg": 0.001006623962894082,
      "objective/train/docs_used": 835219,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2900722026824951,
      "objective/train/original_loss": 1.2900723218917847,
      "objective/train/theoretical_loss": 3.4204036794313675,
      "objective/train/tokens_used": 2314220000,
      "objective/train/value_avg": -0.006023406982421875,
      "objective/train/value_loss": 0.00020214400137774646,
      "objective/train/value_max": -2.6702880859375e-05,
      "objective/train/value_min": -0.407470703125,
      "objective/train/value_reward_corr": 0.6690516829985103,
      "objective/train/value_std": 0.0127410888671875,
      "objective/train/weight_avg": 1.0010955333709717,
      "objective/train/weighted_lm_loss": 1.2918227910995483,
      "objective/train/weights_max": 1.4126380681991577,
      "objective/train/weights_min": 0.30655041337013245,
      "theoretical_loss": 3.4204036794313675,
      "tokens_seen": 2293760000
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003079762477932916,
      "loss": 0.0674,
      "theoretical_loss": 3.4204036794313675,
      "tokens_seen": 2293760000
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003078960038517092,
      "loss": 0.0672,
      "theoretical_loss": 3.4203722244462353,
      "tokens_seen": 2294022144
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030781575991012676,
      "loss": 0.0693,
      "theoretical_loss": 3.4203407740616565,
      "tokens_seen": 2294284288
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003077355159685444,
      "loss": 0.0697,
      "theoretical_loss": 3.4203093282764314,
      "tokens_seen": 2294546432
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.000307655272026962,
      "loss": 0.0652,
      "theoretical_loss": 3.420277887089363,
      "tokens_seen": 2294808576
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003075750280853796,
      "loss": 0.064,
      "theoretical_loss": 3.420246450499253,
      "tokens_seen": 2295070720
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030749478414379716,
      "loss": 0.0642,
      "theoretical_loss": 3.420215018504905,
      "tokens_seen": 2295332864
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030741454020221477,
      "loss": 0.0629,
      "theoretical_loss": 3.4201835911051224,
      "tokens_seen": 2295595008
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030733429626063233,
      "loss": 0.0685,
      "theoretical_loss": 3.4201521682987086,
      "tokens_seen": 2295857152
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003072540523190499,
      "loss": 0.0681,
      "theoretical_loss": 3.420120750084468,
      "tokens_seen": 2296119296
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003071738083774675,
      "loss": 0.0672,
      "theoretical_loss": 3.4200893364612055,
      "tokens_seen": 2296381440
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003070935644358851,
      "loss": 0.0688,
      "theoretical_loss": 3.4200579274277256,
      "tokens_seen": 2296643584
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030701332049430267,
      "loss": 0.0669,
      "theoretical_loss": 3.420026522982835,
      "tokens_seen": 2296905728
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003069330765527203,
      "loss": 0.0657,
      "theoretical_loss": 3.4199951231253394,
      "tokens_seen": 2297167872
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003068528326111379,
      "loss": 0.0666,
      "theoretical_loss": 3.4199637278540447,
      "tokens_seen": 2297430016
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030677258866955545,
      "loss": 0.0647,
      "theoretical_loss": 3.4199323371677584,
      "tokens_seen": 2297692160
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.000306692344727973,
      "loss": 0.066,
      "theoretical_loss": 3.419900951065288,
      "tokens_seen": 2297954304
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003066121007863906,
      "loss": 0.0661,
      "theoretical_loss": 3.419869569545441,
      "tokens_seen": 2298216448
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030653185684480824,
      "loss": 0.0653,
      "theoretical_loss": 3.419838192607026,
      "tokens_seen": 2298478592
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003064516129032258,
      "loss": 0.0629,
      "theoretical_loss": 3.4198068202488514,
      "tokens_seen": 2298740736
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003063713689616434,
      "loss": 0.067,
      "theoretical_loss": 3.419775452469727,
      "tokens_seen": 2299002880
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.000306291125020061,
      "loss": 0.0659,
      "theoretical_loss": 3.4197440892684616,
      "tokens_seen": 2299265024
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003062108810784786,
      "loss": 0.0661,
      "theoretical_loss": 3.4197127306438664,
      "tokens_seen": 2299527168
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030613063713689614,
      "loss": 0.0686,
      "theoretical_loss": 3.4196813765947507,
      "tokens_seen": 2299789312
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030605039319531375,
      "loss": 0.0682,
      "theoretical_loss": 3.4196500271199266,
      "tokens_seen": 2300051456
    },
    {
      "epoch": 0.7,
      "objective/train/advantage_avg": 3.6831370380241424e-05,
      "objective/train/docs_used": 837515,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3956961631774902,
      "objective/train/original_loss": 1.3956961631774902,
      "objective/train/theoretical_loss": 3.4196186822182044,
      "objective/train/tokens_used": 2320773600,
      "objective/train/value_avg": -0.00783538818359375,
      "objective/train/value_loss": 0.0005023198900744319,
      "objective/train/value_max": -3.5643577575683594e-05,
      "objective/train/value_min": -0.91845703125,
      "objective/train/value_reward_corr": 0.6416035836441893,
      "objective/train/value_std": 0.0215606689453125,
      "objective/train/weight_avg": 1.0002703666687012,
      "objective/train/weighted_lm_loss": 1.3962715864181519,
      "objective/train/weights_max": 2.0975584983825684,
      "objective/train/weights_min": 0.368789941072464,
      "theoretical_loss": 3.4196186822182044,
      "tokens_seen": 2300313600
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030597014925373136,
      "loss": 0.0679,
      "theoretical_loss": 3.4196186822182044,
      "tokens_seen": 2300313600
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003058899053121489,
      "loss": 0.0652,
      "theoretical_loss": 3.419587341888397,
      "tokens_seen": 2300575744
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030580966137056653,
      "loss": 0.0644,
      "theoretical_loss": 3.4195560061293158,
      "tokens_seen": 2300837888
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030572941742898415,
      "loss": 0.0675,
      "theoretical_loss": 3.4195246749397743,
      "tokens_seen": 2301100032
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030564917348740176,
      "loss": 0.0641,
      "theoretical_loss": 3.4194933483185856,
      "tokens_seen": 2301362176
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030556892954581926,
      "loss": 0.0661,
      "theoretical_loss": 3.4194620262645627,
      "tokens_seen": 2301624320
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003054886856042369,
      "loss": 0.0686,
      "theoretical_loss": 3.4194307087765203,
      "tokens_seen": 2301886464
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003054084416626545,
      "loss": 0.0698,
      "theoretical_loss": 3.4193993958532722,
      "tokens_seen": 2302148608
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030532819772107205,
      "loss": 0.0649,
      "theoretical_loss": 3.4193680874936345,
      "tokens_seen": 2302410752
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030524795377948966,
      "loss": 0.0671,
      "theoretical_loss": 3.419336783696422,
      "tokens_seen": 2302672896
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030516770983790727,
      "loss": 0.0666,
      "theoretical_loss": 3.41930548446045,
      "tokens_seen": 2302935040
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030508746589632483,
      "loss": 0.0689,
      "theoretical_loss": 3.419274189784536,
      "tokens_seen": 2303197184
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003050072219547424,
      "loss": 0.0677,
      "theoretical_loss": 3.4192428996674957,
      "tokens_seen": 2303459328
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030492697801316,
      "loss": 0.0696,
      "theoretical_loss": 3.4192116141081463,
      "tokens_seen": 2303721472
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003048467340715776,
      "loss": 0.0677,
      "theoretical_loss": 3.4191803331053063,
      "tokens_seen": 2303983616
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030476649012999517,
      "loss": 0.0673,
      "theoretical_loss": 3.419149056657793,
      "tokens_seen": 2304245760
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003046862461884128,
      "loss": 0.0647,
      "theoretical_loss": 3.4191177847644254,
      "tokens_seen": 2304507904
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003046060022468304,
      "loss": 0.0675,
      "theoretical_loss": 3.419086517424022,
      "tokens_seen": 2304770048
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030452575830524796,
      "loss": 0.0666,
      "theoretical_loss": 3.4190552546354023,
      "tokens_seen": 2305032192
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030444551436366557,
      "loss": 0.0673,
      "theoretical_loss": 3.419023996397386,
      "tokens_seen": 2305294336
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003043652704220831,
      "loss": 0.065,
      "theoretical_loss": 3.418992742708794,
      "tokens_seen": 2305556480
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030428502648050074,
      "loss": 0.0654,
      "theoretical_loss": 3.418961493568446,
      "tokens_seen": 2305818624
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003042047825389183,
      "loss": 0.0671,
      "theoretical_loss": 3.4189302489751636,
      "tokens_seen": 2306080768
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003041245385973359,
      "loss": 0.0664,
      "theoretical_loss": 3.418899008927769,
      "tokens_seen": 2306342912
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003040442946557535,
      "loss": 0.0658,
      "theoretical_loss": 3.4188677734250836,
      "tokens_seen": 2306605056
    },
    {
      "epoch": 0.7,
      "objective/train/advantage_avg": 0.001095419516786933,
      "objective/train/docs_used": 839893,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.38678777217865,
      "objective/train/original_loss": 1.3867876529693604,
      "objective/train/theoretical_loss": 3.4188365424659293,
      "objective/train/tokens_used": 2327327200,
      "objective/train/value_avg": -0.0091094970703125,
      "objective/train/value_loss": 0.0002580058353487402,
      "objective/train/value_max": -4.3332576751708984e-05,
      "objective/train/value_min": -0.86572265625,
      "objective/train/value_reward_corr": 0.7183997951826784,
      "objective/train/value_std": 0.0175933837890625,
      "objective/train/weight_avg": 1.0012105703353882,
      "objective/train/weighted_lm_loss": 1.3880761861801147,
      "objective/train/weights_max": 1.3674609661102295,
      "objective/train/weights_min": 0.3692784309387207,
      "theoretical_loss": 3.4188365424659293,
      "tokens_seen": 2306867200
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003039640507141711,
      "loss": 0.0671,
      "theoretical_loss": 3.4188365424659293,
      "tokens_seen": 2306867200
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003038838067725887,
      "loss": 0.066,
      "theoretical_loss": 3.4188053160491303,
      "tokens_seen": 2307129344
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030380356283100625,
      "loss": 0.0663,
      "theoretical_loss": 3.418774094173509,
      "tokens_seen": 2307391488
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030372331888942387,
      "loss": 0.0659,
      "theoretical_loss": 3.4187428768378894,
      "tokens_seen": 2307653632
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003036430749478414,
      "loss": 0.0671,
      "theoretical_loss": 3.4187116640410955,
      "tokens_seen": 2307915776
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030356283100625904,
      "loss": 0.0657,
      "theoretical_loss": 3.418680455781953,
      "tokens_seen": 2308177920
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030348258706467665,
      "loss": 0.0665,
      "theoretical_loss": 3.4186492520592853,
      "tokens_seen": 2308440064
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003034023431230942,
      "loss": 0.0707,
      "theoretical_loss": 3.4186180528719188,
      "tokens_seen": 2308702208
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003033220991815118,
      "loss": 0.0684,
      "theoretical_loss": 3.41858685821868,
      "tokens_seen": 2308964352
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003032418552399294,
      "loss": 0.0682,
      "theoretical_loss": 3.418555668098395,
      "tokens_seen": 2309226496
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030316161129834694,
      "loss": 0.0683,
      "theoretical_loss": 3.41852448250989,
      "tokens_seen": 2309488640
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030308136735676455,
      "loss": 0.0653,
      "theoretical_loss": 3.4184933014519925,
      "tokens_seen": 2309750784
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030300112341518216,
      "loss": 0.065,
      "theoretical_loss": 3.4184621249235305,
      "tokens_seen": 2310012928
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003029208794735998,
      "loss": 0.0659,
      "theoretical_loss": 3.418430952923332,
      "tokens_seen": 2310275072
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030284063553201733,
      "loss": 0.0685,
      "theoretical_loss": 3.418399785450226,
      "tokens_seen": 2310537216
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030276039159043495,
      "loss": 0.0652,
      "theoretical_loss": 3.418368622503041,
      "tokens_seen": 2310799360
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030268014764885256,
      "loss": 0.0659,
      "theoretical_loss": 3.4183374640806066,
      "tokens_seen": 2311061504
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030259990370727006,
      "loss": 0.0671,
      "theoretical_loss": 3.4183063101817526,
      "tokens_seen": 2311323648
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003025196597656877,
      "loss": 0.0686,
      "theoretical_loss": 3.41827516080531,
      "tokens_seen": 2311585792
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003024394158241053,
      "loss": 0.0698,
      "theoretical_loss": 3.418244015950108,
      "tokens_seen": 2311847936
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003023591718825229,
      "loss": 0.0654,
      "theoretical_loss": 3.418212875614979,
      "tokens_seen": 2312110080
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030227892794094046,
      "loss": 0.0663,
      "theoretical_loss": 3.4181817397987553,
      "tokens_seen": 2312372224
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030219868399935807,
      "loss": 0.0669,
      "theoretical_loss": 3.4181506085002673,
      "tokens_seen": 2312634368
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003021184400577757,
      "loss": 0.0668,
      "theoretical_loss": 3.4181194817183487,
      "tokens_seen": 2312896512
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003020381961161932,
      "loss": 0.0669,
      "theoretical_loss": 3.418088359451832,
      "tokens_seen": 2313158656
    },
    {
      "epoch": 0.7,
      "objective/train/advantage_avg": 0.0010467276442795992,
      "objective/train/docs_used": 842373,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.314192771911621,
      "objective/train/original_loss": 1.314192771911621,
      "objective/train/theoretical_loss": 3.41805724169955,
      "objective/train/tokens_used": 2333880800,
      "objective/train/value_avg": -0.005832672119140625,
      "objective/train/value_loss": 0.00015348766464740038,
      "objective/train/value_max": -2.467632293701172e-05,
      "objective/train/value_min": -0.3525390625,
      "objective/train/value_reward_corr": 0.6610346152446436,
      "objective/train/value_std": 0.0113677978515625,
      "objective/train/weight_avg": 1.0011154413223267,
      "objective/train/weighted_lm_loss": 1.3154722452163696,
      "objective/train/weights_max": 1.3709473609924316,
      "objective/train/weights_min": 0.37706345319747925,
      "theoretical_loss": 3.41805724169955,
      "tokens_seen": 2313420800
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003019579521746108,
      "loss": 0.0679,
      "theoretical_loss": 3.41805724169955,
      "tokens_seen": 2313420800
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003018777082330284,
      "loss": 0.0677,
      "theoretical_loss": 3.4180261284603373,
      "tokens_seen": 2313682944
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.000301797464291446,
      "loss": 0.0693,
      "theoretical_loss": 3.417995019733028,
      "tokens_seen": 2313945088
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003017172203498636,
      "loss": 0.0661,
      "theoretical_loss": 3.4179639155164567,
      "tokens_seen": 2314207232
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003016369764082812,
      "loss": 0.0667,
      "theoretical_loss": 3.4179328158094586,
      "tokens_seen": 2314469376
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003015567324666988,
      "loss": 0.0662,
      "theoretical_loss": 3.417901720610869,
      "tokens_seen": 2314731520
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003014764885251163,
      "loss": 0.0642,
      "theoretical_loss": 3.417870629919524,
      "tokens_seen": 2314993664
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003013962445835339,
      "loss": 0.0678,
      "theoretical_loss": 3.41783954373426,
      "tokens_seen": 2315255808
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030131600064195154,
      "loss": 0.0655,
      "theoretical_loss": 3.4178084620539138,
      "tokens_seen": 2315517952
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003012357567003691,
      "loss": 0.0656,
      "theoretical_loss": 3.4177773848773225,
      "tokens_seen": 2315780096
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003011555127587867,
      "loss": 0.068,
      "theoretical_loss": 3.4177463122033243,
      "tokens_seen": 2316042240
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003010752688172043,
      "loss": 0.0672,
      "theoretical_loss": 3.4177152440307568,
      "tokens_seen": 2316304384
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030099502487562194,
      "loss": 0.0672,
      "theoretical_loss": 3.417684180358459,
      "tokens_seen": 2316566528
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003009147809340395,
      "loss": 0.0688,
      "theoretical_loss": 3.417653121185269,
      "tokens_seen": 2316828672
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030083453699245705,
      "loss": 0.0704,
      "theoretical_loss": 3.4176220665100274,
      "tokens_seen": 2317090816
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030075429305087466,
      "loss": 0.0681,
      "theoretical_loss": 3.4175910163315733,
      "tokens_seen": 2317352960
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003006740491092922,
      "loss": 0.0684,
      "theoretical_loss": 3.417559970648747,
      "tokens_seen": 2317615104
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030059380516770984,
      "loss": 0.0691,
      "theoretical_loss": 3.4175289294603894,
      "tokens_seen": 2317877248
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030051356122612745,
      "loss": 0.069,
      "theoretical_loss": 3.4174978927653417,
      "tokens_seen": 2318139392
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030043331728454506,
      "loss": 0.0684,
      "theoretical_loss": 3.4174668605624454,
      "tokens_seen": 2318401536
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003003530733429626,
      "loss": 0.0667,
      "theoretical_loss": 3.4174358328505425,
      "tokens_seen": 2318663680
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003002728294013802,
      "loss": 0.067,
      "theoretical_loss": 3.4174048096284757,
      "tokens_seen": 2318925824
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0003001925854597978,
      "loss": 0.0674,
      "theoretical_loss": 3.4173737908950876,
      "tokens_seen": 2319187968
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030011234151821535,
      "loss": 0.0672,
      "theoretical_loss": 3.417342776649221,
      "tokens_seen": 2319450112
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00030003209757663296,
      "loss": 0.0698,
      "theoretical_loss": 3.4173117668897204,
      "tokens_seen": 2319712256
    },
    {
      "epoch": 0.7,
      "objective/train/advantage_avg": 0.0005700485780835152,
      "objective/train/docs_used": 844817,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.1989997625350952,
      "objective/train/original_loss": 1.1989996433258057,
      "objective/train/theoretical_loss": 3.4172807616154293,
      "objective/train/tokens_used": 2340434400,
      "objective/train/value_avg": -0.00766754150390625,
      "objective/train/value_loss": 0.00026566028827801347,
      "objective/train/value_max": -1.1861324310302734e-05,
      "objective/train/value_min": -0.97998046875,
      "objective/train/value_reward_corr": 0.7219315572173486,
      "objective/train/value_std": 0.01800537109375,
      "objective/train/weight_avg": 1.0006872415542603,
      "objective/train/weighted_lm_loss": 1.1993837356567383,
      "objective/train/weights_max": 2.4944355487823486,
      "objective/train/weights_min": 0.14852790534496307,
      "theoretical_loss": 3.4172807616154293,
      "tokens_seen": 2319974400
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0002999518536350506,
      "loss": 0.0653,
      "theoretical_loss": 3.4172807616154293,
      "tokens_seen": 2319974400
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0002998716096934682,
      "loss": 0.0679,
      "theoretical_loss": 3.417249760825193,
      "tokens_seen": 2320236544
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00029979136575188574,
      "loss": 0.065,
      "theoretical_loss": 3.417218764517856,
      "tokens_seen": 2320498688
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00029971112181030336,
      "loss": 0.0688,
      "theoretical_loss": 3.4171877726922633,
      "tokens_seen": 2320760832
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0002996308778687209,
      "loss": 0.067,
      "theoretical_loss": 3.417156785347262,
      "tokens_seen": 2321022976
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0002995506339271385,
      "loss": 0.0672,
      "theoretical_loss": 3.4171258024816975,
      "tokens_seen": 2321285120
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0002994703899855561,
      "loss": 0.0676,
      "theoretical_loss": 3.4170948240944163,
      "tokens_seen": 2321547264
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0002993901460439737,
      "loss": 0.0667,
      "theoretical_loss": 3.4170638501842663,
      "tokens_seen": 2321809408
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00029930990210239126,
      "loss": 0.0677,
      "theoretical_loss": 3.417032880750094,
      "tokens_seen": 2322071552
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00029922965816080887,
      "loss": 0.0658,
      "theoretical_loss": 3.417001915790749,
      "tokens_seen": 2322333696
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0002991494142192265,
      "loss": 0.0648,
      "theoretical_loss": 3.416970955305078,
      "tokens_seen": 2322595840
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00029906917027764404,
      "loss": 0.0676,
      "theoretical_loss": 3.416939999291931,
      "tokens_seen": 2322857984
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0002989889263360616,
      "loss": 0.0681,
      "theoretical_loss": 3.416909047750157,
      "tokens_seen": 2323120128
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0002989086823944792,
      "loss": 0.0674,
      "theoretical_loss": 3.4168781006786055,
      "tokens_seen": 2323382272
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0002988284384528968,
      "loss": 0.0684,
      "theoretical_loss": 3.4168471580761266,
      "tokens_seen": 2323644416
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0002987481945113144,
      "loss": 0.0672,
      "theoretical_loss": 3.416816219941571,
      "tokens_seen": 2323906560
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.000298667950569732,
      "loss": 0.0661,
      "theoretical_loss": 3.4167852862737895,
      "tokens_seen": 2324168704
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0002985877066281496,
      "loss": 0.068,
      "theoretical_loss": 3.4167543570716337,
      "tokens_seen": 2324430848
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00029850746268656717,
      "loss": 0.066,
      "theoretical_loss": 3.416723432333956,
      "tokens_seen": 2324692992
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0002984272187449847,
      "loss": 0.0662,
      "theoretical_loss": 3.416692512059607,
      "tokens_seen": 2324955136
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00029834697480340234,
      "loss": 0.065,
      "theoretical_loss": 3.416661596247441,
      "tokens_seen": 2325217280
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00029826673086181995,
      "loss": 0.0694,
      "theoretical_loss": 3.41663068489631,
      "tokens_seen": 2325479424
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0002981864869202375,
      "loss": 0.0656,
      "theoretical_loss": 3.4165997780050685,
      "tokens_seen": 2325741568
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.0002981062429786551,
      "loss": 0.0675,
      "theoretical_loss": 3.41656887557257,
      "tokens_seen": 2326003712
    },
    {
      "epoch": 0.7,
      "learning_rate": 0.00029802599903707273,
      "loss": 0.0687,
      "theoretical_loss": 3.416537977597668,
      "tokens_seen": 2326265856
    },
    {
      "epoch": 0.7,
      "objective/train/advantage_avg": 0.0007894880254752934,
      "objective/train/docs_used": 847012,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3645339012145996,
      "objective/train/original_loss": 1.3645339012145996,
      "objective/train/theoretical_loss": 3.4165070840792184,
      "objective/train/tokens_used": 2346988000,
      "objective/train/value_avg": -0.00891876220703125,
      "objective/train/value_loss": 0.0002483201096765697,
      "objective/train/value_max": -3.647804260253906e-05,
      "objective/train/value_min": -0.890625,
      "objective/train/value_reward_corr": 0.7192212149918833,
      "objective/train/value_std": 0.0171661376953125,
      "objective/train/weight_avg": 1.0008996725082397,
      "objective/train/weighted_lm_loss": 1.3655369281768799,
      "objective/train/weights_max": 1.242430329322815,
      "objective/train/weights_min": 0.3732339143753052,
      "theoretical_loss": 3.4165070840792184,
      "tokens_seen": 2326528000
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029794575509549035,
      "loss": 0.0674,
      "theoretical_loss": 3.4165070840792184,
      "tokens_seen": 2326528000
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029786551115390785,
      "loss": 0.0663,
      "theoretical_loss": 3.4164761950160765,
      "tokens_seen": 2326790144
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029778526721232546,
      "loss": 0.0661,
      "theoretical_loss": 3.4164453104070973,
      "tokens_seen": 2327052288
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002977050232707431,
      "loss": 0.0645,
      "theoretical_loss": 3.416414430251137,
      "tokens_seen": 2327314432
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029762477932916063,
      "loss": 0.0678,
      "theoretical_loss": 3.4163835545470524,
      "tokens_seen": 2327576576
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029754453538757825,
      "loss": 0.0616,
      "theoretical_loss": 3.4163526832937,
      "tokens_seen": 2327838720
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029746429144599586,
      "loss": 0.0671,
      "theoretical_loss": 3.4163218164899374,
      "tokens_seen": 2328100864
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029738404750441347,
      "loss": 0.0663,
      "theoretical_loss": 3.416290954134622,
      "tokens_seen": 2328363008
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.000297303803562831,
      "loss": 0.068,
      "theoretical_loss": 3.4162600962266128,
      "tokens_seen": 2328625152
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002972235596212486,
      "loss": 0.0672,
      "theoretical_loss": 3.4162292427647674,
      "tokens_seen": 2328887296
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002971433156796662,
      "loss": 0.0665,
      "theoretical_loss": 3.4161983937479454,
      "tokens_seen": 2329149440
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029706307173808376,
      "loss": 0.0658,
      "theoretical_loss": 3.4161675491750056,
      "tokens_seen": 2329411584
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002969828277965014,
      "loss": 0.0683,
      "theoretical_loss": 3.416136709044809,
      "tokens_seen": 2329673728
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.000296902583854919,
      "loss": 0.0651,
      "theoretical_loss": 3.416105873356215,
      "tokens_seen": 2329935872
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029682233991333654,
      "loss": 0.0665,
      "theoretical_loss": 3.4160750421080843,
      "tokens_seen": 2330198016
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002967420959717541,
      "loss": 0.066,
      "theoretical_loss": 3.4160442152992783,
      "tokens_seen": 2330460160
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002966618520301717,
      "loss": 0.0674,
      "theoretical_loss": 3.4160133929286585,
      "tokens_seen": 2330722304
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029658160808858933,
      "loss": 0.0671,
      "theoretical_loss": 3.4159825749950867,
      "tokens_seen": 2330984448
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002965013641470069,
      "loss": 0.0684,
      "theoretical_loss": 3.4159517614974257,
      "tokens_seen": 2331246592
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002964211202054245,
      "loss": 0.0668,
      "theoretical_loss": 3.4159209524345373,
      "tokens_seen": 2331508736
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002963408762638421,
      "loss": 0.0669,
      "theoretical_loss": 3.4158901478052863,
      "tokens_seen": 2331770880
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029626063232225967,
      "loss": 0.0692,
      "theoretical_loss": 3.4158593476085346,
      "tokens_seen": 2332033024
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002961803883806773,
      "loss": 0.0664,
      "theoretical_loss": 3.4158285518431475,
      "tokens_seen": 2332295168
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029610014443909484,
      "loss": 0.0689,
      "theoretical_loss": 3.4157977605079894,
      "tokens_seen": 2332557312
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029601990049751245,
      "loss": 0.0634,
      "theoretical_loss": 3.415766973601924,
      "tokens_seen": 2332819456
    },
    {
      "epoch": 0.71,
      "objective/train/advantage_avg": 0.00032847325201146305,
      "objective/train/docs_used": 849464,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3709079027175903,
      "objective/train/original_loss": 1.3709079027175903,
      "objective/train/theoretical_loss": 3.415736191123818,
      "objective/train/tokens_used": 2353541600,
      "objective/train/value_avg": -0.00714111328125,
      "objective/train/value_loss": 0.00020859837241005152,
      "objective/train/value_max": -2.8848648071289062e-05,
      "objective/train/value_min": -0.58837890625,
      "objective/train/value_reward_corr": 0.6817947286926539,
      "objective/train/value_std": 0.0138092041015625,
      "objective/train/weight_avg": 1.000423550605774,
      "objective/train/weighted_lm_loss": 1.3710265159606934,
      "objective/train/weights_max": 1.3413829803466797,
      "objective/train/weights_min": 0.3711100220680237,
      "theoretical_loss": 3.415736191123818,
      "tokens_seen": 2333081600
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029593965655593,
      "loss": 0.0665,
      "theoretical_loss": 3.415736191123818,
      "tokens_seen": 2333081600
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002958594126143476,
      "loss": 0.0673,
      "theoretical_loss": 3.4157054130725366,
      "tokens_seen": 2333343744
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029577916867276524,
      "loss": 0.0687,
      "theoretical_loss": 3.415674639446946,
      "tokens_seen": 2333605888
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002956989247311828,
      "loss": 0.0674,
      "theoretical_loss": 3.415643870245913,
      "tokens_seen": 2333868032
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002956186807896004,
      "loss": 0.0683,
      "theoretical_loss": 3.4156131054683034,
      "tokens_seen": 2334130176
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029553843684801797,
      "loss": 0.068,
      "theoretical_loss": 3.415582345112986,
      "tokens_seen": 2334392320
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002954581929064356,
      "loss": 0.0716,
      "theoretical_loss": 3.415551589178828,
      "tokens_seen": 2334654464
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029537794896485314,
      "loss": 0.0678,
      "theoretical_loss": 3.415520837664698,
      "tokens_seen": 2334916608
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029529770502327075,
      "loss": 0.0659,
      "theoretical_loss": 3.415490090569464,
      "tokens_seen": 2335178752
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029521746108168836,
      "loss": 0.0686,
      "theoretical_loss": 3.415459347891996,
      "tokens_seen": 2335440896
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002951372171401059,
      "loss": 0.0658,
      "theoretical_loss": 3.4154286096311623,
      "tokens_seen": 2335703040
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029505697319852353,
      "loss": 0.0644,
      "theoretical_loss": 3.4153978757858336,
      "tokens_seen": 2335965184
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029497672925694115,
      "loss": 0.0661,
      "theoretical_loss": 3.41536714635488,
      "tokens_seen": 2336227328
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029489648531535865,
      "loss": 0.0683,
      "theoretical_loss": 3.4153364213371726,
      "tokens_seen": 2336489472
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029481624137377626,
      "loss": 0.0655,
      "theoretical_loss": 3.415305700731582,
      "tokens_seen": 2336751616
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002947359974321939,
      "loss": 0.0683,
      "theoretical_loss": 3.41527498453698,
      "tokens_seen": 2337013760
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002946557534906115,
      "loss": 0.0684,
      "theoretical_loss": 3.4152442727522385,
      "tokens_seen": 2337275904
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029457550954902905,
      "loss": 0.0696,
      "theoretical_loss": 3.41521356537623,
      "tokens_seen": 2337538048
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029449526560744666,
      "loss": 0.0668,
      "theoretical_loss": 3.415182862407827,
      "tokens_seen": 2337800192
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029441502166586427,
      "loss": 0.0663,
      "theoretical_loss": 3.4151521638459035,
      "tokens_seen": 2338062336
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002943347777242818,
      "loss": 0.0698,
      "theoretical_loss": 3.4151214696893324,
      "tokens_seen": 2338324480
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002942545337826994,
      "loss": 0.0672,
      "theoretical_loss": 3.4150907799369876,
      "tokens_seen": 2338586624
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.000294174289841117,
      "loss": 0.0677,
      "theoretical_loss": 3.4150600945877443,
      "tokens_seen": 2338848768
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002940940458995346,
      "loss": 0.0691,
      "theoretical_loss": 3.415029413640477,
      "tokens_seen": 2339110912
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029401380195795217,
      "loss": 0.0672,
      "theoretical_loss": 3.414998737094061,
      "tokens_seen": 2339373056
    },
    {
      "epoch": 0.71,
      "objective/train/advantage_avg": 0.00047332874964922667,
      "objective/train/docs_used": 851815,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3133671283721924,
      "objective/train/original_loss": 1.3133671283721924,
      "objective/train/theoretical_loss": 3.414968064947372,
      "objective/train/tokens_used": 2360095200,
      "objective/train/value_avg": -0.0105438232421875,
      "objective/train/value_loss": 0.0004383322084322572,
      "objective/train/value_max": -3.218650817871094e-05,
      "objective/train/value_min": -0.97802734375,
      "objective/train/value_reward_corr": 0.7493675175185559,
      "objective/train/value_std": 0.0269317626953125,
      "objective/train/weight_avg": 1.0006917715072632,
      "objective/train/weighted_lm_loss": 1.313127875328064,
      "objective/train/weights_max": 2.2706074714660645,
      "objective/train/weights_min": 0.37235790491104126,
      "theoretical_loss": 3.414968064947372,
      "tokens_seen": 2339635200
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002939335580163698,
      "loss": 0.0672,
      "theoretical_loss": 3.414968064947372,
      "tokens_seen": 2339635200
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002938533140747874,
      "loss": 0.0661,
      "theoretical_loss": 3.4149373971992856,
      "tokens_seen": 2339897344
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002937730701332049,
      "loss": 0.0678,
      "theoretical_loss": 3.4149067338486794,
      "tokens_seen": 2340159488
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002936928261916225,
      "loss": 0.0665,
      "theoretical_loss": 3.4148760748944293,
      "tokens_seen": 2340421632
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002936125822500401,
      "loss": 0.0661,
      "theoretical_loss": 3.4148454203354133,
      "tokens_seen": 2340683776
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029353233830845774,
      "loss": 0.0673,
      "theoretical_loss": 3.4148147701705094,
      "tokens_seen": 2340945920
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002934520943668753,
      "loss": 0.0669,
      "theoretical_loss": 3.414784124398595,
      "tokens_seen": 2341208064
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002933718504252929,
      "loss": 0.0688,
      "theoretical_loss": 3.4147534830185493,
      "tokens_seen": 2341470208
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002932916064837105,
      "loss": 0.0676,
      "theoretical_loss": 3.414722846029251,
      "tokens_seen": 2341732352
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002932113625421281,
      "loss": 0.0695,
      "theoretical_loss": 3.4146922134295794,
      "tokens_seen": 2341994496
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029313111860054564,
      "loss": 0.0686,
      "theoretical_loss": 3.414661585218415,
      "tokens_seen": 2342256640
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029305087465896325,
      "loss": 0.0682,
      "theoretical_loss": 3.4146309613946366,
      "tokens_seen": 2342518784
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002929706307173808,
      "loss": 0.0677,
      "theoretical_loss": 3.4146003419571267,
      "tokens_seen": 2342780928
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002928903867757984,
      "loss": 0.0682,
      "theoretical_loss": 3.414569726904765,
      "tokens_seen": 2343043072
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029281014283421604,
      "loss": 0.0672,
      "theoretical_loss": 3.414539116236434,
      "tokens_seen": 2343305216
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029272989889263365,
      "loss": 0.0662,
      "theoretical_loss": 3.4145085099510144,
      "tokens_seen": 2343567360
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002926496549510512,
      "loss": 0.0674,
      "theoretical_loss": 3.4144779080473895,
      "tokens_seen": 2343829504
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029256941100946877,
      "loss": 0.0649,
      "theoretical_loss": 3.4144473105244413,
      "tokens_seen": 2344091648
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002924891670678864,
      "loss": 0.0668,
      "theoretical_loss": 3.4144167173810533,
      "tokens_seen": 2344353792
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029240892312630394,
      "loss": 0.0679,
      "theoretical_loss": 3.414386128616109,
      "tokens_seen": 2344615936
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029232867918472155,
      "loss": 0.0685,
      "theoretical_loss": 3.414355544228492,
      "tokens_seen": 2344878080
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029224843524313916,
      "loss": 0.0697,
      "theoretical_loss": 3.414324964217087,
      "tokens_seen": 2345140224
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002921681913015568,
      "loss": 0.0657,
      "theoretical_loss": 3.414294388580779,
      "tokens_seen": 2345402368
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029208794735997433,
      "loss": 0.0662,
      "theoretical_loss": 3.4142638173184525,
      "tokens_seen": 2345664512
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002920077034183919,
      "loss": 0.0662,
      "theoretical_loss": 3.4142332504289934,
      "tokens_seen": 2345926656
    },
    {
      "epoch": 0.71,
      "objective/train/advantage_avg": 0.0004487581318244338,
      "objective/train/docs_used": 854185,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4248061180114746,
      "objective/train/original_loss": 1.4248061180114746,
      "objective/train/theoretical_loss": 3.414202687911288,
      "objective/train/tokens_used": 2366648800,
      "objective/train/value_avg": -0.00934600830078125,
      "objective/train/value_loss": 0.00034865766065195203,
      "objective/train/value_max": -4.83393669128418e-05,
      "objective/train/value_min": -0.97119140625,
      "objective/train/value_reward_corr": 0.7621885055068649,
      "objective/train/value_std": 0.0230865478515625,
      "objective/train/weight_avg": 1.0006086826324463,
      "objective/train/weighted_lm_loss": 1.4253432750701904,
      "objective/train/weights_max": 1.7495906352996826,
      "objective/train/weights_min": 0.3696533143520355,
      "theoretical_loss": 3.414202687911288,
      "tokens_seen": 2346188800
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002919274594768095,
      "loss": 0.0682,
      "theoretical_loss": 3.414202687911288,
      "tokens_seen": 2346188800
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029184721553522706,
      "loss": 0.0673,
      "theoretical_loss": 3.414172129764222,
      "tokens_seen": 2346450944
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002917669715936447,
      "loss": 0.0682,
      "theoretical_loss": 3.414141575986682,
      "tokens_seen": 2346713088
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002916867276520623,
      "loss": 0.0669,
      "theoretical_loss": 3.4141110265775563,
      "tokens_seen": 2346975232
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002916064837104799,
      "loss": 0.0665,
      "theoretical_loss": 3.414080481535732,
      "tokens_seen": 2347237376
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029152623976889746,
      "loss": 0.0644,
      "theoretical_loss": 3.414049940860097,
      "tokens_seen": 2347499520
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029144599582731507,
      "loss": 0.0666,
      "theoretical_loss": 3.4140194045495393,
      "tokens_seen": 2347761664
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029136575188573263,
      "loss": 0.066,
      "theoretical_loss": 3.413988872602949,
      "tokens_seen": 2348023808
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002912855079441502,
      "loss": 0.066,
      "theoretical_loss": 3.4139583450192137,
      "tokens_seen": 2348285952
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002912052640025678,
      "loss": 0.0657,
      "theoretical_loss": 3.413927821797224,
      "tokens_seen": 2348548096
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002911250200609854,
      "loss": 0.0681,
      "theoretical_loss": 3.41389730293587,
      "tokens_seen": 2348810240
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029104477611940297,
      "loss": 0.0651,
      "theoretical_loss": 3.413866788434042,
      "tokens_seen": 2349072384
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002909645321778206,
      "loss": 0.068,
      "theoretical_loss": 3.413836278290631,
      "tokens_seen": 2349334528
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002908842882362382,
      "loss": 0.0673,
      "theoretical_loss": 3.4138057725045274,
      "tokens_seen": 2349596672
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029080404429465576,
      "loss": 0.0698,
      "theoretical_loss": 3.413775271074624,
      "tokens_seen": 2349858816
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002907238003530733,
      "loss": 0.0672,
      "theoretical_loss": 3.413744773999812,
      "tokens_seen": 2350120960
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002906435564114909,
      "loss": 0.0654,
      "theoretical_loss": 3.413714281278985,
      "tokens_seen": 2350383104
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029056331246990854,
      "loss": 0.0674,
      "theoretical_loss": 3.413683792911035,
      "tokens_seen": 2350645248
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002904830685283261,
      "loss": 0.067,
      "theoretical_loss": 3.413653308894855,
      "tokens_seen": 2350907392
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002904028245867437,
      "loss": 0.0658,
      "theoretical_loss": 3.4136228292293396,
      "tokens_seen": 2351169536
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002903225806451613,
      "loss": 0.0671,
      "theoretical_loss": 3.4135923539133826,
      "tokens_seen": 2351431680
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029024233670357894,
      "loss": 0.067,
      "theoretical_loss": 3.413561882945878,
      "tokens_seen": 2351693824
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029016209276199644,
      "loss": 0.0659,
      "theoretical_loss": 3.4135314163257213,
      "tokens_seen": 2351955968
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029008184882041405,
      "loss": 0.0674,
      "theoretical_loss": 3.4135009540518073,
      "tokens_seen": 2352218112
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00029000160487883166,
      "loss": 0.0657,
      "theoretical_loss": 3.4134704961230327,
      "tokens_seen": 2352480256
    },
    {
      "epoch": 0.71,
      "objective/train/advantage_avg": 0.0002187528443755582,
      "objective/train/docs_used": 856640,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3601878881454468,
      "objective/train/original_loss": 1.3601877689361572,
      "objective/train/theoretical_loss": 3.4134400425382925,
      "objective/train/tokens_used": 2373202400,
      "objective/train/value_avg": -0.0091094970703125,
      "objective/train/value_loss": 0.00019813631661236286,
      "objective/train/value_max": -4.398822784423828e-05,
      "objective/train/value_min": -0.423095703125,
      "objective/train/value_reward_corr": 0.7711988085987357,
      "objective/train/value_std": 0.0168914794921875,
      "objective/train/weight_avg": 1.0003117322921753,
      "objective/train/weighted_lm_loss": 1.3602980375289917,
      "objective/train/weights_max": 1.232275366783142,
      "objective/train/weights_min": 0.3712516129016876,
      "theoretical_loss": 3.4134400425382925,
      "tokens_seen": 2352742400
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002899213609372492,
      "loss": 0.0671,
      "theoretical_loss": 3.4134400425382925,
      "tokens_seen": 2352742400
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00028984111699566684,
      "loss": 0.0668,
      "theoretical_loss": 3.4134095932964836,
      "tokens_seen": 2353004544
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00028976087305408445,
      "loss": 0.068,
      "theoretical_loss": 3.413379148396503,
      "tokens_seen": 2353266688
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00028968062911250206,
      "loss": 0.0653,
      "theoretical_loss": 3.413348707837248,
      "tokens_seen": 2353528832
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00028960038517091957,
      "loss": 0.0667,
      "theoretical_loss": 3.4133182716176167,
      "tokens_seen": 2353790976
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002895201412293372,
      "loss": 0.0651,
      "theoretical_loss": 3.4132878397365065,
      "tokens_seen": 2354053120
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002894398972877548,
      "loss": 0.0665,
      "theoretical_loss": 3.4132574121928165,
      "tokens_seen": 2354315264
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00028935965334617235,
      "loss": 0.0661,
      "theoretical_loss": 3.413226988985446,
      "tokens_seen": 2354577408
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00028927940940458996,
      "loss": 0.0651,
      "theoretical_loss": 3.4131965701132927,
      "tokens_seen": 2354839552
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002891991654630076,
      "loss": 0.0641,
      "theoretical_loss": 3.4131661555752584,
      "tokens_seen": 2355101696
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002891189215214252,
      "loss": 0.0663,
      "theoretical_loss": 3.413135745370242,
      "tokens_seen": 2355363840
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002890386775798427,
      "loss": 0.0676,
      "theoretical_loss": 3.4131053394971445,
      "tokens_seen": 2355625984
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002889584336382603,
      "loss": 0.0669,
      "theoretical_loss": 3.4130749379548666,
      "tokens_seen": 2355888128
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002888781896966779,
      "loss": 0.0671,
      "theoretical_loss": 3.41304454074231,
      "tokens_seen": 2356150272
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002887979457550955,
      "loss": 0.0664,
      "theoretical_loss": 3.4130141478583758,
      "tokens_seen": 2356412416
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002887177018135131,
      "loss": 0.0662,
      "theoretical_loss": 3.4129837593019667,
      "tokens_seen": 2356674560
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002886374578719307,
      "loss": 0.0672,
      "theoretical_loss": 3.412953375071985,
      "tokens_seen": 2356936704
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00028855721393034826,
      "loss": 0.0663,
      "theoretical_loss": 3.412922995167334,
      "tokens_seen": 2357198848
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00028847696998876587,
      "loss": 0.0663,
      "theoretical_loss": 3.4128926195869163,
      "tokens_seen": 2357460992
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00028839672604718343,
      "loss": 0.0635,
      "theoretical_loss": 3.4128622483296365,
      "tokens_seen": 2357723136
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.00028831648210560104,
      "loss": 0.0674,
      "theoretical_loss": 3.412831881394399,
      "tokens_seen": 2357985280
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002882362381640186,
      "loss": 0.0669,
      "theoretical_loss": 3.4128015187801073,
      "tokens_seen": 2358247424
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002881559942224362,
      "loss": 0.0649,
      "theoretical_loss": 3.4127711604856668,
      "tokens_seen": 2358509568
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002880757502808538,
      "loss": 0.0665,
      "theoretical_loss": 3.4127408065099827,
      "tokens_seen": 2358771712
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.0002879955063392714,
      "loss": 0.0666,
      "theoretical_loss": 3.4127104568519613,
      "tokens_seen": 2359033856
    },
    {
      "epoch": 0.71,
      "objective/train/advantage_avg": 0.0003526402579154819,
      "objective/train/docs_used": 858941,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2617586851119995,
      "objective/train/original_loss": 1.261758804321289,
      "objective/train/theoretical_loss": 3.412680111510508,
      "objective/train/tokens_used": 2379756000,
      "objective/train/value_avg": -0.00748443603515625,
      "objective/train/value_loss": 0.00020228726498316973,
      "objective/train/value_max": -3.272294998168945e-05,
      "objective/train/value_min": -0.41162109375,
      "objective/train/value_reward_corr": 0.7067248936371501,
      "objective/train/value_std": 0.0146942138671875,
      "objective/train/weight_avg": 1.0004441738128662,
      "objective/train/weighted_lm_loss": 1.2619435787200928,
      "objective/train/weights_max": 1.3642196655273438,
      "objective/train/weights_min": 0.36796626448631287,
      "theoretical_loss": 3.412680111510508,
      "tokens_seen": 2359296000
    },
    {
      "epoch": 0.71,
      "learning_rate": 0.000287915262397689,
      "loss": 0.0646,
      "theoretical_loss": 3.412680111510508,
      "tokens_seen": 2359296000
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028783501845610655,
      "loss": 0.0677,
      "theoretical_loss": 3.41264977048453,
      "tokens_seen": 2359558144
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028775477451452417,
      "loss": 0.0694,
      "theoretical_loss": 3.412619433772934,
      "tokens_seen": 2359820288
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002876745305729417,
      "loss": 0.0676,
      "theoretical_loss": 3.412589101374627,
      "tokens_seen": 2360082432
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028759428663135934,
      "loss": 0.0665,
      "theoretical_loss": 3.4125587732885174,
      "tokens_seen": 2360344576
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028751404268977695,
      "loss": 0.0663,
      "theoretical_loss": 3.4125284495135126,
      "tokens_seen": 2360606720
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002874337987481945,
      "loss": 0.0666,
      "theoretical_loss": 3.412498130048522,
      "tokens_seen": 2360868864
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002873535548066121,
      "loss": 0.0681,
      "theoretical_loss": 3.412467814892454,
      "tokens_seen": 2361131008
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002872733108650297,
      "loss": 0.0661,
      "theoretical_loss": 3.4124375040442176,
      "tokens_seen": 2361393152
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002871930669234473,
      "loss": 0.0671,
      "theoretical_loss": 3.4124071975027235,
      "tokens_seen": 2361655296
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028711282298186485,
      "loss": 0.0668,
      "theoretical_loss": 3.4123768952668807,
      "tokens_seen": 2361917440
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028703257904028246,
      "loss": 0.0677,
      "theoretical_loss": 3.412346597335601,
      "tokens_seen": 2362179584
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002869523350987001,
      "loss": 0.0684,
      "theoretical_loss": 3.4123163037077937,
      "tokens_seen": 2362441728
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028687209115711763,
      "loss": 0.0659,
      "theoretical_loss": 3.4122860143823717,
      "tokens_seen": 2362703872
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028679184721553525,
      "loss": 0.0649,
      "theoretical_loss": 3.4122557293582454,
      "tokens_seen": 2362966016
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028671160327395286,
      "loss": 0.0663,
      "theoretical_loss": 3.412225448634328,
      "tokens_seen": 2363228160
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028663135933237036,
      "loss": 0.0659,
      "theoretical_loss": 3.4121951722095316,
      "tokens_seen": 2363490304
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.000286551115390788,
      "loss": 0.0687,
      "theoretical_loss": 3.412164900082769,
      "tokens_seen": 2363752448
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002864708714492056,
      "loss": 0.065,
      "theoretical_loss": 3.412134632252953,
      "tokens_seen": 2364014592
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002863906275076232,
      "loss": 0.0642,
      "theoretical_loss": 3.4121043687189982,
      "tokens_seen": 2364276736
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028631038356604076,
      "loss": 0.0693,
      "theoretical_loss": 3.4120741094798186,
      "tokens_seen": 2364538880
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002862301396244584,
      "loss": 0.0656,
      "theoretical_loss": 3.412043854534328,
      "tokens_seen": 2364801024
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.000286149895682876,
      "loss": 0.0657,
      "theoretical_loss": 3.4120136038814417,
      "tokens_seen": 2365063168
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002860696517412935,
      "loss": 0.0674,
      "theoretical_loss": 3.411983357520075,
      "tokens_seen": 2365325312
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002859894077997111,
      "loss": 0.0659,
      "theoretical_loss": 3.4119531154491436,
      "tokens_seen": 2365587456
    },
    {
      "epoch": 0.72,
      "objective/train/advantage_avg": 0.0008152511436492205,
      "objective/train/docs_used": 861421,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.1844955682754517,
      "objective/train/original_loss": 1.1844955682754517,
      "objective/train/theoretical_loss": 3.411922877667563,
      "objective/train/tokens_used": 2386309600,
      "objective/train/value_avg": -0.00860595703125,
      "objective/train/value_loss": 0.00024558245786465704,
      "objective/train/value_max": -2.4318695068359375e-05,
      "objective/train/value_min": -0.7431640625,
      "objective/train/value_reward_corr": 0.7812741348390095,
      "objective/train/value_std": 0.0189056396484375,
      "objective/train/weight_avg": 1.0009241104125977,
      "objective/train/weighted_lm_loss": 1.1853077411651611,
      "objective/train/weights_max": 1.2296350002288818,
      "objective/train/weights_min": 0.3681640923023224,
      "theoretical_loss": 3.411922877667563,
      "tokens_seen": 2365849600
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002859091638581287,
      "loss": 0.0651,
      "theoretical_loss": 3.411922877667563,
      "tokens_seen": 2365849600
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028582891991654633,
      "loss": 0.0663,
      "theoretical_loss": 3.4118926441742508,
      "tokens_seen": 2366111744
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002857486759749639,
      "loss": 0.0655,
      "theoretical_loss": 3.4118624149681227,
      "tokens_seen": 2366373888
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002856684320333815,
      "loss": 0.0651,
      "theoretical_loss": 3.4118321900480963,
      "tokens_seen": 2366636032
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002855881880917991,
      "loss": 0.0659,
      "theoretical_loss": 3.4118019694130894,
      "tokens_seen": 2366898176
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028550794415021667,
      "loss": 0.0692,
      "theoretical_loss": 3.41177175306202,
      "tokens_seen": 2367160320
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028542770020863423,
      "loss": 0.0669,
      "theoretical_loss": 3.4117415409938068,
      "tokens_seen": 2367422464
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028534745626705184,
      "loss": 0.0667,
      "theoretical_loss": 3.411711333207368,
      "tokens_seen": 2367684608
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028526721232546945,
      "loss": 0.0659,
      "theoretical_loss": 3.411681129701623,
      "tokens_seen": 2367946752
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.000285186968383887,
      "loss": 0.0649,
      "theoretical_loss": 3.411650930475491,
      "tokens_seen": 2368208896
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002851067244423046,
      "loss": 0.066,
      "theoretical_loss": 3.4116207355278934,
      "tokens_seen": 2368471040
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028502648050072224,
      "loss": 0.0669,
      "theoretical_loss": 3.411590544857749,
      "tokens_seen": 2368733184
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002849462365591398,
      "loss": 0.0674,
      "theoretical_loss": 3.4115603584639795,
      "tokens_seen": 2368995328
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028486599261755735,
      "loss": 0.0651,
      "theoretical_loss": 3.4115301763455057,
      "tokens_seen": 2369257472
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028478574867597497,
      "loss": 0.0666,
      "theoretical_loss": 3.411499998501249,
      "tokens_seen": 2369519616
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002847055047343925,
      "loss": 0.0668,
      "theoretical_loss": 3.4114698249301316,
      "tokens_seen": 2369781760
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028462526079281014,
      "loss": 0.0657,
      "theoretical_loss": 3.411439655631076,
      "tokens_seen": 2370043904
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028454501685122775,
      "loss": 0.0665,
      "theoretical_loss": 3.4114094906030052,
      "tokens_seen": 2370306048
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028446477290964536,
      "loss": 0.0672,
      "theoretical_loss": 3.4113793298448414,
      "tokens_seen": 2370568192
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002843845289680629,
      "loss": 0.0681,
      "theoretical_loss": 3.411349173355508,
      "tokens_seen": 2370830336
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002843042850264805,
      "loss": 0.0691,
      "theoretical_loss": 3.4113190211339304,
      "tokens_seen": 2371092480
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002842240410848981,
      "loss": 0.0677,
      "theoretical_loss": 3.4112888731790316,
      "tokens_seen": 2371354624
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028414379714331565,
      "loss": 0.0662,
      "theoretical_loss": 3.4112587294897367,
      "tokens_seen": 2371616768
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028406355320173326,
      "loss": 0.0666,
      "theoretical_loss": 3.4112285900649706,
      "tokens_seen": 2371878912
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002839833092601509,
      "loss": 0.0649,
      "theoretical_loss": 3.411198454903659,
      "tokens_seen": 2372141056
    },
    {
      "epoch": 0.72,
      "objective/train/advantage_avg": 0.00061773881316185,
      "objective/train/docs_used": 863878,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.301668643951416,
      "objective/train/original_loss": 1.3016687631607056,
      "objective/train/theoretical_loss": 3.4111683240047275,
      "objective/train/tokens_used": 2392863200,
      "objective/train/value_avg": -0.007083892822265625,
      "objective/train/value_loss": 0.00021664539235644042,
      "objective/train/value_max": -5.346536636352539e-05,
      "objective/train/value_min": -0.90478515625,
      "objective/train/value_reward_corr": 0.6975712737975295,
      "objective/train/value_std": 0.0143280029296875,
      "objective/train/weight_avg": 1.0007141828536987,
      "objective/train/weighted_lm_loss": 1.3028697967529297,
      "objective/train/weights_max": 1.1462258100509644,
      "objective/train/weights_min": 0.37651151418685913,
      "theoretical_loss": 3.4111683240047275,
      "tokens_seen": 2372403200
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002839030653185685,
      "loss": 0.0669,
      "theoretical_loss": 3.4111683240047275,
      "tokens_seen": 2372403200
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028382282137698605,
      "loss": 0.0625,
      "theoretical_loss": 3.4111381973671024,
      "tokens_seen": 2372665344
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028374257743540366,
      "loss": 0.0653,
      "theoretical_loss": 3.4111080749897105,
      "tokens_seen": 2372927488
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002836623334938212,
      "loss": 0.067,
      "theoretical_loss": 3.4110779568714786,
      "tokens_seen": 2373189632
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002835820895522388,
      "loss": 0.0673,
      "theoretical_loss": 3.4110478430113345,
      "tokens_seen": 2373451776
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002835018456106564,
      "loss": 0.0642,
      "theoretical_loss": 3.4110177334082055,
      "tokens_seen": 2373713920
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.000283421601669074,
      "loss": 0.0683,
      "theoretical_loss": 3.4109876280610196,
      "tokens_seen": 2373976064
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002833413577274916,
      "loss": 0.0624,
      "theoretical_loss": 3.4109575269687062,
      "tokens_seen": 2374238208
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028326111378590917,
      "loss": 0.0681,
      "theoretical_loss": 3.4109274301301937,
      "tokens_seen": 2374500352
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002831808698443268,
      "loss": 0.0647,
      "theoretical_loss": 3.410897337544412,
      "tokens_seen": 2374762496
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028310062590274434,
      "loss": 0.0645,
      "theoretical_loss": 3.41086724921029,
      "tokens_seen": 2375024640
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002830203819611619,
      "loss": 0.0669,
      "theoretical_loss": 3.410837165126758,
      "tokens_seen": 2375286784
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002829401380195795,
      "loss": 0.0677,
      "theoretical_loss": 3.4108070852927472,
      "tokens_seen": 2375548928
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028285989407799713,
      "loss": 0.0656,
      "theoretical_loss": 3.4107770097071874,
      "tokens_seen": 2375811072
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002827796501364147,
      "loss": 0.065,
      "theoretical_loss": 3.410746938369011,
      "tokens_seen": 2376073216
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002826994061948323,
      "loss": 0.068,
      "theoretical_loss": 3.410716871277149,
      "tokens_seen": 2376335360
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002826191622532499,
      "loss": 0.065,
      "theoretical_loss": 3.4106868084305337,
      "tokens_seen": 2376597504
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028253891831166747,
      "loss": 0.0676,
      "theoretical_loss": 3.4106567498280973,
      "tokens_seen": 2376859648
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028245867437008503,
      "loss": 0.0673,
      "theoretical_loss": 3.410626695468773,
      "tokens_seen": 2377121792
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028237843042850264,
      "loss": 0.0653,
      "theoretical_loss": 3.410596645351494,
      "tokens_seen": 2377383936
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028229818648692025,
      "loss": 0.065,
      "theoretical_loss": 3.4105665994751933,
      "tokens_seen": 2377646080
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002822179425453378,
      "loss": 0.0644,
      "theoretical_loss": 3.4105365578388054,
      "tokens_seen": 2377908224
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002821376986037554,
      "loss": 0.0674,
      "theoretical_loss": 3.410506520441265,
      "tokens_seen": 2378170368
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028205745466217304,
      "loss": 0.0661,
      "theoretical_loss": 3.410476487281506,
      "tokens_seen": 2378432512
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028197721072059065,
      "loss": 0.0651,
      "theoretical_loss": 3.410446458358464,
      "tokens_seen": 2378694656
    },
    {
      "epoch": 0.72,
      "objective/train/advantage_avg": 0.00045984104508534074,
      "objective/train/docs_used": 866449,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2284327745437622,
      "objective/train/original_loss": 1.2284328937530518,
      "objective/train/theoretical_loss": 3.4104164336710747,
      "objective/train/tokens_used": 2399416800,
      "objective/train/value_avg": -0.00720977783203125,
      "objective/train/value_loss": 0.00012368944589979947,
      "objective/train/value_max": -1.4483928680419922e-05,
      "objective/train/value_min": -0.37353515625,
      "objective/train/value_reward_corr": 0.754171488899678,
      "objective/train/value_std": 0.01329803466796875,
      "objective/train/weight_avg": 1.0005178451538086,
      "objective/train/weighted_lm_loss": 1.2287147045135498,
      "objective/train/weights_max": 1.3266090154647827,
      "objective/train/weights_min": 0.39287444949150085,
      "theoretical_loss": 3.4104164336710747,
      "tokens_seen": 2378956800
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028189696677900815,
      "loss": 0.0651,
      "theoretical_loss": 3.4104164336710747,
      "tokens_seen": 2378956800
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028181672283742577,
      "loss": 0.0649,
      "theoretical_loss": 3.4103864132182733,
      "tokens_seen": 2379218944
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002817364788958434,
      "loss": 0.0668,
      "theoretical_loss": 3.410356396998997,
      "tokens_seen": 2379481088
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028165623495426094,
      "loss": 0.0703,
      "theoretical_loss": 3.410326385012182,
      "tokens_seen": 2379743232
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028157599101267855,
      "loss": 0.0666,
      "theoretical_loss": 3.4102963772567656,
      "tokens_seen": 2380005376
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028149574707109616,
      "loss": 0.0665,
      "theoretical_loss": 3.4102663737316847,
      "tokens_seen": 2380267520
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002814155031295138,
      "loss": 0.0649,
      "theoretical_loss": 3.410236374435878,
      "tokens_seen": 2380529664
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002813352591879313,
      "loss": 0.0653,
      "theoretical_loss": 3.410206379368283,
      "tokens_seen": 2380791808
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002812550152463489,
      "loss": 0.066,
      "theoretical_loss": 3.4101763885278387,
      "tokens_seen": 2381053952
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002811747713047665,
      "loss": 0.0675,
      "theoretical_loss": 3.410146401913484,
      "tokens_seen": 2381316096
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028109452736318406,
      "loss": 0.0654,
      "theoretical_loss": 3.410116419524158,
      "tokens_seen": 2381578240
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002810142834216017,
      "loss": 0.0658,
      "theoretical_loss": 3.4100864413588003,
      "tokens_seen": 2381840384
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002809340394800193,
      "loss": 0.0652,
      "theoretical_loss": 3.410056467416352,
      "tokens_seen": 2382102528
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002808537955384369,
      "loss": 0.067,
      "theoretical_loss": 3.4100264976957524,
      "tokens_seen": 2382364672
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028077355159685446,
      "loss": 0.0665,
      "theoretical_loss": 3.4099965321959433,
      "tokens_seen": 2382626816
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.000280693307655272,
      "loss": 0.0656,
      "theoretical_loss": 3.4099665709158655,
      "tokens_seen": 2382888960
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028061306371368963,
      "loss": 0.0659,
      "theoretical_loss": 3.409936613854461,
      "tokens_seen": 2383151104
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002805328197721072,
      "loss": 0.0676,
      "theoretical_loss": 3.4099066610106714,
      "tokens_seen": 2383413248
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002804525758305248,
      "loss": 0.0658,
      "theoretical_loss": 3.40987671238344,
      "tokens_seen": 2383675392
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002803723318889424,
      "loss": 0.0661,
      "theoretical_loss": 3.4098467679717084,
      "tokens_seen": 2383937536
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028029208794735997,
      "loss": 0.0666,
      "theoretical_loss": 3.4098168277744203,
      "tokens_seen": 2384199680
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002802118440057776,
      "loss": 0.0667,
      "theoretical_loss": 3.4097868917905196,
      "tokens_seen": 2384461824
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028013160006419514,
      "loss": 0.0654,
      "theoretical_loss": 3.40975696001895,
      "tokens_seen": 2384723968
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00028005135612261276,
      "loss": 0.0684,
      "theoretical_loss": 3.4097270324586555,
      "tokens_seen": 2384986112
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002799711121810303,
      "loss": 0.0654,
      "theoretical_loss": 3.4096971091085817,
      "tokens_seen": 2385248256
    },
    {
      "epoch": 0.72,
      "objective/train/advantage_avg": 0.0010171148460358381,
      "objective/train/docs_used": 868828,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3410003185272217,
      "objective/train/original_loss": 1.3410003185272217,
      "objective/train/theoretical_loss": 3.4096671899676725,
      "objective/train/tokens_used": 2405970400,
      "objective/train/value_avg": -0.00829315185546875,
      "objective/train/value_loss": 0.000190845355973579,
      "objective/train/value_max": -5.3048133850097656e-05,
      "objective/train/value_min": -0.82861328125,
      "objective/train/value_reward_corr": 0.6975594542862572,
      "objective/train/value_std": 0.01468658447265625,
      "objective/train/weight_avg": 1.0011066198349,
      "objective/train/weighted_lm_loss": 1.3416869640350342,
      "objective/train/weights_max": 1.9094114303588867,
      "objective/train/weights_min": 0.372306764125824,
      "theoretical_loss": 3.4096671899676725,
      "tokens_seen": 2385510400
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002798908682394479,
      "loss": 0.0687,
      "theoretical_loss": 3.4096671899676725,
      "tokens_seen": 2385510400
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00027981062429786554,
      "loss": 0.0661,
      "theoretical_loss": 3.4096372750348745,
      "tokens_seen": 2385772544
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002797303803562831,
      "loss": 0.0668,
      "theoretical_loss": 3.409607364309133,
      "tokens_seen": 2386034688
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002796501364147007,
      "loss": 0.0699,
      "theoretical_loss": 3.4095774577893936,
      "tokens_seen": 2386296832
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00027956989247311827,
      "loss": 0.0655,
      "theoretical_loss": 3.4095475554746044,
      "tokens_seen": 2386558976
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002794896485315359,
      "loss": 0.0677,
      "theoretical_loss": 3.409517657363711,
      "tokens_seen": 2386821120
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00027940940458995344,
      "loss": 0.0676,
      "theoretical_loss": 3.4094877634556617,
      "tokens_seen": 2387083264
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00027932916064837105,
      "loss": 0.0679,
      "theoretical_loss": 3.4094578737494037,
      "tokens_seen": 2387345408
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00027924891670678866,
      "loss": 0.0681,
      "theoretical_loss": 3.4094279882438854,
      "tokens_seen": 2387607552
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002791686727652062,
      "loss": 0.0637,
      "theoretical_loss": 3.409398106938055,
      "tokens_seen": 2387869696
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00027908842882362384,
      "loss": 0.0679,
      "theoretical_loss": 3.4093682298308616,
      "tokens_seen": 2388131840
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00027900818488204145,
      "loss": 0.0664,
      "theoretical_loss": 3.4093383569212548,
      "tokens_seen": 2388393984
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00027892794094045895,
      "loss": 0.0666,
      "theoretical_loss": 3.4093084882081834,
      "tokens_seen": 2388656128
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00027884769699887657,
      "loss": 0.0718,
      "theoretical_loss": 3.4092786236905983,
      "tokens_seen": 2388918272
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002787674530572942,
      "loss": 0.0649,
      "theoretical_loss": 3.4092487633674495,
      "tokens_seen": 2389180416
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002786872091157118,
      "loss": 0.0668,
      "theoretical_loss": 3.4092189072376877,
      "tokens_seen": 2389442560
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00027860696517412935,
      "loss": 0.0681,
      "theoretical_loss": 3.4091890553002644,
      "tokens_seen": 2389704704
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.00027852672123254696,
      "loss": 0.0677,
      "theoretical_loss": 3.4091592075541306,
      "tokens_seen": 2389966848
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002784464772909646,
      "loss": 0.0655,
      "theoretical_loss": 3.4091293639982387,
      "tokens_seen": 2390228992
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002783662333493821,
      "loss": 0.065,
      "theoretical_loss": 3.4090995246315403,
      "tokens_seen": 2390491136
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002782859894077997,
      "loss": 0.0652,
      "theoretical_loss": 3.4090696894529886,
      "tokens_seen": 2390753280
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002782057454662173,
      "loss": 0.0668,
      "theoretical_loss": 3.409039858461537,
      "tokens_seen": 2391015424
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002781255015246349,
      "loss": 0.0677,
      "theoretical_loss": 3.4090100316561385,
      "tokens_seen": 2391277568
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002780452575830525,
      "loss": 0.0672,
      "theoretical_loss": 3.408980209035746,
      "tokens_seen": 2391539712
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002779650136414701,
      "loss": 0.067,
      "theoretical_loss": 3.4089503905993155,
      "tokens_seen": 2391801856
    },
    {
      "epoch": 0.72,
      "objective/train/advantage_avg": -0.0005384014802984893,
      "objective/train/docs_used": 871244,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4107896089553833,
      "objective/train/original_loss": 1.4107894897460938,
      "objective/train/theoretical_loss": 3.4089205763458,
      "objective/train/tokens_used": 2412524000,
      "objective/train/value_avg": -0.009735107421875,
      "objective/train/value_loss": 0.00018544522754382342,
      "objective/train/value_max": -3.2961368560791016e-05,
      "objective/train/value_min": -0.63037109375,
      "objective/train/value_reward_corr": 0.8186489105089638,
      "objective/train/value_std": 0.0194091796875,
      "objective/train/weight_avg": 0.9995511770248413,
      "objective/train/weighted_lm_loss": 1.4093742370605469,
      "objective/train/weights_max": 1.6551741361618042,
      "objective/train/weights_min": 0.36919885873794556,
      "theoretical_loss": 3.4089205763458,
      "tokens_seen": 2392064000
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002778847696998877,
      "loss": 0.0674,
      "theoretical_loss": 3.4089205763458,
      "tokens_seen": 2392064000
    },
    {
      "epoch": 0.72,
      "learning_rate": 0.0002778045257583052,
      "loss": 0.0676,
      "theoretical_loss": 3.408890766274155,
      "tokens_seen": 2392326144
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002777242818167228,
      "loss": 0.0669,
      "theoretical_loss": 3.408860960383336,
      "tokens_seen": 2392588288
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027764403787514043,
      "loss": 0.066,
      "theoretical_loss": 3.4088311586722986,
      "tokens_seen": 2392850432
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027756379393355804,
      "loss": 0.0665,
      "theoretical_loss": 3.4088013611399983,
      "tokens_seen": 2393112576
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002774835499919756,
      "loss": 0.0671,
      "theoretical_loss": 3.408771567785392,
      "tokens_seen": 2393374720
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002774033060503932,
      "loss": 0.0667,
      "theoretical_loss": 3.4087417786074363,
      "tokens_seen": 2393636864
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002773230621088108,
      "loss": 0.0678,
      "theoretical_loss": 3.4087119936050887,
      "tokens_seen": 2393899008
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002772428181672284,
      "loss": 0.0675,
      "theoretical_loss": 3.4086822127773067,
      "tokens_seen": 2394161152
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027716257422564594,
      "loss": 0.0646,
      "theoretical_loss": 3.4086524361230475,
      "tokens_seen": 2394423296
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027708233028406355,
      "loss": 0.0659,
      "theoretical_loss": 3.40862266364127,
      "tokens_seen": 2394685440
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027700208634248117,
      "loss": 0.0643,
      "theoretical_loss": 3.4085928953309335,
      "tokens_seen": 2394947584
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002769218424008987,
      "loss": 0.0676,
      "theoretical_loss": 3.408563131190996,
      "tokens_seen": 2395209728
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027684159845931634,
      "loss": 0.0662,
      "theoretical_loss": 3.408533371220417,
      "tokens_seen": 2395471872
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027676135451773395,
      "loss": 0.0663,
      "theoretical_loss": 3.408503615418156,
      "tokens_seen": 2395734016
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002766811105761515,
      "loss": 0.0659,
      "theoretical_loss": 3.4084738637831746,
      "tokens_seen": 2395996160
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027660086663456907,
      "loss": 0.0647,
      "theoretical_loss": 3.408444116314432,
      "tokens_seen": 2396258304
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002765206226929867,
      "loss": 0.0677,
      "theoretical_loss": 3.4084143730108893,
      "tokens_seen": 2396520448
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027644037875140424,
      "loss": 0.0672,
      "theoretical_loss": 3.408384633871509,
      "tokens_seen": 2396782592
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027636013480982185,
      "loss": 0.0684,
      "theoretical_loss": 3.408354898895251,
      "tokens_seen": 2397044736
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027627989086823946,
      "loss": 0.067,
      "theoretical_loss": 3.408325168081078,
      "tokens_seen": 2397306880
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002761996469266571,
      "loss": 0.0665,
      "theoretical_loss": 3.4082954414279523,
      "tokens_seen": 2397569024
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027611940298507464,
      "loss": 0.066,
      "theoretical_loss": 3.408265718934837,
      "tokens_seen": 2397831168
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027603915904349225,
      "loss": 0.0686,
      "theoretical_loss": 3.4082360006006955,
      "tokens_seen": 2398093312
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002759589151019098,
      "loss": 0.0662,
      "theoretical_loss": 3.4082062864244906,
      "tokens_seen": 2398355456
    },
    {
      "epoch": 0.73,
      "objective/train/advantage_avg": 0.0008262444753199816,
      "objective/train/docs_used": 873551,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3028806447982788,
      "objective/train/original_loss": 1.3028804063796997,
      "objective/train/theoretical_loss": 3.4081765764051863,
      "objective/train/tokens_used": 2419077600,
      "objective/train/value_avg": -0.00951385498046875,
      "objective/train/value_loss": 0.0004103725077584386,
      "objective/train/value_max": -3.1948089599609375e-05,
      "objective/train/value_min": -0.8154296875,
      "objective/train/value_reward_corr": 0.7364599665479659,
      "objective/train/value_std": 0.0216217041015625,
      "objective/train/weight_avg": 1.0010087490081787,
      "objective/train/weighted_lm_loss": 1.302687644958496,
      "objective/train/weights_max": 1.6794044971466064,
      "objective/train/weights_min": 0.3682406544685364,
      "theoretical_loss": 3.4081765764051863,
      "tokens_seen": 2398617600
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027587867116032736,
      "loss": 0.0653,
      "theoretical_loss": 3.4081765764051863,
      "tokens_seen": 2398617600
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.000275798427218745,
      "loss": 0.0676,
      "theoretical_loss": 3.4081468705417475,
      "tokens_seen": 2398879744
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002757181832771626,
      "loss": 0.0642,
      "theoretical_loss": 3.408117168833138,
      "tokens_seen": 2399141888
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002756379393355802,
      "loss": 0.0649,
      "theoretical_loss": 3.408087471278323,
      "tokens_seen": 2399404032
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027555769539399776,
      "loss": 0.0657,
      "theoretical_loss": 3.4080577778762686,
      "tokens_seen": 2399666176
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002754774514524154,
      "loss": 0.0691,
      "theoretical_loss": 3.4080280886259393,
      "tokens_seen": 2399928320
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027539720751083293,
      "loss": 0.0665,
      "theoretical_loss": 3.4079984035263022,
      "tokens_seen": 2400190464
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002753169635692505,
      "loss": 0.0647,
      "theoretical_loss": 3.4079687225763236,
      "tokens_seen": 2400452608
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002752367196276681,
      "loss": 0.0676,
      "theoretical_loss": 3.40793904577497,
      "tokens_seen": 2400714752
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002751564756860857,
      "loss": 0.0688,
      "theoretical_loss": 3.4079093731212087,
      "tokens_seen": 2400976896
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027507623174450333,
      "loss": 0.0648,
      "theoretical_loss": 3.4078797046140075,
      "tokens_seen": 2401239040
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002749959878029209,
      "loss": 0.067,
      "theoretical_loss": 3.4078500402523346,
      "tokens_seen": 2401501184
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002749157438613385,
      "loss": 0.0655,
      "theoretical_loss": 3.407820380035157,
      "tokens_seen": 2401763328
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027483549991975606,
      "loss": 0.0669,
      "theoretical_loss": 3.4077907239614453,
      "tokens_seen": 2402025472
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002747552559781736,
      "loss": 0.0681,
      "theoretical_loss": 3.4077610720301674,
      "tokens_seen": 2402287616
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027467501203659123,
      "loss": 0.069,
      "theoretical_loss": 3.407731424240293,
      "tokens_seen": 2402549760
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027459476809500884,
      "loss": 0.0658,
      "theoretical_loss": 3.4077017805907923,
      "tokens_seen": 2402811904
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002745145241534264,
      "loss": 0.0707,
      "theoretical_loss": 3.4076721410806345,
      "tokens_seen": 2403074048
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.000274434280211844,
      "loss": 0.0659,
      "theoretical_loss": 3.407642505708791,
      "tokens_seen": 2403336192
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002743540362702616,
      "loss": 0.0674,
      "theoretical_loss": 3.4076128744742324,
      "tokens_seen": 2403598336
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027427379232867924,
      "loss": 0.0664,
      "theoretical_loss": 3.4075832473759298,
      "tokens_seen": 2403860480
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027419354838709674,
      "loss": 0.0715,
      "theoretical_loss": 3.4075536244128553,
      "tokens_seen": 2404122624
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027411330444551435,
      "loss": 0.0702,
      "theoretical_loss": 3.4075240055839804,
      "tokens_seen": 2404384768
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027403306050393197,
      "loss": 0.0667,
      "theoretical_loss": 3.407494390888278,
      "tokens_seen": 2404646912
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002739528165623495,
      "loss": 0.068,
      "theoretical_loss": 3.4074647803247204,
      "tokens_seen": 2404909056
    },
    {
      "epoch": 0.73,
      "objective/train/advantage_avg": 0.0003997816820628941,
      "objective/train/docs_used": 875512,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3201677799224854,
      "objective/train/original_loss": 1.3201675415039062,
      "objective/train/theoretical_loss": 3.407435173892281,
      "objective/train/tokens_used": 2425631200,
      "objective/train/value_avg": -0.006839752197265625,
      "objective/train/value_loss": 0.0002915321383625269,
      "objective/train/value_max": -2.485513687133789e-05,
      "objective/train/value_min": -0.943359375,
      "objective/train/value_reward_corr": 0.6612879561403876,
      "objective/train/value_std": 0.0157928466796875,
      "objective/train/weight_avg": 1.0005264282226562,
      "objective/train/weighted_lm_loss": 1.3203579187393188,
      "objective/train/weights_max": 1.6203583478927612,
      "objective/train/weights_min": 0.36834144592285156,
      "theoretical_loss": 3.407435173892281,
      "tokens_seen": 2405171200
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027387257262076714,
      "loss": 0.064,
      "theoretical_loss": 3.407435173892281,
      "tokens_seen": 2405171200
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027379232867918475,
      "loss": 0.0696,
      "theoretical_loss": 3.407405571589933,
      "tokens_seen": 2405433344
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027371208473760236,
      "loss": 0.0686,
      "theoretical_loss": 3.4073759734166504,
      "tokens_seen": 2405695488
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027363184079601987,
      "loss": 0.0668,
      "theoretical_loss": 3.4073463793714076,
      "tokens_seen": 2405957632
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002735515968544375,
      "loss": 0.0697,
      "theoretical_loss": 3.407316789453179,
      "tokens_seen": 2406219776
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002734713529128551,
      "loss": 0.0663,
      "theoretical_loss": 3.40728720366094,
      "tokens_seen": 2406481920
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027339110897127265,
      "loss": 0.0688,
      "theoretical_loss": 3.407257621993665,
      "tokens_seen": 2406744064
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027331086502969026,
      "loss": 0.0659,
      "theoretical_loss": 3.4072280444503305,
      "tokens_seen": 2407006208
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002732306210881079,
      "loss": 0.0679,
      "theoretical_loss": 3.407198471029912,
      "tokens_seen": 2407268352
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002731503771465255,
      "loss": 0.0677,
      "theoretical_loss": 3.4071689017313864,
      "tokens_seen": 2407530496
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.000273070133204943,
      "loss": 0.0686,
      "theoretical_loss": 3.40713933655373,
      "tokens_seen": 2407792640
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002729898892633606,
      "loss": 0.0713,
      "theoretical_loss": 3.40710977549592,
      "tokens_seen": 2408054784
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002729096453217782,
      "loss": 0.068,
      "theoretical_loss": 3.4070802185569344,
      "tokens_seen": 2408316928
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002728294013801958,
      "loss": 0.0686,
      "theoretical_loss": 3.407050665735751,
      "tokens_seen": 2408579072
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002727491574386134,
      "loss": 0.0672,
      "theoretical_loss": 3.407021117031347,
      "tokens_seen": 2408841216
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.000272668913497031,
      "loss": 0.066,
      "theoretical_loss": 3.4069915724427027,
      "tokens_seen": 2409103360
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002725886695554486,
      "loss": 0.0662,
      "theoretical_loss": 3.4069620319687957,
      "tokens_seen": 2409365504
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027250842561386617,
      "loss": 0.0693,
      "theoretical_loss": 3.406932495608606,
      "tokens_seen": 2409627648
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027242818167228373,
      "loss": 0.0673,
      "theoretical_loss": 3.4069029633611128,
      "tokens_seen": 2409889792
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027234793773070134,
      "loss": 0.0678,
      "theoretical_loss": 3.4068734352252967,
      "tokens_seen": 2410151936
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002722676937891189,
      "loss": 0.0667,
      "theoretical_loss": 3.406843911200138,
      "tokens_seen": 2410414080
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002721874498475365,
      "loss": 0.0686,
      "theoretical_loss": 3.406814391284617,
      "tokens_seen": 2410676224
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027210720590595413,
      "loss": 0.0655,
      "theoretical_loss": 3.4067848754777152,
      "tokens_seen": 2410938368
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002720269619643717,
      "loss": 0.0679,
      "theoretical_loss": 3.4067553637784145,
      "tokens_seen": 2411200512
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002719467180227893,
      "loss": 0.0683,
      "theoretical_loss": 3.406725856185696,
      "tokens_seen": 2411462656
    },
    {
      "epoch": 0.73,
      "objective/train/advantage_avg": 0.0002313482400495559,
      "objective/train/docs_used": 877919,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3729504346847534,
      "objective/train/original_loss": 1.3729503154754639,
      "objective/train/theoretical_loss": 3.4066963526985425,
      "objective/train/tokens_used": 2432184800,
      "objective/train/value_avg": -0.007843017578125,
      "objective/train/value_loss": 0.00015894180978648365,
      "objective/train/value_max": -3.9458274841308594e-05,
      "objective/train/value_min": -0.697265625,
      "objective/train/value_reward_corr": 0.7240007122567931,
      "objective/train/value_std": 0.01336669921875,
      "objective/train/weight_avg": 1.0003055334091187,
      "objective/train/weighted_lm_loss": 1.3730709552764893,
      "objective/train/weights_max": 1.3394194841384888,
      "objective/train/weights_min": 0.3690439462661743,
      "theoretical_loss": 3.4066963526985425,
      "tokens_seen": 2411724800
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027186647408120686,
      "loss": 0.0671,
      "theoretical_loss": 3.4066963526985425,
      "tokens_seen": 2411724800
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027178623013962447,
      "loss": 0.0675,
      "theoretical_loss": 3.4066668533159365,
      "tokens_seen": 2411986944
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027170598619804203,
      "loss": 0.0689,
      "theoretical_loss": 3.4066373580368614,
      "tokens_seen": 2412249088
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027162574225645964,
      "loss": 0.069,
      "theoretical_loss": 3.4066078668602997,
      "tokens_seen": 2412511232
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027154549831487725,
      "loss": 0.0673,
      "theoretical_loss": 3.4065783797852354,
      "tokens_seen": 2412773376
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002714652543732948,
      "loss": 0.0669,
      "theoretical_loss": 3.4065488968106528,
      "tokens_seen": 2413035520
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002713850104317124,
      "loss": 0.0691,
      "theoretical_loss": 3.406519417935536,
      "tokens_seen": 2413297664
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027130476649013004,
      "loss": 0.0686,
      "theoretical_loss": 3.4064899431588698,
      "tokens_seen": 2413559808
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002712245225485476,
      "loss": 0.067,
      "theoretical_loss": 3.4064604724796395,
      "tokens_seen": 2413821952
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027114427860696515,
      "loss": 0.0683,
      "theoretical_loss": 3.4064310058968306,
      "tokens_seen": 2414084096
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027106403466538277,
      "loss": 0.0639,
      "theoretical_loss": 3.406401543409429,
      "tokens_seen": 2414346240
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002709837907238004,
      "loss": 0.0689,
      "theoretical_loss": 3.4063720850164203,
      "tokens_seen": 2414608384
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027090354678221794,
      "loss": 0.0674,
      "theoretical_loss": 3.406342630716792,
      "tokens_seen": 2414870528
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027082330284063555,
      "loss": 0.068,
      "theoretical_loss": 3.406313180509531,
      "tokens_seen": 2415132672
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027074305889905316,
      "loss": 0.0677,
      "theoretical_loss": 3.406283734393624,
      "tokens_seen": 2415394816
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027066281495747067,
      "loss": 0.066,
      "theoretical_loss": 3.4062542923680588,
      "tokens_seen": 2415656960
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002705825710158883,
      "loss": 0.0681,
      "theoretical_loss": 3.406224854431824,
      "tokens_seen": 2415919104
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002705023270743059,
      "loss": 0.0684,
      "theoretical_loss": 3.406195420583907,
      "tokens_seen": 2416181248
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002704220831327235,
      "loss": 0.0662,
      "theoretical_loss": 3.4061659908232973,
      "tokens_seen": 2416443392
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00027034183919114106,
      "loss": 0.0668,
      "theoretical_loss": 3.4061365651489837,
      "tokens_seen": 2416705536
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002702615952495587,
      "loss": 0.0686,
      "theoretical_loss": 3.406107143559956,
      "tokens_seen": 2416967680
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002701813513079763,
      "loss": 0.0678,
      "theoretical_loss": 3.4060777260552038,
      "tokens_seen": 2417229824
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002701011073663938,
      "loss": 0.0685,
      "theoretical_loss": 3.4060483126337173,
      "tokens_seen": 2417491968
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002700208634248114,
      "loss": 0.0695,
      "theoretical_loss": 3.4060189032944868,
      "tokens_seen": 2417754112
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.000269940619483229,
      "loss": 0.066,
      "theoretical_loss": 3.4059894980365035,
      "tokens_seen": 2418016256
    },
    {
      "epoch": 0.73,
      "objective/train/advantage_avg": -0.0001355194835923612,
      "objective/train/docs_used": 880852,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3654475212097168,
      "objective/train/original_loss": 1.3654474020004272,
      "objective/train/theoretical_loss": 3.4059600968587587,
      "objective/train/tokens_used": 2438738400,
      "objective/train/value_avg": -0.007030487060546875,
      "objective/train/value_loss": 0.0004641265550162643,
      "objective/train/value_max": -4.684925079345703e-05,
      "objective/train/value_min": -0.6494140625,
      "objective/train/value_reward_corr": 0.642583948534395,
      "objective/train/value_std": 0.01666259765625,
      "objective/train/weight_avg": 1.0000636577606201,
      "objective/train/weighted_lm_loss": 1.3649080991744995,
      "objective/train/weights_max": 1.517284870147705,
      "objective/train/weights_min": 0.36884480714797974,
      "theoretical_loss": 3.4059600968587587,
      "tokens_seen": 2418278400
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00026986037554164663,
      "loss": 0.0675,
      "theoretical_loss": 3.4059600968587587,
      "tokens_seen": 2418278400
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002697801316000642,
      "loss": 0.0664,
      "theoretical_loss": 3.4059306997602437,
      "tokens_seen": 2418540544
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002696998876584818,
      "loss": 0.066,
      "theoretical_loss": 3.405901306739951,
      "tokens_seen": 2418802688
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002696196437168994,
      "loss": 0.0659,
      "theoretical_loss": 3.405871917796872,
      "tokens_seen": 2419064832
