diff --git "a/checkpoint-5000/trainer_state.json" "b/checkpoint-5000/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/checkpoint-5000/trainer_state.json"
@@ -0,0 +1,37325 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.09929697740000794,
+  "global_step": 5000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "learning_rate": 1.984126984126984e-06,
+      "loss": 11.0771,
+      "theoretical_loss": 20.841282884190854,
+      "tokens_seen": 65536
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.968253968253968e-06,
+      "loss": 11.0714,
+      "theoretical_loss": 17.594666186972347,
+      "tokens_seen": 131072
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 5.9523809523809525e-06,
+      "loss": 10.9477,
+      "theoretical_loss": 15.967942175480273,
+      "tokens_seen": 196608
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 7.936507936507936e-06,
+      "loss": 10.6658,
+      "theoretical_loss": 14.920781838632275,
+      "tokens_seen": 262144
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.92063492063492e-06,
+      "loss": 10.3389,
+      "theoretical_loss": 14.16468202040668,
+      "tokens_seen": 327680
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 1.1904761904761905e-05,
+      "loss": 10.0572,
+      "theoretical_loss": 13.581026555193928,
+      "tokens_seen": 393216
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 1.3888888888888888e-05,
+      "loss": 9.8012,
+      "theoretical_loss": 13.110274084044894,
+      "tokens_seen": 458752
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 1.5873015873015872e-05,
+      "loss": 9.6269,
+      "theoretical_loss": 12.718594708127029,
+      "tokens_seen": 524288
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 1.7857142857142855e-05,
+      "loss": 9.5127,
+      "theoretical_loss": 12.385057546517329,
+      "tokens_seen": 589824
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 1.984126984126984e-05,
+      "loss": 9.3822,
+      "theoretical_loss": 12.095877689678783,
+      "tokens_seen": 655360
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 2.1825396825396824e-05,
+      "loss": 9.3134,
+      "theoretical_loss": 11.841531313745381,
+      "tokens_seen": 720896
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 2.380952380952381e-05,
+      "loss": 9.19,
+      "theoretical_loss": 11.615184291350435,
+      "tokens_seen": 786432
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 2.5793650793650793e-05,
+      "loss": 9.1376,
+      "theoretical_loss": 11.411779222830493,
+      "tokens_seen": 851968
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 2.7777777777777776e-05,
+      "loss": 9.0561,
+      "theoretical_loss": 11.227476784755577,
+      "tokens_seen": 917504
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 2.9761904761904762e-05,
+      "loss": 8.9674,
+      "theoretical_loss": 11.059299000621618,
+      "tokens_seen": 983040
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.1746031746031745e-05,
+      "loss": 8.9181,
+      "theoretical_loss": 10.904893169100655,
+      "tokens_seen": 1048576
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.373015873015873e-05,
+      "loss": 8.9953,
+      "theoretical_loss": 10.76237082270643,
+      "tokens_seen": 1114112
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.571428571428571e-05,
+      "loss": 8.7793,
+      "theoretical_loss": 10.630194958873984,
+      "tokens_seen": 1179648
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.76984126984127e-05,
+      "loss": 8.7566,
+      "theoretical_loss": 10.507099255000346,
+      "tokens_seen": 1245184
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.968253968253968e-05,
+      "loss": 8.7115,
+      "theoretical_loss": 10.392029026407034,
+      "tokens_seen": 1310720
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 4.1666666666666665e-05,
+      "loss": 8.487,
+      "theoretical_loss": 10.28409730354045,
+      "tokens_seen": 1376256
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 4.365079365079365e-05,
+      "loss": 8.5783,
+      "theoretical_loss": 10.182551635913724,
+      "tokens_seen": 1441792
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 4.563492063492063e-05,
+      "loss": 8.5185,
+      "theoretical_loss": 10.086748644376303,
+      "tokens_seen": 1507328
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 4.761904761904762e-05,
+      "loss": 8.4859,
+      "theoretical_loss": 9.996134261483984,
+      "tokens_seen": 1572864
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 4.96031746031746e-05,
+      "loss": 8.3094,
+      "theoretical_loss": 9.910228209036813,
+      "tokens_seen": 1638400
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 5.1587301587301586e-05,
+      "loss": 8.4173,
+      "theoretical_loss": 9.828611674184266,
+      "tokens_seen": 1703936
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 5.357142857142857e-05,
+      "loss": 8.3191,
+      "theoretical_loss": 9.750917429551151,
+      "tokens_seen": 1769472
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 5.555555555555555e-05,
+      "loss": 8.3262,
+      "theoretical_loss": 9.67682184172525,
+      "tokens_seen": 1835008
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 5.753968253968254e-05,
+      "loss": 8.1858,
+      "theoretical_loss": 9.606038353789344,
+      "tokens_seen": 1900544
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 5.9523809523809524e-05,
+      "loss": 8.1585,
+      "theoretical_loss": 9.538312129408556,
+      "tokens_seen": 1966080
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 6.15079365079365e-05,
+      "loss": 8.1146,
+      "theoretical_loss": 9.47341562027564,
+      "tokens_seen": 2031616
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 6.349206349206349e-05,
+      "loss": 8.0534,
+      "theoretical_loss": 9.41114487355416,
+      "tokens_seen": 2097152
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 6.547619047619048e-05,
+      "loss": 7.962,
+      "theoretical_loss": 9.351316436877685,
+      "tokens_seen": 2162688
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 6.746031746031745e-05,
+      "loss": 7.8543,
+      "theoretical_loss": 9.29376474930398,
+      "tokens_seen": 2228224
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 6.944444444444444e-05,
+      "loss": 8.0138,
+      "theoretical_loss": 9.238339930088461,
+      "tokens_seen": 2293760
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 7.142857142857142e-05,
+      "loss": 7.7162,
+      "theoretical_loss": 9.184905895151996,
+      "tokens_seen": 2359296
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 7.341269841269842e-05,
+      "loss": 7.7563,
+      "theoretical_loss": 9.133338745057056,
+      "tokens_seen": 2424832
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 7.53968253968254e-05,
+      "loss": 7.5792,
+      "theoretical_loss": 9.083525379177438,
+      "tokens_seen": 2490368
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 7.738095238095239e-05,
+      "loss": 7.591,
+      "theoretical_loss": 9.035362299287534,
+      "tokens_seen": 2555904
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 7.936507936507937e-05,
+      "loss": 7.6518,
+      "theoretical_loss": 8.988754572553061,
+      "tokens_seen": 2621440
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 8.134920634920635e-05,
+      "loss": 7.4322,
+      "theoretical_loss": 8.943614929284315,
+      "tokens_seen": 2686976
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 8.333333333333333e-05,
+      "loss": 7.4551,
+      "theoretical_loss": 8.899862975121929,
+      "tokens_seen": 2752512
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 8.531746031746032e-05,
+      "loss": 7.24,
+      "theoretical_loss": 8.857424500797205,
+      "tokens_seen": 2818048
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 8.73015873015873e-05,
+      "loss": 7.2072,
+      "theoretical_loss": 8.816230875422118,
+      "tokens_seen": 2883584
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 8.928571428571429e-05,
+      "loss": 7.3823,
+      "theoretical_loss": 8.776218511555161,
+      "tokens_seen": 2949120
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.126984126984126e-05,
+      "loss": 7.1742,
+      "theoretical_loss": 8.737328392164539,
+      "tokens_seen": 3014656
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.325396825396825e-05,
+      "loss": 7.1599,
+      "theoretical_loss": 8.699505651152553,
+      "tokens_seen": 3080192
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.523809523809524e-05,
+      "loss": 7.1427,
+      "theoretical_loss": 8.66269920037918,
+      "tokens_seen": 3145728
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.722222222222223e-05,
+      "loss": 7.0218,
+      "theoretical_loss": 8.62686139718003,
+      "tokens_seen": 3211264
+    },
+    {
+      "epoch": 0.0,
+      "objective/train/advantage_avg": 0.49531280994415283,
+      "objective/train/docs_used": 8371,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 6.728449821472168,
+      "objective/train/original_loss": 6.728449821472168,
+      "objective/train/theoretical_loss": 8.591947747254773,
+      "objective/train/tokens_used": 23736800,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24548619985580444,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0507793426513672,
+      "objective/train/weighted_lm_loss": 7.069992542266846,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 1.0216598510742188,
+      "theoretical_loss": 8.591947747254773,
+      "tokens_seen": 3276800
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.92063492063492e-05,
+      "loss": 6.9774,
+      "theoretical_loss": 8.591947747254773,
+      "tokens_seen": 3276800
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0001011904761904762,
+      "loss": 6.8521,
+      "theoretical_loss": 8.55791663853882,
+      "tokens_seen": 3342336
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00010317460317460317,
+      "loss": 6.7841,
+      "theoretical_loss": 8.524729102289708,
+      "tokens_seen": 3407872
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00010515873015873016,
+      "loss": 6.8152,
+      "theoretical_loss": 8.492348598140893,
+      "tokens_seen": 3473408
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00010714285714285714,
+      "loss": 6.7547,
+      "theoretical_loss": 8.460740820316484,
+      "tokens_seen": 3538944
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00010912698412698413,
+      "loss": 6.7581,
+      "theoretical_loss": 8.429873522574422,
+      "tokens_seen": 3604480
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0001111111111111111,
+      "loss": 6.7221,
+      "theoretical_loss": 8.399716359763914,
+      "tokens_seen": 3670016
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00011309523809523809,
+      "loss": 6.6347,
+      "theoretical_loss": 8.370240744154781,
+      "tokens_seen": 3735552
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00011507936507936508,
+      "loss": 6.6158,
+      "theoretical_loss": 8.341419714929035,
+      "tokens_seen": 3801088
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00011706349206349207,
+      "loss": 6.4159,
+      "theoretical_loss": 8.313227819424887,
+      "tokens_seen": 3866624
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00011904761904761905,
+      "loss": 6.4367,
+      "theoretical_loss": 8.285641004895568,
+      "tokens_seen": 3932160
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00012103174603174604,
+      "loss": 6.5093,
+      "theoretical_loss": 8.25863651969382,
+      "tokens_seen": 3997696
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.000123015873015873,
+      "loss": 6.3352,
+      "theoretical_loss": 8.232192822921675,
+      "tokens_seen": 4063232
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.000125,
+      "loss": 6.3951,
+      "theoretical_loss": 8.206289501696752,
+      "tokens_seen": 4128768
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00012698412698412698,
+      "loss": 6.4355,
+      "theoretical_loss": 8.180907195283321,
+      "tokens_seen": 4194304
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00012896825396825398,
+      "loss": 6.2604,
+      "theoretical_loss": 8.156027525421205,
+      "tokens_seen": 4259840
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00013095238095238096,
+      "loss": 6.2779,
+      "theoretical_loss": 8.131633032259415,
+      "tokens_seen": 4325376
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00013293650793650793,
+      "loss": 6.3027,
+      "theoretical_loss": 8.107707115366345,
+      "tokens_seen": 4390912
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0001349206349206349,
+      "loss": 6.2102,
+      "theoretical_loss": 8.084233979345122,
+      "tokens_seen": 4456448
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0001369047619047619,
+      "loss": 6.1439,
+      "theoretical_loss": 8.061198583632713,
+      "tokens_seen": 4521984
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0001388888888888889,
+      "loss": 6.1627,
+      "theoretical_loss": 8.03858659610554,
+      "tokens_seen": 4587520
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00014087301587301586,
+      "loss": 5.9802,
+      "theoretical_loss": 8.016384350153084,
+      "tokens_seen": 4653056
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00014285714285714284,
+      "loss": 6.1413,
+      "theoretical_loss": 7.9945788049155055,
+      "tokens_seen": 4718592
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00014484126984126984,
+      "loss": 6.2189,
+      "theoretical_loss": 7.973157508411693,
+      "tokens_seen": 4784128
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00014682539682539685,
+      "loss": 6.0561,
+      "theoretical_loss": 7.952108563311223,
+      "tokens_seen": 4849664
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00014880952380952382,
+      "loss": 6.1682,
+      "theoretical_loss": 7.931420595127772,
+      "tokens_seen": 4915200
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0001507936507936508,
+      "loss": 6.0328,
+      "theoretical_loss": 7.911082722632908,
+      "tokens_seen": 4980736
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0001527777777777778,
+      "loss": 5.9525,
+      "theoretical_loss": 7.891084530308333,
+      "tokens_seen": 5046272
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00015476190476190478,
+      "loss": 6.0668,
+      "theoretical_loss": 7.871416042671642,
+      "tokens_seen": 5111808
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00015674603174603175,
+      "loss": 6.054,
+      "theoretical_loss": 7.8520677003260175,
+      "tokens_seen": 5177344
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00015873015873015873,
+      "loss": 5.8692,
+      "theoretical_loss": 7.83303033759787,
+      "tokens_seen": 5242880
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00016071428571428573,
+      "loss": 5.9204,
+      "theoretical_loss": 7.8142951616387775,
+      "tokens_seen": 5308416
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0001626984126984127,
+      "loss": 5.8646,
+      "theoretical_loss": 7.795853732879069,
+      "tokens_seen": 5373952
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00016468253968253969,
+      "loss": 6.0267,
+      "theoretical_loss": 7.777697946730324,
+      "tokens_seen": 5439488
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00016666666666666666,
+      "loss": 6.0271,
+      "theoretical_loss": 7.759820016443023,
+      "tokens_seen": 5505024
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00016865079365079366,
+      "loss": 5.9168,
+      "theoretical_loss": 7.742212457033625,
+      "tokens_seen": 5570560
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00017063492063492064,
+      "loss": 5.8456,
+      "theoretical_loss": 7.724868070202639,
+      "tokens_seen": 5636096
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00017261904761904762,
+      "loss": 5.7787,
+      "theoretical_loss": 7.707779930171886,
+      "tokens_seen": 5701632
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0001746031746031746,
+      "loss": 5.7974,
+      "theoretical_loss": 7.690941370375033,
+      "tokens_seen": 5767168
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0001765873015873016,
+      "loss": 5.8043,
+      "theoretical_loss": 7.674345970940984,
+      "tokens_seen": 5832704
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00017857142857142857,
+      "loss": 5.7159,
+      "theoretical_loss": 7.6579875469145655,
+      "tokens_seen": 5898240
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00018055555555555555,
+      "loss": 5.8569,
+      "theoretical_loss": 7.641860137163401,
+      "tokens_seen": 5963776
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00018253968253968252,
+      "loss": 5.8218,
+      "theoretical_loss": 7.6259579939239845,
+      "tokens_seen": 6029312
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00018452380952380953,
+      "loss": 5.8325,
+      "theoretical_loss": 7.610275572943593,
+      "tokens_seen": 6094848
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0001865079365079365,
+      "loss": 5.7266,
+      "theoretical_loss": 7.5948075241780755,
+      "tokens_seen": 6160384
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0001884920634920635,
+      "loss": 5.7308,
+      "theoretical_loss": 7.579548683008654,
+      "tokens_seen": 6225920
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00019047619047619048,
+      "loss": 5.8053,
+      "theoretical_loss": 7.564494061943624,
+      "tokens_seen": 6291456
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00019246031746031748,
+      "loss": 5.7645,
+      "theoretical_loss": 7.549638842773467,
+      "tokens_seen": 6356992
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00019444444444444446,
+      "loss": 5.6472,
+      "theoretical_loss": 7.534978369150202,
+      "tokens_seen": 6422528
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00019642857142857144,
+      "loss": 5.6673,
+      "theoretical_loss": 7.520508139563954,
+      "tokens_seen": 6488064
+    },
+    {
+      "epoch": 0.0,
+      "objective/train/advantage_avg": 0.4908171594142914,
+      "objective/train/docs_used": 9704,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 5.399485111236572,
+      "objective/train/original_loss": 5.3994855880737305,
+      "objective/train/theoretical_loss": 7.5062238006917354,
+      "objective/train/tokens_used": 27013600,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2412356734275818,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503079891204834,
+      "objective/train/weighted_lm_loss": 5.670967102050781,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 7.5062238006917354,
+      "tokens_seen": 6553600
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0001984126984126984,
+      "loss": 5.7328,
+      "theoretical_loss": 7.5062238006917354,
+      "tokens_seen": 6553600
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00020039682539682542,
+      "loss": 5.7643,
+      "theoretical_loss": 7.492121141095167,
+      "tokens_seen": 6619136
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0002023809523809524,
+      "loss": 5.5097,
+      "theoretical_loss": 7.478196085245621,
+      "tokens_seen": 6684672
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00020436507936507937,
+      "loss": 5.6864,
+      "theoretical_loss": 7.464444687856724,
+      "tokens_seen": 6750208
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00020634920634920634,
+      "loss": 5.7318,
+      "theoretical_loss": 7.45086312850561,
+      "tokens_seen": 6815744
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00020833333333333335,
+      "loss": 5.4749,
+      "theoretical_loss": 7.43744770652558,
+      "tokens_seen": 6881280
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00021031746031746032,
+      "loss": 5.5339,
+      "theoretical_loss": 7.42419483615405,
+      "tokens_seen": 6946816
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0002123015873015873,
+      "loss": 5.6535,
+      "theoretical_loss": 7.4111010419207375,
+      "tokens_seen": 7012352
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00021428571428571427,
+      "loss": 5.7321,
+      "theoretical_loss": 7.398162954262078,
+      "tokens_seen": 7077888
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00021626984126984128,
+      "loss": 5.5421,
+      "theoretical_loss": 7.385377305348793,
+      "tokens_seen": 7143424
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00021825396825396825,
+      "loss": 5.5598,
+      "theoretical_loss": 7.3727409251143925,
+      "tokens_seen": 7208960
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00022023809523809523,
+      "loss": 5.5974,
+      "theoretical_loss": 7.360250737473224,
+      "tokens_seen": 7274496
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0002222222222222222,
+      "loss": 5.5628,
+      "theoretical_loss": 7.347903756717382,
+      "tokens_seen": 7340032
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0002242063492063492,
+      "loss": 5.5724,
+      "theoretical_loss": 7.335697084082529,
+      "tokens_seen": 7405568
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00022619047619047618,
+      "loss": 5.6969,
+      "theoretical_loss": 7.323627904473269,
+      "tokens_seen": 7471104
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00022817460317460316,
+      "loss": 5.3875,
+      "theoretical_loss": 7.311693483339357,
+      "tokens_seen": 7536640
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00023015873015873016,
+      "loss": 5.4919,
+      "theoretical_loss": 7.299891163694537,
+      "tokens_seen": 7602176
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00023214285714285717,
+      "loss": 5.5971,
+      "theoretical_loss": 7.288218363270328,
+      "tokens_seen": 7667712
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00023412698412698414,
+      "loss": 5.5551,
+      "theoretical_loss": 7.276672571797575,
+      "tokens_seen": 7733248
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00023611111111111112,
+      "loss": 5.5232,
+      "theoretical_loss": 7.265251348408967,
+      "tokens_seen": 7798784
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0002380952380952381,
+      "loss": 5.6741,
+      "theoretical_loss": 7.253952319156202,
+      "tokens_seen": 7864320
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0002400793650793651,
+      "loss": 5.5124,
+      "theoretical_loss": 7.242773174635825,
+      "tokens_seen": 7929856
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00024206349206349207,
+      "loss": 5.6728,
+      "theoretical_loss": 7.2317116677181055,
+      "tokens_seen": 7995392
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00024404761904761905,
+      "loss": 5.3354,
+      "theoretical_loss": 7.22076561137372,
+      "tokens_seen": 8060928
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.000246031746031746,
+      "loss": 5.5085,
+      "theoretical_loss": 7.2099328765932205,
+      "tokens_seen": 8126464
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.000248015873015873,
+      "loss": 5.5368,
+      "theoretical_loss": 7.1992113903946695,
+      "tokens_seen": 8192000
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00025,
+      "loss": 5.4071,
+      "theoretical_loss": 7.188599133914977,
+      "tokens_seen": 8257536
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.000251984126984127,
+      "loss": 5.4527,
+      "theoretical_loss": 7.178094140580843,
+      "tokens_seen": 8323072
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00025396825396825396,
+      "loss": 5.36,
+      "theoretical_loss": 7.167694494355343,
+      "tokens_seen": 8388608
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00025595238095238096,
+      "loss": 5.3949,
+      "theoretical_loss": 7.157398328056499,
+      "tokens_seen": 8454144
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00025793650793650796,
+      "loss": 5.4218,
+      "theoretical_loss": 7.147203821744321,
+      "tokens_seen": 8519680
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00025992063492063497,
+      "loss": 5.4023,
+      "theoretical_loss": 7.1371092011730495,
+      "tokens_seen": 8585216
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0002619047619047619,
+      "loss": 5.3711,
+      "theoretical_loss": 7.127112736305475,
+      "tokens_seen": 8650752
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0002638888888888889,
+      "loss": 5.2684,
+      "theoretical_loss": 7.117212739886398,
+      "tokens_seen": 8716288
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00026587301587301587,
+      "loss": 5.2483,
+      "theoretical_loss": 7.107407566072454,
+      "tokens_seen": 8781824
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00026785714285714287,
+      "loss": 5.5102,
+      "theoretical_loss": 7.097695609115671,
+      "tokens_seen": 8847360
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0002698412698412698,
+      "loss": 5.3072,
+      "theoretical_loss": 7.0880753020982725,
+      "tokens_seen": 8912896
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0002718253968253968,
+      "loss": 5.1466,
+      "theoretical_loss": 7.0785451157163735,
+      "tokens_seen": 8978432
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0002738095238095238,
+      "loss": 5.439,
+      "theoretical_loss": 7.06910355711033,
+      "tokens_seen": 9043968
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00027579365079365083,
+      "loss": 5.2897,
+      "theoretical_loss": 7.059749168739638,
+      "tokens_seen": 9109504
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0002777777777777778,
+      "loss": 5.3691,
+      "theoretical_loss": 7.050480527300383,
+      "tokens_seen": 9175040
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0002797619047619048,
+      "loss": 5.3555,
+      "theoretical_loss": 7.041296242683314,
+      "tokens_seen": 9240576
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00028174603174603173,
+      "loss": 5.3703,
+      "theoretical_loss": 7.032194956970776,
+      "tokens_seen": 9306112
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00028373015873015873,
+      "loss": 5.3154,
+      "theoretical_loss": 7.023175343470762,
+      "tokens_seen": 9371648
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0002857142857142857,
+      "loss": 5.3311,
+      "theoretical_loss": 7.014236105786485,
+      "tokens_seen": 9437184
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0002876984126984127,
+      "loss": 5.2603,
+      "theoretical_loss": 7.005375976919894,
+      "tokens_seen": 9502720
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0002896825396825397,
+      "loss": 5.3003,
+      "theoretical_loss": 6.99659371840772,
+      "tokens_seen": 9568256
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0002916666666666667,
+      "loss": 5.3527,
+      "theoretical_loss": 6.987888119488599,
+      "tokens_seen": 9633792
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0002936507936507937,
+      "loss": 5.3101,
+      "theoretical_loss": 6.979257996300014,
+      "tokens_seen": 9699328
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00029563492063492064,
+      "loss": 5.27,
+      "theoretical_loss": 6.970702191103734,
+      "tokens_seen": 9764864
+    },
+    {
+      "epoch": 0.0,
+      "objective/train/advantage_avg": 0.49107539653778076,
+      "objective/train/docs_used": 11015,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 5.094058036804199,
+      "objective/train/original_loss": 5.094058036804199,
+      "objective/train/theoretical_loss": 6.962219571538605,
+      "objective/train/tokens_used": 30290400,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24188564717769623,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503370761871338,
+      "objective/train/weighted_lm_loss": 5.350419521331787,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 6.962219571538605,
+      "tokens_seen": 9830400
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00029761904761904765,
+      "loss": 5.3296,
+      "theoretical_loss": 6.962219571538605,
+      "tokens_seen": 9830400
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0002996031746031746,
+      "loss": 5.3081,
+      "theoretical_loss": 6.953809029899522,
+      "tokens_seen": 9895936
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0003015873015873016,
+      "loss": 5.2714,
+      "theoretical_loss": 6.945469482441503,
+      "tokens_seen": 9961472
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00030357142857142855,
+      "loss": 5.2838,
+      "theoretical_loss": 6.937199868707852,
+      "tokens_seen": 10027008
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0003055555555555556,
+      "loss": 5.3265,
+      "theoretical_loss": 6.928999150881393,
+      "tokens_seen": 10092544
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00030753968253968255,
+      "loss": 5.2493,
+      "theoretical_loss": 6.920866313157865,
+      "tokens_seen": 10158080
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00030952380952380956,
+      "loss": 5.3069,
+      "theoretical_loss": 6.912800361140576,
+      "tokens_seen": 10223616
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0003115079365079365,
+      "loss": 5.4451,
+      "theoretical_loss": 6.904800321255455,
+      "tokens_seen": 10289152
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0003134920634920635,
+      "loss": 5.1631,
+      "theoretical_loss": 6.896865240185702,
+      "tokens_seen": 10354688
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00031547619047619046,
+      "loss": 5.2823,
+      "theoretical_loss": 6.888994184325256,
+      "tokens_seen": 10420224
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00031746031746031746,
+      "loss": 5.2868,
+      "theoretical_loss": 6.881186239250335,
+      "tokens_seen": 10485760
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0003194444444444444,
+      "loss": 5.2119,
+      "theoretical_loss": 6.873440509208347,
+      "tokens_seen": 10551296
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00032142857142857147,
+      "loss": 5.414,
+      "theoretical_loss": 6.8657561166234995,
+      "tokens_seen": 10616832
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0003234126984126984,
+      "loss": 5.0445,
+      "theoretical_loss": 6.858132201618454,
+      "tokens_seen": 10682368
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0003253968253968254,
+      "loss": 5.1709,
+      "theoretical_loss": 6.8505679215514235,
+      "tokens_seen": 10747904
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00032738095238095237,
+      "loss": 5.1904,
+      "theoretical_loss": 6.8430624505681,
+      "tokens_seen": 10813440
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00032936507936507937,
+      "loss": 5.2185,
+      "theoretical_loss": 6.835614979167886,
+      "tokens_seen": 10878976
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0003313492063492063,
+      "loss": 5.0221,
+      "theoretical_loss": 6.82822471378385,
+      "tokens_seen": 10944512
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0003333333333333333,
+      "loss": 5.1341,
+      "theoretical_loss": 6.8208908763759295,
+      "tokens_seen": 11010048
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0003353174603174603,
+      "loss": 4.9061,
+      "theoretical_loss": 6.813612704036853,
+      "tokens_seen": 11075584
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00033730158730158733,
+      "loss": 5.1869,
+      "theoretical_loss": 6.806389448610343,
+      "tokens_seen": 11141120
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00033928571428571433,
+      "loss": 5.1124,
+      "theoretical_loss": 6.799220376321128,
+      "tokens_seen": 11206656
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0003412698412698413,
+      "loss": 5.2221,
+      "theoretical_loss": 6.79210476741633,
+      "tokens_seen": 11272192
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0003432539682539683,
+      "loss": 5.035,
+      "theoretical_loss": 6.785041915817828,
+      "tokens_seen": 11337728
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00034523809523809523,
+      "loss": 5.1263,
+      "theoretical_loss": 6.778031128785191,
+      "tokens_seen": 11403264
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00034722222222222224,
+      "loss": 5.2328,
+      "theoretical_loss": 6.771071726588795,
+      "tokens_seen": 11468800
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0003492063492063492,
+      "loss": 5.1011,
+      "theoretical_loss": 6.76416304219278,
+      "tokens_seen": 11534336
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0003511904761904762,
+      "loss": 5.0111,
+      "theoretical_loss": 6.757304420947466,
+      "tokens_seen": 11599872
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0003531746031746032,
+      "loss": 5.1246,
+      "theoretical_loss": 6.750495220290931,
+      "tokens_seen": 11665408
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0003551587301587302,
+      "loss": 5.0826,
+      "theoretical_loss": 6.743734809459394,
+      "tokens_seen": 11730944
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00035714285714285714,
+      "loss": 4.9246,
+      "theoretical_loss": 6.737022569206117,
+      "tokens_seen": 11796480
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00035912698412698415,
+      "loss": 4.9429,
+      "theoretical_loss": 6.730357891528529,
+      "tokens_seen": 11862016
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0003611111111111111,
+      "loss": 5.1395,
+      "theoretical_loss": 6.723740179403264,
+      "tokens_seen": 11927552
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0003630952380952381,
+      "loss": 5.1806,
+      "theoretical_loss": 6.717168846528885,
+      "tokens_seen": 11993088
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00036507936507936505,
+      "loss": 5.0837,
+      "theoretical_loss": 6.710643317075979,
+      "tokens_seen": 12058624
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00036706349206349205,
+      "loss": 5.0545,
+      "theoretical_loss": 6.70416302544442,
+      "tokens_seen": 12124160
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00036904761904761905,
+      "loss": 5.051,
+      "theoretical_loss": 6.6977274160275195,
+      "tokens_seen": 12189696
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00037103174603174606,
+      "loss": 4.936,
+      "theoretical_loss": 6.691335942982863,
+      "tokens_seen": 12255232
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.000373015873015873,
+      "loss": 5.2205,
+      "theoretical_loss": 6.684988070009584,
+      "tokens_seen": 12320768
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.000375,
+      "loss": 5.1985,
+      "theoretical_loss": 6.6786832701318755,
+      "tokens_seen": 12386304
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.000376984126984127,
+      "loss": 5.0665,
+      "theoretical_loss": 6.672421025488536,
+      "tokens_seen": 12451840
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00037896825396825396,
+      "loss": 5.2492,
+      "theoretical_loss": 6.666200827128331,
+      "tokens_seen": 12517376
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00038095238095238096,
+      "loss": 5.0901,
+      "theoretical_loss": 6.660022174811009,
+      "tokens_seen": 12582912
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00038293650793650797,
+      "loss": 5.0826,
+      "theoretical_loss": 6.653884576813754,
+      "tokens_seen": 12648448
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00038492063492063497,
+      "loss": 4.9834,
+      "theoretical_loss": 6.647787549742928,
+      "tokens_seen": 12713984
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0003869047619047619,
+      "loss": 5.0116,
+      "theoretical_loss": 6.641730618350916,
+      "tokens_seen": 12779520
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0003888888888888889,
+      "loss": 5.0323,
+      "theoretical_loss": 6.6357133153579175,
+      "tokens_seen": 12845056
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00039087301587301587,
+      "loss": 5.0115,
+      "theoretical_loss": 6.629735181278515,
+      "tokens_seen": 12910592
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0003928571428571429,
+      "loss": 4.9832,
+      "theoretical_loss": 6.623795764252885,
+      "tokens_seen": 12976128
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0003948412698412698,
+      "loss": 5.0157,
+      "theoretical_loss": 6.61789461988249,
+      "tokens_seen": 13041664
+    },
+    {
+      "epoch": 0.0,
+      "objective/train/advantage_avg": 0.4912474751472473,
+      "objective/train/docs_used": 12170,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 4.883951187133789,
+      "objective/train/original_loss": 4.883951187133789,
+      "objective/train/theoretical_loss": 6.612031311070119,
+      "objective/train/tokens_used": 33567200,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2418125718832016,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.05035400390625,
+      "objective/train/weighted_lm_loss": 5.129752159118652,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 6.612031311070119,
+      "tokens_seen": 13107200
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0003968253968253968,
+      "loss": 5.0655,
+      "theoretical_loss": 6.612031311070119,
+      "tokens_seen": 13107200
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00039880952380952383,
+      "loss": 4.8914,
+      "theoretical_loss": 6.6062054078641355,
+      "tokens_seen": 13172736
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00040079365079365083,
+      "loss": 5.1053,
+      "theoretical_loss": 6.600416487306803,
+      "tokens_seen": 13238272
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004027777777777778,
+      "loss": 4.9361,
+      "theoretical_loss": 6.594664133286568,
+      "tokens_seen": 13303808
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004047619047619048,
+      "loss": 5.1403,
+      "theoretical_loss": 6.588947936394168,
+      "tokens_seen": 13369344
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00040674603174603173,
+      "loss": 4.9681,
+      "theoretical_loss": 6.58326749378245,
+      "tokens_seen": 13434880
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00040873015873015874,
+      "loss": 5.061,
+      "theoretical_loss": 6.577622409029796,
+      "tokens_seen": 13500416
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004107142857142857,
+      "loss": 4.8943,
+      "theoretical_loss": 6.5720122920070345,
+      "tokens_seen": 13565952
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004126984126984127,
+      "loss": 4.9159,
+      "theoretical_loss": 6.566436758747731,
+      "tokens_seen": 13631488
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004146825396825397,
+      "loss": 4.8941,
+      "theoretical_loss": 6.560895431321773,
+      "tokens_seen": 13697024
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004166666666666667,
+      "loss": 4.8609,
+      "theoretical_loss": 6.5553879377121245,
+      "tokens_seen": 13762560
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004186507936507937,
+      "loss": 4.9139,
+      "theoretical_loss": 6.5499139116946825,
+      "tokens_seen": 13828096
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00042063492063492065,
+      "loss": 4.9303,
+      "theoretical_loss": 6.544472992721121,
+      "tokens_seen": 13893632
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00042261904761904765,
+      "loss": 4.9147,
+      "theoretical_loss": 6.539064825804655,
+      "tokens_seen": 13959168
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004246031746031746,
+      "loss": 5.0321,
+      "theoretical_loss": 6.533689061408615,
+      "tokens_seen": 14024704
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004265873015873016,
+      "loss": 4.9443,
+      "theoretical_loss": 6.528345355337773,
+      "tokens_seen": 14090240
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00042857142857142855,
+      "loss": 4.9354,
+      "theoretical_loss": 6.523033368632323,
+      "tokens_seen": 14155776
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004305555555555556,
+      "loss": 4.9561,
+      "theoretical_loss": 6.51775276746445,
+      "tokens_seen": 14221312
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00043253968253968256,
+      "loss": 4.9391,
+      "theoretical_loss": 6.512503223037404,
+      "tokens_seen": 14286848
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00043452380952380956,
+      "loss": 4.85,
+      "theoretical_loss": 6.507284411487015,
+      "tokens_seen": 14352384
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004365079365079365,
+      "loss": 4.9346,
+      "theoretical_loss": 6.502096013785574,
+      "tokens_seen": 14417920
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004384920634920635,
+      "loss": 4.8673,
+      "theoretical_loss": 6.496937715648013,
+      "tokens_seen": 14483456
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00044047619047619046,
+      "loss": 4.7758,
+      "theoretical_loss": 6.491809207440321,
+      "tokens_seen": 14548992
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00044246031746031746,
+      "loss": 4.8789,
+      "theoretical_loss": 6.486710184090135,
+      "tokens_seen": 14614528
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004444444444444444,
+      "loss": 4.8879,
+      "theoretical_loss": 6.481640344999435,
+      "tokens_seen": 14680064
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00044642857142857147,
+      "loss": 4.9789,
+      "theoretical_loss": 6.476599393959298,
+      "tokens_seen": 14745600
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004484126984126984,
+      "loss": 4.8676,
+      "theoretical_loss": 6.471587039066644,
+      "tokens_seen": 14811136
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004503968253968254,
+      "loss": 4.8724,
+      "theoretical_loss": 6.466602992642919,
+      "tokens_seen": 14876672
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00045238095238095237,
+      "loss": 4.9345,
+      "theoretical_loss": 6.461646971154669,
+      "tokens_seen": 14942208
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00045436507936507937,
+      "loss": 5.0052,
+      "theoretical_loss": 6.456718695135944,
+      "tokens_seen": 15007744
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004563492063492063,
+      "loss": 4.899,
+      "theoretical_loss": 6.45181788911249,
+      "tokens_seen": 15073280
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004583333333333333,
+      "loss": 4.7496,
+      "theoretical_loss": 6.446944281527674,
+      "tokens_seen": 15138816
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00046031746031746033,
+      "loss": 4.7412,
+      "theoretical_loss": 6.442097604670096,
+      "tokens_seen": 15204352
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00046230158730158733,
+      "loss": 4.6114,
+      "theoretical_loss": 6.437277594602845,
+      "tokens_seen": 15269888
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00046428571428571433,
+      "loss": 4.7578,
+      "theoretical_loss": 6.43248399109436,
+      "tokens_seen": 15335424
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004662698412698413,
+      "loss": 4.763,
+      "theoretical_loss": 6.427716537550838,
+      "tokens_seen": 15400960
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004682539682539683,
+      "loss": 4.7541,
+      "theoretical_loss": 6.422974980950157,
+      "tokens_seen": 15466496
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00047023809523809523,
+      "loss": 4.8955,
+      "theoretical_loss": 6.41825907177728,
+      "tokens_seen": 15532032
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00047222222222222224,
+      "loss": 4.9811,
+      "theoretical_loss": 6.413568563961092,
+      "tokens_seen": 15597568
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004742063492063492,
+      "loss": 4.7225,
+      "theoretical_loss": 6.408903214812629,
+      "tokens_seen": 15663104
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004761904761904762,
+      "loss": 4.8861,
+      "theoretical_loss": 6.404262784964672,
+      "tokens_seen": 15728640
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004781746031746032,
+      "loss": 4.7362,
+      "theoretical_loss": 6.399647038312671,
+      "tokens_seen": 15794176
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004801587301587302,
+      "loss": 4.7356,
+      "theoretical_loss": 6.395055741956952,
+      "tokens_seen": 15859712
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00048214285714285715,
+      "loss": 4.887,
+      "theoretical_loss": 6.390488666146185,
+      "tokens_seen": 15925248
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.00048412698412698415,
+      "loss": 4.8904,
+      "theoretical_loss": 6.3859455842220765,
+      "tokens_seen": 15990784
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004861111111111111,
+      "loss": 4.952,
+      "theoretical_loss": 6.381426272565258,
+      "tokens_seen": 16056320
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004880952380952381,
+      "loss": 4.917,
+      "theoretical_loss": 6.376930510542334,
+      "tokens_seen": 16121856
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004900793650793651,
+      "loss": 4.8962,
+      "theoretical_loss": 6.372458080454071,
+      "tokens_seen": 16187392
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.000492063492063492,
+      "loss": 4.8315,
+      "theoretical_loss": 6.368008767484675,
+      "tokens_seen": 16252928
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004940476190476191,
+      "loss": 4.8416,
+      "theoretical_loss": 6.363582359652165,
+      "tokens_seen": 16318464
+    },
+    {
+      "epoch": 0.0,
+      "objective/train/advantage_avg": 0.49277380108833313,
+      "objective/train/docs_used": 13382,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 4.31853723526001,
+      "objective/train/original_loss": 4.318537712097168,
+      "objective/train/theoretical_loss": 6.359178647759789,
+      "objective/train/tokens_used": 36844000,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2433290332555771,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505143404006958,
+      "objective/train/weighted_lm_loss": 4.537021636962891,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 6.359178647759789,
+      "tokens_seen": 16384000
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.000496031746031746,
+      "loss": 4.7369,
+      "theoretical_loss": 6.359178647759789,
+      "tokens_seen": 16384000
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.000498015873015873,
+      "loss": 4.7574,
+      "theoretical_loss": 6.354797425348455,
+      "tokens_seen": 16449536
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005,
+      "loss": 4.61,
+      "theoretical_loss": 6.350438488650175,
+      "tokens_seen": 16515072
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000501984126984127,
+      "loss": 4.6561,
+      "theoretical_loss": 6.3461016365424685,
+      "tokens_seen": 16580608
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000503968253968254,
+      "loss": 4.7203,
+      "theoretical_loss": 6.341786670503726,
+      "tokens_seen": 16646144
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005059523809523809,
+      "loss": 4.7175,
+      "theoretical_loss": 6.33749339456949,
+      "tokens_seen": 16711680
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005079365079365079,
+      "loss": 4.7454,
+      "theoretical_loss": 6.333221615289645,
+      "tokens_seen": 16777216
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005099206349206349,
+      "loss": 4.9234,
+      "theoretical_loss": 6.328971141686479,
+      "tokens_seen": 16842752
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005119047619047619,
+      "loss": 4.7527,
+      "theoretical_loss": 6.324741785213623,
+      "tokens_seen": 16908288
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005138888888888888,
+      "loss": 4.7599,
+      "theoretical_loss": 6.320533359715803,
+      "tokens_seen": 16973824
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005158730158730159,
+      "loss": 4.7576,
+      "theoretical_loss": 6.316345681389436,
+      "tokens_seen": 17039360
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005178571428571429,
+      "loss": 4.7286,
+      "theoretical_loss": 6.312178568744001,
+      "tokens_seen": 17104896
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005198412698412699,
+      "loss": 4.7687,
+      "theoretical_loss": 6.308031842564212,
+      "tokens_seen": 17170432
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005218253968253968,
+      "loss": 4.4944,
+      "theoretical_loss": 6.303905325872933,
+      "tokens_seen": 17235968
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005238095238095238,
+      "loss": 4.9019,
+      "theoretical_loss": 6.2997988438948465,
+      "tokens_seen": 17301504
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005257936507936508,
+      "loss": 4.6807,
+      "theoretical_loss": 6.295712224020842,
+      "tokens_seen": 17367040
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005277777777777778,
+      "loss": 4.7674,
+      "theoretical_loss": 6.2916452957731135,
+      "tokens_seen": 17432576
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005297619047619047,
+      "loss": 4.6331,
+      "theoretical_loss": 6.287597890770951,
+      "tokens_seen": 17498112
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005317460317460317,
+      "loss": 4.5051,
+      "theoretical_loss": 6.283569842697203,
+      "tokens_seen": 17563648
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005337301587301587,
+      "loss": 4.4246,
+      "theoretical_loss": 6.279560987265395,
+      "tokens_seen": 17629184
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005357142857142857,
+      "loss": 4.6671,
+      "theoretical_loss": 6.275571162187505,
+      "tokens_seen": 17694720
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005376984126984126,
+      "loss": 4.7847,
+      "theoretical_loss": 6.271600207142356,
+      "tokens_seen": 17760256
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005396825396825396,
+      "loss": 4.7324,
+      "theoretical_loss": 6.26764796374462,
+      "tokens_seen": 17825792
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005416666666666666,
+      "loss": 4.6055,
+      "theoretical_loss": 6.263714275514436,
+      "tokens_seen": 17891328
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005436507936507936,
+      "loss": 4.5236,
+      "theoretical_loss": 6.259798987847603,
+      "tokens_seen": 17956864
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005456349206349206,
+      "loss": 4.599,
+      "theoretical_loss": 6.255901947986348,
+      "tokens_seen": 18022400
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005476190476190477,
+      "loss": 4.664,
+      "theoretical_loss": 6.25202300499066,
+      "tokens_seen": 18087936
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005496031746031747,
+      "loss": 4.5562,
+      "theoretical_loss": 6.248162009710162,
+      "tokens_seen": 18153472
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005515873015873017,
+      "loss": 4.7609,
+      "theoretical_loss": 6.244318814756527,
+      "tokens_seen": 18219008
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005535714285714287,
+      "loss": 4.6061,
+      "theoretical_loss": 6.240493274476407,
+      "tokens_seen": 18284544
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005555555555555556,
+      "loss": 4.5964,
+      "theoretical_loss": 6.236685244924882,
+      "tokens_seen": 18350080
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005575396825396826,
+      "loss": 4.5874,
+      "theoretical_loss": 6.232894583839405,
+      "tokens_seen": 18415616
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005595238095238096,
+      "loss": 4.4101,
+      "theoretical_loss": 6.229121150614235,
+      "tokens_seen": 18481152
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005615079365079366,
+      "loss": 4.7411,
+      "theoretical_loss": 6.225364806275344,
+      "tokens_seen": 18546688
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005634920634920635,
+      "loss": 4.5962,
+      "theoretical_loss": 6.2216254134558024,
+      "tokens_seen": 18612224
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005654761904761905,
+      "loss": 4.5972,
+      "theoretical_loss": 6.217902836371607,
+      "tokens_seen": 18677760
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005674603174603175,
+      "loss": 4.6108,
+      "theoretical_loss": 6.214196940797967,
+      "tokens_seen": 18743296
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005694444444444445,
+      "loss": 4.582,
+      "theoretical_loss": 6.210507594046018,
+      "tokens_seen": 18808832
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005714285714285714,
+      "loss": 4.7575,
+      "theoretical_loss": 6.206834664939976,
+      "tokens_seen": 18874368
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005734126984126984,
+      "loss": 4.7611,
+      "theoretical_loss": 6.203178023794688,
+      "tokens_seen": 18939904
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005753968253968254,
+      "loss": 4.6871,
+      "theoretical_loss": 6.199537542393618,
+      "tokens_seen": 19005440
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005773809523809524,
+      "loss": 4.6562,
+      "theoretical_loss": 6.195913093967214,
+      "tokens_seen": 19070976
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005793650793650794,
+      "loss": 4.5739,
+      "theoretical_loss": 6.192304553171669,
+      "tokens_seen": 19136512
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005813492063492064,
+      "loss": 4.5292,
+      "theoretical_loss": 6.188711796068077,
+      "tokens_seen": 19202048
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005833333333333334,
+      "loss": 4.3859,
+      "theoretical_loss": 6.185134700101948,
+      "tokens_seen": 19267584
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005853174603174604,
+      "loss": 4.6491,
+      "theoretical_loss": 6.181573144083102,
+      "tokens_seen": 19333120
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005873015873015874,
+      "loss": 4.633,
+      "theoretical_loss": 6.178027008165916,
+      "tokens_seen": 19398656
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005892857142857143,
+      "loss": 4.6405,
+      "theoretical_loss": 6.174496173829926,
+      "tokens_seen": 19464192
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005912698412698413,
+      "loss": 4.602,
+      "theoretical_loss": 6.170980523860769,
+      "tokens_seen": 19529728
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005932539682539683,
+      "loss": 4.4713,
+      "theoretical_loss": 6.167479942331466,
+      "tokens_seen": 19595264
+    },
+    {
+      "epoch": 0.01,
+      "objective/train/advantage_avg": 0.49178165197372437,
+      "objective/train/docs_used": 14656,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 4.28451681137085,
+      "objective/train/original_loss": 4.284516334533691,
+      "objective/train/theoretical_loss": 6.163994314584031,
+      "objective/train/tokens_used": 40120800,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24251613020896912,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504111051559448,
+      "objective/train/weighted_lm_loss": 4.500372409820557,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 6.163994314584031,
+      "tokens_seen": 19660800
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005952380952380953,
+      "loss": 4.5733,
+      "theoretical_loss": 6.163994314584031,
+      "tokens_seen": 19660800
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005972222222222222,
+      "loss": 4.5806,
+      "theoretical_loss": 6.16052352721141,
+      "tokens_seen": 19726336
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0005992063492063492,
+      "loss": 4.5064,
+      "theoretical_loss": 6.157067468039726,
+      "tokens_seen": 19791872
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006011904761904762,
+      "loss": 4.4136,
+      "theoretical_loss": 6.153626026110843,
+      "tokens_seen": 19857408
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006031746031746032,
+      "loss": 4.5153,
+      "theoretical_loss": 6.150199091665225,
+      "tokens_seen": 19922944
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006051587301587301,
+      "loss": 4.5327,
+      "theoretical_loss": 6.146786556125103,
+      "tokens_seen": 19988480
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006071428571428571,
+      "loss": 4.4621,
+      "theoretical_loss": 6.143388312077918,
+      "tokens_seen": 20054016
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006091269841269841,
+      "loss": 4.5998,
+      "theoretical_loss": 6.140004253260065,
+      "tokens_seen": 20119552
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006111111111111112,
+      "loss": 4.6879,
+      "theoretical_loss": 6.136634274540901,
+      "tokens_seen": 20185088
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006130952380952381,
+      "loss": 4.5875,
+      "theoretical_loss": 6.13327827190704,
+      "tokens_seen": 20250624
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006150793650793651,
+      "loss": 4.5669,
+      "theoretical_loss": 6.129936142446908,
+      "tokens_seen": 20316160
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006170634920634921,
+      "loss": 4.475,
+      "theoretical_loss": 6.12660778433556,
+      "tokens_seen": 20381696
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006190476190476191,
+      "loss": 4.4535,
+      "theoretical_loss": 6.123293096819758,
+      "tokens_seen": 20447232
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000621031746031746,
+      "loss": 4.6288,
+      "theoretical_loss": 6.119991980203299,
+      "tokens_seen": 20512768
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000623015873015873,
+      "loss": 4.3293,
+      "theoretical_loss": 6.11670433583259,
+      "tokens_seen": 20578304
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000625,
+      "loss": 4.5311,
+      "theoretical_loss": 6.113430066082468,
+      "tokens_seen": 20643840
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000626984126984127,
+      "loss": 4.4825,
+      "theoretical_loss": 6.1101690743422505,
+      "tokens_seen": 20709376
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000628968253968254,
+      "loss": 4.4968,
+      "theoretical_loss": 6.106921265002029,
+      "tokens_seen": 20774912
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006309523809523809,
+      "loss": 4.4781,
+      "theoretical_loss": 6.103686543439177,
+      "tokens_seen": 20840448
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006329365079365079,
+      "loss": 4.6214,
+      "theoretical_loss": 6.100464816005092,
+      "tokens_seen": 20905984
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006349206349206349,
+      "loss": 4.5962,
+      "theoretical_loss": 6.097255990012153,
+      "tokens_seen": 20971520
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006369047619047619,
+      "loss": 4.5147,
+      "theoretical_loss": 6.094059973720896,
+      "tokens_seen": 21037056
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006388888888888888,
+      "loss": 4.4541,
+      "theoretical_loss": 6.090876676327392,
+      "tokens_seen": 21102592
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006408730158730159,
+      "loss": 4.4346,
+      "theoretical_loss": 6.087706007950839,
+      "tokens_seen": 21168128
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006428571428571429,
+      "loss": 4.5464,
+      "theoretical_loss": 6.084547879621354,
+      "tokens_seen": 21233664
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006448412698412699,
+      "loss": 4.5378,
+      "theoretical_loss": 6.081402203267963,
+      "tokens_seen": 21299200
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006468253968253968,
+      "loss": 4.5303,
+      "theoretical_loss": 6.078268891706788,
+      "tokens_seen": 21364736
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006488095238095238,
+      "loss": 4.4849,
+      "theoretical_loss": 6.075147858629419,
+      "tokens_seen": 21430272
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006507936507936508,
+      "loss": 4.3543,
+      "theoretical_loss": 6.072039018591484,
+      "tokens_seen": 21495808
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006527777777777778,
+      "loss": 4.5641,
+      "theoretical_loss": 6.068942287001391,
+      "tokens_seen": 21561344
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006547619047619047,
+      "loss": 4.4829,
+      "theoretical_loss": 6.065857580109258,
+      "tokens_seen": 21626880
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006567460317460317,
+      "loss": 4.3724,
+      "theoretical_loss": 6.0627848149960135,
+      "tokens_seen": 21692416
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006587301587301587,
+      "loss": 4.5818,
+      "theoretical_loss": 6.059723909562683,
+      "tokens_seen": 21757952
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006607142857142857,
+      "loss": 4.3904,
+      "theoretical_loss": 6.0566747825198295,
+      "tokens_seen": 21823488
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006626984126984126,
+      "loss": 4.4051,
+      "theoretical_loss": 6.0536373533771695,
+      "tokens_seen": 21889024
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006646825396825396,
+      "loss": 4.5032,
+      "theoretical_loss": 6.050611542433352,
+      "tokens_seen": 21954560
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006666666666666666,
+      "loss": 4.3651,
+      "theoretical_loss": 6.047597270765904,
+      "tokens_seen": 22020096
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006686507936507936,
+      "loss": 4.5418,
+      "theoretical_loss": 6.044594460221319,
+      "tokens_seen": 22085632
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006706349206349207,
+      "loss": 4.4352,
+      "theoretical_loss": 6.041603033405314,
+      "tokens_seen": 22151168
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006726190476190477,
+      "loss": 4.3779,
+      "theoretical_loss": 6.038622913673237,
+      "tokens_seen": 22216704
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006746031746031747,
+      "loss": 4.2616,
+      "theoretical_loss": 6.035654025120612,
+      "tokens_seen": 22282240
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006765873015873017,
+      "loss": 4.319,
+      "theoretical_loss": 6.032696292573847,
+      "tokens_seen": 22347776
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006785714285714287,
+      "loss": 4.2787,
+      "theoretical_loss": 6.029749641581073,
+      "tokens_seen": 22413312
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006805555555555556,
+      "loss": 4.343,
+      "theoretical_loss": 6.02681399840313,
+      "tokens_seen": 22478848
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006825396825396826,
+      "loss": 4.4465,
+      "theoretical_loss": 6.023889290004692,
+      "tokens_seen": 22544384
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006845238095238096,
+      "loss": 4.4453,
+      "theoretical_loss": 6.020975444045523,
+      "tokens_seen": 22609920
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006865079365079366,
+      "loss": 4.28,
+      "theoretical_loss": 6.018072388871873,
+      "tokens_seen": 22675456
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006884920634920635,
+      "loss": 4.4475,
+      "theoretical_loss": 6.015180053507997,
+      "tokens_seen": 22740992
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006904761904761905,
+      "loss": 4.2961,
+      "theoretical_loss": 6.012298367647816,
+      "tokens_seen": 22806528
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006924603174603175,
+      "loss": 4.3297,
+      "theoretical_loss": 6.009427261646688,
+      "tokens_seen": 22872064
+    },
+    {
+      "epoch": 0.01,
+      "objective/train/advantage_avg": 0.49345701932907104,
+      "objective/train/docs_used": 15907,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 4.267792701721191,
+      "objective/train/original_loss": 4.267792701721191,
+      "objective/train/theoretical_loss": 6.006566666513313,
+      "objective/train/tokens_used": 43397600,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24379104375839233,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505850315093994,
+      "objective/train/weighted_lm_loss": 4.483581066131592,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 6.006566666513313,
+      "tokens_seen": 22937600
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006944444444444445,
+      "loss": 4.4127,
+      "theoretical_loss": 6.006566666513313,
+      "tokens_seen": 22937600
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006964285714285714,
+      "loss": 4.09,
+      "theoretical_loss": 6.003716513901762,
+      "tokens_seen": 23003136
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0006984126984126984,
+      "loss": 4.2815,
+      "theoretical_loss": 6.000876736103618,
+      "tokens_seen": 23068672
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007003968253968254,
+      "loss": 4.4355,
+      "theoretical_loss": 5.998047266040238,
+      "tokens_seen": 23134208
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007023809523809524,
+      "loss": 4.4496,
+      "theoretical_loss": 5.995228037255137,
+      "tokens_seen": 23199744
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007043650793650794,
+      "loss": 4.3207,
+      "theoretical_loss": 5.992418983906469,
+      "tokens_seen": 23265280
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007063492063492064,
+      "loss": 4.1968,
+      "theoretical_loss": 5.989620040759641,
+      "tokens_seen": 23330816
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007083333333333334,
+      "loss": 4.2346,
+      "theoretical_loss": 5.986831143180019,
+      "tokens_seen": 23396352
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007103174603174604,
+      "loss": 4.2309,
+      "theoretical_loss": 5.9840522271257495,
+      "tokens_seen": 23461888
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007123015873015874,
+      "loss": 4.2863,
+      "theoretical_loss": 5.981283229140683,
+      "tokens_seen": 23527424
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007142857142857143,
+      "loss": 4.2112,
+      "theoretical_loss": 5.978524086347409,
+      "tokens_seen": 23592960
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007162698412698413,
+      "loss": 4.2578,
+      "theoretical_loss": 5.975774736440385,
+      "tokens_seen": 23658496
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007182539682539683,
+      "loss": 4.2903,
+      "theoretical_loss": 5.973035117679164,
+      "tokens_seen": 23724032
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007202380952380953,
+      "loss": 4.3281,
+      "theoretical_loss": 5.970305168881732,
+      "tokens_seen": 23789568
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007222222222222222,
+      "loss": 4.4398,
+      "theoretical_loss": 5.967584829417934,
+      "tokens_seen": 23855104
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007242063492063492,
+      "loss": 4.2352,
+      "theoretical_loss": 5.964874039202989,
+      "tokens_seen": 23920640
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007261904761904762,
+      "loss": 4.2132,
+      "theoretical_loss": 5.96217273869112,
+      "tokens_seen": 23986176
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007281746031746032,
+      "loss": 4.1317,
+      "theoretical_loss": 5.959480868869245,
+      "tokens_seen": 24051712
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007301587301587301,
+      "loss": 4.3792,
+      "theoretical_loss": 5.956798371250791,
+      "tokens_seen": 24117248
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007321428571428571,
+      "loss": 4.1713,
+      "theoretical_loss": 5.954125187869572,
+      "tokens_seen": 24182784
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007341269841269841,
+      "loss": 4.2422,
+      "theoretical_loss": 5.951461261273764,
+      "tokens_seen": 24248320
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007361111111111112,
+      "loss": 4.137,
+      "theoretical_loss": 5.948806534519972,
+      "tokens_seen": 24313856
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007380952380952381,
+      "loss": 4.1628,
+      "theoretical_loss": 5.9461609511673625,
+      "tokens_seen": 24379392
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007400793650793651,
+      "loss": 4.2607,
+      "theoretical_loss": 5.943524455271907,
+      "tokens_seen": 24444928
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007420634920634921,
+      "loss": 4.2082,
+      "theoretical_loss": 5.940896991380676,
+      "tokens_seen": 24510464
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007440476190476191,
+      "loss": 4.3838,
+      "theoretical_loss": 5.938278504526239,
+      "tokens_seen": 24576000
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000746031746031746,
+      "loss": 4.2242,
+      "theoretical_loss": 5.935668940221127,
+      "tokens_seen": 24641536
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000748015873015873,
+      "loss": 4.1961,
+      "theoretical_loss": 5.933068244452382,
+      "tokens_seen": 24707072
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00075,
+      "loss": 4.1996,
+      "theoretical_loss": 5.930476363676177,
+      "tokens_seen": 24772608
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000751984126984127,
+      "loss": 4.2341,
+      "theoretical_loss": 5.927893244812513,
+      "tokens_seen": 24838144
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000753968253968254,
+      "loss": 4.2983,
+      "theoretical_loss": 5.92531883523999,
+      "tokens_seen": 24903680
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007559523809523809,
+      "loss": 4.144,
+      "theoretical_loss": 5.922753082790654,
+      "tokens_seen": 24969216
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007579365079365079,
+      "loss": 4.1519,
+      "theoretical_loss": 5.920195935744903,
+      "tokens_seen": 25034752
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007599206349206349,
+      "loss": 4.3062,
+      "theoretical_loss": 5.917647342826485,
+      "tokens_seen": 25100288
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007619047619047619,
+      "loss": 4.164,
+      "theoretical_loss": 5.915107253197538,
+      "tokens_seen": 25165824
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007638888888888888,
+      "loss": 4.2066,
+      "theoretical_loss": 5.91257561645372,
+      "tokens_seen": 25231360
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007658730158730159,
+      "loss": 4.236,
+      "theoretical_loss": 5.910052382619399,
+      "tokens_seen": 25296896
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007678571428571429,
+      "loss": 4.2914,
+      "theoretical_loss": 5.907537502142899,
+      "tokens_seen": 25362432
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007698412698412699,
+      "loss": 4.2342,
+      "theoretical_loss": 5.905030925891829,
+      "tokens_seen": 25427968
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007718253968253968,
+      "loss": 4.1399,
+      "theoretical_loss": 5.9025326051484575,
+      "tokens_seen": 25493504
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007738095238095238,
+      "loss": 4.3571,
+      "theoretical_loss": 5.900042491605162,
+      "tokens_seen": 25559040
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007757936507936508,
+      "loss": 4.313,
+      "theoretical_loss": 5.897560537359936,
+      "tokens_seen": 25624576
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007777777777777778,
+      "loss": 4.1413,
+      "theoretical_loss": 5.895086694911951,
+      "tokens_seen": 25690112
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007797619047619047,
+      "loss": 4.2811,
+      "theoretical_loss": 5.892620917157192,
+      "tokens_seen": 25755648
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007817460317460317,
+      "loss": 4.1719,
+      "theoretical_loss": 5.8901631573841335,
+      "tokens_seen": 25821184
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007837301587301587,
+      "loss": 4.2168,
+      "theoretical_loss": 5.887713369269493,
+      "tokens_seen": 25886720
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007857142857142857,
+      "loss": 4.2532,
+      "theoretical_loss": 5.88527150687402,
+      "tokens_seen": 25952256
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007876984126984126,
+      "loss": 4.298,
+      "theoretical_loss": 5.88283752463836,
+      "tokens_seen": 26017792
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007896825396825396,
+      "loss": 4.0948,
+      "theoretical_loss": 5.880411377378962,
+      "tokens_seen": 26083328
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007916666666666666,
+      "loss": 4.0214,
+      "theoretical_loss": 5.87799302028405,
+      "tokens_seen": 26148864
+    },
+    {
+      "epoch": 0.01,
+      "objective/train/advantage_avg": 0.4926624894142151,
+      "objective/train/docs_used": 17033,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 4.048245906829834,
+      "objective/train/original_loss": 4.048246383666992,
+      "objective/train/theoretical_loss": 5.8755824089096285,
+      "objective/train/tokens_used": 46674400,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24356918036937714,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050504446029663,
+      "objective/train/weighted_lm_loss": 4.252634048461914,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 5.8755824089096285,
+      "tokens_seen": 26214400
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007936507936507937,
+      "loss": 4.3117,
+      "theoretical_loss": 5.8755824089096285,
+      "tokens_seen": 26214400
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007956349206349207,
+      "loss": 3.9992,
+      "theoretical_loss": 5.873179499175565,
+      "tokens_seen": 26279936
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007976190476190477,
+      "loss": 4.0696,
+      "theoretical_loss": 5.87078424736171,
+      "tokens_seen": 26345472
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0007996031746031747,
+      "loss": 4.1271,
+      "theoretical_loss": 5.868396610104066,
+      "tokens_seen": 26411008
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008015873015873017,
+      "loss": 4.1637,
+      "theoretical_loss": 5.866016544391016,
+      "tokens_seen": 26476544
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008035714285714287,
+      "loss": 4.0211,
+      "theoretical_loss": 5.863644007559595,
+      "tokens_seen": 26542080
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008055555555555556,
+      "loss": 4.1509,
+      "theoretical_loss": 5.861278957291807,
+      "tokens_seen": 26607616
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008075396825396826,
+      "loss": 4.0588,
+      "theoretical_loss": 5.858921351611005,
+      "tokens_seen": 26673152
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008095238095238096,
+      "loss": 4.0002,
+      "theoretical_loss": 5.856571148878293,
+      "tokens_seen": 26738688
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008115079365079366,
+      "loss": 4.1366,
+      "theoretical_loss": 5.854228307788999,
+      "tokens_seen": 26804224
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008134920634920635,
+      "loss": 4.1982,
+      "theoretical_loss": 5.851892787369178,
+      "tokens_seen": 26869760
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008154761904761905,
+      "loss": 4.044,
+      "theoretical_loss": 5.849564546972166,
+      "tokens_seen": 26935296
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008174603174603175,
+      "loss": 3.8531,
+      "theoretical_loss": 5.847243546275179,
+      "tokens_seen": 27000832
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008194444444444445,
+      "loss": 4.0057,
+      "theoretical_loss": 5.844929745275948,
+      "tokens_seen": 27066368
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008214285714285714,
+      "loss": 4.1101,
+      "theoretical_loss": 5.842623104289412,
+      "tokens_seen": 27131904
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008234126984126984,
+      "loss": 3.9774,
+      "theoretical_loss": 5.840323583944434,
+      "tokens_seen": 27197440
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008253968253968254,
+      "loss": 3.8177,
+      "theoretical_loss": 5.838031145180573,
+      "tokens_seen": 27262976
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008273809523809524,
+      "loss": 4.1632,
+      "theoretical_loss": 5.835745749244896,
+      "tokens_seen": 27328512
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008293650793650794,
+      "loss": 4.1018,
+      "theoretical_loss": 5.833467357688816,
+      "tokens_seen": 27394048
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008313492063492064,
+      "loss": 4.1213,
+      "theoretical_loss": 5.8311959323649925,
+      "tokens_seen": 27459584
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008333333333333334,
+      "loss": 4.1399,
+      "theoretical_loss": 5.82893143542425,
+      "tokens_seen": 27525120
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008353174603174604,
+      "loss": 4.0058,
+      "theoretical_loss": 5.826673829312552,
+      "tokens_seen": 27590656
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008373015873015874,
+      "loss": 4.1258,
+      "theoretical_loss": 5.824423076767996,
+      "tokens_seen": 27656192
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008392857142857143,
+      "loss": 3.9901,
+      "theoretical_loss": 5.8221791408178625,
+      "tokens_seen": 27721728
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008412698412698413,
+      "loss": 4.0441,
+      "theoretical_loss": 5.81994198477569,
+      "tokens_seen": 27787264
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008432539682539683,
+      "loss": 4.0542,
+      "theoretical_loss": 5.817711572238393,
+      "tokens_seen": 27852800
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008452380952380953,
+      "loss": 4.0306,
+      "theoretical_loss": 5.815487867083405,
+      "tokens_seen": 27918336
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008472222222222222,
+      "loss": 3.9466,
+      "theoretical_loss": 5.813270833465875,
+      "tokens_seen": 27983872
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008492063492063492,
+      "loss": 3.9006,
+      "theoretical_loss": 5.811060435815881,
+      "tokens_seen": 28049408
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008511904761904762,
+      "loss": 4.0148,
+      "theoretical_loss": 5.808856638835682,
+      "tokens_seen": 28114944
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008531746031746032,
+      "loss": 3.9268,
+      "theoretical_loss": 5.806659407497016,
+      "tokens_seen": 28180480
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008551587301587301,
+      "loss": 4.1383,
+      "theoretical_loss": 5.804468707038416,
+      "tokens_seen": 28246016
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008571428571428571,
+      "loss": 4.0713,
+      "theoretical_loss": 5.802284502962563,
+      "tokens_seen": 28311552
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008591269841269841,
+      "loss": 4.0327,
+      "theoretical_loss": 5.8001067610336765,
+      "tokens_seen": 28377088
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008611111111111112,
+      "loss": 4.1369,
+      "theoretical_loss": 5.797935447274931,
+      "tokens_seen": 28442624
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008630952380952381,
+      "loss": 3.9127,
+      "theoretical_loss": 5.795770527965905,
+      "tokens_seen": 28508160
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008650793650793651,
+      "loss": 3.959,
+      "theoretical_loss": 5.793611969640068,
+      "tokens_seen": 28573696
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008670634920634921,
+      "loss": 3.7726,
+      "theoretical_loss": 5.791459739082283,
+      "tokens_seen": 28639232
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008690476190476191,
+      "loss": 4.043,
+      "theoretical_loss": 5.789313803326356,
+      "tokens_seen": 28704768
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000871031746031746,
+      "loss": 3.9936,
+      "theoretical_loss": 5.787174129652598,
+      "tokens_seen": 28770304
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000873015873015873,
+      "loss": 4.0503,
+      "theoretical_loss": 5.785040685585437,
+      "tokens_seen": 28835840
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000875,
+      "loss": 3.9107,
+      "theoretical_loss": 5.782913438891035,
+      "tokens_seen": 28901376
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000876984126984127,
+      "loss": 3.9906,
+      "theoretical_loss": 5.780792357574953,
+      "tokens_seen": 28966912
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000878968253968254,
+      "loss": 4.0754,
+      "theoretical_loss": 5.778677409879835,
+      "tokens_seen": 29032448
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008809523809523809,
+      "loss": 3.9564,
+      "theoretical_loss": 5.7765685642831155,
+      "tokens_seen": 29097984
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008829365079365079,
+      "loss": 4.0196,
+      "theoretical_loss": 5.7744657894947675,
+      "tokens_seen": 29163520
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008849206349206349,
+      "loss": 4.0888,
+      "theoretical_loss": 5.772369054455065,
+      "tokens_seen": 29229056
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008869047619047619,
+      "loss": 3.9354,
+      "theoretical_loss": 5.770278328332372,
+      "tokens_seen": 29294592
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008888888888888888,
+      "loss": 4.0387,
+      "theoretical_loss": 5.768193580520972,
+      "tokens_seen": 29360128
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008908730158730159,
+      "loss": 3.8796,
+      "theoretical_loss": 5.766114780638902,
+      "tokens_seen": 29425664
+    },
+    {
+      "epoch": 0.01,
+      "objective/train/advantage_avg": 0.49078458547592163,
+      "objective/train/docs_used": 18174,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 3.943504571914673,
+      "objective/train/original_loss": 3.9435043334960938,
+      "objective/train/theoretical_loss": 5.7640418985258295,
+      "objective/train/tokens_used": 49951200,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2413143813610077,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503051280975342,
+      "objective/train/weighted_lm_loss": 4.141692161560059,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 5.7640418985258295,
+      "tokens_seen": 29491200
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008928571428571429,
+      "loss": 4.0427,
+      "theoretical_loss": 5.7640418985258295,
+      "tokens_seen": 29491200
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008948412698412699,
+      "loss": 3.9764,
+      "theoretical_loss": 5.761974904240946,
+      "tokens_seen": 29556736
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008968253968253968,
+      "loss": 3.8796,
+      "theoretical_loss": 5.759913768060882,
+      "tokens_seen": 29622272
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0008988095238095238,
+      "loss": 3.9728,
+      "theoretical_loss": 5.757858460477662,
+      "tokens_seen": 29687808
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009007936507936508,
+      "loss": 4.0234,
+      "theoretical_loss": 5.755808952196658,
+      "tokens_seen": 29753344
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009027777777777778,
+      "loss": 3.8761,
+      "theoretical_loss": 5.753765214134593,
+      "tokens_seen": 29818880
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009047619047619047,
+      "loss": 4.0411,
+      "theoretical_loss": 5.7517272174175496,
+      "tokens_seen": 29884416
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009067460317460317,
+      "loss": 4.0655,
+      "theoretical_loss": 5.7496949333790095,
+      "tokens_seen": 29949952
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009087301587301587,
+      "loss": 3.8995,
+      "theoretical_loss": 5.747668333557913,
+      "tokens_seen": 30015488
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009107142857142857,
+      "loss": 3.9744,
+      "theoretical_loss": 5.745647389696742,
+      "tokens_seen": 30081024
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009126984126984126,
+      "loss": 4.072,
+      "theoretical_loss": 5.743632073739626,
+      "tokens_seen": 30146560
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009146825396825396,
+      "loss": 3.9495,
+      "theoretical_loss": 5.741622357830466,
+      "tokens_seen": 30212096
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009166666666666666,
+      "loss": 3.9492,
+      "theoretical_loss": 5.7396182143110845,
+      "tokens_seen": 30277632
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009186507936507937,
+      "loss": 4.0284,
+      "theoretical_loss": 5.737619615719392,
+      "tokens_seen": 30343168
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009206349206349207,
+      "loss": 3.892,
+      "theoretical_loss": 5.735626534787584,
+      "tokens_seen": 30408704
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009226190476190477,
+      "loss": 3.8598,
+      "theoretical_loss": 5.7336389444403375,
+      "tokens_seen": 30474240
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009246031746031747,
+      "loss": 3.8676,
+      "theoretical_loss": 5.731656817793057,
+      "tokens_seen": 30539776
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009265873015873017,
+      "loss": 3.8372,
+      "theoretical_loss": 5.729680128150115,
+      "tokens_seen": 30605312
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009285714285714287,
+      "loss": 3.7756,
+      "theoretical_loss": 5.727708849003127,
+      "tokens_seen": 30670848
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009305555555555556,
+      "loss": 3.9913,
+      "theoretical_loss": 5.72574295402924,
+      "tokens_seen": 30736384
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009325396825396826,
+      "loss": 3.9164,
+      "theoretical_loss": 5.723782417089441,
+      "tokens_seen": 30801920
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009345238095238096,
+      "loss": 3.8756,
+      "theoretical_loss": 5.721827212226889,
+      "tokens_seen": 30867456
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009365079365079366,
+      "loss": 3.7424,
+      "theoretical_loss": 5.719877313665254,
+      "tokens_seen": 30932992
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009384920634920635,
+      "loss": 3.8924,
+      "theoretical_loss": 5.717932695807093,
+      "tokens_seen": 30998528
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009404761904761905,
+      "loss": 3.9166,
+      "theoretical_loss": 5.715993333232223,
+      "tokens_seen": 31064064
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009424603174603175,
+      "loss": 3.8553,
+      "theoretical_loss": 5.7140592006961315,
+      "tokens_seen": 31129600
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009444444444444445,
+      "loss": 3.832,
+      "theoretical_loss": 5.712130273128388,
+      "tokens_seen": 31195136
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009464285714285714,
+      "loss": 3.7882,
+      "theoretical_loss": 5.710206525631086,
+      "tokens_seen": 31260672
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009484126984126984,
+      "loss": 4.0133,
+      "theoretical_loss": 5.708287933477296,
+      "tokens_seen": 31326208
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009503968253968254,
+      "loss": 4.0449,
+      "theoretical_loss": 5.706374472109536,
+      "tokens_seen": 31391744
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009523809523809524,
+      "loss": 3.8148,
+      "theoretical_loss": 5.704466117138258,
+      "tokens_seen": 31457280
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009543650793650794,
+      "loss": 3.91,
+      "theoretical_loss": 5.702562844340358,
+      "tokens_seen": 31522816
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009563492063492064,
+      "loss": 3.7582,
+      "theoretical_loss": 5.700664629657688,
+      "tokens_seen": 31588352
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009583333333333334,
+      "loss": 4.045,
+      "theoretical_loss": 5.698771449195602,
+      "tokens_seen": 31653888
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009603174603174604,
+      "loss": 3.8957,
+      "theoretical_loss": 5.696883279221504,
+      "tokens_seen": 31719424
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009623015873015874,
+      "loss": 3.8502,
+      "theoretical_loss": 5.695000096163421,
+      "tokens_seen": 31784960
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009642857142857143,
+      "loss": 3.7878,
+      "theoretical_loss": 5.693121876608577,
+      "tokens_seen": 31850496
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009662698412698413,
+      "loss": 3.9058,
+      "theoretical_loss": 5.691248597302009,
+      "tokens_seen": 31916032
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009682539682539683,
+      "loss": 3.8435,
+      "theoretical_loss": 5.689380235145171,
+      "tokens_seen": 31981568
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009702380952380953,
+      "loss": 3.8857,
+      "theoretical_loss": 5.687516767194563,
+      "tokens_seen": 32047104
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009722222222222222,
+      "loss": 3.8421,
+      "theoretical_loss": 5.685658170660382,
+      "tokens_seen": 32112640
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009742063492063492,
+      "loss": 3.9891,
+      "theoretical_loss": 5.683804422905179,
+      "tokens_seen": 32178176
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009761904761904762,
+      "loss": 3.7935,
+      "theoretical_loss": 5.6819555014425305,
+      "tokens_seen": 32243712
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009781746031746032,
+      "loss": 3.7785,
+      "theoretical_loss": 5.680111383935727,
+      "tokens_seen": 32309248
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009801587301587302,
+      "loss": 3.9492,
+      "theoretical_loss": 5.678272048196479,
+      "tokens_seen": 32374784
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009821428571428572,
+      "loss": 3.8507,
+      "theoretical_loss": 5.6764374721836255,
+      "tokens_seen": 32440320
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000984126984126984,
+      "loss": 3.7337,
+      "theoretical_loss": 5.674607634001871,
+      "tokens_seen": 32505856
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009861111111111112,
+      "loss": 3.9994,
+      "theoretical_loss": 5.672782511900524,
+      "tokens_seen": 32571392
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009880952380952382,
+      "loss": 3.9901,
+      "theoretical_loss": 5.670962084272254,
+      "tokens_seen": 32636928
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009900793650793652,
+      "loss": 3.9945,
+      "theoretical_loss": 5.669146329651861,
+      "tokens_seen": 32702464
+    },
+    {
+      "epoch": 0.01,
+      "objective/train/advantage_avg": 0.49417802691459656,
+      "objective/train/docs_used": 19239,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 3.745575189590454,
+      "objective/train/original_loss": 3.745574951171875,
+      "objective/train/theoretical_loss": 5.667335226715059,
+      "objective/train/tokens_used": 53228000,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2445439100265503,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506609678268433,
+      "objective/train/weighted_lm_loss": 3.935272216796875,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 5.667335226715059,
+      "tokens_seen": 32768000
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000992063492063492,
+      "loss": 3.9393,
+      "theoretical_loss": 5.667335226715059,
+      "tokens_seen": 32768000
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000994047619047619,
+      "loss": 3.8418,
+      "theoretical_loss": 5.66552875427727,
+      "tokens_seen": 32833536
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000996031746031746,
+      "loss": 3.8581,
+      "theoretical_loss": 5.663726891292432,
+      "tokens_seen": 32899072
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000998015873015873,
+      "loss": 3.6826,
+      "theoretical_loss": 5.6619296168518245,
+      "tokens_seen": 32964608
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.001,
+      "loss": 3.9296,
+      "theoretical_loss": 5.6601369101828904,
+      "tokens_seen": 33030144
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009999799398194584,
+      "loss": 3.8296,
+      "theoretical_loss": 5.658348750648093,
+      "tokens_seen": 33095680
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009999598796389167,
+      "loss": 3.7304,
+      "theoretical_loss": 5.656565117743766,
+      "tokens_seen": 33161216
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000999939819458375,
+      "loss": 3.7348,
+      "theoretical_loss": 5.654785991098984,
+      "tokens_seen": 33226752
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009999197592778335,
+      "loss": 3.807,
+      "theoretical_loss": 5.6530113504744435,
+      "tokens_seen": 33292288
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000999899699097292,
+      "loss": 3.7786,
+      "theoretical_loss": 5.651241175761357,
+      "tokens_seen": 33357824
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009998796389167502,
+      "loss": 3.79,
+      "theoretical_loss": 5.649475446980356,
+      "tokens_seen": 33423360
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009998595787362088,
+      "loss": 3.9043,
+      "theoretical_loss": 5.647714144280403,
+      "tokens_seen": 33488896
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000999839518555667,
+      "loss": 3.8137,
+      "theoretical_loss": 5.645957247937725,
+      "tokens_seen": 33554432
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009998194583751255,
+      "loss": 3.7915,
+      "theoretical_loss": 5.644204738354742,
+      "tokens_seen": 33619968
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009997993981945839,
+      "loss": 3.837,
+      "theoretical_loss": 5.642456596059025,
+      "tokens_seen": 33685504
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009997793380140422,
+      "loss": 3.7575,
+      "theoretical_loss": 5.6407128017022465,
+      "tokens_seen": 33751040
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009997592778335006,
+      "loss": 3.8424,
+      "theoretical_loss": 5.638973336059157,
+      "tokens_seen": 33816576
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000999739217652959,
+      "loss": 3.6526,
+      "theoretical_loss": 5.637238180026563,
+      "tokens_seen": 33882112
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009997191574724173,
+      "loss": 3.6857,
+      "theoretical_loss": 5.6355073146223145,
+      "tokens_seen": 33947648
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009996990972918757,
+      "loss": 3.6679,
+      "theoretical_loss": 5.633780720984311,
+      "tokens_seen": 34013184
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000999679037111334,
+      "loss": 3.7352,
+      "theoretical_loss": 5.632058380369512,
+      "tokens_seen": 34078720
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009996589769307924,
+      "loss": 3.6703,
+      "theoretical_loss": 5.630340274152954,
+      "tokens_seen": 34144256
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009996389167502508,
+      "loss": 3.8389,
+      "theoretical_loss": 5.62862638382679,
+      "tokens_seen": 34209792
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009996188565697091,
+      "loss": 3.7032,
+      "theoretical_loss": 5.626916690999319,
+      "tokens_seen": 34275328
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009995987963891675,
+      "loss": 3.6495,
+      "theoretical_loss": 5.625211177394046,
+      "tokens_seen": 34340864
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009995787362086258,
+      "loss": 3.6542,
+      "theoretical_loss": 5.6235098248487425,
+      "tokens_seen": 34406400
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009995586760280842,
+      "loss": 3.5303,
+      "theoretical_loss": 5.621812615314505,
+      "tokens_seen": 34471936
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009995386158475428,
+      "loss": 3.752,
+      "theoretical_loss": 5.620119530854848,
+      "tokens_seen": 34537472
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000999518555667001,
+      "loss": 3.7169,
+      "theoretical_loss": 5.618430553644782,
+      "tokens_seen": 34603008
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009994984954864595,
+      "loss": 3.6834,
+      "theoretical_loss": 5.616745665969917,
+      "tokens_seen": 34668544
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009994784353059177,
+      "loss": 3.6833,
+      "theoretical_loss": 5.615064850225565,
+      "tokens_seen": 34734080
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009994583751253762,
+      "loss": 3.7607,
+      "theoretical_loss": 5.613388088915855,
+      "tokens_seen": 34799616
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009994383149448346,
+      "loss": 3.707,
+      "theoretical_loss": 5.611715364652864,
+      "tokens_seen": 34865152
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000999418254764293,
+      "loss": 3.6116,
+      "theoretical_loss": 5.610046660155737,
+      "tokens_seen": 34930688
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009993981945837513,
+      "loss": 3.6668,
+      "theoretical_loss": 5.608381958249839,
+      "tokens_seen": 34996224
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009993781344032097,
+      "loss": 3.6732,
+      "theoretical_loss": 5.6067212418659,
+      "tokens_seen": 35061760
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000999358074222668,
+      "loss": 3.6028,
+      "theoretical_loss": 5.605064494039176,
+      "tokens_seen": 35127296
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009993380140421264,
+      "loss": 3.6866,
+      "theoretical_loss": 5.603411697908612,
+      "tokens_seen": 35192832
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009993179538615848,
+      "loss": 3.6574,
+      "theoretical_loss": 5.601762836716016,
+      "tokens_seen": 35258368
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009992978936810431,
+      "loss": 3.4094,
+      "theoretical_loss": 5.600117893805246,
+      "tokens_seen": 35323904
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009992778335005015,
+      "loss": 3.5417,
+      "theoretical_loss": 5.598476852621397,
+      "tokens_seen": 35389440
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009992577733199599,
+      "loss": 3.5647,
+      "theoretical_loss": 5.596839696710004,
+      "tokens_seen": 35454976
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009992377131394182,
+      "loss": 3.7131,
+      "theoretical_loss": 5.59520640971624,
+      "tokens_seen": 35520512
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009992176529588766,
+      "loss": 3.5971,
+      "theoretical_loss": 5.5935769753841384,
+      "tokens_seen": 35586048
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000999197592778335,
+      "loss": 3.9092,
+      "theoretical_loss": 5.591951377555809,
+      "tokens_seen": 35651584
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009991775325977935,
+      "loss": 3.7656,
+      "theoretical_loss": 5.590329600170675,
+      "tokens_seen": 35717120
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009991574724172517,
+      "loss": 3.8054,
+      "theoretical_loss": 5.588711627264699,
+      "tokens_seen": 35782656
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009991374122367102,
+      "loss": 3.6167,
+      "theoretical_loss": 5.587097442969634,
+      "tokens_seen": 35848192
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009991173520561684,
+      "loss": 3.7877,
+      "theoretical_loss": 5.585487031512276,
+      "tokens_seen": 35913728
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000999097291875627,
+      "loss": 3.7781,
+      "theoretical_loss": 5.583880377213717,
+      "tokens_seen": 35979264
+    },
+    {
+      "epoch": 0.01,
+      "objective/train/advantage_avg": 0.4903879463672638,
+      "objective/train/docs_used": 20470,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 3.424363613128662,
+      "objective/train/original_loss": 3.424363613128662,
+      "objective/train/theoretical_loss": 5.582277464488612,
+      "objective/train/tokens_used": 56504800,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2408606857061386,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0502631664276123,
+      "objective/train/weighted_lm_loss": 3.596794605255127,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 1.0253151655197144,
+      "theoretical_loss": 5.582277464488612,
+      "tokens_seen": 36044800
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009990772316950853,
+      "loss": 3.5566,
+      "theoretical_loss": 5.582277464488612,
+      "tokens_seen": 36044800
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009990571715145437,
+      "loss": 3.8534,
+      "theoretical_loss": 5.58067827784446,
+      "tokens_seen": 36110336
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000999037111334002,
+      "loss": 3.6709,
+      "theoretical_loss": 5.579082801880871,
+      "tokens_seen": 36175872
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009990170511534604,
+      "loss": 3.7757,
+      "theoretical_loss": 5.577491021288861,
+      "tokens_seen": 36241408
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009989969909729188,
+      "loss": 3.5704,
+      "theoretical_loss": 5.57590292085014,
+      "tokens_seen": 36306944
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009989769307923771,
+      "loss": 3.7544,
+      "theoretical_loss": 5.574318485436422,
+      "tokens_seen": 36372480
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009989568706118355,
+      "loss": 3.512,
+      "theoretical_loss": 5.572737700008718,
+      "tokens_seen": 36438016
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009989368104312939,
+      "loss": 3.5182,
+      "theoretical_loss": 5.5711605496166605,
+      "tokens_seen": 36503552
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009989167502507522,
+      "loss": 3.785,
+      "theoretical_loss": 5.569587019397816,
+      "tokens_seen": 36569088
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009988966900702106,
+      "loss": 3.6573,
+      "theoretical_loss": 5.5680170945770175,
+      "tokens_seen": 36634624
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000998876629889669,
+      "loss": 3.6389,
+      "theoretical_loss": 5.56645076046569,
+      "tokens_seen": 36700160
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009988565697091273,
+      "loss": 3.4898,
+      "theoretical_loss": 5.5648880024612,
+      "tokens_seen": 36765696
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009988365095285857,
+      "loss": 3.6444,
+      "theoretical_loss": 5.5633288060461865,
+      "tokens_seen": 36831232
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009988164493480443,
+      "loss": 3.6373,
+      "theoretical_loss": 5.561773156787925,
+      "tokens_seen": 36896768
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009987963891675024,
+      "loss": 3.667,
+      "theoretical_loss": 5.5602210403376775,
+      "tokens_seen": 36962304
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000998776328986961,
+      "loss": 3.5937,
+      "theoretical_loss": 5.558672442430058,
+      "tokens_seen": 37027840
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009987562688064191,
+      "loss": 3.7541,
+      "theoretical_loss": 5.557127348882403,
+      "tokens_seen": 37093376
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009987362086258777,
+      "loss": 3.6276,
+      "theoretical_loss": 5.555585745594145,
+      "tokens_seen": 37158912
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000998716148445336,
+      "loss": 3.6938,
+      "theoretical_loss": 5.554047618546193,
+      "tokens_seen": 37224448
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009986960882647944,
+      "loss": 3.6552,
+      "theoretical_loss": 5.552512953800322,
+      "tokens_seen": 37289984
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009986760280842528,
+      "loss": 3.5607,
+      "theoretical_loss": 5.550981737498564,
+      "tokens_seen": 37355520
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009986559679037112,
+      "loss": 3.7794,
+      "theoretical_loss": 5.549453955862602,
+      "tokens_seen": 37421056
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009986359077231695,
+      "loss": 3.7612,
+      "theoretical_loss": 5.547929595193182,
+      "tokens_seen": 37486592
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009986158475426279,
+      "loss": 3.5894,
+      "theoretical_loss": 5.5464086418695135,
+      "tokens_seen": 37552128
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009985957873620862,
+      "loss": 3.5894,
+      "theoretical_loss": 5.544891082348692,
+      "tokens_seen": 37617664
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009985757271815446,
+      "loss": 3.6129,
+      "theoretical_loss": 5.543376903165111,
+      "tokens_seen": 37683200
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000998555667001003,
+      "loss": 3.5844,
+      "theoretical_loss": 5.5418660909298945,
+      "tokens_seen": 37748736
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009985356068204613,
+      "loss": 3.6763,
+      "theoretical_loss": 5.540358632330321,
+      "tokens_seen": 37814272
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009985155466399197,
+      "loss": 3.4916,
+      "theoretical_loss": 5.538854514129265,
+      "tokens_seen": 37879808
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000998495486459378,
+      "loss": 3.4597,
+      "theoretical_loss": 5.5373537231646335,
+      "tokens_seen": 37945344
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009984754262788366,
+      "loss": 3.5356,
+      "theoretical_loss": 5.535856246348814,
+      "tokens_seen": 38010880
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000998455366098295,
+      "loss": 3.6021,
+      "theoretical_loss": 5.534362070668127,
+      "tokens_seen": 38076416
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009984353059177534,
+      "loss": 3.5672,
+      "theoretical_loss": 5.532871183182278,
+      "tokens_seen": 38141952
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009984152457372117,
+      "loss": 3.5533,
+      "theoretical_loss": 5.53138357102382,
+      "tokens_seen": 38207488
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00099839518555667,
+      "loss": 3.6106,
+      "theoretical_loss": 5.529899221397624,
+      "tokens_seen": 38273024
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009983751253761284,
+      "loss": 3.4664,
+      "theoretical_loss": 5.528418121580339,
+      "tokens_seen": 38338560
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009983550651955868,
+      "loss": 3.3992,
+      "theoretical_loss": 5.526940258919881,
+      "tokens_seen": 38404096
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009983350050150452,
+      "loss": 3.6064,
+      "theoretical_loss": 5.5254656208349004,
+      "tokens_seen": 38469632
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009983149448345035,
+      "loss": 3.5422,
+      "theoretical_loss": 5.523994194814273,
+      "tokens_seen": 38535168
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000998294884653962,
+      "loss": 3.3976,
+      "theoretical_loss": 5.522525968416592,
+      "tokens_seen": 38600704
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009982748244734203,
+      "loss": 3.6676,
+      "theoretical_loss": 5.521060929269655,
+      "tokens_seen": 38666240
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009982547642928786,
+      "loss": 3.5906,
+      "theoretical_loss": 5.519599065069971,
+      "tokens_seen": 38731776
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000998234704112337,
+      "loss": 3.5077,
+      "theoretical_loss": 5.518140363582252,
+      "tokens_seen": 38797312
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009982146439317953,
+      "loss": 3.4583,
+      "theoretical_loss": 5.516684812638932,
+      "tokens_seen": 38862848
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000998194583751254,
+      "loss": 3.6336,
+      "theoretical_loss": 5.515232400139671,
+      "tokens_seen": 38928384
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000998174523570712,
+      "loss": 3.667,
+      "theoretical_loss": 5.5137831140508755,
+      "tokens_seen": 38993920
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009981544633901706,
+      "loss": 3.4816,
+      "theoretical_loss": 5.512336942405216,
+      "tokens_seen": 39059456
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009981344032096288,
+      "loss": 3.5222,
+      "theoretical_loss": 5.510893873301154,
+      "tokens_seen": 39124992
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009981143430290874,
+      "loss": 3.5284,
+      "theoretical_loss": 5.509453894902469,
+      "tokens_seen": 39190528
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009980942828485457,
+      "loss": 3.483,
+      "theoretical_loss": 5.508016995437791,
+      "tokens_seen": 39256064
+    },
+    {
+      "epoch": 0.01,
+      "objective/train/advantage_avg": 0.4876997768878937,
+      "objective/train/docs_used": 21760,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 3.0592448711395264,
+      "objective/train/original_loss": 3.0592451095581055,
+      "objective/train/theoretical_loss": 5.506583163200142,
+      "objective/train/tokens_used": 59781600,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.23880939185619354,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0499838590621948,
+      "objective/train/weighted_lm_loss": 3.212702512741089,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 5.506583163200142,
+      "tokens_seen": 39321600
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000998074222668004,
+      "loss": 3.3564,
+      "theoretical_loss": 5.506583163200142,
+      "tokens_seen": 39321600
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009980541624874625,
+      "loss": 3.5219,
+      "theoretical_loss": 5.505152386546474,
+      "tokens_seen": 39387136
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009980341023069208,
+      "loss": 3.6034,
+      "theoretical_loss": 5.503724653897207,
+      "tokens_seen": 39452672
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009980140421263792,
+      "loss": 3.5868,
+      "theoretical_loss": 5.502299953735795,
+      "tokens_seen": 39518208
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009979939819458375,
+      "loss": 3.43,
+      "theoretical_loss": 5.5008782746082625,
+      "tokens_seen": 39583744
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000997973921765296,
+      "loss": 3.6451,
+      "theoretical_loss": 5.499459605122775,
+      "tokens_seen": 39649280
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009979538615847543,
+      "loss": 3.3693,
+      "theoretical_loss": 5.498043933949187,
+      "tokens_seen": 39714816
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009979338014042126,
+      "loss": 3.5735,
+      "theoretical_loss": 5.496631249818616,
+      "tokens_seen": 39780352
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000997913741223671,
+      "loss": 3.4721,
+      "theoretical_loss": 5.495221541523011,
+      "tokens_seen": 39845888
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009978936810431294,
+      "loss": 3.4922,
+      "theoretical_loss": 5.493814797914718,
+      "tokens_seen": 39911424
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009978736208625877,
+      "loss": 3.6077,
+      "theoretical_loss": 5.49241100790606,
+      "tokens_seen": 39976960
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000997853560682046,
+      "loss": 3.5721,
+      "theoretical_loss": 5.491010160468919,
+      "tokens_seen": 40042496
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009978335005015047,
+      "loss": 3.399,
+      "theoretical_loss": 5.489612244634316,
+      "tokens_seen": 40108032
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009978134403209628,
+      "loss": 3.4734,
+      "theoretical_loss": 5.4882172494919965,
+      "tokens_seen": 40173568
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009977933801404214,
+      "loss": 3.542,
+      "theoretical_loss": 5.486825164190026,
+      "tokens_seen": 40239104
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009977733199598795,
+      "loss": 3.4436,
+      "theoretical_loss": 5.4854359779343795,
+      "tokens_seen": 40304640
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009977532597793381,
+      "loss": 3.5102,
+      "theoretical_loss": 5.48404967998854,
+      "tokens_seen": 40370176
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009977331995987965,
+      "loss": 3.4231,
+      "theoretical_loss": 5.482666259673101,
+      "tokens_seen": 40435712
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009977131394182548,
+      "loss": 3.6534,
+      "theoretical_loss": 5.481285706365366,
+      "tokens_seen": 40501248
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009976930792377132,
+      "loss": 3.5207,
+      "theoretical_loss": 5.479908009498965,
+      "tokens_seen": 40566784
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009976730190571716,
+      "loss": 3.4939,
+      "theoretical_loss": 5.478533158563456,
+      "tokens_seen": 40632320
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00099765295887663,
+      "loss": 3.7361,
+      "theoretical_loss": 5.477161143103945,
+      "tokens_seen": 40697856
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009976328986960883,
+      "loss": 3.4448,
+      "theoretical_loss": 5.4757919527207,
+      "tokens_seen": 40763392
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009976128385155466,
+      "loss": 3.5947,
+      "theoretical_loss": 5.474425577068777,
+      "tokens_seen": 40828928
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000997592778335005,
+      "loss": 3.4699,
+      "theoretical_loss": 5.473062005857637,
+      "tokens_seen": 40894464
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009975727181544634,
+      "loss": 3.6217,
+      "theoretical_loss": 5.471701228850778,
+      "tokens_seen": 40960000
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009975526579739217,
+      "loss": 3.6091,
+      "theoretical_loss": 5.470343235865368,
+      "tokens_seen": 41025536
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00099753259779338,
+      "loss": 3.4589,
+      "theoretical_loss": 5.468988016771869,
+      "tokens_seen": 41091072
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009975125376128385,
+      "loss": 3.4358,
+      "theoretical_loss": 5.467635561493681,
