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": 1.2553194094449283,
+      "epoch": 0.016,
+      "grad_norm": 270.0,
+      "learning_rate": 6.000000000000001e-07,
+      "loss": 41.3914,
+      "mean_token_accuracy": 0.5671123253181577,
+      "num_tokens": 338352.0,
+      "step": 10
+    },
+    {
+      "entropy": 1.266689160466194,
+      "epoch": 0.032,
+      "grad_norm": 264.0,
+      "learning_rate": 1.2666666666666669e-06,
+      "loss": 41.0628,
+      "mean_token_accuracy": 0.5705239847302437,
+      "num_tokens": 671193.0,
+      "step": 20
+    },
+    {
+      "entropy": 1.3143046092242003,
+      "epoch": 0.048,
+      "grad_norm": 210.0,
+      "learning_rate": 1.9333333333333336e-06,
+      "loss": 39.4998,
+      "mean_token_accuracy": 0.5745313983410597,
+      "num_tokens": 1001008.0,
+      "step": 30
+    },
+    {
+      "entropy": 1.4753582045435905,
+      "epoch": 0.064,
+      "grad_norm": 155.0,
+      "learning_rate": 2.6e-06,
+      "loss": 38.0568,
+      "mean_token_accuracy": 0.5745376290753483,
+      "num_tokens": 1331208.0,
+      "step": 40
+    },
+    {
+      "entropy": 1.6769672967493534,
+      "epoch": 0.08,
+      "grad_norm": 99.5,
+      "learning_rate": 3.266666666666667e-06,
+      "loss": 34.3871,
+      "mean_token_accuracy": 0.5940070981159806,
+      "num_tokens": 1667532.0,
+      "step": 50
+    },
+    {
+      "entropy": 1.7719040885567665,
+      "epoch": 0.096,
+      "grad_norm": 55.25,
+      "learning_rate": 3.9333333333333335e-06,
+      "loss": 30.2468,
+      "mean_token_accuracy": 0.6289831601083279,
+      "num_tokens": 2007176.0,
+      "step": 60
+    },
+    {
+      "entropy": 1.7813668169081212,
+      "epoch": 0.112,
+      "grad_norm": 45.0,
+      "learning_rate": 4.600000000000001e-06,
+      "loss": 28.6054,
+      "mean_token_accuracy": 0.6419326957315207,
+      "num_tokens": 2342259.0,
+      "step": 70
+    },
+    {
+      "entropy": 1.7141680859029294,
+      "epoch": 0.128,
+      "grad_norm": 36.75,
+      "learning_rate": 5.2666666666666665e-06,
+      "loss": 27.4799,
+      "mean_token_accuracy": 0.6465466242283583,
+      "num_tokens": 2663987.0,
+      "step": 80
+    },
+    {
+      "entropy": 1.5760046038776636,
+      "epoch": 0.144,
+      "grad_norm": 26.875,
+      "learning_rate": 5.933333333333335e-06,
+      "loss": 24.8601,
+      "mean_token_accuracy": 0.6713848907500506,
+      "num_tokens": 2997824.0,
+      "step": 90
+    },
+    {
+      "entropy": 1.51325889416039,
+      "epoch": 0.16,
+      "grad_norm": 24.625,
+      "learning_rate": 6.600000000000001e-06,
+      "loss": 23.6713,
+      "mean_token_accuracy": 0.6833302665501833,
+      "num_tokens": 3330597.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.16,
+      "eval_biology_entropy": 3.1857855548858645,
+      "eval_biology_loss": 4.065364837646484,
+      "eval_biology_mean_token_accuracy": 0.4020989770889282,
+      "eval_biology_num_tokens": 3330597.0,
+      "eval_biology_runtime": 49.9004,
+      "eval_biology_samples_per_second": 10.02,
+      "eval_biology_steps_per_second": 2.505,
+      "step": 100
+    },
+    {
+      "epoch": 0.16,
+      "eval_chemistry_entropy": 1.8894153318405151,
+      "eval_chemistry_loss": 1.931850790977478,
+      "eval_chemistry_mean_token_accuracy": 0.6162000632286072,
+      "eval_chemistry_num_tokens": 3330597.0,
+      "eval_chemistry_runtime": 61.8293,
+      "eval_chemistry_samples_per_second": 8.087,
+      "eval_chemistry_steps_per_second": 2.022,
+      "step": 100
+    },
+    {
+      "epoch": 0.16,
+      "eval_math_entropy": 1.3163333106040955,
+      "eval_math_loss": 1.5677636861801147,
+      "eval_math_mean_token_accuracy": 0.6788510384559632,
+      "eval_math_num_tokens": 3330597.0,
+      "eval_math_runtime": 63.4108,
+      "eval_math_samples_per_second": 7.885,
+      "eval_math_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "epoch": 0.16,
+      "eval_physics_entropy": 1.495229031085968,
+      "eval_physics_loss": 1.4545849561691284,
+      "eval_physics_mean_token_accuracy": 0.6851646456718444,
+      "eval_physics_num_tokens": 3330597.0,
+      "eval_physics_runtime": 72.2177,
+      "eval_physics_samples_per_second": 6.924,
+      "eval_physics_steps_per_second": 1.731,
+      "step": 100
+    },
+    {
+      "entropy": 1.473308515548706,
+      "epoch": 0.176,
+      "grad_norm": 25.375,
+      "learning_rate": 7.266666666666668e-06,
+      "loss": 22.9019,
+      "mean_token_accuracy": 0.6873474020510912,
+      "num_tokens": 3658264.0,
+      "step": 110
+    },
+    {
+      "entropy": 1.397306866571307,
+      "epoch": 0.192,
+      "grad_norm": 19.625,
+      "learning_rate": 7.933333333333334e-06,
+      "loss": 21.9472,
+      "mean_token_accuracy": 0.6935743600130081,
+      "num_tokens": 3995568.0,
+      "step": 120
+    },
+    {
+      "entropy": 1.3264490943402052,
+      "epoch": 0.208,
+      "grad_norm": 20.5,
+      "learning_rate": 8.6e-06,
+      "loss": 20.7584,
+      "mean_token_accuracy": 0.7065585050731897,
+      "num_tokens": 4321436.0,
+      "step": 130
+    },
+    {
+      "entropy": 1.2879054959863425,
+      "epoch": 0.224,
+      "grad_norm": 18.625,
+      "learning_rate": 9.266666666666667e-06,
+      "loss": 20.3558,
+      "mean_token_accuracy": 0.7094652783125639,
+      "num_tokens": 4648491.0,
+      "step": 140
+    },
+    {
+      "entropy": 1.262782633677125,
+      "epoch": 0.24,
+      "grad_norm": 20.875,
+      "learning_rate": 9.933333333333334e-06,
+      "loss": 20.125,
+      "mean_token_accuracy": 0.7121465969830751,
+      "num_tokens": 4986175.0,
+      "step": 150
+    },
+    {
+      "entropy": 1.249500698968768,
+      "epoch": 0.256,
+      "grad_norm": 20.75,
+      "learning_rate": 1.0600000000000002e-05,
+      "loss": 19.7318,
+      "mean_token_accuracy": 0.716720923408866,
+      "num_tokens": 5329320.0,
+      "step": 160
+    },
+    {
+      "entropy": 1.2423948224633932,
+      "epoch": 0.272,
+      "grad_norm": 21.125,
+      "learning_rate": 1.1266666666666668e-05,
+      "loss": 19.6803,
+      "mean_token_accuracy": 0.7159549534320832,
+      "num_tokens": 5658796.0,
+      "step": 170
+    },
+    {
+      "entropy": 1.2081249222159385,
+      "epoch": 0.288,
+      "grad_norm": 18.5,
+      "learning_rate": 1.1933333333333335e-05,
+      "loss": 19.318,
+      "mean_token_accuracy": 0.7217235699295997,
+      "num_tokens": 5980201.0,
+      "step": 180
+    },
+    {
+      "entropy": 1.2010240752249957,
+      "epoch": 0.304,
+      "grad_norm": 23.125,
+      "learning_rate": 1.2600000000000001e-05,
+      "loss": 19.0924,
+      "mean_token_accuracy": 0.723386499658227,
+      "num_tokens": 6298858.0,
+      "step": 190
+    },
+    {
+      "entropy": 1.1805048868060113,
+      "epoch": 0.32,
+      "grad_norm": 21.5,
+      "learning_rate": 1.3266666666666668e-05,
+      "loss": 18.8302,
+      "mean_token_accuracy": 0.7257850173860788,
+      "num_tokens": 6622798.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.32,
+      "eval_biology_entropy": 3.0335595417022705,
+      "eval_biology_loss": 3.5063252449035645,
+      "eval_biology_mean_token_accuracy": 0.43639786744117737,
+      "eval_biology_num_tokens": 6622798.0,
+      "eval_biology_runtime": 49.8291,
+      "eval_biology_samples_per_second": 10.034,
+      "eval_biology_steps_per_second": 2.509,
+      "step": 200
+    },
+    {
+      "epoch": 0.32,
+      "eval_chemistry_entropy": 1.5769986052513123,
+      "eval_chemistry_loss": 1.6105988025665283,
+      "eval_chemistry_mean_token_accuracy": 0.6568134782314301,
+      "eval_chemistry_num_tokens": 6622798.0,
+      "eval_chemistry_runtime": 61.7607,
+      "eval_chemistry_samples_per_second": 8.096,
+      "eval_chemistry_steps_per_second": 2.024,
+      "step": 200
+    },
+    {
+      "epoch": 0.32,
+      "eval_math_entropy": 1.1430701270103454,
+      "eval_math_loss": 1.3579970598220825,
+      "eval_math_mean_token_accuracy": 0.7044133353233337,
+      "eval_math_num_tokens": 6622798.0,
+      "eval_math_runtime": 63.3226,
+      "eval_math_samples_per_second": 7.896,
+      "eval_math_steps_per_second": 1.974,
+      "step": 200
+    },
+    {
+      "epoch": 0.32,
+      "eval_physics_entropy": 1.1658131113052368,
+      "eval_physics_loss": 1.1756293773651123,
+      "eval_physics_mean_token_accuracy": 0.7273708815574647,
+      "eval_physics_num_tokens": 6622798.0,
+      "eval_physics_runtime": 72.1537,
+      "eval_physics_samples_per_second": 6.93,
+      "eval_physics_steps_per_second": 1.732,
+      "step": 200
+    },
+    {
+      "entropy": 1.1631717231124639,
+      "epoch": 0.336,
+      "grad_norm": 15.8125,
+      "learning_rate": 1.3933333333333334e-05,
+      "loss": 18.617,
+      "mean_token_accuracy": 0.7290044378489255,
+      "num_tokens": 6953611.0,
+      "step": 210
+    },
+    {
+      "entropy": 1.1292029924690723,
+      "epoch": 0.352,
+      "grad_norm": 19.625,
+      "learning_rate": 1.46e-05,
+      "loss": 17.9563,
+      "mean_token_accuracy": 0.7365154311060905,
+      "num_tokens": 7290378.0,
+      "step": 220
+    },
+    {
+      "entropy": 1.1049898941069842,
+      "epoch": 0.368,
+      "grad_norm": 17.75,
+      "learning_rate": 1.5266666666666667e-05,
+      "loss": 17.6986,
+      "mean_token_accuracy": 0.7393099017441273,
+      "num_tokens": 7621082.0,
+      "step": 230
+    },
+    {
+      "entropy": 1.1595781400799752,
+      "epoch": 0.384,
+      "grad_norm": 19.375,
+      "learning_rate": 1.5933333333333336e-05,
+      "loss": 18.4615,
+      "mean_token_accuracy": 0.727893678098917,
+      "num_tokens": 7955570.0,
+      "step": 240
+    },
+    {
+      "entropy": 1.0945355758070945,
+      "epoch": 0.4,
+      "grad_norm": 21.0,
+      "learning_rate": 1.66e-05,
+      "loss": 17.5388,
+      "mean_token_accuracy": 0.7387776046991348,
+      "num_tokens": 8291049.0,