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00026953939977531697,
      "loss": 0.0697,
      "theoretical_loss": 3.405842532930001,
      "tokens_seen": 2419326976
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00026945915583373453,
      "loss": 0.0653,
      "theoretical_loss": 3.4058131521383292,
      "tokens_seen": 2419589120
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00026937891189215214,
      "loss": 0.0652,
      "theoretical_loss": 3.4057837754208515,
      "tokens_seen": 2419851264
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00026929866795056976,
      "loss": 0.0677,
      "theoretical_loss": 3.4057544027765605,
      "tokens_seen": 2420113408
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002692184240089873,
      "loss": 0.07,
      "theoretical_loss": 3.4057250342044516,
      "tokens_seen": 2420375552
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002691381800674049,
      "loss": 0.0652,
      "theoretical_loss": 3.405695669703518,
      "tokens_seen": 2420637696
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00026905793612582254,
      "loss": 0.0662,
      "theoretical_loss": 3.405666309272755,
      "tokens_seen": 2420899840
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002689776921842401,
      "loss": 0.0682,
      "theoretical_loss": 3.4056369529111583,
      "tokens_seen": 2421161984
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00026889744824265766,
      "loss": 0.0679,
      "theoretical_loss": 3.4056076006177234,
      "tokens_seen": 2421424128
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00026881720430107527,
      "loss": 0.0661,
      "theoretical_loss": 3.4055782523914453,
      "tokens_seen": 2421686272
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002687369603594929,
      "loss": 0.0678,
      "theoretical_loss": 3.405548908231321,
      "tokens_seen": 2421948416
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00026865671641791044,
      "loss": 0.0665,
      "theoretical_loss": 3.405519568136347,
      "tokens_seen": 2422210560
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00026857647247632805,
      "loss": 0.0715,
      "theoretical_loss": 3.40549023210552,
      "tokens_seen": 2422472704
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00026849622853474566,
      "loss": 0.0704,
      "theoretical_loss": 3.4054609001378378,
      "tokens_seen": 2422734848
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002684159845931632,
      "loss": 0.0683,
      "theoretical_loss": 3.4054315722322976,
      "tokens_seen": 2422996992
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002683357406515808,
      "loss": 0.0664,
      "theoretical_loss": 3.4054022483878983,
      "tokens_seen": 2423259136
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002682554967099984,
      "loss": 0.0682,
      "theoretical_loss": 3.4053729286036374,
      "tokens_seen": 2423521280
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00026817525276841595,
      "loss": 0.0662,
      "theoretical_loss": 3.405343612878514,
      "tokens_seen": 2423783424
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00026809500882683357,
      "loss": 0.0651,
      "theoretical_loss": 3.405314301211527,
      "tokens_seen": 2424045568
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002680147648852512,
      "loss": 0.0653,
      "theoretical_loss": 3.4052849936016765,
      "tokens_seen": 2424307712
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002679345209436688,
      "loss": 0.0671,
      "theoretical_loss": 3.4052556900479614,
      "tokens_seen": 2424569856
    },
    {
      "epoch": 0.73,
      "objective/train/advantage_avg": 0.0004885383532382548,
      "objective/train/docs_used": 882990,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3813307285308838,
      "objective/train/original_loss": 1.3813307285308838,
      "objective/train/theoretical_loss": 3.405226390549383,
      "objective/train/tokens_used": 2445292000,
      "objective/train/value_avg": -0.00766754150390625,
      "objective/train/value_loss": 0.00018243658996652812,
      "objective/train/value_max": -4.2319297790527344e-05,
      "objective/train/value_min": -0.306640625,
      "objective/train/value_reward_corr": 0.6878939764227319,
      "objective/train/value_std": 0.012908935546875,
      "objective/train/weight_avg": 1.0005711317062378,
      "objective/train/weighted_lm_loss": 1.381544828414917,
      "objective/train/weights_max": 1.1512739658355713,
      "objective/train/weights_min": 0.3689032196998596,
      "theoretical_loss": 3.405226390549383,
      "tokens_seen": 2424832000
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00026785427700208635,
      "loss": 0.0676,
      "theoretical_loss": 3.405226390549383,
      "tokens_seen": 2424832000
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.00026777403306050396,
      "loss": 0.0685,
      "theoretical_loss": 3.4051970951049406,
      "tokens_seen": 2425094144
    },
    {
      "epoch": 0.73,
      "learning_rate": 0.0002676937891189215,
      "loss": 0.0694,
      "theoretical_loss": 3.4051678037136357,
      "tokens_seen": 2425356288
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002676135451773391,
      "loss": 0.0675,
      "theoretical_loss": 3.40513851637447,
      "tokens_seen": 2425618432
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002675333012357567,
      "loss": 0.0679,
      "theoretical_loss": 3.4051092330864443,
      "tokens_seen": 2425880576
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002674530572941743,
      "loss": 0.0674,
      "theoretical_loss": 3.405079953848561,
      "tokens_seen": 2426142720
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002673728133525919,
      "loss": 0.068,
      "theoretical_loss": 3.4050506786598222,
      "tokens_seen": 2426404864
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002672925694110095,
      "loss": 0.0687,
      "theoretical_loss": 3.405021407519231,
      "tokens_seen": 2426667008
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002672123254694271,
      "loss": 0.068,
      "theoretical_loss": 3.4049921404257897,
      "tokens_seen": 2426929152
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026713208152784465,
      "loss": 0.0652,
      "theoretical_loss": 3.4049628773785017,
      "tokens_seen": 2427191296
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002670518375862622,
      "loss": 0.0649,
      "theoretical_loss": 3.4049336183763717,
      "tokens_seen": 2427453440
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002669715936446798,
      "loss": 0.0677,
      "theoretical_loss": 3.4049043634184026,
      "tokens_seen": 2427715584
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026689134970309743,
      "loss": 0.0681,
      "theoretical_loss": 3.4048751125035994,
      "tokens_seen": 2427977728
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026681110576151504,
      "loss": 0.067,
      "theoretical_loss": 3.404845865630967,
      "tokens_seen": 2428239872
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002667308618199326,
      "loss": 0.0675,
      "theoretical_loss": 3.40481662279951,
      "tokens_seen": 2428502016
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002666506178783502,
      "loss": 0.0671,
      "theoretical_loss": 3.4047873840082348,
      "tokens_seen": 2428764160
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002665703739367678,
      "loss": 0.0673,
      "theoretical_loss": 3.404758149256146,
      "tokens_seen": 2429026304
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026649012999518533,
      "loss": 0.0671,
      "theoretical_loss": 3.4047289185422507,
      "tokens_seen": 2429288448
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026640988605360294,
      "loss": 0.0653,
      "theoretical_loss": 3.404699691865555,
      "tokens_seen": 2429550592
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026632964211202055,
      "loss": 0.0681,
      "theoretical_loss": 3.4046704692250658,
      "tokens_seen": 2429812736
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002662493981704381,
      "loss": 0.0663,
      "theoretical_loss": 3.404641250619791,
      "tokens_seen": 2430074880
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002661691542288557,
      "loss": 0.0694,
      "theoretical_loss": 3.404612036048737,
      "tokens_seen": 2430337024
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026608891028727334,
      "loss": 0.0684,
      "theoretical_loss": 3.404582825510913,
      "tokens_seen": 2430599168
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026600866634569095,
      "loss": 0.0668,
      "theoretical_loss": 3.404553619005326,
      "tokens_seen": 2430861312
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026592842240410846,
      "loss": 0.0673,
      "theoretical_loss": 3.404524416530986,
      "tokens_seen": 2431123456
    },
    {
      "epoch": 0.74,
      "objective/train/advantage_avg": 0.0006771566113457084,
      "objective/train/docs_used": 885315,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3741639852523804,
      "objective/train/original_loss": 1.3741638660430908,
      "objective/train/theoretical_loss": 3.4044952180869013,
      "objective/train/tokens_used": 2451845600,
      "objective/train/value_avg": -0.006244659423828125,
      "objective/train/value_loss": 0.00024454991216771305,
      "objective/train/value_max": -2.6702880859375e-05,
      "objective/train/value_min": -0.79736328125,
      "objective/train/value_reward_corr": 0.587969724531287,
      "objective/train/value_std": 0.011932373046875,
      "objective/train/weight_avg": 1.0007866621017456,
      "objective/train/weighted_lm_loss": 1.3749604225158691,
      "objective/train/weights_max": 2.1188037395477295,
      "objective/train/weights_min": 0.369240403175354,
      "theoretical_loss": 3.4044952180869013,
      "tokens_seen": 2431385600
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026584817846252607,
      "loss": 0.0673,
      "theoretical_loss": 3.4044952180869013,
      "tokens_seen": 2431385600
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002657679345209437,
      "loss": 0.0689,
      "theoretical_loss": 3.4044660236720805,
      "tokens_seen": 2431647744
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026568769057936124,
      "loss": 0.0655,
      "theoretical_loss": 3.404436833285535,
      "tokens_seen": 2431909888
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026560744663777885,
      "loss": 0.0671,
      "theoretical_loss": 3.4044076469262734,
      "tokens_seen": 2432172032
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026552720269619646,
      "loss": 0.0653,
      "theoretical_loss": 3.4043784645933064,
      "tokens_seen": 2432434176
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002654469587546141,
      "loss": 0.0671,
      "theoretical_loss": 3.4043492862856457,
      "tokens_seen": 2432696320
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002653667148130316,
      "loss": 0.0684,
      "theoretical_loss": 3.404320112002301,
      "tokens_seen": 2432958464
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002652864708714492,
      "loss": 0.0676,
      "theoretical_loss": 3.4042909417422846,
      "tokens_seen": 2433220608
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002652062269298668,
      "loss": 0.0679,
      "theoretical_loss": 3.404261775504608,
      "tokens_seen": 2433482752
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026512598298828436,
      "loss": 0.0658,
      "theoretical_loss": 3.404232613288283,
      "tokens_seen": 2433744896
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.000265045739046702,
      "loss": 0.0676,
      "theoretical_loss": 3.404203455092323,
      "tokens_seen": 2434007040
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002649654951051196,
      "loss": 0.0652,
      "theoretical_loss": 3.40417430091574,
      "tokens_seen": 2434269184
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002648852511635372,
      "loss": 0.0675,
      "theoretical_loss": 3.404145150757548,
      "tokens_seen": 2434531328
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026480500722195476,
      "loss": 0.0684,
      "theoretical_loss": 3.4041160046167596,
      "tokens_seen": 2434793472
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002647247632803723,
      "loss": 0.0664,
      "theoretical_loss": 3.4040868624923895,
      "tokens_seen": 2435055616
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026464451933878993,
      "loss": 0.0687,
      "theoretical_loss": 3.4040577243834513,
      "tokens_seen": 2435317760
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002645642753972075,
      "loss": 0.0649,
      "theoretical_loss": 3.4040285902889598,
      "tokens_seen": 2435579904
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002644840314556251,
      "loss": 0.0682,
      "theoretical_loss": 3.40399946020793,
      "tokens_seen": 2435842048
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002644037875140427,
      "loss": 0.0664,
      "theoretical_loss": 3.403970334139377,
      "tokens_seen": 2436104192
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026432354357246033,
      "loss": 0.0655,
      "theoretical_loss": 3.403941212082317,
      "tokens_seen": 2436366336
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002642432996308779,
      "loss": 0.0677,
      "theoretical_loss": 3.4039120940357654,
      "tokens_seen": 2436628480
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026416305568929544,
      "loss": 0.0664,
      "theoretical_loss": 3.4038829799987385,
      "tokens_seen": 2436890624
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026408281174771306,
      "loss": 0.0673,
      "theoretical_loss": 3.4038538699702534,
      "tokens_seen": 2437152768
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002640025678061306,
      "loss": 0.0665,
      "theoretical_loss": 3.403824763949326,
      "tokens_seen": 2437414912
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026392232386454823,
      "loss": 0.0673,
      "theoretical_loss": 3.403795661934976,
      "tokens_seen": 2437677056
    },
    {
      "epoch": 0.74,
      "objective/train/advantage_avg": 3.9483591535827145e-05,
      "objective/train/docs_used": 887686,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2737083435058594,
      "objective/train/original_loss": 1.2737083435058594,
      "objective/train/theoretical_loss": 3.4037665639262182,
      "objective/train/tokens_used": 2458399200,
      "objective/train/value_avg": -0.00609588623046875,
      "objective/train/value_loss": 0.00029432339943014085,
      "objective/train/value_max": -4.00543212890625e-05,
      "objective/train/value_min": -0.70263671875,
      "objective/train/value_reward_corr": 0.6688808394913893,
      "objective/train/value_std": 0.0136260986328125,
      "objective/train/weight_avg": 1.0001697540283203,
      "objective/train/weighted_lm_loss": 1.2733510732650757,
      "objective/train/weights_max": 1.6455042362213135,
      "objective/train/weights_min": 0.3695588707923889,
      "theoretical_loss": 3.4037665639262182,
      "tokens_seen": 2437939200
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026384207992296584,
      "loss": 0.0643,
      "theoretical_loss": 3.4037665639262182,
      "tokens_seen": 2437939200
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002637618359813834,
      "loss": 0.0655,
      "theoretical_loss": 3.403737469922073,
      "tokens_seen": 2438201344
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.000263681592039801,
      "loss": 0.0675,
      "theoretical_loss": 3.4037083799215573,
      "tokens_seen": 2438463488
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026360134809821857,
      "loss": 0.0648,
      "theoretical_loss": 3.40367929392369,
      "tokens_seen": 2438725632
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002635211041566362,
      "loss": 0.0661,
      "theoretical_loss": 3.4036502119274914,
      "tokens_seen": 2438987776
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026344086021505374,
      "loss": 0.0633,
      "theoretical_loss": 3.4036211339319795,
      "tokens_seen": 2439249920
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026336061627347135,
      "loss": 0.0658,
      "theoretical_loss": 3.4035920599361753,
      "tokens_seen": 2439512064
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026328037233188897,
      "loss": 0.0686,
      "theoretical_loss": 3.403562989939098,
      "tokens_seen": 2439774208
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002632001283903065,
      "loss": 0.0679,
      "theoretical_loss": 3.4035339239397686,
      "tokens_seen": 2440036352
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026311988444872414,
      "loss": 0.0688,
      "theoretical_loss": 3.4035048619372072,
      "tokens_seen": 2440298496
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026303964050714175,
      "loss": 0.0663,
      "theoretical_loss": 3.403475803930436,
      "tokens_seen": 2440560640
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002629593965655593,
      "loss": 0.0663,
      "theoretical_loss": 3.4034467499184755,
      "tokens_seen": 2440822784
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026287915262397687,
      "loss": 0.0662,
      "theoretical_loss": 3.4034176999003485,
      "tokens_seen": 2441084928
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002627989086823945,
      "loss": 0.0648,
      "theoretical_loss": 3.4033886538750764,
      "tokens_seen": 2441347072
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002627186647408121,
      "loss": 0.0668,
      "theoretical_loss": 3.4033596118416822,
      "tokens_seen": 2441609216
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026263842079922965,
      "loss": 0.0632,
      "theoretical_loss": 3.4033305737991886,
      "tokens_seen": 2441871360
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026255817685764726,
      "loss": 0.0681,
      "theoretical_loss": 3.403301539746619,
      "tokens_seen": 2442133504
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002624779329160649,
      "loss": 0.0665,
      "theoretical_loss": 3.4032725096829965,
      "tokens_seen": 2442395648
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002623976889744824,
      "loss": 0.0669,
      "theoretical_loss": 3.4032434836073455,
      "tokens_seen": 2442657792
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002623174450329,
      "loss": 0.0688,
      "theoretical_loss": 3.40321446151869,
      "tokens_seen": 2442919936
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002622372010913176,
      "loss": 0.0679,
      "theoretical_loss": 3.4031854434160556,
      "tokens_seen": 2443182080
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002621569571497352,
      "loss": 0.0682,
      "theoretical_loss": 3.4031564292984653,
      "tokens_seen": 2443444224
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002620767132081528,
      "loss": 0.0665,
      "theoretical_loss": 3.403127419164946,
      "tokens_seen": 2443706368
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002619964692665704,
      "loss": 0.0681,
      "theoretical_loss": 3.4030984130145225,
      "tokens_seen": 2443968512
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.000261916225324988,
      "loss": 0.069,
      "theoretical_loss": 3.4030694108462214,
      "tokens_seen": 2444230656
    },
    {
      "epoch": 0.74,
      "objective/train/advantage_avg": 0.0003610901185311377,
      "objective/train/docs_used": 890203,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4358042478561401,
      "objective/train/original_loss": 1.4358041286468506,
      "objective/train/theoretical_loss": 3.4030404126590685,
      "objective/train/tokens_used": 2464952800,
      "objective/train/value_avg": -0.009307861328125,
      "objective/train/value_loss": 0.00024326522543560714,
      "objective/train/value_max": -2.9981136322021484e-05,
      "objective/train/value_min": -0.69384765625,
      "objective/train/value_reward_corr": 0.7629903910047869,
      "objective/train/value_std": 0.0187835693359375,
      "objective/train/weight_avg": 1.000473976135254,
      "objective/train/weighted_lm_loss": 1.4363505840301514,
      "objective/train/weights_max": 1.3608444929122925,
      "objective/train/weights_min": 0.37018251419067383,
      "theoretical_loss": 3.4030404126590685,
      "tokens_seen": 2444492800
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026183598138340556,
      "loss": 0.0687,
      "theoretical_loss": 3.4030404126590685,
      "tokens_seen": 2444492800
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002617557374418231,
      "loss": 0.0686,
      "theoretical_loss": 3.403011418452091,
      "tokens_seen": 2444754944
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026167549350024073,
      "loss": 0.0681,
      "theoretical_loss": 3.402982428224315,
      "tokens_seen": 2445017088
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026159524955865834,
      "loss": 0.0688,
      "theoretical_loss": 3.4029534419747685,
      "tokens_seen": 2445279232
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002615150056170759,
      "loss": 0.0678,
      "theoretical_loss": 3.4029244597024793,
      "tokens_seen": 2445541376
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002614347616754935,
      "loss": 0.0658,
      "theoretical_loss": 3.402895481406475,
      "tokens_seen": 2445803520
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026135451773391113,
      "loss": 0.0651,
      "theoretical_loss": 3.4028665070857844,
      "tokens_seen": 2446065664
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002612742737923287,
      "loss": 0.0672,
      "theoretical_loss": 3.402837536739436,
      "tokens_seen": 2446327808
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026119402985074624,
      "loss": 0.067,
      "theoretical_loss": 3.4028085703664583,
      "tokens_seen": 2446589952
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026111378590916386,
      "loss": 0.0671,
      "theoretical_loss": 3.402779607965882,
      "tokens_seen": 2446852096
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026103354196758147,
      "loss": 0.0684,
      "theoretical_loss": 3.4027506495367357,
      "tokens_seen": 2447114240
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026095329802599903,
      "loss": 0.0665,
      "theoretical_loss": 3.4027216950780494,
      "tokens_seen": 2447376384
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026087305408441664,
      "loss": 0.0653,
      "theoretical_loss": 3.4026927445888546,
      "tokens_seen": 2447638528
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026079281014283425,
      "loss": 0.0609,
      "theoretical_loss": 3.4026637980681813,
      "tokens_seen": 2447900672
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002607125662012518,
      "loss": 0.0663,
      "theoretical_loss": 3.4026348555150605,
      "tokens_seen": 2448162816
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026063232225966937,
      "loss": 0.0668,
      "theoretical_loss": 3.402605916928524,
      "tokens_seen": 2448424960
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.000260552078318087,
      "loss": 0.0658,
      "theoretical_loss": 3.4025769823076035,
      "tokens_seen": 2448687104
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002604718343765046,
      "loss": 0.067,
      "theoretical_loss": 3.402548051651331,
      "tokens_seen": 2448949248
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026039159043492215,
      "loss": 0.0689,
      "theoretical_loss": 3.402519124958739,
      "tokens_seen": 2449211392
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026031134649333977,
      "loss": 0.0644,
      "theoretical_loss": 3.4024902022288606,
      "tokens_seen": 2449473536
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002602311025517574,
      "loss": 0.0646,
      "theoretical_loss": 3.4024612834607284,
      "tokens_seen": 2449735680
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026015085861017494,
      "loss": 0.0619,
      "theoretical_loss": 3.402432368653376,
      "tokens_seen": 2449997824
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00026007061466859255,
      "loss": 0.0668,
      "theoretical_loss": 3.4024034578058373,
      "tokens_seen": 2450259968
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002599903707270101,
      "loss": 0.0646,
      "theoretical_loss": 3.4023745509171466,
      "tokens_seen": 2450522112
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00025991012678542767,
      "loss": 0.064,
      "theoretical_loss": 3.4023456479863383,
      "tokens_seen": 2450784256
    },
    {
      "epoch": 0.74,
      "objective/train/advantage_avg": 0.0005494300276041031,
      "objective/train/docs_used": 891914,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3421109914779663,
      "objective/train/original_loss": 1.3421109914779663,
      "objective/train/theoretical_loss": 3.402316749012447,
      "objective/train/tokens_used": 2471506400,
      "objective/train/value_avg": -0.006969451904296875,
      "objective/train/value_loss": 0.0004489516431931406,
      "objective/train/value_max": -2.0623207092285156e-05,
      "objective/train/value_min": -0.72607421875,
      "objective/train/value_reward_corr": 0.6451283943786772,
      "objective/train/value_std": 0.0165252685546875,
      "objective/train/weight_avg": 1.0007450580596924,
      "objective/train/weighted_lm_loss": 1.3425601720809937,
      "objective/train/weights_max": 1.8591458797454834,
      "objective/train/weights_min": 0.37829095125198364,
      "theoretical_loss": 3.402316749012447,
      "tokens_seen": 2451046400
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002598298828438453,
      "loss": 0.0648,
      "theoretical_loss": 3.402316749012447,
      "tokens_seen": 2451046400
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002597496389022629,
      "loss": 0.0623,
      "theoretical_loss": 3.4022878539945087,
      "tokens_seen": 2451308544
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002596693949606805,
      "loss": 0.0663,
      "theoretical_loss": 3.402258962931558,
      "tokens_seen": 2451570688
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00025958915101909806,
      "loss": 0.0645,
      "theoretical_loss": 3.4022300758226307,
      "tokens_seen": 2451832832
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002595089070775157,
      "loss": 0.0657,
      "theoretical_loss": 3.4022011926667637,
      "tokens_seen": 2452094976
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00025942866313593323,
      "loss": 0.0636,
      "theoretical_loss": 3.402172313462993,
      "tokens_seen": 2452357120
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002593484191943508,
      "loss": 0.0651,
      "theoretical_loss": 3.4021434382103557,
      "tokens_seen": 2452619264
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002592681752527684,
      "loss": 0.0642,
      "theoretical_loss": 3.4021145669078887,
      "tokens_seen": 2452881408
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.000259187931311186,
      "loss": 0.0652,
      "theoretical_loss": 3.40208569955463,
      "tokens_seen": 2453143552
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00025910768736960363,
      "loss": 0.0652,
      "theoretical_loss": 3.402056836149617,
      "tokens_seen": 2453405696
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002590274434280212,
      "loss": 0.0635,
      "theoretical_loss": 3.4020279766918886,
      "tokens_seen": 2453667840
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002589471994864388,
      "loss": 0.0663,
      "theoretical_loss": 3.401999121180482,
      "tokens_seen": 2453929984
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00025886695554485636,
      "loss": 0.0604,
      "theoretical_loss": 3.401970269614438,
      "tokens_seen": 2454192128
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002587867116032739,
      "loss": 0.065,
      "theoretical_loss": 3.4019414219927944,
      "tokens_seen": 2454454272
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00025870646766169153,
      "loss": 0.0622,
      "theoretical_loss": 3.4019125783145907,
      "tokens_seen": 2454716416
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00025862622372010914,
      "loss": 0.0629,
      "theoretical_loss": 3.4018837385788676,
      "tokens_seen": 2454978560
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00025854597977852676,
      "loss": 0.0666,
      "theoretical_loss": 3.401854902784665,
      "tokens_seen": 2455240704
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002584657358369443,
      "loss": 0.0653,
      "theoretical_loss": 3.4018260709310235,
      "tokens_seen": 2455502848
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002583854918953619,
      "loss": 0.0634,
      "theoretical_loss": 3.4017972430169836,
      "tokens_seen": 2455764992
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00025830524795377954,
      "loss": 0.0658,
      "theoretical_loss": 3.4017684190415873,
      "tokens_seen": 2456027136
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00025822500401219704,
      "loss": 0.0642,
      "theoretical_loss": 3.4017395990038755,
      "tokens_seen": 2456289280
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00025814476007061466,
      "loss": 0.0656,
      "theoretical_loss": 3.4017107829028905,
      "tokens_seen": 2456551424
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00025806451612903227,
      "loss": 0.0632,
      "theoretical_loss": 3.401681970737674,
      "tokens_seen": 2456813568
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002579842721874498,
      "loss": 0.0653,
      "theoretical_loss": 3.4016531625072695,
      "tokens_seen": 2457075712
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00025790402824586744,
      "loss": 0.0672,
      "theoretical_loss": 3.4016243582107193,
      "tokens_seen": 2457337856
    },
    {
      "epoch": 0.74,
      "objective/train/advantage_avg": 0.0005634583649225533,
      "objective/train/docs_used": 894282,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2582533359527588,
      "objective/train/original_loss": 1.2582532167434692,
      "objective/train/theoretical_loss": 3.4015955578470667,
      "objective/train/tokens_used": 2478060000,
      "objective/train/value_avg": -0.005840301513671875,
      "objective/train/value_loss": 0.00014222218305803835,
      "objective/train/value_max": -2.4497509002685547e-05,
      "objective/train/value_min": -0.39453125,
      "objective/train/value_reward_corr": 0.6571031252196362,
      "objective/train/value_std": 0.010833740234375,
      "objective/train/weight_avg": 1.0006279945373535,
      "objective/train/weighted_lm_loss": 1.2587014436721802,
      "objective/train/weights_max": 1.2988438606262207,
      "objective/train/weights_min": 0.3701740503311157,
      "theoretical_loss": 3.4015955578470667,
      "tokens_seen": 2457600000
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00025782378430428505,
      "loss": 0.063,
      "theoretical_loss": 3.4015955578470667,
      "tokens_seen": 2457600000
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00025774354036270266,
      "loss": 0.0644,
      "theoretical_loss": 3.4015667614153555,
      "tokens_seen": 2457862144
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.00025766329642112017,
      "loss": 0.0632,
      "theoretical_loss": 3.4015379689146292,
      "tokens_seen": 2458124288
    },
    {
      "epoch": 0.74,
      "learning_rate": 0.0002575830524795378,
      "loss": 0.0662,
      "theoretical_loss": 3.4015091803439326,
      "tokens_seen": 2458386432
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002575028085379554,
      "loss": 0.067,
      "theoretical_loss": 3.4014803957023094,
      "tokens_seen": 2458648576
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025742256459637295,
      "loss": 0.0645,
      "theoretical_loss": 3.4014516149888054,
      "tokens_seen": 2458910720
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025734232065479057,
      "loss": 0.0626,
      "theoretical_loss": 3.401422838202466,
      "tokens_seen": 2459172864
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002572620767132082,
      "loss": 0.0618,
      "theoretical_loss": 3.401394065342336,
      "tokens_seen": 2459435008
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002571818327716258,
      "loss": 0.0661,
      "theoretical_loss": 3.4013652964074614,
      "tokens_seen": 2459697152
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025710158883004335,
      "loss": 0.0655,
      "theoretical_loss": 3.4013365313968893,
      "tokens_seen": 2459959296
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002570213448884609,
      "loss": 0.0655,
      "theoretical_loss": 3.4013077703096655,
      "tokens_seen": 2460221440
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002569411009468785,
      "loss": 0.0649,
      "theoretical_loss": 3.4012790131448374,
      "tokens_seen": 2460483584
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002568608570052961,
      "loss": 0.0611,
      "theoretical_loss": 3.4012502599014516,
      "tokens_seen": 2460745728
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002567806130637137,
      "loss": 0.0643,
      "theoretical_loss": 3.4012215105785564,
      "tokens_seen": 2461007872
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002567003691221313,
      "loss": 0.0642,
      "theoretical_loss": 3.401192765175199,
      "tokens_seen": 2461270016
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002566201251805489,
      "loss": 0.0657,
      "theoretical_loss": 3.4011640236904284,
      "tokens_seen": 2461532160
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002565398812389665,
      "loss": 0.0618,
      "theoretical_loss": 3.401135286123293,
      "tokens_seen": 2461794304
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025645963729738403,
      "loss": 0.0647,
      "theoretical_loss": 3.401106552472841,
      "tokens_seen": 2462056448
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025637939335580165,
      "loss": 0.0651,
      "theoretical_loss": 3.4010778227381233,
      "tokens_seen": 2462318592
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002562991494142192,
      "loss": 0.0651,
      "theoretical_loss": 3.4010490969181877,
      "tokens_seen": 2462580736
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002562189054726368,
      "loss": 0.0664,
      "theoretical_loss": 3.401020375012085,
      "tokens_seen": 2462842880
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025613866153105443,
      "loss": 0.066,
      "theoretical_loss": 3.400991657018865,
      "tokens_seen": 2463105024
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.000256058417589472,
      "loss": 0.0661,
      "theoretical_loss": 3.400962942937579,
      "tokens_seen": 2463367168
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002559781736478896,
      "loss": 0.0644,
      "theoretical_loss": 3.400934232767277,
      "tokens_seen": 2463629312
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025589792970630716,
      "loss": 0.063,
      "theoretical_loss": 3.400905526507011,
      "tokens_seen": 2463891456
    },
    {
      "epoch": 0.75,
      "objective/train/advantage_avg": 0.00015025025641079992,
      "objective/train/docs_used": 896700,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2792147397994995,
      "objective/train/original_loss": 1.2792147397994995,
      "objective/train/theoretical_loss": 3.4008768241558323,
      "objective/train/tokens_used": 2484613600,
      "objective/train/value_avg": -0.0105133056640625,
      "objective/train/value_loss": 0.0002328952105017379,
      "objective/train/value_max": -1.8477439880371094e-05,
      "objective/train/value_min": -0.70068359375,
      "objective/train/value_reward_corr": 0.8330814107450704,
      "objective/train/value_std": 0.023284912109375,
      "objective/train/weight_avg": 1.000259518623352,
      "objective/train/weighted_lm_loss": 1.2786136865615845,
      "objective/train/weights_max": 1.8233647346496582,
      "objective/train/weights_min": 0.36843419075012207,
      "theoretical_loss": 3.4008768241558323,
      "tokens_seen": 2464153600
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025581768576472477,
      "loss": 0.0646,
      "theoretical_loss": 3.4008768241558323,
      "tokens_seen": 2464153600
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025573744182314233,
      "loss": 0.066,
      "theoretical_loss": 3.4008481257127925,
      "tokens_seen": 2464415744
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025565719788155994,
      "loss": 0.0631,
      "theoretical_loss": 3.400819431176944,
      "tokens_seen": 2464677888
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025557695393997755,
      "loss": 0.0621,
      "theoretical_loss": 3.4007907405473405,
      "tokens_seen": 2464940032
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002554967099983951,
      "loss": 0.0664,
      "theoretical_loss": 3.400762053823033,
      "tokens_seen": 2465202176
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002554164660568127,
      "loss": 0.064,
      "theoretical_loss": 3.400733371003076,
      "tokens_seen": 2465464320
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025533622211523034,
      "loss": 0.0665,
      "theoretical_loss": 3.4007046920865225,
      "tokens_seen": 2465726464
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002552559781736479,
      "loss": 0.0629,
      "theoretical_loss": 3.4006760170724264,
      "tokens_seen": 2465988608
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025517573423206546,
      "loss": 0.0665,
      "theoretical_loss": 3.4006473459598423,
      "tokens_seen": 2466250752
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025509549029048307,
      "loss": 0.0662,
      "theoretical_loss": 3.4006186787478243,
      "tokens_seen": 2466512896
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002550152463489007,
      "loss": 0.0657,
      "theoretical_loss": 3.4005900154354274,
      "tokens_seen": 2466775040
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025493500240731824,
      "loss": 0.0654,
      "theoretical_loss": 3.400561356021707,
      "tokens_seen": 2467037184
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025485475846573585,
      "loss": 0.0662,
      "theoretical_loss": 3.400532700505719,
      "tokens_seen": 2467299328
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025477451452415346,
      "loss": 0.0668,
      "theoretical_loss": 3.400504048886518,
      "tokens_seen": 2467561472
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.000254694270582571,
      "loss": 0.0644,
      "theoretical_loss": 3.4004754011631615,
      "tokens_seen": 2467823616
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002546140266409886,
      "loss": 0.0644,
      "theoretical_loss": 3.4004467573347053,
      "tokens_seen": 2468085760
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002545337826994062,
      "loss": 0.0659,
      "theoretical_loss": 3.400418117400206,
      "tokens_seen": 2468347904
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002544535387578238,
      "loss": 0.0639,
      "theoretical_loss": 3.4003894813587223,
      "tokens_seen": 2468610048
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025437329481624136,
      "loss": 0.0644,
      "theoretical_loss": 3.4003608492093096,
      "tokens_seen": 2468872192
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.000254293050874659,
      "loss": 0.0653,
      "theoretical_loss": 3.400332220951027,
      "tokens_seen": 2469134336
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002542128069330766,
      "loss": 0.0651,
      "theoretical_loss": 3.400303596582933,
      "tokens_seen": 2469396480
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002541325629914941,
      "loss": 0.0632,
      "theoretical_loss": 3.4002749761040847,
      "tokens_seen": 2469658624
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002540523190499117,
      "loss": 0.0645,
      "theoretical_loss": 3.400246359513542,
      "tokens_seen": 2469920768
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002539720751083293,
      "loss": 0.0677,
      "theoretical_loss": 3.4002177468103643,
      "tokens_seen": 2470182912
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025389183116674693,
      "loss": 0.0646,
      "theoretical_loss": 3.40018913799361,
      "tokens_seen": 2470445056
    },
    {
      "epoch": 0.75,
      "objective/train/advantage_avg": -0.000635757576674223,
      "objective/train/docs_used": 899059,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2872612476348877,
      "objective/train/original_loss": 1.2872612476348877,
      "objective/train/theoretical_loss": 3.4001605330623397,
      "objective/train/tokens_used": 2491167200,
      "objective/train/value_avg": -0.00960540771484375,
      "objective/train/value_loss": 0.00023242933093570173,
      "objective/train/value_max": -2.485513687133789e-05,
      "objective/train/value_min": -0.96484375,
      "objective/train/value_reward_corr": 0.8674741291638919,
      "objective/train/value_std": 0.0236663818359375,
      "objective/train/weight_avg": 0.9994723200798035,
      "objective/train/weighted_lm_loss": 1.2871603965759277,
      "objective/train/weights_max": 2.156646966934204,
      "objective/train/weights_min": 0.37836602330207825,
      "theoretical_loss": 3.4001605330623397,
      "tokens_seen": 2470707200
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002538115872251645,
      "loss": 0.0634,
      "theoretical_loss": 3.4001605330623397,
      "tokens_seen": 2470707200
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002537313432835821,
      "loss": 0.0645,
      "theoretical_loss": 3.400131932015613,
      "tokens_seen": 2470969344
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002536510993419997,
      "loss": 0.0634,
      "theoretical_loss": 3.4001033348524907,
      "tokens_seen": 2471231488
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002535708554004173,
      "loss": 0.0662,
      "theoretical_loss": 3.4000747415720336,
      "tokens_seen": 2471493632
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025349061145883483,
      "loss": 0.0673,
      "theoretical_loss": 3.4000461521733025,
      "tokens_seen": 2471755776
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025341036751725244,
      "loss": 0.0654,
      "theoretical_loss": 3.400017566655359,
      "tokens_seen": 2472017920
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025333012357567006,
      "loss": 0.0648,
      "theoretical_loss": 3.3999889850172655,
      "tokens_seen": 2472280064
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002532498796340876,
      "loss": 0.0643,
      "theoretical_loss": 3.399960407258083,
      "tokens_seen": 2472542208
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025316963569250523,
      "loss": 0.0646,
      "theoretical_loss": 3.3999318333768747,
      "tokens_seen": 2472804352
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025308939175092284,
      "loss": 0.0619,
      "theoretical_loss": 3.3999032633727024,
      "tokens_seen": 2473066496
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002530091478093404,
      "loss": 0.0676,
      "theoretical_loss": 3.3998746972446305,
      "tokens_seen": 2473328640
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025292890386775796,
      "loss": 0.0673,
      "theoretical_loss": 3.3998461349917215,
      "tokens_seen": 2473590784
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025284865992617557,
      "loss": 0.0645,
      "theoretical_loss": 3.3998175766130396,
      "tokens_seen": 2473852928
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002527684159845932,
      "loss": 0.0677,
      "theoretical_loss": 3.399789022107648,
      "tokens_seen": 2474115072
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025268817204301074,
      "loss": 0.0656,
      "theoretical_loss": 3.399760471474612,
      "tokens_seen": 2474377216
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025260792810142835,
      "loss": 0.0652,
      "theoretical_loss": 3.399731924712996,
      "tokens_seen": 2474639360
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025252768415984597,
      "loss": 0.0645,
      "theoretical_loss": 3.3997033818218645,
      "tokens_seen": 2474901504
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002524474402182635,
      "loss": 0.0676,
      "theoretical_loss": 3.3996748428002834,
      "tokens_seen": 2475163648
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025236719627668114,
      "loss": 0.0667,
      "theoretical_loss": 3.3996463076473185,
      "tokens_seen": 2475425792
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002522869523350987,
      "loss": 0.0659,
      "theoretical_loss": 3.399617776362035,
      "tokens_seen": 2475687936
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002522067083935163,
      "loss": 0.0633,
      "theoretical_loss": 3.3995892489435007,
      "tokens_seen": 2475950080
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025212646445193387,
      "loss": 0.065,
      "theoretical_loss": 3.3995607253907805,
      "tokens_seen": 2476212224
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002520462205103515,
      "loss": 0.0659,
      "theoretical_loss": 3.3995322057029425,
      "tokens_seen": 2476474368
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002519659765687691,
      "loss": 0.0641,
      "theoretical_loss": 3.3995036898790536,
      "tokens_seen": 2476736512
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025188573262718665,
      "loss": 0.064,
      "theoretical_loss": 3.3994751779181813,
      "tokens_seen": 2476998656
    },
    {
      "epoch": 0.75,
      "objective/train/advantage_avg": 0.000814015802461654,
      "objective/train/docs_used": 901501,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.241913914680481,
      "objective/train/original_loss": 1.241913914680481,
      "objective/train/theoretical_loss": 3.399446669819394,
      "objective/train/tokens_used": 2497720800,
      "objective/train/value_avg": -0.007843017578125,
      "objective/train/value_loss": 0.00017967262829188257,
      "objective/train/value_max": -4.792213439941406e-05,
      "objective/train/value_min": -0.46923828125,
      "objective/train/value_reward_corr": 0.7246901935578405,
      "objective/train/value_std": 0.01360321044921875,
      "objective/train/weight_avg": 1.000894546508789,
      "objective/train/weighted_lm_loss": 1.2434070110321045,
      "objective/train/weights_max": 1.5335290431976318,
      "objective/train/weights_min": 0.37751826643943787,
      "theoretical_loss": 3.399446669819394,
      "tokens_seen": 2477260800
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025180548868560426,
      "loss": 0.064,
      "theoretical_loss": 3.399446669819394,
      "tokens_seen": 2477260800
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002517252447440218,
      "loss": 0.0659,
      "theoretical_loss": 3.3994181655817592,
      "tokens_seen": 2477522944
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002516450008024394,
      "loss": 0.0634,
      "theoretical_loss": 3.3993896652043465,
      "tokens_seen": 2477785088
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.000251564756860857,
      "loss": 0.0665,
      "theoretical_loss": 3.399361168686224,
      "tokens_seen": 2478047232
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002514845129192746,
      "loss": 0.0654,
      "theoretical_loss": 3.399332676026461,
      "tokens_seen": 2478309376
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002514042689776922,
      "loss": 0.0686,
      "theoretical_loss": 3.399304187224127,
      "tokens_seen": 2478571520
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002513240250361098,
      "loss": 0.0659,
      "theoretical_loss": 3.3992757022782927,
      "tokens_seen": 2478833664
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002512437810945274,
      "loss": 0.0666,
      "theoretical_loss": 3.3992472211880274,
      "tokens_seen": 2479095808
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025116353715294495,
      "loss": 0.0667,
      "theoretical_loss": 3.3992187439524018,
      "tokens_seen": 2479357952
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002510832932113625,
      "loss": 0.0659,
      "theoretical_loss": 3.399190270570487,
      "tokens_seen": 2479620096
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002510030492697801,
      "loss": 0.0666,
      "theoretical_loss": 3.3991618010413536,
      "tokens_seen": 2479882240
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025092280532819773,
      "loss": 0.0675,
      "theoretical_loss": 3.3991333353640742,
      "tokens_seen": 2480144384
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025084256138661534,
      "loss": 0.0689,
      "theoretical_loss": 3.3991048735377194,
      "tokens_seen": 2480406528
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002507623174450329,
      "loss": 0.0664,
      "theoretical_loss": 3.399076415561362,
      "tokens_seen": 2480668672
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002506820735034505,
      "loss": 0.0668,
      "theoretical_loss": 3.399047961434074,
      "tokens_seen": 2480930816
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025060182956186813,
      "loss": 0.0653,
      "theoretical_loss": 3.399019511154929,
      "tokens_seen": 2481192960
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025052158562028563,
      "loss": 0.0647,
      "theoretical_loss": 3.398991064722999,
      "tokens_seen": 2481455104
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025044134167870324,
      "loss": 0.0677,
      "theoretical_loss": 3.3989626221373586,
      "tokens_seen": 2481717248
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025036109773712086,
      "loss": 0.0657,
      "theoretical_loss": 3.3989341833970803,
      "tokens_seen": 2481979392
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025028085379553847,
      "loss": 0.0641,
      "theoretical_loss": 3.398905748501239,
      "tokens_seen": 2482241536
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025020060985395603,
      "loss": 0.0657,
      "theoretical_loss": 3.3988773174489086,
      "tokens_seen": 2482503680
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025012036591237364,
      "loss": 0.0668,
      "theoretical_loss": 3.3988488902391643,
      "tokens_seen": 2482765824
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00025004012197079125,
      "loss": 0.0651,
      "theoretical_loss": 3.3988204668710806,
      "tokens_seen": 2483027968
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002499598780292088,
      "loss": 0.0647,
      "theoretical_loss": 3.398792047343733,
      "tokens_seen": 2483290112
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002498796340876264,
      "loss": 0.0669,
      "theoretical_loss": 3.3987636316561978,
      "tokens_seen": 2483552256
    },
    {
      "epoch": 0.75,
      "objective/train/advantage_avg": 0.0003351647756062448,
      "objective/train/docs_used": 903880,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3003116846084595,
      "objective/train/original_loss": 1.30031156539917,
      "objective/train/theoretical_loss": 3.3987352198075502,
      "objective/train/tokens_used": 2504274400,
      "objective/train/value_avg": -0.00673675537109375,
      "objective/train/value_loss": 0.00015308342699427158,
      "objective/train/value_max": -2.6285648345947266e-05,
      "objective/train/value_min": -0.38037109375,
      "objective/train/value_reward_corr": 0.7115446243621768,
      "objective/train/value_std": 0.01253509521484375,
      "objective/train/weight_avg": 1.0004048347473145,
      "objective/train/weighted_lm_loss": 1.3008301258087158,
      "objective/train/weights_max": 1.1297332048416138,
      "objective/train/weights_min": 0.3682253658771515,
      "theoretical_loss": 3.3987352198075502,
      "tokens_seen": 2483814400
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.000249799390146044,
      "loss": 0.0664,
      "theoretical_loss": 3.3987352198075502,
      "tokens_seen": 2483814400
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00024971914620446154,
      "loss": 0.0668,
      "theoretical_loss": 3.3987068117968664,
      "tokens_seen": 2484076544
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00024963890226287915,
      "loss": 0.0634,
      "theoretical_loss": 3.3986784076232235,
      "tokens_seen": 2484338688
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00024955865832129677,
      "loss": 0.0625,
      "theoretical_loss": 3.3986500072856987,
      "tokens_seen": 2484600832
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002494784143797143,
      "loss": 0.0642,
      "theoretical_loss": 3.398621610783368,
      "tokens_seen": 2484862976
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00024939817043813194,
      "loss": 0.0665,
      "theoretical_loss": 3.3985932181153107,
      "tokens_seen": 2485125120
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00024931792649654955,
      "loss": 0.0655,
      "theoretical_loss": 3.398564829280603,
      "tokens_seen": 2485387264
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002492376825549671,
      "loss": 0.0648,
      "theoretical_loss": 3.3985364442783244,
      "tokens_seen": 2485649408
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00024915743861338467,
      "loss": 0.0655,
      "theoretical_loss": 3.398508063107553,
      "tokens_seen": 2485911552
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002490771946718023,
      "loss": 0.0662,
      "theoretical_loss": 3.3984796857673674,
      "tokens_seen": 2486173696
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002489969507302199,
      "loss": 0.0655,
      "theoretical_loss": 3.3984513122568467,
      "tokens_seen": 2486435840
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00024891670678863745,
      "loss": 0.0676,
      "theoretical_loss": 3.3984229425750714,
      "tokens_seen": 2486697984
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00024883646284705506,
      "loss": 0.0654,
      "theoretical_loss": 3.3983945767211203,
      "tokens_seen": 2486960128
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002487562189054726,
      "loss": 0.066,
      "theoretical_loss": 3.3983662146940734,
      "tokens_seen": 2487222272
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00024867597496389023,
      "loss": 0.0661,
      "theoretical_loss": 3.398337856493012,
      "tokens_seen": 2487484416
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002485957310223078,
      "loss": 0.0664,
      "theoretical_loss": 3.3983095021170158,
      "tokens_seen": 2487746560
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002485154870807254,
      "loss": 0.0663,
      "theoretical_loss": 3.398281151565167,
      "tokens_seen": 2488008704
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.000248435243139143,
      "loss": 0.0694,
      "theoretical_loss": 3.398252804836546,
      "tokens_seen": 2488270848
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002483549991975606,
      "loss": 0.0663,
      "theoretical_loss": 3.3982244619302358,
      "tokens_seen": 2488532992
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002482747552559782,
      "loss": 0.0678,
      "theoretical_loss": 3.398196122845317,
      "tokens_seen": 2488795136
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00024819451131439575,
      "loss": 0.0672,
      "theoretical_loss": 3.3981677875808725,
      "tokens_seen": 2489057280
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00024811426737281336,
      "loss": 0.065,
      "theoretical_loss": 3.3981394561359854,
      "tokens_seen": 2489319424
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002480340234312309,
      "loss": 0.0682,
      "theoretical_loss": 3.398111128509738,
      "tokens_seen": 2489581568
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00024795377948964853,
      "loss": 0.065,
      "theoretical_loss": 3.3980828047012146,
      "tokens_seen": 2489843712
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00024787353554806614,
      "loss": 0.0638,
      "theoretical_loss": 3.3980544847094976,
      "tokens_seen": 2490105856
    },
    {
      "epoch": 0.75,
      "objective/train/advantage_avg": 0.0007647961028851569,
      "objective/train/docs_used": 906478,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3144490718841553,
      "objective/train/original_loss": 1.3144491910934448,
      "objective/train/theoretical_loss": 3.3980261685336712,
      "objective/train/tokens_used": 2510828000,
      "objective/train/value_avg": -0.00479888916015625,
      "objective/train/value_loss": 0.00020321377087384462,
      "objective/train/value_max": -2.3365020751953125e-05,
      "objective/train/value_min": -0.38916015625,
      "objective/train/value_reward_corr": 0.5756124228481705,
      "objective/train/value_std": 0.00930023193359375,
      "objective/train/weight_avg": 1.0008529424667358,
      "objective/train/weighted_lm_loss": 1.315671682357788,
      "objective/train/weights_max": 1.3893797397613525,
      "objective/train/weights_min": 0.3708355128765106,
      "theoretical_loss": 3.3980261685336712,
      "tokens_seen": 2490368000
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002477932916064837,
      "loss": 0.0655,
      "theoretical_loss": 3.3980261685336712,
      "tokens_seen": 2490368000
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002477130476649013,
      "loss": 0.0635,
      "theoretical_loss": 3.39799785617282,
      "tokens_seen": 2490630144
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.00024763280372331887,
      "loss": 0.0664,
      "theoretical_loss": 3.3979695476260288,
      "tokens_seen": 2490892288
    },
    {
      "epoch": 0.75,
      "learning_rate": 0.0002475525597817365,
      "loss": 0.0668,
      "theoretical_loss": 3.397941242892382,
      "tokens_seen": 2491154432
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024747231584015404,
      "loss": 0.0697,
      "theoretical_loss": 3.397912941970965,
      "tokens_seen": 2491416576
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024739207189857166,
      "loss": 0.0677,
      "theoretical_loss": 3.3978846448608633,
      "tokens_seen": 2491678720
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024731182795698927,
      "loss": 0.0669,
      "theoretical_loss": 3.3978563515611624,
      "tokens_seen": 2491940864
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024723158401540683,
      "loss": 0.0674,
      "theoretical_loss": 3.397828062070949,
      "tokens_seen": 2492203008
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024715134007382444,
      "loss": 0.0639,
      "theoretical_loss": 3.3977997763893093,
      "tokens_seen": 2492465152
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.000247071096132242,
      "loss": 0.069,
      "theoretical_loss": 3.3977714945153297,
      "tokens_seen": 2492727296
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002469908521906596,
      "loss": 0.0653,
      "theoretical_loss": 3.397743216448098,
      "tokens_seen": 2492989440
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002469106082490772,
      "loss": 0.0669,
      "theoretical_loss": 3.3977149421867012,
      "tokens_seen": 2493251584
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002468303643074948,
      "loss": 0.0684,
      "theoretical_loss": 3.397686671730227,
      "tokens_seen": 2493513728
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002467501203659124,
      "loss": 0.0656,
      "theoretical_loss": 3.3976584050777636,
      "tokens_seen": 2493775872
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024666987642432995,
      "loss": 0.0643,
      "theoretical_loss": 3.397630142228399,
      "tokens_seen": 2494038016
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024658963248274757,
      "loss": 0.0657,
      "theoretical_loss": 3.3976018831812222,
      "tokens_seen": 2494300160
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002465093885411651,
      "loss": 0.0661,
      "theoretical_loss": 3.3975736279353224,
      "tokens_seen": 2494562304
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024642914459958274,
      "loss": 0.0628,
      "theoretical_loss": 3.3975453764897883,
      "tokens_seen": 2494824448
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024634890065800035,
      "loss": 0.0651,
      "theoretical_loss": 3.3975171288437096,
      "tokens_seen": 2495086592
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002462686567164179,
      "loss": 0.0644,
      "theoretical_loss": 3.397488884996177,
      "tokens_seen": 2495348736
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002461884127748355,
      "loss": 0.0658,
      "theoretical_loss": 3.3974606449462796,
      "tokens_seen": 2495610880
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002461081688332531,
      "loss": 0.0655,
      "theoretical_loss": 3.3974324086931085,
      "tokens_seen": 2495873024
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002460279248916707,
      "loss": 0.0662,
      "theoretical_loss": 3.3974041762357547,
      "tokens_seen": 2496135168
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024594768095008825,
      "loss": 0.0658,
      "theoretical_loss": 3.397375947573309,
      "tokens_seen": 2496397312
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024586743700850586,
      "loss": 0.0647,
      "theoretical_loss": 3.397347722704863,
      "tokens_seen": 2496659456
    },
    {
      "epoch": 0.76,
      "objective/train/advantage_avg": 0.0009059284930117428,
      "objective/train/docs_used": 908925,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2855145931243896,
      "objective/train/original_loss": 1.2855145931243896,
      "objective/train/theoretical_loss": 3.397319501629509,
      "objective/train/tokens_used": 2517381600,
      "objective/train/value_avg": -0.00823974609375,
      "objective/train/value_loss": 0.00019895529840141535,
      "objective/train/value_max": -2.86102294921875e-05,
      "objective/train/value_min": -0.343994140625,
      "objective/train/value_reward_corr": 0.6927096300481003,
      "objective/train/value_std": 0.01464080810546875,
      "objective/train/weight_avg": 1.000998854637146,
      "objective/train/weighted_lm_loss": 1.2875512838363647,
      "objective/train/weights_max": 1.373934268951416,
      "objective/train/weights_min": 0.3924909830093384,
      "theoretical_loss": 3.397319501629509,
      "tokens_seen": 2496921600
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002457871930669235,
      "loss": 0.0664,
      "theoretical_loss": 3.397319501629509,
      "tokens_seen": 2496921600
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024570694912534103,
      "loss": 0.0645,
      "theoretical_loss": 3.397291284346338,
      "tokens_seen": 2497183744
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024562670518375865,
      "loss": 0.0652,
      "theoretical_loss": 3.397263070854444,
      "tokens_seen": 2497445888
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002455464612421762,
      "loss": 0.066,
      "theoretical_loss": 3.397234861152918,
      "tokens_seen": 2497708032
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002454662173005938,
      "loss": 0.066,
      "theoretical_loss": 3.397206655240854,
      "tokens_seen": 2497970176
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002453859733590114,
      "loss": 0.0639,
      "theoretical_loss": 3.3971784531173457,
      "tokens_seen": 2498232320
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.000245305729417429,
      "loss": 0.0638,
      "theoretical_loss": 3.397150254781486,
      "tokens_seen": 2498494464
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002452254854758466,
      "loss": 0.0672,
      "theoretical_loss": 3.3971220602323693,
      "tokens_seen": 2498756608
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024514524153426416,
      "loss": 0.0666,
      "theoretical_loss": 3.3970938694690895,
      "tokens_seen": 2499018752
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024506499759268177,
      "loss": 0.0636,
      "theoretical_loss": 3.3970656824907417,
      "tokens_seen": 2499280896
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024498475365109933,
      "loss": 0.0649,
      "theoretical_loss": 3.3970374992964207,
      "tokens_seen": 2499543040
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024490450970951694,
      "loss": 0.0664,
      "theoretical_loss": 3.3970093198852216,
      "tokens_seen": 2499805184
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024482426576793456,
      "loss": 0.0669,
      "theoretical_loss": 3.3969811442562396,
      "tokens_seen": 2500067328
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002447440218263521,
      "loss": 0.0628,
      "theoretical_loss": 3.396952972408571,
      "tokens_seen": 2500329472
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002446637778847697,
      "loss": 0.0656,
      "theoretical_loss": 3.3969248043413125,
      "tokens_seen": 2500591616
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002445835339431873,
      "loss": 0.0642,
      "theoretical_loss": 3.3968966400535594,
      "tokens_seen": 2500853760
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002445032900016049,
      "loss": 0.0647,
      "theoretical_loss": 3.396868479544409,
      "tokens_seen": 2501115904
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024442304606002246,
      "loss": 0.0629,
      "theoretical_loss": 3.3968403228129587,
      "tokens_seen": 2501378048
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024434280211844007,
      "loss": 0.0613,
      "theoretical_loss": 3.396812169858306,
      "tokens_seen": 2501640192
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002442625581768577,
      "loss": 0.0648,
      "theoretical_loss": 3.396784020679547,
      "tokens_seen": 2501902336
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024418231423527524,
      "loss": 0.0661,
      "theoretical_loss": 3.3967558752757823,
      "tokens_seen": 2502164480
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024410207029369282,
      "loss": 0.0682,
      "theoretical_loss": 3.396727733646108,
      "tokens_seen": 2502426624
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024402182635211044,
      "loss": 0.0652,
      "theoretical_loss": 3.396699595789624,
      "tokens_seen": 2502688768
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024394158241052802,
      "loss": 0.0667,
      "theoretical_loss": 3.396671461705429,
      "tokens_seen": 2502950912
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024386133846894558,
      "loss": 0.0652,
      "theoretical_loss": 3.396643331392622,
      "tokens_seen": 2503213056
    },
    {
      "epoch": 0.76,
      "objective/train/advantage_avg": 3.1022223993204534e-05,
      "objective/train/docs_used": 911421,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.350574254989624,
      "objective/train/original_loss": 1.350574254989624,
      "objective/train/theoretical_loss": 3.3966152048503027,
      "objective/train/tokens_used": 2523935200,
      "objective/train/value_avg": -0.0098724365234375,
      "objective/train/value_loss": 0.00019183217955287546,
      "objective/train/value_max": -5.435943603515625e-05,
      "objective/train/value_min": -0.349365234375,
      "objective/train/value_reward_corr": 0.8088936830272246,
      "objective/train/value_std": 0.019378662109375,
      "objective/train/weight_avg": 1.0001190900802612,
      "objective/train/weighted_lm_loss": 1.3499020338058472,
      "objective/train/weights_max": 1.1969894170761108,
      "objective/train/weights_min": 0.3697309195995331,
      "theoretical_loss": 3.3966152048503027,
      "tokens_seen": 2503475200
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002437810945273632,
      "loss": 0.0666,
      "theoretical_loss": 3.3966152048503027,
      "tokens_seen": 2503475200
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024370085058578078,
      "loss": 0.0629,
      "theoretical_loss": 3.396587082077571,
      "tokens_seen": 2503737344
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024362060664419836,
      "loss": 0.0672,
      "theoretical_loss": 3.3965589630735273,
      "tokens_seen": 2503999488
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024354036270261595,
      "loss": 0.0622,
      "theoretical_loss": 3.3965308478372718,
      "tokens_seen": 2504261632
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024346011876103356,
      "loss": 0.0652,
      "theoretical_loss": 3.3965027363679052,
      "tokens_seen": 2504523776
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024337987481945115,
      "loss": 0.0652,
      "theoretical_loss": 3.3964746286645293,
      "tokens_seen": 2504785920
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002432996308778687,
      "loss": 0.0631,
      "theoretical_loss": 3.3964465247262448,
      "tokens_seen": 2505048064
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024321938693628632,
      "loss": 0.0649,
      "theoretical_loss": 3.3964184245521536,
      "tokens_seen": 2505310208
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002431391429947039,
      "loss": 0.0674,
      "theoretical_loss": 3.396390328141358,
      "tokens_seen": 2505572352
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024305889905312152,
      "loss": 0.0655,
      "theoretical_loss": 3.3963622354929597,
      "tokens_seen": 2505834496
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024297865511153908,
      "loss": 0.0653,
      "theoretical_loss": 3.396334146606062,
      "tokens_seen": 2506096640
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024289841116995666,
      "loss": 0.0633,
      "theoretical_loss": 3.396306061479768,
      "tokens_seen": 2506358784
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024281816722837427,
      "loss": 0.0671,
      "theoretical_loss": 3.3962779801131804,
      "tokens_seen": 2506620928
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024273792328679183,
      "loss": 0.0667,
      "theoretical_loss": 3.3962499025054034,
      "tokens_seen": 2506883072
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024265767934520945,
      "loss": 0.0637,
      "theoretical_loss": 3.3962218286555403,
      "tokens_seen": 2507145216
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024257743540362703,
      "loss": 0.0624,
      "theoretical_loss": 3.3961937585626956,
      "tokens_seen": 2507407360
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024249719146204464,
      "loss": 0.0655,
      "theoretical_loss": 3.3961656922259738,
      "tokens_seen": 2507669504
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002424169475204622,
      "loss": 0.0664,
      "theoretical_loss": 3.3961376296444796,
      "tokens_seen": 2507931648
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002423367035788798,
      "loss": 0.0665,
      "theoretical_loss": 3.396109570817318,
      "tokens_seen": 2508193792
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002422564596372974,
      "loss": 0.0655,
      "theoretical_loss": 3.3960815157435946,
      "tokens_seen": 2508455936
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024217621569571499,
      "loss": 0.0652,
      "theoretical_loss": 3.3960534644224154,
      "tokens_seen": 2508718080
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024209597175413257,
      "loss": 0.0671,
      "theoretical_loss": 3.3960254168528863,
      "tokens_seen": 2508980224
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024201572781255016,
      "loss": 0.0622,
      "theoretical_loss": 3.395997373034113,
      "tokens_seen": 2509242368
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024193548387096774,
      "loss": 0.0685,
      "theoretical_loss": 3.395969332965203,
      "tokens_seen": 2509504512
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024185523992938533,
      "loss": 0.0648,
      "theoretical_loss": 3.3959412966452627,
      "tokens_seen": 2509766656
    },
    {
      "epoch": 0.76,
      "objective/train/advantage_avg": -0.0003518318699207157,
      "objective/train/docs_used": 913305,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.370742917060852,
      "objective/train/original_loss": 1.370742917060852,
      "objective/train/theoretical_loss": 3.3959132640733998,
      "objective/train/tokens_used": 2530488800,
      "objective/train/value_avg": -0.00812530517578125,
      "objective/train/value_loss": 0.0003890729567501694,
      "objective/train/value_max": -1.722574234008789e-05,
      "objective/train/value_min": -0.91796875,
      "objective/train/value_reward_corr": 0.7049164524146272,
      "objective/train/value_std": 0.0178070068359375,
      "objective/train/weight_avg": 0.9998188614845276,
      "objective/train/weighted_lm_loss": 1.3701258897781372,
      "objective/train/weights_max": 1.7972010374069214,
      "objective/train/weights_min": 0.36976051330566406,
      "theoretical_loss": 3.3959132640733998,
      "tokens_seen": 2510028800
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002417749959878029,
      "loss": 0.0657,
      "theoretical_loss": 3.3959132640733998,
      "tokens_seen": 2510028800
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024169475204622053,
      "loss": 0.0669,
      "theoretical_loss": 3.395885235248721,
      "tokens_seen": 2510290944
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002416145081046381,
      "loss": 0.0672,
      "theoretical_loss": 3.3958572101703353,
      "tokens_seen": 2510553088
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002415342641630557,
      "loss": 0.0627,
      "theoretical_loss": 3.3958291888373506,
      "tokens_seen": 2510815232
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024145402022147328,
      "loss": 0.0654,
      "theoretical_loss": 3.3958011712488747,
      "tokens_seen": 2511077376
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024137377627989087,
      "loss": 0.0668,
      "theoretical_loss": 3.395773157404017,
      "tokens_seen": 2511339520
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024129353233830848,
      "loss": 0.069,
      "theoretical_loss": 3.395745147301887,
      "tokens_seen": 2511601664
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024121328839672604,
      "loss": 0.0661,
      "theoretical_loss": 3.395717140941593,
      "tokens_seen": 2511863808
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024113304445514365,
      "loss": 0.0668,
      "theoretical_loss": 3.395689138322245,
      "tokens_seen": 2512125952
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024105280051356124,
      "loss": 0.0655,
      "theoretical_loss": 3.3956611394429537,
      "tokens_seen": 2512388096
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002409725565719788,
      "loss": 0.0639,
      "theoretical_loss": 3.395633144302829,
      "tokens_seen": 2512650240
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002408923126303964,
      "loss": 0.0655,
      "theoretical_loss": 3.3956051529009814,
      "tokens_seen": 2512912384
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.000240812068688814,
      "loss": 0.0654,
      "theoretical_loss": 3.3955771652365216,
      "tokens_seen": 2513174528
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002407318247472316,
      "loss": 0.0667,
      "theoretical_loss": 3.395549181308562,
      "tokens_seen": 2513436672
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024065158080564916,
      "loss": 0.0682,
      "theoretical_loss": 3.3955212011162126,
      "tokens_seen": 2513698816
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024057133686406678,
      "loss": 0.0655,
      "theoretical_loss": 3.395493224658586,
      "tokens_seen": 2513960960
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024049109292248436,
      "loss": 0.0627,
      "theoretical_loss": 3.395465251934794,
      "tokens_seen": 2514223104
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024041084898090195,
      "loss": 0.0636,
      "theoretical_loss": 3.39543728294395,
      "tokens_seen": 2514485248
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024033060503931953,
      "loss": 0.0631,
      "theoretical_loss": 3.395409317685165,
      "tokens_seen": 2514747392
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024025036109773712,
      "loss": 0.0661,
      "theoretical_loss": 3.395381356157554,
      "tokens_seen": 2515009536
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024017011715615473,
      "loss": 0.0676,
      "theoretical_loss": 3.3953533983602293,
      "tokens_seen": 2515271680
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024008987321457232,
      "loss": 0.066,
      "theoretical_loss": 3.3953254442923044,
      "tokens_seen": 2515533824
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00024000962927298988,
      "loss": 0.0672,
      "theoretical_loss": 3.3952974939528935,
      "tokens_seen": 2515795968
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002399293853314075,
      "loss": 0.0641,
      "theoretical_loss": 3.395269547341111,
      "tokens_seen": 2516058112
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00023984914138982507,
      "loss": 0.0646,
      "theoretical_loss": 3.3952416044560714,
      "tokens_seen": 2516320256
    },
    {
      "epoch": 0.76,
      "objective/train/advantage_avg": -0.0005290715489536524,
      "objective/train/docs_used": 916228,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3841131925582886,
      "objective/train/original_loss": 1.384113073348999,
      "objective/train/theoretical_loss": 3.3952136652968896,
      "objective/train/tokens_used": 2537042400,
      "objective/train/value_avg": -0.0087890625,
      "objective/train/value_loss": 0.0003163471119478345,
      "objective/train/value_max": -3.5643577575683594e-05,
      "objective/train/value_min": -0.72412109375,
      "objective/train/value_reward_corr": 0.6784622902242732,
      "objective/train/value_std": 0.0157012939453125,
      "objective/train/weight_avg": 0.9996117353439331,
      "objective/train/weighted_lm_loss": 1.383963942527771,
      "objective/train/weights_max": 1.3841378688812256,
      "objective/train/weights_min": 0.3704989552497864,
      "theoretical_loss": 3.3952136652968896,
      "tokens_seen": 2516582400
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00023976889744824266,
      "loss": 0.0687,
      "theoretical_loss": 3.3952136652968896,
      "tokens_seen": 2516582400
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00023968865350666024,
      "loss": 0.0652,
      "theoretical_loss": 3.3951857298626806,
      "tokens_seen": 2516844544
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00023960840956507786,
      "loss": 0.0649,
      "theoretical_loss": 3.3951577981525602,
      "tokens_seen": 2517106688
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00023952816562349544,
      "loss": 0.0657,
      "theoretical_loss": 3.395129870165644,
      "tokens_seen": 2517368832
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.000239447921681913,
      "loss": 0.0666,
      "theoretical_loss": 3.3951019459010476,
      "tokens_seen": 2517630976
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002393676777403306,
      "loss": 0.0636,
      "theoretical_loss": 3.3950740253578875,
      "tokens_seen": 2517893120
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002392874337987482,
      "loss": 0.0651,
      "theoretical_loss": 3.3950461085352814,
      "tokens_seen": 2518155264
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002392071898571658,
      "loss": 0.0659,
      "theoretical_loss": 3.3950181954323453,
      "tokens_seen": 2518417408
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00023912694591558337,
      "loss": 0.0654,
      "theoretical_loss": 3.3949902860481966,
      "tokens_seen": 2518679552
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00023904670197400096,
      "loss": 0.0655,
      "theoretical_loss": 3.394962380381953,
      "tokens_seen": 2518941696
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00023896645803241857,
      "loss": 0.0659,
      "theoretical_loss": 3.3949344784327327,
      "tokens_seen": 2519203840
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00023888621409083613,
      "loss": 0.0663,
      "theoretical_loss": 3.3949065801996534,
      "tokens_seen": 2519465984
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00023880597014925374,
      "loss": 0.0631,
      "theoretical_loss": 3.3948786856818334,
      "tokens_seen": 2519728128
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00023872572620767132,
      "loss": 0.0645,
      "theoretical_loss": 3.3948507948783924,
      "tokens_seen": 2519990272
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00023864548226608894,
      "loss": 0.0634,
      "theoretical_loss": 3.3948229077884484,
      "tokens_seen": 2520252416
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002385652383245065,
      "loss": 0.0659,
      "theoretical_loss": 3.394795024411122,
      "tokens_seen": 2520514560
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00023848499438292408,
      "loss": 0.0665,
      "theoretical_loss": 3.394767144745532,
      "tokens_seen": 2520776704
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002384047504413417,
      "loss": 0.0691,
      "theoretical_loss": 3.394739268790798,
      "tokens_seen": 2521038848
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00023832450649975928,
      "loss": 0.0644,
      "theoretical_loss": 3.3947113965460414,
      "tokens_seen": 2521300992
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00023824426255817686,
      "loss": 0.0674,
      "theoretical_loss": 3.3946835280103826,
      "tokens_seen": 2521563136
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00023816401861659445,
      "loss": 0.066,
      "theoretical_loss": 3.3946556631829417,
      "tokens_seen": 2521825280
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00023808377467501204,
      "loss": 0.0668,
      "theoretical_loss": 3.3946278020628404,
      "tokens_seen": 2522087424
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00023800353073342962,
      "loss": 0.0641,
      "theoretical_loss": 3.3945999446492,
      "tokens_seen": 2522349568
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002379232867918472,
      "loss": 0.0652,
      "theoretical_loss": 3.3945720909411428,
      "tokens_seen": 2522611712
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00023784304285026482,
      "loss": 0.0654,
      "theoretical_loss": 3.3945442409377904,
      "tokens_seen": 2522873856
    },
    {
      "epoch": 0.76,
      "objective/train/advantage_avg": 0.0008710318361409009,
      "objective/train/docs_used": 918647,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.381752371788025,
      "objective/train/original_loss": 1.3817524909973145,
      "objective/train/theoretical_loss": 3.394516394638265,
      "objective/train/tokens_used": 2543596000,
      "objective/train/value_avg": -0.0074615478515625,
      "objective/train/value_loss": 0.00024307436251547188,
      "objective/train/value_max": -4.13060188293457e-05,
      "objective/train/value_min": -0.70166015625,
      "objective/train/value_reward_corr": 0.6666992631073855,
      "objective/train/value_std": 0.013763427734375,
      "objective/train/weight_avg": 1.0009773969650269,
      "objective/train/weighted_lm_loss": 1.382368803024292,
      "objective/train/weights_max": 1.411948561668396,
      "objective/train/weights_min": 0.3690214455127716,
      "theoretical_loss": 3.394516394638265,
      "tokens_seen": 2523136000
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002377627989086824,
      "loss": 0.0673,
      "theoretical_loss": 3.394516394638265,
      "tokens_seen": 2523136000
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.0002376825549671,
      "loss": 0.0643,
      "theoretical_loss": 3.3944885520416896,
      "tokens_seen": 2523398144
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00023760231102551758,
      "loss": 0.0682,
      "theoretical_loss": 3.3944607131471876,
      "tokens_seen": 2523660288
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00023752206708393516,
      "loss": 0.0677,
      "theoretical_loss": 3.3944328779538813,
      "tokens_seen": 2523922432
    },
    {
      "epoch": 0.76,
      "learning_rate": 0.00023744182314235277,
      "loss": 0.0639,
      "theoretical_loss": 3.3944050464608955,
      "tokens_seen": 2524184576
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023736157920077033,
      "loss": 0.0653,
      "theoretical_loss": 3.394377218667353,
      "tokens_seen": 2524446720
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023728133525918795,
      "loss": 0.0628,
      "theoretical_loss": 3.3943493945723784,
      "tokens_seen": 2524708864
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023720109131760553,
      "loss": 0.0677,
      "theoretical_loss": 3.394321574175096,
      "tokens_seen": 2524971008
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.0002371208473760231,
      "loss": 0.0655,
      "theoretical_loss": 3.394293757474631,
      "tokens_seen": 2525233152
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.0002370406034344407,
      "loss": 0.0682,
      "theoretical_loss": 3.394265944470108,
      "tokens_seen": 2525495296
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.0002369603594928583,
      "loss": 0.0629,
      "theoretical_loss": 3.3942381351606525,
      "tokens_seen": 2525757440
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.0002368801155512759,
      "loss": 0.067,
      "theoretical_loss": 3.3942103295453903,
      "tokens_seen": 2526019584
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023679987160969346,
      "loss": 0.0637,
      "theoretical_loss": 3.394182527623448,
      "tokens_seen": 2526281728
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023671962766811107,
      "loss": 0.0677,
      "theoretical_loss": 3.39415472939395,
      "tokens_seen": 2526543872
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023663938372652866,
      "loss": 0.0646,
      "theoretical_loss": 3.3941269348560246,
      "tokens_seen": 2526806016
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023655913978494624,
      "loss": 0.0663,
      "theoretical_loss": 3.3940991440087984,
      "tokens_seen": 2527068160
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023647889584336383,
      "loss": 0.0682,
      "theoretical_loss": 3.3940713568513976,
      "tokens_seen": 2527330304
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.0002363986519017814,
      "loss": 0.0647,
      "theoretical_loss": 3.394043573382951,
      "tokens_seen": 2527592448
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023631840796019903,
      "loss": 0.0656,
      "theoretical_loss": 3.394015793602585,
      "tokens_seen": 2527854592
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023623816401861658,
      "loss": 0.0654,
      "theoretical_loss": 3.393988017509429,
      "tokens_seen": 2528116736
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.0002361579200770342,
      "loss": 0.0665,
      "theoretical_loss": 3.3939602451026096,
      "tokens_seen": 2528378880
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023607767613545178,
      "loss": 0.0646,
      "theoretical_loss": 3.393932476381257,
      "tokens_seen": 2528641024
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023599743219386937,
      "loss": 0.066,
      "theoretical_loss": 3.3939047113445,
      "tokens_seen": 2528903168
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023591718825228695,
      "loss": 0.0658,
      "theoretical_loss": 3.393876949991467,
      "tokens_seen": 2529165312
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023583694431070454,
      "loss": 0.0658,
      "theoretical_loss": 3.393849192321288,
      "tokens_seen": 2529427456
    },
    {
      "epoch": 0.77,
      "objective/train/advantage_avg": 8.658033038955182e-06,
      "objective/train/docs_used": 920541,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.329050064086914,
      "objective/train/original_loss": 1.3290499448776245,
      "objective/train/theoretical_loss": 3.3938214383330925,
      "objective/train/tokens_used": 2550149600,
      "objective/train/value_avg": -0.00835418701171875,
      "objective/train/value_loss": 0.0001360275491606444,
      "objective/train/value_max": -2.467632293701172e-05,
      "objective/train/value_min": -0.390869140625,
      "objective/train/value_reward_corr": 0.8806241136256157,
      "objective/train/value_std": 0.0211029052734375,
      "objective/train/weight_avg": 1.000072956085205,
      "objective/train/weighted_lm_loss": 1.3284049034118652,
      "objective/train/weights_max": 1.2012343406677246,
      "objective/train/weights_min": 0.3818925619125366,
      "theoretical_loss": 3.3938214383330925,
      "tokens_seen": 2529689600
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023575670036912215,
      "loss": 0.0678,
      "theoretical_loss": 3.3938214383330925,
      "tokens_seen": 2529689600
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023567645642753974,
      "loss": 0.0664,
      "theoretical_loss": 3.3937936880260113,
      "tokens_seen": 2529951744
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.0002355962124859573,
      "loss": 0.0674,
      "theoretical_loss": 3.3937659413991743,
      "tokens_seen": 2530213888
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.0002355159685443749,
      "loss": 0.066,
      "theoretical_loss": 3.3937381984517123,
      "tokens_seen": 2530476032
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.0002354357246027925,
      "loss": 0.0634,
      "theoretical_loss": 3.3937104591827563,
      "tokens_seen": 2530738176
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.0002353554806612101,
      "loss": 0.0658,
      "theoretical_loss": 3.3936827235914375,
      "tokens_seen": 2531000320
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023527523671962766,
      "loss": 0.0665,
      "theoretical_loss": 3.3936549916768874,
      "tokens_seen": 2531262464
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023519499277804528,
      "loss": 0.0683,
      "theoretical_loss": 3.3936272634382387,
      "tokens_seen": 2531524608
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023511474883646286,
      "loss": 0.0664,
      "theoretical_loss": 3.3935995388746227,
      "tokens_seen": 2531786752
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023503450489488042,
      "loss": 0.0634,
      "theoretical_loss": 3.393571817985172,
      "tokens_seen": 2532048896
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023495426095329803,
      "loss": 0.0674,
      "theoretical_loss": 3.3935441007690197,
      "tokens_seen": 2532311040
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023487401701171562,
      "loss": 0.0655,
      "theoretical_loss": 3.393516387225299,
      "tokens_seen": 2532573184
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023479377307013323,
      "loss": 0.0654,
      "theoretical_loss": 3.393488677353142,
      "tokens_seen": 2532835328
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.0002347135291285508,
      "loss": 0.0655,
      "theoretical_loss": 3.3934609711516845,
      "tokens_seen": 2533097472
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023463328518696838,
      "loss": 0.0639,
      "theoretical_loss": 3.3934332686200586,
      "tokens_seen": 2533359616
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.000234553041245386,
      "loss": 0.0665,
      "theoretical_loss": 3.3934055697573995,
      "tokens_seen": 2533621760
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023447279730380357,
      "loss": 0.0664,
      "theoretical_loss": 3.393377874562841,
      "tokens_seen": 2533883904
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023439255336222116,
      "loss": 0.0656,
      "theoretical_loss": 3.393350183035519,
      "tokens_seen": 2534146048
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023431230942063874,
      "loss": 0.0644,
      "theoretical_loss": 3.3933224951745675,
      "tokens_seen": 2534408192
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023423206547905636,
      "loss": 0.0656,
      "theoretical_loss": 3.3932948109791226,
      "tokens_seen": 2534670336
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023415182153747392,
      "loss": 0.0667,
      "theoretical_loss": 3.39326713044832,
      "tokens_seen": 2534932480
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.0002340715775958915,
      "loss": 0.0668,
      "theoretical_loss": 3.393239453581295,
      "tokens_seen": 2535194624
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.0002339913336543091,
      "loss": 0.0671,
      "theoretical_loss": 3.393211780377185,
      "tokens_seen": 2535456768
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.0002339110897127267,
      "loss": 0.0645,
      "theoretical_loss": 3.3931841108351257,
      "tokens_seen": 2535718912
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023383084577114428,
      "loss": 0.0646,
      "theoretical_loss": 3.393156444954255,
      "tokens_seen": 2535981056
    },
    {
      "epoch": 0.77,
      "objective/train/advantage_avg": 0.0009138497989624739,
      "objective/train/docs_used": 922895,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4049909114837646,
      "objective/train/original_loss": 1.4049906730651855,
      "objective/train/theoretical_loss": 3.393128782733709,
      "objective/train/tokens_used": 2556703200,
      "objective/train/value_avg": -0.006580352783203125,
      "objective/train/value_loss": 0.00024076194677036256,
      "objective/train/value_max": -2.5272369384765625e-05,
      "objective/train/value_min": -0.70556640625,
      "objective/train/value_reward_corr": 0.7010251590697951,
      "objective/train/value_std": 0.01537322998046875,
      "objective/train/weight_avg": 1.0010218620300293,
      "objective/train/weighted_lm_loss": 1.406140923500061,
      "objective/train/weights_max": 1.4373383522033691,
      "objective/train/weights_min": 0.36942780017852783,
      "theoretical_loss": 3.393128782733709,
      "tokens_seen": 2536243200
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023375060182956187,
      "loss": 0.0671,
      "theoretical_loss": 3.393128782733709,
      "tokens_seen": 2536243200
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023367035788797946,
      "loss": 0.0676,
      "theoretical_loss": 3.3931011241726248,
      "tokens_seen": 2536505344
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023359011394639707,
      "loss": 0.0642,
      "theoretical_loss": 3.393073469270142,
      "tokens_seen": 2536767488
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023350987000481463,
      "loss": 0.0652,
      "theoretical_loss": 3.393045818025397,
      "tokens_seen": 2537029632
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023342962606323224,
      "loss": 0.0663,
      "theoretical_loss": 3.393018170437529,
      "tokens_seen": 2537291776
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023334938212164982,
      "loss": 0.067,
      "theoretical_loss": 3.392990526505676,
      "tokens_seen": 2537553920
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.0002332691381800674,
      "loss": 0.0658,
      "theoretical_loss": 3.3929628862289776,
      "tokens_seen": 2537816064
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.000233188894238485,
      "loss": 0.0661,
      "theoretical_loss": 3.392935249606573,
      "tokens_seen": 2538078208
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023310865029690258,
      "loss": 0.0674,
      "theoretical_loss": 3.3929076166376007,
      "tokens_seen": 2538340352
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.0002330284063553202,
      "loss": 0.0633,
      "theoretical_loss": 3.3928799873212014,
      "tokens_seen": 2538602496
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023294816241373775,
      "loss": 0.0651,
      "theoretical_loss": 3.3928523616565154,
      "tokens_seen": 2538864640
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023286791847215536,
      "loss": 0.0637,
      "theoretical_loss": 3.3928247396426827,
      "tokens_seen": 2539126784
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023278767453057295,
      "loss": 0.063,
      "theoretical_loss": 3.3927971212788437,
      "tokens_seen": 2539388928
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023270743058899054,
      "loss": 0.0658,
      "theoretical_loss": 3.3927695065641394,
      "tokens_seen": 2539651072
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023262718664740812,
      "loss": 0.0644,
      "theoretical_loss": 3.392741895497712,
      "tokens_seen": 2539913216
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.0002325469427058257,
      "loss": 0.0628,
      "theoretical_loss": 3.392714288078702,
      "tokens_seen": 2540175360
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023246669876424332,
      "loss": 0.0648,
      "theoretical_loss": 3.3926866843062515,
      "tokens_seen": 2540437504
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023238645482266088,
      "loss": 0.0642,
      "theoretical_loss": 3.392659084179503,
      "tokens_seen": 2540699648
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.0002323062108810785,
      "loss": 0.0632,
      "theoretical_loss": 3.3926314876975985,
      "tokens_seen": 2540961792
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023222596693949608,
      "loss": 0.0643,
      "theoretical_loss": 3.3926038948596813,
      "tokens_seen": 2541223936
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023214572299791366,
      "loss": 0.0649,
      "theoretical_loss": 3.3925763056648934,
      "tokens_seen": 2541486080
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023206547905633125,
      "loss": 0.0649,
      "theoretical_loss": 3.3925487201123787,
      "tokens_seen": 2541748224
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023198523511474883,
      "loss": 0.0671,
      "theoretical_loss": 3.392521138201281,
      "tokens_seen": 2542010368
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023190499117316645,
      "loss": 0.0651,
      "theoretical_loss": 3.392493559930744,
      "tokens_seen": 2542272512
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023182474723158403,
      "loss": 0.0668,
      "theoretical_loss": 3.3924659852999115,
      "tokens_seen": 2542534656
    },
    {
      "epoch": 0.77,
      "objective/train/advantage_avg": -0.00013724910968448967,
      "objective/train/docs_used": 925229,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.207066535949707,
      "objective/train/original_loss": 1.207066535949707,
      "objective/train/theoretical_loss": 3.3924384143079283,
      "objective/train/tokens_used": 2563256800,
      "objective/train/value_avg": -0.0070953369140625,
      "objective/train/value_loss": 0.00013547865091823041,
      "objective/train/value_max": -3.594160079956055e-05,
      "objective/train/value_min": -0.2379150390625,
      "objective/train/value_reward_corr": 0.7183617630226486,
      "objective/train/value_std": 0.0111541748046875,
      "objective/train/weight_avg": 0.9999263286590576,
      "objective/train/weighted_lm_loss": 1.2072526216506958,
      "objective/train/weights_max": 1.1096680164337158,
      "objective/train/weights_min": 0.3745289444923401,
      "theoretical_loss": 3.3924384143079283,
      "tokens_seen": 2542796800
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.0002317445032900016,
      "loss": 0.0647,
      "theoretical_loss": 3.3924384143079283,
      "tokens_seen": 2542796800
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.0002316642593484192,
      "loss": 0.0685,
      "theoretical_loss": 3.392410846953939,
      "tokens_seen": 2543058944
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.0002315840154068368,
      "loss": 0.0663,
      "theoretical_loss": 3.3923832832370886,
      "tokens_seen": 2543321088
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023150377146525437,
      "loss": 0.0672,
      "theoretical_loss": 3.392355723156523,
      "tokens_seen": 2543583232
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023142352752367196,
      "loss": 0.0663,
      "theoretical_loss": 3.392328166711387,
      "tokens_seen": 2543845376
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023134328358208957,
      "loss": 0.0652,
      "theoretical_loss": 3.3923006139008267,
      "tokens_seen": 2544107520
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023126303964050716,
      "loss": 0.0642,
      "theoretical_loss": 3.3922730647239887,
      "tokens_seen": 2544369664
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023118279569892471,
      "loss": 0.0633,
      "theoretical_loss": 3.392245519180019,
      "tokens_seen": 2544631808
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023110255175734233,
      "loss": 0.0643,
      "theoretical_loss": 3.3922179772680643,
      "tokens_seen": 2544893952
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.0002310223078157599,
      "loss": 0.0658,
      "theoretical_loss": 3.392190438987272,
      "tokens_seen": 2545156096
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023094206387417753,
      "loss": 0.0667,
      "theoretical_loss": 3.39216290433679,
      "tokens_seen": 2545418240
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023086181993259508,
      "loss": 0.0642,
      "theoretical_loss": 3.392135373315764,
      "tokens_seen": 2545680384
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023078157599101267,
      "loss": 0.0636,
      "theoretical_loss": 3.392107845923344,
      "tokens_seen": 2545942528