+      "tokens_seen": 41156608
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009974924774322968,
+      "loss": 3.4588,
+      "theoretical_loss": 5.466285860006783,
+      "tokens_seen": 41222144
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009974724172517554,
+      "loss": 3.6569,
+      "theoretical_loss": 5.46493890233937,
+      "tokens_seen": 41287680
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009974523570712135,
+      "loss": 3.453,
+      "theoretical_loss": 5.4635946785715,
+      "tokens_seen": 41353216
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009974322968906721,
+      "loss": 3.3928,
+      "theoretical_loss": 5.462253178834744,
+      "tokens_seen": 41418752
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009974122367101303,
+      "loss": 3.3296,
+      "theoretical_loss": 5.460914393311839,
+      "tokens_seen": 41484288
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009973921765295888,
+      "loss": 3.4467,
+      "theoretical_loss": 5.4595783122363395,
+      "tokens_seen": 41549824
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009973721163490472,
+      "loss": 3.4153,
+      "theoretical_loss": 5.458244925892275,
+      "tokens_seen": 41615360
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009973520561685056,
+      "loss": 3.4421,
+      "theoretical_loss": 5.456914224613812,
+      "tokens_seen": 41680896
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000997331995987964,
+      "loss": 3.4096,
+      "theoretical_loss": 5.455586198784916,
+      "tokens_seen": 41746432
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009973119358074223,
+      "loss": 3.4896,
+      "theoretical_loss": 5.454260838839016,
+      "tokens_seen": 41811968
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009972918756268807,
+      "loss": 3.5514,
+      "theoretical_loss": 5.452938135258673,
+      "tokens_seen": 41877504
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000997271815446339,
+      "loss": 3.3437,
+      "theoretical_loss": 5.451618078575256,
+      "tokens_seen": 41943040
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009972517552657974,
+      "loss": 3.2631,
+      "theoretical_loss": 5.450300659368612,
+      "tokens_seen": 42008576
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009972316950852557,
+      "loss": 3.3576,
+      "theoretical_loss": 5.448985868266736,
+      "tokens_seen": 42074112
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000997211634904714,
+      "loss": 3.5492,
+      "theoretical_loss": 5.447673695945468,
+      "tokens_seen": 42139648
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009971915747241725,
+      "loss": 3.406,
+      "theoretical_loss": 5.446364133128155,
+      "tokens_seen": 42205184
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009971715145436308,
+      "loss": 3.4232,
+      "theoretical_loss": 5.4450571705853505,
+      "tokens_seen": 42270720
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009971514543630892,
+      "loss": 3.1882,
+      "theoretical_loss": 5.443752799134492,
+      "tokens_seen": 42336256
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009971313941825476,
+      "loss": 3.4595,
+      "theoretical_loss": 5.442451009639596,
+      "tokens_seen": 42401792
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009971113340020061,
+      "loss": 3.3603,
+      "theoretical_loss": 5.44115179301095,
+      "tokens_seen": 42467328
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009970912738214643,
+      "loss": 3.5864,
+      "theoretical_loss": 5.439855140204802,
+      "tokens_seen": 42532864
+    },
+    {
+      "epoch": 0.01,
+      "objective/train/advantage_avg": 0.49280866980552673,
+      "objective/train/docs_used": 22938,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 3.297335386276245,
+      "objective/train/original_loss": 3.297335624694824,
+      "objective/train/theoretical_loss": 5.438561042223066,
+      "objective/train/tokens_used": 63058400,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24311578273773193,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505168437957764,
+      "objective/train/weighted_lm_loss": 3.463775634765625,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 1.0253151655197144,
+      "theoretical_loss": 5.438561042223066,
+      "tokens_seen": 42598400
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009970712136409229,
+      "loss": 3.5201,
+      "theoretical_loss": 5.438561042223066,
+      "tokens_seen": 42598400
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000997051153460381,
+      "loss": 3.3525,
+      "theoretical_loss": 5.437269490113014,
+      "tokens_seen": 42663936
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009970310932798396,
+      "loss": 3.3086,
+      "theoretical_loss": 5.435980474966981,
+      "tokens_seen": 42729472
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000997011033099298,
+      "loss": 3.3753,
+      "theoretical_loss": 5.434693987922072,
+      "tokens_seen": 42795008
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009969909729187563,
+      "loss": 3.3824,
+      "theoretical_loss": 5.433410020159865,
+      "tokens_seen": 42860544
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009969709127382147,
+      "loss": 3.2654,
+      "theoretical_loss": 5.432128562906122,
+      "tokens_seen": 42926080
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000996950852557673,
+      "loss": 3.384,
+      "theoretical_loss": 5.430849607430501,
+      "tokens_seen": 42991616
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009969307923771314,
+      "loss": 3.2841,
+      "theoretical_loss": 5.429573145046266,
+      "tokens_seen": 43057152
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009969107321965898,
+      "loss": 3.422,
+      "theoretical_loss": 5.428299167110012,
+      "tokens_seen": 43122688
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009968906720160481,
+      "loss": 3.3227,
+      "theoretical_loss": 5.427027665021371,
+      "tokens_seen": 43188224
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009968706118355065,
+      "loss": 3.285,
+      "theoretical_loss": 5.425758630222747,
+      "tokens_seen": 43253760
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000996850551654965,
+      "loss": 3.3776,
+      "theoretical_loss": 5.424492054199025,
+      "tokens_seen": 43319296
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009968304914744232,
+      "loss": 3.267,
+      "theoretical_loss": 5.423227928477306,
+      "tokens_seen": 43384832
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009968104312938818,
+      "loss": 3.3353,
+      "theoretical_loss": 5.421966244626631,
+      "tokens_seen": 43450368
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00099679037111334,
+      "loss": 3.5885,
+      "theoretical_loss": 5.42070699425771,
+      "tokens_seen": 43515904
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009967703109327985,
+      "loss": 3.4124,
+      "theoretical_loss": 5.419450169022653,
+      "tokens_seen": 43581440
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009967502507522569,
+      "loss": 3.4499,
+      "theoretical_loss": 5.418195760614707,
+      "tokens_seen": 43646976
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009967301905717152,
+      "loss": 3.395,
+      "theoretical_loss": 5.416943760767989,
+      "tokens_seen": 43712512
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009967101303911736,
+      "loss": 3.4137,
+      "theoretical_loss": 5.415694161257225,
+      "tokens_seen": 43778048
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000996690070210632,
+      "loss": 3.3275,
+      "theoretical_loss": 5.414446953897489,
+      "tokens_seen": 43843584
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009966700100300903,
+      "loss": 3.1889,
+      "theoretical_loss": 5.413202130543947,
+      "tokens_seen": 43909120
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009966499498495487,
+      "loss": 3.2611,
+      "theoretical_loss": 5.411959683091602,
+      "tokens_seen": 43974656
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000996629889669007,
+      "loss": 3.339,
+      "theoretical_loss": 5.410719603475034,
+      "tokens_seen": 44040192
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009966098294884654,
+      "loss": 3.5504,
+      "theoretical_loss": 5.409481883668155,
+      "tokens_seen": 44105728
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009965897693079238,
+      "loss": 3.3063,
+      "theoretical_loss": 5.408246515683956,
+      "tokens_seen": 44171264
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009965697091273821,
+      "loss": 3.145,
+      "theoretical_loss": 5.407013491574263,
+      "tokens_seen": 44236800
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009965496489468405,
+      "loss": 3.3912,
+      "theoretical_loss": 5.405782803429483,
+      "tokens_seen": 44302336
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009965295887662989,
+      "loss": 3.2364,
+      "theoretical_loss": 5.404554443378366,
+      "tokens_seen": 44367872
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009965095285857572,
+      "loss": 3.3157,
+      "theoretical_loss": 5.403328403587762,
+      "tokens_seen": 44433408
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009964894684052158,
+      "loss": 3.2918,
+      "theoretical_loss": 5.402104676262381,
+      "tokens_seen": 44498944
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000996469408224674,
+      "loss": 3.2743,
+      "theoretical_loss": 5.400883253644551,
+      "tokens_seen": 44564480
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009964493480441325,
+      "loss": 3.3715,
+      "theoretical_loss": 5.399664128013988,
+      "tokens_seen": 44630016
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009964292878635907,
+      "loss": 3.2805,
+      "theoretical_loss": 5.398447291687555,
+      "tokens_seen": 44695552
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009964092276830492,
+      "loss": 3.3267,
+      "theoretical_loss": 5.397232737019031,
+      "tokens_seen": 44761088
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009963891675025076,
+      "loss": 3.3381,
+      "theoretical_loss": 5.396020456398885,
+      "tokens_seen": 44826624
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000996369107321966,
+      "loss": 3.1136,
+      "theoretical_loss": 5.394810442254032,
+      "tokens_seen": 44892160
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009963490471414243,
+      "loss": 3.2009,
+      "theoretical_loss": 5.393602687047627,
+      "tokens_seen": 44957696
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009963289869608827,
+      "loss": 3.3298,
+      "theoretical_loss": 5.392397183278822,
+      "tokens_seen": 45023232
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000996308926780341,
+      "loss": 3.1392,
+      "theoretical_loss": 5.391193923482547,
+      "tokens_seen": 45088768
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009962888665997994,
+      "loss": 3.3421,
+      "theoretical_loss": 5.38999290022929,
+      "tokens_seen": 45154304
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009962688064192578,
+      "loss": 3.2676,
+      "theoretical_loss": 5.388794106124874,
+      "tokens_seen": 45219840
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009962487462387161,
+      "loss": 3.2214,
+      "theoretical_loss": 5.387597533810238,
+      "tokens_seen": 45285376
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009962286860581745,
+      "loss": 3.245,
+      "theoretical_loss": 5.386403175961223,
+      "tokens_seen": 45350912
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009962086258776329,
+      "loss": 3.177,
+      "theoretical_loss": 5.385211025288349,
+      "tokens_seen": 45416448
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009961885656970912,
+      "loss": 3.1048,
+      "theoretical_loss": 5.38402107453661,
+      "tokens_seen": 45481984
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009961685055165496,
+      "loss": 3.2362,
+      "theoretical_loss": 5.382833316485255,
+      "tokens_seen": 45547520
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000996148445336008,
+      "loss": 3.3837,
+      "theoretical_loss": 5.381647743947578,
+      "tokens_seen": 45613056
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009961283851554665,
+      "loss": 3.3365,
+      "theoretical_loss": 5.3804643497707145,
+      "tokens_seen": 45678592
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009961083249749247,
+      "loss": 3.2244,
+      "theoretical_loss": 5.379283126835426,
+      "tokens_seen": 45744128
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009960882647943833,
+      "loss": 3.3835,
+      "theoretical_loss": 5.378104068055899,
+      "tokens_seen": 45809664
+    },
+    {
+      "epoch": 0.01,
+      "objective/train/advantage_avg": 0.4905271530151367,
+      "objective/train/docs_used": 24173,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 2.979590654373169,
+      "objective/train/original_loss": 2.97959041595459,
+      "objective/train/theoretical_loss": 5.37692716637954,
+      "objective/train/tokens_used": 66335200,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24144382774829865,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0502800941467285,
+      "objective/train/weighted_lm_loss": 3.129429817199707,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 5.37692716637954,
+      "tokens_seen": 45875200
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009960682046138414,
+      "loss": 3.1644,
+      "theoretical_loss": 5.37692716637954,
+      "tokens_seen": 45875200
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009960481444333,
+      "loss": 3.2088,
+      "theoretical_loss": 5.375752414786769,
+      "tokens_seen": 45940736
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009960280842527583,
+      "loss": 3.3534,
+      "theoretical_loss": 5.374579806290823,
+      "tokens_seen": 46006272
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009960080240722167,
+      "loss": 3.1675,
+      "theoretical_loss": 5.373409333937557,
+      "tokens_seen": 46071808
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000995987963891675,
+      "loss": 3.1467,
+      "theoretical_loss": 5.372240990805237,
+      "tokens_seen": 46137344
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009959679037111334,
+      "loss": 3.2232,
+      "theoretical_loss": 5.37107477000435,
+      "tokens_seen": 46202880
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009959478435305918,
+      "loss": 3.2601,
+      "theoretical_loss": 5.369910664677409,
+      "tokens_seen": 46268416
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009959277833500502,
+      "loss": 2.9964,
+      "theoretical_loss": 5.368748667998759,
+      "tokens_seen": 46333952
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009959077231695085,
+      "loss": 3.1695,
+      "theoretical_loss": 5.367588773174377,
+      "tokens_seen": 46399488
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009958876629889669,
+      "loss": 3.2412,
+      "theoretical_loss": 5.366430973441693,
+      "tokens_seen": 46465024
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009958676028084252,
+      "loss": 3.4332,
+      "theoretical_loss": 5.365275262069391,
+      "tokens_seen": 46530560
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009958475426278836,
+      "loss": 3.0728,
+      "theoretical_loss": 5.364121632357221,
+      "tokens_seen": 46596096
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000995827482447342,
+      "loss": 3.3624,
+      "theoretical_loss": 5.36297007763582,
+      "tokens_seen": 46661632
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009958074222668003,
+      "loss": 3.2145,
+      "theoretical_loss": 5.36182059126652,
+      "tokens_seen": 46727168
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009957873620862587,
+      "loss": 3.2506,
+      "theoretical_loss": 5.360673166641163,
+      "tokens_seen": 46792704
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009957673019057173,
+      "loss": 3.229,
+      "theoretical_loss": 5.359527797181926,
+      "tokens_seen": 46858240
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009957472417251754,
+      "loss": 3.2049,
+      "theoretical_loss": 5.358384476341126,
+      "tokens_seen": 46923776
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000995727181544634,
+      "loss": 3.129,
+      "theoretical_loss": 5.357243197601056,
+      "tokens_seen": 46989312
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009957071213640921,
+      "loss": 3.2963,
+      "theoretical_loss": 5.356103954473793,
+      "tokens_seen": 47054848
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009956870611835507,
+      "loss": 3.3261,
+      "theoretical_loss": 5.35496674050103,
+      "tokens_seen": 47120384
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000995667001003009,
+      "loss": 3.083,
+      "theoretical_loss": 5.353831549253895,
+      "tokens_seen": 47185920
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009956469408224674,
+      "loss": 3.1944,
+      "theoretical_loss": 5.352698374332774,
+      "tokens_seen": 47251456
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009956268806419258,
+      "loss": 3.2186,
+      "theoretical_loss": 5.351567209367143,
+      "tokens_seen": 47316992
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009956068204613842,
+      "loss": 3.387,
+      "theoretical_loss": 5.350438048015396,
+      "tokens_seen": 47382528
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009955867602808425,
+      "loss": 3.0156,
+      "theoretical_loss": 5.349310883964664,
+      "tokens_seen": 47448064
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000995566700100301,
+      "loss": 3.0316,
+      "theoretical_loss": 5.348185710930661,
+      "tokens_seen": 47513600
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009955466399197593,
+      "loss": 3.2195,
+      "theoretical_loss": 5.3470625226575,
+      "tokens_seen": 47579136
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009955265797392176,
+      "loss": 3.1434,
+      "theoretical_loss": 5.345941312917536,
+      "tokens_seen": 47644672
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000995506519558676,
+      "loss": 3.2259,
+      "theoretical_loss": 5.344822075511196,
+      "tokens_seen": 47710208
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009954864593781343,
+      "loss": 3.0835,
+      "theoretical_loss": 5.343704804266816,
+      "tokens_seen": 47775744
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000995466399197593,
+      "loss": 3.1539,
+      "theoretical_loss": 5.342589493040472,
+      "tokens_seen": 47841280
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000995446339017051,
+      "loss": 3.1759,
+      "theoretical_loss": 5.3414761357158245,
+      "tokens_seen": 47906816
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009954262788365096,
+      "loss": 3.0574,
+      "theoretical_loss": 5.340364726203955,
+      "tokens_seen": 47972352
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000995406218655968,
+      "loss": 3.0117,
+      "theoretical_loss": 5.339255258443202,
+      "tokens_seen": 48037888
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009953861584754264,
+      "loss": 2.9506,
+      "theoretical_loss": 5.338147726399006,
+      "tokens_seen": 48103424
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009953660982948847,
+      "loss": 2.8871,
+      "theoretical_loss": 5.33704212406375,
+      "tokens_seen": 48168960
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000995346038114343,
+      "loss": 3.272,
+      "theoretical_loss": 5.3359384454566055,
+      "tokens_seen": 48234496
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009953259779338015,
+      "loss": 3.1474,
+      "theoretical_loss": 5.334836684623371,
+      "tokens_seen": 48300032
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009953059177532598,
+      "loss": 3.013,
+      "theoretical_loss": 5.333736835636325,
+      "tokens_seen": 48365568
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009952858575727182,
+      "loss": 2.9694,
+      "theoretical_loss": 5.332638892594063,
+      "tokens_seen": 48431104
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009952657973921765,
+      "loss": 3.0434,
+      "theoretical_loss": 5.331542849621357,
+      "tokens_seen": 48496640
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000995245737211635,
+      "loss": 3.0976,
+      "theoretical_loss": 5.330448700868995,
+      "tokens_seen": 48562176
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009952256770310933,
+      "loss": 3.132,
+      "theoretical_loss": 5.329356440513633,
+      "tokens_seen": 48627712
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009952056168505516,
+      "loss": 3.1594,
+      "theoretical_loss": 5.32826606275765,
+      "tokens_seen": 48693248
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00099518555667001,
+      "loss": 3.0358,
+      "theoretical_loss": 5.327177561828993,
+      "tokens_seen": 48758784
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009951654964894684,
+      "loss": 3.04,
+      "theoretical_loss": 5.326090931981036,
+      "tokens_seen": 48824320
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000995145436308927,
+      "loss": 3.1244,
+      "theoretical_loss": 5.3250061674924325,
+      "tokens_seen": 48889856
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.000995125376128385,
+      "loss": 3.1184,
+      "theoretical_loss": 5.323923262666968,
+      "tokens_seen": 48955392
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009951053159478437,
+      "loss": 3.13,
+      "theoretical_loss": 5.32284221183342,
+      "tokens_seen": 49020928
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009950852557673018,
+      "loss": 3.1912,
+      "theoretical_loss": 5.321763009345411,
+      "tokens_seen": 49086464
+    },
+    {
+      "epoch": 0.01,
+      "objective/train/advantage_avg": 0.4820156991481781,
+      "objective/train/docs_used": 25394,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 2.984027862548828,
+      "objective/train/original_loss": 2.98402738571167,
+      "objective/train/theoretical_loss": 5.3206856495812715,
+      "objective/train/tokens_used": 69612000,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2343626171350479,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0493927001953125,
+      "objective/train/weighted_lm_loss": 3.1309571266174316,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 5.3206856495812715,
+      "tokens_seen": 49152000
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009950651955867604,
+      "loss": 3.0822,
+      "theoretical_loss": 5.3206856495812715,
+      "tokens_seen": 49152000
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009950451354062187,
+      "loss": 3.0233,
+      "theoretical_loss": 5.3196101269438945,
+      "tokens_seen": 49217536
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009950250752256771,
+      "loss": 2.9487,
+      "theoretical_loss": 5.318536435860599,
+      "tokens_seen": 49283072
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009950050150451355,
+      "loss": 3.0769,
+      "theoretical_loss": 5.317464570782989,
+      "tokens_seen": 49348608
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009949849548645938,
+      "loss": 3.0474,
+      "theoretical_loss": 5.316394526186816,
+      "tokens_seen": 49414144
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0009949648946840522,
+      "loss": 2.7961,
+      "theoretical_loss": 5.3153262965718415,
+      "tokens_seen": 49479680
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009949448345035106,
+      "loss": 2.9753,
+      "theoretical_loss": 5.314259876461705,
+      "tokens_seen": 49545216
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000994924774322969,
+      "loss": 3.0748,
+      "theoretical_loss": 5.3131952604037815,
+      "tokens_seen": 49610752
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009949047141424273,
+      "loss": 2.9606,
+      "theoretical_loss": 5.312132442969055,
+      "tokens_seen": 49676288
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009948846539618856,
+      "loss": 2.9401,
+      "theoretical_loss": 5.3110714187519825,
+      "tokens_seen": 49741824
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000994864593781344,
+      "loss": 2.9654,
+      "theoretical_loss": 5.310012182370359,
+      "tokens_seen": 49807360
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009948445336008024,
+      "loss": 2.8686,
+      "theoretical_loss": 5.308954728465188,
+      "tokens_seen": 49872896
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009948244734202607,
+      "loss": 3.0768,
+      "theoretical_loss": 5.307899051700552,
+      "tokens_seen": 49938432
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000994804413239719,
+      "loss": 2.9913,
+      "theoretical_loss": 5.306845146763486,
+      "tokens_seen": 50003968
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009947843530591777,
+      "loss": 2.9679,
+      "theoretical_loss": 5.305793008363841,
+      "tokens_seen": 50069504
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009947642928786358,
+      "loss": 3.0225,
+      "theoretical_loss": 5.304742631234162,
+      "tokens_seen": 50135040
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009947442326980944,
+      "loss": 3.04,
+      "theoretical_loss": 5.3036940101295595,
+      "tokens_seen": 50200576
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009947241725175525,
+      "loss": 2.8716,
+      "theoretical_loss": 5.302647139827585,
+      "tokens_seen": 50266112
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009947041123370111,
+      "loss": 2.9582,
+      "theoretical_loss": 5.301602015128104,
+      "tokens_seen": 50331648
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009946840521564695,
+      "loss": 3.0407,
+      "theoretical_loss": 5.300558630853171,
+      "tokens_seen": 50397184
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009946639919759278,
+      "loss": 2.9239,
+      "theoretical_loss": 5.299516981846912,
+      "tokens_seen": 50462720
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009946439317953862,
+      "loss": 2.909,
+      "theoretical_loss": 5.29847706297539,
+      "tokens_seen": 50528256
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009946238716148446,
+      "loss": 2.952,
+      "theoretical_loss": 5.297438869126498,
+      "tokens_seen": 50593792
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000994603811434303,
+      "loss": 2.9815,
+      "theoretical_loss": 5.296402395209828,
+      "tokens_seen": 50659328
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009945837512537613,
+      "loss": 2.9763,
+      "theoretical_loss": 5.295367636156552,
+      "tokens_seen": 50724864
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009945636910732197,
+      "loss": 3.1347,
+      "theoretical_loss": 5.294334586919307,
+      "tokens_seen": 50790400
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000994543630892678,
+      "loss": 3.0942,
+      "theoretical_loss": 5.293303242472074,
+      "tokens_seen": 50855936
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009945235707121364,
+      "loss": 2.9039,
+      "theoretical_loss": 5.29227359781006,
+      "tokens_seen": 50921472
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009945035105315947,
+      "loss": 3.0075,
+      "theoretical_loss": 5.29124564794958,
+      "tokens_seen": 50987008
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000994483450351053,
+      "loss": 2.8655,
+      "theoretical_loss": 5.290219387927946,
+      "tokens_seen": 51052544
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009944633901705115,
+      "loss": 2.8902,
+      "theoretical_loss": 5.289194812803347,
+      "tokens_seen": 51118080
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009944433299899698,
+      "loss": 3.0833,
+      "theoretical_loss": 5.288171917654733,
+      "tokens_seen": 51183616
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009944232698094284,
+      "loss": 2.801,
+      "theoretical_loss": 5.287150697581708,
+      "tokens_seen": 51249152
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009944032096288866,
+      "loss": 2.7836,
+      "theoretical_loss": 5.286131147704413,
+      "tokens_seen": 51314688
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009943831494483451,
+      "loss": 2.9264,
+      "theoretical_loss": 5.285113263163414,
+      "tokens_seen": 51380224
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009943630892678035,
+      "loss": 2.9479,
+      "theoretical_loss": 5.284097039119587,
+      "tokens_seen": 51445760
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009943430290872619,
+      "loss": 3.0413,
+      "theoretical_loss": 5.283082470754016,
+      "tokens_seen": 51511296
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009943229689067202,
+      "loss": 2.8968,
+      "theoretical_loss": 5.282069553267874,
+      "tokens_seen": 51576832
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009943029087261786,
+      "loss": 2.7193,
+      "theoretical_loss": 5.2810582818823235,
+      "tokens_seen": 51642368
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000994282848545637,
+      "loss": 2.9009,
+      "theoretical_loss": 5.2800486518383964,
+      "tokens_seen": 51707904
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009942627883650953,
+      "loss": 2.8084,
+      "theoretical_loss": 5.279040658396896,
+      "tokens_seen": 51773440
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009942427281845537,
+      "loss": 2.9595,
+      "theoretical_loss": 5.278034296838284,
+      "tokens_seen": 51838976
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000994222668004012,
+      "loss": 2.8929,
+      "theoretical_loss": 5.27702956246258,
+      "tokens_seen": 51904512
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009942026078234704,
+      "loss": 2.8877,
+      "theoretical_loss": 5.276026450589249,
+      "tokens_seen": 51970048
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009941825476429288,
+      "loss": 2.9187,
+      "theoretical_loss": 5.275024956557102,
+      "tokens_seen": 52035584
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009941624874623871,
+      "loss": 2.7947,
+      "theoretical_loss": 5.274025075724189,
+      "tokens_seen": 52101120
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009941424272818455,
+      "loss": 2.7595,
+      "theoretical_loss": 5.273026803467695,
+      "tokens_seen": 52166656
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009941223671013038,
+      "loss": 2.9168,
+      "theoretical_loss": 5.272030135183842,
+      "tokens_seen": 52232192
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009941023069207622,
+      "loss": 2.872,
+      "theoretical_loss": 5.271035066287779,
+      "tokens_seen": 52297728
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009940822467402206,
+      "loss": 2.9237,
+      "theoretical_loss": 5.27004159221349,
+      "tokens_seen": 52363264
+    },
+    {
+      "epoch": 0.02,
+      "objective/train/advantage_avg": 0.49426913261413574,
+      "objective/train/docs_used": 26541,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 2.603318452835083,
+      "objective/train/original_loss": 2.603318691253662,
+      "objective/train/theoretical_loss": 5.269049708413682,
+      "objective/train/tokens_used": 72888800,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24451129138469696,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506700277328491,
+      "objective/train/weighted_lm_loss": 2.735368251800537,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 1.0304546356201172,
+      "theoretical_loss": 5.269049708413682,
+      "tokens_seen": 52428800
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009940621865596791,
+      "loss": 2.7216,
+      "theoretical_loss": 5.269049708413682,
+      "tokens_seen": 52428800
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009940421263791373,
+      "loss": 2.857,
+      "theoretical_loss": 5.268059410359697,
+      "tokens_seen": 52494336
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009940220661985959,
+      "loss": 2.9196,
+      "theoretical_loss": 5.267070693541403,
+      "tokens_seen": 52559872
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009940020060180542,
+      "loss": 2.8979,
+      "theoretical_loss": 5.2660835534671,
+      "tokens_seen": 52625408
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009939819458375126,
+      "loss": 2.862,
+      "theoretical_loss": 5.265097985663418,
+      "tokens_seen": 52690944
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000993961885656971,
+      "loss": 2.9804,
+      "theoretical_loss": 5.264113985675227,
+      "tokens_seen": 52756480
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009939418254764293,
+      "loss": 2.8927,
+      "theoretical_loss": 5.263131549065529,
+      "tokens_seen": 52822016
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009939217652958877,
+      "loss": 2.786,
+      "theoretical_loss": 5.262150671415373,
+      "tokens_seen": 52887552
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000993901705115346,
+      "loss": 2.7832,
+      "theoretical_loss": 5.261171348323755,
+      "tokens_seen": 52953088
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009938816449348044,
+      "loss": 2.9547,
+      "theoretical_loss": 5.2601935754075155,
+      "tokens_seen": 53018624
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009938615847542628,
+      "loss": 2.8122,
+      "theoretical_loss": 5.259217348301259,
+      "tokens_seen": 53084160
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009938415245737211,
+      "loss": 2.7562,
+      "theoretical_loss": 5.25824266265725,
+      "tokens_seen": 53149696
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009938214643931795,
+      "loss": 2.8343,
+      "theoretical_loss": 5.257269514145324,
+      "tokens_seen": 53215232
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000993801404212638,
+      "loss": 2.7475,
+      "theoretical_loss": 5.256297898452793,
+      "tokens_seen": 53280768
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009937813440320962,
+      "loss": 2.7761,
+      "theoretical_loss": 5.2553278112843556,
+      "tokens_seen": 53346304
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009937612838515548,
+      "loss": 2.7001,
+      "theoretical_loss": 5.254359248362006,
+      "tokens_seen": 53411840
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000993741223671013,
+      "loss": 2.8198,
+      "theoretical_loss": 5.2533922054249365,
+      "tokens_seen": 53477376
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009937211634904715,
+      "loss": 2.7642,
+      "theoretical_loss": 5.25242667822946,
+      "tokens_seen": 53542912
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009937011033099299,
+      "loss": 2.8012,
+      "theoretical_loss": 5.251462662548908,
+      "tokens_seen": 53608448
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009936810431293882,
+      "loss": 2.7885,
+      "theoretical_loss": 5.25050015417355,
+      "tokens_seen": 53673984
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009936609829488466,
+      "loss": 2.8942,
+      "theoretical_loss": 5.2495391489104986,
+      "tokens_seen": 53739520
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000993640922768305,
+      "loss": 2.9782,
+      "theoretical_loss": 5.248579642583628,
+      "tokens_seen": 53805056
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009936208625877633,
+      "loss": 2.8523,
+      "theoretical_loss": 5.24762163103348,
+      "tokens_seen": 53870592
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009936008024072217,
+      "loss": 2.7695,
+      "theoretical_loss": 5.246665110117186,
+      "tokens_seen": 53936128
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00099358074222668,
+      "loss": 2.8059,
+      "theoretical_loss": 5.24571007570837,
+      "tokens_seen": 54001664
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009935606820461384,
+      "loss": 2.8622,
+      "theoretical_loss": 5.244756523697072,
+      "tokens_seen": 54067200
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009935406218655968,
+      "loss": 2.8403,
+      "theoretical_loss": 5.243804449989659,
+      "tokens_seen": 54132736
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009935205616850551,
+      "loss": 2.5313,
+      "theoretical_loss": 5.242853850508741,
+      "tokens_seen": 54198272
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009935005015045135,
+      "loss": 2.8811,
+      "theoretical_loss": 5.2419047211930865,
+      "tokens_seen": 54263808
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009934804413239719,
+      "loss": 2.7909,
+      "theoretical_loss": 5.24095705799754,
+      "tokens_seen": 54329344
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009934603811434302,
+      "loss": 2.6954,
+      "theoretical_loss": 5.240010856892939,
+      "tokens_seen": 54394880
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009934403209628888,
+      "loss": 2.7536,
+      "theoretical_loss": 5.239066113866029,
+      "tokens_seen": 54460416
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000993420260782347,
+      "loss": 2.7655,
+      "theoretical_loss": 5.238122824919387,
+      "tokens_seen": 54525952
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009934002006018055,
+      "loss": 2.8289,
+      "theoretical_loss": 5.237180986071335,
+      "tokens_seen": 54591488
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000993380140421264,
+      "loss": 2.7578,
+      "theoretical_loss": 5.236240593355863,
+      "tokens_seen": 54657024
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009933600802407223,
+      "loss": 2.6962,
+      "theoretical_loss": 5.235301642822543,
+      "tokens_seen": 54722560
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009933400200601806,
+      "loss": 2.7935,
+      "theoretical_loss": 5.234364130536457,
+      "tokens_seen": 54788096
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000993319959879639,
+      "loss": 2.7993,
+      "theoretical_loss": 5.2334280525781125,
+      "tokens_seen": 54853632
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009932998996990973,
+      "loss": 2.8282,
+      "theoretical_loss": 5.232493405043362,
+      "tokens_seen": 54919168
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009932798395185557,
+      "loss": 2.8451,
+      "theoretical_loss": 5.231560184043333,
+      "tokens_seen": 54984704
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000993259779338014,
+      "loss": 2.7866,
+      "theoretical_loss": 5.230628385704337,
+      "tokens_seen": 55050240
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009932397191574724,
+      "loss": 2.9565,
+      "theoretical_loss": 5.229698006167809,
+      "tokens_seen": 55115776
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009932196589769308,
+      "loss": 2.5017,
+      "theoretical_loss": 5.228769041590215,
+      "tokens_seen": 55181312
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009931995987963892,
+      "loss": 2.7794,
+      "theoretical_loss": 5.227841488142982,
+      "tokens_seen": 55246848
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009931795386158475,
+      "loss": 2.6695,
+      "theoretical_loss": 5.2269153420124255,
+      "tokens_seen": 55312384
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009931594784353059,
+      "loss": 2.7819,
+      "theoretical_loss": 5.22599059939967,
+      "tokens_seen": 55377920
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009931394182547642,
+      "loss": 2.75,
+      "theoretical_loss": 5.225067256520572,
+      "tokens_seen": 55443456
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009931193580742226,
+      "loss": 2.5643,
+      "theoretical_loss": 5.224145309605651,
+      "tokens_seen": 55508992
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000993099297893681,
+      "loss": 2.7934,
+      "theoretical_loss": 5.223224754900014,
+      "tokens_seen": 55574528
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009930792377131395,
+      "loss": 2.7763,
+      "theoretical_loss": 5.222305588663275,
+      "tokens_seen": 55640064
+    },
+    {
+      "epoch": 0.02,
+      "objective/train/advantage_avg": 0.4954495131969452,
+      "objective/train/docs_used": 27758,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 2.493486166000366,
+      "objective/train/original_loss": 2.493486166000366,
+      "objective/train/theoretical_loss": 5.221387807169494,
+      "objective/train/tokens_used": 76165600,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24568335711956024,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0507941246032715,
+      "objective/train/weighted_lm_loss": 2.6202352046966553,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 5.221387807169494,
+      "tokens_seen": 55705600
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009930591775325977,
+      "loss": 2.6668,
+      "theoretical_loss": 5.221387807169494,
+      "tokens_seen": 55705600
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009930391173520563,
+      "loss": 2.9847,
+      "theoretical_loss": 5.2204714067070945,
+      "tokens_seen": 55771136
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009930190571715146,
+      "loss": 2.6578,
+      "theoretical_loss": 5.219556383578795,
+      "tokens_seen": 55836672
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000992998996990973,
+      "loss": 2.7279,
+      "theoretical_loss": 5.218642734101538,
+      "tokens_seen": 55902208
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009929789368104314,
+      "loss": 2.6868,
+      "theoretical_loss": 5.217730454606417,
+      "tokens_seen": 55967744
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009929588766298897,
+      "loss": 2.79,
+      "theoretical_loss": 5.216819541438607,
+      "tokens_seen": 56033280
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000992938816449348,
+      "loss": 2.7179,
+      "theoretical_loss": 5.215909990957291,
+      "tokens_seen": 56098816
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009929187562688064,
+      "loss": 2.7788,
+      "theoretical_loss": 5.215001799535598,
+      "tokens_seen": 56164352
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009928986960882648,
+      "loss": 2.739,
+      "theoretical_loss": 5.214094963560519,
+      "tokens_seen": 56229888
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009928786359077232,
+      "loss": 2.5606,
+      "theoretical_loss": 5.213189479432854,
+      "tokens_seen": 56295424
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009928585757271815,
+      "loss": 2.734,
+      "theoretical_loss": 5.212285343567135,
+      "tokens_seen": 56360960
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00099283851554664,
+      "loss": 2.5668,
+      "theoretical_loss": 5.211382552391555,
+      "tokens_seen": 56426496
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009928184553660983,
+      "loss": 2.7951,
+      "theoretical_loss": 5.210481102347908,
+      "tokens_seen": 56492032
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009927983951855566,
+      "loss": 2.6594,
+      "theoretical_loss": 5.2095809898915135,
+      "tokens_seen": 56557568
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000992778335005015,
+      "loss": 2.7474,
+      "theoretical_loss": 5.208682211491157,
+      "tokens_seen": 56623104
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009927582748244733,
+      "loss": 2.687,
+      "theoretical_loss": 5.207784763629014,
+      "tokens_seen": 56688640
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009927382146439317,
+      "loss": 2.5976,
+      "theoretical_loss": 5.206888642800598,
+      "tokens_seen": 56754176
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009927181544633903,
+      "loss": 2.6364,
+      "theoretical_loss": 5.205993845514678,
+      "tokens_seen": 56819712
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009926980942828484,
+      "loss": 2.6975,
+      "theoretical_loss": 5.205100368293225,
+      "tokens_seen": 56885248
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000992678034102307,
+      "loss": 2.6616,
+      "theoretical_loss": 5.204208207671343,
+      "tokens_seen": 56950784
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009926579739217654,
+      "loss": 2.6684,
+      "theoretical_loss": 5.203317360197204,
+      "tokens_seen": 57016320
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009926379137412237,
+      "loss": 2.629,
+      "theoretical_loss": 5.202427822431982,
+      "tokens_seen": 57081856
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000992617853560682,
+      "loss": 2.6768,
+      "theoretical_loss": 5.201539590949796,
+      "tokens_seen": 57147392
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009925977933801405,
+      "loss": 2.5754,
+      "theoretical_loss": 5.200652662337637,
+      "tokens_seen": 57212928
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009925777331995988,
+      "loss": 2.5187,
+      "theoretical_loss": 5.199767033195316,
+      "tokens_seen": 57278464
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009925576730190572,
+      "loss": 2.7063,
+      "theoretical_loss": 5.198882700135387,
+      "tokens_seen": 57344000
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009925376128385155,
+      "loss": 2.7405,
+      "theoretical_loss": 5.1979996597831,
+      "tokens_seen": 57409536
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000992517552657974,
+      "loss": 2.6938,
+      "theoretical_loss": 5.1971179087763275,
+      "tokens_seen": 57475072
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009924974924774323,
+      "loss": 2.6516,
+      "theoretical_loss": 5.19623744376551,
+      "tokens_seen": 57540608
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009924774322968906,
+      "loss": 2.6739,
+      "theoretical_loss": 5.195358261413589,
+      "tokens_seen": 57606144
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009924573721163492,
+      "loss": 2.5905,
+      "theoretical_loss": 5.1944803583959525,
+      "tokens_seen": 57671680
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009924373119358074,
+      "loss": 2.6303,
+      "theoretical_loss": 5.193603731400369,
+      "tokens_seen": 57737216
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000992417251755266,
+      "loss": 2.6194,
+      "theoretical_loss": 5.192728377126931,
+      "tokens_seen": 57802752
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009923971915747243,
+      "loss": 2.7367,
+      "theoretical_loss": 5.191854292287992,
+      "tokens_seen": 57868288
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009923771313941827,
+      "loss": 2.7908,
+      "theoretical_loss": 5.190981473608112,
+      "tokens_seen": 57933824
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000992357071213641,
+      "loss": 2.5301,
+      "theoretical_loss": 5.1901099178239924,
+      "tokens_seen": 57999360
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009923370110330994,
+      "loss": 2.5255,
+      "theoretical_loss": 5.189239621684421,
+      "tokens_seen": 58064896
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009923169508525577,
+      "loss": 2.7432,
+      "theoretical_loss": 5.188370581950215,
+      "tokens_seen": 58130432
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009922968906720161,
+      "loss": 2.7216,
+      "theoretical_loss": 5.18750279539416,
+      "tokens_seen": 58195968
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009922768304914745,
+      "loss": 2.493,
+      "theoretical_loss": 5.1866362588009505,
+      "tokens_seen": 58261504
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009922567703109328,
+      "loss": 2.6899,
+      "theoretical_loss": 5.185770968967139,
+      "tokens_seen": 58327040
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009922367101303912,
+      "loss": 2.6763,
+      "theoretical_loss": 5.184906922701076,
+      "tokens_seen": 58392576
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009922166499498496,
+      "loss": 2.5682,
+      "theoretical_loss": 5.184044116822849,
+      "tokens_seen": 58458112
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000992196589769308,
+      "loss": 2.5105,
+      "theoretical_loss": 5.183182548164234,
+      "tokens_seen": 58523648
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009921765295887663,
+      "loss": 2.5142,
+      "theoretical_loss": 5.182322213568634,
+      "tokens_seen": 58589184
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009921564694082246,
+      "loss": 2.7216,
+      "theoretical_loss": 5.181463109891023,
+      "tokens_seen": 58654720
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000992136409227683,
+      "loss": 2.8247,
+      "theoretical_loss": 5.1806052339978965,
+      "tokens_seen": 58720256
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009921163490471414,
+      "loss": 2.5839,
+      "theoretical_loss": 5.179748582767209,
+      "tokens_seen": 58785792
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009920962888666,
+      "loss": 2.6148,
+      "theoretical_loss": 5.178893153088326,
+      "tokens_seen": 58851328
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000992076228686058,
+      "loss": 2.5025,
+      "theoretical_loss": 5.178038941861966,
+      "tokens_seen": 58916864
+    },
+    {
+      "epoch": 0.02,
+      "objective/train/advantage_avg": 0.4918819069862366,
+      "objective/train/docs_used": 29071,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 2.2515406608581543,
+      "objective/train/original_loss": 2.2515406608581543,
+      "objective/train/theoretical_loss": 5.1771859460001455,
+      "objective/train/tokens_used": 79442400,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24229572713375092,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504199266433716,
+      "objective/train/weighted_lm_loss": 2.365377187728882,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 5.1771859460001455,
+      "tokens_seen": 58982400
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009920561685055167,
+      "loss": 2.4743,
+      "theoretical_loss": 5.1771859460001455,
+      "tokens_seen": 58982400
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000992036108324975,
+      "loss": 2.5575,
+      "theoretical_loss": 5.176334162426128,
+      "tokens_seen": 59047936
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009920160481444334,
+      "loss": 2.6685,
+      "theoretical_loss": 5.175483588074371,
+      "tokens_seen": 59113472
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009919959879638918,
+      "loss": 2.5362,
+      "theoretical_loss": 5.17463421989047,
+      "tokens_seen": 59179008
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009919759277833501,
+      "loss": 2.6476,
+      "theoretical_loss": 5.1737860548311065,
+      "tokens_seen": 59244544
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009919558676028085,
+      "loss": 2.5998,
+      "theoretical_loss": 5.172939089863998,
+      "tokens_seen": 59310080
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009919358074222668,
+      "loss": 2.6016,
+      "theoretical_loss": 5.172093321967843,
+      "tokens_seen": 59375616
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009919157472417252,
+      "loss": 2.432,
+      "theoretical_loss": 5.171248748132273,
+      "tokens_seen": 59441152
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009918956870611836,
+      "loss": 2.7428,
+      "theoretical_loss": 5.170405365357794,
+      "tokens_seen": 59506688
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000991875626880642,
+      "loss": 2.5627,
+      "theoretical_loss": 5.169563170655742,
+      "tokens_seen": 59572224
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009918555667001003,
+      "loss": 2.5943,
+      "theoretical_loss": 5.16872216104823,
+      "tokens_seen": 59637760
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009918355065195587,
+      "loss": 2.4533,
+      "theoretical_loss": 5.167882333568095,
+      "tokens_seen": 59703296
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000991815446339017,
+      "loss": 2.6007,
+      "theoretical_loss": 5.167043685258852,
+      "tokens_seen": 59768832
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009917953861584754,
+      "loss": 2.5313,
+      "theoretical_loss": 5.16620621317464,
+      "tokens_seen": 59834368
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009917753259779337,
+      "loss": 2.5483,
+      "theoretical_loss": 5.1653699143801735,
+      "tokens_seen": 59899904
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000991755265797392,
+      "loss": 2.6718,
+      "theoretical_loss": 5.164534785950695,
+      "tokens_seen": 59965440
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009917352056168507,
+      "loss": 2.4902,
+      "theoretical_loss": 5.163700824971922,
+      "tokens_seen": 60030976
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009917151454363088,
+      "loss": 2.7512,
+      "theoretical_loss": 5.162868028540002,
+      "tokens_seen": 60096512
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009916950852557674,
+      "loss": 2.6651,
+      "theoretical_loss": 5.1620363937614595,
+      "tokens_seen": 60162048
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009916750250752258,
+      "loss": 2.6301,
+      "theoretical_loss": 5.161205917753154,
+      "tokens_seen": 60227584
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009916549648946841,
+      "loss": 2.5769,
+      "theoretical_loss": 5.160376597642223,
+      "tokens_seen": 60293120
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009916349047141425,
+      "loss": 2.5911,
+      "theoretical_loss": 5.159548430566044,
+      "tokens_seen": 60358656
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009916148445336009,
+      "loss": 2.6131,
+      "theoretical_loss": 5.158721413672179,
+      "tokens_seen": 60424192
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009915947843530592,
+      "loss": 2.5863,
+      "theoretical_loss": 5.1578955441183325,
+      "tokens_seen": 60489728
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009915747241725176,
+      "loss": 2.5351,
+      "theoretical_loss": 5.157070819072301,
+      "tokens_seen": 60555264
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000991554663991976,
+      "loss": 2.7029,
+      "theoretical_loss": 5.156247235711929,
+      "tokens_seen": 60620800
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009915346038114343,
+      "loss": 2.6004,
+      "theoretical_loss": 5.155424791225061,
+      "tokens_seen": 60686336
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009915145436308927,
+      "loss": 2.6118,
+      "theoretical_loss": 5.1546034828094935,
+      "tokens_seen": 60751872
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000991494483450351,
+      "loss": 2.5272,
+      "theoretical_loss": 5.153783307672935,
+      "tokens_seen": 60817408
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009914744232698094,
+      "loss": 2.6638,
+      "theoretical_loss": 5.152964263032954,
+      "tokens_seen": 60882944
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009914543630892678,
+      "loss": 2.5694,
+      "theoretical_loss": 5.152146346116936,
+      "tokens_seen": 60948480
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009914343029087261,
+      "loss": 2.5069,
+      "theoretical_loss": 5.151329554162039,
+      "tokens_seen": 61014016
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009914142427281847,
+      "loss": 2.6197,
+      "theoretical_loss": 5.150513884415149,
+      "tokens_seen": 61079552
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009913941825476428,
+      "loss": 2.6271,
+      "theoretical_loss": 5.1496993341328325,
+      "tokens_seen": 61145088
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009913741223671014,
+      "loss": 2.5083,
+      "theoretical_loss": 5.148885900581297,
+      "tokens_seen": 61210624
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009913540621865596,
+      "loss": 2.6221,
+      "theoretical_loss": 5.148073581036338,
+      "tokens_seen": 61276160
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009913340020060181,
+      "loss": 2.5627,
+      "theoretical_loss": 5.14726237278331,
+      "tokens_seen": 61341696
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009913139418254765,
+      "loss": 2.5517,
+      "theoretical_loss": 5.146452273117065,
+      "tokens_seen": 61407232
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009912938816449349,
+      "loss": 2.7035,
+      "theoretical_loss": 5.1456432793419244,
+      "tokens_seen": 61472768
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009912738214643932,
+      "loss": 2.5791,
+      "theoretical_loss": 5.144835388771626,
+      "tokens_seen": 61538304
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009912537612838516,
+      "loss": 2.6198,
+      "theoretical_loss": 5.144028598729285,
+      "tokens_seen": 61603840
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00099123370110331,
+      "loss": 2.6126,
+      "theoretical_loss": 5.1432229065473525,
+      "tokens_seen": 61669376
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009912136409227683,
+      "loss": 2.48,
+      "theoretical_loss": 5.142418309567569,
+      "tokens_seen": 61734912
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009911935807422267,
+      "loss": 2.5514,
+      "theoretical_loss": 5.141614805140927,
+      "tokens_seen": 61800448
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000991173520561685,
+      "loss": 2.6091,
+      "theoretical_loss": 5.140812390627624,
+      "tokens_seen": 61865984
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009911534603811434,
+      "loss": 2.5793,
+      "theoretical_loss": 5.1400110633970275,
+      "tokens_seen": 61931520
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009911334002006018,
+      "loss": 2.7038,
+      "theoretical_loss": 5.139210820827625,
+      "tokens_seen": 61997056
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009911133400200601,
+      "loss": 2.5758,
+      "theoretical_loss": 5.138411660306989,
+      "tokens_seen": 62062592
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009910932798395185,
+      "loss": 2.4817,
+      "theoretical_loss": 5.137613579231737,
+      "tokens_seen": 62128128
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009910732196589769,
+      "loss": 2.6186,
+      "theoretical_loss": 5.136816575007483,
+      "tokens_seen": 62193664
+    },
+    {
+      "epoch": 0.02,
+      "objective/train/advantage_avg": 0.4942206144332886,
+      "objective/train/docs_used": 30276,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 2.677966356277466,
+      "objective/train/original_loss": 2.677966594696045,
+      "objective/train/theoretical_loss": 5.136020645048806,
+      "objective/train/tokens_used": 82719200,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24524354934692383,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506689548492432,
+      "objective/train/weighted_lm_loss": 2.8136417865753174,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 5.136020645048806,
+      "tokens_seen": 62259200
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009910531594784354,
+      "loss": 2.6662,
+      "theoretical_loss": 5.136020645048806,
+      "tokens_seen": 62259200
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009910330992978936,
+      "loss": 2.4777,
+      "theoretical_loss": 5.135225786779204,
+      "tokens_seen": 62324736
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009910130391173522,
+      "loss": 2.6514,
+      "theoretical_loss": 5.134431997631053,
+      "tokens_seen": 62390272
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009909929789368103,
+      "loss": 2.4079,
+      "theoretical_loss": 5.1336392750455735,
+      "tokens_seen": 62455808
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009909729187562689,
+      "loss": 2.5303,
+      "theoretical_loss": 5.132847616472784,
+      "tokens_seen": 62521344
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009909528585757272,
+      "loss": 2.5487,
+      "theoretical_loss": 5.132057019371466,
+      "tokens_seen": 62586880
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009909327983951856,
+      "loss": 2.4935,
+      "theoretical_loss": 5.1312674812091235,
+      "tokens_seen": 62652416
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000990912738214644,
+      "loss": 2.4875,
+      "theoretical_loss": 5.130478999461939,
+      "tokens_seen": 62717952
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009908926780341023,
+      "loss": 2.5383,
+      "theoretical_loss": 5.129691571614747,
+      "tokens_seen": 62783488
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009908726178535607,
+      "loss": 2.4761,
+      "theoretical_loss": 5.128905195160981,
+      "tokens_seen": 62849024
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000990852557673019,
+      "loss": 2.589,
+      "theoretical_loss": 5.128119867602646,
+      "tokens_seen": 62914560
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009908324974924774,
+      "loss": 2.601,
+      "theoretical_loss": 5.127335586450272,
+      "tokens_seen": 62980096
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009908124373119358,
+      "loss": 2.634,
+      "theoretical_loss": 5.126552349222884,
+      "tokens_seen": 63045632
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009907923771313941,
+      "loss": 2.5431,
+      "theoretical_loss": 5.12577015344796,
+      "tokens_seen": 63111168
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009907723169508525,
+      "loss": 2.5072,
+      "theoretical_loss": 5.124988996661393,
+      "tokens_seen": 63176704
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000990752256770311,
+      "loss": 2.5105,
+      "theoretical_loss": 5.124208876407453,
+      "tokens_seen": 63242240
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009907321965897692,
+      "loss": 2.4315,
+      "theoretical_loss": 5.1234297902387524,
+      "tokens_seen": 63307776
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009907121364092278,
+      "loss": 2.4456,
+      "theoretical_loss": 5.12265173571621,
+      "tokens_seen": 63373312
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009906920762286862,
+      "loss": 2.5354,
+      "theoretical_loss": 5.121874710409012,
+      "tokens_seen": 63438848
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009906720160481445,
+      "loss": 2.526,
+      "theoretical_loss": 5.121098711894572,
+      "tokens_seen": 63504384
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000990651955867603,
+      "loss": 2.4092,
+      "theoretical_loss": 5.120323737758502,
+      "tokens_seen": 63569920
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009906318956870613,
+      "loss": 2.443,
+      "theoretical_loss": 5.119549785594572,
+      "tokens_seen": 63635456
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009906118355065196,
+      "loss": 2.5223,
+      "theoretical_loss": 5.118776853004677,
+      "tokens_seen": 63700992
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000990591775325978,
+      "loss": 2.5233,
+      "theoretical_loss": 5.118004937598793,
+      "tokens_seen": 63766528
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009905717151454363,
+      "loss": 2.4831,
+      "theoretical_loss": 5.1172340369949545,
+      "tokens_seen": 63832064
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009905516549648947,
+      "loss": 2.5658,
+      "theoretical_loss": 5.116464148819206,
+      "tokens_seen": 63897600
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000990531594784353,
+      "loss": 2.5614,
+      "theoretical_loss": 5.115695270705579,
+      "tokens_seen": 63963136
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009905115346038114,
+      "loss": 2.3751,
+      "theoretical_loss": 5.114927400296046,
+      "tokens_seen": 64028672
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009904914744232698,
+      "loss": 2.4838,
+      "theoretical_loss": 5.114160535240492,
+      "tokens_seen": 64094208
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009904714142427282,
+      "loss": 2.5281,
+      "theoretical_loss": 5.113394673196682,
+      "tokens_seen": 64159744
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009904513540621865,
+      "loss": 2.4431,
+      "theoretical_loss": 5.112629811830217,
+      "tokens_seen": 64225280
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000990431293881645,
+      "loss": 2.4407,
+      "theoretical_loss": 5.111865948814511,
+      "tokens_seen": 64290816
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009904112337011032,
+      "loss": 2.4222,
+      "theoretical_loss": 5.111103081830752,
+      "tokens_seen": 64356352
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009903911735205618,
+      "loss": 2.545,
+      "theoretical_loss": 5.1103412085678634,
+      "tokens_seen": 64421888
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00099037111334002,
+      "loss": 2.5683,
+      "theoretical_loss": 5.10958032672248,
+      "tokens_seen": 64487424
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009903510531594785,
+      "loss": 2.572,
+      "theoretical_loss": 5.108820433998908,
+      "tokens_seen": 64552960
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000990330992978937,
+      "loss": 2.381,
+      "theoretical_loss": 5.108061528109092,
+      "tokens_seen": 64618496
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009903109327983953,
+      "loss": 2.4737,
+      "theoretical_loss": 5.107303606772582,
+      "tokens_seen": 64684032
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009902908726178536,
+      "loss": 2.4322,
+      "theoretical_loss": 5.106546667716508,
+      "tokens_seen": 64749568
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000990270812437312,
+      "loss": 2.592,
+      "theoretical_loss": 5.1057907086755305,
+      "tokens_seen": 64815104
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009902507522567704,
+      "loss": 2.5258,
+      "theoretical_loss": 5.105035727391828,
+      "tokens_seen": 64880640
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009902306920762287,
+      "loss": 2.6152,
+      "theoretical_loss": 5.104281721615046,
+      "tokens_seen": 64946176
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000990210631895687,
+      "loss": 2.5266,
+      "theoretical_loss": 5.103528689102281,
+      "tokens_seen": 65011712
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009901905717151454,
+      "loss": 2.5043,
+      "theoretical_loss": 5.102776627618036,
+      "tokens_seen": 65077248
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009901705115346038,
+      "loss": 2.5268,
+      "theoretical_loss": 5.102025534934196,
+      "tokens_seen": 65142784
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009901504513540622,
+      "loss": 2.5599,
+      "theoretical_loss": 5.10127540882999,
+      "tokens_seen": 65208320
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009901303911735205,
+      "loss": 2.403,
+      "theoretical_loss": 5.100526247091967,
+      "tokens_seen": 65273856
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000990110330992979,
+      "loss": 2.4966,
+      "theoretical_loss": 5.099778047513957,
+      "tokens_seen": 65339392
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009900902708124373,
+      "loss": 2.5187,
+      "theoretical_loss": 5.099030807897048,
+      "tokens_seen": 65404928
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009900702106318958,
+      "loss": 2.5642,
+      "theoretical_loss": 5.098284526049547,
+      "tokens_seen": 65470464
+    },
+    {
+      "epoch": 0.02,
+      "objective/train/advantage_avg": 0.49179795384407043,
+      "objective/train/docs_used": 31526,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 2.5606729984283447,
+      "objective/train/original_loss": 2.5606727600097656,
+      "objective/train/theoretical_loss": 5.097539199786951,
+      "objective/train/tokens_used": 85996000,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24245329201221466,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504124164581299,
+      "objective/train/weighted_lm_loss": 2.6897695064544678,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9753099083900452,
+      "theoretical_loss": 5.097539199786951,
+      "tokens_seen": 65536000
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000990050150451354,
+      "loss": 2.5972,
+      "theoretical_loss": 5.097539199786951,
+      "tokens_seen": 65536000
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009900300902708126,
+      "loss": 2.5759,
+      "theoretical_loss": 5.096794826931921,
+      "tokens_seen": 65601536
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009900100300902707,
+      "loss": 2.5983,
+      "theoretical_loss": 5.096051405314247,
+      "tokens_seen": 65667072
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009899899699097293,
+      "loss": 2.6292,
+      "theoretical_loss": 5.095308932770816,
+      "tokens_seen": 65732608
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009899699097291876,
+      "loss": 2.4965,
+      "theoretical_loss": 5.094567407145588,
+      "tokens_seen": 65798144
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000989949849548646,
+      "loss": 2.4289,
+      "theoretical_loss": 5.093826826289559,
+      "tokens_seen": 65863680
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009899297893681044,
+      "loss": 2.5533,
+      "theoretical_loss": 5.093087188060739,
+      "tokens_seen": 65929216
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009899097291875627,
+      "loss": 2.4959,
+      "theoretical_loss": 5.09234849032411,
+      "tokens_seen": 65994752
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000989889669007021,
+      "loss": 2.4689,
+      "theoretical_loss": 5.09161073095161,
+      "tokens_seen": 66060288
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009898696088264795,
+      "loss": 2.4506,
+      "theoretical_loss": 5.090873907822093,
+      "tokens_seen": 66125824
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009898495486459378,
+      "loss": 2.408,
+      "theoretical_loss": 5.090138018821309,
+      "tokens_seen": 66191360
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009898294884653962,
+      "loss": 2.4846,
+      "theoretical_loss": 5.089403061841864,
+      "tokens_seen": 66256896
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009898094282848545,
+      "loss": 2.502,
+      "theoretical_loss": 5.0886690347832015,
+      "tokens_seen": 66322432
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000989789368104313,
+      "loss": 2.4468,
+      "theoretical_loss": 5.087935935551566,
+      "tokens_seen": 66387968
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009897693079237713,
+      "loss": 2.4499,
+      "theoretical_loss": 5.0872037620599775,
+      "tokens_seen": 66453504
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009897492477432296,
+      "loss": 2.4336,
+      "theoretical_loss": 5.0864725122282035,
+      "tokens_seen": 66519040
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000989729187562688,
+      "loss": 2.5683,
+      "theoretical_loss": 5.0857421839827275,
+      "tokens_seen": 66584576
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009897091273821466,
+      "loss": 2.4402,
+      "theoretical_loss": 5.085012775256727,
+      "tokens_seen": 66650112
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009896890672016047,
+      "loss": 2.441,
+      "theoretical_loss": 5.084284283990035,
+      "tokens_seen": 66715648
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009896690070210633,
+      "loss": 2.4647,
+      "theoretical_loss": 5.083556708129126,
+      "tokens_seen": 66781184
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009896489468405214,
+      "loss": 2.3144,
+      "theoretical_loss": 5.082830045627072,
+      "tokens_seen": 66846720
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00098962888665998,
+      "loss": 2.4836,
+      "theoretical_loss": 5.08210429444353,
+      "tokens_seen": 66912256
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009896088264794384,
+      "loss": 2.4112,
+      "theoretical_loss": 5.081379452544702,
+      "tokens_seen": 66977792
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009895887662988967,
+      "loss": 2.5422,
+      "theoretical_loss": 5.080655517903319,
+      "tokens_seen": 67043328
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000989568706118355,
+      "loss": 2.4475,
+      "theoretical_loss": 5.079932488498602,
+      "tokens_seen": 67108864
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009895486459378135,
+      "loss": 2.6382,
+      "theoretical_loss": 5.079210362316245,
+      "tokens_seen": 67174400
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009895285857572718,
+      "loss": 2.4146,
+      "theoretical_loss": 5.078489137348383,
+      "tokens_seen": 67239936
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009895085255767302,
+      "loss": 2.4907,
+      "theoretical_loss": 5.077768811593563,
+      "tokens_seen": 67305472
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009894884653961886,
+      "loss": 2.3637,
+      "theoretical_loss": 5.077049383056725,
+      "tokens_seen": 67371008
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000989468405215647,
+      "loss": 2.3986,
+      "theoretical_loss": 5.076330849749166,
+      "tokens_seen": 67436544
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009894483450351055,
+      "loss": 2.5128,
+      "theoretical_loss": 5.0756132096885205,
+      "tokens_seen": 67502080
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009894282848545636,
+      "loss": 2.6255,
+      "theoretical_loss": 5.0748964608987315,
+      "tokens_seen": 67567616
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009894082246740222,
+      "loss": 2.5026,
+      "theoretical_loss": 5.074180601410026,
+      "tokens_seen": 67633152
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009893881644934804,
+      "loss": 2.3306,
+      "theoretical_loss": 5.073465629258886,
+      "tokens_seen": 67698688
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000989368104312939,
+      "loss": 2.4388,
+      "theoretical_loss": 5.072751542488024,
+      "tokens_seen": 67764224
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009893480441323973,
+      "loss": 2.4484,
+      "theoretical_loss": 5.072038339146357,
+      "tokens_seen": 67829760
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009893279839518557,
+      "loss": 2.3594,
+      "theoretical_loss": 5.0713260172889845,
+      "tokens_seen": 67895296
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000989307923771314,
+      "loss": 2.4074,
+      "theoretical_loss": 5.070614574977157,
+      "tokens_seen": 67960832
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009892878635907724,
+      "loss": 2.4488,
+      "theoretical_loss": 5.069904010278254,
+      "tokens_seen": 68026368
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009892678034102308,
+      "loss": 2.3926,
+      "theoretical_loss": 5.069194321265758,
+      "tokens_seen": 68091904
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009892477432296891,
+      "loss": 2.2627,
+      "theoretical_loss": 5.068485506019231,
+      "tokens_seen": 68157440
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009892276830491475,
+      "loss": 2.3413,
+      "theoretical_loss": 5.0677775626242845,
+      "tokens_seen": 68222976
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009892076228686058,
+      "loss": 2.4958,
+      "theoretical_loss": 5.067070489172562,
+      "tokens_seen": 68288512
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009891875626880642,
+      "loss": 2.3923,
+      "theoretical_loss": 5.066364283761709,
+      "tokens_seen": 68354048
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009891675025075226,
+      "loss": 2.4773,
+      "theoretical_loss": 5.06565894449535,
+      "tokens_seen": 68419584
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000989147442326981,
+      "loss": 2.3944,
+      "theoretical_loss": 5.064954469483064,
+      "tokens_seen": 68485120
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009891273821464393,
+      "loss": 2.4673,
+      "theoretical_loss": 5.064250856840359,
+      "tokens_seen": 68550656
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009891073219658977,
+      "loss": 2.5153,
+      "theoretical_loss": 5.06354810468865,
+      "tokens_seen": 68616192
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009890872617853562,
+      "loss": 2.367,
+      "theoretical_loss": 5.06284621115523,
+      "tokens_seen": 68681728
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009890672016048144,
+      "loss": 2.4625,
+      "theoretical_loss": 5.062145174373255,
+      "tokens_seen": 68747264
+    },
+    {
+      "epoch": 0.02,
+      "objective/train/advantage_avg": 0.49227413535118103,
+      "objective/train/docs_used": 32629,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 2.2441399097442627,
+      "objective/train/original_loss": 2.2441399097442627,
+      "objective/train/theoretical_loss": 5.061444992481711,
+      "objective/train/tokens_used": 89272800,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24276849627494812,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504614114761353,
+      "objective/train/weighted_lm_loss": 2.3572661876678467,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 5.061444992481711,
+      "tokens_seen": 68812800
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000989047141424273,
+      "loss": 2.4288,
+      "theoretical_loss": 5.061444992481711,
+      "tokens_seen": 68812800
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000989027081243731,
+      "loss": 2.5382,
+      "theoretical_loss": 5.060745663625397,
+      "tokens_seen": 68878336
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009890070210631897,
+      "loss": 2.5068,
+      "theoretical_loss": 5.060047185954893,
+      "tokens_seen": 68943872
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000988986960882648,
+      "loss": 2.3632,
+      "theoretical_loss": 5.059349557626547,
+      "tokens_seen": 69009408
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009889669007021064,
+      "loss": 2.346,
+      "theoretical_loss": 5.058652776802443,
+      "tokens_seen": 69074944
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009889468405215648,
+      "loss": 2.4677,
+      "theoretical_loss": 5.057956841650385,
+      "tokens_seen": 69140480
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009889267803410231,
+      "loss": 2.471,
+      "theoretical_loss": 5.057261750343864,
+      "tokens_seen": 69206016
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009889067201604815,
+      "loss": 2.3835,
+      "theoretical_loss": 5.0565675010620454,
+      "tokens_seen": 69271552
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009888866599799399,
+      "loss": 2.319,
+      "theoretical_loss": 5.05587409198974,
+      "tokens_seen": 69337088
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009888665997993982,
+      "loss": 2.5456,
+      "theoretical_loss": 5.055181521317383,
+      "tokens_seen": 69402624
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009888465396188566,
+      "loss": 2.4407,
+      "theoretical_loss": 5.0544897872410095,
+      "tokens_seen": 69468160
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000988826479438315,
+      "loss": 2.3936,
+      "theoretical_loss": 5.053798887962236,
+      "tokens_seen": 69533696
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009888064192577733,
+      "loss": 2.3531,
+      "theoretical_loss": 5.053108821688232,
+      "tokens_seen": 69599232
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009887863590772317,
+      "loss": 2.4097,
+      "theoretical_loss": 5.052419586631703,
+      "tokens_seen": 69664768
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00098876629889669,
+      "loss": 2.4689,
+      "theoretical_loss": 5.051731181010866,
+      "tokens_seen": 69730304
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009887462387161484,
+      "loss": 2.3541,
+      "theoretical_loss": 5.051043603049426,
+      "tokens_seen": 69795840
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000988726178535607,
+      "loss": 2.3945,
+      "theoretical_loss": 5.0503568509765575,
+      "tokens_seen": 69861376
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009887061183550651,
+      "loss": 2.3833,
+      "theoretical_loss": 5.049670923026879,
+      "tokens_seen": 69926912
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009886860581745237,
+      "loss": 2.3977,
+      "theoretical_loss": 5.048985817440432,
+      "tokens_seen": 69992448
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009886659979939818,
+      "loss": 2.4533,
+      "theoretical_loss": 5.048301532462662,
+      "tokens_seen": 70057984
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009886459378134404,
+      "loss": 2.5417,
+      "theoretical_loss": 5.047618066344392,
+      "tokens_seen": 70123520
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009886258776328988,
+      "loss": 2.3964,
+      "theoretical_loss": 5.046935417341806,
+      "tokens_seen": 70189056
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009886058174523571,
+      "loss": 2.4526,
+      "theoretical_loss": 5.046253583716425,
+      "tokens_seen": 70254592
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009885857572718155,
+      "loss": 2.425,
+      "theoretical_loss": 5.045572563735084,
+      "tokens_seen": 70320128
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009885656970912739,
+      "loss": 2.3847,
+      "theoretical_loss": 5.0448923556699174,
+      "tokens_seen": 70385664
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009885456369107322,
+      "loss": 2.458,
+      "theoretical_loss": 5.044212957798328,
+      "tokens_seen": 70451200
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009885255767301906,
+      "loss": 2.3902,
+      "theoretical_loss": 5.043534368402973,
+      "tokens_seen": 70516736
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000988505516549649,
+      "loss": 2.2822,
+      "theoretical_loss": 5.042856585771746,
+      "tokens_seen": 70582272
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009884854563691073,
+      "loss": 2.3476,
+      "theoretical_loss": 5.0421796081977455,
+      "tokens_seen": 70647808
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009884653961885657,
+      "loss": 2.4328,
+      "theoretical_loss": 5.041503433979264,
+      "tokens_seen": 70713344
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000988445336008024,
+      "loss": 2.5019,
+      "theoretical_loss": 5.040828061419762,
+      "tokens_seen": 70778880
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009884252758274824,
+      "loss": 2.4293,
+      "theoretical_loss": 5.040153488827851,
+      "tokens_seen": 70844416
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009884052156469408,
+      "loss": 2.5365,
+      "theoretical_loss": 5.039479714517272,
+      "tokens_seen": 70909952
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009883851554663991,
+      "loss": 2.4693,
+      "theoretical_loss": 5.038806736806871,
+      "tokens_seen": 70975488
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009883650952858577,
+      "loss": 2.3275,
+      "theoretical_loss": 5.038134554020587,
+      "tokens_seen": 71041024
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009883450351053159,
+      "loss": 2.3314,
+      "theoretical_loss": 5.037463164487425,
+      "tokens_seen": 71106560
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009883249749247744,
+      "loss": 2.3667,
+      "theoretical_loss": 5.036792566541441,
+      "tokens_seen": 71172096
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009883049147442326,
+      "loss": 2.3339,
+      "theoretical_loss": 5.036122758521715,
+      "tokens_seen": 71237632
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009882848545636912,
+      "loss": 2.5154,
+      "theoretical_loss": 5.03545373877234,
+      "tokens_seen": 71303168
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009882647943831495,
+      "loss": 2.4845,
+      "theoretical_loss": 5.034785505642397,
+      "tokens_seen": 71368704
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009882447342026079,
+      "loss": 2.4743,
+      "theoretical_loss": 5.034118057485935,
+      "tokens_seen": 71434240
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009882246740220662,
+      "loss": 2.5096,
+      "theoretical_loss": 5.033451392661957,
+      "tokens_seen": 71499776
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009882046138415246,
+      "loss": 2.3763,
+      "theoretical_loss": 5.032785509534391,
+      "tokens_seen": 71565312
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000988184553660983,
+      "loss": 2.4245,
+      "theoretical_loss": 5.032120406472082,
+      "tokens_seen": 71630848
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009881644934804413,
+      "loss": 2.4523,
+      "theoretical_loss": 5.031456081848761,
+      "tokens_seen": 71696384
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009881444332998997,
+      "loss": 2.2822,
+      "theoretical_loss": 5.030792534043039,
+      "tokens_seen": 71761920
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000988124373119358,
+      "loss": 2.4078,
+      "theoretical_loss": 5.030129761438376,
+      "tokens_seen": 71827456
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009881043129388164,
+      "loss": 2.4901,
+      "theoretical_loss": 5.029467762423069,
+      "tokens_seen": 71892992
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009880842527582748,
+      "loss": 2.4351,
+      "theoretical_loss": 5.028806535390228,
+      "tokens_seen": 71958528
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009880641925777331,
+      "loss": 2.4051,
+      "theoretical_loss": 5.028146078737766,
+      "tokens_seen": 72024064
+    },
+    {
+      "epoch": 0.02,
+      "objective/train/advantage_avg": 0.49358925223350525,
+      "objective/train/docs_used": 33987,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 2.276097536087036,
+      "objective/train/original_loss": 2.2760977745056152,
+      "objective/train/theoretical_loss": 5.0274863908683685,
+      "objective/train/tokens_used": 92549600,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24399147927761078,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050599455833435,
+      "objective/train/weighted_lm_loss": 2.3911280632019043,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 5.0274863908683685,
+      "tokens_seen": 72089600
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009880441323971915,
+      "loss": 2.258,
+      "theoretical_loss": 5.0274863908683685,
+      "tokens_seen": 72089600
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009880240722166499,
+      "loss": 2.4077,
+      "theoretical_loss": 5.026827470189487,
+      "tokens_seen": 72155136
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009880040120361084,
+      "loss": 2.4191,
+      "theoretical_loss": 5.026169315113311,
+      "tokens_seen": 72220672
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009879839518555668,
+      "loss": 2.3911,
+      "theoretical_loss": 5.025511924056753,
+      "tokens_seen": 72286208
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009879638916750252,
+      "loss": 2.5566,
+      "theoretical_loss": 5.024855295441432,
+      "tokens_seen": 72351744
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009879438314944835,
+      "loss": 2.3835,
+      "theoretical_loss": 5.024199427693654,
+      "tokens_seen": 72417280
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000987923771313942,
+      "loss": 2.4704,
+      "theoretical_loss": 5.023544319244391,
+      "tokens_seen": 72482816
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009879037111334003,
+      "loss": 2.3699,
+      "theoretical_loss": 5.0228899685292685,
+      "tokens_seen": 72548352
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009878836509528586,
+      "loss": 2.1989,
+      "theoretical_loss": 5.022236373988544,
+      "tokens_seen": 72613888
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000987863590772317,
+      "loss": 2.4072,
+      "theoretical_loss": 5.021583534067085,
+      "tokens_seen": 72679424
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009878435305917753,
+      "loss": 2.3943,
+      "theoretical_loss": 5.020931447214364,
+      "tokens_seen": 72744960
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009878234704112337,
+      "loss": 2.4555,
+      "theoretical_loss": 5.020280111884426,
+      "tokens_seen": 72810496
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000987803410230692,
+      "loss": 2.3937,
+      "theoretical_loss": 5.01962952653588,
+      "tokens_seen": 72876032
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009877833500501504,
+      "loss": 2.2974,
+      "theoretical_loss": 5.018979689631879,
+      "tokens_seen": 72941568
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009877632898696088,
+      "loss": 2.2553,
+      "theoretical_loss": 5.018330599640101,
+      "tokens_seen": 73007104
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009877432296890674,
+      "loss": 2.3621,
+      "theoretical_loss": 5.017682255032735,
+      "tokens_seen": 73072640
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009877231695085255,
+      "loss": 2.5088,
+      "theoretical_loss": 5.017034654286462,
+      "tokens_seen": 73138176
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000987703109327984,
+      "loss": 2.4084,
+      "theoretical_loss": 5.0163877958824346,
+      "tokens_seen": 73203712
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009876830491474422,
+      "loss": 2.4194,
+      "theoretical_loss": 5.0157416783062665,
+      "tokens_seen": 73269248
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009876629889669008,
+      "loss": 2.4157,
+      "theoretical_loss": 5.015096300048009,
+      "tokens_seen": 73334784
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009876429287863592,
+      "loss": 2.4174,
+      "theoretical_loss": 5.0144516596021385,
+      "tokens_seen": 73400320
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009876228686058175,
+      "loss": 2.205,
+      "theoretical_loss": 5.013807755467537,
+      "tokens_seen": 73465856
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000987602808425276,
+      "loss": 2.2951,
+      "theoretical_loss": 5.013164586147477,
+      "tokens_seen": 73531392
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009875827482447343,
+      "loss": 2.3731,
+      "theoretical_loss": 5.012522150149602,
+      "tokens_seen": 73596928
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009875626880641926,
+      "loss": 2.4183,
+      "theoretical_loss": 5.011880445985916,
+      "tokens_seen": 73662464
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000987542627883651,
+      "loss": 2.3,
+      "theoretical_loss": 5.011239472172758,
+      "tokens_seen": 73728000
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009875225677031094,
+      "loss": 2.5542,
+      "theoretical_loss": 5.010599227230792,
+      "tokens_seen": 73793536
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009875025075225677,
+      "loss": 2.5132,
+      "theoretical_loss": 5.009959709684992,
+      "tokens_seen": 73859072
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000987482447342026,
+      "loss": 2.409,
+      "theoretical_loss": 5.009320918064615,
+      "tokens_seen": 73924608
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009874623871614844,
+      "loss": 2.4014,
+      "theoretical_loss": 5.008682850903201,
+      "tokens_seen": 73990144
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009874423269809428,
+      "loss": 2.4533,
+      "theoretical_loss": 5.00804550673854,
+      "tokens_seen": 74055680
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009874222668004012,
+      "loss": 2.309,
+      "theoretical_loss": 5.007408884112671,
+      "tokens_seen": 74121216
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009874022066198595,
+      "loss": 2.5961,
+      "theoretical_loss": 5.006772981571855,
+      "tokens_seen": 74186752
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009873821464393181,
+      "loss": 2.4414,
+      "theoretical_loss": 5.006137797666559,
+      "tokens_seen": 74252288
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009873620862587763,
+      "loss": 2.2648,
+      "theoretical_loss": 5.005503330951454,
+      "tokens_seen": 74317824
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009873420260782348,
+      "loss": 2.4416,
+      "theoretical_loss": 5.004869579985381,
+      "tokens_seen": 74383360
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000987321965897693,
+      "loss": 2.3453,
+      "theoretical_loss": 5.004236543331345,
+      "tokens_seen": 74448896
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009873019057171516,
+      "loss": 2.3843,
+      "theoretical_loss": 5.0036042195565,
+      "tokens_seen": 74514432
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00098728184553661,
+      "loss": 2.3351,
+      "theoretical_loss": 5.002972607232129,
+      "tokens_seen": 74579968
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009872617853560683,
+      "loss": 2.3041,
+      "theoretical_loss": 5.002341704933631,
+      "tokens_seen": 74645504
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009872417251755266,
+      "loss": 2.4829,
+      "theoretical_loss": 5.001711511240506,
+      "tokens_seen": 74711040
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000987221664994985,
+      "loss": 2.3724,
+      "theoretical_loss": 5.0010820247363394,
+      "tokens_seen": 74776576
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009872016048144434,
+      "loss": 2.3896,
+      "theoretical_loss": 5.000453244008782,
+      "tokens_seen": 74842112
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009871815446339017,
+      "loss": 2.4098,
+      "theoretical_loss": 4.9998251676495435,
+      "tokens_seen": 74907648
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00098716148445336,
+      "loss": 2.425,
+      "theoretical_loss": 4.999197794254371,
+      "tokens_seen": 74973184
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009871414242728185,
+      "loss": 2.5423,
+      "theoretical_loss": 4.998571122423035,
+      "tokens_seen": 75038720
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009871213640922768,
+      "loss": 2.2918,
+      "theoretical_loss": 4.997945150759316,
+      "tokens_seen": 75104256
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009871013039117352,
+      "loss": 2.3774,
+      "theoretical_loss": 4.997319877870986,
+      "tokens_seen": 75169792
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009870812437311935,
+      "loss": 2.3024,
+      "theoretical_loss": 4.9966953023697975,
+      "tokens_seen": 75235328
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000987061183550652,
+      "loss": 2.4583,
+      "theoretical_loss": 4.99607142287147,
+      "tokens_seen": 75300864
+    },
+    {
+      "epoch": 0.02,
+      "objective/train/advantage_avg": 0.49300244450569153,
+      "objective/train/docs_used": 35047,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 2.2772889137268066,
+      "objective/train/original_loss": 2.2772884368896484,
+      "objective/train/theoretical_loss": 4.995448237995667,
+      "objective/train/tokens_used": 95826400,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24339433014392853,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505375862121582,
+      "objective/train/weighted_lm_loss": 2.392406463623047,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.995448237995667,
+      "tokens_seen": 75366400
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009870411233701103,
+      "loss": 2.4079,
+      "theoretical_loss": 4.995448237995667,
+      "tokens_seen": 75366400
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009870210631895688,
+      "loss": 2.3811,
+      "theoretical_loss": 4.994825746365992,
+      "tokens_seen": 75431936
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000987001003009027,
+      "loss": 2.4155,
+      "theoretical_loss": 4.994203946609964,
+      "tokens_seen": 75497472
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009869809428284856,
+      "loss": 2.2558,
+      "theoretical_loss": 4.993582837359012,
+      "tokens_seen": 75563008
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009869608826479437,
+      "loss": 2.4505,
+      "theoretical_loss": 4.992962417248453,
+      "tokens_seen": 75628544
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009869408224674023,
+      "loss": 2.4398,
+      "theoretical_loss": 4.992342684917482,
+      "tokens_seen": 75694080
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009869207622868607,
+      "loss": 2.2416,
+      "theoretical_loss": 4.991723639009154,
+      "tokens_seen": 75759616
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000986900702106319,
+      "loss": 2.2164,
+      "theoretical_loss": 4.991105278170377,
+      "tokens_seen": 75825152
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009868806419257774,
+      "loss": 2.5057,
+      "theoretical_loss": 4.990487601051887,
+      "tokens_seen": 75890688
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009868605817452357,
+      "loss": 2.4679,
+      "theoretical_loss": 4.989870606308244,
+      "tokens_seen": 75956224
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000986840521564694,
+      "loss": 2.3433,
+      "theoretical_loss": 4.989254292597813,
+      "tokens_seen": 76021760
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009868204613841525,
+      "loss": 2.5069,
+      "theoretical_loss": 4.9886386585827465,
+      "tokens_seen": 76087296
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009868004012036108,
+      "loss": 2.4961,
+      "theoretical_loss": 4.988023702928979,
+      "tokens_seen": 76152832
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009867803410230692,
+      "loss": 2.3971,
+      "theoretical_loss": 4.9874094243062075,
+      "tokens_seen": 76218368
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009867602808425276,
+      "loss": 2.4396,
+      "theoretical_loss": 4.986795821387878,
+      "tokens_seen": 76283904
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000986740220661986,
+      "loss": 2.4312,
+      "theoretical_loss": 4.986182892851171,
+      "tokens_seen": 76349440
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009867201604814443,
+      "loss": 2.3369,
+      "theoretical_loss": 4.985570637376993,
+      "tokens_seen": 76414976
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009867001003009026,
+      "loss": 2.3453,
+      "theoretical_loss": 4.984959053649959,
+      "tokens_seen": 76480512
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000986680040120361,
+      "loss": 2.3224,
+      "theoretical_loss": 4.984348140358374,
+      "tokens_seen": 76546048
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009866599799398196,
+      "loss": 2.3089,
+      "theoretical_loss": 4.983737896194229,
+      "tokens_seen": 76611584
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009866399197592777,
+      "loss": 2.3881,
+      "theoretical_loss": 4.983128319853182,
+      "tokens_seen": 76677120
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009866198595787363,
+      "loss": 2.3237,
+      "theoretical_loss": 4.982519410034545,
+      "tokens_seen": 76742656
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009865997993981945,
+      "loss": 2.2656,
+      "theoretical_loss": 4.981911165441273,
+      "tokens_seen": 76808192
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000986579739217653,
+      "loss": 2.358,
+      "theoretical_loss": 4.981303584779946,
+      "tokens_seen": 76873728
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009865596790371114,
+      "loss": 2.4148,
+      "theoretical_loss": 4.980696666760761,
+      "tokens_seen": 76939264
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009865396188565698,
+      "loss": 2.3153,
+      "theoretical_loss": 4.980090410097517,
+      "tokens_seen": 77004800
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009865195586760281,
+      "loss": 2.3638,
+      "theoretical_loss": 4.979484813507599,
+      "tokens_seen": 77070336
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009864994984954865,
+      "loss": 2.3893,
+      "theoretical_loss": 4.97887987571197,
+      "tokens_seen": 77135872
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009864794383149448,
+      "loss": 2.2725,
+      "theoretical_loss": 4.9782755954351545,
+      "tokens_seen": 77201408
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009864593781344032,
+      "loss": 2.3585,
+      "theoretical_loss": 4.977671971405225,
+      "tokens_seen": 77266944
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009864393179538616,
+      "loss": 2.3069,
+      "theoretical_loss": 4.977069002353792,
+      "tokens_seen": 77332480
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00098641925777332,
+      "loss": 2.3565,
+      "theoretical_loss": 4.976466687015991,
+      "tokens_seen": 77398016
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009863991975927785,
+      "loss": 2.375,
+      "theoretical_loss": 4.975865024130467,
+      "tokens_seen": 77463552
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009863791374122367,
+      "loss": 2.2839,
+      "theoretical_loss": 4.975264012439363,
+      "tokens_seen": 77529088
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009863590772316952,
+      "loss": 2.3887,
+      "theoretical_loss": 4.974663650688306,
+      "tokens_seen": 77594624
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009863390170511534,
+      "loss": 2.3619,
+      "theoretical_loss": 4.974063937626402,
+      "tokens_seen": 77660160
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000986318956870612,
+      "loss": 2.3161,
+      "theoretical_loss": 4.973464872006213,
+      "tokens_seen": 77725696
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009862988966900703,
+      "loss": 2.2241,
+      "theoretical_loss": 4.972866452583748,
+      "tokens_seen": 77791232
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009862788365095287,
+      "loss": 2.5561,
+      "theoretical_loss": 4.972268678118454,
+      "tokens_seen": 77856768
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000986258776328987,
+      "loss": 2.2846,
+      "theoretical_loss": 4.971671547373202,
+      "tokens_seen": 77922304
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009862387161484454,
+      "loss": 2.3102,
+      "theoretical_loss": 4.971075059114272,
+      "tokens_seen": 77987840
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009862186559679038,
+      "loss": 2.2737,
+      "theoretical_loss": 4.970479212111342,
+      "tokens_seen": 78053376
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009861985957873621,
+      "loss": 2.2125,
+      "theoretical_loss": 4.969884005137479,
+      "tokens_seen": 78118912
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009861785356068205,
+      "loss": 2.4044,
+      "theoretical_loss": 4.969289436969124,
+      "tokens_seen": 78184448
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009861584754262789,
+      "loss": 2.3445,
+      "theoretical_loss": 4.968695506386077,
+      "tokens_seen": 78249984
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009861384152457372,
+      "loss": 2.1897,
+      "theoretical_loss": 4.968102212171493,
+      "tokens_seen": 78315520
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009861183550651956,
+      "loss": 2.3495,
+      "theoretical_loss": 4.967509553111862,
+      "tokens_seen": 78381056
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000986098294884654,
+      "loss": 2.197,
+      "theoretical_loss": 4.966917527996999,
+      "tokens_seen": 78446592
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009860782347041123,
+      "loss": 2.3427,
+      "theoretical_loss": 4.9663261356200366,
+      "tokens_seen": 78512128
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009860581745235707,
+      "loss": 2.339,
+      "theoretical_loss": 4.965735374777409,
+      "tokens_seen": 78577664
+    },
+    {
+      "epoch": 0.02,
+      "objective/train/advantage_avg": 0.4910324811935425,
+      "objective/train/docs_used": 36308,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 2.3306944370269775,
+      "objective/train/original_loss": 2.3306941986083984,
+      "objective/train/theoretical_loss": 4.96514524426884,
+      "objective/train/tokens_used": 99103200,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24158093333244324,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503313541412354,
+      "objective/train/weighted_lm_loss": 2.448256015777588,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.96514524426884,
+      "tokens_seen": 78643200
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009860381143430292,
+      "loss": 2.4381,
+      "theoretical_loss": 4.96514524426884,
+      "tokens_seen": 78643200
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009860180541624874,
+      "loss": 2.333,
+      "theoretical_loss": 4.964555742897334,
+      "tokens_seen": 78708736
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000985997993981946,
+      "loss": 2.45,
+      "theoretical_loss": 4.963966869469162,
+      "tokens_seen": 78774272
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009859779338014041,
+      "loss": 2.2028,
+      "theoretical_loss": 4.963378622793849,
+      "tokens_seen": 78839808
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009859578736208627,
+      "loss": 2.2797,
+      "theoretical_loss": 4.962791001684167,
+      "tokens_seen": 78905344
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000985937813440321,
+      "loss": 2.2991,
+      "theoretical_loss": 4.962204004956117,
+      "tokens_seen": 78970880
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009859177532597794,
+      "loss": 2.232,
+      "theoretical_loss": 4.961617631428927,
+      "tokens_seen": 79036416
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009858976930792378,
+      "loss": 2.2345,
+      "theoretical_loss": 4.961031879925029,
+      "tokens_seen": 79101952
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009858776328986961,
+      "loss": 2.3468,
+      "theoretical_loss": 4.960446749270055,
+      "tokens_seen": 79167488
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009858575727181545,
+      "loss": 2.348,
+      "theoretical_loss": 4.959862238292827,
+      "tokens_seen": 79233024
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009858375125376129,
+      "loss": 2.469,
+      "theoretical_loss": 4.959278345825336,
+      "tokens_seen": 79298560
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009858174523570712,
+      "loss": 2.3083,
+      "theoretical_loss": 4.958695070702744,
+      "tokens_seen": 79364096
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009857973921765296,
+      "loss": 2.3116,
+      "theoretical_loss": 4.958112411763365,
+      "tokens_seen": 79429632
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000985777331995988,
+      "loss": 2.328,
+      "theoretical_loss": 4.957530367848651,
+      "tokens_seen": 79495168
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009857572718154463,
+      "loss": 2.4212,
+      "theoretical_loss": 4.95694893780319,
+      "tokens_seen": 79560704
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009857372116349047,
+      "loss": 2.325,
+      "theoretical_loss": 4.956368120474689,
+      "tokens_seen": 79626240
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000985717151454363,
+      "loss": 2.3468,
+      "theoretical_loss": 4.955787914713962,
+      "tokens_seen": 79691776
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009856970912738214,
+      "loss": 2.3578,
+      "theoretical_loss": 4.955208319374924,
+      "tokens_seen": 79757312
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00098567703109328,
+      "loss": 2.4362,
+      "theoretical_loss": 4.954629333314573,
+      "tokens_seen": 79822848
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009856569709127381,
+      "loss": 2.2681,
+      "theoretical_loss": 4.954050955392988,
+      "tokens_seen": 79888384
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009856369107321967,
+      "loss": 2.2174,
+      "theoretical_loss": 4.953473184473312,
+      "tokens_seen": 79953920
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009856168505516549,
+      "loss": 2.2807,
+      "theoretical_loss": 4.952896019421743,
+      "tokens_seen": 80019456
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009855967903711134,
+      "loss": 2.329,
+      "theoretical_loss": 4.952319459107521,
+      "tokens_seen": 80084992
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009855767301905718,
+      "loss": 2.2275,
+      "theoretical_loss": 4.951743502402923,
+      "tokens_seen": 80150528
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009855566700100302,
+      "loss": 2.2448,
+      "theoretical_loss": 4.951168148183246,
+      "tokens_seen": 80216064
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009855366098294885,
+      "loss": 2.3235,
+      "theoretical_loss": 4.9505933953268,
+      "tokens_seen": 80281600
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009855165496489469,
+      "loss": 2.3358,
+      "theoretical_loss": 4.9500192427149,
+      "tokens_seen": 80347136
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009854964894684052,
+      "loss": 2.2623,
+      "theoretical_loss": 4.949445689231847,
+      "tokens_seen": 80412672
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009854764292878636,
+      "loss": 2.2786,
+      "theoretical_loss": 4.948872733764926,
+      "tokens_seen": 80478208
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000985456369107322,
+      "loss": 2.1879,
+      "theoretical_loss": 4.9483003752043935,
+      "tokens_seen": 80543744
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009854363089267803,
+      "loss": 2.1886,
+      "theoretical_loss": 4.9477286124434645,
+      "tokens_seen": 80609280
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009854162487462387,
+      "loss": 2.3044,
+      "theoretical_loss": 4.947157444378303,
+      "tokens_seen": 80674816
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000985396188565697,
+      "loss": 2.2717,
+      "theoretical_loss": 4.946586869908014,
+      "tokens_seen": 80740352
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009853761283851554,
+      "loss": 2.268,
+      "theoretical_loss": 4.946016887934633,
+      "tokens_seen": 80805888
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009853560682046138,
+      "loss": 2.37,
+      "theoretical_loss": 4.945447497363112,
+      "tokens_seen": 80871424
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009853360080240721,
+      "loss": 2.3511,
+      "theoretical_loss": 4.944878697101316,
+      "tokens_seen": 80936960
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009853159478435307,
+      "loss": 2.3928,
+      "theoretical_loss": 4.944310486060004,
+      "tokens_seen": 81002496
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009852958876629889,
+      "loss": 2.2323,
+      "theoretical_loss": 4.943742863152828,
+      "tokens_seen": 81068032
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009852758274824474,
+      "loss": 2.203,
+      "theoretical_loss": 4.943175827296319,
+      "tokens_seen": 81133568
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009852557673019056,
+      "loss": 2.3613,
+      "theoretical_loss": 4.942609377409873,
+      "tokens_seen": 81199104
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009852357071213642,
+      "loss": 2.3922,
+      "theoretical_loss": 4.942043512415751,
+      "tokens_seen": 81264640
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009852156469408225,
+      "loss": 2.3077,
+      "theoretical_loss": 4.94147823123906,
+      "tokens_seen": 81330176
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000985195586760281,
+      "loss": 2.4319,
+      "theoretical_loss": 4.940913532807745,
+      "tokens_seen": 81395712
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009851755265797393,
+      "loss": 2.3457,
+      "theoretical_loss": 4.940349416052586,
+      "tokens_seen": 81461248
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009851554663991976,
+      "loss": 2.3576,
+      "theoretical_loss": 4.939785879907176,
+      "tokens_seen": 81526784
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000985135406218656,
+      "loss": 2.1788,
+      "theoretical_loss": 4.939222923307925,
+      "tokens_seen": 81592320
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009851153460381143,
+      "loss": 2.3332,
+      "theoretical_loss": 4.938660545194038,
+      "tokens_seen": 81657856
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009850952858575727,
+      "loss": 2.4885,
+      "theoretical_loss": 4.938098744507517,
+      "tokens_seen": 81723392
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000985075225677031,
+      "loss": 2.4229,
+      "theoretical_loss": 4.937537520193139,
+      "tokens_seen": 81788928
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009850551654964894,
+      "loss": 2.2958,
+      "theoretical_loss": 4.936976871198457,
+      "tokens_seen": 81854464
+    },
+    {
+      "epoch": 0.02,
+      "objective/train/advantage_avg": 0.48865529894828796,
+      "objective/train/docs_used": 37569,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 2.141408681869507,
+      "objective/train/original_loss": 2.141408920288086,
+      "objective/train/theoretical_loss": 4.936416796473786,
+      "objective/train/tokens_used": 102380000,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24008464813232422,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0500857830047607,
+      "objective/train/weighted_lm_loss": 2.2487919330596924,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.936416796473786,
+      "tokens_seen": 81920000
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009850351053159478,
+      "loss": 2.3225,
+      "theoretical_loss": 4.936416796473786,
+      "tokens_seen": 81920000
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009850150451354062,
+      "loss": 2.3456,
+      "theoretical_loss": 4.935857294972193,
+      "tokens_seen": 81985536
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009849949849548645,
+      "loss": 2.1802,
+      "theoretical_loss": 4.93529836564949,
+      "tokens_seen": 82051072
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000984974924774323,
+      "loss": 2.3381,
+      "theoretical_loss": 4.934740007464219,
+      "tokens_seen": 82116608
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009849548645937815,
+      "loss": 2.3167,
+      "theoretical_loss": 4.934182219377654,
+      "tokens_seen": 82182144
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009849348044132398,
