diff --git "a/checkpoint-2500/trainer_state.json" "b/checkpoint-2500/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/checkpoint-2500/trainer_state.json"
@@ -0,0 +1,3634 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 4.0,
+  "eval_steps": 100,
+  "global_step": 2500,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 1.4832931827753781,
+      "epoch": 0.016,
+      "grad_norm": 145.0,
+      "learning_rate": 6.000000000000001e-07,
+      "loss": 35.8107,
+      "mean_token_accuracy": 0.5946215284988284,
+      "num_tokens": 338352.0,
+      "step": 10
+    },
+    {
+      "entropy": 1.4921929031610488,
+      "epoch": 0.032,
+      "grad_norm": 139.0,
+      "learning_rate": 1.2666666666666669e-06,
+      "loss": 35.6429,
+      "mean_token_accuracy": 0.5987227037549019,
+      "num_tokens": 671193.0,
+      "step": 20
+    },
+    {
+      "entropy": 1.5176027774810792,
+      "epoch": 0.048,
+      "grad_norm": 120.0,
+      "learning_rate": 1.9333333333333336e-06,
+      "loss": 34.7928,
+      "mean_token_accuracy": 0.6018522759899497,
+      "num_tokens": 1001008.0,
+      "step": 30
+    },
+    {
+      "entropy": 1.6286215644329787,
+      "epoch": 0.064,
+      "grad_norm": 94.0,
+      "learning_rate": 2.6e-06,
+      "loss": 34.3313,
+      "mean_token_accuracy": 0.5992800608277321,
+      "num_tokens": 1331208.0,
+      "step": 40
+    },
+    {
+      "entropy": 1.6960565708577633,
+      "epoch": 0.08,
+      "grad_norm": 70.0,
+      "learning_rate": 3.266666666666667e-06,
+      "loss": 31.8029,
+      "mean_token_accuracy": 0.6178880449384451,
+      "num_tokens": 1667532.0,
+      "step": 50
+    },
+    {
+      "entropy": 1.6291456781327724,
+      "epoch": 0.096,
+      "grad_norm": 46.25,
+      "learning_rate": 3.9333333333333335e-06,
+      "loss": 28.5587,
+      "mean_token_accuracy": 0.6474138081073761,
+      "num_tokens": 2007176.0,
+      "step": 60
+    },
+    {
+      "entropy": 1.6484372481703757,
+      "epoch": 0.112,
+      "grad_norm": 38.75,
+      "learning_rate": 4.600000000000001e-06,
+      "loss": 27.1023,
+      "mean_token_accuracy": 0.6544294429942965,
+      "num_tokens": 2342259.0,
+      "step": 70
+    },
+    {
+      "entropy": 1.6191057547926904,
+      "epoch": 0.128,
+      "grad_norm": 31.625,
+      "learning_rate": 5.2666666666666665e-06,
+      "loss": 25.9215,
+      "mean_token_accuracy": 0.6600364238023758,
+      "num_tokens": 2663987.0,
+      "step": 80
+    },
+    {
+      "entropy": 1.4830621395260095,
+      "epoch": 0.144,
+      "grad_norm": 23.25,
+      "learning_rate": 5.933333333333335e-06,
+      "loss": 23.6014,
+      "mean_token_accuracy": 0.6834562920033932,
+      "num_tokens": 2997824.0,
+      "step": 90
+    },
+    {
+      "entropy": 1.426737917587161,
+      "epoch": 0.16,
+      "grad_norm": 21.25,
+      "learning_rate": 6.600000000000001e-06,
+      "loss": 22.4685,
+      "mean_token_accuracy": 0.6952182710170746,
+      "num_tokens": 3330597.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.16,
+      "eval_biology_entropy": 3.0376483097076417,
+      "eval_biology_loss": 3.875791549682617,
+      "eval_biology_mean_token_accuracy": 0.41260506939888003,
+      "eval_biology_num_tokens": 3330597.0,
+      "eval_biology_runtime": 49.8433,
+      "eval_biology_samples_per_second": 10.031,
+      "eval_biology_steps_per_second": 2.508,
+      "step": 100
+    },
+    {
+      "epoch": 0.16,
+      "eval_chemistry_entropy": 1.7663879008293153,
+      "eval_chemistry_loss": 1.816375732421875,
+      "eval_chemistry_mean_token_accuracy": 0.6312954430580139,
+      "eval_chemistry_num_tokens": 3330597.0,
+      "eval_chemistry_runtime": 61.8081,
+      "eval_chemistry_samples_per_second": 8.09,
+      "eval_chemistry_steps_per_second": 2.022,
+      "step": 100
+    },
+    {
+      "epoch": 0.16,
+      "eval_math_entropy": 1.1983131990432738,
+      "eval_math_loss": 1.485450029373169,
+      "eval_math_mean_token_accuracy": 0.6917262263298035,
+      "eval_math_num_tokens": 3330597.0,
+      "eval_math_runtime": 63.351,
+      "eval_math_samples_per_second": 7.893,
+      "eval_math_steps_per_second": 1.973,
+      "step": 100
+    },
+    {
+      "epoch": 0.16,
+      "eval_physics_entropy": 1.4052113437652587,
+      "eval_physics_loss": 1.3870269060134888,
+      "eval_physics_mean_token_accuracy": 0.69519624376297,
+      "eval_physics_num_tokens": 3330597.0,
+      "eval_physics_runtime": 72.2152,
+      "eval_physics_samples_per_second": 6.924,
+      "eval_physics_steps_per_second": 1.731,
+      "step": 100
+    },
+    {
+      "entropy": 1.4028880976140499,
+      "epoch": 0.176,
+      "grad_norm": 23.125,
+      "learning_rate": 7.266666666666668e-06,
+      "loss": 21.8268,
+      "mean_token_accuracy": 0.6981963120400906,
+      "num_tokens": 3658264.0,
+      "step": 110
+    },
+    {
+      "entropy": 1.3301444135606288,
+      "epoch": 0.192,
+      "grad_norm": 17.375,
+      "learning_rate": 7.933333333333334e-06,
+      "loss": 20.9681,
+      "mean_token_accuracy": 0.7033475987613201,
+      "num_tokens": 3995568.0,
+      "step": 120
+    },
+    {
+      "entropy": 1.2701803509145975,
+      "epoch": 0.208,
+      "grad_norm": 18.25,
+      "learning_rate": 8.6e-06,
+      "loss": 19.9173,
+      "mean_token_accuracy": 0.7150526810437441,
+      "num_tokens": 4321436.0,
+      "step": 130
+    },
+    {
+      "entropy": 1.2350537855178119,
+      "epoch": 0.224,
+      "grad_norm": 16.625,
+      "learning_rate": 9.266666666666667e-06,
+      "loss": 19.5032,
+      "mean_token_accuracy": 0.717908713966608,
+      "num_tokens": 4648491.0,
+      "step": 140
+    },
+    {
+      "entropy": 1.2139144621789455,
+      "epoch": 0.24,
+      "grad_norm": 20.0,
+      "learning_rate": 9.933333333333334e-06,
+      "loss": 19.3976,
+      "mean_token_accuracy": 0.7190249029546976,
+      "num_tokens": 4986175.0,
+      "step": 150
+    },
+    {
+      "entropy": 1.2034844245761633,
+      "epoch": 0.256,
+      "grad_norm": 18.125,
+      "learning_rate": 1.0600000000000002e-05,
+      "loss": 19.0035,
+      "mean_token_accuracy": 0.7240409322082997,
+      "num_tokens": 5329320.0,
+      "step": 160
+    },
+    {
+      "entropy": 1.19539747312665,
+      "epoch": 0.272,
+      "grad_norm": 19.125,
+      "learning_rate": 1.1266666666666668e-05,
+      "loss": 18.9941,
+      "mean_token_accuracy": 0.7230293404310941,
+      "num_tokens": 5658796.0,
+      "step": 170
+    },
+    {
+      "entropy": 1.1726421393454074,
+      "epoch": 0.288,
+      "grad_norm": 16.75,
+      "learning_rate": 1.1933333333333335e-05,
+      "loss": 18.6802,
+      "mean_token_accuracy": 0.7282810874283314,
+      "num_tokens": 5980201.0,
+      "step": 180
+    },
+    {
+      "entropy": 1.16094581335783,
+      "epoch": 0.304,
+      "grad_norm": 20.5,
+      "learning_rate": 1.2600000000000001e-05,
+      "loss": 18.4063,
+      "mean_token_accuracy": 0.7304705370217561,
+      "num_tokens": 6298858.0,
+      "step": 190
+    },
+    {
+      "entropy": 1.1420196149498225,
+      "epoch": 0.32,
+      "grad_norm": 18.375,
+      "learning_rate": 1.3266666666666668e-05,
+      "loss": 18.2034,
+      "mean_token_accuracy": 0.7318542957305908,
+      "num_tokens": 6622798.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.32,
+      "eval_biology_entropy": 2.920998822212219,
+      "eval_biology_loss": 3.363135814666748,
+      "eval_biology_mean_token_accuracy": 0.4486829369068146,
+      "eval_biology_num_tokens": 6622798.0,
+      "eval_biology_runtime": 49.8146,
+      "eval_biology_samples_per_second": 10.037,
+      "eval_biology_steps_per_second": 2.509,
+      "step": 200
+    },
+    {
+      "epoch": 0.32,
+      "eval_chemistry_entropy": 1.4927990851402282,
+      "eval_chemistry_loss": 1.5301882028579712,
+      "eval_chemistry_mean_token_accuracy": 0.6676614291667938,
+      "eval_chemistry_num_tokens": 6622798.0,
+      "eval_chemistry_runtime": 61.7412,
+      "eval_chemistry_samples_per_second": 8.098,
+      "eval_chemistry_steps_per_second": 2.025,
+      "step": 200
+    },
+    {
+      "epoch": 0.32,
+      "eval_math_entropy": 1.087874176979065,
+      "eval_math_loss": 1.2984504699707031,
+      "eval_math_mean_token_accuracy": 0.7132572102546691,
+      "eval_math_num_tokens": 6622798.0,
+      "eval_math_runtime": 63.3429,
+      "eval_math_samples_per_second": 7.894,
+      "eval_math_steps_per_second": 1.973,
+      "step": 200
+    },
+    {
+      "epoch": 0.32,
+      "eval_physics_entropy": 1.1195718150138856,
+      "eval_physics_loss": 1.1357909440994263,
+      "eval_physics_mean_token_accuracy": 0.7332857141494751,
+      "eval_physics_num_tokens": 6622798.0,
+      "eval_physics_runtime": 72.1176,
+      "eval_physics_samples_per_second": 6.933,
+      "eval_physics_steps_per_second": 1.733,
+      "step": 200
+    },
+    {
+      "entropy": 1.1236778676509858,
+      "epoch": 0.336,
+      "grad_norm": 14.25,
+      "learning_rate": 1.3933333333333334e-05,
+      "loss": 17.9874,
+      "mean_token_accuracy": 0.7355770241469145,
+      "num_tokens": 6953611.0,
+      "step": 210
+    },
+    {
+      "entropy": 1.0880652230232954,
+      "epoch": 0.352,
+      "grad_norm": 18.0,
+      "learning_rate": 1.46e-05,
+      "loss": 17.3415,
+      "mean_token_accuracy": 0.7432329464703799,
+      "num_tokens": 7290378.0,
+      "step": 220
+    },
+    {
+      "entropy": 1.0688236601650716,
+      "epoch": 0.368,
+      "grad_norm": 17.875,
+      "learning_rate": 1.5266666666666667e-05,
+      "loss": 17.0985,
+      "mean_token_accuracy": 0.7455301389098168,
+      "num_tokens": 7621082.0,
+      "step": 230
+    },
+    {
+      "entropy": 1.1190654184669255,
+      "epoch": 0.384,
+      "grad_norm": 17.375,
+      "learning_rate": 1.5933333333333336e-05,
+      "loss": 17.8397,
+      "mean_token_accuracy": 0.7339698273688555,
+      "num_tokens": 7955570.0,
+      "step": 240
+    },
+    {
+      "entropy": 1.0598117262125015,
+      "epoch": 0.4,
+      "grad_norm": 18.0,
+      "learning_rate": 1.66e-05,
+      "loss": 16.9636,
+      "mean_token_accuracy": 0.7449962131679058,
+      "num_tokens": 8291049.0,
+      "step": 250
+    },
+    {
+      "entropy": 1.0420986939221621,
+      "epoch": 0.416,
+      "grad_norm": 16.625,
+      "learning_rate": 1.726666666666667e-05,
+      "loss": 16.6593,
+      "mean_token_accuracy": 0.7490336105227471,
+      "num_tokens": 8627310.0,
+      "step": 260
+    },
+    {
+      "entropy": 1.0523861989378929,
+      "epoch": 0.432,
+      "grad_norm": 19.0,
+      "learning_rate": 1.7933333333333333e-05,
+      "loss": 16.8565,
+      "mean_token_accuracy": 0.7475517597049475,
+      "num_tokens": 8958371.0,
+      "step": 270
+    },
+    {
+      "entropy": 1.0093163922429085,
+      "epoch": 0.448,
+      "grad_norm": 16.75,
+      "learning_rate": 1.86e-05,
+      "loss": 16.2024,
+      "mean_token_accuracy": 0.7556655522435903,
+      "num_tokens": 9294388.0,
+      "step": 280
+    },
+    {
+      "entropy": 1.006963584944606,
+      "epoch": 0.464,
+      "grad_norm": 17.875,
+      "learning_rate": 1.926666666666667e-05,
+      "loss": 16.0211,
+      "mean_token_accuracy": 0.7555320005863905,
+      "num_tokens": 9620366.0,
+      "step": 290
+    },
+    {
+      "entropy": 1.0118312349542975,
+      "epoch": 0.48,
+      "grad_norm": 18.25,
+      "learning_rate": 1.9933333333333334e-05,
+      "loss": 16.2777,
+      "mean_token_accuracy": 0.7526854898780584,
+      "num_tokens": 9955431.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.48,
+      "eval_biology_entropy": 2.8107905416488648,
+      "eval_biology_loss": 3.2179975509643555,
+      "eval_biology_mean_token_accuracy": 0.46178966522216797,
+      "eval_biology_num_tokens": 9955431.0,
+      "eval_biology_runtime": 49.8514,
+      "eval_biology_samples_per_second": 10.03,
+      "eval_biology_steps_per_second": 2.507,
+      "step": 300
+    },
+    {
+      "epoch": 0.48,
+      "eval_chemistry_entropy": 1.406659827709198,
+      "eval_chemistry_loss": 1.4287575483322144,
+      "eval_chemistry_mean_token_accuracy": 0.6827028579711915,
+      "eval_chemistry_num_tokens": 9955431.0,
+      "eval_chemistry_runtime": 61.8676,
+      "eval_chemistry_samples_per_second": 8.082,
+      "eval_chemistry_steps_per_second": 2.02,
+      "step": 300
+    },
+    {
+      "epoch": 0.48,
+      "eval_math_entropy": 1.0421963691711427,
+      "eval_math_loss": 1.2314904928207397,
+      "eval_math_mean_token_accuracy": 0.7236453113555908,
+      "eval_math_num_tokens": 9955431.0,
+      "eval_math_runtime": 63.3661,
+      "eval_math_samples_per_second": 7.891,
+      "eval_math_steps_per_second": 1.973,
+      "step": 300
+    },
+    {
+      "epoch": 0.48,
+      "eval_physics_entropy": 1.0351716361045837,
+      "eval_physics_loss": 1.0356038808822632,
+      "eval_physics_mean_token_accuracy": 0.7498980407714844,
+      "eval_physics_num_tokens": 9955431.0,
+      "eval_physics_runtime": 72.1536,
+      "eval_physics_samples_per_second": 6.93,
+      "eval_physics_steps_per_second": 1.732,
+      "step": 300
+    },
+    {
+      "entropy": 1.0259150266647339,
+      "epoch": 0.496,
+      "grad_norm": 15.8125,
+      "learning_rate": 1.9933333333333334e-05,
+      "loss": 16.4439,
+      "mean_token_accuracy": 0.7514661472290755,
+      "num_tokens": 10289883.0,
+      "step": 310
+    },
+    {
+      "entropy": 1.0062489442527294,
+      "epoch": 0.512,
+      "grad_norm": 19.5,
+      "learning_rate": 1.985925925925926e-05,
+      "loss": 16.2084,
+      "mean_token_accuracy": 0.7547577202320099,