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023070133204943028,
      "loss": 0.0638,
      "theoretical_loss": 3.3920803221586775,
      "tokens_seen": 2546204672
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023062108810784784,
      "loss": 0.0637,
      "theoretical_loss": 3.3920528020209124,
      "tokens_seen": 2546466816
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023054084416626545,
      "loss": 0.0616,
      "theoretical_loss": 3.392025285509198,
      "tokens_seen": 2546728960
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023046060022468304,
      "loss": 0.0633,
      "theoretical_loss": 3.391997772622683,
      "tokens_seen": 2546991104
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00023038035628310065,
      "loss": 0.062,
      "theoretical_loss": 3.3919702633605175,
      "tokens_seen": 2547253248
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.0002303001123415182,
      "loss": 0.0619,
      "theoretical_loss": 3.391942757721851,
      "tokens_seen": 2547515392
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.0002302198683999358,
      "loss": 0.065,
      "theoretical_loss": 3.3919152557058325,
      "tokens_seen": 2547777536
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.0002301396244583534,
      "loss": 0.0637,
      "theoretical_loss": 3.3918877573116126,
      "tokens_seen": 2548039680
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.000230059380516771,
      "loss": 0.066,
      "theoretical_loss": 3.3918602625383425,
      "tokens_seen": 2548301824
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00022997913657518858,
      "loss": 0.0646,
      "theoretical_loss": 3.391832771385172,
      "tokens_seen": 2548563968
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00022989889263360616,
      "loss": 0.0634,
      "theoretical_loss": 3.3918052838512533,
      "tokens_seen": 2548826112
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00022981864869202375,
      "loss": 0.0642,
      "theoretical_loss": 3.3917777999357366,
      "tokens_seen": 2549088256
    },
    {
      "epoch": 0.77,
      "objective/train/advantage_avg": 0.0002214000269304961,
      "objective/train/docs_used": 927515,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.329919695854187,
      "objective/train/original_loss": 1.3299193382263184,
      "objective/train/theoretical_loss": 3.3917503196377745,
      "objective/train/tokens_used": 2569810400,
      "objective/train/value_avg": -0.008148193359375,
      "objective/train/value_loss": 0.00020550192857626826,
      "objective/train/value_max": -2.2292137145996094e-05,
      "objective/train/value_min": -0.71875,
      "objective/train/value_reward_corr": 0.7468589864842972,
      "objective/train/value_std": 0.01580810546875,
      "objective/train/weight_avg": 1.0003150701522827,
      "objective/train/weighted_lm_loss": 1.3300237655639648,
      "objective/train/weights_max": 1.4661985635757446,
      "objective/train/weights_min": 0.37047773599624634,
      "theoretical_loss": 3.3917503196377745,
      "tokens_seen": 2549350400
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00022973840475044136,
      "loss": 0.066,
      "theoretical_loss": 3.3917503196377745,
      "tokens_seen": 2549350400
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00022965816080885892,
      "loss": 0.0647,
      "theoretical_loss": 3.391722842956518,
      "tokens_seen": 2549612544
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00022957791686727653,
      "loss": 0.0627,
      "theoretical_loss": 3.3916953698911203,
      "tokens_seen": 2549874688
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00022949767292569412,
      "loss": 0.064,
      "theoretical_loss": 3.391667900440733,
      "tokens_seen": 2550136832
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.0002294174289841117,
      "loss": 0.0646,
      "theoretical_loss": 3.3916404346045095,
      "tokens_seen": 2550398976
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.0002293371850425293,
      "loss": 0.0642,
      "theoretical_loss": 3.3916129723816026,
      "tokens_seen": 2550661120
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00022925694110094688,
      "loss": 0.0654,
      "theoretical_loss": 3.3915855137711657,
      "tokens_seen": 2550923264
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.0002291766971593645,
      "loss": 0.0618,
      "theoretical_loss": 3.3915580587723526,
      "tokens_seen": 2551185408
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00022909645321778205,
      "loss": 0.0622,
      "theoretical_loss": 3.391530607384317,
      "tokens_seen": 2551447552
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00022901620927619966,
      "loss": 0.0632,
      "theoretical_loss": 3.3915031596062133,
      "tokens_seen": 2551709696
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00022893596533461724,
      "loss": 0.0634,
      "theoretical_loss": 3.391475715437196,
      "tokens_seen": 2551971840
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00022885572139303486,
      "loss": 0.0623,
      "theoretical_loss": 3.39144827487642,
      "tokens_seen": 2552233984
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00022877547745145242,
      "loss": 0.0657,
      "theoretical_loss": 3.3914208379230395,
      "tokens_seen": 2552496128
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00022869523350987,
      "loss": 0.0656,
      "theoretical_loss": 3.3913934045762106,
      "tokens_seen": 2552758272
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00022861498956828761,
      "loss": 0.0624,
      "theoretical_loss": 3.3913659748350895,
      "tokens_seen": 2553020416
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00022853474562670517,
      "loss": 0.066,
      "theoretical_loss": 3.391338548698831,
      "tokens_seen": 2553282560
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00022845450168512278,
      "loss": 0.0644,
      "theoretical_loss": 3.391311126166592,
      "tokens_seen": 2553544704
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00022837425774354037,
      "loss": 0.0624,
      "theoretical_loss": 3.3912837072375286,
      "tokens_seen": 2553806848
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00022829401380195796,
      "loss": 0.0648,
      "theoretical_loss": 3.391256291910798,
      "tokens_seen": 2554068992
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00022821376986037554,
      "loss": 0.064,
      "theoretical_loss": 3.391228880185557,
      "tokens_seen": 2554331136
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00022813352591879313,
      "loss": 0.0625,
      "theoretical_loss": 3.3912014720609625,
      "tokens_seen": 2554593280
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00022805328197721074,
      "loss": 0.0657,
      "theoretical_loss": 3.391174067536173,
      "tokens_seen": 2554855424
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00022797303803562832,
      "loss": 0.0665,
      "theoretical_loss": 3.391146666610346,
      "tokens_seen": 2555117568
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00022789279409404588,
      "loss": 0.0665,
      "theoretical_loss": 3.3911192692826395,
      "tokens_seen": 2555379712
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.0002278125501524635,
      "loss": 0.0647,
      "theoretical_loss": 3.3910918755522124,
      "tokens_seen": 2555641856
    },
    {
      "epoch": 0.77,
      "objective/train/advantage_avg": 0.0013836933067068458,
      "objective/train/docs_used": 929757,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2935469150543213,
      "objective/train/original_loss": 1.2935466766357422,
      "objective/train/theoretical_loss": 3.391064485418223,
      "objective/train/tokens_used": 2576364000,
      "objective/train/value_avg": -0.0069427490234375,
      "objective/train/value_loss": 0.00013478376786224544,
      "objective/train/value_max": -2.8431415557861328e-05,
      "objective/train/value_min": -0.6962890625,
      "objective/train/value_reward_corr": 0.769539782212816,
      "objective/train/value_std": 0.01454925537109375,
      "objective/train/weight_avg": 1.0014495849609375,
      "objective/train/weighted_lm_loss": 1.295472502708435,
      "objective/train/weights_max": 1.4873151779174805,
      "objective/train/weights_min": 0.6114416718482971,
      "theoretical_loss": 3.391064485418223,
      "tokens_seen": 2555904000
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00022773230621088108,
      "loss": 0.0645,
      "theoretical_loss": 3.391064485418223,
      "tokens_seen": 2555904000
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00022765206226929867,
      "loss": 0.0667,
      "theoretical_loss": 3.3910370988798304,
      "tokens_seen": 2556166144
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00022757181832771625,
      "loss": 0.0667,
      "theoretical_loss": 3.3910097159361943,
      "tokens_seen": 2556428288
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00022749157438613386,
      "loss": 0.0634,
      "theoretical_loss": 3.3909823365864744,
      "tokens_seen": 2556690432
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.00022741133044455145,
      "loss": 0.0646,
      "theoretical_loss": 3.3909549608298297,
      "tokens_seen": 2556952576
    },
    {
      "epoch": 0.77,
      "learning_rate": 0.000227331086502969,
      "loss": 0.0647,
      "theoretical_loss": 3.3909275886654213,
      "tokens_seen": 2557214720
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022725084256138662,
      "loss": 0.0653,
      "theoretical_loss": 3.3909002200924094,
      "tokens_seen": 2557476864
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002271705986198042,
      "loss": 0.0655,
      "theoretical_loss": 3.3908728551099543,
      "tokens_seen": 2557739008
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022709035467822182,
      "loss": 0.0656,
      "theoretical_loss": 3.390845493717218,
      "tokens_seen": 2558001152
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022701011073663938,
      "loss": 0.0671,
      "theoretical_loss": 3.3908181359133605,
      "tokens_seen": 2558263296
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.000226929866795057,
      "loss": 0.0642,
      "theoretical_loss": 3.390790781697544,
      "tokens_seen": 2558525440
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022684962285347458,
      "loss": 0.0666,
      "theoretical_loss": 3.390763431068931,
      "tokens_seen": 2558787584
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022676937891189213,
      "loss": 0.0641,
      "theoretical_loss": 3.390736084026683,
      "tokens_seen": 2559049728
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022668913497030975,
      "loss": 0.0609,
      "theoretical_loss": 3.390708740569962,
      "tokens_seen": 2559311872
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022660889102872733,
      "loss": 0.0664,
      "theoretical_loss": 3.3906814006979316,
      "tokens_seen": 2559574016
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022652864708714495,
      "loss": 0.0641,
      "theoretical_loss": 3.390654064409754,
      "tokens_seen": 2559836160
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002264484031455625,
      "loss": 0.0601,
      "theoretical_loss": 3.3906267317045935,
      "tokens_seen": 2560098304
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002263681592039801,
      "loss": 0.0639,
      "theoretical_loss": 3.3905994025816124,
      "tokens_seen": 2560360448
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002262879152623977,
      "loss": 0.0644,
      "theoretical_loss": 3.3905720770399754,
      "tokens_seen": 2560622592
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002262076713208153,
      "loss": 0.0616,
      "theoretical_loss": 3.3905447550788463,
      "tokens_seen": 2560884736
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022612742737923287,
      "loss": 0.0637,
      "theoretical_loss": 3.39051743669739,
      "tokens_seen": 2561146880
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022604718343765046,
      "loss": 0.0638,
      "theoretical_loss": 3.39049012189477,
      "tokens_seen": 2561409024
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022596693949606807,
      "loss": 0.0627,
      "theoretical_loss": 3.3904628106701526,
      "tokens_seen": 2561671168
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022588669555448563,
      "loss": 0.066,
      "theoretical_loss": 3.3904355030227022,
      "tokens_seen": 2561933312
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022580645161290321,
      "loss": 0.0644,
      "theoretical_loss": 3.3904081989515844,
      "tokens_seen": 2562195456
    },
    {
      "epoch": 0.78,
      "objective/train/advantage_avg": 0.00031394496909342706,
      "objective/train/docs_used": 931999,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3014920949935913,
      "objective/train/original_loss": 1.3014919757843018,
      "objective/train/theoretical_loss": 3.390380898455965,
      "objective/train/tokens_used": 2582917600,
      "objective/train/value_avg": -0.005458831787109375,
      "objective/train/value_loss": 0.00011215580889256671,
      "objective/train/value_max": -3.647804260253906e-05,
      "objective/train/value_min": -0.265380859375,
      "objective/train/value_reward_corr": 0.7369466557109846,
      "objective/train/value_std": 0.01082611083984375,
      "objective/train/weight_avg": 1.0003670454025269,
      "objective/train/weighted_lm_loss": 1.30169677734375,
      "objective/train/weights_max": 1.1378111839294434,
      "objective/train/weights_min": 0.3843361437320709,
      "theoretical_loss": 3.390380898455965,
      "tokens_seen": 2562457600
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022572620767132083,
      "loss": 0.0631,
      "theoretical_loss": 3.390380898455965,
      "tokens_seen": 2562457600
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002256459637297384,
      "loss": 0.0665,
      "theoretical_loss": 3.390353601535011,
      "tokens_seen": 2562719744
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.000225565719788156,
      "loss": 0.0643,
      "theoretical_loss": 3.3903263081878876,
      "tokens_seen": 2562981888
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022548547584657358,
      "loss": 0.0635,
      "theoretical_loss": 3.390299018413762,
      "tokens_seen": 2563244032
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022540523190499117,
      "loss": 0.0636,
      "theoretical_loss": 3.3902717322118,
      "tokens_seen": 2563506176
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022532498796340878,
      "loss": 0.0659,
      "theoretical_loss": 3.3902444495811705,
      "tokens_seen": 2563768320
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022524474402182634,
      "loss": 0.0649,
      "theoretical_loss": 3.3902171705210398,
      "tokens_seen": 2564030464
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022516450008024395,
      "loss": 0.0626,
      "theoretical_loss": 3.390189895030576,
      "tokens_seen": 2564292608
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022508425613866154,
      "loss": 0.0635,
      "theoretical_loss": 3.390162623108948,
      "tokens_seen": 2564554752
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022500401219707915,
      "loss": 0.0651,
      "theoretical_loss": 3.390135354755323,
      "tokens_seen": 2564816896
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002249237682554967,
      "loss": 0.0644,
      "theoretical_loss": 3.3901080899688694,
      "tokens_seen": 2565079040
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002248435243139143,
      "loss": 0.0639,
      "theoretical_loss": 3.390080828748757,
      "tokens_seen": 2565341184
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002247632803723319,
      "loss": 0.0642,
      "theoretical_loss": 3.390053571094154,
      "tokens_seen": 2565603328
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022468303643074947,
      "loss": 0.0654,
      "theoretical_loss": 3.3900263170042306,
      "tokens_seen": 2565865472
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022460279248916708,
      "loss": 0.0634,
      "theoretical_loss": 3.3899990664781563,
      "tokens_seen": 2566127616
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022452254854758466,
      "loss": 0.066,
      "theoretical_loss": 3.389971819515101,
      "tokens_seen": 2566389760
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022444230460600225,
      "loss": 0.0662,
      "theoretical_loss": 3.389944576114235,
      "tokens_seen": 2566651904
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022436206066441984,
      "loss": 0.0624,
      "theoretical_loss": 3.3899173362747286,
      "tokens_seen": 2566914048
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022428181672283742,
      "loss": 0.065,
      "theoretical_loss": 3.389890099995753,
      "tokens_seen": 2567176192
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022420157278125503,
      "loss": 0.0661,
      "theoretical_loss": 3.389862867276479,
      "tokens_seen": 2567438336
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022412132883967262,
      "loss": 0.0628,
      "theoretical_loss": 3.3898356381160784,
      "tokens_seen": 2567700480
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002240410848980902,
      "loss": 0.0661,
      "theoretical_loss": 3.3898084125137222,
      "tokens_seen": 2567962624
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002239608409565078,
      "loss": 0.0644,
      "theoretical_loss": 3.389781190468583,
      "tokens_seen": 2568224768
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022388059701492538,
      "loss": 0.0651,
      "theoretical_loss": 3.3897539719798324,
      "tokens_seen": 2568486912
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022380035307334296,
      "loss": 0.0657,
      "theoretical_loss": 3.389726757046643,
      "tokens_seen": 2568749056
    },
    {
      "epoch": 0.78,
      "objective/train/advantage_avg": 0.0011137585388496518,
      "objective/train/docs_used": 934338,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2182008028030396,
      "objective/train/original_loss": 1.2182008028030396,
      "objective/train/theoretical_loss": 3.3896995456681878,
      "objective/train/tokens_used": 2589471200,
      "objective/train/value_avg": -0.0082244873046875,
      "objective/train/value_loss": 0.00020294927526265383,
      "objective/train/value_max": -1.9550323486328125e-05,
      "objective/train/value_min": -0.254150390625,
      "objective/train/value_reward_corr": 0.7417535635411541,
      "objective/train/value_std": 0.015960693359375,
      "objective/train/weight_avg": 1.0012061595916748,
      "objective/train/weighted_lm_loss": 1.2197353839874268,
      "objective/train/weights_max": 1.1746975183486938,
      "objective/train/weights_min": 0.3702559173107147,
      "theoretical_loss": 3.3896995456681878,
      "tokens_seen": 2569011200
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022372010913176055,
      "loss": 0.0633,
      "theoretical_loss": 3.3896995456681878,
      "tokens_seen": 2569011200
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022363986519017816,
      "loss": 0.0691,
      "theoretical_loss": 3.38967233784364,
      "tokens_seen": 2569273344
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022355962124859574,
      "loss": 0.0635,
      "theoretical_loss": 3.3896451335721727,
      "tokens_seen": 2569535488
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002234793773070133,
      "loss": 0.0645,
      "theoretical_loss": 3.3896179328529588,
      "tokens_seen": 2569797632
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022339913336543092,
      "loss": 0.0665,
      "theoretical_loss": 3.3895907356851733,
      "tokens_seen": 2570059776
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002233188894238485,
      "loss": 0.065,
      "theoretical_loss": 3.38956354206799,
      "tokens_seen": 2570321920
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022323864548226611,
      "loss": 0.0667,
      "theoretical_loss": 3.389536352000583,
      "tokens_seen": 2570584064
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022315840154068367,
      "loss": 0.0686,
      "theoretical_loss": 3.389509165482127,
      "tokens_seen": 2570846208
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022307815759910128,
      "loss": 0.0664,
      "theoretical_loss": 3.389481982511797,
      "tokens_seen": 2571108352
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022299791365751887,
      "loss": 0.067,
      "theoretical_loss": 3.3894548030887686,
      "tokens_seen": 2571370496
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022291766971593643,
      "loss": 0.0645,
      "theoretical_loss": 3.3894276272122172,
      "tokens_seen": 2571632640
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022283742577435404,
      "loss": 0.0641,
      "theoretical_loss": 3.389400454881318,
      "tokens_seen": 2571894784
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022275718183277163,
      "loss": 0.067,
      "theoretical_loss": 3.389373286095248,
      "tokens_seen": 2572156928
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022267693789118924,
      "loss": 0.0708,
      "theoretical_loss": 3.389346120853183,
      "tokens_seen": 2572419072
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002225966939496068,
      "loss": 0.0681,
      "theoretical_loss": 3.3893189591543,
      "tokens_seen": 2572681216
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022251645000802438,
      "loss": 0.0686,
      "theoretical_loss": 3.3892918009977753,
      "tokens_seen": 2572943360
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.000222436206066442,
      "loss": 0.0642,
      "theoretical_loss": 3.3892646463827862,
      "tokens_seen": 2573205504
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022235596212485958,
      "loss": 0.0636,
      "theoretical_loss": 3.3892374953085107,
      "tokens_seen": 2573467648
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022227571818327717,
      "loss": 0.0703,
      "theoretical_loss": 3.389210347774126,
      "tokens_seen": 2573729792
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022219547424169475,
      "loss": 0.0646,
      "theoretical_loss": 3.3891832037788103,
      "tokens_seen": 2573991936
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022211523030011236,
      "loss": 0.0631,
      "theoretical_loss": 3.389156063321742,
      "tokens_seen": 2574254080
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022203498635852992,
      "loss": 0.065,
      "theoretical_loss": 3.3891289264020994,
      "tokens_seen": 2574516224
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002219547424169475,
      "loss": 0.0661,
      "theoretical_loss": 3.3891017930190612,
      "tokens_seen": 2574778368
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022187449847536512,
      "loss": 0.0673,
      "theoretical_loss": 3.3890746631718067,
      "tokens_seen": 2575040512
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002217942545337827,
      "loss": 0.0612,
      "theoretical_loss": 3.3890475368595157,
      "tokens_seen": 2575302656
    },
    {
      "epoch": 0.78,
      "objective/train/advantage_avg": 0.0015231725992634892,
      "objective/train/docs_used": 936685,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2750457525253296,
      "objective/train/original_loss": 1.2750457525253296,
      "objective/train/theoretical_loss": 3.389020414081367,
      "objective/train/tokens_used": 2596024800,
      "objective/train/value_avg": -0.008087158203125,
      "objective/train/value_loss": 0.00017865085101220757,
      "objective/train/value_max": -2.86102294921875e-05,
      "objective/train/value_min": -0.2646484375,
      "objective/train/value_reward_corr": 0.6448530001084585,
      "objective/train/value_std": 0.012664794921875,
      "objective/train/weight_avg": 1.0016039609909058,
      "objective/train/weighted_lm_loss": 1.2768608331680298,
      "objective/train/weights_max": 1.1373058557510376,
      "objective/train/weights_min": 0.368586003780365,
      "theoretical_loss": 3.389020414081367,
      "tokens_seen": 2575564800
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002217140105922003,
      "loss": 0.0666,
      "theoretical_loss": 3.389020414081367,
      "tokens_seen": 2575564800
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022163376665061788,
      "loss": 0.0661,
      "theoretical_loss": 3.3889932948365407,
      "tokens_seen": 2575826944
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022155352270903546,
      "loss": 0.0631,
      "theoretical_loss": 3.3889661791242176,
      "tokens_seen": 2576089088
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022147327876745308,
      "loss": 0.0678,
      "theoretical_loss": 3.3889390669435775,
      "tokens_seen": 2576351232
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022139303482587063,
      "loss": 0.0682,
      "theoretical_loss": 3.388911958293802,
      "tokens_seen": 2576613376
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022131279088428825,
      "loss": 0.0616,
      "theoretical_loss": 3.388884853174071,
      "tokens_seen": 2576875520
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022123254694270583,
      "loss": 0.0687,
      "theoretical_loss": 3.3888577515835663,
      "tokens_seen": 2577137664
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022115230300112342,
      "loss": 0.0615,
      "theoretical_loss": 3.3888306535214694,
      "tokens_seen": 2577399808
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.000221072059059541,
      "loss": 0.0656,
      "theoretical_loss": 3.3888035589869627,
      "tokens_seen": 2577661952
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002209918151179586,
      "loss": 0.0651,
      "theoretical_loss": 3.3887764679792274,
      "tokens_seen": 2577924096
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002209115711763762,
      "loss": 0.0655,
      "theoretical_loss": 3.3887493804974462,
      "tokens_seen": 2578186240
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022083132723479376,
      "loss": 0.0692,
      "theoretical_loss": 3.388722296540802,
      "tokens_seen": 2578448384
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022075108329321137,
      "loss": 0.0674,
      "theoretical_loss": 3.3886952161084776,
      "tokens_seen": 2578710528
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022067083935162896,
      "loss": 0.0651,
      "theoretical_loss": 3.3886681391996563,
      "tokens_seen": 2578972672
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022059059541004654,
      "loss": 0.0676,
      "theoretical_loss": 3.3886410658135206,
      "tokens_seen": 2579234816
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022051035146846413,
      "loss": 0.0681,
      "theoretical_loss": 3.388613995949256,
      "tokens_seen": 2579496960
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022043010752688171,
      "loss": 0.0675,
      "theoretical_loss": 3.388586929606045,
      "tokens_seen": 2579759104
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022034986358529933,
      "loss": 0.0643,
      "theoretical_loss": 3.3885598667830727,
      "tokens_seen": 2580021248
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002202696196437169,
      "loss": 0.0619,
      "theoretical_loss": 3.3885328074795233,
      "tokens_seen": 2580283392
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002201893757021345,
      "loss": 0.0648,
      "theoretical_loss": 3.3885057516945816,
      "tokens_seen": 2580545536
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022010913176055208,
      "loss": 0.0674,
      "theoretical_loss": 3.388478699427433,
      "tokens_seen": 2580807680
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00022002888781896967,
      "loss": 0.0639,
      "theoretical_loss": 3.388451650677262,
      "tokens_seen": 2581069824
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00021994864387738725,
      "loss": 0.067,
      "theoretical_loss": 3.388424605443256,
      "tokens_seen": 2581331968
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00021986839993580484,
      "loss": 0.0655,
      "theoretical_loss": 3.3883975637245993,
      "tokens_seen": 2581594112
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00021978815599422245,
      "loss": 0.0639,
      "theoretical_loss": 3.388370525520479,
      "tokens_seen": 2581856256
    },
    {
      "epoch": 0.78,
      "objective/train/advantage_avg": -1.6538759155082516e-05,
      "objective/train/docs_used": 938961,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3801133632659912,
      "objective/train/original_loss": 1.380113124847412,
      "objective/train/theoretical_loss": 3.3883434908300805,
      "objective/train/tokens_used": 2602578400,
      "objective/train/value_avg": -0.00927734375,
      "objective/train/value_loss": 0.0004200639668852091,
      "objective/train/value_max": -3.4809112548828125e-05,
      "objective/train/value_min": -0.662109375,
      "objective/train/value_reward_corr": 0.7653193134682554,
      "objective/train/value_std": 0.021453857421875,
      "objective/train/weight_avg": 1.0001425743103027,
      "objective/train/weighted_lm_loss": 1.3799386024475098,
      "objective/train/weights_max": 1.3826122283935547,
      "objective/train/weights_min": 0.055677253752946854,
      "theoretical_loss": 3.3883434908300805,
      "tokens_seen": 2582118400
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00021970791205264004,
      "loss": 0.0653,
      "theoretical_loss": 3.3883434908300805,
      "tokens_seen": 2582118400
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002196276681110576,
      "loss": 0.0676,
      "theoretical_loss": 3.388316459652591,
      "tokens_seen": 2582380544
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002195474241694752,
      "loss": 0.0639,
      "theoretical_loss": 3.3882894319871983,
      "tokens_seen": 2582642688
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002194671802278928,
      "loss": 0.0654,
      "theoretical_loss": 3.388262407833089,
      "tokens_seen": 2582904832
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002193869362863104,
      "loss": 0.0671,
      "theoretical_loss": 3.3882353871894506,
      "tokens_seen": 2583166976
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00021930669234472797,
      "loss": 0.0657,
      "theoretical_loss": 3.388208370055471,
      "tokens_seen": 2583429120
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00021922644840314558,
      "loss": 0.065,
      "theoretical_loss": 3.388181356430338,
      "tokens_seen": 2583691264
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00021914620446156316,
      "loss": 0.0639,
      "theoretical_loss": 3.3881543463132404,
      "tokens_seen": 2583953408
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00021906596051998072,
      "loss": 0.0645,
      "theoretical_loss": 3.3881273397033667,
      "tokens_seen": 2584215552
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00021898571657839834,
      "loss": 0.0657,
      "theoretical_loss": 3.388100336599906,
      "tokens_seen": 2584477696
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00021890547263681592,
      "loss": 0.0649,
      "theoretical_loss": 3.388073337002047,
      "tokens_seen": 2584739840
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00021882522869523353,
      "loss": 0.0649,
      "theoretical_loss": 3.388046340908979,
      "tokens_seen": 2585001984
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002187449847536511,
      "loss": 0.0653,
      "theoretical_loss": 3.388019348319892,
      "tokens_seen": 2585264128
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002186647408120687,
      "loss": 0.0625,
      "theoretical_loss": 3.3879923592339765,
      "tokens_seen": 2585526272
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002185844968704863,
      "loss": 0.064,
      "theoretical_loss": 3.3879653736504216,
      "tokens_seen": 2585788416
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00021850425292890388,
      "loss": 0.0625,
      "theoretical_loss": 3.3879383915684187,
      "tokens_seen": 2586050560
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00021842400898732146,
      "loss": 0.0695,
      "theoretical_loss": 3.387911412987158,
      "tokens_seen": 2586312704
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00021834376504573905,
      "loss": 0.0652,
      "theoretical_loss": 3.3878844379058313,
      "tokens_seen": 2586574848
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00021826352110415666,
      "loss": 0.0678,
      "theoretical_loss": 3.387857466323629,
      "tokens_seen": 2586836992
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00021818327716257422,
      "loss": 0.0676,
      "theoretical_loss": 3.3878304982397434,
      "tokens_seen": 2587099136
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002181030332209918,
      "loss": 0.0647,
      "theoretical_loss": 3.387803533653366,
      "tokens_seen": 2587361280
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00021802278927940942,
      "loss": 0.0684,
      "theoretical_loss": 3.3877765725636886,
      "tokens_seen": 2587623424
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.000217942545337827,
      "loss": 0.0658,
      "theoretical_loss": 3.3877496149699047,
      "tokens_seen": 2587885568
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002178623013962446,
      "loss": 0.0639,
      "theoretical_loss": 3.387722660871206,
      "tokens_seen": 2588147712
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00021778205745466217,
      "loss": 0.0661,
      "theoretical_loss": 3.387695710266785,
      "tokens_seen": 2588409856
    },
    {
      "epoch": 0.78,
      "objective/train/advantage_avg": -0.00018109590746462345,
      "objective/train/docs_used": 941211,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2806812524795532,
      "objective/train/original_loss": 1.2806812524795532,
      "objective/train/theoretical_loss": 3.387668763155836,
      "objective/train/tokens_used": 2609132000,
      "objective/train/value_avg": -0.007171630859375,
      "objective/train/value_loss": 0.0002567979972809553,
      "objective/train/value_max": -2.7120113372802734e-05,
      "objective/train/value_min": -0.7080078125,
      "objective/train/value_reward_corr": 0.7377811245142272,
      "objective/train/value_std": 0.0166473388671875,
      "objective/train/weight_avg": 0.9999337792396545,
      "objective/train/weighted_lm_loss": 1.2800700664520264,
      "objective/train/weights_max": 1.4415401220321655,
      "objective/train/weights_min": 0.37006106972694397,
      "theoretical_loss": 3.387668763155836,
      "tokens_seen": 2588672000
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00021770181351307978,
      "loss": 0.0682,
      "theoretical_loss": 3.387668763155836,
      "tokens_seen": 2588672000
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00021762156957149737,
      "loss": 0.066,
      "theoretical_loss": 3.387641819537552,
      "tokens_seen": 2588934144
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00021754132562991493,
      "loss": 0.0682,
      "theoretical_loss": 3.3876148794111267,
      "tokens_seen": 2589196288
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00021746108168833254,
      "loss": 0.064,
      "theoretical_loss": 3.3875879427757543,
      "tokens_seen": 2589458432
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.00021738083774675013,
      "loss": 0.0654,
      "theoretical_loss": 3.3875610096306286,
      "tokens_seen": 2589720576
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002173005938051677,
      "loss": 0.0659,
      "theoretical_loss": 3.3875340799749445,
      "tokens_seen": 2589982720
    },
    {
      "epoch": 0.78,
      "learning_rate": 0.0002172203498635853,
      "loss": 0.0612,
      "theoretical_loss": 3.3875071538078965,
      "tokens_seen": 2590244864
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021714010592200288,
      "loss": 0.0623,
      "theoretical_loss": 3.38748023112868,
      "tokens_seen": 2590507008
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.0002170598619804205,
      "loss": 0.0641,
      "theoretical_loss": 3.3874533119364902,
      "tokens_seen": 2590769152
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021697961803883805,
      "loss": 0.0668,
      "theoretical_loss": 3.3874263962305227,
      "tokens_seen": 2591031296
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021689937409725567,
      "loss": 0.0633,
      "theoretical_loss": 3.3873994840099733,
      "tokens_seen": 2591293440
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021681913015567325,
      "loss": 0.0662,
      "theoretical_loss": 3.3873725752740382,
      "tokens_seen": 2591555584
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021673888621409087,
      "loss": 0.0638,
      "theoretical_loss": 3.387345670021914,
      "tokens_seen": 2591817728
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021665864227250842,
      "loss": 0.0658,
      "theoretical_loss": 3.387318768252797,
      "tokens_seen": 2592079872
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.000216578398330926,
      "loss": 0.0653,
      "theoretical_loss": 3.387291869965884,
      "tokens_seen": 2592342016
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021649815438934362,
      "loss": 0.0676,
      "theoretical_loss": 3.3872649751603725,
      "tokens_seen": 2592604160
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021641791044776118,
      "loss": 0.0662,
      "theoretical_loss": 3.3872380838354603,
      "tokens_seen": 2592866304
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.0002163376665061788,
      "loss": 0.0618,
      "theoretical_loss": 3.3872111959903446,
      "tokens_seen": 2593128448
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021625742256459638,
      "loss": 0.0632,
      "theoretical_loss": 3.3871843116242237,
      "tokens_seen": 2593390592
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021617717862301396,
      "loss": 0.0652,
      "theoretical_loss": 3.3871574307362957,
      "tokens_seen": 2593652736
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021609693468143155,
      "loss": 0.0653,
      "theoretical_loss": 3.387130553325759,
      "tokens_seen": 2593914880
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021601669073984913,
      "loss": 0.0658,
      "theoretical_loss": 3.3871036793918123,
      "tokens_seen": 2594177024
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021593644679826675,
      "loss": 0.0661,
      "theoretical_loss": 3.3870768089336556,
      "tokens_seen": 2594439168
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021585620285668433,
      "loss": 0.0664,
      "theoretical_loss": 3.3870499419504867,
      "tokens_seen": 2594701312
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021577595891510192,
      "loss": 0.066,
      "theoretical_loss": 3.3870230784415067,
      "tokens_seen": 2594963456
    },
    {
      "epoch": 0.79,
      "objective/train/advantage_avg": -0.00033047914621420205,
      "objective/train/docs_used": 943650,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4407196044921875,
      "objective/train/original_loss": 1.440719485282898,
      "objective/train/theoretical_loss": 3.3869962184059146,
      "objective/train/tokens_used": 2615685600,
      "objective/train/value_avg": -0.007740020751953125,
      "objective/train/value_loss": 0.000411410495871678,
      "objective/train/value_max": -3.534555435180664e-05,
      "objective/train/value_min": -0.86767578125,
      "objective/train/value_reward_corr": 0.7215062047244037,
      "objective/train/value_std": 0.0186920166015625,
      "objective/train/weight_avg": 0.9998422265052795,
      "objective/train/weighted_lm_loss": 1.4408422708511353,
      "objective/train/weights_max": 1.393456220626831,
      "objective/train/weights_min": 0.23286989331245422,
      "theoretical_loss": 3.3869962184059146,
      "tokens_seen": 2595225600
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.0002156957149735195,
      "loss": 0.0685,
      "theoretical_loss": 3.3869962184059146,
      "tokens_seen": 2595225600
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.0002156154710319371,
      "loss": 0.0694,
      "theoretical_loss": 3.386969361842911,
      "tokens_seen": 2595487744
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.0002155352270903547,
      "loss": 0.0671,
      "theoretical_loss": 3.3869425087516953,
      "tokens_seen": 2595749888
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021545498314877226,
      "loss": 0.0674,
      "theoretical_loss": 3.386915659131469,
      "tokens_seen": 2596012032
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021537473920718987,
      "loss": 0.0681,
      "theoretical_loss": 3.386888812981433,
      "tokens_seen": 2596274176
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021529449526560746,
      "loss": 0.0642,
      "theoretical_loss": 3.3868619703007883,
      "tokens_seen": 2596536320
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021521425132402502,
      "loss": 0.0641,
      "theoretical_loss": 3.386835131088737,
      "tokens_seen": 2596798464
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021513400738244263,
      "loss": 0.065,
      "theoretical_loss": 3.386808295344479,
      "tokens_seen": 2597060608
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021505376344086021,
      "loss": 0.0656,
      "theoretical_loss": 3.386781463067218,
      "tokens_seen": 2597322752
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021497351949927783,
      "loss": 0.0654,
      "theoretical_loss": 3.386754634256156,
      "tokens_seen": 2597584896
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021489327555769539,
      "loss": 0.0644,
      "theoretical_loss": 3.3867278089104946,
      "tokens_seen": 2597847040
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.000214813031616113,
      "loss": 0.066,
      "theoretical_loss": 3.3867009870294376,
      "tokens_seen": 2598109184
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021473278767453058,
      "loss": 0.0685,
      "theoretical_loss": 3.3866741686121875,
      "tokens_seen": 2598371328
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021465254373294817,
      "loss": 0.0645,
      "theoretical_loss": 3.3866473536579473,
      "tokens_seen": 2598633472
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021457229979136576,
      "loss": 0.0684,
      "theoretical_loss": 3.3866205421659217,
      "tokens_seen": 2598895616
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021449205584978334,
      "loss": 0.0675,
      "theoretical_loss": 3.386593734135313,
      "tokens_seen": 2599157760
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021441181190820095,
      "loss": 0.0674,
      "theoretical_loss": 3.3865669295653262,
      "tokens_seen": 2599419904
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.0002143315679666185,
      "loss": 0.0625,
      "theoretical_loss": 3.3865401284551657,
      "tokens_seen": 2599682048
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.0002142513240250361,
      "loss": 0.0662,
      "theoretical_loss": 3.3865133308040356,
      "tokens_seen": 2599944192
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.0002141710800834537,
      "loss": 0.0673,
      "theoretical_loss": 3.3864865366111414,
      "tokens_seen": 2600206336
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.0002140908361418713,
      "loss": 0.0626,
      "theoretical_loss": 3.3864597458756878,
      "tokens_seen": 2600468480
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021401059220028888,
      "loss": 0.0674,
      "theoretical_loss": 3.38643295859688,
      "tokens_seen": 2600730624
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021393034825870647,
      "loss": 0.0695,
      "theoretical_loss": 3.3864061747739242,
      "tokens_seen": 2600992768
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021385010431712408,
      "loss": 0.0695,
      "theoretical_loss": 3.386379394406026,
      "tokens_seen": 2601254912
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021376986037554166,
      "loss": 0.0648,
      "theoretical_loss": 3.3863526174923915,
      "tokens_seen": 2601517056
    },
    {
      "epoch": 0.79,
      "objective/train/advantage_avg": 0.0008119245758280158,
      "objective/train/docs_used": 946148,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.417919635772705,
      "objective/train/original_loss": 1.4179197549819946,
      "objective/train/theoretical_loss": 3.3863258440322275,
      "objective/train/tokens_used": 2622239200,
      "objective/train/value_avg": -0.0088043212890625,
      "objective/train/value_loss": 0.0001911604340421036,
      "objective/train/value_max": -3.707408905029297e-05,
      "objective/train/value_min": -0.51513671875,
      "objective/train/value_reward_corr": 0.7656120068838768,
      "objective/train/value_std": 0.016815185546875,
      "objective/train/weight_avg": 1.0009019374847412,
      "objective/train/weighted_lm_loss": 1.4192315340042114,
      "objective/train/weights_max": 1.3128715753555298,
      "objective/train/weights_min": 0.3708581030368805,
      "theoretical_loss": 3.3863258440322275,
      "tokens_seen": 2601779200
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021368961643395922,
      "loss": 0.0668,
      "theoretical_loss": 3.3863258440322275,
      "tokens_seen": 2601779200
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021360937249237684,
      "loss": 0.0676,
      "theoretical_loss": 3.3862990740247403,
      "tokens_seen": 2602041344
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021352912855079442,
      "loss": 0.0653,
      "theoretical_loss": 3.3862723074691377,
      "tokens_seen": 2602303488
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.000213448884609212,
      "loss": 0.0705,
      "theoretical_loss": 3.3862455443646255,
      "tokens_seen": 2602565632
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.0002133686406676296,
      "loss": 0.0683,
      "theoretical_loss": 3.3862187847104126,
      "tokens_seen": 2602827776
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021328839672604718,
      "loss": 0.0651,
      "theoretical_loss": 3.3861920285057057,
      "tokens_seen": 2603089920
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.0002132081527844648,
      "loss": 0.0661,
      "theoretical_loss": 3.386165275749714,
      "tokens_seen": 2603352064
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021312790884288235,
      "loss": 0.0651,
      "theoretical_loss": 3.3861385264416444,
      "tokens_seen": 2603614208
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021304766490129996,
      "loss": 0.066,
      "theoretical_loss": 3.3861117805807064,
      "tokens_seen": 2603876352
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021296742095971755,
      "loss": 0.0662,
      "theoretical_loss": 3.3860850381661085,
      "tokens_seen": 2604138496
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021288717701813516,
      "loss": 0.0675,
      "theoretical_loss": 3.3860582991970594,
      "tokens_seen": 2604400640
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021280693307655272,
      "loss": 0.0667,
      "theoretical_loss": 3.3860315636727694,
      "tokens_seen": 2604662784
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.0002127266891349703,
      "loss": 0.0668,
      "theoretical_loss": 3.386004831592447,
      "tokens_seen": 2604924928
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021264644519338792,
      "loss": 0.0639,
      "theoretical_loss": 3.385978102955303,
      "tokens_seen": 2605187072
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021256620125180547,
      "loss": 0.0647,
      "theoretical_loss": 3.3859513777605468,
      "tokens_seen": 2605449216
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.0002124859573102231,
      "loss": 0.0651,
      "theoretical_loss": 3.385924656007389,
      "tokens_seen": 2605711360
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021240571336864067,
      "loss": 0.0658,
      "theoretical_loss": 3.38589793769504,
      "tokens_seen": 2605973504
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021232546942705826,
      "loss": 0.0668,
      "theoretical_loss": 3.3858712228227117,
      "tokens_seen": 2606235648
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021224522548547584,
      "loss": 0.0632,
      "theoretical_loss": 3.385844511389614,
      "tokens_seen": 2606497792
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021216498154389343,
      "loss": 0.0694,
      "theoretical_loss": 3.385817803394959,
      "tokens_seen": 2606759936
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021208473760231104,
      "loss": 0.0651,
      "theoretical_loss": 3.3857910988379576,
      "tokens_seen": 2607022080
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021200449366072863,
      "loss": 0.0655,
      "theoretical_loss": 3.385764397717823,
      "tokens_seen": 2607284224
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.0002119242497191462,
      "loss": 0.066,
      "theoretical_loss": 3.3857377000337663,
      "tokens_seen": 2607546368
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.0002118440057775638,
      "loss": 0.0633,
      "theoretical_loss": 3.3857110057850006,
      "tokens_seen": 2607808512
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021176376183598138,
      "loss": 0.0662,
      "theoretical_loss": 3.3856843149707383,
      "tokens_seen": 2608070656
    },
    {
      "epoch": 0.79,
      "objective/train/advantage_avg": 0.0013975916663184762,
      "objective/train/docs_used": 948381,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3176509141921997,
      "objective/train/original_loss": 1.3176510334014893,
      "objective/train/theoretical_loss": 3.3856576275901924,
      "objective/train/tokens_used": 2628792800,
      "objective/train/value_avg": -0.00994873046875,
      "objective/train/value_loss": 0.00038583975401706994,
      "objective/train/value_max": -2.0325183868408203e-05,
      "objective/train/value_min": -0.93505859375,
      "objective/train/value_reward_corr": 0.7545843172896143,
      "objective/train/value_std": 0.0260009765625,
      "objective/train/weight_avg": 1.001584529876709,
      "objective/train/weighted_lm_loss": 1.3192466497421265,
      "objective/train/weights_max": 2.2463696002960205,
      "objective/train/weights_min": 0.371359258890152,
      "theoretical_loss": 3.3856576275901924,
      "tokens_seen": 2608332800
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021168351789439897,
      "loss": 0.0635,
      "theoretical_loss": 3.3856576275901924,
      "tokens_seen": 2608332800
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021160327395281655,
      "loss": 0.0665,
      "theoretical_loss": 3.385630943642576,
      "tokens_seen": 2608594944
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021152303001123417,
      "loss": 0.0648,
      "theoretical_loss": 3.3856042631271026,
      "tokens_seen": 2608857088
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021144278606965175,
      "loss": 0.066,
      "theoretical_loss": 3.385577586042986,
      "tokens_seen": 2609119232
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.0002113625421280693,
      "loss": 0.066,
      "theoretical_loss": 3.385550912389441,
      "tokens_seen": 2609381376
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021128229818648692,
      "loss": 0.0671,
      "theoretical_loss": 3.385524242165681,
      "tokens_seen": 2609643520
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.0002112020542449045,
      "loss": 0.0635,
      "theoretical_loss": 3.3854975753709207,
      "tokens_seen": 2609905664
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021112181030332212,
      "loss": 0.0671,
      "theoretical_loss": 3.3854709120043744,
      "tokens_seen": 2610167808
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021104156636173968,
      "loss": 0.0652,
      "theoretical_loss": 3.3854442520652577,
      "tokens_seen": 2610429952
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.0002109613224201573,
      "loss": 0.0653,
      "theoretical_loss": 3.385417595552786,
      "tokens_seen": 2610692096
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021088107847857488,
      "loss": 0.0673,
      "theoretical_loss": 3.3853909424661746,
      "tokens_seen": 2610954240
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021080083453699246,
      "loss": 0.0677,
      "theoretical_loss": 3.3853642928046397,
      "tokens_seen": 2611216384
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021072059059541005,
      "loss": 0.0668,
      "theoretical_loss": 3.3853376465673968,
      "tokens_seen": 2611478528
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021064034665382763,
      "loss": 0.0699,
      "theoretical_loss": 3.3853110037536625,
      "tokens_seen": 2611740672
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021056010271224525,
      "loss": 0.068,
      "theoretical_loss": 3.3852843643626533,
      "tokens_seen": 2612002816
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.0002104798587706628,
      "loss": 0.0691,
      "theoretical_loss": 3.3852577283935865,
      "tokens_seen": 2612264960
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.0002103996148290804,
      "loss": 0.068,
      "theoretical_loss": 3.385231095845678,
      "tokens_seen": 2612527104
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.000210319370887498,
      "loss": 0.0625,
      "theoretical_loss": 3.385204466718147,
      "tokens_seen": 2612789248
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.0002102391269459156,
      "loss": 0.0661,
      "theoretical_loss": 3.3851778410102096,
      "tokens_seen": 2613051392
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021015888300433317,
      "loss": 0.0673,
      "theoretical_loss": 3.3851512187210844,
      "tokens_seen": 2613313536
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00021007863906275076,
      "loss": 0.066,
      "theoretical_loss": 3.3851245998499895,
      "tokens_seen": 2613575680
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00020999839512116837,
      "loss": 0.0669,
      "theoretical_loss": 3.385097984396143,
      "tokens_seen": 2613837824
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00020991815117958596,
      "loss": 0.0657,
      "theoretical_loss": 3.3850713723587633,
      "tokens_seen": 2614099968
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00020983790723800352,
      "loss": 0.0649,
      "theoretical_loss": 3.38504476373707,
      "tokens_seen": 2614362112
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00020975766329642113,
      "loss": 0.0652,
      "theoretical_loss": 3.3850181585302823,
      "tokens_seen": 2614624256
    },
    {
      "epoch": 0.79,
      "objective/train/advantage_avg": 0.0011500355321913958,
      "objective/train/docs_used": 950739,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3640177249908447,
      "objective/train/original_loss": 1.3640177249908447,
      "objective/train/theoretical_loss": 3.384991556737619,
      "objective/train/tokens_used": 2635346400,
      "objective/train/value_avg": -0.007099151611328125,
      "objective/train/value_loss": 0.00024195798323489726,
      "objective/train/value_max": -2.753734588623047e-05,
      "objective/train/value_min": -0.94287109375,
      "objective/train/value_reward_corr": 0.7123578471830624,
      "objective/train/value_std": 0.0147705078125,
      "objective/train/weight_avg": 1.0012575387954712,
      "objective/train/weighted_lm_loss": 1.365799069404602,
      "objective/train/weights_max": 2.00335693359375,
      "objective/train/weights_min": 0.3683691918849945,
      "theoretical_loss": 3.384991556737619,
      "tokens_seen": 2614886400
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00020967741935483871,
      "loss": 0.0635,
      "theoretical_loss": 3.384991556737619,
      "tokens_seen": 2614886400
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.0002095971754132563,
      "loss": 0.0666,
      "theoretical_loss": 3.3849649583583004,
      "tokens_seen": 2615148544
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00020951693147167389,
      "loss": 0.0657,
      "theoretical_loss": 3.3849383633915457,
      "tokens_seen": 2615410688
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.0002094366875300915,
      "loss": 0.065,
      "theoretical_loss": 3.3849117718365758,
      "tokens_seen": 2615672832
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00020935644358850908,
      "loss": 0.067,
      "theoretical_loss": 3.38488518369261,
      "tokens_seen": 2615934976
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00020927619964692664,
      "loss": 0.0653,
      "theoretical_loss": 3.384858598958871,
      "tokens_seen": 2616197120
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00020919595570534426,
      "loss": 0.0656,
      "theoretical_loss": 3.384832017634578,
      "tokens_seen": 2616459264
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00020911571176376184,
      "loss": 0.065,
      "theoretical_loss": 3.384805439718953,
      "tokens_seen": 2616721408
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00020903546782217945,
      "loss": 0.0656,
      "theoretical_loss": 3.384778865211217,
      "tokens_seen": 2616983552
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.000208955223880597,
      "loss": 0.0662,
      "theoretical_loss": 3.3847522941105925,
      "tokens_seen": 2617245696
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.0002088749799390146,
      "loss": 0.0644,
      "theoretical_loss": 3.384725726416301,
      "tokens_seen": 2617507840
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.0002087947359974322,
      "loss": 0.0647,
      "theoretical_loss": 3.384699162127564,
      "tokens_seen": 2617769984
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00020871449205584977,
      "loss": 0.0672,
      "theoretical_loss": 3.3846726012436057,
      "tokens_seen": 2618032128
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00020863424811426738,
      "loss": 0.0637,
      "theoretical_loss": 3.3846460437636474,
      "tokens_seen": 2618294272
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00020855400417268497,
      "loss": 0.0634,
      "theoretical_loss": 3.3846194896869126,
      "tokens_seen": 2618556416
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00020847376023110258,
      "loss": 0.0683,
      "theoretical_loss": 3.3845929390126246,
      "tokens_seen": 2618818560
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00020839351628952014,
      "loss": 0.0672,
      "theoretical_loss": 3.384566391740007,
      "tokens_seen": 2619080704
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00020831327234793772,
      "loss": 0.0674,
      "theoretical_loss": 3.3845398478682833,
      "tokens_seen": 2619342848
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00020823302840635534,
      "loss": 0.0637,
      "theoretical_loss": 3.384513307396678,
      "tokens_seen": 2619604992
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00020815278446477292,
      "loss": 0.0647,
      "theoretical_loss": 3.384486770324415,
      "tokens_seen": 2619867136
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.0002080725405231905,
      "loss": 0.0658,
      "theoretical_loss": 3.3844602366507184,
      "tokens_seen": 2620129280
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.0002079922965816081,
      "loss": 0.0667,
      "theoretical_loss": 3.3844337063748138,
      "tokens_seen": 2620391424
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00020791205264002568,
      "loss": 0.0682,
      "theoretical_loss": 3.384407179495926,
      "tokens_seen": 2620653568
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00020783180869844326,
      "loss": 0.0634,
      "theoretical_loss": 3.38438065601328,
      "tokens_seen": 2620915712
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00020775156475686085,
      "loss": 0.0629,
      "theoretical_loss": 3.3843541359261016,
      "tokens_seen": 2621177856
    },
    {
      "debugging/Compilability": 1.0,
      "debugging/distinct-1-grams": 0.7381111704358672,
      "debugging/entropy-1-grams": 5.723658601072414,
      "debugging/length": 472.25,
      "debugging/num_segments": 24,
      "debugging/raw_token_scores_avg": 0.007372692227363586,
      "debugging/raw_token_scores_std": 0.024886412546038628,
      "debugging/score": 0.004950147325261966,
      "debugging/score_std": 0.005382910365563692,
      "epoch": 0.79,
      "objective/train/advantage_avg": 0.0008712936541996896,
      "objective/train/docs_used": 953088,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3545817136764526,
      "objective/train/original_loss": 1.3545817136764526,
      "objective/train/theoretical_loss": 3.3843276192336167,
      "objective/train/tokens_used": 2641900000,
      "objective/train/value_avg": -0.00824737548828125,
      "objective/train/value_loss": 0.0002766008547041565,
      "objective/train/value_max": -2.9981136322021484e-05,
      "objective/train/value_min": -0.826171875,
      "objective/train/value_reward_corr": 0.744913716275677,
      "objective/train/value_std": 0.0185394287109375,
      "objective/train/weight_avg": 1.0009992122650146,
      "objective/train/weighted_lm_loss": 1.3555899858474731,
      "objective/train/weights_max": 2.2153494358062744,
      "objective/train/weights_min": 0.38233569264411926,
      "theoretical_loss": 3.3843276192336167,
      "tokens_seen": 2621440000
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00020767132081527846,
      "loss": 0.0665,
      "theoretical_loss": 3.3843276192336167,
      "tokens_seen": 2621440000
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00020759107687369605,
      "loss": 0.0635,
      "theoretical_loss": 3.3843011059350507,
      "tokens_seen": 2621702144
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00020751083293211363,
      "loss": 0.0655,
      "theoretical_loss": 3.3842745960296314,
      "tokens_seen": 2621964288
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00020743058899053122,
      "loss": 0.0652,
      "theoretical_loss": 3.384248089516584,
      "tokens_seen": 2622226432
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.0002073503450489488,
      "loss": 0.0629,
      "theoretical_loss": 3.3842215863951353,
      "tokens_seen": 2622488576
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00020727010110736642,
      "loss": 0.0642,
      "theoretical_loss": 3.384195086664513,
      "tokens_seen": 2622750720
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.00020718985716578397,
      "loss": 0.0647,
      "theoretical_loss": 3.3841685903239447,
      "tokens_seen": 2623012864
    },
    {
      "epoch": 0.79,
      "learning_rate": 0.0002071096132242016,
      "loss": 0.0688,
      "theoretical_loss": 3.3841420973726573,
      "tokens_seen": 2623275008
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020702936928261917,
      "loss": 0.0686,
      "theoretical_loss": 3.3841156078098784,
      "tokens_seen": 2623537152
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020694912534103673,
      "loss": 0.0673,
      "theoretical_loss": 3.384089121634837,
      "tokens_seen": 2623799296
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020686888139945434,
      "loss": 0.066,
      "theoretical_loss": 3.384062638846761,
      "tokens_seen": 2624061440
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020678863745787193,
      "loss": 0.0667,
      "theoretical_loss": 3.3840361594448787,
      "tokens_seen": 2624323584
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020670839351628954,
      "loss": 0.0664,
      "theoretical_loss": 3.3840096834284195,
      "tokens_seen": 2624585728
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0002066281495747071,
      "loss": 0.0668,
      "theoretical_loss": 3.383983210796612,
      "tokens_seen": 2624847872
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0002065479056331247,
      "loss": 0.0687,
      "theoretical_loss": 3.383956741548686,
      "tokens_seen": 2625110016
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0002064676616915423,
      "loss": 0.0667,
      "theoretical_loss": 3.3839302756838707,
      "tokens_seen": 2625372160
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020638741774995988,
      "loss": 0.0675,
      "theoretical_loss": 3.383903813201396,
      "tokens_seen": 2625634304
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020630717380837747,
      "loss": 0.0636,
      "theoretical_loss": 3.383877354100492,
      "tokens_seen": 2625896448
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020622692986679505,
      "loss": 0.0678,
      "theoretical_loss": 3.3838508983803894,
      "tokens_seen": 2626158592
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020614668592521267,
      "loss": 0.0665,
      "theoretical_loss": 3.383824446040319,
      "tokens_seen": 2626420736
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020606644198363025,
      "loss": 0.0647,
      "theoretical_loss": 3.383797997079511,
      "tokens_seen": 2626682880
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0002059861980420478,
      "loss": 0.0664,
      "theoretical_loss": 3.383771551497196,
      "tokens_seen": 2626945024
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020590595410046542,
      "loss": 0.0673,
      "theoretical_loss": 3.383745109292607,
      "tokens_seen": 2627207168
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.000205825710158883,
      "loss": 0.0657,
      "theoretical_loss": 3.3837186704649747,
      "tokens_seen": 2627469312
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0002057454662173006,
      "loss": 0.0654,
      "theoretical_loss": 3.383692235013531,
      "tokens_seen": 2627731456
    },
    {
      "epoch": 0.8,
      "objective/train/advantage_avg": 0.0006061241729184985,
      "objective/train/docs_used": 955375,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3414796590805054,
      "objective/train/original_loss": 1.341479778289795,
      "objective/train/theoretical_loss": 3.3836658029375077,
      "objective/train/tokens_used": 2648453600,
      "objective/train/value_avg": -0.00621795654296875,
      "objective/train/value_loss": 0.00018008060578722507,
      "objective/train/value_max": -4.6133995056152344e-05,
      "objective/train/value_min": -0.488525390625,
      "objective/train/value_reward_corr": 0.7103285800180809,
      "objective/train/value_std": 0.01334381103515625,
      "objective/train/weight_avg": 1.0006861686706543,
      "objective/train/weighted_lm_loss": 1.3422250747680664,
      "objective/train/weights_max": 1.2228349447250366,
      "objective/train/weights_min": 0.37098827958106995,
      "theoretical_loss": 3.3836658029375077,
      "tokens_seen": 2627993600
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020566522227571818,
      "loss": 0.066,
      "theoretical_loss": 3.3836658029375077,
      "tokens_seen": 2627993600
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0002055849783341358,
      "loss": 0.0655,
      "theoretical_loss": 3.3836393742361377,
      "tokens_seen": 2628255744
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020550473439255338,
      "loss": 0.0659,
      "theoretical_loss": 3.3836129489086537,
      "tokens_seen": 2628517888
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020542449045097094,
      "loss": 0.065,
      "theoretical_loss": 3.3835865269542884,
      "tokens_seen": 2628780032
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020534424650938855,
      "loss": 0.0651,
      "theoretical_loss": 3.3835601083722744,
      "tokens_seen": 2629042176
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020526400256780613,
      "loss": 0.0632,
      "theoretical_loss": 3.3835336931618456,
      "tokens_seen": 2629304320
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020518375862622375,
      "loss": 0.0666,
      "theoretical_loss": 3.3835072813222355,
      "tokens_seen": 2629566464
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0002051035146846413,
      "loss": 0.0642,
      "theoretical_loss": 3.3834808728526786,
      "tokens_seen": 2629828608
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0002050232707430589,
      "loss": 0.0657,
      "theoretical_loss": 3.383454467752408,
      "tokens_seen": 2630090752
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0002049430268014765,
      "loss": 0.0644,
      "theoretical_loss": 3.3834280660206586,
      "tokens_seen": 2630352896
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020486278285989406,
      "loss": 0.0628,
      "theoretical_loss": 3.3834016676566647,
      "tokens_seen": 2630615040
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020478253891831167,
      "loss": 0.0684,
      "theoretical_loss": 3.383375272659662,
      "tokens_seen": 2630877184
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020470229497672926,
      "loss": 0.0655,
      "theoretical_loss": 3.3833488810288843,
      "tokens_seen": 2631139328
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020462205103514687,
      "loss": 0.0679,
      "theoretical_loss": 3.3833224927635683,
      "tokens_seen": 2631401472
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020454180709356443,
      "loss": 0.0656,
      "theoretical_loss": 3.3832961078629493,
      "tokens_seen": 2631663616
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020446156315198202,
      "loss": 0.0647,
      "theoretical_loss": 3.3832697263262625,
      "tokens_seen": 2631925760
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020438131921039963,
      "loss": 0.0665,
      "theoretical_loss": 3.3832433481527446,
      "tokens_seen": 2632187904
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020430107526881721,
      "loss": 0.0669,
      "theoretical_loss": 3.383216973341632,
      "tokens_seen": 2632450048
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0002042208313272348,
      "loss": 0.0645,
      "theoretical_loss": 3.383190601892161,
      "tokens_seen": 2632712192
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020414058738565239,
      "loss": 0.0641,
      "theoretical_loss": 3.383164233803569,
      "tokens_seen": 2632974336
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020406034344406997,
      "loss": 0.0631,
      "theoretical_loss": 3.3831378690750924,
      "tokens_seen": 2633236480
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020398009950248756,
      "loss": 0.0639,
      "theoretical_loss": 3.383111507705969,
      "tokens_seen": 2633498624
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020389985556090514,
      "loss": 0.0621,
      "theoretical_loss": 3.3830851496954364,
      "tokens_seen": 2633760768
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020381961161932276,
      "loss": 0.0646,
      "theoretical_loss": 3.383058795042732,
      "tokens_seen": 2634022912
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020373936767774034,
      "loss": 0.0658,
      "theoretical_loss": 3.3830324437470956,
      "tokens_seen": 2634285056
    },
    {
      "epoch": 0.8,
      "objective/train/advantage_avg": -0.000395024981116876,
      "objective/train/docs_used": 957617,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.215680718421936,
      "objective/train/original_loss": 1.2156805992126465,
      "objective/train/theoretical_loss": 3.3830060958077635,
      "objective/train/tokens_used": 2655007200,
      "objective/train/value_avg": -0.00763702392578125,
      "objective/train/value_loss": 0.00011784796515712515,
      "objective/train/value_max": -2.3186206817626953e-05,
      "objective/train/value_min": -0.247314453125,
      "objective/train/value_reward_corr": 0.8557826421660346,
      "objective/train/value_std": 0.0171051025390625,
      "objective/train/weight_avg": 0.9996615052223206,
      "objective/train/weighted_lm_loss": 1.2155773639678955,
      "objective/train/weights_max": 1.1127924919128418,
      "objective/train/weights_min": 0.38944852352142334,
      "theoretical_loss": 3.3830060958077635,
      "tokens_seen": 2634547200
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020365912373615793,
      "loss": 0.064,
      "theoretical_loss": 3.3830060958077635,
      "tokens_seen": 2634547200
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0002035788797945755,
      "loss": 0.0674,
      "theoretical_loss": 3.382979751223975,
      "tokens_seen": 2634809344
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0002034986358529931,
      "loss": 0.0656,
      "theoretical_loss": 3.3829534099949696,
      "tokens_seen": 2635071488
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0002034183919114107,
      "loss": 0.0636,
      "theoretical_loss": 3.382927072119986,
      "tokens_seen": 2635333632
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020333814796982827,
      "loss": 0.0654,
      "theoretical_loss": 3.382900737598263,
      "tokens_seen": 2635595776
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020325790402824588,
      "loss": 0.0666,
      "theoretical_loss": 3.3828744064290412,
      "tokens_seen": 2635857920
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020317766008666347,
      "loss": 0.0637,
      "theoretical_loss": 3.38284807861156,
      "tokens_seen": 2636120064
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020309741614508102,
      "loss": 0.0628,
      "theoretical_loss": 3.382821754145059,
      "tokens_seen": 2636382208
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020301717220349864,
      "loss": 0.066,
      "theoretical_loss": 3.3827954330287797,
      "tokens_seen": 2636644352
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020293692826191622,
      "loss": 0.0662,
      "theoretical_loss": 3.3827691152619614,
      "tokens_seen": 2636906496
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020285668432033384,
      "loss": 0.0649,
      "theoretical_loss": 3.382742800843846,
      "tokens_seen": 2637168640
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0002027764403787514,
      "loss": 0.0641,
      "theoretical_loss": 3.382716489773674,
      "tokens_seen": 2637430784
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.000202696196437169,
      "loss": 0.0657,
      "theoretical_loss": 3.3826901820506867,
      "tokens_seen": 2637692928
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0002026159524955866,
      "loss": 0.0618,
      "theoretical_loss": 3.3826638776741262,
      "tokens_seen": 2637955072
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020253570855400418,
      "loss": 0.0665,
      "theoretical_loss": 3.3826375766432344,
      "tokens_seen": 2638217216
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020245546461242176,
      "loss": 0.064,
      "theoretical_loss": 3.3826112789572527,
      "tokens_seen": 2638479360
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020237522067083935,
      "loss": 0.0645,
      "theoretical_loss": 3.382584984615424,
      "tokens_seen": 2638741504
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020229497672925696,
      "loss": 0.0645,
      "theoretical_loss": 3.3825586936169905,
      "tokens_seen": 2639003648
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020221473278767452,
      "loss": 0.0656,
      "theoretical_loss": 3.382532405961195,
      "tokens_seen": 2639265792
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0002021344888460921,
      "loss": 0.0646,
      "theoretical_loss": 3.3825061216472814,
      "tokens_seen": 2639527936
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020205424490450972,
      "loss": 0.0637,
      "theoretical_loss": 3.382479840674492,
      "tokens_seen": 2639790080
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0002019740009629273,
      "loss": 0.0632,
      "theoretical_loss": 3.3824535630420707,
      "tokens_seen": 2640052224
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0002018937570213449,
      "loss": 0.064,
      "theoretical_loss": 3.3824272887492612,
      "tokens_seen": 2640314368
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020181351307976247,
      "loss": 0.0659,
      "theoretical_loss": 3.382401017795308,
      "tokens_seen": 2640576512
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0002017332691381801,
      "loss": 0.0635,
      "theoretical_loss": 3.382374750179455,
      "tokens_seen": 2640838656
    },
    {
      "epoch": 0.8,
      "objective/train/advantage_avg": 0.00020698863954748958,
      "objective/train/docs_used": 959716,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2583776712417603,
      "objective/train/original_loss": 1.2583776712417603,
      "objective/train/theoretical_loss": 3.3823484859009474,
      "objective/train/tokens_used": 2661560800,
      "objective/train/value_avg": -0.006702423095703125,
      "objective/train/value_loss": 0.00013055212912149727,
      "objective/train/value_max": -2.485513687133789e-05,
      "objective/train/value_min": -0.459716796875,
      "objective/train/value_reward_corr": 0.760377423537848,
      "objective/train/value_std": 0.01239776611328125,
      "objective/train/weight_avg": 1.0002684593200684,
      "objective/train/weighted_lm_loss": 1.2586296796798706,
      "objective/train/weights_max": 1.1409211158752441,
      "objective/train/weights_min": 0.39653629064559937,
      "theoretical_loss": 3.3823484859009474,
      "tokens_seen": 2641100800
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020165302519659767,
      "loss": 0.0615,
      "theoretical_loss": 3.3823484859009474,
      "tokens_seen": 2641100800
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020157278125501523,
      "loss": 0.066,
      "theoretical_loss": 3.3823222249590286,
      "tokens_seen": 2641362944
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020149253731343284,
      "loss": 0.0637,
      "theoretical_loss": 3.3822959673529454,
      "tokens_seen": 2641625088
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020141229337185043,
      "loss": 0.0624,
      "theoretical_loss": 3.3822697130819415,
      "tokens_seen": 2641887232
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020133204943026804,
      "loss": 0.0667,
      "theoretical_loss": 3.3822434621452637,
      "tokens_seen": 2642149376
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0002012518054886856,
      "loss": 0.0638,
      "theoretical_loss": 3.382217214542157,
      "tokens_seen": 2642411520
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0002011715615471032,
      "loss": 0.0613,
      "theoretical_loss": 3.3821909702718678,
      "tokens_seen": 2642673664
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0002010913176055208,
      "loss": 0.0609,
      "theoretical_loss": 3.382164729333642,
      "tokens_seen": 2642935808
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020101107366393836,
      "loss": 0.066,
      "theoretical_loss": 3.3821384917267263,
      "tokens_seen": 2643197952
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020093082972235597,
      "loss": 0.0631,
      "theoretical_loss": 3.382112257450368,
      "tokens_seen": 2643460096
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020085058578077355,
      "loss": 0.0616,
      "theoretical_loss": 3.382086026503813,
      "tokens_seen": 2643722240
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020077034183919117,
      "loss": 0.0643,
      "theoretical_loss": 3.3820597988863095,
      "tokens_seen": 2643984384
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020069009789760873,
      "loss": 0.0658,
      "theoretical_loss": 3.382033574597105,
      "tokens_seen": 2644246528
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0002006098539560263,
      "loss": 0.0634,
      "theoretical_loss": 3.3820073536354465,
      "tokens_seen": 2644508672
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020052961001444392,
      "loss": 0.0652,
      "theoretical_loss": 3.381981136000582,
      "tokens_seen": 2644770816
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0002004493660728615,
      "loss": 0.0659,
      "theoretical_loss": 3.3819549216917606,
      "tokens_seen": 2645032960
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0002003691221312791,
      "loss": 0.0657,
      "theoretical_loss": 3.3819287107082303,
      "tokens_seen": 2645295104
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020028887818969668,
      "loss": 0.0654,
      "theoretical_loss": 3.38190250304924,
      "tokens_seen": 2645557248
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0002002086342481143,
      "loss": 0.0647,
      "theoretical_loss": 3.381876298714038,
      "tokens_seen": 2645819392
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020012839030653185,
      "loss": 0.0627,
      "theoretical_loss": 3.3818500977018746,
      "tokens_seen": 2646081536
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00020004814636494944,
      "loss": 0.0617,
      "theoretical_loss": 3.381823900011998,
      "tokens_seen": 2646343680
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00019996790242336705,
      "loss": 0.0614,
      "theoretical_loss": 3.381797705643659,
      "tokens_seen": 2646605824
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00019988765848178463,
      "loss": 0.0643,
      "theoretical_loss": 3.3817715145961067,
      "tokens_seen": 2646867968
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00019980741454020222,
      "loss": 0.0639,
      "theoretical_loss": 3.3817453268685918,
      "tokens_seen": 2647130112
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0001997271705986198,
      "loss": 0.0634,
      "theoretical_loss": 3.3817191424603648,
      "tokens_seen": 2647392256
    },
    {
      "epoch": 0.8,
      "objective/train/advantage_avg": 0.0011533270590007305,
      "objective/train/docs_used": 961712,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4470112323760986,
      "objective/train/original_loss": 1.4470112323760986,
      "objective/train/theoretical_loss": 3.3816929613706757,
      "objective/train/tokens_used": 2668114400,
      "objective/train/value_avg": -0.007129669189453125,
      "objective/train/value_loss": 0.00019181121024303138,
      "objective/train/value_max": -4.267692565917969e-05,
      "objective/train/value_min": -0.626953125,
      "objective/train/value_reward_corr": 0.645322537903916,
      "objective/train/value_std": 0.01311492919921875,
      "objective/train/weight_avg": 1.0012390613555908,
      "objective/train/weighted_lm_loss": 1.4489718675613403,
      "objective/train/weights_max": 1.632698893547058,
      "objective/train/weights_min": 0.3682536482810974,
      "theoretical_loss": 3.3816929613706757,
      "tokens_seen": 2647654400
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0001996469266570374,
      "loss": 0.0637,
      "theoretical_loss": 3.3816929613706757,
      "tokens_seen": 2647654400
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.000199566682715455,
      "loss": 0.0629,
      "theoretical_loss": 3.381666783598776,
      "tokens_seen": 2647916544
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00019948643877387256,
      "loss": 0.0628,
      "theoretical_loss": 3.381640609143917,
      "tokens_seen": 2648178688
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00019940619483229017,
      "loss": 0.0615,
      "theoretical_loss": 3.3816144380053497,
      "tokens_seen": 2648440832
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00019932595089070776,
      "loss": 0.0646,
      "theoretical_loss": 3.3815882701823257,
      "tokens_seen": 2648702976
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00019924570694912535,
      "loss": 0.0665,
      "theoretical_loss": 3.3815621056740968,
      "tokens_seen": 2648965120
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00019916546300754293,
      "loss": 0.0625,
      "theoretical_loss": 3.381535944479916,
      "tokens_seen": 2649227264
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00019908521906596052,
      "loss": 0.0666,
      "theoretical_loss": 3.3815097865990342,
      "tokens_seen": 2649489408
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00019900497512437813,
      "loss": 0.0639,
      "theoretical_loss": 3.3814836320307053,
      "tokens_seen": 2649751552
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0001989247311827957,
      "loss": 0.0656,
      "theoretical_loss": 3.381457480774182,
      "tokens_seen": 2650013696
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0001988444872412133,
      "loss": 0.0635,
      "theoretical_loss": 3.3814313328287167,
      "tokens_seen": 2650275840
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00019876424329963089,
      "loss": 0.0609,
      "theoretical_loss": 3.381405188193563,
      "tokens_seen": 2650537984
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00019868399935804847,
      "loss": 0.0651,
      "theoretical_loss": 3.381379046867975,
      "tokens_seen": 2650800128
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00019860375541646606,
      "loss": 0.0636,
      "theoretical_loss": 3.381352908851206,
      "tokens_seen": 2651062272
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00019852351147488364,
      "loss": 0.0616,
      "theoretical_loss": 3.38132677414251,
      "tokens_seen": 2651324416
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00019844326753330126,
      "loss": 0.0677,
      "theoretical_loss": 3.381300642741142,
      "tokens_seen": 2651586560
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00019836302359171881,
      "loss": 0.0642,
      "theoretical_loss": 3.3812745146463556,
      "tokens_seen": 2651848704
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00019828277965013643,
      "loss": 0.0634,
      "theoretical_loss": 3.381248389857406,
      "tokens_seen": 2652110848
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.000198202535708554,
      "loss": 0.0644,
      "theoretical_loss": 3.381222268373549,
      "tokens_seen": 2652372992
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0001981222917669716,
      "loss": 0.0654,
      "theoretical_loss": 3.3811961501940386,
      "tokens_seen": 2652635136
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00019804204782538918,
      "loss": 0.0631,
      "theoretical_loss": 3.3811700353181307,
      "tokens_seen": 2652897280
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00019796180388380677,
      "loss": 0.0603,
      "theoretical_loss": 3.3811439237450815,
      "tokens_seen": 2653159424
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00019788155994222438,
      "loss": 0.0623,
      "theoretical_loss": 3.381117815474147,
      "tokens_seen": 2653421568
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00019780131600064197,
      "loss": 0.064,
      "theoretical_loss": 3.381091710504583,
      "tokens_seen": 2653683712
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00019772107205905952,
      "loss": 0.0627,
      "theoretical_loss": 3.381065608835646,
      "tokens_seen": 2653945856
    },
    {
      "epoch": 0.8,
      "objective/train/advantage_avg": 0.00042152591049671173,
      "objective/train/docs_used": 964093,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.1780507564544678,
      "objective/train/original_loss": 1.1780505180358887,
      "objective/train/theoretical_loss": 3.381039510466593,
      "objective/train/tokens_used": 2674668000,
      "objective/train/value_avg": -0.01111602783203125,
      "objective/train/value_loss": 0.0005099286790937185,
      "objective/train/value_max": -4.57763671875e-05,
      "objective/train/value_min": -0.9345703125,
      "objective/train/value_reward_corr": 0.7172054336346624,
      "objective/train/value_std": 0.0220489501953125,
      "objective/train/weight_avg": 1.0006495714187622,
      "objective/train/weighted_lm_loss": 1.1777839660644531,
      "objective/train/weights_max": 1.7102184295654297,
      "objective/train/weights_min": 0.3874923884868622,
      "theoretical_loss": 3.381039510466593,
      "tokens_seen": 2654208000
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00019764082811747714,
      "loss": 0.0643,
      "theoretical_loss": 3.381039510466593,
      "tokens_seen": 2654208000
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00019756058417589472,
      "loss": 0.0641,
      "theoretical_loss": 3.3810134153966804,
      "tokens_seen": 2654470144
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0001974803402343123,
      "loss": 0.0612,
      "theoretical_loss": 3.3809873236251664,
      "tokens_seen": 2654732288
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0001974000962927299,
      "loss": 0.0655,
      "theoretical_loss": 3.380961235151308,
      "tokens_seen": 2654994432
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0001973198523511475,
      "loss": 0.0622,
      "theoretical_loss": 3.3809351499743623,
      "tokens_seen": 2655256576
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.0001972396084095651,
      "loss": 0.0647,
      "theoretical_loss": 3.3809090680935876,
      "tokens_seen": 2655518720
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00019715936446798265,
      "loss": 0.0657,
      "theoretical_loss": 3.3808829895082426,
      "tokens_seen": 2655780864
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00019707912052640026,
      "loss": 0.0658,
      "theoretical_loss": 3.380856914217585,
      "tokens_seen": 2656043008
    },
    {
      "epoch": 0.8,
      "learning_rate": 0.00019699887658481785,
      "loss": 0.0663,
      "theoretical_loss": 3.380830842220874,
      "tokens_seen": 2656305152
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019691863264323546,
      "loss": 0.0643,
      "theoretical_loss": 3.380804773517368,
      "tokens_seen": 2656567296
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019683838870165302,
      "loss": 0.0671,
      "theoretical_loss": 3.380778708106326,
      "tokens_seen": 2656829440
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001967581447600706,
      "loss": 0.0642,
      "theoretical_loss": 3.380752645987008,
      "tokens_seen": 2657091584
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019667790081848822,
      "loss": 0.0674,
      "theoretical_loss": 3.3807265871586734,
      "tokens_seen": 2657353728
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001965976568769058,
      "loss": 0.064,
      "theoretical_loss": 3.380700531620582,
      "tokens_seen": 2657615872
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001965174129353234,
      "loss": 0.0622,
      "theoretical_loss": 3.380674479371993,
      "tokens_seen": 2657878016
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019643716899374097,
      "loss": 0.065,
      "theoretical_loss": 3.3806484304121684,
      "tokens_seen": 2658140160
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001963569250521586,
      "loss": 0.0659,
      "theoretical_loss": 3.3806223847403674,
      "tokens_seen": 2658402304
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019627668111057615,
      "loss": 0.0657,
      "theoretical_loss": 3.380596342355852,
      "tokens_seen": 2658664448
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019619643716899373,
      "loss": 0.0674,
      "theoretical_loss": 3.3805703032578815,
      "tokens_seen": 2658926592
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019611619322741134,
      "loss": 0.0641,
      "theoretical_loss": 3.380544267445719,
      "tokens_seen": 2659188736
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019603594928582893,
      "loss": 0.063,
      "theoretical_loss": 3.380518234918625,
      "tokens_seen": 2659450880
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019595570534424651,
      "loss": 0.0664,
      "theoretical_loss": 3.3804922056758615,
      "tokens_seen": 2659713024
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001958754614026641,
      "loss": 0.0629,
      "theoretical_loss": 3.38046617971669,
      "tokens_seen": 2659975168
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019579521746108169,
      "loss": 0.0651,
      "theoretical_loss": 3.3804401570403737,
      "tokens_seen": 2660237312
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001957149735194993,
      "loss": 0.0658,
      "theoretical_loss": 3.3804141376461745,
      "tokens_seen": 2660499456
    },
    {
      "epoch": 0.81,
      "objective/train/advantage_avg": 1.0176291652896907e-05,
      "objective/train/docs_used": 966512,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2832056283950806,
      "objective/train/original_loss": 1.283205509185791,
      "objective/train/theoretical_loss": 3.3803881215333553,
      "objective/train/tokens_used": 2681221600,
      "objective/train/value_avg": -0.0079803466796875,
      "objective/train/value_loss": 0.00024036792456172407,
      "objective/train/value_max": -4.470348358154297e-05,
      "objective/train/value_min": -0.86083984375,
      "objective/train/value_reward_corr": 0.6866655599257911,
      "objective/train/value_std": 0.01390838623046875,
      "objective/train/weight_avg": 1.0001158714294434,
      "objective/train/weighted_lm_loss": 1.2830950021743774,
      "objective/train/weights_max": 1.3873459100723267,
      "objective/train/weights_min": 0.3687421381473541,
      "theoretical_loss": 3.3803881215333553,
      "tokens_seen": 2660761600
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019563472957791686,
      "loss": 0.0638,
      "theoretical_loss": 3.3803881215333553,
      "tokens_seen": 2660761600
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019555448563633447,
      "loss": 0.0635,
      "theoretical_loss": 3.3803621087011786,
      "tokens_seen": 2661023744
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019547424169475205,
      "loss": 0.0662,
      "theoretical_loss": 3.3803360991489084,
      "tokens_seen": 2661285888
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019539399775316964,
      "loss": 0.0668,
      "theoretical_loss": 3.3803100928758076,
      "tokens_seen": 2661548032
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019531375381158723,
      "loss": 0.0651,
      "theoretical_loss": 3.3802840898811395,
      "tokens_seen": 2661810176
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001952335098700048,
      "loss": 0.0667,
      "theoretical_loss": 3.380258090164169,
      "tokens_seen": 2662072320
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019515326592842242,
      "loss": 0.0666,
      "theoretical_loss": 3.380232093724159,
      "tokens_seen": 2662334464
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019507302198683998,
      "loss": 0.0648,
      "theoretical_loss": 3.380206100560375,
      "tokens_seen": 2662596608
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001949927780452576,
      "loss": 0.0686,
      "theoretical_loss": 3.380180110672081,
      "tokens_seen": 2662858752
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019491253410367518,
      "loss": 0.0653,
      "theoretical_loss": 3.3801541240585418,
      "tokens_seen": 2663120896
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019483229016209277,
      "loss": 0.0662,
      "theoretical_loss": 3.380128140719023,
      "tokens_seen": 2663383040
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019475204622051035,
      "loss": 0.063,
      "theoretical_loss": 3.3801021606527897,
      "tokens_seen": 2663645184
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019467180227892794,
      "loss": 0.0658,
      "theoretical_loss": 3.380076183859107,
      "tokens_seen": 2663907328
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019459155833734555,
      "loss": 0.0654,
      "theoretical_loss": 3.380050210337241,
      "tokens_seen": 2664169472
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001945113143957631,
      "loss": 0.0641,
      "theoretical_loss": 3.380024240086458,
      "tokens_seen": 2664431616
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019443107045418072,
      "loss": 0.0629,
      "theoretical_loss": 3.379998273106024,
      "tokens_seen": 2664693760
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001943508265125983,
      "loss": 0.0661,
      "theoretical_loss": 3.379972309395206,
      "tokens_seen": 2664955904
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001942705825710159,
      "loss": 0.0646,
      "theoretical_loss": 3.3799463489532697,
      "tokens_seen": 2665218048
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019419033862943348,
      "loss": 0.0635,
      "theoretical_loss": 3.379920391779483,
      "tokens_seen": 2665480192
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019411009468785106,
      "loss": 0.0638,
      "theoretical_loss": 3.379894437873113,
      "tokens_seen": 2665742336
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019402985074626867,
      "loss": 0.0649,
      "theoretical_loss": 3.3798684872334266,
      "tokens_seen": 2666004480
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019394960680468626,
      "loss": 0.0661,
      "theoretical_loss": 3.379842539859692,
      "tokens_seen": 2666266624
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019386936286310382,
      "loss": 0.0621,
      "theoretical_loss": 3.379816595751177,
      "tokens_seen": 2666528768
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019378911892152143,
      "loss": 0.064,
      "theoretical_loss": 3.3797906549071506,
      "tokens_seen": 2666790912
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019370887497993902,
      "loss": 0.0607,
      "theoretical_loss": 3.3797647173268794,
      "tokens_seen": 2667053056
    },
    {
      "epoch": 0.81,
      "objective/train/advantage_avg": 0.0007202328997664154,
      "objective/train/docs_used": 968905,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2583950757980347,
      "objective/train/original_loss": 1.2583949565887451,
      "objective/train/theoretical_loss": 3.3797387830096337,
      "objective/train/tokens_used": 2687775200,
      "objective/train/value_avg": -0.00569915771484375,
      "objective/train/value_loss": 0.0001516796910436824,
      "objective/train/value_max": -2.9981136322021484e-05,
      "objective/train/value_min": -0.6875,
      "objective/train/value_reward_corr": 0.6477400895991619,
      "objective/train/value_std": 0.01117706298828125,
      "objective/train/weight_avg": 1.000789761543274,
      "objective/train/weighted_lm_loss": 1.258969783782959,
      "objective/train/weights_max": 1.2237309217453003,
      "objective/train/weights_min": 0.38235902786254883,
      "theoretical_loss": 3.3797387830096337,
      "tokens_seen": 2667315200
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001936286310383566,
      "loss": 0.0653,
      "theoretical_loss": 3.3797387830096337,
      "tokens_seen": 2667315200
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001935483870967742,
      "loss": 0.0651,
      "theoretical_loss": 3.379712851954681,
      "tokens_seen": 2667577344
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001934681431551918,
      "loss": 0.0643,
      "theoretical_loss": 3.3796869241612915,
      "tokens_seen": 2667839488
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019338789921360939,
      "loss": 0.065,
      "theoretical_loss": 3.379660999628734,
      "tokens_seen": 2668101632
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019330765527202694,
      "loss": 0.0617,
      "theoretical_loss": 3.3796350783562783,
      "tokens_seen": 2668363776
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019322741133044456,
      "loss": 0.0636,
      "theoretical_loss": 3.379609160343194,
      "tokens_seen": 2668625920
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019314716738886214,
      "loss": 0.0673,
      "theoretical_loss": 3.3795832455887513,
      "tokens_seen": 2668888064
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019306692344727976,
      "loss": 0.0629,
      "theoretical_loss": 3.3795573340922207,
      "tokens_seen": 2669150208
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019298667950569731,
      "loss": 0.0638,
      "theoretical_loss": 3.379531425852872,
      "tokens_seen": 2669412352
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019290643556411493,
      "loss": 0.0644,
      "theoretical_loss": 3.379505520869977,
      "tokens_seen": 2669674496
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001928261916225325,
      "loss": 0.0628,
      "theoretical_loss": 3.379479619142806,
      "tokens_seen": 2669936640
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019274594768095007,
      "loss": 0.063,
      "theoretical_loss": 3.3794537206706297,
      "tokens_seen": 2670198784
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019266570373936768,
      "loss": 0.0628,
      "theoretical_loss": 3.379427825452721,
      "tokens_seen": 2670460928
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019258545979778527,
      "loss": 0.0661,
      "theoretical_loss": 3.3794019334883503,
      "tokens_seen": 2670723072
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019250521585620288,
      "loss": 0.0646,
      "theoretical_loss": 3.3793760447767904,
      "tokens_seen": 2670985216
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019242497191462044,
      "loss": 0.0659,
      "theoretical_loss": 3.3793501593173128,
      "tokens_seen": 2671247360
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019234472797303802,
      "loss": 0.0616,
      "theoretical_loss": 3.3793242771091903,
      "tokens_seen": 2671509504
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019226448403145564,
      "loss": 0.0632,
      "theoretical_loss": 3.3792983981516955,
      "tokens_seen": 2671771648
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019218424008987322,
      "loss": 0.0653,
      "theoretical_loss": 3.379272522444101,