+      "loss": 2.2076,
+      "theoretical_loss": 4.9336250003537785,
+      "tokens_seen": 82247680
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009849147442326982,
+      "loss": 2.2008,
+      "theoretical_loss": 4.933068349359283,
+      "tokens_seen": 82313216
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0009848946840521565,
+      "loss": 2.255,
+      "theoretical_loss": 4.93251226536356,
+      "tokens_seen": 82378752
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.000984874623871615,
+      "loss": 2.3702,
+      "theoretical_loss": 4.931956747338681,
+      "tokens_seen": 82444288
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009848545636910733,
+      "loss": 2.2657,
+      "theoretical_loss": 4.931401794259408,
+      "tokens_seen": 82509824
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009848345035105316,
+      "loss": 2.3108,
+      "theoretical_loss": 4.93084740510316,
+      "tokens_seen": 82575360
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00098481444332999,
+      "loss": 2.289,
+      "theoretical_loss": 4.930293578850026,
+      "tokens_seen": 82640896
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009847943831494484,
+      "loss": 2.2171,
+      "theoretical_loss": 4.9297403144827445,
+      "tokens_seen": 82706432
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009847743229689067,
+      "loss": 2.2979,
+      "theoretical_loss": 4.929187610986693,
+      "tokens_seen": 82771968
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000984754262788365,
+      "loss": 2.2295,
+      "theoretical_loss": 4.928635467349885,
+      "tokens_seen": 82837504
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009847342026078234,
+      "loss": 2.3672,
+      "theoretical_loss": 4.9280838825629605,
+      "tokens_seen": 82903040
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009847141424272818,
+      "loss": 2.3961,
+      "theoretical_loss": 4.927532855619173,
+      "tokens_seen": 82968576
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009846940822467404,
+      "loss": 2.2037,
+      "theoretical_loss": 4.926982385514384,
+      "tokens_seen": 83034112
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009846740220661985,
+      "loss": 2.3054,
+      "theoretical_loss": 4.92643247124705,
+      "tokens_seen": 83099648
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009846539618856571,
+      "loss": 2.2816,
+      "theoretical_loss": 4.925883111818225,
+      "tokens_seen": 83165184
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009846339017051153,
+      "loss": 2.2191,
+      "theoretical_loss": 4.925334306231532,
+      "tokens_seen": 83230720
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009846138415245738,
+      "loss": 2.36,
+      "theoretical_loss": 4.924786053493177,
+      "tokens_seen": 83296256
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009845937813440322,
+      "loss": 2.3395,
+      "theoretical_loss": 4.924238352611924,
+      "tokens_seen": 83361792
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009845737211634906,
+      "loss": 2.3407,
+      "theoretical_loss": 4.923691202599091,
+      "tokens_seen": 83427328
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000984553660982949,
+      "loss": 2.0856,
+      "theoretical_loss": 4.923144602468545,
+      "tokens_seen": 83492864
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009845336008024073,
+      "loss": 2.2128,
+      "theoretical_loss": 4.922598551236689,
+      "tokens_seen": 83558400
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009845135406218656,
+      "loss": 2.4467,
+      "theoretical_loss": 4.922053047922455,
+      "tokens_seen": 83623936
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000984493480441324,
+      "loss": 2.266,
+      "theoretical_loss": 4.921508091547297,
+      "tokens_seen": 83689472
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009844734202607824,
+      "loss": 2.2651,
+      "theoretical_loss": 4.920963681135177,
+      "tokens_seen": 83755008
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009844533600802407,
+      "loss": 2.2465,
+      "theoretical_loss": 4.9204198157125685,
+      "tokens_seen": 83820544
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000984433299899699,
+      "loss": 2.2151,
+      "theoretical_loss": 4.919876494308432,
+      "tokens_seen": 83886080
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009844132397191575,
+      "loss": 2.3275,
+      "theoretical_loss": 4.91933371595422,
+      "tokens_seen": 83951616
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009843931795386158,
+      "loss": 2.3892,
+      "theoretical_loss": 4.918791479683863,
+      "tokens_seen": 84017152
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009843731193580742,
+      "loss": 2.31,
+      "theoretical_loss": 4.91824978453376,
+      "tokens_seen": 84082688
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009843530591775325,
+      "loss": 2.3452,
+      "theoretical_loss": 4.917708629542775,
+      "tokens_seen": 84148224
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009843329989969911,
+      "loss": 2.2599,
+      "theoretical_loss": 4.917168013752225,
+      "tokens_seen": 84213760
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009843129388164493,
+      "loss": 2.2059,
+      "theoretical_loss": 4.916627936205872,
+      "tokens_seen": 84279296
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009842928786359078,
+      "loss": 2.2837,
+      "theoretical_loss": 4.916088395949915,
+      "tokens_seen": 84344832
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000984272818455366,
+      "loss": 2.4008,
+      "theoretical_loss": 4.915549392032985,
+      "tokens_seen": 84410368
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009842527582748246,
+      "loss": 2.4535,
+      "theoretical_loss": 4.915010923506134,
+      "tokens_seen": 84475904
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000984232698094283,
+      "loss": 2.3609,
+      "theoretical_loss": 4.914472989422826,
+      "tokens_seen": 84541440
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009842126379137413,
+      "loss": 2.1934,
+      "theoretical_loss": 4.913935588838932,
+      "tokens_seen": 84606976
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009841925777331997,
+      "loss": 2.3023,
+      "theoretical_loss": 4.913398720812719,
+      "tokens_seen": 84672512
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000984172517552658,
+      "loss": 2.2574,
+      "theoretical_loss": 4.9128623844048445,
+      "tokens_seen": 84738048
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009841524573721164,
+      "loss": 2.2642,
+      "theoretical_loss": 4.91232657867835,
+      "tokens_seen": 84803584
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009841323971915747,
+      "loss": 2.2327,
+      "theoretical_loss": 4.911791302698646,
+      "tokens_seen": 84869120
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000984112337011033,
+      "loss": 2.3648,
+      "theoretical_loss": 4.9112565555335115,
+      "tokens_seen": 84934656
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009840922768304915,
+      "loss": 2.3775,
+      "theoretical_loss": 4.910722336253086,
+      "tokens_seen": 85000192
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009840722166499498,
+      "loss": 2.2554,
+      "theoretical_loss": 4.910188643929853,
+      "tokens_seen": 85065728
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009840521564694082,
+      "loss": 2.386,
+      "theoretical_loss": 4.909655477638648,
+      "tokens_seen": 85131264
+    },
+    {
+      "epoch": 0.03,
+      "objective/train/advantage_avg": 0.4955958127975464,
+      "objective/train/docs_used": 38678,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 2.1719961166381836,
+      "objective/train/original_loss": 2.1719963550567627,
+      "objective/train/theoretical_loss": 4.909122836456632,
+      "objective/train/tokens_used": 105656800,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24580909311771393,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0508092641830444,
+      "objective/train/weighted_lm_loss": 2.282179594039917,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 1.0304546356201172,
+      "theoretical_loss": 4.909122836456632,
+      "tokens_seen": 85196800
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009840320962888666,
+      "loss": 2.2857,
+      "theoretical_loss": 4.909122836456632,
+      "tokens_seen": 85196800
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000984012036108325,
+      "loss": 2.243,
+      "theoretical_loss": 4.908590719463298,
+      "tokens_seen": 85262336
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009839919759277833,
+      "loss": 2.3113,
+      "theoretical_loss": 4.9080591257404595,
+      "tokens_seen": 85327872
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009839719157472419,
+      "loss": 2.2506,
+      "theoretical_loss": 4.907528054372239,
+      "tokens_seen": 85393408
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009839518555667,
+      "loss": 2.1794,
+      "theoretical_loss": 4.906997504445066,
+      "tokens_seen": 85458944
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009839317953861586,
+      "loss": 2.2323,
+      "theoretical_loss": 4.906467475047664,
+      "tokens_seen": 85524480
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009839117352056167,
+      "loss": 2.4132,
+      "theoretical_loss": 4.905937965271047,
+      "tokens_seen": 85590016
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009838916750250753,
+      "loss": 2.314,
+      "theoretical_loss": 4.905408974208514,
+      "tokens_seen": 85655552
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009838716148445337,
+      "loss": 2.3108,
+      "theoretical_loss": 4.904880500955633,
+      "tokens_seen": 85721088
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000983851554663992,
+      "loss": 2.1642,
+      "theoretical_loss": 4.904352544610242,
+      "tokens_seen": 85786624
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009838314944834504,
+      "loss": 2.354,
+      "theoretical_loss": 4.903825104272437,
+      "tokens_seen": 85852160
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009838114343029088,
+      "loss": 2.2949,
+      "theoretical_loss": 4.903298179044569,
+      "tokens_seen": 85917696
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009837913741223671,
+      "loss": 2.2796,
+      "theoretical_loss": 4.90277176803123,
+      "tokens_seen": 85983232
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009837713139418255,
+      "loss": 2.259,
+      "theoretical_loss": 4.902245870339252,
+      "tokens_seen": 86048768
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009837512537612838,
+      "loss": 2.3348,
+      "theoretical_loss": 4.9017204850776945,
+      "tokens_seen": 86114304
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009837311935807422,
+      "loss": 2.3361,
+      "theoretical_loss": 4.901195611357844,
+      "tokens_seen": 86179840
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009837111334002006,
+      "loss": 2.1493,
+      "theoretical_loss": 4.9006712482931984,
+      "tokens_seen": 86245376
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000983691073219659,
+      "loss": 2.3035,
+      "theoretical_loss": 4.9001473949994665,
+      "tokens_seen": 86310912
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009836710130391173,
+      "loss": 2.1573,
+      "theoretical_loss": 4.899624050594557,
+      "tokens_seen": 86376448
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009836509528585757,
+      "loss": 2.3964,
+      "theoretical_loss": 4.899101214198575,
+      "tokens_seen": 86441984
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000983630892678034,
+      "loss": 2.1748,
+      "theoretical_loss": 4.89857888493381,
+      "tokens_seen": 86507520
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009836108324974926,
+      "loss": 2.2292,
+      "theoretical_loss": 4.898057061924732,
+      "tokens_seen": 86573056
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009835907723169507,
+      "loss": 2.3679,
+      "theoretical_loss": 4.897535744297986,
+      "tokens_seen": 86638592
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009835707121364093,
+      "loss": 2.3989,
+      "theoretical_loss": 4.89701493118238,
+      "tokens_seen": 86704128
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009835506519558675,
+      "loss": 2.1716,
+      "theoretical_loss": 4.896494621708882,
+      "tokens_seen": 86769664
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000983530591775326,
+      "loss": 2.2875,
+      "theoretical_loss": 4.895974815010612,
+      "tokens_seen": 86835200
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009835105315947844,
+      "loss": 2.2411,
+      "theoretical_loss": 4.895455510222835,
+      "tokens_seen": 86900736
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009834904714142428,
+      "loss": 2.2779,
+      "theoretical_loss": 4.8949367064829525,
+      "tokens_seen": 86966272
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009834704112337011,
+      "loss": 2.2055,
+      "theoretical_loss": 4.8944184029305,
+      "tokens_seen": 87031808
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009834503510531595,
+      "loss": 2.3822,
+      "theoretical_loss": 4.893900598707134,
+      "tokens_seen": 87097344
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009834302908726179,
+      "loss": 2.1673,
+      "theoretical_loss": 4.893383292956631,
+      "tokens_seen": 87162880
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009834102306920762,
+      "loss": 2.3037,
+      "theoretical_loss": 4.892866484824879,
+      "tokens_seen": 87228416
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009833901705115346,
+      "loss": 2.1902,
+      "theoretical_loss": 4.892350173459863,
+      "tokens_seen": 87293952
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000983370110330993,
+      "loss": 2.2473,
+      "theoretical_loss": 4.891834358011673,
+      "tokens_seen": 87359488
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009833500501504515,
+      "loss": 2.1875,
+      "theoretical_loss": 4.891319037632486,
+      "tokens_seen": 87425024
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009833299899699097,
+      "loss": 2.2182,
+      "theoretical_loss": 4.890804211476562,
+      "tokens_seen": 87490560
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009833099297893682,
+      "loss": 2.2744,
+      "theoretical_loss": 4.890289878700239,
+      "tokens_seen": 87556096
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009832898696088264,
+      "loss": 2.202,
+      "theoretical_loss": 4.889776038461924,
+      "tokens_seen": 87621632
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000983269809428285,
+      "loss": 2.203,
+      "theoretical_loss": 4.88926268992209,
+      "tokens_seen": 87687168
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009832497492477433,
+      "loss": 2.216,
+      "theoretical_loss": 4.888749832243265,
+      "tokens_seen": 87752704
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009832296890672017,
+      "loss": 2.2196,
+      "theoretical_loss": 4.888237464590028,
+      "tokens_seen": 87818240
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00098320962888666,
+      "loss": 2.3488,
+      "theoretical_loss": 4.887725586129003,
+      "tokens_seen": 87883776
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009831895687061184,
+      "loss": 2.2586,
+      "theoretical_loss": 4.887214196028848,
+      "tokens_seen": 87949312
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009831695085255768,
+      "loss": 2.2706,
+      "theoretical_loss": 4.886703293460259,
+      "tokens_seen": 88014848
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009831494483450351,
+      "loss": 2.2567,
+      "theoretical_loss": 4.8861928775959464,
+      "tokens_seen": 88080384
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009831293881644935,
+      "loss": 2.2866,
+      "theoretical_loss": 4.885682947610648,
+      "tokens_seen": 88145920
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009831093279839519,
+      "loss": 2.2333,
+      "theoretical_loss": 4.885173502681107,
+      "tokens_seen": 88211456
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009830892678034102,
+      "loss": 2.2126,
+      "theoretical_loss": 4.884664541986075,
+      "tokens_seen": 88276992
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009830692076228686,
+      "loss": 2.3505,
+      "theoretical_loss": 4.884156064706302,
+      "tokens_seen": 88342528
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000983049147442327,
+      "loss": 2.1561,
+      "theoretical_loss": 4.883648070024529,
+      "tokens_seen": 88408064
+    },
+    {
+      "epoch": 0.03,
+      "objective/train/advantage_avg": 0.49346664547920227,
+      "objective/train/docs_used": 39491,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 2.0473103523254395,
+      "objective/train/original_loss": 2.0473103523254395,
+      "objective/train/theoretical_loss": 4.8831405571254844,
+      "objective/train/tokens_used": 108933600,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24416816234588623,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505878925323486,
+      "objective/train/weighted_lm_loss": 2.150813102722168,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.8831405571254844,
+      "tokens_seen": 88473600
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009830290872617853,
+      "loss": 2.2569,
+      "theoretical_loss": 4.8831405571254844,
+      "tokens_seen": 88473600
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009830090270812437,
+      "loss": 2.2434,
+      "theoretical_loss": 4.8826335251958755,
+      "tokens_seen": 88539136
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009829889669007023,
+      "loss": 2.2955,
+      "theoretical_loss": 4.882126973424384,
+      "tokens_seen": 88604672
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009829689067201604,
+      "loss": 2.3253,
+      "theoretical_loss": 4.881620901001661,
+      "tokens_seen": 88670208
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000982948846539619,
+      "loss": 2.2168,
+      "theoretical_loss": 4.881115307120314,
+      "tokens_seen": 88735744
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009829287863590771,
+      "loss": 2.155,
+      "theoretical_loss": 4.88061019097491,
+      "tokens_seen": 88801280
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009829087261785357,
+      "loss": 2.3044,
+      "theoretical_loss": 4.880105551761961,
+      "tokens_seen": 88866816
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000982888665997994,
+      "loss": 2.2709,
+      "theoretical_loss": 4.879601388679923,
+      "tokens_seen": 88932352
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009828686058174524,
+      "loss": 2.2732,
+      "theoretical_loss": 4.879097700929192,
+      "tokens_seen": 88997888
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009828485456369108,
+      "loss": 2.1625,
+      "theoretical_loss": 4.878594487712088,
+      "tokens_seen": 89063424
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009828284854563692,
+      "loss": 2.4076,
+      "theoretical_loss": 4.87809174823286,
+      "tokens_seen": 89128960
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009828084252758275,
+      "loss": 2.3207,
+      "theoretical_loss": 4.877589481697674,
+      "tokens_seen": 89194496
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009827883650952859,
+      "loss": 2.1592,
+      "theoretical_loss": 4.877087687314608,
+      "tokens_seen": 89260032
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009827683049147442,
+      "loss": 2.1353,
+      "theoretical_loss": 4.876586364293646,
+      "tokens_seen": 89325568
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009827482447342026,
+      "loss": 2.2606,
+      "theoretical_loss": 4.876085511846673,
+      "tokens_seen": 89391104
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000982728184553661,
+      "loss": 2.2478,
+      "theoretical_loss": 4.875585129187467,
+      "tokens_seen": 89456640
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009827081243731193,
+      "loss": 2.279,
+      "theoretical_loss": 4.875085215531697,
+      "tokens_seen": 89522176
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009826880641925777,
+      "loss": 2.2824,
+      "theoretical_loss": 4.87458577009691,
+      "tokens_seen": 89587712
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000982668004012036,
+      "loss": 2.2527,
+      "theoretical_loss": 4.874086792102535,
+      "tokens_seen": 89653248
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009826479438314944,
+      "loss": 2.2688,
+      "theoretical_loss": 4.873588280769868,
+      "tokens_seen": 89718784
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000982627883650953,
+      "loss": 2.0824,
+      "theoretical_loss": 4.873090235322069,
+      "tokens_seen": 89784320
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009826078234704111,
+      "loss": 2.0821,
+      "theoretical_loss": 4.87259265498416,
+      "tokens_seen": 89849856
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009825877632898697,
+      "loss": 2.2779,
+      "theoretical_loss": 4.872095538983015,
+      "tokens_seen": 89915392
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009825677031093279,
+      "loss": 2.2936,
+      "theoretical_loss": 4.871598886547356,
+      "tokens_seen": 89980928
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009825476429287864,
+      "loss": 2.2121,
+      "theoretical_loss": 4.871102696907745,
+      "tokens_seen": 90046464
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009825275827482448,
+      "loss": 2.2357,
+      "theoretical_loss": 4.870606969296581,
+      "tokens_seen": 90112000
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009825075225677032,
+      "loss": 2.5256,
+      "theoretical_loss": 4.870111702948094,
+      "tokens_seen": 90177536
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009824874623871615,
+      "loss": 2.2016,
+      "theoretical_loss": 4.869616897098336,
+      "tokens_seen": 90243072
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00098246740220662,
+      "loss": 2.1991,
+      "theoretical_loss": 4.869122550985182,
+      "tokens_seen": 90308608
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009824473420260783,
+      "loss": 2.3257,
+      "theoretical_loss": 4.868628663848316,
+      "tokens_seen": 90374144
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009824272818455366,
+      "loss": 2.3091,
+      "theoretical_loss": 4.868135234929232,
+      "tokens_seen": 90439680
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000982407221664995,
+      "loss": 2.3082,
+      "theoretical_loss": 4.867642263471224,
+      "tokens_seen": 90505216
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009823871614844533,
+      "loss": 2.1111,
+      "theoretical_loss": 4.867149748719386,
+      "tokens_seen": 90570752
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009823671013039117,
+      "loss": 2.3432,
+      "theoretical_loss": 4.866657689920601,
+      "tokens_seen": 90636288
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00098234704112337,
+      "loss": 2.1497,
+      "theoretical_loss": 4.866166086323535,
+      "tokens_seen": 90701824
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009823269809428284,
+      "loss": 2.1864,
+      "theoretical_loss": 4.865674937178635,
+      "tokens_seen": 90767360
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009823069207622868,
+      "loss": 2.1773,
+      "theoretical_loss": 4.865184241738126,
+      "tokens_seen": 90832896
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009822868605817452,
+      "loss": 2.2805,
+      "theoretical_loss": 4.864693999255998,
+      "tokens_seen": 90898432
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009822668004012037,
+      "loss": 2.3565,
+      "theoretical_loss": 4.864204208988003,
+      "tokens_seen": 90963968
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009822467402206619,
+      "loss": 2.1671,
+      "theoretical_loss": 4.8637148701916555,
+      "tokens_seen": 91029504
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009822266800401205,
+      "loss": 2.2914,
+      "theoretical_loss": 4.863225982126219,
+      "tokens_seen": 91095040
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009822066198595786,
+      "loss": 2.2608,
+      "theoretical_loss": 4.862737544052706,
+      "tokens_seen": 91160576
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009821865596790372,
+      "loss": 2.1416,
+      "theoretical_loss": 4.86224955523387,
+      "tokens_seen": 91226112
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009821664994984955,
+      "loss": 2.3752,
+      "theoretical_loss": 4.8617620149342,
+      "tokens_seen": 91291648
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000982146439317954,
+      "loss": 2.2634,
+      "theoretical_loss": 4.861274922419918,
+      "tokens_seen": 91357184
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009821263791374123,
+      "loss": 2.322,
+      "theoretical_loss": 4.86078827695897,
+      "tokens_seen": 91422720
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009821063189568706,
+      "loss": 2.2562,
+      "theoretical_loss": 4.860302077821023,
+      "tokens_seen": 91488256
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000982086258776329,
+      "loss": 2.1679,
+      "theoretical_loss": 4.859816324277461,
+      "tokens_seen": 91553792
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009820661985957874,
+      "loss": 2.2282,
+      "theoretical_loss": 4.859331015601375,
+      "tokens_seen": 91619328
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009820461384152457,
+      "loss": 2.1868,
+      "theoretical_loss": 4.858846151067562,
+      "tokens_seen": 91684864
+    },
+    {
+      "epoch": 0.03,
+      "objective/train/advantage_avg": 0.49189305305480957,
+      "objective/train/docs_used": 40748,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.9773281812667847,
+      "objective/train/original_loss": 1.9773280620574951,
+      "objective/train/theoretical_loss": 4.858361729952518,
+      "objective/train/tokens_used": 112210400,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24240271747112274,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504215955734253,
+      "objective/train/weighted_lm_loss": 2.0769896507263184,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.858361729952518,
+      "tokens_seen": 91750400
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000982026078234704,
+      "loss": 2.2005,
+      "theoretical_loss": 4.858361729952518,
+      "tokens_seen": 91750400
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009820060180541624,
+      "loss": 2.199,
+      "theoretical_loss": 4.857877751534435,
+      "tokens_seen": 91815936
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009819859578736208,
+      "loss": 2.2238,
+      "theoretical_loss": 4.857394215093191,
+      "tokens_seen": 91881472
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009819658976930794,
+      "loss": 2.2805,
+      "theoretical_loss": 4.856911119910352,
+      "tokens_seen": 91947008
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009819458375125375,
+      "loss": 2.1968,
+      "theoretical_loss": 4.856428465269159,
+      "tokens_seen": 92012544
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009819257773319961,
+      "loss": 2.1694,
+      "theoretical_loss": 4.85594625045453,
+      "tokens_seen": 92078080
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009819057171514545,
+      "loss": 2.2446,
+      "theoretical_loss": 4.855464474753049,
+      "tokens_seen": 92143616
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009818856569709128,
+      "loss": 2.1264,
+      "theoretical_loss": 4.854983137452964,
+      "tokens_seen": 92209152
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009818655967903712,
+      "loss": 2.2233,
+      "theoretical_loss": 4.8545022378441836,
+      "tokens_seen": 92274688
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009818455366098296,
+      "loss": 2.3523,
+      "theoretical_loss": 4.854021775218266,
+      "tokens_seen": 92340224
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000981825476429288,
+      "loss": 2.2655,
+      "theoretical_loss": 4.853541748868423,
+      "tokens_seen": 92405760
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009818054162487463,
+      "loss": 2.3512,
+      "theoretical_loss": 4.853062158089504,
+      "tokens_seen": 92471296
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009817853560682046,
+      "loss": 2.1191,
+      "theoretical_loss": 4.852583002178001,
+      "tokens_seen": 92536832
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000981765295887663,
+      "loss": 2.1088,
+      "theoretical_loss": 4.852104280432039,
+      "tokens_seen": 92602368
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009817452357071214,
+      "loss": 2.2118,
+      "theoretical_loss": 4.85162599215137,
+      "tokens_seen": 92667904
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009817251755265797,
+      "loss": 2.256,
+      "theoretical_loss": 4.85114813663737,
+      "tokens_seen": 92733440
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000981705115346038,
+      "loss": 2.2957,
+      "theoretical_loss": 4.8506707131930344,
+      "tokens_seen": 92798976
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009816850551654965,
+      "loss": 2.0396,
+      "theoretical_loss": 4.850193721122972,
+      "tokens_seen": 92864512
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009816649949849548,
+      "loss": 2.2755,
+      "theoretical_loss": 4.849717159733401,
+      "tokens_seen": 92930048
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009816449348044134,
+      "loss": 2.2867,
+      "theoretical_loss": 4.849241028332142,
+      "tokens_seen": 92995584
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009816248746238715,
+      "loss": 2.2158,
+      "theoretical_loss": 4.848765326228618,
+      "tokens_seen": 93061120
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009816048144433301,
+      "loss": 2.1635,
+      "theoretical_loss": 4.848290052733843,
+      "tokens_seen": 93126656
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009815847542627883,
+      "loss": 2.186,
+      "theoretical_loss": 4.847815207160422,
+      "tokens_seen": 93192192
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009815646940822468,
+      "loss": 2.2273,
+      "theoretical_loss": 4.847340788822546,
+      "tokens_seen": 93257728
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009815446339017052,
+      "loss": 2.1433,
+      "theoretical_loss": 4.846866797035984,
+      "tokens_seen": 93323264
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009815245737211636,
+      "loss": 2.2482,
+      "theoretical_loss": 4.846393231118083,
+      "tokens_seen": 93388800
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000981504513540622,
+      "loss": 2.2104,
+      "theoretical_loss": 4.845920090387757,
+      "tokens_seen": 93454336
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009814844533600803,
+      "loss": 2.1955,
+      "theoretical_loss": 4.84544737416549,
+      "tokens_seen": 93519872
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009814643931795387,
+      "loss": 2.2442,
+      "theoretical_loss": 4.844975081773322,
+      "tokens_seen": 93585408
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000981444332998997,
+      "loss": 2.1941,
+      "theoretical_loss": 4.844503212534856,
+      "tokens_seen": 93650944
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009814242728184554,
+      "loss": 2.1822,
+      "theoretical_loss": 4.844031765775242,
+      "tokens_seen": 93716480
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009814042126379137,
+      "loss": 2.1861,
+      "theoretical_loss": 4.843560740821177,
+      "tokens_seen": 93782016
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000981384152457372,
+      "loss": 2.3743,
+      "theoretical_loss": 4.843090137000904,
+      "tokens_seen": 93847552
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009813640922768305,
+      "loss": 2.0544,
+      "theoretical_loss": 4.842619953644203,
+      "tokens_seen": 93913088
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009813440320962888,
+      "loss": 2.3207,
+      "theoretical_loss": 4.842150190082384,
+      "tokens_seen": 93978624
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009813239719157472,
+      "loss": 2.0991,
+      "theoretical_loss": 4.841680845648289,
+      "tokens_seen": 94044160
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009813039117352056,
+      "loss": 2.163,
+      "theoretical_loss": 4.841211919676287,
+      "tokens_seen": 94109696
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009812838515546641,
+      "loss": 2.1717,
+      "theoretical_loss": 4.84074341150226,
+      "tokens_seen": 94175232
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009812637913741223,
+      "loss": 2.1603,
+      "theoretical_loss": 4.840275320463608,
+      "tokens_seen": 94240768
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009812437311935809,
+      "loss": 2.2077,
+      "theoretical_loss": 4.839807645899245,
+      "tokens_seen": 94306304
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000981223671013039,
+      "loss": 2.0532,
+      "theoretical_loss": 4.839340387149586,
+      "tokens_seen": 94371840
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009812036108324976,
+      "loss": 2.1507,
+      "theoretical_loss": 4.838873543556552,
+      "tokens_seen": 94437376
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000981183550651956,
+      "loss": 2.1606,
+      "theoretical_loss": 4.838407114463558,
+      "tokens_seen": 94502912
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009811634904714143,
+      "loss": 2.2777,
+      "theoretical_loss": 4.837941099215515,
+      "tokens_seen": 94568448
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009811434302908727,
+      "loss": 2.1049,
+      "theoretical_loss": 4.837475497158817,
+      "tokens_seen": 94633984
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000981123370110331,
+      "loss": 2.2131,
+      "theoretical_loss": 4.83701030764135,
+      "tokens_seen": 94699520
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009811033099297894,
+      "loss": 2.0608,
+      "theoretical_loss": 4.836545530012472,
+      "tokens_seen": 94765056
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009810832497492478,
+      "loss": 2.1397,
+      "theoretical_loss": 4.83608116362302,
+      "tokens_seen": 94830592
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009810631895687061,
+      "loss": 2.1879,
+      "theoretical_loss": 4.835617207825303,
+      "tokens_seen": 94896128
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009810431293881645,
+      "loss": 2.2057,
+      "theoretical_loss": 4.835153661973092,
+      "tokens_seen": 94961664
+    },
+    {
+      "epoch": 0.03,
+      "objective/train/advantage_avg": 0.48443371057510376,
+      "objective/train/docs_used": 42039,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.8953099250793457,
+      "objective/train/original_loss": 1.8953099250793457,
+      "objective/train/theoretical_loss": 4.8346905254216255,
+      "objective/train/tokens_used": 115487200,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.23641405999660492,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0496450662612915,
+      "objective/train/weighted_lm_loss": 1.9899815320968628,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.8346905254216255,
+      "tokens_seen": 95027200
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009810230692076228,
+      "loss": 2.0976,
+      "theoretical_loss": 4.8346905254216255,
+      "tokens_seen": 95027200
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009810030090270812,
+      "loss": 2.1386,
+      "theoretical_loss": 4.834227797527595,
+      "tokens_seen": 95092736
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009809829488465396,
+      "loss": 2.1371,
+      "theoretical_loss": 4.83376547764915,
+      "tokens_seen": 95158272
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000980962888665998,
+      "loss": 2.1342,
+      "theoretical_loss": 4.833303565145885,
+      "tokens_seen": 95223808
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009809428284854563,
+      "loss": 2.2161,
+      "theoretical_loss": 4.832842059378843,
+      "tokens_seen": 95289344
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009809227683049149,
+      "loss": 2.1805,
+      "theoretical_loss": 4.832380959710505,
+      "tokens_seen": 95354880
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000980902708124373,
+      "loss": 2.1964,
+      "theoretical_loss": 4.831920265504792,
+      "tokens_seen": 95420416
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009808826479438316,
+      "loss": 2.1742,
+      "theoretical_loss": 4.831459976127054,
+      "tokens_seen": 95485952
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009808625877632897,
+      "loss": 2.2361,
+      "theoretical_loss": 4.831000090944068,
+      "tokens_seen": 95551488
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009808425275827483,
+      "loss": 2.2147,
+      "theoretical_loss": 4.830540609324041,
+      "tokens_seen": 95617024
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009808224674022067,
+      "loss": 2.2732,
+      "theoretical_loss": 4.830081530636594,
+      "tokens_seen": 95682560
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000980802407221665,
+      "loss": 2.1651,
+      "theoretical_loss": 4.829622854252765,
+      "tokens_seen": 95748096
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009807823470411234,
+      "loss": 2.2456,
+      "theoretical_loss": 4.829164579545006,
+      "tokens_seen": 95813632
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009807622868605818,
+      "loss": 2.1658,
+      "theoretical_loss": 4.828706705887171,
+      "tokens_seen": 95879168
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009807422266800401,
+      "loss": 2.2252,
+      "theoretical_loss": 4.8282492326545245,
+      "tokens_seen": 95944704
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009807221664994985,
+      "loss": 2.2021,
+      "theoretical_loss": 4.827792159223722,
+      "tokens_seen": 96010240
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009807021063189569,
+      "loss": 2.2582,
+      "theoretical_loss": 4.82733548497282,
+      "tokens_seen": 96075776
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009806820461384152,
+      "loss": 2.0183,
+      "theoretical_loss": 4.8268792092812625,
+      "tokens_seen": 96141312
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009806619859578736,
+      "loss": 2.0873,
+      "theoretical_loss": 4.826423331529884,
+      "tokens_seen": 96206848
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000980641925777332,
+      "loss": 2.1451,
+      "theoretical_loss": 4.825967851100897,
+      "tokens_seen": 96272384
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009806218655967903,
+      "loss": 2.3053,
+      "theoretical_loss": 4.825512767377897,
+      "tokens_seen": 96337920
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009806018054162487,
+      "loss": 2.149,
+      "theoretical_loss": 4.825058079745853,
+      "tokens_seen": 96403456
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000980581745235707,
+      "loss": 2.1526,
+      "theoretical_loss": 4.824603787591102,
+      "tokens_seen": 96468992
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009805616850551656,
+      "loss": 2.1577,
+      "theoretical_loss": 4.824149890301353,
+      "tokens_seen": 96534528
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009805416248746238,
+      "loss": 2.1158,
+      "theoretical_loss": 4.823696387265674,
+      "tokens_seen": 96600064
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009805215646940823,
+      "loss": 2.2835,
+      "theoretical_loss": 4.823243277874491,
+      "tokens_seen": 96665600
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009805015045135407,
+      "loss": 2.2474,
+      "theoretical_loss": 4.822790561519591,
+      "tokens_seen": 96731136
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000980481444332999,
+      "loss": 2.1669,
+      "theoretical_loss": 4.822338237594105,
+      "tokens_seen": 96796672
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009804613841524574,
+      "loss": 2.1423,
+      "theoretical_loss": 4.821886305492513,
+      "tokens_seen": 96862208
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009804413239719158,
+      "loss": 2.2884,
+      "theoretical_loss": 4.8214347646106415,
+      "tokens_seen": 96927744
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009804212637913741,
+      "loss": 2.162,
+      "theoretical_loss": 4.8209836143456535,
+      "tokens_seen": 96993280
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009804012036108325,
+      "loss": 2.0766,
+      "theoretical_loss": 4.820532854096047,
+      "tokens_seen": 97058816
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009803811434302909,
+      "loss": 2.2347,
+      "theoretical_loss": 4.820082483261652,
+      "tokens_seen": 97124352
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009803610832497492,
+      "loss": 2.1397,
+      "theoretical_loss": 4.81963250124363,
+      "tokens_seen": 97189888
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009803410230692076,
+      "loss": 2.0618,
+      "theoretical_loss": 4.81918290744446,
+      "tokens_seen": 97255424
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000980320962888666,
+      "loss": 2.0536,
+      "theoretical_loss": 4.818733701267945,
+      "tokens_seen": 97320960
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009803009027081245,
+      "loss": 2.2106,
+      "theoretical_loss": 4.818284882119203,
+      "tokens_seen": 97386496
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009802808425275827,
+      "loss": 2.3802,
+      "theoretical_loss": 4.817836449404666,
+      "tokens_seen": 97452032
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009802607823470413,
+      "loss": 2.2977,
+      "theoretical_loss": 4.817388402532074,
+      "tokens_seen": 97517568
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009802407221664994,
+      "loss": 2.316,
+      "theoretical_loss": 4.81694074091047,
+      "tokens_seen": 97583104
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000980220661985958,
+      "loss": 2.4769,
+      "theoretical_loss": 4.816493463950202,
+      "tokens_seen": 97648640
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009802006018054163,
+      "loss": 2.2114,
+      "theoretical_loss": 4.816046571062912,
+      "tokens_seen": 97714176
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009801805416248747,
+      "loss": 2.2655,
+      "theoretical_loss": 4.815600061661536,
+      "tokens_seen": 97779712
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000980160481444333,
+      "loss": 2.0975,
+      "theoretical_loss": 4.815153935160305,
+      "tokens_seen": 97845248
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009801404212637914,
+      "loss": 2.1909,
+      "theoretical_loss": 4.814708190974729,
+      "tokens_seen": 97910784
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009801203610832498,
+      "loss": 2.2142,
+      "theoretical_loss": 4.8142628285216045,
+      "tokens_seen": 97976320
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009801003009027082,
+      "loss": 2.1922,
+      "theoretical_loss": 4.813817847219008,
+      "tokens_seen": 98041856
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009800802407221665,
+      "loss": 2.1104,
+      "theoretical_loss": 4.81337324648629,
+      "tokens_seen": 98107392
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009800601805416249,
+      "loss": 2.0705,
+      "theoretical_loss": 4.812929025744072,
+      "tokens_seen": 98172928
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009800401203610832,
+      "loss": 2.0202,
+      "theoretical_loss": 4.812485184414244,
+      "tokens_seen": 98238464
+    },
+    {
+      "epoch": 0.03,
+      "objective/train/advantage_avg": 0.4935273230075836,
+      "objective/train/docs_used": 43181,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 2.0869107246398926,
+      "objective/train/original_loss": 2.0869107246398926,
+      "objective/train/theoretical_loss": 4.812041721919962,
+      "objective/train/tokens_used": 118764000,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24405071139335632,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050593376159668,
+      "objective/train/weighted_lm_loss": 2.192361831665039,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.812041721919962,
+      "tokens_seen": 98304000
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009800200601805416,
+      "loss": 2.1919,
+      "theoretical_loss": 4.812041721919962,
+      "tokens_seen": 98304000
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00098,
+      "loss": 2.1269,
+      "theoretical_loss": 4.8115986376856394,
+      "tokens_seen": 98369536
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009799799398194583,
+      "loss": 2.1634,
+      "theoretical_loss": 4.81115593113695,
+      "tokens_seen": 98435072
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009799598796389167,
+      "loss": 2.0142,
+      "theoretical_loss": 4.8107136017008205,
+      "tokens_seen": 98500608
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009799398194583753,
+      "loss": 2.0447,
+      "theoretical_loss": 4.810271648805427,
+      "tokens_seen": 98566144
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009799197592778334,
+      "loss": 2.1669,
+      "theoretical_loss": 4.809830071880194,
+      "tokens_seen": 98631680
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000979899699097292,
+      "loss": 2.2777,
+      "theoretical_loss": 4.809388870355785,
+      "tokens_seen": 98697216
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009798796389167501,
+      "loss": 2.1788,
+      "theoretical_loss": 4.808948043664107,
+      "tokens_seen": 98762752
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009798595787362087,
+      "loss": 2.1686,
+      "theoretical_loss": 4.8085075912383015,
+      "tokens_seen": 98828288
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000979839518555667,
+      "loss": 2.0359,
+      "theoretical_loss": 4.808067512512741,
+      "tokens_seen": 98893824
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009798194583751254,
+      "loss": 2.0967,
+      "theoretical_loss": 4.807627806923029,
+      "tokens_seen": 98959360
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009797993981945838,
+      "loss": 2.2084,
+      "theoretical_loss": 4.807188473905994,
+      "tokens_seen": 99024896
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009797793380140422,
+      "loss": 2.267,
+      "theoretical_loss": 4.806749512899687,
+      "tokens_seen": 99090432
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009797592778335005,
+      "loss": 2.1556,
+      "theoretical_loss": 4.806310923343375,
+      "tokens_seen": 99155968
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000979739217652959,
+      "loss": 2.1009,
+      "theoretical_loss": 4.805872704677542,
+      "tokens_seen": 99221504
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009797191574724173,
+      "loss": 2.1948,
+      "theoretical_loss": 4.805434856343886,
+      "tokens_seen": 99287040
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009796990972918756,
+      "loss": 2.2327,
+      "theoretical_loss": 4.804997377785307,
+      "tokens_seen": 99352576
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000979679037111334,
+      "loss": 2.2464,
+      "theoretical_loss": 4.8045602684459165,
+      "tokens_seen": 99418112
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009796589769307923,
+      "loss": 2.168,
+      "theoretical_loss": 4.804123527771022,
+      "tokens_seen": 99483648
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009796389167502507,
+      "loss": 2.1951,
+      "theoretical_loss": 4.803687155207134,
+      "tokens_seen": 99549184
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000979618856569709,
+      "loss": 2.1594,
+      "theoretical_loss": 4.8032511502019535,
+      "tokens_seen": 99614720
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009795987963891674,
+      "loss": 2.1376,
+      "theoretical_loss": 4.802815512204375,
+      "tokens_seen": 99680256
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000979578736208626,
+      "loss": 2.1674,
+      "theoretical_loss": 4.80238024066448,
+      "tokens_seen": 99745792
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009795586760280842,
+      "loss": 2.0653,
+      "theoretical_loss": 4.801945335033535,
+      "tokens_seen": 99811328
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009795386158475427,
+      "loss": 2.3154,
+      "theoretical_loss": 4.801510794763988,
+      "tokens_seen": 99876864
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000979518555667001,
+      "loss": 2.0517,
+      "theoretical_loss": 4.801076619309466,
+      "tokens_seen": 99942400
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009794984954864595,
+      "loss": 2.0963,
+      "theoretical_loss": 4.800642808124767,
+      "tokens_seen": 100007936
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009794784353059178,
+      "loss": 2.3449,
+      "theoretical_loss": 4.800209360665864,
+      "tokens_seen": 100073472
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009794583751253762,
+      "loss": 2.1891,
+      "theoretical_loss": 4.799776276389897,
+      "tokens_seen": 100139008
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009794383149448345,
+      "loss": 2.183,
+      "theoretical_loss": 4.79934355475517,
+      "tokens_seen": 100204544
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000979418254764293,
+      "loss": 2.1676,
+      "theoretical_loss": 4.79891119522115,
+      "tokens_seen": 100270080
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009793981945837513,
+      "loss": 2.1522,
+      "theoretical_loss": 4.798479197248461,
+      "tokens_seen": 100335616
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009793781344032096,
+      "loss": 2.364,
+      "theoretical_loss": 4.798047560298882,
+      "tokens_seen": 100401152
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000979358074222668,
+      "loss": 2.2696,
+      "theoretical_loss": 4.797616283835346,
+      "tokens_seen": 100466688
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009793380140421264,
+      "loss": 2.0181,
+      "theoretical_loss": 4.797185367321932,
+      "tokens_seen": 100532224
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009793179538615847,
+      "loss": 2.1756,
+      "theoretical_loss": 4.796754810223866,
+      "tokens_seen": 100597760
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000979297893681043,
+      "loss": 2.0873,
+      "theoretical_loss": 4.796324612007515,
+      "tokens_seen": 100663296
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009792778335005014,
+      "loss": 2.1305,
+      "theoretical_loss": 4.795894772140386,
+      "tokens_seen": 100728832
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009792577733199598,
+      "loss": 2.2556,
+      "theoretical_loss": 4.795465290091123,
+      "tokens_seen": 100794368
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009792377131394182,
+      "loss": 2.0559,
+      "theoretical_loss": 4.795036165329499,
+      "tokens_seen": 100859904
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009792176529588767,
+      "loss": 2.1793,
+      "theoretical_loss": 4.794607397326421,
+      "tokens_seen": 100925440
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000979197592778335,
+      "loss": 2.1214,
+      "theoretical_loss": 4.794178985553919,
+      "tokens_seen": 100990976
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009791775325977935,
+      "loss": 2.2544,
+      "theoretical_loss": 4.793750929485149,
+      "tokens_seen": 101056512
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009791574724172518,
+      "loss": 2.3256,
+      "theoretical_loss": 4.793323228594385,
+      "tokens_seen": 101122048
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009791374122367102,
+      "loss": 2.087,
+      "theoretical_loss": 4.792895882357019,
+      "tokens_seen": 101187584
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009791173520561686,
+      "loss": 2.0553,
+      "theoretical_loss": 4.792468890249557,
+      "tokens_seen": 101253120
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000979097291875627,
+      "loss": 2.1112,
+      "theoretical_loss": 4.792042251749617,
+      "tokens_seen": 101318656
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009790772316950853,
+      "loss": 2.1636,
+      "theoretical_loss": 4.791615966335922,
+      "tokens_seen": 101384192
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009790571715145436,
+      "loss": 2.1967,
+      "theoretical_loss": 4.791190033488302,
+      "tokens_seen": 101449728
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000979037111334002,
+      "loss": 2.1536,
+      "theoretical_loss": 4.790764452687689,
+      "tokens_seen": 101515264
+    },
+    {
+      "epoch": 0.03,
+      "objective/train/advantage_avg": 0.49196192622184753,
+      "objective/train/docs_used": 44310,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 2.0242908000946045,
+      "objective/train/original_loss": 2.0242908000946045,
+      "objective/train/theoretical_loss": 4.790339223416113,
+      "objective/train/tokens_used": 122040800,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2426905632019043,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504299402236938,
+      "objective/train/weighted_lm_loss": 2.1261332035064697,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.790339223416113,
+      "tokens_seen": 101580800
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009790170511534604,
+      "loss": 2.0924,
+      "theoretical_loss": 4.790339223416113,
+      "tokens_seen": 101580800
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009789969909729187,
+      "loss": 2.1643,
+      "theoretical_loss": 4.7899143451566975,
+      "tokens_seen": 101646336
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000978976930792377,
+      "loss": 2.1168,
+      "theoretical_loss": 4.7894898173936635,
+      "tokens_seen": 101711872
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009789568706118357,
+      "loss": 2.0482,
+      "theoretical_loss": 4.789065639612316,
+      "tokens_seen": 101777408
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009789368104312938,
+      "loss": 2.17,
+      "theoretical_loss": 4.78864181129905,
+      "tokens_seen": 101842944
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009789167502507524,
+      "loss": 2.0711,
+      "theoretical_loss": 4.788218331941344,
+      "tokens_seen": 101908480
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009788966900702105,
+      "loss": 2.0941,
+      "theoretical_loss": 4.787795201027757,
+      "tokens_seen": 101974016
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009788766298896691,
+      "loss": 2.149,
+      "theoretical_loss": 4.787372418047926,
+      "tokens_seen": 102039552
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009788565697091275,
+      "loss": 2.2792,
+      "theoretical_loss": 4.786949982492559,
+      "tokens_seen": 102105088
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009788365095285858,
+      "loss": 2.1955,
+      "theoretical_loss": 4.786527893853441,
+      "tokens_seen": 102170624
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009788164493480442,
+      "loss": 2.046,
+      "theoretical_loss": 4.786106151623423,
+      "tokens_seen": 102236160
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009787963891675026,
+      "loss": 2.0414,
+      "theoretical_loss": 4.785684755296422,
+      "tokens_seen": 102301696
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000978776328986961,
+      "loss": 2.0363,
+      "theoretical_loss": 4.785263704367418,
+      "tokens_seen": 102367232
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009787562688064193,
+      "loss": 2.0321,
+      "theoretical_loss": 4.7848429983324525,
+      "tokens_seen": 102432768
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009787362086258777,
+      "loss": 2.1357,
+      "theoretical_loss": 4.784422636688622,
+      "tokens_seen": 102498304
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000978716148445336,
+      "loss": 2.168,
+      "theoretical_loss": 4.784002618934078,
+      "tokens_seen": 102563840
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009786960882647944,
+      "loss": 2.0982,
+      "theoretical_loss": 4.783582944568026,
+      "tokens_seen": 102629376
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009786760280842527,
+      "loss": 2.145,
+      "theoretical_loss": 4.783163613090714,
+      "tokens_seen": 102694912
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000978655967903711,
+      "loss": 2.0991,
+      "theoretical_loss": 4.782744624003442,
+      "tokens_seen": 102760448
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009786359077231695,
+      "loss": 2.2595,
+      "theoretical_loss": 4.782325976808549,
+      "tokens_seen": 102825984
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009786158475426278,
+      "loss": 2.2138,
+      "theoretical_loss": 4.7819076710094155,
+      "tokens_seen": 102891520
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009785957873620864,
+      "loss": 2.1458,
+      "theoretical_loss": 4.781489706110458,
+      "tokens_seen": 102957056
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009785757271815446,
+      "loss": 2.0066,
+      "theoretical_loss": 4.781072081617127,
+      "tokens_seen": 103022592
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009785556670010031,
+      "loss": 2.1166,
+      "theoretical_loss": 4.780654797035907,
+      "tokens_seen": 103088128
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009785356068204615,
+      "loss": 2.1703,
+      "theoretical_loss": 4.780237851874308,
+      "tokens_seen": 103153664
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009785155466399199,
+      "loss": 2.1396,
+      "theoretical_loss": 4.779821245640869,
+      "tokens_seen": 103219200
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009784954864593782,
+      "loss": 2.1718,
+      "theoretical_loss": 4.779404977845148,
+      "tokens_seen": 103284736
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009784754262788366,
+      "loss": 2.1923,
+      "theoretical_loss": 4.778989047997728,
+      "tokens_seen": 103350272
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000978455366098295,
+      "loss": 2.0623,
+      "theoretical_loss": 4.778573455610207,
+      "tokens_seen": 103415808
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009784353059177533,
+      "loss": 2.1395,
+      "theoretical_loss": 4.778158200195196,
+      "tokens_seen": 103481344
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009784152457372117,
+      "loss": 2.007,
+      "theoretical_loss": 4.777743281266321,
+      "tokens_seen": 103546880
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00097839518555667,
+      "loss": 2.0959,
+      "theoretical_loss": 4.777328698338218,
+      "tokens_seen": 103612416
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009783751253761284,
+      "loss": 2.146,
+      "theoretical_loss": 4.776914450926526,
+      "tokens_seen": 103677952
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009783550651955868,
+      "loss": 2.1683,
+      "theoretical_loss": 4.776500538547891,
+      "tokens_seen": 103743488
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009783350050150451,
+      "loss": 2.1065,
+      "theoretical_loss": 4.776086960719956,
+      "tokens_seen": 103809024
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009783149448345035,
+      "loss": 2.0547,
+      "theoretical_loss": 4.775673716961369,
+      "tokens_seen": 103874560
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009782948846539618,
+      "loss": 2.2282,
+      "theoretical_loss": 4.775260806791767,
+      "tokens_seen": 103940096
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009782748244734202,
+      "loss": 2.2749,
+      "theoretical_loss": 4.774848229731784,
+      "tokens_seen": 104005632
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009782547642928786,
+      "loss": 2.2745,
+      "theoretical_loss": 4.774435985303043,
+      "tokens_seen": 104071168
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009782347041123371,
+      "loss": 2.1922,
+      "theoretical_loss": 4.774024073028156,
+      "tokens_seen": 104136704
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009782146439317953,
+      "loss": 1.9764,
+      "theoretical_loss": 4.7736124924307175,
+      "tokens_seen": 104202240
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009781945837512539,
+      "loss": 1.9248,
+      "theoretical_loss": 4.773201243035307,
+      "tokens_seen": 104267776
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009781745235707122,
+      "loss": 2.1946,
+      "theoretical_loss": 4.772790324367482,
+      "tokens_seen": 104333312
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009781544633901706,
+      "loss": 2.1539,
+      "theoretical_loss": 4.772379735953779,
+      "tokens_seen": 104398848
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000978134403209629,
+      "loss": 2.1859,
+      "theoretical_loss": 4.771969477321704,
+      "tokens_seen": 104464384
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009781143430290873,
+      "loss": 2.0899,
+      "theoretical_loss": 4.771559547999743,
+      "tokens_seen": 104529920
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009780942828485457,
+      "loss": 2.1757,
+      "theoretical_loss": 4.771149947517346,
+      "tokens_seen": 104595456
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000978074222668004,
+      "loss": 2.159,
+      "theoretical_loss": 4.770740675404928,
+      "tokens_seen": 104660992
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009780541624874624,
+      "loss": 2.044,
+      "theoretical_loss": 4.770331731193872,
+      "tokens_seen": 104726528
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009780341023069208,
+      "loss": 2.0043,
+      "theoretical_loss": 4.769923114416522,
+      "tokens_seen": 104792064
+    },
+    {
+      "epoch": 0.03,
+      "objective/train/advantage_avg": 0.49358728528022766,
+      "objective/train/docs_used": 45444,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 2.1862094402313232,
+      "objective/train/original_loss": 2.1862096786499023,
+      "objective/train/theoretical_loss": 4.769514824606178,
+      "objective/train/tokens_used": 125317600,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2443910837173462,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506013631820679,
+      "objective/train/weighted_lm_loss": 2.2968454360961914,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.769514824606178,
+      "tokens_seen": 104857600
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009780140421263791,
+      "loss": 2.1955,
+      "theoretical_loss": 4.769514824606178,
+      "tokens_seen": 104857600
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009779939819458375,
+      "loss": 2.0942,
+      "theoretical_loss": 4.7691068612971,
+      "tokens_seen": 104923136
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009779739217652959,
+      "loss": 1.9648,
+      "theoretical_loss": 4.768699224024498,
+      "tokens_seen": 104988672
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009779538615847542,
+      "loss": 2.12,
+      "theoretical_loss": 4.768291912324538,
+      "tokens_seen": 105054208
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009779338014042126,
+      "loss": 2.025,
+      "theoretical_loss": 4.7678849257343305,
+      "tokens_seen": 105119744
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000977913741223671,
+      "loss": 2.1114,
+      "theoretical_loss": 4.767478263791933,
+      "tokens_seen": 105185280
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009778936810431293,
+      "loss": 2.0773,
+      "theoretical_loss": 4.7670719260363486,
+      "tokens_seen": 105250816
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009778736208625879,
+      "loss": 2.0793,
+      "theoretical_loss": 4.76666591200752,
+      "tokens_seen": 105316352
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000977853560682046,
+      "loss": 2.0647,
+      "theoretical_loss": 4.766260221246329,
+      "tokens_seen": 105381888
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009778335005015046,
+      "loss": 2.0955,
+      "theoretical_loss": 4.765854853294592,
+      "tokens_seen": 105447424
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000977813440320963,
+      "loss": 2.0974,
+      "theoretical_loss": 4.765449807695062,
+      "tokens_seen": 105512960
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009777933801404213,
+      "loss": 2.1978,
+      "theoretical_loss": 4.765045083991422,
+      "tokens_seen": 105578496
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009777733199598797,
+      "loss": 2.0725,
+      "theoretical_loss": 4.764640681728281,
+      "tokens_seen": 105644032
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000977753259779338,
+      "loss": 2.3367,
+      "theoretical_loss": 4.764236600451178,
+      "tokens_seen": 105709568
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009777331995987964,
+      "loss": 2.19,
+      "theoretical_loss": 4.763832839706574,
+      "tokens_seen": 105775104
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009777131394182548,
+      "loss": 2.0348,
+      "theoretical_loss": 4.763429399041852,
+      "tokens_seen": 105840640
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009776930792377131,
+      "loss": 2.025,
+      "theoretical_loss": 4.763026278005314,
+      "tokens_seen": 105906176
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009776730190571715,
+      "loss": 2.2138,
+      "theoretical_loss": 4.762623476146176,
+      "tokens_seen": 105971712
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009776529588766299,
+      "loss": 2.0778,
+      "theoretical_loss": 4.762220993014572,
+      "tokens_seen": 106037248
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009776328986960882,
+      "loss": 2.0225,
+      "theoretical_loss": 4.761818828161545,
+      "tokens_seen": 106102784
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009776128385155466,
+      "loss": 2.1458,
+      "theoretical_loss": 4.761416981139046,
+      "tokens_seen": 106168320
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000977592778335005,
+      "loss": 2.018,
+      "theoretical_loss": 4.7610154514999365,
+      "tokens_seen": 106233856
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009775727181544633,
+      "loss": 2.1109,
+      "theoretical_loss": 4.760614238797981,
+      "tokens_seen": 106299392
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000977552657973922,
+      "loss": 2.1624,
+      "theoretical_loss": 4.760213342587843,
+      "tokens_seen": 106364928
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00097753259779338,
+      "loss": 2.1445,
+      "theoretical_loss": 4.75981276242509,
+      "tokens_seen": 106430464
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009775125376128386,
+      "loss": 2.0961,
+      "theoretical_loss": 4.759412497866186,
+      "tokens_seen": 106496000
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009774924774322968,
+      "loss": 2.1643,
+      "theoretical_loss": 4.759012548468487,
+      "tokens_seen": 106561536
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009774724172517553,
+      "loss": 2.0578,
+      "theoretical_loss": 4.758612913790243,
+      "tokens_seen": 106627072
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009774523570712137,
+      "loss": 2.0945,
+      "theoretical_loss": 4.758213593390595,
+      "tokens_seen": 106692608
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000977432296890672,
+      "loss": 2.0254,
+      "theoretical_loss": 4.757814586829572,
+      "tokens_seen": 106758144
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009774122367101304,
+      "loss": 2.1265,
+      "theoretical_loss": 4.757415893668089,
+      "tokens_seen": 106823680
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009773921765295888,
+      "loss": 2.1485,
+      "theoretical_loss": 4.75701751346794,
+      "tokens_seen": 106889216
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009773721163490472,
+      "loss": 2.0516,
+      "theoretical_loss": 4.756619445791808,
+      "tokens_seen": 106954752
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009773520561685055,
+      "loss": 2.1517,
+      "theoretical_loss": 4.756221690203244,
+      "tokens_seen": 107020288
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009773319959879639,
+      "loss": 2.2329,
+      "theoretical_loss": 4.755824246266685,
+      "tokens_seen": 107085824
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009773119358074222,
+      "loss": 2.1316,
+      "theoretical_loss": 4.755427113547437,
+      "tokens_seen": 107151360
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009772918756268806,
+      "loss": 2.0784,
+      "theoretical_loss": 4.755030291611678,
+      "tokens_seen": 107216896
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000977271815446339,
+      "loss": 2.1254,
+      "theoretical_loss": 4.754633780026456,
+      "tokens_seen": 107282432
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009772517552657975,
+      "loss": 2.122,
+      "theoretical_loss": 4.754237578359687,
+      "tokens_seen": 107347968
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009772316950852557,
+      "loss": 2.093,
+      "theoretical_loss": 4.753841686180152,
+      "tokens_seen": 107413504
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009772116349047143,
+      "loss": 2.1048,
+      "theoretical_loss": 4.753446103057492,
+      "tokens_seen": 107479040
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009771915747241726,
+      "loss": 2.1523,
+      "theoretical_loss": 4.753050828562211,
+      "tokens_seen": 107544576
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000977171514543631,
+      "loss": 2.1161,
+      "theoretical_loss": 4.752655862265669,
+      "tokens_seen": 107610112
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009771514543630894,
+      "loss": 2.0962,
+      "theoretical_loss": 4.752261203740085,
+      "tokens_seen": 107675648
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009771313941825477,
+      "loss": 2.1709,
+      "theoretical_loss": 4.751866852558529,
+      "tokens_seen": 107741184
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000977111334002006,
+      "loss": 2.0602,
+      "theoretical_loss": 4.751472808294924,
+      "tokens_seen": 107806720
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009770912738214644,
+      "loss": 2.2369,
+      "theoretical_loss": 4.751079070524041,
+      "tokens_seen": 107872256
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009770712136409228,
+      "loss": 2.185,
+      "theoretical_loss": 4.750685638821498,
+      "tokens_seen": 107937792
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009770511534603812,
+      "loss": 2.0072,
+      "theoretical_loss": 4.7502925127637585,
+      "tokens_seen": 108003328
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009770310932798395,
+      "loss": 2.132,
+      "theoretical_loss": 4.749899691928128,
+      "tokens_seen": 108068864
+    },
+    {
+      "epoch": 0.03,
+      "objective/train/advantage_avg": 0.49381500482559204,
+      "objective/train/docs_used": 46682,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 2.180783271789551,
+      "objective/train/original_loss": 2.180783748626709,
+      "objective/train/theoretical_loss": 4.749507175892756,
+      "objective/train/tokens_used": 128594400,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24419686198234558,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506229400634766,
+      "objective/train/weighted_lm_loss": 2.291189432144165,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.749507175892756,
+      "tokens_seen": 108134400
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000977011033099298,
+      "loss": 2.1949,
+      "theoretical_loss": 4.749507175892756,
+      "tokens_seen": 108134400
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009769909729187563,
+      "loss": 2.1011,
+      "theoretical_loss": 4.7491149642366235,
+      "tokens_seen": 108199936
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009769709127382146,
+      "loss": 2.0941,
+      "theoretical_loss": 4.7487230565395535,
+      "tokens_seen": 108265472
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000976950852557673,
+      "loss": 2.1314,
+      "theoretical_loss": 4.7483314523822004,
+      "tokens_seen": 108331008
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009769307923771313,
+      "loss": 2.2006,
+      "theoretical_loss": 4.74794015134605,
+      "tokens_seen": 108396544
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009769107321965897,
+      "loss": 2.0744,
+      "theoretical_loss": 4.74754915301342,
+      "tokens_seen": 108462080
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009768906720160483,
+      "loss": 2.047,
+      "theoretical_loss": 4.747158456967452,
+      "tokens_seen": 108527616
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009768706118355064,
+      "loss": 2.1796,
+      "theoretical_loss": 4.746768062792116,
+      "tokens_seen": 108593152
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000976850551654965,
+      "loss": 2.0949,
+      "theoretical_loss": 4.746377970072205,
+      "tokens_seen": 108658688
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009768304914744234,
+      "loss": 2.1516,
+      "theoretical_loss": 4.745988178393331,
+      "tokens_seen": 108724224
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009768104312938817,
+      "loss": 2.1688,
+      "theoretical_loss": 4.745598687341927,
+      "tokens_seen": 108789760
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00097679037111334,
+      "loss": 2.1327,
+      "theoretical_loss": 4.745209496505241,
+      "tokens_seen": 108855296
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009767703109327985,
+      "loss": 2.1034,
+      "theoretical_loss": 4.744820605471338,
+      "tokens_seen": 108920832
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009767502507522568,
+      "loss": 2.1114,
+      "theoretical_loss": 4.744432013829094,
+      "tokens_seen": 108986368
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009767301905717152,
+      "loss": 2.1984,
+      "theoretical_loss": 4.744043721168196,
+      "tokens_seen": 109051904
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009767101303911735,
+      "loss": 2.0895,
+      "theoretical_loss": 4.7436557270791395,
+      "tokens_seen": 109117440
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000976690070210632,
+      "loss": 2.0292,
+      "theoretical_loss": 4.743268031153226,
+      "tokens_seen": 109182976
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009766700100300903,
+      "loss": 2.0267,
+      "theoretical_loss": 4.742880632982564,
+      "tokens_seen": 109248512
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009766499498495486,
+      "loss": 2.1127,
+      "theoretical_loss": 4.74249353216006,
+      "tokens_seen": 109314048
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000976629889669007,
+      "loss": 1.9842,
+      "theoretical_loss": 4.742106728279422,
+      "tokens_seen": 109379584
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009766098294884654,
+      "loss": 2.1229,
+      "theoretical_loss": 4.741720220935159,
+      "tokens_seen": 109445120
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009765897693079237,
+      "loss": 2.0699,
+      "theoretical_loss": 4.741334009722573,
+      "tokens_seen": 109510656
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009765697091273823,
+      "loss": 2.1166,
+      "theoretical_loss": 4.740948094237761,
+      "tokens_seen": 109576192
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009765496489468406,
+      "loss": 2.1423,
+      "theoretical_loss": 4.7405624740776116,
+      "tokens_seen": 109641728
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009765295887662989,
+      "loss": 2.1827,
+      "theoretical_loss": 4.7401771488398055,
+      "tokens_seen": 109707264
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009765095285857573,
+      "loss": 2.1094,
+      "theoretical_loss": 4.739792118122809,
+      "tokens_seen": 109772800
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009764894684052156,
+      "loss": 2.0291,
+      "theoretical_loss": 4.739407381525874,
+      "tokens_seen": 109838336
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000976469408224674,
+      "loss": 2.1474,
+      "theoretical_loss": 4.73902293864904,
+      "tokens_seen": 109903872
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009764493480441325,
+      "loss": 2.1552,
+      "theoretical_loss": 4.738638789093124,
+      "tokens_seen": 109969408
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009764292878635907,
+      "loss": 2.0372,
+      "theoretical_loss": 4.738254932459727,
+      "tokens_seen": 110034944
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009764092276830492,
+      "loss": 2.1613,
+      "theoretical_loss": 4.7378713683512235,
+      "tokens_seen": 110100480
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009763891675025074,
+      "loss": 2.0998,
+      "theoretical_loss": 4.737488096370768,
+      "tokens_seen": 110166016
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009763691073219659,
+      "loss": 2.0198,
+      "theoretical_loss": 4.737105116122288,
+      "tokens_seen": 110231552
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009763490471414243,
+      "loss": 2.1154,
+      "theoretical_loss": 4.736722427210483,
+      "tokens_seen": 110297088
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009763289869608826,
+      "loss": 2.2146,
+      "theoretical_loss": 4.7363400292408215,
+      "tokens_seen": 110362624
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000976308926780341,
+      "loss": 2.1661,
+      "theoretical_loss": 4.735957921819542,
+      "tokens_seen": 110428160
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009762888665997995,
+      "loss": 2.1288,
+      "theoretical_loss": 4.735576104553646,
+      "tokens_seen": 110493696
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009762688064192577,
+      "loss": 2.0877,
+      "theoretical_loss": 4.735194577050903,
+      "tokens_seen": 110559232
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009762487462387162,
+      "loss": 2.2459,
+      "theoretical_loss": 4.734813338919842,
+      "tokens_seen": 110624768
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009762286860581745,
+      "loss": 2.0061,
+      "theoretical_loss": 4.734432389769756,
+      "tokens_seen": 110690304
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009762086258776329,
+      "loss": 2.0924,
+      "theoretical_loss": 4.734051729210693,
+      "tokens_seen": 110755840
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009761885656970913,
+      "loss": 2.0756,
+      "theoretical_loss": 4.733671356853456,
+      "tokens_seen": 110821376
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009761685055165496,
+      "loss": 2.111,
+      "theoretical_loss": 4.733291272309609,
+      "tokens_seen": 110886912
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000976148445336008,
+      "loss": 2.1974,
+      "theoretical_loss": 4.732911475191464,
+      "tokens_seen": 110952448
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009761283851554664,
+      "loss": 1.9964,
+      "theoretical_loss": 4.732531965112083,
+      "tokens_seen": 111017984
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009761083249749247,
+      "loss": 2.1016,
+      "theoretical_loss": 4.732152741685281,
+      "tokens_seen": 111083520
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009760882647943832,
+      "loss": 2.0433,
+      "theoretical_loss": 4.731773804525616,
+      "tokens_seen": 111149056
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009760682046138415,
+      "loss": 2.0489,
+      "theoretical_loss": 4.731395153248394,
+      "tokens_seen": 111214592
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009760481444332999,
+      "loss": 1.9634,
+      "theoretical_loss": 4.731016787469663,
+      "tokens_seen": 111280128
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009760280842527583,
+      "loss": 2.1297,
+      "theoretical_loss": 4.730638706806213,
+      "tokens_seen": 111345664
+    },
+    {
+      "epoch": 0.03,
+      "objective/train/advantage_avg": 0.49222612380981445,
+      "objective/train/docs_used": 47788,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.9619340896606445,
+      "objective/train/original_loss": 1.961933970451355,
+      "objective/train/theoretical_loss": 4.73026091087557,
+      "objective/train/tokens_used": 131871200,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24286691844463348,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050457239151001,
+      "objective/train/weighted_lm_loss": 2.060546636581421,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.73026091087557,
+      "tokens_seen": 111411200
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009760080240722167,
+      "loss": 2.0969,
+      "theoretical_loss": 4.73026091087557,
+      "tokens_seen": 111411200
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000975987963891675,
+      "loss": 2.1887,
+      "theoretical_loss": 4.729883399296005,
+      "tokens_seen": 111476736
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009759679037111334,
+      "loss": 2.049,
+      "theoretical_loss": 4.729506171686518,
+      "tokens_seen": 111542272
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009759478435305919,
+      "loss": 2.0852,
+      "theoretical_loss": 4.729129227666847,
+      "tokens_seen": 111607808
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009759277833500502,
+      "loss": 2.1017,
+      "theoretical_loss": 4.728752566857459,
+      "tokens_seen": 111673344
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009759077231695086,
+      "loss": 2.1405,
+      "theoretical_loss": 4.728376188879554,
+      "tokens_seen": 111738880
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009758876629889669,
+      "loss": 2.1897,
+      "theoretical_loss": 4.7280000933550586,
+      "tokens_seen": 111804416
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009758676028084253,
+      "loss": 2.1838,
+      "theoretical_loss": 4.727624279906628,
+      "tokens_seen": 111869952
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009758475426278837,
+      "loss": 2.047,
+      "theoretical_loss": 4.72724874815764,
+      "tokens_seen": 111935488
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009758274824473421,
+      "loss": 2.0746,
+      "theoretical_loss": 4.726873497732196,
+      "tokens_seen": 112001024
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009758074222668004,
+      "loss": 2.2697,
+      "theoretical_loss": 4.726498528255121,
+      "tokens_seen": 112066560
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009757873620862589,
+      "loss": 2.1296,
+      "theoretical_loss": 4.7261238393519545,
+      "tokens_seen": 112132096
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009757673019057171,
+      "loss": 2.1274,
+      "theoretical_loss": 4.725749430648958,
+      "tokens_seen": 112197632
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009757472417251756,
+      "loss": 2.0796,
+      "theoretical_loss": 4.725375301773107,
+      "tokens_seen": 112263168
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009757271815446339,
+      "loss": 2.0802,
+      "theoretical_loss": 4.725001452352091,
+      "tokens_seen": 112328704
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009757071213640923,
+      "loss": 2.1455,
+      "theoretical_loss": 4.724627882014311,
+      "tokens_seen": 112394240
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009756870611835507,
+      "loss": 2.0957,
+      "theoretical_loss": 4.724254590388881,
+      "tokens_seen": 112459776
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009756670010030091,
+      "loss": 1.9454,
+      "theoretical_loss": 4.723881577105621,
+      "tokens_seen": 112525312
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009756469408224674,
+      "loss": 2.157,
+      "theoretical_loss": 4.723508841795058,
+      "tokens_seen": 112590848
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009756268806419259,
+      "loss": 2.1352,
+      "theoretical_loss": 4.723136384088427,
+      "tokens_seen": 112656384
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009756068204613841,
+      "loss": 2.078,
+      "theoretical_loss": 4.722764203617663,
+      "tokens_seen": 112721920
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009755867602808426,
+      "loss": 2.1686,
+      "theoretical_loss": 4.722392300015404,
+      "tokens_seen": 112787456
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000975566700100301,
+      "loss": 2.1921,
+      "theoretical_loss": 4.7220206729149865,
+      "tokens_seen": 112852992
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009755466399197593,
+      "loss": 2.0499,
+      "theoretical_loss": 4.72164932195045,
+      "tokens_seen": 112918528
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009755265797392177,
+      "loss": 2.1789,
+      "theoretical_loss": 4.7212782467565235,
+      "tokens_seen": 112984064
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000975506519558676,
+      "loss": 2.0737,
+      "theoretical_loss": 4.720907446968635,
+      "tokens_seen": 113049600
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009754864593781344,
+      "loss": 2.0673,
+      "theoretical_loss": 4.720536922222907,
+      "tokens_seen": 113115136
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009754663991975929,
+      "loss": 2.1072,
+      "theoretical_loss": 4.720166672156147,
+      "tokens_seen": 113180672
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009754463390170511,
+      "loss": 2.0866,
+      "theoretical_loss": 4.719796696405858,
+      "tokens_seen": 113246208
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009754262788365096,
+      "loss": 2.1345,
+      "theoretical_loss": 4.719426994610227,
+      "tokens_seen": 113311744
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009754062186559678,
+      "loss": 2.0008,
+      "theoretical_loss": 4.719057566408132,
+      "tokens_seen": 113377280
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009753861584754263,
+      "loss": 2.0752,
+      "theoretical_loss": 4.718688411439129,
+      "tokens_seen": 113442816
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009753660982948847,
+      "loss": 2.0776,
+      "theoretical_loss": 4.718319529343462,
+      "tokens_seen": 113508352
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.000975346038114343,
+      "loss": 2.1316,
+      "theoretical_loss": 4.7179509197620515,
+      "tokens_seen": 113573888
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009753259779338014,
+      "loss": 2.082,
+      "theoretical_loss": 4.717582582336503,
+      "tokens_seen": 113639424
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009753059177532599,
+      "loss": 1.9908,
+      "theoretical_loss": 4.7172145167090935,
+      "tokens_seen": 113704960
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009752858575727181,
+      "loss": 2.1448,
+      "theoretical_loss": 4.716846722522781,
+      "tokens_seen": 113770496
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009752657973921766,
+      "loss": 2.0554,
+      "theoretical_loss": 4.716479199421195,
+      "tokens_seen": 113836032
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009752457372116349,
+      "loss": 1.9584,
+      "theoretical_loss": 4.716111947048638,
+      "tokens_seen": 113901568
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009752256770310933,
+      "loss": 2.0346,
+      "theoretical_loss": 4.715744965050087,
+      "tokens_seen": 113967104
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009752056168505517,
+      "loss": 2.1214,
+      "theoretical_loss": 4.715378253071181,
+      "tokens_seen": 114032640
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00097518555667001,
+      "loss": 2.01,
+      "theoretical_loss": 4.715011810758236,
+      "tokens_seen": 114098176
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009751654964894684,
+      "loss": 2.0735,
+      "theoretical_loss": 4.714645637758226,
+      "tokens_seen": 114163712
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009751454363089268,
+      "loss": 2.0352,
+      "theoretical_loss": 4.714279733718793,
+      "tokens_seen": 114229248
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009751253761283851,
+      "loss": 2.1201,
+      "theoretical_loss": 4.713914098288242,
+      "tokens_seen": 114294784
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009751053159478436,
+      "loss": 2.0151,
+      "theoretical_loss": 4.71354873111554,
+      "tokens_seen": 114360320
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009750852557673019,
+      "loss": 1.9781,
+      "theoretical_loss": 4.7131836318503115,
+      "tokens_seen": 114425856
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009750651955867603,
+      "loss": 2.0794,
+      "theoretical_loss": 4.7128188001428395,
+      "tokens_seen": 114491392
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009750451354062187,
+      "loss": 2.075,
+      "theoretical_loss": 4.712454235644064,
+      "tokens_seen": 114556928
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009750250752256771,
+      "loss": 2.1529,
+      "theoretical_loss": 4.71208993800558,
+      "tokens_seen": 114622464
+    },
+    {
+      "epoch": 0.03,
+      "objective/train/advantage_avg": 0.49528729915618896,
+      "objective/train/docs_used": 48994,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.978676676750183,
+      "objective/train/original_loss": 1.9786767959594727,
+      "objective/train/theoretical_loss": 4.711725906879634,
+      "objective/train/tokens_used": 135148000,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24554309248924255,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0507770776748657,
+      "objective/train/weighted_lm_loss": 2.0792107582092285,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.711725906879634,
+      "tokens_seen": 114688000
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009750050150451354,
+      "loss": 2.0281,
+      "theoretical_loss": 4.711725906879634,
+      "tokens_seen": 114688000
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009749849548645938,
+      "loss": 2.0641,
+      "theoretical_loss": 4.7113621419191265,
+      "tokens_seen": 114753536
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009749648946840521,
+      "loss": 2.1124,
+      "theoretical_loss": 4.710998642777606,
+      "tokens_seen": 114819072
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009749448345035106,
+      "loss": 2.0097,
+      "theoretical_loss": 4.710635409109269,
+      "tokens_seen": 114884608
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009749247743229689,
+      "loss": 2.2112,
+      "theoretical_loss": 4.710272440568962,
+      "tokens_seen": 114950144
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009749047141424273,
+      "loss": 2.1056,
+      "theoretical_loss": 4.709909736812172,
+      "tokens_seen": 115015680
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009748846539618856,
+      "loss": 2.2072,
+      "theoretical_loss": 4.709547297495034,
+      "tokens_seen": 115081216
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009748645937813441,
+      "loss": 2.1081,
+      "theoretical_loss": 4.709185122274322,
+      "tokens_seen": 115146752
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009748445336008024,
+      "loss": 2.0524,
+      "theoretical_loss": 4.708823210807452,
+      "tokens_seen": 115212288
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009748244734202608,
+      "loss": 1.8597,
+      "theoretical_loss": 4.708461562752479,
+      "tokens_seen": 115277824
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009748044132397191,
+      "loss": 2.0625,
+      "theoretical_loss": 4.708100177768094,
+      "tokens_seen": 115343360
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009747843530591775,
+      "loss": 2.0811,
+      "theoretical_loss": 4.707739055513627,
+      "tokens_seen": 115408896
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0009747642928786359,
+      "loss": 2.183,
+      "theoretical_loss": 4.707378195649038,
+      "tokens_seen": 115474432
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009747442326980943,
+      "loss": 1.9737,
+      "theoretical_loss": 4.7070175978349225,
+      "tokens_seen": 115539968
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009747241725175526,
+      "loss": 2.1049,
+      "theoretical_loss": 4.7066572617325075,
+      "tokens_seen": 115605504
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009747041123370111,
+      "loss": 2.0388,
+      "theoretical_loss": 4.706297187003647,
+      "tokens_seen": 115671040
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009746840521564694,
+      "loss": 2.0434,
+      "theoretical_loss": 4.705937373310828,
+      "tokens_seen": 115736576
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009746639919759278,
+      "loss": 2.2422,
+      "theoretical_loss": 4.705577820317158,
+      "tokens_seen": 115802112
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009746439317953862,
+      "loss": 2.1174,
+      "theoretical_loss": 4.705218527686375,
+      "tokens_seen": 115867648
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009746238716148445,
+      "loss": 2.1772,
+      "theoretical_loss": 4.704859495082837,
+      "tokens_seen": 115933184
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009746038114343029,
+      "loss": 2.1444,
+      "theoretical_loss": 4.704500722171526,
+      "tokens_seen": 115998720
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009745837512537614,
+      "loss": 2.2104,
+      "theoretical_loss": 4.704142208618045,
+      "tokens_seen": 116064256
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009745636910732196,
+      "loss": 2.0155,
+      "theoretical_loss": 4.703783954088612,
+      "tokens_seen": 116129792
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009745436308926781,
+      "loss": 2.3143,
+      "theoretical_loss": 4.703425958250071,
+      "tokens_seen": 116195328
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009745235707121363,
+      "loss": 2.1153,
+      "theoretical_loss": 4.703068220769872,
+      "tokens_seen": 116260864
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009745035105315948,
+      "loss": 2.0486,
+      "theoretical_loss": 4.702710741316087,
+      "tokens_seen": 116326400
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009744834503510532,
+      "loss": 2.1845,
+      "theoretical_loss": 4.702353519557398,
+      "tokens_seen": 116391936
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009744633901705115,
+      "loss": 2.2034,
+      "theoretical_loss": 4.701996555163101,
+      "tokens_seen": 116457472
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00097444332998997,
+      "loss": 2.0476,
+      "theoretical_loss": 4.701639847803098,
+      "tokens_seen": 116523008
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009744232698094282,
+      "loss": 2.1367,
+      "theoretical_loss": 4.701283397147904,
+      "tokens_seen": 116588544
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009744032096288867,
+      "loss": 2.0433,
+      "theoretical_loss": 4.700927202868639,
+      "tokens_seen": 116654080
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009743831494483451,
+      "loss": 2.0795,
+      "theoretical_loss": 4.700571264637029,
+      "tokens_seen": 116719616
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009743630892678034,
+      "loss": 1.985,
+      "theoretical_loss": 4.700215582125406,
+      "tokens_seen": 116785152
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009743430290872618,
+      "loss": 2.0779,
+      "theoretical_loss": 4.699860155006702,
+      "tokens_seen": 116850688
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009743229689067203,
+      "loss": 2.0759,
+      "theoretical_loss": 4.699504982954452,
+      "tokens_seen": 116916224
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009743029087261785,
+      "loss": 2.1299,
+      "theoretical_loss": 4.699150065642794,
+      "tokens_seen": 116981760
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000974282848545637,
+      "loss": 1.9488,
+      "theoretical_loss": 4.698795402746457,
+      "tokens_seen": 117047296
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009742627883650953,
+      "loss": 2.0521,
+      "theoretical_loss": 4.6984409939407765,
+      "tokens_seen": 117112832
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009742427281845537,
+      "loss": 2.3108,
+      "theoretical_loss": 4.698086838901676,
+      "tokens_seen": 117178368
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009742226680040121,
+      "loss": 2.0137,
+      "theoretical_loss": 4.697732937305678,
+      "tokens_seen": 117243904
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009742026078234705,
+      "loss": 2.1527,
+      "theoretical_loss": 4.697379288829895,
+      "tokens_seen": 117309440
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009741825476429288,
+      "loss": 1.9996,
+      "theoretical_loss": 4.697025893152032,
+      "tokens_seen": 117374976
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009741624874623872,
+      "loss": 2.1059,
+      "theoretical_loss": 4.696672749950385,
+      "tokens_seen": 117440512
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009741424272818455,
+      "loss": 1.938,
+      "theoretical_loss": 4.696319858903838,
+      "tokens_seen": 117506048
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000974122367101304,
+      "loss": 2.1274,
+      "theoretical_loss": 4.69596721969186,
+      "tokens_seen": 117571584
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009741023069207623,
+      "loss": 2.1286,
+      "theoretical_loss": 4.69561483199451,
+      "tokens_seen": 117637120
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009740822467402207,
+      "loss": 2.0024,
+      "theoretical_loss": 4.695262695492428,
+      "tokens_seen": 117702656
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009740621865596791,
+      "loss": 1.99,
+      "theoretical_loss": 4.694910809866839,
+      "tokens_seen": 117768192
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009740421263791375,
+      "loss": 2.1947,
+      "theoretical_loss": 4.694559174799549,
+      "tokens_seen": 117833728
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009740220661985958,
+      "loss": 2.0581,
+      "theoretical_loss": 4.694207789972943,
+      "tokens_seen": 117899264
+    },
+    {
+      "epoch": 0.04,
+      "objective/train/advantage_avg": 0.4920607805252075,
+      "objective/train/docs_used": 50256,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 2.054048776626587,
+      "objective/train/original_loss": 2.054048776626587,
+      "objective/train/theoretical_loss": 4.693856655069986,
+      "objective/train/tokens_used": 138424800,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2430429458618164,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504416227340698,
+      "objective/train/weighted_lm_loss": 2.157548666000366,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9048373699188232,
+      "theoretical_loss": 4.693856655069986,
+      "tokens_seen": 117964800
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009740020060180542,
+      "loss": 2.1324,
+      "theoretical_loss": 4.693856655069986,
+      "tokens_seen": 117964800
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009739819458375125,
+      "loss": 2.1721,
+      "theoretical_loss": 4.693505769774223,
+      "tokens_seen": 118030336
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000973961885656971,
+      "loss": 2.1294,
+      "theoretical_loss": 4.6931551337697694,
+      "tokens_seen": 118095872
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009739418254764293,
+      "loss": 2.0215,
+      "theoretical_loss": 4.692804746741322,
+      "tokens_seen": 118161408
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009739217652958877,
+      "loss": 2.1654,
+      "theoretical_loss": 4.692454608374145,
+      "tokens_seen": 118226944
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000973901705115346,
+      "loss": 2.1852,
+      "theoretical_loss": 4.692104718354077,
+      "tokens_seen": 118292480
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009738816449348045,
+      "loss": 1.9563,
+      "theoretical_loss": 4.69175507636753,
+      "tokens_seen": 118358016
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009738615847542628,
+      "loss": 2.0985,
+      "theoretical_loss": 4.691405682101482,
+      "tokens_seen": 118423552
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009738415245737212,
+      "loss": 2.0069,
+      "theoretical_loss": 4.6910565352434785,
+      "tokens_seen": 118489088
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009738214643931796,
+      "loss": 2.1439,
+      "theoretical_loss": 4.690707635481634,
+      "tokens_seen": 118554624
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009738014042126379,
+      "loss": 2.0671,
+      "theoretical_loss": 4.690358982504628,
+      "tokens_seen": 118620160
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009737813440320963,
+      "loss": 1.9935,
+      "theoretical_loss": 4.690010576001702,
+      "tokens_seen": 118685696
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009737612838515547,
+      "loss": 2.0938,
+      "theoretical_loss": 4.689662415662662,
+      "tokens_seen": 118751232
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000973741223671013,
+      "loss": 2.0523,
+      "theoretical_loss": 4.689314501177874,
+      "tokens_seen": 118816768
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009737211634904715,
+      "loss": 2.0498,
+      "theoretical_loss": 4.6889668322382665,
+      "tokens_seen": 118882304
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009737011033099298,
+      "loss": 2.025,
+      "theoretical_loss": 4.688619408535323,
+      "tokens_seen": 118947840
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009736810431293882,
+      "loss": 2.1013,
+      "theoretical_loss": 4.688272229761087,
+      "tokens_seen": 119013376
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009736609829488466,
+      "loss": 2.1606,
+      "theoretical_loss": 4.687925295608157,
+      "tokens_seen": 119078912
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009736409227683049,
+      "loss": 2.1686,
+      "theoretical_loss": 4.687578605769687,
+      "tokens_seen": 119144448
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009736208625877633,
+      "loss": 2.0253,
+      "theoretical_loss": 4.6872321599393825,
+      "tokens_seen": 119209984
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009736008024072218,
+      "loss": 2.0954,
+      "theoretical_loss": 4.686885957811503,
+      "tokens_seen": 119275520
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00097358074222668,
+      "loss": 1.9549,
+      "theoretical_loss": 4.6865399990808605,
+      "tokens_seen": 119341056
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009735606820461385,
+      "loss": 1.9711,
+      "theoretical_loss": 4.686194283442811,
+      "tokens_seen": 119406592
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009735406218655967,
+      "loss": 2.1033,
+      "theoretical_loss": 4.685848810593263,
+      "tokens_seen": 119472128
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009735205616850552,
+      "loss": 1.9815,
+      "theoretical_loss": 4.685503580228671,
+      "tokens_seen": 119537664
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009735005015045136,
+      "loss": 2.1605,
+      "theoretical_loss": 4.685158592046037,
+      "tokens_seen": 119603200
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009734804413239719,
+      "loss": 1.963,
+      "theoretical_loss": 4.684813845742902,
+      "tokens_seen": 119668736
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009734603811434303,
+      "loss": 2.0143,
+      "theoretical_loss": 4.684469341017357,
+      "tokens_seen": 119734272
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009734403209628886,
+      "loss": 2.0491,
+      "theoretical_loss": 4.684125077568028,
+      "tokens_seen": 119799808
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000973420260782347,
+      "loss": 2.2728,
+      "theoretical_loss": 4.683781055094087,
+      "tokens_seen": 119865344
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009734002006018055,
+      "loss": 2.0802,
+      "theoretical_loss": 4.683437273295245,
+      "tokens_seen": 119930880
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009733801404212637,
+      "loss": 2.1258,
+      "theoretical_loss": 4.683093731871746,
+      "tokens_seen": 119996416
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009733600802407222,
+      "loss": 1.933,
+      "theoretical_loss": 4.682750430524376,
+      "tokens_seen": 120061952
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009733400200601806,
+      "loss": 2.0386,
+      "theoretical_loss": 4.682407368954454,
+      "tokens_seen": 120127488
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009733199598796389,
+      "loss": 2.1227,
+      "theoretical_loss": 4.682064546863834,
+      "tokens_seen": 120193024
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009732998996990973,
+      "loss": 2.0572,
+      "theoretical_loss": 4.681721963954903,
+      "tokens_seen": 120258560
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009732798395185557,
+      "loss": 2.2008,
+      "theoretical_loss": 4.6813796199305795,
+      "tokens_seen": 120324096
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000973259779338014,
+      "loss": 2.0648,
+      "theoretical_loss": 4.681037514494313,
+      "tokens_seen": 120389632
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009732397191574725,
+      "loss": 2.1425,
+      "theoretical_loss": 4.680695647350082,
+      "tokens_seen": 120455168
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009732196589769307,
+      "loss": 2.1755,
+      "theoretical_loss": 4.680354018202394,
+      "tokens_seen": 120520704
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009731995987963892,
+      "loss": 2.08,
+      "theoretical_loss": 4.68001262675628,
+      "tokens_seen": 120586240
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009731795386158475,
+      "loss": 1.8476,
+      "theoretical_loss": 4.6796714727173025,
+      "tokens_seen": 120651776
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009731594784353059,
+      "loss": 2.1225,
+      "theoretical_loss": 4.679330555791543,
+      "tokens_seen": 120717312
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009731394182547643,
+      "loss": 2.1533,
+      "theoretical_loss": 4.678989875685607,
+      "tokens_seen": 120782848
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009731193580742227,
+      "loss": 2.0241,
+      "theoretical_loss": 4.678649432106627,
+      "tokens_seen": 120848384
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000973099297893681,
+      "loss": 1.9959,
+      "theoretical_loss": 4.678309224762249,
+      "tokens_seen": 120913920
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009730792377131395,
+      "loss": 2.239,
+      "theoretical_loss": 4.677969253360644,
+      "tokens_seen": 120979456
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009730591775325977,
+      "loss": 2.1644,
+      "theoretical_loss": 4.677629517610498,
+      "tokens_seen": 121044992
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009730391173520562,
+      "loss": 1.9964,
+      "theoretical_loss": 4.677290017221017,
+      "tokens_seen": 121110528
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009730190571715145,
+      "loss": 2.1193,
+      "theoretical_loss": 4.67695075190192,
+      "tokens_seen": 121176064
+    },
+    {
+      "epoch": 0.04,
+      "objective/train/advantage_avg": 0.4931885600090027,
+      "objective/train/docs_used": 51395,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 2.00555157661438,
+      "objective/train/original_loss": 2.00555157661438,
+      "objective/train/theoretical_loss": 4.676611721363443,
+      "objective/train/tokens_used": 141701600,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24360352754592896,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050557255744934,
+      "objective/train/weighted_lm_loss": 2.1069064140319824,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.676611721363443,
+      "tokens_seen": 121241600
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009729989969909729,
+      "loss": 2.107,
+      "theoretical_loss": 4.676611721363443,
+      "tokens_seen": 121241600
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009729789368104313,
+      "loss": 1.9974,
+      "theoretical_loss": 4.6762729253163355,
+      "tokens_seen": 121307136
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009729588766298897,
+      "loss": 2.0427,
+      "theoretical_loss": 4.675934363471857,
+      "tokens_seen": 121372672
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009729388164493481,
+      "loss": 1.9648,
+      "theoretical_loss": 4.675596035541782,
+      "tokens_seen": 121438208
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009729187562688064,
+      "loss": 2.0273,
+      "theoretical_loss": 4.675257941238391,
+      "tokens_seen": 121503744
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009728986960882649,
+      "loss": 2.041,
+      "theoretical_loss": 4.674920080274475,
+      "tokens_seen": 121569280
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009728786359077232,
+      "loss": 2.0907,
+      "theoretical_loss": 4.674582452363334,
+      "tokens_seen": 121634816
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009728585757271816,
+      "loss": 2.0461,
+      "theoretical_loss": 4.674245057218775,
+      "tokens_seen": 121700352
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00097283851554664,
+      "loss": 2.1216,
+      "theoretical_loss": 4.673907894555105,
+      "tokens_seen": 121765888
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009728184553660983,
+      "loss": 1.9807,
+      "theoretical_loss": 4.673570964087141,
+      "tokens_seen": 121831424
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009727983951855567,
+      "loss": 2.1125,
+      "theoretical_loss": 4.673234265530201,
+      "tokens_seen": 121896960
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009727783350050151,
+      "loss": 1.9752,
+      "theoretical_loss": 4.672897798600104,
+      "tokens_seen": 121962496
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009727582748244734,
+      "loss": 2.0491,
+      "theoretical_loss": 4.672561563013172,
+      "tokens_seen": 122028032
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009727382146439319,
+      "loss": 2.019,
+      "theoretical_loss": 4.672225558486223,
+      "tokens_seen": 122093568
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009727181544633902,
+      "loss": 2.064,
+      "theoretical_loss": 4.671889784736576,
+      "tokens_seen": 122159104
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009726980942828486,
+      "loss": 2.0167,
+      "theoretical_loss": 4.671554241482047,
+      "tokens_seen": 122224640
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000972678034102307,
+      "loss": 2.2564,
+      "theoretical_loss": 4.671218928440947,
+      "tokens_seen": 122290176
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009726579739217653,
+      "loss": 2.0625,
+      "theoretical_loss": 4.670883845332084,
+      "tokens_seen": 122355712
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009726379137412237,
+      "loss": 1.9611,
+      "theoretical_loss": 4.670548991874758,
+      "tokens_seen": 122421248
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009726178535606822,
+      "loss": 1.9603,
+      "theoretical_loss": 4.670214367788761,
+      "tokens_seen": 122486784
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009725977933801404,
+      "loss": 2.095,
+      "theoretical_loss": 4.66987997279438,
+      "tokens_seen": 122552320
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009725777331995989,
+      "loss": 1.9932,
+      "theoretical_loss": 4.669545806612389,
+      "tokens_seen": 122617856
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009725576730190571,
+      "loss": 2.0098,
+      "theoretical_loss": 4.669211868964052,
+      "tokens_seen": 122683392
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009725376128385156,
+      "loss": 2.0485,
+      "theoretical_loss": 4.668878159571123,
+      "tokens_seen": 122748928
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000972517552657974,
+      "loss": 2.0402,
+      "theoretical_loss": 4.6685446781558415,
+      "tokens_seen": 122814464
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009724974924774323,
+      "loss": 2.0854,
+      "theoretical_loss": 4.668211424440934,
+      "tokens_seen": 122880000
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009724774322968907,
+      "loss": 2.1726,
+      "theoretical_loss": 4.66787839814961,
+      "tokens_seen": 122945536
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000972457372116349,
+      "loss": 2.0551,
+      "theoretical_loss": 4.667545599005566,
+      "tokens_seen": 123011072
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009724373119358074,
+      "loss": 2.0483,
+      "theoretical_loss": 4.667213026732977,
+      "tokens_seen": 123076608
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009724172517552659,
+      "loss": 1.9799,
+      "theoretical_loss": 4.6668806810565036,
+      "tokens_seen": 123142144
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009723971915747241,
+      "loss": 2.1432,
+      "theoretical_loss": 4.666548561701285,
+      "tokens_seen": 123207680
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009723771313941826,
+      "loss": 2.091,
+      "theoretical_loss": 4.666216668392939,
+      "tokens_seen": 123273216
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000972357071213641,
+      "loss": 2.0325,
+      "theoretical_loss": 4.665885000857562,
+      "tokens_seen": 123338752
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009723370110330993,
+      "loss": 2.0689,
+      "theoretical_loss": 4.665553558821729,
+      "tokens_seen": 123404288