+      "num_tokens": 10619468.0,
+      "step": 320
+    },
+    {
+      "entropy": 1.0211657840758561,
+      "epoch": 0.528,
+      "grad_norm": 17.0,
+      "learning_rate": 1.9785185185185187e-05,
+      "loss": 16.3421,
+      "mean_token_accuracy": 0.75145508274436,
+      "num_tokens": 10944904.0,
+      "step": 330
+    },
+    {
+      "entropy": 1.0024307239800692,
+      "epoch": 0.544,
+      "grad_norm": 17.375,
+      "learning_rate": 1.971111111111111e-05,
+      "loss": 16.1168,
+      "mean_token_accuracy": 0.7545490644872188,
+      "num_tokens": 11266409.0,
+      "step": 340
+    },
+    {
+      "entropy": 0.9814366064965725,
+      "epoch": 0.56,
+      "grad_norm": 15.6875,
+      "learning_rate": 1.963703703703704e-05,
+      "loss": 15.6974,
+      "mean_token_accuracy": 0.759765500202775,
+      "num_tokens": 11605544.0,
+      "step": 350
+    },
+    {
+      "entropy": 0.974255072697997,
+      "epoch": 0.576,
+      "grad_norm": 17.625,
+      "learning_rate": 1.9562962962962964e-05,
+      "loss": 15.6108,
+      "mean_token_accuracy": 0.762177936732769,
+      "num_tokens": 11931715.0,
+      "step": 360
+    },
+    {
+      "entropy": 0.9886789865791797,
+      "epoch": 0.592,
+      "grad_norm": 17.75,
+      "learning_rate": 1.948888888888889e-05,
+      "loss": 15.8091,
+      "mean_token_accuracy": 0.7574512097984553,
+      "num_tokens": 12257243.0,
+      "step": 370
+    },
+    {
+      "entropy": 0.9611615493893624,
+      "epoch": 0.608,
+      "grad_norm": 15.5625,
+      "learning_rate": 1.9414814814814817e-05,
+      "loss": 15.4695,
+      "mean_token_accuracy": 0.7621630631387234,
+      "num_tokens": 12583154.0,
+      "step": 380
+    },
+    {
+      "entropy": 0.9780517250299454,
+      "epoch": 0.624,
+      "grad_norm": 16.25,
+      "learning_rate": 1.9340740740740743e-05,
+      "loss": 15.7482,
+      "mean_token_accuracy": 0.7585811104625464,
+      "num_tokens": 12905392.0,
+      "step": 390
+    },
+    {
+      "entropy": 0.9820186741650104,
+      "epoch": 0.64,
+      "grad_norm": 15.6875,
+      "learning_rate": 1.926666666666667e-05,
+      "loss": 15.7619,
+      "mean_token_accuracy": 0.7589227013289929,
+      "num_tokens": 13232198.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.64,
+      "eval_biology_entropy": 2.7366416778564453,
+      "eval_biology_loss": 3.0968964099884033,
+      "eval_biology_mean_token_accuracy": 0.47410999417304994,
+      "eval_biology_num_tokens": 13232198.0,
+      "eval_biology_runtime": 49.8326,
+      "eval_biology_samples_per_second": 10.034,
+      "eval_biology_steps_per_second": 2.508,
+      "step": 400
+    },
+    {
+      "epoch": 0.64,
+      "eval_chemistry_entropy": 1.3320539832115172,
+      "eval_chemistry_loss": 1.3684444427490234,
+      "eval_chemistry_mean_token_accuracy": 0.693352394104004,
+      "eval_chemistry_num_tokens": 13232198.0,
+      "eval_chemistry_runtime": 61.7471,
+      "eval_chemistry_samples_per_second": 8.098,
+      "eval_chemistry_steps_per_second": 2.024,
+      "step": 400
+    },
+    {
+      "epoch": 0.64,
+      "eval_math_entropy": 0.9903823466300964,
+      "eval_math_loss": 1.1989392042160034,
+      "eval_math_mean_token_accuracy": 0.7292539710998536,
+      "eval_math_num_tokens": 13232198.0,
+      "eval_math_runtime": 63.2924,
+      "eval_math_samples_per_second": 7.9,
+      "eval_math_steps_per_second": 1.975,
+      "step": 400
+    },
+    {
+      "epoch": 0.64,
+      "eval_physics_entropy": 0.9698497958183289,
+      "eval_physics_loss": 0.9801211357116699,
+      "eval_physics_mean_token_accuracy": 0.7597138810157776,
+      "eval_physics_num_tokens": 13232198.0,
+      "eval_physics_runtime": 72.1491,
+      "eval_physics_samples_per_second": 6.93,
+      "eval_physics_steps_per_second": 1.733,
+      "step": 400
+    },
+    {
+      "entropy": 0.9678926695138216,
+      "epoch": 0.656,
+      "grad_norm": 17.625,
+      "learning_rate": 1.9192592592592593e-05,
+      "loss": 15.5332,
+      "mean_token_accuracy": 0.7619860425591469,
+      "num_tokens": 13575902.0,
+      "step": 410
+    },
+    {
+      "entropy": 0.9481775723397732,
+      "epoch": 0.672,
+      "grad_norm": 18.875,
+      "learning_rate": 1.911851851851852e-05,
+      "loss": 15.1784,
+      "mean_token_accuracy": 0.7657071366906166,
+      "num_tokens": 13895997.0,
+      "step": 420
+    },
+    {
+      "entropy": 0.9527117524296045,
+      "epoch": 0.688,
+      "grad_norm": 16.125,
+      "learning_rate": 1.9044444444444446e-05,
+      "loss": 15.2982,
+      "mean_token_accuracy": 0.7649280011653901,
+      "num_tokens": 14234888.0,
+      "step": 430
+    },
+    {
+      "entropy": 0.9615373708307743,
+      "epoch": 0.704,
+      "grad_norm": 17.25,
+      "learning_rate": 1.8970370370370372e-05,
+      "loss": 15.5115,
+      "mean_token_accuracy": 0.7601466745138168,
+      "num_tokens": 14567908.0,
+      "step": 440
+    },
+    {
+      "entropy": 0.9485624451190233,
+      "epoch": 0.72,
+      "grad_norm": 18.5,
+      "learning_rate": 1.8896296296296295e-05,
+      "loss": 15.1719,
+      "mean_token_accuracy": 0.7653266172856092,
+      "num_tokens": 14882927.0,
+      "step": 450
+    },
+    {
+      "entropy": 0.9433321110904217,
+      "epoch": 0.736,
+      "grad_norm": 18.375,
+      "learning_rate": 1.8822222222222225e-05,
+      "loss": 15.1447,
+      "mean_token_accuracy": 0.765366930142045,
+      "num_tokens": 15217342.0,
+      "step": 460
+    },
+    {
+      "entropy": 0.9253723874688149,
+      "epoch": 0.752,
+      "grad_norm": 16.75,
+      "learning_rate": 1.874814814814815e-05,
+      "loss": 14.7655,
+      "mean_token_accuracy": 0.7691089898347855,
+      "num_tokens": 15547710.0,
+      "step": 470
+    },
+    {
+      "entropy": 0.9615132443606853,
+      "epoch": 0.768,
+      "grad_norm": 17.375,
+      "learning_rate": 1.8674074074074075e-05,
+      "loss": 15.4996,
+      "mean_token_accuracy": 0.7604613497853279,
+      "num_tokens": 15877177.0,
+      "step": 480
+    },
+    {
+      "entropy": 0.9211561907082796,
+      "epoch": 0.784,
+      "grad_norm": 14.875,
+      "learning_rate": 1.86e-05,
+      "loss": 14.8121,
+      "mean_token_accuracy": 0.7703294314444065,
+      "num_tokens": 16219640.0,
+      "step": 490
+    },
+    {
+      "entropy": 0.939649223536253,
+      "epoch": 0.8,
+      "grad_norm": 16.75,
+      "learning_rate": 1.8525925925925928e-05,
+      "loss": 15.0708,
+      "mean_token_accuracy": 0.7668382901698351,
+      "num_tokens": 16548261.0,
+      "step": 500
+    },
+    {
+      "epoch": 0.8,
+      "eval_biology_entropy": 2.762967661857605,
+      "eval_biology_loss": 3.090928792953491,
+      "eval_biology_mean_token_accuracy": 0.47848794198036193,
+      "eval_biology_num_tokens": 16548261.0,
+      "eval_biology_runtime": 50.0005,
+      "eval_biology_samples_per_second": 10.0,
+      "eval_biology_steps_per_second": 2.5,
+      "step": 500
+    },
+    {
+      "epoch": 0.8,
+      "eval_chemistry_entropy": 1.333887234210968,
+      "eval_chemistry_loss": 1.340030550956726,
+      "eval_chemistry_mean_token_accuracy": 0.6984462032318115,
+      "eval_chemistry_num_tokens": 16548261.0,
+      "eval_chemistry_runtime": 61.3452,
+      "eval_chemistry_samples_per_second": 8.151,
+      "eval_chemistry_steps_per_second": 2.038,
+      "step": 500
+    },
+    {
+      "epoch": 0.8,
+      "eval_math_entropy": 0.9834395766258239,
+      "eval_math_loss": 1.1809327602386475,
+      "eval_math_mean_token_accuracy": 0.7322900333404541,
+      "eval_math_num_tokens": 16548261.0,
+      "eval_math_runtime": 63.2493,
+      "eval_math_samples_per_second": 7.905,
+      "eval_math_steps_per_second": 1.976,
+      "step": 500
+    },
+    {
+      "epoch": 0.8,
+      "eval_physics_entropy": 0.9553510432243347,
+      "eval_physics_loss": 0.948321521282196,
+      "eval_physics_mean_token_accuracy": 0.7648878183364868,
+      "eval_physics_num_tokens": 16548261.0,
+      "eval_physics_runtime": 72.2375,
+      "eval_physics_samples_per_second": 6.922,
+      "eval_physics_steps_per_second": 1.73,
+      "step": 500
+    },
+    {
+      "entropy": 0.9261549940332771,
+      "epoch": 0.816,
+      "grad_norm": 15.875,
+      "learning_rate": 1.8451851851851855e-05,
+      "loss": 14.9311,
+      "mean_token_accuracy": 0.7663508132100105,
+      "num_tokens": 16870533.0,
+      "step": 510
+    },
+    {
+      "entropy": 0.9406310219317675,
+      "epoch": 0.832,
+      "grad_norm": 17.375,
+      "learning_rate": 1.8377777777777778e-05,
+      "loss": 15.084,
+      "mean_token_accuracy": 0.7663453005254268,
+      "num_tokens": 17196371.0,
+      "step": 520
+    },
+    {
+      "entropy": 0.923311448097229,
+      "epoch": 0.848,
+      "grad_norm": 14.875,
+      "learning_rate": 1.8303703703703704e-05,
+      "loss": 14.8032,
+      "mean_token_accuracy": 0.7687445022165775,
+      "num_tokens": 17527066.0,
+      "step": 530
+    },
+    {
+      "entropy": 0.928763821348548,
+      "epoch": 0.864,
+      "grad_norm": 15.3125,
+      "learning_rate": 1.822962962962963e-05,
+      "loss": 14.9482,
+      "mean_token_accuracy": 0.768150870129466,
+      "num_tokens": 17859605.0,
+      "step": 540
+    },
+    {
+      "entropy": 0.9416768305003643,
+      "epoch": 0.88,
+      "grad_norm": 18.125,
+      "learning_rate": 1.8155555555555557e-05,
+      "loss": 15.1468,
+      "mean_token_accuracy": 0.7654529243707657,
+      "num_tokens": 18187315.0,
+      "step": 550
+    },
+    {
+      "entropy": 0.8831863407045603,
+      "epoch": 0.896,
+      "grad_norm": 15.4375,
+      "learning_rate": 1.8081481481481484e-05,
+      "loss": 14.1869,
+      "mean_token_accuracy": 0.7778372533619404,
+      "num_tokens": 18536790.0,
+      "step": 560
+    },
+    {
+      "entropy": 0.9270613711327315,
+      "epoch": 0.912,
+      "grad_norm": 16.875,
+      "learning_rate": 1.800740740740741e-05,
+      "loss": 14.9239,
+      "mean_token_accuracy": 0.7662449024617672,
+      "num_tokens": 18869344.0,
+      "step": 570
+    },
+    {
+      "entropy": 0.9144386522471905,
+      "epoch": 0.928,
+      "grad_norm": 17.5,
+      "learning_rate": 1.7933333333333333e-05,
+      "loss": 14.7368,
+      "mean_token_accuracy": 0.7688823513686657,
+      "num_tokens": 19199834.0,
+      "step": 580
+    },
+    {
+      "entropy": 0.901116743683815,
+      "epoch": 0.944,
+      "grad_norm": 17.125,
+      "learning_rate": 1.785925925925926e-05,
+      "loss": 14.4066,
+      "mean_token_accuracy": 0.7740797866135836,
+      "num_tokens": 19525385.0,
+      "step": 590
+    },
+    {
+      "entropy": 0.9039361864328385,
+      "epoch": 0.96,
+      "grad_norm": 17.25,
+      "learning_rate": 1.7785185185185186e-05,
+      "loss": 14.5915,
+      "mean_token_accuracy": 0.7717896085232496,
+      "num_tokens": 19856973.0,
+      "step": 600
+    },
+    {
+      "epoch": 0.96,
+      "eval_biology_entropy": 2.677727264404297,
+      "eval_biology_loss": 3.0516796112060547,
+      "eval_biology_mean_token_accuracy": 0.48141900229454043,
+      "eval_biology_num_tokens": 19856973.0,
+      "eval_biology_runtime": 50.8661,
+      "eval_biology_samples_per_second": 9.83,
+      "eval_biology_steps_per_second": 2.457,
+      "step": 600
+    },
+    {
+      "epoch": 0.96,
+      "eval_chemistry_entropy": 1.2866523551940918,
+      "eval_chemistry_loss": 1.317436695098877,
+      "eval_chemistry_mean_token_accuracy": 0.7019241213798523,
+      "eval_chemistry_num_tokens": 19856973.0,
+      "eval_chemistry_runtime": 63.0731,
+      "eval_chemistry_samples_per_second": 7.927,
+      "eval_chemistry_steps_per_second": 1.982,
+      "step": 600
+    },
+    {
+      "epoch": 0.96,
+      "eval_math_entropy": 0.9717253613471984,
+      "eval_math_loss": 1.1687127351760864,
+      "eval_math_mean_token_accuracy": 0.7343572587966919,
+      "eval_math_num_tokens": 19856973.0,
+      "eval_math_runtime": 63.3923,
+      "eval_math_samples_per_second": 7.887,
+      "eval_math_steps_per_second": 1.972,
+      "step": 600
+    },
+    {
+      "epoch": 0.96,
+      "eval_physics_entropy": 0.9267163248062134,
+      "eval_physics_loss": 0.9278588891029358,
+      "eval_physics_mean_token_accuracy": 0.768565085887909,
+      "eval_physics_num_tokens": 19856973.0,
+      "eval_physics_runtime": 72.1894,
+      "eval_physics_samples_per_second": 6.926,
+      "eval_physics_steps_per_second": 1.732,
+      "step": 600
+    },
+    {
+      "entropy": 0.9027096323668957,
+      "epoch": 0.976,
+      "grad_norm": 17.25,
+      "learning_rate": 1.7711111111111113e-05,
+      "loss": 14.4739,
+      "mean_token_accuracy": 0.7739130385220051,
+      "num_tokens": 20190600.0,
+      "step": 610
+    },
+    {
+      "entropy": 0.911938888207078,
+      "epoch": 0.992,
+      "grad_norm": 16.25,
+      "learning_rate": 1.763703703703704e-05,
+      "loss": 14.6737,
+      "mean_token_accuracy": 0.7710133928805589,
+      "num_tokens": 20518014.0,
+      "step": 620
+    },
+    {
+      "entropy": 0.8859073698520661,
+      "epoch": 1.008,
+      "grad_norm": 15.6875,
+      "learning_rate": 1.7562962962962962e-05,
+      "loss": 14.139,
+      "mean_token_accuracy": 0.7765034466981888,
+      "num_tokens": 20859215.0,
+      "step": 630
+    },
+    {
+      "entropy": 0.8853944804519415,
+      "epoch": 1.024,
+      "grad_norm": 16.125,
+      "learning_rate": 1.7488888888888892e-05,
+      "loss": 14.2419,
+      "mean_token_accuracy": 0.7756760004907847,
+      "num_tokens": 21186658.0,
+      "step": 640
+    },
+    {
+      "entropy": 0.8869342122226953,