      "tokens_seen": 2672033792
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001921039961482908,
      "loss": 0.0646,
      "theoretical_loss": 3.3792466499856806,
      "tokens_seen": 2672295936
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001920237522067084,
      "loss": 0.0636,
      "theoretical_loss": 3.3792207807757064,
      "tokens_seen": 2672558080
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.000191943508265126,
      "loss": 0.0633,
      "theoretical_loss": 3.379194914813453,
      "tokens_seen": 2672820224
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019186326432354356,
      "loss": 0.0642,
      "theoretical_loss": 3.379169052098194,
      "tokens_seen": 2673082368
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019178302038196115,
      "loss": 0.0665,
      "theoretical_loss": 3.3791431926292033,
      "tokens_seen": 2673344512
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019170277644037876,
      "loss": 0.0638,
      "theoretical_loss": 3.3791173364057547,
      "tokens_seen": 2673606656
    },
    {
      "epoch": 0.81,
      "objective/train/advantage_avg": 0.0004674495430663228,
      "objective/train/docs_used": 971344,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4315048456192017,
      "objective/train/original_loss": 1.4315046072006226,
      "objective/train/theoretical_loss": 3.3790914834271235,
      "objective/train/tokens_used": 2694328800,
      "objective/train/value_avg": -0.006275177001953125,
      "objective/train/value_loss": 0.00015117165457922965,
      "objective/train/value_max": -1.6450881958007812e-05,
      "objective/train/value_min": -0.2232666015625,
      "objective/train/value_reward_corr": 0.6798207075826601,
      "objective/train/value_std": 0.01152801513671875,
      "objective/train/weight_avg": 1.0005346536636353,
      "objective/train/weighted_lm_loss": 1.4325766563415527,
      "objective/train/weights_max": 1.1521174907684326,
      "objective/train/weights_min": 0.3680993318557739,
      "theoretical_loss": 3.3790914834271235,
      "tokens_seen": 2673868800
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019162253249879635,
      "loss": 0.0642,
      "theoretical_loss": 3.3790914834271235,
      "tokens_seen": 2673868800
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019154228855721393,
      "loss": 0.0659,
      "theoretical_loss": 3.3790656336925844,
      "tokens_seen": 2674130944
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019146204461563152,
      "loss": 0.0628,
      "theoretical_loss": 3.3790397872014117,
      "tokens_seen": 2674393088
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001913818006740491,
      "loss": 0.0624,
      "theoretical_loss": 3.379013943952881,
      "tokens_seen": 2674655232
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019130155673246672,
      "loss": 0.0665,
      "theoretical_loss": 3.3789881039462677,
      "tokens_seen": 2674917376
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019122131279088428,
      "loss": 0.0644,
      "theoretical_loss": 3.3789622671808477,
      "tokens_seen": 2675179520
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001911410688493019,
      "loss": 0.0644,
      "theoretical_loss": 3.3789364336558965,
      "tokens_seen": 2675441664
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019106082490771947,
      "loss": 0.0647,
      "theoretical_loss": 3.3789106033706906,
      "tokens_seen": 2675703808
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001909805809661371,
      "loss": 0.0617,
      "theoretical_loss": 3.378884776324506,
      "tokens_seen": 2675965952
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019090033702455465,
      "loss": 0.0636,
      "theoretical_loss": 3.3788589525166195,
      "tokens_seen": 2676228096
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019082009308297223,
      "loss": 0.067,
      "theoretical_loss": 3.378833131946308,
      "tokens_seen": 2676490240
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019073984914138984,
      "loss": 0.0655,
      "theoretical_loss": 3.378807314612848,
      "tokens_seen": 2676752384
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001906596051998074,
      "loss": 0.0678,
      "theoretical_loss": 3.378781500515518,
      "tokens_seen": 2677014528
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019057936125822501,
      "loss": 0.0661,
      "theoretical_loss": 3.3787556896535937,
      "tokens_seen": 2677276672
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001904991173166426,
      "loss": 0.0641,
      "theoretical_loss": 3.378729882026355,
      "tokens_seen": 2677538816
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019041887337506019,
      "loss": 0.0649,
      "theoretical_loss": 3.3787040776330786,
      "tokens_seen": 2677800960
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019033862943347777,
      "loss": 0.0643,
      "theoretical_loss": 3.3786782764730425,
      "tokens_seen": 2678063104
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019025838549189536,
      "loss": 0.065,
      "theoretical_loss": 3.3786524785455256,
      "tokens_seen": 2678325248
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019017814155031297,
      "loss": 0.0661,
      "theoretical_loss": 3.378626683849807,
      "tokens_seen": 2678587392
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019009789760873055,
      "loss": 0.0682,
      "theoretical_loss": 3.378600892385165,
      "tokens_seen": 2678849536
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00019001765366714814,
      "loss": 0.0671,
      "theoretical_loss": 3.378575104150879,
      "tokens_seen": 2679111680
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00018993740972556573,
      "loss": 0.0641,
      "theoretical_loss": 3.378549319146228,
      "tokens_seen": 2679373824
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001898571657839833,
      "loss": 0.0666,
      "theoretical_loss": 3.3785235373704925,
      "tokens_seen": 2679635968
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001897769218424009,
      "loss": 0.0639,
      "theoretical_loss": 3.3784977588229514,
      "tokens_seen": 2679898112
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00018969667790081848,
      "loss": 0.0651,
      "theoretical_loss": 3.378471983502885,
      "tokens_seen": 2680160256
    },
    {
      "epoch": 0.81,
      "objective/train/advantage_avg": 0.00010337762796552852,
      "objective/train/docs_used": 973756,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3529930114746094,
      "objective/train/original_loss": 1.3529927730560303,
      "objective/train/theoretical_loss": 3.378446211409574,
      "objective/train/tokens_used": 2700882400,
      "objective/train/value_avg": -0.01512908935546875,
      "objective/train/value_loss": 0.0003093662380706519,
      "objective/train/value_max": -2.5093555450439453e-05,
      "objective/train/value_min": -0.56396484375,
      "objective/train/value_reward_corr": 0.94607449504066,
      "objective/train/value_std": 0.051483154296875,
      "objective/train/weight_avg": 1.0002483129501343,
      "objective/train/weighted_lm_loss": 1.352254033088684,
      "objective/train/weights_max": 1.2205979824066162,
      "objective/train/weights_min": 0.3683340847492218,
      "theoretical_loss": 3.378446211409574,
      "tokens_seen": 2680422400
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001896164339592361,
      "loss": 0.0654,
      "theoretical_loss": 3.378446211409574,
      "tokens_seen": 2680422400
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00018953619001765368,
      "loss": 0.0634,
      "theoretical_loss": 3.378420442542299,
      "tokens_seen": 2680684544
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00018945594607607124,
      "loss": 0.0638,
      "theoretical_loss": 3.37839467690034,
      "tokens_seen": 2680946688
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00018937570213448885,
      "loss": 0.0679,
      "theoretical_loss": 3.3783689144829783,
      "tokens_seen": 2681208832
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00018929545819290644,
      "loss": 0.0644,
      "theoretical_loss": 3.3783431552894956,
      "tokens_seen": 2681470976
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00018921521425132405,
      "loss": 0.065,
      "theoretical_loss": 3.378317399319173,
      "tokens_seen": 2681733120
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001891349703097416,
      "loss": 0.063,
      "theoretical_loss": 3.378291646571292,
      "tokens_seen": 2681995264
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00018905472636815922,
      "loss": 0.0667,
      "theoretical_loss": 3.378265897045135,
      "tokens_seen": 2682257408
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001889744824265768,
      "loss": 0.0656,
      "theoretical_loss": 3.3782401507399835,
      "tokens_seen": 2682519552
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00018889423848499436,
      "loss": 0.0636,
      "theoretical_loss": 3.3782144076551206,
      "tokens_seen": 2682781696
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00018881399454341198,
      "loss": 0.0653,
      "theoretical_loss": 3.3781886677898285,
      "tokens_seen": 2683043840
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00018873375060182956,
      "loss": 0.065,
      "theoretical_loss": 3.3781629311433896,
      "tokens_seen": 2683305984
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00018865350666024717,
      "loss": 0.0639,
      "theoretical_loss": 3.378137197715088,
      "tokens_seen": 2683568128
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00018857326271866473,
      "loss": 0.0645,
      "theoretical_loss": 3.3781114675042065,
      "tokens_seen": 2683830272
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00018849301877708232,
      "loss": 0.0671,
      "theoretical_loss": 3.3780857405100284,
      "tokens_seen": 2684092416
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00018841277483549993,
      "loss": 0.0664,
      "theoretical_loss": 3.3780600167318378,
      "tokens_seen": 2684354560
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00018833253089391752,
      "loss": 0.0646,
      "theoretical_loss": 3.3780342961689183,
      "tokens_seen": 2684616704
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001882522869523351,
      "loss": 0.0648,
      "theoretical_loss": 3.3780085788205545,
      "tokens_seen": 2684878848
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001881720430107527,
      "loss": 0.0658,
      "theoretical_loss": 3.3779828646860306,
      "tokens_seen": 2685140992
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001880917990691703,
      "loss": 0.0651,
      "theoretical_loss": 3.377957153764631,
      "tokens_seen": 2685403136
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00018801155512758786,
      "loss": 0.0627,
      "theoretical_loss": 3.3779314460556416,
      "tokens_seen": 2685665280
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00018793131118600544,
      "loss": 0.0652,
      "theoretical_loss": 3.3779057415583464,
      "tokens_seen": 2685927424
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00018785106724442306,
      "loss": 0.0636,
      "theoretical_loss": 3.3778800402720313,
      "tokens_seen": 2686189568
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00018777082330284064,
      "loss": 0.0664,
      "theoretical_loss": 3.377854342195982,
      "tokens_seen": 2686451712
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00018769057936125823,
      "loss": 0.0633,
      "theoretical_loss": 3.3778286473294843,
      "tokens_seen": 2686713856
    },
    {
      "epoch": 0.81,
      "objective/train/advantage_avg": 0.0004787496873177588,
      "objective/train/docs_used": 976171,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2660490274429321,
      "objective/train/original_loss": 1.2660491466522217,
      "objective/train/theoretical_loss": 3.377802955671824,
      "objective/train/tokens_used": 2707436000,
      "objective/train/value_avg": -0.0075531005859375,
      "objective/train/value_loss": 0.00026115079526789486,
      "objective/train/value_max": -3.427267074584961e-05,
      "objective/train/value_min": -0.9697265625,
      "objective/train/value_reward_corr": 0.7488763443348854,
      "objective/train/value_std": 0.01806640625,
      "objective/train/weight_avg": 1.0005987882614136,
      "objective/train/weighted_lm_loss": 1.2669463157653809,
      "objective/train/weights_max": 2.38258957862854,
      "objective/train/weights_min": 0.36950811743736267,
      "theoretical_loss": 3.377802955671824,
      "tokens_seen": 2686976000
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00018761033541967581,
      "loss": 0.0651,
      "theoretical_loss": 3.377802955671824,
      "tokens_seen": 2686976000
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.0001875300914780934,
      "loss": 0.064,
      "theoretical_loss": 3.377777267222287,
      "tokens_seen": 2687238144
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.000187449847536511,
      "loss": 0.0642,
      "theoretical_loss": 3.377751581980161,
      "tokens_seen": 2687500288
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00018736960359492857,
      "loss": 0.0645,
      "theoretical_loss": 3.3777258999447315,
      "tokens_seen": 2687762432
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00018728935965334618,
      "loss": 0.0668,
      "theoretical_loss": 3.377700221115286,
      "tokens_seen": 2688024576
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00018720911571176377,
      "loss": 0.0647,
      "theoretical_loss": 3.3776745454911117,
      "tokens_seen": 2688286720
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00018712887177018135,
      "loss": 0.0662,
      "theoretical_loss": 3.377648873071496,
      "tokens_seen": 2688548864
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00018704862782859894,
      "loss": 0.0648,
      "theoretical_loss": 3.377623203855727,
      "tokens_seen": 2688811008
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00018696838388701652,
      "loss": 0.0663,
      "theoretical_loss": 3.3775975378430916,
      "tokens_seen": 2689073152
    },
    {
      "epoch": 0.81,
      "learning_rate": 0.00018688813994543414,
      "loss": 0.0628,
      "theoretical_loss": 3.3775718750328783,
      "tokens_seen": 2689335296
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001868078960038517,
      "loss": 0.0659,
      "theoretical_loss": 3.3775462154243754,
      "tokens_seen": 2689597440
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001867276520622693,
      "loss": 0.0649,
      "theoretical_loss": 3.3775205590168715,
      "tokens_seen": 2689859584
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001866474081206869,
      "loss": 0.0644,
      "theoretical_loss": 3.377494905809656,
      "tokens_seen": 2690121728
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018656716417910448,
      "loss": 0.0641,
      "theoretical_loss": 3.377469255802017,
      "tokens_seen": 2690383872
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018648692023752206,
      "loss": 0.0647,
      "theoretical_loss": 3.3774436089932443,
      "tokens_seen": 2690646016
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018640667629593965,
      "loss": 0.0641,
      "theoretical_loss": 3.3774179653826266,
      "tokens_seen": 2690908160
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018632643235435726,
      "loss": 0.0644,
      "theoretical_loss": 3.3773923249694544,
      "tokens_seen": 2691170304
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018624618841277485,
      "loss": 0.0628,
      "theoretical_loss": 3.3773666877530175,
      "tokens_seen": 2691432448
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018616594447119243,
      "loss": 0.0635,
      "theoretical_loss": 3.377341053732606,
      "tokens_seen": 2691694592
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018608570052961002,
      "loss": 0.0657,
      "theoretical_loss": 3.37731542290751,
      "tokens_seen": 2691956736
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001860054565880276,
      "loss": 0.0667,
      "theoretical_loss": 3.37728979527702,
      "tokens_seen": 2692218880
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001859252126464452,
      "loss": 0.0612,
      "theoretical_loss": 3.3772641708404274,
      "tokens_seen": 2692481024
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018584496870486278,
      "loss": 0.0636,
      "theoretical_loss": 3.3772385495970227,
      "tokens_seen": 2692743168
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001857647247632804,
      "loss": 0.0639,
      "theoretical_loss": 3.3772129315460977,
      "tokens_seen": 2693005312
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018568448082169797,
      "loss": 0.0644,
      "theoretical_loss": 3.3771873166869435,
      "tokens_seen": 2693267456
    },
    {
      "epoch": 0.82,
      "objective/train/advantage_avg": 0.0006705496925860643,
      "objective/train/docs_used": 978638,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3295810222625732,
      "objective/train/original_loss": 1.3295810222625732,
      "objective/train/theoretical_loss": 3.377161705018852,
      "objective/train/tokens_used": 2713989600,
      "objective/train/value_avg": -0.00616455078125,
      "objective/train/value_loss": 0.0002441635588183999,
      "objective/train/value_max": -1.806020736694336e-05,
      "objective/train/value_min": -0.8525390625,
      "objective/train/value_reward_corr": 0.7188453937477017,
      "objective/train/value_std": 0.0169830322265625,
      "objective/train/weight_avg": 1.0007843971252441,
      "objective/train/weighted_lm_loss": 1.33126962184906,
      "objective/train/weights_max": 1.6763864755630493,
      "objective/train/weights_min": 0.3707548677921295,
      "theoretical_loss": 3.377161705018852,
      "tokens_seen": 2693529600
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018560423688011553,
      "loss": 0.0658,
      "theoretical_loss": 3.377161705018852,
      "tokens_seen": 2693529600
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018552399293853315,
      "loss": 0.066,
      "theoretical_loss": 3.377136096541115,
      "tokens_seen": 2693791744
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018544374899695073,
      "loss": 0.0653,
      "theoretical_loss": 3.377110491253024,
      "tokens_seen": 2694053888
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018536350505536834,
      "loss": 0.0638,
      "theoretical_loss": 3.377084889153873,
      "tokens_seen": 2694316032
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001852832611137859,
      "loss": 0.064,
      "theoretical_loss": 3.3770592902429537,
      "tokens_seen": 2694578176
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018520301717220351,
      "loss": 0.0665,
      "theoretical_loss": 3.377033694519559,
      "tokens_seen": 2694840320
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001851227732306211,
      "loss": 0.0639,
      "theoretical_loss": 3.377008101982982,
      "tokens_seen": 2695102464
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018504252928903866,
      "loss": 0.0668,
      "theoretical_loss": 3.3769825126325164,
      "tokens_seen": 2695364608
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018496228534745627,
      "loss": 0.0652,
      "theoretical_loss": 3.376956926467455,
      "tokens_seen": 2695626752
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018488204140587386,
      "loss": 0.066,
      "theoretical_loss": 3.376931343487092,
      "tokens_seen": 2695888896
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018480179746429147,
      "loss": 0.0645,
      "theoretical_loss": 3.3769057636907216,
      "tokens_seen": 2696151040
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018472155352270903,
      "loss": 0.0644,
      "theoretical_loss": 3.3768801870776377,
      "tokens_seen": 2696413184
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001846413095811266,
      "loss": 0.0639,
      "theoretical_loss": 3.3768546136471347,
      "tokens_seen": 2696675328
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018456106563954423,
      "loss": 0.0675,
      "theoretical_loss": 3.376829043398507,
      "tokens_seen": 2696937472
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001844808216979618,
      "loss": 0.064,
      "theoretical_loss": 3.37680347633105,
      "tokens_seen": 2697199616
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001844005777563794,
      "loss": 0.0677,
      "theoretical_loss": 3.3767779124440587,
      "tokens_seen": 2697461760
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018432033381479698,
      "loss": 0.0664,
      "theoretical_loss": 3.3767523517368288,
      "tokens_seen": 2697723904
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001842400898732146,
      "loss": 0.0639,
      "theoretical_loss": 3.376726794208655,
      "tokens_seen": 2697986048
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018415984593163215,
      "loss": 0.0621,
      "theoretical_loss": 3.3767012398588334,
      "tokens_seen": 2698248192
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018407960199004974,
      "loss": 0.0636,
      "theoretical_loss": 3.3766756886866602,
      "tokens_seen": 2698510336
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018399935804846735,
      "loss": 0.0642,
      "theoretical_loss": 3.3766501406914315,
      "tokens_seen": 2698772480
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018391911410688494,
      "loss": 0.0664,
      "theoretical_loss": 3.376624595872444,
      "tokens_seen": 2699034624
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018383887016530252,
      "loss": 0.0625,
      "theoretical_loss": 3.376599054228994,
      "tokens_seen": 2699296768
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001837586262237201,
      "loss": 0.0627,
      "theoretical_loss": 3.3765735157603785,
      "tokens_seen": 2699558912
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018367838228213772,
      "loss": 0.0667,
      "theoretical_loss": 3.376547980465895,
      "tokens_seen": 2699821056
    },
    {
      "epoch": 0.82,
      "objective/train/advantage_avg": -0.0002805391850415617,
      "objective/train/docs_used": 981079,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3751705884933472,
      "objective/train/original_loss": 1.3751707077026367,
      "objective/train/theoretical_loss": 3.3765224483448404,
      "objective/train/tokens_used": 2720543200,
      "objective/train/value_avg": -0.0082244873046875,
      "objective/train/value_loss": 0.0004667190951295197,
      "objective/train/value_max": -2.5451183319091797e-05,
      "objective/train/value_min": -0.94189453125,
      "objective/train/value_reward_corr": 0.7052109388437202,
      "objective/train/value_std": 0.02203369140625,
      "objective/train/weight_avg": 0.9999330639839172,
      "objective/train/weighted_lm_loss": 1.374388575553894,
      "objective/train/weights_max": 2.1484720706939697,
      "objective/train/weights_min": 0.3696533143520355,
      "theoretical_loss": 3.3765224483448404,
      "tokens_seen": 2700083200
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001835981383405553,
      "loss": 0.0656,
      "theoretical_loss": 3.3765224483448404,
      "tokens_seen": 2700083200
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018351789439897286,
      "loss": 0.0629,
      "theoretical_loss": 3.3764969193965126,
      "tokens_seen": 2700345344
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018343765045739048,
      "loss": 0.066,
      "theoretical_loss": 3.376471393620209,
      "tokens_seen": 2700607488
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018335740651580806,
      "loss": 0.0653,
      "theoretical_loss": 3.3764458710152283,
      "tokens_seen": 2700869632
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018327716257422565,
      "loss": 0.0645,
      "theoretical_loss": 3.376420351580868,
      "tokens_seen": 2701131776
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018319691863264323,
      "loss": 0.0657,
      "theoretical_loss": 3.376394835316427,
      "tokens_seen": 2701393920
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018311667469106082,
      "loss": 0.066,
      "theoretical_loss": 3.3763693222212035,
      "tokens_seen": 2701656064
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018303643074947843,
      "loss": 0.0647,
      "theoretical_loss": 3.3763438122944973,
      "tokens_seen": 2701918208
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.000182956186807896,
      "loss": 0.0636,
      "theoretical_loss": 3.3763183055356065,
      "tokens_seen": 2702180352
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001828759428663136,
      "loss": 0.0634,
      "theoretical_loss": 3.3762928019438316,
      "tokens_seen": 2702442496
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001827956989247312,
      "loss": 0.065,
      "theoretical_loss": 3.376267301518471,
      "tokens_seen": 2702704640
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001827154549831488,
      "loss": 0.0658,
      "theoretical_loss": 3.3762418042588256,
      "tokens_seen": 2702966784
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018263521104156636,
      "loss": 0.0666,
      "theoretical_loss": 3.3762163101641947,
      "tokens_seen": 2703228928
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018255496709998394,
      "loss": 0.064,
      "theoretical_loss": 3.3761908192338788,
      "tokens_seen": 2703491072
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018247472315840156,
      "loss": 0.0638,
      "theoretical_loss": 3.3761653314671785,
      "tokens_seen": 2703753216
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018239447921681912,
      "loss": 0.0661,
      "theoretical_loss": 3.376139846863394,
      "tokens_seen": 2704015360
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018231423527523673,
      "loss": 0.0674,
      "theoretical_loss": 3.3761143654218264,
      "tokens_seen": 2704277504
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018223399133365431,
      "loss": 0.0637,
      "theoretical_loss": 3.3760888871417776,
      "tokens_seen": 2704539648
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001821537473920719,
      "loss": 0.0611,
      "theoretical_loss": 3.376063412022548,
      "tokens_seen": 2704801792
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018207350345048948,
      "loss": 0.0653,
      "theoretical_loss": 3.3760379400634397,
      "tokens_seen": 2705063936
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018199325950890707,
      "loss": 0.0649,
      "theoretical_loss": 3.3760124712637545,
      "tokens_seen": 2705326080
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018191301556732468,
      "loss": 0.066,
      "theoretical_loss": 3.375987005622794,
      "tokens_seen": 2705588224
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018183277162574227,
      "loss": 0.0671,
      "theoretical_loss": 3.3759615431398613,
      "tokens_seen": 2705850368
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018175252768415985,
      "loss": 0.0642,
      "theoretical_loss": 3.3759360838142576,
      "tokens_seen": 2706112512
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018167228374257744,
      "loss": 0.0632,
      "theoretical_loss": 3.375910627645287,
      "tokens_seen": 2706374656
    },
    {
      "epoch": 0.82,
      "objective/train/advantage_avg": -0.000171054300153628,
      "objective/train/docs_used": 983478,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2702794075012207,
      "objective/train/original_loss": 1.2702794075012207,
      "objective/train/theoretical_loss": 3.375885174632251,
      "objective/train/tokens_used": 2727096800,
      "objective/train/value_avg": -0.0064697265625,
      "objective/train/value_loss": 0.00020486005814746022,
      "objective/train/value_max": -2.2292137145996094e-05,
      "objective/train/value_min": -0.6494140625,
      "objective/train/value_reward_corr": 0.7068462014367565,
      "objective/train/value_std": 0.01386260986328125,
      "objective/train/weight_avg": 0.9999198913574219,
      "objective/train/weighted_lm_loss": 1.269945740699768,
      "objective/train/weights_max": 1.7154713869094849,
      "objective/train/weights_min": 0.23434561491012573,
      "theoretical_loss": 3.375885174632251,
      "tokens_seen": 2706636800
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018159203980099502,
      "loss": 0.0658,
      "theoretical_loss": 3.375885174632251,
      "tokens_seen": 2706636800
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018151179585941264,
      "loss": 0.0671,
      "theoretical_loss": 3.375859724774454,
      "tokens_seen": 2706898944
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001814315519178302,
      "loss": 0.0663,
      "theoretical_loss": 3.375834278071199,
      "tokens_seen": 2707161088
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001813513079762478,
      "loss": 0.0657,
      "theoretical_loss": 3.375808834521789,
      "tokens_seen": 2707423232
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001812710640346654,
      "loss": 0.0644,
      "theoretical_loss": 3.375783394125529,
      "tokens_seen": 2707685376
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018119082009308295,
      "loss": 0.0696,
      "theoretical_loss": 3.375757956881721,
      "tokens_seen": 2707947520
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018111057615150057,
      "loss": 0.0715,
      "theoretical_loss": 3.3757325227896713,
      "tokens_seen": 2708209664
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018103033220991815,
      "loss": 0.0688,
      "theoretical_loss": 3.3757070918486836,
      "tokens_seen": 2708471808
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018095008826833576,
      "loss": 0.069,
      "theoretical_loss": 3.3756816640580625,
      "tokens_seen": 2708733952
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018086984432675332,
      "loss": 0.0704,
      "theoretical_loss": 3.375656239417113,
      "tokens_seen": 2708996096
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018078960038517093,
      "loss": 0.0658,
      "theoretical_loss": 3.37563081792514,
      "tokens_seen": 2709258240
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018070935644358852,
      "loss": 0.0681,
      "theoretical_loss": 3.375605399581449,
      "tokens_seen": 2709520384
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001806291125020061,
      "loss": 0.067,
      "theoretical_loss": 3.375579984385346,
      "tokens_seen": 2709782528
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001805488685604237,
      "loss": 0.0641,
      "theoretical_loss": 3.375554572336136,
      "tokens_seen": 2710044672
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018046862461884128,
      "loss": 0.0674,
      "theoretical_loss": 3.3755291634331255,
      "tokens_seen": 2710306816
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001803883806772589,
      "loss": 0.0677,
      "theoretical_loss": 3.3755037576756206,
      "tokens_seen": 2710568960
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018030813673567645,
      "loss": 0.0686,
      "theoretical_loss": 3.3754783550629277,
      "tokens_seen": 2710831104
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018022789279409403,
      "loss": 0.0653,
      "theoretical_loss": 3.375452955594354,
      "tokens_seen": 2711093248
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018014764885251165,
      "loss": 0.069,
      "theoretical_loss": 3.3754275592692053,
      "tokens_seen": 2711355392
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00018006740491092923,
      "loss": 0.0663,
      "theoretical_loss": 3.37540216608679,
      "tokens_seen": 2711617536
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00017998716096934682,
      "loss": 0.0644,
      "theoretical_loss": 3.3753767760464144,
      "tokens_seen": 2711879680
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001799069170277644,
      "loss": 0.0679,
      "theoretical_loss": 3.3753513891473865,
      "tokens_seen": 2712141824
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00017982667308618201,
      "loss": 0.0647,
      "theoretical_loss": 3.375326005389014,
      "tokens_seen": 2712403968
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001797464291445996,
      "loss": 0.0712,
      "theoretical_loss": 3.3753006247706048,
      "tokens_seen": 2712666112
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00017966618520301716,
      "loss": 0.0661,
      "theoretical_loss": 3.3752752472914676,
      "tokens_seen": 2712928256
    },
    {
      "epoch": 0.82,
      "objective/train/advantage_avg": 0.00032114412169903517,
      "objective/train/docs_used": 985832,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3805224895477295,
      "objective/train/original_loss": 1.3805222511291504,
      "objective/train/theoretical_loss": 3.37524987295091,
      "objective/train/tokens_used": 2733650400,
      "objective/train/value_avg": -0.00806427001953125,
      "objective/train/value_loss": 0.0002658286248333752,
      "objective/train/value_max": -3.427267074584961e-05,
      "objective/train/value_min": -0.94384765625,
      "objective/train/value_reward_corr": 0.7371548868423696,
      "objective/train/value_std": 0.0176239013671875,
      "objective/train/weight_avg": 1.0004411935806274,
      "objective/train/weighted_lm_loss": 1.380434513092041,
      "objective/train/weights_max": 1.8222967386245728,
      "objective/train/weights_min": 0.3710845410823822,
      "theoretical_loss": 3.37524987295091,
      "tokens_seen": 2713190400
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00017958594126143477,
      "loss": 0.0704,
      "theoretical_loss": 3.37524987295091,
      "tokens_seen": 2713190400
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00017950569731985236,
      "loss": 0.0649,
      "theoretical_loss": 3.375224501748241,
      "tokens_seen": 2713452544
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00017942545337826994,
      "loss": 0.0666,
      "theoretical_loss": 3.3751991336827696,
      "tokens_seen": 2713714688
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00017934520943668753,
      "loss": 0.0657,
      "theoretical_loss": 3.375173768753805,
      "tokens_seen": 2713976832
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001792649654951051,
      "loss": 0.0627,
      "theoretical_loss": 3.3751484069606565,
      "tokens_seen": 2714238976
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00017918472155352273,
      "loss": 0.068,
      "theoretical_loss": 3.375123048302633,
      "tokens_seen": 2714501120
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00017910447761194028,
      "loss": 0.0645,
      "theoretical_loss": 3.375097692779045,
      "tokens_seen": 2714763264
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001790242336703579,
      "loss": 0.0657,
      "theoretical_loss": 3.3750723403892016,
      "tokens_seen": 2715025408
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00017894398972877548,
      "loss": 0.0634,
      "theoretical_loss": 3.375046991132414,
      "tokens_seen": 2715287552
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001788637457871931,
      "loss": 0.068,
      "theoretical_loss": 3.375021645007992,
      "tokens_seen": 2715549696
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00017878350184561065,
      "loss": 0.0681,
      "theoretical_loss": 3.3749963020152465,
      "tokens_seen": 2715811840
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00017870325790402824,
      "loss": 0.0656,
      "theoretical_loss": 3.3749709621534882,
      "tokens_seen": 2716073984
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00017862301396244585,
      "loss": 0.0673,
      "theoretical_loss": 3.374945625422028,
      "tokens_seen": 2716336128
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001785427700208634,
      "loss": 0.0654,
      "theoretical_loss": 3.374920291820177,
      "tokens_seen": 2716598272
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00017846252607928102,
      "loss": 0.0656,
      "theoretical_loss": 3.374894961347247,
      "tokens_seen": 2716860416
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001783822821376986,
      "loss": 0.0711,
      "theoretical_loss": 3.37486963400255,
      "tokens_seen": 2717122560
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001783020381961162,
      "loss": 0.072,
      "theoretical_loss": 3.3748443097853973,
      "tokens_seen": 2717384704
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00017822179425453378,
      "loss": 0.0672,
      "theoretical_loss": 3.374818988695101,
      "tokens_seen": 2717646848
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00017814155031295136,
      "loss": 0.0648,
      "theoretical_loss": 3.3747936707309747,
      "tokens_seen": 2717908992
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00017806130637136898,
      "loss": 0.0646,
      "theoretical_loss": 3.3747683558923294,
      "tokens_seen": 2718171136
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00017798106242978656,
      "loss": 0.0682,
      "theoretical_loss": 3.3747430441784787,
      "tokens_seen": 2718433280
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00017790081848820415,
      "loss": 0.0653,
      "theoretical_loss": 3.3747177355887352,
      "tokens_seen": 2718695424
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00017782057454662173,
      "loss": 0.0622,
      "theoretical_loss": 3.374692430122413,
      "tokens_seen": 2718957568
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00017774033060503932,
      "loss": 0.0681,
      "theoretical_loss": 3.3746671277788245,
      "tokens_seen": 2719219712
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001776600866634569,
      "loss": 0.0678,
      "theoretical_loss": 3.3746418285572837,
      "tokens_seen": 2719481856
    },
    {
      "epoch": 0.82,
      "objective/train/advantage_avg": -0.0001307609345531091,
      "objective/train/docs_used": 988161,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2664278745651245,
      "objective/train/original_loss": 1.266427755355835,
      "objective/train/theoretical_loss": 3.3746165324571047,
      "objective/train/tokens_used": 2740204000,
      "objective/train/value_avg": -0.00821685791015625,
      "objective/train/value_loss": 0.00019688457541633397,
      "objective/train/value_max": -2.9981136322021484e-05,
      "objective/train/value_min": -0.94091796875,
      "objective/train/value_reward_corr": 0.8155150163275279,
      "objective/train/value_std": 0.0184478759765625,
      "objective/train/weight_avg": 0.9999621510505676,
      "objective/train/weighted_lm_loss": 1.2667092084884644,
      "objective/train/weights_max": 1.6394892930984497,
      "objective/train/weights_min": 0.38045015931129456,
      "theoretical_loss": 3.3746165324571047,
      "tokens_seen": 2719744000
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001775798427218745,
      "loss": 0.0647,
      "theoretical_loss": 3.3746165324571047,
      "tokens_seen": 2719744000
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001774995987802921,
      "loss": 0.0657,
      "theoretical_loss": 3.374591239477602,
      "tokens_seen": 2720006144
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001774193548387097,
      "loss": 0.0698,
      "theoretical_loss": 3.3745659496180886,
      "tokens_seen": 2720268288
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00017733911089712725,
      "loss": 0.0684,
      "theoretical_loss": 3.3745406628778802,
      "tokens_seen": 2720530432
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00017725886695554486,
      "loss": 0.0684,
      "theoretical_loss": 3.374515379256291,
      "tokens_seen": 2720792576
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00017717862301396244,
      "loss": 0.0639,
      "theoretical_loss": 3.374490098752636,
      "tokens_seen": 2721054720
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00017709837907238006,
      "loss": 0.0657,
      "theoretical_loss": 3.3744648213662307,
      "tokens_seen": 2721316864
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00017701813513079762,
      "loss": 0.0673,
      "theoretical_loss": 3.37443954709639,
      "tokens_seen": 2721579008
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00017693789118921523,
      "loss": 0.0639,
      "theoretical_loss": 3.37441427594243,
      "tokens_seen": 2721841152
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.00017685764724763281,
      "loss": 0.0667,
      "theoretical_loss": 3.3743890079036665,
      "tokens_seen": 2722103296
    },
    {
      "epoch": 0.82,
      "learning_rate": 0.0001767774033060504,
      "loss": 0.0619,
      "theoretical_loss": 3.374363742979415,
      "tokens_seen": 2722365440
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017669715936446798,
      "loss": 0.0664,
      "theoretical_loss": 3.3743384811689925,
      "tokens_seen": 2722627584
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017661691542288557,
      "loss": 0.0641,
      "theoretical_loss": 3.3743132224717147,
      "tokens_seen": 2722889728
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017653667148130318,
      "loss": 0.0646,
      "theoretical_loss": 3.374287966886899,
      "tokens_seen": 2723151872
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017645642753972074,
      "loss": 0.0635,
      "theoretical_loss": 3.3742627144138617,
      "tokens_seen": 2723414016
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017637618359813833,
      "loss": 0.067,
      "theoretical_loss": 3.3742374650519205,
      "tokens_seen": 2723676160
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017629593965655594,
      "loss": 0.0651,
      "theoretical_loss": 3.3742122188003925,
      "tokens_seen": 2723938304
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017621569571497352,
      "loss": 0.0648,
      "theoretical_loss": 3.3741869756585947,
      "tokens_seen": 2724200448
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001761354517733911,
      "loss": 0.0638,
      "theoretical_loss": 3.374161735625846,
      "tokens_seen": 2724462592
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001760552078318087,
      "loss": 0.0653,
      "theoretical_loss": 3.3741364987014633,
      "tokens_seen": 2724724736
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001759749638902263,
      "loss": 0.0672,
      "theoretical_loss": 3.3741112648847653,
      "tokens_seen": 2724986880
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001758947199486439,
      "loss": 0.0636,
      "theoretical_loss": 3.374086034175071,
      "tokens_seen": 2725249024
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017581447600706145,
      "loss": 0.0659,
      "theoretical_loss": 3.3740608065716975,
      "tokens_seen": 2725511168
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017573423206547907,
      "loss": 0.065,
      "theoretical_loss": 3.374035582073965,
      "tokens_seen": 2725773312
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017565398812389665,
      "loss": 0.0638,
      "theoretical_loss": 3.374010360681192,
      "tokens_seen": 2726035456
    },
    {
      "epoch": 0.83,
      "objective/train/advantage_avg": 0.00022230023751035333,
      "objective/train/docs_used": 990556,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2834949493408203,
      "objective/train/original_loss": 1.2834949493408203,
      "objective/train/theoretical_loss": 3.373985142392698,
      "objective/train/tokens_used": 2746757600,
      "objective/train/value_avg": -0.007114410400390625,
      "objective/train/value_loss": 0.00023567344760522246,
      "objective/train/value_max": -2.753734588623047e-05,
      "objective/train/value_min": -0.27734375,
      "objective/train/value_reward_corr": 0.7233507761066421,
      "objective/train/value_std": 0.016265869140625,
      "objective/train/weight_avg": 1.0003267526626587,
      "objective/train/weighted_lm_loss": 1.2832162380218506,
      "objective/train/weights_max": 1.2012343406677246,
      "objective/train/weights_min": 0.36834847927093506,
      "theoretical_loss": 3.373985142392698,
      "tokens_seen": 2726297600
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017557374418231424,
      "loss": 0.065,
      "theoretical_loss": 3.373985142392698,
      "tokens_seen": 2726297600
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017549350024073182,
      "loss": 0.0668,
      "theoretical_loss": 3.3739599272078022,
      "tokens_seen": 2726559744
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017541325629914943,
      "loss": 0.0653,
      "theoretical_loss": 3.3739347151258245,
      "tokens_seen": 2726821888
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017533301235756702,
      "loss": 0.0641,
      "theoretical_loss": 3.3739095061460853,
      "tokens_seen": 2727084032
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017525276841598458,
      "loss": 0.0653,
      "theoretical_loss": 3.3738843002679038,
      "tokens_seen": 2727346176
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001751725244744022,
      "loss": 0.0671,
      "theoretical_loss": 3.3738590974906004,
      "tokens_seen": 2727608320
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017509228053281978,
      "loss": 0.0652,
      "theoretical_loss": 3.3738338978134967,
      "tokens_seen": 2727870464
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001750120365912374,
      "loss": 0.0691,
      "theoretical_loss": 3.3738087012359124,
      "tokens_seen": 2728132608
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017493179264965495,
      "loss": 0.0657,
      "theoretical_loss": 3.3737835077571687,
      "tokens_seen": 2728394752
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017485154870807253,
      "loss": 0.0655,
      "theoretical_loss": 3.3737583173765873,
      "tokens_seen": 2728656896
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017477130476649015,
      "loss": 0.066,
      "theoretical_loss": 3.3737331300934894,
      "tokens_seen": 2728919040
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001746910608249077,
      "loss": 0.0654,
      "theoretical_loss": 3.3737079459071966,
      "tokens_seen": 2729181184
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017461081688332532,
      "loss": 0.0659,
      "theoretical_loss": 3.37368276481703,
      "tokens_seen": 2729443328
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001745305729417429,
      "loss": 0.0635,
      "theoretical_loss": 3.373657586822313,
      "tokens_seen": 2729705472
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017445032900016051,
      "loss": 0.0679,
      "theoretical_loss": 3.373632411922367,
      "tokens_seen": 2729967616
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017437008505857807,
      "loss": 0.0665,
      "theoretical_loss": 3.373607240116515,
      "tokens_seen": 2730229760
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017428984111699566,
      "loss": 0.0638,
      "theoretical_loss": 3.373582071404079,
      "tokens_seen": 2730491904
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017420959717541327,
      "loss": 0.0693,
      "theoretical_loss": 3.3735569057843824,
      "tokens_seen": 2730754048
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017412935323383086,
      "loss": 0.0667,
      "theoretical_loss": 3.373531743256748,
      "tokens_seen": 2731016192
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017404910929224844,
      "loss": 0.064,
      "theoretical_loss": 3.3735065838205,
      "tokens_seen": 2731278336
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017396886535066603,
      "loss": 0.0663,
      "theoretical_loss": 3.3734814274749607,
      "tokens_seen": 2731540480
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001738886214090836,
      "loss": 0.0649,
      "theoretical_loss": 3.3734562742194547,
      "tokens_seen": 2731802624
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001738083774675012,
      "loss": 0.0666,
      "theoretical_loss": 3.373431124053306,
      "tokens_seen": 2732064768
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017372813352591878,
      "loss": 0.0654,
      "theoretical_loss": 3.373405976975838,
      "tokens_seen": 2732326912
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001736478895843364,
      "loss": 0.0672,
      "theoretical_loss": 3.373380832986376,
      "tokens_seen": 2732589056
    },
    {
      "epoch": 0.83,
      "objective/train/advantage_avg": -0.00045747478725388646,
      "objective/train/docs_used": 992760,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4617254734039307,
      "objective/train/original_loss": 1.4617254734039307,
      "objective/train/theoretical_loss": 3.3733556920842442,
      "objective/train/tokens_used": 2753311200,
      "objective/train/value_avg": -0.0057525634765625,
      "objective/train/value_loss": 0.00013484605005942285,
      "objective/train/value_max": -4.363059997558594e-05,
      "objective/train/value_min": -0.63623046875,
      "objective/train/value_reward_corr": 0.7118297628842131,
      "objective/train/value_std": 0.01107025146484375,
      "objective/train/weight_avg": 0.9996037483215332,
      "objective/train/weighted_lm_loss": 1.4614465236663818,
      "objective/train/weights_max": 1.3987005949020386,
      "objective/train/weights_min": 0.36880823969841003,
      "theoretical_loss": 3.3733556920842442,
      "tokens_seen": 2732851200
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017356764564275398,
      "loss": 0.0681,
      "theoretical_loss": 3.3733556920842442,
      "tokens_seen": 2732851200
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017348740170117157,
      "loss": 0.0658,
      "theoretical_loss": 3.373330554268768,
      "tokens_seen": 2733113344
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017340715775958915,
      "loss": 0.0627,
      "theoretical_loss": 3.373305419539271,
      "tokens_seen": 2733375488
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017332691381800674,
      "loss": 0.0677,
      "theoretical_loss": 3.37328028789508,
      "tokens_seen": 2733637632
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017324666987642435,
      "loss": 0.0646,
      "theoretical_loss": 3.37325515933552,
      "tokens_seen": 2733899776
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001731664259348419,
      "loss": 0.0662,
      "theoretical_loss": 3.3732300338599157,
      "tokens_seen": 2734161920
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017308618199325952,
      "loss": 0.0639,
      "theoretical_loss": 3.3732049114675946,
      "tokens_seen": 2734424064
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001730059380516771,
      "loss": 0.066,
      "theoretical_loss": 3.373179792157882,
      "tokens_seen": 2734686208
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017292569411009467,
      "loss": 0.0655,
      "theoretical_loss": 3.3731546759301034,
      "tokens_seen": 2734948352
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017284545016851228,
      "loss": 0.0627,
      "theoretical_loss": 3.373129562783587,
      "tokens_seen": 2735210496
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017276520622692986,
      "loss": 0.0661,
      "theoretical_loss": 3.373104452717658,
      "tokens_seen": 2735472640
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017268496228534748,
      "loss": 0.0642,
      "theoretical_loss": 3.3730793457316444,
      "tokens_seen": 2735734784
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017260471834376504,
      "loss": 0.0666,
      "theoretical_loss": 3.3730542418248732,
      "tokens_seen": 2735996928
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017252447440218265,
      "loss": 0.0647,
      "theoretical_loss": 3.373029140996671,
      "tokens_seen": 2736259072
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017244423046060023,
      "loss": 0.0624,
      "theoretical_loss": 3.3730040432463664,
      "tokens_seen": 2736521216
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017236398651901782,
      "loss": 0.0651,
      "theoretical_loss": 3.372978948573287,
      "tokens_seen": 2736783360
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001722837425774354,
      "loss": 0.0633,
      "theoretical_loss": 3.3729538569767596,
      "tokens_seen": 2737045504
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.000172203498635853,
      "loss": 0.0642,
      "theoretical_loss": 3.3729287684561142,
      "tokens_seen": 2737307648
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001721232546942706,
      "loss": 0.0644,
      "theoretical_loss": 3.3729036830106778,
      "tokens_seen": 2737569792
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001720430107526882,
      "loss": 0.0681,
      "theoretical_loss": 3.3728786006397797,
      "tokens_seen": 2737831936
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017196276681110575,
      "loss": 0.0658,
      "theoretical_loss": 3.372853521342749,
      "tokens_seen": 2738094080
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017188252286952336,
      "loss": 0.0648,
      "theoretical_loss": 3.3728284451189143,
      "tokens_seen": 2738356224
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017180227892794094,
      "loss": 0.0627,
      "theoretical_loss": 3.3728033719676045,
      "tokens_seen": 2738618368
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017172203498635853,
      "loss": 0.064,
      "theoretical_loss": 3.3727783018881503,
      "tokens_seen": 2738880512
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017164179104477612,
      "loss": 0.0669,
      "theoretical_loss": 3.3727532348798803,
      "tokens_seen": 2739142656
    },
    {
      "epoch": 0.83,
      "objective/train/advantage_avg": -0.00029088734299875796,
      "objective/train/docs_used": 995217,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3371988534927368,
      "objective/train/original_loss": 1.3371988534927368,
      "objective/train/theoretical_loss": 3.372728170942125,
      "objective/train/tokens_used": 2759864800,
      "objective/train/value_avg": -0.005031585693359375,
      "objective/train/value_loss": 0.00014523128629662097,
      "objective/train/value_max": -2.6464462280273438e-05,
      "objective/train/value_min": -0.346923828125,
      "objective/train/value_reward_corr": 0.6469031074700431,
      "objective/train/value_std": 0.00977325439453125,
      "objective/train/weight_avg": 0.9997756481170654,
      "objective/train/weighted_lm_loss": 1.3366706371307373,
      "objective/train/weights_max": 1.358128309249878,
      "objective/train/weights_min": 0.3687118589878082,
      "theoretical_loss": 3.372728170942125,
      "tokens_seen": 2739404800
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017156154710319373,
      "loss": 0.0647,
      "theoretical_loss": 3.372728170942125,
      "tokens_seen": 2739404800
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017148130316161131,
      "loss": 0.0678,
      "theoretical_loss": 3.3727031100742133,
      "tokens_seen": 2739666944
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017140105922002887,
      "loss": 0.062,
      "theoretical_loss": 3.372678052275478,
      "tokens_seen": 2739929088
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017132081527844648,
      "loss": 0.0626,
      "theoretical_loss": 3.372652997545247,
      "tokens_seen": 2740191232
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017124057133686407,
      "loss": 0.0647,
      "theoretical_loss": 3.3726279458828525,
      "tokens_seen": 2740453376
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017116032739528168,
      "loss": 0.0664,
      "theoretical_loss": 3.3726028972876247,
      "tokens_seen": 2740715520
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017108008345369924,
      "loss": 0.0648,
      "theoretical_loss": 3.3725778517588956,
      "tokens_seen": 2740977664
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017099983951211683,
      "loss": 0.0634,
      "theoretical_loss": 3.372552809295996,
      "tokens_seen": 2741239808
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017091959557053444,
      "loss": 0.0655,
      "theoretical_loss": 3.372527769898258,
      "tokens_seen": 2741501952
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.000170839351628952,
      "loss": 0.0656,
      "theoretical_loss": 3.372502733565012,
      "tokens_seen": 2741764096
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001707591076873696,
      "loss": 0.0649,
      "theoretical_loss": 3.372477700295592,
      "tokens_seen": 2742026240
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001706788637457872,
      "loss": 0.0644,
      "theoretical_loss": 3.372452670089329,
      "tokens_seen": 2742288384
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001705986198042048,
      "loss": 0.0619,
      "theoretical_loss": 3.3724276429455555,
      "tokens_seen": 2742550528
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017051837586262237,
      "loss": 0.0649,
      "theoretical_loss": 3.3724026188636045,
      "tokens_seen": 2742812672
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017043813192103995,
      "loss": 0.0657,
      "theoretical_loss": 3.3723775978428083,
      "tokens_seen": 2743074816
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017035788797945757,
      "loss": 0.0659,
      "theoretical_loss": 3.3723525798825005,
      "tokens_seen": 2743336960
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017027764403787515,
      "loss": 0.0642,
      "theoretical_loss": 3.372327564982014,
      "tokens_seen": 2743599104
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017019740009629274,
      "loss": 0.0628,
      "theoretical_loss": 3.372302553140682,
      "tokens_seen": 2743861248
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00017011715615471032,
      "loss": 0.0637,
      "theoretical_loss": 3.3722775443578388,
      "tokens_seen": 2744123392
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001700369122131279,
      "loss": 0.0642,
      "theoretical_loss": 3.372252538632818,
      "tokens_seen": 2744385536
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001699566682715455,
      "loss": 0.0653,
      "theoretical_loss": 3.372227535964954,
      "tokens_seen": 2744647680
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00016987642432996308,
      "loss": 0.0658,
      "theoretical_loss": 3.372202536353581,
      "tokens_seen": 2744909824
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001697961803883807,
      "loss": 0.0606,
      "theoretical_loss": 3.3721775397980323,
      "tokens_seen": 2745171968
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00016971593644679828,
      "loss": 0.064,
      "theoretical_loss": 3.3721525462976443,
      "tokens_seen": 2745434112
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00016963569250521586,
      "loss": 0.0655,
      "theoretical_loss": 3.372127555851751,
      "tokens_seen": 2745696256
    },
    {
      "epoch": 0.83,
      "objective/train/advantage_avg": 0.00043947980157099664,
      "objective/train/docs_used": 997819,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2819221019744873,
      "objective/train/original_loss": 1.2819222211837769,
      "objective/train/theoretical_loss": 3.372102568459688,
      "objective/train/tokens_used": 2766418400,
      "objective/train/value_avg": -0.0078277587890625,
      "objective/train/value_loss": 0.00024490803480148315,
      "objective/train/value_max": -2.2292137145996094e-05,
      "objective/train/value_min": -0.515625,
      "objective/train/value_reward_corr": 0.666362487041277,
      "objective/train/value_std": 0.01556396484375,
      "objective/train/weight_avg": 1.0005496740341187,
      "objective/train/weighted_lm_loss": 1.2824876308441162,
      "objective/train/weights_max": 1.6491175889968872,
      "objective/train/weights_min": 0.37588581442832947,
      "theoretical_loss": 3.372102568459688,
      "tokens_seen": 2745958400
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00016955544856363345,
      "loss": 0.0635,
      "theoretical_loss": 3.372102568459688,
      "tokens_seen": 2745958400
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00016947520462205103,
      "loss": 0.0661,
      "theoretical_loss": 3.372077584120791,
      "tokens_seen": 2746220544
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00016939496068046865,
      "loss": 0.0622,
      "theoretical_loss": 3.372052602834394,
      "tokens_seen": 2746482688
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001693147167388862,
      "loss": 0.0665,
      "theoretical_loss": 3.372027624599834,
      "tokens_seen": 2746744832
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00016923447279730382,
      "loss": 0.0624,
      "theoretical_loss": 3.372002649416447,
      "tokens_seen": 2747006976
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001691542288557214,
      "loss": 0.0629,
      "theoretical_loss": 3.3719776772835686,
      "tokens_seen": 2747269120
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00016907398491413896,
      "loss": 0.0645,
      "theoretical_loss": 3.3719527082005354,
      "tokens_seen": 2747531264
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00016899374097255657,
      "loss": 0.0656,
      "theoretical_loss": 3.371927742166684,
      "tokens_seen": 2747793408
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00016891349703097416,
      "loss": 0.0631,
      "theoretical_loss": 3.371902779181352,
      "tokens_seen": 2748055552
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00016883325308939177,
      "loss": 0.0662,
      "theoretical_loss": 3.3718778192438745,
      "tokens_seen": 2748317696
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00016875300914780933,
      "loss": 0.0646,
      "theoretical_loss": 3.371852862353591,
      "tokens_seen": 2748579840
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00016867276520622694,
      "loss": 0.0653,
      "theoretical_loss": 3.371827908509837,
      "tokens_seen": 2748841984
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00016859252126464453,
      "loss": 0.0649,
      "theoretical_loss": 3.371802957711951,
      "tokens_seen": 2749104128
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001685122773230621,
      "loss": 0.0656,
      "theoretical_loss": 3.3717780099592707,
      "tokens_seen": 2749366272
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001684320333814797,
      "loss": 0.0661,
      "theoretical_loss": 3.371753065251134,
      "tokens_seen": 2749628416
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00016835178943989728,
      "loss": 0.0628,
      "theoretical_loss": 3.3717281235868795,
      "tokens_seen": 2749890560
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001682715454983149,
      "loss": 0.0633,
      "theoretical_loss": 3.3717031849658454,
      "tokens_seen": 2750152704
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00016819130155673246,
      "loss": 0.0648,
      "theoretical_loss": 3.3716782493873705,
      "tokens_seen": 2750414848
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00016811105761515004,
      "loss": 0.0649,
      "theoretical_loss": 3.371653316850794,
      "tokens_seen": 2750676992
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00016803081367356765,
      "loss": 0.066,
      "theoretical_loss": 3.371628387355454,
      "tokens_seen": 2750939136
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00016795056973198524,
      "loss": 0.0624,
      "theoretical_loss": 3.3716034609006904,
      "tokens_seen": 2751201280
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00016787032579040282,
      "loss": 0.0661,
      "theoretical_loss": 3.3715785374858425,
      "tokens_seen": 2751463424
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001677900818488204,
      "loss": 0.0626,
      "theoretical_loss": 3.3715536171102505,
      "tokens_seen": 2751725568
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00016770983790723802,
      "loss": 0.0624,
      "theoretical_loss": 3.3715286997732536,
      "tokens_seen": 2751987712
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001676295939656556,
      "loss": 0.0638,
      "theoretical_loss": 3.3715037854741925,
      "tokens_seen": 2752249856
    },
    {
      "epoch": 0.83,
      "objective/train/advantage_avg": 0.0006543775671161711,
      "objective/train/docs_used": 1000102,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3108874559402466,
      "objective/train/original_loss": 1.3108872175216675,
      "objective/train/theoretical_loss": 3.371478874212407,
      "objective/train/tokens_used": 2772972000,
      "objective/train/value_avg": -0.00687408447265625,
      "objective/train/value_loss": 0.00016098450578283519,
      "objective/train/value_max": -3.349781036376953e-05,
      "objective/train/value_min": -0.63232421875,
      "objective/train/value_reward_corr": 0.7514518038976055,
      "objective/train/value_std": 0.0152587890625,
      "objective/train/weight_avg": 1.0007280111312866,
      "objective/train/weighted_lm_loss": 1.3118845224380493,
      "objective/train/weights_max": 1.4685527086257935,
      "objective/train/weights_min": 0.37259092926979065,
      "theoretical_loss": 3.371478874212407,
      "tokens_seen": 2752512000
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00016754935002407317,
      "loss": 0.0649,
      "theoretical_loss": 3.371478874212407,
      "tokens_seen": 2752512000
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00016746910608249078,
      "loss": 0.0645,
      "theoretical_loss": 3.371453965987238,
      "tokens_seen": 2752774144
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00016738886214090836,
      "loss": 0.0645,
      "theoretical_loss": 3.371429060798026,
      "tokens_seen": 2753036288
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00016730861819932598,
      "loss": 0.0676,
      "theoretical_loss": 3.371404158644112,
      "tokens_seen": 2753298432
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00016722837425774354,
      "loss": 0.0648,
      "theoretical_loss": 3.3713792595248373,
      "tokens_seen": 2753560576
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00016714813031616112,
      "loss": 0.0639,
      "theoretical_loss": 3.3713543634395426,
      "tokens_seen": 2753822720
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.00016706788637457873,
      "loss": 0.0638,
      "theoretical_loss": 3.3713294703875705,
      "tokens_seen": 2754084864
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001669876424329963,
      "loss": 0.064,
      "theoretical_loss": 3.371304580368262,
      "tokens_seen": 2754347008
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001669073984914139,
      "loss": 0.0634,
      "theoretical_loss": 3.3712796933809592,
      "tokens_seen": 2754609152
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001668271545498315,
      "loss": 0.0644,
      "theoretical_loss": 3.3712548094250043,
      "tokens_seen": 2754871296
    },
    {
      "epoch": 0.83,
      "learning_rate": 0.0001667469106082491,
      "loss": 0.0638,
      "theoretical_loss": 3.37122992849974,
      "tokens_seen": 2755133440
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016666666666666666,
      "loss": 0.0677,
      "theoretical_loss": 3.371205050604508,
      "tokens_seen": 2755395584
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016658642272508425,
      "loss": 0.0654,
      "theoretical_loss": 3.371180175738652,
      "tokens_seen": 2755657728
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016650617878350186,
      "loss": 0.067,
      "theoretical_loss": 3.371155303901514,
      "tokens_seen": 2755919872
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016642593484191944,
      "loss": 0.0653,
      "theoretical_loss": 3.3711304350924385,
      "tokens_seen": 2756182016
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016634569090033703,
      "loss": 0.0641,
      "theoretical_loss": 3.371105569310768,
      "tokens_seen": 2756444160
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016626544695875462,
      "loss": 0.0642,
      "theoretical_loss": 3.371080706555846,
      "tokens_seen": 2756706304
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016618520301717223,
      "loss": 0.0637,
      "theoretical_loss": 3.3710558468270166,
      "tokens_seen": 2756968448
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.0001661049590755898,
      "loss": 0.0682,
      "theoretical_loss": 3.371030990123624,
      "tokens_seen": 2757230592
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016602471513400737,
      "loss": 0.0663,
      "theoretical_loss": 3.371006136445012,
      "tokens_seen": 2757492736
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016594447119242498,
      "loss": 0.0663,
      "theoretical_loss": 3.3709812857905255,
      "tokens_seen": 2757754880
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016586422725084257,
      "loss": 0.0679,
      "theoretical_loss": 3.3709564381595087,
      "tokens_seen": 2758017024
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016578398330926016,
      "loss": 0.0654,
      "theoretical_loss": 3.370931593551307,
      "tokens_seen": 2758279168
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016570373936767774,
      "loss": 0.0653,
      "theoretical_loss": 3.3709067519652645,
      "tokens_seen": 2758541312
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016562349542609533,
      "loss": 0.0662,
      "theoretical_loss": 3.370881913400727,
      "tokens_seen": 2758803456
    },
    {
      "epoch": 0.84,
      "objective/train/advantage_avg": -0.0003122219059150666,
      "objective/train/docs_used": 1002432,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3762762546539307,
      "objective/train/original_loss": 1.3762761354446411,
      "objective/train/theoretical_loss": 3.37085707785704,
      "objective/train/tokens_used": 2779525600,
      "objective/train/value_avg": -0.007312774658203125,
      "objective/train/value_loss": 0.0001845854421844706,
      "objective/train/value_max": -1.895427703857422e-05,
      "objective/train/value_min": -0.83056640625,
      "objective/train/value_reward_corr": 0.757421269593654,
      "objective/train/value_std": 0.014129638671875,
      "objective/train/weight_avg": 0.9997718334197998,
      "objective/train/weighted_lm_loss": 1.3756195306777954,
      "objective/train/weights_max": 1.1826108694076538,
      "objective/train/weights_min": 0.38129866123199463,
      "theoretical_loss": 3.37085707785704,
      "tokens_seen": 2759065600
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016554325148451294,
      "loss": 0.0672,
      "theoretical_loss": 3.37085707785704,
      "tokens_seen": 2759065600
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.0001654630075429305,
      "loss": 0.0641,
      "theoretical_loss": 3.3708322453335486,
      "tokens_seen": 2759327744
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.0001653827636013481,
      "loss": 0.0652,
      "theoretical_loss": 3.3708074158295993,
      "tokens_seen": 2759589888
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.0001653025196597657,
      "loss": 0.0614,
      "theoretical_loss": 3.3707825893445382,
      "tokens_seen": 2759852032
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016522227571818328,
      "loss": 0.0662,
      "theoretical_loss": 3.3707577658777113,
      "tokens_seen": 2760114176
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016514203177660087,
      "loss": 0.0641,
      "theoretical_loss": 3.3707329454284647,
      "tokens_seen": 2760376320
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016506178783501845,
      "loss": 0.0675,
      "theoretical_loss": 3.3707081279961453,
      "tokens_seen": 2760638464
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016498154389343607,
      "loss": 0.0659,
      "theoretical_loss": 3.3706833135801,
      "tokens_seen": 2760900608
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016490129995185362,
      "loss": 0.0636,
      "theoretical_loss": 3.370658502179676,
      "tokens_seen": 2761162752
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016482105601027124,
      "loss": 0.066,
      "theoretical_loss": 3.370633693794221,
      "tokens_seen": 2761424896
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016474081206868882,
      "loss": 0.0675,
      "theoretical_loss": 3.3706088884230816,
      "tokens_seen": 2761687040
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.0001646605681271064,
      "loss": 0.0635,
      "theoretical_loss": 3.3705840860656053,
      "tokens_seen": 2761949184
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.000164580324185524,
      "loss": 0.0622,
      "theoretical_loss": 3.370559286721141,
      "tokens_seen": 2762211328
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016450008024394158,
      "loss": 0.0646,
      "theoretical_loss": 3.3705344903890366,
      "tokens_seen": 2762473472
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.0001644198363023592,
      "loss": 0.0663,
      "theoretical_loss": 3.3705096970686395,
      "tokens_seen": 2762735616
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016433959236077675,
      "loss": 0.0674,
      "theoretical_loss": 3.370484906759299,
      "tokens_seen": 2762997760
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016425934841919436,
      "loss": 0.0612,
      "theoretical_loss": 3.370460119460364,
      "tokens_seen": 2763259904
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016417910447761195,
      "loss": 0.0639,
      "theoretical_loss": 3.3704353351711824,
      "tokens_seen": 2763522048
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016409886053602953,
      "loss": 0.0645,
      "theoretical_loss": 3.3704105538911038,
      "tokens_seen": 2763784192
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016401861659444712,
      "loss": 0.0657,
      "theoretical_loss": 3.3703857756194777,
      "tokens_seen": 2764046336
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.0001639383726528647,
      "loss": 0.0663,
      "theoretical_loss": 3.3703610003556537,
      "tokens_seen": 2764308480
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016385812871128232,
      "loss": 0.0659,
      "theoretical_loss": 3.370336228098981,
      "tokens_seen": 2764570624
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.0001637778847696999,
      "loss": 0.064,
      "theoretical_loss": 3.3703114588488097,
      "tokens_seen": 2764832768
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016369764082811746,
      "loss": 0.0611,
      "theoretical_loss": 3.3702866926044903,
      "tokens_seen": 2765094912
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016361739688653507,
      "loss": 0.0647,
      "theoretical_loss": 3.3702619293653724,
      "tokens_seen": 2765357056
    },
    {
      "epoch": 0.84,
      "objective/train/advantage_avg": 0.00030220934422686696,
      "objective/train/docs_used": 1004597,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.231160283088684,
      "objective/train/original_loss": 1.231160044670105,
      "objective/train/theoretical_loss": 3.370237169130807,
      "objective/train/tokens_used": 2786079200,
      "objective/train/value_avg": -0.006072998046875,
      "objective/train/value_loss": 0.00020748592214658856,
      "objective/train/value_max": -2.5093555450439453e-05,
      "objective/train/value_min": -0.3544921875,
      "objective/train/value_reward_corr": 0.6472541412830914,
      "objective/train/value_std": 0.0118408203125,
      "objective/train/weight_avg": 1.0003962516784668,
      "objective/train/weighted_lm_loss": 1.23166823387146,
      "objective/train/weights_max": 1.314475178718567,
      "objective/train/weights_min": 0.3690946400165558,
      "theoretical_loss": 3.370237169130807,
      "tokens_seen": 2765619200
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016353715294495266,
      "loss": 0.0636,
      "theoretical_loss": 3.370237169130807,
      "tokens_seen": 2765619200
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016345690900337024,
      "loss": 0.0639,
      "theoretical_loss": 3.3702124119001446,
      "tokens_seen": 2765881344
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016337666506178783,
      "loss": 0.0659,
      "theoretical_loss": 3.3701876576727368,
      "tokens_seen": 2766143488
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016329642112020544,
      "loss": 0.0657,
      "theoretical_loss": 3.3701629064479333,
      "tokens_seen": 2766405632
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016321617717862303,
      "loss": 0.0662,
      "theoretical_loss": 3.370138158225087,
      "tokens_seen": 2766667776
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016313593323704059,
      "loss": 0.0639,
      "theoretical_loss": 3.370113413003548,
      "tokens_seen": 2766929920
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.0001630556892954582,
      "loss": 0.0641,
      "theoretical_loss": 3.3700886707826694,
      "tokens_seen": 2767192064
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016297544535387578,
      "loss": 0.0652,
      "theoretical_loss": 3.370063931561802,
      "tokens_seen": 2767454208
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.0001628952014122934,
      "loss": 0.0627,
      "theoretical_loss": 3.3700391953402984,
      "tokens_seen": 2767716352
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016281495747071096,
      "loss": 0.0661,
      "theoretical_loss": 3.3700144621175108,
      "tokens_seen": 2767978496
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016273471352912854,
      "loss": 0.0638,
      "theoretical_loss": 3.369989731892792,
      "tokens_seen": 2768240640
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016265446958754615,
      "loss": 0.0635,
      "theoretical_loss": 3.3699650046654948,
      "tokens_seen": 2768502784
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016257422564596374,
      "loss": 0.0642,
      "theoretical_loss": 3.3699402804349714,
      "tokens_seen": 2768764928
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016249398170438132,
      "loss": 0.0679,
      "theoretical_loss": 3.3699155592005754,
      "tokens_seen": 2769027072
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.0001624137377627989,
      "loss": 0.066,
      "theoretical_loss": 3.36989084096166,
      "tokens_seen": 2769289216
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016233349382121652,
      "loss": 0.068,
      "theoretical_loss": 3.3698661257175795,
      "tokens_seen": 2769551360
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016225324987963408,
      "loss": 0.0671,
      "theoretical_loss": 3.369841413467687,
      "tokens_seen": 2769813504
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016217300593805167,
      "loss": 0.0644,
      "theoretical_loss": 3.3698167042113356,
      "tokens_seen": 2770075648
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016209276199646928,
      "loss": 0.0664,
      "theoretical_loss": 3.369791997947881,
      "tokens_seen": 2770337792
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016201251805488686,
      "loss": 0.0651,
      "theoretical_loss": 3.3697672946766764,
      "tokens_seen": 2770599936
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016193227411330445,
      "loss": 0.0655,
      "theoretical_loss": 3.369742594397077,
      "tokens_seen": 2770862080
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016185203017172204,
      "loss": 0.0631,
      "theoretical_loss": 3.369717897108437,
      "tokens_seen": 2771124224
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016177178623013962,
      "loss": 0.0646,
      "theoretical_loss": 3.3696932028101116,
      "tokens_seen": 2771386368
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016169154228855723,
      "loss": 0.0656,
      "theoretical_loss": 3.369668511501456,
      "tokens_seen": 2771648512
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.0001616112983469748,
      "loss": 0.0642,
      "theoretical_loss": 3.3696438231818258,
      "tokens_seen": 2771910656
    },
    {
      "epoch": 0.84,
      "objective/train/advantage_avg": 0.0006500775343738496,
      "objective/train/docs_used": 1006627,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2970798015594482,
      "objective/train/original_loss": 1.2970798015594482,
      "objective/train/theoretical_loss": 3.369619137850576,
      "objective/train/tokens_used": 2792632800,
      "objective/train/value_avg": -0.00782012939453125,
      "objective/train/value_loss": 0.0002618881990201771,
      "objective/train/value_max": -1.8656253814697266e-05,
      "objective/train/value_min": -0.422607421875,
      "objective/train/value_reward_corr": 0.6451092115126782,
      "objective/train/value_std": 0.01337432861328125,
      "objective/train/weight_avg": 1.0007460117340088,
      "objective/train/weighted_lm_loss": 1.2972346544265747,
      "objective/train/weights_max": 1.3180100917816162,
      "objective/train/weights_min": 0.05597710609436035,
      "theoretical_loss": 3.369619137850576,
      "tokens_seen": 2772172800
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.0001615310544053924,
      "loss": 0.0682,
      "theoretical_loss": 3.369619137850576,
      "tokens_seen": 2772172800
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016145081046381,
      "loss": 0.0639,
      "theoretical_loss": 3.3695944555070625,
      "tokens_seen": 2772434944
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016137056652222758,
      "loss": 0.062,
      "theoretical_loss": 3.3695697761506413,
      "tokens_seen": 2772697088
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016129032258064516,
      "loss": 0.0647,
      "theoretical_loss": 3.3695450997806686,
      "tokens_seen": 2772959232
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016121007863906275,
      "loss": 0.0653,
      "theoretical_loss": 3.369520426396501,
      "tokens_seen": 2773221376
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016112983469748036,
      "loss": 0.0659,
      "theoretical_loss": 3.3694957559974936,
      "tokens_seen": 2773483520
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016104959075589792,
      "loss": 0.063,
      "theoretical_loss": 3.3694710885830053,
      "tokens_seen": 2773745664
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016096934681431553,
      "loss": 0.0674,
      "theoretical_loss": 3.3694464241523914,
      "tokens_seen": 2774007808
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016088910287273312,
      "loss": 0.0647,
      "theoretical_loss": 3.36942176270501,
      "tokens_seen": 2774269952
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.0001608088589311507,
      "loss": 0.0626,
      "theoretical_loss": 3.3693971042402175,
      "tokens_seen": 2774532096
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.0001607286149895683,
      "loss": 0.0624,
      "theoretical_loss": 3.369372448757372,
      "tokens_seen": 2774794240
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016064837104798587,
      "loss": 0.0637,
      "theoretical_loss": 3.3693477962558314,
      "tokens_seen": 2775056384
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016056812710640348,
      "loss": 0.0636,
      "theoretical_loss": 3.3693231467349536,
      "tokens_seen": 2775318528
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016048788316482104,
      "loss": 0.066,
      "theoretical_loss": 3.3692985001940965,
      "tokens_seen": 2775580672
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016040763922323866,
      "loss": 0.0647,
      "theoretical_loss": 3.369273856632618,
      "tokens_seen": 2775842816
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016032739528165624,
      "loss": 0.0653,
      "theoretical_loss": 3.3692492160498775,
      "tokens_seen": 2776104960
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00016024715134007383,
      "loss": 0.0641,
      "theoretical_loss": 3.369224578445233,
      "tokens_seen": 2776367104
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.0001601669073984914,
      "loss": 0.0634,
      "theoretical_loss": 3.369199943818044,
      "tokens_seen": 2776629248
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.000160086663456909,
      "loss": 0.064,
      "theoretical_loss": 3.3691753121676693,
      "tokens_seen": 2776891392
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.0001600064195153266,
      "loss": 0.0676,
      "theoretical_loss": 3.3691506834934684,
      "tokens_seen": 2777153536
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.0001599261755737442,
      "loss": 0.0652,
      "theoretical_loss": 3.3691260577948006,
      "tokens_seen": 2777415680
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015984593163216175,
      "loss": 0.0639,
      "theoretical_loss": 3.3691014350710256,
      "tokens_seen": 2777677824
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015976568769057937,
      "loss": 0.0638,
      "theoretical_loss": 3.3690768153215034,
      "tokens_seen": 2777939968
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015968544374899695,
      "loss": 0.0645,
      "theoretical_loss": 3.3690521985455946,
      "tokens_seen": 2778202112
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015960519980741454,
      "loss": 0.064,
      "theoretical_loss": 3.3690275847426587,
      "tokens_seen": 2778464256
    },
    {
      "epoch": 0.84,
      "objective/train/advantage_avg": -0.0006413300288841128,
      "objective/train/docs_used": 1009547,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2469478845596313,
      "objective/train/original_loss": 1.246948003768921,
      "objective/train/theoretical_loss": 3.3690029739120564,
      "objective/train/tokens_used": 2799186400,
      "objective/train/value_avg": -0.00994110107421875,
      "objective/train/value_loss": 0.00023580322158522904,
      "objective/train/value_max": -3.707408905029297e-05,
      "objective/train/value_min": -0.44580078125,
      "objective/train/value_reward_corr": 0.7511882765787193,
      "objective/train/value_std": 0.0173492431640625,
      "objective/train/weight_avg": 0.9994668960571289,
      "objective/train/weighted_lm_loss": 1.2461657524108887,
      "objective/train/weights_max": 1.2655079364776611,
      "objective/train/weights_min": 0.3680661618709564,
      "theoretical_loss": 3.3690029739120564,
      "tokens_seen": 2778726400
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015952495586583212,
      "loss": 0.0672,
      "theoretical_loss": 3.3690029739120564,
      "tokens_seen": 2778726400
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015944471192424974,
      "loss": 0.0663,
      "theoretical_loss": 3.368978366053149,
      "tokens_seen": 2778988544
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015936446798266732,
      "loss": 0.0611,
      "theoretical_loss": 3.3689537611652964,
      "tokens_seen": 2779250688
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015928422404108488,
      "loss": 0.066,
      "theoretical_loss": 3.3689291592478607,
      "tokens_seen": 2779512832
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.0001592039800995025,
      "loss": 0.064,
      "theoretical_loss": 3.3689045603002024,
      "tokens_seen": 2779774976
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015912373615792008,
      "loss": 0.062,
      "theoretical_loss": 3.3688799643216836,
      "tokens_seen": 2780037120
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.0001590434922163377,
      "loss": 0.0655,
      "theoretical_loss": 3.368855371311666,
      "tokens_seen": 2780299264
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015896324827475525,
      "loss": 0.0659,
      "theoretical_loss": 3.368830781269511,
      "tokens_seen": 2780561408
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015888300433317283,
      "loss": 0.0647,
      "theoretical_loss": 3.3688061941945806,
      "tokens_seen": 2780823552
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015880276039159045,
      "loss": 0.0649,
      "theoretical_loss": 3.3687816100862378,
      "tokens_seen": 2781085696
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.000158722516450008,
      "loss": 0.0676,
      "theoretical_loss": 3.368757028943845,
      "tokens_seen": 2781347840
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015864227250842562,
      "loss": 0.0658,
      "theoretical_loss": 3.3687324507667644,
      "tokens_seen": 2781609984
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.0001585620285668432,
      "loss": 0.0655,
      "theoretical_loss": 3.368707875554359,
      "tokens_seen": 2781872128
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015848178462526082,
      "loss": 0.0652,
      "theoretical_loss": 3.3686833033059917,
      "tokens_seen": 2782134272
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015840154068367837,
      "loss": 0.0646,
      "theoretical_loss": 3.368658734021026,
      "tokens_seen": 2782396416
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015832129674209596,
      "loss": 0.064,
      "theoretical_loss": 3.3686341676988256,
      "tokens_seen": 2782658560
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015824105280051357,
      "loss": 0.0664,
      "theoretical_loss": 3.3686096043387543,
      "tokens_seen": 2782920704
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015816080885893116,
      "loss": 0.0649,
      "theoretical_loss": 3.3685850439401754,
      "tokens_seen": 2783182848
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015808056491734874,
      "loss": 0.0656,
      "theoretical_loss": 3.368560486502453,
      "tokens_seen": 2783444992
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015800032097576633,
      "loss": 0.0668,
      "theoretical_loss": 3.3685359320249515,
      "tokens_seen": 2783707136
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015792007703418394,
      "loss": 0.0625,
      "theoretical_loss": 3.3685113805070355,
      "tokens_seen": 2783969280
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.0001578398330926015,
      "loss": 0.0628,
      "theoretical_loss": 3.3684868319480694,
      "tokens_seen": 2784231424
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015775958915101909,
      "loss": 0.0654,
      "theoretical_loss": 3.3684622863474183,
      "tokens_seen": 2784493568
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.0001576793452094367,
      "loss": 0.0655,
      "theoretical_loss": 3.368437743704447,
      "tokens_seen": 2784755712
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015759910126785428,
      "loss": 0.0668,
      "theoretical_loss": 3.3684132040185206,
      "tokens_seen": 2785017856
    },
    {
      "epoch": 0.84,
      "objective/train/advantage_avg": 2.9359511245274916e-05,
      "objective/train/docs_used": 1011592,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3140312433242798,
      "objective/train/original_loss": 1.3140308856964111,
      "objective/train/theoretical_loss": 3.3683886672890053,
      "objective/train/tokens_used": 2805740000,
      "objective/train/value_avg": -0.007099151611328125,
      "objective/train/value_loss": 0.0002247738157166168,
      "objective/train/value_max": -2.7954578399658203e-05,
      "objective/train/value_min": -0.99365234375,
      "objective/train/value_reward_corr": 0.6858024849139678,
      "objective/train/value_std": 0.014373779296875,
      "objective/train/weight_avg": 1.0001325607299805,
      "objective/train/weighted_lm_loss": 1.3137497901916504,
      "objective/train/weights_max": 2.3321189880371094,
      "objective/train/weights_min": 0.36934465169906616,
      "theoretical_loss": 3.3683886672890053,
      "tokens_seen": 2785280000
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015751885732627187,
      "loss": 0.0649,
      "theoretical_loss": 3.3683886672890053,
      "tokens_seen": 2785280000
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015743861338468946,
      "loss": 0.0661,
      "theoretical_loss": 3.368364133515266,
      "tokens_seen": 2785542144
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015735836944310704,
      "loss": 0.0642,
      "theoretical_loss": 3.3683396026966688,
      "tokens_seen": 2785804288
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015727812550152465,
      "loss": 0.0664,
      "theoretical_loss": 3.368315074832579,
      "tokens_seen": 2786066432
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.0001571978815599422,
      "loss": 0.0639,
      "theoretical_loss": 3.3682905499223645,
      "tokens_seen": 2786328576
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015711763761835982,
      "loss": 0.0654,
      "theoretical_loss": 3.36826602796539,
      "tokens_seen": 2786590720
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.0001570373936767774,
      "loss": 0.0654,
      "theoretical_loss": 3.368241508961023,
      "tokens_seen": 2786852864
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015695714973519502,
      "loss": 0.0637,
      "theoretical_loss": 3.36821699290863,
      "tokens_seen": 2787115008
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015687690579361258,
      "loss": 0.0665,
      "theoretical_loss": 3.368192479807578,
      "tokens_seen": 2787377152
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015679666185203017,
      "loss": 0.0667,
      "theoretical_loss": 3.368167969657234,
      "tokens_seen": 2787639296
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015671641791044778,
      "loss": 0.0642,
      "theoretical_loss": 3.368143462456966,
      "tokens_seen": 2787901440
    },
    {
      "epoch": 0.84,
      "learning_rate": 0.00015663617396886534,
      "loss": 0.0618,
      "theoretical_loss": 3.3681189582061415,
      "tokens_seen": 2788163584
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015655593002728295,
      "loss": 0.0657,
      "theoretical_loss": 3.3680944569041276,
      "tokens_seen": 2788425728
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015647568608570054,
      "loss": 0.064,
      "theoretical_loss": 3.3680699585502927,
      "tokens_seen": 2788687872
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015639544214411812,
      "loss": 0.0629,
      "theoretical_loss": 3.3680454631440053,
      "tokens_seen": 2788950016
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001563151982025357,
      "loss": 0.0672,
      "theoretical_loss": 3.368020970684633,
      "tokens_seen": 2789212160
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001562349542609533,
      "loss": 0.0638,
      "theoretical_loss": 3.367996481171545,
      "tokens_seen": 2789474304
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001561547103193709,
      "loss": 0.0678,
      "theoretical_loss": 3.3679719946041096,
      "tokens_seen": 2789736448
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001560744663777885,
      "loss": 0.0673,
      "theoretical_loss": 3.367947510981696,
      "tokens_seen": 2789998592
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015599422243620608,
      "loss": 0.0656,
      "theoretical_loss": 3.3679230303036736,
      "tokens_seen": 2790260736
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015591397849462366,
      "loss": 0.0649,
      "theoretical_loss": 3.367898552569411,
      "tokens_seen": 2790522880
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015583373455304125,
      "loss": 0.0646,
      "theoretical_loss": 3.3678740777782785,
      "tokens_seen": 2790785024
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015575349061145883,
      "loss": 0.0647,
      "theoretical_loss": 3.367849605929645,
      "tokens_seen": 2791047168
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015567324666987642,
      "loss": 0.0651,
      "theoretical_loss": 3.3678251370228813,
      "tokens_seen": 2791309312
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015559300272829403,
      "loss": 0.0667,
      "theoretical_loss": 3.367800671057357,
      "tokens_seen": 2791571456
    },
    {
      "epoch": 0.85,
      "objective/train/advantage_avg": -2.4366694560740143e-05,
      "objective/train/docs_used": 1014349,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3677260875701904,
      "objective/train/original_loss": 1.3677259683609009,
      "objective/train/theoretical_loss": 3.3677762080324425,
      "objective/train/tokens_used": 2812293600,
      "objective/train/value_avg": -0.006603240966796875,
      "objective/train/value_loss": 0.0001228439505212009,
      "objective/train/value_max": -2.485513687133789e-05,
      "objective/train/value_min": -0.468505859375,
      "objective/train/value_reward_corr": 0.7589092297420921,
      "objective/train/value_std": 0.012939453125,
      "objective/train/weight_avg": 1.000034213066101,
      "objective/train/weighted_lm_loss": 1.3680479526519775,
      "objective/train/weights_max": 1.1594135761260986,
      "objective/train/weights_min": 0.38203245401382446,
      "theoretical_loss": 3.3677762080324425,
      "tokens_seen": 2791833600
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015551275878671162,
      "loss": 0.0673,
      "theoretical_loss": 3.3677762080324425,
      "tokens_seen": 2791833600
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015543251484512917,
      "loss": 0.0626,
      "theoretical_loss": 3.3677517479475085,
      "tokens_seen": 2792095744
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001553522709035468,
      "loss": 0.0664,
      "theoretical_loss": 3.367727290801925,
      "tokens_seen": 2792357888
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015527202696196437,
      "loss": 0.0667,
      "theoretical_loss": 3.367702836595064,
      "tokens_seen": 2792620032
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015519178302038198,
      "loss": 0.0654,
      "theoretical_loss": 3.367678385326296,
      "tokens_seen": 2792882176
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015511153907879954,
      "loss": 0.0642,
      "theoretical_loss": 3.3676539369949916,
      "tokens_seen": 2793144320
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015503129513721716,
      "loss": 0.064,
      "theoretical_loss": 3.367629491600524,
      "tokens_seen": 2793406464
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015495105119563474,
      "loss": 0.0648,
      "theoretical_loss": 3.3676050491422624,
      "tokens_seen": 2793668608
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001548708072540523,
      "loss": 0.0643,
      "theoretical_loss": 3.367580609619581,
      "tokens_seen": 2793930752
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001547905633124699,
      "loss": 0.0645,
      "theoretical_loss": 3.3675561730318506,
      "tokens_seen": 2794192896
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001547103193708875,
      "loss": 0.0656,
      "theoretical_loss": 3.3675317393784434,
      "tokens_seen": 2794455040
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001546300754293051,
      "loss": 0.0663,
      "theoretical_loss": 3.3675073086587326,
      "tokens_seen": 2794717184
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015454983148772267,
      "loss": 0.0659,
      "theoretical_loss": 3.3674828808720902,
      "tokens_seen": 2794979328
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015446958754614025,
      "loss": 0.0619,
      "theoretical_loss": 3.3674584560178893,
      "tokens_seen": 2795241472
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015438934360455787,
      "loss": 0.0667,
      "theoretical_loss": 3.3674340340955027,
      "tokens_seen": 2795503616
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015430909966297545,
      "loss": 0.0655,
      "theoretical_loss": 3.367409615104304,
      "tokens_seen": 2795765760
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015422885572139304,
      "loss": 0.0646,
      "theoretical_loss": 3.3673851990436656,
      "tokens_seen": 2796027904
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015414861177981062,
      "loss": 0.0646,
      "theoretical_loss": 3.3673607859129624,
      "tokens_seen": 2796290048
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015406836783822824,
      "loss": 0.0637,
      "theoretical_loss": 3.3673363757115675,
      "tokens_seen": 2796552192
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001539881238966458,
      "loss": 0.0666,
      "theoretical_loss": 3.367311968438855,
      "tokens_seen": 2796814336
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015390787995506338,
      "loss": 0.066,
      "theoretical_loss": 3.367287564094199,
      "tokens_seen": 2797076480
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.000153827636013481,
      "loss": 0.0663,
      "theoretical_loss": 3.3672631626769736,
      "tokens_seen": 2797338624
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015374739207189858,
      "loss": 0.0667,
      "theoretical_loss": 3.367238764186554,
      "tokens_seen": 2797600768
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015366714813031616,
      "loss": 0.0665,
      "theoretical_loss": 3.3672143686223146,
      "tokens_seen": 2797862912
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015358690418873375,
      "loss": 0.066,
      "theoretical_loss": 3.36718997598363,
      "tokens_seen": 2798125056
    },
    {
      "epoch": 0.85,
      "objective/train/advantage_avg": 0.0007134893676266074,
      "objective/train/docs_used": 1016369,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2880847454071045,
      "objective/train/original_loss": 1.288084626197815,
      "objective/train/theoretical_loss": 3.367165586269876,
      "objective/train/tokens_used": 2818847200,
      "objective/train/value_avg": -0.00850677490234375,