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009723169508525577,
+      "loss": 2.125,
+      "theoretical_loss": 4.665222342012491,
+      "tokens_seen": 123469824
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009722968906720161,
+      "loss": 1.9438,
+      "theoretical_loss": 4.664891350157374,
+      "tokens_seen": 123535360
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009722768304914744,
+      "loss": 2.0849,
+      "theoretical_loss": 4.664560582984374,
+      "tokens_seen": 123600896
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009722567703109329,
+      "loss": 1.9947,
+      "theoretical_loss": 4.664230040221966,
+      "tokens_seen": 123666432
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009722367101303911,
+      "loss": 2.2468,
+      "theoretical_loss": 4.663899721599093,
+      "tokens_seen": 123731968
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009722166499498496,
+      "loss": 1.9996,
+      "theoretical_loss": 4.66356962684517,
+      "tokens_seen": 123797504
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009721965897693079,
+      "loss": 2.03,
+      "theoretical_loss": 4.663239755690082,
+      "tokens_seen": 123863040
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009721765295887663,
+      "loss": 2.0711,
+      "theoretical_loss": 4.66291010786418,
+      "tokens_seen": 123928576
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009721564694082247,
+      "loss": 1.9784,
+      "theoretical_loss": 4.6625806830982865,
+      "tokens_seen": 123994112
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009721364092276831,
+      "loss": 1.9567,
+      "theoretical_loss": 4.662251481123688,
+      "tokens_seen": 124059648
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009721163490471414,
+      "loss": 2.0798,
+      "theoretical_loss": 4.661922501672136,
+      "tokens_seen": 124125184
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009720962888665999,
+      "loss": 2.0622,
+      "theoretical_loss": 4.66159374447585,
+      "tokens_seen": 124190720
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009720762286860581,
+      "loss": 2.0201,
+      "theoretical_loss": 4.661265209267507,
+      "tokens_seen": 124256256
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009720561685055166,
+      "loss": 1.9692,
+      "theoretical_loss": 4.660936895780253,
+      "tokens_seen": 124321792
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009720361083249749,
+      "loss": 2.1423,
+      "theoretical_loss": 4.660608803747688,
+      "tokens_seen": 124387328
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009720160481444333,
+      "loss": 2.0592,
+      "theoretical_loss": 4.660280932903879,
+      "tokens_seen": 124452864
+    },
+    {
+      "epoch": 0.04,
+      "objective/train/advantage_avg": 0.4943527281284332,
+      "objective/train/docs_used": 52572,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.8934916257858276,
+      "objective/train/original_loss": 1.8934917449951172,
+      "objective/train/theoretical_loss": 4.659953282983348,
+      "objective/train/tokens_used": 144978400,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2446935772895813,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050679326057434,
+      "objective/train/weighted_lm_loss": 1.9894511699676514,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.659953282983348,
+      "tokens_seen": 124518400
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009719959879638917,
+      "loss": 1.9832,
+      "theoretical_loss": 4.659953282983348,
+      "tokens_seen": 124518400
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009719759277833501,
+      "loss": 2.0029,
+      "theoretical_loss": 4.659625853721076,
+      "tokens_seen": 124583936
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009719558676028084,
+      "loss": 2.167,
+      "theoretical_loss": 4.659298644852502,
+      "tokens_seen": 124649472
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009719358074222668,
+      "loss": 2.0752,
+      "theoretical_loss": 4.65897165611352,
+      "tokens_seen": 124715008
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009719157472417252,
+      "loss": 2.1872,
+      "theoretical_loss": 4.658644887240481,
+      "tokens_seen": 124780544
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009718956870611836,
+      "loss": 2.0446,
+      "theoretical_loss": 4.6583183379701865,
+      "tokens_seen": 124846080
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009718756268806419,
+      "loss": 2.0463,
+      "theoretical_loss": 4.657992008039894,
+      "tokens_seen": 124911616
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009718555667001004,
+      "loss": 2.1342,
+      "theoretical_loss": 4.657665897187313,
+      "tokens_seen": 124977152
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009718355065195586,
+      "loss": 2.1419,
+      "theoretical_loss": 4.657340005150602,
+      "tokens_seen": 125042688
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009718154463390171,
+      "loss": 2.1212,
+      "theoretical_loss": 4.657014331668371,
+      "tokens_seen": 125108224
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009717953861584754,
+      "loss": 2.106,
+      "theoretical_loss": 4.656688876479679,
+      "tokens_seen": 125173760
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009717753259779338,
+      "loss": 2.0815,
+      "theoretical_loss": 4.656363639324032,
+      "tokens_seen": 125239296
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009717552657973922,
+      "loss": 2.0729,
+      "theoretical_loss": 4.656038619941382,
+      "tokens_seen": 125304832
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009717352056168506,
+      "loss": 1.8985,
+      "theoretical_loss": 4.655713818072132,
+      "tokens_seen": 125370368
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009717151454363089,
+      "loss": 2.0584,
+      "theoretical_loss": 4.655389233457123,
+      "tokens_seen": 125435904
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009716950852557674,
+      "loss": 2.067,
+      "theoretical_loss": 4.6550648658376454,
+      "tokens_seen": 125501440
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009716750250752256,
+      "loss": 2.0179,
+      "theoretical_loss": 4.654740714955429,
+      "tokens_seen": 125566976
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009716549648946841,
+      "loss": 2.0203,
+      "theoretical_loss": 4.654416780552648,
+      "tokens_seen": 125632512
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009716349047141424,
+      "loss": 2.1031,
+      "theoretical_loss": 4.654093062371917,
+      "tokens_seen": 125698048
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009716148445336008,
+      "loss": 2.1167,
+      "theoretical_loss": 4.65376956015629,
+      "tokens_seen": 125763584
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009715947843530592,
+      "loss": 2.0359,
+      "theoretical_loss": 4.653446273649259,
+      "tokens_seen": 125829120
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009715747241725175,
+      "loss": 2.0423,
+      "theoretical_loss": 4.653123202594757,
+      "tokens_seen": 125894656
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009715546639919759,
+      "loss": 2.1098,
+      "theoretical_loss": 4.652800346737152,
+      "tokens_seen": 125960192
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009715346038114344,
+      "loss": 2.062,
+      "theoretical_loss": 4.652477705821248,
+      "tokens_seen": 126025728
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009715145436308926,
+      "loss": 1.9808,
+      "theoretical_loss": 4.652155279592286,
+      "tokens_seen": 126091264
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009714944834503511,
+      "loss": 2.0248,
+      "theoretical_loss": 4.651833067795939,
+      "tokens_seen": 126156800
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009714744232698093,
+      "loss": 1.9461,
+      "theoretical_loss": 4.651511070178314,
+      "tokens_seen": 126222336
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009714543630892678,
+      "loss": 2.1108,
+      "theoretical_loss": 4.65118928648595,
+      "tokens_seen": 126287872
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009714343029087263,
+      "loss": 2.0036,
+      "theoretical_loss": 4.650867716465819,
+      "tokens_seen": 126353408
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009714142427281845,
+      "loss": 1.8282,
+      "theoretical_loss": 4.65054635986532,
+      "tokens_seen": 126418944
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000971394182547643,
+      "loss": 2.0785,
+      "theoretical_loss": 4.650225216432285,
+      "tokens_seen": 126484480
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009713741223671014,
+      "loss": 2.1061,
+      "theoretical_loss": 4.649904285914971,
+      "tokens_seen": 126550016
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009713540621865597,
+      "loss": 2.1424,
+      "theoretical_loss": 4.6495835680620665,
+      "tokens_seen": 126615552
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009713340020060181,
+      "loss": 2.0539,
+      "theoretical_loss": 4.649263062622682,
+      "tokens_seen": 126681088
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009713139418254765,
+      "loss": 1.9926,
+      "theoretical_loss": 4.648942769346356,
+      "tokens_seen": 126746624
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009712938816449348,
+      "loss": 2.0468,
+      "theoretical_loss": 4.648622687983053,
+      "tokens_seen": 126812160
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009712738214643933,
+      "loss": 1.9613,
+      "theoretical_loss": 4.648302818283158,
+      "tokens_seen": 126877696
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009712537612838515,
+      "loss": 2.1172,
+      "theoretical_loss": 4.64798315999748,
+      "tokens_seen": 126943232
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00097123370110331,
+      "loss": 2.1304,
+      "theoretical_loss": 4.647663712877252,
+      "tokens_seen": 127008768
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009712136409227683,
+      "loss": 2.0174,
+      "theoretical_loss": 4.6473444766741245,
+      "tokens_seen": 127074304
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009711935807422267,
+      "loss": 2.11,
+      "theoretical_loss": 4.64702545114017,
+      "tokens_seen": 127139840
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009711735205616851,
+      "loss": 2.0202,
+      "theoretical_loss": 4.646706636027881,
+      "tokens_seen": 127205376
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009711534603811435,
+      "loss": 2.0189,
+      "theoretical_loss": 4.646388031090163,
+      "tokens_seen": 127270912
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009711334002006018,
+      "loss": 2.0739,
+      "theoretical_loss": 4.646069636080346,
+      "tokens_seen": 127336448
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009711133400200603,
+      "loss": 2.0184,
+      "theoretical_loss": 4.645751450752172,
+      "tokens_seen": 127401984
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009710932798395185,
+      "loss": 1.9785,
+      "theoretical_loss": 4.645433474859798,
+      "tokens_seen": 127467520
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000971073219658977,
+      "loss": 1.9877,
+      "theoretical_loss": 4.645115708157796,
+      "tokens_seen": 127533056
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009710531594784353,
+      "loss": 1.9167,
+      "theoretical_loss": 4.6447981504011535,
+      "tokens_seen": 127598592
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009710330992978937,
+      "loss": 2.11,
+      "theoretical_loss": 4.644480801345268,
+      "tokens_seen": 127664128
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009710130391173521,
+      "loss": 2.0886,
+      "theoretical_loss": 4.644163660745951,
+      "tokens_seen": 127729664
+    },
+    {
+      "epoch": 0.04,
+      "objective/train/advantage_avg": 0.49203452467918396,
+      "objective/train/docs_used": 53741,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.9939597845077515,
+      "objective/train/original_loss": 1.993959665298462,
+      "objective/train/theoretical_loss": 4.6438467283594225,
+      "objective/train/tokens_used": 148255200,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2424752116203308,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050436019897461,
+      "objective/train/weighted_lm_loss": 2.0946204662323,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.6438467283594225,
+      "tokens_seen": 127795200
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009709929789368105,
+      "loss": 2.1437,
+      "theoretical_loss": 4.6438467283594225,
+      "tokens_seen": 127795200
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009709729187562688,
+      "loss": 2.1764,
+      "theoretical_loss": 4.643530003942313,
+      "tokens_seen": 127860736
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009709528585757272,
+      "loss": 1.9627,
+      "theoretical_loss": 4.643213487251664,
+      "tokens_seen": 127926272
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009709327983951856,
+      "loss": 1.9454,
+      "theoretical_loss": 4.642897178044924,
+      "tokens_seen": 127991808
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000970912738214644,
+      "loss": 1.9753,
+      "theoretical_loss": 4.6425810760799475,
+      "tokens_seen": 128057344
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009708926780341023,
+      "loss": 2.1277,
+      "theoretical_loss": 4.642265181114996,
+      "tokens_seen": 128122880
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009708726178535608,
+      "loss": 1.9976,
+      "theoretical_loss": 4.641949492908737,
+      "tokens_seen": 128188416
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000970852557673019,
+      "loss": 2.0455,
+      "theoretical_loss": 4.641634011220242,
+      "tokens_seen": 128253952
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009708324974924775,
+      "loss": 2.1228,
+      "theoretical_loss": 4.641318735808988,
+      "tokens_seen": 128319488
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009708124373119358,
+      "loss": 1.9554,
+      "theoretical_loss": 4.641003666434852,
+      "tokens_seen": 128385024
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009707923771313942,
+      "loss": 2.0163,
+      "theoretical_loss": 4.640688802858113,
+      "tokens_seen": 128450560
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009707723169508526,
+      "loss": 2.0832,
+      "theoretical_loss": 4.640374144839454,
+      "tokens_seen": 128516096
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000970752256770311,
+      "loss": 2.1972,
+      "theoretical_loss": 4.640059692139955,
+      "tokens_seen": 128581632
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009707321965897693,
+      "loss": 1.9591,
+      "theoretical_loss": 4.639745444521098,
+      "tokens_seen": 128647168
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009707121364092278,
+      "loss": 2.0579,
+      "theoretical_loss": 4.63943140174476,
+      "tokens_seen": 128712704
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000970692076228686,
+      "loss": 2.126,
+      "theoretical_loss": 4.63911756357322,
+      "tokens_seen": 128778240
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009706720160481445,
+      "loss": 2.0433,
+      "theoretical_loss": 4.63880392976915,
+      "tokens_seen": 128843776
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009706519558676028,
+      "loss": 1.974,
+      "theoretical_loss": 4.638490500095619,
+      "tokens_seen": 128909312
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009706318956870612,
+      "loss": 2.0815,
+      "theoretical_loss": 4.6381772743160905,
+      "tokens_seen": 128974848
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009706118355065196,
+      "loss": 2.1851,
+      "theoretical_loss": 4.637864252194423,
+      "tokens_seen": 129040384
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009705917753259779,
+      "loss": 2.0548,
+      "theoretical_loss": 4.637551433494869,
+      "tokens_seen": 129105920
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009705717151454363,
+      "loss": 2.0605,
+      "theoretical_loss": 4.637238817982071,
+      "tokens_seen": 129171456
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009705516549648948,
+      "loss": 1.9794,
+      "theoretical_loss": 4.636926405421065,
+      "tokens_seen": 129236992
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000970531594784353,
+      "loss": 2.1056,
+      "theoretical_loss": 4.636614195577276,
+      "tokens_seen": 129302528
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009705115346038115,
+      "loss": 1.8769,
+      "theoretical_loss": 4.63630218821652,
+      "tokens_seen": 129368064
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009704914744232697,
+      "loss": 2.0787,
+      "theoretical_loss": 4.635990383105003,
+      "tokens_seen": 129433600
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009704714142427282,
+      "loss": 1.9702,
+      "theoretical_loss": 4.635678780009318,
+      "tokens_seen": 129499136
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009704513540621866,
+      "loss": 2.0665,
+      "theoretical_loss": 4.6353673786964436,
+      "tokens_seen": 129564672
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009704312938816449,
+      "loss": 1.9906,
+      "theoretical_loss": 4.635056178933747,
+      "tokens_seen": 129630208
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009704112337011033,
+      "loss": 2.0691,
+      "theoretical_loss": 4.6347451804889825,
+      "tokens_seen": 129695744
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009703911735205618,
+      "loss": 1.9572,
+      "theoretical_loss": 4.634434383130284,
+      "tokens_seen": 129761280
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00097037111334002,
+      "loss": 2.1197,
+      "theoretical_loss": 4.634123786626175,
+      "tokens_seen": 129826816
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009703510531594785,
+      "loss": 2.0133,
+      "theoretical_loss": 4.633813390745559,
+      "tokens_seen": 129892352
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009703309929789367,
+      "loss": 2.1973,
+      "theoretical_loss": 4.633503195257724,
+      "tokens_seen": 129957888
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009703109327983952,
+      "loss": 2.1009,
+      "theoretical_loss": 4.633193199932336,
+      "tokens_seen": 130023424
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009702908726178536,
+      "loss": 2.1452,
+      "theoretical_loss": 4.632883404539445,
+      "tokens_seen": 130088960
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009702708124373119,
+      "loss": 2.0481,
+      "theoretical_loss": 4.632573808849479,
+      "tokens_seen": 130154496
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009702507522567703,
+      "loss": 2.0503,
+      "theoretical_loss": 4.632264412633246,
+      "tokens_seen": 130220032
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009702306920762287,
+      "loss": 2.0984,
+      "theoretical_loss": 4.631955215661932,
+      "tokens_seen": 130285568
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000970210631895687,
+      "loss": 2.1714,
+      "theoretical_loss": 4.6316462177071,
+      "tokens_seen": 130351104
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009701905717151455,
+      "loss": 2.1332,
+      "theoretical_loss": 4.63133741854069,
+      "tokens_seen": 130416640
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009701705115346038,
+      "loss": 2.0043,
+      "theoretical_loss": 4.631028817935018,
+      "tokens_seen": 130482176
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009701504513540622,
+      "loss": 1.9436,
+      "theoretical_loss": 4.630720415662774,
+      "tokens_seen": 130547712
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009701303911735205,
+      "loss": 1.9561,
+      "theoretical_loss": 4.630412211497022,
+      "tokens_seen": 130613248
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000970110330992979,
+      "loss": 1.8725,
+      "theoretical_loss": 4.630104205211202,
+      "tokens_seen": 130678784
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009700902708124373,
+      "loss": 2.0146,
+      "theoretical_loss": 4.629796396579124,
+      "tokens_seen": 130744320
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009700702106318957,
+      "loss": 1.9803,
+      "theoretical_loss": 4.629488785374969,
+      "tokens_seen": 130809856
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000970050150451354,
+      "loss": 2.1308,
+      "theoretical_loss": 4.629181371373292,
+      "tokens_seen": 130875392
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009700300902708125,
+      "loss": 2.0169,
+      "theoretical_loss": 4.628874154349017,
+      "tokens_seen": 130940928
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009700100300902708,
+      "loss": 1.9983,
+      "theoretical_loss": 4.628567134077436,
+      "tokens_seen": 131006464
+    },
+    {
+      "epoch": 0.04,
+      "objective/train/advantage_avg": 0.49247100949287415,
+      "objective/train/docs_used": 54897,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.9558112621307373,
+      "objective/train/original_loss": 1.9558112621307373,
+      "objective/train/theoretical_loss": 4.628260310334209,
+      "objective/train/tokens_used": 151532000,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24331486225128174,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.05048406124115,
+      "objective/train/weighted_lm_loss": 2.0544002056121826,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9048373699188232,
+      "theoretical_loss": 4.628260310334209,
+      "tokens_seen": 131072000
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009699899699097292,
+      "loss": 1.9695,
+      "theoretical_loss": 4.628260310334209,
+      "tokens_seen": 131072000
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009699699097291876,
+      "loss": 2.2511,
+      "theoretical_loss": 4.6279536828953685,
+      "tokens_seen": 131137536
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000969949849548646,
+      "loss": 1.9499,
+      "theoretical_loss": 4.627647251537309,
+      "tokens_seen": 131203072
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009699297893681044,
+      "loss": 1.9427,
+      "theoretical_loss": 4.627341016036794,
+      "tokens_seen": 131268608
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009699097291875627,
+      "loss": 2.0919,
+      "theoretical_loss": 4.6270349761709495,
+      "tokens_seen": 131334144
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009698896690070212,
+      "loss": 1.9728,
+      "theoretical_loss": 4.626729131717271,
+      "tokens_seen": 131399680
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009698696088264794,
+      "loss": 1.9052,
+      "theoretical_loss": 4.626423482453612,
+      "tokens_seen": 131465216
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009698495486459379,
+      "loss": 2.1468,
+      "theoretical_loss": 4.626118028158196,
+      "tokens_seen": 131530752
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009698294884653962,
+      "loss": 1.9891,
+      "theoretical_loss": 4.625812768609601,
+      "tokens_seen": 131596288
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009698094282848546,
+      "loss": 1.9941,
+      "theoretical_loss": 4.6255077035867735,
+      "tokens_seen": 131661824
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000969789368104313,
+      "loss": 2.2123,
+      "theoretical_loss": 4.625202832869016,
+      "tokens_seen": 131727360
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009697693079237714,
+      "loss": 2.0563,
+      "theoretical_loss": 4.624898156235994,
+      "tokens_seen": 131792896
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009697492477432297,
+      "loss": 2.0261,
+      "theoretical_loss": 4.62459367346773,
+      "tokens_seen": 131858432
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009697291875626882,
+      "loss": 2.0436,
+      "theoretical_loss": 4.624289384344609,
+      "tokens_seen": 131923968
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009697091273821464,
+      "loss": 2.1076,
+      "theoretical_loss": 4.623985288647368,
+      "tokens_seen": 131989504
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009696890672016049,
+      "loss": 1.9861,
+      "theoretical_loss": 4.623681386157104,
+      "tokens_seen": 132055040
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009696690070210632,
+      "loss": 2.0016,
+      "theoretical_loss": 4.623377676655271,
+      "tokens_seen": 132120576
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009696489468405216,
+      "loss": 2.0137,
+      "theoretical_loss": 4.623074159923678,
+      "tokens_seen": 132186112
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00096962888665998,
+      "loss": 2.0859,
+      "theoretical_loss": 4.622770835744487,
+      "tokens_seen": 132251648
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009696088264794383,
+      "loss": 2.0853,
+      "theoretical_loss": 4.622467703900216,
+      "tokens_seen": 132317184
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009695887662988967,
+      "loss": 2.1195,
+      "theoretical_loss": 4.622164764173735,
+      "tokens_seen": 132382720
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009695687061183552,
+      "loss": 2.0064,
+      "theoretical_loss": 4.621862016348267,
+      "tokens_seen": 132448256
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009695486459378134,
+      "loss": 2.1665,
+      "theoretical_loss": 4.621559460207386,
+      "tokens_seen": 132513792
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009695285857572719,
+      "loss": 2.141,
+      "theoretical_loss": 4.621257095535019,
+      "tokens_seen": 132579328
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009695085255767301,
+      "loss": 2.0802,
+      "theoretical_loss": 4.62095492211544,
+      "tokens_seen": 132644864
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009694884653961886,
+      "loss": 2.1012,
+      "theoretical_loss": 4.620652939733276,
+      "tokens_seen": 132710400
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000969468405215647,
+      "loss": 2.2375,
+      "theoretical_loss": 4.6203511481735,
+      "tokens_seen": 132775936
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009694483450351053,
+      "loss": 2.2068,
+      "theoretical_loss": 4.6200495472214325,
+      "tokens_seen": 132841472
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009694282848545637,
+      "loss": 2.1315,
+      "theoretical_loss": 4.6197481366627455,
+      "tokens_seen": 132907008
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009694082246740222,
+      "loss": 2.1076,
+      "theoretical_loss": 4.6194469162834535,
+      "tokens_seen": 132972544
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009693881644934804,
+      "loss": 2.0728,
+      "theoretical_loss": 4.6191458858699175,
+      "tokens_seen": 133038080
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009693681043129389,
+      "loss": 2.0247,
+      "theoretical_loss": 4.618845045208845,
+      "tokens_seen": 133103616
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009693480441323971,
+      "loss": 2.0239,
+      "theoretical_loss": 4.618544394087287,
+      "tokens_seen": 133169152
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009693279839518556,
+      "loss": 2.0453,
+      "theoretical_loss": 4.618243932292639,
+      "tokens_seen": 133234688
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000969307923771314,
+      "loss": 2.0592,
+      "theoretical_loss": 4.617943659612637,
+      "tokens_seen": 133300224
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009692878635907723,
+      "loss": 1.944,
+      "theoretical_loss": 4.617643575835361,
+      "tokens_seen": 133365760
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009692678034102307,
+      "loss": 1.8841,
+      "theoretical_loss": 4.617343680749233,
+      "tokens_seen": 133431296
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009692477432296891,
+      "loss": 2.0001,
+      "theoretical_loss": 4.617043974143014,
+      "tokens_seen": 133496832
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009692276830491474,
+      "loss": 1.9382,
+      "theoretical_loss": 4.616744455805806,
+      "tokens_seen": 133562368
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009692076228686059,
+      "loss": 2.1043,
+      "theoretical_loss": 4.616445125527051,
+      "tokens_seen": 133627904
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009691875626880642,
+      "loss": 1.9803,
+      "theoretical_loss": 4.61614598309653,
+      "tokens_seen": 133693440
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009691675025075226,
+      "loss": 2.0213,
+      "theoretical_loss": 4.6158470283043584,
+      "tokens_seen": 133758976
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009691474423269809,
+      "loss": 1.9767,
+      "theoretical_loss": 4.615548260940992,
+      "tokens_seen": 133824512
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009691273821464394,
+      "loss": 2.1192,
+      "theoretical_loss": 4.615249680797223,
+      "tokens_seen": 133890048
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009691073219658977,
+      "loss": 2.1185,
+      "theoretical_loss": 4.614951287664179,
+      "tokens_seen": 133955584
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009690872617853561,
+      "loss": 2.1068,
+      "theoretical_loss": 4.614653081333321,
+      "tokens_seen": 134021120
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009690672016048144,
+      "loss": 2.0312,
+      "theoretical_loss": 4.614355061596447,
+      "tokens_seen": 134086656
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009690471414242729,
+      "loss": 2.1178,
+      "theoretical_loss": 4.614057228245686,
+      "tokens_seen": 134152192
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009690270812437312,
+      "loss": 2.1038,
+      "theoretical_loss": 4.613759581073502,
+      "tokens_seen": 134217728
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009690070210631896,
+      "loss": 1.9887,
+      "theoretical_loss": 4.61346211987269,
+      "tokens_seen": 134283264
+    },
+    {
+      "epoch": 0.04,
+      "objective/train/advantage_avg": 0.49365389347076416,
+      "objective/train/docs_used": 56110,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.9194551706314087,
+      "objective/train/original_loss": 1.9194552898406982,
+      "objective/train/theoretical_loss": 4.613164844436379,
+      "objective/train/tokens_used": 154808800,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24395570158958435,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506056547164917,
+      "objective/train/weighted_lm_loss": 2.0164098739624023,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 1.0168063640594482,
+      "theoretical_loss": 4.613164844436379,
+      "tokens_seen": 134348800
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009689869608826479,
+      "loss": 2.1666,
+      "theoretical_loss": 4.613164844436379,
+      "tokens_seen": 134348800
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009689669007021064,
+      "loss": 2.1292,
+      "theoretical_loss": 4.6128677545580254,
+      "tokens_seen": 134414336
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009689468405215647,
+      "loss": 2.0502,
+      "theoretical_loss": 4.612570850031418,
+      "tokens_seen": 134479872
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009689267803410231,
+      "loss": 2.0157,
+      "theoretical_loss": 4.612274130650675,
+      "tokens_seen": 134545408
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009689067201604814,
+      "loss": 1.9991,
+      "theoretical_loss": 4.611977596210242,
+      "tokens_seen": 134610944
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009688866599799398,
+      "loss": 1.9506,
+      "theoretical_loss": 4.6116812465048955,
+      "tokens_seen": 134676480
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009688665997993982,
+      "loss": 2.0571,
+      "theoretical_loss": 4.611385081329736,
+      "tokens_seen": 134742016
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009688465396188566,
+      "loss": 2.0394,
+      "theoretical_loss": 4.611089100480193,
+      "tokens_seen": 134807552
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009688264794383149,
+      "loss": 1.8767,
+      "theoretical_loss": 4.610793303752023,
+      "tokens_seen": 134873088
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009688064192577734,
+      "loss": 2.0354,
+      "theoretical_loss": 4.610497690941305,
+      "tokens_seen": 134938624
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009687863590772317,
+      "loss": 2.1914,
+      "theoretical_loss": 4.610202261844444,
+      "tokens_seen": 135004160
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009687662988966901,
+      "loss": 2.0961,
+      "theoretical_loss": 4.609907016258171,
+      "tokens_seen": 135069696
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009687462387161484,
+      "loss": 2.1632,
+      "theoretical_loss": 4.609611953979537,
+      "tokens_seen": 135135232
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009687261785356068,
+      "loss": 1.9911,
+      "theoretical_loss": 4.609317074805919,
+      "tokens_seen": 135200768
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009687061183550652,
+      "loss": 1.9436,
+      "theoretical_loss": 4.6090223785350135,
+      "tokens_seen": 135266304
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009686860581745236,
+      "loss": 1.974,
+      "theoretical_loss": 4.608727864964839,
+      "tokens_seen": 135331840
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009686659979939819,
+      "loss": 1.9855,
+      "theoretical_loss": 4.608433533893738,
+      "tokens_seen": 135397376
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009686459378134404,
+      "loss": 2.0362,
+      "theoretical_loss": 4.608139385120367,
+      "tokens_seen": 135462912
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009686258776328986,
+      "loss": 2.0603,
+      "theoretical_loss": 4.607845418443706,
+      "tokens_seen": 135528448
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009686058174523571,
+      "loss": 1.95,
+      "theoretical_loss": 4.6075516336630535,
+      "tokens_seen": 135593984
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009685857572718155,
+      "loss": 2.0118,
+      "theoretical_loss": 4.607258030578024,
+      "tokens_seen": 135659520
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009685656970912738,
+      "loss": 2.0902,
+      "theoretical_loss": 4.606964608988553,
+      "tokens_seen": 135725056
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009685456369107322,
+      "loss": 2.266,
+      "theoretical_loss": 4.606671368694888,
+      "tokens_seen": 135790592
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009685255767301905,
+      "loss": 1.9943,
+      "theoretical_loss": 4.6063783094975985,
+      "tokens_seen": 135856128
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009685055165496489,
+      "loss": 2.0378,
+      "theoretical_loss": 4.606085431197563,
+      "tokens_seen": 135921664
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009684854563691074,
+      "loss": 2.0629,
+      "theoretical_loss": 4.6057927335959805,
+      "tokens_seen": 135987200
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009684653961885657,
+      "loss": 2.0653,
+      "theoretical_loss": 4.6055002164943595,
+      "tokens_seen": 136052736
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009684453360080241,
+      "loss": 2.1102,
+      "theoretical_loss": 4.605207879694525,
+      "tokens_seen": 136118272
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009684252758274826,
+      "loss": 1.9009,
+      "theoretical_loss": 4.604915722998614,
+      "tokens_seen": 136183808
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009684052156469408,
+      "loss": 1.9044,
+      "theoretical_loss": 4.604623746209077,
+      "tokens_seen": 136249344
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009683851554663993,
+      "loss": 2.0011,
+      "theoretical_loss": 4.604331949128672,
+      "tokens_seen": 136314880
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009683650952858575,
+      "loss": 2.173,
+      "theoretical_loss": 4.604040331560473,
+      "tokens_seen": 136380416
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000968345035105316,
+      "loss": 2.0807,
+      "theoretical_loss": 4.603748893307861,
+      "tokens_seen": 136445952
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009683249749247744,
+      "loss": 2.007,
+      "theoretical_loss": 4.603457634174527,
+      "tokens_seen": 136511488
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009683049147442327,
+      "loss": 2.0617,
+      "theoretical_loss": 4.603166553964474,
+      "tokens_seen": 136577024
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009682848545636911,
+      "loss": 2.2422,
+      "theoretical_loss": 4.602875652482009,
+      "tokens_seen": 136642560
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009682647943831495,
+      "loss": 1.9923,
+      "theoretical_loss": 4.60258492953175,
+      "tokens_seen": 136708096
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009682447342026078,
+      "loss": 2.0762,
+      "theoretical_loss": 4.60229438491862,
+      "tokens_seen": 136773632
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009682246740220663,
+      "loss": 2.1774,
+      "theoretical_loss": 4.60200401844785,
+      "tokens_seen": 136839168
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009682046138415246,
+      "loss": 1.9386,
+      "theoretical_loss": 4.601713829924977,
+      "tokens_seen": 136904704
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000968184553660983,
+      "loss": 2.037,
+      "theoretical_loss": 4.601423819155842,
+      "tokens_seen": 136970240
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009681644934804413,
+      "loss": 1.9515,
+      "theoretical_loss": 4.601133985946591,
+      "tokens_seen": 137035776
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009681444332998998,
+      "loss": 2.0307,
+      "theoretical_loss": 4.6008443301036746,
+      "tokens_seen": 137101312
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009681243731193581,
+      "loss": 1.9943,
+      "theoretical_loss": 4.600554851433845,
+      "tokens_seen": 137166848
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009681043129388165,
+      "loss": 2.1094,
+      "theoretical_loss": 4.60026554974416,
+      "tokens_seen": 137232384
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009680842527582748,
+      "loss": 2.0925,
+      "theoretical_loss": 4.599976424841977,
+      "tokens_seen": 137297920
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009680641925777333,
+      "loss": 2.0283,
+      "theoretical_loss": 4.5996874765349585,
+      "tokens_seen": 137363456
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009680441323971916,
+      "loss": 2.0397,
+      "theoretical_loss": 4.599398704631062,
+      "tokens_seen": 137428992
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00096802407221665,
+      "loss": 1.9583,
+      "theoretical_loss": 4.599110108938549,
+      "tokens_seen": 137494528
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009680040120361083,
+      "loss": 1.9251,
+      "theoretical_loss": 4.598821689265982,
+      "tokens_seen": 137560064
+    },
+    {
+      "epoch": 0.04,
+      "objective/train/advantage_avg": 0.49249258637428284,
+      "objective/train/docs_used": 57180,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 2.132284641265869,
+      "objective/train/original_loss": 2.1322851181030273,
+      "objective/train/theoretical_loss": 4.598533445422221,
+      "objective/train/tokens_used": 158085600,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24288922548294067,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.05048406124115,
+      "objective/train/weighted_lm_loss": 2.2398180961608887,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 1.0168063640594482,
+      "theoretical_loss": 4.598533445422221,
+      "tokens_seen": 137625600
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009679839518555668,
+      "loss": 2.1706,
+      "theoretical_loss": 4.598533445422221,
+      "tokens_seen": 137625600
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009679638916750251,
+      "loss": 1.9657,
+      "theoretical_loss": 4.598245377216422,
+      "tokens_seen": 137691136
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009679438314944835,
+      "loss": 1.9385,
+      "theoretical_loss": 4.5979574844580435,
+      "tokens_seen": 137756672
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009679237713139418,
+      "loss": 2.0471,
+      "theoretical_loss": 4.597669766956837,
+      "tokens_seen": 137822208
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009679037111334002,
+      "loss": 2.1599,
+      "theoretical_loss": 4.597382224522855,
+      "tokens_seen": 137887744
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009678836509528586,
+      "loss": 1.9596,
+      "theoretical_loss": 4.597094856966441,
+      "tokens_seen": 137953280
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000967863590772317,
+      "loss": 1.9701,
+      "theoretical_loss": 4.596807664098238,
+      "tokens_seen": 138018816
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009678435305917753,
+      "loss": 2.0581,
+      "theoretical_loss": 4.596520645729181,
+      "tokens_seen": 138084352
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009678234704112338,
+      "loss": 2.1893,
+      "theoretical_loss": 4.596233801670502,
+      "tokens_seen": 138149888
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009678034102306921,
+      "loss": 1.8744,
+      "theoretical_loss": 4.595947131733725,
+      "tokens_seen": 138215424
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009677833500501505,
+      "loss": 2.1235,
+      "theoretical_loss": 4.595660635730665,
+      "tokens_seen": 138280960
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009677632898696088,
+      "loss": 2.1337,
+      "theoretical_loss": 4.595374313473435,
+      "tokens_seen": 138346496
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009677432296890672,
+      "loss": 1.9614,
+      "theoretical_loss": 4.595088164774435,
+      "tokens_seen": 138412032
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009677231695085256,
+      "loss": 1.9998,
+      "theoretical_loss": 4.594802189446356,
+      "tokens_seen": 138477568
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000967703109327984,
+      "loss": 2.0338,
+      "theoretical_loss": 4.594516387302184,
+      "tokens_seen": 138543104
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009676830491474423,
+      "loss": 2.118,
+      "theoretical_loss": 4.5942307581551916,
+      "tokens_seen": 138608640
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009676629889669008,
+      "loss": 2.167,
+      "theoretical_loss": 4.593945301818941,
+      "tokens_seen": 138674176
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000967642928786359,
+      "loss": 1.9812,
+      "theoretical_loss": 4.593660018107286,
+      "tokens_seen": 138739712
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009676228686058175,
+      "loss": 1.9162,
+      "theoretical_loss": 4.593374906834365,
+      "tokens_seen": 138805248
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009676028084252759,
+      "loss": 1.8815,
+      "theoretical_loss": 4.593089967814606,
+      "tokens_seen": 138870784
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009675827482447342,
+      "loss": 2.0387,
+      "theoretical_loss": 4.592805200862726,
+      "tokens_seen": 138936320
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009675626880641926,
+      "loss": 2.0424,
+      "theoretical_loss": 4.592520605793725,
+      "tokens_seen": 139001856
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009675426278836509,
+      "loss": 2.0179,
+      "theoretical_loss": 4.592236182422891,
+      "tokens_seen": 139067392
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009675225677031093,
+      "loss": 1.9477,
+      "theoretical_loss": 4.591951930565798,
+      "tokens_seen": 139132928
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009675025075225678,
+      "loss": 1.9914,
+      "theoretical_loss": 4.591667850038302,
+      "tokens_seen": 139198464
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000967482447342026,
+      "loss": 2.1217,
+      "theoretical_loss": 4.591383940656547,
+      "tokens_seen": 139264000
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009674623871614845,
+      "loss": 2.0834,
+      "theoretical_loss": 4.59110020223696,
+      "tokens_seen": 139329536
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009674423269809429,
+      "loss": 2.0082,
+      "theoretical_loss": 4.590816634596247,
+      "tokens_seen": 139395072
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009674222668004012,
+      "loss": 1.883,
+      "theoretical_loss": 4.590533237551401,
+      "tokens_seen": 139460608
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009674022066198596,
+      "loss": 2.0717,
+      "theoretical_loss": 4.590250010919696,
+      "tokens_seen": 139526144
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000967382146439318,
+      "loss": 1.941,
+      "theoretical_loss": 4.5899669545186885,
+      "tokens_seen": 139591680
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009673620862587763,
+      "loss": 2.1246,
+      "theoretical_loss": 4.589684068166212,
+      "tokens_seen": 139657216
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009673420260782348,
+      "loss": 1.9705,
+      "theoretical_loss": 4.589401351680385,
+      "tokens_seen": 139722752
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000967321965897693,
+      "loss": 2.0961,
+      "theoretical_loss": 4.5891188048796,
+      "tokens_seen": 139788288
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009673019057171515,
+      "loss": 2.0898,
+      "theoretical_loss": 4.588836427582536,
+      "tokens_seen": 139853824
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009672818455366098,
+      "loss": 1.9653,
+      "theoretical_loss": 4.588554219608145,
+      "tokens_seen": 139919360
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009672617853560682,
+      "loss": 2.0289,
+      "theoretical_loss": 4.588272180775659,
+      "tokens_seen": 139984896
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009672417251755266,
+      "loss": 2.0156,
+      "theoretical_loss": 4.587990310904587,
+      "tokens_seen": 140050432
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000967221664994985,
+      "loss": 1.8817,
+      "theoretical_loss": 4.587708609814715,
+      "tokens_seen": 140115968
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009672016048144433,
+      "loss": 2.0011,
+      "theoretical_loss": 4.587427077326108,
+      "tokens_seen": 140181504
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009671815446339017,
+      "loss": 2.0467,
+      "theoretical_loss": 4.587145713259102,
+      "tokens_seen": 140247040
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00096716148445336,
+      "loss": 1.9799,
+      "theoretical_loss": 4.586864517434312,
+      "tokens_seen": 140312576
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009671414242728185,
+      "loss": 2.0688,
+      "theoretical_loss": 4.586583489672625,
+      "tokens_seen": 140378112
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009671213640922768,
+      "loss": 1.9412,
+      "theoretical_loss": 4.586302629795206,
+      "tokens_seen": 140443648
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009671013039117352,
+      "loss": 1.9925,
+      "theoretical_loss": 4.5860219376234905,
+      "tokens_seen": 140509184
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009670812437311936,
+      "loss": 2.0053,
+      "theoretical_loss": 4.585741412979187,
+      "tokens_seen": 140574720
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000967061183550652,
+      "loss": 1.9645,
+      "theoretical_loss": 4.585461055684279,
+      "tokens_seen": 140640256
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009670411233701103,
+      "loss": 2.204,
+      "theoretical_loss": 4.58518086556102,
+      "tokens_seen": 140705792
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009670210631895687,
+      "loss": 1.9912,
+      "theoretical_loss": 4.584900842431934,
+      "tokens_seen": 140771328
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000967001003009027,
+      "loss": 2.0303,
+      "theoretical_loss": 4.58462098611982,
+      "tokens_seen": 140836864
+    },
+    {
+      "epoch": 0.04,
+      "objective/train/advantage_avg": 0.49588945508003235,
+      "objective/train/docs_used": 58312,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.7948875427246094,
+      "objective/train/original_loss": 1.7948875427246094,
+      "objective/train/theoretical_loss": 4.584341296447742,
+      "objective/train/tokens_used": 161362400,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24614021182060242,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0508403778076172,
+      "objective/train/weighted_lm_loss": 1.8861085176467896,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.584341296447742,
+      "tokens_seen": 140902400
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009669809428284855,
+      "loss": 1.8735,
+      "theoretical_loss": 4.584341296447742,
+      "tokens_seen": 140902400
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009669608826479439,
+      "loss": 2.0184,
+      "theoretical_loss": 4.584061773239039,
+      "tokens_seen": 140967936
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009669408224674022,
+      "loss": 1.9653,
+      "theoretical_loss": 4.583782416317316,
+      "tokens_seen": 141033472
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009669207622868606,
+      "loss": 1.8443,
+      "theoretical_loss": 4.583503225506448,
+      "tokens_seen": 141099008
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000966900702106319,
+      "loss": 2.1068,
+      "theoretical_loss": 4.583224200630577,
+      "tokens_seen": 141164544
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009668806419257774,
+      "loss": 2.0342,
+      "theoretical_loss": 4.582945341514114,
+      "tokens_seen": 141230080
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009668605817452357,
+      "loss": 2.0212,
+      "theoretical_loss": 4.582666647981739,
+      "tokens_seen": 141295616
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009668405215646942,
+      "loss": 2.0599,
+      "theoretical_loss": 4.582388119858393,
+      "tokens_seen": 141361152
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009668204613841525,
+      "loss": 2.0357,
+      "theoretical_loss": 4.582109756969289,
+      "tokens_seen": 141426688
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009668004012036109,
+      "loss": 1.7923,
+      "theoretical_loss": 4.581831559139902,
+      "tokens_seen": 141492224
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009667803410230693,
+      "loss": 1.9931,
+      "theoretical_loss": 4.581553526195974,
+      "tokens_seen": 141557760
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009667602808425276,
+      "loss": 2.0577,
+      "theoretical_loss": 4.581275657963511,
+      "tokens_seen": 141623296
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000966740220661986,
+      "loss": 2.0293,
+      "theoretical_loss": 4.580997954268781,
+      "tokens_seen": 141688832
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009667201604814444,
+      "loss": 1.9801,
+      "theoretical_loss": 4.5807204149383205,
+      "tokens_seen": 141754368
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009667001003009027,
+      "loss": 2.028,
+      "theoretical_loss": 4.580443039798922,
+      "tokens_seen": 141819904
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009666800401203612,
+      "loss": 2.027,
+      "theoretical_loss": 4.580165828677647,
+      "tokens_seen": 141885440
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009666599799398194,
+      "loss": 2.0958,
+      "theoretical_loss": 4.579888781401816,
+      "tokens_seen": 141950976
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009666399197592779,
+      "loss": 2.0398,
+      "theoretical_loss": 4.579611897799008,
+      "tokens_seen": 142016512
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009666198595787363,
+      "loss": 1.9667,
+      "theoretical_loss": 4.57933517769707,
+      "tokens_seen": 142082048
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009665997993981946,
+      "loss": 2.0251,
+      "theoretical_loss": 4.579058620924103,
+      "tokens_seen": 142147584
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000966579739217653,
+      "loss": 1.9583,
+      "theoretical_loss": 4.578782227308471,
+      "tokens_seen": 142213120
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009665596790371113,
+      "loss": 2.063,
+      "theoretical_loss": 4.578505996678796,
+      "tokens_seen": 142278656
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009665396188565697,
+      "loss": 2.1415,
+      "theoretical_loss": 4.578229928863959,
+      "tokens_seen": 142344192
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009665195586760282,
+      "loss": 2.0813,
+      "theoretical_loss": 4.577954023693101,
+      "tokens_seen": 142409728
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009664994984954864,
+      "loss": 1.9436,
+      "theoretical_loss": 4.577678280995619,
+      "tokens_seen": 142475264
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009664794383149449,
+      "loss": 1.9627,
+      "theoretical_loss": 4.577402700601168,
+      "tokens_seen": 142540800
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009664593781344033,
+      "loss": 2.0339,
+      "theoretical_loss": 4.57712728233966,
+      "tokens_seen": 142606336
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009664393179538616,
+      "loss": 1.9898,
+      "theoretical_loss": 4.576852026041261,
+      "tokens_seen": 142671872
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00096641925777332,
+      "loss": 1.9128,
+      "theoretical_loss": 4.5765769315363976,
+      "tokens_seen": 142737408
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009663991975927783,
+      "loss": 2.0191,
+      "theoretical_loss": 4.576301998655747,
+      "tokens_seen": 142802944
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009663791374122367,
+      "loss": 1.9703,
+      "theoretical_loss": 4.576027227230245,
+      "tokens_seen": 142868480
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009663590772316952,
+      "loss": 2.0606,
+      "theoretical_loss": 4.5757526170910765,
+      "tokens_seen": 142934016
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009663390170511534,
+      "loss": 2.0374,
+      "theoretical_loss": 4.575478168069687,
+      "tokens_seen": 142999552
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009663189568706119,
+      "loss": 2.121,
+      "theoretical_loss": 4.575203879997771,
+      "tokens_seen": 143065088
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009662988966900702,
+      "loss": 2.0917,
+      "theoretical_loss": 4.574929752707274,
+      "tokens_seen": 143130624
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009662788365095286,
+      "loss": 1.9206,
+      "theoretical_loss": 4.5746557860304,
+      "tokens_seen": 143196160
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000966258776328987,
+      "loss": 2.0839,
+      "theoretical_loss": 4.574381979799599,
+      "tokens_seen": 143261696
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009662387161484454,
+      "loss": 2.0296,
+      "theoretical_loss": 4.574108333847575,
+      "tokens_seen": 143327232
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009662186559679037,
+      "loss": 1.9963,
+      "theoretical_loss": 4.573834848007284,
+      "tokens_seen": 143392768
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009661985957873621,
+      "loss": 2.0374,
+      "theoretical_loss": 4.573561522111927,
+      "tokens_seen": 143458304
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009661785356068204,
+      "loss": 2.0905,
+      "theoretical_loss": 4.573288355994963,
+      "tokens_seen": 143523840
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009661584754262789,
+      "loss": 1.8426,
+      "theoretical_loss": 4.573015349490094,
+      "tokens_seen": 143589376
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009661384152457372,
+      "loss": 2.0254,
+      "theoretical_loss": 4.572742502431272,
+      "tokens_seen": 143654912
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009661183550651956,
+      "loss": 2.0699,
+      "theoretical_loss": 4.572469814652698,
+      "tokens_seen": 143720448
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000966098294884654,
+      "loss": 2.0789,
+      "theoretical_loss": 4.572197285988824,
+      "tokens_seen": 143785984
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009660782347041124,
+      "loss": 1.925,
+      "theoretical_loss": 4.571924916274344,
+      "tokens_seen": 143851520
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009660581745235707,
+      "loss": 1.8722,
+      "theoretical_loss": 4.571652705344202,
+      "tokens_seen": 143917056
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009660381143430291,
+      "loss": 2.0093,
+      "theoretical_loss": 4.571380653033588,
+      "tokens_seen": 143982592
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009660180541624874,
+      "loss": 2.0186,
+      "theoretical_loss": 4.571108759177939,
+      "tokens_seen": 144048128
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009659979939819459,
+      "loss": 2.0186,
+      "theoretical_loss": 4.570837023612935,
+      "tokens_seen": 144113664
+    },
+    {
+      "epoch": 0.04,
+      "objective/train/advantage_avg": 0.4903619587421417,
+      "objective/train/docs_used": 59583,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.8303672075271606,
+      "objective/train/original_loss": 1.830367088317871,
+      "objective/train/theoretical_loss": 4.570565446174504,
+      "objective/train/tokens_used": 164639200,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2413649708032608,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0502631664276123,
+      "objective/train/weighted_lm_loss": 1.9222012758255005,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.570565446174504,
+      "tokens_seen": 144179200
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009659779338014042,
+      "loss": 2.0127,
+      "theoretical_loss": 4.570565446174504,
+      "tokens_seen": 144179200
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009659578736208626,
+      "loss": 1.9161,
+      "theoretical_loss": 4.570294026698815,
+      "tokens_seen": 144244736
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009659378134403209,
+      "loss": 2.0468,
+      "theoretical_loss": 4.570022765022285,
+      "tokens_seen": 144310272
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009659177532597794,
+      "loss": 2.0074,
+      "theoretical_loss": 4.569751660981572,
+      "tokens_seen": 144375808
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009658976930792377,
+      "loss": 1.9636,
+      "theoretical_loss": 4.569480714413578,
+      "tokens_seen": 144441344
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009658776328986961,
+      "loss": 2.0375,
+      "theoretical_loss": 4.569209925155445,
+      "tokens_seen": 144506880
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009658575727181545,
+      "loss": 1.9634,
+      "theoretical_loss": 4.568939293044564,
+      "tokens_seen": 144572416
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009658375125376129,
+      "loss": 2.0068,
+      "theoretical_loss": 4.56866881791856,
+      "tokens_seen": 144637952
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009658174523570712,
+      "loss": 2.078,
+      "theoretical_loss": 4.568398499615305,
+      "tokens_seen": 144703488
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009657973921765297,
+      "loss": 2.0601,
+      "theoretical_loss": 4.568128337972907,
+      "tokens_seen": 144769024
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009657773319959879,
+      "loss": 2.0646,
+      "theoretical_loss": 4.567858332829719,
+      "tokens_seen": 144834560
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009657572718154464,
+      "loss": 1.9944,
+      "theoretical_loss": 4.567588484024331,
+      "tokens_seen": 144900096
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009657372116349047,
+      "loss": 2.0121,
+      "theoretical_loss": 4.56731879139557,
+      "tokens_seen": 144965632
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009657171514543631,
+      "loss": 2.0377,
+      "theoretical_loss": 4.567049254782509,
+      "tokens_seen": 145031168
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009656970912738215,
+      "loss": 1.9483,
+      "theoretical_loss": 4.566779874024453,
+      "tokens_seen": 145096704
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009656770310932798,
+      "loss": 1.8921,
+      "theoretical_loss": 4.566510648960948,
+      "tokens_seen": 145162240
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009656569709127382,
+      "loss": 1.9893,
+      "theoretical_loss": 4.566241579431776,
+      "tokens_seen": 145227776
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009656369107321967,
+      "loss": 1.9413,
+      "theoretical_loss": 4.565972665276957,
+      "tokens_seen": 145293312
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009656168505516549,
+      "loss": 2.1933,
+      "theoretical_loss": 4.56570390633675,
+      "tokens_seen": 145358848
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009655967903711134,
+      "loss": 1.882,
+      "theoretical_loss": 4.565435302451645,
+      "tokens_seen": 145424384
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009655767301905716,
+      "loss": 2.0421,
+      "theoretical_loss": 4.565166853462371,
+      "tokens_seen": 145489920
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009655566700100301,
+      "loss": 2.0148,
+      "theoretical_loss": 4.56489855920989,
+      "tokens_seen": 145555456
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009655366098294885,
+      "loss": 2.1477,
+      "theoretical_loss": 4.564630419535405,
+      "tokens_seen": 145620992
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009655165496489468,
+      "loss": 1.9786,
+      "theoretical_loss": 4.564362434280344,
+      "tokens_seen": 145686528
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009654964894684052,
+      "loss": 2.0122,
+      "theoretical_loss": 4.564094603286375,
+      "tokens_seen": 145752064
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009654764292878637,
+      "loss": 2.0539,
+      "theoretical_loss": 4.5638269263954,
+      "tokens_seen": 145817600
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000965456369107322,
+      "loss": 2.0813,
+      "theoretical_loss": 4.563559403449551,
+      "tokens_seen": 145883136
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009654363089267804,
+      "loss": 1.826,
+      "theoretical_loss": 4.563292034291193,
+      "tokens_seen": 145948672
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009654162487462387,
+      "loss": 1.9389,
+      "theoretical_loss": 4.5630248187629245,
+      "tokens_seen": 146014208
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009653961885656971,
+      "loss": 1.9886,
+      "theoretical_loss": 4.562757756707574,
+      "tokens_seen": 146079744
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009653761283851556,
+      "loss": 2.0087,
+      "theoretical_loss": 4.5624908479682045,
+      "tokens_seen": 146145280
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009653560682046138,
+      "loss": 1.9831,
+      "theoretical_loss": 4.562224092388105,
+      "tokens_seen": 146210816
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009653360080240723,
+      "loss": 1.9913,
+      "theoretical_loss": 4.561957489810798,
+      "tokens_seen": 146276352
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009653159478435306,
+      "loss": 1.9859,
+      "theoretical_loss": 4.561691040080035,
+      "tokens_seen": 146341888
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000965295887662989,
+      "loss": 1.9011,
+      "theoretical_loss": 4.561424743039797,
+      "tokens_seen": 146407424
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009652758274824474,
+      "loss": 2.0157,
+      "theoretical_loss": 4.561158598534295,
+      "tokens_seen": 146472960
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009652557673019058,
+      "loss": 1.9363,
+      "theoretical_loss": 4.5608926064079665,
+      "tokens_seen": 146538496
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009652357071213641,
+      "loss": 2.0811,
+      "theoretical_loss": 4.560626766505479,
+      "tokens_seen": 146604032
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009652156469408225,
+      "loss": 1.9238,
+      "theoretical_loss": 4.5603610786717255,
+      "tokens_seen": 146669568
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009651955867602808,
+      "loss": 2.1267,
+      "theoretical_loss": 4.56009554275183,
+      "tokens_seen": 146735104
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009651755265797393,
+      "loss": 1.9993,
+      "theoretical_loss": 4.559830158591139,
+      "tokens_seen": 146800640
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009651554663991976,
+      "loss": 1.9902,
+      "theoretical_loss": 4.559564926035229,
+      "tokens_seen": 146866176
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000965135406218656,
+      "loss": 1.8406,
+      "theoretical_loss": 4.5592998449299,
+      "tokens_seen": 146931712
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009651153460381144,
+      "loss": 1.9684,
+      "theoretical_loss": 4.559034915121179,
+      "tokens_seen": 146997248
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009650952858575728,
+      "loss": 1.9249,
+      "theoretical_loss": 4.558770136455316,
+      "tokens_seen": 147062784
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009650752256770311,
+      "loss": 2.013,
+      "theoretical_loss": 4.55850550877879,
+      "tokens_seen": 147128320
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009650551654964895,
+      "loss": 2.001,
+      "theoretical_loss": 4.558241031938298,
+      "tokens_seen": 147193856
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009650351053159478,
+      "loss": 1.8982,
+      "theoretical_loss": 4.557976705780766,
+      "tokens_seen": 147259392
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009650150451354063,
+      "loss": 1.9986,
+      "theoretical_loss": 4.557712530153342,
+      "tokens_seen": 147324928
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009649949849548646,
+      "loss": 1.932,
+      "theoretical_loss": 4.557448504903396,
+      "tokens_seen": 147390464
+    },
+    {
+      "epoch": 0.04,
+      "objective/train/advantage_avg": 0.493988960981369,
+      "objective/train/docs_used": 60747,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.7616697549819946,
+      "objective/train/original_loss": 1.761669635772705,
+      "objective/train/theoretical_loss": 4.557184629878522,
+      "objective/train/tokens_used": 167916000,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2443990260362625,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050641417503357,
+      "objective/train/weighted_lm_loss": 1.8508374691009521,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 1.0,
+      "theoretical_loss": 4.557184629878522,
+      "tokens_seen": 147456000
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.000964974924774323,
+      "loss": 1.8862,
+      "theoretical_loss": 4.557184629878522,
+      "tokens_seen": 147456000
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009649548645937813,
+      "loss": 1.9551,
+      "theoretical_loss": 4.556920904926534,
+      "tokens_seen": 147521536
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009649348044132398,
+      "loss": 1.9787,
+      "theoretical_loss": 4.556657329895469,
+      "tokens_seen": 147587072
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009649147442326981,
+      "loss": 2.0888,
+      "theoretical_loss": 4.556393904633584,
+      "tokens_seen": 147652608
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009648946840521565,
+      "loss": 1.9027,
+      "theoretical_loss": 4.5561306289893615,
+      "tokens_seen": 147718144
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009648746238716149,
+      "loss": 1.9725,
+      "theoretical_loss": 4.555867502811498,
+      "tokens_seen": 147783680
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009648545636910733,
+      "loss": 2.0147,
+      "theoretical_loss": 4.5556045259489135,
+      "tokens_seen": 147849216
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009648345035105316,
+      "loss": 2.0284,
+      "theoretical_loss": 4.555341698250748,
+      "tokens_seen": 147914752
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00096481444332999,
+      "loss": 2.0678,
+      "theoretical_loss": 4.555079019566359,
+      "tokens_seen": 147980288
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009647943831494483,
+      "loss": 1.9821,
+      "theoretical_loss": 4.554816489745324,
+      "tokens_seen": 148045824
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009647743229689068,
+      "loss": 1.9895,
+      "theoretical_loss": 4.554554108637437,
+      "tokens_seen": 148111360
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009647542627883651,
+      "loss": 2.0104,
+      "theoretical_loss": 4.554291876092713,
+      "tokens_seen": 148176896
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009647342026078235,
+      "loss": 1.8984,
+      "theoretical_loss": 4.554029791961382,
+      "tokens_seen": 148242432
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009647141424272819,
+      "loss": 1.936,
+      "theoretical_loss": 4.5537678560938915,
+      "tokens_seen": 148307968
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009646940822467402,
+      "loss": 2.0408,
+      "theoretical_loss": 4.553506068340907,
+      "tokens_seen": 148373504
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0009646740220661986,
+      "loss": 1.88,
+      "theoretical_loss": 4.55324442855331,
+      "tokens_seen": 148439040
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009646539618856571,
+      "loss": 1.8749,
+      "theoretical_loss": 4.552982936582194,
+      "tokens_seen": 148504576
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009646339017051153,
+      "loss": 2.0451,
+      "theoretical_loss": 4.552721592278875,
+      "tokens_seen": 148570112
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009646138415245738,
+      "loss": 1.9803,
+      "theoretical_loss": 4.552460395494878,
+      "tokens_seen": 148635648
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000964593781344032,
+      "loss": 1.9701,
+      "theoretical_loss": 4.552199346081946,
+      "tokens_seen": 148701184
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009645737211634905,
+      "loss": 2.0254,
+      "theoretical_loss": 4.551938443892035,
+      "tokens_seen": 148766720
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009645536609829489,
+      "loss": 1.8414,
+      "theoretical_loss": 4.551677688777316,
+      "tokens_seen": 148832256
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009645336008024072,
+      "loss": 2.0927,
+      "theoretical_loss": 4.55141708059017,
+      "tokens_seen": 148897792
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009645135406218656,
+      "loss": 1.9209,
+      "theoretical_loss": 4.551156619183196,
+      "tokens_seen": 148963328
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009644934804413241,
+      "loss": 1.7724,
+      "theoretical_loss": 4.5508963044092035,
+      "tokens_seen": 149028864
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009644734202607823,
+      "loss": 2.0814,
+      "theoretical_loss": 4.550636136121211,
+      "tokens_seen": 149094400
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009644533600802408,
+      "loss": 1.8497,
+      "theoretical_loss": 4.5503761141724555,
+      "tokens_seen": 149159936
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000964433299899699,
+      "loss": 1.8274,
+      "theoretical_loss": 4.550116238416379,
+      "tokens_seen": 149225472
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009644132397191575,
+      "loss": 1.9875,
+      "theoretical_loss": 4.54985650870664,
+      "tokens_seen": 149291008
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009643931795386159,
+      "loss": 1.929,
+      "theoretical_loss": 4.549596924897102,
+      "tokens_seen": 149356544
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009643731193580742,
+      "loss": 1.9954,
+      "theoretical_loss": 4.549337486841843,
+      "tokens_seen": 149422080
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009643530591775326,
+      "loss": 1.8849,
+      "theoretical_loss": 4.54907819439515,
+      "tokens_seen": 149487616
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000964332998996991,
+      "loss": 1.9833,
+      "theoretical_loss": 4.54881904741152,
+      "tokens_seen": 149553152
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009643129388164493,
+      "loss": 1.8151,
+      "theoretical_loss": 4.548560045745656,
+      "tokens_seen": 149618688
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009642928786359078,
+      "loss": 2.1873,
+      "theoretical_loss": 4.548301189252473,
+      "tokens_seen": 149684224
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000964272818455366,
+      "loss": 1.9426,
+      "theoretical_loss": 4.548042477787093,
+      "tokens_seen": 149749760
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009642527582748245,
+      "loss": 2.0007,
+      "theoretical_loss": 4.547783911204846,
+      "tokens_seen": 149815296
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009642326980942828,
+      "loss": 2.0073,
+      "theoretical_loss": 4.54752548936127,
+      "tokens_seen": 149880832
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009642126379137412,
+      "loss": 1.9919,
+      "theoretical_loss": 4.54726721211211,
+      "tokens_seen": 149946368
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009641925777331996,
+      "loss": 1.9581,
+      "theoretical_loss": 4.547009079313316,
+      "tokens_seen": 150011904
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000964172517552658,
+      "loss": 1.9835,
+      "theoretical_loss": 4.546751090821047,
+      "tokens_seen": 150077440
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009641524573721163,
+      "loss": 1.9467,
+      "theoretical_loss": 4.546493246491666,
+      "tokens_seen": 150142976
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009641323971915748,
+      "loss": 1.88,
+      "theoretical_loss": 4.546235546181743,
+      "tokens_seen": 150208512
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009641123370110331,
+      "loss": 2.0288,
+      "theoretical_loss": 4.545977989748052,
+      "tokens_seen": 150274048
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009640922768304915,
+      "loss": 1.8078,
+      "theoretical_loss": 4.5457205770475735,
+      "tokens_seen": 150339584
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009640722166499498,
+      "loss": 1.8783,
+      "theoretical_loss": 4.5454633079374895,
+      "tokens_seen": 150405120
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009640521564694082,
+      "loss": 1.853,
+      "theoretical_loss": 4.545206182275189,
+      "tokens_seen": 150470656
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009640320962888666,
+      "loss": 1.8686,
+      "theoretical_loss": 4.544949199918264,
+      "tokens_seen": 150536192
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000964012036108325,
+      "loss": 1.9122,
+      "theoretical_loss": 4.544692360724509,
+      "tokens_seen": 150601728
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009639919759277833,
+      "loss": 1.8843,
+      "theoretical_loss": 4.544435664551921,
+      "tokens_seen": 150667264
+    },
+    {
+      "epoch": 0.05,
+      "objective/train/advantage_avg": 0.49405166506767273,
+      "objective/train/docs_used": 61752,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.8966041803359985,
+      "objective/train/original_loss": 1.8966041803359985,
+      "objective/train/theoretical_loss": 4.5441791112587016,
+      "objective/train/tokens_used": 171192800,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2443924993276596,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506476163864136,
+      "objective/train/weighted_lm_loss": 1.9925545454025269,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 1.0,
+      "theoretical_loss": 4.5441791112587016,
+      "tokens_seen": 150732800
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009639719157472417,
+      "loss": 1.9816,
+      "theoretical_loss": 4.5441791112587016,
+      "tokens_seen": 150732800
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009639518555667002,
+      "loss": 2.1063,
+      "theoretical_loss": 4.543922700703251,
+      "tokens_seen": 150798336
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009639317953861585,
+      "loss": 2.0471,
+      "theoretical_loss": 4.5436664327441765,
+      "tokens_seen": 150863872
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009639117352056169,
+      "loss": 2.0619,
+      "theoretical_loss": 4.543410307240281,
+      "tokens_seen": 150929408
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009638916750250753,
+      "loss": 1.9035,
+      "theoretical_loss": 4.5431543240505725,
+      "tokens_seen": 150994944
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009638716148445337,
+      "loss": 1.9168,
+      "theoretical_loss": 4.542898483034259,
+      "tokens_seen": 151060480
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000963851554663992,
+      "loss": 1.9746,
+      "theoretical_loss": 4.542642784050747,
+      "tokens_seen": 151126016
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009638314944834505,
+      "loss": 2.0691,
+      "theoretical_loss": 4.542387226959644,
+      "tokens_seen": 151191552
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009638114343029087,
+      "loss": 1.9151,
+      "theoretical_loss": 4.5421318116207585,
+      "tokens_seen": 151257088
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009637913741223672,
+      "loss": 1.8876,
+      "theoretical_loss": 4.541876537894095,
+      "tokens_seen": 151322624
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009637713139418255,
+      "loss": 1.8877,
+      "theoretical_loss": 4.54162140563986,
+      "tokens_seen": 151388160
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009637512537612839,
+      "loss": 1.8514,
+      "theoretical_loss": 4.541366414718456,
+      "tokens_seen": 151453696
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009637311935807423,
+      "loss": 1.9069,
+      "theoretical_loss": 4.541111564990485,
+      "tokens_seen": 151519232
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009637111334002006,
+      "loss": 1.9673,
+      "theoretical_loss": 4.540856856316747,
+      "tokens_seen": 151584768
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000963691073219659,
+      "loss": 1.8871,
+      "theoretical_loss": 4.540602288558238,
+      "tokens_seen": 151650304
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009636710130391175,
+      "loss": 1.8173,
+      "theoretical_loss": 4.5403478615761514,
+      "tokens_seen": 151715840
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009636509528585757,
+      "loss": 2.0094,
+      "theoretical_loss": 4.540093575231879,
+      "tokens_seen": 151781376
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009636308926780342,
+      "loss": 1.8458,
+      "theoretical_loss": 4.539839429387005,
+      "tokens_seen": 151846912
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009636108324974924,
+      "loss": 1.9317,
+      "theoretical_loss": 4.539585423903313,
+      "tokens_seen": 151912448
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009635907723169509,
+      "loss": 1.9469,
+      "theoretical_loss": 4.539331558642782,
+      "tokens_seen": 151977984
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009635707121364093,
+      "loss": 1.8159,
+      "theoretical_loss": 4.539077833467582,
+      "tokens_seen": 152043520
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009635506519558676,
+      "loss": 2.0562,
+      "theoretical_loss": 4.5388242482400845,
+      "tokens_seen": 152109056
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000963530591775326,
+      "loss": 1.9007,
+      "theoretical_loss": 4.53857080282285,
+      "tokens_seen": 152174592
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009635105315947845,
+      "loss": 1.8905,
+      "theoretical_loss": 4.538317497078634,
+      "tokens_seen": 152240128
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009634904714142427,
+      "loss": 1.8761,
+      "theoretical_loss": 4.538064330870389,
+      "tokens_seen": 152305664
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009634704112337012,
+      "loss": 2.1206,
+      "theoretical_loss": 4.537811304061257,
+      "tokens_seen": 152371200
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009634503510531594,
+      "loss": 1.8516,
+      "theoretical_loss": 4.537558416514575,
+      "tokens_seen": 152436736
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009634302908726179,
+      "loss": 1.955,
+      "theoretical_loss": 4.537305668093871,
+      "tokens_seen": 152502272
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009634102306920763,
+      "loss": 2.0363,
+      "theoretical_loss": 4.537053058662869,
+      "tokens_seen": 152567808
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009633901705115346,
+      "loss": 2.0132,
+      "theoretical_loss": 4.53680058808548,
+      "tokens_seen": 152633344
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000963370110330993,
+      "loss": 1.9282,
+      "theoretical_loss": 4.53654825622581,
+      "tokens_seen": 152698880
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009633500501504514,
+      "loss": 1.9805,
+      "theoretical_loss": 4.536296062948156,
+      "tokens_seen": 152764416
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009633299899699097,
+      "loss": 1.8875,
+      "theoretical_loss": 4.536044008117005,
+      "tokens_seen": 152829952
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009633099297893682,
+      "loss": 2.0565,
+      "theoretical_loss": 4.535792091597035,
+      "tokens_seen": 152895488
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009632898696088264,
+      "loss": 2.0856,
+      "theoretical_loss": 4.535540313253112,
+      "tokens_seen": 152961024
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009632698094282849,
+      "loss": 1.9599,
+      "theoretical_loss": 4.535288672950296,
+      "tokens_seen": 153026560
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009632497492477432,
+      "loss": 1.9393,
+      "theoretical_loss": 4.535037170553833,
+      "tokens_seen": 153092096
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009632296890672016,
+      "loss": 1.9247,
+      "theoretical_loss": 4.534785805929161,
+      "tokens_seen": 153157632
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00096320962888666,
+      "loss": 1.8164,
+      "theoretical_loss": 4.534534578941903,
+      "tokens_seen": 153223168
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009631895687061184,
+      "loss": 2.0316,
+      "theoretical_loss": 4.534283489457875,
+      "tokens_seen": 153288704
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009631695085255767,
+      "loss": 1.9613,
+      "theoretical_loss": 4.534032537343078,
+      "tokens_seen": 153354240
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009631494483450352,
+      "loss": 1.8766,
+      "theoretical_loss": 4.5337817224637025,
+      "tokens_seen": 153419776
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009631293881644935,
+      "loss": 2.105,
+      "theoretical_loss": 4.533531044686123,
+      "tokens_seen": 153485312
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009631093279839519,
+      "loss": 1.89,
+      "theoretical_loss": 4.533280503876906,
+      "tokens_seen": 153550848
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009630892678034102,
+      "loss": 2.0764,
+      "theoretical_loss": 4.533030099902803,
+      "tokens_seen": 153616384
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009630692076228686,
+      "loss": 1.9391,
+      "theoretical_loss": 4.532779832630748,
+      "tokens_seen": 153681920
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000963049147442327,
+      "loss": 2.0657,
+      "theoretical_loss": 4.532529701927868,
+      "tokens_seen": 153747456
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009630290872617854,
+      "loss": 1.9845,
+      "theoretical_loss": 4.53227970766147,
+      "tokens_seen": 153812992
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009630090270812437,
+      "loss": 1.9766,
+      "theoretical_loss": 4.53202984969905,
+      "tokens_seen": 153878528
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009629889669007021,
+      "loss": 1.949,
+      "theoretical_loss": 4.531780127908286,
+      "tokens_seen": 153944064
+    },
+    {
+      "epoch": 0.05,
+      "objective/train/advantage_avg": 0.49506837129592896,
+      "objective/train/docs_used": 63027,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 2.0724551677703857,
+      "objective/train/original_loss": 2.0724549293518066,
+      "objective/train/theoretical_loss": 4.531530542157043,
+      "objective/train/tokens_used": 174469600,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24586887657642365,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0507569313049316,
+      "objective/train/weighted_lm_loss": 2.1776838302612305,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.531530542157043,
+      "tokens_seen": 154009600
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009629689067201605,
+      "loss": 2.0624,
+      "theoretical_loss": 4.531530542157043,
+      "tokens_seen": 154009600
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009629488465396189,
+      "loss": 1.997,
+      "theoretical_loss": 4.53128109231337,
+      "tokens_seen": 154075136
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009629287863590772,
+      "loss": 2.1624,
+      "theoretical_loss": 4.531031778245499,
+      "tokens_seen": 154140672
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009629087261785357,
+      "loss": 1.9835,
+      "theoretical_loss": 4.530782599821848,
+      "tokens_seen": 154206208
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009628886659979939,
+      "loss": 1.8497,
+      "theoretical_loss": 4.5305335569110134,
+      "tokens_seen": 154271744
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009628686058174524,
+      "loss": 1.9732,
+      "theoretical_loss": 4.530284649381782,
+      "tokens_seen": 154337280
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009628485456369107,
+      "loss": 2.0882,
+      "theoretical_loss": 4.530035877103115,
+      "tokens_seen": 154402816
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009628284854563691,
+      "loss": 2.0315,
+      "theoretical_loss": 4.529787239944165,
+      "tokens_seen": 154468352
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009628084252758275,
+      "loss": 1.9269,
+      "theoretical_loss": 4.529538737774259,
+      "tokens_seen": 154533888
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009627883650952859,
+      "loss": 2.0243,
+      "theoretical_loss": 4.52929037046291,
+      "tokens_seen": 154599424
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009627683049147442,
+      "loss": 1.9704,
+      "theoretical_loss": 4.529042137879809,
+      "tokens_seen": 154664960
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009627482447342027,
+      "loss": 1.9537,
+      "theoretical_loss": 4.528794039894832,
+      "tokens_seen": 154730496
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009627281845536609,
+      "loss": 1.9413,
+      "theoretical_loss": 4.528546076378033,
+      "tokens_seen": 154796032
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009627081243731194,
+      "loss": 1.8682,
+      "theoretical_loss": 4.528298247199649,
+      "tokens_seen": 154861568
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009626880641925777,
+      "loss": 1.9414,
+      "theoretical_loss": 4.528050552230092,
+      "tokens_seen": 154927104
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009626680040120361,
+      "loss": 1.8703,
+      "theoretical_loss": 4.52780299133996,
+      "tokens_seen": 154992640
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009626479438314945,
+      "loss": 2.079,
+      "theoretical_loss": 4.527555564400026,
+      "tokens_seen": 155058176
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009626278836509528,
+      "loss": 2.0089,
+      "theoretical_loss": 4.527308271281243,
+      "tokens_seen": 155123712
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009626078234704112,
+      "loss": 1.9527,
+      "theoretical_loss": 4.527061111854746,
+      "tokens_seen": 155189248
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009625877632898697,
+      "loss": 2.0116,
+      "theoretical_loss": 4.526814085991843,
+      "tokens_seen": 155254784
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009625677031093279,
+      "loss": 2.0765,
+      "theoretical_loss": 4.5265671935640235,
+      "tokens_seen": 155320320
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009625476429287864,
+      "loss": 1.9764,
+      "theoretical_loss": 4.526320434442955,
+      "tokens_seen": 155385856
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009625275827482446,
+      "loss": 1.9078,
+      "theoretical_loss": 4.526073808500481,
+      "tokens_seen": 155451392
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009625075225677031,
+      "loss": 1.8425,
+      "theoretical_loss": 4.525827315608623,
+      "tokens_seen": 155516928
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009624874623871615,
+      "loss": 1.7777,
+      "theoretical_loss": 4.5255809556395805,
+      "tokens_seen": 155582464
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009624674022066198,
+      "loss": 2.0199,
+      "theoretical_loss": 4.525334728465727,
+      "tokens_seen": 155648000
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009624473420260783,
+      "loss": 2.0546,
+      "theoretical_loss": 4.525088633959613,
+      "tokens_seen": 155713536
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009624272818455367,
+      "loss": 1.8745,
+      "theoretical_loss": 4.524842671993966,
+      "tokens_seen": 155779072
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000962407221664995,
+      "loss": 1.9245,
+      "theoretical_loss": 4.524596842441689,
+      "tokens_seen": 155844608
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009623871614844534,
+      "loss": 2.1329,
+      "theoretical_loss": 4.524351145175858,
+      "tokens_seen": 155910144
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009623671013039118,
+      "loss": 1.8084,
+      "theoretical_loss": 4.524105580069728,
+      "tokens_seen": 155975680
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009623470411233701,
+      "loss": 1.9775,
+      "theoretical_loss": 4.523860146996725,
+      "tokens_seen": 156041216
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009623269809428286,
+      "loss": 1.8121,
+      "theoretical_loss": 4.5236148458304495,
+      "tokens_seen": 156106752
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009623069207622868,
+      "loss": 1.9081,
+      "theoretical_loss": 4.523369676444679,
+      "tokens_seen": 156172288
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009622868605817453,
+      "loss": 1.9997,
+      "theoretical_loss": 4.523124638713361,
+      "tokens_seen": 156237824
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009622668004012036,
+      "loss": 2.0708,
+      "theoretical_loss": 4.52287973251062,
+      "tokens_seen": 156303360
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000962246740220662,
+      "loss": 1.901,
+      "theoretical_loss": 4.52263495771075,
+      "tokens_seen": 156368896
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009622266800401204,
+      "loss": 1.9521,
+      "theoretical_loss": 4.522390314188221,
+      "tokens_seen": 156434432
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009622066198595788,
+      "loss": 1.9751,
+      "theoretical_loss": 4.522145801817673,
+      "tokens_seen": 156499968
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009621865596790371,
+      "loss": 1.9347,
+      "theoretical_loss": 4.5219014204739185,
+      "tokens_seen": 156565504
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009621664994984956,
+      "loss": 1.7942,
+      "theoretical_loss": 4.521657170031944,
+      "tokens_seen": 156631040
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009621464393179539,
+      "loss": 1.9491,
+      "theoretical_loss": 4.521413050366905,
+      "tokens_seen": 156696576
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009621263791374123,
+      "loss": 1.8449,
+      "theoretical_loss": 4.521169061354129,
+      "tokens_seen": 156762112
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009621063189568706,
+      "loss": 1.8015,
+      "theoretical_loss": 4.520925202869115,
+      "tokens_seen": 156827648
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000962086258776329,
+      "loss": 1.9523,
+      "theoretical_loss": 4.520681474787532,
+      "tokens_seen": 156893184
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009620661985957874,
+      "loss": 1.9431,
+      "theoretical_loss": 4.520437876985218,
+      "tokens_seen": 156958720
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009620461384152458,
+      "loss": 1.8395,
+      "theoretical_loss": 4.520194409338185,
+      "tokens_seen": 157024256
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009620260782347041,
+      "loss": 2.0112,
+      "theoretical_loss": 4.519951071722611,
+      "tokens_seen": 157089792
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009620060180541625,
+      "loss": 1.8646,
+      "theoretical_loss": 4.519707864014844,
+      "tokens_seen": 157155328
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009619859578736209,
+      "loss": 1.8881,
+      "theoretical_loss": 4.519464786091401,
+      "tokens_seen": 157220864
+    },
+    {
+      "epoch": 0.05,
+      "objective/train/advantage_avg": 0.4935777187347412,
+      "objective/train/docs_used": 64097,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 2.037578582763672,
+      "objective/train/original_loss": 2.037578582763672,
+      "objective/train/theoretical_loss": 4.519221837828971,
+      "objective/train/tokens_used": 177746400,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24397633969783783,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505980253219604,
+      "objective/train/weighted_lm_loss": 2.1407289505004883,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 1.0168063640594482,
+      "theoretical_loss": 4.519221837828971,
+      "tokens_seen": 157286400
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009619658976930793,
+      "loss": 2.1463,
+      "theoretical_loss": 4.519221837828971,
+      "tokens_seen": 157286400
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009619458375125376,
+      "loss": 1.8261,
+      "theoretical_loss": 4.518979019104407,
+      "tokens_seen": 157351936
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009619257773319961,
+      "loss": 2.0235,
+      "theoretical_loss": 4.518736329794732,
+      "tokens_seen": 157417472
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009619057171514543,
+      "loss": 1.9399,
+      "theoretical_loss": 4.5184937697771375,
+      "tokens_seen": 157483008
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009618856569709128,
+      "loss": 2.0169,
+      "theoretical_loss": 4.51825133892898,
+      "tokens_seen": 157548544
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009618655967903711,
+      "loss": 1.9519,
+      "theoretical_loss": 4.51800903712779,
+      "tokens_seen": 157614080
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009618455366098295,
+      "loss": 1.9318,
+      "theoretical_loss": 4.517766864251255,
+      "tokens_seen": 157679616
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009618254764292879,
+      "loss": 1.878,
+      "theoretical_loss": 4.517524820177239,
+      "tokens_seen": 157745152
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009618054162487463,
+      "loss": 1.8226,
+      "theoretical_loss": 4.517282904783764,
+      "tokens_seen": 157810688
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009617853560682046,
+      "loss": 1.9889,
+      "theoretical_loss": 4.5170411179490255,
+      "tokens_seen": 157876224
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009617652958876631,
+      "loss": 1.9731,
+      "theoretical_loss": 4.516799459551378,
+      "tokens_seen": 157941760
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009617452357071213,
+      "loss": 1.8943,
+      "theoretical_loss": 4.516557929469347,
+      "tokens_seen": 158007296
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009617251755265798,
+      "loss": 1.9482,
+      "theoretical_loss": 4.516316527581621,
+      "tokens_seen": 158072832
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009617051153460381,
+      "loss": 1.9452,
+      "theoretical_loss": 4.5160752537670525,
+      "tokens_seen": 158138368
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009616850551654965,
+      "loss": 1.9644,
+      "theoretical_loss": 4.515834107904659,
+      "tokens_seen": 158203904
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009616649949849549,
+      "loss": 2.0442,
+      "theoretical_loss": 4.515593089873624,
+      "tokens_seen": 158269440
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009616449348044132,
+      "loss": 1.9659,
+      "theoretical_loss": 4.515352199553295,
+      "tokens_seen": 158334976
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009616248746238716,
+      "loss": 2.0096,
+      "theoretical_loss": 4.51511143682318,
+      "tokens_seen": 158400512
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009616048144433301,
+      "loss": 1.9158,
+      "theoretical_loss": 4.514870801562955,
+      "tokens_seen": 158466048
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009615847542627883,
+      "loss": 1.904,
+      "theoretical_loss": 4.514630293652454,
+      "tokens_seen": 158531584
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009615646940822468,
+      "loss": 1.9249,
+      "theoretical_loss": 4.514389912971679,
+      "tokens_seen": 158597120
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000961544633901705,
+      "loss": 1.8967,
+      "theoretical_loss": 4.5141496594007915,
+      "tokens_seen": 158662656
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009615245737211635,
+      "loss": 1.9135,
+      "theoretical_loss": 4.513909532820117,
+      "tokens_seen": 158728192
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009615045135406219,
+      "loss": 1.9641,
+      "theoretical_loss": 4.513669533110141,
+      "tokens_seen": 158793728
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009614844533600802,
+      "loss": 1.9775,
+      "theoretical_loss": 4.513429660151513,
+      "tokens_seen": 158859264
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009614643931795386,
+      "loss": 1.9063,
+      "theoretical_loss": 4.513189913825043,
+      "tokens_seen": 158924800
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009614443329989971,
+      "loss": 1.9086,
+      "theoretical_loss": 4.5129502940117,
+      "tokens_seen": 158990336
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009614242728184553,
+      "loss": 1.803,
+      "theoretical_loss": 4.512710800592618,
+      "tokens_seen": 159055872
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009614042126379138,
+      "loss": 1.919,
+      "theoretical_loss": 4.51247143344909,
+      "tokens_seen": 159121408
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000961384152457372,
+      "loss": 1.8886,
+      "theoretical_loss": 4.512232192462566,
+      "tokens_seen": 159186944
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009613640922768305,
+      "loss": 1.9788,
+      "theoretical_loss": 4.511993077514662,
+      "tokens_seen": 159252480
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009613440320962889,
+      "loss": 1.9061,
+      "theoretical_loss": 4.51175408848715,
+      "tokens_seen": 159318016
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009613239719157472,
+      "loss": 1.9861,
+      "theoretical_loss": 4.511515225261961,
+      "tokens_seen": 159383552
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009613039117352056,
+      "loss": 1.8786,
+      "theoretical_loss": 4.511276487721187,
+      "tokens_seen": 159449088
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000961283851554664,
+      "loss": 1.9009,
+      "theoretical_loss": 4.511037875747078,
+      "tokens_seen": 159514624
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009612637913741223,
+      "loss": 1.9976,
+      "theoretical_loss": 4.5107993892220435,
+      "tokens_seen": 159580160
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009612437311935808,
+      "loss": 1.9536,
+      "theoretical_loss": 4.5105610280286506,
+      "tokens_seen": 159645696
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009612236710130391,
+      "loss": 1.9695,
+      "theoretical_loss": 4.5103227920496245,
+      "tokens_seen": 159711232
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009612036108324975,
+      "loss": 1.9963,
+      "theoretical_loss": 4.510084681167848,
+      "tokens_seen": 159776768
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009611835506519559,
+      "loss": 1.8545,
+      "theoretical_loss": 4.509846695266361,
+      "tokens_seen": 159842304
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009611634904714143,
+      "loss": 1.9444,
+      "theoretical_loss": 4.509608834228365,
+      "tokens_seen": 159907840
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009611434302908726,
+      "loss": 1.897,
+      "theoretical_loss": 4.50937109793721,
+      "tokens_seen": 159973376
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000961123370110331,
+      "loss": 1.7706,
+      "theoretical_loss": 4.5091334862764105,
+      "tokens_seen": 160038912
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009611033099297893,
+      "loss": 1.9551,
+      "theoretical_loss": 4.508895999129634,
+      "tokens_seen": 160104448
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009610832497492478,
+      "loss": 1.9135,
+      "theoretical_loss": 4.508658636380705,
+      "tokens_seen": 160169984
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009610631895687061,
+      "loss": 1.8921,
+      "theoretical_loss": 4.508421397913603,
+      "tokens_seen": 160235520
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009610431293881645,
+      "loss": 2.0087,
+      "theoretical_loss": 4.508184283612462,
+      "tokens_seen": 160301056
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009610230692076228,
+      "loss": 1.9208,
+      "theoretical_loss": 4.5079472933615765,
+      "tokens_seen": 160366592
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009610030090270813,
+      "loss": 2.1351,
+      "theoretical_loss": 4.507710427045389,
+      "tokens_seen": 160432128
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009609829488465396,
+      "loss": 2.0943,
+      "theoretical_loss": 4.507473684548502,
+      "tokens_seen": 160497664
+    },
+    {
+      "epoch": 0.05,
+      "objective/train/advantage_avg": 0.49687397480010986,
+      "objective/train/docs_used": 65316,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.8788886070251465,
+      "objective/train/original_loss": 1.8788886070251465,
+      "objective/train/theoretical_loss": 4.507237065755672,
+      "objective/train/tokens_used": 181023200,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24701941013336182,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0509432554244995,
+      "objective/train/weighted_lm_loss": 1.97454035282135,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.507237065755672,
+      "tokens_seen": 160563200
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000960962888665998,
+      "loss": 1.985,
+      "theoretical_loss": 4.507237065755672,
+      "tokens_seen": 160563200
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009609428284854565,
+      "loss": 1.8503,
+      "theoretical_loss": 4.507000570551806,
+      "tokens_seen": 160628736
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009609227683049147,
+      "loss": 1.8316,
+      "theoretical_loss": 4.50676419882197,
+      "tokens_seen": 160694272
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009609027081243732,
+      "loss": 2.0776,
+      "theoretical_loss": 4.506527950451379,
+      "tokens_seen": 160759808
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009608826479438315,
+      "loss": 1.8979,
+      "theoretical_loss": 4.506291825325405,
+      "tokens_seen": 160825344
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009608625877632899,
+      "loss": 1.9678,
+      "theoretical_loss": 4.5060558233295716,
+      "tokens_seen": 160890880
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009608425275827483,
+      "loss": 1.8766,
+      "theoretical_loss": 4.505819944349556,
+      "tokens_seen": 160956416
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009608224674022067,
+      "loss": 1.95,
+      "theoretical_loss": 4.505584188271186,
+      "tokens_seen": 161021952
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000960802407221665,
+      "loss": 1.9329,
+      "theoretical_loss": 4.505348554980444,
+      "tokens_seen": 161087488
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009607823470411235,
+      "loss": 1.8921,
+      "theoretical_loss": 4.5051130443634655,
+      "tokens_seen": 161153024
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009607622868605817,
+      "loss": 1.7781,
+      "theoretical_loss": 4.504877656306535,
+      "tokens_seen": 161218560
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009607422266800402,
+      "loss": 2.0076,
+      "theoretical_loss": 4.504642390696089,
+      "tokens_seen": 161284096
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009607221664994985,
+      "loss": 1.9829,
+      "theoretical_loss": 4.504407247418717,
+      "tokens_seen": 161349632
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009607021063189569,
+      "loss": 1.9316,
+      "theoretical_loss": 4.50417222636116,
+      "tokens_seen": 161415168
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009606820461384153,
+      "loss": 1.973,
+      "theoretical_loss": 4.503937327410306,
+      "tokens_seen": 161480704
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009606619859578736,
+      "loss": 1.9379,
+      "theoretical_loss": 4.5037025504531965,
+      "tokens_seen": 161546240
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000960641925777332,
+      "loss": 1.9915,
+      "theoretical_loss": 4.503467895377025,
+      "tokens_seen": 161611776
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009606218655967905,
+      "loss": 1.9256,
+      "theoretical_loss": 4.50323336206913,
+      "tokens_seen": 161677312
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009606018054162487,
+      "loss": 1.984,
+      "theoretical_loss": 4.502998950417004,
+      "tokens_seen": 161742848
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009605817452357072,
+      "loss": 1.8998,
+      "theoretical_loss": 4.502764660308286,
+      "tokens_seen": 161808384
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009605616850551654,
+      "loss": 2.0802,
+      "theoretical_loss": 4.502530491630768,
+      "tokens_seen": 161873920
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009605416248746239,
+      "loss": 2.0452,
+      "theoretical_loss": 4.502296444272387,
+      "tokens_seen": 161939456
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009605215646940823,
+      "loss": 1.8631,
+      "theoretical_loss": 4.502062518121232,
+      "tokens_seen": 162004992
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009605015045135406,
+      "loss": 1.9151,
+      "theoretical_loss": 4.501828713065537,
+      "tokens_seen": 162070528
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000960481444332999,
+      "loss": 1.8775,
+      "theoretical_loss": 4.501595028993686,
+      "tokens_seen": 162136064
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009604613841524575,
+      "loss": 1.9531,
+      "theoretical_loss": 4.501361465794213,
+      "tokens_seen": 162201600
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009604413239719157,
+      "loss": 1.9826,
+      "theoretical_loss": 4.501128023355796,
+      "tokens_seen": 162267136
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009604212637913742,
+      "loss": 1.9779,
+      "theoretical_loss": 4.500894701567263,
+      "tokens_seen": 162332672
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009604012036108325,
+      "loss": 1.8989,
+      "theoretical_loss": 4.500661500317587,
+      "tokens_seen": 162398208
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009603811434302909,
+      "loss": 1.8547,
+      "theoretical_loss": 4.500428419495892,
+      "tokens_seen": 162463744
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009603610832497493,
+      "loss": 1.9796,
+      "theoretical_loss": 4.500195458991443,
+      "tokens_seen": 162529280
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009603410230692076,
+      "loss": 2.0481,
+      "theoretical_loss": 4.4999626186936545,
+      "tokens_seen": 162594816
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000960320962888666,
+      "loss": 2.0033,
+      "theoretical_loss": 4.499729898492089,
+      "tokens_seen": 162660352
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009603009027081244,
+      "loss": 2.0799,
+      "theoretical_loss": 4.4994972982764505,
+      "tokens_seen": 162725888
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009602808425275827,
+      "loss": 1.8236,
+      "theoretical_loss": 4.499264817936593,
+      "tokens_seen": 162791424
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009602607823470412,
+      "loss": 2.1797,
+      "theoretical_loss": 4.499032457362514,
+      "tokens_seen": 162856960
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009602407221664995,
+      "loss": 1.9804,
+      "theoretical_loss": 4.498800216444353,
+      "tokens_seen": 162922496
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009602206619859579,
+      "loss": 1.9246,
+      "theoretical_loss": 4.498568095072401,
+      "tokens_seen": 162988032
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009602006018054163,
+      "loss": 1.9251,
+      "theoretical_loss": 4.498336093137089,
+      "tokens_seen": 163053568
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009601805416248747,
+      "loss": 1.9896,
+      "theoretical_loss": 4.498104210528993,
+      "tokens_seen": 163119104
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000960160481444333,
+      "loss": 1.9048,
+      "theoretical_loss": 4.497872447138834,
+      "tokens_seen": 163184640
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009601404212637914,
+      "loss": 1.9159,
+      "theoretical_loss": 4.497640802857477,
+      "tokens_seen": 163250176
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009601203610832497,
+      "loss": 2.0265,
+      "theoretical_loss": 4.49740927757593,
+      "tokens_seen": 163315712
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009601003009027082,
+      "loss": 1.7832,
+      "theoretical_loss": 4.497177871185343,
+      "tokens_seen": 163381248
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009600802407221665,
+      "loss": 2.0497,
+      "theoretical_loss": 4.4969465835770155,
+      "tokens_seen": 163446784
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009600601805416249,
+      "loss": 1.8603,
+      "theoretical_loss": 4.496715414642381,
+      "tokens_seen": 163512320
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009600401203610832,
+      "loss": 1.9919,
+      "theoretical_loss": 4.496484364273021,
+      "tokens_seen": 163577856
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009600200601805417,
+      "loss": 2.0578,
+      "theoretical_loss": 4.49625343236066,
+      "tokens_seen": 163643392
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00096,
+      "loss": 1.97,
+      "theoretical_loss": 4.496022618797161,
+      "tokens_seen": 163708928
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009599799398194584,
+      "loss": 1.8953,
+      "theoretical_loss": 4.495791923474533,
+      "tokens_seen": 163774464
+    },
+    {
+      "epoch": 0.05,
+      "objective/train/advantage_avg": 0.4943779706954956,
+      "objective/train/docs_used": 66569,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.7600229978561401,
+      "objective/train/original_loss": 1.7600231170654297,
+      "objective/train/theoretical_loss": 4.495561346284925,
+      "objective/train/tokens_used": 184300000,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2446385622024536,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050681471824646,
+      "objective/train/weighted_lm_loss": 1.849135160446167,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.495561346284925,
+      "tokens_seen": 163840000
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009599598796389167,
+      "loss": 1.7395,
+      "theoretical_loss": 4.495561346284925,
+      "tokens_seen": 163840000
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009599398194583751,
+      "loss": 1.8734,
+      "theoretical_loss": 4.495330887120624,
+      "tokens_seen": 163905536
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009599197592778335,
+      "loss": 2.0076,
+      "theoretical_loss": 4.495100545874065,
+      "tokens_seen": 163971072
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009598996990972919,
+      "loss": 1.955,
+      "theoretical_loss": 4.4948703224378175,
+      "tokens_seen": 164036608
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009598796389167502,
+      "loss": 1.9929,
+      "theoretical_loss": 4.494640216704598,
+      "tokens_seen": 164102144
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009598595787362087,
+      "loss": 2.0399,
+      "theoretical_loss": 4.494410228567258,
+      "tokens_seen": 164167680
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000959839518555667,