+      "epoch": 1.04,
+      "grad_norm": 15.625,
+      "learning_rate": 1.7414814814814815e-05,
+      "loss": 14.2479,
+      "mean_token_accuracy": 0.7757201492786407,
+      "num_tokens": 21517237.0,
+      "step": 650
+    },
+    {
+      "entropy": 0.8696011204272509,
+      "epoch": 1.056,
+      "grad_norm": 16.0,
+      "learning_rate": 1.7340740740740742e-05,
+      "loss": 13.9805,
+      "mean_token_accuracy": 0.779824410751462,
+      "num_tokens": 21852528.0,
+      "step": 660
+    },
+    {
+      "entropy": 0.8888028047978878,
+      "epoch": 1.072,
+      "grad_norm": 14.75,
+      "learning_rate": 1.726666666666667e-05,
+      "loss": 14.2631,
+      "mean_token_accuracy": 0.775978759303689,
+      "num_tokens": 22192380.0,
+      "step": 670
+    },
+    {
+      "entropy": 0.8862319141626358,
+      "epoch": 1.088,
+      "grad_norm": 18.125,
+      "learning_rate": 1.7192592592592595e-05,
+      "loss": 14.2198,
+      "mean_token_accuracy": 0.776379656419158,
+      "num_tokens": 22517971.0,
+      "step": 680
+    },
+    {
+      "entropy": 0.8709126710891724,
+      "epoch": 1.104,
+      "grad_norm": 14.8125,
+      "learning_rate": 1.711851851851852e-05,
+      "loss": 14.002,
+      "mean_token_accuracy": 0.7780292049050331,
+      "num_tokens": 22848694.0,
+      "step": 690
+    },
+    {
+      "entropy": 0.8840448129922152,
+      "epoch": 1.12,
+      "grad_norm": 16.375,
+      "learning_rate": 1.7044444444444445e-05,
+      "loss": 14.264,
+      "mean_token_accuracy": 0.7752253647893668,
+      "num_tokens": 23183889.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.12,
+      "eval_biology_entropy": 2.627446077346802,
+      "eval_biology_loss": 3.060105085372925,
+      "eval_biology_mean_token_accuracy": 0.4816924409866333,
+      "eval_biology_num_tokens": 23183889.0,
+      "eval_biology_runtime": 50.5624,
+      "eval_biology_samples_per_second": 9.889,
+      "eval_biology_steps_per_second": 2.472,
+      "step": 700
+    },
+    {
+      "epoch": 1.12,
+      "eval_chemistry_entropy": 1.257134593486786,
+      "eval_chemistry_loss": 1.306192398071289,
+      "eval_chemistry_mean_token_accuracy": 0.7035604500770569,
+      "eval_chemistry_num_tokens": 23183889.0,
+      "eval_chemistry_runtime": 61.7362,
+      "eval_chemistry_samples_per_second": 8.099,
+      "eval_chemistry_steps_per_second": 2.025,
+      "step": 700
+    },
+    {
+      "epoch": 1.12,
+      "eval_math_entropy": 0.9475419321060181,
+      "eval_math_loss": 1.162497639656067,
+      "eval_math_mean_token_accuracy": 0.735723156452179,
+      "eval_math_num_tokens": 23183889.0,
+      "eval_math_runtime": 63.3016,
+      "eval_math_samples_per_second": 7.899,
+      "eval_math_steps_per_second": 1.975,
+      "step": 700
+    },
+    {
+      "epoch": 1.12,
+      "eval_physics_entropy": 0.9011278233528137,
+      "eval_physics_loss": 0.9133982062339783,
+      "eval_physics_mean_token_accuracy": 0.7716574244499207,
+      "eval_physics_num_tokens": 23183889.0,
+      "eval_physics_runtime": 72.1196,
+      "eval_physics_samples_per_second": 6.933,
+      "eval_physics_steps_per_second": 1.733,
+      "step": 700
+    },
+    {
+      "entropy": 0.8449788447469473,
+      "epoch": 1.1360000000000001,
+      "grad_norm": 14.5,
+      "learning_rate": 1.697037037037037e-05,
+      "loss": 13.571,
+      "mean_token_accuracy": 0.7838629804551601,
+      "num_tokens": 23523903.0,
+      "step": 710
+    },
+    {
+      "entropy": 0.8750829499214887,
+      "epoch": 1.152,
+      "grad_norm": 18.125,
+      "learning_rate": 1.6896296296296298e-05,
+      "loss": 14.0535,
+      "mean_token_accuracy": 0.7792033236473799,
+      "num_tokens": 23848848.0,
+      "step": 720
+    },
+    {
+      "entropy": 0.8909317035228014,
+      "epoch": 1.168,
+      "grad_norm": 17.125,
+      "learning_rate": 1.6822222222222224e-05,
+      "loss": 14.2108,
+      "mean_token_accuracy": 0.7754395861178637,
+      "num_tokens": 24174589.0,
+      "step": 730
+    },
+    {
+      "entropy": 0.8695099549368024,
+      "epoch": 1.184,
+      "grad_norm": 17.5,
+      "learning_rate": 1.6748148148148147e-05,
+      "loss": 14.0066,
+      "mean_token_accuracy": 0.7783829633146524,
+      "num_tokens": 24495981.0,
+      "step": 740
+    },
+    {
+      "entropy": 0.8525647681206465,
+      "epoch": 1.2,
+      "grad_norm": 17.25,
+      "learning_rate": 1.6674074074074077e-05,
+      "loss": 13.7868,
+      "mean_token_accuracy": 0.7816724233329296,
+      "num_tokens": 24826245.0,
+      "step": 750
+    },
+    {
+      "entropy": 0.8792602054774761,
+      "epoch": 1.216,
+      "grad_norm": 20.25,
+      "learning_rate": 1.66e-05,
+      "loss": 14.0613,
+      "mean_token_accuracy": 0.7764780439436436,
+      "num_tokens": 25147775.0,
+      "step": 760
+    },
+    {
+      "entropy": 0.8663859061896801,
+      "epoch": 1.232,
+      "grad_norm": 17.125,
+      "learning_rate": 1.6525925925925927e-05,
+      "loss": 13.9083,
+      "mean_token_accuracy": 0.7811534900218249,
+      "num_tokens": 25480033.0,
+      "step": 770
+    },
+    {
+      "entropy": 0.8663209840655327,
+      "epoch": 1.248,
+      "grad_norm": 17.25,
+      "learning_rate": 1.6451851851851853e-05,
+      "loss": 13.8854,
+      "mean_token_accuracy": 0.7800497461110354,
+      "num_tokens": 25811978.0,
+      "step": 780
+    },
+    {
+      "entropy": 0.856946163251996,
+      "epoch": 1.264,
+      "grad_norm": 17.75,
+      "learning_rate": 1.637777777777778e-05,
+      "loss": 13.8279,
+      "mean_token_accuracy": 0.7810843765735627,
+      "num_tokens": 26146299.0,
+      "step": 790
+    },
+    {
+      "entropy": 0.8911789767444134,
+      "epoch": 1.28,
+      "grad_norm": 17.0,
+      "learning_rate": 1.6303703703703706e-05,
+      "loss": 14.3363,
+      "mean_token_accuracy": 0.7737658254802227,
+      "num_tokens": 26478207.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.28,
+      "eval_biology_entropy": 2.59397083568573,
+      "eval_biology_loss": 3.0740294456481934,
+      "eval_biology_mean_token_accuracy": 0.4842085497379303,
+      "eval_biology_num_tokens": 26478207.0,
+      "eval_biology_runtime": 49.7902,
+      "eval_biology_samples_per_second": 10.042,
+      "eval_biology_steps_per_second": 2.511,
+      "step": 800
+    },
+    {
+      "epoch": 1.28,
+      "eval_chemistry_entropy": 1.2249236187934875,
+      "eval_chemistry_loss": 1.2970499992370605,
+      "eval_chemistry_mean_token_accuracy": 0.7058695545196533,
+      "eval_chemistry_num_tokens": 26478207.0,
+      "eval_chemistry_runtime": 61.7992,
+      "eval_chemistry_samples_per_second": 8.091,
+      "eval_chemistry_steps_per_second": 2.023,
+      "step": 800
+    },
+    {
+      "epoch": 1.28,
+      "eval_math_entropy": 0.9352578225135804,
+      "eval_math_loss": 1.1583356857299805,
+      "eval_math_mean_token_accuracy": 0.7370887098312378,
+      "eval_math_num_tokens": 26478207.0,
+      "eval_math_runtime": 63.2982,
+      "eval_math_samples_per_second": 7.899,
+      "eval_math_steps_per_second": 1.975,
+      "step": 800
+    },
+    {
+      "epoch": 1.28,
+      "eval_physics_entropy": 0.8803811745643616,
+      "eval_physics_loss": 0.9023361802101135,
+      "eval_physics_mean_token_accuracy": 0.7736921410560608,
+      "eval_physics_num_tokens": 26478207.0,
+      "eval_physics_runtime": 72.1443,
+      "eval_physics_samples_per_second": 6.931,
+      "eval_physics_steps_per_second": 1.733,
+      "step": 800
+    },
+    {
+      "entropy": 0.8500199999660254,
+      "epoch": 1.296,
+      "grad_norm": 17.375,
+      "learning_rate": 1.622962962962963e-05,
+      "loss": 13.6729,
+      "mean_token_accuracy": 0.7830070056021213,
+      "num_tokens": 26806477.0,
+      "step": 810
+    },
+    {
+      "entropy": 0.8865886203944683,
+      "epoch": 1.312,
+      "grad_norm": 18.0,
+      "learning_rate": 1.6155555555555556e-05,
+      "loss": 14.1733,
+      "mean_token_accuracy": 0.7760876413434744,
+      "num_tokens": 27137528.0,
+      "step": 820
+    },
+    {
+      "entropy": 0.8618463430553674,
+      "epoch": 1.328,
+      "grad_norm": 18.75,
+      "learning_rate": 1.6081481481481482e-05,
+      "loss": 13.8875,
+      "mean_token_accuracy": 0.7796577278524637,
+      "num_tokens": 27465859.0,
+      "step": 830
+    },
+    {
+      "entropy": 0.8727495713159442,
+      "epoch": 1.3439999999999999,
+      "grad_norm": 18.0,
+      "learning_rate": 1.600740740740741e-05,
+      "loss": 14.0654,
+      "mean_token_accuracy": 0.7770151857286691,
+      "num_tokens": 27795568.0,
+      "step": 840
+    },
+    {
+      "entropy": 0.8694380177184939,
+      "epoch": 1.3599999999999999,
+      "grad_norm": 17.125,
+      "learning_rate": 1.5933333333333336e-05,
+      "loss": 13.9903,
+      "mean_token_accuracy": 0.7778833784162998,
+      "num_tokens": 28129988.0,
+      "step": 850
+    },
+    {
+      "entropy": 0.8656781645491719,
+      "epoch": 1.376,
+      "grad_norm": 18.875,
+      "learning_rate": 1.5859259259259262e-05,
+      "loss": 13.8313,
+      "mean_token_accuracy": 0.7801834944635629,
+      "num_tokens": 28465903.0,
+      "step": 860
+    },
+    {
+      "entropy": 0.8433871390298009,
+      "epoch": 1.392,
+      "grad_norm": 17.875,
+      "learning_rate": 1.5785185185185185e-05,
+      "loss": 13.5092,
+      "mean_token_accuracy": 0.7850066836923361,
+      "num_tokens": 28801554.0,
+      "step": 870
+    },
+    {
+      "entropy": 0.8494528220966459,
+      "epoch": 1.408,
+      "grad_norm": 16.875,
+      "learning_rate": 1.571111111111111e-05,
+      "loss": 13.7082,
+      "mean_token_accuracy": 0.7817110113799572,
+      "num_tokens": 29125114.0,
+      "step": 880
+    },
+    {
+      "entropy": 0.88326512016356,
+      "epoch": 1.424,
+      "grad_norm": 16.125,
+      "learning_rate": 1.5637037037037038e-05,
+      "loss": 14.2467,
+      "mean_token_accuracy": 0.7743587661534548,
+      "num_tokens": 29454334.0,
+      "step": 890
+    },
+    {
+      "entropy": 0.8315953454002738,
+      "epoch": 1.44,
+      "grad_norm": 16.25,
+      "learning_rate": 1.5562962962962965e-05,
+      "loss": 13.3551,
+      "mean_token_accuracy": 0.7856587562710047,
+      "num_tokens": 29789253.0,
+      "step": 900
+    },
+    {
+      "epoch": 1.44,
+      "eval_biology_entropy": 2.6302443017959596,
+      "eval_biology_loss": 3.0371100902557373,
+      "eval_biology_mean_token_accuracy": 0.48449889016151426,
+      "eval_biology_num_tokens": 29789253.0,
+      "eval_biology_runtime": 49.9234,
+      "eval_biology_samples_per_second": 10.015,
+      "eval_biology_steps_per_second": 2.504,
+      "step": 900
+    },
+    {
+      "epoch": 1.44,
+      "eval_chemistry_entropy": 1.228422354221344,
+      "eval_chemistry_loss": 1.2853810787200928,
+      "eval_chemistry_mean_token_accuracy": 0.7073793506622315,
+      "eval_chemistry_num_tokens": 29789253.0,
+      "eval_chemistry_runtime": 61.8217,
+      "eval_chemistry_samples_per_second": 8.088,
+      "eval_chemistry_steps_per_second": 2.022,
+      "step": 900
+    },
+    {
+      "epoch": 1.44,
+      "eval_math_entropy": 0.9464052295684815,
+      "eval_math_loss": 1.1523010730743408,
+      "eval_math_mean_token_accuracy": 0.7370671148300171,
+      "eval_math_num_tokens": 29789253.0,
+      "eval_math_runtime": 63.3892,
+      "eval_math_samples_per_second": 7.888,
+      "eval_math_steps_per_second": 1.972,
+      "step": 900
+    },
+    {
+      "epoch": 1.44,
+      "eval_physics_entropy": 0.8797389287948608,
+      "eval_physics_loss": 0.8921938538551331,
+      "eval_physics_mean_token_accuracy": 0.7752533774375916,
+      "eval_physics_num_tokens": 29789253.0,
+      "eval_physics_runtime": 72.2272,
+      "eval_physics_samples_per_second": 6.923,
+      "eval_physics_steps_per_second": 1.731,
+      "step": 900
+    },
+    {
+      "entropy": 0.8377421669661999,
+      "epoch": 1.456,
+      "grad_norm": 14.75,
+      "learning_rate": 1.548888888888889e-05,
+      "loss": 13.5018,
+      "mean_token_accuracy": 0.7858919285237789,
+      "num_tokens": 30128576.0,
+      "step": 910
+    },
+    {
+      "entropy": 0.8895138014107943,
+      "epoch": 1.472,
+      "grad_norm": 18.5,
+      "learning_rate": 1.5414814814814814e-05,
+      "loss": 14.3264,
+      "mean_token_accuracy": 0.7742519076913595,
+      "num_tokens": 30452719.0,
+      "step": 920
+    },
+    {
+      "entropy": 0.8661323908716441,
+      "epoch": 1.488,
+      "grad_norm": 16.375,
+      "learning_rate": 1.5340740740740744e-05,
+      "loss": 13.9071,
+      "mean_token_accuracy": 0.7790522363036871,
+      "num_tokens": 30782343.0,
+      "step": 930
+    },
+    {
+      "entropy": 0.8495663637295365,
+      "epoch": 1.504,
+      "grad_norm": 15.3125,
+      "learning_rate": 1.5266666666666667e-05,
+      "loss": 13.6016,
+      "mean_token_accuracy": 0.7836384274065494,
+      "num_tokens": 31113238.0,
+      "step": 940
+    },
+    {
+      "entropy": 0.8703195974230766,
+      "epoch": 1.52,
+      "grad_norm": 17.25,
+      "learning_rate": 1.5192592592592594e-05,
+      "loss": 13.9447,
+      "mean_token_accuracy": 0.7785820949822664,
+      "num_tokens": 31444589.0,
+      "step": 950
+    },
+    {
+      "entropy": 0.8683951802551746,
+      "epoch": 1.536,
+      "grad_norm": 18.25,
+      "learning_rate": 1.5118518518518519e-05,
+      "loss": 14.0464,
+      "mean_token_accuracy": 0.7769809190183878,
+      "num_tokens": 31774137.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.8442707929760218,
+      "epoch": 1.552,
+      "grad_norm": 17.5,