      "objective/train/value_loss": 0.0001657002285355702,
      "objective/train/value_max": -3.647804260253906e-05,
      "objective/train/value_min": -0.268798828125,
      "objective/train/value_reward_corr": 0.7366273106085361,
      "objective/train/value_std": 0.01476287841796875,
      "objective/train/weight_avg": 1.0007914304733276,
      "objective/train/weighted_lm_loss": 1.2886732816696167,
      "objective/train/weights_max": 1.2570312023162842,
      "objective/train/weights_min": 0.37699440121650696,
      "theoretical_loss": 3.367165586269876,
      "tokens_seen": 2798387200
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015350666024715133,
      "loss": 0.0639,
      "theoretical_loss": 3.367165586269876,
      "tokens_seen": 2798387200
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015342641630556895,
      "loss": 0.0654,
      "theoretical_loss": 3.3671411994804274,
      "tokens_seen": 2798649344
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001533461723639865,
      "loss": 0.0621,
      "theoretical_loss": 3.36711681561466,
      "tokens_seen": 2798911488
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015326592842240412,
      "loss": 0.0655,
      "theoretical_loss": 3.3670924346719495,
      "tokens_seen": 2799173632
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001531856844808217,
      "loss": 0.0655,
      "theoretical_loss": 3.3670680566516715,
      "tokens_seen": 2799435776
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001531054405392393,
      "loss": 0.0628,
      "theoretical_loss": 3.367043681553202,
      "tokens_seen": 2799697920
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015302519659765687,
      "loss": 0.0683,
      "theoretical_loss": 3.367019309375918,
      "tokens_seen": 2799960064
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015294495265607446,
      "loss": 0.0665,
      "theoretical_loss": 3.3669949401191954,
      "tokens_seen": 2800222208
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015286470871449207,
      "loss": 0.0651,
      "theoretical_loss": 3.366970573782411,
      "tokens_seen": 2800484352
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015278446477290963,
      "loss": 0.0624,
      "theoretical_loss": 3.366946210364942,
      "tokens_seen": 2800746496
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015270422083132724,
      "loss": 0.0657,
      "theoretical_loss": 3.3669218498661646,
      "tokens_seen": 2801008640
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015262397688974483,
      "loss": 0.0643,
      "theoretical_loss": 3.3668974922854567,
      "tokens_seen": 2801270784
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015254373294816242,
      "loss": 0.0685,
      "theoretical_loss": 3.366873137622196,
      "tokens_seen": 2801532928
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015246348900658,
      "loss": 0.0679,
      "theoretical_loss": 3.366848785875759,
      "tokens_seen": 2801795072
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015238324506499759,
      "loss": 0.0669,
      "theoretical_loss": 3.3668244370455245,
      "tokens_seen": 2802057216
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001523030011234152,
      "loss": 0.0663,
      "theoretical_loss": 3.3668000911308704,
      "tokens_seen": 2802319360
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015222275718183278,
      "loss": 0.0625,
      "theoretical_loss": 3.3667757481311744,
      "tokens_seen": 2802581504
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015214251324025037,
      "loss": 0.064,
      "theoretical_loss": 3.3667514080458156,
      "tokens_seen": 2802843648
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015206226929866796,
      "loss": 0.0668,
      "theoretical_loss": 3.3667270708741714,
      "tokens_seen": 2803105792
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015198202535708554,
      "loss": 0.0659,
      "theoretical_loss": 3.366702736615622,
      "tokens_seen": 2803367936
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015190178141550313,
      "loss": 0.0663,
      "theoretical_loss": 3.3666784052695453,
      "tokens_seen": 2803630080
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001518215374739207,
      "loss": 0.0632,
      "theoretical_loss": 3.366654076835321,
      "tokens_seen": 2803892224
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015174129353233832,
      "loss": 0.0671,
      "theoretical_loss": 3.366629751312328,
      "tokens_seen": 2804154368
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001516610495907559,
      "loss": 0.0648,
      "theoretical_loss": 3.3666054286999465,
      "tokens_seen": 2804416512
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015158080564917347,
      "loss": 0.0667,
      "theoretical_loss": 3.3665811089975555,
      "tokens_seen": 2804678656
    },
    {
      "epoch": 0.85,
      "objective/train/advantage_avg": 0.00020279214368201792,
      "objective/train/docs_used": 1018816,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3492354154586792,
      "objective/train/original_loss": 1.3492355346679688,
      "objective/train/theoretical_loss": 3.366556792204535,
      "objective/train/tokens_used": 2825400800,
      "objective/train/value_avg": -0.008026123046875,
      "objective/train/value_loss": 0.00025738292606547475,
      "objective/train/value_max": -3.1948089599609375e-05,
      "objective/train/value_min": -0.366455078125,
      "objective/train/value_reward_corr": 0.7410373552381239,
      "objective/train/value_std": 0.017486572265625,
      "objective/train/weight_avg": 1.0003200769424438,
      "objective/train/weighted_lm_loss": 1.3495210409164429,
      "objective/train/weights_max": 1.3756163120269775,
      "objective/train/weights_min": 0.3690573275089264,
      "theoretical_loss": 3.366556792204535,
      "tokens_seen": 2804940800
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015150056170759108,
      "loss": 0.0646,
      "theoretical_loss": 3.366556792204535,
      "tokens_seen": 2804940800
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015142031776600867,
      "loss": 0.0661,
      "theoretical_loss": 3.3665324783202655,
      "tokens_seen": 2805202944
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015134007382442628,
      "loss": 0.0652,
      "theoretical_loss": 3.3665081673441275,
      "tokens_seen": 2805465088
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015125982988284384,
      "loss": 0.0665,
      "theoretical_loss": 3.3664838592755006,
      "tokens_seen": 2805727232
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015117958594126145,
      "loss": 0.0663,
      "theoretical_loss": 3.366459554113766,
      "tokens_seen": 2805989376
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015109934199967904,
      "loss": 0.0642,
      "theoretical_loss": 3.3664352518583045,
      "tokens_seen": 2806251520
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001510190980580966,
      "loss": 0.0645,
      "theoretical_loss": 3.366410952508497,
      "tokens_seen": 2806513664
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001509388541165142,
      "loss": 0.0639,
      "theoretical_loss": 3.366386656063725,
      "tokens_seen": 2806775808
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001508586101749318,
      "loss": 0.0663,
      "theoretical_loss": 3.3663623625233696,
      "tokens_seen": 2807037952
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001507783662333494,
      "loss": 0.061,
      "theoretical_loss": 3.3663380718868128,
      "tokens_seen": 2807300096
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015069812229176696,
      "loss": 0.0669,
      "theoretical_loss": 3.3663137841534363,
      "tokens_seen": 2807562240
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015061787835018455,
      "loss": 0.0646,
      "theoretical_loss": 3.3662894993226216,
      "tokens_seen": 2807824384
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015053763440860216,
      "loss": 0.0665,
      "theoretical_loss": 3.3662652173937517,
      "tokens_seen": 2808086528
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015045739046701975,
      "loss": 0.0671,
      "theoretical_loss": 3.366240938366208,
      "tokens_seen": 2808348672
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015037714652543733,
      "loss": 0.065,
      "theoretical_loss": 3.366216662239374,
      "tokens_seen": 2808610816
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015029690258385492,
      "loss": 0.0644,
      "theoretical_loss": 3.3661923890126326,
      "tokens_seen": 2808872960
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015021665864227253,
      "loss": 0.0662,
      "theoretical_loss": 3.366168118685365,
      "tokens_seen": 2809135104
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001501364147006901,
      "loss": 0.0639,
      "theoretical_loss": 3.3661438512569566,
      "tokens_seen": 2809397248
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00015005617075910767,
      "loss": 0.0665,
      "theoretical_loss": 3.3661195867267892,
      "tokens_seen": 2809659392
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001499759268175253,
      "loss": 0.0631,
      "theoretical_loss": 3.3660953250942467,
      "tokens_seen": 2809921536
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014989568287594287,
      "loss": 0.0631,
      "theoretical_loss": 3.366071066358713,
      "tokens_seen": 2810183680
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014981543893436046,
      "loss": 0.0628,
      "theoretical_loss": 3.3660468105195713,
      "tokens_seen": 2810445824
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014973519499277804,
      "loss": 0.0618,
      "theoretical_loss": 3.3660225575762066,
      "tokens_seen": 2810707968
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014965495105119563,
      "loss": 0.0654,
      "theoretical_loss": 3.365998307528003,
      "tokens_seen": 2810970112
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014957470710961324,
      "loss": 0.0684,
      "theoretical_loss": 3.3659740603743447,
      "tokens_seen": 2811232256
    },
    {
      "epoch": 0.85,
      "objective/train/advantage_avg": 2.8910437322338112e-05,
      "objective/train/docs_used": 1021297,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2576926946640015,
      "objective/train/original_loss": 1.2576926946640015,
      "objective/train/theoretical_loss": 3.365949816114616,
      "objective/train/tokens_used": 2831954400,
      "objective/train/value_avg": -0.00836181640625,
      "objective/train/value_loss": 0.0002257737360196188,
      "objective/train/value_max": -2.300739288330078e-05,
      "objective/train/value_min": -0.485595703125,
      "objective/train/value_reward_corr": 0.7908450645758498,
      "objective/train/value_std": 0.0178680419921875,
      "objective/train/weight_avg": 1.0001318454742432,
      "objective/train/weighted_lm_loss": 1.2580550909042358,
      "objective/train/weights_max": 1.375653624534607,
      "objective/train/weights_min": 0.3706671893596649,
      "theoretical_loss": 3.365949816114616,
      "tokens_seen": 2811494400
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001494944631680308,
      "loss": 0.0671,
      "theoretical_loss": 3.365949816114616,
      "tokens_seen": 2811494400
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001494142192264484,
      "loss": 0.0656,
      "theoretical_loss": 3.3659255747482018,
      "tokens_seen": 2811756544
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.000149333975284866,
      "loss": 0.0634,
      "theoretical_loss": 3.365901336274488,
      "tokens_seen": 2812018688
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014925373134328358,
      "loss": 0.0651,
      "theoretical_loss": 3.3658771006928587,
      "tokens_seen": 2812280832
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014917348740170117,
      "loss": 0.0655,
      "theoretical_loss": 3.3658528680027002,
      "tokens_seen": 2812542976
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014909324346011875,
      "loss": 0.0657,
      "theoretical_loss": 3.3658286382033973,
      "tokens_seen": 2812805120
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014901299951853637,
      "loss": 0.0652,
      "theoretical_loss": 3.365804411294336,
      "tokens_seen": 2813067264
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014893275557695393,
      "loss": 0.0662,
      "theoretical_loss": 3.3657801872749027,
      "tokens_seen": 2813329408
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014885251163537154,
      "loss": 0.0646,
      "theoretical_loss": 3.3657559661444827,
      "tokens_seen": 2813591552
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014877226769378912,
      "loss": 0.0639,
      "theoretical_loss": 3.3657317479024633,
      "tokens_seen": 2813853696
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014869202375220674,
      "loss": 0.0641,
      "theoretical_loss": 3.3657075325482304,
      "tokens_seen": 2814115840
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001486117798106243,
      "loss": 0.0652,
      "theoretical_loss": 3.3656833200811707,
      "tokens_seen": 2814377984
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014853153586904188,
      "loss": 0.0669,
      "theoretical_loss": 3.3656591105006712,
      "tokens_seen": 2814640128
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001484512919274595,
      "loss": 0.0661,
      "theoretical_loss": 3.3656349038061193,
      "tokens_seen": 2814902272
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014837104798587705,
      "loss": 0.0641,
      "theoretical_loss": 3.3656106999969015,
      "tokens_seen": 2815164416
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014829080404429466,
      "loss": 0.0645,
      "theoretical_loss": 3.365586499072406,
      "tokens_seen": 2815426560
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014821056010271225,
      "loss": 0.0681,
      "theoretical_loss": 3.36556230103202,
      "tokens_seen": 2815688704
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014813031616112983,
      "loss": 0.0642,
      "theoretical_loss": 3.3655381058751317,
      "tokens_seen": 2815950848
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014805007221954742,
      "loss": 0.0678,
      "theoretical_loss": 3.3655139136011285,
      "tokens_seen": 2816212992
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.000147969828277965,
      "loss": 0.0675,
      "theoretical_loss": 3.3654897242093993,
      "tokens_seen": 2816475136
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014788958433638262,
      "loss": 0.0641,
      "theoretical_loss": 3.365465537699332,
      "tokens_seen": 2816737280
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001478093403948002,
      "loss": 0.0676,
      "theoretical_loss": 3.3654413540703154,
      "tokens_seen": 2816999424
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001477290964532178,
      "loss": 0.0669,
      "theoretical_loss": 3.365417173321738,
      "tokens_seen": 2817261568
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014764885251163538,
      "loss": 0.067,
      "theoretical_loss": 3.3653929954529893,
      "tokens_seen": 2817523712
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014756860857005296,
      "loss": 0.0669,
      "theoretical_loss": 3.3653688204634578,
      "tokens_seen": 2817785856
    },
    {
      "epoch": 0.85,
      "objective/train/advantage_avg": 0.0007476872997358441,
      "objective/train/docs_used": 1024155,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2878682613372803,
      "objective/train/original_loss": 1.2878682613372803,
      "objective/train/theoretical_loss": 3.3653446483525333,
      "objective/train/tokens_used": 2838508000,
      "objective/train/value_avg": -0.0068511962890625,
      "objective/train/value_loss": 0.0002798317582346499,
      "objective/train/value_max": -3.1948089599609375e-05,
      "objective/train/value_min": -0.99462890625,
      "objective/train/value_reward_corr": 0.7159308124042012,
      "objective/train/value_std": 0.0169219970703125,
      "objective/train/weight_avg": 1.0008704662322998,
      "objective/train/weighted_lm_loss": 1.2883635759353638,
      "objective/train/weights_max": 1.835062026977539,
      "objective/train/weights_min": 0.3006945848464966,
      "theoretical_loss": 3.3653446483525333,
      "tokens_seen": 2818048000
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014748836462847057,
      "loss": 0.0658,
      "theoretical_loss": 3.3653446483525333,
      "tokens_seen": 2818048000
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014740812068688813,
      "loss": 0.0668,
      "theoretical_loss": 3.365320479119605,
      "tokens_seen": 2818310144
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014732787674530574,
      "loss": 0.0656,
      "theoretical_loss": 3.3652963127640625,
      "tokens_seen": 2818572288
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014724763280372333,
      "loss": 0.0665,
      "theoretical_loss": 3.365272149285296,
      "tokens_seen": 2818834432
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001471673888621409,
      "loss": 0.0685,
      "theoretical_loss": 3.3652479886826954,
      "tokens_seen": 2819096576
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001470871449205585,
      "loss": 0.0635,
      "theoretical_loss": 3.3652238309556513,
      "tokens_seen": 2819358720
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014700690097897609,
      "loss": 0.0673,
      "theoretical_loss": 3.365199676103553,
      "tokens_seen": 2819620864
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.0001469266570373937,
      "loss": 0.067,
      "theoretical_loss": 3.3651755241257924,
      "tokens_seen": 2819883008
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014684641309581126,
      "loss": 0.063,
      "theoretical_loss": 3.36515137502176,
      "tokens_seen": 2820145152
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014676616915422887,
      "loss": 0.0672,
      "theoretical_loss": 3.3651272287908465,
      "tokens_seen": 2820407296
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014668592521264646,
      "loss": 0.064,
      "theoretical_loss": 3.365103085432443,
      "tokens_seen": 2820669440
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014660568127106404,
      "loss": 0.0651,
      "theoretical_loss": 3.365078944945941,
      "tokens_seen": 2820931584
    },
    {
      "epoch": 0.85,
      "learning_rate": 0.00014652543732948163,
      "loss": 0.0643,
      "theoretical_loss": 3.3650548073307327,
      "tokens_seen": 2821193728
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.0001464451933878992,
      "loss": 0.0682,
      "theoretical_loss": 3.365030672586209,
      "tokens_seen": 2821455872
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014636494944631682,
      "loss": 0.065,
      "theoretical_loss": 3.3650065407117618,
      "tokens_seen": 2821718016
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014628470550473438,
      "loss": 0.0646,
      "theoretical_loss": 3.3649824117067832,
      "tokens_seen": 2821980160
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014620446156315197,
      "loss": 0.067,
      "theoretical_loss": 3.3649582855706663,
      "tokens_seen": 2822242304
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014612421762156958,
      "loss": 0.0647,
      "theoretical_loss": 3.364934162302803,
      "tokens_seen": 2822504448
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014604397367998717,
      "loss": 0.0665,
      "theoretical_loss": 3.3649100419025855,
      "tokens_seen": 2822766592
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014596372973840475,
      "loss": 0.0647,
      "theoretical_loss": 3.364885924369408,
      "tokens_seen": 2823028736
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014588348579682234,
      "loss": 0.064,
      "theoretical_loss": 3.3648618097026617,
      "tokens_seen": 2823290880
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014580324185523995,
      "loss": 0.0654,
      "theoretical_loss": 3.3648376979017414,
      "tokens_seen": 2823553024
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014572299791365754,
      "loss": 0.0654,
      "theoretical_loss": 3.3648135889660398,
      "tokens_seen": 2823815168
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.0001456427539720751,
      "loss": 0.0633,
      "theoretical_loss": 3.3647894828949503,
      "tokens_seen": 2824077312
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.0001455625100304927,
      "loss": 0.0663,
      "theoretical_loss": 3.364765379687867,
      "tokens_seen": 2824339456
    },
    {
      "epoch": 0.86,
      "objective/train/advantage_avg": -0.0002453245688229799,
      "objective/train/docs_used": 1026596,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.294011116027832,
      "objective/train/original_loss": 1.2940109968185425,
      "objective/train/theoretical_loss": 3.3647412793441838,
      "objective/train/tokens_used": 2845061600,
      "objective/train/value_avg": -0.00684356689453125,
      "objective/train/value_loss": 0.0001892550935735926,
      "objective/train/value_max": -3.170967102050781e-05,
      "objective/train/value_min": -0.68603515625,
      "objective/train/value_reward_corr": 0.7367796299909823,
      "objective/train/value_std": 0.01493072509765625,
      "objective/train/weight_avg": 0.9998413920402527,
      "objective/train/weighted_lm_loss": 1.2936891317367554,
      "objective/train/weights_max": 1.3273746967315674,
      "objective/train/weights_min": 0.36991289258003235,
      "theoretical_loss": 3.3647412793441838,
      "tokens_seen": 2824601600
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.0001454822660889103,
      "loss": 0.0631,
      "theoretical_loss": 3.3647412793441838,
      "tokens_seen": 2824601600
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014540202214732788,
      "loss": 0.0628,
      "theoretical_loss": 3.364717181863295,
      "tokens_seen": 2824863744
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014532177820574546,
      "loss": 0.064,
      "theoretical_loss": 3.3646930872445946,
      "tokens_seen": 2825125888
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014524153426416305,
      "loss": 0.0642,
      "theoretical_loss": 3.3646689954874773,
      "tokens_seen": 2825388032
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014516129032258066,
      "loss": 0.0664,
      "theoretical_loss": 3.3646449065913377,
      "tokens_seen": 2825650176
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014508104638099822,
      "loss": 0.0633,
      "theoretical_loss": 3.364620820555571,
      "tokens_seen": 2825912320
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014500080243941583,
      "loss": 0.0634,
      "theoretical_loss": 3.3645967373795718,
      "tokens_seen": 2826174464
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014492055849783342,
      "loss": 0.0678,
      "theoretical_loss": 3.3645726570627357,
      "tokens_seen": 2826436608
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014484031455625103,
      "loss": 0.065,
      "theoretical_loss": 3.3645485796044574,
      "tokens_seen": 2826698752
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.0001447600706146686,
      "loss": 0.0638,
      "theoretical_loss": 3.3645245050041335,
      "tokens_seen": 2826960896
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014467982667308617,
      "loss": 0.0645,
      "theoretical_loss": 3.364500433261159,
      "tokens_seen": 2827223040
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.0001445995827315038,
      "loss": 0.0639,
      "theoretical_loss": 3.3644763643749305,
      "tokens_seen": 2827485184
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014451933878992135,
      "loss": 0.0655,
      "theoretical_loss": 3.364452298344844,
      "tokens_seen": 2827747328
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014443909484833896,
      "loss": 0.067,
      "theoretical_loss": 3.3644282351702954,
      "tokens_seen": 2828009472
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014435885090675654,
      "loss": 0.0643,
      "theoretical_loss": 3.364404174850682,
      "tokens_seen": 2828271616
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014427860696517413,
      "loss": 0.065,
      "theoretical_loss": 3.3643801173853998,
      "tokens_seen": 2828533760
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014419836302359171,
      "loss": 0.0642,
      "theoretical_loss": 3.364356062773846,
      "tokens_seen": 2828795904
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.0001441181190820093,
      "loss": 0.0676,
      "theoretical_loss": 3.3643320110154176,
      "tokens_seen": 2829058048
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.0001440378751404269,
      "loss": 0.0668,
      "theoretical_loss": 3.364307962109512,
      "tokens_seen": 2829320192
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.0001439576311988445,
      "loss": 0.0652,
      "theoretical_loss": 3.364283916055527,
      "tokens_seen": 2829582336
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014387738725726208,
      "loss": 0.0635,
      "theoretical_loss": 3.364259872852859,
      "tokens_seen": 2829844480
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014379714331567967,
      "loss": 0.0648,
      "theoretical_loss": 3.364235832500907,
      "tokens_seen": 2830106624
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014371689937409725,
      "loss": 0.0645,
      "theoretical_loss": 3.3642117949990684,
      "tokens_seen": 2830368768
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014363665543251484,
      "loss": 0.064,
      "theoretical_loss": 3.364187760346742,
      "tokens_seen": 2830630912
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014355641149093243,
      "loss": 0.0636,
      "theoretical_loss": 3.3641637285433257,
      "tokens_seen": 2830893056
    },
    {
      "epoch": 0.86,
      "objective/train/advantage_avg": 0.0009019442368298769,
      "objective/train/docs_used": 1029215,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2658255100250244,
      "objective/train/original_loss": 1.2658252716064453,
      "objective/train/theoretical_loss": 3.364139699588218,
      "objective/train/tokens_used": 2851615200,
      "objective/train/value_avg": -0.0078277587890625,
      "objective/train/value_loss": 0.0002247198426630348,
      "objective/train/value_max": -1.9371509552001953e-05,
      "objective/train/value_min": -0.9765625,
      "objective/train/value_reward_corr": 0.6742154938115128,
      "objective/train/value_std": 0.01406097412109375,
      "objective/train/weight_avg": 1.0009979009628296,
      "objective/train/weighted_lm_loss": 1.266892671585083,
      "objective/train/weights_max": 1.2235815525054932,
      "objective/train/weights_min": 0.24483628571033478,
      "theoretical_loss": 3.364139699588218,
      "tokens_seen": 2831155200
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014347616754935004,
      "loss": 0.065,
      "theoretical_loss": 3.364139699588218,
      "tokens_seen": 2831155200
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014339592360776762,
      "loss": 0.0651,
      "theoretical_loss": 3.3641156734808173,
      "tokens_seen": 2831417344
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014331567966618518,
      "loss": 0.0662,
      "theoretical_loss": 3.364091650220523,
      "tokens_seen": 2831679488
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.0001432354357246028,
      "loss": 0.0643,
      "theoretical_loss": 3.3640676298067347,
      "tokens_seen": 2831941632
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014315519178302038,
      "loss": 0.0662,
      "theoretical_loss": 3.3640436122388504,
      "tokens_seen": 2832203776
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.000143074947841438,
      "loss": 0.0661,
      "theoretical_loss": 3.364019597516271,
      "tokens_seen": 2832465920
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014299470389985555,
      "loss": 0.064,
      "theoretical_loss": 3.3639955856383947,
      "tokens_seen": 2832728064
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014291445995827316,
      "loss": 0.0676,
      "theoretical_loss": 3.363971576604622,
      "tokens_seen": 2832990208
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014283421601669075,
      "loss": 0.0664,
      "theoretical_loss": 3.3639475704143535,
      "tokens_seen": 2833252352
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014275397207510833,
      "loss": 0.0653,
      "theoretical_loss": 3.3639235670669883,
      "tokens_seen": 2833514496
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014267372813352592,
      "loss": 0.0666,
      "theoretical_loss": 3.3638995665619276,
      "tokens_seen": 2833776640
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.0001425934841919435,
      "loss": 0.0646,
      "theoretical_loss": 3.363875568898571,
      "tokens_seen": 2834038784
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014251324025036112,
      "loss": 0.0646,
      "theoretical_loss": 3.36385157407632,
      "tokens_seen": 2834300928
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014243299630877868,
      "loss": 0.0637,
      "theoretical_loss": 3.3638275820945758,
      "tokens_seen": 2834563072
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014235275236719626,
      "loss": 0.0651,
      "theoretical_loss": 3.3638035929527392,
      "tokens_seen": 2834825216
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014227250842561388,
      "loss": 0.0659,
      "theoretical_loss": 3.3637796066502106,
      "tokens_seen": 2835087360
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014219226448403146,
      "loss": 0.0657,
      "theoretical_loss": 3.3637556231863925,
      "tokens_seen": 2835349504
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014211202054244905,
      "loss": 0.0651,
      "theoretical_loss": 3.363731642560686,
      "tokens_seen": 2835611648
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014203177660086663,
      "loss": 0.0677,
      "theoretical_loss": 3.3637076647724937,
      "tokens_seen": 2835873792
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014195153265928424,
      "loss": 0.0673,
      "theoretical_loss": 3.3636836898212166,
      "tokens_seen": 2836135936
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014187128871770183,
      "loss": 0.066,
      "theoretical_loss": 3.3636597177062573,
      "tokens_seen": 2836398080
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.0001417910447761194,
      "loss": 0.0662,
      "theoretical_loss": 3.363635748427018,
      "tokens_seen": 2836660224
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.000141710800834537,
      "loss": 0.0682,
      "theoretical_loss": 3.363611781982902,
      "tokens_seen": 2836922368
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014163055689295459,
      "loss": 0.0651,
      "theoretical_loss": 3.3635878183733112,
      "tokens_seen": 2837184512
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014155031295137217,
      "loss": 0.0662,
      "theoretical_loss": 3.3635638575976485,
      "tokens_seen": 2837446656
    },
    {
      "epoch": 0.86,
      "objective/train/advantage_avg": 0.0008682442712597549,
      "objective/train/docs_used": 1031594,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2491042613983154,
      "objective/train/original_loss": 1.2491042613983154,
      "objective/train/theoretical_loss": 3.3635398996553176,
      "objective/train/tokens_used": 2858168800,
      "objective/train/value_avg": -0.0068206787109375,
      "objective/train/value_loss": 0.00021514028776437044,
      "objective/train/value_max": -3.9458274841308594e-05,
      "objective/train/value_min": -0.81103515625,
      "objective/train/value_reward_corr": 0.643333056599688,
      "objective/train/value_std": 0.01387786865234375,
      "objective/train/weight_avg": 1.0009657144546509,
      "objective/train/weighted_lm_loss": 1.2500951290130615,
      "objective/train/weights_max": 1.7786227464675903,
      "objective/train/weights_min": 0.3690904378890991,
      "theoretical_loss": 3.3635398996553176,
      "tokens_seen": 2837708800
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014147006900978976,
      "loss": 0.0652,
      "theoretical_loss": 3.3635398996553176,
      "tokens_seen": 2837708800
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014138982506820734,
      "loss": 0.067,
      "theoretical_loss": 3.3635159445457212,
      "tokens_seen": 2837970944
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014130958112662496,
      "loss": 0.0647,
      "theoretical_loss": 3.363491992268263,
      "tokens_seen": 2838233088
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014122933718504251,
      "loss": 0.0644,
      "theoretical_loss": 3.3634680428223467,
      "tokens_seen": 2838495232
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014114909324346013,
      "loss": 0.0611,
      "theoretical_loss": 3.3634440962073757,
      "tokens_seen": 2838757376
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.0001410688493018777,
      "loss": 0.0642,
      "theoretical_loss": 3.3634201524227545,
      "tokens_seen": 2839019520
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014098860536029532,
      "loss": 0.0665,
      "theoretical_loss": 3.363396211467887,
      "tokens_seen": 2839281664
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014090836141871288,
      "loss": 0.0655,
      "theoretical_loss": 3.3633722733421774,
      "tokens_seen": 2839543808
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014082811747713047,
      "loss": 0.0656,
      "theoretical_loss": 3.363348338045031,
      "tokens_seen": 2839805952
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014074787353554808,
      "loss": 0.0656,
      "theoretical_loss": 3.3633244055758516,
      "tokens_seen": 2840068096
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014066762959396564,
      "loss": 0.0646,
      "theoretical_loss": 3.363300475934045,
      "tokens_seen": 2840330240
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014058738565238325,
      "loss": 0.0646,
      "theoretical_loss": 3.363276549119015,
      "tokens_seen": 2840592384
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014050714171080084,
      "loss": 0.0648,
      "theoretical_loss": 3.3632526251301678,
      "tokens_seen": 2840854528
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014042689776921845,
      "loss": 0.0653,
      "theoretical_loss": 3.3632287039669087,
      "tokens_seen": 2841116672
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.000140346653827636,
      "loss": 0.0663,
      "theoretical_loss": 3.363204785628643,
      "tokens_seen": 2841378816
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.0001402664098860536,
      "loss": 0.0672,
      "theoretical_loss": 3.363180870114777,
      "tokens_seen": 2841640960
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.0001401861659444712,
      "loss": 0.0656,
      "theoretical_loss": 3.3631569574247164,
      "tokens_seen": 2841903104
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.0001401059220028888,
      "loss": 0.0661,
      "theoretical_loss": 3.3631330475578674,
      "tokens_seen": 2842165248
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00014002567806130638,
      "loss": 0.0652,
      "theoretical_loss": 3.363109140513636,
      "tokens_seen": 2842427392
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013994543411972396,
      "loss": 0.0663,
      "theoretical_loss": 3.3630852362914294,
      "tokens_seen": 2842689536
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013986519017814155,
      "loss": 0.0655,
      "theoretical_loss": 3.363061334890654,
      "tokens_seen": 2842951680
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013978494623655913,
      "loss": 0.0647,
      "theoretical_loss": 3.363037436310716,
      "tokens_seen": 2843213824
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013970470229497672,
      "loss": 0.0645,
      "theoretical_loss": 3.363013540551023,
      "tokens_seen": 2843475968
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013962445835339433,
      "loss": 0.0669,
      "theoretical_loss": 3.362989647610983,
      "tokens_seen": 2843738112
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013954421441181192,
      "loss": 0.0658,
      "theoretical_loss": 3.362965757490002,
      "tokens_seen": 2844000256
    },
    {
      "epoch": 0.86,
      "objective/train/advantage_avg": 0.0007202853448688984,
      "objective/train/docs_used": 1033870,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3085803985595703,
      "objective/train/original_loss": 1.3085805177688599,
      "objective/train/theoretical_loss": 3.3629418701874885,
      "objective/train/tokens_used": 2864722400,
      "objective/train/value_avg": -0.005878448486328125,
      "objective/train/value_loss": 0.00014473906776402146,
      "objective/train/value_max": -2.7120113372802734e-05,
      "objective/train/value_min": -0.4228515625,
      "objective/train/value_reward_corr": 0.6706889494136361,
      "objective/train/value_std": 0.01087188720703125,
      "objective/train/weight_avg": 1.0007864236831665,
      "objective/train/weighted_lm_loss": 1.3093208074569702,
      "objective/train/weights_max": 1.4480531215667725,
      "objective/train/weights_min": 0.3685835599899292,
      "theoretical_loss": 3.3629418701874885,
      "tokens_seen": 2844262400
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013946397047022948,
      "loss": 0.0664,
      "theoretical_loss": 3.3629418701874885,
      "tokens_seen": 2844262400
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.0001393837265286471,
      "loss": 0.0656,
      "theoretical_loss": 3.3629179857028495,
      "tokens_seen": 2844524544
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013930348258706467,
      "loss": 0.0649,
      "theoretical_loss": 3.362894104035494,
      "tokens_seen": 2844786688
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.0001392232386454823,
      "loss": 0.0642,
      "theoretical_loss": 3.3628702251848295,
      "tokens_seen": 2845048832
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013914299470389985,
      "loss": 0.0643,
      "theoretical_loss": 3.3628463491502645,
      "tokens_seen": 2845310976
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013906275076231746,
      "loss": 0.0664,
      "theoretical_loss": 3.362822475931207,
      "tokens_seen": 2845573120
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013898250682073504,
      "loss": 0.066,
      "theoretical_loss": 3.362798605527066,
      "tokens_seen": 2845835264
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.0001389022628791526,
      "loss": 0.0654,
      "theoretical_loss": 3.362774737937251,
      "tokens_seen": 2846097408
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013882201893757021,
      "loss": 0.0651,
      "theoretical_loss": 3.3627508731611697,
      "tokens_seen": 2846359552
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.0001387417749959878,
      "loss": 0.0645,
      "theoretical_loss": 3.362727011198232,
      "tokens_seen": 2846621696
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.0001386615310544054,
      "loss": 0.0633,
      "theoretical_loss": 3.362703152047848,
      "tokens_seen": 2846883840
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013858128711282297,
      "loss": 0.0639,
      "theoretical_loss": 3.362679295709426,
      "tokens_seen": 2847145984
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013850104317124058,
      "loss": 0.0638,
      "theoretical_loss": 3.3626554421823758,
      "tokens_seen": 2847408128
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013842079922965817,
      "loss": 0.0667,
      "theoretical_loss": 3.3626315914661085,
      "tokens_seen": 2847670272
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013834055528807575,
      "loss": 0.0634,
      "theoretical_loss": 3.362607743560033,
      "tokens_seen": 2847932416
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013826031134649334,
      "loss": 0.0655,
      "theoretical_loss": 3.36258389846356,
      "tokens_seen": 2848194560
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013818006740491093,
      "loss": 0.0642,
      "theoretical_loss": 3.3625600561761,
      "tokens_seen": 2848456704
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013809982346332854,
      "loss": 0.0636,
      "theoretical_loss": 3.362536216697064,
      "tokens_seen": 2848718848
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013801957952174612,
      "loss": 0.0652,
      "theoretical_loss": 3.362512380025862,
      "tokens_seen": 2848980992
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013793933558016368,
      "loss": 0.0656,
      "theoretical_loss": 3.3624885461619054,
      "tokens_seen": 2849243136
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.0001378590916385813,
      "loss": 0.0661,
      "theoretical_loss": 3.362464715104605,
      "tokens_seen": 2849505280
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013777884769699888,
      "loss": 0.0649,
      "theoretical_loss": 3.362440886853373,
      "tokens_seen": 2849767424
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013769860375541647,
      "loss": 0.0661,
      "theoretical_loss": 3.36241706140762,
      "tokens_seen": 2850029568
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013761835981383405,
      "loss": 0.0663,
      "theoretical_loss": 3.3623932387667583,
      "tokens_seen": 2850291712
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013753811587225166,
      "loss": 0.0652,
      "theoretical_loss": 3.362369418930199,
      "tokens_seen": 2850553856
    },
    {
      "epoch": 0.86,
      "objective/train/advantage_avg": 0.0005318819312378764,
      "objective/train/docs_used": 1036421,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2573963403701782,
      "objective/train/original_loss": 1.2573959827423096,
      "objective/train/theoretical_loss": 3.3623456018973554,
      "objective/train/tokens_used": 2871276000,
      "objective/train/value_avg": -0.007602691650390625,
      "objective/train/value_loss": 0.0002422871912131086,
      "objective/train/value_max": -3.737211227416992e-05,
      "objective/train/value_min": -0.87158203125,
      "objective/train/value_reward_corr": 0.7889085844853957,
      "objective/train/value_std": 0.0200347900390625,
      "objective/train/weight_avg": 1.0006413459777832,
      "objective/train/weighted_lm_loss": 1.257759690284729,
      "objective/train/weights_max": 1.4836885929107666,
      "objective/train/weights_min": 0.36959269642829895,
      "theoretical_loss": 3.3623456018973554,
      "tokens_seen": 2850816000
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013745787193066925,
      "loss": 0.0657,
      "theoretical_loss": 3.3623456018973554,
      "tokens_seen": 2850816000
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.0001373776279890868,
      "loss": 0.0641,
      "theoretical_loss": 3.362321787667639,
      "tokens_seen": 2851078144
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013729738404750442,
      "loss": 0.0637,
      "theoretical_loss": 3.3622979762404617,
      "tokens_seen": 2851340288
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.000137217140105922,
      "loss": 0.0633,
      "theoretical_loss": 3.3622741676152366,
      "tokens_seen": 2851602432
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013713689616433962,
      "loss": 0.0628,
      "theoretical_loss": 3.3622503617913764,
      "tokens_seen": 2851864576
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013705665222275718,
      "loss": 0.0637,
      "theoretical_loss": 3.362226558768294,
      "tokens_seen": 2852126720
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013697640828117476,
      "loss": 0.0644,
      "theoretical_loss": 3.3622027585454024,
      "tokens_seen": 2852388864
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013689616433959238,
      "loss": 0.0665,
      "theoretical_loss": 3.3621789611221153,
      "tokens_seen": 2852651008
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013681592039800993,
      "loss": 0.0656,
      "theoretical_loss": 3.362155166497846,
      "tokens_seen": 2852913152
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013673567645642755,
      "loss": 0.0641,
      "theoretical_loss": 3.362131374672008,
      "tokens_seen": 2853175296
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013665543251484513,
      "loss": 0.0644,
      "theoretical_loss": 3.362107585644015,
      "tokens_seen": 2853437440
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.00013657518857326274,
      "loss": 0.0646,
      "theoretical_loss": 3.362083799413281,
      "tokens_seen": 2853699584
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.0001364949446316803,
      "loss": 0.0646,
      "theoretical_loss": 3.3620600159792207,
      "tokens_seen": 2853961728
    },
    {
      "epoch": 0.86,
      "learning_rate": 0.0001364147006900979,
      "loss": 0.0662,
      "theoretical_loss": 3.3620362353412476,
      "tokens_seen": 2854223872
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001363344567485155,
      "loss": 0.066,
      "theoretical_loss": 3.362012457498777,
      "tokens_seen": 2854486016
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013625421280693309,
      "loss": 0.0662,
      "theoretical_loss": 3.361988682451223,
      "tokens_seen": 2854748160
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013617396886535067,
      "loss": 0.0652,
      "theoretical_loss": 3.361964910198001,
      "tokens_seen": 2855010304
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013609372492376826,
      "loss": 0.0651,
      "theoretical_loss": 3.3619411407385256,
      "tokens_seen": 2855272448
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013601348098218584,
      "loss": 0.066,
      "theoretical_loss": 3.3619173740722124,
      "tokens_seen": 2855534592
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013593323704060343,
      "loss": 0.0653,
      "theoretical_loss": 3.361893610198476,
      "tokens_seen": 2855796736
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013585299309902101,
      "loss": 0.0684,
      "theoretical_loss": 3.3618698491167334,
      "tokens_seen": 2856058880
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013577274915743863,
      "loss": 0.063,
      "theoretical_loss": 3.3618460908263987,
      "tokens_seen": 2856321024
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001356925052158562,
      "loss": 0.0672,
      "theoretical_loss": 3.361822335326889,
      "tokens_seen": 2856583168
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001356122612742738,
      "loss": 0.065,
      "theoretical_loss": 3.3617985826176207,
      "tokens_seen": 2856845312
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013553201733269138,
      "loss": 0.0641,
      "theoretical_loss": 3.361774832698009,
      "tokens_seen": 2857107456
    },
    {
      "epoch": 0.87,
      "objective/train/advantage_avg": 0.00029029804863967,
      "objective/train/docs_used": 1038225,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3740379810333252,
      "objective/train/original_loss": 1.374037742614746,
      "objective/train/theoretical_loss": 3.3617510855674704,
      "objective/train/tokens_used": 2877829600,
      "objective/train/value_avg": -0.007160186767578125,
      "objective/train/value_loss": 0.00025967363035306334,
      "objective/train/value_max": -4.470348358154297e-05,
      "objective/train/value_min": -0.67431640625,
      "objective/train/value_reward_corr": 0.6895487345317597,
      "objective/train/value_std": 0.013885498046875,
      "objective/train/weight_avg": 1.0004048347473145,
      "objective/train/weighted_lm_loss": 1.3736540079116821,
      "objective/train/weights_max": 1.2676682472229004,
      "objective/train/weights_min": 0.37812936305999756,
      "theoretical_loss": 3.3617510855674704,
      "tokens_seen": 2857369600
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013545177339110897,
      "loss": 0.0651,
      "theoretical_loss": 3.3617510855674704,
      "tokens_seen": 2857369600
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013537152944952658,
      "loss": 0.0651,
      "theoretical_loss": 3.3617273412254223,
      "tokens_seen": 2857631744
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013529128550794414,
      "loss": 0.065,
      "theoretical_loss": 3.361703599671281,
      "tokens_seen": 2857893888
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013521104156636175,
      "loss": 0.0679,
      "theoretical_loss": 3.361679860904464,
      "tokens_seen": 2858156032
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013513079762477934,
      "loss": 0.0639,
      "theoretical_loss": 3.3616561249243877,
      "tokens_seen": 2858418176
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001350505536831969,
      "loss": 0.0677,
      "theoretical_loss": 3.36163239173047,
      "tokens_seen": 2858680320
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001349703097416145,
      "loss": 0.0672,
      "theoretical_loss": 3.361608661322128,
      "tokens_seen": 2858942464
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001348900658000321,
      "loss": 0.067,
      "theoretical_loss": 3.3615849336987798,
      "tokens_seen": 2859204608
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001348098218584497,
      "loss": 0.0647,
      "theoretical_loss": 3.361561208859843,
      "tokens_seen": 2859466752
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013472957791686727,
      "loss": 0.0649,
      "theoretical_loss": 3.361537486804736,
      "tokens_seen": 2859728896
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013464933397528488,
      "loss": 0.0657,
      "theoretical_loss": 3.3615137675328763,
      "tokens_seen": 2859991040
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013456909003370246,
      "loss": 0.066,
      "theoretical_loss": 3.361490051043683,
      "tokens_seen": 2860253184
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013448884609212005,
      "loss": 0.0652,
      "theoretical_loss": 3.361466337336574,
      "tokens_seen": 2860515328
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013440860215053763,
      "loss": 0.066,
      "theoretical_loss": 3.3614426264109687,
      "tokens_seen": 2860777472
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013432835820895522,
      "loss": 0.0663,
      "theoretical_loss": 3.3614189182662857,
      "tokens_seen": 2861039616
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013424811426737283,
      "loss": 0.0649,
      "theoretical_loss": 3.361395212901944,
      "tokens_seen": 2861301760
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001341678703257904,
      "loss": 0.0646,
      "theoretical_loss": 3.3613715103173627,
      "tokens_seen": 2861563904
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013408762638420798,
      "loss": 0.0658,
      "theoretical_loss": 3.361347810511962,
      "tokens_seen": 2861826048
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001340073824426256,
      "loss": 0.0629,
      "theoretical_loss": 3.36132411348516,
      "tokens_seen": 2862088192
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013392713850104317,
      "loss": 0.0657,
      "theoretical_loss": 3.3613004192363785,
      "tokens_seen": 2862350336
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013384689455946076,
      "loss": 0.0644,
      "theoretical_loss": 3.361276727765036,
      "tokens_seen": 2862612480
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013376665061787835,
      "loss": 0.0656,
      "theoretical_loss": 3.361253039070553,
      "tokens_seen": 2862874624
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013368640667629596,
      "loss": 0.0641,
      "theoretical_loss": 3.36122935315235,
      "tokens_seen": 2863136768
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013360616273471354,
      "loss": 0.0665,
      "theoretical_loss": 3.361205670009847,
      "tokens_seen": 2863398912
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001335259187931311,
      "loss": 0.0651,
      "theoretical_loss": 3.361181989642465,
      "tokens_seen": 2863661056
    },
    {
      "epoch": 0.87,
      "objective/train/advantage_avg": 0.00021905562607571483,
      "objective/train/docs_used": 1040737,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.214687466621399,
      "objective/train/original_loss": 1.214687466621399,
      "objective/train/theoretical_loss": 3.3611583120496253,
      "objective/train/tokens_used": 2884383200,
      "objective/train/value_avg": -0.006732940673828125,
      "objective/train/value_loss": 0.0001407399686286226,
      "objective/train/value_max": -1.895427703857422e-05,
      "objective/train/value_min": -0.482177734375,
      "objective/train/value_reward_corr": 0.780282191310512,
      "objective/train/value_std": 0.01496124267578125,
      "objective/train/weight_avg": 1.00028657913208,
      "objective/train/weighted_lm_loss": 1.214595079421997,
      "objective/train/weights_max": 1.286849856376648,
      "objective/train/weights_min": 0.42503201961517334,
      "theoretical_loss": 3.3611583120496253,
      "tokens_seen": 2863923200
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013344567485154871,
      "loss": 0.0604,
      "theoretical_loss": 3.3611583120496253,
      "tokens_seen": 2863923200
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001333654309099663,
      "loss": 0.0661,
      "theoretical_loss": 3.361134637230748,
      "tokens_seen": 2864185344
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001332851869683839,
      "loss": 0.0637,
      "theoretical_loss": 3.361110965185255,
      "tokens_seen": 2864447488
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013320494302680147,
      "loss": 0.0662,
      "theoretical_loss": 3.361087295912567,
      "tokens_seen": 2864709632
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013312469908521906,
      "loss": 0.0655,
      "theoretical_loss": 3.3610636294121057,
      "tokens_seen": 2864971776
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013304445514363667,
      "loss": 0.0649,
      "theoretical_loss": 3.361039965683293,
      "tokens_seen": 2865233920
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013296421120205423,
      "loss": 0.0657,
      "theoretical_loss": 3.361016304725551,
      "tokens_seen": 2865496064
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013288396726047184,
      "loss": 0.0666,
      "theoretical_loss": 3.360992646538301,
      "tokens_seen": 2865758208
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013280372331888943,
      "loss": 0.0656,
      "theoretical_loss": 3.3609689911209664,
      "tokens_seen": 2866020352
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013272347937730704,
      "loss": 0.0632,
      "theoretical_loss": 3.360945338472968,
      "tokens_seen": 2866282496
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001326432354357246,
      "loss": 0.0669,
      "theoretical_loss": 3.36092168859373,
      "tokens_seen": 2866544640
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013256299149414218,
      "loss": 0.0642,
      "theoretical_loss": 3.3608980414826735,
      "tokens_seen": 2866806784
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001324827475525598,
      "loss": 0.0644,
      "theoretical_loss": 3.3608743971392228,
      "tokens_seen": 2867068928
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013240250361097738,
      "loss": 0.0653,
      "theoretical_loss": 3.3608507555628,
      "tokens_seen": 2867331072
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013232225966939497,
      "loss": 0.0644,
      "theoretical_loss": 3.360827116752829,
      "tokens_seen": 2867593216
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013224201572781255,
      "loss": 0.0644,
      "theoretical_loss": 3.360803480708733,
      "tokens_seen": 2867855360
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013216177178623016,
      "loss": 0.064,
      "theoretical_loss": 3.360779847429936,
      "tokens_seen": 2868117504
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013208152784464772,
      "loss": 0.0633,
      "theoretical_loss": 3.3607562169158607,
      "tokens_seen": 2868379648
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001320012839030653,
      "loss": 0.0652,
      "theoretical_loss": 3.360732589165932,
      "tokens_seen": 2868641792
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013192103996148292,
      "loss": 0.0676,
      "theoretical_loss": 3.3607089641795738,
      "tokens_seen": 2868903936
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001318407960199005,
      "loss": 0.0646,
      "theoretical_loss": 3.36068534195621,
      "tokens_seen": 2869166080
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001317605520783181,
      "loss": 0.0654,
      "theoretical_loss": 3.3606617224952657,
      "tokens_seen": 2869428224
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013168030813673568,
      "loss": 0.0667,
      "theoretical_loss": 3.3606381057961654,
      "tokens_seen": 2869690368
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013160006419515326,
      "loss": 0.0632,
      "theoretical_loss": 3.3606144918583336,
      "tokens_seen": 2869952512
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013151982025357088,
      "loss": 0.0633,
      "theoretical_loss": 3.3605908806811953,
      "tokens_seen": 2870214656
    },
    {
      "epoch": 0.87,
      "objective/train/advantage_avg": 0.0008621090091764927,
      "objective/train/docs_used": 1043211,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.284454345703125,
      "objective/train/original_loss": 1.284454345703125,
      "objective/train/theoretical_loss": 3.3605672722641757,
      "objective/train/tokens_used": 2890936800,
      "objective/train/value_avg": -0.0079803466796875,
      "objective/train/value_loss": 0.00011206701310584322,
      "objective/train/value_max": -3.045797348022461e-05,
      "objective/train/value_min": -0.25439453125,
      "objective/train/value_reward_corr": 0.7807823403804984,
      "objective/train/value_std": 0.01438140869140625,
      "objective/train/weight_avg": 1.0009177923202515,
      "objective/train/weighted_lm_loss": 1.2848042249679565,
      "objective/train/weights_max": 1.1853408813476562,
      "objective/train/weights_min": 0.8233441710472107,
      "theoretical_loss": 3.3605672722641757,
      "tokens_seen": 2870476800
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013143957631198843,
      "loss": 0.0626,
      "theoretical_loss": 3.3605672722641757,
      "tokens_seen": 2870476800
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013135933237040605,
      "loss": 0.0648,
      "theoretical_loss": 3.3605436666067003,
      "tokens_seen": 2870738944
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013127908842882363,
      "loss": 0.0629,
      "theoretical_loss": 3.3605200637081944,
      "tokens_seen": 2871001088
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001311988444872412,
      "loss": 0.0659,
      "theoretical_loss": 3.360496463568084,
      "tokens_seen": 2871263232
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001311186005456588,
      "loss": 0.0676,
      "theoretical_loss": 3.360472866185794,
      "tokens_seen": 2871525376
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001310383566040764,
      "loss": 0.0671,
      "theoretical_loss": 3.3604492715607517,
      "tokens_seen": 2871787520
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.000130958112662494,
      "loss": 0.0681,
      "theoretical_loss": 3.3604256796923826,
      "tokens_seen": 2872049664
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013087786872091156,
      "loss": 0.0651,
      "theoretical_loss": 3.360402090580113,
      "tokens_seen": 2872311808
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013079762477932917,
      "loss": 0.0642,
      "theoretical_loss": 3.3603785042233696,
      "tokens_seen": 2872573952
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013071738083774676,
      "loss": 0.0616,
      "theoretical_loss": 3.360354920621579,
      "tokens_seen": 2872836096
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013063713689616434,
      "loss": 0.0667,
      "theoretical_loss": 3.3603313397741683,
      "tokens_seen": 2873098240
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013055689295458193,
      "loss": 0.0681,
      "theoretical_loss": 3.3603077616805646,
      "tokens_seen": 2873360384
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013047664901299951,
      "loss": 0.0654,
      "theoretical_loss": 3.3602841863401944,
      "tokens_seen": 2873622528
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013039640507141713,
      "loss": 0.0618,
      "theoretical_loss": 3.360260613752486,
      "tokens_seen": 2873884672
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013031616112983468,
      "loss": 0.0653,
      "theoretical_loss": 3.3602370439168663,
      "tokens_seen": 2874146816
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001302359171882523,
      "loss": 0.066,
      "theoretical_loss": 3.3602134768327634,
      "tokens_seen": 2874408960
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013015567324666988,
      "loss": 0.0646,
      "theoretical_loss": 3.360189912499605,
      "tokens_seen": 2874671104
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00013007542930508747,
      "loss": 0.0657,
      "theoretical_loss": 3.360166350916819,
      "tokens_seen": 2874933248
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012999518536350505,
      "loss": 0.0672,
      "theoretical_loss": 3.360142792083834,
      "tokens_seen": 2875195392
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012991494142192264,
      "loss": 0.0672,
      "theoretical_loss": 3.3601192360000782,
      "tokens_seen": 2875457536
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012983469748034025,
      "loss": 0.0647,
      "theoretical_loss": 3.360095682664981,
      "tokens_seen": 2875719680
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012975445353875784,
      "loss": 0.065,
      "theoretical_loss": 3.3600721320779696,
      "tokens_seen": 2875981824
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001296742095971754,
      "loss": 0.0663,
      "theoretical_loss": 3.3600485842384735,
      "tokens_seen": 2876243968
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.000129593965655593,
      "loss": 0.0663,
      "theoretical_loss": 3.3600250391459223,
      "tokens_seen": 2876506112
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001295137217140106,
      "loss": 0.0644,
      "theoretical_loss": 3.360001496799746,
      "tokens_seen": 2876768256
    },
    {
      "epoch": 0.87,
      "objective/train/advantage_avg": 0.0009444629540666938,
      "objective/train/docs_used": 1045577,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3567603826522827,
      "objective/train/original_loss": 1.3567602634429932,
      "objective/train/theoretical_loss": 3.359977957199372,
      "objective/train/tokens_used": 2897490400,
      "objective/train/value_avg": -0.007648468017578125,
      "objective/train/value_loss": 0.00022198569786269218,
      "objective/train/value_max": -1.817941665649414e-05,
      "objective/train/value_min": -0.666015625,
      "objective/train/value_reward_corr": 0.7108661837889094,
      "objective/train/value_std": 0.01611328125,
      "objective/train/weight_avg": 1.0010464191436768,
      "objective/train/weighted_lm_loss": 1.3576266765594482,
      "objective/train/weights_max": 1.7288668155670166,
      "objective/train/weights_min": 0.36866194009780884,
      "theoretical_loss": 3.359977957199372,
      "tokens_seen": 2877030400
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012943347777242818,
      "loss": 0.0676,
      "theoretical_loss": 3.359977957199372,
      "tokens_seen": 2877030400
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012935323383084577,
      "loss": 0.0666,
      "theoretical_loss": 3.3599544203442306,
      "tokens_seen": 2877292544
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012927298988926338,
      "loss": 0.0641,
      "theoretical_loss": 3.359930886233753,
      "tokens_seen": 2877554688
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012919274594768096,
      "loss": 0.065,
      "theoretical_loss": 3.3599073548673672,
      "tokens_seen": 2877816832
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012911250200609852,
      "loss": 0.0668,
      "theoretical_loss": 3.3598838262445048,
      "tokens_seen": 2878078976
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012903225806451613,
      "loss": 0.0679,
      "theoretical_loss": 3.359860300364595,
      "tokens_seen": 2878341120
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012895201412293372,
      "loss": 0.0623,
      "theoretical_loss": 3.359836777227069,
      "tokens_seen": 2878603264
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012887177018135133,
      "loss": 0.0687,
      "theoretical_loss": 3.359813256831357,
      "tokens_seen": 2878865408
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001287915262397689,
      "loss": 0.0653,
      "theoretical_loss": 3.3597897391768905,
      "tokens_seen": 2879127552
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012871128229818648,
      "loss": 0.0647,
      "theoretical_loss": 3.3597662242630992,
      "tokens_seen": 2879389696
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001286310383566041,
      "loss": 0.0657,
      "theoretical_loss": 3.359742712089415,
      "tokens_seen": 2879651840
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012855079441502167,
      "loss": 0.0659,
      "theoretical_loss": 3.3597192026552696,
      "tokens_seen": 2879913984
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012847055047343926,
      "loss": 0.0661,
      "theoretical_loss": 3.3596956959600943,
      "tokens_seen": 2880176128
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012839030653185685,
      "loss": 0.0676,
      "theoretical_loss": 3.3596721920033197,
      "tokens_seen": 2880438272
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012831006259027446,
      "loss": 0.0637,
      "theoretical_loss": 3.359648690784379,
      "tokens_seen": 2880700416
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012822981864869202,
      "loss": 0.0648,
      "theoretical_loss": 3.3596251923027034,
      "tokens_seen": 2880962560
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001281495747071096,
      "loss": 0.0642,
      "theoretical_loss": 3.3596016965577253,
      "tokens_seen": 2881224704
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012806933076552721,
      "loss": 0.0646,
      "theoretical_loss": 3.359578203548877,
      "tokens_seen": 2881486848
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001279890868239448,
      "loss": 0.0656,
      "theoretical_loss": 3.3595547132755907,
      "tokens_seen": 2881748992
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012790884288236239,
      "loss": 0.0672,
      "theoretical_loss": 3.3595312257372996,
      "tokens_seen": 2882011136
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012782859894077997,
      "loss": 0.0681,
      "theoretical_loss": 3.359507740933436,
      "tokens_seen": 2882273280
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012774835499919756,
      "loss": 0.0642,
      "theoretical_loss": 3.359484258863434,
      "tokens_seen": 2882535424
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012766811105761517,
      "loss": 0.0649,
      "theoretical_loss": 3.3594607795267253,
      "tokens_seen": 2882797568
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012758786711603273,
      "loss": 0.0658,
      "theoretical_loss": 3.359437302922744,
      "tokens_seen": 2883059712
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012750762317445034,
      "loss": 0.0637,
      "theoretical_loss": 3.3594138290509235,
      "tokens_seen": 2883321856
    },
    {
      "epoch": 0.87,
      "objective/train/advantage_avg": 0.0009214555029757321,
      "objective/train/docs_used": 1048090,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.292439579963684,
      "objective/train/original_loss": 1.2924394607543945,
      "objective/train/theoretical_loss": 3.3593903579106974,
      "objective/train/tokens_used": 2904044000,
      "objective/train/value_avg": -0.00852203369140625,
      "objective/train/value_loss": 0.00013700471026822925,
      "objective/train/value_max": -1.6033649444580078e-05,
      "objective/train/value_min": -0.274658203125,
      "objective/train/value_reward_corr": 0.8968270892350809,
      "objective/train/value_std": 0.0258941650390625,
      "objective/train/weight_avg": 1.0009853839874268,
      "objective/train/weighted_lm_loss": 1.2942968606948853,
      "objective/train/weights_max": 1.2442163228988647,
      "objective/train/weights_min": 0.3884040415287018,
      "theoretical_loss": 3.3593903579106974,
      "tokens_seen": 2883584000
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012742737923286793,
      "loss": 0.0652,
      "theoretical_loss": 3.3593903579106974,
      "tokens_seen": 2883584000
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001273471352912855,
      "loss": 0.0641,
      "theoretical_loss": 3.3593668895014996,
      "tokens_seen": 2883846144
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001272668913497031,
      "loss": 0.0645,
      "theoretical_loss": 3.3593434238227644,
      "tokens_seen": 2884108288
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012718664740812068,
      "loss": 0.0658,
      "theoretical_loss": 3.3593199608739255,
      "tokens_seen": 2884370432
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001271064034665383,
      "loss": 0.0648,
      "theoretical_loss": 3.359296500654418,
      "tokens_seen": 2884632576
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012702615952495585,
      "loss": 0.0639,
      "theoretical_loss": 3.3592730431636753,
      "tokens_seen": 2884894720
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012694591558337347,
      "loss": 0.0653,
      "theoretical_loss": 3.359249588401133,
      "tokens_seen": 2885156864
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012686567164179105,
      "loss": 0.0663,
      "theoretical_loss": 3.3592261363662255,
      "tokens_seen": 2885419008
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012678542770020864,
      "loss": 0.0645,
      "theoretical_loss": 3.359202687058388,
      "tokens_seen": 2885681152
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012670518375862622,
      "loss": 0.0683,
      "theoretical_loss": 3.3591792404770557,
      "tokens_seen": 2885943296
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001266249398170438,
      "loss": 0.065,
      "theoretical_loss": 3.359155796621664,
      "tokens_seen": 2886205440
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012654469587546142,
      "loss": 0.0657,
      "theoretical_loss": 3.3591323554916483,
      "tokens_seen": 2886467584
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012646445193387898,
      "loss": 0.0648,
      "theoretical_loss": 3.3591089170864445,
      "tokens_seen": 2886729728
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.0001263842079922966,
      "loss": 0.0634,
      "theoretical_loss": 3.3590854814054882,
      "tokens_seen": 2886991872
    },
    {
      "epoch": 0.87,
      "learning_rate": 0.00012630396405071418,
      "loss": 0.0643,
      "theoretical_loss": 3.359062048448216,
      "tokens_seen": 2887254016
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012622372010913176,
      "loss": 0.0656,
      "theoretical_loss": 3.3590386182140635,
      "tokens_seen": 2887516160
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012614347616754935,
      "loss": 0.0658,
      "theoretical_loss": 3.3590151907024675,
      "tokens_seen": 2887778304
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012606323222596693,
      "loss": 0.0685,
      "theoretical_loss": 3.3589917659128634,
      "tokens_seen": 2888040448
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012598298828438455,
      "loss": 0.0639,
      "theoretical_loss": 3.3589683438446896,
      "tokens_seen": 2888302592
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012590274434280213,
      "loss": 0.0647,
      "theoretical_loss": 3.358944924497382,
      "tokens_seen": 2888564736
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.0001258225004012197,
      "loss": 0.0656,
      "theoretical_loss": 3.3589215078703782,
      "tokens_seen": 2888826880
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.0001257422564596373,
      "loss": 0.0638,
      "theoretical_loss": 3.3588980939631146,
      "tokens_seen": 2889089024
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.0001256620125180549,
      "loss": 0.0657,
      "theoretical_loss": 3.358874682775029,
      "tokens_seen": 2889351168
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012558176857647247,
      "loss": 0.0654,
      "theoretical_loss": 3.358851274305559,
      "tokens_seen": 2889613312
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012550152463489006,
      "loss": 0.0653,
      "theoretical_loss": 3.3588278685541426,
      "tokens_seen": 2889875456
    },
    {
      "epoch": 0.88,
      "objective/train/advantage_avg": 9.303795377491042e-05,
      "objective/train/docs_used": 1050561,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3007079362869263,
      "objective/train/original_loss": 1.3007079362869263,
      "objective/train/theoretical_loss": 3.358804465520217,
      "objective/train/tokens_used": 2910597600,
      "objective/train/value_avg": -0.0085906982421875,
      "objective/train/value_loss": 0.00047798847663216293,
      "objective/train/value_max": -1.1205673217773438e-05,
      "objective/train/value_min": -0.9912109375,
      "objective/train/value_reward_corr": 0.7482428192656447,
      "objective/train/value_std": 0.0229644775390625,
      "objective/train/weight_avg": 1.0002946853637695,
      "objective/train/weighted_lm_loss": 1.2997008562088013,
      "objective/train/weights_max": 2.048863172531128,
      "objective/train/weights_min": 0.22682039439678192,
      "theoretical_loss": 3.358804465520217,
      "tokens_seen": 2890137600
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012542128069330767,
      "loss": 0.0652,
      "theoretical_loss": 3.358804465520217,
      "tokens_seen": 2890137600
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012534103675172526,
      "loss": 0.064,
      "theoretical_loss": 3.358781065203221,
      "tokens_seen": 2890399744
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012526079281014282,
      "loss": 0.0664,
      "theoretical_loss": 3.358757667602592,
      "tokens_seen": 2890661888
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012518054886856043,
      "loss": 0.0651,
      "theoretical_loss": 3.3587342727177694,
      "tokens_seen": 2890924032
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012510030492697801,
      "loss": 0.0647,
      "theoretical_loss": 3.3587108805481907,
      "tokens_seen": 2891186176
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012502006098539563,
      "loss": 0.0635,
      "theoretical_loss": 3.358687491093295,
      "tokens_seen": 2891448320
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.0001249398170438132,
      "loss": 0.0646,
      "theoretical_loss": 3.3586641043525214,
      "tokens_seen": 2891710464
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012485957310223077,
      "loss": 0.0665,
      "theoretical_loss": 3.358640720325309,
      "tokens_seen": 2891972608
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012477932916064838,
      "loss": 0.0657,
      "theoretical_loss": 3.3586173390110967,
      "tokens_seen": 2892234752
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012469908521906597,
      "loss": 0.063,
      "theoretical_loss": 3.358593960409324,
      "tokens_seen": 2892496896
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012461884127748355,
      "loss": 0.065,
      "theoretical_loss": 3.3585705845194305,
      "tokens_seen": 2892759040
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012453859733590114,
      "loss": 0.0657,
      "theoretical_loss": 3.3585472113408557,
      "tokens_seen": 2893021184
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012445835339431873,
      "loss": 0.0643,
      "theoretical_loss": 3.35852384087304,
      "tokens_seen": 2893283328
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.0001243781094527363,
      "loss": 0.0644,
      "theoretical_loss": 3.3585004731154227,
      "tokens_seen": 2893545472
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.0001242978655111539,
      "loss": 0.0647,
      "theoretical_loss": 3.358477108067445,
      "tokens_seen": 2893807616
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.0001242176215695715,
      "loss": 0.0669,
      "theoretical_loss": 3.3584537457285464,
      "tokens_seen": 2894069760
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.0001241373776279891,
      "loss": 0.0654,
      "theoretical_loss": 3.3584303860981675,
      "tokens_seen": 2894331904
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012405713368640668,
      "loss": 0.0682,
      "theoretical_loss": 3.3584070291757495,
      "tokens_seen": 2894594048
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012397688974482427,
      "loss": 0.0641,
      "theoretical_loss": 3.358383674960733,
      "tokens_seen": 2894856192
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012389664580324185,
      "loss": 0.0658,
      "theoretical_loss": 3.358360323452559,
      "tokens_seen": 2895118336
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012381640186165944,
      "loss": 0.0641,
      "theoretical_loss": 3.358336974650669,
      "tokens_seen": 2895380480
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012373615792007702,
      "loss": 0.0618,
      "theoretical_loss": 3.3583136285545043,
      "tokens_seen": 2895642624
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012365591397849463,
      "loss": 0.0646,
      "theoretical_loss": 3.3582902851635064,
      "tokens_seen": 2895904768
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012357567003691222,
      "loss": 0.0651,
      "theoretical_loss": 3.3582669444771165,
      "tokens_seen": 2896166912
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.0001234954260953298,
      "loss": 0.0636,
      "theoretical_loss": 3.3582436064947774,
      "tokens_seen": 2896429056
    },
    {
      "epoch": 0.88,
      "objective/train/advantage_avg": 0.000869004346895963,
      "objective/train/docs_used": 1053062,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2724812030792236,
      "objective/train/original_loss": 1.2724812030792236,
      "objective/train/theoretical_loss": 3.358220271215931,
      "objective/train/tokens_used": 2917151200,
      "objective/train/value_avg": -0.0066986083984375,
      "objective/train/value_loss": 0.000132927336380817,
      "objective/train/value_max": -1.7523765563964844e-05,
      "objective/train/value_min": -0.335693359375,
      "objective/train/value_reward_corr": 0.7094154754724421,
      "objective/train/value_std": 0.01219940185546875,
      "objective/train/weight_avg": 1.0009315013885498,
      "objective/train/weighted_lm_loss": 1.2730143070220947,
      "objective/train/weights_max": 1.165982961654663,
      "objective/train/weights_min": 0.3854755163192749,
      "theoretical_loss": 3.358220271215931,
      "tokens_seen": 2896691200
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.0001234151821537474,
      "loss": 0.0642,
      "theoretical_loss": 3.358220271215931,
      "tokens_seen": 2896691200
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012333493821216498,
      "loss": 0.0645,
      "theoretical_loss": 3.358196938640018,
      "tokens_seen": 2896953344
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012325469427058256,
      "loss": 0.068,
      "theoretical_loss": 3.358173608766483,
      "tokens_seen": 2897215488
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012317445032900017,
      "loss": 0.0683,
      "theoretical_loss": 3.358150281594767,
      "tokens_seen": 2897477632
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012309420638741776,
      "loss": 0.0646,
      "theoretical_loss": 3.358126957124313,
      "tokens_seen": 2897739776
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012301396244583535,
      "loss": 0.065,
      "theoretical_loss": 3.3581036353545644,
      "tokens_seen": 2898001920
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012293371850425293,
      "loss": 0.0643,
      "theoretical_loss": 3.3580803162849637,
      "tokens_seen": 2898264064
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012285347456267052,
      "loss": 0.0639,
      "theoretical_loss": 3.3580569999149548,
      "tokens_seen": 2898526208
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.0001227732306210881,
      "loss": 0.0632,
      "theoretical_loss": 3.35803368624398,
      "tokens_seen": 2898788352
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.0001226929866795057,
      "loss": 0.0625,
      "theoretical_loss": 3.358010375271484,
      "tokens_seen": 2899050496
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.0001226127427379233,
      "loss": 0.066,
      "theoretical_loss": 3.357987066996909,
      "tokens_seen": 2899312640
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012253249879634089,
      "loss": 0.0652,
      "theoretical_loss": 3.3579637614197004,
      "tokens_seen": 2899574784
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012245225485475847,
      "loss": 0.0649,
      "theoretical_loss": 3.3579404585393013,
      "tokens_seen": 2899836928
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012237201091317606,
      "loss": 0.0672,
      "theoretical_loss": 3.357917158355156,
      "tokens_seen": 2900099072
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012229176697159364,
      "loss": 0.0667,
      "theoretical_loss": 3.3578938608667097,
      "tokens_seen": 2900361216
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012221152303001123,
      "loss": 0.0624,
      "theoretical_loss": 3.3578705660734056,
      "tokens_seen": 2900623360
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012213127908842884,
      "loss": 0.0633,
      "theoretical_loss": 3.3578472739746896,
      "tokens_seen": 2900885504
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012205103514684641,
      "loss": 0.064,
      "theoretical_loss": 3.3578239845700053,
      "tokens_seen": 2901147648
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012197079120526401,
      "loss": 0.0642,
      "theoretical_loss": 3.357800697858799,
      "tokens_seen": 2901409792
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.0001218905472636816,
      "loss": 0.0643,
      "theoretical_loss": 3.3577774138405148,
      "tokens_seen": 2901671936
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012181030332209918,
      "loss": 0.0645,
      "theoretical_loss": 3.3577541325145988,
      "tokens_seen": 2901934080
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012173005938051678,
      "loss": 0.0649,
      "theoretical_loss": 3.357730853880496,
      "tokens_seen": 2902196224
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012164981543893435,
      "loss": 0.0628,
      "theoretical_loss": 3.3577075779376524,
      "tokens_seen": 2902458368
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012156957149735195,
      "loss": 0.0656,
      "theoretical_loss": 3.3576843046855136,
      "tokens_seen": 2902720512
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012148932755576954,
      "loss": 0.0638,
      "theoretical_loss": 3.3576610341235256,
      "tokens_seen": 2902982656
    },
    {
      "epoch": 0.88,
      "objective/train/advantage_avg": 9.657556074671447e-05,
      "objective/train/docs_used": 1055596,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.330022931098938,
      "objective/train/original_loss": 1.3300228118896484,
      "objective/train/theoretical_loss": 3.357637766251135,
      "objective/train/tokens_used": 2923704800,
      "objective/train/value_avg": -0.00839996337890625,
      "objective/train/value_loss": 0.00018951736274175346,
      "objective/train/value_max": -3.3736228942871094e-05,
      "objective/train/value_min": -0.25439453125,
      "objective/train/value_reward_corr": 0.7220051445696589,
      "objective/train/value_std": 0.01517486572265625,
      "objective/train/weight_avg": 1.000184178352356,
      "objective/train/weighted_lm_loss": 1.3297585248947144,
      "objective/train/weights_max": 1.1480686664581299,
      "objective/train/weights_min": 0.36959975957870483,
      "theoretical_loss": 3.357637766251135,
      "tokens_seen": 2903244800
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012140908361418714,
      "loss": 0.0647,
      "theoretical_loss": 3.357637766251135,
      "tokens_seen": 2903244800
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012132883967260472,
      "loss": 0.0663,
      "theoretical_loss": 3.3576145010677876,
      "tokens_seen": 2903506944
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012124859573102232,
      "loss": 0.0622,
      "theoretical_loss": 3.35759123857293,
      "tokens_seen": 2903769088
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.0001211683517894399,
      "loss": 0.0656,
      "theoretical_loss": 3.357567978766009,
      "tokens_seen": 2904031232
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012108810784785749,
      "loss": 0.0656,
      "theoretical_loss": 3.3575447216464718,
      "tokens_seen": 2904293376
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012100786390627508,
      "loss": 0.0632,
      "theoretical_loss": 3.3575214672137648,
      "tokens_seen": 2904555520
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012092761996469266,
      "loss": 0.0653,
      "theoretical_loss": 3.357498215467335,
      "tokens_seen": 2904817664
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012084737602311026,
      "loss": 0.06,
      "theoretical_loss": 3.35747496640663,
      "tokens_seen": 2905079808
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012076713208152785,
      "loss": 0.0666,
      "theoretical_loss": 3.3574517200310976,
      "tokens_seen": 2905341952
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012068688813994543,
      "loss": 0.0611,
      "theoretical_loss": 3.3574284763401847,
      "tokens_seen": 2905604096
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012060664419836302,
      "loss": 0.0661,
      "theoretical_loss": 3.35740523533334,
      "tokens_seen": 2905866240
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012052640025678062,
      "loss": 0.0648,
      "theoretical_loss": 3.3573819970100107,
      "tokens_seen": 2906128384
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.0001204461563151982,
      "loss": 0.0651,
      "theoretical_loss": 3.357358761369645,
      "tokens_seen": 2906390528
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.0001203659123736158,
      "loss": 0.0635,
      "theoretical_loss": 3.357335528411692,
      "tokens_seen": 2906652672
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012028566843203339,
      "loss": 0.0677,
      "theoretical_loss": 3.357312298135599,
      "tokens_seen": 2906914816
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012020542449045097,
      "loss": 0.0631,
      "theoretical_loss": 3.3572890705408156,
      "tokens_seen": 2907176960
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012012518054886856,
      "loss": 0.065,
      "theoretical_loss": 3.35726584562679,
      "tokens_seen": 2907439104
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00012004493660728616,
      "loss": 0.0653,
      "theoretical_loss": 3.357242623392971,
      "tokens_seen": 2907701248
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011996469266570374,
      "loss": 0.0636,
      "theoretical_loss": 3.3572194038388083,
      "tokens_seen": 2907963392
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011988444872412133,
      "loss": 0.0647,
      "theoretical_loss": 3.3571961869637508,
      "tokens_seen": 2908225536
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011980420478253893,
      "loss": 0.0633,
      "theoretical_loss": 3.357172972767248,
      "tokens_seen": 2908487680
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.0001197239608409565,
      "loss": 0.0637,
      "theoretical_loss": 3.357149761248749,
      "tokens_seen": 2908749824
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.0001196437168993741,
      "loss": 0.065,
      "theoretical_loss": 3.357126552407705,
      "tokens_seen": 2909011968
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011956347295779169,
      "loss": 0.066,
      "theoretical_loss": 3.3571033462435644,
      "tokens_seen": 2909274112
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011948322901620928,
      "loss": 0.0667,
      "theoretical_loss": 3.3570801427557777,
      "tokens_seen": 2909536256
    },
    {
      "epoch": 0.88,
      "objective/train/advantage_avg": -0.0011890976456925273,
      "objective/train/docs_used": 1057798,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2657890319824219,
      "objective/train/original_loss": 1.2657889127731323,
      "objective/train/theoretical_loss": 3.357056941943796,
      "objective/train/tokens_used": 2930258400,
      "objective/train/value_avg": -0.009063720703125,
      "objective/train/value_loss": 0.0001873580040410161,
      "objective/train/value_max": -2.2649765014648438e-05,
      "objective/train/value_min": -0.325439453125,
      "objective/train/value_reward_corr": 0.9082124581128177,
      "objective/train/value_std": 0.023529052734375,
      "objective/train/weight_avg": 0.9988996982574463,
      "objective/train/weighted_lm_loss": 1.2639284133911133,
      "objective/train/weights_max": 1.3067957162857056,
      "objective/train/weights_min": 0.37186679244041443,
      "theoretical_loss": 3.357056941943796,
      "tokens_seen": 2909798400
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011940298507462687,
      "loss": 0.0629,
      "theoretical_loss": 3.357056941943796,
      "tokens_seen": 2909798400
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011932274113304447,
      "loss": 0.0643,
      "theoretical_loss": 3.3570337438070683,
      "tokens_seen": 2910060544
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011924249719146204,
      "loss": 0.0667,
      "theoretical_loss": 3.357010548345046,
      "tokens_seen": 2910322688
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011916225324987964,
      "loss": 0.0668,
      "theoretical_loss": 3.3569873555571803,
      "tokens_seen": 2910584832
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011908200930829723,
      "loss": 0.066,
      "theoretical_loss": 3.3569641654429208,
      "tokens_seen": 2910846976
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011900176536671481,
      "loss": 0.0643,
      "theoretical_loss": 3.3569409780017194,
      "tokens_seen": 2911109120
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011892152142513241,
      "loss": 0.0648,
      "theoretical_loss": 3.356917793233027,
      "tokens_seen": 2911371264
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011884127748355,
      "loss": 0.0656,
      "theoretical_loss": 3.356894611136296,
      "tokens_seen": 2911633408
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011876103354196758,
      "loss": 0.0648,
      "theoretical_loss": 3.3568714317109762,
      "tokens_seen": 2911895552
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011868078960038517,
      "loss": 0.0627,
      "theoretical_loss": 3.3568482549565206,
      "tokens_seen": 2912157696
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011860054565880277,
      "loss": 0.0639,
      "theoretical_loss": 3.3568250808723805,
      "tokens_seen": 2912419840
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011852030171722035,
      "loss": 0.0671,
      "theoretical_loss": 3.356801909458008,
      "tokens_seen": 2912681984
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011844005777563795,
      "loss": 0.0643,
      "theoretical_loss": 3.3567787407128558,
      "tokens_seen": 2912944128
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011835981383405554,
      "loss": 0.065,
      "theoretical_loss": 3.3567555746363755,
      "tokens_seen": 2913206272
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011827956989247312,
      "loss": 0.0666,
      "theoretical_loss": 3.35673241122802,
      "tokens_seen": 2913468416
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.0001181993259508907,
      "loss": 0.0664,
      "theoretical_loss": 3.356709250487242,
      "tokens_seen": 2913730560
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011811908200930829,
      "loss": 0.0636,
      "theoretical_loss": 3.356686092413494,
      "tokens_seen": 2913992704
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011803883806772589,
      "loss": 0.064,
      "theoretical_loss": 3.3566629370062295,
      "tokens_seen": 2914254848
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011795859412614348,
      "loss": 0.0669,
      "theoretical_loss": 3.3566397842649014,
      "tokens_seen": 2914516992
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011787835018456108,
      "loss": 0.0666,
      "theoretical_loss": 3.356616634188963,
      "tokens_seen": 2914779136
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011779810624297865,
      "loss": 0.0652,
      "theoretical_loss": 3.356593486777868,
      "tokens_seen": 2915041280
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011771786230139625,
      "loss": 0.0641,
      "theoretical_loss": 3.3565703420310697,
      "tokens_seen": 2915303424
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011763761835981383,
      "loss": 0.066,
      "theoretical_loss": 3.356547199948022,
      "tokens_seen": 2915565568
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011755737441823143,
      "loss": 0.0662,
      "theoretical_loss": 3.356524060528179,
      "tokens_seen": 2915827712
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011747713047664902,
      "loss": 0.0642,
      "theoretical_loss": 3.356500923770995,
      "tokens_seen": 2916089856
    },
    {
      "epoch": 0.88,
      "objective/train/advantage_avg": 0.00014633123646490276,
      "objective/train/docs_used": 1060149,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2154415845870972,
      "objective/train/original_loss": 1.2154414653778076,
      "objective/train/theoretical_loss": 3.3564777896759237,
      "objective/train/tokens_used": 2936812000,
      "objective/train/value_avg": -0.00882720947265625,
      "objective/train/value_loss": 0.00020796981698367745,
      "objective/train/value_max": -1.6570091247558594e-05,
      "objective/train/value_min": -0.29736328125,
      "objective/train/value_reward_corr": 0.8001300778151015,
      "objective/train/value_std": 0.01922607421875,
      "objective/train/weight_avg": 1.0002456903457642,
      "objective/train/weighted_lm_loss": 1.215817928314209,
      "objective/train/weights_max": 1.2599588632583618,
      "objective/train/weights_min": 0.3954486548900604,
      "theoretical_loss": 3.3564777896759237,
      "tokens_seen": 2916352000
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011739688653506662,
      "loss": 0.066,
      "theoretical_loss": 3.3564777896759237,
      "tokens_seen": 2916352000
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011731664259348419,
      "loss": 0.0654,
      "theoretical_loss": 3.35645465824242,
      "tokens_seen": 2916614144
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011723639865190179,
      "loss": 0.0634,
      "theoretical_loss": 3.3564315294699383,
      "tokens_seen": 2916876288
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011715615471031937,
      "loss": 0.0652,
      "theoretical_loss": 3.3564084033579333,
      "tokens_seen": 2917138432
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011707591076873696,
      "loss": 0.0655,
      "theoretical_loss": 3.3563852799058607,
      "tokens_seen": 2917400576
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011699566682715456,
      "loss": 0.0633,
      "theoretical_loss": 3.3563621591131745,
      "tokens_seen": 2917662720
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011691542288557214,
      "loss": 0.0644,
      "theoretical_loss": 3.3563390409793303,
      "tokens_seen": 2917924864
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011683517894398973,
      "loss": 0.0663,
      "theoretical_loss": 3.356315925503784,
      "tokens_seen": 2918187008
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011675493500240731,
      "loss": 0.0656,
      "theoretical_loss": 3.356292812685991,
      "tokens_seen": 2918449152
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011667469106082491,
      "loss": 0.0634,
      "theoretical_loss": 3.3562697025254065,
      "tokens_seen": 2918711296
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.0001165944471192425,
      "loss": 0.0644,
      "theoretical_loss": 3.356246595021487,
      "tokens_seen": 2918973440
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.0001165142031776601,
      "loss": 0.0681,
      "theoretical_loss": 3.356223490173688,
      "tokens_seen": 2919235584
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011643395923607768,
      "loss": 0.0647,
      "theoretical_loss": 3.356200387981466,
      "tokens_seen": 2919497728
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011635371529449527,
      "loss": 0.0668,
      "theoretical_loss": 3.3561772884442775,
      "tokens_seen": 2919759872
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011627347135291285,
      "loss": 0.0636,
      "theoretical_loss": 3.356154191561579,
      "tokens_seen": 2920022016
    },
    {
      "epoch": 0.88,
      "learning_rate": 0.00011619322741133044,
      "loss": 0.066,
      "theoretical_loss": 3.356131097332827,
      "tokens_seen": 2920284160
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011611298346974804,
      "loss": 0.0662,
      "theoretical_loss": 3.356108005757479,
      "tokens_seen": 2920546304
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011603273952816562,
      "loss": 0.0686,
      "theoretical_loss": 3.3560849168349907,
      "tokens_seen": 2920808448
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011595249558658322,
      "loss": 0.0648,
      "theoretical_loss": 3.3560618305648204,
      "tokens_seen": 2921070592
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.0001158722516450008,
      "loss": 0.0636,
      "theoretical_loss": 3.3560387469464255,
      "tokens_seen": 2921332736
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.0001157920077034184,
      "loss": 0.0671,
      "theoretical_loss": 3.356015665979262,
      "tokens_seen": 2921594880
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011571176376183598,
      "loss": 0.065,
      "theoretical_loss": 3.3559925876627896,
      "tokens_seen": 2921857024
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011563151982025358,
      "loss": 0.0652,
      "theoretical_loss": 3.3559695119964648,
      "tokens_seen": 2922119168
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011555127587867116,
      "loss": 0.0641,
      "theoretical_loss": 3.355946438979746,
      "tokens_seen": 2922381312
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011547103193708876,
      "loss": 0.0648,
      "theoretical_loss": 3.3559233686120913,
      "tokens_seen": 2922643456
    },
    {
      "epoch": 0.89,
      "objective/train/advantage_avg": 0.00019381623133085668,
      "objective/train/docs_used": 1062456,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2408287525177002,
      "objective/train/original_loss": 1.2408287525177002,
      "objective/train/theoretical_loss": 3.3559003008929587,
      "objective/train/tokens_used": 2943365600,
      "objective/train/value_avg": -0.00800323486328125,
      "objective/train/value_loss": 0.00016172441246453673,
      "objective/train/value_max": -2.7120113372802734e-05,
      "objective/train/value_min": -0.5068359375,
      "objective/train/value_reward_corr": 0.7553830884393562,
      "objective/train/value_std": 0.0154571533203125,
      "objective/train/weight_avg": 1.0002673864364624,
      "objective/train/weighted_lm_loss": 1.2412976026535034,
      "objective/train/weights_max": 1.332244634628296,
      "objective/train/weights_min": 0.3701034188270569,
      "theoretical_loss": 3.3559003008929587,
      "tokens_seen": 2922905600
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011539078799550633,
      "loss": 0.0644,
      "theoretical_loss": 3.3559003008929587,
      "tokens_seen": 2922905600
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011531054405392392,
      "loss": 0.0637,
      "theoretical_loss": 3.355877235821807,
      "tokens_seen": 2923167744
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011523030011234152,
      "loss": 0.0647,
      "theoretical_loss": 3.355854173398095,
      "tokens_seen": 2923429888
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.0001151500561707591,
      "loss": 0.067,
      "theoretical_loss": 3.3558311136212806,
      "tokens_seen": 2923692032
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.0001150698122291767,
      "loss": 0.0648,
      "theoretical_loss": 3.355808056490824,
      "tokens_seen": 2923954176
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011498956828759429,
      "loss": 0.0677,
      "theoretical_loss": 3.3557850020061832,
      "tokens_seen": 2924216320