+      "loss": 1.9737,
+      "theoretical_loss": 4.494180357918792,
+      "tokens_seen": 164233216
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009598194583751254,
+      "loss": 1.9149,
+      "theoretical_loss": 4.493950604652333,
+      "tokens_seen": 164298752
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009597993981945838,
+      "loss": 1.9785,
+      "theoretical_loss": 4.493720968661158,
+      "tokens_seen": 164364288
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009597793380140421,
+      "loss": 1.8819,
+      "theoretical_loss": 4.493491449838676,
+      "tokens_seen": 164429824
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009597592778335005,
+      "loss": 1.9185,
+      "theoretical_loss": 4.493262048078442,
+      "tokens_seen": 164495360
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000959739217652959,
+      "loss": 1.928,
+      "theoretical_loss": 4.493032763274148,
+      "tokens_seen": 164560896
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009597191574724172,
+      "loss": 1.9115,
+      "theoretical_loss": 4.492803595319623,
+      "tokens_seen": 164626432
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009596990972918757,
+      "loss": 1.8792,
+      "theoretical_loss": 4.492574544108838,
+      "tokens_seen": 164691968
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009596790371113339,
+      "loss": 1.7851,
+      "theoretical_loss": 4.492345609535899,
+      "tokens_seen": 164757504
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009596589769307924,
+      "loss": 2.0436,
+      "theoretical_loss": 4.492116791495052,
+      "tokens_seen": 164823040
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009596389167502508,
+      "loss": 1.8955,
+      "theoretical_loss": 4.49188808988068,
+      "tokens_seen": 164888576
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009596188565697091,
+      "loss": 1.7672,
+      "theoretical_loss": 4.491659504587304,
+      "tokens_seen": 164954112
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009595987963891675,
+      "loss": 1.8993,
+      "theoretical_loss": 4.491431035509584,
+      "tokens_seen": 165019648
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009595787362086258,
+      "loss": 1.8166,
+      "theoretical_loss": 4.4912026825423155,
+      "tokens_seen": 165085184
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009595586760280842,
+      "loss": 1.8962,
+      "theoretical_loss": 4.490974445580429,
+      "tokens_seen": 165150720
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009595386158475427,
+      "loss": 1.8735,
+      "theoretical_loss": 4.490746324518995,
+      "tokens_seen": 165216256
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009595185556670009,
+      "loss": 1.8615,
+      "theoretical_loss": 4.490518319253221,
+      "tokens_seen": 165281792
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009594984954864594,
+      "loss": 1.8063,
+      "theoretical_loss": 4.490290429678447,
+      "tokens_seen": 165347328
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009594784353059178,
+      "loss": 1.9261,
+      "theoretical_loss": 4.490062655690153,
+      "tokens_seen": 165412864
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009594583751253761,
+      "loss": 1.8684,
+      "theoretical_loss": 4.4898349971839515,
+      "tokens_seen": 165478400
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009594383149448346,
+      "loss": 1.7628,
+      "theoretical_loss": 4.489607454055593,
+      "tokens_seen": 165543936
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009594182547642929,
+      "loss": 1.9646,
+      "theoretical_loss": 4.489380026200961,
+      "tokens_seen": 165609472
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009593981945837513,
+      "loss": 1.9232,
+      "theoretical_loss": 4.489152713516077,
+      "tokens_seen": 165675008
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009593781344032097,
+      "loss": 1.7531,
+      "theoretical_loss": 4.488925515897095,
+      "tokens_seen": 165740544
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000959358074222668,
+      "loss": 1.6834,
+      "theoretical_loss": 4.488698433240304,
+      "tokens_seen": 165806080
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009593380140421264,
+      "loss": 1.949,
+      "theoretical_loss": 4.48847146544213,
+      "tokens_seen": 165871616
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009593179538615848,
+      "loss": 1.8811,
+      "theoretical_loss": 4.488244612399129,
+      "tokens_seen": 165937152
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009592978936810431,
+      "loss": 1.8608,
+      "theoretical_loss": 4.488017874007993,
+      "tokens_seen": 166002688
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009592778335005016,
+      "loss": 1.873,
+      "theoretical_loss": 4.487791250165549,
+      "tokens_seen": 166068224
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009592577733199599,
+      "loss": 1.8062,
+      "theoretical_loss": 4.487564740768756,
+      "tokens_seen": 166133760
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009592377131394183,
+      "loss": 1.8608,
+      "theoretical_loss": 4.487338345714707,
+      "tokens_seen": 166199296
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009592176529588767,
+      "loss": 1.9687,
+      "theoretical_loss": 4.487112064900627,
+      "tokens_seen": 166264832
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009591975927783351,
+      "loss": 1.9348,
+      "theoretical_loss": 4.486885898223873,
+      "tokens_seen": 166330368
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009591775325977934,
+      "loss": 1.8068,
+      "theoretical_loss": 4.48665984558194,
+      "tokens_seen": 166395904
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009591574724172518,
+      "loss": 1.8497,
+      "theoretical_loss": 4.486433906872448,
+      "tokens_seen": 166461440
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009591374122367101,
+      "loss": 1.971,
+      "theoretical_loss": 4.486208081993155,
+      "tokens_seen": 166526976
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009591173520561686,
+      "loss": 1.9106,
+      "theoretical_loss": 4.485982370841946,
+      "tokens_seen": 166592512
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009590972918756269,
+      "loss": 1.878,
+      "theoretical_loss": 4.485756773316844,
+      "tokens_seen": 166658048
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009590772316950853,
+      "loss": 1.9501,
+      "theoretical_loss": 4.485531289315997,
+      "tokens_seen": 166723584
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009590571715145436,
+      "loss": 1.8681,
+      "theoretical_loss": 4.485305918737687,
+      "tokens_seen": 166789120
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009590371113340021,
+      "loss": 1.8118,
+      "theoretical_loss": 4.48508066148033,
+      "tokens_seen": 166854656
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009590170511534604,
+      "loss": 1.8352,
+      "theoretical_loss": 4.484855517442467,
+      "tokens_seen": 166920192
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009589969909729188,
+      "loss": 1.7397,
+      "theoretical_loss": 4.484630486522775,
+      "tokens_seen": 166985728
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009589769307923771,
+      "loss": 1.9343,
+      "theoretical_loss": 4.4844055686200575,
+      "tokens_seen": 167051264
+    },
+    {
+      "epoch": 0.05,
+      "objective/train/advantage_avg": 0.4909823536872864,
+      "objective/train/docs_used": 67770,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.918248176574707,
+      "objective/train/original_loss": 1.9182484149932861,
+      "objective/train/theoretical_loss": 4.48418076363325,
+      "objective/train/tokens_used": 187576800,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2417370080947876,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503270626068115,
+      "objective/train/weighted_lm_loss": 2.014829158782959,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.48418076363325,
+      "tokens_seen": 167116800
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009589568706118355,
+      "loss": 1.9665,
+      "theoretical_loss": 4.48418076363325,
+      "tokens_seen": 167116800
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009589368104312939,
+      "loss": 1.9255,
+      "theoretical_loss": 4.483956071461419,
+      "tokens_seen": 167182336
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009589167502507523,
+      "loss": 2.0147,
+      "theoretical_loss": 4.483731492003757,
+      "tokens_seen": 167247872
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009588966900702106,
+      "loss": 1.825,
+      "theoretical_loss": 4.483507025159591,
+      "tokens_seen": 167313408
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009588766298896691,
+      "loss": 1.8508,
+      "theoretical_loss": 4.483282670828374,
+      "tokens_seen": 167378944
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009588565697091274,
+      "loss": 1.9944,
+      "theoretical_loss": 4.483058428909688,
+      "tokens_seen": 167444480
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009588365095285858,
+      "loss": 1.8854,
+      "theoretical_loss": 4.482834299303246,
+      "tokens_seen": 167510016
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009588164493480442,
+      "loss": 1.9292,
+      "theoretical_loss": 4.482610281908888,
+      "tokens_seen": 167575552
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009587963891675025,
+      "loss": 1.9696,
+      "theoretical_loss": 4.482386376626582,
+      "tokens_seen": 167641088
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009587763289869609,
+      "loss": 1.8661,
+      "theoretical_loss": 4.482162583356427,
+      "tokens_seen": 167706624
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009587562688064194,
+      "loss": 1.7834,
+      "theoretical_loss": 4.481938901998647,
+      "tokens_seen": 167772160
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009587362086258776,
+      "loss": 2.0171,
+      "theoretical_loss": 4.481715332453595,
+      "tokens_seen": 167837696
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009587161484453361,
+      "loss": 1.8928,
+      "theoretical_loss": 4.481491874621751,
+      "tokens_seen": 167903232
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009586960882647943,
+      "loss": 1.6732,
+      "theoretical_loss": 4.481268528403723,
+      "tokens_seen": 167968768
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009586760280842528,
+      "loss": 1.8741,
+      "theoretical_loss": 4.481045293700248,
+      "tokens_seen": 168034304
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009586559679037112,
+      "loss": 1.8851,
+      "theoretical_loss": 4.480822170412186,
+      "tokens_seen": 168099840
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009586359077231695,
+      "loss": 1.8634,
+      "theoretical_loss": 4.480599158440526,
+      "tokens_seen": 168165376
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009586158475426279,
+      "loss": 1.7684,
+      "theoretical_loss": 4.480376257686384,
+      "tokens_seen": 168230912
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009585957873620862,
+      "loss": 2.012,
+      "theoretical_loss": 4.480153468051001,
+      "tokens_seen": 168296448
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009585757271815446,
+      "loss": 1.9126,
+      "theoretical_loss": 4.479930789435746,
+      "tokens_seen": 168361984
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009585556670010031,
+      "loss": 1.9654,
+      "theoretical_loss": 4.479708221742111,
+      "tokens_seen": 168427520
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009585356068204613,
+      "loss": 1.8602,
+      "theoretical_loss": 4.479485764871717,
+      "tokens_seen": 168493056
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009585155466399198,
+      "loss": 1.9912,
+      "theoretical_loss": 4.4792634187263065,
+      "tokens_seen": 168558592
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009584954864593782,
+      "loss": 2.0017,
+      "theoretical_loss": 4.479041183207752,
+      "tokens_seen": 168624128
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009584754262788365,
+      "loss": 1.8021,
+      "theoretical_loss": 4.478819058218047,
+      "tokens_seen": 168689664
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009584553660982949,
+      "loss": 1.9251,
+      "theoretical_loss": 4.478597043659311,
+      "tokens_seen": 168755200
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009584353059177533,
+      "loss": 1.9435,
+      "theoretical_loss": 4.4783751394337905,
+      "tokens_seen": 168820736
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009584152457372116,
+      "loss": 1.7251,
+      "theoretical_loss": 4.478153345443854,
+      "tokens_seen": 168886272
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009583951855566701,
+      "loss": 1.964,
+      "theoretical_loss": 4.477931661591995,
+      "tokens_seen": 168951808
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009583751253761283,
+      "loss": 1.7283,
+      "theoretical_loss": 4.47771008778083,
+      "tokens_seen": 169017344
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009583550651955868,
+      "loss": 1.7976,
+      "theoretical_loss": 4.4774886239131,
+      "tokens_seen": 169082880
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009583350050150451,
+      "loss": 2.0903,
+      "theoretical_loss": 4.477267269891671,
+      "tokens_seen": 169148416
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009583149448345035,
+      "loss": 1.8024,
+      "theoretical_loss": 4.477046025619531,
+      "tokens_seen": 169213952
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009582948846539619,
+      "loss": 1.9521,
+      "theoretical_loss": 4.47682489099979,
+      "tokens_seen": 169279488
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009582748244734203,
+      "loss": 1.9436,
+      "theoretical_loss": 4.476603865935683,
+      "tokens_seen": 169345024
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009582547642928786,
+      "loss": 1.9069,
+      "theoretical_loss": 4.476382950330569,
+      "tokens_seen": 169410560
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009582347041123371,
+      "loss": 1.9511,
+      "theoretical_loss": 4.476162144087925,
+      "tokens_seen": 169476096
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009582146439317953,
+      "loss": 1.9537,
+      "theoretical_loss": 4.475941447111355,
+      "tokens_seen": 169541632
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009581945837512538,
+      "loss": 2.0403,
+      "theoretical_loss": 4.475720859304583,
+      "tokens_seen": 169607168
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009581745235707121,
+      "loss": 1.9219,
+      "theoretical_loss": 4.4755003805714555,
+      "tokens_seen": 169672704
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009581544633901705,
+      "loss": 1.8615,
+      "theoretical_loss": 4.47528001081594,
+      "tokens_seen": 169738240
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009581344032096289,
+      "loss": 1.9764,
+      "theoretical_loss": 4.475059749942126,
+      "tokens_seen": 169803776
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009581143430290873,
+      "loss": 1.6865,
+      "theoretical_loss": 4.474839597854226,
+      "tokens_seen": 169869312
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009580942828485456,
+      "loss": 2.0214,
+      "theoretical_loss": 4.474619554456571,
+      "tokens_seen": 169934848
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000958074222668004,
+      "loss": 1.8969,
+      "theoretical_loss": 4.4743996196536155,
+      "tokens_seen": 170000384
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009580541624874624,
+      "loss": 1.9962,
+      "theoretical_loss": 4.474179793349933,
+      "tokens_seen": 170065920
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009580341023069208,
+      "loss": 1.7917,
+      "theoretical_loss": 4.473960075450218,
+      "tokens_seen": 170131456
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009580140421263791,
+      "loss": 1.9129,
+      "theoretical_loss": 4.473740465859285,
+      "tokens_seen": 170196992
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009579939819458375,
+      "loss": 1.993,
+      "theoretical_loss": 4.47352096448207,
+      "tokens_seen": 170262528
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009579739217652958,
+      "loss": 2.0726,
+      "theoretical_loss": 4.473301571223629,
+      "tokens_seen": 170328064
+    },
+    {
+      "epoch": 0.05,
+      "objective/train/advantage_avg": 0.4946981370449066,
+      "objective/train/docs_used": 69022,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.9940156936645508,
+      "objective/train/original_loss": 1.9940155744552612,
+      "objective/train/theoretical_loss": 4.473082285989134,
+      "objective/train/tokens_used": 190853600,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24500080943107605,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0507153272628784,
+      "objective/train/weighted_lm_loss": 2.094956636428833,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.473082285989134,
+      "tokens_seen": 170393600
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009579538615847543,
+      "loss": 1.9899,
+      "theoretical_loss": 4.473082285989134,
+      "tokens_seen": 170393600
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009579338014042127,
+      "loss": 2.0756,
+      "theoretical_loss": 4.472863108683884,
+      "tokens_seen": 170459136
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000957913741223671,
+      "loss": 1.9973,
+      "theoretical_loss": 4.47264403921329,
+      "tokens_seen": 170524672
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009578936810431295,
+      "loss": 1.8127,
+      "theoretical_loss": 4.472425077482886,
+      "tokens_seen": 170590208
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009578736208625878,
+      "loss": 1.8438,
+      "theoretical_loss": 4.472206223398325,
+      "tokens_seen": 170655744
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009578535606820462,
+      "loss": 1.9837,
+      "theoretical_loss": 4.471987476865376,
+      "tokens_seen": 170721280
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009578335005015046,
+      "loss": 1.8991,
+      "theoretical_loss": 4.471768837789931,
+      "tokens_seen": 170786816
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009578134403209629,
+      "loss": 1.8756,
+      "theoretical_loss": 4.4715503060779955,
+      "tokens_seen": 170852352
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009577933801404213,
+      "loss": 1.8956,
+      "theoretical_loss": 4.471331881635698,
+      "tokens_seen": 170917888
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009577733199598798,
+      "loss": 1.9264,
+      "theoretical_loss": 4.471113564369282,
+      "tokens_seen": 170983424
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000957753259779338,
+      "loss": 1.7323,
+      "theoretical_loss": 4.47089535418511,
+      "tokens_seen": 171048960
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009577331995987965,
+      "loss": 2.1019,
+      "theoretical_loss": 4.470677250989661,
+      "tokens_seen": 171114496
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009577131394182547,
+      "loss": 1.8691,
+      "theoretical_loss": 4.470459254689533,
+      "tokens_seen": 171180032
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009576930792377132,
+      "loss": 1.8631,
+      "theoretical_loss": 4.470241365191441,
+      "tokens_seen": 171245568
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009576730190571716,
+      "loss": 1.9141,
+      "theoretical_loss": 4.4700235824022165,
+      "tokens_seen": 171311104
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009576529588766299,
+      "loss": 1.9757,
+      "theoretical_loss": 4.469805906228807,
+      "tokens_seen": 171376640
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009576328986960883,
+      "loss": 1.8581,
+      "theoretical_loss": 4.469588336578277,
+      "tokens_seen": 171442176
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009576128385155466,
+      "loss": 1.905,
+      "theoretical_loss": 4.469370873357811,
+      "tokens_seen": 171507712
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000957592778335005,
+      "loss": 1.7967,
+      "theoretical_loss": 4.469153516474703,
+      "tokens_seen": 171573248
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009575727181544635,
+      "loss": 1.893,
+      "theoretical_loss": 4.468936265836371,
+      "tokens_seen": 171638784
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009575526579739217,
+      "loss": 1.8793,
+      "theoretical_loss": 4.468719121350343,
+      "tokens_seen": 171704320
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009575325977933802,
+      "loss": 1.816,
+      "theoretical_loss": 4.468502082924266,
+      "tokens_seen": 171769856
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009575125376128386,
+      "loss": 1.905,
+      "theoretical_loss": 4.468285150465899,
+      "tokens_seen": 171835392
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009574924774322969,
+      "loss": 1.888,
+      "theoretical_loss": 4.468068323883122,
+      "tokens_seen": 171900928
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009574724172517553,
+      "loss": 1.8005,
+      "theoretical_loss": 4.467851603083923,
+      "tokens_seen": 171966464
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009574523570712137,
+      "loss": 1.8414,
+      "theoretical_loss": 4.467634987976411,
+      "tokens_seen": 172032000
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000957432296890672,
+      "loss": 2.0145,
+      "theoretical_loss": 4.467418478468808,
+      "tokens_seen": 172097536
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009574122367101305,
+      "loss": 1.9675,
+      "theoretical_loss": 4.467202074469449,
+      "tokens_seen": 172163072
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009573921765295887,
+      "loss": 1.8405,
+      "theoretical_loss": 4.466985775886784,
+      "tokens_seen": 172228608
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009573721163490472,
+      "loss": 1.9227,
+      "theoretical_loss": 4.4667695826293805,
+      "tokens_seen": 172294144
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009573520561685055,
+      "loss": 1.8589,
+      "theoretical_loss": 4.4665534946059156,
+      "tokens_seen": 172359680
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009573319959879639,
+      "loss": 1.9755,
+      "theoretical_loss": 4.4663375117251825,
+      "tokens_seen": 172425216
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009573119358074223,
+      "loss": 1.8297,
+      "theoretical_loss": 4.466121633896087,
+      "tokens_seen": 172490752
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009572918756268807,
+      "loss": 2.0146,
+      "theoretical_loss": 4.4659058610276485,
+      "tokens_seen": 172556288
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000957271815446339,
+      "loss": 1.837,
+      "theoretical_loss": 4.465690193029003,
+      "tokens_seen": 172621824
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009572517552657975,
+      "loss": 2.0488,
+      "theoretical_loss": 4.465474629809394,
+      "tokens_seen": 172687360
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009572316950852557,
+      "loss": 1.8981,
+      "theoretical_loss": 4.465259171278182,
+      "tokens_seen": 172752896
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009572116349047142,
+      "loss": 1.9211,
+      "theoretical_loss": 4.4650438173448395,
+      "tokens_seen": 172818432
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009571915747241725,
+      "loss": 1.886,
+      "theoretical_loss": 4.464828567918951,
+      "tokens_seen": 172883968
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009571715145436309,
+      "loss": 1.8676,
+      "theoretical_loss": 4.464613422910213,
+      "tokens_seen": 172949504
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009571514543630893,
+      "loss": 1.8263,
+      "theoretical_loss": 4.464398382228435,
+      "tokens_seen": 173015040
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009571313941825477,
+      "loss": 1.9294,
+      "theoretical_loss": 4.464183445783537,
+      "tokens_seen": 173080576
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000957111334002006,
+      "loss": 1.9157,
+      "theoretical_loss": 4.463968613485555,
+      "tokens_seen": 173146112
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009570912738214644,
+      "loss": 2.0705,
+      "theoretical_loss": 4.463753885244631,
+      "tokens_seen": 173211648
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009570712136409228,
+      "loss": 1.8366,
+      "theoretical_loss": 4.463539260971023,
+      "tokens_seen": 173277184
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009570511534603812,
+      "loss": 2.0209,
+      "theoretical_loss": 4.463324740575098,
+      "tokens_seen": 173342720
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009570310932798395,
+      "loss": 1.8134,
+      "theoretical_loss": 4.463110323967335,
+      "tokens_seen": 173408256
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000957011033099298,
+      "loss": 2.0315,
+      "theoretical_loss": 4.462896011058323,
+      "tokens_seen": 173473792
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009569909729187562,
+      "loss": 1.936,
+      "theoretical_loss": 4.462681801758762,
+      "tokens_seen": 173539328
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009569709127382147,
+      "loss": 1.8807,
+      "theoretical_loss": 4.4624676959794645,
+      "tokens_seen": 173604864
+    },
+    {
+      "epoch": 0.05,
+      "objective/train/advantage_avg": 0.4944310486316681,
+      "objective/train/docs_used": 70257,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.9424092769622803,
+      "objective/train/original_loss": 1.9424092769622803,
+      "objective/train/theoretical_loss": 4.462253693631352,
+      "objective/train/tokens_used": 194130400,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24486778676509857,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506879091262817,
+      "objective/train/weighted_lm_loss": 2.0407896041870117,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9834713935852051,
+      "theoretical_loss": 4.462253693631352,
+      "tokens_seen": 173670400
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000956950852557673,
+      "loss": 1.9977,
+      "theoretical_loss": 4.462253693631352,
+      "tokens_seen": 173670400
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009569307923771314,
+      "loss": 1.9687,
+      "theoretical_loss": 4.462039794625454,
+      "tokens_seen": 173735936
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009569107321965898,
+      "loss": 1.9434,
+      "theoretical_loss": 4.461825998872914,
+      "tokens_seen": 173801472
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009568906720160482,
+      "loss": 1.8063,
+      "theoretical_loss": 4.461612306284984,
+      "tokens_seen": 173867008
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009568706118355065,
+      "loss": 1.8382,
+      "theoretical_loss": 4.461398716773022,
+      "tokens_seen": 173932544
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000956850551654965,
+      "loss": 1.9081,
+      "theoretical_loss": 4.461185230248503,
+      "tokens_seen": 173998080
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009568304914744232,
+      "loss": 1.7065,
+      "theoretical_loss": 4.460971846623005,
+      "tokens_seen": 174063616
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009568104312938817,
+      "loss": 1.854,
+      "theoretical_loss": 4.4607585658082165,
+      "tokens_seen": 174129152
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00095679037111334,
+      "loss": 1.8371,
+      "theoretical_loss": 4.460545387715937,
+      "tokens_seen": 174194688
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009567703109327984,
+      "loss": 1.9075,
+      "theoretical_loss": 4.460332312258075,
+      "tokens_seen": 174260224
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009567502507522568,
+      "loss": 1.9443,
+      "theoretical_loss": 4.460119339346643,
+      "tokens_seen": 174325760
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009567301905717151,
+      "loss": 1.9899,
+      "theoretical_loss": 4.4599064688937675,
+      "tokens_seen": 174391296
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009567101303911735,
+      "loss": 1.7698,
+      "theoretical_loss": 4.45969370081168,
+      "tokens_seen": 174456832
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000956690070210632,
+      "loss": 1.7344,
+      "theoretical_loss": 4.459481035012722,
+      "tokens_seen": 174522368
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009566700100300902,
+      "loss": 1.8361,
+      "theoretical_loss": 4.45926847140934,
+      "tokens_seen": 174587904
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009566499498495487,
+      "loss": 1.8969,
+      "theoretical_loss": 4.459056009914093,
+      "tokens_seen": 174653440
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009566298896690069,
+      "loss": 1.7649,
+      "theoretical_loss": 4.458843650439643,
+      "tokens_seen": 174718976
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009566098294884654,
+      "loss": 1.8389,
+      "theoretical_loss": 4.458631392898763,
+      "tokens_seen": 174784512
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009565897693079238,
+      "loss": 1.9442,
+      "theoretical_loss": 4.45841923720433,
+      "tokens_seen": 174850048
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009565697091273821,
+      "loss": 2.0796,
+      "theoretical_loss": 4.45820718326933,
+      "tokens_seen": 174915584
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009565496489468405,
+      "loss": 2.0647,
+      "theoretical_loss": 4.457995231006857,
+      "tokens_seen": 174981120
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000956529588766299,
+      "loss": 1.8885,
+      "theoretical_loss": 4.457783380330108,
+      "tokens_seen": 175046656
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009565095285857572,
+      "loss": 1.8958,
+      "theoretical_loss": 4.4575716311523905,
+      "tokens_seen": 175112192
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009564894684052157,
+      "loss": 1.8034,
+      "theoretical_loss": 4.457359983387116,
+      "tokens_seen": 175177728
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000956469408224674,
+      "loss": 1.9068,
+      "theoretical_loss": 4.457148436947804,
+      "tokens_seen": 175243264
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009564493480441324,
+      "loss": 1.9427,
+      "theoretical_loss": 4.456936991748078,
+      "tokens_seen": 175308800
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009564292878635909,
+      "loss": 1.9902,
+      "theoretical_loss": 4.456725647701669,
+      "tokens_seen": 175374336
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009564092276830491,
+      "loss": 1.8762,
+      "theoretical_loss": 4.456514404722413,
+      "tokens_seen": 175439872
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009563891675025076,
+      "loss": 1.9261,
+      "theoretical_loss": 4.456303262724252,
+      "tokens_seen": 175505408
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009563691073219659,
+      "loss": 1.8886,
+      "theoretical_loss": 4.456092221621233,
+      "tokens_seen": 175570944
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009563490471414243,
+      "loss": 1.8757,
+      "theoretical_loss": 4.455881281327508,
+      "tokens_seen": 175636480
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009563289869608827,
+      "loss": 1.9739,
+      "theoretical_loss": 4.455670441757334,
+      "tokens_seen": 175702016
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009563089267803411,
+      "loss": 1.8367,
+      "theoretical_loss": 4.455459702825074,
+      "tokens_seen": 175767552
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009562888665997994,
+      "loss": 1.9812,
+      "theoretical_loss": 4.455249064445195,
+      "tokens_seen": 175833088
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009562688064192579,
+      "loss": 1.9976,
+      "theoretical_loss": 4.4550385265322685,
+      "tokens_seen": 175898624
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009562487462387161,
+      "loss": 1.9515,
+      "theoretical_loss": 4.454828089000969,
+      "tokens_seen": 175964160
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009562286860581746,
+      "loss": 1.8709,
+      "theoretical_loss": 4.454617751766078,
+      "tokens_seen": 176029696
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009562086258776329,
+      "loss": 1.8376,
+      "theoretical_loss": 4.454407514742479,
+      "tokens_seen": 176095232
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009561885656970913,
+      "loss": 1.9017,
+      "theoretical_loss": 4.45419737784516,
+      "tokens_seen": 176160768
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009561685055165497,
+      "loss": 1.9527,
+      "theoretical_loss": 4.4539873409892135,
+      "tokens_seen": 176226304
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009561484453360081,
+      "loss": 2.0055,
+      "theoretical_loss": 4.453777404089832,
+      "tokens_seen": 176291840
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009561283851554664,
+      "loss": 1.8616,
+      "theoretical_loss": 4.453567567062318,
+      "tokens_seen": 176357376
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009561083249749248,
+      "loss": 1.9151,
+      "theoretical_loss": 4.45335782982207,
+      "tokens_seen": 176422912
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009560882647943832,
+      "loss": 1.865,
+      "theoretical_loss": 4.453148192284595,
+      "tokens_seen": 176488448
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009560682046138416,
+      "loss": 2.0749,
+      "theoretical_loss": 4.4529386543655,
+      "tokens_seen": 176553984
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009560481444332999,
+      "loss": 2.0762,
+      "theoretical_loss": 4.452729215980495,
+      "tokens_seen": 176619520
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009560280842527584,
+      "loss": 1.8006,
+      "theoretical_loss": 4.452519877045393,
+      "tokens_seen": 176685056
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009560080240722166,
+      "loss": 2.0119,
+      "theoretical_loss": 4.452310637476112,
+      "tokens_seen": 176750592
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009559879638916751,
+      "loss": 1.9248,
+      "theoretical_loss": 4.452101497188666,
+      "tokens_seen": 176816128
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009559679037111334,
+      "loss": 2.002,
+      "theoretical_loss": 4.451892456099176,
+      "tokens_seen": 176881664
+    },
+    {
+      "epoch": 0.05,
+      "objective/train/advantage_avg": 0.49354079365730286,
+      "objective/train/docs_used": 71338,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.6465367078781128,
+      "objective/train/original_loss": 1.6465368270874023,
+      "objective/train/theoretical_loss": 4.451683514123864,
+      "objective/train/tokens_used": 197407200,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2438439577817917,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505937337875366,
+      "objective/train/weighted_lm_loss": 1.7298942804336548,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.451683514123864,
+      "tokens_seen": 176947200
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009559478435305918,
+      "loss": 1.8323,
+      "theoretical_loss": 4.451683514123864,
+      "tokens_seen": 176947200
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009559277833500502,
+      "loss": 1.8433,
+      "theoretical_loss": 4.451474671179055,
+      "tokens_seen": 177012736
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009559077231695086,
+      "loss": 1.9551,
+      "theoretical_loss": 4.451265927181171,
+      "tokens_seen": 177078272
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009558876629889669,
+      "loss": 2.0386,
+      "theoretical_loss": 4.451057282046741,
+      "tokens_seen": 177143808
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009558676028084254,
+      "loss": 1.8875,
+      "theoretical_loss": 4.450848735692391,
+      "tokens_seen": 177209344
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009558475426278836,
+      "loss": 2.0172,
+      "theoretical_loss": 4.45064028803485,
+      "tokens_seen": 177274880
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009558274824473421,
+      "loss": 1.8569,
+      "theoretical_loss": 4.450431938990949,
+      "tokens_seen": 177340416
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009558074222668004,
+      "loss": 1.8493,
+      "theoretical_loss": 4.450223688477617,
+      "tokens_seen": 177405952
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009557873620862588,
+      "loss": 1.9685,
+      "theoretical_loss": 4.450015536411886,
+      "tokens_seen": 177471488
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009557673019057172,
+      "loss": 1.906,
+      "theoretical_loss": 4.449807482710888,
+      "tokens_seen": 177537024
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009557472417251755,
+      "loss": 1.8314,
+      "theoretical_loss": 4.449599527291854,
+      "tokens_seen": 177602560
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009557271815446339,
+      "loss": 1.8091,
+      "theoretical_loss": 4.449391670072117,
+      "tokens_seen": 177668096
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009557071213640924,
+      "loss": 1.9665,
+      "theoretical_loss": 4.449183910969108,
+      "tokens_seen": 177733632
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009556870611835506,
+      "loss": 1.9209,
+      "theoretical_loss": 4.44897624990036,
+      "tokens_seen": 177799168
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009556670010030091,
+      "loss": 1.7531,
+      "theoretical_loss": 4.448768686783504,
+      "tokens_seen": 177864704
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009556469408224673,
+      "loss": 1.8657,
+      "theoretical_loss": 4.448561221536272,
+      "tokens_seen": 177930240
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009556268806419258,
+      "loss": 1.8918,
+      "theoretical_loss": 4.448353854076494,
+      "tokens_seen": 177995776
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009556068204613842,
+      "loss": 1.9234,
+      "theoretical_loss": 4.4481465843220995,
+      "tokens_seen": 178061312
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009555867602808425,
+      "loss": 1.9232,
+      "theoretical_loss": 4.447939412191117,
+      "tokens_seen": 178126848
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009555667001003009,
+      "loss": 1.9091,
+      "theoretical_loss": 4.447732337601675,
+      "tokens_seen": 178192384
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009555466399197594,
+      "loss": 1.8798,
+      "theoretical_loss": 4.4475253604719995,
+      "tokens_seen": 178257920
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009555265797392176,
+      "loss": 1.7858,
+      "theoretical_loss": 4.447318480720416,
+      "tokens_seen": 178323456
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009555065195586761,
+      "loss": 2.0034,
+      "theoretical_loss": 4.447111698265346,
+      "tokens_seen": 178388992
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009554864593781343,
+      "loss": 1.7928,
+      "theoretical_loss": 4.446905013025313,
+      "tokens_seen": 178454528
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009554663991975928,
+      "loss": 1.7196,
+      "theoretical_loss": 4.446698424918937,
+      "tokens_seen": 178520064
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009554463390170512,
+      "loss": 1.9755,
+      "theoretical_loss": 4.446491933864936,
+      "tokens_seen": 178585600
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009554262788365095,
+      "loss": 1.849,
+      "theoretical_loss": 4.4462855397821235,
+      "tokens_seen": 178651136
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009554062186559679,
+      "loss": 1.9696,
+      "theoretical_loss": 4.4460792425894144,
+      "tokens_seen": 178716672
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009553861584754263,
+      "loss": 1.8554,
+      "theoretical_loss": 4.44587304220582,
+      "tokens_seen": 178782208
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009553660982948846,
+      "loss": 1.7727,
+      "theoretical_loss": 4.445666938550447,
+      "tokens_seen": 178847744
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009553460381143431,
+      "loss": 1.8321,
+      "theoretical_loss": 4.445460931542502,
+      "tokens_seen": 178913280
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009553259779338014,
+      "loss": 1.6822,
+      "theoretical_loss": 4.4452550211012865,
+      "tokens_seen": 178978816
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009553059177532598,
+      "loss": 1.9822,
+      "theoretical_loss": 4.4450492071462,
+      "tokens_seen": 179044352
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009552858575727181,
+      "loss": 1.8321,
+      "theoretical_loss": 4.444843489596739,
+      "tokens_seen": 179109888
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009552657973921765,
+      "loss": 2.0049,
+      "theoretical_loss": 4.444637868372495,
+      "tokens_seen": 179175424
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009552457372116349,
+      "loss": 1.9795,
+      "theoretical_loss": 4.444432343393158,
+      "tokens_seen": 179240960
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009552256770310933,
+      "loss": 2.046,
+      "theoretical_loss": 4.444226914578513,
+      "tokens_seen": 179306496
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009552056168505516,
+      "loss": 2.0291,
+      "theoretical_loss": 4.444021581848442,
+      "tokens_seen": 179372032
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009551855566700101,
+      "loss": 2.0813,
+      "theoretical_loss": 4.443816345122922,
+      "tokens_seen": 179437568
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009551654964894684,
+      "loss": 1.7821,
+      "theoretical_loss": 4.443611204322026,
+      "tokens_seen": 179503104
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009551454363089268,
+      "loss": 1.916,
+      "theoretical_loss": 4.4434061593659235,
+      "tokens_seen": 179568640
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009551253761283851,
+      "loss": 1.9274,
+      "theoretical_loss": 4.4432012101748795,
+      "tokens_seen": 179634176
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009551053159478436,
+      "loss": 1.8894,
+      "theoretical_loss": 4.442996356669253,
+      "tokens_seen": 179699712
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009550852557673019,
+      "loss": 1.9832,
+      "theoretical_loss": 4.4427915987695,
+      "tokens_seen": 179765248
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009550651955867603,
+      "loss": 1.8851,
+      "theoretical_loss": 4.442586936396171,
+      "tokens_seen": 179830784
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009550451354062186,
+      "loss": 1.8177,
+      "theoretical_loss": 4.442382369469909,
+      "tokens_seen": 179896320
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000955025075225677,
+      "loss": 1.9614,
+      "theoretical_loss": 4.442177897911458,
+      "tokens_seen": 179961856
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009550050150451354,
+      "loss": 1.9223,
+      "theoretical_loss": 4.441973521641649,
+      "tokens_seen": 180027392
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009549849548645938,
+      "loss": 1.9658,
+      "theoretical_loss": 4.441769240581412,
+      "tokens_seen": 180092928
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009549648946840521,
+      "loss": 2.0016,
+      "theoretical_loss": 4.441565054651772,
+      "tokens_seen": 180158464
+    },
+    {
+      "epoch": 0.05,
+      "objective/train/advantage_avg": 0.4942334294319153,
+      "objective/train/docs_used": 72550,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.7501457929611206,
+      "objective/train/original_loss": 1.7501459121704102,
+      "objective/train/theoretical_loss": 4.441360963773846,
+      "objective/train/tokens_used": 200684000,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2446201890707016,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506670475006104,
+      "objective/train/weighted_lm_loss": 1.8388445377349854,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.441360963773846,
+      "tokens_seen": 180224000
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009549448345035106,
+      "loss": 1.9136,
+      "theoretical_loss": 4.441360963773846,
+      "tokens_seen": 180224000
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000954924774322969,
+      "loss": 2.0606,
+      "theoretical_loss": 4.441156967868846,
+      "tokens_seen": 180289536
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009549047141424273,
+      "loss": 2.0416,
+      "theoretical_loss": 4.440953066858077,
+      "tokens_seen": 180355072
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009548846539618858,
+      "loss": 2.0186,
+      "theoretical_loss": 4.440749260662939,
+      "tokens_seen": 180420608
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000954864593781344,
+      "loss": 1.8847,
+      "theoretical_loss": 4.440545549204925,
+      "tokens_seen": 180486144
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009548445336008025,
+      "loss": 1.7716,
+      "theoretical_loss": 4.440341932405623,
+      "tokens_seen": 180551680
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009548244734202608,
+      "loss": 1.8777,
+      "theoretical_loss": 4.4401384101867105,
+      "tokens_seen": 180617216
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009548044132397192,
+      "loss": 1.8929,
+      "theoretical_loss": 4.439934982469962,
+      "tokens_seen": 180682752
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009547843530591776,
+      "loss": 1.9978,
+      "theoretical_loss": 4.439731649177244,
+      "tokens_seen": 180748288
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009547642928786359,
+      "loss": 1.9419,
+      "theoretical_loss": 4.439528410230514,
+      "tokens_seen": 180813824
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009547442326980943,
+      "loss": 1.767,
+      "theoretical_loss": 4.439325265551826,
+      "tokens_seen": 180879360
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009547241725175528,
+      "loss": 1.9141,
+      "theoretical_loss": 4.439122215063323,
+      "tokens_seen": 180944896
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.000954704112337011,
+      "loss": 1.9685,
+      "theoretical_loss": 4.438919258687242,
+      "tokens_seen": 181010432
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009546840521564695,
+      "loss": 1.826,
+      "theoretical_loss": 4.438716396345913,
+      "tokens_seen": 181075968
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009546639919759277,
+      "loss": 1.9419,
+      "theoretical_loss": 4.438513627961757,
+      "tokens_seen": 181141504
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009546439317953862,
+      "loss": 1.8724,
+      "theoretical_loss": 4.438310953457286,
+      "tokens_seen": 181207040
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009546238716148446,
+      "loss": 1.9956,
+      "theoretical_loss": 4.438108372755108,
+      "tokens_seen": 181272576
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009546038114343029,
+      "loss": 1.9181,
+      "theoretical_loss": 4.43790588577792,
+      "tokens_seen": 181338112
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009545837512537613,
+      "loss": 1.9048,
+      "theoretical_loss": 4.437703492448509,
+      "tokens_seen": 181403648
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0009545636910732198,
+      "loss": 2.0959,
+      "theoretical_loss": 4.437501192689755,
+      "tokens_seen": 181469184
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000954543630892678,
+      "loss": 1.9178,
+      "theoretical_loss": 4.437298986424631,
+      "tokens_seen": 181534720
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009545235707121365,
+      "loss": 1.9351,
+      "theoretical_loss": 4.437096873576199,
+      "tokens_seen": 181600256
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009545035105315947,
+      "loss": 1.8552,
+      "theoretical_loss": 4.436894854067614,
+      "tokens_seen": 181665792
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009544834503510532,
+      "loss": 1.8738,
+      "theoretical_loss": 4.4366929278221185,
+      "tokens_seen": 181731328
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009544633901705116,
+      "loss": 2.0089,
+      "theoretical_loss": 4.436491094763049,
+      "tokens_seen": 181796864
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009544433299899699,
+      "loss": 1.8173,
+      "theoretical_loss": 4.436289354813832,
+      "tokens_seen": 181862400
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009544232698094283,
+      "loss": 1.9785,
+      "theoretical_loss": 4.436087707897984,
+      "tokens_seen": 181927936
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009544032096288867,
+      "loss": 1.8956,
+      "theoretical_loss": 4.435886153939111,
+      "tokens_seen": 181993472
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000954383149448345,
+      "loss": 1.901,
+      "theoretical_loss": 4.435684692860912,
+      "tokens_seen": 182059008
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009543630892678035,
+      "loss": 1.9325,
+      "theoretical_loss": 4.435483324587172,
+      "tokens_seen": 182124544
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009543430290872618,
+      "loss": 1.9463,
+      "theoretical_loss": 4.435282049041769,
+      "tokens_seen": 182190080
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009543229689067202,
+      "loss": 1.8567,
+      "theoretical_loss": 4.43508086614867,
+      "tokens_seen": 182255616
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009543029087261785,
+      "loss": 2.0057,
+      "theoretical_loss": 4.4348797758319325,
+      "tokens_seen": 182321152
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000954282848545637,
+      "loss": 1.8366,
+      "theoretical_loss": 4.434678778015701,
+      "tokens_seen": 182386688
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009542627883650953,
+      "loss": 1.8694,
+      "theoretical_loss": 4.434477872624212,
+      "tokens_seen": 182452224
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009542427281845537,
+      "loss": 2.0037,
+      "theoretical_loss": 4.434277059581789,
+      "tokens_seen": 182517760
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000954222668004012,
+      "loss": 1.925,
+      "theoretical_loss": 4.434076338812848,
+      "tokens_seen": 182583296
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009542026078234705,
+      "loss": 1.9509,
+      "theoretical_loss": 4.43387571024189,
+      "tokens_seen": 182648832
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009541825476429288,
+      "loss": 1.9147,
+      "theoretical_loss": 4.433675173793507,
+      "tokens_seen": 182714368
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009541624874623872,
+      "loss": 2.0302,
+      "theoretical_loss": 4.433474729392379,
+      "tokens_seen": 182779904
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009541424272818455,
+      "loss": 2.053,
+      "theoretical_loss": 4.433274376963276,
+      "tokens_seen": 182845440
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000954122367101304,
+      "loss": 1.8329,
+      "theoretical_loss": 4.4330741164310545,
+      "tokens_seen": 182910976
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009541023069207623,
+      "loss": 1.7745,
+      "theoretical_loss": 4.43287394772066,
+      "tokens_seen": 182976512
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009540822467402207,
+      "loss": 1.9235,
+      "theoretical_loss": 4.432673870757127,
+      "tokens_seen": 183042048
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000954062186559679,
+      "loss": 1.956,
+      "theoretical_loss": 4.432473885465576,
+      "tokens_seen": 183107584
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009540421263791374,
+      "loss": 2.0946,
+      "theoretical_loss": 4.432273991771218,
+      "tokens_seen": 183173120
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009540220661985958,
+      "loss": 1.9536,
+      "theoretical_loss": 4.43207418959935,
+      "tokens_seen": 183238656
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009540020060180542,
+      "loss": 1.8331,
+      "theoretical_loss": 4.431874478875356,
+      "tokens_seen": 183304192
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009539819458375125,
+      "loss": 1.8043,
+      "theoretical_loss": 4.431674859524709,
+      "tokens_seen": 183369728
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000953961885656971,
+      "loss": 1.8608,
+      "theoretical_loss": 4.43147533147297,
+      "tokens_seen": 183435264
+    },
+    {
+      "epoch": 0.06,
+      "objective/train/advantage_avg": 0.4887767434120178,
+      "objective/train/docs_used": 73723,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.8926588296890259,
+      "objective/train/original_loss": 1.8926584720611572,
+      "objective/train/theoretical_loss": 4.431275894645784,
+      "objective/train/tokens_used": 203960800,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24062563478946686,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0501008033752441,
+      "objective/train/weighted_lm_loss": 1.9868710041046143,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.431275894645784,
+      "tokens_seen": 183500800
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009539418254764293,
+      "loss": 1.91,
+      "theoretical_loss": 4.431275894645784,
+      "tokens_seen": 183500800
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009539217652958877,
+      "loss": 1.9714,
+      "theoretical_loss": 4.431076548968887,
+      "tokens_seen": 183566336
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000953901705115346,
+      "loss": 1.9316,
+      "theoretical_loss": 4.430877294368099,
+      "tokens_seen": 183631872
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009538816449348044,
+      "loss": 1.8594,
+      "theoretical_loss": 4.430678130769329,
+      "tokens_seen": 183697408
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009538615847542628,
+      "loss": 1.8918,
+      "theoretical_loss": 4.43047905809857,
+      "tokens_seen": 183762944
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009538415245737212,
+      "loss": 1.9531,
+      "theoretical_loss": 4.430280076281904,
+      "tokens_seen": 183828480
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009538214643931795,
+      "loss": 1.7753,
+      "theoretical_loss": 4.430081185245498,
+      "tokens_seen": 183894016
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000953801404212638,
+      "loss": 1.9356,
+      "theoretical_loss": 4.429882384915607,
+      "tokens_seen": 183959552
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009537813440320962,
+      "loss": 1.9047,
+      "theoretical_loss": 4.42968367521857,
+      "tokens_seen": 184025088
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009537612838515547,
+      "loss": 1.9452,
+      "theoretical_loss": 4.4294850560808126,
+      "tokens_seen": 184090624
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009537412236710131,
+      "loss": 1.8756,
+      "theoretical_loss": 4.429286527428848,
+      "tokens_seen": 184156160
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009537211634904714,
+      "loss": 1.9095,
+      "theoretical_loss": 4.429088089189273,
+      "tokens_seen": 184221696
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009537011033099298,
+      "loss": 1.9531,
+      "theoretical_loss": 4.428889741288771,
+      "tokens_seen": 184287232
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009536810431293881,
+      "loss": 2.0226,
+      "theoretical_loss": 4.428691483654109,
+      "tokens_seen": 184352768
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009536609829488465,
+      "loss": 1.9104,
+      "theoretical_loss": 4.428493316212144,
+      "tokens_seen": 184418304
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000953640922768305,
+      "loss": 1.9173,
+      "theoretical_loss": 4.428295238889815,
+      "tokens_seen": 184483840
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009536208625877632,
+      "loss": 1.9288,
+      "theoretical_loss": 4.428097251614145,
+      "tokens_seen": 184549376
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009536008024072217,
+      "loss": 1.8834,
+      "theoretical_loss": 4.4278993543122445,
+      "tokens_seen": 184614912
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009535807422266801,
+      "loss": 2.0234,
+      "theoretical_loss": 4.427701546911307,
+      "tokens_seen": 184680448
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009535606820461384,
+      "loss": 1.846,
+      "theoretical_loss": 4.427503829338613,
+      "tokens_seen": 184745984
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009535406218655968,
+      "loss": 1.8591,
+      "theoretical_loss": 4.427306201521524,
+      "tokens_seen": 184811520
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009535205616850551,
+      "loss": 1.8564,
+      "theoretical_loss": 4.427108663387489,
+      "tokens_seen": 184877056
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009535005015045135,
+      "loss": 1.85,
+      "theoretical_loss": 4.426911214864043,
+      "tokens_seen": 184942592
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000953480441323972,
+      "loss": 1.8551,
+      "theoretical_loss": 4.426713855878798,
+      "tokens_seen": 185008128
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009534603811434302,
+      "loss": 2.0234,
+      "theoretical_loss": 4.426516586359458,
+      "tokens_seen": 185073664
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009534403209628887,
+      "loss": 2.0515,
+      "theoretical_loss": 4.426319406233807,
+      "tokens_seen": 185139200
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009534202607823471,
+      "loss": 1.9792,
+      "theoretical_loss": 4.426122315429713,
+      "tokens_seen": 185204736
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009534002006018054,
+      "loss": 1.8541,
+      "theoretical_loss": 4.425925313875129,
+      "tokens_seen": 185270272
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009533801404212639,
+      "loss": 1.8656,
+      "theoretical_loss": 4.425728401498089,
+      "tokens_seen": 185335808
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009533600802407222,
+      "loss": 2.041,
+      "theoretical_loss": 4.425531578226715,
+      "tokens_seen": 185401344
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009533400200601806,
+      "loss": 1.9088,
+      "theoretical_loss": 4.425334843989207,
+      "tokens_seen": 185466880
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009533199598796389,
+      "loss": 2.0045,
+      "theoretical_loss": 4.425138198713853,
+      "tokens_seen": 185532416
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009532998996990973,
+      "loss": 1.8641,
+      "theoretical_loss": 4.424941642329019,
+      "tokens_seen": 185597952
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009532798395185557,
+      "loss": 2.0065,
+      "theoretical_loss": 4.42474517476316,
+      "tokens_seen": 185663488
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009532597793380141,
+      "loss": 1.8994,
+      "theoretical_loss": 4.424548795944808,
+      "tokens_seen": 185729024
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009532397191574724,
+      "loss": 1.8996,
+      "theoretical_loss": 4.424352505802582,
+      "tokens_seen": 185794560
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009532196589769309,
+      "loss": 1.898,
+      "theoretical_loss": 4.42415630426518,
+      "tokens_seen": 185860096
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009531995987963892,
+      "loss": 1.9314,
+      "theoretical_loss": 4.423960191261387,
+      "tokens_seen": 185925632
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009531795386158476,
+      "loss": 1.8369,
+      "theoretical_loss": 4.423764166720066,
+      "tokens_seen": 185991168
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009531594784353059,
+      "loss": 1.8829,
+      "theoretical_loss": 4.4235682305701625,
+      "tokens_seen": 186056704
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009531394182547644,
+      "loss": 1.8661,
+      "theoretical_loss": 4.423372382740707,
+      "tokens_seen": 186122240
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009531193580742227,
+      "loss": 1.8,
+      "theoretical_loss": 4.423176623160811,
+      "tokens_seen": 186187776
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009530992978936811,
+      "loss": 1.8912,
+      "theoretical_loss": 4.422980951759665,
+      "tokens_seen": 186253312
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009530792377131394,
+      "loss": 1.8467,
+      "theoretical_loss": 4.4227853684665455,
+      "tokens_seen": 186318848
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009530591775325978,
+      "loss": 1.8339,
+      "theoretical_loss": 4.422589873210806,
+      "tokens_seen": 186384384
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009530391173520562,
+      "loss": 1.8041,
+      "theoretical_loss": 4.422394465921887,
+      "tokens_seen": 186449920
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009530190571715146,
+      "loss": 1.8145,
+      "theoretical_loss": 4.422199146529305,
+      "tokens_seen": 186515456
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009529989969909729,
+      "loss": 1.8298,
+      "theoretical_loss": 4.42200391496266,
+      "tokens_seen": 186580992
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009529789368104314,
+      "loss": 1.9237,
+      "theoretical_loss": 4.4218087711516345,
+      "tokens_seen": 186646528
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009529588766298897,
+      "loss": 1.9574,
+      "theoretical_loss": 4.421613715025988,
+      "tokens_seen": 186712064
+    },
+    {
+      "epoch": 0.06,
+      "objective/train/advantage_avg": 0.49304935336112976,
+      "objective/train/docs_used": 74904,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.7934205532073975,
+      "objective/train/original_loss": 1.7934205532073975,
+      "objective/train/theoretical_loss": 4.421418746515567,
+      "objective/train/tokens_used": 207237600,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24341203272342682,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505423545837402,
+      "objective/train/weighted_lm_loss": 1.8839329481124878,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.421418746515567,
+      "tokens_seen": 186777600
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009529388164493481,
+      "loss": 1.8573,
+      "theoretical_loss": 4.421418746515567,
+      "tokens_seen": 186777600
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009529187562688064,
+      "loss": 2.0024,
+      "theoretical_loss": 4.421223865550291,
+      "tokens_seen": 186843136
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009528986960882648,
+      "loss": 1.9979,
+      "theoretical_loss": 4.421029072060167,
+      "tokens_seen": 186908672
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009528786359077232,
+      "loss": 2.0378,
+      "theoretical_loss": 4.420834365975278,
+      "tokens_seen": 186974208
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009528585757271816,
+      "loss": 1.8248,
+      "theoretical_loss": 4.42063974722579,
+      "tokens_seen": 187039744
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009528385155466399,
+      "loss": 1.8471,
+      "theoretical_loss": 4.420445215741949,
+      "tokens_seen": 187105280
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009528184553660984,
+      "loss": 1.8749,
+      "theoretical_loss": 4.420250771454078,
+      "tokens_seen": 187170816
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009527983951855566,
+      "loss": 1.9967,
+      "theoretical_loss": 4.420056414292584,
+      "tokens_seen": 187236352
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009527783350050151,
+      "loss": 1.9333,
+      "theoretical_loss": 4.419862144187952,
+      "tokens_seen": 187301888
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009527582748244735,
+      "loss": 1.93,
+      "theoretical_loss": 4.419667961070747,
+      "tokens_seen": 187367424
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009527382146439318,
+      "loss": 2.0093,
+      "theoretical_loss": 4.419473864871613,
+      "tokens_seen": 187432960
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009527181544633902,
+      "loss": 1.9498,
+      "theoretical_loss": 4.419279855521276,
+      "tokens_seen": 187498496
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009526980942828485,
+      "loss": 2.0281,
+      "theoretical_loss": 4.419085932950537,
+      "tokens_seen": 187564032
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009526780341023069,
+      "loss": 1.9989,
+      "theoretical_loss": 4.41889209709028,
+      "tokens_seen": 187629568
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009526579739217654,
+      "loss": 1.8987,
+      "theoretical_loss": 4.4186983478714685,
+      "tokens_seen": 187695104
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009526379137412236,
+      "loss": 1.9608,
+      "theoretical_loss": 4.418504685225142,
+      "tokens_seen": 187760640
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009526178535606821,
+      "loss": 1.9515,
+      "theoretical_loss": 4.4183111090824205,
+      "tokens_seen": 187826176
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009525977933801405,
+      "loss": 2.0248,
+      "theoretical_loss": 4.418117619374502,
+      "tokens_seen": 187891712
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009525777331995988,
+      "loss": 1.9202,
+      "theoretical_loss": 4.417924216032667,
+      "tokens_seen": 187957248
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009525576730190572,
+      "loss": 1.9305,
+      "theoretical_loss": 4.417730898988268,
+      "tokens_seen": 188022784
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009525376128385155,
+      "loss": 1.9592,
+      "theoretical_loss": 4.417537668172743,
+      "tokens_seen": 188088320
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009525175526579739,
+      "loss": 1.8921,
+      "theoretical_loss": 4.417344523517602,
+      "tokens_seen": 188153856
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009524974924774324,
+      "loss": 1.9449,
+      "theoretical_loss": 4.417151464954437,
+      "tokens_seen": 188219392
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009524774322968906,
+      "loss": 1.9775,
+      "theoretical_loss": 4.416958492414918,
+      "tokens_seen": 188284928
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009524573721163491,
+      "loss": 1.9599,
+      "theoretical_loss": 4.416765605830792,
+      "tokens_seen": 188350464
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009524373119358074,
+      "loss": 1.8162,
+      "theoretical_loss": 4.4165728051338835,
+      "tokens_seen": 188416000
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009524172517552658,
+      "loss": 1.9151,
+      "theoretical_loss": 4.416380090256095,
+      "tokens_seen": 188481536
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009523971915747242,
+      "loss": 1.8405,
+      "theoretical_loss": 4.416187461129409,
+      "tokens_seen": 188547072
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009523771313941826,
+      "loss": 2.0353,
+      "theoretical_loss": 4.415994917685881,
+      "tokens_seen": 188612608
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009523570712136409,
+      "loss": 2.0348,
+      "theoretical_loss": 4.415802459857648,
+      "tokens_seen": 188678144
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009523370110330993,
+      "loss": 2.0916,
+      "theoretical_loss": 4.415610087576923,
+      "tokens_seen": 188743680
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009523169508525576,
+      "loss": 1.8956,
+      "theoretical_loss": 4.4154178007759946,
+      "tokens_seen": 188809216
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009522968906720161,
+      "loss": 2.0073,
+      "theoretical_loss": 4.415225599387231,
+      "tokens_seen": 188874752
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009522768304914744,
+      "loss": 1.916,
+      "theoretical_loss": 4.415033483343075,
+      "tokens_seen": 188940288
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009522567703109328,
+      "loss": 1.9299,
+      "theoretical_loss": 4.414841452576049,
+      "tokens_seen": 189005824
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009522367101303912,
+      "loss": 1.9453,
+      "theoretical_loss": 4.4146495070187495,
+      "tokens_seen": 189071360
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009522166499498496,
+      "loss": 1.961,
+      "theoretical_loss": 4.41445764660385,
+      "tokens_seen": 189136896
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009521965897693079,
+      "loss": 1.7822,
+      "theoretical_loss": 4.414265871264102,
+      "tokens_seen": 189202432
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009521765295887663,
+      "loss": 1.8333,
+      "theoretical_loss": 4.414074180932333,
+      "tokens_seen": 189267968
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009521564694082246,
+      "loss": 1.9526,
+      "theoretical_loss": 4.413882575541446,
+      "tokens_seen": 189333504
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009521364092276831,
+      "loss": 1.8715,
+      "theoretical_loss": 4.41369105502442,
+      "tokens_seen": 189399040
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009521163490471414,
+      "loss": 1.8555,
+      "theoretical_loss": 4.41349961931431,
+      "tokens_seen": 189464576
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009520962888665998,
+      "loss": 1.8327,
+      "theoretical_loss": 4.413308268344249,
+      "tokens_seen": 189530112
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009520762286860581,
+      "loss": 1.989,
+      "theoretical_loss": 4.413117002047443,
+      "tokens_seen": 189595648
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009520561685055166,
+      "loss": 1.9247,
+      "theoretical_loss": 4.4129258203571755,
+      "tokens_seen": 189661184
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009520361083249749,
+      "loss": 1.8879,
+      "theoretical_loss": 4.412734723206805,
+      "tokens_seen": 189726720
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009520160481444333,
+      "loss": 1.9158,
+      "theoretical_loss": 4.412543710529766,
+      "tokens_seen": 189792256
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009519959879638917,
+      "loss": 1.9396,
+      "theoretical_loss": 4.412352782259568,
+      "tokens_seen": 189857792
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009519759277833501,
+      "loss": 1.7889,
+      "theoretical_loss": 4.4121619383297945,
+      "tokens_seen": 189923328
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009519558676028084,
+      "loss": 2.0035,
+      "theoretical_loss": 4.411971178674106,
+      "tokens_seen": 189988864
+    },
+    {
+      "epoch": 0.06,
+      "objective/train/advantage_avg": 0.4934732913970947,
+      "objective/train/docs_used": 76077,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.8127355575561523,
+      "objective/train/original_loss": 1.8127355575561523,
+      "objective/train/theoretical_loss": 4.411780503226238,
+      "objective/train/tokens_used": 210514400,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24385586380958557,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505869388580322,
+      "objective/train/weighted_lm_loss": 1.9044004678726196,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.411780503226238,
+      "tokens_seen": 190054400
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009519358074222668,
+      "loss": 1.9662,
+      "theoretical_loss": 4.411780503226238,
+      "tokens_seen": 190054400
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009519157472417252,
+      "loss": 1.9113,
+      "theoretical_loss": 4.411589911919999,
+      "tokens_seen": 190119936
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009518956870611836,
+      "loss": 1.9649,
+      "theoretical_loss": 4.411399404689275,
+      "tokens_seen": 190185472
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000951875626880642,
+      "loss": 2.0698,
+      "theoretical_loss": 4.411208981468025,
+      "tokens_seen": 190251008
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009518555667001003,
+      "loss": 1.8534,
+      "theoretical_loss": 4.4110186421902835,
+      "tokens_seen": 190316544
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009518355065195588,
+      "loss": 1.7358,
+      "theoretical_loss": 4.410828386790158,
+      "tokens_seen": 190382080
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000951815446339017,
+      "loss": 1.7901,
+      "theoretical_loss": 4.410638215201831,
+      "tokens_seen": 190447616
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009517953861584755,
+      "loss": 1.9016,
+      "theoretical_loss": 4.410448127359561,
+      "tokens_seen": 190513152
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009517753259779339,
+      "loss": 2.0252,
+      "theoretical_loss": 4.4102581231976785,
+      "tokens_seen": 190578688
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009517552657973922,
+      "loss": 1.9728,
+      "theoretical_loss": 4.410068202650588,
+      "tokens_seen": 190644224
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009517352056168506,
+      "loss": 1.9752,
+      "theoretical_loss": 4.40987836565277,
+      "tokens_seen": 190709760
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009517151454363089,
+      "loss": 1.8996,
+      "theoretical_loss": 4.409688612138778,
+      "tokens_seen": 190775296
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009516950852557673,
+      "loss": 1.9769,
+      "theoretical_loss": 4.409498942043237,
+      "tokens_seen": 190840832
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009516750250752258,
+      "loss": 2.0145,
+      "theoretical_loss": 4.4093093553008496,
+      "tokens_seen": 190906368
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000951654964894684,
+      "loss": 1.9347,
+      "theoretical_loss": 4.409119851846388,
+      "tokens_seen": 190971904
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009516349047141425,
+      "loss": 1.8087,
+      "theoretical_loss": 4.4089304316147,
+      "tokens_seen": 191037440
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009516148445336009,
+      "loss": 1.9947,
+      "theoretical_loss": 4.408741094540707,
+      "tokens_seen": 191102976
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009515947843530592,
+      "loss": 1.8363,
+      "theoretical_loss": 4.408551840559403,
+      "tokens_seen": 191168512
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009515747241725176,
+      "loss": 1.8745,
+      "theoretical_loss": 4.408362669605854,
+      "tokens_seen": 191234048
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000951554663991976,
+      "loss": 1.9859,
+      "theoretical_loss": 4.408173581615199,
+      "tokens_seen": 191299584
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009515346038114343,
+      "loss": 1.8372,
+      "theoretical_loss": 4.407984576522653,
+      "tokens_seen": 191365120
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009515145436308928,
+      "loss": 2.0925,
+      "theoretical_loss": 4.4077956542635,
+      "tokens_seen": 191430656
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000951494483450351,
+      "loss": 1.9144,
+      "theoretical_loss": 4.407606814773099,
+      "tokens_seen": 191496192
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009514744232698095,
+      "loss": 1.7939,
+      "theoretical_loss": 4.407418057986881,
+      "tokens_seen": 191561728
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009514543630892678,
+      "loss": 1.9196,
+      "theoretical_loss": 4.407229383840347,
+      "tokens_seen": 191627264
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009514343029087262,
+      "loss": 1.9596,
+      "theoretical_loss": 4.407040792269075,
+      "tokens_seen": 191692800
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009514142427281846,
+      "loss": 1.9563,
+      "theoretical_loss": 4.406852283208711,
+      "tokens_seen": 191758336
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000951394182547643,
+      "loss": 1.8223,
+      "theoretical_loss": 4.406663856594977,
+      "tokens_seen": 191823872
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009513741223671013,
+      "loss": 1.8939,
+      "theoretical_loss": 4.406475512363663,
+      "tokens_seen": 191889408
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009513540621865597,
+      "loss": 1.8934,
+      "theoretical_loss": 4.4062872504506325,
+      "tokens_seen": 191954944
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000951334002006018,
+      "loss": 1.7905,
+      "theoretical_loss": 4.406099070791822,
+      "tokens_seen": 192020480
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009513139418254765,
+      "loss": 1.768,
+      "theoretical_loss": 4.405910973323239,
+      "tokens_seen": 192086016
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009512938816449348,
+      "loss": 1.8877,
+      "theoretical_loss": 4.405722957980962,
+      "tokens_seen": 192151552
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009512738214643932,
+      "loss": 1.9167,
+      "theoretical_loss": 4.405535024701141,
+      "tokens_seen": 192217088
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009512537612838516,
+      "loss": 1.9675,
+      "theoretical_loss": 4.405347173419998,
+      "tokens_seen": 192282624
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00095123370110331,
+      "loss": 1.9097,
+      "theoretical_loss": 4.405159404073826,
+      "tokens_seen": 192348160
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009512136409227683,
+      "loss": 1.8425,
+      "theoretical_loss": 4.40497171659899,
+      "tokens_seen": 192413696
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009511935807422267,
+      "loss": 2.008,
+      "theoretical_loss": 4.404784110931924,
+      "tokens_seen": 192479232
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000951173520561685,
+      "loss": 1.7656,
+      "theoretical_loss": 4.404596587009136,
+      "tokens_seen": 192544768
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009511534603811435,
+      "loss": 1.7931,
+      "theoretical_loss": 4.404409144767201,
+      "tokens_seen": 192610304
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009511334002006018,
+      "loss": 1.8794,
+      "theoretical_loss": 4.404221784142768,
+      "tokens_seen": 192675840
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009511133400200602,
+      "loss": 1.8634,
+      "theoretical_loss": 4.404034505072556,
+      "tokens_seen": 192741376
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009510932798395185,
+      "loss": 1.8689,
+      "theoretical_loss": 4.403847307493353,
+      "tokens_seen": 192806912
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000951073219658977,
+      "loss": 1.9959,
+      "theoretical_loss": 4.40366019134202,
+      "tokens_seen": 192872448
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009510531594784353,
+      "loss": 1.9952,
+      "theoretical_loss": 4.403473156555487,
+      "tokens_seen": 192937984
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009510330992978937,
+      "loss": 1.9028,
+      "theoretical_loss": 4.403286203070751,
+      "tokens_seen": 193003520
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009510130391173521,
+      "loss": 1.7787,
+      "theoretical_loss": 4.403099330824887,
+      "tokens_seen": 193069056
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009509929789368105,
+      "loss": 1.9463,
+      "theoretical_loss": 4.4029125397550315,
+      "tokens_seen": 193134592
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009509729187562688,
+      "loss": 1.834,
+      "theoretical_loss": 4.402725829798397,
+      "tokens_seen": 193200128
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009509528585757272,
+      "loss": 1.9024,
+      "theoretical_loss": 4.402539200892264,
+      "tokens_seen": 193265664
+    },
+    {
+      "epoch": 0.06,
+      "objective/train/advantage_avg": 0.49331486225128174,
+      "objective/train/docs_used": 77266,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.8808577060699463,
+      "objective/train/original_loss": 1.8808577060699463,
+      "objective/train/theoretical_loss": 4.40235265297398,
+      "objective/train/tokens_used": 213791200,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24369393289089203,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505704879760742,
+      "objective/train/weighted_lm_loss": 1.9757757186889648,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.40235265297398,
+      "tokens_seen": 193331200
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009509327983951855,
+      "loss": 1.8993,
+      "theoretical_loss": 4.40235265297398,
+      "tokens_seen": 193331200
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000950912738214644,
+      "loss": 1.8289,
+      "theoretical_loss": 4.402166185980967,
+      "tokens_seen": 193396736
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009508926780341023,
+      "loss": 1.8115,
+      "theoretical_loss": 4.4019797998507135,
+      "tokens_seen": 193462272
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009508726178535607,
+      "loss": 1.8387,
+      "theoretical_loss": 4.401793494520778,
+      "tokens_seen": 193527808
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009508525576730191,
+      "loss": 1.7944,
+      "theoretical_loss": 4.401607269928788,
+      "tokens_seen": 193593344
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009508324974924774,
+      "loss": 1.7242,
+      "theoretical_loss": 4.4014211260124405,
+      "tokens_seen": 193658880
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009508124373119358,
+      "loss": 1.8879,
+      "theoretical_loss": 4.401235062709502,
+      "tokens_seen": 193724416
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009507923771313943,
+      "loss": 1.8485,
+      "theoretical_loss": 4.401049079957809,
+      "tokens_seen": 193789952
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009507723169508525,
+      "loss": 1.8177,
+      "theoretical_loss": 4.400863177695264,
+      "tokens_seen": 193855488
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000950752256770311,
+      "loss": 1.8554,
+      "theoretical_loss": 4.400677355859841,
+      "tokens_seen": 193921024
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009507321965897692,
+      "loss": 2.0593,
+      "theoretical_loss": 4.400491614389582,
+      "tokens_seen": 193986560
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009507121364092277,
+      "loss": 1.8376,
+      "theoretical_loss": 4.400305953222596,
+      "tokens_seen": 194052096
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009506920762286861,
+      "loss": 1.9119,
+      "theoretical_loss": 4.4001203722970645,
+      "tokens_seen": 194117632
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009506720160481444,
+      "loss": 1.8258,
+      "theoretical_loss": 4.399934871551234,
+      "tokens_seen": 194183168
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009506519558676028,
+      "loss": 1.6996,
+      "theoretical_loss": 4.3997494509234185,
+      "tokens_seen": 194248704
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009506318956870613,
+      "loss": 1.8192,
+      "theoretical_loss": 4.399564110352005,
+      "tokens_seen": 194314240
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009506118355065195,
+      "loss": 1.9074,
+      "theoretical_loss": 4.399378849775443,
+      "tokens_seen": 194379776
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000950591775325978,
+      "loss": 1.8622,
+      "theoretical_loss": 4.399193669132255,
+      "tokens_seen": 194445312
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009505717151454362,
+      "loss": 1.9398,
+      "theoretical_loss": 4.399008568361027,
+      "tokens_seen": 194510848
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009505516549648947,
+      "loss": 1.8967,
+      "theoretical_loss": 4.398823547400417,
+      "tokens_seen": 194576384
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009505315947843531,
+      "loss": 1.9654,
+      "theoretical_loss": 4.398638606189149,
+      "tokens_seen": 194641920
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009505115346038114,
+      "loss": 1.8578,
+      "theoretical_loss": 4.398453744666012,
+      "tokens_seen": 194707456
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009504914744232698,
+      "loss": 1.9715,
+      "theoretical_loss": 4.398268962769867,
+      "tokens_seen": 194772992
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009504714142427282,
+      "loss": 2.0093,
+      "theoretical_loss": 4.39808426043964,
+      "tokens_seen": 194838528
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009504513540621865,
+      "loss": 1.8911,
+      "theoretical_loss": 4.397899637614326,
+      "tokens_seen": 194904064
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000950431293881645,
+      "loss": 1.9758,
+      "theoretical_loss": 4.397715094232984,
+      "tokens_seen": 194969600
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009504112337011034,
+      "loss": 1.9613,
+      "theoretical_loss": 4.397530630234744,
+      "tokens_seen": 195035136
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009503911735205617,
+      "loss": 1.9392,
+      "theoretical_loss": 4.3973462455588,
+      "tokens_seen": 195100672
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009503711133400201,
+      "loss": 1.8308,
+      "theoretical_loss": 4.397161940144415,
+      "tokens_seen": 195166208
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009503510531594784,
+      "loss": 1.7817,
+      "theoretical_loss": 4.396977713930919,
+      "tokens_seen": 195231744
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009503309929789369,
+      "loss": 1.9156,
+      "theoretical_loss": 4.396793566857708,
+      "tokens_seen": 195297280
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009503109327983952,
+      "loss": 1.8466,
+      "theoretical_loss": 4.396609498864243,
+      "tokens_seen": 195362816
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009502908726178536,
+      "loss": 1.8075,
+      "theoretical_loss": 4.396425509890054,
+      "tokens_seen": 195428352
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000950270812437312,
+      "loss": 1.8059,
+      "theoretical_loss": 4.396241599874738,
+      "tokens_seen": 195493888
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009502507522567704,
+      "loss": 1.8906,
+      "theoretical_loss": 4.396057768757957,
+      "tokens_seen": 195559424
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009502306920762287,
+      "loss": 1.8618,
+      "theoretical_loss": 4.395874016479438,
+      "tokens_seen": 195624960
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009502106318956871,
+      "loss": 1.8689,
+      "theoretical_loss": 4.395690342978977,
+      "tokens_seen": 195690496
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009501905717151454,
+      "loss": 1.7489,
+      "theoretical_loss": 4.395506748196435,
+      "tokens_seen": 195756032
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009501705115346039,
+      "loss": 2.0602,
+      "theoretical_loss": 4.395323232071737,
+      "tokens_seen": 195821568
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009501504513540622,
+      "loss": 1.8469,
+      "theoretical_loss": 4.395139794544878,
+      "tokens_seen": 195887104
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009501303911735206,
+      "loss": 1.9692,
+      "theoretical_loss": 4.394956435555916,
+      "tokens_seen": 195952640
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009501103309929789,
+      "loss": 1.9052,
+      "theoretical_loss": 4.394773155044976,
+      "tokens_seen": 196018176
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009500902708124374,
+      "loss": 1.8058,
+      "theoretical_loss": 4.394589952952247,
+      "tokens_seen": 196083712
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009500702106318957,
+      "loss": 1.787,
+      "theoretical_loss": 4.394406829217983,
+      "tokens_seen": 196149248
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009500501504513541,
+      "loss": 1.8961,
+      "theoretical_loss": 4.394223783782509,
+      "tokens_seen": 196214784
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009500300902708125,
+      "loss": 1.7633,
+      "theoretical_loss": 4.394040816586209,
+      "tokens_seen": 196280320
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009500100300902709,
+      "loss": 1.7421,
+      "theoretical_loss": 4.393857927569534,
+      "tokens_seen": 196345856
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009499899699097292,
+      "loss": 1.9565,
+      "theoretical_loss": 4.393675116673003,
+      "tokens_seen": 196411392
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009499699097291876,
+      "loss": 1.8156,
+      "theoretical_loss": 4.393492383837196,
+      "tokens_seen": 196476928
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009499498495486459,
+      "loss": 1.8552,
+      "theoretical_loss": 4.393309729002761,
+      "tokens_seen": 196542464
+    },
+    {
+      "epoch": 0.06,
+      "objective/train/advantage_avg": 0.4933859407901764,
+      "objective/train/docs_used": 78313,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.5727931261062622,
+      "objective/train/original_loss": 1.5727932453155518,
+      "objective/train/theoretical_loss": 4.393127152110409,
+      "objective/train/tokens_used": 217068000,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24412523210048676,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.05057954788208,
+      "objective/train/weighted_lm_loss": 1.652202844619751,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.393127152110409,
+      "tokens_seen": 196608000
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009499297893681044,
+      "loss": 1.7561,
+      "theoretical_loss": 4.393127152110409,
+      "tokens_seen": 196608000
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009499097291875627,
+      "loss": 1.855,
+      "theoretical_loss": 4.392944653100918,
+      "tokens_seen": 196673536
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009498896690070211,
+      "loss": 1.9183,
+      "theoretical_loss": 4.392762231915128,
+      "tokens_seen": 196739072
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009498696088264795,
+      "loss": 1.8333,
+      "theoretical_loss": 4.392579888493946,
+      "tokens_seen": 196804608
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009498495486459378,
+      "loss": 1.8654,
+      "theoretical_loss": 4.392397622778343,
+      "tokens_seen": 196870144
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009498294884653962,
+      "loss": 1.8546,
+      "theoretical_loss": 4.392215434709351,
+      "tokens_seen": 196935680
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009498094282848547,
+      "loss": 1.9817,
+      "theoretical_loss": 4.392033324228073,
+      "tokens_seen": 197001216
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009497893681043129,
+      "loss": 1.8226,
+      "theoretical_loss": 4.391851291275671,
+      "tokens_seen": 197066752
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009497693079237714,
+      "loss": 1.8504,
+      "theoretical_loss": 4.391669335793372,
+      "tokens_seen": 197132288
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009497492477432296,
+      "loss": 1.7561,
+      "theoretical_loss": 4.391487457722469,
+      "tokens_seen": 197197824
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009497291875626881,
+      "loss": 1.671,
+      "theoretical_loss": 4.391305657004317,
+      "tokens_seen": 197263360
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009497091273821465,
+      "loss": 1.9367,
+      "theoretical_loss": 4.391123933580336,
+      "tokens_seen": 197328896
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009496890672016048,
+      "loss": 1.9203,
+      "theoretical_loss": 4.39094228739201,
+      "tokens_seen": 197394432
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009496690070210632,
+      "loss": 1.8818,
+      "theoretical_loss": 4.390760718380884,
+      "tokens_seen": 197459968
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009496489468405217,
+      "loss": 1.8564,
+      "theoretical_loss": 4.390579226488571,
+      "tokens_seen": 197525504
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009496288866599799,
+      "loss": 1.9317,
+      "theoretical_loss": 4.3903978116567455,
+      "tokens_seen": 197591040
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009496088264794384,
+      "loss": 1.9909,
+      "theoretical_loss": 4.390216473827143,
+      "tokens_seen": 197656576
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009495887662988966,
+      "loss": 1.7772,
+      "theoretical_loss": 4.390035212941568,
+      "tokens_seen": 197722112
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009495687061183551,
+      "loss": 1.7826,
+      "theoretical_loss": 4.3898540289418815,
+      "tokens_seen": 197787648
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009495486459378135,
+      "loss": 1.8518,
+      "theoretical_loss": 4.389672921770013,
+      "tokens_seen": 197853184
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009495285857572718,
+      "loss": 1.7901,
+      "theoretical_loss": 4.389491891367953,
+      "tokens_seen": 197918720
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009495085255767302,
+      "loss": 1.8022,
+      "theoretical_loss": 4.389310937677754,
+      "tokens_seen": 197984256
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009494884653961886,
+      "loss": 1.9155,
+      "theoretical_loss": 4.389130060641534,
+      "tokens_seen": 198049792
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009494684052156469,
+      "loss": 1.8942,
+      "theoretical_loss": 4.388949260201471,
+      "tokens_seen": 198115328
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009494483450351054,
+      "loss": 1.8778,
+      "theoretical_loss": 4.388768536299808,
+      "tokens_seen": 198180864
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009494282848545636,
+      "loss": 1.9305,
+      "theoretical_loss": 4.388587888878849,
+      "tokens_seen": 198246400
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009494082246740221,
+      "loss": 1.9257,
+      "theoretical_loss": 4.388407317880962,
+      "tokens_seen": 198311936
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009493881644934804,
+      "loss": 1.9663,
+      "theoretical_loss": 4.388226823248576,
+      "tokens_seen": 198377472
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009493681043129388,
+      "loss": 1.9069,
+      "theoretical_loss": 4.388046404924184,
+      "tokens_seen": 198443008
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009493480441323972,
+      "loss": 1.8286,
+      "theoretical_loss": 4.387866062850339,
+      "tokens_seen": 198508544
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009493279839518556,
+      "loss": 1.915,
+      "theoretical_loss": 4.3876857969696585,
+      "tokens_seen": 198574080
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009493079237713139,
+      "loss": 1.8648,
+      "theoretical_loss": 4.387505607224821,
+      "tokens_seen": 198639616
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009492878635907724,
+      "loss": 1.8246,
+      "theoretical_loss": 4.387325493558566,
+      "tokens_seen": 198705152
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009492678034102307,
+      "loss": 1.8195,
+      "theoretical_loss": 4.387145455913698,
+      "tokens_seen": 198770688
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009492477432296891,
+      "loss": 1.7988,
+      "theoretical_loss": 4.38696549423308,
+      "tokens_seen": 198836224
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009492276830491474,
+      "loss": 1.9393,
+      "theoretical_loss": 4.386785608459639,
+      "tokens_seen": 198901760
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009492076228686058,
+      "loss": 1.9559,
+      "theoretical_loss": 4.386605798536362,
+      "tokens_seen": 198967296
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009491875626880642,
+      "loss": 1.945,
+      "theoretical_loss": 4.3864260644063,
+      "tokens_seen": 199032832
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009491675025075226,
+      "loss": 1.8763,
+      "theoretical_loss": 4.386246406012562,
+      "tokens_seen": 199098368
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009491474423269809,
+      "loss": 1.9405,
+      "theoretical_loss": 4.3860668232983215,
+      "tokens_seen": 199163904
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009491273821464393,
+      "loss": 1.962,
+      "theoretical_loss": 4.385887316206812,
+      "tokens_seen": 199229440
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009491073219658977,
+      "loss": 1.8279,
+      "theoretical_loss": 4.385707884681327,
+      "tokens_seen": 199294976
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009490872617853561,
+      "loss": 1.7523,
+      "theoretical_loss": 4.385528528665225,
+      "tokens_seen": 199360512
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009490672016048144,
+      "loss": 1.8032,
+      "theoretical_loss": 4.385349248101922,
+      "tokens_seen": 199426048
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009490471414242729,
+      "loss": 1.8202,
+      "theoretical_loss": 4.385170042934896,
+      "tokens_seen": 199491584
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009490270812437311,
+      "loss": 1.8708,
+      "theoretical_loss": 4.384990913107686,
+      "tokens_seen": 199557120
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009490070210631896,
+      "loss": 1.8222,
+      "theoretical_loss": 4.384811858563891,
+      "tokens_seen": 199622656
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009489869608826479,
+      "loss": 1.7764,
+      "theoretical_loss": 4.384632879247173,
+      "tokens_seen": 199688192
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009489669007021063,
+      "loss": 1.865,
+      "theoretical_loss": 4.384453975101251,
+      "tokens_seen": 199753728
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009489468405215648,
+      "loss": 1.8772,
+      "theoretical_loss": 4.384275146069909,
+      "tokens_seen": 199819264
+    },
+    {
+      "epoch": 0.06,
+      "objective/train/advantage_avg": 0.4942682981491089,
+      "objective/train/docs_used": 79488,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.6611281633377075,
+      "objective/train/original_loss": 1.661128282546997,
+      "objective/train/theoretical_loss": 4.384096392096988,
+      "objective/train/tokens_used": 220344800,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24450337886810303,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.05066978931427,
+      "objective/train/weighted_lm_loss": 1.7454839944839478,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.384096392096988,
+      "tokens_seen": 199884800
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009489267803410231,
+      "loss": 1.8157,
+      "theoretical_loss": 4.384096392096988,
+      "tokens_seen": 199884800
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009489067201604815,
+      "loss": 1.9058,
+      "theoretical_loss": 4.38391771312639,
+      "tokens_seen": 199950336
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009488866599799399,
+      "loss": 1.9625,
+      "theoretical_loss": 4.38373910910208,
+      "tokens_seen": 200015872
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009488665997993982,
+      "loss": 1.8666,
+      "theoretical_loss": 4.383560579968078,
+      "tokens_seen": 200081408
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009488465396188566,
+      "loss": 1.7205,
+      "theoretical_loss": 4.3833821256684695,
+      "tokens_seen": 200146944
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009488264794383151,
+      "loss": 1.8827,
+      "theoretical_loss": 4.383203746147397,
+      "tokens_seen": 200212480
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009488064192577733,
+      "loss": 1.9402,
+      "theoretical_loss": 4.383025441349063,
+      "tokens_seen": 200278016
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009487863590772318,
+      "loss": 1.8922,
+      "theoretical_loss": 4.382847211217731,
+      "tokens_seen": 200343552
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00094876629889669,
+      "loss": 1.9755,
+      "theoretical_loss": 4.3826690556977255,
+      "tokens_seen": 200409088
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009487462387161485,
+      "loss": 1.9237,
+      "theoretical_loss": 4.382490974733425,
+      "tokens_seen": 200474624
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009487261785356069,
+      "loss": 2.0019,
+      "theoretical_loss": 4.382312968269276,
+      "tokens_seen": 200540160
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009487061183550652,
+      "loss": 1.8139,
+      "theoretical_loss": 4.382135036249777,
+      "tokens_seen": 200605696
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009486860581745236,
+      "loss": 1.8938,
+      "theoretical_loss": 4.3819571786194915,
+      "tokens_seen": 200671232
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009486659979939821,
+      "loss": 1.8056,
+      "theoretical_loss": 4.381779395323038,
+      "tokens_seen": 200736768
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009486459378134403,
+      "loss": 1.8419,
+      "theoretical_loss": 4.381601686305098,
+      "tokens_seen": 200802304
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009486258776328988,
+      "loss": 1.7723,
+      "theoretical_loss": 4.381424051510409,
+      "tokens_seen": 200867840
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000948605817452357,
+      "loss": 1.8334,
+      "theoretical_loss": 4.38124649088377,
+      "tokens_seen": 200933376
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009485857572718155,
+      "loss": 1.8382,
+      "theoretical_loss": 4.381069004370039,
+      "tokens_seen": 200998912
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009485656970912739,
+      "loss": 1.8665,
+      "theoretical_loss": 4.38089159191413,
+      "tokens_seen": 201064448
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009485456369107322,
+      "loss": 1.7653,
+      "theoretical_loss": 4.38071425346102,
+      "tokens_seen": 201129984
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009485255767301906,
+      "loss": 1.8926,
+      "theoretical_loss": 4.380536988955742,
+      "tokens_seen": 201195520
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000948505516549649,
+      "loss": 1.7422,
+      "theoretical_loss": 4.380359798343388,
+      "tokens_seen": 201261056
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009484854563691073,
+      "loss": 1.707,
+      "theoretical_loss": 4.380182681569111,
+      "tokens_seen": 201326592
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009484653961885658,
+      "loss": 1.8416,
+      "theoretical_loss": 4.380005638578119,
+      "tokens_seen": 201392128
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000948445336008024,
+      "loss": 1.8569,
+      "theoretical_loss": 4.37982866931568,
+      "tokens_seen": 201457664
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009484252758274825,
+      "loss": 1.82,
+      "theoretical_loss": 4.379651773727122,
+      "tokens_seen": 201523200
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009484052156469408,
+      "loss": 1.978,
+      "theoretical_loss": 4.379474951757829,
+      "tokens_seen": 201588736
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009483851554663992,
+      "loss": 1.7683,
+      "theoretical_loss": 4.379298203353244,
+      "tokens_seen": 201654272
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009483650952858576,
+      "loss": 1.9227,
+      "theoretical_loss": 4.379121528458869,
+      "tokens_seen": 201719808
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000948345035105316,
+      "loss": 1.8948,
+      "theoretical_loss": 4.378944927020262,
+      "tokens_seen": 201785344
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009483249749247743,
+      "loss": 1.939,
+      "theoretical_loss": 4.378768398983042,
+      "tokens_seen": 201850880
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009483049147442328,
+      "loss": 1.8403,
+      "theoretical_loss": 4.378591944292882,
+      "tokens_seen": 201916416
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000948284854563691,
+      "loss": 1.8222,
+      "theoretical_loss": 4.378415562895518,
+      "tokens_seen": 201981952
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009482647943831495,
+      "loss": 1.7433,
+      "theoretical_loss": 4.378239254736738,
+      "tokens_seen": 202047488
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009482447342026078,
+      "loss": 1.9094,
+      "theoretical_loss": 4.378063019762392,
+      "tokens_seen": 202113024
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009482246740220662,
+      "loss": 1.7928,
+      "theoretical_loss": 4.377886857918385,
+      "tokens_seen": 202178560
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009482046138415246,
+      "loss": 1.8734,
+      "theoretical_loss": 4.377710769150681,
+      "tokens_seen": 202244096
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000948184553660983,
+      "loss": 1.9099,
+      "theoretical_loss": 4.3775347534053015,
+      "tokens_seen": 202309632
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009481644934804413,
+      "loss": 1.7542,
+      "theoretical_loss": 4.377358810628324,
+      "tokens_seen": 202375168
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009481444332998997,
+      "loss": 1.8399,
+      "theoretical_loss": 4.377182940765883,
+      "tokens_seen": 202440704
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009481243731193581,
+      "loss": 1.7524,
+      "theoretical_loss": 4.377007143764175,
+      "tokens_seen": 202506240
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009481043129388165,
+      "loss": 1.8736,
+      "theoretical_loss": 4.376831419569445,
+      "tokens_seen": 202571776
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009480842527582748,
+      "loss": 1.9028,
+      "theoretical_loss": 4.3766557681280025,
+      "tokens_seen": 202637312
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009480641925777333,
+      "loss": 1.9198,
+      "theoretical_loss": 4.3764801893862115,
+      "tokens_seen": 202702848
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009480441323971915,
+      "loss": 1.8034,
+      "theoretical_loss": 4.376304683290491,
+      "tokens_seen": 202768384
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00094802407221665,
+      "loss": 1.8115,
+      "theoretical_loss": 4.37612924978732,
+      "tokens_seen": 202833920
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009480040120361083,
+      "loss": 1.8673,
+      "theoretical_loss": 4.375953888823233,
+      "tokens_seen": 202899456
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009479839518555667,
+      "loss": 1.7649,
+      "theoretical_loss": 4.375778600344819,
+      "tokens_seen": 202964992
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009479638916750251,
+      "loss": 1.7969,
+      "theoretical_loss": 4.375603384298726,
+      "tokens_seen": 203030528
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009479438314944835,
+      "loss": 1.8562,
+      "theoretical_loss": 4.375428240631659,
+      "tokens_seen": 203096064
+    },
+    {
+      "epoch": 0.06,
+      "objective/train/advantage_avg": 0.49045994877815247,
+      "objective/train/docs_used": 80605,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.8460792303085327,
+      "objective/train/original_loss": 1.8460791110992432,
+      "objective/train/theoretical_loss": 4.375253169290376,
+      "objective/train/tokens_used": 223621600,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24119678139686584,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050271987915039,
+      "objective/train/weighted_lm_loss": 1.9390350580215454,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.375253169290376,
+      "tokens_seen": 203161600
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009479237713139418,
+      "loss": 1.9466,
+      "theoretical_loss": 4.375253169290376,
+      "tokens_seen": 203161600
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009479037111334003,
+      "loss": 1.9524,
+      "theoretical_loss": 4.3750781702216965,
+      "tokens_seen": 203227136
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009478836509528585,
+      "loss": 1.7591,
+      "theoretical_loss": 4.37490324337249,
+      "tokens_seen": 203292672
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000947863590772317,
+      "loss": 1.8594,
+      "theoretical_loss": 4.374728388689688,
+      "tokens_seen": 203358208
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009478435305917753,
+      "loss": 1.8902,
+      "theoretical_loss": 4.374553606120274,
+      "tokens_seen": 203423744
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009478234704112337,
+      "loss": 1.8355,
+      "theoretical_loss": 4.374378895611289,
+      "tokens_seen": 203489280
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009478034102306921,
+      "loss": 1.9065,
+      "theoretical_loss": 4.374204257109832,
+      "tokens_seen": 203554816
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009477833500501504,
+      "loss": 1.7489,
+      "theoretical_loss": 4.374029690563053,
+      "tokens_seen": 203620352
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009477632898696088,
+      "loss": 1.8441,
+      "theoretical_loss": 4.373855195918162,
+      "tokens_seen": 203685888
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009477432296890673,
+      "loss": 1.9567,
+      "theoretical_loss": 4.373680773122424,
+      "tokens_seen": 203751424
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009477231695085255,
+      "loss": 1.8593,
+      "theoretical_loss": 4.3735064221231585,
+      "tokens_seen": 203816960
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000947703109327984,
+      "loss": 1.9252,
+      "theoretical_loss": 4.373332142867739,
+      "tokens_seen": 203882496
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009476830491474422,
+      "loss": 1.8711,
+      "theoretical_loss": 4.3731579353036,
+      "tokens_seen": 203948032
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009476629889669007,
+      "loss": 1.8338,
+      "theoretical_loss": 4.372983799378225,
+      "tokens_seen": 204013568
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009476429287863591,
+      "loss": 1.8558,
+      "theoretical_loss": 4.372809735039157,
+      "tokens_seen": 204079104
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009476228686058174,
+      "loss": 1.8181,
+      "theoretical_loss": 4.372635742233991,
+      "tokens_seen": 204144640
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009476028084252758,
+      "loss": 1.7501,
+      "theoretical_loss": 4.372461820910382,
+      "tokens_seen": 204210176
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009475827482447343,
+      "loss": 1.8748,
+      "theoretical_loss": 4.372287971016036,
+      "tokens_seen": 204275712
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009475626880641925,
+      "loss": 1.7815,
+      "theoretical_loss": 4.372114192498714,
+      "tokens_seen": 204341248
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000947542627883651,
+      "loss": 1.9619,
+      "theoretical_loss": 4.371940485306234,
+      "tokens_seen": 204406784
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009475225677031093,
+      "loss": 1.8881,
+      "theoretical_loss": 4.371766849386468,
+      "tokens_seen": 204472320
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009475025075225677,
+      "loss": 1.9036,
+      "theoretical_loss": 4.371593284687343,
+      "tokens_seen": 204537856
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009474824473420261,
+      "loss": 1.7205,
+      "theoretical_loss": 4.371419791156841,
+      "tokens_seen": 204603392
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009474623871614844,
+      "loss": 1.8684,
+      "theoretical_loss": 4.371246368742996,
+      "tokens_seen": 204668928
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009474423269809429,
+      "loss": 2.0226,
+      "theoretical_loss": 4.3710730173939005,
+      "tokens_seen": 204734464
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009474222668004012,
+      "loss": 1.7672,
+      "theoretical_loss": 4.3708997370577,
+      "tokens_seen": 204800000
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009474022066198596,
+      "loss": 1.9649,
+      "theoretical_loss": 4.370726527682593,
+      "tokens_seen": 204865536
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000947382146439318,
+      "loss": 1.9361,
+      "theoretical_loss": 4.370553389216834,
+      "tokens_seen": 204931072
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009473620862587764,
+      "loss": 1.8473,
+      "theoretical_loss": 4.370380321608731,
+      "tokens_seen": 204996608
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009473420260782347,
+      "loss": 1.978,
+      "theoretical_loss": 4.370207324806645,
+      "tokens_seen": 205062144
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009473219658976932,
+      "loss": 1.8565,
+      "theoretical_loss": 4.370034398758996,
+      "tokens_seen": 205127680
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009473019057171515,
+      "loss": 1.7406,
+      "theoretical_loss": 4.369861543414251,
+      "tokens_seen": 205193216
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009472818455366099,
+      "loss": 2.0285,
+      "theoretical_loss": 4.369688758720937,
+      "tokens_seen": 205258752
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009472617853560682,
+      "loss": 1.851,
+      "theoretical_loss": 4.36951604462763,