+      "step": 250
+    },
+    {
+      "entropy": 1.0819710556417703,
+      "epoch": 0.416,
+      "grad_norm": 18.375,
+      "learning_rate": 1.726666666666667e-05,
+      "loss": 17.251,
+      "mean_token_accuracy": 0.7430898856371642,
+      "num_tokens": 8627310.0,
+      "step": 260
+    },
+    {
+      "entropy": 1.089613201841712,
+      "epoch": 0.432,
+      "grad_norm": 20.75,
+      "learning_rate": 1.7933333333333333e-05,
+      "loss": 17.4707,
+      "mean_token_accuracy": 0.7406438160687685,
+      "num_tokens": 8958371.0,
+      "step": 270
+    },
+    {
+      "entropy": 1.0421824473887682,
+      "epoch": 0.448,
+      "grad_norm": 18.25,
+      "learning_rate": 1.86e-05,
+      "loss": 16.7755,
+      "mean_token_accuracy": 0.7496135134249926,
+      "num_tokens": 9294388.0,
+      "step": 280
+    },
+    {
+      "entropy": 1.0451348841190338,
+      "epoch": 0.464,
+      "grad_norm": 19.625,
+      "learning_rate": 1.926666666666667e-05,
+      "loss": 16.5784,
+      "mean_token_accuracy": 0.7497908163815736,
+      "num_tokens": 9620366.0,
+      "step": 290
+    },
+    {
+      "entropy": 1.0491984207183123,
+      "epoch": 0.48,
+      "grad_norm": 21.0,
+      "learning_rate": 1.9933333333333334e-05,
+      "loss": 16.8506,
+      "mean_token_accuracy": 0.7469458010047674,
+      "num_tokens": 9955431.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.48,
+      "eval_biology_entropy": 2.9388448543548584,
+      "eval_biology_loss": 3.3656885623931885,
+      "eval_biology_mean_token_accuracy": 0.45114623188972475,
+      "eval_biology_num_tokens": 9955431.0,
+      "eval_biology_runtime": 49.8383,
+      "eval_biology_samples_per_second": 10.032,
+      "eval_biology_steps_per_second": 2.508,
+      "step": 300
+    },
+    {
+      "epoch": 0.48,
+      "eval_chemistry_entropy": 1.491808822631836,
+      "eval_chemistry_loss": 1.5070267915725708,
+      "eval_chemistry_mean_token_accuracy": 0.6716848533153534,
+      "eval_chemistry_num_tokens": 9955431.0,
+      "eval_chemistry_runtime": 61.771,
+      "eval_chemistry_samples_per_second": 8.094,
+      "eval_chemistry_steps_per_second": 2.024,
+      "step": 300
+    },
+    {
+      "epoch": 0.48,
+      "eval_math_entropy": 1.1062884669303894,
+      "eval_math_loss": 1.285144567489624,
+      "eval_math_mean_token_accuracy": 0.7163918890953064,
+      "eval_math_num_tokens": 9955431.0,
+      "eval_math_runtime": 63.3221,
+      "eval_math_samples_per_second": 7.896,
+      "eval_math_steps_per_second": 1.974,
+      "step": 300
+    },
+    {
+      "epoch": 0.48,
+      "eval_physics_entropy": 1.0871692457199096,
+      "eval_physics_loss": 1.0718116760253906,
+      "eval_physics_mean_token_accuracy": 0.7436052026748657,
+      "eval_physics_num_tokens": 9955431.0,
+      "eval_physics_runtime": 72.1598,
+      "eval_physics_samples_per_second": 6.929,
+      "eval_physics_steps_per_second": 1.732,
+      "step": 300
+    },
+    {
+      "entropy": 1.0622025161981583,
+      "epoch": 0.496,
+      "grad_norm": 17.625,
+      "learning_rate": 1.9933333333333334e-05,
+      "loss": 17.0457,
+      "mean_token_accuracy": 0.7451069477945567,
+      "num_tokens": 10289883.0,
+      "step": 310
+    },
+    {
+      "entropy": 1.0407135926187039,
+      "epoch": 0.512,
+      "grad_norm": 21.375,
+      "learning_rate": 1.985925925925926e-05,
+      "loss": 16.7612,
+      "mean_token_accuracy": 0.7487055193632841,
+      "num_tokens": 10619468.0,
+      "step": 320
+    },
+    {
+      "entropy": 1.060495626181364,
+      "epoch": 0.528,
+      "grad_norm": 19.625,
+      "learning_rate": 1.9785185185185187e-05,
+      "loss": 16.9124,
+      "mean_token_accuracy": 0.7458847668021917,
+      "num_tokens": 10944904.0,
+      "step": 330
+    },
+    {
+      "entropy": 1.0363870609551669,
+      "epoch": 0.544,
+      "grad_norm": 19.125,
+      "learning_rate": 1.971111111111111e-05,
+      "loss": 16.6786,
+      "mean_token_accuracy": 0.7487504351884127,
+      "num_tokens": 11266409.0,
+      "step": 340
+    },
+    {
+      "entropy": 1.01834645755589,
+      "epoch": 0.56,
+      "grad_norm": 17.125,
+      "learning_rate": 1.963703703703704e-05,
+      "loss": 16.224,
+      "mean_token_accuracy": 0.7543694581836462,
+      "num_tokens": 11605544.0,
+      "step": 350
+    },
+    {
+      "entropy": 1.0103376515209674,
+      "epoch": 0.576,
+      "grad_norm": 21.75,
+      "learning_rate": 1.9562962962962964e-05,
+      "loss": 16.1635,
+      "mean_token_accuracy": 0.7562580808997155,
+      "num_tokens": 11931715.0,
+      "step": 360
+    },
+    {
+      "entropy": 1.0192645259201527,
+      "epoch": 0.592,
+      "grad_norm": 19.625,
+      "learning_rate": 1.948888888888889e-05,
+      "loss": 16.3501,
+      "mean_token_accuracy": 0.7522288095206022,
+      "num_tokens": 12257243.0,
+      "step": 370
+    },
+    {
+      "entropy": 0.9979648280888795,
+      "epoch": 0.608,
+      "grad_norm": 17.25,
+      "learning_rate": 1.9414814814814817e-05,
+      "loss": 16.0193,
+      "mean_token_accuracy": 0.7565512597560883,
+      "num_tokens": 12583154.0,
+      "step": 380
+    },
+    {
+      "entropy": 1.0146858640015126,
+      "epoch": 0.624,
+      "grad_norm": 18.375,
+      "learning_rate": 1.9340740740740743e-05,
+      "loss": 16.2917,
+      "mean_token_accuracy": 0.7530687287449837,
+      "num_tokens": 12905392.0,
+      "step": 390
+    },
+    {
+      "entropy": 1.0172333989292384,
+      "epoch": 0.64,
+      "grad_norm": 18.0,
+      "learning_rate": 1.926666666666667e-05,
+      "loss": 16.311,
+      "mean_token_accuracy": 0.753251226246357,
+      "num_tokens": 13232198.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.64,
+      "eval_biology_entropy": 2.8556682786941527,
+      "eval_biology_loss": 3.2405552864074707,
+      "eval_biology_mean_token_accuracy": 0.46047434616088867,
+      "eval_biology_num_tokens": 13232198.0,
+      "eval_biology_runtime": 49.8137,
+      "eval_biology_samples_per_second": 10.037,
+      "eval_biology_steps_per_second": 2.509,
+      "step": 400
+    },
+    {
+      "epoch": 0.64,
+      "eval_chemistry_entropy": 1.4023492636680603,
+      "eval_chemistry_loss": 1.4446464776992798,
+      "eval_chemistry_mean_token_accuracy": 0.6815885419845581,
+      "eval_chemistry_num_tokens": 13232198.0,
+      "eval_chemistry_runtime": 61.7651,
+      "eval_chemistry_samples_per_second": 8.095,
+      "eval_chemistry_steps_per_second": 2.024,
+      "step": 400
+    },
+    {
+      "epoch": 0.64,
+      "eval_math_entropy": 1.0369325885772704,
+      "eval_math_loss": 1.2500724792480469,
+      "eval_math_mean_token_accuracy": 0.7222203068733215,
+      "eval_math_num_tokens": 13232198.0,
+      "eval_math_runtime": 63.311,
+      "eval_math_samples_per_second": 7.898,
+      "eval_math_steps_per_second": 1.974,
+      "step": 400
+    },
+    {
+      "epoch": 0.64,
+      "eval_physics_entropy": 1.0061357097625732,
+      "eval_physics_loss": 1.0139094591140747,
+      "eval_physics_mean_token_accuracy": 0.7539046273231507,
+      "eval_physics_num_tokens": 13232198.0,
+      "eval_physics_runtime": 72.128,
+      "eval_physics_samples_per_second": 6.932,
+      "eval_physics_steps_per_second": 1.733,
+      "step": 400
+    },
+    {
+      "entropy": 1.0025192126631737,
+      "epoch": 0.656,
+      "grad_norm": 19.0,
+      "learning_rate": 1.9192592592592593e-05,
+      "loss": 16.0757,
+      "mean_token_accuracy": 0.7558728314936161,
+      "num_tokens": 13575902.0,
+      "step": 410
+    },
+    {
+      "entropy": 0.9805797968059778,
+      "epoch": 0.672,
+      "grad_norm": 20.75,
+      "learning_rate": 1.911851851851852e-05,
+      "loss": 15.6938,
+      "mean_token_accuracy": 0.7607984621077776,
+      "num_tokens": 13895997.0,
+      "step": 420
+    },
+    {
+      "entropy": 0.98455760627985,
+      "epoch": 0.688,
+      "grad_norm": 17.5,
+      "learning_rate": 1.9044444444444446e-05,
+      "loss": 15.8079,
+      "mean_token_accuracy": 0.7591990135610104,
+      "num_tokens": 14234888.0,
+      "step": 430
+    },
+    {
+      "entropy": 0.995212909206748,
+      "epoch": 0.704,
+      "grad_norm": 19.5,
+      "learning_rate": 1.8970370370370372e-05,
+      "loss": 16.0276,
+      "mean_token_accuracy": 0.7551241502165794,
+      "num_tokens": 14567908.0,
+      "step": 440
+    },
+    {
+      "entropy": 0.9859623953700065,
+      "epoch": 0.72,
+      "grad_norm": 19.875,
+      "learning_rate": 1.8896296296296295e-05,
+      "loss": 15.7158,
+      "mean_token_accuracy": 0.759292322024703,
+      "num_tokens": 14882927.0,
+      "step": 450
+    },
+    {
+      "entropy": 0.9759141772985458,
+      "epoch": 0.736,
+      "grad_norm": 20.625,
+      "learning_rate": 1.8822222222222225e-05,
+      "loss": 15.6771,
+      "mean_token_accuracy": 0.759360334649682,
+      "num_tokens": 15217342.0,
+      "step": 460
+    },
+    {
+      "entropy": 0.9576828479766846,
+      "epoch": 0.752,
+      "grad_norm": 19.5,
+      "learning_rate": 1.874814814814815e-05,
+      "loss": 15.2707,
+      "mean_token_accuracy": 0.764180276170373,
+      "num_tokens": 15547710.0,
+      "step": 470
+    },
+    {
+      "entropy": 0.9969257492572069,
+      "epoch": 0.768,
+      "grad_norm": 19.625,
+      "learning_rate": 1.8674074074074075e-05,
+      "loss": 16.0499,
+      "mean_token_accuracy": 0.7543311327695846,
+      "num_tokens": 15877177.0,
+      "step": 480
+    },
+    {
+      "entropy": 0.9526452720165253,
+      "epoch": 0.784,
+      "grad_norm": 15.8125,
+      "learning_rate": 1.86e-05,
+      "loss": 15.3172,
+      "mean_token_accuracy": 0.7652558576315641,
+      "num_tokens": 16219640.0,
+      "step": 490
+    },
+    {
+      "entropy": 0.9730141185224056,
+      "epoch": 0.8,
+      "grad_norm": 19.25,
+      "learning_rate": 1.8525925925925928e-05,
+      "loss": 15.5783,