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011490932434601187,
      "loss": 0.066,
      "theoretical_loss": 3.355761950166818,
      "tokens_seen": 2924478464
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011482908040442946,
      "loss": 0.0693,
      "theoretical_loss": 3.355738900972187,
      "tokens_seen": 2924740608
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011474883646284706,
      "loss": 0.0644,
      "theoretical_loss": 3.355715854421751,
      "tokens_seen": 2925002752
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011466859252126464,
      "loss": 0.0662,
      "theoretical_loss": 3.3556928105149693,
      "tokens_seen": 2925264896
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011458834857968224,
      "loss": 0.0638,
      "theoretical_loss": 3.355669769251301,
      "tokens_seen": 2925527040
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011450810463809983,
      "loss": 0.0648,
      "theoretical_loss": 3.355646730630207,
      "tokens_seen": 2925789184
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011442786069651743,
      "loss": 0.0669,
      "theoretical_loss": 3.3556236946511473,
      "tokens_seen": 2926051328
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.000114347616754935,
      "loss": 0.0621,
      "theoretical_loss": 3.355600661313582,
      "tokens_seen": 2926313472
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011426737281335259,
      "loss": 0.0642,
      "theoretical_loss": 3.3555776306169722,
      "tokens_seen": 2926575616
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011418712887177019,
      "loss": 0.0639,
      "theoretical_loss": 3.355554602560778,
      "tokens_seen": 2926837760
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011410688493018777,
      "loss": 0.0642,
      "theoretical_loss": 3.3555315771444603,
      "tokens_seen": 2927099904
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011402664098860537,
      "loss": 0.0669,
      "theoretical_loss": 3.3555085543674803,
      "tokens_seen": 2927362048
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011394639704702294,
      "loss": 0.0669,
      "theoretical_loss": 3.355485534229299,
      "tokens_seen": 2927624192
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011386615310544054,
      "loss": 0.064,
      "theoretical_loss": 3.3554625167293777,
      "tokens_seen": 2927886336
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011378590916385813,
      "loss": 0.0655,
      "theoretical_loss": 3.355439501867178,
      "tokens_seen": 2928148480
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011370566522227573,
      "loss": 0.0635,
      "theoretical_loss": 3.3554164896421614,
      "tokens_seen": 2928410624
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011362542128069331,
      "loss": 0.066,
      "theoretical_loss": 3.35539348005379,
      "tokens_seen": 2928672768
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011354517733911091,
      "loss": 0.0656,
      "theoretical_loss": 3.3553704731015253,
      "tokens_seen": 2928934912
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.0001134649333975285,
      "loss": 0.063,
      "theoretical_loss": 3.3553474687848297,
      "tokens_seen": 2929197056
    },
    {
      "epoch": 0.89,
      "objective/train/advantage_avg": 0.0012888078344985843,
      "objective/train/docs_used": 1064613,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.31832754611969,
      "objective/train/original_loss": 1.31832754611969,
      "objective/train/theoretical_loss": 3.355324467103165,
      "objective/train/tokens_used": 2949919200,
      "objective/train/value_avg": -0.00921630859375,
      "objective/train/value_loss": 0.0002458643866702914,
      "objective/train/value_max": -2.9981136322021484e-05,
      "objective/train/value_min": -0.345703125,
      "objective/train/value_reward_corr": 0.6926176165417648,
      "objective/train/value_std": 0.015594482421875,
      "objective/train/weight_avg": 1.0013985633850098,
      "objective/train/weighted_lm_loss": 1.319889783859253,
      "objective/train/weights_max": 1.2300209999084473,
      "objective/train/weights_min": 0.3723408579826355,
      "theoretical_loss": 3.355324467103165,
      "tokens_seen": 2929459200
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011338468945594607,
      "loss": 0.0656,
      "theoretical_loss": 3.355324467103165,
      "tokens_seen": 2929459200
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011330444551436367,
      "loss": 0.0661,
      "theoretical_loss": 3.3553014680559943,
      "tokens_seen": 2929721344
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011322420157278125,
      "loss": 0.0675,
      "theoretical_loss": 3.3552784716427797,
      "tokens_seen": 2929983488
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011314395763119885,
      "loss": 0.0626,
      "theoretical_loss": 3.355255477862984,
      "tokens_seen": 2930245632
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011306371368961644,
      "loss": 0.0649,
      "theoretical_loss": 3.35523248671607,
      "tokens_seen": 2930507776
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011298346974803404,
      "loss": 0.068,
      "theoretical_loss": 3.3552094982015013,
      "tokens_seen": 2930769920
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011290322580645161,
      "loss": 0.0665,
      "theoretical_loss": 3.35518651231874,
      "tokens_seen": 2931032064
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.0001128229818648692,
      "loss": 0.0656,
      "theoretical_loss": 3.3551635290672506,
      "tokens_seen": 2931294208
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011274273792328679,
      "loss": 0.0674,
      "theoretical_loss": 3.3551405484464962,
      "tokens_seen": 2931556352
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011266249398170439,
      "loss": 0.0622,
      "theoretical_loss": 3.35511757045594,
      "tokens_seen": 2931818496
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011258225004012198,
      "loss": 0.0631,
      "theoretical_loss": 3.3550945950950473,
      "tokens_seen": 2932080640
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011250200609853958,
      "loss": 0.0632,
      "theoretical_loss": 3.35507162236328,
      "tokens_seen": 2932342784
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011242176215695715,
      "loss": 0.0668,
      "theoretical_loss": 3.3550486522601037,
      "tokens_seen": 2932604928
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011234151821537473,
      "loss": 0.0666,
      "theoretical_loss": 3.355025684784982,
      "tokens_seen": 2932867072
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011226127427379233,
      "loss": 0.0638,
      "theoretical_loss": 3.35500271993738,
      "tokens_seen": 2933129216
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011218103033220992,
      "loss": 0.0634,
      "theoretical_loss": 3.354979757716762,
      "tokens_seen": 2933391360
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011210078639062752,
      "loss": 0.0636,
      "theoretical_loss": 3.3549567981225925,
      "tokens_seen": 2933653504
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.0001120205424490451,
      "loss": 0.0651,
      "theoretical_loss": 3.354933841154337,
      "tokens_seen": 2933915648
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011194029850746269,
      "loss": 0.0654,
      "theoretical_loss": 3.35491088681146,
      "tokens_seen": 2934177792
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011186005456588027,
      "loss": 0.0654,
      "theoretical_loss": 3.3548879350934273,
      "tokens_seen": 2934439936
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011177981062429787,
      "loss": 0.065,
      "theoretical_loss": 3.3548649859997033,
      "tokens_seen": 2934702080
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011169956668271546,
      "loss": 0.0658,
      "theoretical_loss": 3.354842039529755,
      "tokens_seen": 2934964224
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011161932274113306,
      "loss": 0.0658,
      "theoretical_loss": 3.3548190956830473,
      "tokens_seen": 2935226368
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011153907879955064,
      "loss": 0.0631,
      "theoretical_loss": 3.354796154459046,
      "tokens_seen": 2935488512
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011145883485796821,
      "loss": 0.0643,
      "theoretical_loss": 3.354773215857217,
      "tokens_seen": 2935750656
    },
    {
      "epoch": 0.89,
      "objective/train/advantage_avg": 0.0004477580660022795,
      "objective/train/docs_used": 1067139,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2940055131912231,
      "objective/train/original_loss": 1.2940055131912231,
      "objective/train/theoretical_loss": 3.354750279877027,
      "objective/train/tokens_used": 2956472800,
      "objective/train/value_avg": -0.008087158203125,
      "objective/train/value_loss": 0.00039907018071971834,
      "objective/train/value_max": -3.8504600524902344e-05,
      "objective/train/value_min": -0.497802734375,
      "objective/train/value_reward_corr": 0.6092954334011763,
      "objective/train/value_std": 0.01544952392578125,
      "objective/train/weight_avg": 1.000596284866333,
      "objective/train/weighted_lm_loss": 1.2948415279388428,
      "objective/train/weights_max": 1.4681942462921143,
      "objective/train/weights_min": 0.05272703617811203,
      "theoretical_loss": 3.354750279877027,
      "tokens_seen": 2936012800
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011137859091638581,
      "loss": 0.0647,
      "theoretical_loss": 3.354750279877027,
      "tokens_seen": 2936012800
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.0001112983469748034,
      "loss": 0.0648,
      "theoretical_loss": 3.3547273465179424,
      "tokens_seen": 2936274944
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.000111218103033221,
      "loss": 0.0622,
      "theoretical_loss": 3.354704415779429,
      "tokens_seen": 2936537088
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011113785909163858,
      "loss": 0.0635,
      "theoretical_loss": 3.3546814876609536,
      "tokens_seen": 2936799232
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011105761515005618,
      "loss": 0.0664,
      "theoretical_loss": 3.354658562161984,
      "tokens_seen": 2937061376
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011097737120847375,
      "loss": 0.0647,
      "theoretical_loss": 3.354635639281986,
      "tokens_seen": 2937323520
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011089712726689135,
      "loss": 0.0656,
      "theoretical_loss": 3.3546127190204276,
      "tokens_seen": 2937585664
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011081688332530894,
      "loss": 0.0664,
      "theoretical_loss": 3.354589801376775,
      "tokens_seen": 2937847808
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011073663938372654,
      "loss": 0.0641,
      "theoretical_loss": 3.3545668863504963,
      "tokens_seen": 2938109952
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011065639544214412,
      "loss": 0.066,
      "theoretical_loss": 3.3545439739410594,
      "tokens_seen": 2938372096
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011057615150056171,
      "loss": 0.0607,
      "theoretical_loss": 3.3545210641479315,
      "tokens_seen": 2938634240
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.0001104959075589793,
      "loss": 0.0643,
      "theoretical_loss": 3.354498156970581,
      "tokens_seen": 2938896384
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011041566361739688,
      "loss": 0.0653,
      "theoretical_loss": 3.3544752524084753,
      "tokens_seen": 2939158528
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011033541967581448,
      "loss": 0.0655,
      "theoretical_loss": 3.3544523504610826,
      "tokens_seen": 2939420672
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011025517573423206,
      "loss": 0.0662,
      "theoretical_loss": 3.354429451127872,
      "tokens_seen": 2939682816
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011017493179264966,
      "loss": 0.0669,
      "theoretical_loss": 3.354406554408312,
      "tokens_seen": 2939944960
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011009468785106725,
      "loss": 0.0638,
      "theoretical_loss": 3.3543836603018704,
      "tokens_seen": 2940207104
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00011001444390948483,
      "loss": 0.0655,
      "theoretical_loss": 3.354360768808017,
      "tokens_seen": 2940469248
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010993419996790242,
      "loss": 0.0655,
      "theoretical_loss": 3.35433787992622,
      "tokens_seen": 2940731392
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010985395602632002,
      "loss": 0.0667,
      "theoretical_loss": 3.354314993655949,
      "tokens_seen": 2940993536
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.0001097737120847376,
      "loss": 0.064,
      "theoretical_loss": 3.354292109996673,
      "tokens_seen": 2941255680
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.0001096934681431552,
      "loss": 0.0646,
      "theoretical_loss": 3.354269228947862,
      "tokens_seen": 2941517824
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010961322420157279,
      "loss": 0.0663,
      "theoretical_loss": 3.354246350508985,
      "tokens_seen": 2941779968
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010953298025999036,
      "loss": 0.0628,
      "theoretical_loss": 3.354223474679512,
      "tokens_seen": 2942042112
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010945273631840796,
      "loss": 0.0648,
      "theoretical_loss": 3.354200601458913,
      "tokens_seen": 2942304256
    },
    {
      "epoch": 0.89,
      "objective/train/advantage_avg": 0.000351957103703171,
      "objective/train/docs_used": 1069745,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.43468177318573,
      "objective/train/original_loss": 1.43468177318573,
      "objective/train/theoretical_loss": 3.354177730846658,
      "objective/train/tokens_used": 2963026400,
      "objective/train/value_avg": -0.007633209228515625,
      "objective/train/value_loss": 0.0003218376077711582,
      "objective/train/value_max": -3.349781036376953e-05,
      "objective/train/value_min": -0.71435546875,
      "objective/train/value_reward_corr": 0.7403361391563008,
      "objective/train/value_std": 0.0193939208984375,
      "objective/train/weight_avg": 1.0004934072494507,
      "objective/train/weighted_lm_loss": 1.4347542524337769,
      "objective/train/weights_max": 1.8865798711776733,
      "objective/train/weights_min": 0.23063500225543976,
      "theoretical_loss": 3.354177730846658,
      "tokens_seen": 2942566400
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010937249237682555,
      "loss": 0.0688,
      "theoretical_loss": 3.354177730846658,
      "tokens_seen": 2942566400
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010929224843524314,
      "loss": 0.0654,
      "theoretical_loss": 3.354154862842217,
      "tokens_seen": 2942828544
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010921200449366073,
      "loss": 0.0664,
      "theoretical_loss": 3.354131997445061,
      "tokens_seen": 2943090688
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010913176055207833,
      "loss": 0.0644,
      "theoretical_loss": 3.3541091346546597,
      "tokens_seen": 2943352832
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.0001090515166104959,
      "loss": 0.0651,
      "theoretical_loss": 3.3540862744704842,
      "tokens_seen": 2943614976
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.0001089712726689135,
      "loss": 0.0661,
      "theoretical_loss": 3.3540634168920063,
      "tokens_seen": 2943877120
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010889102872733109,
      "loss": 0.0668,
      "theoretical_loss": 3.354040561918695,
      "tokens_seen": 2944139264
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010881078478574869,
      "loss": 0.0681,
      "theoretical_loss": 3.354017709550023,
      "tokens_seen": 2944401408
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010873054084416627,
      "loss": 0.0643,
      "theoretical_loss": 3.353994859785461,
      "tokens_seen": 2944663552
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010865029690258386,
      "loss": 0.0668,
      "theoretical_loss": 3.353972012624481,
      "tokens_seen": 2944925696
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010857005296100144,
      "loss": 0.0654,
      "theoretical_loss": 3.3539491680665545,
      "tokens_seen": 2945187840
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010848980901941903,
      "loss": 0.066,
      "theoretical_loss": 3.3539263261111523,
      "tokens_seen": 2945449984
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010840956507783663,
      "loss": 0.067,
      "theoretical_loss": 3.3539034867577473,
      "tokens_seen": 2945712128
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010832932113625421,
      "loss": 0.0655,
      "theoretical_loss": 3.353880650005811,
      "tokens_seen": 2945974272
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010824907719467181,
      "loss": 0.0649,
      "theoretical_loss": 3.3538578158548167,
      "tokens_seen": 2946236416
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.0001081688332530894,
      "loss": 0.0662,
      "theoretical_loss": 3.3538349843042354,
      "tokens_seen": 2946498560
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010808858931150698,
      "loss": 0.0654,
      "theoretical_loss": 3.3538121553535407,
      "tokens_seen": 2946760704
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010800834536992457,
      "loss": 0.0671,
      "theoretical_loss": 3.353789329002205,
      "tokens_seen": 2947022848
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010792810142834217,
      "loss": 0.0656,
      "theoretical_loss": 3.3537665052497005,
      "tokens_seen": 2947284992
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010784785748675975,
      "loss": 0.0645,
      "theoretical_loss": 3.353743684095501,
      "tokens_seen": 2947547136
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010776761354517735,
      "loss": 0.0649,
      "theoretical_loss": 3.3537208655390796,
      "tokens_seen": 2947809280
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010768736960359494,
      "loss": 0.0653,
      "theoretical_loss": 3.353698049579909,
      "tokens_seen": 2948071424
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010760712566201251,
      "loss": 0.0631,
      "theoretical_loss": 3.3536752362174633,
      "tokens_seen": 2948333568
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010752688172043011,
      "loss": 0.0629,
      "theoretical_loss": 3.3536524254512163,
      "tokens_seen": 2948595712
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010744663777884769,
      "loss": 0.0651,
      "theoretical_loss": 3.353629617280641,
      "tokens_seen": 2948857856
    },
    {
      "epoch": 0.89,
      "objective/train/advantage_avg": 0.0009776019724085927,
      "objective/train/docs_used": 1072071,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.184030532836914,
      "objective/train/original_loss": 1.1840304136276245,
      "objective/train/theoretical_loss": 3.3536068117052116,
      "objective/train/tokens_used": 2969580000,
      "objective/train/value_avg": -0.0084228515625,
      "objective/train/value_loss": 0.0003648189886007458,
      "objective/train/value_max": -2.0325183868408203e-05,
      "objective/train/value_min": -0.89697265625,
      "objective/train/value_reward_corr": 0.6921030821127245,
      "objective/train/value_std": 0.02081298828125,
      "objective/train/weight_avg": 1.0011451244354248,
      "objective/train/weighted_lm_loss": 1.1850285530090332,
      "objective/train/weights_max": 2.162311553955078,
      "objective/train/weights_min": 0.37333929538726807,
      "theoretical_loss": 3.3536068117052116,
      "tokens_seen": 2949120000
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010736639383726529,
      "loss": 0.0643,
      "theoretical_loss": 3.3536068117052116,
      "tokens_seen": 2949120000
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010728614989568288,
      "loss": 0.0615,
      "theoretical_loss": 3.3535840087244027,
      "tokens_seen": 2949382144
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010720590595410048,
      "loss": 0.0661,
      "theoretical_loss": 3.3535612083376876,
      "tokens_seen": 2949644288
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010712566201251805,
      "loss": 0.0618,
      "theoretical_loss": 3.353538410544542,
      "tokens_seen": 2949906432
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010704541807093565,
      "loss": 0.0645,
      "theoretical_loss": 3.3535156153444388,
      "tokens_seen": 2950168576
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010696517412935323,
      "loss": 0.0647,
      "theoretical_loss": 3.353492822736854,
      "tokens_seen": 2950430720
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010688493018777083,
      "loss": 0.0629,
      "theoretical_loss": 3.353470032721262,
      "tokens_seen": 2950692864
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010680468624618842,
      "loss": 0.066,
      "theoretical_loss": 3.3534472452971373,
      "tokens_seen": 2950955008
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.000106724442304606,
      "loss": 0.0622,
      "theoretical_loss": 3.3534244604639563,
      "tokens_seen": 2951217152
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010664419836302359,
      "loss": 0.064,
      "theoretical_loss": 3.353401678221193,
      "tokens_seen": 2951479296
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010656395442144117,
      "loss": 0.0636,
      "theoretical_loss": 3.3533788985683235,
      "tokens_seen": 2951741440
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010648371047985877,
      "loss": 0.0677,
      "theoretical_loss": 3.3533561215048233,
      "tokens_seen": 2952003584
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010640346653827636,
      "loss": 0.0635,
      "theoretical_loss": 3.3533333470301683,
      "tokens_seen": 2952265728
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010632322259669396,
      "loss": 0.0658,
      "theoretical_loss": 3.3533105751438343,
      "tokens_seen": 2952527872
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010624297865511154,
      "loss": 0.0631,
      "theoretical_loss": 3.3532878058452975,
      "tokens_seen": 2952790016
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010616273471352913,
      "loss": 0.0651,
      "theoretical_loss": 3.3532650391340337,
      "tokens_seen": 2953052160
    },
    {
      "epoch": 0.89,
      "learning_rate": 0.00010608249077194671,
      "loss": 0.0654,
      "theoretical_loss": 3.35324227500952,
      "tokens_seen": 2953314304
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010600224683036431,
      "loss": 0.0614,
      "theoretical_loss": 3.353219513471232,
      "tokens_seen": 2953576448
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.0001059220028887819,
      "loss": 0.0642,
      "theoretical_loss": 3.3531967545186467,
      "tokens_seen": 2953838592
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010584175894719948,
      "loss": 0.0649,
      "theoretical_loss": 3.3531739981512416,
      "tokens_seen": 2954100736
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010576151500561708,
      "loss": 0.063,
      "theoretical_loss": 3.353151244368493,
      "tokens_seen": 2954362880
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010568127106403466,
      "loss": 0.0627,
      "theoretical_loss": 3.353128493169878,
      "tokens_seen": 2954625024
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010560102712245225,
      "loss": 0.0625,
      "theoretical_loss": 3.353105744554875,
      "tokens_seen": 2954887168
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010552078318086984,
      "loss": 0.0641,
      "theoretical_loss": 3.3530829985229595,
      "tokens_seen": 2955149312
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010544053923928744,
      "loss": 0.0638,
      "theoretical_loss": 3.3530602550736104,
      "tokens_seen": 2955411456
    },
    {
      "epoch": 0.9,
      "objective/train/advantage_avg": 0.00048578670248389244,
      "objective/train/docs_used": 1074453,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3792080879211426,
      "objective/train/original_loss": 1.3792080879211426,
      "objective/train/theoretical_loss": 3.3530375142063056,
      "objective/train/tokens_used": 2976133600,
      "objective/train/value_avg": -0.0086669921875,
      "objective/train/value_loss": 0.00022539905330631882,
      "objective/train/value_max": -2.8848648071289062e-05,
      "objective/train/value_min": -0.93212890625,
      "objective/train/value_reward_corr": 0.7546802600898205,
      "objective/train/value_std": 0.016845703125,
      "objective/train/weight_avg": 1.000587821006775,
      "objective/train/weighted_lm_loss": 1.3799760341644287,
      "objective/train/weights_max": 1.2848964929580688,
      "objective/train/weights_min": 0.3680596351623535,
      "theoretical_loss": 3.3530375142063056,
      "tokens_seen": 2955673600
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010536029529770502,
      "loss": 0.0664,
      "theoretical_loss": 3.3530375142063056,
      "tokens_seen": 2955673600
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010528005135612262,
      "loss": 0.0648,
      "theoretical_loss": 3.353014775920522,
      "tokens_seen": 2955935744
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.0001051998074145402,
      "loss": 0.065,
      "theoretical_loss": 3.3529920402157387,
      "tokens_seen": 2956197888
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.0001051195634729578,
      "loss": 0.0638,
      "theoretical_loss": 3.352969307091433,
      "tokens_seen": 2956460032
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010503931953137538,
      "loss": 0.0651,
      "theoretical_loss": 3.352946576547084,
      "tokens_seen": 2956722176
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010495907558979298,
      "loss": 0.0648,
      "theoretical_loss": 3.3529238485821695,
      "tokens_seen": 2956984320
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010487883164821056,
      "loss": 0.0675,
      "theoretical_loss": 3.352901123196169,
      "tokens_seen": 2957246464
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010479858770662815,
      "loss": 0.0664,
      "theoretical_loss": 3.3528784003885606,
      "tokens_seen": 2957508608
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010471834376504575,
      "loss": 0.0614,
      "theoretical_loss": 3.3528556801588234,
      "tokens_seen": 2957770752
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010463809982346332,
      "loss": 0.0632,
      "theoretical_loss": 3.3528329625064366,
      "tokens_seen": 2958032896
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010455785588188092,
      "loss": 0.0654,
      "theoretical_loss": 3.3528102474308796,
      "tokens_seen": 2958295040
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.0001044776119402985,
      "loss": 0.0662,
      "theoretical_loss": 3.3527875349316316,
      "tokens_seen": 2958557184
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.0001043973679987161,
      "loss": 0.0651,
      "theoretical_loss": 3.3527648250081725,
      "tokens_seen": 2958819328
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010431712405713369,
      "loss": 0.0638,
      "theoretical_loss": 3.3527421176599814,
      "tokens_seen": 2959081472
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010423688011555129,
      "loss": 0.0658,
      "theoretical_loss": 3.3527194128865387,
      "tokens_seen": 2959343616
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010415663617396886,
      "loss": 0.0648,
      "theoretical_loss": 3.352696710687324,
      "tokens_seen": 2959605760
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010407639223238646,
      "loss": 0.0655,
      "theoretical_loss": 3.352674011061818,
      "tokens_seen": 2959867904
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010399614829080405,
      "loss": 0.0644,
      "theoretical_loss": 3.352651314009501,
      "tokens_seen": 2960130048
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010391590434922163,
      "loss": 0.0656,
      "theoretical_loss": 3.3526286195298525,
      "tokens_seen": 2960392192
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010383566040763923,
      "loss": 0.0629,
      "theoretical_loss": 3.352605927622354,
      "tokens_seen": 2960654336
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010375541646605682,
      "loss": 0.0647,
      "theoretical_loss": 3.352583238286486,
      "tokens_seen": 2960916480
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.0001036751725244744,
      "loss": 0.0635,
      "theoretical_loss": 3.35256055152173,
      "tokens_seen": 2961178624
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010359492858289199,
      "loss": 0.062,
      "theoretical_loss": 3.352537867327566,
      "tokens_seen": 2961440768
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010351468464130959,
      "loss": 0.0634,
      "theoretical_loss": 3.3525151857034765,
      "tokens_seen": 2961702912
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010343444069972717,
      "loss": 0.0652,
      "theoretical_loss": 3.3524925066489417,
      "tokens_seen": 2961965056
    },
    {
      "epoch": 0.9,
      "objective/train/advantage_avg": 0.000486421340610832,
      "objective/train/docs_used": 1076860,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.347232460975647,
      "objective/train/original_loss": 1.3472323417663574,
      "objective/train/theoretical_loss": 3.3524698301634435,
      "objective/train/tokens_used": 2982687200,
      "objective/train/value_avg": -0.00687408447265625,
      "objective/train/value_loss": 0.000281084852758795,
      "objective/train/value_max": -3.534555435180664e-05,
      "objective/train/value_min": -0.646484375,
      "objective/train/value_reward_corr": 0.6846278441731948,
      "objective/train/value_std": 0.01476287841796875,
      "objective/train/weight_avg": 1.0006098747253418,
      "objective/train/weighted_lm_loss": 1.3480408191680908,
      "objective/train/weights_max": 1.5544962882995605,
      "objective/train/weights_min": 0.3717136085033417,
      "theoretical_loss": 3.3524698301634435,
      "tokens_seen": 2962227200
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010335419675814477,
      "loss": 0.0652,
      "theoretical_loss": 3.3524698301634435,
      "tokens_seen": 2962227200
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010327395281656236,
      "loss": 0.0663,
      "theoretical_loss": 3.352447156246464,
      "tokens_seen": 2962489344
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010319370887497994,
      "loss": 0.0643,
      "theoretical_loss": 3.3524244848974845,
      "tokens_seen": 2962751488
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010311346493339753,
      "loss": 0.0656,
      "theoretical_loss": 3.352401816115987,
      "tokens_seen": 2963013632
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010303322099181513,
      "loss": 0.0646,
      "theoretical_loss": 3.352379149901454,
      "tokens_seen": 2963275776
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010295297705023271,
      "loss": 0.0642,
      "theoretical_loss": 3.3523564862533677,
      "tokens_seen": 2963537920
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.0001028727331086503,
      "loss": 0.0645,
      "theoretical_loss": 3.3523338251712103,
      "tokens_seen": 2963800064
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.0001027924891670679,
      "loss": 0.0637,
      "theoretical_loss": 3.3523111666544643,
      "tokens_seen": 2964062208
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010271224522548547,
      "loss": 0.0634,
      "theoretical_loss": 3.352288510702613,
      "tokens_seen": 2964324352
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010263200128390307,
      "loss": 0.0655,
      "theoretical_loss": 3.3522658573151385,
      "tokens_seen": 2964586496
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010255175734232065,
      "loss": 0.0654,
      "theoretical_loss": 3.3522432064915244,
      "tokens_seen": 2964848640
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010247151340073825,
      "loss": 0.0641,
      "theoretical_loss": 3.3522205582312536,
      "tokens_seen": 2965110784
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010239126945915584,
      "loss": 0.0652,
      "theoretical_loss": 3.3521979125338097,
      "tokens_seen": 2965372928
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010231102551757344,
      "loss": 0.0653,
      "theoretical_loss": 3.3521752693986757,
      "tokens_seen": 2965635072
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010223078157599101,
      "loss": 0.0636,
      "theoretical_loss": 3.3521526288253356,
      "tokens_seen": 2965897216
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010215053763440861,
      "loss": 0.064,
      "theoretical_loss": 3.352129990813273,
      "tokens_seen": 2966159360
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010207029369282619,
      "loss": 0.0643,
      "theoretical_loss": 3.3521073553619725,
      "tokens_seen": 2966421504
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010199004975124378,
      "loss": 0.0619,
      "theoretical_loss": 3.352084722470917,
      "tokens_seen": 2966683648
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010190980580966138,
      "loss": 0.0631,
      "theoretical_loss": 3.352062092139591,
      "tokens_seen": 2966945792
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010182956186807896,
      "loss": 0.0615,
      "theoretical_loss": 3.35203946436748,
      "tokens_seen": 2967207936
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010174931792649655,
      "loss": 0.0657,
      "theoretical_loss": 3.3520168391540675,
      "tokens_seen": 2967470080
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010166907398491413,
      "loss": 0.0639,
      "theoretical_loss": 3.3519942164988383,
      "tokens_seen": 2967732224
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010158883004333173,
      "loss": 0.067,
      "theoretical_loss": 3.3519715964012775,
      "tokens_seen": 2967994368
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010150858610174932,
      "loss": 0.0646,
      "theoretical_loss": 3.3519489788608694,
      "tokens_seen": 2968256512
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010142834216016692,
      "loss": 0.0648,
      "theoretical_loss": 3.3519263638770997,
      "tokens_seen": 2968518656
    },
    {
      "epoch": 0.9,
      "objective/train/advantage_avg": 0.0002028256276389584,
      "objective/train/docs_used": 1079221,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2540658712387085,
      "objective/train/original_loss": 1.254065752029419,
      "objective/train/theoretical_loss": 3.351903751449454,
      "objective/train/tokens_used": 2989240800,
      "objective/train/value_avg": -0.007740020751953125,
      "objective/train/value_loss": 0.00018627307144924998,
      "objective/train/value_max": -2.0444393157958984e-05,
      "objective/train/value_min": -0.6416015625,
      "objective/train/value_reward_corr": 0.7760167147041913,
      "objective/train/value_std": 0.01641845703125,
      "objective/train/weight_avg": 1.000290036201477,
      "objective/train/weighted_lm_loss": 1.2542730569839478,
      "objective/train/weights_max": 1.706524133682251,
      "objective/train/weights_min": 0.36923545598983765,
      "theoretical_loss": 3.351903751449454,
      "tokens_seen": 2968780800
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.0001013480982185845,
      "loss": 0.0631,
      "theoretical_loss": 3.351903751449454,
      "tokens_seen": 2968780800
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010126785427700209,
      "loss": 0.0641,
      "theoretical_loss": 3.351881141577417,
      "tokens_seen": 2969042944
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010118761033541967,
      "loss": 0.0631,
      "theoretical_loss": 3.3518585342604745,
      "tokens_seen": 2969305088
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010110736639383726,
      "loss": 0.0626,
      "theoretical_loss": 3.3518359294981126,
      "tokens_seen": 2969567232
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010102712245225486,
      "loss": 0.0626,
      "theoretical_loss": 3.3518133272898165,
      "tokens_seen": 2969829376
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010094687851067244,
      "loss": 0.0653,
      "theoretical_loss": 3.3517907276350725,
      "tokens_seen": 2970091520
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010086663456909004,
      "loss": 0.0642,
      "theoretical_loss": 3.351768130533367,
      "tokens_seen": 2970353664
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010078639062750762,
      "loss": 0.0646,
      "theoretical_loss": 3.351745535984186,
      "tokens_seen": 2970615808
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010070614668592521,
      "loss": 0.0641,
      "theoretical_loss": 3.3517229439870166,
      "tokens_seen": 2970877952
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.0001006259027443428,
      "loss": 0.0631,
      "theoretical_loss": 3.3517003545413444,
      "tokens_seen": 2971140096
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.0001005456588027604,
      "loss": 0.0654,
      "theoretical_loss": 3.351677767646657,
      "tokens_seen": 2971402240
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010046541486117798,
      "loss": 0.0633,
      "theoretical_loss": 3.351655183302441,
      "tokens_seen": 2971664384
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010038517091959558,
      "loss": 0.0639,
      "theoretical_loss": 3.3516326015081828,
      "tokens_seen": 2971926528
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010030492697801316,
      "loss": 0.062,
      "theoretical_loss": 3.3516100222633707,
      "tokens_seen": 2972188672
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010022468303643075,
      "loss": 0.0619,
      "theoretical_loss": 3.351587445567491,
      "tokens_seen": 2972450816
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010014443909484834,
      "loss": 0.0651,
      "theoretical_loss": 3.351564871420033,
      "tokens_seen": 2972712960
    },
    {
      "epoch": 0.9,
      "learning_rate": 0.00010006419515326593,
      "loss": 0.0623,
      "theoretical_loss": 3.351542299820482,
      "tokens_seen": 2972975104
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.998395121168352e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.3515197307683273,
      "tokens_seen": 2973237248
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.990370727010111e-05,
      "loss": 0.0651,
      "theoretical_loss": 3.3514971642630567,
      "tokens_seen": 2973499392
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.98234633285187e-05,
      "loss": 0.065,
      "theoretical_loss": 3.3514746003041576,
      "tokens_seen": 2973761536
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.974321938693628e-05,
      "loss": 0.0657,
      "theoretical_loss": 3.351452038891119,
      "tokens_seen": 2974023680
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.966297544535388e-05,
      "loss": 0.0651,
      "theoretical_loss": 3.3514294800234286,
      "tokens_seen": 2974285824
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.958273150377147e-05,
      "loss": 0.0659,
      "theoretical_loss": 3.3514069237005755,
      "tokens_seen": 2974547968
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.950248756218906e-05,
      "loss": 0.0632,
      "theoretical_loss": 3.3513843699220485,
      "tokens_seen": 2974810112
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.942224362060665e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.351361818687336,
      "tokens_seen": 2975072256
    },
    {
      "epoch": 0.9,
      "objective/train/advantage_avg": 0.0007520467042922974,
      "objective/train/docs_used": 1081640,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2578017711639404,
      "objective/train/original_loss": 1.2578016519546509,
      "objective/train/theoretical_loss": 3.3513392699959272,
      "objective/train/tokens_used": 2995794400,
      "objective/train/value_avg": -0.007770538330078125,
      "objective/train/value_loss": 0.00020639115246012807,
      "objective/train/value_max": -3.594160079956055e-05,
      "objective/train/value_min": -0.487060546875,
      "objective/train/value_reward_corr": 0.7141149475744621,
      "objective/train/value_std": 0.01519012451171875,
      "objective/train/weight_avg": 1.0008457899093628,
      "objective/train/weighted_lm_loss": 1.2584537267684937,
      "objective/train/weights_max": 1.2506414651870728,
      "objective/train/weights_min": 0.37683340907096863,
      "theoretical_loss": 3.3513392699959272,
      "tokens_seen": 2975334400
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.934199967902424e-05,
      "loss": 0.0632,
      "theoretical_loss": 3.3513392699959272,
      "tokens_seen": 2975334400
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.926175573744182e-05,
      "loss": 0.0646,
      "theoretical_loss": 3.3513167238473107,
      "tokens_seen": 2975596544
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.918151179585941e-05,
      "loss": 0.0622,
      "theoretical_loss": 3.3512941802409766,
      "tokens_seen": 2975858688
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.9101267854277e-05,
      "loss": 0.0657,
      "theoretical_loss": 3.3512716391764137,
      "tokens_seen": 2976120832
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.902102391269459e-05,
      "loss": 0.0634,
      "theoretical_loss": 3.351249100653112,
      "tokens_seen": 2976382976
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.894077997111219e-05,
      "loss": 0.0678,
      "theoretical_loss": 3.3512265646705615,
      "tokens_seen": 2976645120
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.886053602952976e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.3512040312282507,
      "tokens_seen": 2976907264
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.878029208794736e-05,
      "loss": 0.0645,
      "theoretical_loss": 3.351181500325671,
      "tokens_seen": 2977169408
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.870004814636495e-05,
      "loss": 0.0626,
      "theoretical_loss": 3.351158971962312,
      "tokens_seen": 2977431552
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.861980420478255e-05,
      "loss": 0.0685,
      "theoretical_loss": 3.351136446137664,
      "tokens_seen": 2977693696
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.853956026320013e-05,
      "loss": 0.0673,
      "theoretical_loss": 3.3511139228512175,
      "tokens_seen": 2977955840
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.845931632161773e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.3510914021024636,
      "tokens_seen": 2978217984
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.83790723800353e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.3510688838908917,
      "tokens_seen": 2978480128
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.82988284384529e-05,
      "loss": 0.0671,
      "theoretical_loss": 3.3510463682159943,
      "tokens_seen": 2978742272
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.821858449687049e-05,
      "loss": 0.0646,
      "theoretical_loss": 3.3510238550772615,
      "tokens_seen": 2979004416
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.813834055528807e-05,
      "loss": 0.0623,
      "theoretical_loss": 3.3510013444741844,
      "tokens_seen": 2979266560
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.805809661370567e-05,
      "loss": 0.0645,
      "theoretical_loss": 3.350978836406255,
      "tokens_seen": 2979528704
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.797785267212326e-05,
      "loss": 0.0652,
      "theoretical_loss": 3.3509563308729646,
      "tokens_seen": 2979790848
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.789760873054084e-05,
      "loss": 0.0629,
      "theoretical_loss": 3.350933827873804,
      "tokens_seen": 2980052992
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.781736478895843e-05,
      "loss": 0.0669,
      "theoretical_loss": 3.3509113274082662,
      "tokens_seen": 2980315136
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.773712084737603e-05,
      "loss": 0.0658,
      "theoretical_loss": 3.350888829475842,
      "tokens_seen": 2980577280
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.765687690579361e-05,
      "loss": 0.0624,
      "theoretical_loss": 3.3508663340760245,
      "tokens_seen": 2980839424
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.757663296421121e-05,
      "loss": 0.0677,
      "theoretical_loss": 3.3508438412083055,
      "tokens_seen": 2981101568
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.74963890226288e-05,
      "loss": 0.0646,
      "theoretical_loss": 3.350821350872177,
      "tokens_seen": 2981363712
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.741614508104638e-05,
      "loss": 0.0632,
      "theoretical_loss": 3.3507988630671317,
      "tokens_seen": 2981625856
    },
    {
      "epoch": 0.9,
      "objective/train/advantage_avg": 0.0002680000616237521,
      "objective/train/docs_used": 1084154,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2389239072799683,
      "objective/train/original_loss": 1.2389240264892578,
      "objective/train/theoretical_loss": 3.3507763777926627,
      "objective/train/tokens_used": 3002348000,
      "objective/train/value_avg": -0.0086822509765625,
      "objective/train/value_loss": 0.00023992547357920557,
      "objective/train/value_max": -2.2292137145996094e-05,
      "objective/train/value_min": -0.9775390625,
      "objective/train/value_reward_corr": 0.7576909276402002,
      "objective/train/value_std": 0.019012451171875,
      "objective/train/weight_avg": 1.0003770589828491,
      "objective/train/weighted_lm_loss": 1.2388797998428345,
      "objective/train/weights_max": 1.7507752180099487,
      "objective/train/weights_min": 0.36900806427001953,
      "theoretical_loss": 3.3507763777926627,
      "tokens_seen": 2981888000
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.733590113946397e-05,
      "loss": 0.0636,
      "theoretical_loss": 3.3507763777926627,
      "tokens_seen": 2981888000
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.725565719788155e-05,
      "loss": 0.0629,
      "theoretical_loss": 3.3507538950482623,
      "tokens_seen": 2982150144
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.717541325629915e-05,
      "loss": 0.0646,
      "theoretical_loss": 3.3507314148334233,
      "tokens_seen": 2982412288
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.709516931471674e-05,
      "loss": 0.0653,
      "theoretical_loss": 3.3507089371476395,
      "tokens_seen": 2982674432
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.701492537313434e-05,
      "loss": 0.0653,
      "theoretical_loss": 3.3506864619904038,
      "tokens_seen": 2982936576
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.693468143155191e-05,
      "loss": 0.0624,
      "theoretical_loss": 3.3506639893612093,
      "tokens_seen": 2983198720
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.685443748996951e-05,
      "loss": 0.0636,
      "theoretical_loss": 3.3506415192595496,
      "tokens_seen": 2983460864
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.67741935483871e-05,
      "loss": 0.0626,
      "theoretical_loss": 3.350619051684919,
      "tokens_seen": 2983723008
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.669394960680469e-05,
      "loss": 0.0632,
      "theoretical_loss": 3.3505965866368106,
      "tokens_seen": 2983985152
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.661370566522228e-05,
      "loss": 0.0614,
      "theoretical_loss": 3.3505741241147184,
      "tokens_seen": 2984247296
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.653346172363988e-05,
      "loss": 0.0625,
      "theoretical_loss": 3.350551664118137,
      "tokens_seen": 2984509440
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.645321778205746e-05,
      "loss": 0.0638,
      "theoretical_loss": 3.35052920664656,
      "tokens_seen": 2984771584
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.637297384047504e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.350506751699483,
      "tokens_seen": 2985033728
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.629272989889263e-05,
      "loss": 0.0621,
      "theoretical_loss": 3.3504842992763995,
      "tokens_seen": 2985295872
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.621248595731022e-05,
      "loss": 0.0646,
      "theoretical_loss": 3.350461849376804,
      "tokens_seen": 2985558016
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.613224201572782e-05,
      "loss": 0.0658,
      "theoretical_loss": 3.350439402000192,
      "tokens_seen": 2985820160
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.60519980741454e-05,
      "loss": 0.0631,
      "theoretical_loss": 3.3504169571460585,
      "tokens_seen": 2986082304
    },
    {
      "epoch": 0.9,
      "learning_rate": 9.5971754132563e-05,
      "loss": 0.0648,
      "theoretical_loss": 3.3503945148138983,
      "tokens_seen": 2986344448
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.589151019098058e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.350372075003207,
      "tokens_seen": 2986606592
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.581126624939817e-05,
      "loss": 0.0629,
      "theoretical_loss": 3.3503496377134794,
      "tokens_seen": 2986868736
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.573102230781576e-05,
      "loss": 0.0629,
      "theoretical_loss": 3.3503272029442117,
      "tokens_seen": 2987130880
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.565077836623336e-05,
      "loss": 0.0632,
      "theoretical_loss": 3.3503047706948994,
      "tokens_seen": 2987393024
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.557053442465094e-05,
      "loss": 0.0617,
      "theoretical_loss": 3.3502823409650384,
      "tokens_seen": 2987655168
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.549029048306854e-05,
      "loss": 0.0662,
      "theoretical_loss": 3.3502599137541247,
      "tokens_seen": 2987917312
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.541004654148612e-05,
      "loss": 0.0656,
      "theoretical_loss": 3.350237489061654,
      "tokens_seen": 2988179456
    },
    {
      "epoch": 0.91,
      "objective/train/advantage_avg": 0.00044840245391242206,
      "objective/train/docs_used": 1086412,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2609559297561646,
      "objective/train/original_loss": 1.260955810546875,
      "objective/train/theoretical_loss": 3.350215066887124,
      "objective/train/tokens_used": 3008901600,
      "objective/train/value_avg": -0.00728607177734375,
      "objective/train/value_loss": 0.00024669128470122814,
      "objective/train/value_max": -1.9669532775878906e-05,
      "objective/train/value_min": -0.939453125,
      "objective/train/value_reward_corr": 0.7119615799601327,
      "objective/train/value_std": 0.01641845703125,
      "objective/train/weight_avg": 1.0005605220794678,
      "objective/train/weighted_lm_loss": 1.261648178100586,
      "objective/train/weights_max": 2.1778111457824707,
      "objective/train/weights_min": 0.37020087242126465,
      "theoretical_loss": 3.350215066887124,
      "tokens_seen": 2988441600
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.53298025999037e-05,
      "loss": 0.0647,
      "theoretical_loss": 3.350215066887124,
      "tokens_seen": 2988441600
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.52495586583213e-05,
      "loss": 0.0631,
      "theoretical_loss": 3.3501926472300294,
      "tokens_seen": 2988703744
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.516931471673889e-05,
      "loss": 0.0633,
      "theoretical_loss": 3.3501702300898675,
      "tokens_seen": 2988965888
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.508907077515648e-05,
      "loss": 0.0638,
      "theoretical_loss": 3.3501478154661353,
      "tokens_seen": 2989228032
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.500882683357407e-05,
      "loss": 0.064,
      "theoretical_loss": 3.3501254033583296,
      "tokens_seen": 2989490176
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.492858289199166e-05,
      "loss": 0.0659,
      "theoretical_loss": 3.350102993765947,
      "tokens_seen": 2989752320
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.484833895040924e-05,
      "loss": 0.0665,
      "theoretical_loss": 3.3500805866884855,
      "tokens_seen": 2990014464
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.476809500882684e-05,
      "loss": 0.0599,
      "theoretical_loss": 3.350058182125441,
      "tokens_seen": 2990276608
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.468785106724443e-05,
      "loss": 0.0633,
      "theoretical_loss": 3.3500357800763125,
      "tokens_seen": 2990538752
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.460760712566202e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.3500133805405965,
      "tokens_seen": 2990800896
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.452736318407961e-05,
      "loss": 0.0636,
      "theoretical_loss": 3.3499909835177912,
      "tokens_seen": 2991063040
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.444711924249718e-05,
      "loss": 0.0665,
      "theoretical_loss": 3.3499685890073945,
      "tokens_seen": 2991325184
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.436687530091478e-05,
      "loss": 0.063,
      "theoretical_loss": 3.3499461970089044,
      "tokens_seen": 2991587328
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.428663135933237e-05,
      "loss": 0.0637,
      "theoretical_loss": 3.349923807521819,
      "tokens_seen": 2991849472
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.420638741774997e-05,
      "loss": 0.0663,
      "theoretical_loss": 3.3499014205456366,
      "tokens_seen": 2992111616
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.412614347616755e-05,
      "loss": 0.0633,
      "theoretical_loss": 3.349879036079856,
      "tokens_seen": 2992373760
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.404589953458515e-05,
      "loss": 0.0622,
      "theoretical_loss": 3.349856654123975,
      "tokens_seen": 2992635904
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.396565559300272e-05,
      "loss": 0.0622,
      "theoretical_loss": 3.349834274677493,
      "tokens_seen": 2992898048
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.388541165142032e-05,
      "loss": 0.0619,
      "theoretical_loss": 3.349811897739909,
      "tokens_seen": 2993160192
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.380516770983791e-05,
      "loss": 0.0615,
      "theoretical_loss": 3.3497895233107218,
      "tokens_seen": 2993422336
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.37249237682555e-05,
      "loss": 0.0623,
      "theoretical_loss": 3.3497671513894307,
      "tokens_seen": 2993684480
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.364467982667309e-05,
      "loss": 0.0653,
      "theoretical_loss": 3.3497447819755344,
      "tokens_seen": 2993946624
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.356443588509068e-05,
      "loss": 0.063,
      "theoretical_loss": 3.349722415068533,
      "tokens_seen": 2994208768
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.348419194350826e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.3497000506679266,
      "tokens_seen": 2994470912
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.340394800192585e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.349677688773214,
      "tokens_seen": 2994733056
    },
    {
      "epoch": 0.91,
      "objective/train/advantage_avg": 4.877928859059466e-06,
      "objective/train/docs_used": 1088743,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2250750064849854,
      "objective/train/original_loss": 1.2250750064849854,
      "objective/train/theoretical_loss": 3.3496553293838955,
      "objective/train/tokens_used": 3015455200,
      "objective/train/value_avg": -0.007190704345703125,
      "objective/train/value_loss": 0.0002711678680498153,
      "objective/train/value_max": -2.092123031616211e-05,
      "objective/train/value_min": -0.77490234375,
      "objective/train/value_reward_corr": 0.736253916918936,
      "objective/train/value_std": 0.01544189453125,
      "objective/train/weight_avg": 1.000122308731079,
      "objective/train/weighted_lm_loss": 1.2247551679611206,
      "objective/train/weights_max": 1.774009346961975,
      "objective/train/weights_min": 0.3681500554084778,
      "theoretical_loss": 3.3496553293838955,
      "tokens_seen": 2994995200
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.332370406034345e-05,
      "loss": 0.0622,
      "theoretical_loss": 3.3496553293838955,
      "tokens_seen": 2994995200
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.324346011876103e-05,
      "loss": 0.0645,
      "theoretical_loss": 3.349632972499471,
      "tokens_seen": 2995257344
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.316321617717863e-05,
      "loss": 0.0602,
      "theoretical_loss": 3.3496106181194407,
      "tokens_seen": 2995519488
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.308297223559622e-05,
      "loss": 0.0626,
      "theoretical_loss": 3.3495882662433054,
      "tokens_seen": 2995781632
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.30027282940138e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.349565916870565,
      "tokens_seen": 2996043776
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.292248435243139e-05,
      "loss": 0.0654,
      "theoretical_loss": 3.34954357000072,
      "tokens_seen": 2996305920
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.284224041084899e-05,
      "loss": 0.0636,
      "theoretical_loss": 3.3495212256332723,
      "tokens_seen": 2996568064
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.276199646926657e-05,
      "loss": 0.0631,
      "theoretical_loss": 3.3494988837677213,
      "tokens_seen": 2996830208
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.268175252768417e-05,
      "loss": 0.0637,
      "theoretical_loss": 3.349476544403569,
      "tokens_seen": 2997092352
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.260150858610176e-05,
      "loss": 0.0629,
      "theoretical_loss": 3.349454207540316,
      "tokens_seen": 2997354496
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.252126464451933e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.3494318731774646,
      "tokens_seen": 2997616640
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.244102070293693e-05,
      "loss": 0.0625,
      "theoretical_loss": 3.3494095413145155,
      "tokens_seen": 2997878784
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.236077676135451e-05,
      "loss": 0.0628,
      "theoretical_loss": 3.3493872119509702,
      "tokens_seen": 2998140928
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.228053281977211e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.349364885086331,
      "tokens_seen": 2998403072
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.22002888781897e-05,
      "loss": 0.0632,
      "theoretical_loss": 3.3493425607200993,
      "tokens_seen": 2998665216
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.21200449366073e-05,
      "loss": 0.0655,
      "theoretical_loss": 3.349320238851777,
      "tokens_seen": 2998927360
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.203980099502487e-05,
      "loss": 0.063,
      "theoretical_loss": 3.3492979194808674,
      "tokens_seen": 2999189504
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.195955705344247e-05,
      "loss": 0.0629,
      "theoretical_loss": 3.349275602606872,
      "tokens_seen": 2999451648
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.187931311186005e-05,
      "loss": 0.064,
      "theoretical_loss": 3.3492532882292934,
      "tokens_seen": 2999713792
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.179906917027765e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.3492309763476342,
      "tokens_seen": 2999975936
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.171882522869524e-05,
      "loss": 0.0639,
      "theoretical_loss": 3.349208666961397,
      "tokens_seen": 3000238080
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.163858128711282e-05,
      "loss": 0.0623,
      "theoretical_loss": 3.349186360070085,
      "tokens_seen": 3000500224
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.155833734553041e-05,
      "loss": 0.0628,
      "theoretical_loss": 3.349164055673201,
      "tokens_seen": 3000762368
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.1478093403948e-05,
      "loss": 0.0653,
      "theoretical_loss": 3.3491417537702484,
      "tokens_seen": 3001024512
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.13978494623656e-05,
      "loss": 0.0626,
      "theoretical_loss": 3.3491194543607308,
      "tokens_seen": 3001286656
    },
    {
      "epoch": 0.91,
      "objective/train/advantage_avg": 0.0006810373160988092,
      "objective/train/docs_used": 1091037,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2039985656738281,
      "objective/train/original_loss": 1.2039985656738281,
      "objective/train/theoretical_loss": 3.349097157444151,
      "objective/train/tokens_used": 3022008800,
      "objective/train/value_avg": -0.007076263427734375,
      "objective/train/value_loss": 0.00011562102008610964,
      "objective/train/value_max": -1.9252300262451172e-05,
      "objective/train/value_min": -0.258056640625,
      "objective/train/value_reward_corr": 0.7941427158484182,
      "objective/train/value_std": 0.0141754150390625,
      "objective/train/weight_avg": 1.000735878944397,
      "objective/train/weighted_lm_loss": 1.2046757936477661,
      "objective/train/weights_max": 1.2743442058563232,
      "objective/train/weights_min": 0.3687484562397003,
      "theoretical_loss": 3.349097157444151,
      "tokens_seen": 3001548800
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.131760552078318e-05,
      "loss": 0.0638,
      "theoretical_loss": 3.349097157444151,
      "tokens_seen": 3001548800
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.123736157920078e-05,
      "loss": 0.0656,
      "theoretical_loss": 3.349074863020013,
      "tokens_seen": 3001810944
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.115711763761836e-05,
      "loss": 0.063,
      "theoretical_loss": 3.3490525710878205,
      "tokens_seen": 3002073088
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.107687369603595e-05,
      "loss": 0.0639,
      "theoretical_loss": 3.3490302816470776,
      "tokens_seen": 3002335232
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.099662975445354e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.349007994697288,
      "tokens_seen": 3002597376
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.091638581287113e-05,
      "loss": 0.0632,
      "theoretical_loss": 3.348985710237956,
      "tokens_seen": 3002859520
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.083614187128872e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.348963428268586,
      "tokens_seen": 3003121664
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.075589792970632e-05,
      "loss": 0.0631,
      "theoretical_loss": 3.3489411487886827,
      "tokens_seen": 3003383808
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.06756539881239e-05,
      "loss": 0.0628,
      "theoretical_loss": 3.34891887179775,
      "tokens_seen": 3003645952
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.059541004654148e-05,
      "loss": 0.0633,
      "theoretical_loss": 3.348896597295293,
      "tokens_seen": 3003908096
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.051516610495908e-05,
      "loss": 0.062,
      "theoretical_loss": 3.3488743252808173,
      "tokens_seen": 3004170240
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.043492216337666e-05,
      "loss": 0.0648,
      "theoretical_loss": 3.3488520557538273,
      "tokens_seen": 3004432384
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.035467822179426e-05,
      "loss": 0.0638,
      "theoretical_loss": 3.348829788713828,
      "tokens_seen": 3004694528
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.027443428021185e-05,
      "loss": 0.0632,
      "theoretical_loss": 3.348807524160325,
      "tokens_seen": 3004956672
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.019419033862944e-05,
      "loss": 0.0626,
      "theoretical_loss": 3.3487852620928233,
      "tokens_seen": 3005218816
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.011394639704702e-05,
      "loss": 0.0616,
      "theoretical_loss": 3.3487630025108293,
      "tokens_seen": 3005480960
    },
    {
      "epoch": 0.91,
      "learning_rate": 9.003370245546462e-05,
      "loss": 0.0621,
      "theoretical_loss": 3.3487407454138483,
      "tokens_seen": 3005743104
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.99534585138822e-05,
      "loss": 0.0633,
      "theoretical_loss": 3.348718490801386,
      "tokens_seen": 3006005248
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.98732145722998e-05,
      "loss": 0.0655,
      "theoretical_loss": 3.348696238672949,
      "tokens_seen": 3006267392
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.979297063071739e-05,
      "loss": 0.0629,
      "theoretical_loss": 3.3486739890280433,
      "tokens_seen": 3006529536
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.971272668913497e-05,
      "loss": 0.0662,
      "theoretical_loss": 3.3486517418661745,
      "tokens_seen": 3006791680
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.963248274755256e-05,
      "loss": 0.0653,
      "theoretical_loss": 3.3486294971868498,
      "tokens_seen": 3007053824
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.955223880597014e-05,
      "loss": 0.0633,
      "theoretical_loss": 3.3486072549895756,
      "tokens_seen": 3007315968
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.947199486438774e-05,
      "loss": 0.0657,
      "theoretical_loss": 3.3485850152738585,
      "tokens_seen": 3007578112
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.939175092280533e-05,
      "loss": 0.0624,
      "theoretical_loss": 3.3485627780392058,
      "tokens_seen": 3007840256
    },
    {
      "epoch": 0.91,
      "objective/train/advantage_avg": 0.00021146050130482763,
      "objective/train/docs_used": 1093533,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.1856549978256226,
      "objective/train/original_loss": 1.185654878616333,
      "objective/train/theoretical_loss": 3.348540543285124,
      "objective/train/tokens_used": 3028562400,
      "objective/train/value_avg": -0.00530242919921875,
      "objective/train/value_loss": 0.0002320571948075667,
      "objective/train/value_max": -1.1205673217773438e-05,
      "objective/train/value_min": -0.82958984375,
      "objective/train/value_reward_corr": 0.5928466282023085,
      "objective/train/value_std": 0.01294708251953125,
      "objective/train/weight_avg": 1.0003186464309692,
      "objective/train/weighted_lm_loss": 1.185418725013733,
      "objective/train/weights_max": 2.2635436058044434,
      "objective/train/weights_min": 0.36832353472709656,
      "theoretical_loss": 3.348540543285124,
      "tokens_seen": 3008102400
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.931150698122293e-05,
      "loss": 0.0614,
      "theoretical_loss": 3.348540543285124,
      "tokens_seen": 3008102400
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.923126303964051e-05,
      "loss": 0.0632,
      "theoretical_loss": 3.3485183110111203,
      "tokens_seen": 3008364544
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.91510190980581e-05,
      "loss": 0.0618,
      "theoretical_loss": 3.348496081216702,
      "tokens_seen": 3008626688
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.907077515647568e-05,
      "loss": 0.0663,
      "theoretical_loss": 3.3484738539013774,
      "tokens_seen": 3008888832
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.899053121489328e-05,
      "loss": 0.0638,
      "theoretical_loss": 3.3484516290646527,
      "tokens_seen": 3009150976
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.891028727331087e-05,
      "loss": 0.0662,
      "theoretical_loss": 3.3484294067060367,
      "tokens_seen": 3009413120
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.883004333172845e-05,
      "loss": 0.0627,
      "theoretical_loss": 3.3484071868250367,
      "tokens_seen": 3009675264
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.874979939014605e-05,
      "loss": 0.0647,
      "theoretical_loss": 3.348384969421161,
      "tokens_seen": 3009937408
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.866955544856362e-05,
      "loss": 0.0622,
      "theoretical_loss": 3.3483627544939174,
      "tokens_seen": 3010199552
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.858931150698122e-05,
      "loss": 0.0654,
      "theoretical_loss": 3.3483405420428145,
      "tokens_seen": 3010461696
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.850906756539881e-05,
      "loss": 0.0628,
      "theoretical_loss": 3.34831833206736,
      "tokens_seen": 3010723840
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.842882362381641e-05,
      "loss": 0.063,
      "theoretical_loss": 3.3482961245670637,
      "tokens_seen": 3010985984
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.834857968223399e-05,
      "loss": 0.0628,
      "theoretical_loss": 3.348273919541434,
      "tokens_seen": 3011248128
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.826833574065159e-05,
      "loss": 0.066,
      "theoretical_loss": 3.348251716989979,
      "tokens_seen": 3011510272
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.818809179906916e-05,
      "loss": 0.064,
      "theoretical_loss": 3.3482295169122076,
      "tokens_seen": 3011772416
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.810784785748676e-05,
      "loss": 0.0675,
      "theoretical_loss": 3.3482073193076296,
      "tokens_seen": 3012034560
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.802760391590435e-05,
      "loss": 0.062,
      "theoretical_loss": 3.3481851241757545,
      "tokens_seen": 3012296704
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.794735997432195e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.3481629315160912,
      "tokens_seen": 3012558848
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.786711603273953e-05,
      "loss": 0.0645,
      "theoretical_loss": 3.348140741328149,
      "tokens_seen": 3012820992
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.778687209115712e-05,
      "loss": 0.0652,
      "theoretical_loss": 3.348118553611438,
      "tokens_seen": 3013083136
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.770662814957472e-05,
      "loss": 0.0664,
      "theoretical_loss": 3.348096368365468,
      "tokens_seen": 3013345280
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.762638420799229e-05,
      "loss": 0.0641,
      "theoretical_loss": 3.348074185589749,
      "tokens_seen": 3013607424
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.754614026640989e-05,
      "loss": 0.0644,
      "theoretical_loss": 3.3480520052837908,
      "tokens_seen": 3013869568
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.746589632482747e-05,
      "loss": 0.0629,
      "theoretical_loss": 3.348029827447104,
      "tokens_seen": 3014131712
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.738565238324507e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.3480076520791986,
      "tokens_seen": 3014393856
    },
    {
      "epoch": 0.91,
      "objective/train/advantage_avg": 6.449154170695692e-05,
      "objective/train/docs_used": 1095882,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3385285139083862,
      "objective/train/original_loss": 1.3385286331176758,
      "objective/train/theoretical_loss": 3.3479854791795853,
      "objective/train/tokens_used": 3035116000,
      "objective/train/value_avg": -0.0070648193359375,
      "objective/train/value_loss": 0.00022063912183512002,
      "objective/train/value_max": -1.8477439880371094e-05,
      "objective/train/value_min": -0.66064453125,
      "objective/train/value_reward_corr": 0.7004792297942277,
      "objective/train/value_std": 0.013946533203125,
      "objective/train/weight_avg": 1.0001661777496338,
      "objective/train/weighted_lm_loss": 1.3388984203338623,
      "objective/train/weights_max": 1.2268719673156738,
      "objective/train/weights_min": 0.3715803325176239,
      "theoretical_loss": 3.3479854791795853,
      "tokens_seen": 3014656000
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.730540844166266e-05,
      "loss": 0.0638,
      "theoretical_loss": 3.3479854791795853,
      "tokens_seen": 3014656000
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.722516450008026e-05,
      "loss": 0.0634,
      "theoretical_loss": 3.347963308747775,
      "tokens_seen": 3014918144
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.714492055849783e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.347941140783278,
      "tokens_seen": 3015180288
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.706467661691543e-05,
      "loss": 0.0647,
      "theoretical_loss": 3.347918975285606,
      "tokens_seen": 3015442432
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.698443267533301e-05,
      "loss": 0.0625,
      "theoretical_loss": 3.347896812254269,
      "tokens_seen": 3015704576
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.69041887337506e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.3478746516887794,
      "tokens_seen": 3015966720
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.68239447921682e-05,
      "loss": 0.0608,
      "theoretical_loss": 3.347852493588648,
      "tokens_seen": 3016228864
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.674370085058578e-05,
      "loss": 0.0628,
      "theoretical_loss": 3.347830337953386,
      "tokens_seen": 3016491008
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.666345690900337e-05,
      "loss": 0.0645,
      "theoretical_loss": 3.3478081847825054,
      "tokens_seen": 3016753152
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.658321296742095e-05,
      "loss": 0.066,
      "theoretical_loss": 3.3477860340755177,
      "tokens_seen": 3017015296
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.650296902583855e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.3477638858319354,
      "tokens_seen": 3017277440
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.642272508425614e-05,
      "loss": 0.0646,
      "theoretical_loss": 3.34774174005127,
      "tokens_seen": 3017539584
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.634248114267374e-05,
      "loss": 0.0628,
      "theoretical_loss": 3.347719596733034,
      "tokens_seen": 3017801728
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.626223720109132e-05,
      "loss": 0.0613,
      "theoretical_loss": 3.3476974558767396,
      "tokens_seen": 3018063872
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.618199325950891e-05,
      "loss": 0.0615,
      "theoretical_loss": 3.347675317481899,
      "tokens_seen": 3018326016
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.61017493179265e-05,
      "loss": 0.0667,
      "theoretical_loss": 3.3476531815480253,
      "tokens_seen": 3018588160
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.60215053763441e-05,
      "loss": 0.0659,
      "theoretical_loss": 3.347631048074631,
      "tokens_seen": 3018850304
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.594126143476168e-05,
      "loss": 0.0634,
      "theoretical_loss": 3.347608917061229,
      "tokens_seen": 3019112448
    },
    {
      "epoch": 0.91,
      "learning_rate": 8.586101749317927e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.347586788507332,
      "tokens_seen": 3019374592
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.578077355159686e-05,
      "loss": 0.0614,
      "theoretical_loss": 3.347564662412454,
      "tokens_seen": 3019636736
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.570052961001444e-05,
      "loss": 0.0623,
      "theoretical_loss": 3.347542538776108,
      "tokens_seen": 3019898880
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.562028566843204e-05,
      "loss": 0.0651,
      "theoretical_loss": 3.347520417597807,
      "tokens_seen": 3020161024
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.554004172684962e-05,
      "loss": 0.064,
      "theoretical_loss": 3.3474982988770647,
      "tokens_seen": 3020423168
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.545979778526722e-05,
      "loss": 0.0625,
      "theoretical_loss": 3.347476182613395,
      "tokens_seen": 3020685312
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.53795538436848e-05,
      "loss": 0.0632,
      "theoretical_loss": 3.3474540688063117,
      "tokens_seen": 3020947456
    },
    {
      "epoch": 0.92,
      "objective/train/advantage_avg": 0.0008990961359813809,
      "objective/train/docs_used": 1098075,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.220808982849121,
      "objective/train/original_loss": 1.220808982849121,
      "objective/train/theoretical_loss": 3.347431957455329,
      "objective/train/tokens_used": 3041669600,
      "objective/train/value_avg": -0.00785064697265625,
      "objective/train/value_loss": 0.0002740893396548927,
      "objective/train/value_max": -2.3365020751953125e-05,
      "objective/train/value_min": -0.482177734375,
      "objective/train/value_reward_corr": 0.6300108547976242,
      "objective/train/value_std": 0.0153656005859375,
      "objective/train/weight_avg": 1.0010228157043457,
      "objective/train/weighted_lm_loss": 1.222151517868042,
      "objective/train/weights_max": 1.6195976734161377,
      "objective/train/weights_min": 0.3715973496437073,
      "theoretical_loss": 3.347431957455329,
      "tokens_seen": 3021209600
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.52993099021024e-05,
      "loss": 0.0622,
      "theoretical_loss": 3.347431957455329,
      "tokens_seen": 3021209600
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.521906596051998e-05,
      "loss": 0.0637,
      "theoretical_loss": 3.347409848559961,
      "tokens_seen": 3021471744
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.513882201893758e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.347387742119721,
      "tokens_seen": 3021733888
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.505857807735516e-05,
      "loss": 0.065,
      "theoretical_loss": 3.347365638134125,
      "tokens_seen": 3021996032
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.497833413577275e-05,
      "loss": 0.063,
      "theoretical_loss": 3.3473435366026867,
      "tokens_seen": 3022258176
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.489809019419035e-05,
      "loss": 0.065,
      "theoretical_loss": 3.3473214375249207,
      "tokens_seen": 3022520320
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.481784625260793e-05,
      "loss": 0.0613,
      "theoretical_loss": 3.347299340900342,
      "tokens_seen": 3022782464
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.473760231102552e-05,
      "loss": 0.0614,
      "theoretical_loss": 3.3472772467284657,
      "tokens_seen": 3023044608
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.46573583694431e-05,
      "loss": 0.0639,
      "theoretical_loss": 3.3472551550088063,
      "tokens_seen": 3023306752
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.45771144278607e-05,
      "loss": 0.0658,
      "theoretical_loss": 3.34723306574088,
      "tokens_seen": 3023568896
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.449687048627829e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.3472109789242017,
      "tokens_seen": 3023831040
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.441662654469589e-05,
      "loss": 0.065,
      "theoretical_loss": 3.347188894558287,
      "tokens_seen": 3024093184
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.433638260311347e-05,
      "loss": 0.0647,
      "theoretical_loss": 3.347166812642651,
      "tokens_seen": 3024355328
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.425613866153106e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.34714473317681,
      "tokens_seen": 3024617472
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.417589471994864e-05,
      "loss": 0.0633,
      "theoretical_loss": 3.34712265616028,
      "tokens_seen": 3024879616
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.409565077836623e-05,
      "loss": 0.0628,
      "theoretical_loss": 3.347100581592577,
      "tokens_seen": 3025141760
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.401540683678383e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.347078509473217,
      "tokens_seen": 3025403904
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.393516289520141e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.3470564398017166,
      "tokens_seen": 3025666048
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.385491895361901e-05,
      "loss": 0.0665,
      "theoretical_loss": 3.3470343725775917,
      "tokens_seen": 3025928192
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.377467501203658e-05,
      "loss": 0.0658,
      "theoretical_loss": 3.34701230780036,
      "tokens_seen": 3026190336
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.369443107045418e-05,
      "loss": 0.0622,
      "theoretical_loss": 3.3469902454695375,
      "tokens_seen": 3026452480
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.361418712887177e-05,
      "loss": 0.0656,
      "theoretical_loss": 3.346968185584641,
      "tokens_seen": 3026714624
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.353394318728937e-05,
      "loss": 0.0645,
      "theoretical_loss": 3.3469461281451878,
      "tokens_seen": 3026976768
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.345369924570695e-05,
      "loss": 0.0662,
      "theoretical_loss": 3.346924073150695,
      "tokens_seen": 3027238912
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.337345530412455e-05,
      "loss": 0.0648,
      "theoretical_loss": 3.34690202060068,
      "tokens_seen": 3027501056
    },
    {
      "epoch": 0.92,
      "objective/train/advantage_avg": -8.04458322818391e-05,
      "objective/train/docs_used": 1100401,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.253341555595398,
      "objective/train/original_loss": 1.2533414363861084,
      "objective/train/theoretical_loss": 3.34687997049466,
      "objective/train/tokens_used": 3048223200,
      "objective/train/value_avg": -0.0073089599609375,
      "objective/train/value_loss": 0.0001552473404444754,
      "objective/train/value_max": -3.4809112548828125e-05,
      "objective/train/value_min": -0.321533203125,
      "objective/train/value_reward_corr": 0.7404640557884099,
      "objective/train/value_std": 0.0133514404296875,
      "objective/train/weight_avg": 0.9999929666519165,
      "objective/train/weighted_lm_loss": 1.2528541088104248,
      "objective/train/weights_max": 1.2134668827056885,
      "objective/train/weights_min": 0.36880823969841003,
      "theoretical_loss": 3.34687997049466,
      "tokens_seen": 3027763200
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.329321136254212e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.34687997049466,
      "tokens_seen": 3027763200
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.321296742095972e-05,
      "loss": 0.061,
      "theoretical_loss": 3.346857922832153,
      "tokens_seen": 3028025344
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.313272347937731e-05,
      "loss": 0.0648,
      "theoretical_loss": 3.346835877612676,
      "tokens_seen": 3028287488
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.30524795377949e-05,
      "loss": 0.0656,
      "theoretical_loss": 3.3468138348357472,
      "tokens_seen": 3028549632
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.297223559621249e-05,
      "loss": 0.0631,
      "theoretical_loss": 3.346791794500885,
      "tokens_seen": 3028811776
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.289199165463008e-05,
      "loss": 0.0648,
      "theoretical_loss": 3.3467697566076073,
      "tokens_seen": 3029073920
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.281174771304766e-05,
      "loss": 0.0671,
      "theoretical_loss": 3.346747721155432,
      "tokens_seen": 3029336064
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.273150377146525e-05,
      "loss": 0.064,
      "theoretical_loss": 3.346725688143878,
      "tokens_seen": 3029598208
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.265125982988285e-05,
      "loss": 0.0632,
      "theoretical_loss": 3.3467036575724634,
      "tokens_seen": 3029860352
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.257101588830043e-05,
      "loss": 0.0636,
      "theoretical_loss": 3.346681629440707,
      "tokens_seen": 3030122496
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.249077194671803e-05,
      "loss": 0.0633,
      "theoretical_loss": 3.3466596037481278,
      "tokens_seen": 3030384640
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.241052800513562e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.346637580494244,
      "tokens_seen": 3030646784
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.23302840635532e-05,
      "loss": 0.0659,
      "theoretical_loss": 3.3466155596785763,
      "tokens_seen": 3030908928
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.225004012197079e-05,
      "loss": 0.0655,
      "theoretical_loss": 3.3465935413006425,
      "tokens_seen": 3031171072
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.216979618038837e-05,
      "loss": 0.0651,
      "theoretical_loss": 3.3465715253599617,
      "tokens_seen": 3031433216
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.208955223880597e-05,
      "loss": 0.064,
      "theoretical_loss": 3.346549511856055,
      "tokens_seen": 3031695360
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.200930829722356e-05,
      "loss": 0.0645,
      "theoretical_loss": 3.3465275007884405,
      "tokens_seen": 3031957504
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.192906435564116e-05,
      "loss": 0.0646,
      "theoretical_loss": 3.3465054921566386,
      "tokens_seen": 3032219648
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.184882041405873e-05,
      "loss": 0.0637,
      "theoretical_loss": 3.346483485960169,
      "tokens_seen": 3032481792
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.176857647247633e-05,
      "loss": 0.0658,
      "theoretical_loss": 3.3464614821985523,
      "tokens_seen": 3032743936
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.168833253089391e-05,
      "loss": 0.0663,
      "theoretical_loss": 3.3464394808713083,
      "tokens_seen": 3033006080
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.160808858931151e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.3464174819779564,
      "tokens_seen": 3033268224
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.15278446477291e-05,
      "loss": 0.0653,
      "theoretical_loss": 3.3463954855180185,
      "tokens_seen": 3033530368
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.14476007061467e-05,
      "loss": 0.0641,
      "theoretical_loss": 3.346373491491015,
      "tokens_seen": 3033792512
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.136735676456427e-05,
      "loss": 0.0671,
      "theoretical_loss": 3.346351499896465,
      "tokens_seen": 3034054656
    },
    {
      "epoch": 0.92,
      "objective/train/advantage_avg": 0.0005928638274781406,
      "objective/train/docs_used": 1102938,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2831909656524658,
      "objective/train/original_loss": 1.2831907272338867,
      "objective/train/theoretical_loss": 3.3463295107338915,
      "objective/train/tokens_used": 3054776800,
      "objective/train/value_avg": -0.00801849365234375,
      "objective/train/value_loss": 0.0004268684715498239,
      "objective/train/value_max": -2.3186206817626953e-05,
      "objective/train/value_min": -0.8818359375,
      "objective/train/value_reward_corr": 0.6643995542878625,
      "objective/train/value_std": 0.0186004638671875,
      "objective/train/weight_avg": 1.0007858276367188,
      "objective/train/weighted_lm_loss": 1.2838879823684692,
      "objective/train/weights_max": 2.341017961502075,
      "objective/train/weights_min": 0.3796498477458954,
      "theoretical_loss": 3.3463295107338915,
      "tokens_seen": 3034316800
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.128711282298187e-05,
      "loss": 0.0655,
      "theoretical_loss": 3.3463295107338915,
      "tokens_seen": 3034316800
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.120686888139945e-05,
      "loss": 0.065,
      "theoretical_loss": 3.346307524002814,
      "tokens_seen": 3034578944
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.112662493981704e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.346285539702754,
      "tokens_seen": 3034841088
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.104638099823464e-05,
      "loss": 0.0613,
      "theoretical_loss": 3.3462635578332334,
      "tokens_seen": 3035103232
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.096613705665222e-05,
      "loss": 0.0644,
      "theoretical_loss": 3.3462415783937725,
      "tokens_seen": 3035365376
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.088589311506981e-05,
      "loss": 0.0654,
      "theoretical_loss": 3.3462196013838934,
      "tokens_seen": 3035627520
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.08056491734874e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.3461976268031184,
      "tokens_seen": 3035889664
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.0725405231905e-05,
      "loss": 0.062,
      "theoretical_loss": 3.3461756546509682,
      "tokens_seen": 3036151808
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.064516129032258e-05,
      "loss": 0.0639,
      "theoretical_loss": 3.346153684926965,
      "tokens_seen": 3036413952
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.056491734874018e-05,
      "loss": 0.0638,
      "theoretical_loss": 3.3461317176306316,
      "tokens_seen": 3036676096
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.048467340715777e-05,
      "loss": 0.0662,
      "theoretical_loss": 3.3461097527614894,
      "tokens_seen": 3036938240
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.040442946557535e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.3460877903190607,
      "tokens_seen": 3037200384
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.032418552399294e-05,
      "loss": 0.0627,
      "theoretical_loss": 3.3460658303028685,
      "tokens_seen": 3037462528
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.024394158241052e-05,
      "loss": 0.064,
      "theoretical_loss": 3.3460438727124355,
      "tokens_seen": 3037724672
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.016369764082812e-05,
      "loss": 0.0656,
      "theoretical_loss": 3.346021917547284,
      "tokens_seen": 3037986816
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.00834536992457e-05,
      "loss": 0.0641,
      "theoretical_loss": 3.345999964806937,
      "tokens_seen": 3038248960
    },
    {
      "epoch": 0.92,
      "learning_rate": 8.00032097576633e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.3459780144909175,
      "tokens_seen": 3038511104
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.992296581608088e-05,
      "loss": 0.0605,
      "theoretical_loss": 3.3459560665987484,
      "tokens_seen": 3038773248
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.984272187449848e-05,
      "loss": 0.0623,
      "theoretical_loss": 3.3459341211299534,
      "tokens_seen": 3039035392
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.976247793291606e-05,
      "loss": 0.0634,
      "theoretical_loss": 3.345912178084056,
      "tokens_seen": 3039297536
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.968223399133366e-05,
      "loss": 0.0653,
      "theoretical_loss": 3.3458902374605795,
      "tokens_seen": 3039559680
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.960199004975125e-05,
      "loss": 0.0647,
      "theoretical_loss": 3.3458682992590476,
      "tokens_seen": 3039821824
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.952174610816885e-05,
      "loss": 0.0634,
      "theoretical_loss": 3.345846363478984,
      "tokens_seen": 3040083968
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.944150216658642e-05,
      "loss": 0.063,
      "theoretical_loss": 3.345824430119913,
      "tokens_seen": 3040346112
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.9361258225004e-05,
      "loss": 0.0652,
      "theoretical_loss": 3.3458024991813584,
      "tokens_seen": 3040608256
    },
    {
      "epoch": 0.92,
      "objective/train/advantage_avg": 0.0005279024480842054,
      "objective/train/docs_used": 1105174,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3808923959732056,
      "objective/train/original_loss": 1.3808923959732056,
      "objective/train/theoretical_loss": 3.3457805706628445,
      "objective/train/tokens_used": 3061330400,
      "objective/train/value_avg": -0.00731658935546875,
      "objective/train/value_loss": 0.00018059344438370317,
      "objective/train/value_max": -2.753734588623047e-05,
      "objective/train/value_min": -0.9140625,
      "objective/train/value_reward_corr": 0.7750596911751518,
      "objective/train/value_std": 0.016510009765625,
      "objective/train/weight_avg": 1.0006126165390015,
      "objective/train/weighted_lm_loss": 1.3814222812652588,
      "objective/train/weights_max": 1.484392523765564,
      "objective/train/weights_min": 0.4039379060268402,
      "theoretical_loss": 3.3457805706628445,
      "tokens_seen": 3040870400
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.92810142834216e-05,
      "loss": 0.0651,
      "theoretical_loss": 3.3457805706628445,
      "tokens_seen": 3040870400
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.920077034183919e-05,
      "loss": 0.0632,
      "theoretical_loss": 3.345758644563896,
      "tokens_seen": 3041132544
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.912052640025679e-05,
      "loss": 0.0613,
      "theoretical_loss": 3.345736720884037,
      "tokens_seen": 3041394688
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.904028245867437e-05,
      "loss": 0.0617,
      "theoretical_loss": 3.345714799622792,
      "tokens_seen": 3041656832
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.896003851709197e-05,
      "loss": 0.064,
      "theoretical_loss": 3.345692880779686,
      "tokens_seen": 3041918976
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.887979457550954e-05,
      "loss": 0.064,
      "theoretical_loss": 3.345670964354244,
      "tokens_seen": 3042181120
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.879955063392714e-05,
      "loss": 0.063,
      "theoretical_loss": 3.3456490503459912,
      "tokens_seen": 3042443264
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.871930669234473e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.3456271387544523,
      "tokens_seen": 3042705408
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.863906275076233e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.345605229579153,
      "tokens_seen": 3042967552
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.855881880917991e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.345583322819618,
      "tokens_seen": 3043229696
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.847857486759751e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.345561418475374,
      "tokens_seen": 3043491840
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.839833092601508e-05,
      "loss": 0.0609,
      "theoretical_loss": 3.3455395165459456,
      "tokens_seen": 3043753984
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.831808698443267e-05,
      "loss": 0.0625,
      "theoretical_loss": 3.3455176170308594,
      "tokens_seen": 3044016128
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.823784304285027e-05,
      "loss": 0.0638,
      "theoretical_loss": 3.3454957199296413,
      "tokens_seen": 3044278272
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.815759910126785e-05,
      "loss": 0.0651,
      "theoretical_loss": 3.3454738252418172,
      "tokens_seen": 3044540416
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.807735515968545e-05,
      "loss": 0.0653,
      "theoretical_loss": 3.3454519329669132,
      "tokens_seen": 3044802560
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.799711121810304e-05,
      "loss": 0.0646,
      "theoretical_loss": 3.345430043104456,
      "tokens_seen": 3045064704
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.791686727652062e-05,
      "loss": 0.0645,
      "theoretical_loss": 3.345408155653972,
      "tokens_seen": 3045326848
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.783662333493821e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.3453862706149877,
      "tokens_seen": 3045588992
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.775637939335581e-05,
      "loss": 0.0654,
      "theoretical_loss": 3.34536438798703,
      "tokens_seen": 3045851136
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.76761354517734e-05,
      "loss": 0.0625,
      "theoretical_loss": 3.345342507769626,
      "tokens_seen": 3046113280
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.759589151019099e-05,
      "loss": 0.0636,
      "theoretical_loss": 3.345320629962302,
      "tokens_seen": 3046375424
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.751564756860858e-05,
      "loss": 0.0617,
      "theoretical_loss": 3.345298754564586,
      "tokens_seen": 3046637568
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.743540362702615e-05,
      "loss": 0.063,
      "theoretical_loss": 3.3452768815760052,
      "tokens_seen": 3046899712
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.735515968544375e-05,
      "loss": 0.0646,
      "theoretical_loss": 3.3452550109960866,
      "tokens_seen": 3047161856
    },
    {
      "epoch": 0.92,
      "objective/train/advantage_avg": 0.00029660866130143404,
      "objective/train/docs_used": 1107536,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.388244390487671,
      "objective/train/original_loss": 1.3882445096969604,
      "objective/train/theoretical_loss": 3.3452331428243585,
      "objective/train/tokens_used": 3067884000,
      "objective/train/value_avg": -0.007091522216796875,
      "objective/train/value_loss": 0.00014839263167232275,
      "objective/train/value_max": -1.9550323486328125e-05,
      "objective/train/value_min": -0.384521484375,
      "objective/train/value_reward_corr": 0.7271648855293683,
      "objective/train/value_std": 0.01367950439453125,
      "objective/train/weight_avg": 1.0003644227981567,
      "objective/train/weighted_lm_loss": 1.3882359266281128,
      "objective/train/weights_max": 1.3238147497177124,
      "objective/train/weights_min": 0.3687090575695038,
      "theoretical_loss": 3.3452331428243585,
      "tokens_seen": 3047424000
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.727491574386133e-05,
      "loss": 0.0663,
      "theoretical_loss": 3.3452331428243585,
      "tokens_seen": 3047424000
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.719467180227893e-05,
      "loss": 0.0628,
      "theoretical_loss": 3.345211277060348,
      "tokens_seen": 3047686144
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.711442786069652e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.345189413703583,
      "tokens_seen": 3047948288
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.703418391911412e-05,
      "loss": 0.0653,
      "theoretical_loss": 3.3451675527535913,
      "tokens_seen": 3048210432
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.695393997753169e-05,
      "loss": 0.0641,
      "theoretical_loss": 3.3451456942099016,
      "tokens_seen": 3048472576
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.687369603594929e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.3451238380720416,
      "tokens_seen": 3048734720
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.679345209436687e-05,
      "loss": 0.063,
      "theoretical_loss": 3.34510198433954,
      "tokens_seen": 3048996864
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.671320815278447e-05,
      "loss": 0.0633,
      "theoretical_loss": 3.3450801330119253,
      "tokens_seen": 3049259008
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.663296421120206e-05,
      "loss": 0.0621,
      "theoretical_loss": 3.3450582840887257,
      "tokens_seen": 3049521152
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.655272026961964e-05,
      "loss": 0.0647,
      "theoretical_loss": 3.3450364375694708,
      "tokens_seen": 3049783296
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.647247632803723e-05,
      "loss": 0.0626,
      "theoretical_loss": 3.345014593453689,
      "tokens_seen": 3050045440
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.639223238645482e-05,
      "loss": 0.0657,
      "theoretical_loss": 3.344992751740909,
      "tokens_seen": 3050307584
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.631198844487241e-05,
      "loss": 0.0631,
      "theoretical_loss": 3.344970912430661,
      "tokens_seen": 3050569728