+      "tokens_seen": 205324288
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009472417251755266,
+      "loss": 1.8151,
+      "theoretical_loss": 4.369343401082966,
+      "tokens_seen": 205389824
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000947221664994985,
+      "loss": 1.7439,
+      "theoretical_loss": 4.369170828035628,
+      "tokens_seen": 205455360
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009472016048144434,
+      "loss": 1.8962,
+      "theoretical_loss": 4.368998325434355,
+      "tokens_seen": 205520896
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009471815446339017,
+      "loss": 1.8172,
+      "theoretical_loss": 4.368825893227943,
+      "tokens_seen": 205586432
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009471614844533601,
+      "loss": 1.7986,
+      "theoretical_loss": 4.368653531365237,
+      "tokens_seen": 205651968
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009471414242728185,
+      "loss": 1.8557,
+      "theoretical_loss": 4.3684812397951385,
+      "tokens_seen": 205717504
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009471213640922769,
+      "loss": 1.7482,
+      "theoretical_loss": 4.3683090184666,
+      "tokens_seen": 205783040
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009471013039117352,
+      "loss": 1.7962,
+      "theoretical_loss": 4.368136867328628,
+      "tokens_seen": 205848576
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009470812437311937,
+      "loss": 1.7665,
+      "theoretical_loss": 4.367964786330285,
+      "tokens_seen": 205914112
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009470611835506519,
+      "loss": 1.8658,
+      "theoretical_loss": 4.367792775420682,
+      "tokens_seen": 205979648
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009470411233701104,
+      "loss": 1.7218,
+      "theoretical_loss": 4.367620834548987,
+      "tokens_seen": 206045184
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009470210631895687,
+      "loss": 1.8305,
+      "theoretical_loss": 4.367448963664419,
+      "tokens_seen": 206110720
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009470010030090271,
+      "loss": 1.8604,
+      "theoretical_loss": 4.367277162716249,
+      "tokens_seen": 206176256
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009469809428284855,
+      "loss": 1.8282,
+      "theoretical_loss": 4.367105431653806,
+      "tokens_seen": 206241792
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009469608826479439,
+      "loss": 1.8322,
+      "theoretical_loss": 4.3669337704264635,
+      "tokens_seen": 206307328
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009469408224674022,
+      "loss": 1.891,
+      "theoretical_loss": 4.366762178983656,
+      "tokens_seen": 206372864
+    },
+    {
+      "epoch": 0.06,
+      "objective/train/advantage_avg": 0.4905458092689514,
+      "objective/train/docs_used": 81724,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.7321252822875977,
+      "objective/train/original_loss": 1.7321252822875977,
+      "objective/train/theoretical_loss": 4.366590657274865,
+      "objective/train/tokens_used": 226898400,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24100235104560852,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0502796173095703,
+      "objective/train/weighted_lm_loss": 1.8194676637649536,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 1.0304546356201172,
+      "theoretical_loss": 4.366590657274865,
+      "tokens_seen": 206438400
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009469207622868607,
+      "loss": 1.82,
+      "theoretical_loss": 4.366590657274865,
+      "tokens_seen": 206438400
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009469007021063189,
+      "loss": 1.8988,
+      "theoretical_loss": 4.366419205249628,
+      "tokens_seen": 206503936
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009468806419257774,
+      "loss": 1.9824,
+      "theoretical_loss": 4.366247822857533,
+      "tokens_seen": 206569472
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009468605817452357,
+      "loss": 1.8839,
+      "theoretical_loss": 4.36607651004822,
+      "tokens_seen": 206635008
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009468405215646941,
+      "loss": 1.8377,
+      "theoretical_loss": 4.365905266771383,
+      "tokens_seen": 206700544
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009468204613841525,
+      "loss": 1.9625,
+      "theoretical_loss": 4.36573409297677,
+      "tokens_seen": 206766080
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009468004012036108,
+      "loss": 1.7877,
+      "theoretical_loss": 4.365562988614176,
+      "tokens_seen": 206831616
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009467803410230692,
+      "loss": 1.9814,
+      "theoretical_loss": 4.365391953633453,
+      "tokens_seen": 206897152
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009467602808425277,
+      "loss": 1.8504,
+      "theoretical_loss": 4.365220987984502,
+      "tokens_seen": 206962688
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009467402206619859,
+      "loss": 1.5948,
+      "theoretical_loss": 4.365050091617277,
+      "tokens_seen": 207028224
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009467201604814444,
+      "loss": 1.8719,
+      "theoretical_loss": 4.364879264481787,
+      "tokens_seen": 207093760
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009467001003009026,
+      "loss": 1.9276,
+      "theoretical_loss": 4.364708506528087,
+      "tokens_seen": 207159296
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009466800401203611,
+      "loss": 1.7526,
+      "theoretical_loss": 4.364537817706289,
+      "tokens_seen": 207224832
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009466599799398195,
+      "loss": 1.8202,
+      "theoretical_loss": 4.364367197966552,
+      "tokens_seen": 207290368
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009466399197592778,
+      "loss": 1.7798,
+      "theoretical_loss": 4.364196647259092,
+      "tokens_seen": 207355904
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009466198595787362,
+      "loss": 1.8522,
+      "theoretical_loss": 4.364026165534174,
+      "tokens_seen": 207421440
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009465997993981947,
+      "loss": 1.7636,
+      "theoretical_loss": 4.363855752742112,
+      "tokens_seen": 207486976
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009465797392176529,
+      "loss": 1.9009,
+      "theoretical_loss": 4.363685408833276,
+      "tokens_seen": 207552512
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009465596790371114,
+      "loss": 1.9654,
+      "theoretical_loss": 4.363515133758084,
+      "tokens_seen": 207618048
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009465396188565697,
+      "loss": 1.9998,
+      "theoretical_loss": 4.363344927467009,
+      "tokens_seen": 207683584
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009465195586760281,
+      "loss": 1.9099,
+      "theoretical_loss": 4.36317478991057,
+      "tokens_seen": 207749120
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009464994984954865,
+      "loss": 1.8827,
+      "theoretical_loss": 4.363004721039341,
+      "tokens_seen": 207814656
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009464794383149448,
+      "loss": 1.8781,
+      "theoretical_loss": 4.3628347208039475,
+      "tokens_seen": 207880192
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009464593781344032,
+      "loss": 1.7533,
+      "theoretical_loss": 4.362664789155063,
+      "tokens_seen": 207945728
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009464393179538616,
+      "loss": 1.661,
+      "theoretical_loss": 4.362494926043414,
+      "tokens_seen": 208011264
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009464192577733199,
+      "loss": 1.7946,
+      "theoretical_loss": 4.362325131419779,
+      "tokens_seen": 208076800
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009463991975927784,
+      "loss": 1.9939,
+      "theoretical_loss": 4.362155405234985,
+      "tokens_seen": 208142336
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009463791374122367,
+      "loss": 1.8894,
+      "theoretical_loss": 4.361985747439911,
+      "tokens_seen": 208207872
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009463590772316951,
+      "loss": 1.9331,
+      "theoretical_loss": 4.361816157985485,
+      "tokens_seen": 208273408
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009463390170511535,
+      "loss": 1.8647,
+      "theoretical_loss": 4.361646636822689,
+      "tokens_seen": 208338944
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009463189568706119,
+      "loss": 2.0205,
+      "theoretical_loss": 4.361477183902554,
+      "tokens_seen": 208404480
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009462988966900702,
+      "loss": 1.8844,
+      "theoretical_loss": 4.361307799176159,
+      "tokens_seen": 208470016
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009462788365095286,
+      "loss": 1.8218,
+      "theoretical_loss": 4.361138482594637,
+      "tokens_seen": 208535552
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009462587763289869,
+      "loss": 1.8755,
+      "theoretical_loss": 4.360969234109169,
+      "tokens_seen": 208601088
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009462387161484454,
+      "loss": 1.8704,
+      "theoretical_loss": 4.360800053670989,
+      "tokens_seen": 208666624
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009462186559679037,
+      "loss": 1.7776,
+      "theoretical_loss": 4.360630941231378,
+      "tokens_seen": 208732160
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009461985957873621,
+      "loss": 1.7952,
+      "theoretical_loss": 4.360461896741668,
+      "tokens_seen": 208797696
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009461785356068204,
+      "loss": 1.8615,
+      "theoretical_loss": 4.360292920153244,
+      "tokens_seen": 208863232
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009461584754262789,
+      "loss": 1.8697,
+      "theoretical_loss": 4.360124011417536,
+      "tokens_seen": 208928768
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009461384152457372,
+      "loss": 1.9524,
+      "theoretical_loss": 4.359955170486029,
+      "tokens_seen": 208994304
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009461183550651956,
+      "loss": 1.7694,
+      "theoretical_loss": 4.359786397310254,
+      "tokens_seen": 209059840
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000946098294884654,
+      "loss": 1.7871,
+      "theoretical_loss": 4.359617691841795,
+      "tokens_seen": 209125376
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009460782347041123,
+      "loss": 1.9203,
+      "theoretical_loss": 4.359449054032282,
+      "tokens_seen": 209190912
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009460581745235707,
+      "loss": 1.8939,
+      "theoretical_loss": 4.3592804838334,
+      "tokens_seen": 209256448
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009460381143430291,
+      "loss": 1.7299,
+      "theoretical_loss": 4.359111981196877,
+      "tokens_seen": 209321984
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009460180541624874,
+      "loss": 1.9227,
+      "theoretical_loss": 4.358943546074497,
+      "tokens_seen": 209387520
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009459979939819459,
+      "loss": 1.8355,
+      "theoretical_loss": 4.358775178418089,
+      "tokens_seen": 209453056
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009459779338014042,
+      "loss": 1.9658,
+      "theoretical_loss": 4.358606878179533,
+      "tokens_seen": 209518592
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009459578736208626,
+      "loss": 1.8921,
+      "theoretical_loss": 4.358438645310758,
+      "tokens_seen": 209584128
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009459378134403211,
+      "loss": 1.7757,
+      "theoretical_loss": 4.3582704797637435,
+      "tokens_seen": 209649664
+    },
+    {
+      "epoch": 0.06,
+      "objective/train/advantage_avg": 0.49177494645118713,
+      "objective/train/docs_used": 82917,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.6986033916473389,
+      "objective/train/original_loss": 1.6986033916473389,
+      "objective/train/theoretical_loss": 4.358102381490517,
+      "objective/train/tokens_used": 230175200,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24229730665683746,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050409197807312,
+      "objective/train/weighted_lm_loss": 1.7842323780059814,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.358102381490517,
+      "tokens_seen": 209715200
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009459177532597793,
+      "loss": 1.8477,
+      "theoretical_loss": 4.358102381490517,
+      "tokens_seen": 209715200
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009458976930792378,
+      "loss": 1.8678,
+      "theoretical_loss": 4.357934350443156,
+      "tokens_seen": 209780736
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009458776328986961,
+      "loss": 1.8636,
+      "theoretical_loss": 4.357766386573785,
+      "tokens_seen": 209846272
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009458575727181545,
+      "loss": 1.8975,
+      "theoretical_loss": 4.357598489834581,
+      "tokens_seen": 209911808
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009458375125376129,
+      "loss": 1.8138,
+      "theoretical_loss": 4.3574306601777675,
+      "tokens_seen": 209977344
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009458174523570712,
+      "loss": 1.8093,
+      "theoretical_loss": 4.357262897555617,
+      "tokens_seen": 210042880
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009457973921765296,
+      "loss": 1.8011,
+      "theoretical_loss": 4.357095201920452,
+      "tokens_seen": 210108416
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009457773319959881,
+      "loss": 1.7639,
+      "theoretical_loss": 4.356927573224643,
+      "tokens_seen": 210173952
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009457572718154463,
+      "loss": 1.7829,
+      "theoretical_loss": 4.356760011420608,
+      "tokens_seen": 210239488
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009457372116349048,
+      "loss": 1.9724,
+      "theoretical_loss": 4.356592516460817,
+      "tokens_seen": 210305024
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000945717151454363,
+      "loss": 1.992,
+      "theoretical_loss": 4.356425088297786,
+      "tokens_seen": 210370560
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009456970912738215,
+      "loss": 1.8751,
+      "theoretical_loss": 4.3562577268840785,
+      "tokens_seen": 210436096
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009456770310932799,
+      "loss": 1.9101,
+      "theoretical_loss": 4.3560904321723095,
+      "tokens_seen": 210501632
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009456569709127382,
+      "loss": 1.9617,
+      "theoretical_loss": 4.355923204115141,
+      "tokens_seen": 210567168
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009456369107321966,
+      "loss": 1.9317,
+      "theoretical_loss": 4.3557560426652815,
+      "tokens_seen": 210632704
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009456168505516551,
+      "loss": 1.7983,
+      "theoretical_loss": 4.355588947775491,
+      "tokens_seen": 210698240
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009455967903711133,
+      "loss": 1.8471,
+      "theoretical_loss": 4.355421919398576,
+      "tokens_seen": 210763776
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009455767301905718,
+      "loss": 1.8127,
+      "theoretical_loss": 4.35525495748739,
+      "tokens_seen": 210829312
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00094555667001003,
+      "loss": 1.8958,
+      "theoretical_loss": 4.355088061994837,
+      "tokens_seen": 210894848
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009455366098294885,
+      "loss": 1.9029,
+      "theoretical_loss": 4.354921232873868,
+      "tokens_seen": 210960384
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009455165496489469,
+      "loss": 1.876,
+      "theoretical_loss": 4.354754470077481,
+      "tokens_seen": 211025920
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009454964894684052,
+      "loss": 1.9396,
+      "theoretical_loss": 4.354587773558722,
+      "tokens_seen": 211091456
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009454764292878636,
+      "loss": 1.8428,
+      "theoretical_loss": 4.354421143270687,
+      "tokens_seen": 211156992
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000945456369107322,
+      "loss": 1.7118,
+      "theoretical_loss": 4.354254579166517,
+      "tokens_seen": 211222528
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009454363089267803,
+      "loss": 1.8104,
+      "theoretical_loss": 4.354088081199402,
+      "tokens_seen": 211288064
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009454162487462388,
+      "loss": 1.8452,
+      "theoretical_loss": 4.353921649322579,
+      "tokens_seen": 211353600
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009453961885656971,
+      "loss": 1.8087,
+      "theoretical_loss": 4.353755283489334,
+      "tokens_seen": 211419136
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009453761283851555,
+      "loss": 1.8845,
+      "theoretical_loss": 4.353588983652999,
+      "tokens_seen": 211484672
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009453560682046139,
+      "loss": 1.8368,
+      "theoretical_loss": 4.3534227497669535,
+      "tokens_seen": 211550208
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009453360080240723,
+      "loss": 1.9266,
+      "theoretical_loss": 4.353256581784625,
+      "tokens_seen": 211615744
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009453159478435306,
+      "loss": 1.7949,
+      "theoretical_loss": 4.353090479659488,
+      "tokens_seen": 211681280
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000945295887662989,
+      "loss": 1.8264,
+      "theoretical_loss": 4.3529244433450645,
+      "tokens_seen": 211746816
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009452758274824473,
+      "loss": 1.7892,
+      "theoretical_loss": 4.352758472794923,
+      "tokens_seen": 211812352
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009452557673019058,
+      "loss": 1.8234,
+      "theoretical_loss": 4.35259256796268,
+      "tokens_seen": 211877888
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009452357071213641,
+      "loss": 1.8501,
+      "theoretical_loss": 4.352426728801999,
+      "tokens_seen": 211943424
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009452156469408225,
+      "loss": 1.8889,
+      "theoretical_loss": 4.352260955266589,
+      "tokens_seen": 212008960
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009451955867602808,
+      "loss": 1.9609,
+      "theoretical_loss": 4.352095247310208,
+      "tokens_seen": 212074496
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009451755265797393,
+      "loss": 1.7887,
+      "theoretical_loss": 4.351929604886658,
+      "tokens_seen": 212140032
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009451554663991976,
+      "loss": 1.8833,
+      "theoretical_loss": 4.351764027949793,
+      "tokens_seen": 212205568
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000945135406218656,
+      "loss": 1.9028,
+      "theoretical_loss": 4.3515985164535085,
+      "tokens_seen": 212271104
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009451153460381143,
+      "loss": 1.8628,
+      "theoretical_loss": 4.351433070351748,
+      "tokens_seen": 212336640
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009450952858575727,
+      "loss": 1.9107,
+      "theoretical_loss": 4.351267689598503,
+      "tokens_seen": 212402176
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009450752256770311,
+      "loss": 1.9297,
+      "theoretical_loss": 4.3511023741478105,
+      "tokens_seen": 212467712
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009450551654964895,
+      "loss": 1.8527,
+      "theoretical_loss": 4.350937123953755,
+      "tokens_seen": 212533248
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009450351053159478,
+      "loss": 1.8359,
+      "theoretical_loss": 4.350771938970466,
+      "tokens_seen": 212598784
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009450150451354063,
+      "loss": 1.9545,
+      "theoretical_loss": 4.35060681915212,
+      "tokens_seen": 212664320
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009449949849548646,
+      "loss": 1.6537,
+      "theoretical_loss": 4.35044176445294,
+      "tokens_seen": 212729856
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000944974924774323,
+      "loss": 1.9184,
+      "theoretical_loss": 4.350276774827195,
+      "tokens_seen": 212795392
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009449548645937814,
+      "loss": 1.7896,
+      "theoretical_loss": 4.350111850229202,
+      "tokens_seen": 212860928
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009449348044132397,
+      "loss": 1.8893,
+      "theoretical_loss": 4.3499469906133195,
+      "tokens_seen": 212926464
+    },
+    {
+      "epoch": 0.06,
+      "objective/train/advantage_avg": 0.49271607398986816,
+      "objective/train/docs_used": 84140,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.6249154806137085,
+      "objective/train/original_loss": 1.624915599822998,
+      "objective/train/theoretical_loss": 4.349782195933957,
+      "objective/train/tokens_used": 233452000,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24310225248336792,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505075454711914,
+      "objective/train/weighted_lm_loss": 1.7071648836135864,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.349782195933957,
+      "tokens_seen": 212992000
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009449147442326981,
+      "loss": 1.723,
+      "theoretical_loss": 4.349782195933957,
+      "tokens_seen": 212992000
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009448946840521565,
+      "loss": 1.7778,
+      "theoretical_loss": 4.349617466145568,
+      "tokens_seen": 213057536
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009448746238716148,
+      "loss": 1.7859,
+      "theoretical_loss": 4.34945280120265,
+      "tokens_seen": 213123072
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009448545636910733,
+      "loss": 1.8328,
+      "theoretical_loss": 4.349288201059751,
+      "tokens_seen": 213188608
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009448345035105315,
+      "loss": 1.7755,
+      "theoretical_loss": 4.34912366567146,
+      "tokens_seen": 213254144
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00094481444332999,
+      "loss": 1.9151,
+      "theoretical_loss": 4.348959194992415,
+      "tokens_seen": 213319680
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009447943831494484,
+      "loss": 1.8385,
+      "theoretical_loss": 4.348794788977298,
+      "tokens_seen": 213385216
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009447743229689067,
+      "loss": 1.8569,
+      "theoretical_loss": 4.348630447580839,
+      "tokens_seen": 213450752
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009447542627883651,
+      "loss": 1.8746,
+      "theoretical_loss": 4.34846617075781,
+      "tokens_seen": 213516288
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009447342026078234,
+      "loss": 1.7508,
+      "theoretical_loss": 4.34830195846303,
+      "tokens_seen": 213581824
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009447141424272818,
+      "loss": 1.9108,
+      "theoretical_loss": 4.348137810651366,
+      "tokens_seen": 213647360
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009446940822467403,
+      "loss": 1.8,
+      "theoretical_loss": 4.347973727277726,
+      "tokens_seen": 213712896
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009446740220661985,
+      "loss": 1.8573,
+      "theoretical_loss": 4.347809708297067,
+      "tokens_seen": 213778432
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000944653961885657,
+      "loss": 1.7998,
+      "theoretical_loss": 4.347645753664389,
+      "tokens_seen": 213843968
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009446339017051154,
+      "loss": 1.7913,
+      "theoretical_loss": 4.347481863334738,
+      "tokens_seen": 213909504
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009446138415245737,
+      "loss": 1.849,
+      "theoretical_loss": 4.347318037263206,
+      "tokens_seen": 213975040
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009445937813440321,
+      "loss": 1.8244,
+      "theoretical_loss": 4.34715427540493,
+      "tokens_seen": 214040576
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009445737211634905,
+      "loss": 1.8991,
+      "theoretical_loss": 4.34699057771509,
+      "tokens_seen": 214106112
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009445536609829488,
+      "loss": 1.8225,
+      "theoretical_loss": 4.346826944148912,
+      "tokens_seen": 214171648
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009445336008024073,
+      "loss": 1.8634,
+      "theoretical_loss": 4.346663374661669,
+      "tokens_seen": 214237184
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009445135406218655,
+      "loss": 1.8842,
+      "theoretical_loss": 4.346499869208676,
+      "tokens_seen": 214302720
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000944493480441324,
+      "loss": 1.888,
+      "theoretical_loss": 4.346336427745294,
+      "tokens_seen": 214368256
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009444734202607823,
+      "loss": 1.8777,
+      "theoretical_loss": 4.3461730502269305,
+      "tokens_seen": 214433792
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0009444533600802407,
+      "loss": 1.9699,
+      "theoretical_loss": 4.346009736609034,
+      "tokens_seen": 214499328
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009444332998996992,
+      "loss": 1.7312,
+      "theoretical_loss": 4.345846486847099,
+      "tokens_seen": 214564864
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009444132397191575,
+      "loss": 1.89,
+      "theoretical_loss": 4.345683300896668,
+      "tokens_seen": 214630400
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009443931795386159,
+      "loss": 1.7744,
+      "theoretical_loss": 4.345520178713323,
+      "tokens_seen": 214695936
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009443731193580743,
+      "loss": 1.8952,
+      "theoretical_loss": 4.345357120252692,
+      "tokens_seen": 214761472
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009443530591775327,
+      "loss": 1.988,
+      "theoretical_loss": 4.34519412547045,
+      "tokens_seen": 214827008
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000944332998996991,
+      "loss": 1.908,
+      "theoretical_loss": 4.345031194322313,
+      "tokens_seen": 214892544
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009443129388164494,
+      "loss": 1.9555,
+      "theoretical_loss": 4.344868326764045,
+      "tokens_seen": 214958080
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009442928786359077,
+      "loss": 1.9617,
+      "theoretical_loss": 4.344705522751449,
+      "tokens_seen": 215023616
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009442728184553662,
+      "loss": 1.8555,
+      "theoretical_loss": 4.344542782240376,
+      "tokens_seen": 215089152
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009442527582748245,
+      "loss": 1.8447,
+      "theoretical_loss": 4.344380105186721,
+      "tokens_seen": 215154688
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009442326980942829,
+      "loss": 1.8825,
+      "theoretical_loss": 4.344217491546422,
+      "tokens_seen": 215220224
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009442126379137412,
+      "loss": 2.0065,
+      "theoretical_loss": 4.344054941275459,
+      "tokens_seen": 215285760
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009441925777331997,
+      "loss": 1.8606,
+      "theoretical_loss": 4.343892454329863,
+      "tokens_seen": 215351296
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000944172517552658,
+      "loss": 1.8549,
+      "theoretical_loss": 4.343730030665699,
+      "tokens_seen": 215416832
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009441524573721164,
+      "loss": 1.691,
+      "theoretical_loss": 4.343567670239084,
+      "tokens_seen": 215482368
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009441323971915747,
+      "loss": 1.7313,
+      "theoretical_loss": 4.343405373006175,
+      "tokens_seen": 215547904
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009441123370110331,
+      "loss": 1.9053,
+      "theoretical_loss": 4.343243138923173,
+      "tokens_seen": 215613440
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009440922768304915,
+      "loss": 2.0116,
+      "theoretical_loss": 4.343080967946323,
+      "tokens_seen": 215678976
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009440722166499499,
+      "loss": 1.9259,
+      "theoretical_loss": 4.342918860031914,
+      "tokens_seen": 215744512
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009440521564694082,
+      "loss": 1.7324,
+      "theoretical_loss": 4.342756815136278,
+      "tokens_seen": 215810048
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009440320962888667,
+      "loss": 1.9033,
+      "theoretical_loss": 4.3425948332157915,
+      "tokens_seen": 215875584
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000944012036108325,
+      "loss": 1.7331,
+      "theoretical_loss": 4.3424329142268725,
+      "tokens_seen": 215941120
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009439919759277834,
+      "loss": 1.7929,
+      "theoretical_loss": 4.342271058125983,
+      "tokens_seen": 216006656
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009439719157472418,
+      "loss": 1.9002,
+      "theoretical_loss": 4.342109264869631,
+      "tokens_seen": 216072192
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009439518555667001,
+      "loss": 1.9297,
+      "theoretical_loss": 4.341947534414364,
+      "tokens_seen": 216137728
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009439317953861585,
+      "loss": 1.83,
+      "theoretical_loss": 4.341785866716774,
+      "tokens_seen": 216203264
+    },
+    {
+      "epoch": 0.07,
+      "objective/train/advantage_avg": 0.4911940395832062,
+      "objective/train/docs_used": 85191,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.79802668094635,
+      "objective/train/original_loss": 1.79802668094635,
+      "objective/train/theoretical_loss": 4.341624261733497,
+      "objective/train/tokens_used": 236728800,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24179859459400177,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503485202789307,
+      "objective/train/weighted_lm_loss": 1.8884522914886475,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.341624261733497,
+      "tokens_seen": 216268800
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000943911735205617,
+      "loss": 1.8136,
+      "theoretical_loss": 4.341624261733497,
+      "tokens_seen": 216268800
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009438916750250752,
+      "loss": 1.8533,
+      "theoretical_loss": 4.341462719421211,
+      "tokens_seen": 216334336
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009438716148445337,
+      "loss": 1.8464,
+      "theoretical_loss": 4.341301239736639,
+      "tokens_seen": 216399872
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009438515546639919,
+      "loss": 1.9912,
+      "theoretical_loss": 4.341139822636544,
+      "tokens_seen": 216465408
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009438314944834504,
+      "loss": 1.8442,
+      "theoretical_loss": 4.340978468077735,
+      "tokens_seen": 216530944
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009438114343029088,
+      "loss": 1.9045,
+      "theoretical_loss": 4.34081717601706,
+      "tokens_seen": 216596480
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009437913741223671,
+      "loss": 1.7926,
+      "theoretical_loss": 4.340655946411413,
+      "tokens_seen": 216662016
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009437713139418255,
+      "loss": 1.896,
+      "theoretical_loss": 4.340494779217732,
+      "tokens_seen": 216727552
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009437512537612838,
+      "loss": 1.8232,
+      "theoretical_loss": 4.340333674392992,
+      "tokens_seen": 216793088
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009437311935807422,
+      "loss": 1.9165,
+      "theoretical_loss": 4.340172631894218,
+      "tokens_seen": 216858624
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009437111334002007,
+      "loss": 1.9249,
+      "theoretical_loss": 4.340011651678472,
+      "tokens_seen": 216924160
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009436910732196589,
+      "loss": 1.757,
+      "theoretical_loss": 4.33985073370286,
+      "tokens_seen": 216989696
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009436710130391174,
+      "loss": 1.8661,
+      "theoretical_loss": 4.339689877924531,
+      "tokens_seen": 217055232
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009436509528585758,
+      "loss": 1.8468,
+      "theoretical_loss": 4.339529084300676,
+      "tokens_seen": 217120768
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009436308926780341,
+      "loss": 1.8,
+      "theoretical_loss": 4.339368352788531,
+      "tokens_seen": 217186304
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009436108324974925,
+      "loss": 1.8137,
+      "theoretical_loss": 4.339207683345369,
+      "tokens_seen": 217251840
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009435907723169509,
+      "loss": 1.8983,
+      "theoretical_loss": 4.3390470759285105,
+      "tokens_seen": 217317376
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009435707121364092,
+      "loss": 1.8911,
+      "theoretical_loss": 4.338886530495314,
+      "tokens_seen": 217382912
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009435506519558677,
+      "loss": 1.8343,
+      "theoretical_loss": 4.338726047003183,
+      "tokens_seen": 217448448
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009435305917753259,
+      "loss": 1.797,
+      "theoretical_loss": 4.338565625409563,
+      "tokens_seen": 217513984
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009435105315947844,
+      "loss": 1.8644,
+      "theoretical_loss": 4.338405265671941,
+      "tokens_seen": 217579520
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009434904714142427,
+      "loss": 1.6793,
+      "theoretical_loss": 4.338244967747844,
+      "tokens_seen": 217645056
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009434704112337011,
+      "loss": 1.834,
+      "theoretical_loss": 4.338084731594844,
+      "tokens_seen": 217710592
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009434503510531595,
+      "loss": 1.7865,
+      "theoretical_loss": 4.3379245571705525,
+      "tokens_seen": 217776128
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009434302908726179,
+      "loss": 1.7227,
+      "theoretical_loss": 4.337764444432625,
+      "tokens_seen": 217841664
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009434102306920762,
+      "loss": 1.8882,
+      "theoretical_loss": 4.337604393338758,
+      "tokens_seen": 217907200
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009433901705115347,
+      "loss": 1.8603,
+      "theoretical_loss": 4.337444403846687,
+      "tokens_seen": 217972736
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000943370110330993,
+      "loss": 1.9331,
+      "theoretical_loss": 4.3372844759141955,
+      "tokens_seen": 218038272
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009433500501504514,
+      "loss": 1.7618,
+      "theoretical_loss": 4.337124609499101,
+      "tokens_seen": 218103808
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009433299899699097,
+      "loss": 1.8607,
+      "theoretical_loss": 4.336964804559269,
+      "tokens_seen": 218169344
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009433099297893681,
+      "loss": 1.8718,
+      "theoretical_loss": 4.336805061052601,
+      "tokens_seen": 218234880
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009432898696088265,
+      "loss": 1.6282,
+      "theoretical_loss": 4.336645378937046,
+      "tokens_seen": 218300416
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009432698094282849,
+      "loss": 1.7722,
+      "theoretical_loss": 4.336485758170589,
+      "tokens_seen": 218365952
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009432497492477432,
+      "loss": 1.8395,
+      "theoretical_loss": 4.336326198711259,
+      "tokens_seen": 218431488
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009432296890672016,
+      "loss": 1.8712,
+      "theoretical_loss": 4.336166700517125,
+      "tokens_seen": 218497024
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00094320962888666,
+      "loss": 1.8655,
+      "theoretical_loss": 4.3360072635463,
+      "tokens_seen": 218562560
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009431895687061184,
+      "loss": 1.989,
+      "theoretical_loss": 4.335847887756934,
+      "tokens_seen": 218628096
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009431695085255767,
+      "loss": 1.8191,
+      "theoretical_loss": 4.335688573107221,
+      "tokens_seen": 218693632
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009431494483450351,
+      "loss": 1.8258,
+      "theoretical_loss": 4.3355293195553966,
+      "tokens_seen": 218759168
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009431293881644934,
+      "loss": 1.8348,
+      "theoretical_loss": 4.335370127059735,
+      "tokens_seen": 218824704
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009431093279839519,
+      "loss": 1.7862,
+      "theoretical_loss": 4.335210995578553,
+      "tokens_seen": 218890240
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009430892678034102,
+      "loss": 1.8832,
+      "theoretical_loss": 4.335051925070209,
+      "tokens_seen": 218955776
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009430692076228686,
+      "loss": 1.8708,
+      "theoretical_loss": 4.334892915493099,
+      "tokens_seen": 219021312
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000943049147442327,
+      "loss": 1.73,
+      "theoretical_loss": 4.334733966805665,
+      "tokens_seen": 219086848
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009430290872617854,
+      "loss": 1.9105,
+      "theoretical_loss": 4.334575078966383,
+      "tokens_seen": 219152384
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009430090270812437,
+      "loss": 1.8589,
+      "theoretical_loss": 4.334416251933778,
+      "tokens_seen": 219217920
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009429889669007022,
+      "loss": 1.8095,
+      "theoretical_loss": 4.334257485666408,
+      "tokens_seen": 219283456
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009429689067201604,
+      "loss": 1.9426,
+      "theoretical_loss": 4.334098780122876,
+      "tokens_seen": 219348992
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009429488465396189,
+      "loss": 1.9143,
+      "theoretical_loss": 4.333940135261823,
+      "tokens_seen": 219414528
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009429287863590774,
+      "loss": 1.8288,
+      "theoretical_loss": 4.333781551041935,
+      "tokens_seen": 219480064
+    },
+    {
+      "epoch": 0.07,
+      "objective/train/advantage_avg": 0.4948246479034424,
+      "objective/train/docs_used": 86440,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.6815640926361084,
+      "objective/train/original_loss": 1.6815643310546875,
+      "objective/train/theoretical_loss": 4.3336230274219325,
+      "objective/train/tokens_used": 240005600,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24505706131458282,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0507283210754395,
+      "objective/train/weighted_lm_loss": 1.7667181491851807,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.3336230274219325,
+      "tokens_seen": 219545600
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009429087261785356,
+      "loss": 1.7774,
+      "theoretical_loss": 4.3336230274219325,
+      "tokens_seen": 219545600
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009428886659979941,
+      "loss": 1.7581,
+      "theoretical_loss": 4.333464564360581,
+      "tokens_seen": 219611136
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009428686058174523,
+      "loss": 1.9298,
+      "theoretical_loss": 4.333306161816684,
+      "tokens_seen": 219676672
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009428485456369108,
+      "loss": 1.9582,
+      "theoretical_loss": 4.333147819749085,
+      "tokens_seen": 219742208
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009428284854563692,
+      "loss": 1.9185,
+      "theoretical_loss": 4.332989538116671,
+      "tokens_seen": 219807744
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009428084252758275,
+      "loss": 1.8663,
+      "theoretical_loss": 4.332831316878364,
+      "tokens_seen": 219873280
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009427883650952859,
+      "loss": 1.8499,
+      "theoretical_loss": 4.332673155993131,
+      "tokens_seen": 219938816
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009427683049147442,
+      "loss": 1.894,
+      "theoretical_loss": 4.332515055419976,
+      "tokens_seen": 220004352
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009427482447342026,
+      "loss": 1.7753,
+      "theoretical_loss": 4.332357015117946,
+      "tokens_seen": 220069888
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009427281845536611,
+      "loss": 1.8143,
+      "theoretical_loss": 4.332199035046124,
+      "tokens_seen": 220135424
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009427081243731193,
+      "loss": 1.7558,
+      "theoretical_loss": 4.332041115163636,
+      "tokens_seen": 220200960
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009426880641925778,
+      "loss": 1.8822,
+      "theoretical_loss": 4.3318832554296485,
+      "tokens_seen": 220266496
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009426680040120362,
+      "loss": 1.8592,
+      "theoretical_loss": 4.3317254558033635,
+      "tokens_seen": 220332032
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009426479438314945,
+      "loss": 1.9003,
+      "theoretical_loss": 4.331567716244027,
+      "tokens_seen": 220397568
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009426278836509529,
+      "loss": 1.9003,
+      "theoretical_loss": 4.331410036710925,
+      "tokens_seen": 220463104
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009426078234704113,
+      "loss": 1.792,
+      "theoretical_loss": 4.331252417163379,
+      "tokens_seen": 220528640
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009425877632898696,
+      "loss": 1.8564,
+      "theoretical_loss": 4.331094857560754,
+      "tokens_seen": 220594176
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009425677031093281,
+      "loss": 1.8892,
+      "theoretical_loss": 4.330937357862453,
+      "tokens_seen": 220659712
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009425476429287863,
+      "loss": 1.8259,
+      "theoretical_loss": 4.330779918027919,
+      "tokens_seen": 220725248
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009425275827482448,
+      "loss": 1.9108,
+      "theoretical_loss": 4.330622538016635,
+      "tokens_seen": 220790784
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009425075225677031,
+      "loss": 1.8322,
+      "theoretical_loss": 4.330465217788122,
+      "tokens_seen": 220856320
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009424874623871615,
+      "loss": 1.933,
+      "theoretical_loss": 4.33030795730194,
+      "tokens_seen": 220921856
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009424674022066199,
+      "loss": 1.808,
+      "theoretical_loss": 4.330150756517692,
+      "tokens_seen": 220987392
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009424473420260783,
+      "loss": 1.716,
+      "theoretical_loss": 4.329993615395017,
+      "tokens_seen": 221052928
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009424272818455366,
+      "loss": 1.8106,
+      "theoretical_loss": 4.329836533893593,
+      "tokens_seen": 221118464
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009424072216649951,
+      "loss": 1.9543,
+      "theoretical_loss": 4.329679511973138,
+      "tokens_seen": 221184000
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009423871614844533,
+      "loss": 1.8496,
+      "theoretical_loss": 4.3295225495934115,
+      "tokens_seen": 221249536
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009423671013039118,
+      "loss": 1.6599,
+      "theoretical_loss": 4.3293656467142085,
+      "tokens_seen": 221315072
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009423470411233701,
+      "loss": 1.8111,
+      "theoretical_loss": 4.329208803295365,
+      "tokens_seen": 221380608
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009423269809428285,
+      "loss": 1.734,
+      "theoretical_loss": 4.329052019296754,
+      "tokens_seen": 221446144
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009423069207622869,
+      "loss": 1.795,
+      "theoretical_loss": 4.328895294678292,
+      "tokens_seen": 221511680
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009422868605817453,
+      "loss": 1.9343,
+      "theoretical_loss": 4.328738629399928,
+      "tokens_seen": 221577216
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009422668004012036,
+      "loss": 1.7639,
+      "theoretical_loss": 4.328582023421655,
+      "tokens_seen": 221642752
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000942246740220662,
+      "loss": 1.8256,
+      "theoretical_loss": 4.328425476703503,
+      "tokens_seen": 221708288
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009422266800401204,
+      "loss": 1.8873,
+      "theoretical_loss": 4.32826898920554,
+      "tokens_seen": 221773824
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009422066198595788,
+      "loss": 1.8476,
+      "theoretical_loss": 4.328112560887875,
+      "tokens_seen": 221839360
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009421865596790371,
+      "loss": 1.9737,
+      "theoretical_loss": 4.327956191710651,
+      "tokens_seen": 221904896
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009421664994984955,
+      "loss": 1.884,
+      "theoretical_loss": 4.327799881634056,
+      "tokens_seen": 221970432
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009421464393179538,
+      "loss": 1.8819,
+      "theoretical_loss": 4.3276436306183115,
+      "tokens_seen": 222035968
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009421263791374123,
+      "loss": 1.7373,
+      "theoretical_loss": 4.327487438623679,
+      "tokens_seen": 222101504
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009421063189568706,
+      "loss": 1.8465,
+      "theoretical_loss": 4.32733130561046,
+      "tokens_seen": 222167040
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000942086258776329,
+      "loss": 1.7857,
+      "theoretical_loss": 4.327175231538991,
+      "tokens_seen": 222232576
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009420661985957874,
+      "loss": 1.7946,
+      "theoretical_loss": 4.327019216369651,
+      "tokens_seen": 222298112
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009420461384152458,
+      "loss": 1.8871,
+      "theoretical_loss": 4.326863260062854,
+      "tokens_seen": 222363648
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009420260782347041,
+      "loss": 1.7786,
+      "theoretical_loss": 4.326707362579054,
+      "tokens_seen": 222429184
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009420060180541626,
+      "loss": 1.7791,
+      "theoretical_loss": 4.326551523878742,
+      "tokens_seen": 222494720
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009419859578736208,
+      "loss": 1.7992,
+      "theoretical_loss": 4.32639574392245,
+      "tokens_seen": 222560256
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009419658976930793,
+      "loss": 1.735,
+      "theoretical_loss": 4.326240022670743,
+      "tokens_seen": 222625792
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009419458375125376,
+      "loss": 1.8356,
+      "theoretical_loss": 4.326084360084229,
+      "tokens_seen": 222691328
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000941925777331996,
+      "loss": 1.72,
+      "theoretical_loss": 4.325928756123551,
+      "tokens_seen": 222756864
+    },
+    {
+      "epoch": 0.07,
+      "objective/train/advantage_avg": 0.4933840334415436,
+      "objective/train/docs_used": 87601,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.578743815422058,
+      "objective/train/original_loss": 1.5787436962127686,
+      "objective/train/theoretical_loss": 4.325773210749392,
+      "objective/train/tokens_used": 243282400,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24385935068130493,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505781173706055,
+      "objective/train/weighted_lm_loss": 1.658752679824829,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.325773210749392,
+      "tokens_seen": 222822400
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009419057171514544,
+      "loss": 1.7028,
+      "theoretical_loss": 4.325773210749392,
+      "tokens_seen": 222822400
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009418856569709127,
+      "loss": 1.9186,
+      "theoretical_loss": 4.325617723922472,
+      "tokens_seen": 222887936
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009418655967903711,
+      "loss": 1.7035,
+      "theoretical_loss": 4.325462295603547,
+      "tokens_seen": 222953472
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009418455366098296,
+      "loss": 1.8367,
+      "theoretical_loss": 4.325306925753415,
+      "tokens_seen": 223019008
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009418254764292878,
+      "loss": 1.8699,
+      "theoretical_loss": 4.325151614332908,
+      "tokens_seen": 223084544
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009418054162487463,
+      "loss": 1.8548,
+      "theoretical_loss": 4.324996361302899,
+      "tokens_seen": 223150080
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009417853560682045,
+      "loss": 1.8911,
+      "theoretical_loss": 4.324841166624294,
+      "tokens_seen": 223215616
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000941765295887663,
+      "loss": 1.8347,
+      "theoretical_loss": 4.324686030258042,
+      "tokens_seen": 223281152
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009417452357071214,
+      "loss": 1.8127,
+      "theoretical_loss": 4.3245309521651265,
+      "tokens_seen": 223346688
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009417251755265797,
+      "loss": 1.8398,
+      "theoretical_loss": 4.324375932306568,
+      "tokens_seen": 223412224
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009417051153460381,
+      "loss": 1.8697,
+      "theoretical_loss": 4.324220970643426,
+      "tokens_seen": 223477760
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009416850551654966,
+      "loss": 1.8429,
+      "theoretical_loss": 4.324066067136798,
+      "tokens_seen": 223543296
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009416649949849548,
+      "loss": 1.807,
+      "theoretical_loss": 4.323911221747817,
+      "tokens_seen": 223608832
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009416449348044133,
+      "loss": 1.9073,
+      "theoretical_loss": 4.323756434437655,
+      "tokens_seen": 223674368
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009416248746238715,
+      "loss": 1.7969,
+      "theoretical_loss": 4.323601705167521,
+      "tokens_seen": 223739904
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00094160481444333,
+      "loss": 1.8434,
+      "theoretical_loss": 4.323447033898661,
+      "tokens_seen": 223805440
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009415847542627884,
+      "loss": 1.9763,
+      "theoretical_loss": 4.323292420592356,
+      "tokens_seen": 223870976
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009415646940822467,
+      "loss": 1.7727,
+      "theoretical_loss": 4.323137865209928,
+      "tokens_seen": 223936512
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009415446339017051,
+      "loss": 1.8935,
+      "theoretical_loss": 4.322983367712735,
+      "tokens_seen": 224002048
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009415245737211635,
+      "loss": 1.7683,
+      "theoretical_loss": 4.322828928062171,
+      "tokens_seen": 224067584
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009415045135406218,
+      "loss": 1.918,
+      "theoretical_loss": 4.322674546219666,
+      "tokens_seen": 224133120
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009414844533600803,
+      "loss": 1.8897,
+      "theoretical_loss": 4.322520222146689,
+      "tokens_seen": 224198656
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009414643931795386,
+      "loss": 1.811,
+      "theoretical_loss": 4.322365955804749,
+      "tokens_seen": 224264192
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000941444332998997,
+      "loss": 1.8297,
+      "theoretical_loss": 4.322211747155383,
+      "tokens_seen": 224329728
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009414242728184555,
+      "loss": 1.8643,
+      "theoretical_loss": 4.322057596160174,
+      "tokens_seen": 224395264
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009414042126379137,
+      "loss": 1.6737,
+      "theoretical_loss": 4.321903502780736,
+      "tokens_seen": 224460800
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009413841524573722,
+      "loss": 1.8232,
+      "theoretical_loss": 4.321749466978722,
+      "tokens_seen": 224526336
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009413640922768305,
+      "loss": 1.7283,
+      "theoretical_loss": 4.321595488715822,
+      "tokens_seen": 224591872
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009413440320962889,
+      "loss": 1.8719,
+      "theoretical_loss": 4.321441567953762,
+      "tokens_seen": 224657408
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009413239719157473,
+      "loss": 1.8208,
+      "theoretical_loss": 4.3212877046543055,
+      "tokens_seen": 224722944
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009413039117352057,
+      "loss": 1.7285,
+      "theoretical_loss": 4.32113389877925,
+      "tokens_seen": 224788480
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000941283851554664,
+      "loss": 1.7993,
+      "theoretical_loss": 4.320980150290432,
+      "tokens_seen": 224854016
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009412637913741224,
+      "loss": 1.9261,
+      "theoretical_loss": 4.320826459149725,
+      "tokens_seen": 224919552
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009412437311935808,
+      "loss": 1.911,
+      "theoretical_loss": 4.320672825319035,
+      "tokens_seen": 224985088
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009412236710130392,
+      "loss": 1.9533,
+      "theoretical_loss": 4.320519248760309,
+      "tokens_seen": 225050624
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009412036108324975,
+      "loss": 1.9988,
+      "theoretical_loss": 4.3203657294355295,
+      "tokens_seen": 225116160
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000941183550651956,
+      "loss": 1.8758,
+      "theoretical_loss": 4.3202122673067125,
+      "tokens_seen": 225181696
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009411634904714142,
+      "loss": 1.8191,
+      "theoretical_loss": 4.3200588623359115,
+      "tokens_seen": 225247232
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009411434302908727,
+      "loss": 1.8512,
+      "theoretical_loss": 4.319905514485218,
+      "tokens_seen": 225312768
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000941123370110331,
+      "loss": 1.7275,
+      "theoretical_loss": 4.319752223716758,
+      "tokens_seen": 225378304
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009411033099297894,
+      "loss": 1.7934,
+      "theoretical_loss": 4.319598989992695,
+      "tokens_seen": 225443840
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009410832497492478,
+      "loss": 1.8376,
+      "theoretical_loss": 4.319445813275225,
+      "tokens_seen": 225509376
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009410631895687062,
+      "loss": 2.0349,
+      "theoretical_loss": 4.3192926935265845,
+      "tokens_seen": 225574912
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009410431293881645,
+      "loss": 1.9335,
+      "theoretical_loss": 4.319139630709043,
+      "tokens_seen": 225640448
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000941023069207623,
+      "loss": 1.7679,
+      "theoretical_loss": 4.318986624784908,
+      "tokens_seen": 225705984
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009410030090270812,
+      "loss": 1.8803,
+      "theoretical_loss": 4.3188336757165215,
+      "tokens_seen": 225771520
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009409829488465397,
+      "loss": 1.9487,
+      "theoretical_loss": 4.318680783466261,
+      "tokens_seen": 225837056
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000940962888665998,
+      "loss": 1.7355,
+      "theoretical_loss": 4.318527947996542,
+      "tokens_seen": 225902592
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009409428284854564,
+      "loss": 1.8677,
+      "theoretical_loss": 4.318375169269813,
+      "tokens_seen": 225968128
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009409227683049148,
+      "loss": 1.8113,
+      "theoretical_loss": 4.3182224472485595,
+      "tokens_seen": 226033664
+    },
+    {
+      "epoch": 0.07,
+      "objective/train/advantage_avg": 0.49421149492263794,
+      "objective/train/docs_used": 88830,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.722211480140686,
+      "objective/train/original_loss": 1.7222118377685547,
+      "objective/train/theoretical_loss": 4.3180697818953035,
+      "objective/train/tokens_used": 246559200,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24472033977508545,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050665259361267,
+      "objective/train/weighted_lm_loss": 1.809336543083191,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.3180697818953035,
+      "tokens_seen": 226099200
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009409027081243731,
+      "loss": 1.7996,
+      "theoretical_loss": 4.3180697818953035,
+      "tokens_seen": 226099200
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009408826479438315,
+      "loss": 1.9476,
+      "theoretical_loss": 4.317917173172602,
+      "tokens_seen": 226164736
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00094086258776329,
+      "loss": 1.7379,
+      "theoretical_loss": 4.317764621043046,
+      "tokens_seen": 226230272
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009408425275827482,
+      "loss": 1.8394,
+      "theoretical_loss": 4.317612125469264,
+      "tokens_seen": 226295808
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009408224674022067,
+      "loss": 1.7664,
+      "theoretical_loss": 4.31745968641392,
+      "tokens_seen": 226361344
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009408024072216649,
+      "loss": 1.8748,
+      "theoretical_loss": 4.317307303839712,
+      "tokens_seen": 226426880
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009407823470411234,
+      "loss": 1.7081,
+      "theoretical_loss": 4.317154977709375,
+      "tokens_seen": 226492416
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009407622868605818,
+      "loss": 1.6976,
+      "theoretical_loss": 4.3170027079856785,
+      "tokens_seen": 226557952
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009407422266800401,
+      "loss": 1.795,
+      "theoretical_loss": 4.316850494631427,
+      "tokens_seen": 226623488
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009407221664994985,
+      "loss": 1.881,
+      "theoretical_loss": 4.316698337609461,
+      "tokens_seen": 226689024
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000940702106318957,
+      "loss": 1.8121,
+      "theoretical_loss": 4.3165462368826555,
+      "tokens_seen": 226754560
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009406820461384152,
+      "loss": 1.729,
+      "theoretical_loss": 4.316394192413922,
+      "tokens_seen": 226820096
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009406619859578737,
+      "loss": 1.8137,
+      "theoretical_loss": 4.316242204166205,
+      "tokens_seen": 226885632
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000940641925777332,
+      "loss": 1.8044,
+      "theoretical_loss": 4.316090272102486,
+      "tokens_seen": 226951168
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009406218655967904,
+      "loss": 1.8453,
+      "theoretical_loss": 4.315938396185782,
+      "tokens_seen": 227016704
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009406018054162488,
+      "loss": 1.7808,
+      "theoretical_loss": 4.315786576379143,
+      "tokens_seen": 227082240
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009405817452357071,
+      "loss": 1.7636,
+      "theoretical_loss": 4.315634812645654,
+      "tokens_seen": 227147776
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009405616850551655,
+      "loss": 1.7784,
+      "theoretical_loss": 4.315483104948438,
+      "tokens_seen": 227213312
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009405416248746239,
+      "loss": 1.885,
+      "theoretical_loss": 4.315331453250648,
+      "tokens_seen": 227278848
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009405215646940822,
+      "loss": 1.7154,
+      "theoretical_loss": 4.315179857515477,
+      "tokens_seen": 227344384
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009405015045135407,
+      "loss": 1.7473,
+      "theoretical_loss": 4.315028317706149,
+      "tokens_seen": 227409920
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000940481444332999,
+      "loss": 1.8366,
+      "theoretical_loss": 4.314876833785926,
+      "tokens_seen": 227475456
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009404613841524574,
+      "loss": 1.8375,
+      "theoretical_loss": 4.314725405718099,
+      "tokens_seen": 227540992
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009404413239719157,
+      "loss": 1.7129,
+      "theoretical_loss": 4.314574033466001,
+      "tokens_seen": 227606528
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009404212637913741,
+      "loss": 1.789,
+      "theoretical_loss": 4.314422716992994,
+      "tokens_seen": 227672064
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009404012036108325,
+      "loss": 1.8073,
+      "theoretical_loss": 4.314271456262479,
+      "tokens_seen": 227737600
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009403811434302909,
+      "loss": 1.8551,
+      "theoretical_loss": 4.314120251237887,
+      "tokens_seen": 227803136
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009403610832497492,
+      "loss": 1.8417,
+      "theoretical_loss": 4.313969101882687,
+      "tokens_seen": 227868672
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009403410230692077,
+      "loss": 1.7539,
+      "theoretical_loss": 4.31381800816038,
+      "tokens_seen": 227934208
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000940320962888666,
+      "loss": 1.8444,
+      "theoretical_loss": 4.313666970034505,
+      "tokens_seen": 227999744
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009403009027081244,
+      "loss": 1.9733,
+      "theoretical_loss": 4.31351598746863,
+      "tokens_seen": 228065280
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009402808425275827,
+      "loss": 1.7642,
+      "theoretical_loss": 4.313365060426363,
+      "tokens_seen": 228130816
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009402607823470412,
+      "loss": 1.7757,
+      "theoretical_loss": 4.313214188871341,
+      "tokens_seen": 228196352
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009402407221664995,
+      "loss": 1.8742,
+      "theoretical_loss": 4.31306337276724,
+      "tokens_seen": 228261888
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009402206619859579,
+      "loss": 1.8931,
+      "theoretical_loss": 4.312912612077767,
+      "tokens_seen": 228327424
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009402006018054162,
+      "loss": 1.848,
+      "theoretical_loss": 4.312761906766664,
+      "tokens_seen": 228392960
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009401805416248746,
+      "loss": 1.8055,
+      "theoretical_loss": 4.312611256797708,
+      "tokens_seen": 228458496
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000940160481444333,
+      "loss": 1.8979,
+      "theoretical_loss": 4.3124606621347095,
+      "tokens_seen": 228524032
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009401404212637914,
+      "loss": 1.8403,
+      "theoretical_loss": 4.312310122741512,
+      "tokens_seen": 228589568
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009401203610832497,
+      "loss": 1.8505,
+      "theoretical_loss": 4.312159638581995,
+      "tokens_seen": 228655104
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009401003009027082,
+      "loss": 1.8639,
+      "theoretical_loss": 4.31200920962007,
+      "tokens_seen": 228720640
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009400802407221665,
+      "loss": 1.8759,
+      "theoretical_loss": 4.311858835819684,
+      "tokens_seen": 228786176
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009400601805416249,
+      "loss": 1.8338,
+      "theoretical_loss": 4.311708517144817,
+      "tokens_seen": 228851712
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009400401203610832,
+      "loss": 1.838,
+      "theoretical_loss": 4.3115582535594825,
+      "tokens_seen": 228917248
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009400200601805416,
+      "loss": 1.796,
+      "theoretical_loss": 4.311408045027728,
+      "tokens_seen": 228982784
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00094,
+      "loss": 1.9516,
+      "theoretical_loss": 4.3112578915136375,
+      "tokens_seen": 229048320
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009399799398194584,
+      "loss": 1.8333,
+      "theoretical_loss": 4.311107792981323,
+      "tokens_seen": 229113856
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009399598796389167,
+      "loss": 1.8905,
+      "theoretical_loss": 4.310957749394935,
+      "tokens_seen": 229179392
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009399398194583752,
+      "loss": 1.8143,
+      "theoretical_loss": 4.310807760718656,
+      "tokens_seen": 229244928
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009399197592778335,
+      "loss": 1.8554,
+      "theoretical_loss": 4.310657826916701,
+      "tokens_seen": 229310464
+    },
+    {
+      "epoch": 0.07,
+      "objective/train/advantage_avg": 0.4928604066371918,
+      "objective/train/docs_used": 90061,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.8705188035964966,
+      "objective/train/original_loss": 1.8705189228057861,
+      "objective/train/theoretical_loss": 4.310507947953321,
+      "objective/train/tokens_used": 249836000,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24335232377052307,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050523281097412,
+      "objective/train/weighted_lm_loss": 1.9649579524993896,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.310507947953321,
+      "tokens_seen": 229376000
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009398996990972919,
+      "loss": 1.8946,
+      "theoretical_loss": 4.310507947953321,
+      "tokens_seen": 229376000
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009398796389167504,
+      "loss": 1.7751,
+      "theoretical_loss": 4.310358123792798,
+      "tokens_seen": 229441536
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009398595787362086,
+      "loss": 1.9335,
+      "theoretical_loss": 4.31020835439945,
+      "tokens_seen": 229507072
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009398395185556671,
+      "loss": 1.9429,
+      "theoretical_loss": 4.310058639737626,
+      "tokens_seen": 229572608
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009398194583751253,
+      "loss": 1.8902,
+      "theoretical_loss": 4.309908979771709,
+      "tokens_seen": 229638144
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009397993981945838,
+      "loss": 1.7913,
+      "theoretical_loss": 4.3097593744661165,
+      "tokens_seen": 229703680
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009397793380140422,
+      "loss": 1.9064,
+      "theoretical_loss": 4.309609823785298,
+      "tokens_seen": 229769216
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009397592778335005,
+      "loss": 1.8685,
+      "theoretical_loss": 4.309460327693737,
+      "tokens_seen": 229834752
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009397392176529589,
+      "loss": 1.9508,
+      "theoretical_loss": 4.3093108861559495,
+      "tokens_seen": 229900288
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009397191574724174,
+      "loss": 1.839,
+      "theoretical_loss": 4.3091614991364855,
+      "tokens_seen": 229965824
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009396990972918756,
+      "loss": 1.8797,
+      "theoretical_loss": 4.309012166599929,
+      "tokens_seen": 230031360
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009396790371113341,
+      "loss": 1.8448,
+      "theoretical_loss": 4.308862888510893,
+      "tokens_seen": 230096896
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009396589769307923,
+      "loss": 1.7767,
+      "theoretical_loss": 4.308713664834029,
+      "tokens_seen": 230162432
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009396389167502508,
+      "loss": 1.6558,
+      "theoretical_loss": 4.308564495534017,
+      "tokens_seen": 230227968
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009396188565697092,
+      "loss": 1.8051,
+      "theoretical_loss": 4.308415380575573,
+      "tokens_seen": 230293504
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009395987963891675,
+      "loss": 1.71,
+      "theoretical_loss": 4.308266319923445,
+      "tokens_seen": 230359040
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009395787362086259,
+      "loss": 1.669,
+      "theoretical_loss": 4.308117313542413,
+      "tokens_seen": 230424576
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009395586760280843,
+      "loss": 1.7442,
+      "theoretical_loss": 4.3079683613972914,
+      "tokens_seen": 230490112
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009395386158475426,
+      "loss": 2.0136,
+      "theoretical_loss": 4.307819463452925,
+      "tokens_seen": 230555648
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009395185556670011,
+      "loss": 1.7937,
+      "theoretical_loss": 4.3076706196741945,
+      "tokens_seen": 230621184
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009394984954864594,
+      "loss": 1.8568,
+      "theoretical_loss": 4.30752183002601,
+      "tokens_seen": 230686720
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009394784353059178,
+      "loss": 1.7888,
+      "theoretical_loss": 4.307373094473318,
+      "tokens_seen": 230752256
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009394583751253761,
+      "loss": 1.758,
+      "theoretical_loss": 4.307224412981095,
+      "tokens_seen": 230817792
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009394383149448345,
+      "loss": 1.7212,
+      "theoretical_loss": 4.3070757855143516,
+      "tokens_seen": 230883328
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009394182547642929,
+      "loss": 1.8631,
+      "theoretical_loss": 4.3069272120381275,
+      "tokens_seen": 230948864
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009393981945837513,
+      "loss": 1.7568,
+      "theoretical_loss": 4.3067786925175,
+      "tokens_seen": 231014400
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009393781344032096,
+      "loss": 1.8251,
+      "theoretical_loss": 4.3066302269175765,
+      "tokens_seen": 231079936
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009393580742226681,
+      "loss": 1.9332,
+      "theoretical_loss": 4.306481815203496,
+      "tokens_seen": 231145472
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009393380140421264,
+      "loss": 1.8374,
+      "theoretical_loss": 4.30633345734043,
+      "tokens_seen": 231211008
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009393179538615848,
+      "loss": 1.8336,
+      "theoretical_loss": 4.306185153293584,
+      "tokens_seen": 231276544
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009392978936810431,
+      "loss": 1.6981,
+      "theoretical_loss": 4.306036903028196,
+      "tokens_seen": 231342080
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009392778335005016,
+      "loss": 1.7224,
+      "theoretical_loss": 4.305888706509534,
+      "tokens_seen": 231407616
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009392577733199599,
+      "loss": 1.9788,
+      "theoretical_loss": 4.3057405637029,
+      "tokens_seen": 231473152
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009392377131394183,
+      "loss": 1.7976,
+      "theoretical_loss": 4.305592474573627,
+      "tokens_seen": 231538688
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009392176529588766,
+      "loss": 1.9443,
+      "theoretical_loss": 4.305444439087082,
+      "tokens_seen": 231604224
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000939197592778335,
+      "loss": 1.7595,
+      "theoretical_loss": 4.305296457208662,
+      "tokens_seen": 231669760
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009391775325977934,
+      "loss": 1.8406,
+      "theoretical_loss": 4.305148528903798,
+      "tokens_seen": 231735296
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009391574724172518,
+      "loss": 1.9642,
+      "theoretical_loss": 4.3050006541379515,
+      "tokens_seen": 231800832
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009391374122367101,
+      "loss": 1.7075,
+      "theoretical_loss": 4.304852832876618,
+      "tokens_seen": 231866368
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009391173520561686,
+      "loss": 1.7245,
+      "theoretical_loss": 4.304705065085322,
+      "tokens_seen": 231931904
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009390972918756269,
+      "loss": 1.972,
+      "theoretical_loss": 4.304557350729623,
+      "tokens_seen": 231997440
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009390772316950853,
+      "loss": 1.9232,
+      "theoretical_loss": 4.304409689775111,
+      "tokens_seen": 232062976
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009390571715145436,
+      "loss": 1.8148,
+      "theoretical_loss": 4.304262082187407,
+      "tokens_seen": 232128512
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000939037111334002,
+      "loss": 1.6574,
+      "theoretical_loss": 4.304114527932166,
+      "tokens_seen": 232194048
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009390170511534604,
+      "loss": 1.8246,
+      "theoretical_loss": 4.303967026975072,
+      "tokens_seen": 232259584
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009389969909729188,
+      "loss": 1.7932,
+      "theoretical_loss": 4.303819579281844,
+      "tokens_seen": 232325120
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009389769307923771,
+      "loss": 1.8462,
+      "theoretical_loss": 4.30367218481823,
+      "tokens_seen": 232390656
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009389568706118356,
+      "loss": 1.8204,
+      "theoretical_loss": 4.30352484355001,
+      "tokens_seen": 232456192
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009389368104312938,
+      "loss": 1.9215,
+      "theoretical_loss": 4.303377555442998,
+      "tokens_seen": 232521728
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009389167502507523,
+      "loss": 1.7366,
+      "theoretical_loss": 4.3032303204630376,
+      "tokens_seen": 232587264
+    },
+    {
+      "epoch": 0.07,
+      "objective/train/advantage_avg": 0.49400627613067627,
+      "objective/train/docs_used": 91258,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.6414120197296143,
+      "objective/train/original_loss": 1.6414121389389038,
+      "objective/train/theoretical_loss": 4.303083138576003,
+      "objective/train/tokens_used": 253112800,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24473948776721954,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506447553634644,
+      "objective/train/weighted_lm_loss": 1.7243413925170898,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9048373699188232,
+      "theoretical_loss": 4.303083138576003,
+      "tokens_seen": 232652800
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009388966900702107,
+      "loss": 1.8242,
+      "theoretical_loss": 4.303083138576003,
+      "tokens_seen": 232652800
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000938876629889669,
+      "loss": 1.7083,
+      "theoretical_loss": 4.302936009747803,
+      "tokens_seen": 232718336
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009388565697091274,
+      "loss": 1.7864,
+      "theoretical_loss": 4.302788933944375,
+      "tokens_seen": 232783872
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009388365095285857,
+      "loss": 1.8209,
+      "theoretical_loss": 4.302641911131689,
+      "tokens_seen": 232849408
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009388164493480441,
+      "loss": 1.8624,
+      "theoretical_loss": 4.3024949412757465,
+      "tokens_seen": 232914944
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009387963891675026,
+      "loss": 1.7925,
+      "theoretical_loss": 4.302348024342581,
+      "tokens_seen": 232980480
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009387763289869608,
+      "loss": 1.8408,
+      "theoretical_loss": 4.302201160298255,
+      "tokens_seen": 233046016
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009387562688064193,
+      "loss": 1.8639,
+      "theoretical_loss": 4.302054349108865,
+      "tokens_seen": 233111552
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009387362086258777,
+      "loss": 1.7454,
+      "theoretical_loss": 4.301907590740538,
+      "tokens_seen": 233177088
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000938716148445336,
+      "loss": 1.7967,
+      "theoretical_loss": 4.301760885159431,
+      "tokens_seen": 233242624
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009386960882647944,
+      "loss": 1.8907,
+      "theoretical_loss": 4.301614232331733,
+      "tokens_seen": 233308160
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009386760280842527,
+      "loss": 1.817,
+      "theoretical_loss": 4.301467632223664,
+      "tokens_seen": 233373696
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009386559679037111,
+      "loss": 1.9417,
+      "theoretical_loss": 4.301321084801478,
+      "tokens_seen": 233439232
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009386359077231696,
+      "loss": 1.744,
+      "theoretical_loss": 4.301174590031453,
+      "tokens_seen": 233504768
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009386158475426278,
+      "loss": 1.7738,
+      "theoretical_loss": 4.301028147879904,
+      "tokens_seen": 233570304
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009385957873620863,
+      "loss": 1.6949,
+      "theoretical_loss": 4.300881758313178,
+      "tokens_seen": 233635840
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009385757271815446,
+      "loss": 1.7507,
+      "theoretical_loss": 4.300735421297647,
+      "tokens_seen": 233701376
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000938555667001003,
+      "loss": 1.7811,
+      "theoretical_loss": 4.300589136799719,
+      "tokens_seen": 233766912
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009385356068204614,
+      "loss": 1.8286,
+      "theoretical_loss": 4.300442904785831,
+      "tokens_seen": 233832448
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009385155466399198,
+      "loss": 1.7496,
+      "theoretical_loss": 4.300296725222449,
+      "tokens_seen": 233897984
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009384954864593781,
+      "loss": 1.8731,
+      "theoretical_loss": 4.300150598076074,
+      "tokens_seen": 233963520
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009384754262788365,
+      "loss": 1.8883,
+      "theoretical_loss": 4.300004523313237,
+      "tokens_seen": 234029056
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009384553660982948,
+      "loss": 1.8071,
+      "theoretical_loss": 4.299858500900495,
+      "tokens_seen": 234094592
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009384353059177533,
+      "loss": 1.9614,
+      "theoretical_loss": 4.299712530804441,
+      "tokens_seen": 234160128
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009384152457372117,
+      "loss": 1.6487,
+      "theoretical_loss": 4.299566612991695,
+      "tokens_seen": 234225664
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00093839518555667,
+      "loss": 1.8937,
+      "theoretical_loss": 4.299420747428911,
+      "tokens_seen": 234291200
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009383751253761285,
+      "loss": 1.867,
+      "theoretical_loss": 4.2992749340827725,
+      "tokens_seen": 234356736
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009383550651955868,
+      "loss": 1.828,
+      "theoretical_loss": 4.299129172919991,
+      "tokens_seen": 234422272
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009383350050150452,
+      "loss": 1.8159,
+      "theoretical_loss": 4.2989834639073115,
+      "tokens_seen": 234487808
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009383149448345035,
+      "loss": 1.8431,
+      "theoretical_loss": 4.298837807011508,
+      "tokens_seen": 234553344
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000938294884653962,
+      "loss": 1.955,
+      "theoretical_loss": 4.298692202199386,
+      "tokens_seen": 234618880
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009382748244734203,
+      "loss": 1.9081,
+      "theoretical_loss": 4.29854664943778,
+      "tokens_seen": 234684416
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009382547642928787,
+      "loss": 1.7294,
+      "theoretical_loss": 4.298401148693557,
+      "tokens_seen": 234749952
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000938234704112337,
+      "loss": 1.8228,
+      "theoretical_loss": 4.298255699933612,
+      "tokens_seen": 234815488
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009382146439317954,
+      "loss": 1.8377,
+      "theoretical_loss": 4.298110303124871,
+      "tokens_seen": 234881024
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009381945837512538,
+      "loss": 1.9015,
+      "theoretical_loss": 4.297964958234292,
+      "tokens_seen": 234946560
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009381745235707122,
+      "loss": 1.8647,
+      "theoretical_loss": 4.297819665228861,
+      "tokens_seen": 235012096
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009381544633901705,
+      "loss": 1.8087,
+      "theoretical_loss": 4.297674424075594,
+      "tokens_seen": 235077632
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000938134403209629,
+      "loss": 1.812,
+      "theoretical_loss": 4.29752923474154,
+      "tokens_seen": 235143168
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009381143430290873,
+      "loss": 1.823,
+      "theoretical_loss": 4.297384097193775,
+      "tokens_seen": 235208704
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009380942828485457,
+      "loss": 1.8137,
+      "theoretical_loss": 4.297239011399407,
+      "tokens_seen": 235274240
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000938074222668004,
+      "loss": 1.8754,
+      "theoretical_loss": 4.297093977325573,
+      "tokens_seen": 235339776
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009380541624874624,
+      "loss": 1.7469,
+      "theoretical_loss": 4.29694899493944,
+      "tokens_seen": 235405312
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009380341023069208,
+      "loss": 1.7643,
+      "theoretical_loss": 4.296804064208207,
+      "tokens_seen": 235470848
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009380140421263792,
+      "loss": 1.8098,
+      "theoretical_loss": 4.296659185099099,
+      "tokens_seen": 235536384
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009379939819458375,
+      "loss": 1.8432,
+      "theoretical_loss": 4.296514357579375,
+      "tokens_seen": 235601920
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000937973921765296,
+      "loss": 1.9429,
+      "theoretical_loss": 4.2963695816163225,
+      "tokens_seen": 235667456
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009379538615847542,
+      "loss": 1.7737,
+      "theoretical_loss": 4.296224857177257,
+      "tokens_seen": 235732992
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009379338014042127,
+      "loss": 1.8518,
+      "theoretical_loss": 4.296080184229526,
+      "tokens_seen": 235798528
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009379137412236711,
+      "loss": 1.7253,
+      "theoretical_loss": 4.295935562740505,
+      "tokens_seen": 235864064
+    },
+    {
+      "epoch": 0.07,
+      "objective/train/advantage_avg": 0.4898257851600647,
+      "objective/train/docs_used": 92471,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.8431349992752075,
+      "objective/train/original_loss": 1.843135118484497,
+      "objective/train/theoretical_loss": 4.295790992677603,
+      "objective/train/tokens_used": 256389600,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24101541936397552,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0502077341079712,
+      "objective/train/weighted_lm_loss": 1.9356428384780884,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.295790992677603,
+      "tokens_seen": 235929600
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009378936810431294,
+      "loss": 1.856,
+      "theoretical_loss": 4.295790992677603,
+      "tokens_seen": 235929600
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009378736208625878,
+      "loss": 1.8341,
+      "theoretical_loss": 4.295646474008252,
+      "tokens_seen": 235995136
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009378535606820461,
+      "loss": 1.9499,
+      "theoretical_loss": 4.29550200669992,
+      "tokens_seen": 236060672
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009378335005015045,
+      "loss": 1.7604,
+      "theoretical_loss": 4.295357590720101,
+      "tokens_seen": 236126208
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000937813440320963,
+      "loss": 1.7573,
+      "theoretical_loss": 4.2952132260363225,
+      "tokens_seen": 236191744
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009377933801404212,
+      "loss": 1.8687,
+      "theoretical_loss": 4.295068912616135,
+      "tokens_seen": 236257280
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009377733199598797,
+      "loss": 1.7367,
+      "theoretical_loss": 4.294924650427125,
+      "tokens_seen": 236322816
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009377532597793381,
+      "loss": 1.7755,
+      "theoretical_loss": 4.294780439436904,
+      "tokens_seen": 236388352
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009377331995987964,
+      "loss": 1.7929,
+      "theoretical_loss": 4.294636279613117,
+      "tokens_seen": 236453888
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009377131394182548,
+      "loss": 1.8914,
+      "theoretical_loss": 4.2944921709234345,
+      "tokens_seen": 236519424
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009376930792377131,
+      "loss": 1.897,
+      "theoretical_loss": 4.294348113335559,
+      "tokens_seen": 236584960
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009376730190571715,
+      "loss": 2.0073,
+      "theoretical_loss": 4.29420410681722,
+      "tokens_seen": 236650496
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00093765295887663,
+      "loss": 1.7469,
+      "theoretical_loss": 4.294060151336178,
+      "tokens_seen": 236716032
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009376328986960882,
+      "loss": 1.8601,
+      "theoretical_loss": 4.293916246860224,
+      "tokens_seen": 236781568
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009376128385155467,
+      "loss": 1.8224,
+      "theoretical_loss": 4.293772393357176,
+      "tokens_seen": 236847104
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000937592778335005,
+      "loss": 1.9164,
+      "theoretical_loss": 4.2936285907948815,
+      "tokens_seen": 236912640
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009375727181544634,
+      "loss": 1.8051,
+      "theoretical_loss": 4.293484839141217,
+      "tokens_seen": 236978176
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009375526579739218,
+      "loss": 1.7483,
+      "theoretical_loss": 4.293341138364091,
+      "tokens_seen": 237043712
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009375325977933802,
+      "loss": 1.7834,
+      "theoretical_loss": 4.293197488431435,
+      "tokens_seen": 237109248
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009375125376128385,
+      "loss": 1.8088,
+      "theoretical_loss": 4.293053889311217,
+      "tokens_seen": 237174784
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009374924774322969,
+      "loss": 1.8689,
+      "theoretical_loss": 4.29291034097143,
+      "tokens_seen": 237240320
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009374724172517552,
+      "loss": 1.7728,
+      "theoretical_loss": 4.292766843380094,
+      "tokens_seen": 237305856
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009374523570712137,
+      "loss": 1.7375,
+      "theoretical_loss": 4.292623396505262,
+      "tokens_seen": 237371392
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000937432296890672,
+      "loss": 1.7408,
+      "theoretical_loss": 4.292480000315014,
+      "tokens_seen": 237436928
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009374122367101304,
+      "loss": 1.7242,
+      "theoretical_loss": 4.2923366547774595,
+      "tokens_seen": 237502464
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009373921765295888,
+      "loss": 1.7053,
+      "theoretical_loss": 4.292193359860737,
+      "tokens_seen": 237568000
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009373721163490472,
+      "loss": 1.7752,
+      "theoretical_loss": 4.292050115533012,
+      "tokens_seen": 237633536
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009373520561685055,
+      "loss": 1.7755,
+      "theoretical_loss": 4.291906921762481,
+      "tokens_seen": 237699072
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009373319959879639,
+      "loss": 1.9598,
+      "theoretical_loss": 4.2917637785173675,
+      "tokens_seen": 237764608
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009373119358074222,
+      "loss": 1.7152,
+      "theoretical_loss": 4.291620685765926,
+      "tokens_seen": 237830144
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009372918756268807,
+      "loss": 1.884,
+      "theoretical_loss": 4.291477643476438,
+      "tokens_seen": 237895680
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000937271815446339,
+      "loss": 1.8008,
+      "theoretical_loss": 4.291334651617213,
+      "tokens_seen": 237961216
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009372517552657974,
+      "loss": 1.9097,
+      "theoretical_loss": 4.291191710156591,
+      "tokens_seen": 238026752
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009372316950852557,
+      "loss": 1.8682,
+      "theoretical_loss": 4.291048819062939,
+      "tokens_seen": 238092288
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009372116349047142,
+      "loss": 1.7768,
+      "theoretical_loss": 4.290905978304655,
+      "tokens_seen": 238157824
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009371915747241725,
+      "loss": 1.7492,
+      "theoretical_loss": 4.290763187850161,
+      "tokens_seen": 238223360
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009371715145436309,
+      "loss": 1.8498,
+      "theoretical_loss": 4.290620447667912,
+      "tokens_seen": 238288896
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009371514543630893,
+      "loss": 1.7209,
+      "theoretical_loss": 4.29047775772639,
+      "tokens_seen": 238354432
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009371313941825476,
+      "loss": 1.7331,
+      "theoretical_loss": 4.290335117994104,
+      "tokens_seen": 238419968
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000937111334002006,
+      "loss": 1.738,
+      "theoretical_loss": 4.290192528439594,
+      "tokens_seen": 238485504
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009370912738214644,
+      "loss": 1.9051,
+      "theoretical_loss": 4.290049989031424,
+      "tokens_seen": 238551040
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009370712136409227,
+      "loss": 1.9187,
+      "theoretical_loss": 4.289907499738192,
+      "tokens_seen": 238616576
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009370511534603812,
+      "loss": 1.8264,
+      "theoretical_loss": 4.289765060528521,
+      "tokens_seen": 238682112
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009370310932798395,
+      "loss": 1.9482,
+      "theoretical_loss": 4.2896226713710615,
+      "tokens_seen": 238747648
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009370110330992979,
+      "loss": 1.8173,
+      "theoretical_loss": 4.289480332234493,
+      "tokens_seen": 238813184
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009369909729187563,
+      "loss": 1.7753,
+      "theoretical_loss": 4.289338043087527,
+      "tokens_seen": 238878720
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009369709127382146,
+      "loss": 1.7691,
+      "theoretical_loss": 4.289195803898897,
+      "tokens_seen": 238944256
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000936950852557673,
+      "loss": 1.9161,
+      "theoretical_loss": 4.289053614637368,
+      "tokens_seen": 239009792
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009369307923771315,
+      "loss": 1.8554,
+      "theoretical_loss": 4.288911475271731,
+      "tokens_seen": 239075328
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009369107321965898,
+      "loss": 1.8071,
+      "theoretical_loss": 4.288769385770809,
+      "tokens_seen": 239140864
+    },
+    {
+      "epoch": 0.07,
+      "objective/train/advantage_avg": 0.49386152625083923,
+      "objective/train/docs_used": 93570,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.8087902069091797,
+      "objective/train/original_loss": 1.8087904453277588,
+      "objective/train/theoretical_loss": 4.28862734610345,
+      "objective/train/tokens_used": 259666400,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24422965943813324,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506278276443481,
+      "objective/train/weighted_lm_loss": 1.9002867937088013,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.28862734610345,
+      "tokens_seen": 239206400
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009368906720160482,
+      "loss": 1.8793,
+      "theoretical_loss": 4.28862734610345,
+      "tokens_seen": 239206400
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009368706118355065,
+      "loss": 1.8012,
+      "theoretical_loss": 4.288485356238529,
+      "tokens_seen": 239271936
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009368505516549649,
+      "loss": 1.7934,
+      "theoretical_loss": 4.288343416144952,
+      "tokens_seen": 239337472
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009368304914744234,
+      "loss": 1.9502,
+      "theoretical_loss": 4.288201525791651,
+      "tokens_seen": 239403008
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009368104312938816,
+      "loss": 1.8676,
+      "theoretical_loss": 4.288059685147586,
+      "tokens_seen": 239468544
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009367903711133401,
+      "loss": 1.8661,
+      "theoretical_loss": 4.287917894181746,
+      "tokens_seen": 239534080
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009367703109327985,
+      "loss": 1.7173,
+      "theoretical_loss": 4.287776152863146,
+      "tokens_seen": 239599616
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009367502507522568,
+      "loss": 1.8081,
+      "theoretical_loss": 4.28763446116083,
+      "tokens_seen": 239665152
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009367301905717152,
+      "loss": 1.8705,
+      "theoretical_loss": 4.28749281904387,
+      "tokens_seen": 239730688
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009367101303911735,
+      "loss": 1.8234,
+      "theoretical_loss": 4.287351226481366,
+      "tokens_seen": 239796224
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009366900702106319,
+      "loss": 1.8895,
+      "theoretical_loss": 4.287209683442444,
+      "tokens_seen": 239861760
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009366700100300904,
+      "loss": 1.8936,
+      "theoretical_loss": 4.2870681898962575,
+      "tokens_seen": 239927296
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009366499498495486,
+      "loss": 1.8068,
+      "theoretical_loss": 4.286926745811991,
+      "tokens_seen": 239992832
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009366298896690071,
+      "loss": 1.7208,
+      "theoretical_loss": 4.286785351158852,
+      "tokens_seen": 240058368
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009366098294884654,
+      "loss": 1.8455,
+      "theoretical_loss": 4.286644005906081,
+      "tokens_seen": 240123904
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009365897693079238,
+      "loss": 1.8656,
+      "theoretical_loss": 4.2865027100229405,
+      "tokens_seen": 240189440
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009365697091273822,
+      "loss": 1.9201,
+      "theoretical_loss": 4.286361463478723,
+      "tokens_seen": 240254976
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009365496489468406,
+      "loss": 1.8351,
+      "theoretical_loss": 4.28622026624275,
+      "tokens_seen": 240320512
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009365295887662989,
+      "loss": 1.7223,
+      "theoretical_loss": 4.286079118284368,
+      "tokens_seen": 240386048
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009365095285857573,
+      "loss": 1.7905,
+      "theoretical_loss": 4.285938019572951,
+      "tokens_seen": 240451584
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009364894684052156,
+      "loss": 1.922,
+      "theoretical_loss": 4.285796970077903,
+      "tokens_seen": 240517120
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009364694082246741,
+      "loss": 1.8986,
+      "theoretical_loss": 4.285655969768651,
+      "tokens_seen": 240582656
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009364493480441324,
+      "loss": 1.9152,
+      "theoretical_loss": 4.285515018614655,
+      "tokens_seen": 240648192
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009364292878635908,
+      "loss": 1.7653,
+      "theoretical_loss": 4.2853741165853965,
+      "tokens_seen": 240713728
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009364092276830492,
+      "loss": 1.6739,
+      "theoretical_loss": 4.285233263650387,
+      "tokens_seen": 240779264
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009363891675025076,
+      "loss": 1.7319,
+      "theoretical_loss": 4.285092459779166,
+      "tokens_seen": 240844800
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009363691073219659,
+      "loss": 1.7374,
+      "theoretical_loss": 4.2849517049412995,
+      "tokens_seen": 240910336
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009363490471414243,
+      "loss": 1.7947,
+      "theoretical_loss": 4.28481099910638,
+      "tokens_seen": 240975872
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009363289869608826,
+      "loss": 1.7014,
+      "theoretical_loss": 4.284670342244027,
+      "tokens_seen": 241041408
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009363089267803411,
+      "loss": 1.8506,
+      "theoretical_loss": 4.284529734323887,
+      "tokens_seen": 241106944
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009362888665997994,
+      "loss": 1.7271,
+      "theoretical_loss": 4.284389175315636,
+      "tokens_seen": 241172480
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009362688064192578,
+      "loss": 1.7538,
+      "theoretical_loss": 4.284248665188974,
+      "tokens_seen": 241238016
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009362487462387161,
+      "loss": 1.7851,
+      "theoretical_loss": 4.284108203913629,
+      "tokens_seen": 241303552
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009362286860581746,
+      "loss": 1.8025,
+      "theoretical_loss": 4.283967791459356,
+      "tokens_seen": 241369088
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009362086258776329,
+      "loss": 1.8707,
+      "theoretical_loss": 4.283827427795939,
+      "tokens_seen": 241434624
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009361885656970913,
+      "loss": 1.8208,
+      "theoretical_loss": 4.283687112893184,
+      "tokens_seen": 241500160
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009361685055165497,
+      "loss": 1.7022,
+      "theoretical_loss": 4.283546846720929,
+      "tokens_seen": 241565696
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000936148445336008,
+      "loss": 1.798,
+      "theoretical_loss": 4.283406629249036,
+      "tokens_seen": 241631232
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009361283851554664,
+      "loss": 1.7899,
+      "theoretical_loss": 4.283266460447394,
+      "tokens_seen": 241696768
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009361083249749248,
+      "loss": 1.7569,
+      "theoretical_loss": 4.28312634028592,
+      "tokens_seen": 241762304
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009360882647943831,
+      "loss": 1.7866,
+      "theoretical_loss": 4.282986268734556,
+      "tokens_seen": 241827840
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009360682046138416,
+      "loss": 1.6588,
+      "theoretical_loss": 4.282846245763273,
+      "tokens_seen": 241893376
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009360481444332999,
+      "loss": 1.7699,
+      "theoretical_loss": 4.282706271342066,
+      "tokens_seen": 241958912
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009360280842527583,
+      "loss": 1.8742,
+      "theoretical_loss": 4.2825663454409595,
+      "tokens_seen": 242024448
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009360080240722167,
+      "loss": 1.8847,
+      "theoretical_loss": 4.2824264680300015,
+      "tokens_seen": 242089984
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000935987963891675,
+      "loss": 1.9179,
+      "theoretical_loss": 4.2822866390792695,
+      "tokens_seen": 242155520
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009359679037111334,
+      "loss": 1.8009,
+      "theoretical_loss": 4.282146858558866,
+      "tokens_seen": 242221056
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009359478435305919,
+      "loss": 1.7736,
+      "theoretical_loss": 4.28200712643892,
+      "tokens_seen": 242286592
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009359277833500501,
+      "loss": 1.627,
+      "theoretical_loss": 4.281867442689587,
+      "tokens_seen": 242352128
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009359077231695086,
+      "loss": 1.87,
+      "theoretical_loss": 4.281727807281051,
+      "tokens_seen": 242417664
+    },
+    {
+      "epoch": 0.07,
+      "objective/train/advantage_avg": 0.49168694019317627,
+      "objective/train/docs_used": 94878,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.7456352710723877,
+      "objective/train/original_loss": 1.7456352710723877,
+      "objective/train/theoretical_loss": 4.281588220183519,
+      "objective/train/tokens_used": 262943200,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24250559508800507,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504014492034912,
+      "objective/train/weighted_lm_loss": 1.8333992958068848,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.281588220183519,
+      "tokens_seen": 242483200
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009358876629889668,
+      "loss": 1.7932,
+      "theoretical_loss": 4.281588220183519,
+      "tokens_seen": 242483200
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009358676028084253,
+      "loss": 1.8316,
+      "theoretical_loss": 4.281448681367226,
+      "tokens_seen": 242548736
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009358475426278837,
+      "loss": 1.7968,
+      "theoretical_loss": 4.2813091908024345,
+      "tokens_seen": 242614272
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000935827482447342,
+      "loss": 1.7241,
+      "theoretical_loss": 4.281169748459432,
+      "tokens_seen": 242679808
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009358074222668004,
+      "loss": 1.8417,
+      "theoretical_loss": 4.281030354308533,
+      "tokens_seen": 242745344
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009357873620862589,
+      "loss": 1.811,
+      "theoretical_loss": 4.2808910083200775,
+      "tokens_seen": 242810880
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009357673019057171,
+      "loss": 1.7931,
+      "theoretical_loss": 4.280751710464432,
+      "tokens_seen": 242876416
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009357472417251756,
+      "loss": 1.7608,
+      "theoretical_loss": 4.280612460711989,
+      "tokens_seen": 242941952
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009357271815446338,
+      "loss": 1.8087,
+      "theoretical_loss": 4.280473259033169,
+      "tokens_seen": 243007488
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009357071213640923,
+      "loss": 1.8108,
+      "theoretical_loss": 4.280334105398417,
+      "tokens_seen": 243073024
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009356870611835507,
+      "loss": 1.7234,
+      "theoretical_loss": 4.280194999778203,
+      "tokens_seen": 243138560
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000935667001003009,
+      "loss": 1.7381,
+      "theoretical_loss": 4.280055942143027,
+      "tokens_seen": 243204096
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009356469408224674,
+      "loss": 1.7984,
+      "theoretical_loss": 4.27991693246341,
+      "tokens_seen": 243269632
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009356268806419258,
+      "loss": 1.8265,
+      "theoretical_loss": 4.279777970709902,
+      "tokens_seen": 243335168
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009356068204613841,
+      "loss": 1.8135,
+      "theoretical_loss": 4.27963905685308,
+      "tokens_seen": 243400704
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009355867602808426,
+      "loss": 1.7835,
+      "theoretical_loss": 4.279500190863544,
+      "tokens_seen": 243466240
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009355667001003008,
+      "loss": 1.7395,
+      "theoretical_loss": 4.279361372711923,
+      "tokens_seen": 243531776
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009355466399197593,
+      "loss": 1.899,
+      "theoretical_loss": 4.27922260236887,
+      "tokens_seen": 243597312
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009355265797392176,
+      "loss": 1.8655,
+      "theoretical_loss": 4.279083879805064,
+      "tokens_seen": 243662848
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000935506519558676,
+      "loss": 1.8957,
+      "theoretical_loss": 4.278945204991212,
+      "tokens_seen": 243728384
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009354864593781344,
+      "loss": 1.8112,
+      "theoretical_loss": 4.278806577898042,
+      "tokens_seen": 243793920
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009354663991975928,
+      "loss": 1.7751,
+      "theoretical_loss": 4.278667998496313,
+      "tokens_seen": 243859456
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009354463390170511,
+      "loss": 1.8183,
+      "theoretical_loss": 4.278529466756806,
+      "tokens_seen": 243924992
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009354262788365096,
+      "loss": 1.7717,
+      "theoretical_loss": 4.278390982650333,
+      "tokens_seen": 243990528
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000935406218655968,
+      "loss": 1.7982,
+      "theoretical_loss": 4.278252546147724,
+      "tokens_seen": 244056064
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009353861584754263,
+      "loss": 1.9031,
+      "theoretical_loss": 4.278114157219839,
+      "tokens_seen": 244121600
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009353660982948847,
+      "loss": 1.7776,
+      "theoretical_loss": 4.277975815837567,
+      "tokens_seen": 244187136
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000935346038114343,
+      "loss": 1.8307,
+      "theoretical_loss": 4.277837521971816,
+      "tokens_seen": 244252672
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009353259779338015,
+      "loss": 1.7035,
+      "theoretical_loss": 4.277699275593523,
+      "tokens_seen": 244318208
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009353059177532598,
+      "loss": 1.6069,
+      "theoretical_loss": 4.277561076673652,
+      "tokens_seen": 244383744
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009352858575727182,
+      "loss": 1.7675,
+      "theoretical_loss": 4.277422925183188,
+      "tokens_seen": 244449280
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009352657973921765,
+      "loss": 1.7285,
+      "theoretical_loss": 4.277284821093147,
+      "tokens_seen": 244514816
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000935245737211635,
+      "loss": 1.7922,
+      "theoretical_loss": 4.277146764374566,
+      "tokens_seen": 244580352
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009352256770310933,
+      "loss": 1.7639,
+      "theoretical_loss": 4.27700875499851,
+      "tokens_seen": 244645888
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009352056168505517,
+      "loss": 1.8802,
+      "theoretical_loss": 4.2768707929360685,
+      "tokens_seen": 244711424
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00093518555667001,
+      "loss": 1.7328,
+      "theoretical_loss": 4.276732878158356,
+      "tokens_seen": 244776960
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009351654964894684,
+      "loss": 1.8951,
+      "theoretical_loss": 4.276595010636514,
+      "tokens_seen": 244842496
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009351454363089268,
+      "loss": 1.7935,
+      "theoretical_loss": 4.276457190341707,
+      "tokens_seen": 244908032
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009351253761283852,
+      "loss": 1.8199,
+      "theoretical_loss": 4.2763194172451255,
+      "tokens_seen": 244973568
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009351053159478435,
+      "loss": 1.7277,
+      "theoretical_loss": 4.276181691317987,
+      "tokens_seen": 245039104
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000935085255767302,
+      "loss": 1.8886,
+      "theoretical_loss": 4.276044012531534,
+      "tokens_seen": 245104640
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009350651955867603,
+      "loss": 2.0016,
+      "theoretical_loss": 4.275906380857031,
+      "tokens_seen": 245170176
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009350451354062187,
+      "loss": 1.7677,
+      "theoretical_loss": 4.275768796265771,
+      "tokens_seen": 245235712
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009350250752256771,
+      "loss": 1.7328,
+      "theoretical_loss": 4.275631258729071,
+      "tokens_seen": 245301248
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009350050150451354,
+      "loss": 1.83,
+      "theoretical_loss": 4.275493768218274,
+      "tokens_seen": 245366784
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009349849548645938,
+      "loss": 1.7851,
+      "theoretical_loss": 4.275356324704745,
+      "tokens_seen": 245432320
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009349648946840523,
+      "loss": 1.8365,
+      "theoretical_loss": 4.27521892815988,
+      "tokens_seen": 245497856
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009349448345035105,
+      "loss": 1.6734,
+      "theoretical_loss": 4.275081578555094,
+      "tokens_seen": 245563392
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000934924774322969,
+      "loss": 1.7928,
+      "theoretical_loss": 4.274944275861828,
+      "tokens_seen": 245628928
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009349047141424272,
+      "loss": 1.7817,
+      "theoretical_loss": 4.274807020051552,
+      "tokens_seen": 245694464
+    },
+    {
+      "epoch": 0.07,
+      "objective/train/advantage_avg": 0.4900068938732147,
+      "objective/train/docs_used": 96014,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.7985907793045044,
+      "objective/train/original_loss": 1.798590898513794,
+      "objective/train/theoretical_loss": 4.274669811095759,
+      "objective/train/tokens_used": 266220000,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24235199391841888,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0502325296401978,
+      "objective/train/weighted_lm_loss": 1.888869047164917,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.274669811095759,
+      "tokens_seen": 245760000
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009348846539618857,
+      "loss": 1.7632,
+      "theoretical_loss": 4.274669811095759,
+      "tokens_seen": 245760000
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009348645937813441,
+      "loss": 1.7704,
+      "theoretical_loss": 4.274532648965964,
+      "tokens_seen": 245825536
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009348445336008024,
+      "loss": 1.7771,
+      "theoretical_loss": 4.274395533633712,
+      "tokens_seen": 245891072
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009348244734202608,
+      "loss": 1.7508,
+      "theoretical_loss": 4.274258465070567,
+      "tokens_seen": 245956608
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009348044132397193,
+      "loss": 1.8499,
+      "theoretical_loss": 4.274121443248122,
+      "tokens_seen": 246022144
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009347843530591775,
+      "loss": 1.7992,
+      "theoretical_loss": 4.273984468137995,
+      "tokens_seen": 246087680
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000934764292878636,
+      "loss": 1.5857,
+      "theoretical_loss": 4.273847539711825,
+      "tokens_seen": 246153216