+      "learning_rate": 1.5044444444444445e-05,
+      "loss": 13.4756,
+      "mean_token_accuracy": 0.7855445120483637,
+      "num_tokens": 32101776.0,
+      "step": 970
+    },
+    {
+      "entropy": 0.8429634341970086,
+      "epoch": 1.568,
+      "grad_norm": 16.75,
+      "learning_rate": 1.497037037037037e-05,
+      "loss": 13.5629,
+      "mean_token_accuracy": 0.7842523407191038,
+      "num_tokens": 32428021.0,
+      "step": 980
+    },
+    {
+      "entropy": 0.8717077055945992,
+      "epoch": 1.584,
+      "grad_norm": 17.25,
+      "learning_rate": 1.4896296296296298e-05,
+      "loss": 14.057,
+      "mean_token_accuracy": 0.7751756615936756,
+      "num_tokens": 32752735.0,
+      "step": 990
+    },
+    {
+      "entropy": 0.8670839730650186,
+      "epoch": 1.6,
+      "grad_norm": 16.75,
+      "learning_rate": 1.4822222222222225e-05,
+      "loss": 13.8733,
+      "mean_token_accuracy": 0.7796571251004935,
+      "num_tokens": 33075822.0,
+      "step": 1000
+    },
+    {
+      "epoch": 1.6,
+      "eval_biology_entropy": 2.5688937005996704,
+      "eval_biology_loss": 3.027909994125366,
+      "eval_biology_mean_token_accuracy": 0.4863655905723572,
+      "eval_biology_num_tokens": 33075822.0,
+      "eval_biology_runtime": 49.9141,
+      "eval_biology_samples_per_second": 10.017,
+      "eval_biology_steps_per_second": 2.504,
+      "step": 1000
+    },
+    {
+      "epoch": 1.6,
+      "eval_chemistry_entropy": 1.2045018038749695,
+      "eval_chemistry_loss": 1.281038761138916,
+      "eval_chemistry_mean_token_accuracy": 0.707553605556488,
+      "eval_chemistry_num_tokens": 33075822.0,
+      "eval_chemistry_runtime": 61.4373,
+      "eval_chemistry_samples_per_second": 8.138,
+      "eval_chemistry_steps_per_second": 2.035,
+      "step": 1000
+    },
+    {
+      "epoch": 1.6,
+      "eval_math_entropy": 0.9304646430015564,
+      "eval_math_loss": 1.1519391536712646,
+      "eval_math_mean_token_accuracy": 0.7375589380264282,
+      "eval_math_num_tokens": 33075822.0,
+      "eval_math_runtime": 63.2339,
+      "eval_math_samples_per_second": 7.907,
+      "eval_math_steps_per_second": 1.977,
+      "step": 1000
+    },
+    {
+      "epoch": 1.6,
+      "eval_physics_entropy": 0.8599038424491883,
+      "eval_physics_loss": 0.885993480682373,
+      "eval_physics_mean_token_accuracy": 0.7767008581161499,
+      "eval_physics_num_tokens": 33075822.0,
+      "eval_physics_runtime": 72.1075,
+      "eval_physics_samples_per_second": 6.934,
+      "eval_physics_steps_per_second": 1.734,
+      "step": 1000
+    },
+    {
+      "entropy": 0.8235349113121628,
+      "epoch": 1.616,
+      "grad_norm": 17.5,
+      "learning_rate": 1.474814814814815e-05,
+      "loss": 13.3243,
+      "mean_token_accuracy": 0.7873352337628603,
+      "num_tokens": 33411850.0,
+      "step": 1010
+    },
+    {
+      "entropy": 0.8450564581900835,
+      "epoch": 1.6320000000000001,
+      "grad_norm": 17.375,
+      "learning_rate": 1.4674074074074076e-05,
+      "loss": 13.5454,
+      "mean_token_accuracy": 0.7839825380593538,
+      "num_tokens": 33739456.0,
+      "step": 1020
+    },
+    {
+      "entropy": 0.8137065678834915,
+      "epoch": 1.6480000000000001,
+      "grad_norm": 16.625,
+      "learning_rate": 1.46e-05,
+      "loss": 13.0726,
+      "mean_token_accuracy": 0.7894206911325454,
+      "num_tokens": 34072022.0,
+      "step": 1030
+    },
+    {
+      "entropy": 0.8465166810899973,
+      "epoch": 1.6640000000000001,
+      "grad_norm": 18.625,
+      "learning_rate": 1.4525925925925927e-05,
+      "loss": 13.565,
+      "mean_token_accuracy": 0.7846160337328911,
+      "num_tokens": 34398959.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.8240113854408264,
+      "epoch": 1.6800000000000002,
+      "grad_norm": 16.125,
+      "learning_rate": 1.4451851851851852e-05,
+      "loss": 13.3644,
+      "mean_token_accuracy": 0.786401529237628,
+      "num_tokens": 34732284.0,
+      "step": 1050
+    },
+    {
+      "entropy": 0.8502342702820898,
+      "epoch": 1.696,
+      "grad_norm": 17.625,
+      "learning_rate": 1.4377777777777779e-05,
+      "loss": 13.5728,
+      "mean_token_accuracy": 0.7834321312606335,
+      "num_tokens": 35053712.0,
+      "step": 1060
+    },
+    {
+      "entropy": 0.8552145214751363,
+      "epoch": 1.712,
+      "grad_norm": 18.375,
+      "learning_rate": 1.4303703703703703e-05,
+      "loss": 13.7943,
+      "mean_token_accuracy": 0.7798704143613577,
+      "num_tokens": 35393088.0,
+      "step": 1070
+    },
+    {
+      "entropy": 0.8368038963526487,
+      "epoch": 1.728,
+      "grad_norm": 17.0,
+      "learning_rate": 1.4229629629629632e-05,
+      "loss": 13.42,
+      "mean_token_accuracy": 0.7853118624538183,
+      "num_tokens": 35729007.0,
+      "step": 1080
+    },
+    {
+      "entropy": 0.8420057747513056,
+      "epoch": 1.744,
+      "grad_norm": 17.75,
+      "learning_rate": 1.4155555555555556e-05,
+      "loss": 13.5284,
+      "mean_token_accuracy": 0.7845802444964647,
+      "num_tokens": 36072489.0,
+      "step": 1090
+    },
+    {
+      "entropy": 0.8349423199892044,
+      "epoch": 1.76,
+      "grad_norm": 16.875,
+      "learning_rate": 1.4081481481481483e-05,
+      "loss": 13.3871,
+      "mean_token_accuracy": 0.7846417732536792,
+      "num_tokens": 36398285.0,
+      "step": 1100
+    },
+    {
+      "epoch": 1.76,
+      "eval_biology_entropy": 2.555653926849365,
+      "eval_biology_loss": 3.0194432735443115,
+      "eval_biology_mean_token_accuracy": 0.4884138941764832,
+      "eval_biology_num_tokens": 36398285.0,
+      "eval_biology_runtime": 49.8041,
+      "eval_biology_samples_per_second": 10.039,
+      "eval_biology_steps_per_second": 2.51,
+      "step": 1100
+    },
+    {
+      "epoch": 1.76,
+      "eval_chemistry_entropy": 1.1974089312553406,
+      "eval_chemistry_loss": 1.2756012678146362,
+      "eval_chemistry_mean_token_accuracy": 0.7093247275352478,
+      "eval_chemistry_num_tokens": 36398285.0,
+      "eval_chemistry_runtime": 62.0097,
+      "eval_chemistry_samples_per_second": 8.063,
+      "eval_chemistry_steps_per_second": 2.016,
+      "step": 1100
+    },
+    {
+      "epoch": 1.76,
+      "eval_math_entropy": 0.9289093766212463,
+      "eval_math_loss": 1.1466560363769531,
+      "eval_math_mean_token_accuracy": 0.7382361373901367,
+      "eval_math_num_tokens": 36398285.0,
+      "eval_math_runtime": 63.3872,
+      "eval_math_samples_per_second": 7.888,
+      "eval_math_steps_per_second": 1.972,
+      "step": 1100
+    },
+    {
+      "epoch": 1.76,
+      "eval_physics_entropy": 0.855881242275238,
+      "eval_physics_loss": 0.880148708820343,
+      "eval_physics_mean_token_accuracy": 0.7775014510154724,
+      "eval_physics_num_tokens": 36398285.0,
+      "eval_physics_runtime": 72.1346,
+      "eval_physics_samples_per_second": 6.931,
+      "eval_physics_steps_per_second": 1.733,
+      "step": 1100
+    },
+    {
+      "entropy": 0.8676729390397668,
+      "epoch": 1.776,
+      "grad_norm": 16.875,
+      "learning_rate": 1.400740740740741e-05,
+      "loss": 13.9824,
+      "mean_token_accuracy": 0.7772907704114914,
+      "num_tokens": 36730479.0,
+      "step": 1110
+    },
+    {
+      "entropy": 0.8294270180165768,
+      "epoch": 1.792,
+      "grad_norm": 20.625,
+      "learning_rate": 1.3933333333333334e-05,
+      "loss": 13.3857,
+      "mean_token_accuracy": 0.7853363569825887,
+      "num_tokens": 37057425.0,
+      "step": 1120
+    },
+    {
+      "entropy": 0.8251689115539194,
+      "epoch": 1.808,
+      "grad_norm": 17.125,
+      "learning_rate": 1.385925925925926e-05,
+      "loss": 13.2245,
+      "mean_token_accuracy": 0.7889013431966305,
+      "num_tokens": 37395449.0,
+      "step": 1130
+    },
+    {
+      "entropy": 0.8390438333153725,
+      "epoch": 1.8239999999999998,
+      "grad_norm": 17.0,
+      "learning_rate": 1.3785185185185186e-05,
+      "loss": 13.4561,
+      "mean_token_accuracy": 0.7853912770748138,
+      "num_tokens": 37729535.0,
+      "step": 1140
+    },
+    {
+      "entropy": 0.836634048819542,
+      "epoch": 1.8399999999999999,
+      "grad_norm": 17.625,
+      "learning_rate": 1.3711111111111112e-05,
+      "loss": 13.4197,
+      "mean_token_accuracy": 0.7858145259320736,
+      "num_tokens": 38057839.0,
+      "step": 1150
+    },
+    {
+      "entropy": 0.8325331646949052,
+      "epoch": 1.8559999999999999,
+      "grad_norm": 16.875,
+      "learning_rate": 1.3637037037037037e-05,
+      "loss": 13.4691,
+      "mean_token_accuracy": 0.78513167463243,
+      "num_tokens": 38398033.0,
+      "step": 1160
+    },
+    {
+      "entropy": 0.8332853283733129,
+      "epoch": 1.8719999999999999,
+      "grad_norm": 16.375,
+      "learning_rate": 1.3562962962962965e-05,
+      "loss": 13.3377,
+      "mean_token_accuracy": 0.7850610390305519,
+      "num_tokens": 38729203.0,
+      "step": 1170
+    },
+    {
+      "entropy": 0.8507293021306396,
+      "epoch": 1.888,
+      "grad_norm": 17.75,
+      "learning_rate": 1.3488888888888888e-05,
+      "loss": 13.7269,
+      "mean_token_accuracy": 0.7809833358973265,
+      "num_tokens": 39058791.0,
+      "step": 1180
+    },
+    {
+      "entropy": 0.8589327406138182,
+      "epoch": 1.904,
+      "grad_norm": 17.625,
+      "learning_rate": 1.3414814814814817e-05,
+      "loss": 13.9451,
+      "mean_token_accuracy": 0.7795876618474722,
+      "num_tokens": 39386170.0,
+      "step": 1190
+    },
+    {
+      "entropy": 0.8757910108193755,
+      "epoch": 1.92,
+      "grad_norm": 17.25,
+      "learning_rate": 1.3340740740740741e-05,
+      "loss": 13.9202,
+      "mean_token_accuracy": 0.7796670190989972,
+      "num_tokens": 39710381.0,
+      "step": 1200
+    },
+    {
+      "epoch": 1.92,
+      "eval_biology_entropy": 2.5806034650802614,
+      "eval_biology_loss": 2.9976675510406494,
+      "eval_biology_mean_token_accuracy": 0.489067396402359,
+      "eval_biology_num_tokens": 39710381.0,
+      "eval_biology_runtime": 49.9313,
+      "eval_biology_samples_per_second": 10.014,
+      "eval_biology_steps_per_second": 2.503,
+      "step": 1200
+    },
+    {
+      "epoch": 1.92,
+      "eval_chemistry_entropy": 1.189872570991516,
+      "eval_chemistry_loss": 1.2703872919082642,
+      "eval_chemistry_mean_token_accuracy": 0.7104025254249573,
+      "eval_chemistry_num_tokens": 39710381.0,
+      "eval_chemistry_runtime": 61.7636,
+      "eval_chemistry_samples_per_second": 8.095,
+      "eval_chemistry_steps_per_second": 2.024,
+      "step": 1200
+    },
+    {
+      "epoch": 1.92,
+      "eval_math_entropy": 0.9202042121887207,
+      "eval_math_loss": 1.1432040929794312,
+      "eval_math_mean_token_accuracy": 0.7393214902877807,
+      "eval_math_num_tokens": 39710381.0,
+      "eval_math_runtime": 63.3516,
+      "eval_math_samples_per_second": 7.892,
+      "eval_math_steps_per_second": 1.973,
+      "step": 1200
+    },
+    {
+      "epoch": 1.92,
+      "eval_physics_entropy": 0.8482071080207825,
+      "eval_physics_loss": 0.8750180602073669,
+      "eval_physics_mean_token_accuracy": 0.7784125237464905,
+      "eval_physics_num_tokens": 39710381.0,
+      "eval_physics_runtime": 72.1428,
+      "eval_physics_samples_per_second": 6.931,
+      "eval_physics_steps_per_second": 1.733,
+      "step": 1200
+    },
+    {
+      "entropy": 0.8378945112228393,
+      "epoch": 1.936,
+      "grad_norm": 17.375,
+      "learning_rate": 1.3266666666666668e-05,
+      "loss": 13.5373,
+      "mean_token_accuracy": 0.7823467884212733,
+      "num_tokens": 40043354.0,
+      "step": 1210
+    },
+    {
+      "entropy": 0.8021389935165644,
+      "epoch": 1.952,
+      "grad_norm": 16.5,
+      "learning_rate": 1.3192592592592594e-05,
+      "loss": 12.8276,
+      "mean_token_accuracy": 0.7932522356510162,
+      "num_tokens": 40375790.0,
+      "step": 1220
+    },
+    {
+      "entropy": 0.8322626123204827,
+      "epoch": 1.968,
+      "grad_norm": 16.875,
+      "learning_rate": 1.311851851851852e-05,
+      "loss": 13.4748,
+      "mean_token_accuracy": 0.7855609647929669,
+      "num_tokens": 40705451.0,
+      "step": 1230
+    },
+    {
+      "entropy": 0.8527568828314542,
+      "epoch": 1.984,
+      "grad_norm": 18.375,
+      "learning_rate": 1.3044444444444446e-05,
+      "loss": 13.6777,
+      "mean_token_accuracy": 0.7815342791378498,
+      "num_tokens": 41040313.0,
+      "step": 1240
+    },
+    {
+      "entropy": 0.8515846587717533,
+      "epoch": 2.0,
+      "grad_norm": 16.75,
+      "learning_rate": 1.297037037037037e-05,
+      "loss": 13.6678,
+      "mean_token_accuracy": 0.7835123162716627,
+      "num_tokens": 41373608.0,
+      "step": 1250
+    },
+    {
+      "entropy": 0.8105791673064232,
+      "epoch": 2.016,
+      "grad_norm": 14.75,
+      "learning_rate": 1.2896296296296299e-05,
+      "loss": 12.94,
+      "mean_token_accuracy": 0.7908968936651946,
+      "num_tokens": 41708226.0,
+      "step": 1260
+    },
+    {
+      "entropy": 0.810917510278523,
+      "epoch": 2.032,
+      "grad_norm": 16.875,
+      "learning_rate": 1.2822222222222222e-05,
+      "loss": 13.0378,
+      "mean_token_accuracy": 0.7905050247907639,
+      "num_tokens": 42045433.0,
+      "step": 1270
+    },
+    {
+      "entropy": 0.8347248645499349,
+      "epoch": 2.048,
+      "grad_norm": 18.0,
+      "learning_rate": 1.274814814814815e-05,
+      "loss": 13.5033,
+      "mean_token_accuracy": 0.7835712507367134,
+      "num_tokens": 42372694.0,
+      "step": 1280
+    },
+    {
+      "entropy": 0.8173987654969096,
+      "epoch": 2.064,