+      "mean_token_accuracy": 0.7610898818820715,
+      "num_tokens": 16548261.0,
+      "step": 500
+    },
+    {
+      "epoch": 0.8,
+      "eval_biology_entropy": 2.8879401025772093,
+      "eval_biology_loss": 3.2465980052948,
+      "eval_biology_mean_token_accuracy": 0.4656795439720154,
+      "eval_biology_num_tokens": 16548261.0,
+      "eval_biology_runtime": 50.0039,
+      "eval_biology_samples_per_second": 9.999,
+      "eval_biology_steps_per_second": 2.5,
+      "step": 500
+    },
+    {
+      "epoch": 0.8,
+      "eval_chemistry_entropy": 1.4028268342018126,
+      "eval_chemistry_loss": 1.4141976833343506,
+      "eval_chemistry_mean_token_accuracy": 0.6879124054908753,
+      "eval_chemistry_num_tokens": 16548261.0,
+      "eval_chemistry_runtime": 61.4753,
+      "eval_chemistry_samples_per_second": 8.133,
+      "eval_chemistry_steps_per_second": 2.033,
+      "step": 500
+    },
+    {
+      "epoch": 0.8,
+      "eval_math_entropy": 1.036446961402893,
+      "eval_math_loss": 1.2320657968521118,
+      "eval_math_mean_token_accuracy": 0.7248759384155273,
+      "eval_math_num_tokens": 16548261.0,
+      "eval_math_runtime": 63.2484,
+      "eval_math_samples_per_second": 7.905,
+      "eval_math_steps_per_second": 1.976,
+      "step": 500
+    },
+    {
+      "epoch": 0.8,
+      "eval_physics_entropy": 0.9860591797828674,
+      "eval_physics_loss": 0.9807776808738708,
+      "eval_physics_mean_token_accuracy": 0.7594618821144103,
+      "eval_physics_num_tokens": 16548261.0,
+      "eval_physics_runtime": 72.1552,
+      "eval_physics_samples_per_second": 6.93,
+      "eval_physics_steps_per_second": 1.732,
+      "step": 500
+    },
+    {
+      "entropy": 0.9596989471465349,
+      "epoch": 0.816,
+      "grad_norm": 17.5,
+      "learning_rate": 1.8451851851851855e-05,
+      "loss": 15.4334,
+      "mean_token_accuracy": 0.7610802616924047,
+      "num_tokens": 16870533.0,
+      "step": 510
+    },
+    {
+      "entropy": 0.9716402690857648,
+      "epoch": 0.832,
+      "grad_norm": 18.25,
+      "learning_rate": 1.8377777777777778e-05,
+      "loss": 15.5823,
+      "mean_token_accuracy": 0.7609996516257524,
+      "num_tokens": 17196371.0,
+      "step": 520
+    },
+    {
+      "entropy": 0.9541628547012806,
+      "epoch": 0.848,
+      "grad_norm": 16.5,
+      "learning_rate": 1.8303703703703704e-05,
+      "loss": 15.3143,
+      "mean_token_accuracy": 0.7634800903499126,
+      "num_tokens": 17527066.0,
+      "step": 530
+    },
+    {
+      "entropy": 0.9640029605478049,
+      "epoch": 0.864,
+      "grad_norm": 17.375,
+      "learning_rate": 1.822962962962963e-05,
+      "loss": 15.4415,
+      "mean_token_accuracy": 0.7621862847357989,
+      "num_tokens": 17859605.0,
+      "step": 540
+    },
+    {
+      "entropy": 0.9721696741878987,
+      "epoch": 0.88,
+      "grad_norm": 20.25,
+      "learning_rate": 1.8155555555555557e-05,
+      "loss": 15.653,
+      "mean_token_accuracy": 0.7603420093655586,
+      "num_tokens": 18187315.0,
+      "step": 550
+    },
+    {
+      "entropy": 0.9153296016156673,
+      "epoch": 0.896,
+      "grad_norm": 17.25,
+      "learning_rate": 1.8081481481481484e-05,
+      "loss": 14.6671,
+      "mean_token_accuracy": 0.7723869703710079,
+      "num_tokens": 18536790.0,
+      "step": 560
+    },
+    {
+      "entropy": 0.958526523038745,
+      "epoch": 0.912,
+      "grad_norm": 18.75,
+      "learning_rate": 1.800740740740741e-05,
+      "loss": 15.4361,
+      "mean_token_accuracy": 0.7610937509685755,
+      "num_tokens": 18869344.0,
+      "step": 570
+    },
+    {
+      "entropy": 0.9475870430469513,
+      "epoch": 0.928,
+      "grad_norm": 19.125,
+      "learning_rate": 1.7933333333333333e-05,
+      "loss": 15.2185,
+      "mean_token_accuracy": 0.7640813227742911,
+      "num_tokens": 19199834.0,
+      "step": 580
+    },
+    {
+      "entropy": 0.9297015912830829,
+      "epoch": 0.944,
+      "grad_norm": 18.75,
+      "learning_rate": 1.785925925925926e-05,
+      "loss": 14.8769,
+      "mean_token_accuracy": 0.7691585179418325,
+      "num_tokens": 19525385.0,
+      "step": 590
+    },
+    {
+      "entropy": 0.9348739337176084,
+      "epoch": 0.96,
+      "grad_norm": 19.0,
+      "learning_rate": 1.7785185185185186e-05,
+      "loss": 15.0455,
+      "mean_token_accuracy": 0.7664772134274245,
+      "num_tokens": 19856973.0,
+      "step": 600
+    },
+    {
+      "epoch": 0.96,
+      "eval_biology_entropy": 2.7933437023162844,
+      "eval_biology_loss": 3.194451332092285,
+      "eval_biology_mean_token_accuracy": 0.46872161483764646,
+      "eval_biology_num_tokens": 19856973.0,
+      "eval_biology_runtime": 49.8208,
+      "eval_biology_samples_per_second": 10.036,
+      "eval_biology_steps_per_second": 2.509,
+      "step": 600
+    },
+    {
+      "epoch": 0.96,
+      "eval_chemistry_entropy": 1.3487913212776184,
+      "eval_chemistry_loss": 1.3904978036880493,
+      "eval_chemistry_mean_token_accuracy": 0.6907124133110046,
+      "eval_chemistry_num_tokens": 19856973.0,
+      "eval_chemistry_runtime": 62.2399,
+      "eval_chemistry_samples_per_second": 8.033,
+      "eval_chemistry_steps_per_second": 2.008,
+      "step": 600
+    },
+    {
+      "epoch": 0.96,
+      "eval_math_entropy": 1.0215094347000122,
+      "eval_math_loss": 1.2187507152557373,
+      "eval_math_mean_token_accuracy": 0.727313805103302,
+      "eval_math_num_tokens": 19856973.0,
+      "eval_math_runtime": 63.311,
+      "eval_math_samples_per_second": 7.898,
+      "eval_math_steps_per_second": 1.974,
+      "step": 600
+    },
+    {
+      "epoch": 0.96,
+      "eval_physics_entropy": 0.9578838763237,
+      "eval_physics_loss": 0.9591855406761169,
+      "eval_physics_mean_token_accuracy": 0.7633539743423462,
+      "eval_physics_num_tokens": 19856973.0,
+      "eval_physics_runtime": 72.6851,
+      "eval_physics_samples_per_second": 6.879,
+      "eval_physics_steps_per_second": 1.72,
+      "step": 600
+    },
+    {
+      "entropy": 0.9349439710378646,
+      "epoch": 0.976,
+      "grad_norm": 19.125,
+      "learning_rate": 1.7711111111111113e-05,
+      "loss": 14.9645,
+      "mean_token_accuracy": 0.7687592633068562,
+      "num_tokens": 20190600.0,
+      "step": 610
+    },
+    {
+      "entropy": 0.9438869334757328,
+      "epoch": 0.992,
+      "grad_norm": 18.625,
+      "learning_rate": 1.763703703703704e-05,
+      "loss": 15.1663,
+      "mean_token_accuracy": 0.765958558395505,
+      "num_tokens": 20518014.0,
+      "step": 620
+    },
+    {
+      "entropy": 0.9152817223221064,
+      "epoch": 1.008,
+      "grad_norm": 17.0,
+      "learning_rate": 1.7562962962962962e-05,
+      "loss": 14.5949,
+      "mean_token_accuracy": 0.7715727590024472,
+      "num_tokens": 20859215.0,
+      "step": 630
+    },
+    {
+      "entropy": 0.9109054185450077,
+      "epoch": 1.024,
+      "grad_norm": 18.0,
+      "learning_rate": 1.7488888888888892e-05,
+      "loss": 14.683,
+      "mean_token_accuracy": 0.7712803717702628,
+      "num_tokens": 21186658.0,
+      "step": 640
+    },
+    {
+      "entropy": 0.9192747503519059,
+      "epoch": 1.04,
+      "grad_norm": 17.375,
+      "learning_rate": 1.7414814814814815e-05,
+      "loss": 14.7149,
+      "mean_token_accuracy": 0.7704985976219177,
+      "num_tokens": 21517237.0,
+      "step": 650
+    },
+    {
+      "entropy": 0.8997716940939426,
+      "epoch": 1.056,
+      "grad_norm": 18.5,
+      "learning_rate": 1.7340740740740742e-05,
+      "loss": 14.4443,
+      "mean_token_accuracy": 0.7745653819292784,
+      "num_tokens": 21852528.0,
+      "step": 660
+    },
+    {
+      "entropy": 0.9201500467956066,
+      "epoch": 1.072,
+      "grad_norm": 16.375,
+      "learning_rate": 1.726666666666667e-05,
+      "loss": 14.7404,
+      "mean_token_accuracy": 0.7707031384110451,
+      "num_tokens": 22192380.0,
+      "step": 670
+    },
+    {
+      "entropy": 0.9155960086733103,
+      "epoch": 1.088,
+      "grad_norm": 20.0,
+      "learning_rate": 1.7192592592592595e-05,
+      "loss": 14.6895,
+      "mean_token_accuracy": 0.771490203589201,
+      "num_tokens": 22517971.0,
+      "step": 680
+    },
+    {
+      "entropy": 0.8976538371294737,
+      "epoch": 1.104,
+      "grad_norm": 16.625,
+      "learning_rate": 1.711851851851852e-05,
+      "loss": 14.4464,
+      "mean_token_accuracy": 0.7735383935272694,
+      "num_tokens": 22848694.0,
+      "step": 690
+    },
+    {
+      "entropy": 0.9162428434938192,
+      "epoch": 1.12,
+      "grad_norm": 17.5,
+      "learning_rate": 1.7044444444444445e-05,
+      "loss": 14.7282,
+      "mean_token_accuracy": 0.7701425358653069,
+      "num_tokens": 23183889.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.12,
+      "eval_biology_entropy": 2.7417425537109374,
+      "eval_biology_loss": 3.198380947113037,
+      "eval_biology_mean_token_accuracy": 0.46923186993598937,
+      "eval_biology_num_tokens": 23183889.0,
+      "eval_biology_runtime": 49.9576,
+      "eval_biology_samples_per_second": 10.008,
+      "eval_biology_steps_per_second": 2.502,
+      "step": 700
+    },
+    {
+      "epoch": 1.12,
+      "eval_chemistry_entropy": 1.3224085516929627,
+      "eval_chemistry_loss": 1.3783864974975586,
+      "eval_chemistry_mean_token_accuracy": 0.6928294057846069,
+      "eval_chemistry_num_tokens": 23183889.0,
+      "eval_chemistry_runtime": 62.3468,
+      "eval_chemistry_samples_per_second": 8.02,
+      "eval_chemistry_steps_per_second": 2.005,
+      "step": 700
+    },
+    {
+      "epoch": 1.12,
+      "eval_math_entropy": 0.9974914565086365,
+      "eval_math_loss": 1.2118620872497559,
+      "eval_math_mean_token_accuracy": 0.7289316186904907,
+      "eval_math_num_tokens": 23183889.0,
+      "eval_math_runtime": 63.342,
+      "eval_math_samples_per_second": 7.894,
+      "eval_math_steps_per_second": 1.973,
+      "step": 700
+    },
+    {