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009347442326980942,
+      "loss": 1.8579,
+      "theoretical_loss": 4.273710657941282,
+      "tokens_seen": 246218752
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009347241725175527,
+      "loss": 1.8407,
+      "theoretical_loss": 4.273573822798054,
+      "tokens_seen": 246284288
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009347041123370111,
+      "loss": 1.839,
+      "theoretical_loss": 4.273437034253859,
+      "tokens_seen": 246349824
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009346840521564694,
+      "loss": 1.8247,
+      "theoretical_loss": 4.273300292280435,
+      "tokens_seen": 246415360
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009346639919759278,
+      "loss": 1.8677,
+      "theoretical_loss": 4.273163596849549,
+      "tokens_seen": 246480896
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009346439317953862,
+      "loss": 1.8076,
+      "theoretical_loss": 4.27302694793299,
+      "tokens_seen": 246546432
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009346238716148445,
+      "loss": 1.748,
+      "theoretical_loss": 4.272890345502573,
+      "tokens_seen": 246611968
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000934603811434303,
+      "loss": 1.8006,
+      "theoretical_loss": 4.272753789530134,
+      "tokens_seen": 246677504
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009345837512537612,
+      "loss": 1.6839,
+      "theoretical_loss": 4.272617279987541,
+      "tokens_seen": 246743040
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009345636910732197,
+      "loss": 1.6944,
+      "theoretical_loss": 4.272480816846678,
+      "tokens_seen": 246808576
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000934543630892678,
+      "loss": 1.7108,
+      "theoretical_loss": 4.27234440007946,
+      "tokens_seen": 246874112
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009345235707121364,
+      "loss": 1.8806,
+      "theoretical_loss": 4.272208029657822,
+      "tokens_seen": 246939648
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009345035105315948,
+      "loss": 1.8091,
+      "theoretical_loss": 4.272071705553726,
+      "tokens_seen": 247005184
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009344834503510532,
+      "loss": 1.7406,
+      "theoretical_loss": 4.271935427739158,
+      "tokens_seen": 247070720
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009344633901705115,
+      "loss": 1.7875,
+      "theoretical_loss": 4.2717991961861275,
+      "tokens_seen": 247136256
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00093444332998997,
+      "loss": 1.7592,
+      "theoretical_loss": 4.271663010866669,
+      "tokens_seen": 247201792
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009344232698094283,
+      "loss": 1.9044,
+      "theoretical_loss": 4.27152687175284,
+      "tokens_seen": 247267328
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009344032096288867,
+      "loss": 1.8186,
+      "theoretical_loss": 4.271390778816727,
+      "tokens_seen": 247332864
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000934383149448345,
+      "loss": 1.8104,
+      "theoretical_loss": 4.2712547320304335,
+      "tokens_seen": 247398400
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0009343630892678034,
+      "loss": 1.7732,
+      "theoretical_loss": 4.2711187313660925,
+      "tokens_seen": 247463936
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009343430290872618,
+      "loss": 1.7146,
+      "theoretical_loss": 4.27098277679586,
+      "tokens_seen": 247529472
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009343229689067202,
+      "loss": 1.8197,
+      "theoretical_loss": 4.270846868291914,
+      "tokens_seen": 247595008
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009343029087261785,
+      "loss": 1.8096,
+      "theoretical_loss": 4.270711005826461,
+      "tokens_seen": 247660544
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009342828485456369,
+      "loss": 1.8219,
+      "theoretical_loss": 4.270575189371727,
+      "tokens_seen": 247726080
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009342627883650953,
+      "loss": 1.8408,
+      "theoretical_loss": 4.270439418899966,
+      "tokens_seen": 247791616
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009342427281845537,
+      "loss": 1.772,
+      "theoretical_loss": 4.270303694383453,
+      "tokens_seen": 247857152
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000934222668004012,
+      "loss": 1.7264,
+      "theoretical_loss": 4.270168015794489,
+      "tokens_seen": 247922688
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009342026078234705,
+      "loss": 1.8814,
+      "theoretical_loss": 4.270032383105398,
+      "tokens_seen": 247988224
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009341825476429287,
+      "loss": 1.7944,
+      "theoretical_loss": 4.269896796288528,
+      "tokens_seen": 248053760
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009341624874623872,
+      "loss": 1.7638,
+      "theoretical_loss": 4.269761255316252,
+      "tokens_seen": 248119296
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009341424272818455,
+      "loss": 1.7488,
+      "theoretical_loss": 4.269625760160965,
+      "tokens_seen": 248184832
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009341223671013039,
+      "loss": 1.7874,
+      "theoretical_loss": 4.269490310795089,
+      "tokens_seen": 248250368
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009341023069207623,
+      "loss": 1.899,
+      "theoretical_loss": 4.269354907191065,
+      "tokens_seen": 248315904
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009340822467402207,
+      "loss": 1.9713,
+      "theoretical_loss": 4.269219549321364,
+      "tokens_seen": 248381440
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000934062186559679,
+      "loss": 1.8625,
+      "theoretical_loss": 4.269084237158476,
+      "tokens_seen": 248446976
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009340421263791375,
+      "loss": 1.8055,
+      "theoretical_loss": 4.268948970674917,
+      "tokens_seen": 248512512
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009340220661985957,
+      "loss": 2.1709,
+      "theoretical_loss": 4.268813749843226,
+      "tokens_seen": 248578048
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009340020060180542,
+      "loss": 1.72,
+      "theoretical_loss": 4.268678574635967,
+      "tokens_seen": 248643584
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009339819458375125,
+      "loss": 1.7354,
+      "theoretical_loss": 4.268543445025724,
+      "tokens_seen": 248709120
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009339618856569709,
+      "loss": 1.8053,
+      "theoretical_loss": 4.268408360985109,
+      "tokens_seen": 248774656
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009339418254764293,
+      "loss": 1.9108,
+      "theoretical_loss": 4.268273322486757,
+      "tokens_seen": 248840192
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009339217652958876,
+      "loss": 1.775,
+      "theoretical_loss": 4.268138329503325,
+      "tokens_seen": 248905728
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009339017051153461,
+      "loss": 1.8734,
+      "theoretical_loss": 4.2680033820074925,
+      "tokens_seen": 248971264
+    },
+    {
+      "epoch": 0.08,
+      "objective/train/advantage_avg": 0.4903026223182678,
+      "objective/train/docs_used": 97104,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.707381010055542,
+      "objective/train/original_loss": 1.7073808908462524,
+      "objective/train/theoretical_loss": 4.267868479971966,
+      "objective/train/tokens_used": 269496800,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24140579998493195,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0502574443817139,
+      "objective/train/weighted_lm_loss": 1.793129563331604,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.267868479971966,
+      "tokens_seen": 249036800
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009338816449348045,
+      "loss": 1.7747,
+      "theoretical_loss": 4.267868479971966,
+      "tokens_seen": 249036800
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009338615847542628,
+      "loss": 1.7815,
+      "theoretical_loss": 4.267733623369474,
+      "tokens_seen": 249102336
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009338415245737212,
+      "loss": 1.8381,
+      "theoretical_loss": 4.267598812172769,
+      "tokens_seen": 249167872
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009338214643931797,
+      "loss": 1.966,
+      "theoretical_loss": 4.2674640463546245,
+      "tokens_seen": 249233408
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009338014042126379,
+      "loss": 1.752,
+      "theoretical_loss": 4.267329325887841,
+      "tokens_seen": 249298944
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009337813440320964,
+      "loss": 1.5597,
+      "theoretical_loss": 4.26719465074524,
+      "tokens_seen": 249364480
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009337612838515546,
+      "loss": 1.719,
+      "theoretical_loss": 4.267060020899668,
+      "tokens_seen": 249430016
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009337412236710131,
+      "loss": 1.804,
+      "theoretical_loss": 4.2669254363239935,
+      "tokens_seen": 249495552
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009337211634904715,
+      "loss": 1.7289,
+      "theoretical_loss": 4.266790896991109,
+      "tokens_seen": 249561088
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009337011033099298,
+      "loss": 1.8028,
+      "theoretical_loss": 4.266656402873932,
+      "tokens_seen": 249626624
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009336810431293882,
+      "loss": 1.9201,
+      "theoretical_loss": 4.2665219539454,
+      "tokens_seen": 249692160
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009336609829488466,
+      "loss": 1.9164,
+      "theoretical_loss": 4.266387550178476,
+      "tokens_seen": 249757696
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009336409227683049,
+      "loss": 1.8046,
+      "theoretical_loss": 4.266253191546146,
+      "tokens_seen": 249823232
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009336208625877634,
+      "loss": 1.8337,
+      "theoretical_loss": 4.266118878021419,
+      "tokens_seen": 249888768
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009336008024072216,
+      "loss": 1.9622,
+      "theoretical_loss": 4.265984609577327,
+      "tokens_seen": 249954304
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009335807422266801,
+      "loss": 1.8541,
+      "theoretical_loss": 4.265850386186925,
+      "tokens_seen": 250019840
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009335606820461384,
+      "loss": 1.6784,
+      "theoretical_loss": 4.265716207823292,
+      "tokens_seen": 250085376
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009335406218655968,
+      "loss": 1.7275,
+      "theoretical_loss": 4.265582074459531,
+      "tokens_seen": 250150912
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009335205616850552,
+      "loss": 1.8329,
+      "theoretical_loss": 4.265447986068765,
+      "tokens_seen": 250216448
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009335005015045136,
+      "loss": 1.8422,
+      "theoretical_loss": 4.265313942624143,
+      "tokens_seen": 250281984
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009334804413239719,
+      "loss": 1.9325,
+      "theoretical_loss": 4.2651799440988345,
+      "tokens_seen": 250347520
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009334603811434304,
+      "loss": 1.8524,
+      "theoretical_loss": 4.265045990466035,
+      "tokens_seen": 250413056
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009334403209628887,
+      "loss": 1.8634,
+      "theoretical_loss": 4.264912081698962,
+      "tokens_seen": 250478592
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009334202607823471,
+      "loss": 1.7535,
+      "theoretical_loss": 4.264778217770854,
+      "tokens_seen": 250544128
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009334002006018054,
+      "loss": 1.7067,
+      "theoretical_loss": 4.2646443986549745,
+      "tokens_seen": 250609664
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009333801404212638,
+      "loss": 1.767,
+      "theoretical_loss": 4.264510624324609,
+      "tokens_seen": 250675200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009333600802407222,
+      "loss": 1.9334,
+      "theoretical_loss": 4.2643768947530685,
+      "tokens_seen": 250740736
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009333400200601806,
+      "loss": 1.8349,
+      "theoretical_loss": 4.2642432099136816,
+      "tokens_seen": 250806272
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009333199598796389,
+      "loss": 1.751,
+      "theoretical_loss": 4.264109569779803,
+      "tokens_seen": 250871808
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009332998996990973,
+      "loss": 1.9473,
+      "theoretical_loss": 4.263975974324813,
+      "tokens_seen": 250937344
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009332798395185557,
+      "loss": 1.7429,
+      "theoretical_loss": 4.263842423522109,
+      "tokens_seen": 251002880
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009332597793380141,
+      "loss": 1.8841,
+      "theoretical_loss": 4.263708917345115,
+      "tokens_seen": 251068416
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009332397191574724,
+      "loss": 1.9271,
+      "theoretical_loss": 4.263575455767277,
+      "tokens_seen": 251133952
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009332196589769309,
+      "loss": 1.9296,
+      "theoretical_loss": 4.2634420387620615,
+      "tokens_seen": 251199488
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009331995987963891,
+      "loss": 1.8366,
+      "theoretical_loss": 4.2633086663029625,
+      "tokens_seen": 251265024
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009331795386158476,
+      "loss": 1.8976,
+      "theoretical_loss": 4.263175338363491,
+      "tokens_seen": 251330560
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009331594784353059,
+      "loss": 1.8301,
+      "theoretical_loss": 4.263042054917186,
+      "tokens_seen": 251396096
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009331394182547643,
+      "loss": 2.0389,
+      "theoretical_loss": 4.262908815937604,
+      "tokens_seen": 251461632
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009331193580742227,
+      "loss": 2.0028,
+      "theoretical_loss": 4.262775621398329,
+      "tokens_seen": 251527168
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009330992978936811,
+      "loss": 2.0408,
+      "theoretical_loss": 4.262642471272963,
+      "tokens_seen": 251592704
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009330792377131394,
+      "loss": 2.0631,
+      "theoretical_loss": 4.262509365535134,
+      "tokens_seen": 251658240
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009330591775325979,
+      "loss": 1.8172,
+      "theoretical_loss": 4.262376304158492,
+      "tokens_seen": 251723776
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009330391173520561,
+      "loss": 1.9694,
+      "theoretical_loss": 4.262243287116707,
+      "tokens_seen": 251789312
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009330190571715146,
+      "loss": 1.8517,
+      "theoretical_loss": 4.2621103143834755,
+      "tokens_seen": 251854848
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000932998996990973,
+      "loss": 2.0222,
+      "theoretical_loss": 4.261977385932512,
+      "tokens_seen": 251920384
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009329789368104313,
+      "loss": 1.9252,
+      "theoretical_loss": 4.261844501737556,
+      "tokens_seen": 251985920
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009329588766298897,
+      "loss": 1.9154,
+      "theoretical_loss": 4.26171166177237,
+      "tokens_seen": 252051456
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000932938816449348,
+      "loss": 1.8763,
+      "theoretical_loss": 4.261578866010739,
+      "tokens_seen": 252116992
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009329187562688064,
+      "loss": 1.9258,
+      "theoretical_loss": 4.261446114426466,
+      "tokens_seen": 252182528
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009328986960882649,
+      "loss": 1.823,
+      "theoretical_loss": 4.261313406993382,
+      "tokens_seen": 252248064
+    },
+    {
+      "epoch": 0.08,
+      "objective/train/advantage_avg": 0.49495235085487366,
+      "objective/train/docs_used": 98257,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.8033281564712524,
+      "objective/train/original_loss": 1.803328037261963,
+      "objective/train/theoretical_loss": 4.261180743685337,
+      "objective/train/tokens_used": 272773600,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24527713656425476,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0507421493530273,
+      "objective/train/weighted_lm_loss": 1.8945033550262451,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 1.0253151655197144,
+      "theoretical_loss": 4.261180743685337,
+      "tokens_seen": 252313600
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009328786359077231,
+      "loss": 1.8192,
+      "theoretical_loss": 4.261180743685337,
+      "tokens_seen": 252313600
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009328585757271816,
+      "loss": 1.9158,
+      "theoretical_loss": 4.261048124476204,
+      "tokens_seen": 252379136
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009328385155466398,
+      "loss": 1.9291,
+      "theoretical_loss": 4.260915549339879,
+      "tokens_seen": 252444672
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009328184553660983,
+      "loss": 1.7672,
+      "theoretical_loss": 4.2607830182502795,
+      "tokens_seen": 252510208
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009327983951855567,
+      "loss": 1.8606,
+      "theoretical_loss": 4.260650531181345,
+      "tokens_seen": 252575744
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000932778335005015,
+      "loss": 1.8812,
+      "theoretical_loss": 4.260518088107038,
+      "tokens_seen": 252641280
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009327582748244734,
+      "loss": 1.9065,
+      "theoretical_loss": 4.2603856890013425,
+      "tokens_seen": 252706816
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009327382146439319,
+      "loss": 1.8938,
+      "theoretical_loss": 4.260253333838264,
+      "tokens_seen": 252772352
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009327181544633901,
+      "loss": 1.9165,
+      "theoretical_loss": 4.260121022591832,
+      "tokens_seen": 252837888
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009326980942828486,
+      "loss": 1.7946,
+      "theoretical_loss": 4.259988755236097,
+      "tokens_seen": 252903424
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009326780341023069,
+      "loss": 1.9426,
+      "theoretical_loss": 4.25985653174513,
+      "tokens_seen": 252968960
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009326579739217653,
+      "loss": 1.7934,
+      "theoretical_loss": 4.259724352093027,
+      "tokens_seen": 253034496
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009326379137412237,
+      "loss": 1.7676,
+      "theoretical_loss": 4.259592216253903,
+      "tokens_seen": 253100032
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000932617853560682,
+      "loss": 1.8153,
+      "theoretical_loss": 4.259460124201899,
+      "tokens_seen": 253165568
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009325977933801404,
+      "loss": 1.7746,
+      "theoretical_loss": 4.259328075911173,
+      "tokens_seen": 253231104
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009325777331995988,
+      "loss": 1.7761,
+      "theoretical_loss": 4.259196071355909,
+      "tokens_seen": 253296640
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009325576730190571,
+      "loss": 1.9124,
+      "theoretical_loss": 4.259064110510311,
+      "tokens_seen": 253362176
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009325376128385156,
+      "loss": 1.8265,
+      "theoretical_loss": 4.258932193348604,
+      "tokens_seen": 253427712
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009325175526579739,
+      "loss": 1.9741,
+      "theoretical_loss": 4.258800319845038,
+      "tokens_seen": 253493248
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009324974924774323,
+      "loss": 1.8248,
+      "theoretical_loss": 4.258668489973882,
+      "tokens_seen": 253558784
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009324774322968907,
+      "loss": 1.8307,
+      "theoretical_loss": 4.258536703709426,
+      "tokens_seen": 253624320
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000932457372116349,
+      "loss": 1.7819,
+      "theoretical_loss": 4.258404961025986,
+      "tokens_seen": 253689856
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009324373119358074,
+      "loss": 1.9135,
+      "theoretical_loss": 4.258273261897896,
+      "tokens_seen": 253755392
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009324172517552658,
+      "loss": 1.8765,
+      "theoretical_loss": 4.258141606299514,
+      "tokens_seen": 253820928
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009323971915747242,
+      "loss": 1.7368,
+      "theoretical_loss": 4.258009994205218,
+      "tokens_seen": 253886464
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009323771313941826,
+      "loss": 1.8066,
+      "theoretical_loss": 4.257878425589407,
+      "tokens_seen": 253952000
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000932357071213641,
+      "loss": 1.8179,
+      "theoretical_loss": 4.257746900426506,
+      "tokens_seen": 254017536
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009323370110330993,
+      "loss": 1.7851,
+      "theoretical_loss": 4.257615418690956,
+      "tokens_seen": 254083072
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009323169508525577,
+      "loss": 1.7856,
+      "theoretical_loss": 4.2574839803572235,
+      "tokens_seen": 254148608
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009322968906720161,
+      "loss": 1.8776,
+      "theoretical_loss": 4.257352585399795,
+      "tokens_seen": 254214144
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009322768304914745,
+      "loss": 1.8318,
+      "theoretical_loss": 4.25722123379318,
+      "tokens_seen": 254279680
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009322567703109328,
+      "loss": 1.952,
+      "theoretical_loss": 4.257089925511907,
+      "tokens_seen": 254345216
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009322367101303913,
+      "loss": 2.0118,
+      "theoretical_loss": 4.256958660530529,
+      "tokens_seen": 254410752
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009322166499498495,
+      "loss": 1.8912,
+      "theoretical_loss": 4.256827438823617,
+      "tokens_seen": 254476288
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000932196589769308,
+      "loss": 1.7353,
+      "theoretical_loss": 4.256696260365768,
+      "tokens_seen": 254541824
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009321765295887663,
+      "loss": 1.8728,
+      "theoretical_loss": 4.256565125131597,
+      "tokens_seen": 254607360
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009321564694082247,
+      "loss": 1.8088,
+      "theoretical_loss": 4.25643403309574,
+      "tokens_seen": 254672896
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009321364092276831,
+      "loss": 1.8988,
+      "theoretical_loss": 4.256302984232858,
+      "tokens_seen": 254738432
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009321163490471415,
+      "loss": 1.8722,
+      "theoretical_loss": 4.256171978517629,
+      "tokens_seen": 254803968
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009320962888665998,
+      "loss": 1.6762,
+      "theoretical_loss": 4.256041015924756,
+      "tokens_seen": 254869504
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009320762286860583,
+      "loss": 1.8221,
+      "theoretical_loss": 4.255910096428962,
+      "tokens_seen": 254935040
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009320561685055165,
+      "loss": 1.8804,
+      "theoretical_loss": 4.255779220004991,
+      "tokens_seen": 255000576
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000932036108324975,
+      "loss": 1.6887,
+      "theoretical_loss": 4.255648386627607,
+      "tokens_seen": 255066112
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009320160481444333,
+      "loss": 1.8639,
+      "theoretical_loss": 4.255517596271598,
+      "tokens_seen": 255131648
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009319959879638917,
+      "loss": 1.8197,
+      "theoretical_loss": 4.255386848911772,
+      "tokens_seen": 255197184
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009319759277833501,
+      "loss": 1.7893,
+      "theoretical_loss": 4.255256144522958,
+      "tokens_seen": 255262720
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009319558676028084,
+      "loss": 1.8523,
+      "theoretical_loss": 4.255125483080007,
+      "tokens_seen": 255328256
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009319358074222668,
+      "loss": 1.8729,
+      "theoretical_loss": 4.254994864557789,
+      "tokens_seen": 255393792
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009319157472417253,
+      "loss": 1.7798,
+      "theoretical_loss": 4.254864288931198,
+      "tokens_seen": 255459328
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009318956870611835,
+      "loss": 1.8593,
+      "theoretical_loss": 4.254733756175147,
+      "tokens_seen": 255524864
+    },
+    {
+      "epoch": 0.08,
+      "objective/train/advantage_avg": 0.4922325015068054,
+      "objective/train/docs_used": 99403,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.7772718667984009,
+      "objective/train/original_loss": 1.7772717475891113,
+      "objective/train/theoretical_loss": 4.254603266264572,
+      "objective/train/tokens_used": 276050400,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24300870299339294,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050458550453186,
+      "objective/train/weighted_lm_loss": 1.8668968677520752,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.254603266264572,
+      "tokens_seen": 255590400
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000931875626880642,
+      "loss": 1.7993,
+      "theoretical_loss": 4.254603266264572,
+      "tokens_seen": 255590400
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009318555667001002,
+      "loss": 1.7996,
+      "theoretical_loss": 4.254472819174428,
+      "tokens_seen": 255655936
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009318355065195587,
+      "loss": 1.6432,
+      "theoretical_loss": 4.254342414879692,
+      "tokens_seen": 255721472
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009318154463390171,
+      "loss": 1.833,
+      "theoretical_loss": 4.254212053355362,
+      "tokens_seen": 255787008
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009317953861584754,
+      "loss": 1.8269,
+      "theoretical_loss": 4.254081734576458,
+      "tokens_seen": 255852544
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009317753259779338,
+      "loss": 1.7435,
+      "theoretical_loss": 4.25395145851802,
+      "tokens_seen": 255918080
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009317552657973923,
+      "loss": 1.8863,
+      "theoretical_loss": 4.253821225155107,
+      "tokens_seen": 255983616
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009317352056168505,
+      "loss": 1.7373,
+      "theoretical_loss": 4.253691034462804,
+      "tokens_seen": 256049152
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000931715145436309,
+      "loss": 1.8167,
+      "theoretical_loss": 4.253560886416212,
+      "tokens_seen": 256114688
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009316950852557673,
+      "loss": 1.8065,
+      "theoretical_loss": 4.253430780990455,
+      "tokens_seen": 256180224
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009316750250752257,
+      "loss": 1.7302,
+      "theoretical_loss": 4.253300718160678,
+      "tokens_seen": 256245760
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009316549648946841,
+      "loss": 1.8542,
+      "theoretical_loss": 4.253170697902046,
+      "tokens_seen": 256311296
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009316349047141424,
+      "loss": 1.7814,
+      "theoretical_loss": 4.253040720189746,
+      "tokens_seen": 256376832
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009316148445336008,
+      "loss": 1.9602,
+      "theoretical_loss": 4.252910784998984,
+      "tokens_seen": 256442368
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009315947843530592,
+      "loss": 1.6887,
+      "theoretical_loss": 4.2527808923049895,
+      "tokens_seen": 256507904
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009315747241725175,
+      "loss": 1.8397,
+      "theoretical_loss": 4.252651042083009,
+      "tokens_seen": 256573440
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000931554663991976,
+      "loss": 1.7413,
+      "theoretical_loss": 4.252521234308315,
+      "tokens_seen": 256638976
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009315346038114343,
+      "loss": 1.7299,
+      "theoretical_loss": 4.252391468956194,
+      "tokens_seen": 256704512
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009315145436308927,
+      "loss": 1.7748,
+      "theoretical_loss": 4.25226174600196,
+      "tokens_seen": 256770048
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009314944834503511,
+      "loss": 1.7904,
+      "theoretical_loss": 4.252132065420942,
+      "tokens_seen": 256835584
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009314744232698095,
+      "loss": 1.8599,
+      "theoretical_loss": 4.2520024271884935,
+      "tokens_seen": 256901120
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009314543630892678,
+      "loss": 1.6606,
+      "theoretical_loss": 4.251872831279986,
+      "tokens_seen": 256966656
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009314343029087262,
+      "loss": 1.7963,
+      "theoretical_loss": 4.251743277670815,
+      "tokens_seen": 257032192
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009314142427281845,
+      "loss": 1.9087,
+      "theoretical_loss": 4.251613766336391,
+      "tokens_seen": 257097728
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000931394182547643,
+      "loss": 1.9016,
+      "theoretical_loss": 4.251484297252151,
+      "tokens_seen": 257163264
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009313741223671013,
+      "loss": 1.9086,
+      "theoretical_loss": 4.25135487039355,
+      "tokens_seen": 257228800
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009313540621865597,
+      "loss": 1.9209,
+      "theoretical_loss": 4.251225485736063,
+      "tokens_seen": 257294336
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000931334002006018,
+      "loss": 1.7854,
+      "theoretical_loss": 4.251096143255185,
+      "tokens_seen": 257359872
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009313139418254765,
+      "loss": 1.9528,
+      "theoretical_loss": 4.250966842926434,
+      "tokens_seen": 257425408
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009312938816449348,
+      "loss": 1.808,
+      "theoretical_loss": 4.250837584725346,
+      "tokens_seen": 257490944
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009312738214643932,
+      "loss": 1.7127,
+      "theoretical_loss": 4.250708368627477,
+      "tokens_seen": 257556480
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009312537612838515,
+      "loss": 1.9612,
+      "theoretical_loss": 4.250579194608408,
+      "tokens_seen": 257622016
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009312337011033099,
+      "loss": 1.8438,
+      "theoretical_loss": 4.250450062643734,
+      "tokens_seen": 257687552
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009312136409227683,
+      "loss": 1.7651,
+      "theoretical_loss": 4.2503209727090745,
+      "tokens_seen": 257753088
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009311935807422267,
+      "loss": 1.8181,
+      "theoretical_loss": 4.250191924780069,
+      "tokens_seen": 257818624
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000931173520561685,
+      "loss": 1.781,
+      "theoretical_loss": 4.250062918832374,
+      "tokens_seen": 257884160
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009311534603811435,
+      "loss": 1.6716,
+      "theoretical_loss": 4.249933954841672,
+      "tokens_seen": 257949696
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009311334002006018,
+      "loss": 2.0471,
+      "theoretical_loss": 4.249805032783661,
+      "tokens_seen": 258015232
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009311133400200602,
+      "loss": 1.8217,
+      "theoretical_loss": 4.249676152634061,
+      "tokens_seen": 258080768
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009310932798395186,
+      "loss": 1.9101,
+      "theoretical_loss": 4.249547314368613,
+      "tokens_seen": 258146304
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009310732196589769,
+      "loss": 1.8241,
+      "theoretical_loss": 4.2494185179630755,
+      "tokens_seen": 258211840
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009310531594784353,
+      "loss": 1.8213,
+      "theoretical_loss": 4.24928976339323,
+      "tokens_seen": 258277376
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009310330992978937,
+      "loss": 1.8407,
+      "theoretical_loss": 4.249161050634876,
+      "tokens_seen": 258342912
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000931013039117352,
+      "loss": 1.9322,
+      "theoretical_loss": 4.249032379663836,
+      "tokens_seen": 258408448
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009309929789368105,
+      "loss": 1.7703,
+      "theoretical_loss": 4.24890375045595,
+      "tokens_seen": 258473984
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009309729187562687,
+      "loss": 1.8058,
+      "theoretical_loss": 4.248775162987078,
+      "tokens_seen": 258539520
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009309528585757272,
+      "loss": 1.7423,
+      "theoretical_loss": 4.248646617233103,
+      "tokens_seen": 258605056
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009309327983951856,
+      "loss": 1.7474,
+      "theoretical_loss": 4.248518113169924,
+      "tokens_seen": 258670592
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009309127382146439,
+      "loss": 1.9037,
+      "theoretical_loss": 4.248389650773463,
+      "tokens_seen": 258736128
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009308926780341024,
+      "loss": 1.7221,
+      "theoretical_loss": 4.248261230019661,
+      "tokens_seen": 258801664
+    },
+    {
+      "epoch": 0.08,
+      "objective/train/advantage_avg": 0.4922111928462982,
+      "objective/train/docs_used": 100739,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.5991153717041016,
+      "objective/train/original_loss": 1.5991151332855225,
+      "objective/train/theoretical_loss": 4.24813285088448,
+      "objective/train/tokens_used": 279327200,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24256941676139832,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504542589187622,
+      "objective/train/weighted_lm_loss": 1.679618239402771,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 1.0253151655197144,
+      "theoretical_loss": 4.24813285088448,
+      "tokens_seen": 258867200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009308726178535606,
+      "loss": 1.7685,
+      "theoretical_loss": 4.24813285088448,
+      "tokens_seen": 258867200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009308525576730191,
+      "loss": 1.8756,
+      "theoretical_loss": 4.248004513343899,
+      "tokens_seen": 258932736
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009308324974924775,
+      "loss": 1.9389,
+      "theoretical_loss": 4.24787621737392,
+      "tokens_seen": 258998272
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009308124373119358,
+      "loss": 1.7622,
+      "theoretical_loss": 4.247747962950564,
+      "tokens_seen": 259063808
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009307923771313942,
+      "loss": 1.8245,
+      "theoretical_loss": 4.247619750049871,
+      "tokens_seen": 259129344
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009307723169508527,
+      "loss": 1.7996,
+      "theoretical_loss": 4.247491578647903,
+      "tokens_seen": 259194880
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009307522567703109,
+      "loss": 1.7617,
+      "theoretical_loss": 4.247363448720739,
+      "tokens_seen": 259260416
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009307321965897694,
+      "loss": 1.7627,
+      "theoretical_loss": 4.24723536024448,
+      "tokens_seen": 259325952
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009307121364092277,
+      "loss": 1.8327,
+      "theoretical_loss": 4.247107313195246,
+      "tokens_seen": 259391488
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009306920762286861,
+      "loss": 1.8597,
+      "theoretical_loss": 4.246979307549177,
+      "tokens_seen": 259457024
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009306720160481445,
+      "loss": 1.8707,
+      "theoretical_loss": 4.246851343282432,
+      "tokens_seen": 259522560
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009306519558676028,
+      "loss": 1.9826,
+      "theoretical_loss": 4.246723420371191,
+      "tokens_seen": 259588096
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009306318956870612,
+      "loss": 1.828,
+      "theoretical_loss": 4.246595538791653,
+      "tokens_seen": 259653632
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009306118355065196,
+      "loss": 1.8696,
+      "theoretical_loss": 4.246467698520037,
+      "tokens_seen": 259719168
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009305917753259779,
+      "loss": 1.823,
+      "theoretical_loss": 4.246339899532582,
+      "tokens_seen": 259784704
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009305717151454364,
+      "loss": 1.872,
+      "theoretical_loss": 4.246212141805544,
+      "tokens_seen": 259850240
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009305516549648947,
+      "loss": 1.8403,
+      "theoretical_loss": 4.246084425315203,
+      "tokens_seen": 259915776
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009305315947843531,
+      "loss": 1.6529,
+      "theoretical_loss": 4.245956750037855,
+      "tokens_seen": 259981312
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009305115346038115,
+      "loss": 1.7805,
+      "theoretical_loss": 4.245829115949818,
+      "tokens_seen": 260046848
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009304914744232699,
+      "loss": 1.9065,
+      "theoretical_loss": 4.245701523027429,
+      "tokens_seen": 260112384
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009304714142427282,
+      "loss": 1.8109,
+      "theoretical_loss": 4.245573971247042,
+      "tokens_seen": 260177920
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009304513540621866,
+      "loss": 1.8887,
+      "theoretical_loss": 4.245446460585035,
+      "tokens_seen": 260243456
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009304312938816449,
+      "loss": 1.886,
+      "theoretical_loss": 4.245318991017802,
+      "tokens_seen": 260308992
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009304112337011034,
+      "loss": 1.7386,
+      "theoretical_loss": 4.245191562521757,
+      "tokens_seen": 260374528
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009303911735205617,
+      "loss": 1.8463,
+      "theoretical_loss": 4.2450641750733356,
+      "tokens_seen": 260440064
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009303711133400201,
+      "loss": 1.7975,
+      "theoretical_loss": 4.24493682864899,
+      "tokens_seen": 260505600
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009303510531594784,
+      "loss": 1.8787,
+      "theoretical_loss": 4.244809523225195,
+      "tokens_seen": 260571136
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009303309929789369,
+      "loss": 1.7446,
+      "theoretical_loss": 4.244682258778441,
+      "tokens_seen": 260636672
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009303109327983952,
+      "loss": 1.8998,
+      "theoretical_loss": 4.2445550352852415,
+      "tokens_seen": 260702208
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009302908726178536,
+      "loss": 1.919,
+      "theoretical_loss": 4.244427852722127,
+      "tokens_seen": 260767744
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000930270812437312,
+      "loss": 1.8169,
+      "theoretical_loss": 4.244300711065646,
+      "tokens_seen": 260833280
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009302507522567703,
+      "loss": 1.785,
+      "theoretical_loss": 4.244173610292372,
+      "tokens_seen": 260898816
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009302306920762287,
+      "loss": 1.685,
+      "theoretical_loss": 4.244046550378892,
+      "tokens_seen": 260964352
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009302106318956871,
+      "loss": 1.6913,
+      "theoretical_loss": 4.243919531301814,
+      "tokens_seen": 261029888
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009301905717151454,
+      "loss": 1.7615,
+      "theoretical_loss": 4.243792553037767,
+      "tokens_seen": 261095424
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009301705115346039,
+      "loss": 1.8247,
+      "theoretical_loss": 4.243665615563396,
+      "tokens_seen": 261160960
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009301504513540622,
+      "loss": 1.8357,
+      "theoretical_loss": 4.243538718855369,
+      "tokens_seen": 261226496
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009301303911735206,
+      "loss": 1.8834,
+      "theoretical_loss": 4.243411862890371,
+      "tokens_seen": 261292032
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000930110330992979,
+      "loss": 1.7083,
+      "theoretical_loss": 4.243285047645106,
+      "tokens_seen": 261357568
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009300902708124373,
+      "loss": 1.8269,
+      "theoretical_loss": 4.243158273096297,
+      "tokens_seen": 261423104
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009300702106318957,
+      "loss": 1.9182,
+      "theoretical_loss": 4.243031539220689,
+      "tokens_seen": 261488640
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009300501504513541,
+      "loss": 1.9445,
+      "theoretical_loss": 4.242904845995041,
+      "tokens_seen": 261554176
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009300300902708124,
+      "loss": 1.8878,
+      "theoretical_loss": 4.242778193396136,
+      "tokens_seen": 261619712
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009300100300902709,
+      "loss": 1.8315,
+      "theoretical_loss": 4.242651581400774,
+      "tokens_seen": 261685248
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009299899699097291,
+      "loss": 1.8104,
+      "theoretical_loss": 4.242525009985773,
+      "tokens_seen": 261750784
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009299699097291876,
+      "loss": 1.8935,
+      "theoretical_loss": 4.242398479127972,
+      "tokens_seen": 261816320
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000929949849548646,
+      "loss": 1.7999,
+      "theoretical_loss": 4.242271988804228,
+      "tokens_seen": 261881856
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009299297893681043,
+      "loss": 1.7533,
+      "theoretical_loss": 4.242145538991417,
+      "tokens_seen": 261947392
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009299097291875627,
+      "loss": 1.6952,
+      "theoretical_loss": 4.242019129666434,
+      "tokens_seen": 262012928
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000929889669007021,
+      "loss": 1.688,
+      "theoretical_loss": 4.241892760806194,
+      "tokens_seen": 262078464
+    },
+    {
+      "epoch": 0.08,
+      "objective/train/advantage_avg": 0.4959133267402649,
+      "objective/train/docs_used": 101934,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.8622653484344482,
+      "objective/train/original_loss": 1.8622653484344482,
+      "objective/train/theoretical_loss": 4.241766432387629,
+      "objective/train/tokens_used": 282604000,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24602378904819489,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050842046737671,
+      "objective/train/weighted_lm_loss": 1.957066535949707,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 1.0396549701690674,
+      "theoretical_loss": 4.241766432387629,
+      "tokens_seen": 262144000
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009298696088264794,
+      "loss": 1.8692,
+      "theoretical_loss": 4.241766432387629,
+      "tokens_seen": 262144000
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009298495486459379,
+      "loss": 1.8105,
+      "theoretical_loss": 4.241640144387691,
+      "tokens_seen": 262209536
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009298294884653961,
+      "loss": 1.7178,
+      "theoretical_loss": 4.241513896783352,
+      "tokens_seen": 262275072
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009298094282848546,
+      "loss": 1.8063,
+      "theoretical_loss": 4.2413876895516,
+      "tokens_seen": 262340608
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000929789368104313,
+      "loss": 1.8252,
+      "theoretical_loss": 4.241261522669445,
+      "tokens_seen": 262406144
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009297693079237713,
+      "loss": 1.8267,
+      "theoretical_loss": 4.241135396113914,
+      "tokens_seen": 262471680
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009297492477432297,
+      "loss": 1.9232,
+      "theoretical_loss": 4.2410093098620525,
+      "tokens_seen": 262537216
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000929729187562688,
+      "loss": 1.7269,
+      "theoretical_loss": 4.240883263890925,
+      "tokens_seen": 262602752
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009297091273821464,
+      "loss": 1.9214,
+      "theoretical_loss": 4.240757258177617,
+      "tokens_seen": 262668288
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009296890672016049,
+      "loss": 1.8207,
+      "theoretical_loss": 4.240631292699231,
+      "tokens_seen": 262733824
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009296690070210631,
+      "loss": 1.7695,
+      "theoretical_loss": 4.2405053674328865,
+      "tokens_seen": 262799360
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009296489468405216,
+      "loss": 1.789,
+      "theoretical_loss": 4.240379482355724,
+      "tokens_seen": 262864896
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009296288866599799,
+      "loss": 1.6906,
+      "theoretical_loss": 4.240253637444903,
+      "tokens_seen": 262930432
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009296088264794383,
+      "loss": 1.689,
+      "theoretical_loss": 4.2401278326776,
+      "tokens_seen": 262995968
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009295887662988967,
+      "loss": 1.7909,
+      "theoretical_loss": 4.2400020680310115,
+      "tokens_seen": 263061504
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009295687061183551,
+      "loss": 1.7541,
+      "theoretical_loss": 4.239876343482352,
+      "tokens_seen": 263127040
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009295486459378134,
+      "loss": 1.8034,
+      "theoretical_loss": 4.239750659008854,
+      "tokens_seen": 263192576
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009295285857572719,
+      "loss": 1.7027,
+      "theoretical_loss": 4.23962501458777,
+      "tokens_seen": 263258112
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009295085255767301,
+      "loss": 1.923,
+      "theoretical_loss": 4.2394994101963706,
+      "tokens_seen": 263323648
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009294884653961886,
+      "loss": 1.7502,
+      "theoretical_loss": 4.239373845811944,
+      "tokens_seen": 263389184
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009294684052156469,
+      "loss": 1.8158,
+      "theoretical_loss": 4.2392483214117975,
+      "tokens_seen": 263454720
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009294483450351053,
+      "loss": 1.9041,
+      "theoretical_loss": 4.239122836973258,
+      "tokens_seen": 263520256
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009294282848545638,
+      "loss": 1.8612,
+      "theoretical_loss": 4.238997392473669,
+      "tokens_seen": 263585792
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009294082246740221,
+      "loss": 1.8119,
+      "theoretical_loss": 4.238871987890395,
+      "tokens_seen": 263651328
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009293881644934805,
+      "loss": 1.8197,
+      "theoretical_loss": 4.238746623200815,
+      "tokens_seen": 263716864
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009293681043129388,
+      "loss": 1.8277,
+      "theoretical_loss": 4.238621298382331,
+      "tokens_seen": 263782400
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009293480441323973,
+      "loss": 1.8537,
+      "theoretical_loss": 4.23849601341236,
+      "tokens_seen": 263847936
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009293279839518556,
+      "loss": 1.7895,
+      "theoretical_loss": 4.2383707682683385,
+      "tokens_seen": 263913472
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000929307923771314,
+      "loss": 1.8325,
+      "theoretical_loss": 4.238245562927722,
+      "tokens_seen": 263979008
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009292878635907723,
+      "loss": 1.7732,
+      "theoretical_loss": 4.238120397367984,
+      "tokens_seen": 264044544
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009292678034102307,
+      "loss": 1.8699,
+      "theoretical_loss": 4.237995271566616,
+      "tokens_seen": 264110080
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009292477432296891,
+      "loss": 1.9023,
+      "theoretical_loss": 4.237870185501127,
+      "tokens_seen": 264175616
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009292276830491475,
+      "loss": 1.7925,
+      "theoretical_loss": 4.237745139149047,
+      "tokens_seen": 264241152
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009292076228686058,
+      "loss": 1.825,
+      "theoretical_loss": 4.237620132487923,
+      "tokens_seen": 264306688
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009291875626880643,
+      "loss": 1.7765,
+      "theoretical_loss": 4.237495165495317,
+      "tokens_seen": 264372224
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009291675025075226,
+      "loss": 1.7354,
+      "theoretical_loss": 4.237370238148814,
+      "tokens_seen": 264437760
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000929147442326981,
+      "loss": 1.7925,
+      "theoretical_loss": 4.237245350426015,
+      "tokens_seen": 264503296
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009291273821464394,
+      "loss": 1.6433,
+      "theoretical_loss": 4.23712050230454,
+      "tokens_seen": 264568832
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009291073219658977,
+      "loss": 1.8064,
+      "theoretical_loss": 4.236995693762026,
+      "tokens_seen": 264634368
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009290872617853561,
+      "loss": 1.6874,
+      "theoretical_loss": 4.236870924776128,
+      "tokens_seen": 264699904
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009290672016048145,
+      "loss": 1.8004,
+      "theoretical_loss": 4.236746195324523,
+      "tokens_seen": 264765440
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009290471414242728,
+      "loss": 1.8509,
+      "theoretical_loss": 4.2366215053849,
+      "tokens_seen": 264830976
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009290270812437313,
+      "loss": 1.8856,
+      "theoretical_loss": 4.23649685493497,
+      "tokens_seen": 264896512
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009290070210631895,
+      "loss": 1.8162,
+      "theoretical_loss": 4.236372243952461,
+      "tokens_seen": 264962048
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000928986960882648,
+      "loss": 1.9025,
+      "theoretical_loss": 4.2362476724151215,
+      "tokens_seen": 265027584
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009289669007021064,
+      "loss": 1.8805,
+      "theoretical_loss": 4.236123140300713,
+      "tokens_seen": 265093120
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009289468405215647,
+      "loss": 1.7836,
+      "theoretical_loss": 4.235998647587021,
+      "tokens_seen": 265158656
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009289267803410231,
+      "loss": 1.8182,
+      "theoretical_loss": 4.235874194251843,
+      "tokens_seen": 265224192
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009289067201604814,
+      "loss": 1.737,
+      "theoretical_loss": 4.235749780272998,
+      "tokens_seen": 265289728
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009288866599799398,
+      "loss": 1.8422,
+      "theoretical_loss": 4.2356254056283245,
+      "tokens_seen": 265355264
+    },
+    {
+      "epoch": 0.08,
+      "objective/train/advantage_avg": 0.49507758021354675,
+      "objective/train/docs_used": 103026,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.5715179443359375,
+      "objective/train/original_loss": 1.5715179443359375,
+      "objective/train/theoretical_loss": 4.235501070295674,
+      "objective/train/tokens_used": 285880800,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24532651901245117,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0507550239562988,
+      "objective/train/weighted_lm_loss": 1.6512619256973267,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.235501070295674,
+      "tokens_seen": 265420800
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009288665997993983,
+      "loss": 1.8102,
+      "theoretical_loss": 4.235501070295674,
+      "tokens_seen": 265420800
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009288465396188565,
+      "loss": 1.8941,
+      "theoretical_loss": 4.235376774252922,
+      "tokens_seen": 265486336
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000928826479438315,
+      "loss": 1.9101,
+      "theoretical_loss": 4.235252517477956,
+      "tokens_seen": 265551872
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009288064192577734,
+      "loss": 1.7015,
+      "theoretical_loss": 4.235128299948684,
+      "tokens_seen": 265617408
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009287863590772317,
+      "loss": 1.7076,
+      "theoretical_loss": 4.235004121643034,
+      "tokens_seen": 265682944
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009287662988966901,
+      "loss": 1.8181,
+      "theoretical_loss": 4.234879982538949,
+      "tokens_seen": 265748480
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009287462387161485,
+      "loss": 1.7664,
+      "theoretical_loss": 4.23475588261439,
+      "tokens_seen": 265814016
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009287261785356068,
+      "loss": 1.787,
+      "theoretical_loss": 4.234631821847337,
+      "tokens_seen": 265879552
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009287061183550653,
+      "loss": 1.7446,
+      "theoretical_loss": 4.234507800215788,
+      "tokens_seen": 265945088
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009286860581745235,
+      "loss": 1.8445,
+      "theoretical_loss": 4.234383817697755,
+      "tokens_seen": 266010624
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000928665997993982,
+      "loss": 1.777,
+      "theoretical_loss": 4.234259874271275,
+      "tokens_seen": 266076160
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009286459378134403,
+      "loss": 1.8054,
+      "theoretical_loss": 4.234135969914396,
+      "tokens_seen": 266141696
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009286258776328987,
+      "loss": 1.9136,
+      "theoretical_loss": 4.234012104605187,
+      "tokens_seen": 266207232
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009286058174523571,
+      "loss": 1.7999,
+      "theoretical_loss": 4.233888278321734,
+      "tokens_seen": 266272768
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009285857572718155,
+      "loss": 1.6735,
+      "theoretical_loss": 4.23376449104214,
+      "tokens_seen": 266338304
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009285656970912738,
+      "loss": 1.8224,
+      "theoretical_loss": 4.233640742744528,
+      "tokens_seen": 266403840
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009285456369107323,
+      "loss": 1.9214,
+      "theoretical_loss": 4.233517033407035,
+      "tokens_seen": 266469376
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009285255767301905,
+      "loss": 1.8953,
+      "theoretical_loss": 4.23339336300782,
+      "tokens_seen": 266534912
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000928505516549649,
+      "loss": 1.8393,
+      "theoretical_loss": 4.233269731525055,
+      "tokens_seen": 266600448
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009284854563691073,
+      "loss": 1.8253,
+      "theoretical_loss": 4.233146138936933,
+      "tokens_seen": 266665984
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009284653961885657,
+      "loss": 1.8904,
+      "theoretical_loss": 4.233022585221662,
+      "tokens_seen": 266731520
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009284453360080241,
+      "loss": 1.8195,
+      "theoretical_loss": 4.232899070357472,
+      "tokens_seen": 266797056
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009284252758274825,
+      "loss": 1.7398,
+      "theoretical_loss": 4.232775594322605,
+      "tokens_seen": 266862592
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009284052156469408,
+      "loss": 1.8531,
+      "theoretical_loss": 4.232652157095323,
+      "tokens_seen": 266928128
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009283851554663992,
+      "loss": 1.8012,
+      "theoretical_loss": 4.232528758653906,
+      "tokens_seen": 266993664
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009283650952858576,
+      "loss": 1.7835,
+      "theoretical_loss": 4.232405398976653,
+      "tokens_seen": 267059200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000928345035105316,
+      "loss": 1.7591,
+      "theoretical_loss": 4.232282078041876,
+      "tokens_seen": 267124736
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009283249749247743,
+      "loss": 1.7493,
+      "theoretical_loss": 4.232158795827908,
+      "tokens_seen": 267190272
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009283049147442327,
+      "loss": 1.8331,
+      "theoretical_loss": 4.232035552313097,
+      "tokens_seen": 267255808
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000928284854563691,
+      "loss": 1.759,
+      "theoretical_loss": 4.231912347475812,
+      "tokens_seen": 267321344
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009282647943831495,
+      "loss": 1.9049,
+      "theoretical_loss": 4.231789181294436,
+      "tokens_seen": 267386880
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009282447342026078,
+      "loss": 1.8436,
+      "theoretical_loss": 4.231666053747371,
+      "tokens_seen": 267452416
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009282246740220662,
+      "loss": 1.9009,
+      "theoretical_loss": 4.231542964813034,
+      "tokens_seen": 267517952
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009282046138415246,
+      "loss": 1.7458,
+      "theoretical_loss": 4.231419914469864,
+      "tokens_seen": 267583488
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000928184553660983,
+      "loss": 1.9132,
+      "theoretical_loss": 4.231296902696314,
+      "tokens_seen": 267649024
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009281644934804413,
+      "loss": 1.738,
+      "theoretical_loss": 4.231173929470853,
+      "tokens_seen": 267714560
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009281444332998998,
+      "loss": 1.8208,
+      "theoretical_loss": 4.231050994771971,
+      "tokens_seen": 267780096
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000928124373119358,
+      "loss": 1.7115,
+      "theoretical_loss": 4.230928098578174,
+      "tokens_seen": 267845632
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009281043129388165,
+      "loss": 1.7666,
+      "theoretical_loss": 4.230805240867982,
+      "tokens_seen": 267911168
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009280842527582748,
+      "loss": 1.8364,
+      "theoretical_loss": 4.230682421619936,
+      "tokens_seen": 267976704
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009280641925777332,
+      "loss": 1.7936,
+      "theoretical_loss": 4.2305596408125945,
+      "tokens_seen": 268042240
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009280441323971916,
+      "loss": 1.7574,
+      "theoretical_loss": 4.23043689842453,
+      "tokens_seen": 268107776
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009280240722166499,
+      "loss": 1.7779,
+      "theoretical_loss": 4.230314194434336,
+      "tokens_seen": 268173312
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009280040120361083,
+      "loss": 1.7898,
+      "theoretical_loss": 4.23019152882062,
+      "tokens_seen": 268238848
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009279839518555668,
+      "loss": 1.8302,
+      "theoretical_loss": 4.2300689015620065,
+      "tokens_seen": 268304384
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000927963891675025,
+      "loss": 1.8138,
+      "theoretical_loss": 4.229946312637141,
+      "tokens_seen": 268369920
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009279438314944835,
+      "loss": 1.7802,
+      "theoretical_loss": 4.229823762024681,
+      "tokens_seen": 268435456
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009279237713139418,
+      "loss": 1.8906,
+      "theoretical_loss": 4.2297012497033055,
+      "tokens_seen": 268500992
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009279037111334002,
+      "loss": 1.8555,
+      "theoretical_loss": 4.229578775651707,
+      "tokens_seen": 268566528
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009278836509528587,
+      "loss": 1.7775,
+      "theoretical_loss": 4.229456339848598,
+      "tokens_seen": 268632064
+    },
+    {
+      "epoch": 0.08,
+      "objective/train/advantage_avg": 0.49237704277038574,
+      "objective/train/docs_used": 104173,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.5822020769119263,
+      "objective/train/original_loss": 1.5822019577026367,
+      "objective/train/theoretical_loss": 4.2293339422727065,
+      "objective/train/tokens_used": 289157600,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24295735359191895,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504728555679321,
+      "objective/train/weighted_lm_loss": 1.661965250968933,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.2293339422727065,
+      "tokens_seen": 268697600
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009278635907723169,
+      "loss": 1.677,
+      "theoretical_loss": 4.2293339422727065,
+      "tokens_seen": 268697600
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009278435305917754,
+      "loss": 1.771,
+      "theoretical_loss": 4.229211582902778,
+      "tokens_seen": 268763136
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009278234704112338,
+      "loss": 1.8272,
+      "theoretical_loss": 4.229089261717574,
+      "tokens_seen": 268828672
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009278034102306921,
+      "loss": 1.8982,
+      "theoretical_loss": 4.2289669786958735,
+      "tokens_seen": 268894208
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009277833500501505,
+      "loss": 1.8134,
+      "theoretical_loss": 4.228844733816474,
+      "tokens_seen": 268959744
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009277632898696089,
+      "loss": 1.8508,
+      "theoretical_loss": 4.228722527058188,
+      "tokens_seen": 269025280
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009277432296890672,
+      "loss": 1.7465,
+      "theoretical_loss": 4.228600358399845,
+      "tokens_seen": 269090816
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009277231695085257,
+      "loss": 1.8031,
+      "theoretical_loss": 4.228478227820292,
+      "tokens_seen": 269156352
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009277031093279839,
+      "loss": 1.6002,
+      "theoretical_loss": 4.228356135298394,
+      "tokens_seen": 269221888
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009276830491474424,
+      "loss": 1.7915,
+      "theoretical_loss": 4.2282340808130305,
+      "tokens_seen": 269287424
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009276629889669007,
+      "loss": 1.7165,
+      "theoretical_loss": 4.2281120643431,
+      "tokens_seen": 269352960
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009276429287863591,
+      "loss": 1.6739,
+      "theoretical_loss": 4.227990085867517,
+      "tokens_seen": 269418496
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009276228686058175,
+      "loss": 1.7462,
+      "theoretical_loss": 4.227868145365211,
+      "tokens_seen": 269484032
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009276028084252759,
+      "loss": 1.7257,
+      "theoretical_loss": 4.227746242815132,
+      "tokens_seen": 269549568
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009275827482447342,
+      "loss": 1.8566,
+      "theoretical_loss": 4.227624378196245,
+      "tokens_seen": 269615104
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009275626880641927,
+      "loss": 1.8052,
+      "theoretical_loss": 4.227502551487531,
+      "tokens_seen": 269680640
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000927542627883651,
+      "loss": 1.804,
+      "theoretical_loss": 4.227380762667987,
+      "tokens_seen": 269746176
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009275225677031094,
+      "loss": 1.8553,
+      "theoretical_loss": 4.2272590117166295,
+      "tokens_seen": 269811712
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009275025075225677,
+      "loss": 1.682,
+      "theoretical_loss": 4.22713729861249,
+      "tokens_seen": 269877248
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009274824473420261,
+      "loss": 1.7293,
+      "theoretical_loss": 4.2270156233346174,
+      "tokens_seen": 269942784
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009274623871614845,
+      "loss": 1.7395,
+      "theoretical_loss": 4.226893985862076,
+      "tokens_seen": 270008320
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009274423269809429,
+      "loss": 1.8387,
+      "theoretical_loss": 4.226772386173948,
+      "tokens_seen": 270073856
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009274222668004012,
+      "loss": 1.8202,
+      "theoretical_loss": 4.226650824249333,
+      "tokens_seen": 270139392
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009274022066198596,
+      "loss": 1.8298,
+      "theoretical_loss": 4.226529300067345,
+      "tokens_seen": 270204928
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000927382146439318,
+      "loss": 1.8401,
+      "theoretical_loss": 4.226407813607116,
+      "tokens_seen": 270270464
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009273620862587764,
+      "loss": 1.831,
+      "theoretical_loss": 4.226286364847795,
+      "tokens_seen": 270336000
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009273420260782347,
+      "loss": 1.7314,
+      "theoretical_loss": 4.226164953768545,
+      "tokens_seen": 270401536
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009273219658976931,
+      "loss": 1.7272,
+      "theoretical_loss": 4.226043580348549,
+      "tokens_seen": 270467072
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009273019057171514,
+      "loss": 1.9179,
+      "theoretical_loss": 4.2259222445670055,
+      "tokens_seen": 270532608
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009272818455366099,
+      "loss": 1.8673,
+      "theoretical_loss": 4.225800946403127,
+      "tokens_seen": 270598144
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009272617853560682,
+      "loss": 1.6766,
+      "theoretical_loss": 4.2256796858361465,
+      "tokens_seen": 270663680
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009272417251755266,
+      "loss": 1.7084,
+      "theoretical_loss": 4.225558462845311,
+      "tokens_seen": 270729216
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000927221664994985,
+      "loss": 1.6541,
+      "theoretical_loss": 4.225437277409885,
+      "tokens_seen": 270794752
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009272016048144434,
+      "loss": 1.7796,
+      "theoretical_loss": 4.225316129509147,
+      "tokens_seen": 270860288
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009271815446339017,
+      "loss": 1.6994,
+      "theoretical_loss": 4.2251950191223955,
+      "tokens_seen": 270925824
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009271614844533602,
+      "loss": 1.8671,
+      "theoretical_loss": 4.225073946228944,
+      "tokens_seen": 270991360
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009271414242728184,
+      "loss": 1.698,
+      "theoretical_loss": 4.224952910808122,
+      "tokens_seen": 271056896
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009271213640922769,
+      "loss": 1.7853,
+      "theoretical_loss": 4.224831912839275,
+      "tokens_seen": 271122432
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009271013039117352,
+      "loss": 1.8293,
+      "theoretical_loss": 4.224710952301766,
+      "tokens_seen": 271187968
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009270812437311936,
+      "loss": 1.8736,
+      "theoretical_loss": 4.2245900291749745,
+      "tokens_seen": 271253504
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000927061183550652,
+      "loss": 1.7211,
+      "theoretical_loss": 4.224469143438294,
+      "tokens_seen": 271319040
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009270411233701103,
+      "loss": 1.9201,
+      "theoretical_loss": 4.224348295071138,
+      "tokens_seen": 271384576
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009270210631895687,
+      "loss": 1.8416,
+      "theoretical_loss": 4.224227484052934,
+      "tokens_seen": 271450112
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009270010030090272,
+      "loss": 1.7762,
+      "theoretical_loss": 4.224106710363124,
+      "tokens_seen": 271515648
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009269809428284854,
+      "loss": 1.9026,
+      "theoretical_loss": 4.223985973981171,
+      "tokens_seen": 271581184
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009269608826479439,
+      "loss": 1.9119,
+      "theoretical_loss": 4.223865274886551,
+      "tokens_seen": 271646720
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009269408224674021,
+      "loss": 1.8214,
+      "theoretical_loss": 4.223744613058755,
+      "tokens_seen": 271712256
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009269207622868606,
+      "loss": 1.7376,
+      "theoretical_loss": 4.223623988477295,
+      "tokens_seen": 271777792
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000926900702106319,
+      "loss": 1.9123,
+      "theoretical_loss": 4.223503401121693,
+      "tokens_seen": 271843328
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009268806419257773,
+      "loss": 1.741,
+      "theoretical_loss": 4.223382850971494,
+      "tokens_seen": 271908864
+    },
+    {
+      "epoch": 0.08,
+      "objective/train/advantage_avg": 0.4953818619251251,
+      "objective/train/docs_used": 105443,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.6509478092193604,
+      "objective/train/original_loss": 1.6509475708007812,
+      "objective/train/theoretical_loss": 4.223262338006254,
+      "objective/train/tokens_used": 292434400,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24556824564933777,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0507866144180298,
+      "objective/train/weighted_lm_loss": 1.7348129749298096,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9753099083900452,
+      "theoretical_loss": 4.223262338006254,
+      "tokens_seen": 271974400
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009268605817452357,
+      "loss": 1.7963,
+      "theoretical_loss": 4.223262338006254,
+      "tokens_seen": 271974400
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009268405215646942,
+      "loss": 1.6685,
+      "theoretical_loss": 4.223141862205546,
+      "tokens_seen": 272039936
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009268204613841524,
+      "loss": 1.7575,
+      "theoretical_loss": 4.223021423548962,
+      "tokens_seen": 272105472
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009268004012036109,
+      "loss": 1.7426,
+      "theoretical_loss": 4.2229010220161065,
+      "tokens_seen": 272171008
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009267803410230691,
+      "loss": 1.7231,
+      "theoretical_loss": 4.222780657586601,
+      "tokens_seen": 272236544
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009267602808425276,
+      "loss": 1.7428,
+      "theoretical_loss": 4.222660330240086,
+      "tokens_seen": 272302080
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000926740220661986,
+      "loss": 1.9007,
+      "theoretical_loss": 4.222540039956215,
+      "tokens_seen": 272367616
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009267201604814443,
+      "loss": 1.795,
+      "theoretical_loss": 4.222419786714657,
+      "tokens_seen": 272433152
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009267001003009027,
+      "loss": 1.768,
+      "theoretical_loss": 4.2222995704951005,
+      "tokens_seen": 272498688
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009266800401203611,
+      "loss": 1.7197,
+      "theoretical_loss": 4.222179391277247,
+      "tokens_seen": 272564224
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009266599799398194,
+      "loss": 1.76,
+      "theoretical_loss": 4.222059249040814,
+      "tokens_seen": 272629760
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009266399197592779,
+      "loss": 1.7811,
+      "theoretical_loss": 4.221939143765538,
+      "tokens_seen": 272695296
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009266198595787362,
+      "loss": 1.842,
+      "theoretical_loss": 4.221819075431168,
+      "tokens_seen": 272760832
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009265997993981946,
+      "loss": 1.7936,
+      "theoretical_loss": 4.221699044017471,
+      "tokens_seen": 272826368
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009265797392176529,
+      "loss": 1.904,
+      "theoretical_loss": 4.2215790495042285,
+      "tokens_seen": 272891904
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009265596790371113,
+      "loss": 1.8888,
+      "theoretical_loss": 4.22145909187124,
+      "tokens_seen": 272957440
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009265396188565697,
+      "loss": 1.8071,
+      "theoretical_loss": 4.221339171098318,
+      "tokens_seen": 273022976
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009265195586760281,
+      "loss": 1.6943,
+      "theoretical_loss": 4.221219287165294,
+      "tokens_seen": 273088512
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009264994984954864,
+      "loss": 1.9064,
+      "theoretical_loss": 4.221099440052014,
+      "tokens_seen": 273154048
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009264794383149449,
+      "loss": 1.8892,
+      "theoretical_loss": 4.220979629738339,
+      "tokens_seen": 273219584
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009264593781344032,
+      "loss": 1.7186,
+      "theoretical_loss": 4.220859856204147,
+      "tokens_seen": 273285120
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009264393179538616,
+      "loss": 1.7486,
+      "theoretical_loss": 4.22074011942933,
+      "tokens_seen": 273350656
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00092641925777332,
+      "loss": 1.7085,
+      "theoretical_loss": 4.220620419393799,
+      "tokens_seen": 273416192
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009263991975927784,
+      "loss": 1.7723,
+      "theoretical_loss": 4.220500756077479,
+      "tokens_seen": 273481728
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009263791374122368,
+      "loss": 1.8846,
+      "theoretical_loss": 4.22038112946031,
+      "tokens_seen": 273547264
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009263590772316951,
+      "loss": 1.875,
+      "theoretical_loss": 4.220261539522249,
+      "tokens_seen": 273612800
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009263390170511535,
+      "loss": 1.7537,
+      "theoretical_loss": 4.220141986243268,
+      "tokens_seen": 273678336
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009263189568706118,
+      "loss": 1.7963,
+      "theoretical_loss": 4.220022469603356,
+      "tokens_seen": 273743872
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009262988966900703,
+      "loss": 1.7591,
+      "theoretical_loss": 4.219902989582516,
+      "tokens_seen": 273809408
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009262788365095286,
+      "loss": 1.8612,
+      "theoretical_loss": 4.219783546160768,
+      "tokens_seen": 273874944
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000926258776328987,
+      "loss": 1.8849,
+      "theoretical_loss": 4.219664139318145,
+      "tokens_seen": 273940480
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009262387161484454,
+      "loss": 1.7991,
+      "theoretical_loss": 4.219544769034701,
+      "tokens_seen": 274006016
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009262186559679038,
+      "loss": 1.7852,
+      "theoretical_loss": 4.2194254352905,
+      "tokens_seen": 274071552
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009261985957873621,
+      "loss": 1.7718,
+      "theoretical_loss": 4.219306138065626,
+      "tokens_seen": 274137088
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009261785356068206,
+      "loss": 1.8751,
+      "theoretical_loss": 4.219186877340174,
+      "tokens_seen": 274202624
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009261584754262788,
+      "loss": 1.7946,
+      "theoretical_loss": 4.219067653094261,
+      "tokens_seen": 274268160
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009261384152457373,
+      "loss": 1.6745,
+      "theoretical_loss": 4.218948465308013,
+      "tokens_seen": 274333696
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009261183550651956,
+      "loss": 1.7044,
+      "theoretical_loss": 4.218829313961575,
+      "tokens_seen": 274399232
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000926098294884654,
+      "loss": 1.8374,
+      "theoretical_loss": 4.218710199035108,
+      "tokens_seen": 274464768
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009260782347041124,
+      "loss": 1.6713,
+      "theoretical_loss": 4.218591120508787,
+      "tokens_seen": 274530304
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009260581745235707,
+      "loss": 1.7884,
+      "theoretical_loss": 4.218472078362803,
+      "tokens_seen": 274595840
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009260381143430291,
+      "loss": 1.7148,
+      "theoretical_loss": 4.2183530725773615,
+      "tokens_seen": 274661376
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009260180541624876,
+      "loss": 1.8029,
+      "theoretical_loss": 4.218234103132686,
+      "tokens_seen": 274726912
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009259979939819458,
+      "loss": 1.9025,
+      "theoretical_loss": 4.218115170009014,
+      "tokens_seen": 274792448
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009259779338014043,
+      "loss": 1.7295,
+      "theoretical_loss": 4.217996273186598,
+      "tokens_seen": 274857984
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009259578736208625,
+      "loss": 1.7892,
+      "theoretical_loss": 4.217877412645706,
+      "tokens_seen": 274923520
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000925937813440321,
+      "loss": 1.8208,
+      "theoretical_loss": 4.217758588366623,
+      "tokens_seen": 274989056
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009259177532597794,
+      "loss": 1.8495,
+      "theoretical_loss": 4.217639800329647,
+      "tokens_seen": 275054592
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009258976930792377,
+      "loss": 1.6701,
+      "theoretical_loss": 4.217521048515094,
+      "tokens_seen": 275120128
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009258776328986961,
+      "loss": 1.7104,
+      "theoretical_loss": 4.217402332903292,
+      "tokens_seen": 275185664
+    },
+    {
+      "epoch": 0.08,
+      "objective/train/advantage_avg": 0.4921969771385193,
+      "objective/train/docs_used": 106703,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.6392239332199097,
+      "objective/train/original_loss": 1.6392236948013306,
+      "objective/train/theoretical_loss": 4.217283653474588,
+      "objective/train/tokens_used": 295711200,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.243023082613945,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050455093383789,
+      "objective/train/weighted_lm_loss": 1.7219140529632568,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.217283653474588,
+      "tokens_seen": 275251200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009258575727181546,
+      "loss": 1.7057,
+      "theoretical_loss": 4.217283653474588,
+      "tokens_seen": 275251200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009258375125376128,
+      "loss": 1.752,
+      "theoretical_loss": 4.217165010209343,
+      "tokens_seen": 275316736
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009258174523570713,
+      "loss": 1.8469,
+      "theoretical_loss": 4.217046403087933,
+      "tokens_seen": 275382272
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009257973921765295,
+      "loss": 1.7028,
+      "theoretical_loss": 4.2169278320907475,
+      "tokens_seen": 275447808
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000925777331995988,
+      "loss": 1.86,
+      "theoretical_loss": 4.216809297198195,
+      "tokens_seen": 275513344
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009257572718154464,
+      "loss": 1.8148,
+      "theoretical_loss": 4.216690798390697,
+      "tokens_seen": 275578880
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009257372116349047,
+      "loss": 1.7633,
+      "theoretical_loss": 4.216572335648692,
+      "tokens_seen": 275644416
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009257171514543631,
+      "loss": 1.8091,
+      "theoretical_loss": 4.21645390895263,
+      "tokens_seen": 275709952
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009256970912738215,
+      "loss": 1.8062,
+      "theoretical_loss": 4.21633551828298,
+      "tokens_seen": 275775488
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009256770310932798,
+      "loss": 1.6688,
+      "theoretical_loss": 4.216217163620225,
+      "tokens_seen": 275841024
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009256569709127383,
+      "loss": 1.8757,
+      "theoretical_loss": 4.216098844944863,
+      "tokens_seen": 275906560
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009256369107321966,
+      "loss": 1.8344,
+      "theoretical_loss": 4.215980562237409,
+      "tokens_seen": 275972096
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000925616850551655,
+      "loss": 1.904,
+      "theoretical_loss": 4.215862315478388,
+      "tokens_seen": 276037632
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009255967903711133,
+      "loss": 1.8534,
+      "theoretical_loss": 4.215744104648347,
+      "tokens_seen": 276103168
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009255767301905717,
+      "loss": 1.7778,
+      "theoretical_loss": 4.2156259297278424,
+      "tokens_seen": 276168704
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009255566700100301,
+      "loss": 1.9598,
+      "theoretical_loss": 4.215507790697449,
+      "tokens_seen": 276234240
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009255366098294885,
+      "loss": 1.7881,
+      "theoretical_loss": 4.2153896875377574,
+      "tokens_seen": 276299776
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009255165496489468,
+      "loss": 1.8789,
+      "theoretical_loss": 4.21527162022937,
+      "tokens_seen": 276365312
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009254964894684053,
+      "loss": 1.8463,
+      "theoretical_loss": 4.215153588752907,
+      "tokens_seen": 276430848
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009254764292878636,
+      "loss": 1.9038,
+      "theoretical_loss": 4.215035593089002,
+      "tokens_seen": 276496384
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000925456369107322,
+      "loss": 1.6358,
+      "theoretical_loss": 4.214917633218304,
+      "tokens_seen": 276561920
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009254363089267803,
+      "loss": 1.8301,
+      "theoretical_loss": 4.214799709121479,
+      "tokens_seen": 276627456
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009254162487462388,
+      "loss": 1.8829,
+      "theoretical_loss": 4.214681820779205,
+      "tokens_seen": 276692992
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009253961885656971,
+      "loss": 1.7861,
+      "theoretical_loss": 4.214563968172178,
+      "tokens_seen": 276758528
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009253761283851555,
+      "loss": 1.7876,
+      "theoretical_loss": 4.214446151281106,
+      "tokens_seen": 276824064
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009253560682046138,
+      "loss": 1.8599,
+      "theoretical_loss": 4.214328370086713,
+      "tokens_seen": 276889600
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009253360080240722,
+      "loss": 1.6949,
+      "theoretical_loss": 4.21421062456974,
+      "tokens_seen": 276955136
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009253159478435306,
+      "loss": 1.7641,
+      "theoretical_loss": 4.214092914710941,
+      "tokens_seen": 277020672
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000925295887662989,
+      "loss": 1.7158,
+      "theoretical_loss": 4.213975240491084,
+      "tokens_seen": 277086208
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009252758274824473,
+      "loss": 1.8548,
+      "theoretical_loss": 4.213857601890955,
+      "tokens_seen": 277151744
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009252557673019058,
+      "loss": 1.8456,
+      "theoretical_loss": 4.213739998891352,
+      "tokens_seen": 277217280
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000925235707121364,
+      "loss": 1.7495,
+      "theoretical_loss": 4.213622431473088,
+      "tokens_seen": 277282816
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009252156469408225,
+      "loss": 1.7643,
+      "theoretical_loss": 4.213504899616995,
+      "tokens_seen": 277348352
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009251955867602808,
+      "loss": 1.7701,
+      "theoretical_loss": 4.213387403303912,
+      "tokens_seen": 277413888
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009251755265797392,
+      "loss": 1.7438,
+      "theoretical_loss": 4.213269942514701,
+      "tokens_seen": 277479424
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009251554663991976,
+      "loss": 1.8916,
+      "theoretical_loss": 4.2131525172302355,
+      "tokens_seen": 277544960
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000925135406218656,
+      "loss": 1.8483,
+      "theoretical_loss": 4.213035127431402,
+      "tokens_seen": 277610496
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009251153460381143,
+      "loss": 1.7789,
+      "theoretical_loss": 4.212917773099105,
+      "tokens_seen": 277676032
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009250952858575728,
+      "loss": 1.7993,
+      "theoretical_loss": 4.2128004542142605,
+      "tokens_seen": 277741568
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000925075225677031,
+      "loss": 1.8079,
+      "theoretical_loss": 4.212683170757802,
+      "tokens_seen": 277807104
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009250551654964895,
+      "loss": 1.8363,
+      "theoretical_loss": 4.212565922710677,
+      "tokens_seen": 277872640
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009250351053159479,
+      "loss": 1.8011,
+      "theoretical_loss": 4.212448710053847,
+      "tokens_seen": 277938176
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009250150451354062,
+      "loss": 1.8875,
+      "theoretical_loss": 4.212331532768289,
+      "tokens_seen": 278003712
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009249949849548646,
+      "loss": 1.7518,
+      "theoretical_loss": 4.212214390834994,
+      "tokens_seen": 278069248
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009249749247743229,
+      "loss": 1.7445,
+      "theoretical_loss": 4.21209728423497,
+      "tokens_seen": 278134784
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009249548645937813,
+      "loss": 1.9204,
+      "theoretical_loss": 4.211980212949234,
+      "tokens_seen": 278200320
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009249348044132398,
+      "loss": 1.7544,
+      "theoretical_loss": 4.2118631769588255,
+      "tokens_seen": 278265856
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009249147442326981,
+      "loss": 1.8568,
+      "theoretical_loss": 4.211746176244792,
+      "tokens_seen": 278331392
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009248946840521565,
+      "loss": 1.8516,
+      "theoretical_loss": 4.2116292107882,
+      "tokens_seen": 278396928
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000924874623871615,
+      "loss": 1.7939,
+      "theoretical_loss": 4.211512280570127,
+      "tokens_seen": 278462464
+    },
+    {
+      "epoch": 0.08,
+      "objective/train/advantage_avg": 0.4929666519165039,
+      "objective/train/docs_used": 107929,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.715563178062439,
+      "objective/train/original_loss": 1.7155632972717285,
+      "objective/train/theoretical_loss": 4.211395385571668,
+      "objective/train/tokens_used": 298988000,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2441163808107376,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505377054214478,
+      "objective/train/weighted_lm_loss": 1.8023045063018799,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.211395385571668,
+      "tokens_seen": 278528000
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009248545636910732,
+      "loss": 1.8125,
+      "theoretical_loss": 4.211395385571668,
+      "tokens_seen": 278528000
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009248345035105317,
+      "loss": 1.8219,
+      "theoretical_loss": 4.211278525773933,
+      "tokens_seen": 278593536
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00092481444332999,
+      "loss": 1.838,
+      "theoretical_loss": 4.211161701158042,
+      "tokens_seen": 278659072
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009247943831494484,
+      "loss": 1.8463,
+      "theoretical_loss": 4.211044911705135,
+      "tokens_seen": 278724608
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009247743229689068,
+      "loss": 1.8635,
+      "theoretical_loss": 4.2109281573963635,
+      "tokens_seen": 278790144
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009247542627883651,
+      "loss": 1.8983,
+      "theoretical_loss": 4.210811438212894,
+      "tokens_seen": 278855680
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009247342026078235,
+      "loss": 1.739,
+      "theoretical_loss": 4.2106947541359085,
+      "tokens_seen": 278921216
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009247141424272819,
+      "loss": 1.9169,
+      "theoretical_loss": 4.2105781051466025,
+      "tokens_seen": 278986752
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009246940822467402,
+      "loss": 1.7906,
+      "theoretical_loss": 4.2104614912261855,
+      "tokens_seen": 279052288
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009246740220661987,
+      "loss": 1.6945,
+      "theoretical_loss": 4.210344912355883,
+      "tokens_seen": 279117824
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000924653961885657,
+      "loss": 1.7868,
+      "theoretical_loss": 4.210228368516935,
+      "tokens_seen": 279183360
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009246339017051154,
+      "loss": 1.8477,
+      "theoretical_loss": 4.210111859690594,
+      "tokens_seen": 279248896
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009246138415245737,
+      "loss": 1.9463,
+      "theoretical_loss": 4.209995385858129,
+      "tokens_seen": 279314432
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009245937813440321,
+      "loss": 1.9702,
+      "theoretical_loss": 4.20987894700082,
+      "tokens_seen": 279379968
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009245737211634905,
+      "loss": 1.8656,
+      "theoretical_loss": 4.209762543099966,
+      "tokens_seen": 279445504
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009245536609829489,
+      "loss": 1.821,
+      "theoretical_loss": 4.209646174136879,
+      "tokens_seen": 279511040
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009245336008024072,
+      "loss": 1.8142,
+      "theoretical_loss": 4.209529840092884,
+      "tokens_seen": 279576576
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009245135406218657,
+      "loss": 1.911,
+      "theoretical_loss": 4.209413540949319,
+      "tokens_seen": 279642112
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000924493480441324,
+      "loss": 1.7045,
+      "theoretical_loss": 4.209297276687541,
+      "tokens_seen": 279707648
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009244734202607824,
+      "loss": 1.8239,
+      "theoretical_loss": 4.209181047288917,
+      "tokens_seen": 279773184
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009244533600802407,
+      "loss": 1.7827,
+      "theoretical_loss": 4.209064852734831,
+      "tokens_seen": 279838720
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009244332998996992,
+      "loss": 1.962,
+      "theoretical_loss": 4.20894869300668,
+      "tokens_seen": 279904256
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009244132397191575,
+      "loss": 1.8998,
+      "theoretical_loss": 4.2088325680858745,
+      "tokens_seen": 279969792
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009243931795386159,
+      "loss": 1.7249,
+      "theoretical_loss": 4.208716477953843,
+      "tokens_seen": 280035328
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009243731193580742,
+      "loss": 1.8494,
+      "theoretical_loss": 4.2086004225920215,
+      "tokens_seen": 280100864
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009243530591775326,
+      "loss": 1.8086,
+      "theoretical_loss": 4.208484401981868,
+      "tokens_seen": 280166400
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000924332998996991,
+      "loss": 1.7255,
+      "theoretical_loss": 4.208368416104849,
+      "tokens_seen": 280231936
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009243129388164494,
+      "loss": 1.7742,
+      "theoretical_loss": 4.208252464942449,
+      "tokens_seen": 280297472
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009242928786359077,
+      "loss": 1.8617,
+      "theoretical_loss": 4.208136548476163,
+      "tokens_seen": 280363008
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009242728184553662,
+      "loss": 1.8351,
+      "theoretical_loss": 4.208020666687503,
+      "tokens_seen": 280428544
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0009242527582748244,
+      "loss": 1.9316,
+      "theoretical_loss": 4.207904819557995,
+      "tokens_seen": 280494080
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009242326980942829,
+      "loss": 1.7367,
+      "theoretical_loss": 4.207789007069177,
+      "tokens_seen": 280559616
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009242126379137412,
+      "loss": 1.8257,
+      "theoretical_loss": 4.207673229202604,
+      "tokens_seen": 280625152
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009241925777331996,
+      "loss": 1.7784,
+      "theoretical_loss": 4.2075574859398435,
+      "tokens_seen": 280690688
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000924172517552658,
+      "loss": 1.8542,
+      "theoretical_loss": 4.207441777262477,
+      "tokens_seen": 280756224
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009241524573721164,
+      "loss": 1.703,
+      "theoretical_loss": 4.2073261031521,
+      "tokens_seen": 280821760
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009241323971915747,
+      "loss": 1.7389,
+      "theoretical_loss": 4.207210463590323,
+      "tokens_seen": 280887296
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009241123370110332,
+      "loss": 1.9065,
+      "theoretical_loss": 4.207094858558771,
+      "tokens_seen": 280952832
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009240922768304914,
+      "loss": 1.774,
+      "theoretical_loss": 4.206979288039081,
+      "tokens_seen": 281018368
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009240722166499499,
+      "loss": 1.7895,
+      "theoretical_loss": 4.2068637520129055,
+      "tokens_seen": 281083904
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009240521564694083,
+      "loss": 1.7552,
+      "theoretical_loss": 4.206748250461911,
+      "tokens_seen": 281149440
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009240320962888666,
+      "loss": 1.771,
+      "theoretical_loss": 4.206632783367777,
+      "tokens_seen": 281214976
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000924012036108325,
+      "loss": 1.7158,
+      "theoretical_loss": 4.206517350712199,
+      "tokens_seen": 281280512
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009239919759277833,
+      "loss": 1.9191,
+      "theoretical_loss": 4.2064019524768845,
+      "tokens_seen": 281346048
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009239719157472417,
+      "loss": 1.8433,
+      "theoretical_loss": 4.206286588643556,
+      "tokens_seen": 281411584
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009239518555667002,
+      "loss": 1.7216,
+      "theoretical_loss": 4.206171259193948,
+      "tokens_seen": 281477120
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009239317953861584,
+      "loss": 1.7115,
+      "theoretical_loss": 4.206055964109813,
+      "tokens_seen": 281542656
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009239117352056169,
+      "loss": 1.705,
+      "theoretical_loss": 4.205940703372915,
+      "tokens_seen": 281608192
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009238916750250753,
+      "loss": 1.7713,
+      "theoretical_loss": 4.20582547696503,
+      "tokens_seen": 281673728
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009238716148445336,
+      "loss": 1.8652,
+      "theoretical_loss": 4.205710284867951,
+      "tokens_seen": 281739264
+    },
+    {
+      "epoch": 0.09,
+      "objective/train/advantage_avg": 0.49427810311317444,
+      "objective/train/docs_used": 109222,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.6125668287277222,
+      "objective/train/original_loss": 1.6125669479370117,
+      "objective/train/theoretical_loss": 4.205595127063485,
+      "objective/train/tokens_used": 302264800,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24455593526363373,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050671100616455,
+      "objective/train/weighted_lm_loss": 1.694186806678772,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.205595127063485,
+      "tokens_seen": 281804800
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000923851554663992,
+      "loss": 1.7848,
+      "theoretical_loss": 4.205595127063485,
+      "tokens_seen": 281804800
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009238314944834503,
+      "loss": 1.7997,
+      "theoretical_loss": 4.20548000353345,
+      "tokens_seen": 281870336
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009238114343029087,
+      "loss": 1.7206,
+      "theoretical_loss": 4.205364914259679,
+      "tokens_seen": 281935872
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009237913741223672,
+      "loss": 1.7657,
+      "theoretical_loss": 4.205249859224021,
+      "tokens_seen": 282001408
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009237713139418254,
+      "loss": 1.7049,
+      "theoretical_loss": 4.205134838408337,
+      "tokens_seen": 282066944
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009237512537612839,
+      "loss": 1.7584,
+      "theoretical_loss": 4.2050198517945,
+      "tokens_seen": 282132480
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009237311935807422,
+      "loss": 1.7654,
+      "theoretical_loss": 4.204904899364402,
+      "tokens_seen": 282198016
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009237111334002006,
+      "loss": 1.7786,
+      "theoretical_loss": 4.2047899810999425,
+      "tokens_seen": 282263552
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000923691073219659,
+      "loss": 1.8459,
+      "theoretical_loss": 4.20467509698304,
+      "tokens_seen": 282329088
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009236710130391174,
+      "loss": 1.7859,
+      "theoretical_loss": 4.204560246995625,
+      "tokens_seen": 282394624
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009236509528585757,
+      "loss": 1.7339,
+      "theoretical_loss": 4.204445431119639,
+      "tokens_seen": 282460160
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009236308926780341,
+      "loss": 1.7904,
+      "theoretical_loss": 4.204330649337042,
+      "tokens_seen": 282525696
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009236108324974924,
+      "loss": 1.7929,
+      "theoretical_loss": 4.204215901629803,
+      "tokens_seen": 282591232
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009235907723169509,
+      "loss": 1.8564,
+      "theoretical_loss": 4.2041011879799095,
+      "tokens_seen": 282656768
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009235707121364092,
+      "loss": 1.7214,
+      "theoretical_loss": 4.2039865083693595,
+      "tokens_seen": 282722304
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009235506519558676,
+      "loss": 1.8249,
+      "theoretical_loss": 4.203871862780165,
+      "tokens_seen": 282787840
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000923530591775326,
+      "loss": 1.7751,
+      "theoretical_loss": 4.203757251194353,
+      "tokens_seen": 282853376
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009235105315947844,
+      "loss": 1.7531,
+      "theoretical_loss": 4.203642673593962,
+      "tokens_seen": 282918912
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009234904714142427,
+      "loss": 1.7935,
+      "theoretical_loss": 4.203528129961047,
+      "tokens_seen": 282984448
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009234704112337011,
+      "loss": 1.7967,
+      "theoretical_loss": 4.203413620277673,
+      "tokens_seen": 283049984
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009234503510531594,
+      "loss": 1.7649,
+      "theoretical_loss": 4.203299144525923,
+      "tokens_seen": 283115520
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009234302908726179,
+      "loss": 1.8882,
+      "theoretical_loss": 4.20318470268789,
+      "tokens_seen": 283181056
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009234102306920763,
+      "loss": 1.7575,
+      "theoretical_loss": 4.203070294745682,
+      "tokens_seen": 283246592
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009233901705115346,
+      "loss": 1.7621,
+      "theoretical_loss": 4.202955920681421,
+      "tokens_seen": 283312128
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000923370110330993,
+      "loss": 1.7461,
+      "theoretical_loss": 4.202841580477241,
+      "tokens_seen": 283377664
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009233500501504514,
+      "loss": 1.9422,
+      "theoretical_loss": 4.202727274115292,
+      "tokens_seen": 283443200
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009233299899699098,
+      "loss": 1.798,
+      "theoretical_loss": 4.202613001577735,
+      "tokens_seen": 283508736
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009233099297893681,
+      "loss": 1.8495,
+      "theoretical_loss": 4.202498762846745,
+      "tokens_seen": 283574272
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009232898696088266,
+      "loss": 1.7852,
+      "theoretical_loss": 4.202384557904513,
+      "tokens_seen": 283639808
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009232698094282848,
+      "loss": 1.7845,
+      "theoretical_loss": 4.202270386733241,
+      "tokens_seen": 283705344
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009232497492477433,
+      "loss": 1.7888,
+      "theoretical_loss": 4.202156249315145,
+      "tokens_seen": 283770880
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009232296890672016,
+      "loss": 1.8123,
+      "theoretical_loss": 4.202042145632453,
+      "tokens_seen": 283836416
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00092320962888666,
+      "loss": 1.7403,
+      "theoretical_loss": 4.201928075667411,
+      "tokens_seen": 283901952
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009231895687061184,
+      "loss": 1.735,
+      "theoretical_loss": 4.201814039402273,
+      "tokens_seen": 283967488
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009231695085255768,
+      "loss": 1.7603,
+      "theoretical_loss": 4.201700036819311,
+      "tokens_seen": 284033024
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009231494483450351,
+      "loss": 1.9366,
+      "theoretical_loss": 4.201586067900806,
+      "tokens_seen": 284098560
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009231293881644936,
+      "loss": 1.6707,
+      "theoretical_loss": 4.201472132629057,
+      "tokens_seen": 284164096
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009231093279839518,
+      "loss": 1.9417,
+      "theoretical_loss": 4.201358230986372,
+      "tokens_seen": 284229632
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009230892678034103,
+      "loss": 1.82,
+      "theoretical_loss": 4.201244362955077,
+      "tokens_seen": 284295168
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009230692076228687,
+      "loss": 1.8958,
+      "theoretical_loss": 4.201130528517506,
+      "tokens_seen": 284360704
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000923049147442327,
+      "loss": 1.8833,
+      "theoretical_loss": 4.201016727656012,
+      "tokens_seen": 284426240
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009230290872617854,
+      "loss": 1.784,
+      "theoretical_loss": 4.200902960352956,
+      "tokens_seen": 284491776
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009230090270812437,
+      "loss": 1.8496,
+      "theoretical_loss": 4.200789226590717,
+      "tokens_seen": 284557312
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009229889669007021,
+      "loss": 1.7737,
+      "theoretical_loss": 4.200675526351684,
+      "tokens_seen": 284622848
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009229689067201606,
+      "loss": 1.893,
+      "theoretical_loss": 4.2005618596182615,
+      "tokens_seen": 284688384
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009229488465396188,
+      "loss": 1.8352,
+      "theoretical_loss": 4.200448226372865,
+      "tokens_seen": 284753920
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009229287863590773,
+      "loss": 1.8389,
+      "theoretical_loss": 4.200334626597925,
+      "tokens_seen": 284819456
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009229087261785357,
+      "loss": 1.8186,
+      "theoretical_loss": 4.200221060275885,
+      "tokens_seen": 284884992
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000922888665997994,
+      "loss": 1.8161,
+      "theoretical_loss": 4.200107527389202,
+      "tokens_seen": 284950528
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009228686058174524,
+      "loss": 1.7495,
+      "theoretical_loss": 4.1999940279203445,
+      "tokens_seen": 285016064
+    },
+    {
+      "epoch": 0.09,
+      "objective/train/advantage_avg": 0.4956050515174866,
+      "objective/train/docs_used": 110428,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.646780252456665,
+      "objective/train/original_loss": 1.646780252456665,
+      "objective/train/theoretical_loss": 4.1998805618517965,
+      "objective/train/tokens_used": 305541600,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24578233063220978,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0508100986480713,
+      "objective/train/weighted_lm_loss": 1.730436086654663,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.1998805618517965,
+      "tokens_seen": 285081600
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009228485456369107,
+      "loss": 1.7924,
+      "theoretical_loss": 4.1998805618517965,
+      "tokens_seen": 285081600
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009228284854563691,
+      "loss": 1.7654,
+      "theoretical_loss": 4.199767129166054,
+      "tokens_seen": 285147136
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009228084252758276,
+      "loss": 1.871,
+      "theoretical_loss": 4.199653729845626,
+      "tokens_seen": 285212672
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009227883650952858,
+      "loss": 1.8636,
+      "theoretical_loss": 4.199540363873036,
+      "tokens_seen": 285278208
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009227683049147443,
+      "loss": 1.7923,
+      "theoretical_loss": 4.19942703123082,
+      "tokens_seen": 285343744
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009227482447342026,
+      "loss": 1.7764,
+      "theoretical_loss": 4.199313731901524,
+      "tokens_seen": 285409280
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000922728184553661,
+      "loss": 1.791,
+      "theoretical_loss": 4.199200465867714,
+      "tokens_seen": 285474816
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009227081243731194,
+      "loss": 1.8081,
+      "theoretical_loss": 4.199087233111962,
+      "tokens_seen": 285540352
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009226880641925778,
+      "loss": 1.6973,
+      "theoretical_loss": 4.198974033616859,
+      "tokens_seen": 285605888
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009226680040120361,
+      "loss": 1.8117,
+      "theoretical_loss": 4.1988608673650045,
+      "tokens_seen": 285671424
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009226479438314945,
+      "loss": 1.8261,
+      "theoretical_loss": 4.198747734339013,
+      "tokens_seen": 285736960
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009226278836509528,
+      "loss": 1.8351,
+      "theoretical_loss": 4.198634634521513,
+      "tokens_seen": 285802496
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009226078234704113,
+      "loss": 1.7645,
+      "theoretical_loss": 4.198521567895144,
+      "tokens_seen": 285868032
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009225877632898696,
+      "loss": 1.7425,
+      "theoretical_loss": 4.198408534442561,
+      "tokens_seen": 285933568
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000922567703109328,
+      "loss": 1.7141,
+      "theoretical_loss": 4.198295534146429,
+      "tokens_seen": 285999104
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009225476429287864,
+      "loss": 1.7746,
+      "theoretical_loss": 4.19818256698943,
+      "tokens_seen": 286064640
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009225275827482448,
+      "loss": 1.7737,
+      "theoretical_loss": 4.198069632954255,
+      "tokens_seen": 286130176
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009225075225677031,
+      "loss": 1.7182,
+      "theoretical_loss": 4.19795673202361,
+      "tokens_seen": 286195712
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009224874623871615,
+      "loss": 1.8036,
+      "theoretical_loss": 4.197843864180214,
+      "tokens_seen": 286261248
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009224674022066198,
+      "loss": 1.8232,
+      "theoretical_loss": 4.1977310294068,
+      "tokens_seen": 286326784
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009224473420260783,
+      "loss": 1.7982,
+      "theoretical_loss": 4.19761822768611,
+      "tokens_seen": 286392320
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009224272818455366,
+      "loss": 1.735,
+      "theoretical_loss": 4.1975054590009035,
+      "tokens_seen": 286457856
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000922407221664995,
+      "loss": 1.7124,
+      "theoretical_loss": 4.197392723333951,
+      "tokens_seen": 286523392
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009223871614844533,
+      "loss": 1.8675,
+      "theoretical_loss": 4.1972800206680345,
+      "tokens_seen": 286588928
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009223671013039118,
+      "loss": 1.7668,
+      "theoretical_loss": 4.197167350985952,
+      "tokens_seen": 286654464
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009223470411233701,
+      "loss": 1.812,
+      "theoretical_loss": 4.197054714270513,
+      "tokens_seen": 286720000
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009223269809428285,
+      "loss": 1.8564,
+      "theoretical_loss": 4.196942110504538,
+      "tokens_seen": 286785536
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009223069207622869,
+      "loss": 1.7916,
+      "theoretical_loss": 4.196829539670864,
+      "tokens_seen": 286851072
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009222868605817452,
+      "loss": 1.7991,
+      "theoretical_loss": 4.1967170017523365,
+      "tokens_seen": 286916608
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009222668004012036,
+      "loss": 1.7301,
+      "theoretical_loss": 4.196604496731818,
+      "tokens_seen": 286982144
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000922246740220662,
+      "loss": 1.8278,
+      "theoretical_loss": 4.196492024592183,
+      "tokens_seen": 287047680
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009222266800401203,
+      "loss": 1.891,
+      "theoretical_loss": 4.196379585316315,
+      "tokens_seen": 287113216
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009222066198595788,
+      "loss": 1.7694,
+      "theoretical_loss": 4.1962671788871155,
+      "tokens_seen": 287178752
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009221865596790371,
+      "loss": 1.7081,
+      "theoretical_loss": 4.196154805287497,