+      "grad_norm": 16.125,
+      "learning_rate": 1.2674074074074075e-05,
+      "loss": 13.044,
+      "mean_token_accuracy": 0.7898124691098929,
+      "num_tokens": 42710240.0,
+      "step": 1290
+    },
+    {
+      "entropy": 0.8316412966698408,
+      "epoch": 2.08,
+      "grad_norm": 16.25,
+      "learning_rate": 1.2600000000000001e-05,
+      "loss": 13.3883,
+      "mean_token_accuracy": 0.7858571786433458,
+      "num_tokens": 43037844.0,
+      "step": 1300
+    },
+    {
+      "epoch": 2.08,
+      "eval_biology_entropy": 2.5521722202301027,
+      "eval_biology_loss": 3.0149927139282227,
+      "eval_biology_mean_token_accuracy": 0.4876684684753418,
+      "eval_biology_num_tokens": 43037844.0,
+      "eval_biology_runtime": 49.814,
+      "eval_biology_samples_per_second": 10.037,
+      "eval_biology_steps_per_second": 2.509,
+      "step": 1300
+    },
+    {
+      "epoch": 2.08,
+      "eval_chemistry_entropy": 1.176895384788513,
+      "eval_chemistry_loss": 1.271460771560669,
+      "eval_chemistry_mean_token_accuracy": 0.709957902431488,
+      "eval_chemistry_num_tokens": 43037844.0,
+      "eval_chemistry_runtime": 61.7814,
+      "eval_chemistry_samples_per_second": 8.093,
+      "eval_chemistry_steps_per_second": 2.023,
+      "step": 1300
+    },
+    {
+      "epoch": 2.08,
+      "eval_math_entropy": 0.9160779042243957,
+      "eval_math_loss": 1.144756555557251,
+      "eval_math_mean_token_accuracy": 0.7386934194564819,
+      "eval_math_num_tokens": 43037844.0,
+      "eval_math_runtime": 63.3256,
+      "eval_math_samples_per_second": 7.896,
+      "eval_math_steps_per_second": 1.974,
+      "step": 1300
+    },
+    {
+      "epoch": 2.08,
+      "eval_physics_entropy": 0.8399050388336181,
+      "eval_physics_loss": 0.8716227412223816,
+      "eval_physics_mean_token_accuracy": 0.7787612590789795,
+      "eval_physics_num_tokens": 43037844.0,
+      "eval_physics_runtime": 72.3388,
+      "eval_physics_samples_per_second": 6.912,
+      "eval_physics_steps_per_second": 1.728,
+      "step": 1300
+    },
+    {
+      "entropy": 0.8293728087097406,
+      "epoch": 2.096,
+      "grad_norm": 17.75,
+      "learning_rate": 1.2525925925925928e-05,
+      "loss": 13.2913,
+      "mean_token_accuracy": 0.7866353880614042,
+      "num_tokens": 43368675.0,
+      "step": 1310
+    },
+    {
+      "entropy": 0.8099473301321268,
+      "epoch": 2.112,
+      "grad_norm": 16.625,
+      "learning_rate": 1.2451851851851853e-05,
+      "loss": 13.0447,
+      "mean_token_accuracy": 0.7900471687316895,
+      "num_tokens": 43699544.0,
+      "step": 1320
+    },
+    {
+      "entropy": 0.8236053828150034,
+      "epoch": 2.128,
+      "grad_norm": 17.5,
+      "learning_rate": 1.237777777777778e-05,
+      "loss": 13.1993,
+      "mean_token_accuracy": 0.7866601809859276,
+      "num_tokens": 44012632.0,
+      "step": 1330
+    },
+    {
+      "entropy": 0.8101952638477087,
+      "epoch": 2.144,
+      "grad_norm": 16.125,
+      "learning_rate": 1.2303703703703704e-05,
+      "loss": 13.1048,
+      "mean_token_accuracy": 0.7884116806089878,
+      "num_tokens": 44349732.0,
+      "step": 1340
+    },
+    {
+      "entropy": 0.8528481485322118,
+      "epoch": 2.16,
+      "grad_norm": 17.375,
+      "learning_rate": 1.222962962962963e-05,
+      "loss": 13.706,
+      "mean_token_accuracy": 0.7812714777886868,
+      "num_tokens": 44680410.0,
+      "step": 1350
+    },
+    {
+      "entropy": 0.8150178231298923,
+      "epoch": 2.176,
+      "grad_norm": 16.125,
+      "learning_rate": 1.2155555555555555e-05,
+      "loss": 13.0736,
+      "mean_token_accuracy": 0.789815029129386,
+      "num_tokens": 45017922.0,
+      "step": 1360
+    },
+    {
+      "entropy": 0.8112508501857519,
+      "epoch": 2.192,
+      "grad_norm": 16.75,
+      "learning_rate": 1.2081481481481484e-05,
+      "loss": 12.962,
+      "mean_token_accuracy": 0.7903204344213008,
+      "num_tokens": 45348179.0,
+      "step": 1370
+    },
+    {
+      "entropy": 0.8060232829302549,
+      "epoch": 2.208,
+      "grad_norm": 18.75,
+      "learning_rate": 1.2007407407407408e-05,
+      "loss": 12.9808,
+      "mean_token_accuracy": 0.7912364609539508,
+      "num_tokens": 45678397.0,
+      "step": 1380
+    },
+    {
+      "entropy": 0.801847311295569,
+      "epoch": 2.224,
+      "grad_norm": 15.4375,
+      "learning_rate": 1.1933333333333335e-05,
+      "loss": 12.9152,
+      "mean_token_accuracy": 0.7912111356854439,
+      "num_tokens": 46010052.0,
+      "step": 1390
+    },
+    {
+      "entropy": 0.8080146696418524,
+      "epoch": 2.24,
+      "grad_norm": 16.375,
+      "learning_rate": 1.185925925925926e-05,
+      "loss": 12.9823,
+      "mean_token_accuracy": 0.790325066447258,
+      "num_tokens": 46347116.0,
+      "step": 1400
+    },
+    {
+      "epoch": 2.24,
+      "eval_biology_entropy": 2.5362069902420044,
+      "eval_biology_loss": 3.014728307723999,
+      "eval_biology_mean_token_accuracy": 0.489130656003952,
+      "eval_biology_num_tokens": 46347116.0,
+      "eval_biology_runtime": 49.8869,
+      "eval_biology_samples_per_second": 10.023,
+      "eval_biology_steps_per_second": 2.506,
+      "step": 1400
+    },
+    {
+      "epoch": 2.24,
+      "eval_chemistry_entropy": 1.1747996592521668,
+      "eval_chemistry_loss": 1.2685794830322266,
+      "eval_chemistry_mean_token_accuracy": 0.7108006677627563,
+      "eval_chemistry_num_tokens": 46347116.0,
+      "eval_chemistry_runtime": 61.8867,
+      "eval_chemistry_samples_per_second": 8.079,
+      "eval_chemistry_steps_per_second": 2.02,
+      "step": 1400
+    },
+    {
+      "epoch": 2.24,
+      "eval_math_entropy": 0.9145697908401489,
+      "eval_math_loss": 1.1440415382385254,
+      "eval_math_mean_token_accuracy": 0.7391237573623657,
+      "eval_math_num_tokens": 46347116.0,
+      "eval_math_runtime": 63.3857,
+      "eval_math_samples_per_second": 7.888,
+      "eval_math_steps_per_second": 1.972,
+      "step": 1400
+    },
+    {
+      "epoch": 2.24,
+      "eval_physics_entropy": 0.8387198600769042,
+      "eval_physics_loss": 0.8690110445022583,
+      "eval_physics_mean_token_accuracy": 0.7796805229187012,
+      "eval_physics_num_tokens": 46347116.0,
+      "eval_physics_runtime": 72.3681,
+      "eval_physics_samples_per_second": 6.909,
+      "eval_physics_steps_per_second": 1.727,
+      "step": 1400
+    },
+    {
+      "entropy": 0.8407515358179808,
+      "epoch": 2.2560000000000002,
+      "grad_norm": 17.5,
+      "learning_rate": 1.1785185185185186e-05,
+      "loss": 13.5299,
+      "mean_token_accuracy": 0.7844579003751277,
+      "num_tokens": 46673074.0,
+      "step": 1410
+    },
+    {
+      "entropy": 0.7986599007621408,
+      "epoch": 2.2720000000000002,
+      "grad_norm": 17.125,
+      "learning_rate": 1.1711111111111113e-05,
+      "loss": 12.801,
+      "mean_token_accuracy": 0.7931681673973798,
+      "num_tokens": 47009386.0,
+      "step": 1420
+    },
+    {
+      "entropy": 0.832692601531744,
+      "epoch": 2.288,
+      "grad_norm": 17.625,
+      "learning_rate": 1.1637037037037037e-05,
+      "loss": 13.3743,
+      "mean_token_accuracy": 0.7854725748300553,
+      "num_tokens": 47344231.0,
+      "step": 1430
+    },
+    {
+      "entropy": 0.822732993401587,
+      "epoch": 2.304,
+      "grad_norm": 18.25,
+      "learning_rate": 1.1562962962962964e-05,
+      "loss": 13.2472,
+      "mean_token_accuracy": 0.787301842868328,
+      "num_tokens": 47658392.0,
+      "step": 1440
+    },
+    {
+      "entropy": 0.816081466525793,
+      "epoch": 2.32,
+      "grad_norm": 15.75,
+      "learning_rate": 1.1488888888888889e-05,
+      "loss": 13.0771,
+      "mean_token_accuracy": 0.7877950664609671,
+      "num_tokens": 47995015.0,
+      "step": 1450
+    },
+    {
+      "entropy": 0.7912156611680985,
+      "epoch": 2.336,
+      "grad_norm": 16.5,
+      "learning_rate": 1.1414814814814817e-05,
+      "loss": 12.7775,
+      "mean_token_accuracy": 0.7924822468310595,
+      "num_tokens": 48336708.0,
+      "step": 1460
+    },
+    {
+      "entropy": 0.7956573387607933,
+      "epoch": 2.352,
+      "grad_norm": 16.0,
+      "learning_rate": 1.1340740740740742e-05,
+      "loss": 12.6835,
+      "mean_token_accuracy": 0.7935895614326001,
+      "num_tokens": 48665311.0,
+      "step": 1470
+    },
+    {
+      "entropy": 0.8168220078572631,
+      "epoch": 2.368,
+      "grad_norm": 17.0,
+      "learning_rate": 1.1266666666666668e-05,
+      "loss": 13.1569,
+      "mean_token_accuracy": 0.7885483611375094,
+      "num_tokens": 49005334.0,
+      "step": 1480
+    },
+    {
+      "entropy": 0.8288948491215706,
+      "epoch": 2.384,
+      "grad_norm": 18.25,
+      "learning_rate": 1.1192592592592593e-05,
+      "loss": 13.3051,
+      "mean_token_accuracy": 0.7858261972665787,
+      "num_tokens": 49331936.0,
+      "step": 1490
+    },
+    {
+      "entropy": 0.8256178490817547,
+      "epoch": 2.4,
+      "grad_norm": 17.125,
+      "learning_rate": 1.111851851851852e-05,
+      "loss": 13.2133,
+      "mean_token_accuracy": 0.7874985858798027,
+      "num_tokens": 49671048.0,
+      "step": 1500
+    },
+    {
+      "epoch": 2.4,
+      "eval_biology_entropy": 2.5136468687057496,
+      "eval_biology_loss": 3.0059711933135986,
+      "eval_biology_mean_token_accuracy": 0.48898950242996214,
+      "eval_biology_num_tokens": 49671048.0,
+      "eval_biology_runtime": 49.8803,
+      "eval_biology_samples_per_second": 10.024,
+      "eval_biology_steps_per_second": 2.506,
+      "step": 1500
+    },
+    {
+      "epoch": 2.4,
+      "eval_chemistry_entropy": 1.1712419180870055,
+      "eval_chemistry_loss": 1.2678245306015015,
+      "eval_chemistry_mean_token_accuracy": 0.7101301984786987,
+      "eval_chemistry_num_tokens": 49671048.0,
+      "eval_chemistry_runtime": 61.3799,
+      "eval_chemistry_samples_per_second": 8.146,
+      "eval_chemistry_steps_per_second": 2.036,
+      "step": 1500
+    },
+    {
+      "epoch": 2.4,
+      "eval_math_entropy": 0.9104172115325928,
+      "eval_math_loss": 1.1439919471740723,
+      "eval_math_mean_token_accuracy": 0.7392911777496338,
+      "eval_math_num_tokens": 49671048.0,
+      "eval_math_runtime": 63.275,
+      "eval_math_samples_per_second": 7.902,
+      "eval_math_steps_per_second": 1.976,
+      "step": 1500
+    },
+    {
+      "epoch": 2.4,
+      "eval_physics_entropy": 0.8380388784408569,
+      "eval_physics_loss": 0.8667259812355042,
+      "eval_physics_mean_token_accuracy": 0.7799293994903564,
+      "eval_physics_num_tokens": 49671048.0,
+      "eval_physics_runtime": 72.3034,
+      "eval_physics_samples_per_second": 6.915,
+      "eval_physics_steps_per_second": 1.729,
+      "step": 1500
+    },
+    {
+      "entropy": 0.8068066079169511,
+      "epoch": 2.416,
+      "grad_norm": 16.75,
+      "learning_rate": 1.1044444444444444e-05,
+      "loss": 13.0195,
+      "mean_token_accuracy": 0.7899820879101753,
+      "num_tokens": 50018182.0,
+      "step": 1510
+    },
+    {
+      "entropy": 0.8188721906393767,
+      "epoch": 2.432,
+      "grad_norm": 18.5,
+      "learning_rate": 1.0970370370370371e-05,
+      "loss": 13.1238,
+      "mean_token_accuracy": 0.789657450094819,
+      "num_tokens": 50341187.0,
+      "step": 1520
+    },
+    {
+      "entropy": 0.8350767683237791,
+      "epoch": 2.448,
+      "grad_norm": 16.375,
+      "learning_rate": 1.0896296296296298e-05,
+      "loss": 13.4868,
+      "mean_token_accuracy": 0.7843202859163284,
+      "num_tokens": 50667019.0,
+      "step": 1530
+    },
+    {
+      "entropy": 0.8286553211510181,
+      "epoch": 2.464,
+      "grad_norm": 18.0,
+      "learning_rate": 1.0822222222222222e-05,
+      "loss": 13.2374,
+      "mean_token_accuracy": 0.787796426564455,
+      "num_tokens": 50991113.0,
+      "step": 1540
+    },
+    {
+      "entropy": 0.8140665043145419,
+      "epoch": 2.48,
+      "grad_norm": 17.25,
+      "learning_rate": 1.074814814814815e-05,
+      "loss": 13.1668,
+      "mean_token_accuracy": 0.7878118209540844,
+      "num_tokens": 51319624.0,
+      "step": 1550
+    },
+    {
+      "entropy": 0.8287572022527456,
+      "epoch": 2.496,
+      "grad_norm": 16.375,
+      "learning_rate": 1.0674074074074074e-05,
+      "loss": 13.2657,
+      "mean_token_accuracy": 0.7874793462455273,
+      "num_tokens": 51654633.0,
+      "step": 1560
+    },
+    {
+      "entropy": 0.8008298767730594,
+      "epoch": 2.512,
+      "grad_norm": 18.875,
+      "learning_rate": 1.0600000000000002e-05,
+      "loss": 12.9202,
+      "mean_token_accuracy": 0.791350769251585,
+      "num_tokens": 51980051.0,
+      "step": 1570
+    },
+    {
+      "entropy": 0.8046136885881424,
+      "epoch": 2.528,
+      "grad_norm": 16.25,
+      "learning_rate": 1.0525925925925927e-05,
+      "loss": 12.933,
+      "mean_token_accuracy": 0.7911055952310562,
+      "num_tokens": 52316109.0,
+      "step": 1580
+    },
+    {
+      "entropy": 0.8215494090691209,
+      "epoch": 2.544,
+      "grad_norm": 18.75,
+      "learning_rate": 1.0451851851851853e-05,
+      "loss": 13.2795,
+      "mean_token_accuracy": 0.786406721919775,
+      "num_tokens": 52634528.0,
+      "step": 1590
+    },
+    {
+      "entropy": 0.8231798518449068,
+      "epoch": 2.56,
+      "grad_norm": 18.125,
+      "learning_rate": 1.0377777777777778e-05,
+      "loss": 13.0478,
+      "mean_token_accuracy": 0.7896040949970484,
+      "num_tokens": 52964035.0,
+      "step": 1600
+    },
+    {
+      "epoch": 2.56,
+      "eval_biology_entropy": 2.51357483959198,