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.623174450329e-05,
      "loss": 0.0647,
      "theoretical_loss": 3.344949075522473,
      "tokens_seen": 3050831872
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.61515005617076e-05,
      "loss": 0.0633,
      "theoretical_loss": 3.3449272410158755,
      "tokens_seen": 3051094016
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.607125662012518e-05,
      "loss": 0.0671,
      "theoretical_loss": 3.3449054089103973,
      "tokens_seen": 3051356160
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.599101267854277e-05,
      "loss": 0.0624,
      "theoretical_loss": 3.3448835792055687,
      "tokens_seen": 3051618304
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.591076873696036e-05,
      "loss": 0.0626,
      "theoretical_loss": 3.344861751900919,
      "tokens_seen": 3051880448
    },
    {
      "epoch": 0.92,
      "learning_rate": 7.583052479537795e-05,
      "loss": 0.0651,
      "theoretical_loss": 3.344839926995979,
      "tokens_seen": 3052142592
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.575028085379554e-05,
      "loss": 0.0638,
      "theoretical_loss": 3.3448181044902783,
      "tokens_seen": 3052404736
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.567003691221314e-05,
      "loss": 0.0663,
      "theoretical_loss": 3.3447962843833468,
      "tokens_seen": 3052666880
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.558979297063073e-05,
      "loss": 0.0633,
      "theoretical_loss": 3.344774466674715,
      "tokens_seen": 3052929024
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.55095490290483e-05,
      "loss": 0.0652,
      "theoretical_loss": 3.3447526513639136,
      "tokens_seen": 3053191168
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.54293050874659e-05,
      "loss": 0.0636,
      "theoretical_loss": 3.3447308384504737,
      "tokens_seen": 3053453312
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.534906114588348e-05,
      "loss": 0.0634,
      "theoretical_loss": 3.3447090279339253,
      "tokens_seen": 3053715456
    },
    {
      "epoch": 0.93,
      "objective/train/advantage_avg": 0.0005677227163687348,
      "objective/train/docs_used": 1109702,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2256311178207397,
      "objective/train/original_loss": 1.2256312370300293,
      "objective/train/theoretical_loss": 3.344687219813799,
      "objective/train/tokens_used": 3074437600,
      "objective/train/value_avg": -0.0052642822265625,
      "objective/train/value_loss": 0.00013756824773736298,
      "objective/train/value_max": -3.2961368560791016e-05,
      "objective/train/value_min": -0.323486328125,
      "objective/train/value_reward_corr": 0.6535918841042918,
      "objective/train/value_std": 0.01059722900390625,
      "objective/train/weight_avg": 1.000628113746643,
      "objective/train/weighted_lm_loss": 1.226385474205017,
      "objective/train/weights_max": 1.242395043373108,
      "objective/train/weights_min": 0.3693855106830597,
      "theoretical_loss": 3.344687219813799,
      "tokens_seen": 3053977600
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.526881720430108e-05,
      "loss": 0.0632,
      "theoretical_loss": 3.344687219813799,
      "tokens_seen": 3053977600
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.518857326271867e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.344665414089627,
      "tokens_seen": 3054239744
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.510832932113627e-05,
      "loss": 0.0641,
      "theoretical_loss": 3.3446436107609396,
      "tokens_seen": 3054501888
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.502808537955384e-05,
      "loss": 0.0638,
      "theoretical_loss": 3.3446218098272684,
      "tokens_seen": 3054764032
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.494784143797144e-05,
      "loss": 0.0654,
      "theoretical_loss": 3.344600011288145,
      "tokens_seen": 3055026176
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.486759749638902e-05,
      "loss": 0.0669,
      "theoretical_loss": 3.3445782151431005,
      "tokens_seen": 3055288320
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.478735355480662e-05,
      "loss": 0.0636,
      "theoretical_loss": 3.3445564213916668,
      "tokens_seen": 3055550464
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.47071096132242e-05,
      "loss": 0.0644,
      "theoretical_loss": 3.3445346300333756,
      "tokens_seen": 3055812608
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.462686567164179e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.344512841067759,
      "tokens_seen": 3056074752
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.454662173005938e-05,
      "loss": 0.0615,
      "theoretical_loss": 3.344491054494349,
      "tokens_seen": 3056336896
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.446637778847696e-05,
      "loss": 0.064,
      "theoretical_loss": 3.3444692703126777,
      "tokens_seen": 3056599040
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.438613384689456e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.3444474885222784,
      "tokens_seen": 3056861184
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.430588990531215e-05,
      "loss": 0.0626,
      "theoretical_loss": 3.3444257091226817,
      "tokens_seen": 3057123328
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.422564596372975e-05,
      "loss": 0.0647,
      "theoretical_loss": 3.344403932113422,
      "tokens_seen": 3057385472
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.414540202214733e-05,
      "loss": 0.0638,
      "theoretical_loss": 3.344382157494031,
      "tokens_seen": 3057647616
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.406515808056492e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.344360385264042,
      "tokens_seen": 3057909760
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.39849141389825e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.3443386154229877,
      "tokens_seen": 3058171904
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.39046701974001e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.344316847970402,
      "tokens_seen": 3058434048
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.382442625581769e-05,
      "loss": 0.0634,
      "theoretical_loss": 3.344295082905817,
      "tokens_seen": 3058696192
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.374418231423529e-05,
      "loss": 0.0656,
      "theoretical_loss": 3.344273320228767,
      "tokens_seen": 3058958336
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.366393837265287e-05,
      "loss": 0.0608,
      "theoretical_loss": 3.344251559938785,
      "tokens_seen": 3059220480
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.358369443107044e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.3442298020354047,
      "tokens_seen": 3059482624
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.350345048948804e-05,
      "loss": 0.0641,
      "theoretical_loss": 3.3442080465181605,
      "tokens_seen": 3059744768
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.342320654790563e-05,
      "loss": 0.064,
      "theoretical_loss": 3.3441862933865854,
      "tokens_seen": 3060006912
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.334296260632323e-05,
      "loss": 0.0615,
      "theoretical_loss": 3.3441645426402142,
      "tokens_seen": 3060269056
    },
    {
      "epoch": 0.93,
      "objective/train/advantage_avg": 0.0004281561414245516,
      "objective/train/docs_used": 1112043,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2398734092712402,
      "objective/train/original_loss": 1.2398734092712402,
      "objective/train/theoretical_loss": 3.344142794278581,
      "objective/train/tokens_used": 3080991200,
      "objective/train/value_avg": -0.00775909423828125,
      "objective/train/value_loss": 0.00013446260709315538,
      "objective/train/value_max": -3.2961368560791016e-05,
      "objective/train/value_min": -0.2418212890625,
      "objective/train/value_reward_corr": 0.7998507260566405,
      "objective/train/value_std": 0.015167236328125,
      "objective/train/weight_avg": 1.0004942417144775,
      "objective/train/weighted_lm_loss": 1.2405970096588135,
      "objective/train/weights_max": 1.2352036237716675,
      "objective/train/weights_min": 0.6098437309265137,
      "theoretical_loss": 3.344142794278581,
      "tokens_seen": 3060531200
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.326271866474081e-05,
      "loss": 0.0621,
      "theoretical_loss": 3.344142794278581,
      "tokens_seen": 3060531200
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.318247472315841e-05,
      "loss": 0.0633,
      "theoretical_loss": 3.34412104830122,
      "tokens_seen": 3060793344
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.310223078157598e-05,
      "loss": 0.062,
      "theoretical_loss": 3.344099304707665,
      "tokens_seen": 3061055488
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.302198683999358e-05,
      "loss": 0.062,
      "theoretical_loss": 3.3440775634974513,
      "tokens_seen": 3061317632
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.294174289841117e-05,
      "loss": 0.0631,
      "theoretical_loss": 3.3440558246701135,
      "tokens_seen": 3061579776
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.286149895682877e-05,
      "loss": 0.0644,
      "theoretical_loss": 3.3440340882251864,
      "tokens_seen": 3061841920
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.278125501524635e-05,
      "loss": 0.0654,
      "theoretical_loss": 3.344012354162205,
      "tokens_seen": 3062104064
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.270101107366394e-05,
      "loss": 0.063,
      "theoretical_loss": 3.3439906224807046,
      "tokens_seen": 3062366208
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.262076713208152e-05,
      "loss": 0.0645,
      "theoretical_loss": 3.34396889318022,
      "tokens_seen": 3062628352
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.254052319049911e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.3439471662602864,
      "tokens_seen": 3062890496
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.246027924891671e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.34392544172044,
      "tokens_seen": 3063152640
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.23800353073343e-05,
      "loss": 0.0636,
      "theoretical_loss": 3.3439037195602155,
      "tokens_seen": 3063414784
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.22997913657519e-05,
      "loss": 0.0648,
      "theoretical_loss": 3.3438819997791493,
      "tokens_seen": 3063676928
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.221954742416948e-05,
      "loss": 0.0645,
      "theoretical_loss": 3.343860282376778,
      "tokens_seen": 3063939072
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.213930348258706e-05,
      "loss": 0.0614,
      "theoretical_loss": 3.343838567352636,
      "tokens_seen": 3064201216
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.205905954100465e-05,
      "loss": 0.0619,
      "theoretical_loss": 3.34381685470626,
      "tokens_seen": 3064463360
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.197881559942225e-05,
      "loss": 0.0661,
      "theoretical_loss": 3.3437951444371867,
      "tokens_seen": 3064725504
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.189857165783983e-05,
      "loss": 0.0647,
      "theoretical_loss": 3.3437734365449527,
      "tokens_seen": 3064987648
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.181832771625742e-05,
      "loss": 0.0651,
      "theoretical_loss": 3.3437517310290934,
      "tokens_seen": 3065249792
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.173808377467502e-05,
      "loss": 0.0626,
      "theoretical_loss": 3.343730027889147,
      "tokens_seen": 3065511936
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.165783983309259e-05,
      "loss": 0.0634,
      "theoretical_loss": 3.3437083271246486,
      "tokens_seen": 3065774080
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.157759589151019e-05,
      "loss": 0.0624,
      "theoretical_loss": 3.343686628735137,
      "tokens_seen": 3066036224
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.149735194992778e-05,
      "loss": 0.0623,
      "theoretical_loss": 3.3436649327201473,
      "tokens_seen": 3066298368
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.141710800834537e-05,
      "loss": 0.0674,
      "theoretical_loss": 3.343643239079218,
      "tokens_seen": 3066560512
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.133686406676296e-05,
      "loss": 0.0615,
      "theoretical_loss": 3.343621547811886,
      "tokens_seen": 3066822656
    },
    {
      "epoch": 0.93,
      "objective/train/advantage_avg": 0.0005850589368492365,
      "objective/train/docs_used": 1114524,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3618371486663818,
      "objective/train/original_loss": 1.3618371486663818,
      "objective/train/theoretical_loss": 3.3435998589176883,
      "objective/train/tokens_used": 3087544800,
      "objective/train/value_avg": -0.007190704345703125,
      "objective/train/value_loss": 0.00024140205641742796,
      "objective/train/value_max": -2.586841583251953e-05,
      "objective/train/value_min": -0.97705078125,
      "objective/train/value_reward_corr": 0.685630743050589,
      "objective/train/value_std": 0.01404571533203125,
      "objective/train/weight_avg": 1.0006887912750244,
      "objective/train/weighted_lm_loss": 1.3621577024459839,
      "objective/train/weights_max": 1.189415454864502,
      "objective/train/weights_min": 0.2359674870967865,
      "theoretical_loss": 3.3435998589176883,
      "tokens_seen": 3067084800
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.125662012518056e-05,
      "loss": 0.0651,
      "theoretical_loss": 3.3435998589176883,
      "tokens_seen": 3067084800
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.117637618359813e-05,
      "loss": 0.061,
      "theoretical_loss": 3.343578172396163,
      "tokens_seen": 3067346944
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.109613224201573e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.3435564882468483,
      "tokens_seen": 3067609088
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.101588830043332e-05,
      "loss": 0.0623,
      "theoretical_loss": 3.343534806469281,
      "tokens_seen": 3067871232
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.093564435885091e-05,
      "loss": 0.0656,
      "theoretical_loss": 3.343513127063,
      "tokens_seen": 3068133376
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.08554004172685e-05,
      "loss": 0.0648,
      "theoretical_loss": 3.343491450027542,
      "tokens_seen": 3068395520
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.077515647568609e-05,
      "loss": 0.0652,
      "theoretical_loss": 3.343469775362447,
      "tokens_seen": 3068657664
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.069491253410367e-05,
      "loss": 0.0628,
      "theoretical_loss": 3.3434481030672516,
      "tokens_seen": 3068919808
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.061466859252126e-05,
      "loss": 0.0629,
      "theoretical_loss": 3.343426433141496,
      "tokens_seen": 3069181952
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.053442465093886e-05,
      "loss": 0.0615,
      "theoretical_loss": 3.343404765584717,
      "tokens_seen": 3069444096
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.045418070935644e-05,
      "loss": 0.0636,
      "theoretical_loss": 3.343383100396455,
      "tokens_seen": 3069706240
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.037393676777404e-05,
      "loss": 0.0631,
      "theoretical_loss": 3.343361437576248,
      "tokens_seen": 3069968384
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.029369282619163e-05,
      "loss": 0.0652,
      "theoretical_loss": 3.343339777123635,
      "tokens_seen": 3070230528
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.021344888460923e-05,
      "loss": 0.0651,
      "theoretical_loss": 3.343318119038155,
      "tokens_seen": 3070492672
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.01332049430268e-05,
      "loss": 0.0662,
      "theoretical_loss": 3.343296463319348,
      "tokens_seen": 3070754816
    },
    {
      "epoch": 0.93,
      "learning_rate": 7.00529610014444e-05,
      "loss": 0.0646,
      "theoretical_loss": 3.3432748099667524,
      "tokens_seen": 3071016960
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.997271705986198e-05,
      "loss": 0.0616,
      "theoretical_loss": 3.343253158979908,
      "tokens_seen": 3071279104
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.989247311827957e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.343231510358355,
      "tokens_seen": 3071541248
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.981222917669717e-05,
      "loss": 0.0637,
      "theoretical_loss": 3.343209864101633,
      "tokens_seen": 3071803392
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.973198523511474e-05,
      "loss": 0.066,
      "theoretical_loss": 3.343188220209281,
      "tokens_seen": 3072065536
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.965174129353234e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.34316657868084,
      "tokens_seen": 3072327680
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.957149735194992e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.34314493951585,
      "tokens_seen": 3072589824
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.949125341036752e-05,
      "loss": 0.0656,
      "theoretical_loss": 3.343123302713851,
      "tokens_seen": 3072851968
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.941100946878511e-05,
      "loss": 0.0658,
      "theoretical_loss": 3.3431016682743833,
      "tokens_seen": 3073114112
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.93307655272027e-05,
      "loss": 0.0645,
      "theoretical_loss": 3.343080036196988,
      "tokens_seen": 3073376256
    },
    {
      "epoch": 0.93,
      "objective/train/advantage_avg": 0.0006747455336153507,
      "objective/train/docs_used": 1116766,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.1027328968048096,
      "objective/train/original_loss": 1.1027328968048096,
      "objective/train/theoretical_loss": 3.3430584064812052,
      "objective/train/tokens_used": 3094098400,
      "objective/train/value_avg": -0.0059967041015625,
      "objective/train/value_loss": 0.0001406289666192606,
      "objective/train/value_max": -2.6047229766845703e-05,
      "objective/train/value_min": -0.935546875,
      "objective/train/value_reward_corr": 0.7394594899401773,
      "objective/train/value_std": 0.01275634765625,
      "objective/train/weight_avg": 1.000738263130188,
      "objective/train/weighted_lm_loss": 1.1030726432800293,
      "objective/train/weights_max": 1.1870945692062378,
      "objective/train/weights_min": 0.3722897469997406,
      "theoretical_loss": 3.3430584064812052,
      "tokens_seen": 3073638400
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.925052158562029e-05,
      "loss": 0.0608,
      "theoretical_loss": 3.3430584064812052,
      "tokens_seen": 3073638400
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.917027764403788e-05,
      "loss": 0.0622,
      "theoretical_loss": 3.3430367791265763,
      "tokens_seen": 3073900544
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.909003370245546e-05,
      "loss": 0.0625,
      "theoretical_loss": 3.3430151541326416,
      "tokens_seen": 3074162688
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.900978976087306e-05,
      "loss": 0.0659,
      "theoretical_loss": 3.3429935314989425,
      "tokens_seen": 3074424832
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.892954581929065e-05,
      "loss": 0.0612,
      "theoretical_loss": 3.34297191122502,
      "tokens_seen": 3074686976
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.884930187770823e-05,
      "loss": 0.063,
      "theoretical_loss": 3.342950293310415,
      "tokens_seen": 3074949120
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.876905793612583e-05,
      "loss": 0.0628,
      "theoretical_loss": 3.3429286777546703,
      "tokens_seen": 3075211264
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.86888139945434e-05,
      "loss": 0.0623,
      "theoretical_loss": 3.342907064557326,
      "tokens_seen": 3075473408
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.8608570052961e-05,
      "loss": 0.0662,
      "theoretical_loss": 3.3428854537179244,
      "tokens_seen": 3075735552
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.852832611137859e-05,
      "loss": 0.0607,
      "theoretical_loss": 3.3428638452360078,
      "tokens_seen": 3075997696
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.844808216979619e-05,
      "loss": 0.063,
      "theoretical_loss": 3.342842239111117,
      "tokens_seen": 3076259840
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.836783822821377e-05,
      "loss": 0.0608,
      "theoretical_loss": 3.342820635342795,
      "tokens_seen": 3076521984
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.828759428663137e-05,
      "loss": 0.0637,
      "theoretical_loss": 3.342799033930584,
      "tokens_seen": 3076784128
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.820735034504894e-05,
      "loss": 0.0644,
      "theoretical_loss": 3.3427774348740256,
      "tokens_seen": 3077046272
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.812710640346654e-05,
      "loss": 0.0637,
      "theoretical_loss": 3.342755838172663,
      "tokens_seen": 3077308416
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.804686246188413e-05,
      "loss": 0.0658,
      "theoretical_loss": 3.3427342438260386,
      "tokens_seen": 3077570560
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.796661852030171e-05,
      "loss": 0.0658,
      "theoretical_loss": 3.342712651833695,
      "tokens_seen": 3077832704
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.788637457871931e-05,
      "loss": 0.0648,
      "theoretical_loss": 3.342691062195175,
      "tokens_seen": 3078094848
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.78061306371369e-05,
      "loss": 0.066,
      "theoretical_loss": 3.3426694749100214,
      "tokens_seen": 3078356992
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.772588669555448e-05,
      "loss": 0.0617,
      "theoretical_loss": 3.342647889977778,
      "tokens_seen": 3078619136
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.764564275397207e-05,
      "loss": 0.0653,
      "theoretical_loss": 3.342626307397987,
      "tokens_seen": 3078881280
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.756539881238967e-05,
      "loss": 0.064,
      "theoretical_loss": 3.342604727170193,
      "tokens_seen": 3079143424
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.748515487080725e-05,
      "loss": 0.0615,
      "theoretical_loss": 3.3425831492939384,
      "tokens_seen": 3079405568
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.740491092922485e-05,
      "loss": 0.0629,
      "theoretical_loss": 3.3425615737687675,
      "tokens_seen": 3079667712
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.732466698764244e-05,
      "loss": 0.0633,
      "theoretical_loss": 3.342540000594224,
      "tokens_seen": 3079929856
    },
    {
      "epoch": 0.93,
      "objective/train/advantage_avg": -0.00011591265501920134,
      "objective/train/docs_used": 1119127,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.1800016164779663,
      "objective/train/original_loss": 1.1800014972686768,
      "objective/train/theoretical_loss": 3.342518429769851,
      "objective/train/tokens_used": 3100652000,
      "objective/train/value_avg": -0.0055389404296875,
      "objective/train/value_loss": 0.00014903591363690794,
      "objective/train/value_max": -2.3365020751953125e-05,
      "objective/train/value_min": -0.31640625,
      "objective/train/value_reward_corr": 0.6550612583763221,
      "objective/train/value_std": 0.0103302001953125,
      "objective/train/weight_avg": 0.9999504685401917,
      "objective/train/weighted_lm_loss": 1.1799626350402832,
      "objective/train/weights_max": 1.2208958864212036,
      "objective/train/weights_min": 0.37342190742492676,
      "theoretical_loss": 3.342518429769851,
      "tokens_seen": 3080192000
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.724442304606002e-05,
      "loss": 0.0611,
      "theoretical_loss": 3.342518429769851,
      "tokens_seen": 3080192000
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.716417910447761e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.342496861295193,
      "tokens_seen": 3080454144
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.70839351628952e-05,
      "loss": 0.0656,
      "theoretical_loss": 3.342475295169794,
      "tokens_seen": 3080716288
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.70036912213128e-05,
      "loss": 0.0607,
      "theoretical_loss": 3.342453731393199,
      "tokens_seen": 3080978432
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.692344727973038e-05,
      "loss": 0.0644,
      "theoretical_loss": 3.342432169964951,
      "tokens_seen": 3081240576
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.684320333814798e-05,
      "loss": 0.06,
      "theoretical_loss": 3.3424106108845955,
      "tokens_seen": 3081502720
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.676295939656555e-05,
      "loss": 0.0656,
      "theoretical_loss": 3.342389054151677,
      "tokens_seen": 3081764864
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.668271545498315e-05,
      "loss": 0.0608,
      "theoretical_loss": 3.34236749976574,
      "tokens_seen": 3082027008
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.660247151340074e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.34234594772633,
      "tokens_seen": 3082289152
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.652222757181833e-05,
      "loss": 0.061,
      "theoretical_loss": 3.342324398032991,
      "tokens_seen": 3082551296
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.644198363023592e-05,
      "loss": 0.0616,
      "theoretical_loss": 3.342302850685269,
      "tokens_seen": 3082813440
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.636173968865352e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.3422813056827088,
      "tokens_seen": 3083075584
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.628149574707109e-05,
      "loss": 0.0657,
      "theoretical_loss": 3.342259763024856,
      "tokens_seen": 3083337728
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.620125180548869e-05,
      "loss": 0.0625,
      "theoretical_loss": 3.3422382227112557,
      "tokens_seen": 3083599872
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.612100786390628e-05,
      "loss": 0.0611,
      "theoretical_loss": 3.3422166847414543,
      "tokens_seen": 3083862016
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.604076392232386e-05,
      "loss": 0.0633,
      "theoretical_loss": 3.3421951491149966,
      "tokens_seen": 3084124160
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.596051998074146e-05,
      "loss": 0.0644,
      "theoretical_loss": 3.3421736158314292,
      "tokens_seen": 3084386304
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.588027603915905e-05,
      "loss": 0.0631,
      "theoretical_loss": 3.3421520848902984,
      "tokens_seen": 3084648448
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.580003209757663e-05,
      "loss": 0.0624,
      "theoretical_loss": 3.3421305562911496,
      "tokens_seen": 3084910592
    },
    {
      "epoch": 0.93,
      "learning_rate": 6.571978815599422e-05,
      "loss": 0.0622,
      "theoretical_loss": 3.342109030033529,
      "tokens_seen": 3085172736
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.563954421441182e-05,
      "loss": 0.0641,
      "theoretical_loss": 3.342087506116984,
      "tokens_seen": 3085434880
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.55593002728294e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.3420659845410605,
      "tokens_seen": 3085697024
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.5479056331247e-05,
      "loss": 0.0639,
      "theoretical_loss": 3.342044465305305,
      "tokens_seen": 3085959168
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.539881238966459e-05,
      "loss": 0.0646,
      "theoretical_loss": 3.3420229484092645,
      "tokens_seen": 3086221312
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.531856844808217e-05,
      "loss": 0.0659,
      "theoretical_loss": 3.3420014338524857,
      "tokens_seen": 3086483456
    },
    {
      "epoch": 0.94,
      "objective/train/advantage_avg": 0.0005402964889071882,
      "objective/train/docs_used": 1121586,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2228138446807861,
      "objective/train/original_loss": 1.2228137254714966,
      "objective/train/theoretical_loss": 3.3419799216345156,
      "objective/train/tokens_used": 3107205600,
      "objective/train/value_avg": -0.005706787109375,
      "objective/train/value_loss": 0.0002355128963245079,
      "objective/train/value_max": -4.2319297790527344e-05,
      "objective/train/value_min": -0.6748046875,
      "objective/train/value_reward_corr": 0.653781937380487,
      "objective/train/value_std": 0.01299285888671875,
      "objective/train/weight_avg": 1.0006425380706787,
      "objective/train/weighted_lm_loss": 1.223292350769043,
      "objective/train/weights_max": 1.7227903604507446,
      "objective/train/weights_min": 0.3708694577217102,
      "theoretical_loss": 3.3419799216345156,
      "tokens_seen": 3086745600
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.523832450649976e-05,
      "loss": 0.0629,
      "theoretical_loss": 3.3419799216345156,
      "tokens_seen": 3086745600
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.515808056491734e-05,
      "loss": 0.0639,
      "theoretical_loss": 3.341958411754902,
      "tokens_seen": 3087007744
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.507783662333494e-05,
      "loss": 0.0628,
      "theoretical_loss": 3.3419369042131915,
      "tokens_seen": 3087269888
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.499759268175253e-05,
      "loss": 0.066,
      "theoretical_loss": 3.341915399008932,
      "tokens_seen": 3087532032
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.491734874017013e-05,
      "loss": 0.0644,
      "theoretical_loss": 3.3418938961416704,
      "tokens_seen": 3087794176
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.48371047985877e-05,
      "loss": 0.0645,
      "theoretical_loss": 3.341872395610955,
      "tokens_seen": 3088056320
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.47568608570053e-05,
      "loss": 0.0632,
      "theoretical_loss": 3.3418508974163332,
      "tokens_seen": 3088318464
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.467661691542288e-05,
      "loss": 0.064,
      "theoretical_loss": 3.3418294015573533,
      "tokens_seen": 3088580608
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.459637297384048e-05,
      "loss": 0.0614,
      "theoretical_loss": 3.3418079080335628,
      "tokens_seen": 3088842752
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.451612903225807e-05,
      "loss": 0.0639,
      "theoretical_loss": 3.34178641684451,
      "tokens_seen": 3089104896
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.443588509067567e-05,
      "loss": 0.0651,
      "theoretical_loss": 3.3417649279897437,
      "tokens_seen": 3089367040
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.435564114909324e-05,
      "loss": 0.0608,
      "theoretical_loss": 3.341743441468812,
      "tokens_seen": 3089629184
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.427539720751084e-05,
      "loss": 0.0637,
      "theoretical_loss": 3.3417219572812633,
      "tokens_seen": 3089891328
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.419515326592842e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.3417004754266464,
      "tokens_seen": 3090153472
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.411490932434601e-05,
      "loss": 0.0632,
      "theoretical_loss": 3.34167899590451,
      "tokens_seen": 3090415616
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.403466538276361e-05,
      "loss": 0.063,
      "theoretical_loss": 3.3416575187144035,
      "tokens_seen": 3090677760
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.395442144118119e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.341636043855875,
      "tokens_seen": 3090939904
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.387417749959878e-05,
      "loss": 0.0637,
      "theoretical_loss": 3.3416145713284746,
      "tokens_seen": 3091202048
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.379393355801636e-05,
      "loss": 0.0669,
      "theoretical_loss": 3.341593101131751,
      "tokens_seen": 3091464192
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.371368961643396e-05,
      "loss": 0.0662,
      "theoretical_loss": 3.341571633265254,
      "tokens_seen": 3091726336
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.363344567485155e-05,
      "loss": 0.0674,
      "theoretical_loss": 3.341550167728533,
      "tokens_seen": 3091988480
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.355320173326915e-05,
      "loss": 0.0629,
      "theoretical_loss": 3.3415287045211377,
      "tokens_seen": 3092250624
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.347295779168673e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.3415072436426176,
      "tokens_seen": 3092512768
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.339271385010432e-05,
      "loss": 0.0656,
      "theoretical_loss": 3.341485785092523,
      "tokens_seen": 3092774912
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.33124699085219e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.341464328870404,
      "tokens_seen": 3093037056
    },
    {
      "epoch": 0.94,
      "objective/train/advantage_avg": -8.85079352883622e-05,
      "objective/train/docs_used": 1123676,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.302232027053833,
      "objective/train/original_loss": 1.302232027053833,
      "objective/train/theoretical_loss": 3.3414428749758107,
      "objective/train/tokens_used": 3113759200,
      "objective/train/value_avg": -0.005947113037109375,
      "objective/train/value_loss": 0.00013723468873649836,
      "objective/train/value_max": -1.6987323760986328e-05,
      "objective/train/value_min": -0.43994140625,
      "objective/train/value_reward_corr": 0.8200767261251052,
      "objective/train/value_std": 0.01557159423828125,
      "objective/train/weight_avg": 0.999978244304657,
      "objective/train/weighted_lm_loss": 1.302291989326477,
      "objective/train/weights_max": 1.2921568155288696,
      "objective/train/weights_min": 0.6140877604484558,
      "theoretical_loss": 3.3414428749758107,
      "tokens_seen": 3093299200
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.323222596693949e-05,
      "loss": 0.064,
      "theoretical_loss": 3.3414428749758107,
      "tokens_seen": 3093299200
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.315198202535709e-05,
      "loss": 0.0666,
      "theoretical_loss": 3.341421423408293,
      "tokens_seen": 3093561344
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.307173808377467e-05,
      "loss": 0.0663,
      "theoretical_loss": 3.341399974167402,
      "tokens_seen": 3093823488
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.299149414219227e-05,
      "loss": 0.0646,
      "theoretical_loss": 3.3413785272526875,
      "tokens_seen": 3094085632
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.291125020060985e-05,
      "loss": 0.0632,
      "theoretical_loss": 3.341357082663701,
      "tokens_seen": 3094347776
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.283100625902744e-05,
      "loss": 0.064,
      "theoretical_loss": 3.3413356403999925,
      "tokens_seen": 3094609920
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.275076231744503e-05,
      "loss": 0.0689,
      "theoretical_loss": 3.341314200461113,
      "tokens_seen": 3094872064
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.267051837586263e-05,
      "loss": 0.0647,
      "theoretical_loss": 3.3412927628466145,
      "tokens_seen": 3095134208
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.259027443428021e-05,
      "loss": 0.0628,
      "theoretical_loss": 3.341271327556047,
      "tokens_seen": 3095396352
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.251003049269781e-05,
      "loss": 0.0632,
      "theoretical_loss": 3.341249894588963,
      "tokens_seen": 3095658496
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.242978655111539e-05,
      "loss": 0.0634,
      "theoretical_loss": 3.3412284639449124,
      "tokens_seen": 3095920640
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.234954260953298e-05,
      "loss": 0.0627,
      "theoretical_loss": 3.341207035623448,
      "tokens_seen": 3096182784
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.226929866795057e-05,
      "loss": 0.0633,
      "theoretical_loss": 3.341185609624121,
      "tokens_seen": 3096444928
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.218905472636816e-05,
      "loss": 0.0633,
      "theoretical_loss": 3.341164185946483,
      "tokens_seen": 3096707072
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.210881078478575e-05,
      "loss": 0.0639,
      "theoretical_loss": 3.3411427645900864,
      "tokens_seen": 3096969216
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.202856684320334e-05,
      "loss": 0.066,
      "theoretical_loss": 3.341121345554483,
      "tokens_seen": 3097231360
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.194832290162093e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.341099928839225,
      "tokens_seen": 3097493504
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.186807896003851e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.3410785144438644,
      "tokens_seen": 3097755648
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.178783501845611e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.3410571023679543,
      "tokens_seen": 3098017792
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.17075910768737e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.3410356926110465,
      "tokens_seen": 3098279936
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.162734713529128e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.3410142851726943,
      "tokens_seen": 3098542080
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.154710319370888e-05,
      "loss": 0.0644,
      "theoretical_loss": 3.3409928800524495,
      "tokens_seen": 3098804224
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.146685925212647e-05,
      "loss": 0.065,
      "theoretical_loss": 3.3409714772498664,
      "tokens_seen": 3099066368
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.138661531054405e-05,
      "loss": 0.0615,
      "theoretical_loss": 3.3409500767644973,
      "tokens_seen": 3099328512
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.130637136896165e-05,
      "loss": 0.0627,
      "theoretical_loss": 3.340928678595895,
      "tokens_seen": 3099590656
    },
    {
      "epoch": 0.94,
      "objective/train/advantage_avg": 0.0005747837130911648,
      "objective/train/docs_used": 1126162,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2767504453659058,
      "objective/train/original_loss": 1.2767504453659058,
      "objective/train/theoretical_loss": 3.3409072827436135,
      "objective/train/tokens_used": 3120312800,
      "objective/train/value_avg": -0.00778961181640625,
      "objective/train/value_loss": 0.000372168084140867,
      "objective/train/value_max": -2.7298927307128906e-05,
      "objective/train/value_min": -0.73974609375,
      "objective/train/value_reward_corr": 0.6550666805696803,
      "objective/train/value_std": 0.01641845703125,
      "objective/train/weight_avg": 1.0007363557815552,
      "objective/train/weighted_lm_loss": 1.2775542736053467,
      "objective/train/weights_max": 2.0023789405822754,
      "objective/train/weights_min": 0.3681640923023224,
      "theoretical_loss": 3.3409072827436135,
      "tokens_seen": 3099852800
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.122612742737924e-05,
      "loss": 0.0638,
      "theoretical_loss": 3.3409072827436135,
      "tokens_seen": 3099852800
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.114588348579682e-05,
      "loss": 0.0659,
      "theoretical_loss": 3.3408858892072058,
      "tokens_seen": 3100114944
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.106563954421442e-05,
      "loss": 0.065,
      "theoretical_loss": 3.3408644979862254,
      "tokens_seen": 3100377088
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.0985395602632006e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.340843109080226,
      "tokens_seen": 3100639232
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.090515166104959e-05,
      "loss": 0.0638,
      "theoretical_loss": 3.340821722488762,
      "tokens_seen": 3100901376
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.082490771946718e-05,
      "loss": 0.0636,
      "theoretical_loss": 3.3408003382113862,
      "tokens_seen": 3101163520
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.074466377788477e-05,
      "loss": 0.0633,
      "theoretical_loss": 3.3407789562476538,
      "tokens_seen": 3101425664
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.066441983630236e-05,
      "loss": 0.0652,
      "theoretical_loss": 3.3407575765971176,
      "tokens_seen": 3101687808
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.058417589471995e-05,
      "loss": 0.0662,
      "theoretical_loss": 3.340736199259333,
      "tokens_seen": 3101949952
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.050393195313754e-05,
      "loss": 0.0626,
      "theoretical_loss": 3.3407148242338542,
      "tokens_seen": 3102212096
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.042368801155513e-05,
      "loss": 0.0655,
      "theoretical_loss": 3.3406934515202353,
      "tokens_seen": 3102474240
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.034344406997272e-05,
      "loss": 0.067,
      "theoretical_loss": 3.3406720811180315,
      "tokens_seen": 3102736384
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.026320012839031e-05,
      "loss": 0.0625,
      "theoretical_loss": 3.340650713026797,
      "tokens_seen": 3102998528
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.01829561868079e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.340629347246087,
      "tokens_seen": 3103260672
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.010271224522549e-05,
      "loss": 0.0637,
      "theoretical_loss": 3.3406079837754565,
      "tokens_seen": 3103522816
    },
    {
      "epoch": 0.94,
      "learning_rate": 6.002246830364308e-05,
      "loss": 0.0645,
      "theoretical_loss": 3.3405866226144605,
      "tokens_seen": 3103784960
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.9942224362060665e-05,
      "loss": 0.0654,
      "theoretical_loss": 3.3405652637626546,
      "tokens_seen": 3104047104
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.986198042047825e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.340543907219594,
      "tokens_seen": 3104309248
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.978173647889584e-05,
      "loss": 0.0651,
      "theoretical_loss": 3.340522552984834,
      "tokens_seen": 3104571392
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.9701492537313435e-05,
      "loss": 0.0632,
      "theoretical_loss": 3.3405012010579305,
      "tokens_seen": 3104833536
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.962124859573102e-05,
      "loss": 0.0629,
      "theoretical_loss": 3.340479851438439,
      "tokens_seen": 3105095680
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.954100465414861e-05,
      "loss": 0.0645,
      "theoretical_loss": 3.340458504125916,
      "tokens_seen": 3105357824
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.9460760712566205e-05,
      "loss": 0.0639,
      "theoretical_loss": 3.3404371591199165,
      "tokens_seen": 3105619968
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.938051677098379e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.3404158164199975,
      "tokens_seen": 3105882112
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.930027282940138e-05,
      "loss": 0.0685,
      "theoretical_loss": 3.3403944760257147,
      "tokens_seen": 3106144256
    },
    {
      "epoch": 0.94,
      "objective/train/advantage_avg": 0.00033568739308975637,
      "objective/train/docs_used": 1128625,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2322754859924316,
      "objective/train/original_loss": 1.2322752475738525,
      "objective/train/theoretical_loss": 3.340373137936625,
      "objective/train/tokens_used": 3126866400,
      "objective/train/value_avg": -0.0088043212890625,
      "objective/train/value_loss": 0.0002217359869973734,
      "objective/train/value_max": -3.5643577575683594e-05,
      "objective/train/value_min": -0.6103515625,
      "objective/train/value_reward_corr": 0.7731854039618011,
      "objective/train/value_std": 0.0171966552734375,
      "objective/train/weight_avg": 1.0004370212554932,
      "objective/train/weighted_lm_loss": 1.23274827003479,
      "objective/train/weights_max": 1.7944713830947876,
      "objective/train/weights_min": 0.3681126534938812,
      "theoretical_loss": 3.340373137936625,
      "tokens_seen": 3106406400
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.9220028887818975e-05,
      "loss": 0.0646,
      "theoretical_loss": 3.340373137936625,
      "tokens_seen": 3106406400
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.913978494623656e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.340351802152284,
      "tokens_seen": 3106668544
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.9059541004654146e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.3403304686722497,
      "tokens_seen": 3106930688
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.897929706307174e-05,
      "loss": 0.0623,
      "theoretical_loss": 3.3403091374960776,
      "tokens_seen": 3107192832
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.8899053121489324e-05,
      "loss": 0.066,
      "theoretical_loss": 3.340287808623325,
      "tokens_seen": 3107454976
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.8818809179906916e-05,
      "loss": 0.0652,
      "theoretical_loss": 3.340266482053549,
      "tokens_seen": 3107717120
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.873856523832451e-05,
      "loss": 0.0637,
      "theoretical_loss": 3.3402451577863066,
      "tokens_seen": 3107979264
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.8658321296742094e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.340223835821155,
      "tokens_seen": 3108241408
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.8578077355159686e-05,
      "loss": 0.0628,
      "theoretical_loss": 3.3402025161576514,
      "tokens_seen": 3108503552
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.849783341357728e-05,
      "loss": 0.0619,
      "theoretical_loss": 3.340181198795354,
      "tokens_seen": 3108765696
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.8417589471994864e-05,
      "loss": 0.0638,
      "theoretical_loss": 3.340159883733819,
      "tokens_seen": 3109027840
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.8337345530412456e-05,
      "loss": 0.0654,
      "theoretical_loss": 3.3401385709726052,
      "tokens_seen": 3109289984
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.825710158883005e-05,
      "loss": 0.0634,
      "theoretical_loss": 3.3401172605112706,
      "tokens_seen": 3109552128
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.8176857647247634e-05,
      "loss": 0.0682,
      "theoretical_loss": 3.3400959523493725,
      "tokens_seen": 3109814272
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.809661370566522e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.3400746464864692,
      "tokens_seen": 3110076416
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.801636976408281e-05,
      "loss": 0.0657,
      "theoretical_loss": 3.340053342922119,
      "tokens_seen": 3110338560
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.79361258225004e-05,
      "loss": 0.0645,
      "theoretical_loss": 3.3400320416558804,
      "tokens_seen": 3110600704
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.785588188091799e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.340010742687311,
      "tokens_seen": 3110862848
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.777563793933558e-05,
      "loss": 0.0655,
      "theoretical_loss": 3.339989446015971,
      "tokens_seen": 3111124992
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.769539399775317e-05,
      "loss": 0.0653,
      "theoretical_loss": 3.3399681516414175,
      "tokens_seen": 3111387136
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.761515005617076e-05,
      "loss": 0.066,
      "theoretical_loss": 3.33994685956321,
      "tokens_seen": 3111649280
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.753490611458835e-05,
      "loss": 0.0683,
      "theoretical_loss": 3.3399255697809074,
      "tokens_seen": 3111911424
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.745466217300594e-05,
      "loss": 0.0675,
      "theoretical_loss": 3.3399042822940688,
      "tokens_seen": 3112173568
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.737441823142353e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.339882997102253,
      "tokens_seen": 3112435712
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.729417428984112e-05,
      "loss": 0.0651,
      "theoretical_loss": 3.3398617142050195,
      "tokens_seen": 3112697856
    },
    {
      "epoch": 0.94,
      "objective/train/advantage_avg": 0.0003075933491345495,
      "objective/train/docs_used": 1130876,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2553244829177856,
      "objective/train/original_loss": 1.2553246021270752,
      "objective/train/theoretical_loss": 3.3398404336019283,
      "objective/train/tokens_used": 3133420000,
      "objective/train/value_avg": -0.0066986083984375,
      "objective/train/value_loss": 0.00021400822151917964,
      "objective/train/value_max": -1.9848346710205078e-05,
      "objective/train/value_min": -0.6826171875,
      "objective/train/value_reward_corr": 0.7202309173108771,
      "objective/train/value_std": 0.0164794921875,
      "objective/train/weight_avg": 1.0004074573516846,
      "objective/train/weighted_lm_loss": 1.2557744979858398,
      "objective/train/weights_max": 1.742195725440979,
      "objective/train/weights_min": 0.3772936761379242,
      "theoretical_loss": 3.3398404336019283,
      "tokens_seen": 3112960000
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.7213930348258714e-05,
      "loss": 0.065,
      "theoretical_loss": 3.3398404336019283,
      "tokens_seen": 3112960000
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.713368640667629e-05,
      "loss": 0.0647,
      "theoretical_loss": 3.3398191552925383,
      "tokens_seen": 3113222144
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.7053442465093885e-05,
      "loss": 0.0631,
      "theoretical_loss": 3.33979787927641,
      "tokens_seen": 3113484288
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.697319852351147e-05,
      "loss": 0.0624,
      "theoretical_loss": 3.3397766055531015,
      "tokens_seen": 3113746432
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.689295458192906e-05,
      "loss": 0.0654,
      "theoretical_loss": 3.3397553341221746,
      "tokens_seen": 3114008576
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.6812710640346655e-05,
      "loss": 0.0617,
      "theoretical_loss": 3.3397340649831877,
      "tokens_seen": 3114270720
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.673246669876425e-05,
      "loss": 0.0619,
      "theoretical_loss": 3.3397127981357024,
      "tokens_seen": 3114532864
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.665222275718183e-05,
      "loss": 0.065,
      "theoretical_loss": 3.339691533579278,
      "tokens_seen": 3114795008
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.6571978815599425e-05,
      "loss": 0.0644,
      "theoretical_loss": 3.3396702713134756,
      "tokens_seen": 3115057152
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.649173487401702e-05,
      "loss": 0.0661,
      "theoretical_loss": 3.3396490113378547,
      "tokens_seen": 3115319296
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.64114909324346e-05,
      "loss": 0.0644,
      "theoretical_loss": 3.3396277536519774,
      "tokens_seen": 3115581440
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.6331246990852196e-05,
      "loss": 0.0644,
      "theoretical_loss": 3.339606498255403,
      "tokens_seen": 3115843584
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.625100304926979e-05,
      "loss": 0.064,
      "theoretical_loss": 3.3395852451476937,
      "tokens_seen": 3116105728
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.6170759107687367e-05,
      "loss": 0.0647,
      "theoretical_loss": 3.339563994328409,
      "tokens_seen": 3116367872
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.609051516610496e-05,
      "loss": 0.0652,
      "theoretical_loss": 3.339542745797112,
      "tokens_seen": 3116630016
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.601027122452255e-05,
      "loss": 0.0674,
      "theoretical_loss": 3.3395214995533617,
      "tokens_seen": 3116892160
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.5930027282940137e-05,
      "loss": 0.0674,
      "theoretical_loss": 3.339500255596721,
      "tokens_seen": 3117154304
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.584978334135773e-05,
      "loss": 0.0703,
      "theoretical_loss": 3.339479013926751,
      "tokens_seen": 3117416448
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.576953939977532e-05,
      "loss": 0.0663,
      "theoretical_loss": 3.3394577745430136,
      "tokens_seen": 3117678592
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.568929545819291e-05,
      "loss": 0.0692,
      "theoretical_loss": 3.33943653744507,
      "tokens_seen": 3117940736
    },
    {
      "epoch": 0.94,
      "learning_rate": 5.56090515166105e-05,
      "loss": 0.067,
      "theoretical_loss": 3.339415302632482,
      "tokens_seen": 3118202880
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.552880757502809e-05,
      "loss": 0.0691,
      "theoretical_loss": 3.339394070104812,
      "tokens_seen": 3118465024
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.544856363344568e-05,
      "loss": 0.0655,
      "theoretical_loss": 3.339372839861622,
      "tokens_seen": 3118727168
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.536831969186327e-05,
      "loss": 0.066,
      "theoretical_loss": 3.339351611902474,
      "tokens_seen": 3118989312
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.5288075750280855e-05,
      "loss": 0.0638,
      "theoretical_loss": 3.339330386226931,
      "tokens_seen": 3119251456
    },
    {
      "epoch": 0.95,
      "objective/train/advantage_avg": 0.0004081982187926769,
      "objective/train/docs_used": 1133291,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.438138484954834,
      "objective/train/original_loss": 1.438138484954834,
      "objective/train/theoretical_loss": 3.3393091628345544,
      "objective/train/tokens_used": 3139973600,
      "objective/train/value_avg": -0.00905609130859375,
      "objective/train/value_loss": 0.00029880873626098037,
      "objective/train/value_max": -2.014636993408203e-05,
      "objective/train/value_min": -0.681640625,
      "objective/train/value_reward_corr": 0.6956209491124092,
      "objective/train/value_std": 0.0166778564453125,
      "objective/train/weight_avg": 1.0005412101745605,
      "objective/train/weighted_lm_loss": 1.4387654066085815,
      "objective/train/weights_max": 1.4040422439575195,
      "objective/train/weights_min": 0.37822747230529785,
      "theoretical_loss": 3.3393091628345544,
      "tokens_seen": 3119513600
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.520783180869844e-05,
      "loss": 0.0672,
      "theoretical_loss": 3.3393091628345544,
      "tokens_seen": 3119513600
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.512758786711603e-05,
      "loss": 0.063,
      "theoretical_loss": 3.3392879417249075,
      "tokens_seen": 3119775744
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.5047343925533625e-05,
      "loss": 0.0697,
      "theoretical_loss": 3.3392667228975528,
      "tokens_seen": 3120037888
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.496709998395121e-05,
      "loss": 0.0652,
      "theoretical_loss": 3.339245506352053,
      "tokens_seen": 3120300032
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.48868560423688e-05,
      "loss": 0.0685,
      "theoretical_loss": 3.339224292087972,
      "tokens_seen": 3120562176
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.4806612100786395e-05,
      "loss": 0.0675,
      "theoretical_loss": 3.339203080104871,
      "tokens_seen": 3120824320
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.472636815920398e-05,
      "loss": 0.0648,
      "theoretical_loss": 3.3391818704023146,
      "tokens_seen": 3121086464
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.464612421762157e-05,
      "loss": 0.0657,
      "theoretical_loss": 3.339160662979866,
      "tokens_seen": 3121348608
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.4565880276039165e-05,
      "loss": 0.0631,
      "theoretical_loss": 3.3391394578370877,
      "tokens_seen": 3121610752
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.448563633445675e-05,
      "loss": 0.0661,
      "theoretical_loss": 3.3391182549735445,
      "tokens_seen": 3121872896
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.440539239287434e-05,
      "loss": 0.0651,
      "theoretical_loss": 3.339097054388799,
      "tokens_seen": 3122135040
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.432514845129193e-05,
      "loss": 0.0674,
      "theoretical_loss": 3.3390758560824154,
      "tokens_seen": 3122397184
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.4244904509709514e-05,
      "loss": 0.0644,
      "theoretical_loss": 3.3390546600539577,
      "tokens_seen": 3122659328
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.4164660568127106e-05,
      "loss": 0.0672,
      "theoretical_loss": 3.33903346630299,
      "tokens_seen": 3122921472
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.40844166265447e-05,
      "loss": 0.0648,
      "theoretical_loss": 3.3390122748290763,
      "tokens_seen": 3123183616
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.4004172684962284e-05,
      "loss": 0.064,
      "theoretical_loss": 3.338991085631781,
      "tokens_seen": 3123445760
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.3923928743379876e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.3389698987106673,
      "tokens_seen": 3123707904
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.384368480179747e-05,
      "loss": 0.0679,
      "theoretical_loss": 3.338948714065302,
      "tokens_seen": 3123970048
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.3763440860215054e-05,
      "loss": 0.0659,
      "theoretical_loss": 3.3389275316952474,
      "tokens_seen": 3124232192
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.3683196918632646e-05,
      "loss": 0.066,
      "theoretical_loss": 3.3389063516000697,
      "tokens_seen": 3124494336
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.360295297705024e-05,
      "loss": 0.0677,
      "theoretical_loss": 3.3388851737793335,
      "tokens_seen": 3124756480
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.3522709035467824e-05,
      "loss": 0.0637,
      "theoretical_loss": 3.338863998232603,
      "tokens_seen": 3125018624
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.3442465093885416e-05,
      "loss": 0.0669,
      "theoretical_loss": 3.338842824959445,
      "tokens_seen": 3125280768
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.3362221152303e-05,
      "loss": 0.0652,
      "theoretical_loss": 3.338821653959423,
      "tokens_seen": 3125542912
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.328197721072059e-05,
      "loss": 0.0659,
      "theoretical_loss": 3.3388004852321025,
      "tokens_seen": 3125805056
    },
    {
      "epoch": 0.95,
      "objective/train/advantage_avg": 0.0002322028885828331,
      "objective/train/docs_used": 1135598,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3199049234390259,
      "objective/train/original_loss": 1.3199048042297363,
      "objective/train/theoretical_loss": 3.33877931877705,
      "objective/train/tokens_used": 3146527200,
      "objective/train/value_avg": -0.010223388671875,
      "objective/train/value_loss": 0.0003747669979929924,
      "objective/train/value_max": -2.9087066650390625e-05,
      "objective/train/value_min": -0.9521484375,
      "objective/train/value_reward_corr": 0.7227508381490007,
      "objective/train/value_std": 0.0213623046875,
      "objective/train/weight_avg": 1.0004043579101562,
      "objective/train/weighted_lm_loss": 1.3192325830459595,
      "objective/train/weights_max": 2.4473495483398438,
      "objective/train/weights_min": 0.36831334233283997,
      "theoretical_loss": 3.33877931877705,
      "tokens_seen": 3126067200
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.320173326913818e-05,
      "loss": 0.0645,
      "theoretical_loss": 3.33877931877705,
      "tokens_seen": 3126067200
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.312148932755577e-05,
      "loss": 0.0671,
      "theoretical_loss": 3.3387581545938305,
      "tokens_seen": 3126329344
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.304124538597336e-05,
      "loss": 0.0645,
      "theoretical_loss": 3.3387369926820094,
      "tokens_seen": 3126591488
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.296100144439095e-05,
      "loss": 0.0671,
      "theoretical_loss": 3.338715833041153,
      "tokens_seen": 3126853632
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.288075750280854e-05,
      "loss": 0.0637,
      "theoretical_loss": 3.338694675670827,
      "tokens_seen": 3127115776
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.280051356122613e-05,
      "loss": 0.0676,
      "theoretical_loss": 3.3386735205705977,
      "tokens_seen": 3127377920
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.272026961964372e-05,
      "loss": 0.0655,
      "theoretical_loss": 3.338652367740031,
      "tokens_seen": 3127640064
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.264002567806131e-05,
      "loss": 0.0673,
      "theoretical_loss": 3.338631217178693,
      "tokens_seen": 3127902208
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.25597817364789e-05,
      "loss": 0.0621,
      "theoretical_loss": 3.3386100688861506,
      "tokens_seen": 3128164352
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.247953779489649e-05,
      "loss": 0.0621,
      "theoretical_loss": 3.33858892286197,
      "tokens_seen": 3128426496
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.2399293853314075e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.338567779105718,
      "tokens_seen": 3128688640
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.231904991173166e-05,
      "loss": 0.0651,
      "theoretical_loss": 3.3385466376169615,
      "tokens_seen": 3128950784
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.223880597014925e-05,
      "loss": 0.0646,
      "theoretical_loss": 3.338525498395267,
      "tokens_seen": 3129212928
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.2158562028566845e-05,
      "loss": 0.0636,
      "theoretical_loss": 3.338504361440202,
      "tokens_seen": 3129475072
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.207831808698443e-05,
      "loss": 0.0636,
      "theoretical_loss": 3.3384832267513334,
      "tokens_seen": 3129737216
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.199807414540202e-05,
      "loss": 0.0682,
      "theoretical_loss": 3.338462094328228,
      "tokens_seen": 3129999360
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.1917830203819615e-05,
      "loss": 0.0659,
      "theoretical_loss": 3.338440964170454,
      "tokens_seen": 3130261504
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.18375862622372e-05,
      "loss": 0.0672,
      "theoretical_loss": 3.338419836277578,
      "tokens_seen": 3130523648
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.175734232065479e-05,
      "loss": 0.0653,
      "theoretical_loss": 3.338398710649168,
      "tokens_seen": 3130785792
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.1677098379072385e-05,
      "loss": 0.0631,
      "theoretical_loss": 3.3383775872847923,
      "tokens_seen": 3131047936
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.159685443748997e-05,
      "loss": 0.0672,
      "theoretical_loss": 3.3383564661840177,
      "tokens_seen": 3131310080
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.151661049590756e-05,
      "loss": 0.062,
      "theoretical_loss": 3.3383353473464132,
      "tokens_seen": 3131572224
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.143636655432515e-05,
      "loss": 0.0648,
      "theoretical_loss": 3.3383142307715463,
      "tokens_seen": 3131834368
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.1356122612742734e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.338293116458985,
      "tokens_seen": 3132096512
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.1275878671160326e-05,
      "loss": 0.0665,
      "theoretical_loss": 3.3382720044082976,
      "tokens_seen": 3132358656
    },
    {
      "epoch": 0.95,
      "objective/train/advantage_avg": 0.00031480283359996974,
      "objective/train/docs_used": 1137994,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3578956127166748,
      "objective/train/original_loss": 1.3578956127166748,
      "objective/train/theoretical_loss": 3.3382508946190534,
      "objective/train/tokens_used": 3153080800,
      "objective/train/value_avg": -0.007335662841796875,
      "objective/train/value_loss": 0.0002345583779970184,
      "objective/train/value_max": -3.7610530853271484e-05,
      "objective/train/value_min": -0.30712890625,
      "objective/train/value_reward_corr": 0.6553938720168582,
      "objective/train/value_std": 0.01401519775390625,
      "objective/train/weight_avg": 1.0004183053970337,
      "objective/train/weighted_lm_loss": 1.3580381870269775,
      "objective/train/weights_max": 1.2803038358688354,
      "objective/train/weights_min": 0.3710845410823822,
      "theoretical_loss": 3.3382508946190534,
      "tokens_seen": 3132620800
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.119563472957792e-05,
      "loss": 0.065,
      "theoretical_loss": 3.3382508946190534,
      "tokens_seen": 3132620800
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.1115390787995504e-05,
      "loss": 0.0641,
      "theoretical_loss": 3.33822978709082,
      "tokens_seen": 3132882944
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.1035146846413096e-05,
      "loss": 0.0637,
      "theoretical_loss": 3.3382086818231667,
      "tokens_seen": 3133145088
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.095490290483069e-05,
      "loss": 0.063,
      "theoretical_loss": 3.3381875788156616,
      "tokens_seen": 3133407232
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.0874658963248274e-05,
      "loss": 0.0611,
      "theoretical_loss": 3.3381664780678744,
      "tokens_seen": 3133669376
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.0794415021665867e-05,
      "loss": 0.065,
      "theoretical_loss": 3.338145379579373,
      "tokens_seen": 3133931520
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.071417108008346e-05,
      "loss": 0.064,
      "theoretical_loss": 3.338124283349728,
      "tokens_seen": 3134193664
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.0633927138501044e-05,
      "loss": 0.0664,
      "theoretical_loss": 3.338103189378508,
      "tokens_seen": 3134455808
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.055368319691863e-05,
      "loss": 0.0641,
      "theoretical_loss": 3.338082097665282,
      "tokens_seen": 3134717952
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.047343925533622e-05,
      "loss": 0.0658,
      "theoretical_loss": 3.33806100820962,
      "tokens_seen": 3134980096
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.039319531375381e-05,
      "loss": 0.0638,
      "theoretical_loss": 3.338039921011091,
      "tokens_seen": 3135242240
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.03129513721714e-05,
      "loss": 0.0638,
      "theoretical_loss": 3.3380188360692657,
      "tokens_seen": 3135504384
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.023270743058899e-05,
      "loss": 0.0672,
      "theoretical_loss": 3.3379977533837133,
      "tokens_seen": 3135766528
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.015246348900658e-05,
      "loss": 0.0691,
      "theoretical_loss": 3.3379766729540035,
      "tokens_seen": 3136028672
    },
    {
      "epoch": 0.95,
      "learning_rate": 5.007221954742417e-05,
      "loss": 0.0622,
      "theoretical_loss": 3.3379555947797073,
      "tokens_seen": 3136290816
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.999197560584176e-05,
      "loss": 0.0654,
      "theoretical_loss": 3.337934518860394,
      "tokens_seen": 3136552960
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.991173166425935e-05,
      "loss": 0.0653,
      "theoretical_loss": 3.3379134451956345,
      "tokens_seen": 3136815104
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.983148772267694e-05,
      "loss": 0.0624,
      "theoretical_loss": 3.337892373784999,
      "tokens_seen": 3137077248
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.975124378109453e-05,
      "loss": 0.0671,
      "theoretical_loss": 3.337871304628058,
      "tokens_seen": 3137339392
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.967099983951212e-05,
      "loss": 0.0685,
      "theoretical_loss": 3.3378502377243824,
      "tokens_seen": 3137601536
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.9590755897929703e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.3378291730735428,
      "tokens_seen": 3137863680
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.9510511956347296e-05,
      "loss": 0.0651,
      "theoretical_loss": 3.3378081106751103,
      "tokens_seen": 3138125824
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.943026801476488e-05,
      "loss": 0.0637,
      "theoretical_loss": 3.3377870505286555,
      "tokens_seen": 3138387968
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.9350024073182473e-05,
      "loss": 0.0655,
      "theoretical_loss": 3.3377659926337504,
      "tokens_seen": 3138650112
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.9269780131600066e-05,
      "loss": 0.0638,
      "theoretical_loss": 3.3377449369899654,
      "tokens_seen": 3138912256
    },
    {
      "epoch": 0.95,
      "objective/train/advantage_avg": 2.2450554752140306e-05,
      "objective/train/docs_used": 1140387,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2468316555023193,
      "objective/train/original_loss": 1.2468316555023193,
      "objective/train/theoretical_loss": 3.337723883596872,
      "objective/train/tokens_used": 3159634400,
      "objective/train/value_avg": -0.00865936279296875,
      "objective/train/value_loss": 0.0004632585623767227,
      "objective/train/value_max": -1.3530254364013672e-05,
      "objective/train/value_min": -0.7841796875,
      "objective/train/value_reward_corr": 0.6389569430252178,
      "objective/train/value_std": 0.0171966552734375,
      "objective/train/weight_avg": 1.0002206563949585,
      "objective/train/weighted_lm_loss": 1.2470567226409912,
      "objective/train/weights_max": 1.8875012397766113,
      "objective/train/weights_min": 0.3693002760410309,
      "theoretical_loss": 3.337723883596872,
      "tokens_seen": 3139174400
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.918953619001765e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.337723883596872,
      "tokens_seen": 3139174400
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.9109292248435244e-05,
      "loss": 0.0648,
      "theoretical_loss": 3.337702832454042,
      "tokens_seen": 3139436544
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.9029048306852836e-05,
      "loss": 0.0626,
      "theoretical_loss": 3.337681783561047,
      "tokens_seen": 3139698688
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.894880436527042e-05,
      "loss": 0.0658,
      "theoretical_loss": 3.3376607369174587,
      "tokens_seen": 3139960832
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.8868560423688014e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.337639692522849,
      "tokens_seen": 3140222976
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.8788316482105606e-05,
      "loss": 0.0673,
      "theoretical_loss": 3.3376186503767897,
      "tokens_seen": 3140485120
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.870807254052319e-05,
      "loss": 0.0657,
      "theoretical_loss": 3.3375976104788525,
      "tokens_seen": 3140747264
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.862782859894078e-05,
      "loss": 0.0652,
      "theoretical_loss": 3.3375765728286106,
      "tokens_seen": 3141009408
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.854758465735837e-05,
      "loss": 0.0656,
      "theoretical_loss": 3.3375555374256356,
      "tokens_seen": 3141271552
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.8467340715775955e-05,
      "loss": 0.0632,
      "theoretical_loss": 3.3375345042695,
      "tokens_seen": 3141533696
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.838709677419355e-05,
      "loss": 0.063,
      "theoretical_loss": 3.3375134733597767,
      "tokens_seen": 3141795840
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.830685283261114e-05,
      "loss": 0.0647,
      "theoretical_loss": 3.337492444696038,
      "tokens_seen": 3142057984
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.822660889102873e-05,
      "loss": 0.0666,
      "theoretical_loss": 3.3374714182778566,
      "tokens_seen": 3142320128
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.814636494944632e-05,
      "loss": 0.0667,
      "theoretical_loss": 3.337450394104806,
      "tokens_seen": 3142582272
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.806612100786391e-05,
      "loss": 0.0675,
      "theoretical_loss": 3.3374293721764587,
      "tokens_seen": 3142844416
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.79858770662815e-05,
      "loss": 0.0624,
      "theoretical_loss": 3.337408352492388,
      "tokens_seen": 3143106560
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.790563312469909e-05,
      "loss": 0.0655,
      "theoretical_loss": 3.3373873350521674,
      "tokens_seen": 3143368704
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.782538918311668e-05,
      "loss": 0.0639,
      "theoretical_loss": 3.3373663198553696,
      "tokens_seen": 3143630848
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.774514524153427e-05,
      "loss": 0.0664,
      "theoretical_loss": 3.3373453069015686,
      "tokens_seen": 3143892992
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.766490129995185e-05,
      "loss": 0.0656,
      "theoretical_loss": 3.337324296190338,
      "tokens_seen": 3144155136
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.758465735836944e-05,
      "loss": 0.0648,
      "theoretical_loss": 3.3373032877212516,
      "tokens_seen": 3144417280
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.7504413416787035e-05,
      "loss": 0.0645,
      "theoretical_loss": 3.3372822814938825,
      "tokens_seen": 3144679424
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.742416947520462e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.337261277507806,
      "tokens_seen": 3144941568
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.734392553362221e-05,
      "loss": 0.0646,
      "theoretical_loss": 3.3372402757625945,
      "tokens_seen": 3145203712
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.7263681592039805e-05,
      "loss": 0.0657,
      "theoretical_loss": 3.337219276257824,
      "tokens_seen": 3145465856
    },
    {
      "epoch": 0.95,
      "objective/train/advantage_avg": 0.0006178407347761095,
      "objective/train/docs_used": 1142692,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2589737176895142,
      "objective/train/original_loss": 1.258973479270935,
      "objective/train/theoretical_loss": 3.3371982789930668,
      "objective/train/tokens_used": 3166188000,
      "objective/train/value_avg": -0.006740570068359375,
      "objective/train/value_loss": 0.00018427084432914853,
      "objective/train/value_max": -2.9087066650390625e-05,
      "objective/train/value_min": -0.6171875,
      "objective/train/value_reward_corr": 0.7213768801644866,
      "objective/train/value_std": 0.014190673828125,
      "objective/train/weight_avg": 1.000701904296875,
      "objective/train/weighted_lm_loss": 1.2594910860061646,
      "objective/train/weights_max": 1.3403171300888062,
      "objective/train/weights_min": 0.36926716566085815,
      "theoretical_loss": 3.3371982789930668,
      "tokens_seen": 3145728000
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.718343765045739e-05,
      "loss": 0.0619,
      "theoretical_loss": 3.3371982789930668,
      "tokens_seen": 3145728000
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.710319370887498e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.3371772839678995,
      "tokens_seen": 3145990144
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.7022949767292575e-05,
      "loss": 0.0647,
      "theoretical_loss": 3.337156291181895,
      "tokens_seen": 3146252288
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.694270582571016e-05,
      "loss": 0.0653,
      "theoretical_loss": 3.3371353006346283,
      "tokens_seen": 3146514432
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.686246188412775e-05,
      "loss": 0.0622,
      "theoretical_loss": 3.3371143123256743,
      "tokens_seen": 3146776576
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.678221794254534e-05,
      "loss": 0.0658,
      "theoretical_loss": 3.3370933262546085,
      "tokens_seen": 3147038720
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.6701974000962924e-05,
      "loss": 0.0652,
      "theoretical_loss": 3.337072342421005,
      "tokens_seen": 3147300864
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.6621730059380516e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.3370513608244394,
      "tokens_seen": 3147563008
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.654148611779811e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.3370303814644866,
      "tokens_seen": 3147825152
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.6461242176215694e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.337009404340722,
      "tokens_seen": 3148087296
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.6380998234633286e-05,
      "loss": 0.0639,
      "theoretical_loss": 3.3369884294527217,
      "tokens_seen": 3148349440
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.630075429305088e-05,
      "loss": 0.0672,
      "theoretical_loss": 3.3369674568000605,
      "tokens_seen": 3148611584
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.6220510351468464e-05,
      "loss": 0.0647,
      "theoretical_loss": 3.336946486382314,
      "tokens_seen": 3148873728
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.6140266409886056e-05,
      "loss": 0.0608,
      "theoretical_loss": 3.336925518199059,
      "tokens_seen": 3149135872
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.606002246830365e-05,
      "loss": 0.0654,
      "theoretical_loss": 3.336904552249871,
      "tokens_seen": 3149398016
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.5979778526721234e-05,
      "loss": 0.0636,
      "theoretical_loss": 3.3368835885343255,
      "tokens_seen": 3149660160
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.5899534585138826e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.3368626270519988,
      "tokens_seen": 3149922304
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.581929064355641e-05,
      "loss": 0.065,
      "theoretical_loss": 3.3368416678024677,
      "tokens_seen": 3150184448
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.5739046701974e-05,
      "loss": 0.0655,
      "theoretical_loss": 3.3368207107853083,
      "tokens_seen": 3150446592
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.565880276039159e-05,
      "loss": 0.0638,
      "theoretical_loss": 3.3367997560000973,
      "tokens_seen": 3150708736
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.557855881880918e-05,
      "loss": 0.0654,
      "theoretical_loss": 3.3367788034464105,
      "tokens_seen": 3150970880
    },
    {
      "epoch": 0.95,
      "learning_rate": 4.549831487722677e-05,
      "loss": 0.0627,
      "theoretical_loss": 3.336757853123826,
      "tokens_seen": 3151233024
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.541807093564436e-05,
      "loss": 0.0659,
      "theoretical_loss": 3.3367369050319193,
      "tokens_seen": 3151495168
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.533782699406195e-05,
      "loss": 0.064,
      "theoretical_loss": 3.3367159591702684,
      "tokens_seen": 3151757312
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.525758305247954e-05,
      "loss": 0.0631,
      "theoretical_loss": 3.33669501553845,
      "tokens_seen": 3152019456
    },
    {
      "epoch": 0.96,
      "objective/train/advantage_avg": -0.00011186428309883922,
      "objective/train/docs_used": 1145159,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2790420055389404,
      "objective/train/original_loss": 1.2790418863296509,
      "objective/train/theoretical_loss": 3.336674074136041,
      "objective/train/tokens_used": 3172741600,
      "objective/train/value_avg": -0.00704193115234375,
      "objective/train/value_loss": 0.00023630002397112548,
      "objective/train/value_max": -2.390146255493164e-05,
      "objective/train/value_min": -0.98193359375,
      "objective/train/value_reward_corr": 0.8191802793083187,
      "objective/train/value_std": 0.02069091796875,
      "objective/train/weight_avg": 0.9999978542327881,
      "objective/train/weighted_lm_loss": 1.2786507606506348,
      "objective/train/weights_max": 1.631901741027832,
      "objective/train/weights_min": 0.373872309923172,
      "theoretical_loss": 3.336674074136041,
      "tokens_seen": 3152281600
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.517733911089713e-05,
      "loss": 0.065,
      "theoretical_loss": 3.336674074136041,
      "tokens_seen": 3152281600
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.509709516931472e-05,
      "loss": 0.0621,
      "theoretical_loss": 3.336653134962619,
      "tokens_seen": 3152543744
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.501685122773231e-05,
      "loss": 0.0656,
      "theoretical_loss": 3.3366321980177616,
      "tokens_seen": 3152805888
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.49366072861499e-05,
      "loss": 0.0623,
      "theoretical_loss": 3.3366112633010463,
      "tokens_seen": 3153068032
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.4856363344567486e-05,
      "loss": 0.0658,
      "theoretical_loss": 3.3365903308120504,
      "tokens_seen": 3153330176
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.477611940298507e-05,
      "loss": 0.0627,
      "theoretical_loss": 3.336569400550352,
      "tokens_seen": 3153592320
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.469587546140266e-05,
      "loss": 0.0608,
      "theoretical_loss": 3.3365484725155286,
      "tokens_seen": 3153854464
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.4615631519820256e-05,
      "loss": 0.063,
      "theoretical_loss": 3.3365275467071585,
      "tokens_seen": 3154116608
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.453538757823784e-05,
      "loss": 0.0615,
      "theoretical_loss": 3.33650662312482,
      "tokens_seen": 3154378752
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.445514363665543e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.3364857017680913,
      "tokens_seen": 3154640896
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.4374899695073026e-05,
      "loss": 0.0654,
      "theoretical_loss": 3.3364647826365506,
      "tokens_seen": 3154903040
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.429465575349061e-05,
      "loss": 0.0657,
      "theoretical_loss": 3.3364438657297764,
      "tokens_seen": 3155165184
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.4214411811908203e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.3364229510473473,
      "tokens_seen": 3155427328
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.4134167870325796e-05,
      "loss": 0.0647,
      "theoretical_loss": 3.3364020385888415,
      "tokens_seen": 3155689472
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.405392392874338e-05,
      "loss": 0.0658,
      "theoretical_loss": 3.3363811283538385,
      "tokens_seen": 3155951616
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.3973679987160974e-05,
      "loss": 0.0621,
      "theoretical_loss": 3.336360220341917,
      "tokens_seen": 3156213760
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.389343604557856e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.3363393145526565,
      "tokens_seen": 3156475904
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.3813192103996145e-05,
      "loss": 0.0645,
      "theoretical_loss": 3.336318410985635,
      "tokens_seen": 3156738048
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.373294816241374e-05,
      "loss": 0.0659,
      "theoretical_loss": 3.336297509640433,
      "tokens_seen": 3157000192
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.365270422083133e-05,
      "loss": 0.0639,
      "theoretical_loss": 3.336276610516629,
      "tokens_seen": 3157262336
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.3572460279248915e-05,
      "loss": 0.0657,
      "theoretical_loss": 3.3362557136138027,
      "tokens_seen": 3157524480
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.349221633766651e-05,
      "loss": 0.0629,
      "theoretical_loss": 3.3362348189315343,
      "tokens_seen": 3157786624
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.34119723960841e-05,
      "loss": 0.0618,
      "theoretical_loss": 3.3362139264694024,
      "tokens_seen": 3158048768
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.3331728454501685e-05,
      "loss": 0.0667,
      "theoretical_loss": 3.3361930362269883,
      "tokens_seen": 3158310912
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.325148451291928e-05,
      "loss": 0.0633,
      "theoretical_loss": 3.336172148203871,
      "tokens_seen": 3158573056
    },
    {
      "epoch": 0.96,
      "objective/train/advantage_avg": 0.0004656356177292764,
      "objective/train/docs_used": 1147634,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2245007753372192,
      "objective/train/original_loss": 1.2245006561279297,
      "objective/train/theoretical_loss": 3.33615126239963,
      "objective/train/tokens_used": 3179295200,
      "objective/train/value_avg": -0.006221771240234375,
      "objective/train/value_loss": 0.00020156294340267777,
      "objective/train/value_max": -1.1563301086425781e-05,
      "objective/train/value_min": -0.97802734375,
      "objective/train/value_reward_corr": 0.617885544008769,
      "objective/train/value_std": 0.0126495361328125,
      "objective/train/weight_avg": 1.0005531311035156,
      "objective/train/weighted_lm_loss": 1.2248764038085938,
      "objective/train/weights_max": 1.505949854850769,
      "objective/train/weights_min": 0.3681730329990387,
      "theoretical_loss": 3.33615126239963,
      "tokens_seen": 3158835200
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.317124057133687e-05,
      "loss": 0.0664,
      "theoretical_loss": 3.33615126239963,
      "tokens_seen": 3158835200
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.3090996629754455e-05,
      "loss": 0.0644,
      "theoretical_loss": 3.336130378813847,
      "tokens_seen": 3159097344
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.301075268817205e-05,
      "loss": 0.0628,
      "theoretical_loss": 3.3361094974461016,
      "tokens_seen": 3159359488
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.293050874658963e-05,
      "loss": 0.0661,
      "theoretical_loss": 3.3360886182959737,
      "tokens_seen": 3159621632
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.285026480500722e-05,
      "loss": 0.0648,
      "theoretical_loss": 3.3360677413630446,
      "tokens_seen": 3159883776
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.277002086342481e-05,
      "loss": 0.0681,
      "theoretical_loss": 3.3360468666468943,
      "tokens_seen": 3160145920
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.26897769218424e-05,
      "loss": 0.063,
      "theoretical_loss": 3.336025994147104,
      "tokens_seen": 3160408064
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.260953298025999e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.3360051238632544,
      "tokens_seen": 3160670208
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.252928903867758e-05,
      "loss": 0.0648,
      "theoretical_loss": 3.3359842557949264,
      "tokens_seen": 3160932352
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.244904509709517e-05,
      "loss": 0.0618,
      "theoretical_loss": 3.3359633899417016,
      "tokens_seen": 3161194496
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.236880115551276e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.335942526303161,
      "tokens_seen": 3161456640
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.228855721393035e-05,
      "loss": 0.0615,
      "theoretical_loss": 3.335921664878885,
      "tokens_seen": 3161718784
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.220831327234794e-05,
      "loss": 0.0654,
      "theoretical_loss": 3.3359008056684565,
      "tokens_seen": 3161980928
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.212806933076553e-05,
      "loss": 0.0619,
      "theoretical_loss": 3.335879948671456,
      "tokens_seen": 3162243072
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.2047825389183114e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.3358590938874655,
      "tokens_seen": 3162505216
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.1967581447600706e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.335838241316067,
      "tokens_seen": 3162767360
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.188733750601829e-05,
      "loss": 0.0639,
      "theoretical_loss": 3.335817390956842,
      "tokens_seen": 3163029504
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.1807093564435884e-05,
      "loss": 0.0613,
      "theoretical_loss": 3.3357965428093728,
      "tokens_seen": 3163291648
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.1726849622853476e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.3357756968732413,
      "tokens_seen": 3163553792
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.164660568127106e-05,
      "loss": 0.0659,
      "theoretical_loss": 3.33575485314803,
      "tokens_seen": 3163815936
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.1566361739688654e-05,
      "loss": 0.065,
      "theoretical_loss": 3.335734011633321,
      "tokens_seen": 3164078080
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.1486117798106246e-05,
      "loss": 0.0661,
      "theoretical_loss": 3.3357131723286972,
      "tokens_seen": 3164340224
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.140587385652383e-05,
      "loss": 0.0639,
      "theoretical_loss": 3.33569233523374,
      "tokens_seen": 3164602368
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.1325629914941424e-05,
      "loss": 0.0661,
      "theoretical_loss": 3.3356715003480337,
      "tokens_seen": 3164864512
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.1245385973359016e-05,
      "loss": 0.065,
      "theoretical_loss": 3.3356506676711604,
      "tokens_seen": 3165126656
    },
    {
      "epoch": 0.96,
      "objective/train/advantage_avg": -0.00016843386401887983,
      "objective/train/docs_used": 1150157,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2679888010025024,
      "objective/train/original_loss": 1.267988920211792,
      "objective/train/theoretical_loss": 3.3356298372027027,
      "objective/train/tokens_used": 3185848800,
      "objective/train/value_avg": -0.006591796875,
      "objective/train/value_loss": 0.00019000943575520068,
      "objective/train/value_max": -3.11732292175293e-05,
      "objective/train/value_min": -0.6064453125,
      "objective/train/value_reward_corr": 0.6557403656697849,
      "objective/train/value_std": 0.012451171875,
      "objective/train/weight_avg": 0.9999187588691711,
      "objective/train/weighted_lm_loss": 1.2679390907287598,
      "objective/train/weights_max": 1.687270164489746,
      "objective/train/weights_min": 0.3683769106864929,
      "theoretical_loss": 3.3356298372027027,
      "tokens_seen": 3165388800
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.11651420317766e-05,
      "loss": 0.0659,
      "theoretical_loss": 3.3356298372027027,
      "tokens_seen": 3165388800
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.108489809019419e-05,
      "loss": 0.0629,
      "theoretical_loss": 3.335609008942244,
      "tokens_seen": 3165650944
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.100465414861178e-05,
      "loss": 0.0623,
      "theoretical_loss": 3.335588182889367,
      "tokens_seen": 3165913088
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.0924410207029365e-05,
      "loss": 0.0623,
      "theoretical_loss": 3.3355673590436554,
      "tokens_seen": 3166175232
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.084416626544696e-05,
      "loss": 0.0615,
      "theoretical_loss": 3.3355465374046926,
      "tokens_seen": 3166437376
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.076392232386455e-05,
      "loss": 0.066,
      "theoretical_loss": 3.335525717972062,
      "tokens_seen": 3166699520
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.0683678382282135e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.335504900745347,
      "tokens_seen": 3166961664
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.060343444069973e-05,
      "loss": 0.0656,
      "theoretical_loss": 3.335484085724132,
      "tokens_seen": 3167223808
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.052319049911732e-05,
      "loss": 0.062,
      "theoretical_loss": 3.335463272907999,
      "tokens_seen": 3167485952
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.0442946557534905e-05,
      "loss": 0.0653,
      "theoretical_loss": 3.335442462296534,
      "tokens_seen": 3167748096
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.03627026159525e-05,
      "loss": 0.0629,
      "theoretical_loss": 3.3354216538893207,
      "tokens_seen": 3168010240
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.028245867437009e-05,
      "loss": 0.0671,
      "theoretical_loss": 3.335400847685942,
      "tokens_seen": 3168272384
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.0202214732787675e-05,
      "loss": 0.0656,
      "theoretical_loss": 3.335380043685983,
      "tokens_seen": 3168534528
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.012197079120526e-05,
      "loss": 0.0647,
      "theoretical_loss": 3.3353592418890283,
      "tokens_seen": 3168796672
    },
    {
      "epoch": 0.96,
      "learning_rate": 4.004172684962285e-05,
      "loss": 0.0617,
      "theoretical_loss": 3.3353384422946624,
      "tokens_seen": 3169058816
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.996148290804044e-05,
      "loss": 0.0631,
      "theoretical_loss": 3.335317644902469,
      "tokens_seen": 3169320960
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.988123896645803e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.335296849712034,
      "tokens_seen": 3169583104
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.980099502487562e-05,
      "loss": 0.0641,
      "theoretical_loss": 3.335276056722942,
      "tokens_seen": 3169845248
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.972075108329321e-05,
      "loss": 0.0651,
      "theoretical_loss": 3.3352552659347774,
      "tokens_seen": 3170107392
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.96405071417108e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.335234477347125,
      "tokens_seen": 3170369536
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.956026320012839e-05,
      "loss": 0.0615,
      "theoretical_loss": 3.3352136909595713,
      "tokens_seen": 3170631680
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.9480019258545986e-05,
      "loss": 0.0638,
      "theoretical_loss": 3.3351929067717005,
      "tokens_seen": 3170893824
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.939977531696357e-05,
      "loss": 0.0641,
      "theoretical_loss": 3.335172124783098,
      "tokens_seen": 3171155968
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.931953137538116e-05,
      "loss": 0.0655,
      "theoretical_loss": 3.33515134499335,
      "tokens_seen": 3171418112
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.9239287433798756e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.3351305674020417,
      "tokens_seen": 3171680256
    },
    {
      "epoch": 0.96,
      "objective/train/advantage_avg": 0.0003502919862512499,
      "objective/train/docs_used": 1152731,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2793630361557007,
      "objective/train/original_loss": 1.2793629169464111,
      "objective/train/theoretical_loss": 3.3351097920087587,
      "objective/train/tokens_used": 3192402400,
      "objective/train/value_avg": -0.00690460205078125,
      "objective/train/value_loss": 0.00017006472626235336,
      "objective/train/value_max": -2.9087066650390625e-05,
      "objective/train/value_min": -0.81591796875,
      "objective/train/value_reward_corr": 0.7036529803661943,
      "objective/train/value_std": 0.01306915283203125,
      "objective/train/weight_avg": 1.0004277229309082,
      "objective/train/weighted_lm_loss": 1.2795383930206299,
      "objective/train/weights_max": 1.226123332977295,
      "objective/train/weights_min": 0.3684985041618347,
      "theoretical_loss": 3.3351097920087587,
      "tokens_seen": 3171942400
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.9159043492216334e-05,
      "loss": 0.0648,
      "theoretical_loss": 3.3351097920087587,
      "tokens_seen": 3171942400
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.907879955063393e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.335089018813087,
      "tokens_seen": 3172204544
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.899855560905152e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.3350682478146125,
      "tokens_seen": 3172466688
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.8918311667469104e-05,
      "loss": 0.0658,
      "theoretical_loss": 3.3350474790129216,
      "tokens_seen": 3172728832
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.88380677258867e-05,
      "loss": 0.062,
      "theoretical_loss": 3.3350267124076,
      "tokens_seen": 3172990976
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.875782378430429e-05,
      "loss": 0.0668,
      "theoretical_loss": 3.335005947998235,
      "tokens_seen": 3173253120
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.8677579842721875e-05,
      "loss": 0.0641,
      "theoretical_loss": 3.3349851857844115,
      "tokens_seen": 3173515264
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.859733590113947e-05,
      "loss": 0.0629,
      "theoretical_loss": 3.3349644257657167,
      "tokens_seen": 3173777408
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.851709195955706e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.3349436679417375,
      "tokens_seen": 3174039552
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.8436848017974645e-05,
      "loss": 0.063,
      "theoretical_loss": 3.3349229123120603,
      "tokens_seen": 3174301696
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.835660407639224e-05,
      "loss": 0.0661,
      "theoretical_loss": 3.334902158876272,
      "tokens_seen": 3174563840
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.827636013480982e-05,
      "loss": 0.0644,
      "theoretical_loss": 3.33488140763396,
      "tokens_seen": 3174825984
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.819611619322741e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.334860658584711,
      "tokens_seen": 3175088128
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.8115872251645e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.3348399117281122,
      "tokens_seen": 3175350272
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.803562831006259e-05,
      "loss": 0.0633,
      "theoretical_loss": 3.3348191670637513,
      "tokens_seen": 3175612416
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.795538436848018e-05,
      "loss": 0.066,
      "theoretical_loss": 3.334798424591215,
      "tokens_seen": 3175874560
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.787514042689777e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.334777684310091,
      "tokens_seen": 3176136704
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.779489648531536e-05,
      "loss": 0.0666,
      "theoretical_loss": 3.3347569462199678,