+      "epoch": 1.12,
+      "eval_physics_entropy": 0.9341713042259217,
+      "eval_physics_loss": 0.9430394172668457,
+      "eval_physics_mean_token_accuracy": 0.7666752815246582,
+      "eval_physics_num_tokens": 23183889.0,
+      "eval_physics_runtime": 72.1566,
+      "eval_physics_samples_per_second": 6.929,
+      "eval_physics_steps_per_second": 1.732,
+      "step": 700
+    },
+    {
+      "entropy": 0.8717782899737359,
+      "epoch": 1.1360000000000001,
+      "grad_norm": 15.9375,
+      "learning_rate": 1.697037037037037e-05,
+      "loss": 14.027,
+      "mean_token_accuracy": 0.7789868891239167,
+      "num_tokens": 23523903.0,
+      "step": 710
+    },
+    {
+      "entropy": 0.9066614583134651,
+      "epoch": 1.152,
+      "grad_norm": 20.0,
+      "learning_rate": 1.6896296296296298e-05,
+      "loss": 14.5252,
+      "mean_token_accuracy": 0.7747723516076803,
+      "num_tokens": 23848848.0,
+      "step": 720
+    },
+    {
+      "entropy": 0.9192185271531343,
+      "epoch": 1.168,
+      "grad_norm": 19.0,
+      "learning_rate": 1.6822222222222224e-05,
+      "loss": 14.6797,
+      "mean_token_accuracy": 0.7702812351286411,
+      "num_tokens": 24174589.0,
+      "step": 730
+    },
+    {
+      "entropy": 0.8968648243695497,
+      "epoch": 1.184,
+      "grad_norm": 19.25,
+      "learning_rate": 1.6748148148148147e-05,
+      "loss": 14.4441,
+      "mean_token_accuracy": 0.7734088286757469,
+      "num_tokens": 24495981.0,
+      "step": 740
+    },
+    {
+      "entropy": 0.8798596655949951,
+      "epoch": 1.2,
+      "grad_norm": 18.625,
+      "learning_rate": 1.6674074074074077e-05,
+      "loss": 14.2088,
+      "mean_token_accuracy": 0.7773797914385796,
+      "num_tokens": 24826245.0,
+      "step": 750
+    },
+    {
+      "entropy": 0.9085365436971188,
+      "epoch": 1.216,
+      "grad_norm": 23.625,
+      "learning_rate": 1.66e-05,
+      "loss": 14.5158,
+      "mean_token_accuracy": 0.7717528346925974,
+      "num_tokens": 25147775.0,
+      "step": 760
+    },
+    {
+      "entropy": 0.8942177969962358,
+      "epoch": 1.232,
+      "grad_norm": 19.375,
+      "learning_rate": 1.6525925925925927e-05,
+      "loss": 14.35,
+      "mean_token_accuracy": 0.7760623764246702,
+      "num_tokens": 25480033.0,
+      "step": 770
+    },
+    {
+      "entropy": 0.8957445651292801,
+      "epoch": 1.248,
+      "grad_norm": 19.625,
+      "learning_rate": 1.6451851851851853e-05,
+      "loss": 14.3489,
+      "mean_token_accuracy": 0.7746611826121808,
+      "num_tokens": 25811978.0,
+      "step": 780
+    },
+    {
+      "entropy": 0.8850473642349244,
+      "epoch": 1.264,
+      "grad_norm": 19.375,
+      "learning_rate": 1.637777777777778e-05,
+      "loss": 14.2639,
+      "mean_token_accuracy": 0.7760811734944582,
+      "num_tokens": 26146299.0,
+      "step": 790
+    },
+    {
+      "entropy": 0.9187131529673934,
+      "epoch": 1.28,
+      "grad_norm": 18.875,
+      "learning_rate": 1.6303703703703706e-05,
+      "loss": 14.7971,
+      "mean_token_accuracy": 0.7688343476504087,
+      "num_tokens": 26478207.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.28,
+      "eval_biology_entropy": 2.7373362789154054,
+      "eval_biology_loss": 3.2188053131103516,
+      "eval_biology_mean_token_accuracy": 0.4715359935760498,
+      "eval_biology_num_tokens": 26478207.0,
+      "eval_biology_runtime": 49.8242,
+      "eval_biology_samples_per_second": 10.035,
+      "eval_biology_steps_per_second": 2.509,
+      "step": 800
+    },
+    {
+      "epoch": 1.28,
+      "eval_chemistry_entropy": 1.2954678382873535,
+      "eval_chemistry_loss": 1.3689188957214355,
+      "eval_chemistry_mean_token_accuracy": 0.6956489958763122,
+      "eval_chemistry_num_tokens": 26478207.0,
+      "eval_chemistry_runtime": 61.7443,
+      "eval_chemistry_samples_per_second": 8.098,
+      "eval_chemistry_steps_per_second": 2.024,
+      "step": 800
+    },
+    {
+      "epoch": 1.28,
+      "eval_math_entropy": 0.9856116833686829,
+      "eval_math_loss": 1.2060915231704712,
+      "eval_math_mean_token_accuracy": 0.7302432260513305,
+      "eval_math_num_tokens": 26478207.0,
+      "eval_math_runtime": 63.311,
+      "eval_math_samples_per_second": 7.898,
+      "eval_math_steps_per_second": 1.974,
+      "step": 800
+    },
+    {
+      "epoch": 1.28,
+      "eval_physics_entropy": 0.9129857659339905,
+      "eval_physics_loss": 0.9318634271621704,
+      "eval_physics_mean_token_accuracy": 0.7681499900817871,
+      "eval_physics_num_tokens": 26478207.0,
+      "eval_physics_runtime": 72.1104,
+      "eval_physics_samples_per_second": 6.934,
+      "eval_physics_steps_per_second": 1.733,
+      "step": 800
+    },
+    {
+      "entropy": 0.8775971475988626,
+      "epoch": 1.296,
+      "grad_norm": 19.625,
+      "learning_rate": 1.622962962962963e-05,
+      "loss": 14.0996,
+      "mean_token_accuracy": 0.7783409550786018,
+      "num_tokens": 26806477.0,
+      "step": 810
+    },
+    {
+      "entropy": 0.9151829794049263,
+      "epoch": 1.312,
+      "grad_norm": 20.125,
+      "learning_rate": 1.6155555555555556e-05,
+      "loss": 14.6154,
+      "mean_token_accuracy": 0.7715053271502257,
+      "num_tokens": 27137528.0,
+      "step": 820
+    },
+    {
+      "entropy": 0.8893604565411806,
+      "epoch": 1.328,
+      "grad_norm": 21.375,
+      "learning_rate": 1.6081481481481482e-05,
+      "loss": 14.3336,
+      "mean_token_accuracy": 0.7749421369284392,
+      "num_tokens": 27465859.0,
+      "step": 830
+    },
+    {
+      "entropy": 0.9043792076408863,
+      "epoch": 1.3439999999999999,
+      "grad_norm": 20.25,
+      "learning_rate": 1.600740740740741e-05,
+      "loss": 14.5396,
+      "mean_token_accuracy": 0.7718299012631178,
+      "num_tokens": 27795568.0,
+      "step": 840
+    },
+    {
+      "entropy": 0.8975318446755409,
+      "epoch": 1.3599999999999999,
+      "grad_norm": 18.75,
+      "learning_rate": 1.5933333333333336e-05,
+      "loss": 14.4516,
+      "mean_token_accuracy": 0.7730446238070726,
+      "num_tokens": 28129988.0,
+      "step": 850
+    },
+    {
+      "entropy": 0.894953028857708,
+      "epoch": 1.376,
+      "grad_norm": 20.625,
+      "learning_rate": 1.5859259259259262e-05,
+      "loss": 14.2654,
+      "mean_token_accuracy": 0.7754011388868094,
+      "num_tokens": 28465903.0,
+      "step": 860
+    },
+    {
+      "entropy": 0.8700921148061752,
+      "epoch": 1.392,
+      "grad_norm": 19.875,
+      "learning_rate": 1.5785185185185185e-05,
+      "loss": 13.9603,
+      "mean_token_accuracy": 0.7800745751708746,
+      "num_tokens": 28801554.0,
+      "step": 870
+    },
+    {
+      "entropy": 0.8798988396301866,
+      "epoch": 1.408,
+      "grad_norm": 18.75,
+      "learning_rate": 1.571111111111111e-05,
+      "loss": 14.1608,
+      "mean_token_accuracy": 0.7768289878964424,
+      "num_tokens": 29125114.0,
+      "step": 880
+    },
+    {
+      "entropy": 0.9143678206950426,
+      "epoch": 1.424,
+      "grad_norm": 17.875,
+      "learning_rate": 1.5637037037037038e-05,
+      "loss": 14.7235,
+      "mean_token_accuracy": 0.7692738968878985,
+      "num_tokens": 29454334.0,
+      "step": 890
+    },
+    {
+      "entropy": 0.8575869327411055,
+      "epoch": 1.44,
+      "grad_norm": 17.75,
+      "learning_rate": 1.5562962962962965e-05,
+      "loss": 13.7841,
+      "mean_token_accuracy": 0.7813448339700699,
+      "num_tokens": 29789253.0,
+      "step": 900
+    },
+    {
+      "epoch": 1.44,
+      "eval_biology_entropy": 2.765932272911072,
+      "eval_biology_loss": 3.186239242553711,
+      "eval_biology_mean_token_accuracy": 0.4713315246105194,
+      "eval_biology_num_tokens": 29789253.0,
+      "eval_biology_runtime": 49.9865,
+      "eval_biology_samples_per_second": 10.003,
+      "eval_biology_steps_per_second": 2.501,
+      "step": 900
+    },
+    {
+      "epoch": 1.44,
+      "eval_chemistry_entropy": 1.293758502960205,
+      "eval_chemistry_loss": 1.3550570011138916,
+      "eval_chemistry_mean_token_accuracy": 0.6970094337463378,
+      "eval_chemistry_num_tokens": 29789253.0,
+      "eval_chemistry_runtime": 61.9517,
+      "eval_chemistry_samples_per_second": 8.071,
+      "eval_chemistry_steps_per_second": 2.018,
+      "step": 900
+    },
+    {
+      "epoch": 1.44,
+      "eval_math_entropy": 0.9867032241821289,
+      "eval_math_loss": 1.201257348060608,
+      "eval_math_mean_token_accuracy": 0.7303560256958008,
+      "eval_math_num_tokens": 29789253.0,
+      "eval_math_runtime": 63.5245,
+      "eval_math_samples_per_second": 7.871,
+      "eval_math_steps_per_second": 1.968,
+      "step": 900
+    },
+    {
+      "epoch": 1.44,
+      "eval_physics_entropy": 0.9099465346336365,
+      "eval_physics_loss": 0.9209814071655273,
+      "eval_physics_mean_token_accuracy": 0.7701499495506287,
+      "eval_physics_num_tokens": 29789253.0,
+      "eval_physics_runtime": 72.3521,
+      "eval_physics_samples_per_second": 6.911,
+      "eval_physics_steps_per_second": 1.728,
+      "step": 900
+    },
+    {
+      "entropy": 0.8655804365873336,
+      "epoch": 1.456,
+      "grad_norm": 15.875,
+      "learning_rate": 1.548888888888889e-05,
+      "loss": 13.9105,
+      "mean_token_accuracy": 0.7815111551433802,
+      "num_tokens": 30128576.0,
+      "step": 910
+    },
+    {
+      "entropy": 0.9171438716351986,
+      "epoch": 1.472,
+      "grad_norm": 20.375,
+      "learning_rate": 1.5414814814814814e-05,
+      "loss": 14.7757,
+      "mean_token_accuracy": 0.7698528375476599,
+      "num_tokens": 30452719.0,
+      "step": 920
+    },
+    {
+      "entropy": 0.8968168456107378,
+      "epoch": 1.488,
+      "grad_norm": 18.375,
+      "learning_rate": 1.5340740740740744e-05,
+      "loss": 14.385,
+      "mean_token_accuracy": 0.7741949297487736,
+      "num_tokens": 30782343.0,
+      "step": 930
+    },
+    {