+      "tokens_seen": 287244288
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009221664994984955,
+      "loss": 1.7177,
+      "theoretical_loss": 4.196042464500382,
+      "tokens_seen": 287309824
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009221464393179539,
+      "loss": 1.7366,
+      "theoretical_loss": 4.195930156508711,
+      "tokens_seen": 287375360
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009221263791374122,
+      "loss": 1.7228,
+      "theoretical_loss": 4.195817881295431,
+      "tokens_seen": 287440896
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009221063189568706,
+      "loss": 1.8829,
+      "theoretical_loss": 4.195705638843508,
+      "tokens_seen": 287506432
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000922086258776329,
+      "loss": 1.7452,
+      "theoretical_loss": 4.195593429135916,
+      "tokens_seen": 287571968
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009220661985957873,
+      "loss": 2.0069,
+      "theoretical_loss": 4.195481252155645,
+      "tokens_seen": 287637504
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009220461384152458,
+      "loss": 1.9142,
+      "theoretical_loss": 4.1953691078856945,
+      "tokens_seen": 287703040
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000922026078234704,
+      "loss": 1.8221,
+      "theoretical_loss": 4.19525699630908,
+      "tokens_seen": 287768576
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009220060180541625,
+      "loss": 1.823,
+      "theoretical_loss": 4.195144917408828,
+      "tokens_seen": 287834112
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009219859578736209,
+      "loss": 1.7773,
+      "theoretical_loss": 4.195032871167977,
+      "tokens_seen": 287899648
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009219658976930792,
+      "loss": 1.706,
+      "theoretical_loss": 4.1949208575695796,
+      "tokens_seen": 287965184
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009219458375125376,
+      "loss": 1.7919,
+      "theoretical_loss": 4.1948088765967,
+      "tokens_seen": 288030720
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009219257773319961,
+      "loss": 1.7851,
+      "theoretical_loss": 4.194696928232417,
+      "tokens_seen": 288096256
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009219057171514544,
+      "loss": 1.7673,
+      "theoretical_loss": 4.194585012459818,
+      "tokens_seen": 288161792
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009218856569709128,
+      "loss": 1.7555,
+      "theoretical_loss": 4.194473129262008,
+      "tokens_seen": 288227328
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009218655967903711,
+      "loss": 1.7409,
+      "theoretical_loss": 4.194361278622099,
+      "tokens_seen": 288292864
+    },
+    {
+      "epoch": 0.09,
+      "objective/train/advantage_avg": 0.4943329095840454,
+      "objective/train/docs_used": 111591,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.6546730995178223,
+      "objective/train/original_loss": 1.6546730995178223,
+      "objective/train/theoretical_loss": 4.194249460523222,
+      "objective/train/tokens_used": 308818400,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2445804327726364,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050676703453064,
+      "objective/train/weighted_lm_loss": 1.7384644746780396,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.194249460523222,
+      "tokens_seen": 288358400
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009218455366098295,
+      "loss": 1.713,
+      "theoretical_loss": 4.194249460523222,
+      "tokens_seen": 288358400
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000921825476429288,
+      "loss": 1.8013,
+      "theoretical_loss": 4.194137674948516,
+      "tokens_seen": 288423936
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009218054162487462,
+      "loss": 1.8124,
+      "theoretical_loss": 4.194025921881134,
+      "tokens_seen": 288489472
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009217853560682047,
+      "loss": 1.7621,
+      "theoretical_loss": 4.19391420130424,
+      "tokens_seen": 288555008
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000921765295887663,
+      "loss": 1.8619,
+      "theoretical_loss": 4.193802513201015,
+      "tokens_seen": 288620544
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009217452357071214,
+      "loss": 1.8636,
+      "theoretical_loss": 4.193690857554646,
+      "tokens_seen": 288686080
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009217251755265798,
+      "loss": 1.9187,
+      "theoretical_loss": 4.193579234348338,
+      "tokens_seen": 288751616
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009217051153460382,
+      "loss": 1.8693,
+      "theoretical_loss": 4.193467643565306,
+      "tokens_seen": 288817152
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009216850551654965,
+      "loss": 1.8022,
+      "theoretical_loss": 4.193356085188778,
+      "tokens_seen": 288882688
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009216649949849549,
+      "loss": 1.7514,
+      "theoretical_loss": 4.193244559201995,
+      "tokens_seen": 288948224
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009216449348044132,
+      "loss": 1.8518,
+      "theoretical_loss": 4.193133065588209,
+      "tokens_seen": 289013760
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009216248746238717,
+      "loss": 2.0078,
+      "theoretical_loss": 4.193021604330686,
+      "tokens_seen": 289079296
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00092160481444333,
+      "loss": 1.8522,
+      "theoretical_loss": 4.1929101754127025,
+      "tokens_seen": 289144832
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009215847542627884,
+      "loss": 1.7833,
+      "theoretical_loss": 4.192798778817551,
+      "tokens_seen": 289210368
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009215646940822468,
+      "loss": 1.8018,
+      "theoretical_loss": 4.192687414528532,
+      "tokens_seen": 289275904
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009215446339017052,
+      "loss": 1.8112,
+      "theoretical_loss": 4.192576082528962,
+      "tokens_seen": 289341440
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009215245737211635,
+      "loss": 1.8178,
+      "theoretical_loss": 4.192464782802167,
+      "tokens_seen": 289406976
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009215045135406219,
+      "loss": 1.8044,
+      "theoretical_loss": 4.192353515331489,
+      "tokens_seen": 289472512
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009214844533600802,
+      "loss": 1.7338,
+      "theoretical_loss": 4.192242280100279,
+      "tokens_seen": 289538048
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009214643931795387,
+      "loss": 1.8753,
+      "theoretical_loss": 4.192131077091901,
+      "tokens_seen": 289603584
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000921444332998997,
+      "loss": 1.7773,
+      "theoretical_loss": 4.192019906289733,
+      "tokens_seen": 289669120
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009214242728184554,
+      "loss": 1.9292,
+      "theoretical_loss": 4.191908767677163,
+      "tokens_seen": 289734656
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009214042126379137,
+      "loss": 1.7483,
+      "theoretical_loss": 4.191797661237594,
+      "tokens_seen": 289800192
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009213841524573722,
+      "loss": 1.8835,
+      "theoretical_loss": 4.191686586954439,
+      "tokens_seen": 289865728
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009213640922768305,
+      "loss": 1.9106,
+      "theoretical_loss": 4.1915755448111245,
+      "tokens_seen": 289931264
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009213440320962889,
+      "loss": 1.8306,
+      "theoretical_loss": 4.191464534791089,
+      "tokens_seen": 289996800
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009213239719157473,
+      "loss": 1.8738,
+      "theoretical_loss": 4.191353556877782,
+      "tokens_seen": 290062336
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009213039117352056,
+      "loss": 1.7921,
+      "theoretical_loss": 4.191242611054668,
+      "tokens_seen": 290127872
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000921283851554664,
+      "loss": 1.8376,
+      "theoretical_loss": 4.191131697305222,
+      "tokens_seen": 290193408
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009212637913741224,
+      "loss": 1.8509,
+      "theoretical_loss": 4.191020815612932,
+      "tokens_seen": 290258944
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009212437311935807,
+      "loss": 1.8397,
+      "theoretical_loss": 4.190909965961296,
+      "tokens_seen": 290324480
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009212236710130392,
+      "loss": 1.7438,
+      "theoretical_loss": 4.190799148333827,
+      "tokens_seen": 290390016
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009212036108324975,
+      "loss": 1.9381,
+      "theoretical_loss": 4.1906883627140505,
+      "tokens_seen": 290455552
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009211835506519559,
+      "loss": 1.7074,
+      "theoretical_loss": 4.1905776090855,
+      "tokens_seen": 290521088
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009211634904714143,
+      "loss": 1.7709,
+      "theoretical_loss": 4.190466887431725,
+      "tokens_seen": 290586624
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009211434302908726,
+      "loss": 1.8717,
+      "theoretical_loss": 4.190356197736288,
+      "tokens_seen": 290652160
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000921123370110331,
+      "loss": 1.7057,
+      "theoretical_loss": 4.19024553998276,
+      "tokens_seen": 290717696
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009211033099297895,
+      "loss": 1.8125,
+      "theoretical_loss": 4.190134914154727,
+      "tokens_seen": 290783232
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009210832497492477,
+      "loss": 1.889,
+      "theoretical_loss": 4.190024320235784,
+      "tokens_seen": 290848768
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009210631895687062,
+      "loss": 1.8423,
+      "theoretical_loss": 4.189913758209543,
+      "tokens_seen": 290914304
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009210431293881644,
+      "loss": 1.7914,
+      "theoretical_loss": 4.189803228059623,
+      "tokens_seen": 290979840
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009210230692076229,
+      "loss": 1.9284,
+      "theoretical_loss": 4.189692729769659,
+      "tokens_seen": 291045376
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009210030090270813,
+      "loss": 1.9621,
+      "theoretical_loss": 4.189582263323296,
+      "tokens_seen": 291110912
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009209829488465396,
+      "loss": 1.7024,
+      "theoretical_loss": 4.1894718287041925,
+      "tokens_seen": 291176448
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000920962888665998,
+      "loss": 1.7627,
+      "theoretical_loss": 4.189361425896016,
+      "tokens_seen": 291241984
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009209428284854565,
+      "loss": 1.7543,
+      "theoretical_loss": 4.18925105488245,
+      "tokens_seen": 291307520
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009209227683049147,
+      "loss": 1.8712,
+      "theoretical_loss": 4.189140715647189,
+      "tokens_seen": 291373056
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009209027081243732,
+      "loss": 1.7619,
+      "theoretical_loss": 4.189030408173936,
+      "tokens_seen": 291438592
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009208826479438314,
+      "loss": 1.7172,
+      "theoretical_loss": 4.188920132446411,
+      "tokens_seen": 291504128
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009208625877632899,
+      "loss": 1.8162,
+      "theoretical_loss": 4.1888098884483425,
+      "tokens_seen": 291569664
+    },
+    {
+      "epoch": 0.09,
+      "objective/train/advantage_avg": 0.49275243282318115,
+      "objective/train/docs_used": 112844,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.677362322807312,
+      "objective/train/original_loss": 1.6773622035980225,
+      "objective/train/theoretical_loss": 4.188699676163473,
+      "objective/train/tokens_used": 312095200,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24318408966064453,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505115985870361,
+      "objective/train/weighted_lm_loss": 1.7620242834091187,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.188699676163473,
+      "tokens_seen": 291635200
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009208425275827483,
+      "loss": 1.865,
+      "theoretical_loss": 4.188699676163473,
+      "tokens_seen": 291635200
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009208224674022066,
+      "loss": 1.8001,
+      "theoretical_loss": 4.188589495575557,
+      "tokens_seen": 291700736
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000920802407221665,
+      "loss": 1.7084,
+      "theoretical_loss": 4.188479346668359,
+      "tokens_seen": 291766272
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009207823470411234,
+      "loss": 1.6633,
+      "theoretical_loss": 4.188369229425657,
+      "tokens_seen": 291831808
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009207622868605817,
+      "loss": 1.9012,
+      "theoretical_loss": 4.188259143831241,
+      "tokens_seen": 291897344
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009207422266800402,
+      "loss": 1.8598,
+      "theoretical_loss": 4.188149089868912,
+      "tokens_seen": 291962880
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009207221664994984,
+      "loss": 1.7914,
+      "theoretical_loss": 4.188039067522484,
+      "tokens_seen": 292028416
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009207021063189569,
+      "loss": 1.818,
+      "theoretical_loss": 4.1879290767757835,
+      "tokens_seen": 292093952
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009206820461384152,
+      "loss": 1.8285,
+      "theoretical_loss": 4.187819117612645,
+      "tokens_seen": 292159488
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009206619859578736,
+      "loss": 1.7891,
+      "theoretical_loss": 4.18770919001692,
+      "tokens_seen": 292225024
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000920641925777332,
+      "loss": 1.8172,
+      "theoretical_loss": 4.18759929397247,
+      "tokens_seen": 292290560
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009206218655967904,
+      "loss": 1.8366,
+      "theoretical_loss": 4.187489429463166,
+      "tokens_seen": 292356096
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009206018054162487,
+      "loss": 1.87,
+      "theoretical_loss": 4.187379596472894,
+      "tokens_seen": 292421632
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009205817452357072,
+      "loss": 1.8121,
+      "theoretical_loss": 4.18726979498555,
+      "tokens_seen": 292487168
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009205616850551655,
+      "loss": 1.8013,
+      "theoretical_loss": 4.187160024985044,
+      "tokens_seen": 292552704
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009205416248746239,
+      "loss": 1.8087,
+      "theoretical_loss": 4.187050286455294,
+      "tokens_seen": 292618240
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009205215646940822,
+      "loss": 1.9665,
+      "theoretical_loss": 4.1869405793802335,
+      "tokens_seen": 292683776
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009205015045135406,
+      "loss": 1.8839,
+      "theoretical_loss": 4.1868309037438065,
+      "tokens_seen": 292749312
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000920481444332999,
+      "loss": 1.7737,
+      "theoretical_loss": 4.1867212595299685,
+      "tokens_seen": 292814848
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009204613841524574,
+      "loss": 1.7755,
+      "theoretical_loss": 4.186611646722687,
+      "tokens_seen": 292880384
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009204413239719157,
+      "loss": 1.9252,
+      "theoretical_loss": 4.186502065305942,
+      "tokens_seen": 292945920
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009204212637913741,
+      "loss": 1.7296,
+      "theoretical_loss": 4.186392515263723,
+      "tokens_seen": 293011456
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009204012036108326,
+      "loss": 1.8694,
+      "theoretical_loss": 4.186282996580034,
+      "tokens_seen": 293076992
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009203811434302909,
+      "loss": 1.9082,
+      "theoretical_loss": 4.1861735092388885,
+      "tokens_seen": 293142528
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009203610832497493,
+      "loss": 1.9323,
+      "theoretical_loss": 4.186064053224314,
+      "tokens_seen": 293208064
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009203410230692077,
+      "loss": 1.7652,
+      "theoretical_loss": 4.185954628520347,
+      "tokens_seen": 293273600
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000920320962888666,
+      "loss": 1.8631,
+      "theoretical_loss": 4.185845235111037,
+      "tokens_seen": 293339136
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009203009027081244,
+      "loss": 1.7549,
+      "theoretical_loss": 4.185735872980447,
+      "tokens_seen": 293404672
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009202808425275828,
+      "loss": 1.927,
+      "theoretical_loss": 4.185626542112649,
+      "tokens_seen": 293470208
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009202607823470411,
+      "loss": 1.7839,
+      "theoretical_loss": 4.185517242491728,
+      "tokens_seen": 293535744
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009202407221664996,
+      "loss": 1.8171,
+      "theoretical_loss": 4.185407974101779,
+      "tokens_seen": 293601280
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009202206619859579,
+      "loss": 1.8146,
+      "theoretical_loss": 4.185298736926912,
+      "tokens_seen": 293666816
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009202006018054163,
+      "loss": 1.8778,
+      "theoretical_loss": 4.185189530951245,
+      "tokens_seen": 293732352
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009201805416248747,
+      "loss": 1.6603,
+      "theoretical_loss": 4.18508035615891,
+      "tokens_seen": 293797888
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000920160481444333,
+      "loss": 1.813,
+      "theoretical_loss": 4.184971212534048,
+      "tokens_seen": 293863424
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009201404212637914,
+      "loss": 1.8153,
+      "theoretical_loss": 4.184862100060816,
+      "tokens_seen": 293928960
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009201203610832499,
+      "loss": 1.8382,
+      "theoretical_loss": 4.184753018723379,
+      "tokens_seen": 293994496
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009201003009027081,
+      "loss": 1.7472,
+      "theoretical_loss": 4.184643968505914,
+      "tokens_seen": 294060032
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009200802407221666,
+      "loss": 1.8905,
+      "theoretical_loss": 4.184534949392611,
+      "tokens_seen": 294125568
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009200601805416248,
+      "loss": 1.7706,
+      "theoretical_loss": 4.18442596136767,
+      "tokens_seen": 294191104
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009200401203610833,
+      "loss": 1.8976,
+      "theoretical_loss": 4.184317004415303,
+      "tokens_seen": 294256640
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009200200601805417,
+      "loss": 1.8011,
+      "theoretical_loss": 4.184208078519735,
+      "tokens_seen": 294322176
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00092,
+      "loss": 1.7645,
+      "theoretical_loss": 4.184099183665199,
+      "tokens_seen": 294387712
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009199799398194584,
+      "loss": 1.9067,
+      "theoretical_loss": 4.183990319835944,
+      "tokens_seen": 294453248
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009199598796389169,
+      "loss": 1.7968,
+      "theoretical_loss": 4.183881487016227,
+      "tokens_seen": 294518784
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009199398194583751,
+      "loss": 1.856,
+      "theoretical_loss": 4.183772685190318,
+      "tokens_seen": 294584320
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009199197592778336,
+      "loss": 1.925,
+      "theoretical_loss": 4.1836639143425,
+      "tokens_seen": 294649856
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009198996990972918,
+      "loss": 1.787,
+      "theoretical_loss": 4.183555174457063,
+      "tokens_seen": 294715392
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009198796389167503,
+      "loss": 1.7992,
+      "theoretical_loss": 4.183446465518312,
+      "tokens_seen": 294780928
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009198595787362087,
+      "loss": 1.7961,
+      "theoretical_loss": 4.183337787510563,
+      "tokens_seen": 294846464
+    },
+    {
+      "epoch": 0.09,
+      "objective/train/advantage_avg": 0.4940253496170044,
+      "objective/train/docs_used": 114046,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.7510039806365967,
+      "objective/train/original_loss": 1.7510039806365967,
+      "objective/train/theoretical_loss": 4.1832291404181445,
+      "objective/train/tokens_used": 315372000,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24425998330116272,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506443977355957,
+      "objective/train/weighted_lm_loss": 1.83975088596344,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 1.0,
+      "theoretical_loss": 4.1832291404181445,
+      "tokens_seen": 294912000
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000919839518555667,
+      "loss": 1.8329,
+      "theoretical_loss": 4.1832291404181445,
+      "tokens_seen": 294912000
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009198194583751254,
+      "loss": 1.7224,
+      "theoretical_loss": 4.183120524225393,
+      "tokens_seen": 294977536
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009197993981945838,
+      "loss": 1.9731,
+      "theoretical_loss": 4.183011938916657,
+      "tokens_seen": 295043072
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009197793380140421,
+      "loss": 1.7446,
+      "theoretical_loss": 4.182903384476301,
+      "tokens_seen": 295108608
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009197592778335006,
+      "loss": 1.7929,
+      "theoretical_loss": 4.182794860888696,
+      "tokens_seen": 295174144
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009197392176529588,
+      "loss": 1.8788,
+      "theoretical_loss": 4.182686368138226,
+      "tokens_seen": 295239680
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009197191574724173,
+      "loss": 1.8534,
+      "theoretical_loss": 4.182577906209287,
+      "tokens_seen": 295305216
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009196990972918756,
+      "loss": 1.7757,
+      "theoretical_loss": 4.182469475086286,
+      "tokens_seen": 295370752
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000919679037111334,
+      "loss": 1.8442,
+      "theoretical_loss": 4.18236107475364,
+      "tokens_seen": 295436288
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009196589769307924,
+      "loss": 1.9694,
+      "theoretical_loss": 4.182252705195778,
+      "tokens_seen": 295501824
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009196389167502508,
+      "loss": 1.6922,
+      "theoretical_loss": 4.182144366397143,
+      "tokens_seen": 295567360
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009196188565697091,
+      "loss": 1.8983,
+      "theoretical_loss": 4.182036058342186,
+      "tokens_seen": 295632896
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009195987963891676,
+      "loss": 1.8164,
+      "theoretical_loss": 4.18192778101537,
+      "tokens_seen": 295698432
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009195787362086259,
+      "loss": 1.8424,
+      "theoretical_loss": 4.1818195344011695,
+      "tokens_seen": 295763968
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009195586760280843,
+      "loss": 1.7505,
+      "theoretical_loss": 4.181711318484071,
+      "tokens_seen": 295829504
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009195386158475426,
+      "loss": 1.769,
+      "theoretical_loss": 4.181603133248572,
+      "tokens_seen": 295895040
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000919518555667001,
+      "loss": 1.7624,
+      "theoretical_loss": 4.181494978679181,
+      "tokens_seen": 295960576
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009194984954864594,
+      "loss": 1.7339,
+      "theoretical_loss": 4.181386854760417,
+      "tokens_seen": 296026112
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009194784353059178,
+      "loss": 1.8652,
+      "theoretical_loss": 4.181278761476811,
+      "tokens_seen": 296091648
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009194583751253761,
+      "loss": 1.7627,
+      "theoretical_loss": 4.181170698812906,
+      "tokens_seen": 296157184
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009194383149448345,
+      "loss": 1.8077,
+      "theoretical_loss": 4.181062666753256,
+      "tokens_seen": 296222720
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009194182547642929,
+      "loss": 1.8082,
+      "theoretical_loss": 4.180954665282423,
+      "tokens_seen": 296288256
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009193981945837513,
+      "loss": 1.7497,
+      "theoretical_loss": 4.180846694384986,
+      "tokens_seen": 296353792
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009193781344032096,
+      "loss": 1.8343,
+      "theoretical_loss": 4.18073875404553,
+      "tokens_seen": 296419328
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000919358074222668,
+      "loss": 1.7863,
+      "theoretical_loss": 4.180630844248653,
+      "tokens_seen": 296484864
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009193380140421263,
+      "loss": 1.8319,
+      "theoretical_loss": 4.180522964978966,
+      "tokens_seen": 296550400
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009193179538615848,
+      "loss": 1.7604,
+      "theoretical_loss": 4.180415116221088,
+      "tokens_seen": 296615936
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009192978936810431,
+      "loss": 1.9332,
+      "theoretical_loss": 4.180307297959651,
+      "tokens_seen": 296681472
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009192778335005015,
+      "loss": 1.8325,
+      "theoretical_loss": 4.180199510179299,
+      "tokens_seen": 296747008
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009192577733199599,
+      "loss": 1.6463,
+      "theoretical_loss": 4.180091752864684,
+      "tokens_seen": 296812544
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009192377131394183,
+      "loss": 1.8372,
+      "theoretical_loss": 4.179984026000473,
+      "tokens_seen": 296878080
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009192176529588766,
+      "loss": 1.7951,
+      "theoretical_loss": 4.179876329571341,
+      "tokens_seen": 296943616
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009191975927783351,
+      "loss": 1.867,
+      "theoretical_loss": 4.179768663561975,
+      "tokens_seen": 297009152
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009191775325977933,
+      "loss": 1.7988,
+      "theoretical_loss": 4.179661027957074,
+      "tokens_seen": 297074688
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009191574724172518,
+      "loss": 1.8474,
+      "theoretical_loss": 4.179553422741347,
+      "tokens_seen": 297140224
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009191374122367101,
+      "loss": 1.725,
+      "theoretical_loss": 4.179445847899515,
+      "tokens_seen": 297205760
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009191173520561685,
+      "loss": 1.603,
+      "theoretical_loss": 4.1793383034163085,
+      "tokens_seen": 297271296
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009190972918756269,
+      "loss": 1.6909,
+      "theoretical_loss": 4.17923078927647,
+      "tokens_seen": 297336832
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009190772316950852,
+      "loss": 1.7892,
+      "theoretical_loss": 4.1791233054647545,
+      "tokens_seen": 297402368
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009190571715145436,
+      "loss": 1.8126,
+      "theoretical_loss": 4.179015851965926,
+      "tokens_seen": 297467904
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009190371113340021,
+      "loss": 1.7874,
+      "theoretical_loss": 4.178908428764759,
+      "tokens_seen": 297533440
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009190170511534603,
+      "loss": 1.7865,
+      "theoretical_loss": 4.178801035846041,
+      "tokens_seen": 297598976
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009189969909729188,
+      "loss": 1.81,
+      "theoretical_loss": 4.178693673194569,
+      "tokens_seen": 297664512
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000918976930792377,
+      "loss": 1.7445,
+      "theoretical_loss": 4.178586340795152,
+      "tokens_seen": 297730048
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009189568706118355,
+      "loss": 1.8662,
+      "theoretical_loss": 4.17847903863261,
+      "tokens_seen": 297795584
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009189368104312939,
+      "loss": 1.7746,
+      "theoretical_loss": 4.178371766691773,
+      "tokens_seen": 297861120
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009189167502507522,
+      "loss": 1.7787,
+      "theoretical_loss": 4.1782645249574815,
+      "tokens_seen": 297926656
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009188966900702107,
+      "loss": 1.8718,
+      "theoretical_loss": 4.178157313414589,
+      "tokens_seen": 297992192
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009188766298896691,
+      "loss": 1.8258,
+      "theoretical_loss": 4.178050132047958,
+      "tokens_seen": 298057728
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009188565697091274,
+      "loss": 1.7314,
+      "theoretical_loss": 4.177942980842464,
+      "tokens_seen": 298123264
+    },
+    {
+      "epoch": 0.09,
+      "objective/train/advantage_avg": 0.4918665587902069,
+      "objective/train/docs_used": 115247,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.66420578956604,
+      "objective/train/original_loss": 1.66420578956604,
+      "objective/train/theoretical_loss": 4.1778358597829905,
+      "objective/train/tokens_used": 318648800,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2422984093427658,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0504183769226074,
+      "objective/train/weighted_lm_loss": 1.7481639385223389,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.1778358597829905,
+      "tokens_seen": 298188800
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009188365095285858,
+      "loss": 1.825,
+      "theoretical_loss": 4.1778358597829905,
+      "tokens_seen": 298188800
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009188164493480442,
+      "loss": 1.7152,
+      "theoretical_loss": 4.1777287688544344,
+      "tokens_seen": 298254336
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009187963891675025,
+      "loss": 1.8607,
+      "theoretical_loss": 4.177621708041703,
+      "tokens_seen": 298319872
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000918776328986961,
+      "loss": 1.8997,
+      "theoretical_loss": 4.177514677329713,
+      "tokens_seen": 298385408
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009187562688064192,
+      "loss": 1.8691,
+      "theoretical_loss": 4.177407676703393,
+      "tokens_seen": 298450944
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009187362086258777,
+      "loss": 1.7494,
+      "theoretical_loss": 4.177300706147683,
+      "tokens_seen": 298516480
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000918716148445336,
+      "loss": 1.8863,
+      "theoretical_loss": 4.177193765647534,
+      "tokens_seen": 298582016
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009186960882647944,
+      "loss": 1.7996,
+      "theoretical_loss": 4.1770868551879055,
+      "tokens_seen": 298647552
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009186760280842528,
+      "loss": 1.8353,
+      "theoretical_loss": 4.17697997475377,
+      "tokens_seen": 298713088
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009186559679037112,
+      "loss": 1.8755,
+      "theoretical_loss": 4.176873124330111,
+      "tokens_seen": 298778624
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009186359077231695,
+      "loss": 2.0181,
+      "theoretical_loss": 4.176766303901922,
+      "tokens_seen": 298844160
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000918615847542628,
+      "loss": 2.0517,
+      "theoretical_loss": 4.176659513454207,
+      "tokens_seen": 298909696
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009185957873620863,
+      "loss": 1.8836,
+      "theoretical_loss": 4.176552752971981,
+      "tokens_seen": 298975232
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009185757271815447,
+      "loss": 1.875,
+      "theoretical_loss": 4.1764460224402695,
+      "tokens_seen": 299040768
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000918555667001003,
+      "loss": 1.8465,
+      "theoretical_loss": 4.17633932184411,
+      "tokens_seen": 299106304
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009185356068204614,
+      "loss": 1.9421,
+      "theoretical_loss": 4.176232651168551,
+      "tokens_seen": 299171840
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009185155466399198,
+      "loss": 1.8236,
+      "theoretical_loss": 4.176126010398647,
+      "tokens_seen": 299237376
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009184954864593782,
+      "loss": 1.9052,
+      "theoretical_loss": 4.176019399519471,
+      "tokens_seen": 299302912
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009184754262788365,
+      "loss": 1.8264,
+      "theoretical_loss": 4.1759128185161005,
+      "tokens_seen": 299368448
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009184553660982949,
+      "loss": 1.7301,
+      "theoretical_loss": 4.175806267373627,
+      "tokens_seen": 299433984
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009184353059177533,
+      "loss": 1.8618,
+      "theoretical_loss": 4.17569974607715,
+      "tokens_seen": 299499520
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009184152457372117,
+      "loss": 1.7564,
+      "theoretical_loss": 4.1755932546117815,
+      "tokens_seen": 299565056
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00091839518555667,
+      "loss": 1.7976,
+      "theoretical_loss": 4.175486792962646,
+      "tokens_seen": 299630592
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009183751253761285,
+      "loss": 1.8856,
+      "theoretical_loss": 4.175380361114874,
+      "tokens_seen": 299696128
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009183550651955867,
+      "loss": 1.7783,
+      "theoretical_loss": 4.1752739590536105,
+      "tokens_seen": 299761664
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009183350050150452,
+      "loss": 1.7971,
+      "theoretical_loss": 4.17516758676401,
+      "tokens_seen": 299827200
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009183149448345035,
+      "loss": 1.924,
+      "theoretical_loss": 4.175061244231237,
+      "tokens_seen": 299892736
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009182948846539619,
+      "loss": 1.9111,
+      "theoretical_loss": 4.174954931440468,
+      "tokens_seen": 299958272
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009182748244734203,
+      "loss": 1.947,
+      "theoretical_loss": 4.174848648376889,
+      "tokens_seen": 300023808
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009182547642928787,
+      "loss": 1.8443,
+      "theoretical_loss": 4.174742395025696,
+      "tokens_seen": 300089344
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000918234704112337,
+      "loss": 1.8847,
+      "theoretical_loss": 4.174636171372097,
+      "tokens_seen": 300154880
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009182146439317955,
+      "loss": 1.8018,
+      "theoretical_loss": 4.174529977401312,
+      "tokens_seen": 300220416
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009181945837512537,
+      "loss": 1.7974,
+      "theoretical_loss": 4.174423813098566,
+      "tokens_seen": 300285952
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009181745235707122,
+      "loss": 1.8399,
+      "theoretical_loss": 4.174317678449101,
+      "tokens_seen": 300351488
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009181544633901705,
+      "loss": 1.7105,
+      "theoretical_loss": 4.174211573438166,
+      "tokens_seen": 300417024
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009181344032096289,
+      "loss": 1.9141,
+      "theoretical_loss": 4.174105498051022,
+      "tokens_seen": 300482560
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009181143430290873,
+      "loss": 1.9205,
+      "theoretical_loss": 4.173999452272939,
+      "tokens_seen": 300548096
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009180942828485456,
+      "loss": 1.9612,
+      "theoretical_loss": 4.173893436089199,
+      "tokens_seen": 300613632
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000918074222668004,
+      "loss": 1.92,
+      "theoretical_loss": 4.173787449485094,
+      "tokens_seen": 300679168
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009180541624874625,
+      "loss": 2.085,
+      "theoretical_loss": 4.173681492445926,
+      "tokens_seen": 300744704
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009180341023069207,
+      "loss": 1.9086,
+      "theoretical_loss": 4.173575564957009,
+      "tokens_seen": 300810240
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009180140421263792,
+      "loss": 1.8754,
+      "theoretical_loss": 4.1734696670036655,
+      "tokens_seen": 300875776
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009179939819458374,
+      "loss": 2.1293,
+      "theoretical_loss": 4.17336379857123,
+      "tokens_seen": 300941312
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009179739217652959,
+      "loss": 1.9747,
+      "theoretical_loss": 4.1732579596450465,
+      "tokens_seen": 301006848
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009179538615847543,
+      "loss": 1.8285,
+      "theoretical_loss": 4.17315215021047,
+      "tokens_seen": 301072384
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009179338014042126,
+      "loss": 2.0572,
+      "theoretical_loss": 4.173046370252866,
+      "tokens_seen": 301137920
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000917913741223671,
+      "loss": 1.832,
+      "theoretical_loss": 4.172940619757611,
+      "tokens_seen": 301203456
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009178936810431295,
+      "loss": 1.8726,
+      "theoretical_loss": 4.172834898710089,
+      "tokens_seen": 301268992
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009178736208625877,
+      "loss": 1.8577,
+      "theoretical_loss": 4.1727292070957,
+      "tokens_seen": 301334528
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009178535606820462,
+      "loss": 1.934,
+      "theoretical_loss": 4.172623544899849,
+      "tokens_seen": 301400064
+    },
+    {
+      "epoch": 0.09,
+      "objective/train/advantage_avg": 0.4897353947162628,
+      "objective/train/docs_used": 116418,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.8441132307052612,
+      "objective/train/original_loss": 1.8441133499145508,
+      "objective/train/theoretical_loss": 4.172517912107954,
+      "objective/train/tokens_used": 321925600,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24035406112670898,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050195336341858,
+      "objective/train/weighted_lm_loss": 1.9364567995071411,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.172517912107954,
+      "tokens_seen": 301465600
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009178335005015045,
+      "loss": 2.0693,
+      "theoretical_loss": 4.172517912107954,
+      "tokens_seen": 301465600
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009178134403209629,
+      "loss": 1.8463,
+      "theoretical_loss": 4.172412308705443,
+      "tokens_seen": 301531136
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009177933801404213,
+      "loss": 1.8559,
+      "theoretical_loss": 4.172306734677753,
+      "tokens_seen": 301596672
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009177733199598796,
+      "loss": 1.9022,
+      "theoretical_loss": 4.172201190010334,
+      "tokens_seen": 301662208
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000917753259779338,
+      "loss": 1.814,
+      "theoretical_loss": 4.172095674688645,
+      "tokens_seen": 301727744
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009177331995987964,
+      "loss": 2.0133,
+      "theoretical_loss": 4.171990188698155,
+      "tokens_seen": 301793280
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009177131394182547,
+      "loss": 1.8285,
+      "theoretical_loss": 4.171884732024344,
+      "tokens_seen": 301858816
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009176930792377132,
+      "loss": 1.8609,
+      "theoretical_loss": 4.171779304652702,
+      "tokens_seen": 301924352
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009176730190571715,
+      "loss": 1.8516,
+      "theoretical_loss": 4.171673906568729,
+      "tokens_seen": 301989888
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009176529588766299,
+      "loss": 1.8591,
+      "theoretical_loss": 4.171568537757936,
+      "tokens_seen": 302055424
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009176328986960883,
+      "loss": 2.047,
+      "theoretical_loss": 4.171463198205844,
+      "tokens_seen": 302120960
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009176128385155467,
+      "loss": 1.8739,
+      "theoretical_loss": 4.171357887897985,
+      "tokens_seen": 302186496
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000917592778335005,
+      "loss": 1.8573,
+      "theoretical_loss": 4.171252606819899,
+      "tokens_seen": 302252032
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009175727181544634,
+      "loss": 1.9282,
+      "theoretical_loss": 4.17114735495714,
+      "tokens_seen": 302317568
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009175526579739217,
+      "loss": 1.8217,
+      "theoretical_loss": 4.171042132295268,
+      "tokens_seen": 302383104
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009175325977933802,
+      "loss": 1.8833,
+      "theoretical_loss": 4.170936938819857,
+      "tokens_seen": 302448640
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009175125376128385,
+      "loss": 1.779,
+      "theoretical_loss": 4.170831774516489,
+      "tokens_seen": 302514176
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009174924774322969,
+      "loss": 1.8598,
+      "theoretical_loss": 4.170726639370757,
+      "tokens_seen": 302579712
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009174724172517552,
+      "loss": 1.855,
+      "theoretical_loss": 4.170621533368265,
+      "tokens_seen": 302645248
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009174523570712137,
+      "loss": 1.8664,
+      "theoretical_loss": 4.170516456494624,
+      "tokens_seen": 302710784
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000917432296890672,
+      "loss": 1.853,
+      "theoretical_loss": 4.170411408735461,
+      "tokens_seen": 302776320
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009174122367101304,
+      "loss": 1.7922,
+      "theoretical_loss": 4.170306390076408,
+      "tokens_seen": 302841856
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009173921765295889,
+      "loss": 2.0018,
+      "theoretical_loss": 4.1702014005031085,
+      "tokens_seen": 302907392
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009173721163490471,
+      "loss": 1.7791,
+      "theoretical_loss": 4.170096440001217,
+      "tokens_seen": 302972928
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009173520561685056,
+      "loss": 1.9441,
+      "theoretical_loss": 4.169991508556398,
+      "tokens_seen": 303038464
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009173319959879639,
+      "loss": 1.8292,
+      "theoretical_loss": 4.169886606154328,
+      "tokens_seen": 303104000
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009173119358074223,
+      "loss": 1.8355,
+      "theoretical_loss": 4.169781732780688,
+      "tokens_seen": 303169536
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009172918756268807,
+      "loss": 1.861,
+      "theoretical_loss": 4.169676888421175,
+      "tokens_seen": 303235072
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009172718154463391,
+      "loss": 1.7574,
+      "theoretical_loss": 4.169572073061493,
+      "tokens_seen": 303300608
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009172517552657974,
+      "loss": 1.9617,
+      "theoretical_loss": 4.169467286687359,
+      "tokens_seen": 303366144
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009172316950852559,
+      "loss": 1.8277,
+      "theoretical_loss": 4.169362529284497,
+      "tokens_seen": 303431680
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009172116349047141,
+      "loss": 1.8347,
+      "theoretical_loss": 4.169257800838642,
+      "tokens_seen": 303497216
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009171915747241726,
+      "loss": 2.0021,
+      "theoretical_loss": 4.16915310133554,
+      "tokens_seen": 303562752
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000917171514543631,
+      "loss": 1.9621,
+      "theoretical_loss": 4.169048430760946,
+      "tokens_seen": 303628288
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009171514543630893,
+      "loss": 1.8753,
+      "theoretical_loss": 4.168943789100626,
+      "tokens_seen": 303693824
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009171313941825477,
+      "loss": 1.8549,
+      "theoretical_loss": 4.168839176340356,
+      "tokens_seen": 303759360
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000917111334002006,
+      "loss": 1.7648,
+      "theoretical_loss": 4.1687345924659205,
+      "tokens_seen": 303824896
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009170912738214644,
+      "loss": 1.7129,
+      "theoretical_loss": 4.168630037463117,
+      "tokens_seen": 303890432
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009170712136409229,
+      "loss": 1.8143,
+      "theoretical_loss": 4.168525511317751,
+      "tokens_seen": 303955968
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009170511534603811,
+      "loss": 1.8852,
+      "theoretical_loss": 4.168421014015637,
+      "tokens_seen": 304021504
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009170310932798396,
+      "loss": 1.8363,
+      "theoretical_loss": 4.168316545542602,
+      "tokens_seen": 304087040
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009170110330992978,
+      "loss": 1.8495,
+      "theoretical_loss": 4.168212105884482,
+      "tokens_seen": 304152576
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009169909729187563,
+      "loss": 1.8456,
+      "theoretical_loss": 4.168107695027123,
+      "tokens_seen": 304218112
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009169709127382147,
+      "loss": 1.7831,
+      "theoretical_loss": 4.1680033129563805,
+      "tokens_seen": 304283648
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000916950852557673,
+      "loss": 1.75,
+      "theoretical_loss": 4.167898959658121,
+      "tokens_seen": 304349184
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009169307923771314,
+      "loss": 1.7698,
+      "theoretical_loss": 4.167794635118219,
+      "tokens_seen": 304414720
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009169107321965899,
+      "loss": 1.8279,
+      "theoretical_loss": 4.167690339322562,
+      "tokens_seen": 304480256
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009168906720160481,
+      "loss": 1.6521,
+      "theoretical_loss": 4.167586072257047,
+      "tokens_seen": 304545792
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009168706118355066,
+      "loss": 1.7137,
+      "theoretical_loss": 4.167481833907576,
+      "tokens_seen": 304611328
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009168505516549649,
+      "loss": 1.931,
+      "theoretical_loss": 4.167377624260068,
+      "tokens_seen": 304676864
+    },
+    {
+      "epoch": 0.09,
+      "objective/train/advantage_avg": 0.4910045862197876,
+      "objective/train/docs_used": 117554,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.6303279399871826,
+      "objective/train/original_loss": 1.6303277015686035,
+      "objective/train/theoretical_loss": 4.167273443300447,
+      "objective/train/tokens_used": 325202400,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.241532102227211,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050328254699707,
+      "objective/train/weighted_lm_loss": 1.7122821807861328,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.167273443300447,
+      "tokens_seen": 304742400
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009168304914744233,
+      "loss": 1.7162,
+      "theoretical_loss": 4.167273443300447,
+      "tokens_seen": 304742400
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009168104312938817,
+      "loss": 1.9333,
+      "theoretical_loss": 4.167169291014649,
+      "tokens_seen": 304807936
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00091679037111334,
+      "loss": 2.0274,
+      "theoretical_loss": 4.16706516738862,
+      "tokens_seen": 304873472
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009167703109327984,
+      "loss": 1.7796,
+      "theoretical_loss": 4.1669610724083155,
+      "tokens_seen": 304939008
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009167502507522568,
+      "loss": 1.8187,
+      "theoretical_loss": 4.1668570060597006,
+      "tokens_seen": 305004544
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009167301905717151,
+      "loss": 1.8278,
+      "theoretical_loss": 4.16675296832875,
+      "tokens_seen": 305070080
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009167101303911736,
+      "loss": 1.7686,
+      "theoretical_loss": 4.166648959201449,
+      "tokens_seen": 305135616
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009166900702106319,
+      "loss": 1.9126,
+      "theoretical_loss": 4.1665449786637945,
+      "tokens_seen": 305201152
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009166700100300903,
+      "loss": 1.8431,
+      "theoretical_loss": 4.166441026701788,
+      "tokens_seen": 305266688
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009166499498495487,
+      "loss": 1.8467,
+      "theoretical_loss": 4.166337103301447,
+      "tokens_seen": 305332224
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000916629889669007,
+      "loss": 1.8731,
+      "theoretical_loss": 4.166233208448794,
+      "tokens_seen": 305397760
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009166098294884654,
+      "loss": 1.8152,
+      "theoretical_loss": 4.166129342129865,
+      "tokens_seen": 305463296
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009165897693079238,
+      "loss": 1.8298,
+      "theoretical_loss": 4.166025504330703,
+      "tokens_seen": 305528832
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009165697091273821,
+      "loss": 1.8831,
+      "theoretical_loss": 4.165921695037363,
+      "tokens_seen": 305594368
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009165496489468406,
+      "loss": 1.8718,
+      "theoretical_loss": 4.165817914235908,
+      "tokens_seen": 305659904
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009165295887662989,
+      "loss": 1.8804,
+      "theoretical_loss": 4.165714161912412,
+      "tokens_seen": 305725440
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009165095285857573,
+      "loss": 1.7289,
+      "theoretical_loss": 4.165610438052958,
+      "tokens_seen": 305790976
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009164894684052156,
+      "loss": 1.7683,
+      "theoretical_loss": 4.1655067426436405,
+      "tokens_seen": 305856512
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009164694082246741,
+      "loss": 1.8967,
+      "theoretical_loss": 4.165403075670562,
+      "tokens_seen": 305922048
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009164493480441324,
+      "loss": 1.8074,
+      "theoretical_loss": 4.165299437119834,
+      "tokens_seen": 305987584
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009164292878635908,
+      "loss": 1.8561,
+      "theoretical_loss": 4.1651958269775795,
+      "tokens_seen": 306053120
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009164092276830491,
+      "loss": 1.7487,
+      "theoretical_loss": 4.165092245229932,
+      "tokens_seen": 306118656
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009163891675025075,
+      "loss": 1.7637,
+      "theoretical_loss": 4.164988691863032,
+      "tokens_seen": 306184192
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009163691073219659,
+      "loss": 1.9437,
+      "theoretical_loss": 4.164885166863032,
+      "tokens_seen": 306249728
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009163490471414243,
+      "loss": 1.8181,
+      "theoretical_loss": 4.164781670216092,
+      "tokens_seen": 306315264
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009163289869608826,
+      "loss": 1.7637,
+      "theoretical_loss": 4.1646782019083854,
+      "tokens_seen": 306380800
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009163089267803411,
+      "loss": 1.6582,
+      "theoretical_loss": 4.164574761926092,
+      "tokens_seen": 306446336
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009162888665997994,
+      "loss": 1.76,
+      "theoretical_loss": 4.164471350255402,
+      "tokens_seen": 306511872
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009162688064192578,
+      "loss": 2.0577,
+      "theoretical_loss": 4.164367966882517,
+      "tokens_seen": 306577408
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009162487462387162,
+      "loss": 1.574,
+      "theoretical_loss": 4.164264611793644,
+      "tokens_seen": 306642944
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009162286860581745,
+      "loss": 1.8299,
+      "theoretical_loss": 4.164161284975005,
+      "tokens_seen": 306708480
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009162086258776329,
+      "loss": 1.7307,
+      "theoretical_loss": 4.164057986412828,
+      "tokens_seen": 306774016
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009161885656970913,
+      "loss": 1.8212,
+      "theoretical_loss": 4.163954716093354,
+      "tokens_seen": 306839552
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009161685055165496,
+      "loss": 1.8538,
+      "theoretical_loss": 4.163851474002829,
+      "tokens_seen": 306905088
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009161484453360081,
+      "loss": 1.7853,
+      "theoretical_loss": 4.1637482601275115,
+      "tokens_seen": 306970624
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009161283851554663,
+      "loss": 1.8917,
+      "theoretical_loss": 4.16364507445367,
+      "tokens_seen": 307036160
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009161083249749248,
+      "loss": 1.831,
+      "theoretical_loss": 4.163541916967581,
+      "tokens_seen": 307101696
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009160882647943832,
+      "loss": 1.9265,
+      "theoretical_loss": 4.163438787655533,
+      "tokens_seen": 307167232
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009160682046138415,
+      "loss": 1.8527,
+      "theoretical_loss": 4.163335686503822,
+      "tokens_seen": 307232768
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009160481444332999,
+      "loss": 1.868,
+      "theoretical_loss": 4.163232613498753,
+      "tokens_seen": 307298304
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009160280842527582,
+      "loss": 1.7818,
+      "theoretical_loss": 4.163129568626642,
+      "tokens_seen": 307363840
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009160080240722166,
+      "loss": 1.7847,
+      "theoretical_loss": 4.1630265518738145,
+      "tokens_seen": 307429376
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009159879638916751,
+      "loss": 1.8922,
+      "theoretical_loss": 4.162923563226607,
+      "tokens_seen": 307494912
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009159679037111333,
+      "loss": 1.7713,
+      "theoretical_loss": 4.162820602671362,
+      "tokens_seen": 307560448
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009159478435305918,
+      "loss": 1.8753,
+      "theoretical_loss": 4.162717670194433,
+      "tokens_seen": 307625984
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009159277833500502,
+      "loss": 1.673,
+      "theoretical_loss": 4.162614765782186,
+      "tokens_seen": 307691520
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009159077231695085,
+      "loss": 1.805,
+      "theoretical_loss": 4.1625118894209905,
+      "tokens_seen": 307757056
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000915887662988967,
+      "loss": 1.7618,
+      "theoretical_loss": 4.1624090410972325,
+      "tokens_seen": 307822592
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009158676028084253,
+      "loss": 1.7028,
+      "theoretical_loss": 4.162306220797301,
+      "tokens_seen": 307888128
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009158475426278837,
+      "loss": 1.8686,
+      "theoretical_loss": 4.162203428507601,
+      "tokens_seen": 307953664
+    },
+    {
+      "epoch": 0.09,
+      "objective/train/advantage_avg": 0.49156174063682556,
+      "objective/train/docs_used": 118729,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.7690374851226807,
+      "objective/train/original_loss": 1.7690374851226807,
+      "objective/train/theoretical_loss": 4.16210066421454,
+      "objective/train/tokens_used": 328479200,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2423403114080429,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050388216972351,
+      "objective/train/weighted_lm_loss": 1.858426570892334,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.16210066421454,
+      "tokens_seen": 308019200
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009158274824473421,
+      "loss": 1.7856,
+      "theoretical_loss": 4.16210066421454,
+      "tokens_seen": 308019200
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009158074222668004,
+      "loss": 1.8842,
+      "theoretical_loss": 4.16199792790454,
+      "tokens_seen": 308084736
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009157873620862588,
+      "loss": 1.8463,
+      "theoretical_loss": 4.161895219564032,
+      "tokens_seen": 308150272
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009157673019057172,
+      "loss": 1.7313,
+      "theoretical_loss": 4.161792539179454,
+      "tokens_seen": 308215808
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009157472417251755,
+      "loss": 1.7317,
+      "theoretical_loss": 4.161689886737255,
+      "tokens_seen": 308281344
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000915727181544634,
+      "loss": 1.9005,
+      "theoretical_loss": 4.161587262223895,
+      "tokens_seen": 308346880
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009157071213640923,
+      "loss": 1.7685,
+      "theoretical_loss": 4.161484665625839,
+      "tokens_seen": 308412416
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009156870611835507,
+      "loss": 1.8842,
+      "theoretical_loss": 4.161382096929565,
+      "tokens_seen": 308477952
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009156670010030091,
+      "loss": 1.8179,
+      "theoretical_loss": 4.161279556121562,
+      "tokens_seen": 308543488
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009156469408224675,
+      "loss": 1.8155,
+      "theoretical_loss": 4.1611770431883235,
+      "tokens_seen": 308609024
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009156268806419258,
+      "loss": 1.8621,
+      "theoretical_loss": 4.161074558116356,
+      "tokens_seen": 308674560
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009156068204613842,
+      "loss": 1.8453,
+      "theoretical_loss": 4.160972100892174,
+      "tokens_seen": 308740096
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009155867602808425,
+      "loss": 1.737,
+      "theoretical_loss": 4.160869671502302,
+      "tokens_seen": 308805632
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000915566700100301,
+      "loss": 1.767,
+      "theoretical_loss": 4.160767269933274,
+      "tokens_seen": 308871168
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009155466399197593,
+      "loss": 1.814,
+      "theoretical_loss": 4.160664896171632,
+      "tokens_seen": 308936704
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009155265797392177,
+      "loss": 1.8207,
+      "theoretical_loss": 4.160562550203928,
+      "tokens_seen": 309002240
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000915506519558676,
+      "loss": 1.9213,
+      "theoretical_loss": 4.160460232016725,
+      "tokens_seen": 309067776
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009154864593781345,
+      "loss": 1.8559,
+      "theoretical_loss": 4.160357941596594,
+      "tokens_seen": 309133312
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009154663991975928,
+      "loss": 1.7862,
+      "theoretical_loss": 4.160255678930115,
+      "tokens_seen": 309198848
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009154463390170512,
+      "loss": 1.8416,
+      "theoretical_loss": 4.1601534440038765,
+      "tokens_seen": 309264384
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009154262788365095,
+      "loss": 1.8704,
+      "theoretical_loss": 4.16005123680448,
+      "tokens_seen": 309329920
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009154062186559679,
+      "loss": 1.7641,
+      "theoretical_loss": 4.159949057318532,
+      "tokens_seen": 309395456
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009153861584754263,
+      "loss": 1.7451,
+      "theoretical_loss": 4.1598469055326515,
+      "tokens_seen": 309460992
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009153660982948847,
+      "loss": 1.7089,
+      "theoretical_loss": 4.159744781433464,
+      "tokens_seen": 309526528
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000915346038114343,
+      "loss": 1.8765,
+      "theoretical_loss": 4.159642685007606,
+      "tokens_seen": 309592064
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009153259779338015,
+      "loss": 1.8021,
+      "theoretical_loss": 4.159540616241723,
+      "tokens_seen": 309657600
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009153059177532598,
+      "loss": 1.7355,
+      "theoretical_loss": 4.159438575122471,
+      "tokens_seen": 309723136
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009152858575727182,
+      "loss": 1.7621,
+      "theoretical_loss": 4.159336561636513,
+      "tokens_seen": 309788672
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009152657973921766,
+      "loss": 1.7003,
+      "theoretical_loss": 4.1592345757705225,
+      "tokens_seen": 309854208
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009152457372116349,
+      "loss": 1.6826,
+      "theoretical_loss": 4.159132617511181,
+      "tokens_seen": 309919744
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009152256770310933,
+      "loss": 1.7535,
+      "theoretical_loss": 4.159030686845182,
+      "tokens_seen": 309985280
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009152056168505517,
+      "loss": 1.776,
+      "theoretical_loss": 4.158928783759226,
+      "tokens_seen": 310050816
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00091518555667001,
+      "loss": 1.7716,
+      "theoretical_loss": 4.158826908240022,
+      "tokens_seen": 310116352
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009151654964894685,
+      "loss": 1.7732,
+      "theoretical_loss": 4.158725060274291,
+      "tokens_seen": 310181888
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009151454363089267,
+      "loss": 1.7397,
+      "theoretical_loss": 4.158623239848762,
+      "tokens_seen": 310247424
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009151253761283852,
+      "loss": 1.7299,
+      "theoretical_loss": 4.15852144695017,
+      "tokens_seen": 310312960
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009151053159478436,
+      "loss": 1.8316,
+      "theoretical_loss": 4.158419681565265,
+      "tokens_seen": 310378496
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009150852557673019,
+      "loss": 1.7404,
+      "theoretical_loss": 4.158317943680801,
+      "tokens_seen": 310444032
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009150651955867603,
+      "loss": 1.6477,
+      "theoretical_loss": 4.158216233283545,
+      "tokens_seen": 310509568
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009150451354062186,
+      "loss": 1.8772,
+      "theoretical_loss": 4.158114550360271,
+      "tokens_seen": 310575104
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000915025075225677,
+      "loss": 1.925,
+      "theoretical_loss": 4.1580128948977615,
+      "tokens_seen": 310640640
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009150050150451355,
+      "loss": 1.7451,
+      "theoretical_loss": 4.157911266882811,
+      "tokens_seen": 310706176
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009149849548645937,
+      "loss": 1.7826,
+      "theoretical_loss": 4.157809666302221,
+      "tokens_seen": 310771712
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009149648946840522,
+      "loss": 1.7177,
+      "theoretical_loss": 4.157708093142801,
+      "tokens_seen": 310837248
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009149448345035106,
+      "loss": 1.9009,
+      "theoretical_loss": 4.157606547391374,
+      "tokens_seen": 310902784
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009149247743229689,
+      "loss": 1.7678,
+      "theoretical_loss": 4.157505029034767,
+      "tokens_seen": 310968320
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009149047141424273,
+      "loss": 1.9484,
+      "theoretical_loss": 4.157403538059819,
+      "tokens_seen": 311033856
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009148846539618857,
+      "loss": 1.8541,
+      "theoretical_loss": 4.157302074453379,
+      "tokens_seen": 311099392
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000914864593781344,
+      "loss": 1.8846,
+      "theoretical_loss": 4.157200638202301,
+      "tokens_seen": 311164928
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009148445336008025,
+      "loss": 1.5425,
+      "theoretical_loss": 4.157099229293453,
+      "tokens_seen": 311230464
+    },
+    {
+      "epoch": 0.09,
+      "objective/train/advantage_avg": 0.4924384355545044,
+      "objective/train/docs_used": 120076,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.819303274154663,
+      "objective/train/original_loss": 1.8193035125732422,
+      "objective/train/theoretical_loss": 4.156997847713709,
+      "objective/train/tokens_used": 331756000,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24328577518463135,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050480604171753,
+      "objective/train/weighted_lm_loss": 1.911105751991272,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.156997847713709,
+      "tokens_seen": 311296000
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009148244734202607,
+      "loss": 1.7748,
+      "theoretical_loss": 4.156997847713709,
+      "tokens_seen": 311296000
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009148044132397192,
+      "loss": 1.7985,
+      "theoretical_loss": 4.156896493449951,
+      "tokens_seen": 311361536
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009147843530591775,
+      "loss": 1.7454,
+      "theoretical_loss": 4.156795166489074,
+      "tokens_seen": 311427072
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009147642928786359,
+      "loss": 1.8097,
+      "theoretical_loss": 4.15669386681798,
+      "tokens_seen": 311492608
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009147442326980943,
+      "loss": 1.8527,
+      "theoretical_loss": 4.156592594423579,
+      "tokens_seen": 311558144
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009147241725175527,
+      "loss": 1.7922,
+      "theoretical_loss": 4.156491349292789,
+      "tokens_seen": 311623680
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000914704112337011,
+      "loss": 1.7492,
+      "theoretical_loss": 4.156390131412543,
+      "tokens_seen": 311689216
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009146840521564695,
+      "loss": 1.6228,
+      "theoretical_loss": 4.156288940769778,
+      "tokens_seen": 311754752
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009146639919759277,
+      "loss": 1.8304,
+      "theoretical_loss": 4.156187777351438,
+      "tokens_seen": 311820288
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009146439317953862,
+      "loss": 1.7774,
+      "theoretical_loss": 4.156086641144483,
+      "tokens_seen": 311885824
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009146238716148445,
+      "loss": 1.8537,
+      "theoretical_loss": 4.155985532135875,
+      "tokens_seen": 311951360
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009146038114343029,
+      "loss": 1.7901,
+      "theoretical_loss": 4.155884450312589,
+      "tokens_seen": 312016896
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009145837512537613,
+      "loss": 1.8438,
+      "theoretical_loss": 4.155783395661609,
+      "tokens_seen": 312082432
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009145636910732197,
+      "loss": 1.8641,
+      "theoretical_loss": 4.155682368169925,
+      "tokens_seen": 312147968
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000914543630892678,
+      "loss": 1.8986,
+      "theoretical_loss": 4.1555813678245395,
+      "tokens_seen": 312213504
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009145235707121364,
+      "loss": 1.9761,
+      "theoretical_loss": 4.155480394612462,
+      "tokens_seen": 312279040
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009145035105315948,
+      "loss": 1.644,
+      "theoretical_loss": 4.15537944852071,
+      "tokens_seen": 312344576
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009144834503510532,
+      "loss": 1.8129,
+      "theoretical_loss": 4.155278529536313,
+      "tokens_seen": 312410112
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009144633901705115,
+      "loss": 1.6904,
+      "theoretical_loss": 4.155177637646306,
+      "tokens_seen": 312475648
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00091444332998997,
+      "loss": 1.8086,
+      "theoretical_loss": 4.1550767728377345,
+      "tokens_seen": 312541184
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009144232698094282,
+      "loss": 1.619,
+      "theoretical_loss": 4.154975935097655,
+      "tokens_seen": 312606720
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009144032096288867,
+      "loss": 1.7375,
+      "theoretical_loss": 4.154875124413128,
+      "tokens_seen": 312672256
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009143831494483451,
+      "loss": 1.779,
+      "theoretical_loss": 4.154774340771228,
+      "tokens_seen": 312737792
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009143630892678034,
+      "loss": 1.6979,
+      "theoretical_loss": 4.154673584159036,
+      "tokens_seen": 312803328
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009143430290872619,
+      "loss": 1.8726,
+      "theoretical_loss": 4.15457285456364,
+      "tokens_seen": 312868864
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009143229689067202,
+      "loss": 1.7921,
+      "theoretical_loss": 4.154472151972142,
+      "tokens_seen": 312934400
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009143029087261786,
+      "loss": 1.8944,
+      "theoretical_loss": 4.154371476371646,
+      "tokens_seen": 312999936
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000914282848545637,
+      "loss": 1.9317,
+      "theoretical_loss": 4.154270827749271,
+      "tokens_seen": 313065472
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009142627883650953,
+      "loss": 1.8379,
+      "theoretical_loss": 4.154170206092143,
+      "tokens_seen": 313131008
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009142427281845537,
+      "loss": 1.6154,
+      "theoretical_loss": 4.154069611387394,
+      "tokens_seen": 313196544
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009142226680040121,
+      "loss": 1.7201,
+      "theoretical_loss": 4.153969043622169,
+      "tokens_seen": 313262080
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009142026078234704,
+      "loss": 1.9128,
+      "theoretical_loss": 4.153868502783618,
+      "tokens_seen": 313327616
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009141825476429289,
+      "loss": 1.868,
+      "theoretical_loss": 4.153767988858904,
+      "tokens_seen": 313393152
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0009141624874623871,
+      "loss": 1.7828,
+      "theoretical_loss": 4.153667501835195,
+      "tokens_seen": 313458688
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009141424272818456,
+      "loss": 1.7955,
+      "theoretical_loss": 4.15356704169967,
+      "tokens_seen": 313524224
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.000914122367101304,
+      "loss": 1.8222,
+      "theoretical_loss": 4.1534666084395155,
+      "tokens_seen": 313589760
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009141023069207623,
+      "loss": 1.7689,
+      "theoretical_loss": 4.153366202041927,
+      "tokens_seen": 313655296
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009140822467402207,
+      "loss": 1.7823,
+      "theoretical_loss": 4.153265822494111,
+      "tokens_seen": 313720832
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.000914062186559679,
+      "loss": 2.0244,
+      "theoretical_loss": 4.153165469783279,
+      "tokens_seen": 313786368
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009140421263791374,
+      "loss": 1.8247,
+      "theoretical_loss": 4.153065143896655,
+      "tokens_seen": 313851904
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009140220661985959,
+      "loss": 1.9073,
+      "theoretical_loss": 4.152964844821469,
+      "tokens_seen": 313917440
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009140020060180541,
+      "loss": 1.8277,
+      "theoretical_loss": 4.15286457254496,
+      "tokens_seen": 313982976
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009139819458375126,
+      "loss": 1.8345,
+      "theoretical_loss": 4.152764327054376,
+      "tokens_seen": 314048512
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.000913961885656971,
+      "loss": 1.7599,
+      "theoretical_loss": 4.152664108336976,
+      "tokens_seen": 314114048
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009139418254764293,
+      "loss": 1.8354,
+      "theoretical_loss": 4.152563916380026,
+      "tokens_seen": 314179584
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009139217652958877,
+      "loss": 1.818,
+      "theoretical_loss": 4.152463751170799,
+      "tokens_seen": 314245120
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.000913901705115346,
+      "loss": 1.778,
+      "theoretical_loss": 4.152363612696579,
+      "tokens_seen": 314310656
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009138816449348044,
+      "loss": 1.8639,
+      "theoretical_loss": 4.152263500944658,
+      "tokens_seen": 314376192
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009138615847542629,
+      "loss": 1.7843,
+      "theoretical_loss": 4.152163415902336,
+      "tokens_seen": 314441728
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009138415245737211,
+      "loss": 1.6987,
+      "theoretical_loss": 4.152063357556923,
+      "tokens_seen": 314507264
+    },
+    {
+      "epoch": 0.1,
+      "objective/train/advantage_avg": 0.49289390444755554,
+      "objective/train/docs_used": 121169,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.7357332706451416,
+      "objective/train/original_loss": 1.735733151435852,
+      "objective/train/theoretical_loss": 4.151963325895737,
+      "objective/train/tokens_used": 335032800,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24352145195007324,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0505274534225464,
+      "objective/train/weighted_lm_loss": 1.8233038187026978,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.151963325895737,
+      "tokens_seen": 314572800
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009138214643931796,
+      "loss": 1.7403,
+      "theoretical_loss": 4.151963325895737,
+      "tokens_seen": 314572800
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009138014042126379,
+      "loss": 1.7728,
+      "theoretical_loss": 4.151863320906105,
+      "tokens_seen": 314638336
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009137813440320963,
+      "loss": 1.7417,
+      "theoretical_loss": 4.151763342575362,
+      "tokens_seen": 314703872
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009137612838515547,
+      "loss": 1.7573,
+      "theoretical_loss": 4.151663390890851,
+      "tokens_seen": 314769408
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009137412236710131,
+      "loss": 1.8931,
+      "theoretical_loss": 4.151563465839927,
+      "tokens_seen": 314834944
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009137211634904714,
+      "loss": 1.7439,
+      "theoretical_loss": 4.151463567409949,
+      "tokens_seen": 314900480
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009137011033099299,
+      "loss": 1.8137,
+      "theoretical_loss": 4.151363695588287,
+      "tokens_seen": 314966016
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009136810431293881,
+      "loss": 1.7893,
+      "theoretical_loss": 4.151263850362321,
+      "tokens_seen": 315031552
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009136609829488466,
+      "loss": 1.844,
+      "theoretical_loss": 4.151164031719437,
+      "tokens_seen": 315097088
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009136409227683049,
+      "loss": 1.8051,
+      "theoretical_loss": 4.151064239647032,
+      "tokens_seen": 315162624
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009136208625877633,
+      "loss": 1.7564,
+      "theoretical_loss": 4.1509644741325085,
+      "tokens_seen": 315228160
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009136008024072217,
+      "loss": 1.8875,
+      "theoretical_loss": 4.150864735163281,
+      "tokens_seen": 315293696
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009135807422266801,
+      "loss": 1.8228,
+      "theoretical_loss": 4.15076502272677,
+      "tokens_seen": 315359232
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009135606820461384,
+      "loss": 1.7781,
+      "theoretical_loss": 4.150665336810405,
+      "tokens_seen": 315424768
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009135406218655968,
+      "loss": 1.7661,
+      "theoretical_loss": 4.150565677401627,
+      "tokens_seen": 315490304
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009135205616850552,
+      "loss": 1.7718,
+      "theoretical_loss": 4.15046604448788,
+      "tokens_seen": 315555840
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009135005015045136,
+      "loss": 1.7745,
+      "theoretical_loss": 4.150366438056622,
+      "tokens_seen": 315621376
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009134804413239719,
+      "loss": 1.7982,
+      "theoretical_loss": 4.150266858095317,
+      "tokens_seen": 315686912
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009134603811434303,
+      "loss": 1.674,
+      "theoretical_loss": 4.150167304591437,
+      "tokens_seen": 315752448
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009134403209628886,
+      "loss": 1.7676,
+      "theoretical_loss": 4.150067777532464,
+      "tokens_seen": 315817984
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009134202607823471,
+      "loss": 1.7478,
+      "theoretical_loss": 4.149968276905888,
+      "tokens_seen": 315883520
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009134002006018054,
+      "loss": 1.797,
+      "theoretical_loss": 4.149868802699206,
+      "tokens_seen": 315949056
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009133801404212638,
+      "loss": 2.1356,
+      "theoretical_loss": 4.149769354899927,
+      "tokens_seen": 316014592
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009133600802407222,
+      "loss": 1.8429,
+      "theoretical_loss": 4.149669933495565,
+      "tokens_seen": 316080128
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009133400200601806,
+      "loss": 1.7527,
+      "theoretical_loss": 4.149570538473644,
+      "tokens_seen": 316145664
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009133199598796389,
+      "loss": 1.731,
+      "theoretical_loss": 4.149471169821696,
+      "tokens_seen": 316211200
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009132998996990974,
+      "loss": 1.7687,
+      "theoretical_loss": 4.149371827527264,
+      "tokens_seen": 316276736
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009132798395185556,
+      "loss": 1.6522,
+      "theoretical_loss": 4.149272511577895,
+      "tokens_seen": 316342272
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009132597793380141,
+      "loss": 1.7597,
+      "theoretical_loss": 4.149173221961146,
+      "tokens_seen": 316407808
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009132397191574724,
+      "loss": 1.6395,
+      "theoretical_loss": 4.1490739586645855,
+      "tokens_seen": 316473344
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009132196589769308,
+      "loss": 1.87,
+      "theoretical_loss": 4.148974721675787,
+      "tokens_seen": 316538880
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009131995987963892,
+      "loss": 1.7709,
+      "theoretical_loss": 4.148875510982333,
+      "tokens_seen": 316604416
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009131795386158475,
+      "loss": 1.8424,
+      "theoretical_loss": 4.1487763265718165,
+      "tokens_seen": 316669952
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009131594784353059,
+      "loss": 1.7587,
+      "theoretical_loss": 4.148677168431836,
+      "tokens_seen": 316735488
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009131394182547644,
+      "loss": 1.7732,
+      "theoretical_loss": 4.148578036550001,
+      "tokens_seen": 316801024
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009131193580742226,
+      "loss": 1.8576,
+      "theoretical_loss": 4.148478930913927,
+      "tokens_seen": 316866560
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009130992978936811,
+      "loss": 1.7591,
+      "theoretical_loss": 4.148379851511241,
+      "tokens_seen": 316932096
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009130792377131393,
+      "loss": 1.8007,
+      "theoretical_loss": 4.148280798329575,
+      "tokens_seen": 316997632
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009130591775325978,
+      "loss": 1.8282,
+      "theoretical_loss": 4.148181771356571,
+      "tokens_seen": 317063168
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009130391173520562,
+      "loss": 1.7429,
+      "theoretical_loss": 4.14808277057988,
+      "tokens_seen": 317128704
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009130190571715145,
+      "loss": 1.7629,
+      "theoretical_loss": 4.147983795987161,
+      "tokens_seen": 317194240
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009129989969909729,
+      "loss": 1.7354,
+      "theoretical_loss": 4.147884847566079,
+      "tokens_seen": 317259776
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009129789368104314,
+      "loss": 1.7621,
+      "theoretical_loss": 4.147785925304311,
+      "tokens_seen": 317325312
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009129588766298896,
+      "loss": 1.7322,
+      "theoretical_loss": 4.147687029189541,
+      "tokens_seen": 317390848
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009129388164493481,
+      "loss": 1.901,
+      "theoretical_loss": 4.14758815920946,
+      "tokens_seen": 317456384
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009129187562688063,
+      "loss": 1.7886,
+      "theoretical_loss": 4.147489315351769,
+      "tokens_seen": 317521920
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009128986960882648,
+      "loss": 1.6663,
+      "theoretical_loss": 4.147390497604177,
+      "tokens_seen": 317587456
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009128786359077233,
+      "loss": 1.7448,
+      "theoretical_loss": 4.147291705954401,
+      "tokens_seen": 317652992
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009128585757271815,
+      "loss": 1.717,
+      "theoretical_loss": 4.147192940390165,
+      "tokens_seen": 317718528
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00091283851554664,
+      "loss": 1.947,
+      "theoretical_loss": 4.147094200899205,
+      "tokens_seen": 317784064
+    },
+    {
+      "epoch": 0.1,
+      "objective/train/advantage_avg": 0.49450305104255676,
+      "objective/train/docs_used": 122440,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.54131281375885,
+      "objective/train/original_loss": 1.5413126945495605,
+      "objective/train/theoretical_loss": 4.146995487469262,
+      "objective/train/tokens_used": 338309600,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24474811553955078,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0506945848464966,
+      "objective/train/weighted_lm_loss": 1.619470477104187,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 1.0,
+      "theoretical_loss": 4.146995487469262,
+      "tokens_seen": 317849600
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009128184553660983,
+      "loss": 1.7732,
+      "theoretical_loss": 4.146995487469262,
+      "tokens_seen": 317849600
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009127983951855567,
+      "loss": 1.8489,
+      "theoretical_loss": 4.146896800088085,
+      "tokens_seen": 317915136
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009127783350050151,
+      "loss": 1.7643,
+      "theoretical_loss": 4.146798138743433,
+      "tokens_seen": 317980672
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009127582748244735,
+      "loss": 1.707,
+      "theoretical_loss": 4.146699503423074,
+      "tokens_seen": 318046208
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009127382146439318,
+      "loss": 1.6486,
+      "theoretical_loss": 4.1466008941147825,
+      "tokens_seen": 318111744
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009127181544633903,
+      "loss": 1.8341,
+      "theoretical_loss": 4.146502310806342,
+      "tokens_seen": 318177280
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009126980942828485,
+      "loss": 1.6462,
+      "theoretical_loss": 4.146403753485544,
+      "tokens_seen": 318242816
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.000912678034102307,
+      "loss": 1.7509,
+      "theoretical_loss": 4.146305222140187,
+      "tokens_seen": 318308352
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009126579739217653,
+      "loss": 1.7517,
+      "theoretical_loss": 4.146206716758081,
+      "tokens_seen": 318373888
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009126379137412237,
+      "loss": 1.7286,
+      "theoretical_loss": 4.1461082373270415,
+      "tokens_seen": 318439424
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009126178535606821,
+      "loss": 1.8237,
+      "theoretical_loss": 4.146009783834892,
+      "tokens_seen": 318504960
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009125977933801405,
+      "loss": 1.8733,
+      "theoretical_loss": 4.145911356269468,
+      "tokens_seen": 318570496
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009125777331995988,
+      "loss": 1.8255,
+      "theoretical_loss": 4.1458129546186075,
+      "tokens_seen": 318636032
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009125576730190572,
+      "loss": 1.7995,
+      "theoretical_loss": 4.1457145788701615,
+      "tokens_seen": 318701568
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009125376128385156,
+      "loss": 1.7193,
+      "theoretical_loss": 4.145616229011987,
+      "tokens_seen": 318767104
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.000912517552657974,
+      "loss": 1.828,
+      "theoretical_loss": 4.145517905031948,
+      "tokens_seen": 318832640
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009124974924774323,
+      "loss": 1.7877,
+      "theoretical_loss": 4.145419606917921,
+      "tokens_seen": 318898176
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009124774322968907,
+      "loss": 1.9088,
+      "theoretical_loss": 4.145321334657785,
+      "tokens_seen": 318963712
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.000912457372116349,
+      "loss": 1.7268,
+      "theoretical_loss": 4.145223088239432,
+      "tokens_seen": 319029248
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009124373119358075,
+      "loss": 1.801,
+      "theoretical_loss": 4.145124867650759,
+      "tokens_seen": 319094784
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009124172517552658,
+      "loss": 1.7628,
+      "theoretical_loss": 4.145026672879674,
+      "tokens_seen": 319160320
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009123971915747242,
+      "loss": 1.7966,
+      "theoretical_loss": 4.14492850391409,
+      "tokens_seen": 319225856
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009123771313941826,
+      "loss": 1.6315,
+      "theoretical_loss": 4.14483036074193,
+      "tokens_seen": 319291392
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.000912357071213641,
+      "loss": 1.7785,
+      "theoretical_loss": 4.144732243351125,
+      "tokens_seen": 319356928
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009123370110330993,
+      "loss": 1.8006,
+      "theoretical_loss": 4.144634151729615,
+      "tokens_seen": 319422464
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009123169508525578,
+      "loss": 1.7751,
+      "theoretical_loss": 4.144536085865344,
+      "tokens_seen": 319488000
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.000912296890672016,
+      "loss": 1.9455,
+      "theoretical_loss": 4.14443804574627,
+      "tokens_seen": 319553536
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009122768304914745,
+      "loss": 1.7975,
+      "theoretical_loss": 4.144340031360354,
+      "tokens_seen": 319619072
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009122567703109328,
+      "loss": 1.8911,
+      "theoretical_loss": 4.14424204269557,
+      "tokens_seen": 319684608
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009122367101303912,
+      "loss": 1.6413,
+      "theoretical_loss": 4.1441440797398945,
+      "tokens_seen": 319750144
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009122166499498496,
+      "loss": 1.6841,
+      "theoretical_loss": 4.144046142481317,
+      "tokens_seen": 319815680
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009121965897693079,
+      "loss": 1.9622,
+      "theoretical_loss": 4.143948230907832,
+      "tokens_seen": 319881216
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009121765295887663,
+      "loss": 1.7783,
+      "theoretical_loss": 4.1438503450074435,
+      "tokens_seen": 319946752
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009121564694082248,
+      "loss": 1.7984,
+      "theoretical_loss": 4.1437524847681635,
+      "tokens_seen": 320012288
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.000912136409227683,
+      "loss": 1.7974,
+      "theoretical_loss": 4.143654650178012,
+      "tokens_seen": 320077824
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009121163490471415,
+      "loss": 1.8931,
+      "theoretical_loss": 4.143556841225015,
+      "tokens_seen": 320143360
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009120962888665997,
+      "loss": 1.8957,
+      "theoretical_loss": 4.14345905789721,
+      "tokens_seen": 320208896
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009120762286860582,
+      "loss": 1.9032,
+      "theoretical_loss": 4.14336130018264,
+      "tokens_seen": 320274432
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009120561685055166,
+      "loss": 1.6911,
+      "theoretical_loss": 4.143263568069358,
+      "tokens_seen": 320339968
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009120361083249749,
+      "loss": 1.7567,
+      "theoretical_loss": 4.143165861545421,
+      "tokens_seen": 320405504
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009120160481444333,
+      "loss": 1.6795,
+      "theoretical_loss": 4.143068180598902,
+      "tokens_seen": 320471040
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009119959879638918,
+      "loss": 1.7473,
+      "theoretical_loss": 4.142970525217872,
+      "tokens_seen": 320536576
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00091197592778335,
+      "loss": 1.8411,
+      "theoretical_loss": 4.142872895390417,
+      "tokens_seen": 320602112
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009119558676028085,
+      "loss": 1.8678,
+      "theoretical_loss": 4.142775291104629,
+      "tokens_seen": 320667648
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009119358074222667,
+      "loss": 1.7909,
+      "theoretical_loss": 4.142677712348608,
+      "tokens_seen": 320733184
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009119157472417252,
+      "loss": 1.748,
+      "theoretical_loss": 4.14258015911046,
+      "tokens_seen": 320798720
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009118956870611836,
+      "loss": 1.7048,
+      "theoretical_loss": 4.142482631378303,
+      "tokens_seen": 320864256
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009118756268806419,
+      "loss": 1.7556,
+      "theoretical_loss": 4.142385129140259,
+      "tokens_seen": 320929792
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009118555667001003,
+      "loss": 1.6847,
+      "theoretical_loss": 4.142287652384462,
+      "tokens_seen": 320995328
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009118355065195587,
+      "loss": 1.6888,
+      "theoretical_loss": 4.142190201099049,
+      "tokens_seen": 321060864
+    },
+    {
+      "epoch": 0.1,
+      "objective/train/advantage_avg": 0.48887574672698975,
+      "objective/train/docs_used": 123582,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.6981501579284668,
+      "objective/train/original_loss": 1.6981501579284668,
+      "objective/train/theoretical_loss": 4.142092775272169,
+      "objective/train/tokens_used": 341586400,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.23958638310432434,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.050105333328247,
+      "objective/train/weighted_lm_loss": 1.7833449840545654,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.142092775272169,
+      "tokens_seen": 321126400
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.000911815446339017,
+      "loss": 1.7979,
+      "theoretical_loss": 4.142092775272169,
+      "tokens_seen": 321126400
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009117953861584755,
+      "loss": 1.783,
+      "theoretical_loss": 4.141995374891978,
+      "tokens_seen": 321191936
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009117753259779338,
+      "loss": 1.5579,
+      "theoretical_loss": 4.1418979999466385,
+      "tokens_seen": 321257472
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009117552657973922,
+      "loss": 1.656,
+      "theoretical_loss": 4.141800650424322,
+      "tokens_seen": 321323008
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009117352056168505,
+      "loss": 1.7395,
+      "theoretical_loss": 4.141703326313209,
+      "tokens_seen": 321388544
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.000911715145436309,
+      "loss": 1.8497,
+      "theoretical_loss": 4.141606027601485,
+      "tokens_seen": 321454080
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009116950852557673,
+      "loss": 1.7712,
+      "theoretical_loss": 4.141508754277346,
+      "tokens_seen": 321519616
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009116750250752257,
+      "loss": 1.6483,
+      "theoretical_loss": 4.141411506328995,
+      "tokens_seen": 321585152
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.000911654964894684,
+      "loss": 1.8046,
+      "theoretical_loss": 4.141314283744643,
+      "tokens_seen": 321650688
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009116349047141425,
+      "loss": 1.8253,
+      "theoretical_loss": 4.141217086512508,
+      "tokens_seen": 321716224
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009116148445336008,
+      "loss": 1.8028,
+      "theoretical_loss": 4.141119914620817,
+      "tokens_seen": 321781760
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009115947843530592,
+      "loss": 1.7437,
+      "theoretical_loss": 4.141022768057805,
+      "tokens_seen": 321847296
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009115747241725175,
+      "loss": 1.6649,
+      "theoretical_loss": 4.140925646811714,
+      "tokens_seen": 321912832
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.000911554663991976,
+      "loss": 1.7697,
+      "theoretical_loss": 4.140828550870793,
+      "tokens_seen": 321978368
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009115346038114343,
+      "loss": 1.9109,
+      "theoretical_loss": 4.140731480223303,
+      "tokens_seen": 322043904
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009115145436308927,
+      "loss": 1.7703,
+      "theoretical_loss": 4.140634434857506,
+      "tokens_seen": 322109440
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.000911494483450351,
+      "loss": 1.7073,
+      "theoretical_loss": 4.1405374147616785,
+      "tokens_seen": 322174976
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009114744232698094,
+      "loss": 1.7863,
+      "theoretical_loss": 4.140440419924101,
+      "tokens_seen": 322240512
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009114543630892678,
+      "loss": 1.8862,
+      "theoretical_loss": 4.140343450333064,
+      "tokens_seen": 322306048
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009114343029087262,
+      "loss": 1.5811,
+      "theoretical_loss": 4.1402465059768625,
+      "tokens_seen": 322371584
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009114142427281845,
+      "loss": 1.7835,
+      "theoretical_loss": 4.140149586843803,
+      "tokens_seen": 322437120
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.000911394182547643,
+      "loss": 1.7408,
+      "theoretical_loss": 4.140052692922199,
+      "tokens_seen": 322502656
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009113741223671014,
+      "loss": 1.8122,
+      "theoretical_loss": 4.139955824200369,
+      "tokens_seen": 322568192
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009113540621865597,
+      "loss": 1.8127,
+      "theoretical_loss": 4.1398589806666415,
+      "tokens_seen": 322633728
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009113340020060182,
+      "loss": 1.7441,
+      "theoretical_loss": 4.139762162309355,
+      "tokens_seen": 322699264
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009113139418254764,
+      "loss": 1.8992,
+      "theoretical_loss": 4.13966536911685,
+      "tokens_seen": 322764800
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009112938816449349,
+      "loss": 1.7507,
+      "theoretical_loss": 4.139568601077481,
+      "tokens_seen": 322830336
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009112738214643932,
+      "loss": 1.7393,
+      "theoretical_loss": 4.139471858179606,
+      "tokens_seen": 322895872
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009112537612838516,
+      "loss": 1.6676,
+      "theoretical_loss": 4.139375140411592,
+      "tokens_seen": 322961408
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00091123370110331,
+      "loss": 1.7302,
+      "theoretical_loss": 4.139278447761814,
+      "tokens_seen": 323026944
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009112136409227683,
+      "loss": 1.6805,
+      "theoretical_loss": 4.139181780218656,
+      "tokens_seen": 323092480
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009111935807422267,
+      "loss": 1.6964,
+      "theoretical_loss": 4.139085137770507,
+      "tokens_seen": 323158016
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009111735205616852,
+      "loss": 1.7933,
+      "theoretical_loss": 4.138988520405764,
+      "tokens_seen": 323223552
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009111534603811434,
+      "loss": 1.6671,
+      "theoretical_loss": 4.138891928112835,
+      "tokens_seen": 323289088
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009111334002006019,
+      "loss": 1.8928,
+      "theoretical_loss": 4.138795360880132,
+      "tokens_seen": 323354624
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009111133400200601,
+      "loss": 1.6507,
+      "theoretical_loss": 4.138698818696076,
+      "tokens_seen": 323420160
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009110932798395186,
+      "loss": 1.7278,
+      "theoretical_loss": 4.138602301549097,
+      "tokens_seen": 323485696
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.000911073219658977,
+      "loss": 1.7415,
+      "theoretical_loss": 4.138505809427632,
+      "tokens_seen": 323551232
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009110531594784353,
+      "loss": 1.8598,
+      "theoretical_loss": 4.138409342320124,
+      "tokens_seen": 323616768
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009110330992978937,
+      "loss": 1.5982,
+      "theoretical_loss": 4.138312900215024,
+      "tokens_seen": 323682304
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009110130391173522,
+      "loss": 1.7677,
+      "theoretical_loss": 4.138216483100795,
+      "tokens_seen": 323747840
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009109929789368104,
+      "loss": 1.6368,
+      "theoretical_loss": 4.1381200909659,
+      "tokens_seen": 323813376
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009109729187562689,
+      "loss": 1.7068,
+      "theoretical_loss": 4.138023723798817,
+      "tokens_seen": 323878912
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009109528585757271,
+      "loss": 1.7321,
+      "theoretical_loss": 4.137927381588026,
+      "tokens_seen": 323944448
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009109327983951856,
+      "loss": 1.7413,
+      "theoretical_loss": 4.137831064322021,
+      "tokens_seen": 324009984
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.000910912738214644,
+      "loss": 1.8037,
+      "theoretical_loss": 4.1377347719892965,
+      "tokens_seen": 324075520
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009108926780341023,
+      "loss": 1.6356,
+      "theoretical_loss": 4.137638504578359,
+      "tokens_seen": 324141056
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009108726178535607,
+      "loss": 1.7395,
+      "theoretical_loss": 4.137542262077721,
+      "tokens_seen": 324206592
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009108525576730191,
+      "loss": 1.6488,
+      "theoretical_loss": 4.1374460444759045,
+      "tokens_seen": 324272128
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009108324974924774,
+      "loss": 1.7526,
+      "theoretical_loss": 4.137349851761437,
+      "tokens_seen": 324337664
+    },
+    {
+      "epoch": 0.1,
+      "objective/train/advantage_avg": 0.49557074904441833,
+      "objective/train/docs_used": 124782,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.604373812675476,
+      "objective/train/original_loss": 1.6043736934661865,
+      "objective/train/theoretical_loss": 4.137253683922854,
+      "objective/train/tokens_used": 344863200,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.2460704892873764,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0508081912994385,
+      "objective/train/weighted_lm_loss": 1.6858175992965698,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.9048373699188232,
+      "theoretical_loss": 4.137253683922854,
+      "tokens_seen": 324403200
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009108124373119359,
+      "loss": 1.7217,
+      "theoretical_loss": 4.137253683922854,
+      "tokens_seen": 324403200
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009107923771313942,
+      "loss": 1.7176,
+      "theoretical_loss": 4.137157540948699,
+      "tokens_seen": 324468736
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009107723169508526,
+      "loss": 1.7777,
+      "theoretical_loss": 4.137061422827525,
+      "tokens_seen": 324534272
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009107522567703109,
+      "loss": 1.817,
+      "theoretical_loss": 4.136965329547888,
+      "tokens_seen": 324599808
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009107321965897693,
+      "loss": 1.6928,
+      "theoretical_loss": 4.136869261098356,
+      "tokens_seen": 324665344
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009107121364092277,
+      "loss": 1.7311,
+      "theoretical_loss": 4.136773217467502,
+      "tokens_seen": 324730880
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009106920762286861,
+      "loss": 1.863,
+      "theoretical_loss": 4.136677198643908,
+      "tokens_seen": 324796416
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009106720160481444,
+      "loss": 1.7369,
+      "theoretical_loss": 4.136581204616162,
+      "tokens_seen": 324861952
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009106519558676029,
+      "loss": 1.7076,
+      "theoretical_loss": 4.136485235372861,
+      "tokens_seen": 324927488
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009106318956870612,
+      "loss": 1.7839,
+      "theoretical_loss": 4.13638929090261,
+      "tokens_seen": 324993024
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009106118355065196,
+      "loss": 1.6871,
+      "theoretical_loss": 4.13629337119402,
+      "tokens_seen": 325058560
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009105917753259779,
+      "loss": 1.6558,
+      "theoretical_loss": 4.136197476235708,
+      "tokens_seen": 325124096
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009105717151454364,
+      "loss": 1.8245,
+      "theoretical_loss": 4.136101606016304,
+      "tokens_seen": 325189632
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009105516549648947,
+      "loss": 1.7017,
+      "theoretical_loss": 4.136005760524439,
+      "tokens_seen": 325255168
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009105315947843531,
+      "loss": 1.7789,
+      "theoretical_loss": 4.135909939748757,
+      "tokens_seen": 325320704
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009105115346038114,
+      "loss": 1.7688,
+      "theoretical_loss": 4.135814143677906,
+      "tokens_seen": 325386240
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009104914744232698,
+      "loss": 1.7062,
+      "theoretical_loss": 4.135718372300543,
+      "tokens_seen": 325451776
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009104714142427282,
+      "loss": 1.7736,
+      "theoretical_loss": 4.135622625605333,
+      "tokens_seen": 325517312
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009104513540621866,
+      "loss": 1.698,
+      "theoretical_loss": 4.135526903580946,
+      "tokens_seen": 325582848
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009104312938816449,
+      "loss": 1.8322,
+      "theoretical_loss": 4.135431206216061,
+      "tokens_seen": 325648384
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009104112337011034,
+      "loss": 1.7194,
+      "theoretical_loss": 4.135335533499366,
+      "tokens_seen": 325713920
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009103911735205616,
+      "loss": 1.7395,
+      "theoretical_loss": 4.135239885419554,
+      "tokens_seen": 325779456
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009103711133400201,
+      "loss": 1.7968,
+      "theoretical_loss": 4.135144261965327,
+      "tokens_seen": 325844992
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009103510531594784,
+      "loss": 1.666,
+      "theoretical_loss": 4.135048663125394,
+      "tokens_seen": 325910528
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009103309929789368,
+      "loss": 1.8125,
+      "theoretical_loss": 4.134953088888471,
+      "tokens_seen": 325976064
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009103109327983952,
+      "loss": 1.8514,
+      "theoretical_loss": 4.134857539243283,
+      "tokens_seen": 326041600
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009102908726178536,
+      "loss": 1.6628,
+      "theoretical_loss": 4.134762014178559,
+      "tokens_seen": 326107136
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009102708124373119,
+      "loss": 1.8064,
+      "theoretical_loss": 4.13466651368304,
+      "tokens_seen": 326172672
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009102507522567704,
+      "loss": 1.748,
+      "theoretical_loss": 4.134571037745471,
+      "tokens_seen": 326238208
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009102306920762286,
+      "loss": 1.6509,
+      "theoretical_loss": 4.134475586354606,
+      "tokens_seen": 326303744
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009102106318956871,
+      "loss": 1.7845,
+      "theoretical_loss": 4.134380159499204,
+      "tokens_seen": 326369280
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009101905717151455,
+      "loss": 1.783,
+      "theoretical_loss": 4.134284757168037,
+      "tokens_seen": 326434816
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009101705115346038,
+      "loss": 1.6616,
+      "theoretical_loss": 4.134189379349877,
+      "tokens_seen": 326500352
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009101504513540622,
+      "loss": 1.7684,
+      "theoretical_loss": 4.13409402603351,
+      "tokens_seen": 326565888
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009101303911735205,
+      "loss": 1.8304,
+      "theoretical_loss": 4.1339986972077245,
+      "tokens_seen": 326631424
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009101103309929789,
+      "loss": 1.6475,
+      "theoretical_loss": 4.13390339286132,
+      "tokens_seen": 326696960
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009100902708124374,
+      "loss": 1.7013,
+      "theoretical_loss": 4.1338081129831,
+      "tokens_seen": 326762496
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009100702106318956,
+      "loss": 1.7019,
+      "theoretical_loss": 4.1337128575618785,
+      "tokens_seen": 326828032
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009100501504513541,
+      "loss": 1.6178,
+      "theoretical_loss": 4.133617626586475,
+      "tokens_seen": 326893568
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009100300902708125,
+      "loss": 1.6345,
+      "theoretical_loss": 4.1335224200457175,
+      "tokens_seen": 326959104
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009100100300902708,
+      "loss": 1.7228,
+      "theoretical_loss": 4.133427237928439,
+      "tokens_seen": 327024640
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009099899699097292,
+      "loss": 1.7379,
+      "theoretical_loss": 4.133332080223483,
+      "tokens_seen": 327090176
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009099699097291875,
+      "loss": 1.8473,
+      "theoretical_loss": 4.133236946919698,
+      "tokens_seen": 327155712
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009099498495486459,
+      "loss": 2.0025,
+      "theoretical_loss": 4.133141838005942,
+      "tokens_seen": 327221248
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009099297893681044,
+      "loss": 1.7865,
+      "theoretical_loss": 4.133046753471077,
+      "tokens_seen": 327286784
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009099097291875627,
+      "loss": 1.7868,
+      "theoretical_loss": 4.132951693303976,
+      "tokens_seen": 327352320
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009098896690070211,
+      "loss": 1.7852,
+      "theoretical_loss": 4.132856657493516,
+      "tokens_seen": 327417856
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009098696088264795,
+      "loss": 1.8734,
+      "theoretical_loss": 4.132761646028585,
+      "tokens_seen": 327483392
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009098495486459378,
+      "loss": 1.7436,
+      "theoretical_loss": 4.132666658898074,
+      "tokens_seen": 327548928
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009098294884653963,
+      "loss": 1.7704,
+      "theoretical_loss": 4.132571696090885,
+      "tokens_seen": 327614464
+    },
+    {
+      "debugging/Compilability": 1.0,
+      "debugging/distinct-1-grams": 0.7246169962973871,
+      "debugging/entropy-1-grams": 5.13190351820591,
+      "debugging/length": 472.8,
+      "debugging/num_segments": 10,
+      "debugging/raw_token_scores_avg": 0.008663635700941086,
+      "debugging/raw_token_scores_std": 0.023182114586234093,
+      "debugging/score": 0.0033104506818385365,
+      "debugging/score_std": 0.0017105448100298692,
+      "epoch": 0.1,
+      "objective/train/advantage_avg": 0.4913388192653656,
+      "objective/train/docs_used": 126013,
+      "objective/train/instantaneous_batch_size": 32,
+      "objective/train/instantaneous_microbatch_size": 32768,
+      "objective/train/lm_loss": 1.739636778831482,
+      "objective/train/original_loss": 1.7396366596221924,
+      "objective/train/theoretical_loss": 4.132476757595925,
+      "objective/train/tokens_used": 348140000,
+      "objective/train/value_avg": -0.5,
+      "objective/train/value_loss": 0.24195149540901184,
+      "objective/train/value_max": -0.5,
+      "objective/train/value_min": -0.5,
+      "objective/train/value_reward_corr": NaN,
+      "objective/train/value_std": 0.0,
+      "objective/train/weight_avg": 1.0503638982772827,
+      "objective/train/weighted_lm_loss": 1.8274545669555664,
+      "objective/train/weights_max": 1.0512710809707642,
+      "objective/train/weights_min": 0.951229453086853,
+      "theoretical_loss": 4.132476757595925,
+      "tokens_seen": 327680000
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0009098094282848546,
+      "loss": 1.8537,
+      "theoretical_loss": 4.132476757595925,
+      "tokens_seen": 327680000
+    }
+  ],
+  "max_steps": 50354,
+  "num_train_epochs": 9223372036854775807,
+  "total_flos": 1.6722841042944e+17,
+  "trial_name": null,
+  "trial_params": null
+}