+      "eval_biology_loss": 3.008274793624878,
+      "eval_biology_mean_token_accuracy": 0.4892403426170349,
+      "eval_biology_num_tokens": 52964035.0,
+      "eval_biology_runtime": 50.0462,
+      "eval_biology_samples_per_second": 9.991,
+      "eval_biology_steps_per_second": 2.498,
+      "step": 1600
+    },
+    {
+      "epoch": 2.56,
+      "eval_chemistry_entropy": 1.1565793409347533,
+      "eval_chemistry_loss": 1.2655391693115234,
+      "eval_chemistry_mean_token_accuracy": 0.7113845334053039,
+      "eval_chemistry_num_tokens": 52964035.0,
+      "eval_chemistry_runtime": 61.9092,
+      "eval_chemistry_samples_per_second": 8.076,
+      "eval_chemistry_steps_per_second": 2.019,
+      "step": 1600
+    },
+    {
+      "epoch": 2.56,
+      "eval_math_entropy": 0.902139217376709,
+      "eval_math_loss": 1.1428073644638062,
+      "eval_math_mean_token_accuracy": 0.739846918106079,
+      "eval_math_num_tokens": 52964035.0,
+      "eval_math_runtime": 63.492,
+      "eval_math_samples_per_second": 7.875,
+      "eval_math_steps_per_second": 1.969,
+      "step": 1600
+    },
+    {
+      "epoch": 2.56,
+      "eval_physics_entropy": 0.8253551514148713,
+      "eval_physics_loss": 0.8642763495445251,
+      "eval_physics_mean_token_accuracy": 0.7805017833709716,
+      "eval_physics_num_tokens": 52964035.0,
+      "eval_physics_runtime": 72.994,
+      "eval_physics_samples_per_second": 6.85,
+      "eval_physics_steps_per_second": 1.712,
+      "step": 1600
+    },
+    {
+      "entropy": 0.7979077056050301,
+      "epoch": 2.576,
+      "grad_norm": 18.375,
+      "learning_rate": 1.0303703703703705e-05,
+      "loss": 12.8472,
+      "mean_token_accuracy": 0.7914734333753586,
+      "num_tokens": 53293345.0,
+      "step": 1610
+    },
+    {
+      "entropy": 0.8134908404201269,
+      "epoch": 2.592,
+      "grad_norm": 14.8125,
+      "learning_rate": 1.0229629629629631e-05,
+      "loss": 13.0572,
+      "mean_token_accuracy": 0.7907055057585239,
+      "num_tokens": 53620896.0,
+      "step": 1620
+    },
+    {
+      "entropy": 0.7859407916665078,
+      "epoch": 2.608,
+      "grad_norm": 16.375,
+      "learning_rate": 1.0155555555555556e-05,
+      "loss": 12.7158,
+      "mean_token_accuracy": 0.7942901607602835,
+      "num_tokens": 53950989.0,
+      "step": 1630
+    },
+    {
+      "entropy": 0.8090131890028716,
+      "epoch": 2.624,
+      "grad_norm": 16.75,
+      "learning_rate": 1.0081481481481484e-05,
+      "loss": 13.0312,
+      "mean_token_accuracy": 0.7899593055248261,
+      "num_tokens": 54282732.0,
+      "step": 1640
+    },
+    {
+      "entropy": 0.8099012725055218,
+      "epoch": 2.64,
+      "grad_norm": 17.625,
+      "learning_rate": 1.0007407407407407e-05,
+      "loss": 13.0171,
+      "mean_token_accuracy": 0.7907546170055866,
+      "num_tokens": 54609748.0,
+      "step": 1650
+    },
+    {
+      "entropy": 0.8474382936954499,
+      "epoch": 2.656,
+      "grad_norm": 17.875,
+      "learning_rate": 9.933333333333334e-06,
+      "loss": 13.6484,
+      "mean_token_accuracy": 0.7823971558362246,
+      "num_tokens": 54953583.0,
+      "step": 1660
+    },
+    {
+      "entropy": 0.8184938356280327,
+      "epoch": 2.672,
+      "grad_norm": 17.75,
+      "learning_rate": 9.85925925925926e-06,
+      "loss": 13.1878,
+      "mean_token_accuracy": 0.7879120871424675,
+      "num_tokens": 55281945.0,
+      "step": 1670
+    },
+    {
+      "entropy": 0.7947592413052916,
+      "epoch": 2.6879999999999997,
+      "grad_norm": 17.625,
+      "learning_rate": 9.785185185185187e-06,
+      "loss": 12.6828,
+      "mean_token_accuracy": 0.7949017755687237,
+      "num_tokens": 55612391.0,
+      "step": 1680
+    },
+    {
+      "entropy": 0.8030332267284394,
+      "epoch": 2.7039999999999997,
+      "grad_norm": 16.0,
+      "learning_rate": 9.711111111111111e-06,
+      "loss": 13.0333,
+      "mean_token_accuracy": 0.7908818315714597,
+      "num_tokens": 55949791.0,
+      "step": 1690
+    },
+    {
+      "entropy": 0.813567541912198,
+      "epoch": 2.7199999999999998,
+      "grad_norm": 19.25,
+      "learning_rate": 9.637037037037038e-06,
+      "loss": 12.9476,
+      "mean_token_accuracy": 0.7907637499272824,
+      "num_tokens": 56282204.0,
+      "step": 1700
+    },
+    {
+      "epoch": 2.7199999999999998,
+      "eval_biology_entropy": 2.4613250551223755,
+      "eval_biology_loss": 3.027848958969116,
+      "eval_biology_mean_token_accuracy": 0.48847622537612917,
+      "eval_biology_num_tokens": 56282204.0,
+      "eval_biology_runtime": 50.4799,
+      "eval_biology_samples_per_second": 9.905,
+      "eval_biology_steps_per_second": 2.476,
+      "step": 1700
+    },
+    {
+      "epoch": 2.7199999999999998,
+      "eval_chemistry_entropy": 1.1412961044311523,
+      "eval_chemistry_loss": 1.2647969722747803,
+      "eval_chemistry_mean_token_accuracy": 0.7117968564033508,
+      "eval_chemistry_num_tokens": 56282204.0,
+      "eval_chemistry_runtime": 62.0012,
+      "eval_chemistry_samples_per_second": 8.064,
+      "eval_chemistry_steps_per_second": 2.016,
+      "step": 1700
+    },
+    {
+      "epoch": 2.7199999999999998,
+      "eval_math_entropy": 0.8986919198036194,
+      "eval_math_loss": 1.14143705368042,
+      "eval_math_mean_token_accuracy": 0.7399654779434204,
+      "eval_math_num_tokens": 56282204.0,
+      "eval_math_runtime": 63.4321,
+      "eval_math_samples_per_second": 7.882,
+      "eval_math_steps_per_second": 1.971,
+      "step": 1700
+    },
+    {
+      "epoch": 2.7199999999999998,
+      "eval_physics_entropy": 0.8178641319274902,
+      "eval_physics_loss": 0.862618625164032,
+      "eval_physics_mean_token_accuracy": 0.7811606640815735,
+      "eval_physics_num_tokens": 56282204.0,
+      "eval_physics_runtime": 72.6927,
+      "eval_physics_samples_per_second": 6.878,
+      "eval_physics_steps_per_second": 1.72,
+      "step": 1700
+    },
+    {
+      "entropy": 0.833752753958106,
+      "epoch": 2.7359999999999998,
+      "grad_norm": 17.625,
+      "learning_rate": 9.562962962962965e-06,
+      "loss": 13.5456,
+      "mean_token_accuracy": 0.7830413796007634,
+      "num_tokens": 56605478.0,
+      "step": 1710
+    },
+    {
+      "entropy": 0.8262990560382605,
+      "epoch": 2.752,
+      "grad_norm": 16.5,
+      "learning_rate": 9.48888888888889e-06,
+      "loss": 13.2637,
+      "mean_token_accuracy": 0.787689670175314,
+      "num_tokens": 56935874.0,
+      "step": 1720
+    },
+    {
+      "entropy": 0.8006830558180809,
+      "epoch": 2.768,
+      "grad_norm": 18.0,
+      "learning_rate": 9.414814814814816e-06,
+      "loss": 12.8185,
+      "mean_token_accuracy": 0.7925874296575784,
+      "num_tokens": 57270552.0,
+      "step": 1730
+    },
+    {
+      "entropy": 0.8151656309142709,
+      "epoch": 2.784,
+      "grad_norm": 17.875,
+      "learning_rate": 9.34074074074074e-06,
+      "loss": 13.0998,
+      "mean_token_accuracy": 0.788911472633481,
+      "num_tokens": 57602968.0,
+      "step": 1740
+    },
+    {
+      "entropy": 0.8569422919303179,
+      "epoch": 2.8,
+      "grad_norm": 18.25,
+      "learning_rate": 9.266666666666667e-06,
+      "loss": 13.8572,
+      "mean_token_accuracy": 0.7788108222186565,
+      "num_tokens": 57925886.0,
+      "step": 1750
+    },
+    {
+      "entropy": 0.8001469722017646,
+      "epoch": 2.816,
+      "grad_norm": 16.75,
+      "learning_rate": 9.192592592592594e-06,
+      "loss": 12.6895,
+      "mean_token_accuracy": 0.7940869968384504,
+      "num_tokens": 58262960.0,
+      "step": 1760
+    },
+    {
+      "entropy": 0.8144458586350083,
+      "epoch": 2.832,
+      "grad_norm": 17.125,
+      "learning_rate": 9.118518518518518e-06,
+      "loss": 13.258,
+      "mean_token_accuracy": 0.7859390754252672,
+      "num_tokens": 58586324.0,
+      "step": 1770
+    },
+    {
+      "entropy": 0.8187485463917256,
+      "epoch": 2.848,
+      "grad_norm": 16.25,
+      "learning_rate": 9.044444444444445e-06,
+      "loss": 13.0923,
+      "mean_token_accuracy": 0.7889409240335226,
+      "num_tokens": 58912949.0,
+      "step": 1780
+    },
+    {
+      "entropy": 0.8013842988759279,
+      "epoch": 2.864,
+      "grad_norm": 15.9375,
+      "learning_rate": 8.970370370370372e-06,
+      "loss": 12.9237,
+      "mean_token_accuracy": 0.7916461985558272,
+      "num_tokens": 59252494.0,
+      "step": 1790
+    },
+    {
+      "entropy": 0.8138244587928056,
+      "epoch": 2.88,
+      "grad_norm": 20.625,
+      "learning_rate": 8.896296296296298e-06,
+      "loss": 13.081,
+      "mean_token_accuracy": 0.7896058142185212,
+      "num_tokens": 59574951.0,
+      "step": 1800
+    },
+    {
+      "epoch": 2.88,
+      "eval_biology_entropy": 2.517832468032837,
+      "eval_biology_loss": 2.9910948276519775,
+      "eval_biology_mean_token_accuracy": 0.491009064912796,
+      "eval_biology_num_tokens": 59574951.0,
+      "eval_biology_runtime": 50.0281,
+      "eval_biology_samples_per_second": 9.994,
+      "eval_biology_steps_per_second": 2.499,
+      "step": 1800
+    },
+    {
+      "epoch": 2.88,
+      "eval_chemistry_entropy": 1.1583227319717406,
+      "eval_chemistry_loss": 1.2615896463394165,
+      "eval_chemistry_mean_token_accuracy": 0.7119167141914368,
+      "eval_chemistry_num_tokens": 59574951.0,
+      "eval_chemistry_runtime": 61.944,
+      "eval_chemistry_samples_per_second": 8.072,
+      "eval_chemistry_steps_per_second": 2.018,
+      "step": 1800
+    },
+    {
+      "epoch": 2.88,
+      "eval_math_entropy": 0.9031689348220825,
+      "eval_math_loss": 1.1404438018798828,
+      "eval_math_mean_token_accuracy": 0.7399563612937927,
+      "eval_math_num_tokens": 59574951.0,
+      "eval_math_runtime": 63.8736,
+      "eval_math_samples_per_second": 7.828,
+      "eval_math_steps_per_second": 1.957,
+      "step": 1800
+    },
+    {
+      "epoch": 2.88,
+      "eval_physics_entropy": 0.8247539691925049,
+      "eval_physics_loss": 0.8607794046401978,
+      "eval_physics_mean_token_accuracy": 0.7810482344627381,
+      "eval_physics_num_tokens": 59574951.0,
+      "eval_physics_runtime": 72.4938,
+      "eval_physics_samples_per_second": 6.897,
+      "eval_physics_steps_per_second": 1.724,
+      "step": 1800
+    },
+    {
+      "entropy": 0.8088404158130288,
+      "epoch": 2.896,
+      "grad_norm": 17.5,
+      "learning_rate": 8.822222222222223e-06,
+      "loss": 12.9999,
+      "mean_token_accuracy": 0.7911344520747662,
+      "num_tokens": 59903274.0,
+      "step": 1810
+    },
+    {
+      "entropy": 0.8048945292830467,
+      "epoch": 2.912,
+      "grad_norm": 17.5,
+      "learning_rate": 8.74814814814815e-06,
+      "loss": 12.9083,
+      "mean_token_accuracy": 0.7916154999285936,
+      "num_tokens": 60234203.0,
+      "step": 1820
+    },
+    {
+      "entropy": 0.7997408710420132,
+      "epoch": 2.928,
+      "grad_norm": 17.625,
+      "learning_rate": 8.674074074074074e-06,
+      "loss": 12.8469,
+      "mean_token_accuracy": 0.7927624244242907,
+      "num_tokens": 60559123.0,
+      "step": 1830
+    },
+    {
+      "entropy": 0.799671963043511,
+      "epoch": 2.944,
+      "grad_norm": 19.5,
+      "learning_rate": 8.6e-06,
+      "loss": 12.8479,
+      "mean_token_accuracy": 0.7929738946259022,
+      "num_tokens": 60893423.0,
+      "step": 1840
+    },
+    {
+      "entropy": 0.8103774160146713,
+      "epoch": 2.96,
+      "grad_norm": 16.375,
+      "learning_rate": 8.525925925925927e-06,
+      "loss": 13.0871,
+      "mean_token_accuracy": 0.789578789100051,
+      "num_tokens": 61228692.0,
+      "step": 1850
+    },
+    {
+      "entropy": 0.813435361161828,
+      "epoch": 2.976,
+      "grad_norm": 16.875,
+      "learning_rate": 8.451851851851852e-06,
+      "loss": 13.0494,
+      "mean_token_accuracy": 0.7886098399758339,
+      "num_tokens": 61560852.0,
+      "step": 1860
+    },
+    {
+      "entropy": 0.7873949326574803,
+      "epoch": 2.992,
+      "grad_norm": 17.0,
+      "learning_rate": 8.377777777777779e-06,
+      "loss": 12.6489,
+      "mean_token_accuracy": 0.7942363824695349,
+      "num_tokens": 61898951.0,
+      "step": 1870
+    },
+    {
+      "entropy": 0.7897056974470615,
+      "epoch": 3.008,
+      "grad_norm": 16.125,
+      "learning_rate": 8.303703703703705e-06,
+      "loss": 12.6796,
+      "mean_token_accuracy": 0.7947975434362888,
+      "num_tokens": 62223875.0,
+      "step": 1880
+    },
+    {
+      "entropy": 0.8028375385329127,
+      "epoch": 3.024,
+      "grad_norm": 16.875,
+      "learning_rate": 8.229629629629632e-06,
+      "loss": 12.8036,
+      "mean_token_accuracy": 0.792441963031888,
+      "num_tokens": 62553166.0,
+      "step": 1890
+    },
+    {
+      "entropy": 0.7828883344307542,
+      "epoch": 3.04,
+      "grad_norm": 17.5,
+      "learning_rate": 8.155555555555556e-06,
+      "loss": 12.6348,
+      "mean_token_accuracy": 0.7942964274436235,
+      "num_tokens": 62893113.0,
+      "step": 1900
+    },
+    {
+      "epoch": 3.04,
+      "eval_biology_entropy": 2.50356245803833,
+      "eval_biology_loss": 3.003526449203491,
+      "eval_biology_mean_token_accuracy": 0.4901131451129913,
+      "eval_biology_num_tokens": 62893113.0,
+      "eval_biology_runtime": 49.8515,
+      "eval_biology_samples_per_second": 10.03,
+      "eval_biology_steps_per_second": 2.507,
+      "step": 1900
+    },
+    {
+      "epoch": 3.04,
+      "eval_chemistry_entropy": 1.150402109146118,
+      "eval_chemistry_loss": 1.2631772756576538,
+      "eval_chemistry_mean_token_accuracy": 0.7117339906692505,
+      "eval_chemistry_num_tokens": 62893113.0,