+      "entropy": 0.8784723855555058,
+      "epoch": 1.504,
+      "grad_norm": 16.875,
+      "learning_rate": 1.5266666666666667e-05,
+      "loss": 14.0301,
+      "mean_token_accuracy": 0.7789344146847725,
+      "num_tokens": 31113238.0,
+      "step": 940
+    },
+    {
+      "entropy": 0.8974394641816617,
+      "epoch": 1.52,
+      "grad_norm": 18.875,
+      "learning_rate": 1.5192592592592594e-05,
+      "loss": 14.3838,
+      "mean_token_accuracy": 0.7739370841532945,
+      "num_tokens": 31444589.0,
+      "step": 950
+    },
+    {
+      "entropy": 0.8973223175853491,
+      "epoch": 1.536,
+      "grad_norm": 20.625,
+      "learning_rate": 1.5118518518518519e-05,
+      "loss": 14.4894,
+      "mean_token_accuracy": 0.7721214085817337,
+      "num_tokens": 31774137.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.8697954054921866,
+      "epoch": 1.552,
+      "grad_norm": 19.375,
+      "learning_rate": 1.5044444444444445e-05,
+      "loss": 13.8974,
+      "mean_token_accuracy": 0.78069927431643,
+      "num_tokens": 32101776.0,
+      "step": 970
+    },
+    {
+      "entropy": 0.8705245990306139,
+      "epoch": 1.568,
+      "grad_norm": 19.125,
+      "learning_rate": 1.497037037037037e-05,
+      "loss": 13.979,
+      "mean_token_accuracy": 0.7797447606921196,
+      "num_tokens": 32428021.0,
+      "step": 980
+    },
+    {
+      "entropy": 0.8989604346454143,
+      "epoch": 1.584,
+      "grad_norm": 18.375,
+      "learning_rate": 1.4896296296296298e-05,
+      "loss": 14.4829,
+      "mean_token_accuracy": 0.7704301360994578,
+      "num_tokens": 32752735.0,
+      "step": 990
+    },
+    {
+      "entropy": 0.8908115413039923,
+      "epoch": 1.6,
+      "grad_norm": 17.75,
+      "learning_rate": 1.4822222222222225e-05,
+      "loss": 14.2986,
+      "mean_token_accuracy": 0.7751044664531946,
+      "num_tokens": 33075822.0,
+      "step": 1000
+    },
+    {
+      "epoch": 1.6,
+      "eval_biology_entropy": 2.702871771812439,
+      "eval_biology_loss": 3.161149501800537,
+      "eval_biology_mean_token_accuracy": 0.47477258658409116,
+      "eval_biology_num_tokens": 33075822.0,
+      "eval_biology_runtime": 50.0794,
+      "eval_biology_samples_per_second": 9.984,
+      "eval_biology_steps_per_second": 2.496,
+      "step": 1000
+    },
+    {
+      "epoch": 1.6,
+      "eval_chemistry_entropy": 1.265580331325531,
+      "eval_chemistry_loss": 1.3501311540603638,
+      "eval_chemistry_mean_token_accuracy": 0.6981884994506836,
+      "eval_chemistry_num_tokens": 33075822.0,
+      "eval_chemistry_runtime": 61.4474,
+      "eval_chemistry_samples_per_second": 8.137,
+      "eval_chemistry_steps_per_second": 2.034,
+      "step": 1000
+    },
+    {
+      "epoch": 1.6,
+      "eval_math_entropy": 0.9701343817710877,
+      "eval_math_loss": 1.2001285552978516,
+      "eval_math_mean_token_accuracy": 0.7312138648033142,
+      "eval_math_num_tokens": 33075822.0,
+      "eval_math_runtime": 63.2376,
+      "eval_math_samples_per_second": 7.907,
+      "eval_math_steps_per_second": 1.977,
+      "step": 1000
+    },
+    {
+      "epoch": 1.6,
+      "eval_physics_entropy": 0.8882069087028504,
+      "eval_physics_loss": 0.9141198992729187,
+      "eval_physics_mean_token_accuracy": 0.7712050895690918,
+      "eval_physics_num_tokens": 33075822.0,
+      "eval_physics_runtime": 72.1374,
+      "eval_physics_samples_per_second": 6.931,
+      "eval_physics_steps_per_second": 1.733,
+      "step": 1000
+    },
+    {
+      "entropy": 0.852231327816844,
+      "epoch": 1.616,
+      "grad_norm": 19.5,
+      "learning_rate": 1.474814814814815e-05,
+      "loss": 13.7455,
+      "mean_token_accuracy": 0.7828571233898401,
+      "num_tokens": 33411850.0,
+      "step": 1010
+    },
+    {
+      "entropy": 0.8694836173206568,
+      "epoch": 1.6320000000000001,
+      "grad_norm": 19.375,
+      "learning_rate": 1.4674074074074076e-05,
+      "loss": 13.9574,
+      "mean_token_accuracy": 0.7792858675122261,
+      "num_tokens": 33739456.0,
+      "step": 1020
+    },
+    {
+      "entropy": 0.8416479125618934,
+      "epoch": 1.6480000000000001,
+      "grad_norm": 18.375,
+      "learning_rate": 1.46e-05,
+      "loss": 13.4841,
+      "mean_token_accuracy": 0.7848905235528946,
+      "num_tokens": 34072022.0,
+      "step": 1030
+    },
+    {
+      "entropy": 0.8727968029677868,
+      "epoch": 1.6640000000000001,
+      "grad_norm": 20.875,
+      "learning_rate": 1.4525925925925927e-05,
+      "loss": 13.9919,
+      "mean_token_accuracy": 0.7798548869788646,
+      "num_tokens": 34398959.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.8499273512512445,
+      "epoch": 1.6800000000000002,
+      "grad_norm": 17.875,
+      "learning_rate": 1.4451851851851852e-05,
+      "loss": 13.7693,
+      "mean_token_accuracy": 0.7817482702434063,
+      "num_tokens": 34732284.0,
+      "step": 1050
+    },
+    {
+      "entropy": 0.8752633452415466,
+      "epoch": 1.696,
+      "grad_norm": 19.25,
+      "learning_rate": 1.4377777777777779e-05,
+      "loss": 13.9838,
+      "mean_token_accuracy": 0.7789037089794875,
+      "num_tokens": 35053712.0,
+      "step": 1060
+    },
+    {
+      "entropy": 0.8853001151233911,
+      "epoch": 1.712,
+      "grad_norm": 20.0,
+      "learning_rate": 1.4303703703703703e-05,
+      "loss": 14.2107,
+      "mean_token_accuracy": 0.7755775731056929,
+      "num_tokens": 35393088.0,
+      "step": 1070
+    },
+    {
+      "entropy": 0.8599673572927713,
+      "epoch": 1.728,
+      "grad_norm": 18.5,
+      "learning_rate": 1.4229629629629632e-05,
+      "loss": 13.8371,
+      "mean_token_accuracy": 0.7809670843183995,
+      "num_tokens": 35729007.0,
+      "step": 1080
+    },
+    {
+      "entropy": 0.867702866345644,
+      "epoch": 1.744,
+      "grad_norm": 19.125,
+      "learning_rate": 1.4155555555555556e-05,
+      "loss": 13.9299,
+      "mean_token_accuracy": 0.7801329270005226,
+      "num_tokens": 36072489.0,
+      "step": 1090
+    },
+    {
+      "entropy": 0.8642092518508434,
+      "epoch": 1.76,
+      "grad_norm": 18.625,
+      "learning_rate": 1.4081481481481483e-05,
+      "loss": 13.8054,
+      "mean_token_accuracy": 0.7801033087074757,
+      "num_tokens": 36398285.0,
+      "step": 1100
+    },
+    {
+      "epoch": 1.76,
+      "eval_biology_entropy": 2.676761179924011,
+      "eval_biology_loss": 3.160200834274292,
+      "eval_biology_mean_token_accuracy": 0.47685695719718935,
+      "eval_biology_num_tokens": 36398285.0,
+      "eval_biology_runtime": 49.8309,
+      "eval_biology_samples_per_second": 10.034,
+      "eval_biology_steps_per_second": 2.508,
+      "step": 1100
+    },
+    {
+      "epoch": 1.76,
+      "eval_chemistry_entropy": 1.250497200012207,
+      "eval_chemistry_loss": 1.3435908555984497,
+      "eval_chemistry_mean_token_accuracy": 0.7001057581901551,
+      "eval_chemistry_num_tokens": 36398285.0,
+      "eval_chemistry_runtime": 62.0199,
+      "eval_chemistry_samples_per_second": 8.062,
+      "eval_chemistry_steps_per_second": 2.015,
+      "step": 1100
+    },
+    {
+      "epoch": 1.76,
+      "eval_math_entropy": 0.9594085035324097,
+      "eval_math_loss": 1.1948144435882568,
+      "eval_math_mean_token_accuracy": 0.7322173299789428,
+      "eval_math_num_tokens": 36398285.0,
+      "eval_math_runtime": 63.3342,
+      "eval_math_samples_per_second": 7.895,
+      "eval_math_steps_per_second": 1.974,
+      "step": 1100
+    },
+    {
+      "epoch": 1.76,
+      "eval_physics_entropy": 0.8774992027282715,
+      "eval_physics_loss": 0.9080738425254822,
+      "eval_physics_mean_token_accuracy": 0.7721537561416626,
+      "eval_physics_num_tokens": 36398285.0,
+      "eval_physics_runtime": 72.1564,
+      "eval_physics_samples_per_second": 6.929,
+      "eval_physics_steps_per_second": 1.732,
+      "step": 1100
+    },
+    {
+      "entropy": 0.895181929320097,
+      "epoch": 1.776,
+      "grad_norm": 19.0,
+      "learning_rate": 1.400740740740741e-05,
+      "loss": 14.4407,
+      "mean_token_accuracy": 0.7728559862822294,
+      "num_tokens": 36730479.0,
+      "step": 1110
+    },
+    {
+      "entropy": 0.8563918903470039,
+      "epoch": 1.792,
+      "grad_norm": 23.5,
+      "learning_rate": 1.3933333333333334e-05,
+      "loss": 13.803,
+      "mean_token_accuracy": 0.7809405270963907,
+      "num_tokens": 37057425.0,
+      "step": 1120
+    },
+    {
+      "entropy": 0.8510978795588017,
+      "epoch": 1.808,
+      "grad_norm": 18.875,
+      "learning_rate": 1.385925925925926e-05,
+      "loss": 13.6333,
+      "mean_token_accuracy": 0.784480345249176,
+      "num_tokens": 37395449.0,
+      "step": 1130
+    },
+    {
+      "entropy": 0.8645633378997445,
+      "epoch": 1.8239999999999998,
+      "grad_norm": 18.875,
+      "learning_rate": 1.3785185185185186e-05,
+      "loss": 13.872,
+      "mean_token_accuracy": 0.7808197818696498,
+      "num_tokens": 37729535.0,
+      "step": 1140
+    },
+    {
+      "entropy": 0.8631477665156126,
+      "epoch": 1.8399999999999999,
+      "grad_norm": 19.625,
+      "learning_rate": 1.3711111111111112e-05,
+      "loss": 13.8345,
+      "mean_token_accuracy": 0.7815310817211867,
+      "num_tokens": 38057839.0,
+      "step": 1150
+    },
+    {
+      "entropy": 0.8587682455778122,
+      "epoch": 1.8559999999999999,
+      "grad_norm": 18.875,
+      "learning_rate": 1.3637037037037037e-05,
+      "loss": 13.8867,
+      "mean_token_accuracy": 0.780883165821433,
+      "num_tokens": 38398033.0,
+      "step": 1160
+    },
+    {
+      "entropy": 0.859000613167882,
+      "epoch": 1.8719999999999999,
+      "grad_norm": 17.875,
+      "learning_rate": 1.3562962962962965e-05,
+      "loss": 13.7667,
+      "mean_token_accuracy": 0.7798742011189461,
+      "num_tokens": 38729203.0,
+      "step": 1170
+    },
+    {
+      "entropy": 0.8777886133641004,
+      "epoch": 1.888,