      "tokens_seen": 3176398848
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.771465254373295e-05,
      "loss": 0.0669,
      "theoretical_loss": 3.3347362103204317,
      "tokens_seen": 3176660992
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.763440860215054e-05,
      "loss": 0.0632,
      "theoretical_loss": 3.3347154766110716,
      "tokens_seen": 3176923136
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.755416466056813e-05,
      "loss": 0.0631,
      "theoretical_loss": 3.3346947450914755,
      "tokens_seen": 3177185280
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.747392071898572e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.334674015761231,
      "tokens_seen": 3177447424
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.739367677740331e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.3346532886199265,
      "tokens_seen": 3177709568
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.7313432835820896e-05,
      "loss": 0.0636,
      "theoretical_loss": 3.3346325636671503,
      "tokens_seen": 3177971712
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.723318889423848e-05,
      "loss": 0.0634,
      "theoretical_loss": 3.334611840902491,
      "tokens_seen": 3178233856
    },
    {
      "epoch": 0.96,
      "objective/train/advantage_avg": 2.5288290999014862e-05,
      "objective/train/docs_used": 1155119,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3060029745101929,
      "objective/train/original_loss": 1.3060028553009033,
      "objective/train/theoretical_loss": 3.3345911203255367,
      "objective/train/tokens_used": 3198956000,
      "objective/train/value_avg": -0.00821685791015625,
      "objective/train/value_loss": 0.00039423233829438686,
      "objective/train/value_max": -2.092123031616211e-05,
      "objective/train/value_min": -0.9677734375,
      "objective/train/value_reward_corr": 0.6660958908671027,
      "objective/train/value_std": 0.0167999267578125,
      "objective/train/weight_avg": 1.0001922845840454,
      "objective/train/weighted_lm_loss": 1.3057191371917725,
      "objective/train/weights_max": 1.4924074411392212,
      "objective/train/weights_min": 0.3686675727367401,
      "theoretical_loss": 3.3345911203255367,
      "tokens_seen": 3178496000
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.7152944952656074e-05,
      "loss": 0.0628,
      "theoretical_loss": 3.3345911203255367,
      "tokens_seen": 3178496000
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.7072701011073666e-05,
      "loss": 0.065,
      "theoretical_loss": 3.334570401935877,
      "tokens_seen": 3178758144
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.699245706949125e-05,
      "loss": 0.0641,
      "theoretical_loss": 3.334549685733099,
      "tokens_seen": 3179020288
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.6912213127908844e-05,
      "loss": 0.0639,
      "theoretical_loss": 3.334528971716793,
      "tokens_seen": 3179282432
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.6831969186326436e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.3345082598865474,
      "tokens_seen": 3179544576
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.675172524474402e-05,
      "loss": 0.0648,
      "theoretical_loss": 3.3344875502419513,
      "tokens_seen": 3179806720
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.6671481303161614e-05,
      "loss": 0.0634,
      "theoretical_loss": 3.3344668427825943,
      "tokens_seen": 3180068864
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.6591237361579206e-05,
      "loss": 0.0675,
      "theoretical_loss": 3.334446137508065,
      "tokens_seen": 3180331008
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.651099341999679e-05,
      "loss": 0.0638,
      "theoretical_loss": 3.3344254344179536,
      "tokens_seen": 3180593152
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.6430749478414384e-05,
      "loss": 0.0651,
      "theoretical_loss": 3.334404733511849,
      "tokens_seen": 3180855296
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.635050553683197e-05,
      "loss": 0.0654,
      "theoretical_loss": 3.334384034789341,
      "tokens_seen": 3181117440
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.6270261595249555e-05,
      "loss": 0.0638,
      "theoretical_loss": 3.33436333825002,
      "tokens_seen": 3181379584
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.619001765366715e-05,
      "loss": 0.0685,
      "theoretical_loss": 3.334342643893475,
      "tokens_seen": 3181641728
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.610977371208474e-05,
      "loss": 0.0656,
      "theoretical_loss": 3.3343219517192964,
      "tokens_seen": 3181903872
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.6029529770502325e-05,
      "loss": 0.0659,
      "theoretical_loss": 3.334301261727074,
      "tokens_seen": 3182166016
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.594928582891992e-05,
      "loss": 0.0664,
      "theoretical_loss": 3.3342805739163985,
      "tokens_seen": 3182428160
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.586904188733751e-05,
      "loss": 0.0644,
      "theoretical_loss": 3.3342598882868595,
      "tokens_seen": 3182690304
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.5788797945755095e-05,
      "loss": 0.0662,
      "theoretical_loss": 3.334239204838048,
      "tokens_seen": 3182952448
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.570855400417269e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.3342185235695543,
      "tokens_seen": 3183214592
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.562831006259028e-05,
      "loss": 0.067,
      "theoretical_loss": 3.334197844480969,
      "tokens_seen": 3183476736
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.5548066121007865e-05,
      "loss": 0.0673,
      "theoretical_loss": 3.334177167571883,
      "tokens_seen": 3183738880
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.546782217942546e-05,
      "loss": 0.0641,
      "theoretical_loss": 3.334156492841887,
      "tokens_seen": 3184001024
    },
    {
      "epoch": 0.96,
      "learning_rate": 3.538757823784304e-05,
      "loss": 0.0665,
      "theoretical_loss": 3.3341358202905718,
      "tokens_seen": 3184263168
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.530733429626063e-05,
      "loss": 0.0652,
      "theoretical_loss": 3.334115149917529,
      "tokens_seen": 3184525312
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.522709035467822e-05,
      "loss": 0.0658,
      "theoretical_loss": 3.3340944817223495,
      "tokens_seen": 3184787456
    },
    {
      "epoch": 0.97,
      "objective/train/advantage_avg": 0.0005753776640631258,
      "objective/train/docs_used": 1157476,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.251091718673706,
      "objective/train/original_loss": 1.251091718673706,
      "objective/train/theoretical_loss": 3.334073815704625,
      "objective/train/tokens_used": 3205509600,
      "objective/train/value_avg": -0.00763702392578125,
      "objective/train/value_loss": 0.0001186743684229441,
      "objective/train/value_max": -2.4318695068359375e-05,
      "objective/train/value_min": -0.250732421875,
      "objective/train/value_reward_corr": 0.8103845478100814,
      "objective/train/value_std": 0.0146942138671875,
      "objective/train/weight_avg": 1.0006309747695923,
      "objective/train/weighted_lm_loss": 1.2523066997528076,
      "objective/train/weights_max": 1.284966230392456,
      "objective/train/weights_min": 0.37953981757164,
      "theoretical_loss": 3.334073815704625,
      "tokens_seen": 3185049600
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.514684641309581e-05,
      "loss": 0.0626,
      "theoretical_loss": 3.334073815704625,
      "tokens_seen": 3185049600
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.50666024715134e-05,
      "loss": 0.0669,
      "theoretical_loss": 3.3340531518639462,
      "tokens_seen": 3185311744
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.498635852993099e-05,
      "loss": 0.0671,
      "theoretical_loss": 3.3340324901999048,
      "tokens_seen": 3185573888
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.490611458834858e-05,
      "loss": 0.0632,
      "theoretical_loss": 3.3340118307120927,
      "tokens_seen": 3185836032
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.482587064676617e-05,
      "loss": 0.066,
      "theoretical_loss": 3.3339911734001015,
      "tokens_seen": 3186098176
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.474562670518376e-05,
      "loss": 0.0664,
      "theoretical_loss": 3.3339705182635235,
      "tokens_seen": 3186360320
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.466538276360135e-05,
      "loss": 0.0667,
      "theoretical_loss": 3.3339498653019497,
      "tokens_seen": 3186622464
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.458513882201894e-05,
      "loss": 0.0658,
      "theoretical_loss": 3.3339292145149733,
      "tokens_seen": 3186884608
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.450489488043653e-05,
      "loss": 0.065,
      "theoretical_loss": 3.3339085659021857,
      "tokens_seen": 3187146752
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.4424650938854116e-05,
      "loss": 0.0627,
      "theoretical_loss": 3.333887919463179,
      "tokens_seen": 3187408896
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.43444069972717e-05,
      "loss": 0.0645,
      "theoretical_loss": 3.333867275197546,
      "tokens_seen": 3187671040
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.4264163055689294e-05,
      "loss": 0.0652,
      "theoretical_loss": 3.33384663310488,
      "tokens_seen": 3187933184
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.4183919114106887e-05,
      "loss": 0.0637,
      "theoretical_loss": 3.3338259931847727,
      "tokens_seen": 3188195328
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.410367517252447e-05,
      "loss": 0.0655,
      "theoretical_loss": 3.3338053554368163,
      "tokens_seen": 3188457472
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.4023431230942064e-05,
      "loss": 0.0675,
      "theoretical_loss": 3.333784719860605,
      "tokens_seen": 3188719616
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.394318728935966e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.333764086455731,
      "tokens_seen": 3188981760
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.386294334777724e-05,
      "loss": 0.0622,
      "theoretical_loss": 3.333743455221787,
      "tokens_seen": 3189243904
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.3782699406194834e-05,
      "loss": 0.0631,
      "theoretical_loss": 3.333722826158367,
      "tokens_seen": 3189506048
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.370245546461243e-05,
      "loss": 0.064,
      "theoretical_loss": 3.3337021992650637,
      "tokens_seen": 3189768192
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.362221152303001e-05,
      "loss": 0.0626,
      "theoretical_loss": 3.3336815745414707,
      "tokens_seen": 3190030336
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.35419675814476e-05,
      "loss": 0.0644,
      "theoretical_loss": 3.3336609519871816,
      "tokens_seen": 3190292480
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.346172363986519e-05,
      "loss": 0.0636,
      "theoretical_loss": 3.33364033160179,
      "tokens_seen": 3190554624
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.3381479698282775e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.3336197133848895,
      "tokens_seen": 3190816768
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.330123575670037e-05,
      "loss": 0.0651,
      "theoretical_loss": 3.333599097336074,
      "tokens_seen": 3191078912
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.322099181511796e-05,
      "loss": 0.0653,
      "theoretical_loss": 3.3335784834549367,
      "tokens_seen": 3191341056
    },
    {
      "epoch": 0.97,
      "objective/train/advantage_avg": -0.0003126697556581348,
      "objective/train/docs_used": 1159894,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2759596109390259,
      "objective/train/original_loss": 1.2759596109390259,
      "objective/train/theoretical_loss": 3.333557871741073,
      "objective/train/tokens_used": 3212063200,
      "objective/train/value_avg": -0.007526397705078125,
      "objective/train/value_loss": 0.0002957968390546739,
      "objective/train/value_max": -1.3649463653564453e-05,
      "objective/train/value_min": -0.6982421875,
      "objective/train/value_reward_corr": 0.7892284286544861,
      "objective/train/value_std": 0.0211029052734375,
      "objective/train/weight_avg": 0.9998196959495544,
      "objective/train/weighted_lm_loss": 1.2751232385635376,
      "objective/train/weights_max": 1.3489364385604858,
      "objective/train/weights_min": 0.3683748245239258,
      "theoretical_loss": 3.333557871741073,
      "tokens_seen": 3191603200
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.3140747873535546e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.333557871741073,
      "tokens_seen": 3191603200
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.306050393195314e-05,
      "loss": 0.0629,
      "theoretical_loss": 3.3335372621940764,
      "tokens_seen": 3191865344
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.298025999037073e-05,
      "loss": 0.0637,
      "theoretical_loss": 3.333516654813541,
      "tokens_seen": 3192127488
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.2900016048788316e-05,
      "loss": 0.064,
      "theoretical_loss": 3.3334960495990615,
      "tokens_seen": 3192389632
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.281977210720591e-05,
      "loss": 0.0638,
      "theoretical_loss": 3.3334754465502323,
      "tokens_seen": 3192651776
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.27395281656235e-05,
      "loss": 0.0657,
      "theoretical_loss": 3.3334548456666475,
      "tokens_seen": 3192913920
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.2659284224041086e-05,
      "loss": 0.0651,
      "theoretical_loss": 3.3334342469479026,
      "tokens_seen": 3193176064
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.257904028245867e-05,
      "loss": 0.0647,
      "theoretical_loss": 3.3334136503935916,
      "tokens_seen": 3193438208
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.2498796340876264e-05,
      "loss": 0.0659,
      "theoretical_loss": 3.3333930560033105,
      "tokens_seen": 3193700352
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.241855239929385e-05,
      "loss": 0.0644,
      "theoretical_loss": 3.3333724637766533,
      "tokens_seen": 3193962496
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.233830845771144e-05,
      "loss": 0.0683,
      "theoretical_loss": 3.333351873713216,
      "tokens_seen": 3194224640
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.2258064516129034e-05,
      "loss": 0.0647,
      "theoretical_loss": 3.333331285812593,
      "tokens_seen": 3194486784
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.217782057454662e-05,
      "loss": 0.0626,
      "theoretical_loss": 3.33331070007438,
      "tokens_seen": 3194748928
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.209757663296421e-05,
      "loss": 0.0639,
      "theoretical_loss": 3.3332901164981728,
      "tokens_seen": 3195011072
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.2017332691381804e-05,
      "loss": 0.0628,
      "theoretical_loss": 3.3332695350835664,
      "tokens_seen": 3195273216
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.193708874979939e-05,
      "loss": 0.0646,
      "theoretical_loss": 3.333248955830157,
      "tokens_seen": 3195535360
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.185684480821698e-05,
      "loss": 0.0646,
      "theoretical_loss": 3.33322837873754,
      "tokens_seen": 3195797504
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.1776600866634574e-05,
      "loss": 0.0636,
      "theoretical_loss": 3.3332078038053115,
      "tokens_seen": 3196059648
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.169635692505216e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.3331872310330675,
      "tokens_seen": 3196321792
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.1616112983469745e-05,
      "loss": 0.0618,
      "theoretical_loss": 3.333166660420404,
      "tokens_seen": 3196583936
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.153586904188734e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.333146091966918,
      "tokens_seen": 3196846080
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.145562510030492e-05,
      "loss": 0.0634,
      "theoretical_loss": 3.3331255256722043,
      "tokens_seen": 3197108224
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.1375381158722515e-05,
      "loss": 0.0633,
      "theoretical_loss": 3.3331049615358608,
      "tokens_seen": 3197370368
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.129513721714011e-05,
      "loss": 0.065,
      "theoretical_loss": 3.333084399557483,
      "tokens_seen": 3197632512
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.121489327555769e-05,
      "loss": 0.0626,
      "theoretical_loss": 3.3330638397366683,
      "tokens_seen": 3197894656
    },
    {
      "epoch": 0.97,
      "objective/train/advantage_avg": 0.00034200731897726655,
      "objective/train/docs_used": 1162228,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.1946808099746704,
      "objective/train/original_loss": 1.1946806907653809,
      "objective/train/theoretical_loss": 3.3330432820730134,
      "objective/train/tokens_used": 3218616800,
      "objective/train/value_avg": -0.007663726806640625,
      "objective/train/value_loss": 0.00036682773497886956,
      "objective/train/value_max": -2.753734588623047e-05,
      "objective/train/value_min": -0.97705078125,
      "objective/train/value_reward_corr": 0.6712799686665404,
      "objective/train/value_std": 0.01715087890625,
      "objective/train/weight_avg": 1.000498652458191,
      "objective/train/weighted_lm_loss": 1.1946550607681274,
      "objective/train/weights_max": 1.920973539352417,
      "objective/train/weights_min": 0.2709263861179352,
      "theoretical_loss": 3.3330432820730134,
      "tokens_seen": 3198156800
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.1134649333975285e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.3330432820730134,
      "tokens_seen": 3198156800
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.105440539239288e-05,
      "loss": 0.0651,
      "theoretical_loss": 3.3330227265661145,
      "tokens_seen": 3198418944
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.097416145081046e-05,
      "loss": 0.0647,
      "theoretical_loss": 3.3330021732155695,
      "tokens_seen": 3198681088
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.0893917509228055e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.332981622020975,
      "tokens_seen": 3198943232
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.081367356764564e-05,
      "loss": 0.0623,
      "theoretical_loss": 3.332961072981928,
      "tokens_seen": 3199205376
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.073342962606323e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.3329405260980267,
      "tokens_seen": 3199467520
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.0653185684480825e-05,
      "loss": 0.0631,
      "theoretical_loss": 3.3329199813688675,
      "tokens_seen": 3199729664
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.057294174289841e-05,
      "loss": 0.0653,
      "theoretical_loss": 3.3328994387940485,
      "tokens_seen": 3199991808
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.0492697801316003e-05,
      "loss": 0.066,
      "theoretical_loss": 3.332878898373167,
      "tokens_seen": 3200253952
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.041245385973359e-05,
      "loss": 0.0636,
      "theoretical_loss": 3.332858360105821,
      "tokens_seen": 3200516096
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.033220991815118e-05,
      "loss": 0.0631,
      "theoretical_loss": 3.3328378239916083,
      "tokens_seen": 3200778240
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.025196597656877e-05,
      "loss": 0.0648,
      "theoretical_loss": 3.3328172900301265,
      "tokens_seen": 3201040384
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.017172203498636e-05,
      "loss": 0.0626,
      "theoretical_loss": 3.3327967582209745,
      "tokens_seen": 3201302528
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.009147809340395e-05,
      "loss": 0.0637,
      "theoretical_loss": 3.33277622856375,
      "tokens_seen": 3201564672
    },
    {
      "epoch": 0.97,
      "learning_rate": 3.001123415182154e-05,
      "loss": 0.0636,
      "theoretical_loss": 3.332755701058051,
      "tokens_seen": 3201826816
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.9930990210239125e-05,
      "loss": 0.0646,
      "theoretical_loss": 3.332735175703476,
      "tokens_seen": 3202088960
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.9850746268656717e-05,
      "loss": 0.0637,
      "theoretical_loss": 3.3327146524996243,
      "tokens_seen": 3202351104
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.9770502327074306e-05,
      "loss": 0.0628,
      "theoretical_loss": 3.332694131446093,
      "tokens_seen": 3202613248
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.9690258385491895e-05,
      "loss": 0.0618,
      "theoretical_loss": 3.3326736125424827,
      "tokens_seen": 3202875392
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.9610014443909487e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.3326530957883906,
      "tokens_seen": 3203137536
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.9529770502327073e-05,
      "loss": 0.0633,
      "theoretical_loss": 3.3326325811834163,
      "tokens_seen": 3203399680
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.9449526560744662e-05,
      "loss": 0.064,
      "theoretical_loss": 3.332612068727159,
      "tokens_seen": 3203661824
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.9369282619162254e-05,
      "loss": 0.0651,
      "theoretical_loss": 3.3325915584192174,
      "tokens_seen": 3203923968
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.9289038677579843e-05,
      "loss": 0.0637,
      "theoretical_loss": 3.332571050259191,
      "tokens_seen": 3204186112
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.9208794735997432e-05,
      "loss": 0.0596,
      "theoretical_loss": 3.3325505442466796,
      "tokens_seen": 3204448256
    },
    {
      "epoch": 0.97,
      "objective/train/advantage_avg": -0.00047033251030370593,
      "objective/train/docs_used": 1164453,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.324925184249878,
      "objective/train/original_loss": 1.3249250650405884,
      "objective/train/theoretical_loss": 3.3325300403812816,
      "objective/train/tokens_used": 3225170400,
      "objective/train/value_avg": -0.006519317626953125,
      "objective/train/value_loss": 0.00018281102529726923,
      "objective/train/value_max": -2.7120113372802734e-05,
      "objective/train/value_min": -0.314208984375,
      "objective/train/value_reward_corr": 0.8121445704998986,
      "objective/train/value_std": 0.01412200927734375,
      "objective/train/weight_avg": 0.9996156096458435,
      "objective/train/weighted_lm_loss": 1.3245081901550293,
      "objective/train/weights_max": 1.179583191871643,
      "objective/train/weights_min": 0.3681163489818573,
      "theoretical_loss": 3.3325300403812816,
      "tokens_seen": 3204710400
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.9128550794415024e-05,
      "loss": 0.0657,
      "theoretical_loss": 3.3325300403812816,
      "tokens_seen": 3204710400
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.904830685283261e-05,
      "loss": 0.0618,
      "theoretical_loss": 3.332509538662598,
      "tokens_seen": 3204972544
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.89680629112502e-05,
      "loss": 0.0601,
      "theoretical_loss": 3.332489039090227,
      "tokens_seen": 3205234688
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.888781896966779e-05,
      "loss": 0.062,
      "theoretical_loss": 3.3324685416637694,
      "tokens_seen": 3205496832
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.880757502808538e-05,
      "loss": 0.0633,
      "theoretical_loss": 3.3324480463828245,
      "tokens_seen": 3205758976
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.872733108650297e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.3324275532469927,
      "tokens_seen": 3206021120
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.864708714492056e-05,
      "loss": 0.0629,
      "theoretical_loss": 3.332407062255874,
      "tokens_seen": 3206283264
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.8566843203338146e-05,
      "loss": 0.065,
      "theoretical_loss": 3.3323865734090687,
      "tokens_seen": 3206545408
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.8486599261755735e-05,
      "loss": 0.0625,
      "theoretical_loss": 3.332366086706177,
      "tokens_seen": 3206807552
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.8406355320173328e-05,
      "loss": 0.0653,
      "theoretical_loss": 3.3323456021467996,
      "tokens_seen": 3207069696
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.8326111378590917e-05,
      "loss": 0.0622,
      "theoretical_loss": 3.3323251197305366,
      "tokens_seen": 3207331840
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.824586743700851e-05,
      "loss": 0.0645,
      "theoretical_loss": 3.3323046394569893,
      "tokens_seen": 3207593984
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.8165623495426098e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.3322841613257577,
      "tokens_seen": 3207856128
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.8085379553843683e-05,
      "loss": 0.0629,
      "theoretical_loss": 3.332263685336443,
      "tokens_seen": 3208118272
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.8005135612261276e-05,
      "loss": 0.0659,
      "theoretical_loss": 3.3322432114886458,
      "tokens_seen": 3208380416
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.7924891670678864e-05,
      "loss": 0.0627,
      "theoretical_loss": 3.332222739781968,
      "tokens_seen": 3208642560
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.7844647729096453e-05,
      "loss": 0.0638,
      "theoretical_loss": 3.33220227021601,
      "tokens_seen": 3208904704
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.7764403787514046e-05,
      "loss": 0.0641,
      "theoretical_loss": 3.332181802790374,
      "tokens_seen": 3209166848
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.7684159845931635e-05,
      "loss": 0.0634,
      "theoretical_loss": 3.3321613375046604,
      "tokens_seen": 3209428992
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.760391590434922e-05,
      "loss": 0.063,
      "theoretical_loss": 3.3321408743584713,
      "tokens_seen": 3209691136
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.7523671962766812e-05,
      "loss": 0.0641,
      "theoretical_loss": 3.332120413351408,
      "tokens_seen": 3209953280
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.74434280211844e-05,
      "loss": 0.0646,
      "theoretical_loss": 3.3320999544830725,
      "tokens_seen": 3210215424
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.736318407960199e-05,
      "loss": 0.0639,
      "theoretical_loss": 3.3320794977530666,
      "tokens_seen": 3210477568
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.7282940138019582e-05,
      "loss": 0.0615,
      "theoretical_loss": 3.332059043160992,
      "tokens_seen": 3210739712
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.720269619643717e-05,
      "loss": 0.0626,
      "theoretical_loss": 3.3320385907064507,
      "tokens_seen": 3211001856
    },
    {
      "epoch": 0.97,
      "objective/train/advantage_avg": 0.001090014586225152,
      "objective/train/docs_used": 1166890,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3051739931106567,
      "objective/train/original_loss": 1.3051739931106567,
      "objective/train/theoretical_loss": 3.3320181403890454,
      "objective/train/tokens_used": 3231724000,
      "objective/train/value_avg": -0.0057220458984375,
      "objective/train/value_loss": 0.00018149317475035787,
      "objective/train/value_max": -2.1636486053466797e-05,
      "objective/train/value_min": -0.97509765625,
      "objective/train/value_reward_corr": 0.5922709854946266,
      "objective/train/value_std": 0.01177215576171875,
      "objective/train/weight_avg": 1.0011723041534424,
      "objective/train/weighted_lm_loss": 1.306033730506897,
      "objective/train/weights_max": 1.4535712003707886,
      "objective/train/weights_min": 0.22798629105091095,
      "theoretical_loss": 3.3320181403890454,
      "tokens_seen": 3211264000
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.7122452254854757e-05,
      "loss": 0.0626,
      "theoretical_loss": 3.3320181403890454,
      "tokens_seen": 3211264000
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.704220831327235e-05,
      "loss": 0.0654,
      "theoretical_loss": 3.3319976922083776,
      "tokens_seen": 3211526144
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.6961964371689938e-05,
      "loss": 0.0605,
      "theoretical_loss": 3.33197724616405,
      "tokens_seen": 3211788288
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.6881720430107527e-05,
      "loss": 0.0665,
      "theoretical_loss": 3.3319568022556654,
      "tokens_seen": 3212050432
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.680147648852512e-05,
      "loss": 0.0662,
      "theoretical_loss": 3.331936360482826,
      "tokens_seen": 3212312576
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.6721232546942708e-05,
      "loss": 0.0634,
      "theoretical_loss": 3.3319159208451343,
      "tokens_seen": 3212574720
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.6640988605360294e-05,
      "loss": 0.066,
      "theoretical_loss": 3.3318954833421937,
      "tokens_seen": 3212836864
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.6560744663777886e-05,
      "loss": 0.0626,
      "theoretical_loss": 3.3318750479736066,
      "tokens_seen": 3213099008
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.6480500722195475e-05,
      "loss": 0.0622,
      "theoretical_loss": 3.3318546147389756,
      "tokens_seen": 3213361152
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.6400256780613064e-05,
      "loss": 0.0634,
      "theoretical_loss": 3.3318341836379046,
      "tokens_seen": 3213623296
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.6320012839030656e-05,
      "loss": 0.0646,
      "theoretical_loss": 3.3318137546699966,
      "tokens_seen": 3213885440
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.6239768897448245e-05,
      "loss": 0.0625,
      "theoretical_loss": 3.3317933278348546,
      "tokens_seen": 3214147584
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.615952495586583e-05,
      "loss": 0.062,
      "theoretical_loss": 3.331772903132083,
      "tokens_seen": 3214409728
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.6079281014283423e-05,
      "loss": 0.0607,
      "theoretical_loss": 3.331752480561284,
      "tokens_seen": 3214671872
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.599903707270101e-05,
      "loss": 0.0625,
      "theoretical_loss": 3.3317320601220617,
      "tokens_seen": 3214934016
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.59187931311186e-05,
      "loss": 0.065,
      "theoretical_loss": 3.33171164181402,
      "tokens_seen": 3215196160
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.5838549189536193e-05,
      "loss": 0.0628,
      "theoretical_loss": 3.331691225636763,
      "tokens_seen": 3215458304
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.575830524795378e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.331670811589894,
      "tokens_seen": 3215720448
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.5678061306371367e-05,
      "loss": 0.0622,
      "theoretical_loss": 3.3316503996730176,
      "tokens_seen": 3215982592
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.559781736478896e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.3316299898857373,
      "tokens_seen": 3216244736
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.5517573423206548e-05,
      "loss": 0.0632,
      "theoretical_loss": 3.3316095822276584,
      "tokens_seen": 3216506880
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.5437329481624137e-05,
      "loss": 0.064,
      "theoretical_loss": 3.3315891766983845,
      "tokens_seen": 3216769024
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.535708554004173e-05,
      "loss": 0.0651,
      "theoretical_loss": 3.3315687732975205,
      "tokens_seen": 3217031168
    },
    {
      "epoch": 0.97,
      "learning_rate": 2.5276841598459315e-05,
      "loss": 0.0616,
      "theoretical_loss": 3.3315483720246704,
      "tokens_seen": 3217293312
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.5196597656876904e-05,
      "loss": 0.0657,
      "theoretical_loss": 3.3315279728794396,
      "tokens_seen": 3217555456
    },
    {
      "epoch": 0.98,
      "objective/train/advantage_avg": 0.00010754424874903634,
      "objective/train/docs_used": 1169144,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.370460867881775,
      "objective/train/original_loss": 1.3704607486724854,
      "objective/train/theoretical_loss": 3.3315075758614325,
      "objective/train/tokens_used": 3238277600,
      "objective/train/value_avg": -0.007785797119140625,
      "objective/train/value_loss": 0.0004237206594552845,
      "objective/train/value_max": -1.8477439880371094e-05,
      "objective/train/value_min": -0.67578125,
      "objective/train/value_reward_corr": 0.6450809449913903,
      "objective/train/value_std": 0.015899658203125,
      "objective/train/weight_avg": 1.0002824068069458,
      "objective/train/weighted_lm_loss": 1.3699983358383179,
      "objective/train/weights_max": 1.5724527835845947,
      "objective/train/weights_min": 0.2259981781244278,
      "theoretical_loss": 3.3315075758614325,
      "tokens_seen": 3217817600
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.5116353715294496e-05,
      "loss": 0.0674,
      "theoretical_loss": 3.3315075758614325,
      "tokens_seen": 3217817600
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.5036109773712085e-05,
      "loss": 0.0652,
      "theoretical_loss": 3.331487180970254,
      "tokens_seen": 3218079744
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.4955865832129674e-05,
      "loss": 0.0657,
      "theoretical_loss": 3.331466788205509,
      "tokens_seen": 3218341888
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.4875621890547266e-05,
      "loss": 0.0608,
      "theoretical_loss": 3.331446397566803,
      "tokens_seen": 3218604032
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.4795377948964852e-05,
      "loss": 0.0646,
      "theoretical_loss": 3.3314260090537413,
      "tokens_seen": 3218866176
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.471513400738244e-05,
      "loss": 0.0617,
      "theoretical_loss": 3.3314056226659283,
      "tokens_seen": 3219128320
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.4634890065800033e-05,
      "loss": 0.0648,
      "theoretical_loss": 3.3313852384029707,
      "tokens_seen": 3219390464
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.4554646124217622e-05,
      "loss": 0.0627,
      "theoretical_loss": 3.331364856264473,
      "tokens_seen": 3219652608
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.447440218263521e-05,
      "loss": 0.0652,
      "theoretical_loss": 3.3313444762500413,
      "tokens_seen": 3219914752
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.4394158241052803e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.331324098359281,
      "tokens_seen": 3220176896
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.431391429947039e-05,
      "loss": 0.0645,
      "theoretical_loss": 3.331303722591799,
      "tokens_seen": 3220439040
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.4233670357887977e-05,
      "loss": 0.0663,
      "theoretical_loss": 3.3312833489472,
      "tokens_seen": 3220701184
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.415342641630557e-05,
      "loss": 0.065,
      "theoretical_loss": 3.331262977425091,
      "tokens_seen": 3220963328
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.407318247472316e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.331242608025077,
      "tokens_seen": 3221225472
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.399293853314075e-05,
      "loss": 0.064,
      "theoretical_loss": 3.331222240746765,
      "tokens_seen": 3221487616
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.391269459155834e-05,
      "loss": 0.0651,
      "theoretical_loss": 3.331201875589762,
      "tokens_seen": 3221749760
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.3832450649975925e-05,
      "loss": 0.0613,
      "theoretical_loss": 3.3311815125536737,
      "tokens_seen": 3222011904
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.3752206708393518e-05,
      "loss": 0.0632,
      "theoretical_loss": 3.3311611516381063,
      "tokens_seen": 3222274048
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.3671962766811106e-05,
      "loss": 0.0638,
      "theoretical_loss": 3.3311407928426675,
      "tokens_seen": 3222536192
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.3591718825228695e-05,
      "loss": 0.0637,
      "theoretical_loss": 3.3311204361669637,
      "tokens_seen": 3222798336
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.3511474883646288e-05,
      "loss": 0.0648,
      "theoretical_loss": 3.331100081610602,
      "tokens_seen": 3223060480
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.3431230942063876e-05,
      "loss": 0.0636,
      "theoretical_loss": 3.3310797291731884,
      "tokens_seen": 3223322624
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.3350987000481462e-05,
      "loss": 0.0662,
      "theoretical_loss": 3.3310593788543312,
      "tokens_seen": 3223584768
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.3270743058899054e-05,
      "loss": 0.0618,
      "theoretical_loss": 3.331039030653637,
      "tokens_seen": 3223846912
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.3190499117316643e-05,
      "loss": 0.0652,
      "theoretical_loss": 3.3310186845707137,
      "tokens_seen": 3224109056
    },
    {
      "epoch": 0.98,
      "objective/train/advantage_avg": 7.147617725422606e-05,
      "objective/train/docs_used": 1171684,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3322689533233643,
      "objective/train/original_loss": 1.3322687149047852,
      "objective/train/theoretical_loss": 3.330998340605168,
      "objective/train/tokens_used": 3244831200,
      "objective/train/value_avg": -0.0081939697265625,
      "objective/train/value_loss": 0.0003085145144723356,
      "objective/train/value_max": -2.1755695343017578e-05,
      "objective/train/value_min": -0.78173828125,
      "objective/train/value_reward_corr": 0.7559338916277085,
      "objective/train/value_std": 0.0196380615234375,
      "objective/train/weight_avg": 1.0002161264419556,
      "objective/train/weighted_lm_loss": 1.3316999673843384,
      "objective/train/weights_max": 1.965567946434021,
      "objective/train/weights_min": 0.3757654130458832,
      "theoretical_loss": 3.330998340605168,
      "tokens_seen": 3224371200
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.3110255175734232e-05,
      "loss": 0.0658,
      "theoretical_loss": 3.330998340605168,
      "tokens_seen": 3224371200
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.3030011234151824e-05,
      "loss": 0.0654,
      "theoretical_loss": 3.3309779987566075,
      "tokens_seen": 3224633344
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.2949767292569413e-05,
      "loss": 0.0633,
      "theoretical_loss": 3.330957659024641,
      "tokens_seen": 3224895488
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.2869523350987e-05,
      "loss": 0.0625,
      "theoretical_loss": 3.330937321408874,
      "tokens_seen": 3225157632
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.278927940940459e-05,
      "loss": 0.0639,
      "theoretical_loss": 3.3309169859089165,
      "tokens_seen": 3225419776
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.270903546782218e-05,
      "loss": 0.0646,
      "theoretical_loss": 3.3308966525243755,
      "tokens_seen": 3225681920
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.262879152623977e-05,
      "loss": 0.0659,
      "theoretical_loss": 3.330876321254859,
      "tokens_seen": 3225944064
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.254854758465736e-05,
      "loss": 0.0625,
      "theoretical_loss": 3.3308559920999756,
      "tokens_seen": 3226206208
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.246830364307495e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.3308356650593334,
      "tokens_seen": 3226468352
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.2388059701492536e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.33081534013254,
      "tokens_seen": 3226730496
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.2307815759910128e-05,
      "loss": 0.0618,
      "theoretical_loss": 3.3307950173192054,
      "tokens_seen": 3226992640
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.2227571818327717e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.3307746966189367,
      "tokens_seen": 3227254784
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.2147327876745306e-05,
      "loss": 0.0608,
      "theoretical_loss": 3.3307543780313433,
      "tokens_seen": 3227516928
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.2067083935162898e-05,
      "loss": 0.0662,
      "theoretical_loss": 3.330734061556034,
      "tokens_seen": 3227779072
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.1986839993580487e-05,
      "loss": 0.0648,
      "theoretical_loss": 3.3307137471926174,
      "tokens_seen": 3228041216
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.1906596051998072e-05,
      "loss": 0.0652,
      "theoretical_loss": 3.330693434940703,
      "tokens_seen": 3228303360
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.1826352110415665e-05,
      "loss": 0.0666,
      "theoretical_loss": 3.3306731247998993,
      "tokens_seen": 3228565504
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.1746108168833253e-05,
      "loss": 0.0633,
      "theoretical_loss": 3.330652816769816,
      "tokens_seen": 3228827648
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.1665864227250842e-05,
      "loss": 0.0669,
      "theoretical_loss": 3.330632510850062,
      "tokens_seen": 3229089792
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.1585620285668435e-05,
      "loss": 0.0644,
      "theoretical_loss": 3.3306122070402466,
      "tokens_seen": 3229351936
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.1505376344086024e-05,
      "loss": 0.0634,
      "theoretical_loss": 3.33059190533998,
      "tokens_seen": 3229614080
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.142513240250361e-05,
      "loss": 0.0646,
      "theoretical_loss": 3.3305716057488706,
      "tokens_seen": 3229876224
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.13448884609212e-05,
      "loss": 0.0627,
      "theoretical_loss": 3.33055130826653,
      "tokens_seen": 3230138368
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.126464451933879e-05,
      "loss": 0.0615,
      "theoretical_loss": 3.3305310128925663,
      "tokens_seen": 3230400512
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.118440057775638e-05,
      "loss": 0.0653,
      "theoretical_loss": 3.33051071962659,
      "tokens_seen": 3230662656
    },
    {
      "epoch": 0.98,
      "objective/train/advantage_avg": 0.0006097581936046481,
      "objective/train/docs_used": 1174146,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.445149302482605,
      "objective/train/original_loss": 1.4451491832733154,
      "objective/train/theoretical_loss": 3.3304904284682113,
      "objective/train/tokens_used": 3251384800,
      "objective/train/value_avg": -0.006221771240234375,
      "objective/train/value_loss": 0.00014050577010493726,
      "objective/train/value_max": -1.9252300262451172e-05,
      "objective/train/value_min": -0.70556640625,
      "objective/train/value_reward_corr": 0.7495420834654312,
      "objective/train/value_std": 0.01364898681640625,
      "objective/train/weight_avg": 1.000675082206726,
      "objective/train/weighted_lm_loss": 1.4458887577056885,
      "objective/train/weights_max": 1.349595308303833,
      "objective/train/weights_min": 0.36818361282348633,
      "theoretical_loss": 3.3304904284682113,
      "tokens_seen": 3230924800
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.110415663617397e-05,
      "loss": 0.0668,
      "theoretical_loss": 3.3304904284682113,
      "tokens_seen": 3230924800
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.1023912694591557e-05,
      "loss": 0.0652,
      "theoretical_loss": 3.3304701394170406,
      "tokens_seen": 3231186944
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.0943668753009146e-05,
      "loss": 0.064,
      "theoretical_loss": 3.3304498524726873,
      "tokens_seen": 3231449088
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.0863424811426738e-05,
      "loss": 0.0639,
      "theoretical_loss": 3.3304295676347624,
      "tokens_seen": 3231711232
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.0783180869844327e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.3304092849028764,
      "tokens_seen": 3231973376
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.0702936928261916e-05,
      "loss": 0.0676,
      "theoretical_loss": 3.3303890042766393,
      "tokens_seen": 3232235520
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.0622692986679508e-05,
      "loss": 0.0656,
      "theoretical_loss": 3.3303687257556622,
      "tokens_seen": 3232497664
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.0542449045097094e-05,
      "loss": 0.0628,
      "theoretical_loss": 3.3303484493395556,
      "tokens_seen": 3232759808
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.0462205103514683e-05,
      "loss": 0.0637,
      "theoretical_loss": 3.3303281750279305,
      "tokens_seen": 3233021952
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.0381961161932275e-05,
      "loss": 0.0626,
      "theoretical_loss": 3.3303079028203983,
      "tokens_seen": 3233284096
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.0301717220349864e-05,
      "loss": 0.0655,
      "theoretical_loss": 3.330287632716569,
      "tokens_seen": 3233546240
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.0221473278767453e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.330267364716055,
      "tokens_seen": 3233808384
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.0141229337185045e-05,
      "loss": 0.0652,
      "theoretical_loss": 3.3302470988184667,
      "tokens_seen": 3234070528
    },
    {
      "epoch": 0.98,
      "learning_rate": 2.006098539560263e-05,
      "loss": 0.0634,
      "theoretical_loss": 3.3302268350234154,
      "tokens_seen": 3234332672
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.998074145402022e-05,
      "loss": 0.0644,
      "theoretical_loss": 3.3302065733305133,
      "tokens_seen": 3234594816
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.990049751243781e-05,
      "loss": 0.0652,
      "theoretical_loss": 3.3301863137393717,
      "tokens_seen": 3234856960
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.98202535708554e-05,
      "loss": 0.0619,
      "theoretical_loss": 3.330166056249602,
      "tokens_seen": 3235119104
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.9740009629272993e-05,
      "loss": 0.0659,
      "theoretical_loss": 3.3301458008608162,
      "tokens_seen": 3235381248
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.965976568769058e-05,
      "loss": 0.0637,
      "theoretical_loss": 3.3301255475726266,
      "tokens_seen": 3235643392
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.9579521746108167e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.3301052963846445,
      "tokens_seen": 3235905536
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.949927780452576e-05,
      "loss": 0.0654,
      "theoretical_loss": 3.3300850472964822,
      "tokens_seen": 3236167680
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.941903386294335e-05,
      "loss": 0.0619,
      "theoretical_loss": 3.3300648003077518,
      "tokens_seen": 3236429824
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.9338789921360937e-05,
      "loss": 0.0651,
      "theoretical_loss": 3.330044555418066,
      "tokens_seen": 3236691968
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.925854597977853e-05,
      "loss": 0.0652,
      "theoretical_loss": 3.330024312627037,
      "tokens_seen": 3236954112
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.917830203819612e-05,
      "loss": 0.0643,
      "theoretical_loss": 3.330004071934278,
      "tokens_seen": 3237216256
    },
    {
      "epoch": 0.98,
      "objective/train/advantage_avg": -0.0003025984624400735,
      "objective/train/docs_used": 1176230,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.234129548072815,
      "objective/train/original_loss": 1.2341294288635254,
      "objective/train/theoretical_loss": 3.3299838333394,
      "objective/train/tokens_used": 3257938400,
      "objective/train/value_avg": -0.0086822509765625,
      "objective/train/value_loss": 0.0002719620242714882,
      "objective/train/value_max": -1.9371509552001953e-05,
      "objective/train/value_min": -0.94677734375,
      "objective/train/value_reward_corr": 0.8143104825805222,
      "objective/train/value_std": 0.0205230712890625,
      "objective/train/weight_avg": 0.9998201727867126,
      "objective/train/weighted_lm_loss": 1.2336772680282593,
      "objective/train/weights_max": 1.4542574882507324,
      "objective/train/weights_min": 0.23121991753578186,
      "theoretical_loss": 3.3299838333394,
      "tokens_seen": 3237478400
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.9098058096613704e-05,
      "loss": 0.0632,
      "theoretical_loss": 3.3299838333394,
      "tokens_seen": 3237478400
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.9017814155031296e-05,
      "loss": 0.0666,
      "theoretical_loss": 3.329963596842017,
      "tokens_seen": 3237740544
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.8937570213448885e-05,
      "loss": 0.0629,
      "theoretical_loss": 3.3299433624417416,
      "tokens_seen": 3238002688
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.8857326271866474e-05,
      "loss": 0.0609,
      "theoretical_loss": 3.3299231301381864,
      "tokens_seen": 3238264832
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.8777082330284066e-05,
      "loss": 0.0617,
      "theoretical_loss": 3.3299028999309646,
      "tokens_seen": 3238526976
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.8696838388701655e-05,
      "loss": 0.0624,
      "theoretical_loss": 3.3298826718196897,
      "tokens_seen": 3238789120
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.861659444711924e-05,
      "loss": 0.064,
      "theoretical_loss": 3.3298624458039745,
      "tokens_seen": 3239051264
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.8536350505536833e-05,
      "loss": 0.0624,
      "theoretical_loss": 3.3298422218834323,
      "tokens_seen": 3239313408
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.8456106563954422e-05,
      "loss": 0.0644,
      "theoretical_loss": 3.3298220000576766,
      "tokens_seen": 3239575552
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.837586262237201e-05,
      "loss": 0.0637,
      "theoretical_loss": 3.329801780326321,
      "tokens_seen": 3239837696
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.8295618680789603e-05,
      "loss": 0.0625,
      "theoretical_loss": 3.3297815626889795,
      "tokens_seen": 3240099840
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.8215374739207192e-05,
      "loss": 0.064,
      "theoretical_loss": 3.329761347145265,
      "tokens_seen": 3240361984
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.8135130797624777e-05,
      "loss": 0.064,
      "theoretical_loss": 3.3297411336947924,
      "tokens_seen": 3240624128
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.805488685604237e-05,
      "loss": 0.0641,
      "theoretical_loss": 3.3297209223371746,
      "tokens_seen": 3240886272
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.797464291445996e-05,
      "loss": 0.0652,
      "theoretical_loss": 3.329700713072026,
      "tokens_seen": 3241148416
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.7894398972877548e-05,
      "loss": 0.0662,
      "theoretical_loss": 3.3296805058989616,
      "tokens_seen": 3241410560
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.781415503129514e-05,
      "loss": 0.0665,
      "theoretical_loss": 3.3296603008175945,
      "tokens_seen": 3241672704
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.773391108971273e-05,
      "loss": 0.0627,
      "theoretical_loss": 3.32964009782754,
      "tokens_seen": 3241934848
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.7653667148130314e-05,
      "loss": 0.0644,
      "theoretical_loss": 3.3296198969284116,
      "tokens_seen": 3242196992
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.7573423206547907e-05,
      "loss": 0.0625,
      "theoretical_loss": 3.3295996981198246,
      "tokens_seen": 3242459136
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.7493179264965495e-05,
      "loss": 0.0645,
      "theoretical_loss": 3.329579501401393,
      "tokens_seen": 3242721280
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.7412935323383084e-05,
      "loss": 0.0627,
      "theoretical_loss": 3.3295593067727323,
      "tokens_seen": 3242983424
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.7332691381800677e-05,
      "loss": 0.0627,
      "theoretical_loss": 3.329539114233457,
      "tokens_seen": 3243245568
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.7252447440218265e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.3295189237831826,
      "tokens_seen": 3243507712
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.717220349863585e-05,
      "loss": 0.0628,
      "theoretical_loss": 3.329498735421523,
      "tokens_seen": 3243769856
    },
    {
      "epoch": 0.98,
      "objective/train/advantage_avg": 0.0003678802167996764,
      "objective/train/docs_used": 1178511,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2842475175857544,
      "objective/train/original_loss": 1.2842473983764648,
      "objective/train/theoretical_loss": 3.3294785491480945,
      "objective/train/tokens_used": 3264492000,
      "objective/train/value_avg": -0.00807952880859375,
      "objective/train/value_loss": 0.0004038362530991435,
      "objective/train/value_max": -2.6285648345947266e-05,
      "objective/train/value_min": -0.525390625,
      "objective/train/value_reward_corr": 0.6182115186902631,
      "objective/train/value_std": 0.0156707763671875,
      "objective/train/weight_avg": 1.0005180835723877,
      "objective/train/weighted_lm_loss": 1.284083366394043,
      "objective/train/weights_max": 1.6044148206710815,
      "objective/train/weights_min": 0.053332287818193436,
      "theoretical_loss": 3.3294785491480945,
      "tokens_seen": 3244032000
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.7091959557053443e-05,
      "loss": 0.0644,
      "theoretical_loss": 3.3294785491480945,
      "tokens_seen": 3244032000
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.7011715615471032e-05,
      "loss": 0.0638,
      "theoretical_loss": 3.3294583649625116,
      "tokens_seen": 3244294144
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.693147167388862e-05,
      "loss": 0.0637,
      "theoretical_loss": 3.3294381828643904,
      "tokens_seen": 3244556288
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.6851227732306213e-05,
      "loss": 0.063,
      "theoretical_loss": 3.329418002853346,
      "tokens_seen": 3244818432
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.67709837907238e-05,
      "loss": 0.0614,
      "theoretical_loss": 3.3293978249289937,
      "tokens_seen": 3245080576
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.6690739849141388e-05,
      "loss": 0.064,
      "theoretical_loss": 3.32937764909095,
      "tokens_seen": 3245342720
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.661049590755898e-05,
      "loss": 0.0639,
      "theoretical_loss": 3.32935747533883,
      "tokens_seen": 3245604864
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.653025196597657e-05,
      "loss": 0.0629,
      "theoretical_loss": 3.32933730367225,
      "tokens_seen": 3245867008
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.6450008024394158e-05,
      "loss": 0.065,
      "theoretical_loss": 3.329317134090825,
      "tokens_seen": 3246129152
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.636976408281175e-05,
      "loss": 0.0655,
      "theoretical_loss": 3.3292969665941725,
      "tokens_seen": 3246391296
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.6289520141229336e-05,
      "loss": 0.0638,
      "theoretical_loss": 3.3292768011819076,
      "tokens_seen": 3246653440
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.6209276199646925e-05,
      "loss": 0.0641,
      "theoretical_loss": 3.3292566378536472,
      "tokens_seen": 3246915584
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.6129032258064517e-05,
      "loss": 0.0652,
      "theoretical_loss": 3.329236476609008,
      "tokens_seen": 3247177728
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.6048788316482106e-05,
      "loss": 0.0652,
      "theoretical_loss": 3.3292163174476057,
      "tokens_seen": 3247439872
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.5968544374899695e-05,
      "loss": 0.0656,
      "theoretical_loss": 3.329196160369057,
      "tokens_seen": 3247702016
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.5888300433317287e-05,
      "loss": 0.0654,
      "theoretical_loss": 3.3291760053729798,
      "tokens_seen": 3247964160
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.5808056491734872e-05,
      "loss": 0.0648,
      "theoretical_loss": 3.329155852458989,
      "tokens_seen": 3248226304
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.572781255015246e-05,
      "loss": 0.0653,
      "theoretical_loss": 3.329135701626703,
      "tokens_seen": 3248488448
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.5647568608570054e-05,
      "loss": 0.0654,
      "theoretical_loss": 3.329115552875738,
      "tokens_seen": 3248750592
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.5567324666987642e-05,
      "loss": 0.0667,
      "theoretical_loss": 3.3290954062057114,
      "tokens_seen": 3249012736
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.548708072540523e-05,
      "loss": 0.0651,
      "theoretical_loss": 3.3290752616162402,
      "tokens_seen": 3249274880
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.540683678382282e-05,
      "loss": 0.0652,
      "theoretical_loss": 3.3290551191069424,
      "tokens_seen": 3249537024
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.5326592842240413e-05,
      "loss": 0.0648,
      "theoretical_loss": 3.3290349786774347,
      "tokens_seen": 3249799168
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.5246348900658001e-05,
      "loss": 0.065,
      "theoretical_loss": 3.3290148403273347,
      "tokens_seen": 3250061312
    },
    {
      "epoch": 0.98,
      "learning_rate": 1.516610495907559e-05,
      "loss": 0.0609,
      "theoretical_loss": 3.3289947040562606,
      "tokens_seen": 3250323456
    },
    {
      "epoch": 0.98,
      "objective/train/advantage_avg": 0.000409378350013867,
      "objective/train/docs_used": 1181030,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2445560693740845,
      "objective/train/original_loss": 1.244555950164795,
      "objective/train/theoretical_loss": 3.3289745698638296,
      "objective/train/tokens_used": 3271045600,
      "objective/train/value_avg": -0.0084991455078125,
      "objective/train/value_loss": 0.00021049527276773006,
      "objective/train/value_max": -2.485513687133789e-05,
      "objective/train/value_min": -0.2293701171875,
      "objective/train/value_reward_corr": 0.7449387464870404,
      "objective/train/value_std": 0.016754150390625,
      "objective/train/weight_avg": 1.0005031824111938,
      "objective/train/weighted_lm_loss": 1.2445323467254639,
      "objective/train/weights_max": 1.112181305885315,
      "objective/train/weights_min": 0.36861130595207214,
      "theoretical_loss": 3.3289745698638296,
      "tokens_seen": 3250585600
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.508586101749318e-05,
      "loss": 0.064,
      "theoretical_loss": 3.3289745698638296,
      "tokens_seen": 3250585600
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.500561707591077e-05,
      "loss": 0.0636,
      "theoretical_loss": 3.328954437749659,
      "tokens_seen": 3250847744
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.4925373134328359e-05,
      "loss": 0.0631,
      "theoretical_loss": 3.328934307713368,
      "tokens_seen": 3251109888
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.4845129192745948e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.328914179754573,
      "tokens_seen": 3251372032
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.4764885251163536e-05,
      "loss": 0.0634,
      "theoretical_loss": 3.328894053872894,
      "tokens_seen": 3251634176
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.4684641309581127e-05,
      "loss": 0.0623,
      "theoretical_loss": 3.3288739300679477,
      "tokens_seen": 3251896320
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.4604397367998716e-05,
      "loss": 0.0629,
      "theoretical_loss": 3.3288538083393533,
      "tokens_seen": 3252158464
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.4524153426416305e-05,
      "loss": 0.0661,
      "theoretical_loss": 3.328833688686729,
      "tokens_seen": 3252420608
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.4443909484833895e-05,
      "loss": 0.0646,
      "theoretical_loss": 3.328813571109693,
      "tokens_seen": 3252682752
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.4363665543251484e-05,
      "loss": 0.0621,
      "theoretical_loss": 3.3287934556078644,
      "tokens_seen": 3252944896
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.4283421601669073e-05,
      "loss": 0.062,
      "theoretical_loss": 3.3287733421808614,
      "tokens_seen": 3253207040
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.4203177660086664e-05,
      "loss": 0.0658,
      "theoretical_loss": 3.328753230828303,
      "tokens_seen": 3253469184
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.4122933718504254e-05,
      "loss": 0.0642,
      "theoretical_loss": 3.3287331215498086,
      "tokens_seen": 3253731328
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.4042689776921842e-05,
      "loss": 0.063,
      "theoretical_loss": 3.328713014344997,
      "tokens_seen": 3253993472
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.3962445835339432e-05,
      "loss": 0.0658,
      "theoretical_loss": 3.328692909213487,
      "tokens_seen": 3254255616
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.3882201893757023e-05,
      "loss": 0.0607,
      "theoretical_loss": 3.3286728061548976,
      "tokens_seen": 3254517760
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.380195795217461e-05,
      "loss": 0.0624,
      "theoretical_loss": 3.328652705168849,
      "tokens_seen": 3254779904
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.37217140105922e-05,
      "loss": 0.0632,
      "theoretical_loss": 3.3286326062549603,
      "tokens_seen": 3255042048
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.3641470069009791e-05,
      "loss": 0.0657,
      "theoretical_loss": 3.3286125094128507,
      "tokens_seen": 3255304192
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.3561226127427378e-05,
      "loss": 0.0604,
      "theoretical_loss": 3.32859241464214,
      "tokens_seen": 3255566336
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.3480982185844969e-05,
      "loss": 0.0635,
      "theoretical_loss": 3.328572321942448,
      "tokens_seen": 3255828480
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.340073824426256e-05,
      "loss": 0.0633,
      "theoretical_loss": 3.328552231313395,
      "tokens_seen": 3256090624
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.3320494302680147e-05,
      "loss": 0.0634,
      "theoretical_loss": 3.3285321427545997,
      "tokens_seen": 3256352768
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.3240250361097737e-05,
      "loss": 0.0663,
      "theoretical_loss": 3.3285120562656836,
      "tokens_seen": 3256614912
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.3160006419515328e-05,
      "loss": 0.0654,
      "theoretical_loss": 3.3284919718462653,
      "tokens_seen": 3256877056
    },
    {
      "epoch": 0.99,
      "objective/train/advantage_avg": 0.00017991720233112574,
      "objective/train/docs_used": 1183352,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2570805549621582,
      "objective/train/original_loss": 1.2570806741714478,
      "objective/train/theoretical_loss": 3.3284718894959666,
      "objective/train/tokens_used": 3277599200,
      "objective/train/value_avg": -0.00946044921875,
      "objective/train/value_loss": 0.00020266363571863621,
      "objective/train/value_max": -3.0934810638427734e-05,
      "objective/train/value_min": -0.92529296875,
      "objective/train/value_reward_corr": 0.7796501256549795,
      "objective/train/value_std": 0.018463134765625,
      "objective/train/weight_avg": 1.0002756118774414,
      "objective/train/weighted_lm_loss": 1.2571203708648682,
      "objective/train/weights_max": 1.4121947288513184,
      "objective/train/weights_min": 0.3786547780036926,
      "theoretical_loss": 3.3284718894959666,
      "tokens_seen": 3257139200
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.3079762477932915e-05,
      "loss": 0.0655,
      "theoretical_loss": 3.3284718894959666,
      "tokens_seen": 3257139200
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.2999518536350506e-05,
      "loss": 0.0616,
      "theoretical_loss": 3.328451809214407,
      "tokens_seen": 3257401344
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.2919274594768096e-05,
      "loss": 0.0633,
      "theoretical_loss": 3.328431731001207,
      "tokens_seen": 3257663488
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.2839030653185684e-05,
      "loss": 0.0645,
      "theoretical_loss": 3.328411654855987,
      "tokens_seen": 3257925632
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.2758786711603274e-05,
      "loss": 0.0598,
      "theoretical_loss": 3.3283915807783675,
      "tokens_seen": 3258187776
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.2678542770020865e-05,
      "loss": 0.062,
      "theoretical_loss": 3.3283715087679697,
      "tokens_seen": 3258449920
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.2598298828438452e-05,
      "loss": 0.0616,
      "theoretical_loss": 3.3283514388244146,
      "tokens_seen": 3258712064
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.2518054886856043e-05,
      "loss": 0.063,
      "theoretical_loss": 3.3283313709473226,
      "tokens_seen": 3258974208
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.2437810945273633e-05,
      "loss": 0.0657,
      "theoretical_loss": 3.3283113051363147,
      "tokens_seen": 3259236352
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.235756700369122e-05,
      "loss": 0.064,
      "theoretical_loss": 3.3282912413910126,
      "tokens_seen": 3259498496
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.2277323062108811e-05,
      "loss": 0.0621,
      "theoretical_loss": 3.328271179711037,
      "tokens_seen": 3259760640
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.2197079120526401e-05,
      "loss": 0.0631,
      "theoretical_loss": 3.3282511200960094,
      "tokens_seen": 3260022784
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.2116835178943989e-05,
      "loss": 0.0622,
      "theoretical_loss": 3.3282310625455516,
      "tokens_seen": 3260284928
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.203659123736158e-05,
      "loss": 0.062,
      "theoretical_loss": 3.3282110070592847,
      "tokens_seen": 3260547072
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.195634729577917e-05,
      "loss": 0.0631,
      "theoretical_loss": 3.32819095363683,
      "tokens_seen": 3260809216
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.1876103354196759e-05,
      "loss": 0.0623,
      "theoretical_loss": 3.3281709022778103,
      "tokens_seen": 3261071360
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.1795859412614348e-05,
      "loss": 0.0615,
      "theoretical_loss": 3.3281508529818464,
      "tokens_seen": 3261333504
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.1715615471031938e-05,
      "loss": 0.0622,
      "theoretical_loss": 3.3281308057485606,
      "tokens_seen": 3261595648
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.1635371529449527e-05,
      "loss": 0.0628,
      "theoretical_loss": 3.3281107605775753,
      "tokens_seen": 3261857792
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.1555127587867116e-05,
      "loss": 0.0626,
      "theoretical_loss": 3.328090717468512,
      "tokens_seen": 3262119936
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.1474883646284707e-05,
      "loss": 0.0621,
      "theoretical_loss": 3.3280706764209933,
      "tokens_seen": 3262382080
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.1394639704702296e-05,
      "loss": 0.0624,
      "theoretical_loss": 3.3280506374346417,
      "tokens_seen": 3262644224
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.1314395763119884e-05,
      "loss": 0.0604,
      "theoretical_loss": 3.3280306005090794,
      "tokens_seen": 3262906368
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.1234151821537475e-05,
      "loss": 0.0649,
      "theoretical_loss": 3.3280105656439285,
      "tokens_seen": 3263168512
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.1153907879955064e-05,
      "loss": 0.0617,
      "theoretical_loss": 3.3279905328388124,
      "tokens_seen": 3263430656
    },
    {
      "epoch": 0.99,
      "objective/train/advantage_avg": -9.282096289098263e-05,
      "objective/train/docs_used": 1185836,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.4554698467254639,
      "objective/train/original_loss": 1.4554697275161743,
      "objective/train/theoretical_loss": 3.327970502093353,
      "objective/train/tokens_used": 3284152800,
      "objective/train/value_avg": -0.0071563720703125,
      "objective/train/value_loss": 0.00044870967394672334,
      "objective/train/value_max": -1.722574234008789e-05,
      "objective/train/value_min": -0.908203125,
      "objective/train/value_reward_corr": 0.7138698933127893,
      "objective/train/value_std": 0.021087646484375,
      "objective/train/weight_avg": 1.000106930732727,
      "objective/train/weighted_lm_loss": 1.4546253681182861,
      "objective/train/weights_max": 2.210451126098633,
      "objective/train/weights_min": 0.3683115839958191,
      "theoretical_loss": 3.327970502093353,
      "tokens_seen": 3263692800
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.1073663938372653e-05,
      "loss": 0.0655,
      "theoretical_loss": 3.327970502093353,
      "tokens_seen": 3263692800
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.0993419996790243e-05,
      "loss": 0.0625,
      "theoretical_loss": 3.3279504734071743,
      "tokens_seen": 3263954944
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.0913176055207832e-05,
      "loss": 0.0632,
      "theoretical_loss": 3.3279304467798987,
      "tokens_seen": 3264217088
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.0832932113625421e-05,
      "loss": 0.062,
      "theoretical_loss": 3.3279104222111484,
      "tokens_seen": 3264479232
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.0752688172043012e-05,
      "loss": 0.0647,
      "theoretical_loss": 3.3278903997005473,
      "tokens_seen": 3264741376
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.06724442304606e-05,
      "loss": 0.0633,
      "theoretical_loss": 3.327870379247719,
      "tokens_seen": 3265003520
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.059220028887819e-05,
      "loss": 0.0644,
      "theoretical_loss": 3.327850360852286,
      "tokens_seen": 3265265664
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.0511956347295778e-05,
      "loss": 0.0637,
      "theoretical_loss": 3.327830344513872,
      "tokens_seen": 3265527808
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.0431712405713369e-05,
      "loss": 0.0621,
      "theoretical_loss": 3.3278103302321007,
      "tokens_seen": 3265789952
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.0351468464130958e-05,
      "loss": 0.0618,
      "theoretical_loss": 3.327790318006596,
      "tokens_seen": 3266052096
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.0271224522548547e-05,
      "loss": 0.0623,
      "theoretical_loss": 3.327770307836981,
      "tokens_seen": 3266314240
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.0190980580966137e-05,
      "loss": 0.063,
      "theoretical_loss": 3.3277502997228794,
      "tokens_seen": 3266576384
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.0110736639383726e-05,
      "loss": 0.0639,
      "theoretical_loss": 3.327730293663916,
      "tokens_seen": 3266838528
    },
    {
      "epoch": 0.99,
      "learning_rate": 1.0030492697801315e-05,
      "loss": 0.0667,
      "theoretical_loss": 3.327710289659714,
      "tokens_seen": 3267100672
    },
    {
      "epoch": 0.99,
      "learning_rate": 9.950248756218906e-06,
      "loss": 0.0659,
      "theoretical_loss": 3.327690287709898,
      "tokens_seen": 3267362816
    },
    {
      "epoch": 0.99,
      "learning_rate": 9.870004814636496e-06,
      "loss": 0.0623,
      "theoretical_loss": 3.327670287814092,
      "tokens_seen": 3267624960
    },
    {
      "epoch": 0.99,
      "learning_rate": 9.789760873054084e-06,
      "loss": 0.0626,
      "theoretical_loss": 3.3276502899719205,
      "tokens_seen": 3267887104
    },
    {
      "epoch": 0.99,
      "learning_rate": 9.709516931471674e-06,
      "loss": 0.0639,
      "theoretical_loss": 3.3276302941830074,
      "tokens_seen": 3268149248
    },
    {
      "epoch": 0.99,
      "learning_rate": 9.629272989889265e-06,
      "loss": 0.0617,
      "theoretical_loss": 3.327610300446978,
      "tokens_seen": 3268411392
    },
    {
      "epoch": 0.99,
      "learning_rate": 9.549029048306852e-06,
      "loss": 0.0642,
      "theoretical_loss": 3.3275903087634564,
      "tokens_seen": 3268673536
    },
    {
      "epoch": 0.99,
      "learning_rate": 9.468785106724443e-06,
      "loss": 0.0621,
      "theoretical_loss": 3.327570319132067,
      "tokens_seen": 3268935680
    },
    {
      "epoch": 0.99,
      "learning_rate": 9.388541165142033e-06,
      "loss": 0.0662,
      "theoretical_loss": 3.3275503315524357,
      "tokens_seen": 3269197824
    },
    {
      "epoch": 0.99,
      "learning_rate": 9.30829722355962e-06,
      "loss": 0.0616,
      "theoretical_loss": 3.3275303460241865,
      "tokens_seen": 3269459968
    },
    {
      "epoch": 0.99,
      "learning_rate": 9.228053281977211e-06,
      "loss": 0.0636,
      "theoretical_loss": 3.327510362546944,
      "tokens_seen": 3269722112
    },
    {
      "epoch": 0.99,
      "learning_rate": 9.147809340394802e-06,
      "loss": 0.0636,
      "theoretical_loss": 3.3274903811203345,
      "tokens_seen": 3269984256
    },
    {
      "epoch": 0.99,
      "objective/train/advantage_avg": 5.63900321139954e-05,
      "objective/train/docs_used": 1188255,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3194350004196167,
      "objective/train/original_loss": 1.3194348812103271,
      "objective/train/theoretical_loss": 3.327470401743983,
      "objective/train/tokens_used": 3290706400,
      "objective/train/value_avg": -0.007358551025390625,
      "objective/train/value_loss": 0.00023028827854432166,
      "objective/train/value_max": -3.427267074584961e-05,
      "objective/train/value_min": -0.39013671875,
      "objective/train/value_reward_corr": 0.6854116826893848,
      "objective/train/value_std": 0.01427459716796875,
      "objective/train/weight_avg": 1.0001581907272339,
      "objective/train/weighted_lm_loss": 1.3194934129714966,
      "objective/train/weights_max": 1.4432260990142822,
      "objective/train/weights_min": 0.3683066666126251,
      "theoretical_loss": 3.327470401743983,
      "tokens_seen": 3270246400
    },
    {
      "epoch": 0.99,
      "learning_rate": 9.067565398812389e-06,
      "loss": 0.0637,
      "theoretical_loss": 3.327470401743983,
      "tokens_seen": 3270246400
    },
    {
      "epoch": 0.99,
      "learning_rate": 8.98732145722998e-06,
      "loss": 0.0615,
      "theoretical_loss": 3.327450424417514,
      "tokens_seen": 3270508544
    },
    {
      "epoch": 0.99,
      "learning_rate": 8.90707751564757e-06,
      "loss": 0.0646,
      "theoretical_loss": 3.3274304491405533,
      "tokens_seen": 3270770688
    },
    {
      "epoch": 0.99,
      "learning_rate": 8.826833574065157e-06,
      "loss": 0.0628,
      "theoretical_loss": 3.3274104759127265,
      "tokens_seen": 3271032832
    },
    {
      "epoch": 0.99,
      "learning_rate": 8.746589632482748e-06,
      "loss": 0.0633,
      "theoretical_loss": 3.3273905047336596,
      "tokens_seen": 3271294976
    },
    {
      "epoch": 0.99,
      "learning_rate": 8.666345690900338e-06,
      "loss": 0.0632,
      "theoretical_loss": 3.3273705356029772,
      "tokens_seen": 3271557120
    },
    {
      "epoch": 0.99,
      "learning_rate": 8.586101749317925e-06,
      "loss": 0.0642,
      "theoretical_loss": 3.327350568520306,
      "tokens_seen": 3271819264
    },
    {
      "epoch": 0.99,
      "learning_rate": 8.505857807735516e-06,
      "loss": 0.0625,
      "theoretical_loss": 3.327330603485272,
      "tokens_seen": 3272081408
    },
    {
      "epoch": 0.99,
      "learning_rate": 8.425613866153107e-06,
      "loss": 0.0629,
      "theoretical_loss": 3.3273106404975,
      "tokens_seen": 3272343552
    },
    {
      "epoch": 0.99,
      "learning_rate": 8.345369924570694e-06,
      "loss": 0.0631,
      "theoretical_loss": 3.3272906795566177,
      "tokens_seen": 3272605696
    },
    {
      "epoch": 0.99,
      "learning_rate": 8.265125982988284e-06,
      "loss": 0.063,
      "theoretical_loss": 3.3272707206622503,
      "tokens_seen": 3272867840
    },
    {
      "epoch": 0.99,
      "learning_rate": 8.184882041405875e-06,
      "loss": 0.0629,
      "theoretical_loss": 3.327250763814024,
      "tokens_seen": 3273129984
    },
    {
      "epoch": 0.99,
      "learning_rate": 8.104638099823462e-06,
      "loss": 0.0642,
      "theoretical_loss": 3.327230809011566,
      "tokens_seen": 3273392128
    },
    {
      "epoch": 0.99,
      "learning_rate": 8.024394158241053e-06,
      "loss": 0.0635,
      "theoretical_loss": 3.327210856254502,
      "tokens_seen": 3273654272
    },
    {
      "epoch": 0.99,
      "learning_rate": 7.944150216658643e-06,
      "loss": 0.0644,
      "theoretical_loss": 3.3271909055424587,
      "tokens_seen": 3273916416
    },
    {
      "epoch": 0.99,
      "learning_rate": 7.86390627507623e-06,
      "loss": 0.0621,
      "theoretical_loss": 3.327170956875063,
      "tokens_seen": 3274178560
    },
    {
      "epoch": 0.99,
      "learning_rate": 7.783662333493821e-06,
      "loss": 0.0632,
      "theoretical_loss": 3.3271510102519413,
      "tokens_seen": 3274440704
    },
    {
      "epoch": 0.99,
      "learning_rate": 7.70341839191141e-06,
      "loss": 0.0621,
      "theoretical_loss": 3.327131065672721,
      "tokens_seen": 3274702848
    },
    {
      "epoch": 0.99,
      "learning_rate": 7.623174450329001e-06,
      "loss": 0.0638,
      "theoretical_loss": 3.327111123137029,
      "tokens_seen": 3274964992
    },
    {
      "epoch": 0.99,
      "learning_rate": 7.54293050874659e-06,
      "loss": 0.0612,
      "theoretical_loss": 3.327091182644492,
      "tokens_seen": 3275227136
    },
    {
      "epoch": 0.99,
      "learning_rate": 7.462686567164179e-06,
      "loss": 0.0609,
      "theoretical_loss": 3.3270712441947374,
      "tokens_seen": 3275489280
    },
    {
      "epoch": 0.99,
      "learning_rate": 7.382442625581768e-06,
      "loss": 0.0627,
      "theoretical_loss": 3.327051307787393,
      "tokens_seen": 3275751424
    },
    {
      "epoch": 0.99,
      "learning_rate": 7.302198683999358e-06,
      "loss": 0.0613,
      "theoretical_loss": 3.327031373422085,
      "tokens_seen": 3276013568
    },
    {
      "epoch": 0.99,
      "learning_rate": 7.221954742416948e-06,
      "loss": 0.0643,
      "theoretical_loss": 3.3270114410984415,
      "tokens_seen": 3276275712
    },
    {
      "epoch": 0.99,
      "learning_rate": 7.141710800834537e-06,
      "loss": 0.0645,
      "theoretical_loss": 3.3269915108160903,
      "tokens_seen": 3276537856
    },
    {
      "debugging/Compilability": 1.0,
      "debugging/distinct-1-grams": 0.7547341464283754,
      "debugging/entropy-1-grams": 5.388662098705375,
      "debugging/length": 454.42105263157896,
      "debugging/num_segments": 19,
      "debugging/raw_token_scores_avg": 0.006775569636374712,
      "debugging/raw_token_scores_std": 0.018655896186828613,
      "debugging/score": 0.008703016206509368,
      "debugging/score_std": 0.010562036912131832,
      "epoch": 0.99,
      "objective/train/advantage_avg": 0.0003331316402181983,
      "objective/train/docs_used": 1190634,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.3039427995681763,
      "objective/train/original_loss": 1.3039426803588867,
      "objective/train/theoretical_loss": 3.3269715825746586,
      "objective/train/tokens_used": 3297260000,
      "objective/train/value_avg": -0.007110595703125,
      "objective/train/value_loss": 0.00014388281852006912,
      "objective/train/value_max": -2.467632293701172e-05,
      "objective/train/value_min": -0.85791015625,
      "objective/train/value_reward_corr": 0.7662398062523196,
      "objective/train/value_std": 0.01428985595703125,
      "objective/train/weight_avg": 1.000400424003601,
      "objective/train/weighted_lm_loss": 1.3044633865356445,
      "objective/train/weights_max": 1.2395168542861938,
      "objective/train/weights_min": 0.37148112058639526,
      "theoretical_loss": 3.3269715825746586,
      "tokens_seen": 3276800000
    },
    {
      "epoch": 0.99,
      "learning_rate": 7.061466859252127e-06,
      "loss": 0.0623,
      "theoretical_loss": 3.3269715825746586,
      "tokens_seen": 3276800000
    },
    {
      "epoch": 0.99,
      "learning_rate": 6.981222917669716e-06,
      "loss": 0.0634,
      "theoretical_loss": 3.3269516563737747,
      "tokens_seen": 3277062144
    },
    {
      "epoch": 0.99,
      "learning_rate": 6.900978976087305e-06,
      "loss": 0.0667,
      "theoretical_loss": 3.326931732213066,
      "tokens_seen": 3277324288
    },
    {
      "epoch": 0.99,
      "learning_rate": 6.820735034504896e-06,
      "loss": 0.063,
      "theoretical_loss": 3.32691181009216,
      "tokens_seen": 3277586432
    },
    {
      "epoch": 0.99,
      "learning_rate": 6.7404910929224845e-06,
      "loss": 0.0625,
      "theoretical_loss": 3.326891890010686,
      "tokens_seen": 3277848576
    },
    {
      "epoch": 0.99,
      "learning_rate": 6.660247151340073e-06,
      "loss": 0.0624,
      "theoretical_loss": 3.326871971968271,
      "tokens_seen": 3278110720
    },
    {
      "epoch": 0.99,
      "learning_rate": 6.580003209757664e-06,
      "loss": 0.0645,
      "theoretical_loss": 3.326852055964544,
      "tokens_seen": 3278372864
    },
    {
      "epoch": 0.99,
      "learning_rate": 6.499759268175253e-06,
      "loss": 0.065,
      "theoretical_loss": 3.3268321419991325,
      "tokens_seen": 3278635008
    },
    {
      "epoch": 0.99,
      "learning_rate": 6.419515326592842e-06,
      "loss": 0.0617,
      "theoretical_loss": 3.326812230071666,
      "tokens_seen": 3278897152
    },
    {
      "epoch": 0.99,
      "learning_rate": 6.339271385010432e-06,
      "loss": 0.0626,
      "theoretical_loss": 3.326792320181772,
      "tokens_seen": 3279159296
    },
    {
      "epoch": 0.99,
      "learning_rate": 6.259027443428021e-06,
      "loss": 0.061,
      "theoretical_loss": 3.3267724123290803,
      "tokens_seen": 3279421440
    },
    {
      "epoch": 0.99,
      "learning_rate": 6.17878350184561e-06,
      "loss": 0.0644,
      "theoretical_loss": 3.3267525065132184,
      "tokens_seen": 3279683584
    },
    {
      "epoch": 0.99,
      "learning_rate": 6.098539560263201e-06,
      "loss": 0.0637,
      "theoretical_loss": 3.326732602733816,
      "tokens_seen": 3279945728
    },
    {
      "epoch": 0.99,
      "learning_rate": 6.01829561868079e-06,
      "loss": 0.0657,
      "theoretical_loss": 3.3267127009905018,
      "tokens_seen": 3280207872
    },
    {
      "epoch": 0.99,
      "learning_rate": 5.938051677098379e-06,
      "loss": 0.0649,
      "theoretical_loss": 3.3266928012829045,
      "tokens_seen": 3280470016
    },
    {
      "epoch": 0.99,
      "learning_rate": 5.857807735515969e-06,
      "loss": 0.0649,
      "theoretical_loss": 3.3266729036106533,
      "tokens_seen": 3280732160
    },
    {
      "epoch": 0.99,
      "learning_rate": 5.777563793933558e-06,
      "loss": 0.0607,
      "theoretical_loss": 3.326653007973378,
      "tokens_seen": 3280994304
    },
    {
      "epoch": 0.99,
      "learning_rate": 5.697319852351148e-06,
      "loss": 0.0608,
      "theoretical_loss": 3.3266331143707073,
      "tokens_seen": 3281256448
    },
    {
      "epoch": 0.99,
      "learning_rate": 5.6170759107687375e-06,
      "loss": 0.0632,
      "theoretical_loss": 3.326613222802271,
      "tokens_seen": 3281518592
    },
    {
      "epoch": 0.99,
      "learning_rate": 5.536831969186326e-06,
      "loss": 0.0649,
      "theoretical_loss": 3.326593333267698,
      "tokens_seen": 3281780736
    },
    {
      "epoch": 0.99,
      "learning_rate": 5.456588027603916e-06,
      "loss": 0.0636,
      "theoretical_loss": 3.3265734457666185,
      "tokens_seen": 3282042880
    },
    {
      "epoch": 0.99,
      "learning_rate": 5.376344086021506e-06,
      "loss": 0.063,
      "theoretical_loss": 3.3265535602986622,
      "tokens_seen": 3282305024
    },
    {
      "epoch": 0.99,
      "learning_rate": 5.296100144439095e-06,
      "loss": 0.0608,
      "theoretical_loss": 3.3265336768634586,
      "tokens_seen": 3282567168
    },
    {
      "epoch": 0.99,
      "learning_rate": 5.2158562028566845e-06,
      "loss": 0.0636,
      "theoretical_loss": 3.3265137954606376,
      "tokens_seen": 3282829312
    },
    {
      "epoch": 0.99,
      "learning_rate": 5.135612261274273e-06,
      "loss": 0.0624,
      "theoretical_loss": 3.3264939160898295,
      "tokens_seen": 3283091456
    },
    {
      "epoch": 0.99,
      "objective/train/advantage_avg": 0.0002915835939347744,
      "objective/train/docs_used": 1192893,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.233608603477478,
      "objective/train/original_loss": 1.2336084842681885,
      "objective/train/theoretical_loss": 3.326474038750664,
      "objective/train/tokens_used": 3303813600,
      "objective/train/value_avg": -0.00603485107421875,
      "objective/train/value_loss": 0.0001250050845555961,
      "objective/train/value_max": -3.272294998168945e-05,
      "objective/train/value_min": -0.449951171875,
      "objective/train/value_reward_corr": 0.695894549483661,
      "objective/train/value_std": 0.010955810546875,
      "objective/train/weight_avg": 1.0003503561019897,
      "objective/train/weighted_lm_loss": 1.23422110080719,
      "objective/train/weights_max": 1.2648903131484985,
      "objective/train/weights_min": 0.5307503342628479,
      "theoretical_loss": 3.326474038750664,
      "tokens_seen": 3283353600
    },
    {
      "epoch": 0.99,
      "learning_rate": 5.055368319691863e-06,
      "loss": 0.0616,
      "theoretical_loss": 3.326474038750664,
      "tokens_seen": 3283353600
    },
    {
      "epoch": 1.0,
      "learning_rate": 4.975124378109453e-06,
      "loss": 0.0634,
      "theoretical_loss": 3.3264541634427713,
      "tokens_seen": 3283615744
    },
    {
      "epoch": 1.0,
      "learning_rate": 4.894880436527042e-06,
      "loss": 0.0633,
      "theoretical_loss": 3.3264342901657824,
      "tokens_seen": 3283877888
    },
    {
      "epoch": 1.0,
      "learning_rate": 4.814636494944632e-06,
      "loss": 0.0632,
      "theoretical_loss": 3.326414418919327,
      "tokens_seen": 3284140032
    },
    {
      "epoch": 1.0,
      "learning_rate": 4.734392553362221e-06,
      "loss": 0.0613,
      "theoretical_loss": 3.3263945497030356,
      "tokens_seen": 3284402176
    },
    {
      "epoch": 1.0,
      "learning_rate": 4.65414861177981e-06,
      "loss": 0.0648,
      "theoretical_loss": 3.326374682516539,
      "tokens_seen": 3284664320
    },
    {
      "epoch": 1.0,
      "learning_rate": 4.573904670197401e-06,
      "loss": 0.0636,
      "theoretical_loss": 3.3263548173594675,
      "tokens_seen": 3284926464
    },
    {
      "epoch": 1.0,
      "learning_rate": 4.49366072861499e-06,
      "loss": 0.0645,
      "theoretical_loss": 3.3263349542314526,
      "tokens_seen": 3285188608
    },
    {
      "epoch": 1.0,
      "learning_rate": 4.4134167870325786e-06,
      "loss": 0.0639,
      "theoretical_loss": 3.3263150931321244,
      "tokens_seen": 3285450752
    },
    {
      "epoch": 1.0,
      "learning_rate": 4.333172845450169e-06,
      "loss": 0.0615,
      "theoretical_loss": 3.326295234061114,
      "tokens_seen": 3285712896
    },
    {
      "epoch": 1.0,
      "learning_rate": 4.252928903867758e-06,
      "loss": 0.0624,
      "theoretical_loss": 3.326275377018053,
      "tokens_seen": 3285975040
    },
    {
      "epoch": 1.0,
      "learning_rate": 4.172684962285347e-06,
      "loss": 0.0625,
      "theoretical_loss": 3.326255522002572,
      "tokens_seen": 3286237184
    },
    {
      "epoch": 1.0,
      "learning_rate": 4.0924410207029375e-06,
      "loss": 0.0635,
      "theoretical_loss": 3.3262356690143022,
      "tokens_seen": 3286499328
    },
    {
      "epoch": 1.0,
      "learning_rate": 4.012197079120526e-06,
      "loss": 0.0633,
      "theoretical_loss": 3.3262158180528756,
      "tokens_seen": 3286761472
    },
    {
      "epoch": 1.0,
      "learning_rate": 3.931953137538115e-06,
      "loss": 0.0604,
      "theoretical_loss": 3.326195969117923,
      "tokens_seen": 3287023616
    },
    {
      "epoch": 1.0,
      "learning_rate": 3.851709195955705e-06,
      "loss": 0.0644,
      "theoretical_loss": 3.326176122209076,
      "tokens_seen": 3287285760
    },
    {
      "epoch": 1.0,
      "learning_rate": 3.771465254373295e-06,
      "loss": 0.0656,
      "theoretical_loss": 3.3261562773259667,
      "tokens_seen": 3287547904
    },
    {
      "epoch": 1.0,
      "learning_rate": 3.691221312790884e-06,
      "loss": 0.0643,
      "theoretical_loss": 3.3261364344682267,
      "tokens_seen": 3287810048
    },
    {
      "epoch": 1.0,
      "learning_rate": 3.610977371208474e-06,
      "loss": 0.0638,
      "theoretical_loss": 3.3261165936354873,
      "tokens_seen": 3288072192
    },
    {
      "epoch": 1.0,
      "learning_rate": 3.5307334296260636e-06,
      "loss": 0.0627,
      "theoretical_loss": 3.3260967548273808,
      "tokens_seen": 3288334336
    },
    {
      "epoch": 1.0,
      "learning_rate": 3.4504894880436525e-06,
      "loss": 0.0626,
      "theoretical_loss": 3.326076918043539,
      "tokens_seen": 3288596480
    },
    {
      "epoch": 1.0,
      "learning_rate": 3.3702455464612422e-06,
      "loss": 0.0631,
      "theoretical_loss": 3.3260570832835943,
      "tokens_seen": 3288858624
    },
    {
      "epoch": 1.0,
      "learning_rate": 3.290001604878832e-06,
      "loss": 0.0633,
      "theoretical_loss": 3.326037250547179,
      "tokens_seen": 3289120768
    },
    {
      "epoch": 1.0,
      "learning_rate": 3.209757663296421e-06,
      "loss": 0.0602,
      "theoretical_loss": 3.326017419833925,
      "tokens_seen": 3289382912
    },
    {
      "epoch": 1.0,
      "learning_rate": 3.1295137217140106e-06,
      "loss": 0.0603,
      "theoretical_loss": 3.3259975911434654,
      "tokens_seen": 3289645056
    },
    {
      "epoch": 1.0,
      "objective/train/advantage_avg": 0.0004319137951824814,
      "objective/train/docs_used": 1195204,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.2409605979919434,
      "objective/train/original_loss": 1.2409604787826538,
      "objective/train/theoretical_loss": 3.3259777644754323,
      "objective/train/tokens_used": 3310367200,
      "objective/train/value_avg": -0.005619049072265625,
      "objective/train/value_loss": 0.00011758451000787318,
      "objective/train/value_max": -2.1457672119140625e-05,
      "objective/train/value_min": -0.44287109375,
      "objective/train/value_reward_corr": 0.7364096476912639,
      "objective/train/value_std": 0.0122528076171875,
      "objective/train/weight_avg": 1.0004870891571045,
      "objective/train/weighted_lm_loss": 1.2417213916778564,
      "objective/train/weights_max": 1.5326497554779053,
      "objective/train/weights_min": 0.37251704931259155,
      "theoretical_loss": 3.3259777644754323,
      "tokens_seen": 3289907200
    },
    {
      "epoch": 1.0,
      "learning_rate": 3.0492697801316004e-06,
      "loss": 0.063,
      "theoretical_loss": 3.3259777644754323,
      "tokens_seen": 3289907200
    },
    {
      "epoch": 1.0,
      "learning_rate": 2.9690258385491897e-06,
      "loss": 0.0647,
      "theoretical_loss": 3.3259579398294576,
      "tokens_seen": 3290169344
    },
    {
      "epoch": 1.0,
      "learning_rate": 2.888781896966779e-06,
      "loss": 0.0621,
      "theoretical_loss": 3.325938117205175,
      "tokens_seen": 3290431488
    },
    {
      "epoch": 1.0,
      "learning_rate": 2.8085379553843687e-06,
      "loss": 0.0612,
      "theoretical_loss": 3.3259182966022167,
      "tokens_seen": 3290693632
    },
    {
      "epoch": 1.0,
      "learning_rate": 2.728294013801958e-06,
      "loss": 0.0617,
      "theoretical_loss": 3.3258984780202163,
      "tokens_seen": 3290955776
    },
    {
      "epoch": 1.0,
      "learning_rate": 2.6480500722195474e-06,
      "loss": 0.0622,
      "theoretical_loss": 3.325878661458806,
      "tokens_seen": 3291217920
    },
    {
      "epoch": 1.0,
      "learning_rate": 2.5678061306371367e-06,
      "loss": 0.0639,
      "theoretical_loss": 3.325858846917619,
      "tokens_seen": 3291480064
    },
    {
      "epoch": 1.0,
      "learning_rate": 2.4875621890547264e-06,
      "loss": 0.0626,
      "theoretical_loss": 3.3258390343962887,
      "tokens_seen": 3291742208
    },
    {
      "epoch": 1.0,
      "learning_rate": 2.407318247472316e-06,
      "loss": 0.0638,
      "theoretical_loss": 3.3258192238944484,
      "tokens_seen": 3292004352
    },
    {
      "epoch": 1.0,
      "learning_rate": 2.327074305889905e-06,
      "loss": 0.0606,
      "theoretical_loss": 3.3257994154117316,
      "tokens_seen": 3292266496
    },
    {
      "epoch": 1.0,
      "learning_rate": 2.246830364307495e-06,
      "loss": 0.0625,
      "theoretical_loss": 3.325779608947771,
      "tokens_seen": 3292528640
    },
    {
      "epoch": 1.0,
      "learning_rate": 2.1665864227250846e-06,
      "loss": 0.0625,
      "theoretical_loss": 3.3257598045022014,
      "tokens_seen": 3292790784
    },
    {
      "epoch": 1.0,
      "learning_rate": 2.0863424811426735e-06,
      "loss": 0.0638,
      "theoretical_loss": 3.3257400020746553,
      "tokens_seen": 3293052928
    },
    {
      "epoch": 1.0,
      "learning_rate": 2.006098539560263e-06,
      "loss": 0.0632,
      "theoretical_loss": 3.325720201664767,
      "tokens_seen": 3293315072
    },
    {
      "epoch": 1.0,
      "learning_rate": 1.9258545979778525e-06,
      "loss": 0.0623,
      "theoretical_loss": 3.32570040327217,
      "tokens_seen": 3293577216
    },
    {
      "epoch": 1.0,
      "learning_rate": 1.845610656395442e-06,
      "loss": 0.0625,
      "theoretical_loss": 3.325680606896499,
      "tokens_seen": 3293839360
    },
    {
      "epoch": 1.0,
      "learning_rate": 1.7653667148130318e-06,
      "loss": 0.0639,
      "theoretical_loss": 3.325660812537387,
      "tokens_seen": 3294101504
    },
    {
      "epoch": 1.0,
      "learning_rate": 1.6851227732306211e-06,
      "loss": 0.0648,
      "theoretical_loss": 3.3256410201944693,
      "tokens_seen": 3294363648
    },
    {
      "epoch": 1.0,
      "learning_rate": 1.6048788316482104e-06,
      "loss": 0.063,
      "theoretical_loss": 3.325621229867379,
      "tokens_seen": 3294625792
    },
    {
      "epoch": 1.0,
      "learning_rate": 1.5246348900658002e-06,
      "loss": 0.0644,
      "theoretical_loss": 3.3256014415557504,
      "tokens_seen": 3294887936
    },
    {
      "epoch": 1.0,
      "learning_rate": 1.4443909484833895e-06,
      "loss": 0.0631,
      "theoretical_loss": 3.325581655259219,
      "tokens_seen": 3295150080
    },
    {
      "epoch": 1.0,
      "learning_rate": 1.364147006900979e-06,
      "loss": 0.0624,
      "theoretical_loss": 3.3255618709774186,
      "tokens_seen": 3295412224
    },
    {
      "epoch": 1.0,
      "learning_rate": 1.2839030653185684e-06,
      "loss": 0.0656,
      "theoretical_loss": 3.3255420887099842,
      "tokens_seen": 3295674368
    },
    {
      "epoch": 1.0,
      "learning_rate": 1.203659123736158e-06,
      "loss": 0.0612,
      "theoretical_loss": 3.32552230845655,
      "tokens_seen": 3295936512
    },
    {
      "epoch": 1.0,
      "learning_rate": 1.1234151821537474e-06,
      "loss": 0.0634,
      "theoretical_loss": 3.3255025302167507,
      "tokens_seen": 3296198656
    },
    {
      "epoch": 1.0,
      "objective/train/advantage_avg": 0.00032849638955667615,
      "objective/train/docs_used": 1197549,
      "objective/train/instantaneous_batch_size": 64,
      "objective/train/instantaneous_microbatch_size": 65536,
      "objective/train/lm_loss": 1.261033296585083,
      "objective/train/original_loss": 1.2610334157943726,
      "objective/train/theoretical_loss": 3.3254827539902223,
      "objective/train/tokens_used": 3316920800,
      "objective/train/value_avg": -0.0079345703125,
      "objective/train/value_loss": 0.00026692645042203367,
      "objective/train/value_max": -2.6285648345947266e-05,
      "objective/train/value_min": -0.73681640625,
      "objective/train/value_reward_corr": 0.7701488647382327,
      "objective/train/value_std": 0.0187225341796875,
      "objective/train/weight_avg": 1.000449299812317,
      "objective/train/weighted_lm_loss": 1.2615493535995483,
      "objective/train/weights_max": 1.6513952016830444,
      "objective/train/weights_min": 0.36833932995796204,
      "theoretical_loss": 3.3254827539902223,
      "tokens_seen": 3296460800
    },
    {
      "epoch": 1.0,
      "learning_rate": 1.0431712405713367e-06,
      "loss": 0.063,
      "theoretical_loss": 3.3254827539902223,
      "tokens_seen": 3296460800
    },
    {
      "epoch": 1.0,
      "learning_rate": 9.629272989889263e-07,
      "loss": 0.0665,
      "theoretical_loss": 3.3254629797765984,
      "tokens_seen": 3296722944
    },
    {
      "epoch": 1.0,
      "learning_rate": 8.826833574065159e-07,
      "loss": 0.0624,
      "theoretical_loss": 3.325443207575515,
      "tokens_seen": 3296985088
    },
    {
      "epoch": 1.0,
      "learning_rate": 8.024394158241052e-07,
      "loss": 0.0681,
      "theoretical_loss": 3.3254234373866067,
      "tokens_seen": 3297247232
    },
    {
      "epoch": 1.0,
      "learning_rate": 7.221954742416948e-07,
      "loss": 0.0619,
      "theoretical_loss": 3.325403669209509,
      "tokens_seen": 3297509376
    },
    {
      "epoch": 1.0,
      "learning_rate": 6.419515326592842e-07,
      "loss": 0.0643,
      "theoretical_loss": 3.3253839030438574,
      "tokens_seen": 3297771520
    },
    {
      "epoch": 1.0,
      "learning_rate": 5.617075910768737e-07,
      "loss": 0.0637,
      "theoretical_loss": 3.325364138889287,
      "tokens_seen": 3298033664
    },
    {
      "epoch": 1.0,
      "learning_rate": 4.814636494944631e-07,
      "loss": 0.0675,
      "theoretical_loss": 3.3253443767454343,
      "tokens_seen": 3298295808
    },
    {
      "epoch": 1.0,
      "learning_rate": 4.012197079120526e-07,
      "loss": 0.0648,
      "theoretical_loss": 3.325324616611934,
      "tokens_seen": 3298557952
    },
    {
      "epoch": 1.0,
      "learning_rate": 3.209757663296421e-07,
      "loss": 0.0614,
      "theoretical_loss": 3.325304858488422,
      "tokens_seen": 3298820096
    },
    {
      "epoch": 1.0,
      "learning_rate": 2.4073182474723157e-07,
      "loss": 0.0608,
      "theoretical_loss": 3.3252851023745347,
      "tokens_seen": 3299082240
    },
    {
      "epoch": 1.0,
      "learning_rate": 1.6048788316482104e-07,
      "loss": 0.062,
      "theoretical_loss": 3.325265348269907,
      "tokens_seen": 3299344384
    },
    {
      "epoch": 1.0,
      "learning_rate": 8.024394158241052e-08,
      "loss": 0.0618,
      "theoretical_loss": 3.325245596174176,
      "tokens_seen": 3299606528
    },
    {
      "epoch": 1.0,
      "learning_rate": 0.0,
      "loss": 0.0602,
      "theoretical_loss": 3.325225846086977,
      "tokens_seen": 3299868672
    }
  ],
  "max_steps": 12588,
  "num_train_epochs": 9223372036854775807,
  "total_flos": 1.6840569843886326e+18,
  "trial_name": null,
  "trial_params": null
}