+      "eval_chemistry_runtime": 61.7676,
+      "eval_chemistry_samples_per_second": 8.095,
+      "eval_chemistry_steps_per_second": 2.024,
+      "step": 1900
+    },
+    {
+      "epoch": 3.04,
+      "eval_math_entropy": 0.8996168537139893,
+      "eval_math_loss": 1.1427340507507324,
+      "eval_math_mean_token_accuracy": 0.7398220405578614,
+      "eval_math_num_tokens": 62893113.0,
+      "eval_math_runtime": 63.662,
+      "eval_math_samples_per_second": 7.854,
+      "eval_math_steps_per_second": 1.963,
+      "step": 1900
+    },
+    {
+      "epoch": 3.04,
+      "eval_physics_entropy": 0.818979761838913,
+      "eval_physics_loss": 0.8601692914962769,
+      "eval_physics_mean_token_accuracy": 0.7813335127830505,
+      "eval_physics_num_tokens": 62893113.0,
+      "eval_physics_runtime": 72.4072,
+      "eval_physics_samples_per_second": 6.905,
+      "eval_physics_steps_per_second": 1.726,
+      "step": 1900
+    },
+    {
+      "entropy": 0.8113107359036803,
+      "epoch": 3.056,
+      "grad_norm": 18.375,
+      "learning_rate": 8.081481481481483e-06,
+      "loss": 12.9961,
+      "mean_token_accuracy": 0.7898518111556768,
+      "num_tokens": 63218088.0,
+      "step": 1910
+    },
+    {
+      "entropy": 0.805760845914483,
+      "epoch": 3.072,
+      "grad_norm": 18.75,
+      "learning_rate": 8.007407407407408e-06,
+      "loss": 12.9227,
+      "mean_token_accuracy": 0.7900473494082689,
+      "num_tokens": 63551201.0,
+      "step": 1920
+    },
+    {
+      "entropy": 0.8170072333887219,
+      "epoch": 3.088,
+      "grad_norm": 17.875,
+      "learning_rate": 7.933333333333334e-06,
+      "loss": 13.2105,
+      "mean_token_accuracy": 0.7866270847618579,
+      "num_tokens": 63876362.0,
+      "step": 1930
+    },
+    {
+      "entropy": 0.7876848684623837,
+      "epoch": 3.104,
+      "grad_norm": 17.375,
+      "learning_rate": 7.859259259259259e-06,
+      "loss": 12.5329,
+      "mean_token_accuracy": 0.7958414383232594,
+      "num_tokens": 64208171.0,
+      "step": 1940
+    },
+    {
+      "entropy": 0.784156777150929,
+      "epoch": 3.12,
+      "grad_norm": 17.25,
+      "learning_rate": 7.785185185185185e-06,
+      "loss": 12.5755,
+      "mean_token_accuracy": 0.7954553622752428,
+      "num_tokens": 64544493.0,
+      "step": 1950
+    },
+    {
+      "entropy": 0.8146663185209035,
+      "epoch": 3.136,
+      "grad_norm": 17.875,
+      "learning_rate": 7.711111111111112e-06,
+      "loss": 13.2047,
+      "mean_token_accuracy": 0.7866963110864162,
+      "num_tokens": 64872488.0,
+      "step": 1960
+    },
+    {
+      "entropy": 0.7842913199216127,
+      "epoch": 3.152,
+      "grad_norm": 17.375,
+      "learning_rate": 7.637037037037037e-06,
+      "loss": 12.5105,
+      "mean_token_accuracy": 0.7967761281877757,
+      "num_tokens": 65201124.0,
+      "step": 1970
+    },
+    {
+      "entropy": 0.8004832187667489,
+      "epoch": 3.168,
+      "grad_norm": 16.125,
+      "learning_rate": 7.562962962962963e-06,
+      "loss": 12.9651,
+      "mean_token_accuracy": 0.7915269497781992,
+      "num_tokens": 65525868.0,
+      "step": 1980
+    },
+    {
+      "entropy": 0.790750872157514,
+      "epoch": 3.184,
+      "grad_norm": 17.875,
+      "learning_rate": 7.48888888888889e-06,
+      "loss": 12.6417,
+      "mean_token_accuracy": 0.7964012406766414,
+      "num_tokens": 65861603.0,
+      "step": 1990
+    },
+    {
+      "entropy": 0.8021171147003769,
+      "epoch": 3.2,
+      "grad_norm": 16.625,
+      "learning_rate": 7.4148148148148155e-06,
+      "loss": 12.912,
+      "mean_token_accuracy": 0.7916875142604113,
+      "num_tokens": 66193675.0,
+      "step": 2000
+    },
+    {
+      "epoch": 3.2,
+      "eval_biology_entropy": 2.470094334602356,
+      "eval_biology_loss": 3.0078163146972656,
+      "eval_biology_mean_token_accuracy": 0.49014620304107664,
+      "eval_biology_num_tokens": 66193675.0,
+      "eval_biology_runtime": 49.9577,
+      "eval_biology_samples_per_second": 10.008,
+      "eval_biology_steps_per_second": 2.502,
+      "step": 2000
+    },
+    {
+      "epoch": 3.2,
+      "eval_chemistry_entropy": 1.1386250944137573,
+      "eval_chemistry_loss": 1.2633483409881592,
+      "eval_chemistry_mean_token_accuracy": 0.71189799118042,
+      "eval_chemistry_num_tokens": 66193675.0,
+      "eval_chemistry_runtime": 61.4075,
+      "eval_chemistry_samples_per_second": 8.142,
+      "eval_chemistry_steps_per_second": 2.036,
+      "step": 2000
+    },
+    {
+      "epoch": 3.2,
+      "eval_math_entropy": 0.8969358072280884,
+      "eval_math_loss": 1.1421797275543213,
+      "eval_math_mean_token_accuracy": 0.7401115093231201,
+      "eval_math_num_tokens": 66193675.0,
+      "eval_math_runtime": 63.315,
+      "eval_math_samples_per_second": 7.897,
+      "eval_math_steps_per_second": 1.974,
+      "step": 2000
+    },
+    {
+      "epoch": 3.2,
+      "eval_physics_entropy": 0.8138435740470886,
+      "eval_physics_loss": 0.8593137860298157,
+      "eval_physics_mean_token_accuracy": 0.7811777157783508,
+      "eval_physics_num_tokens": 66193675.0,
+      "eval_physics_runtime": 72.1799,
+      "eval_physics_samples_per_second": 6.927,
+      "eval_physics_steps_per_second": 1.732,
+      "step": 2000
+    },
+    {
+      "entropy": 0.7921612739562989,
+      "epoch": 3.216,
+      "grad_norm": 18.125,
+      "learning_rate": 7.340740740740742e-06,
+      "loss": 12.7707,
+      "mean_token_accuracy": 0.7925254683941603,
+      "num_tokens": 66522835.0,
+      "step": 2010
+    },
+    {
+      "entropy": 0.829443721473217,
+      "epoch": 3.232,
+      "grad_norm": 17.75,
+      "learning_rate": 7.266666666666668e-06,
+      "loss": 13.3302,
+      "mean_token_accuracy": 0.7857878033071757,
+      "num_tokens": 66853693.0,
+      "step": 2020
+    },
+    {
+      "entropy": 0.7663173181936145,
+      "epoch": 3.248,
+      "grad_norm": 18.375,
+      "learning_rate": 7.192592592592593e-06,
+      "loss": 12.3159,
+      "mean_token_accuracy": 0.7985509064048528,
+      "num_tokens": 67192299.0,
+      "step": 2030
+    },
+    {
+      "entropy": 0.7895545920357108,
+      "epoch": 3.2640000000000002,
+      "grad_norm": 17.75,
+      "learning_rate": 7.118518518518519e-06,
+      "loss": 12.7685,
+      "mean_token_accuracy": 0.7944346476346255,
+      "num_tokens": 67517068.0,
+      "step": 2040
+    },
+    {
+      "entropy": 0.801218194887042,
+      "epoch": 3.2800000000000002,
+      "grad_norm": 17.625,
+      "learning_rate": 7.044444444444445e-06,
+      "loss": 12.7416,
+      "mean_token_accuracy": 0.7935682334005832,
+      "num_tokens": 67841380.0,
+      "step": 2050
+    },
+    {
+      "entropy": 0.786173316463828,
+      "epoch": 3.296,
+      "grad_norm": 16.5,
+      "learning_rate": 6.97037037037037e-06,
+      "loss": 12.695,
+      "mean_token_accuracy": 0.79406350068748,
+      "num_tokens": 68175287.0,
+      "step": 2060
+    },
+    {
+      "entropy": 0.7954026944935322,
+      "epoch": 3.312,
+      "grad_norm": 18.0,
+      "learning_rate": 6.896296296296297e-06,
+      "loss": 12.7967,
+      "mean_token_accuracy": 0.7934347588568926,
+      "num_tokens": 68513030.0,
+      "step": 2070
+    },
+    {
+      "entropy": 0.8265415087342263,
+      "epoch": 3.328,
+      "grad_norm": 18.5,
+      "learning_rate": 6.8222222222222225e-06,
+      "loss": 13.3081,
+      "mean_token_accuracy": 0.7859519779682159,
+      "num_tokens": 68839883.0,
+      "step": 2080
+    },
+    {
+      "entropy": 0.8022375550121069,
+      "epoch": 3.344,
+      "grad_norm": 19.25,
+      "learning_rate": 6.748148148148149e-06,
+      "loss": 12.8368,
+      "mean_token_accuracy": 0.791323509812355,
+      "num_tokens": 69173450.0,
+      "step": 2090
+    },
+    {
+      "entropy": 0.7997627248987556,
+      "epoch": 3.36,
+      "grad_norm": 17.625,
+      "learning_rate": 6.674074074074075e-06,
+      "loss": 12.905,
+      "mean_token_accuracy": 0.7927103146910668,
+      "num_tokens": 69510512.0,
+      "step": 2100
+    },
+    {
+      "epoch": 3.36,
+      "eval_biology_entropy": 2.480081204414368,
+      "eval_biology_loss": 3.014317274093628,
+      "eval_biology_mean_token_accuracy": 0.49056104898452757,
+      "eval_biology_num_tokens": 69510512.0,
+      "eval_biology_runtime": 49.8261,
+      "eval_biology_samples_per_second": 10.035,
+      "eval_biology_steps_per_second": 2.509,
+      "step": 2100
+    },
+    {
+      "epoch": 3.36,
+      "eval_chemistry_entropy": 1.1460778632164002,
+      "eval_chemistry_loss": 1.262313723564148,
+      "eval_chemistry_mean_token_accuracy": 0.7124251260757446,
+      "eval_chemistry_num_tokens": 69510512.0,
+      "eval_chemistry_runtime": 61.7856,
+      "eval_chemistry_samples_per_second": 8.092,
+      "eval_chemistry_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "epoch": 3.36,
+      "eval_math_entropy": 0.9010528683662414,
+      "eval_math_loss": 1.1404390335083008,
+      "eval_math_mean_token_accuracy": 0.7402434344291687,
+      "eval_math_num_tokens": 69510512.0,
+      "eval_math_runtime": 63.5967,
+      "eval_math_samples_per_second": 7.862,
+      "eval_math_steps_per_second": 1.966,
+      "step": 2100
+    },
+    {
+      "epoch": 3.36,
+      "eval_physics_entropy": 0.8187825033664703,
+      "eval_physics_loss": 0.858341634273529,
+      "eval_physics_mean_token_accuracy": 0.7814857354164123,
+      "eval_physics_num_tokens": 69510512.0,
+      "eval_physics_runtime": 72.1906,
+      "eval_physics_samples_per_second": 6.926,
+      "eval_physics_steps_per_second": 1.732,
+      "step": 2100
+    },
+    {
+      "entropy": 0.772541718557477,
+      "epoch": 3.376,
+      "grad_norm": 16.125,
+      "learning_rate": 6.600000000000001e-06,
+      "loss": 12.4046,
+      "mean_token_accuracy": 0.7991201918572187,
+      "num_tokens": 69853992.0,
+      "step": 2110
+    },
+    {
+      "entropy": 0.824666154384613,
+      "epoch": 3.392,
+      "grad_norm": 17.875,
+      "learning_rate": 6.525925925925927e-06,
+      "loss": 13.2013,
+      "mean_token_accuracy": 0.7854650605469942,
+      "num_tokens": 70178388.0,
+      "step": 2120
+    },
+    {
+      "entropy": 0.787809319794178,
+      "epoch": 3.408,
+      "grad_norm": 18.25,
+      "learning_rate": 6.4518518518518525e-06,
+      "loss": 12.6522,
+      "mean_token_accuracy": 0.794764668866992,
+      "num_tokens": 70507903.0,
+      "step": 2130
+    },
+    {
+      "entropy": 0.7877073505893349,
+      "epoch": 3.424,
+      "grad_norm": 18.0,
+      "learning_rate": 6.377777777777778e-06,
+      "loss": 12.6351,
+      "mean_token_accuracy": 0.7949817195534706,
+      "num_tokens": 70835982.0,
+      "step": 2140
+    },
+    {
+      "entropy": 0.8121343495324254,
+      "epoch": 3.44,
+      "grad_norm": 17.5,
+      "learning_rate": 6.303703703703704e-06,
+      "loss": 13.1399,
+      "mean_token_accuracy": 0.7877719759941101,
+      "num_tokens": 71172338.0,
+      "step": 2150
+    },
+    {
+      "entropy": 0.8116744419559836,
+      "epoch": 3.456,
+      "grad_norm": 18.25,
+      "learning_rate": 6.2296296296296295e-06,
+      "loss": 13.0305,
+      "mean_token_accuracy": 0.7885824784636497,
+      "num_tokens": 71506904.0,
+      "step": 2160
+    },
+    {
+      "entropy": 0.8163298014551401,
+      "epoch": 3.472,
+      "grad_norm": 18.125,
+      "learning_rate": 6.155555555555556e-06,
+      "loss": 13.1392,
+      "mean_token_accuracy": 0.7880806677043438,
+      "num_tokens": 71830111.0,
+      "step": 2170
+    },
+    {
+      "entropy": 0.7843415776267648,
+      "epoch": 3.488,
+      "grad_norm": 17.625,
+      "learning_rate": 6.081481481481482e-06,
+      "loss": 12.5195,
+      "mean_token_accuracy": 0.7982313647866249,
+      "num_tokens": 72162307.0,
+      "step": 2180
+    },
+    {
+      "entropy": 0.8081060107797384,
+      "epoch": 3.504,
+      "grad_norm": 18.75,
+      "learning_rate": 6.007407407407407e-06,
+      "loss": 13.0465,
+      "mean_token_accuracy": 0.7899345368146896,
+      "num_tokens": 72492324.0,
+      "step": 2190
+    },
+    {
+      "entropy": 0.7964632350951433,
+      "epoch": 3.52,
+      "grad_norm": 18.25,
+      "learning_rate": 5.933333333333335e-06,
+      "loss": 12.8622,
+      "mean_token_accuracy": 0.7926870550960302,
+      "num_tokens": 72814868.0,
+      "step": 2200
+    },
+    {
+      "epoch": 3.52,
+      "eval_biology_entropy": 2.484531102180481,
+      "eval_biology_loss": 3.003647804260254,
+      "eval_biology_mean_token_accuracy": 0.4901256234645843,
+      "eval_biology_num_tokens": 72814868.0,
+      "eval_biology_runtime": 49.9141,
+      "eval_biology_samples_per_second": 10.017,
+      "eval_biology_steps_per_second": 2.504,
+      "step": 2200
+    },
+    {
+      "epoch": 3.52,
+      "eval_chemistry_entropy": 1.1463134574890137,
+      "eval_chemistry_loss": 1.261918306350708,
+      "eval_chemistry_mean_token_accuracy": 0.7117061018943787,
+      "eval_chemistry_num_tokens": 72814868.0,
+      "eval_chemistry_runtime": 61.8204,
+      "eval_chemistry_samples_per_second": 8.088,
+      "eval_chemistry_steps_per_second": 2.022,
+      "step": 2200
+    },
+    {
+      "epoch": 3.52,
+      "eval_math_entropy": 0.8994995903968811,
+      "eval_math_loss": 1.1406532526016235,
+      "eval_math_mean_token_accuracy": 0.7399827222824097,
+      "eval_math_num_tokens": 72814868.0,
+      "eval_math_runtime": 63.5794,
+      "eval_math_samples_per_second": 7.864,
+      "eval_math_steps_per_second": 1.966,
+      "step": 2200
+    },
+    {
+      "epoch": 3.52,
+      "eval_physics_entropy": 0.8177375242710113,