+      "grad_norm": 19.625,
+      "learning_rate": 1.3488888888888888e-05,
+      "loss": 14.1636,
+      "mean_token_accuracy": 0.7765467241406441,
+      "num_tokens": 39058791.0,
+      "step": 1180
+    },
+    {
+      "entropy": 0.8881994623690843,
+      "epoch": 1.904,
+      "grad_norm": 20.125,
+      "learning_rate": 1.3414814814814817e-05,
+      "loss": 14.3781,
+      "mean_token_accuracy": 0.7752389151602983,
+      "num_tokens": 39386170.0,
+      "step": 1190
+    },
+    {
+      "entropy": 0.9030840078368783,
+      "epoch": 1.92,
+      "grad_norm": 19.5,
+      "learning_rate": 1.3340740740740741e-05,
+      "loss": 14.3564,
+      "mean_token_accuracy": 0.7748420361429453,
+      "num_tokens": 39710381.0,
+      "step": 1200
+    },
+    {
+      "epoch": 1.92,
+      "eval_biology_entropy": 2.697428282737732,
+      "eval_biology_loss": 3.138005256652832,
+      "eval_biology_mean_token_accuracy": 0.47676516604423524,
+      "eval_biology_num_tokens": 39710381.0,
+      "eval_biology_runtime": 49.822,
+      "eval_biology_samples_per_second": 10.036,
+      "eval_biology_steps_per_second": 2.509,
+      "step": 1200
+    },
+    {
+      "epoch": 1.92,
+      "eval_chemistry_entropy": 1.2482068204879762,
+      "eval_chemistry_loss": 1.3385791778564453,
+      "eval_chemistry_mean_token_accuracy": 0.7002091021537781,
+      "eval_chemistry_num_tokens": 39710381.0,
+      "eval_chemistry_runtime": 61.7709,
+      "eval_chemistry_samples_per_second": 8.094,
+      "eval_chemistry_steps_per_second": 2.024,
+      "step": 1200
+    },
+    {
+      "epoch": 1.92,
+      "eval_math_entropy": 0.9588226528167725,
+      "eval_math_loss": 1.190582036972046,
+      "eval_math_mean_token_accuracy": 0.7330412902832031,
+      "eval_math_num_tokens": 39710381.0,
+      "eval_math_runtime": 63.3698,
+      "eval_math_samples_per_second": 7.89,
+      "eval_math_steps_per_second": 1.973,
+      "step": 1200
+    },
+    {
+      "epoch": 1.92,
+      "eval_physics_entropy": 0.8757374310493469,
+      "eval_physics_loss": 0.9024468064308167,
+      "eval_physics_mean_token_accuracy": 0.7736027889251709,
+      "eval_physics_num_tokens": 39710381.0,
+      "eval_physics_runtime": 72.1456,
+      "eval_physics_samples_per_second": 6.93,
+      "eval_physics_steps_per_second": 1.733,
+      "step": 1200
+    },
+    {
+      "entropy": 0.8624950293451548,
+      "epoch": 1.936,
+      "grad_norm": 19.375,
+      "learning_rate": 1.3266666666666668e-05,
+      "loss": 13.9438,
+      "mean_token_accuracy": 0.7779798842966557,
+      "num_tokens": 40043354.0,
+      "step": 1210
+    },
+    {
+      "entropy": 0.8265730097889901,
+      "epoch": 1.952,
+      "grad_norm": 18.75,
+      "learning_rate": 1.3192592592592594e-05,
+      "loss": 13.1879,
+      "mean_token_accuracy": 0.7890967357903719,
+      "num_tokens": 40375790.0,
+      "step": 1220
+    },
+    {
+      "entropy": 0.8566216921433807,
+      "epoch": 1.968,
+      "grad_norm": 18.375,
+      "learning_rate": 1.311851851851852e-05,
+      "loss": 13.8742,
+      "mean_token_accuracy": 0.780952601134777,
+      "num_tokens": 40705451.0,
+      "step": 1230
+    },
+    {
+      "entropy": 0.8784732647240162,
+      "epoch": 1.984,
+      "grad_norm": 20.625,
+      "learning_rate": 1.3044444444444446e-05,
+      "loss": 14.1064,
+      "mean_token_accuracy": 0.776585515961051,
+      "num_tokens": 41040313.0,
+      "step": 1240
+    },
+    {
+      "entropy": 0.8804466765373945,
+      "epoch": 2.0,
+      "grad_norm": 18.5,
+      "learning_rate": 1.297037037037037e-05,
+      "loss": 14.0925,
+      "mean_token_accuracy": 0.7788621835410595,
+      "num_tokens": 41373608.0,
+      "step": 1250
+    },
+    {
+      "entropy": 0.8347923129796981,
+      "epoch": 2.016,
+      "grad_norm": 16.125,
+      "learning_rate": 1.2896296296296299e-05,
+      "loss": 13.3304,
+      "mean_token_accuracy": 0.7866924706846475,
+      "num_tokens": 41708226.0,
+      "step": 1260
+    },
+    {
+      "entropy": 0.8343301571905613,
+      "epoch": 2.032,
+      "grad_norm": 18.5,
+      "learning_rate": 1.2822222222222222e-05,
+      "loss": 13.426,
+      "mean_token_accuracy": 0.7863655511289835,
+      "num_tokens": 42045433.0,
+      "step": 1270
+    },
+    {
+      "entropy": 0.8626088196411729,
+      "epoch": 2.048,
+      "grad_norm": 19.75,
+      "learning_rate": 1.274814814814815e-05,
+      "loss": 13.9137,
+      "mean_token_accuracy": 0.7789722084999084,
+      "num_tokens": 42372694.0,
+      "step": 1280
+    },
+    {
+      "entropy": 0.8391540687531233,
+      "epoch": 2.064,
+      "grad_norm": 17.75,
+      "learning_rate": 1.2674074074074075e-05,
+      "loss": 13.4434,
+      "mean_token_accuracy": 0.7854840617626906,
+      "num_tokens": 42710240.0,
+      "step": 1290
+    },
+    {
+      "entropy": 0.8561890006065369,
+      "epoch": 2.08,
+      "grad_norm": 17.5,
+      "learning_rate": 1.2600000000000001e-05,
+      "loss": 13.7956,
+      "mean_token_accuracy": 0.7818153217434883,
+      "num_tokens": 43037844.0,
+      "step": 1300
+    },
+    {
+      "epoch": 2.08,
+      "eval_biology_entropy": 2.680125617027283,
+      "eval_biology_loss": 3.154866933822632,
+      "eval_biology_mean_token_accuracy": 0.4771236026287079,
+      "eval_biology_num_tokens": 43037844.0,
+      "eval_biology_runtime": 49.8062,
+      "eval_biology_samples_per_second": 10.039,
+      "eval_biology_steps_per_second": 2.51,
+      "step": 1300
+    },
+    {
+      "epoch": 2.08,
+      "eval_chemistry_entropy": 1.23085289478302,
+      "eval_chemistry_loss": 1.339719533920288,
+      "eval_chemistry_mean_token_accuracy": 0.700504720211029,
+      "eval_chemistry_num_tokens": 43037844.0,
+      "eval_chemistry_runtime": 61.7767,
+      "eval_chemistry_samples_per_second": 8.094,
+      "eval_chemistry_steps_per_second": 2.023,
+      "step": 1300
+    },
+    {
+      "epoch": 2.08,
+      "eval_math_entropy": 0.9530718994140625,
+      "eval_math_loss": 1.1914963722229004,
+      "eval_math_mean_token_accuracy": 0.732671552658081,
+      "eval_math_num_tokens": 43037844.0,
+      "eval_math_runtime": 63.3384,
+      "eval_math_samples_per_second": 7.894,
+      "eval_math_steps_per_second": 1.974,
+      "step": 1300
+    },
+    {
+      "epoch": 2.08,
+      "eval_physics_entropy": 0.863683602809906,
+      "eval_physics_loss": 0.898903489112854,
+      "eval_physics_mean_token_accuracy": 0.7743104872703552,
+      "eval_physics_num_tokens": 43037844.0,
+      "eval_physics_runtime": 72.1486,
+      "eval_physics_samples_per_second": 6.93,
+      "eval_physics_steps_per_second": 1.733,
+      "step": 1300
+    },
+    {
+      "entropy": 0.8574293557554483,
+      "epoch": 2.096,
+      "grad_norm": 19.5,
+      "learning_rate": 1.2525925925925928e-05,
+      "loss": 13.697,
+      "mean_token_accuracy": 0.7824351858347655,
+      "num_tokens": 43368675.0,
+      "step": 1310
+    },
+    {
+      "entropy": 0.8351046226918697,
+      "epoch": 2.112,
+      "grad_norm": 18.375,
+      "learning_rate": 1.2451851851851853e-05,
+      "loss": 13.452,
+      "mean_token_accuracy": 0.7861107390373945,
+      "num_tokens": 43699544.0,
+      "step": 1320
+    },
+    {
+      "entropy": 0.848756181821227,
+      "epoch": 2.128,
+      "grad_norm": 19.375,
+      "learning_rate": 1.237777777777778e-05,
+      "loss": 13.6014,
+      "mean_token_accuracy": 0.7825081821531057,
+      "num_tokens": 44012632.0,
+      "step": 1330
+    },
+    {
+      "entropy": 0.8376241635531188,
+      "epoch": 2.144,
+      "grad_norm": 17.875,
+      "learning_rate": 1.2303703703703704e-05,
+      "loss": 13.5012,
+      "mean_token_accuracy": 0.7842539783567191,
+      "num_tokens": 44349732.0,
+      "step": 1340
+    },
+    {
+      "entropy": 0.8786815196275711,
+      "epoch": 2.16,
+      "grad_norm": 19.75,
+      "learning_rate": 1.222962962962963e-05,
+      "loss": 14.1221,
+      "mean_token_accuracy": 0.7767074994742871,
+      "num_tokens": 44680410.0,
+      "step": 1350
+    },
+    {
+      "entropy": 0.8388923730701208,
+      "epoch": 2.176,
+      "grad_norm": 17.75,
+      "learning_rate": 1.2155555555555555e-05,
+      "loss": 13.4611,
+      "mean_token_accuracy": 0.7856471933424473,
+      "num_tokens": 45017922.0,
+      "step": 1360
+    },
+    {
+      "entropy": 0.8349485833197832,
+      "epoch": 2.192,
+      "grad_norm": 18.625,
+      "learning_rate": 1.2081481481481484e-05,
+      "loss": 13.3623,
+      "mean_token_accuracy": 0.7861445616930723,
+      "num_tokens": 45348179.0,
+      "step": 1370
+    },
+    {
+      "entropy": 0.8328462485224009,
+      "epoch": 2.208,
+      "grad_norm": 21.0,
+      "learning_rate": 1.2007407407407408e-05,
+      "loss": 13.3785,
+      "mean_token_accuracy": 0.7866237808018923,
+      "num_tokens": 45678397.0,
+      "step": 1380
+    },
+    {
+      "entropy": 0.8291925746947527,
+      "epoch": 2.224,
+      "grad_norm": 17.75,
+      "learning_rate": 1.1933333333333335e-05,
+      "loss": 13.328,
+      "mean_token_accuracy": 0.7863113507628441,
+      "num_tokens": 46010052.0,
+      "step": 1390
+    },
+    {
+      "entropy": 0.8287136562168598,
+      "epoch": 2.24,
+      "grad_norm": 18.0,
+      "learning_rate": 1.185925925925926e-05,
+      "loss": 13.3318,
+      "mean_token_accuracy": 0.7863692469894886,
+      "num_tokens": 46347116.0,
+      "step": 1400
+    },
+    {
+      "epoch": 2.24,
+      "eval_biology_entropy": 2.6612108364105223,
+      "eval_biology_loss": 3.153256416320801,
+      "eval_biology_mean_token_accuracy": 0.4783942770957947,
+      "eval_biology_num_tokens": 46347116.0,
+      "eval_biology_runtime": 49.9149,
+      "eval_biology_samples_per_second": 10.017,
+      "eval_biology_steps_per_second": 2.504,
+      "step": 1400
+    },
+    {
+      "epoch": 2.24,
+      "eval_chemistry_entropy": 1.2294214334487914,