+      "eval_physics_loss": 0.8577298521995544,
+      "eval_physics_mean_token_accuracy": 0.7814569849967956,
+      "eval_physics_num_tokens": 72814868.0,
+      "eval_physics_runtime": 72.6689,
+      "eval_physics_samples_per_second": 6.881,
+      "eval_physics_steps_per_second": 1.72,
+      "step": 2200
+    },
+    {
+      "entropy": 0.8043171558529139,
+      "epoch": 3.536,
+      "grad_norm": 16.25,
+      "learning_rate": 5.85925925925926e-06,
+      "loss": 12.9022,
+      "mean_token_accuracy": 0.7911262758076191,
+      "num_tokens": 73144676.0,
+      "step": 2210
+    },
+    {
+      "entropy": 0.8001729313284158,
+      "epoch": 3.552,
+      "grad_norm": 15.75,
+      "learning_rate": 5.785185185185186e-06,
+      "loss": 12.7439,
+      "mean_token_accuracy": 0.7945221561938525,
+      "num_tokens": 73474150.0,
+      "step": 2220
+    },
+    {
+      "entropy": 0.7879581512883306,
+      "epoch": 3.568,
+      "grad_norm": 18.375,
+      "learning_rate": 5.711111111111112e-06,
+      "loss": 12.6712,
+      "mean_token_accuracy": 0.7943290121853351,
+      "num_tokens": 73814337.0,
+      "step": 2230
+    },
+    {
+      "entropy": 0.8236944487318396,
+      "epoch": 3.584,
+      "grad_norm": 17.75,
+      "learning_rate": 5.637037037037037e-06,
+      "loss": 13.4206,
+      "mean_token_accuracy": 0.7838957156985998,
+      "num_tokens": 74139916.0,
+      "step": 2240
+    },
+    {
+      "entropy": 0.7940308267250658,
+      "epoch": 3.6,
+      "grad_norm": 18.625,
+      "learning_rate": 5.562962962962963e-06,
+      "loss": 12.5958,
+      "mean_token_accuracy": 0.7954386703670024,
+      "num_tokens": 74481410.0,
+      "step": 2250
+    },
+    {
+      "entropy": 0.7940249260514974,
+      "epoch": 3.616,
+      "grad_norm": 17.375,
+      "learning_rate": 5.4888888888888895e-06,
+      "loss": 12.7409,
+      "mean_token_accuracy": 0.7927557546645403,
+      "num_tokens": 74818399.0,
+      "step": 2260
+    },
+    {
+      "entropy": 0.7933648740872741,
+      "epoch": 3.632,
+      "grad_norm": 17.25,
+      "learning_rate": 5.414814814814815e-06,
+      "loss": 12.7148,
+      "mean_token_accuracy": 0.7937173150479794,
+      "num_tokens": 75148445.0,
+      "step": 2270
+    },
+    {
+      "entropy": 0.8004923764616251,
+      "epoch": 3.648,
+      "grad_norm": 17.625,
+      "learning_rate": 5.340740740740741e-06,
+      "loss": 12.9717,
+      "mean_token_accuracy": 0.7907850630581379,
+      "num_tokens": 75475448.0,
+      "step": 2280
+    },
+    {
+      "entropy": 0.7910401994362474,
+      "epoch": 3.664,
+      "grad_norm": 16.75,
+      "learning_rate": 5.2666666666666665e-06,
+      "loss": 12.6453,
+      "mean_token_accuracy": 0.7947485193610191,
+      "num_tokens": 75811516.0,
+      "step": 2290
+    },
+    {
+      "entropy": 0.8122350828722119,
+      "epoch": 3.68,
+      "grad_norm": 17.375,
+      "learning_rate": 5.192592592592594e-06,
+      "loss": 13.1267,
+      "mean_token_accuracy": 0.7880386158823967,
+      "num_tokens": 76131750.0,
+      "step": 2300
+    },
+    {
+      "epoch": 3.68,
+      "eval_biology_entropy": 2.4879787940979003,
+      "eval_biology_loss": 3.0091187953948975,
+      "eval_biology_mean_token_accuracy": 0.4910128047466278,
+      "eval_biology_num_tokens": 76131750.0,
+      "eval_biology_runtime": 49.8735,
+      "eval_biology_samples_per_second": 10.025,
+      "eval_biology_steps_per_second": 2.506,
+      "step": 2300
+    },
+    {
+      "epoch": 3.68,
+      "eval_chemistry_entropy": 1.146693642616272,
+      "eval_chemistry_loss": 1.2613716125488281,
+      "eval_chemistry_mean_token_accuracy": 0.7124113698005676,
+      "eval_chemistry_num_tokens": 76131750.0,
+      "eval_chemistry_runtime": 61.8096,
+      "eval_chemistry_samples_per_second": 8.089,
+      "eval_chemistry_steps_per_second": 2.022,
+      "step": 2300
+    },
+    {
+      "epoch": 3.68,
+      "eval_math_entropy": 0.8972381176948547,
+      "eval_math_loss": 1.141001582145691,
+      "eval_math_mean_token_accuracy": 0.7400426187515259,
+      "eval_math_num_tokens": 76131750.0,
+      "eval_math_runtime": 63.4249,
+      "eval_math_samples_per_second": 7.883,
+      "eval_math_steps_per_second": 1.971,
+      "step": 2300
+    },
+    {
+      "epoch": 3.68,
+      "eval_physics_entropy": 0.8186536066532135,
+      "eval_physics_loss": 0.8571553230285645,
+      "eval_physics_mean_token_accuracy": 0.7815062079429627,
+      "eval_physics_num_tokens": 76131750.0,
+      "eval_physics_runtime": 72.2085,
+      "eval_physics_samples_per_second": 6.924,
+      "eval_physics_steps_per_second": 1.731,
+      "step": 2300
+    },
+    {
+      "entropy": 0.8261356288567185,
+      "epoch": 3.6959999999999997,
+      "grad_norm": 18.125,
+      "learning_rate": 5.1185185185185195e-06,
+      "loss": 13.2921,
+      "mean_token_accuracy": 0.7866732176393271,
+      "num_tokens": 76460225.0,
+      "step": 2310
+    },
+    {
+      "entropy": 0.7932228075340391,
+      "epoch": 3.7119999999999997,
+      "grad_norm": 19.125,
+      "learning_rate": 5.044444444444445e-06,
+      "loss": 12.7199,
+      "mean_token_accuracy": 0.7937006145715714,
+      "num_tokens": 76793614.0,
+      "step": 2320
+    },
+    {
+      "entropy": 0.8164208957925438,
+      "epoch": 3.7279999999999998,
+      "grad_norm": 17.0,
+      "learning_rate": 4.970370370370371e-06,
+      "loss": 13.1811,
+      "mean_token_accuracy": 0.7872148770838976,
+      "num_tokens": 77121009.0,
+      "step": 2330
+    },
+    {
+      "entropy": 0.8028501404449344,
+      "epoch": 3.7439999999999998,
+      "grad_norm": 18.125,
+      "learning_rate": 4.8962962962962965e-06,
+      "loss": 12.8896,
+      "mean_token_accuracy": 0.7913309291005135,
+      "num_tokens": 77446311.0,
+      "step": 2340
+    },
+    {
+      "entropy": 0.8100664691999555,
+      "epoch": 3.76,
+      "grad_norm": 17.375,
+      "learning_rate": 4.822222222222222e-06,
+      "loss": 12.9901,
+      "mean_token_accuracy": 0.789349377527833,
+      "num_tokens": 77781488.0,
+      "step": 2350
+    },
+    {
+      "entropy": 0.8052188634872437,
+      "epoch": 3.776,
+      "grad_norm": 18.75,
+      "learning_rate": 4.748148148148149e-06,
+      "loss": 12.9672,
+      "mean_token_accuracy": 0.7903082646429539,
+      "num_tokens": 78108138.0,
+      "step": 2360
+    },
+    {
+      "entropy": 0.7854003531858325,
+      "epoch": 3.792,
+      "grad_norm": 17.125,
+      "learning_rate": 4.674074074074074e-06,
+      "loss": 12.6051,
+      "mean_token_accuracy": 0.7969572100788355,
+      "num_tokens": 78449177.0,
+      "step": 2370
+    },
+    {
+      "entropy": 0.804776719585061,
+      "epoch": 3.808,
+      "grad_norm": 16.875,
+      "learning_rate": 4.600000000000001e-06,
+      "loss": 12.9076,
+      "mean_token_accuracy": 0.7905494842678309,
+      "num_tokens": 78778761.0,
+      "step": 2380
+    },
+    {
+      "entropy": 0.817225544527173,
+      "epoch": 3.824,
+      "grad_norm": 17.875,
+      "learning_rate": 4.5259259259259265e-06,
+      "loss": 13.1296,
+      "mean_token_accuracy": 0.7879973672330379,
+      "num_tokens": 79102912.0,
+      "step": 2390
+    },
+    {
+      "entropy": 0.7974447071552276,
+      "epoch": 3.84,
+      "grad_norm": 17.5,
+      "learning_rate": 4.451851851851852e-06,
+      "loss": 12.8469,
+      "mean_token_accuracy": 0.7920668609440327,
+      "num_tokens": 79436236.0,
+      "step": 2400
+    },
+    {
+      "epoch": 3.84,
+      "eval_biology_entropy": 2.486952880859375,
+      "eval_biology_loss": 3.0059683322906494,
+      "eval_biology_mean_token_accuracy": 0.49017577290534975,
+      "eval_biology_num_tokens": 79436236.0,
+      "eval_biology_runtime": 50.069,
+      "eval_biology_samples_per_second": 9.986,
+      "eval_biology_steps_per_second": 2.497,
+      "step": 2400
+    },
+    {
+      "epoch": 3.84,
+      "eval_chemistry_entropy": 1.1472082695961,
+      "eval_chemistry_loss": 1.2613314390182495,
+      "eval_chemistry_mean_token_accuracy": 0.7123120713233948,
+      "eval_chemistry_num_tokens": 79436236.0,
+      "eval_chemistry_runtime": 61.8189,
+      "eval_chemistry_samples_per_second": 8.088,
+      "eval_chemistry_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "epoch": 3.84,
+      "eval_math_entropy": 0.9007806525230407,
+      "eval_math_loss": 1.1408295631408691,
+      "eval_math_mean_token_accuracy": 0.7399366483688354,
+      "eval_math_num_tokens": 79436236.0,
+      "eval_math_runtime": 63.3314,
+      "eval_math_samples_per_second": 7.895,
+      "eval_math_steps_per_second": 1.974,
+      "step": 2400
+    },
+    {
+      "epoch": 3.84,
+      "eval_physics_entropy": 0.8190191547870636,
+      "eval_physics_loss": 0.8569294214248657,
+      "eval_physics_mean_token_accuracy": 0.7818823776245117,
+      "eval_physics_num_tokens": 79436236.0,
+      "eval_physics_runtime": 72.5771,
+      "eval_physics_samples_per_second": 6.889,
+      "eval_physics_steps_per_second": 1.722,
+      "step": 2400
+    },
+    {
+      "entropy": 0.7811166772618889,
+      "epoch": 3.856,
+      "grad_norm": 17.0,
+      "learning_rate": 4.377777777777778e-06,
+      "loss": 12.5334,
+      "mean_token_accuracy": 0.7964712589979172,
+      "num_tokens": 79772163.0,
+      "step": 2410
+    },
+    {
+      "entropy": 0.8073545757681131,
+      "epoch": 3.872,
+      "grad_norm": 18.5,
+      "learning_rate": 4.3037037037037035e-06,
+      "loss": 12.8932,
+      "mean_token_accuracy": 0.7909647975116968,
+      "num_tokens": 80099967.0,
+      "step": 2420
+    },
+    {
+      "entropy": 0.8008806597441435,
+      "epoch": 3.888,
+      "grad_norm": 18.375,
+      "learning_rate": 4.22962962962963e-06,
+      "loss": 12.9345,
+      "mean_token_accuracy": 0.7915704354643822,
+      "num_tokens": 80430055.0,
+      "step": 2430
+    },
+    {
+      "entropy": 0.8328071933239698,
+      "epoch": 3.904,
+      "grad_norm": 20.125,
+      "learning_rate": 4.155555555555556e-06,
+      "loss": 13.4493,
+      "mean_token_accuracy": 0.7839287407696247,
+      "num_tokens": 80763923.0,
+      "step": 2440
+    },
+    {
+      "entropy": 0.7978027690201998,
+      "epoch": 3.92,
+      "grad_norm": 18.875,
+      "learning_rate": 4.081481481481482e-06,
+      "loss": 12.7918,
+      "mean_token_accuracy": 0.7934749405831099,
+      "num_tokens": 81098568.0,
+      "step": 2450
+    },
+    {
+      "entropy": 0.8075996682047843,
+      "epoch": 3.936,
+      "grad_norm": 17.5,
+      "learning_rate": 4.007407407407408e-06,
+      "loss": 12.9714,
+      "mean_token_accuracy": 0.7904196504503489,
+      "num_tokens": 81423109.0,
+      "step": 2460
+    },
+    {
+      "entropy": 0.8154539056122303,
+      "epoch": 3.952,
+      "grad_norm": 19.75,
+      "learning_rate": 3.9333333333333335e-06,
+      "loss": 13.1029,
+      "mean_token_accuracy": 0.7884767297655344,
+      "num_tokens": 81753214.0,
+      "step": 2470
+    },
+    {
+      "entropy": 0.7908139565959573,
+      "epoch": 3.968,
+      "grad_norm": 19.875,
+      "learning_rate": 3.85925925925926e-06,
+      "loss": 12.6902,
+      "mean_token_accuracy": 0.7936715561896562,
+      "num_tokens": 82076472.0,
+      "step": 2480
+    },
+    {
+      "entropy": 0.7808545615524054,
+      "epoch": 3.984,
+      "grad_norm": 16.875,
+      "learning_rate": 3.7851851851851857e-06,
+      "loss": 12.4941,
+      "mean_token_accuracy": 0.7965636719018221,
+      "num_tokens": 82418472.0,
+      "step": 2490
+    },
+    {
+      "entropy": 0.7908104166388512,
+      "epoch": 4.0,
+      "grad_norm": 18.875,
+      "learning_rate": 3.7111111111111113e-06,
+      "loss": 12.7267,
+      "mean_token_accuracy": 0.7932475034147501,
+      "num_tokens": 82747216.0,
+      "step": 2500
+    },
+    {
+      "epoch": 4.0,
+      "eval_biology_entropy": 2.4771251888275145,
+      "eval_biology_loss": 3.006816864013672,
+      "eval_biology_mean_token_accuracy": 0.49003281354904177,
+      "eval_biology_num_tokens": 82747216.0,
+      "eval_biology_runtime": 49.8413,
+      "eval_biology_samples_per_second": 10.032,
+      "eval_biology_steps_per_second": 2.508,
+      "step": 2500
+    },
+    {
+      "epoch": 4.0,
+      "eval_chemistry_entropy": 1.1426447529792785,
+      "eval_chemistry_loss": 1.2616218328475952,
+      "eval_chemistry_mean_token_accuracy": 0.7119665017127991,
+      "eval_chemistry_num_tokens": 82747216.0,
+      "eval_chemistry_runtime": 61.4018,
+      "eval_chemistry_samples_per_second": 8.143,
+      "eval_chemistry_steps_per_second": 2.036,
+      "step": 2500
+    },
+    {
+      "epoch": 4.0,
+      "eval_math_entropy": 0.898317488193512,
+      "eval_math_loss": 1.1417402029037476,
+      "eval_math_mean_token_accuracy": 0.7399371833801269,
+      "eval_math_num_tokens": 82747216.0,
+      "eval_math_runtime": 63.2462,
+      "eval_math_samples_per_second": 7.906,
+      "eval_math_steps_per_second": 1.976,
+      "step": 2500
+    },
+    {
+      "epoch": 4.0,
+      "eval_physics_entropy": 0.8161102538108825,
+      "eval_physics_loss": 0.8568674325942993,
+      "eval_physics_mean_token_accuracy": 0.7817758975028992,
+      "eval_physics_num_tokens": 82747216.0,
+      "eval_physics_runtime": 72.3314,
+      "eval_physics_samples_per_second": 6.913,
+      "eval_physics_steps_per_second": 1.728,
+      "step": 2500
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 3000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 5,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 8.272635579032348e+18,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}