+      "eval_chemistry_loss": 1.3362380266189575,
+      "eval_chemistry_mean_token_accuracy": 0.701572069644928,
+      "eval_chemistry_num_tokens": 46347116.0,
+      "eval_chemistry_runtime": 61.8559,
+      "eval_chemistry_samples_per_second": 8.083,
+      "eval_chemistry_steps_per_second": 2.021,
+      "step": 1400
+    },
+    {
+      "epoch": 2.24,
+      "eval_math_entropy": 0.9489249901771546,
+      "eval_math_loss": 1.1918082237243652,
+      "eval_math_mean_token_accuracy": 0.7329039392471314,
+      "eval_math_num_tokens": 46347116.0,
+      "eval_math_runtime": 63.4236,
+      "eval_math_samples_per_second": 7.884,
+      "eval_math_steps_per_second": 1.971,
+      "step": 1400
+    },
+    {
+      "epoch": 2.24,
+      "eval_physics_entropy": 0.8618579044342041,
+      "eval_physics_loss": 0.8958038687705994,
+      "eval_physics_mean_token_accuracy": 0.7750196180343628,
+      "eval_physics_num_tokens": 46347116.0,
+      "eval_physics_runtime": 72.3632,
+      "eval_physics_samples_per_second": 6.91,
+      "eval_physics_steps_per_second": 1.727,
+      "step": 1400
+    },
+    {
+      "entropy": 0.8701908178627491,
+      "epoch": 2.2560000000000002,
+      "grad_norm": 19.25,
+      "learning_rate": 1.1785185185185186e-05,
+      "loss": 13.9577,
+      "mean_token_accuracy": 0.7795143604278565,
+      "num_tokens": 46673074.0,
+      "step": 1410
+    },
+    {
+      "entropy": 0.8190190486609936,
+      "epoch": 2.2720000000000002,
+      "grad_norm": 18.875,
+      "learning_rate": 1.1711111111111113e-05,
+      "loss": 13.1852,
+      "mean_token_accuracy": 0.7890499792993069,
+      "num_tokens": 47009386.0,
+      "step": 1420
+    },
+    {
+      "entropy": 0.8609435971826315,
+      "epoch": 2.288,
+      "grad_norm": 19.75,
+      "learning_rate": 1.1637037037037037e-05,
+      "loss": 13.786,
+      "mean_token_accuracy": 0.7809262648224831,
+      "num_tokens": 47344231.0,
+      "step": 1430
+    },
+    {
+      "entropy": 0.8457425959408283,
+      "epoch": 2.304,
+      "grad_norm": 20.125,
+      "learning_rate": 1.1562962962962964e-05,
+      "loss": 13.6528,
+      "mean_token_accuracy": 0.7828251335769891,
+      "num_tokens": 47658392.0,
+      "step": 1440
+    },
+    {
+      "entropy": 0.8437560629099607,
+      "epoch": 2.32,
+      "grad_norm": 17.25,
+      "learning_rate": 1.1488888888888889e-05,
+      "loss": 13.4697,
+      "mean_token_accuracy": 0.7838904622942209,
+      "num_tokens": 47995015.0,
+      "step": 1450
+    },
+    {
+      "entropy": 0.8140388157218694,
+      "epoch": 2.336,
+      "grad_norm": 17.875,
+      "learning_rate": 1.1414814814814817e-05,
+      "loss": 13.1535,
+      "mean_token_accuracy": 0.7884997557848692,
+      "num_tokens": 48336708.0,
+      "step": 1460
+    },
+    {
+      "entropy": 0.8194112915545702,
+      "epoch": 2.352,
+      "grad_norm": 17.5,
+      "learning_rate": 1.1340740740740742e-05,
+      "loss": 13.0636,
+      "mean_token_accuracy": 0.7896034978330135,
+      "num_tokens": 48665311.0,
+      "step": 1470
+    },
+    {
+      "entropy": 0.8433867402374744,
+      "epoch": 2.368,
+      "grad_norm": 18.625,
+      "learning_rate": 1.1266666666666668e-05,
+      "loss": 13.5528,
+      "mean_token_accuracy": 0.7844488836824894,
+      "num_tokens": 49005334.0,
+      "step": 1480
+    },
+    {
+      "entropy": 0.8540257982909679,
+      "epoch": 2.384,
+      "grad_norm": 20.625,
+      "learning_rate": 1.1192592592592593e-05,
+      "loss": 13.7212,
+      "mean_token_accuracy": 0.781413323059678,
+      "num_tokens": 49331936.0,
+      "step": 1490
+    },
+    {
+      "entropy": 0.8498314812779426,
+      "epoch": 2.4,
+      "grad_norm": 19.25,
+      "learning_rate": 1.111851851851852e-05,
+      "loss": 13.604,
+      "mean_token_accuracy": 0.7828326590359211,
+      "num_tokens": 49671048.0,
+      "step": 1500
+    },
+    {
+      "epoch": 2.4,
+      "eval_biology_entropy": 2.621306725502014,
+      "eval_biology_loss": 3.1416707038879395,
+      "eval_biology_mean_token_accuracy": 0.4782486002445221,
+      "eval_biology_num_tokens": 49671048.0,
+      "eval_biology_runtime": 49.9245,
+      "eval_biology_samples_per_second": 10.015,
+      "eval_biology_steps_per_second": 2.504,
+      "step": 1500
+    },
+    {
+      "epoch": 2.4,
+      "eval_chemistry_entropy": 1.223756856918335,
+      "eval_chemistry_loss": 1.3346976041793823,
+      "eval_chemistry_mean_token_accuracy": 0.7012901954650879,
+      "eval_chemistry_num_tokens": 49671048.0,
+      "eval_chemistry_runtime": 61.475,
+      "eval_chemistry_samples_per_second": 8.133,
+      "eval_chemistry_steps_per_second": 2.033,
+      "step": 1500
+    },
+    {
+      "epoch": 2.4,
+      "eval_math_entropy": 0.9483405375480651,
+      "eval_math_loss": 1.19205641746521,
+      "eval_math_mean_token_accuracy": 0.7328525404930115,
+      "eval_math_num_tokens": 49671048.0,
+      "eval_math_runtime": 63.2408,
+      "eval_math_samples_per_second": 7.906,
+      "eval_math_steps_per_second": 1.977,
+      "step": 1500
+    },
+    {
+      "epoch": 2.4,
+      "eval_physics_entropy": 0.8615480012893677,
+      "eval_physics_loss": 0.8934000134468079,
+      "eval_physics_mean_token_accuracy": 0.7753945760726929,
+      "eval_physics_num_tokens": 49671048.0,
+      "eval_physics_runtime": 72.1452,
+      "eval_physics_samples_per_second": 6.93,
+      "eval_physics_steps_per_second": 1.733,
+      "step": 1500
+    },
+    {
+      "entropy": 0.8307130090892315,
+      "epoch": 2.416,
+      "grad_norm": 17.5,
+      "learning_rate": 1.1044444444444444e-05,
+      "loss": 13.3888,
+      "mean_token_accuracy": 0.7861995015293359,
+      "num_tokens": 50018182.0,
+      "step": 1510
+    },
+    {
+      "entropy": 0.8422547899186611,
+      "epoch": 2.432,
+      "grad_norm": 20.125,
+      "learning_rate": 1.0970370370370371e-05,
+      "loss": 13.4992,
+      "mean_token_accuracy": 0.7854960076510906,
+      "num_tokens": 50341187.0,
+      "step": 1520
+    },
+    {
+      "entropy": 0.8614559264853596,
+      "epoch": 2.448,
+      "grad_norm": 18.0,
+      "learning_rate": 1.0896296296296298e-05,
+      "loss": 13.915,
+      "mean_token_accuracy": 0.7800705958157778,
+      "num_tokens": 50667019.0,
+      "step": 1530
+    },
+    {
+      "entropy": 0.8549725610762835,
+      "epoch": 2.464,
+      "grad_norm": 19.75,
+      "learning_rate": 1.0822222222222222e-05,
+      "loss": 13.6406,
+      "mean_token_accuracy": 0.7828543521463871,
+      "num_tokens": 50991113.0,
+      "step": 1540
+    },
+    {
+      "entropy": 0.8388963419944048,
+      "epoch": 2.48,
+      "grad_norm": 18.875,
+      "learning_rate": 1.074814814814815e-05,
+      "loss": 13.562,
+      "mean_token_accuracy": 0.7833587605506181,
+      "num_tokens": 51319624.0,
+      "step": 1550
+    },
+    {
+      "entropy": 0.8547497015446425,
+      "epoch": 2.496,
+      "grad_norm": 17.875,
+      "learning_rate": 1.0674074074074074e-05,
+      "loss": 13.6692,
+      "mean_token_accuracy": 0.7828979298472405,
+      "num_tokens": 51654633.0,
+      "step": 1560
+    },
+    {
+      "entropy": 0.8275569640100002,
+      "epoch": 2.512,
+      "grad_norm": 20.75,
+      "learning_rate": 1.0600000000000002e-05,
+      "loss": 13.3135,
+      "mean_token_accuracy": 0.7875300694257021,
+      "num_tokens": 51980051.0,
+      "step": 1570
+    },
+    {
+      "entropy": 0.8288551384583116,
+      "epoch": 2.528,
+      "grad_norm": 17.25,
+      "learning_rate": 1.0525925925925927e-05,
+      "loss": 13.3319,
+      "mean_token_accuracy": 0.7867311630398035,
+      "num_tokens": 52316109.0,
+      "step": 1580
+    },
+    {
+      "entropy": 0.8469713069498539,
+      "epoch": 2.544,
+      "grad_norm": 20.375,
+      "learning_rate": 1.0451851851851853e-05,
+      "loss": 13.664,
+      "mean_token_accuracy": 0.7819825943559409,
+      "num_tokens": 52634528.0,
+      "step": 1590
+    },
+    {
+      "entropy": 0.8472158204764128,
+      "epoch": 2.56,
+      "grad_norm": 20.5,
+      "learning_rate": 1.0377777777777778e-05,
+      "loss": 13.4411,
+      "mean_token_accuracy": 0.7855797190219164,
+      "num_tokens": 52964035.0,
+      "step": 1600
+    },
+    {
+      "epoch": 2.56,
+      "eval_biology_entropy": 2.636871319770813,
+      "eval_biology_loss": 3.152658224105835,
+      "eval_biology_mean_token_accuracy": 0.4772343513965607,
+      "eval_biology_num_tokens": 52964035.0,
+      "eval_biology_runtime": 50.2549,
+      "eval_biology_samples_per_second": 9.949,
+      "eval_biology_steps_per_second": 2.487,
+      "step": 1600
+    },
+    {
+      "epoch": 2.56,
+      "eval_chemistry_entropy": 1.2150323610305787,
+      "eval_chemistry_loss": 1.332220435142517,
+      "eval_chemistry_mean_token_accuracy": 0.7014281511306762,
+      "eval_chemistry_num_tokens": 52964035.0,
+      "eval_chemistry_runtime": 62.5654,
+      "eval_chemistry_samples_per_second": 7.992,
+      "eval_chemistry_steps_per_second": 1.998,
+      "step": 1600
+    },
+    {
+      "epoch": 2.56,
+      "eval_math_entropy": 0.9430220746994018,
+      "eval_math_loss": 1.1897228956222534,
+      "eval_math_mean_token_accuracy": 0.7337738809585571,
+      "eval_math_num_tokens": 52964035.0,
+      "eval_math_runtime": 63.65,
+      "eval_math_samples_per_second": 7.855,
+      "eval_math_steps_per_second": 1.964,
+      "step": 1600
+    },
+    {
+      "epoch": 2.56,
+      "eval_physics_entropy": 0.8505253372192383,
+      "eval_physics_loss": 0.8909267783164978,
+      "eval_physics_mean_token_accuracy": 0.775915168762207,
+      "eval_physics_num_tokens": 52964035.0,
+      "eval_physics_runtime": 72.3847,
+      "eval_physics_samples_per_second": 6.908,
+      "eval_physics_steps_per_second": 1.727,
+      "step": 1600
+    },
+    {
+      "entropy": 0.8204018078744412,
+      "epoch": 2.576,
+      "grad_norm": 20.25,
+      "learning_rate": 1.0303703703703705e-05,
+      "loss": 13.224,
+      "mean_token_accuracy": 0.7873958699405194,
+      "num_tokens": 53293345.0,
+      "step": 1610
+    },
+    {
+      "entropy": 0.8367499966174364,
+      "epoch": 2.592,
+      "grad_norm": 16.0,
+      "learning_rate": 1.0229629629629631e-05,
+      "loss": 13.4302,
+      "mean_token_accuracy": 0.7862669274210929,
+      "num_tokens": 53620896.0,
+      "step": 1620
+    },
+    {
+      "entropy": 0.8095174605026841,
+      "epoch": 2.608,
+      "grad_norm": 18.125,
+      "learning_rate": 1.0155555555555556e-05,
+      "loss": 13.0771,
+      "mean_token_accuracy": 0.7906746052205562,
+      "num_tokens": 53950989.0,
+      "step": 1630
+    },
+    {
+      "entropy": 0.8320839395746589,
+      "epoch": 2.624,
+      "grad_norm": 18.625,
+      "learning_rate": 1.0081481481481484e-05,
+      "loss": 13.417,
+      "mean_token_accuracy": 0.7861438430845737,
+      "num_tokens": 54282732.0,
+      "step": 1640
+    },
+    {
+      "entropy": 0.8359093256294727,
+      "epoch": 2.64,
+      "grad_norm": 19.375,
+      "learning_rate": 1.0007407407407407e-05,
+      "loss": 13.4375,
+      "mean_token_accuracy": 0.7856674194335938,
+      "num_tokens": 54609748.0,
+      "step": 1650
+    },
+    {
+      "entropy": 0.8712694313377142,
+      "epoch": 2.656,
+      "grad_norm": 19.5,
+      "learning_rate": 9.933333333333334e-06,
+      "loss": 14.0354,
+      "mean_token_accuracy": 0.778102644905448,
+      "num_tokens": 54953583.0,
+      "step": 1660
+    },
+    {
+      "entropy": 0.8449299931526184,
+      "epoch": 2.672,
+      "grad_norm": 19.75,
+      "learning_rate": 9.85925925925926e-06,
+      "loss": 13.577,
+      "mean_token_accuracy": 0.7835838712751866,
+      "num_tokens": 55281945.0,
+      "step": 1670
+    },
+    {
+      "entropy": 0.8184873767197132,
+      "epoch": 2.6879999999999997,
+      "grad_norm": 19.625,
+      "learning_rate": 9.785185185185187e-06,
+      "loss": 13.0566,
+      "mean_token_accuracy": 0.7910151798278093,
+      "num_tokens": 55612391.0,
+      "step": 1680
+    },
+    {
+      "entropy": 0.827063605375588,
+      "epoch": 2.7039999999999997,
+      "grad_norm": 17.625,
+      "learning_rate": 9.711111111111111e-06,
+      "loss": 13.4211,
+      "mean_token_accuracy": 0.7862589839845896,
+      "num_tokens": 55949791.0,
+      "step": 1690
+    },
+    {
+      "entropy": 0.839899223856628,
+      "epoch": 2.7199999999999998,
+      "grad_norm": 21.0,
+      "learning_rate": 9.637037037037038e-06,
+      "loss": 13.3423,
+      "mean_token_accuracy": 0.7863923076540231,
+      "num_tokens": 56282204.0,
+      "step": 1700
+    },
+    {
+      "epoch": 2.7199999999999998,
+      "eval_biology_entropy": 2.5845158891677857,
+      "eval_biology_loss": 3.161525011062622,
+      "eval_biology_mean_token_accuracy": 0.47775351333618166,
+      "eval_biology_num_tokens": 56282204.0,
+      "eval_biology_runtime": 50.3616,
+      "eval_biology_samples_per_second": 9.928,
+      "eval_biology_steps_per_second": 2.482,
+      "step": 1700
+    },
+    {
+      "epoch": 2.7199999999999998,
+      "eval_chemistry_entropy": 1.2002769947052,
+      "eval_chemistry_loss": 1.3316991329193115,
+      "eval_chemistry_mean_token_accuracy": 0.7020627927780151,
+      "eval_chemistry_num_tokens": 56282204.0,
+      "eval_chemistry_runtime": 62.7009,
+      "eval_chemistry_samples_per_second": 7.974,
+      "eval_chemistry_steps_per_second": 1.994,
+      "step": 1700
+    },
+    {
+      "epoch": 2.7199999999999998,
+      "eval_math_entropy": 0.9366156039237976,
+      "eval_math_loss": 1.1879971027374268,
+      "eval_math_mean_token_accuracy": 0.7340071239471435,
+      "eval_math_num_tokens": 56282204.0,
+      "eval_math_runtime": 63.516,
+      "eval_math_samples_per_second": 7.872,
+      "eval_math_steps_per_second": 1.968,
+      "step": 1700
+    },
+    {
+      "epoch": 2.7199999999999998,
+      "eval_physics_entropy": 0.8443611931800842,
+      "eval_physics_loss": 0.8887805342674255,
+      "eval_physics_mean_token_accuracy": 0.7763362097740173,
+      "eval_physics_num_tokens": 56282204.0,
+      "eval_physics_runtime": 72.6341,
+      "eval_physics_samples_per_second": 6.884,
+      "eval_physics_steps_per_second": 1.721,
+      "step": 1700
+    },
+    {
+      "entropy": 0.8606409896165133,
+      "epoch": 2.7359999999999998,
+      "grad_norm": 19.25,
+      "learning_rate": 9.562962962962965e-06,
+      "loss": 13.9963,
+      "mean_token_accuracy": 0.7781527418643236,
+      "num_tokens": 56605478.0,
+      "step": 1710
+    },
+    {
+      "entropy": 0.8518755808472633,
+      "epoch": 2.752,
+      "grad_norm": 18.25,
+      "learning_rate": 9.48888888888889e-06,
+      "loss": 13.6621,
+      "mean_token_accuracy": 0.7833192612975836,
+      "num_tokens": 56935874.0,
+      "step": 1720
+    },
+    {
+      "entropy": 0.8237534828484059,
+      "epoch": 2.768,
+      "grad_norm": 19.625,
+      "learning_rate": 9.414814814814816e-06,
+      "loss": 13.2122,
+      "mean_token_accuracy": 0.7885770782828331,
+      "num_tokens": 57270552.0,
+      "step": 1730
+    },
+    {
+      "entropy": 0.8422744061797858,
+      "epoch": 2.784,
+      "grad_norm": 19.75,
+      "learning_rate": 9.34074074074074e-06,
+      "loss": 13.4941,
+      "mean_token_accuracy": 0.7843061711639165,
+      "num_tokens": 57602968.0,
+      "step": 1740
+    },
+    {
+      "entropy": 0.8839327782392502,
+      "epoch": 2.8,
+      "grad_norm": 20.375,
+      "learning_rate": 9.266666666666667e-06,
+      "loss": 14.309,
+      "mean_token_accuracy": 0.7745042227208614,
+      "num_tokens": 57925886.0,
+      "step": 1750
+    },
+    {
+      "entropy": 0.8240670500323176,
+      "epoch": 2.816,
+      "grad_norm": 18.375,
+      "learning_rate": 9.192592592592594e-06,
+      "loss": 13.0772,
+      "mean_token_accuracy": 0.7897984176874161,
+      "num_tokens": 58262960.0,
+      "step": 1760
+    },
+    {
+      "entropy": 0.8412995642051101,
+      "epoch": 2.832,
+      "grad_norm": 18.75,
+      "learning_rate": 9.118518518518518e-06,
+      "loss": 13.6619,
+      "mean_token_accuracy": 0.781951853260398,
+      "num_tokens": 58586324.0,
+      "step": 1770
+    },
+    {
+      "entropy": 0.842792528681457,
+      "epoch": 2.848,
+      "grad_norm": 18.5,
+      "learning_rate": 9.044444444444445e-06,
+      "loss": 13.4669,
+      "mean_token_accuracy": 0.7849409289658069,
+      "num_tokens": 58912949.0,
+      "step": 1780
+    },
+    {
+      "entropy": 0.8247644234448671,
+      "epoch": 2.864,
+      "grad_norm": 17.375,
+      "learning_rate": 8.970370370370372e-06,
+      "loss": 13.3001,
+      "mean_token_accuracy": 0.7879149217158556,
+      "num_tokens": 59252494.0,
+      "step": 1790
+    },
+    {
+      "entropy": 0.8388911845162511,
+      "epoch": 2.88,
+      "grad_norm": 22.75,
+      "learning_rate": 8.896296296296298e-06,
+      "loss": 13.4581,
+      "mean_token_accuracy": 0.7853025410324335,
+      "num_tokens": 59574951.0,
+      "step": 1800
+    },
+    {
+      "epoch": 2.88,
+      "eval_biology_entropy": 2.6436156339645387,
+      "eval_biology_loss": 3.1234591007232666,
+      "eval_biology_mean_token_accuracy": 0.47976515793800356,
+      "eval_biology_num_tokens": 59574951.0,
+      "eval_biology_runtime": 49.9354,
+      "eval_biology_samples_per_second": 10.013,
+      "eval_biology_steps_per_second": 2.503,
+      "step": 1800
+    },
+    {
+      "epoch": 2.88,
+      "eval_chemistry_entropy": 1.2157120180130006,
+      "eval_chemistry_loss": 1.327707052230835,
+      "eval_chemistry_mean_token_accuracy": 0.7026024498939514,
+      "eval_chemistry_num_tokens": 59574951.0,
+      "eval_chemistry_runtime": 61.896,
+      "eval_chemistry_samples_per_second": 8.078,
+      "eval_chemistry_steps_per_second": 2.02,
+      "step": 1800
+    },
+    {
+      "epoch": 2.88,
+      "eval_math_entropy": 0.9411606469154358,
+      "eval_math_loss": 1.1865390539169312,
+      "eval_math_mean_token_accuracy": 0.733888931274414,
+      "eval_math_num_tokens": 59574951.0,
+      "eval_math_runtime": 63.4329,
+      "eval_math_samples_per_second": 7.882,
+      "eval_math_steps_per_second": 1.971,
+      "step": 1800
+    },
+    {
+      "epoch": 2.88,
+      "eval_physics_entropy": 0.8502533750534058,
+      "eval_physics_loss": 0.8870522379875183,
+      "eval_physics_mean_token_accuracy": 0.7766082410812378,
+      "eval_physics_num_tokens": 59574951.0,
+      "eval_physics_runtime": 72.3169,
+      "eval_physics_samples_per_second": 6.914,
+      "eval_physics_steps_per_second": 1.729,
+      "step": 1800
+    },
+    {
+      "entropy": 0.831662517786026,
+      "epoch": 2.896,
+      "grad_norm": 19.125,
+      "learning_rate": 8.822222222222223e-06,
+      "loss": 13.3788,
+      "mean_token_accuracy": 0.7865715123713016,
+      "num_tokens": 59903274.0,
+      "step": 1810
+    },
+    {
+      "entropy": 0.8292084554210305,
+      "epoch": 2.912,
+      "grad_norm": 19.625,
+      "learning_rate": 8.74814814814815e-06,
+      "loss": 13.3071,
+      "mean_token_accuracy": 0.7875285055488348,
+      "num_tokens": 60234203.0,
+      "step": 1820
+    },
+    {
+      "entropy": 0.8230883669108152,
+      "epoch": 2.928,
+      "grad_norm": 19.0,
+      "learning_rate": 8.674074074074074e-06,
+      "loss": 13.2304,
+      "mean_token_accuracy": 0.7883638426661491,
+      "num_tokens": 60559123.0,
+      "step": 1830
+    },
+    {
+      "entropy": 0.8234717439860105,
+      "epoch": 2.944,
+      "grad_norm": 21.125,
+      "learning_rate": 8.6e-06,
+      "loss": 13.2157,
+      "mean_token_accuracy": 0.7892068441957235,
+      "num_tokens": 60893423.0,
+      "step": 1840
+    },
+    {
+      "entropy": 0.834986207447946,
+      "epoch": 2.96,
+      "grad_norm": 17.5,
+      "learning_rate": 8.525925925925927e-06,
+      "loss": 13.4812,
+      "mean_token_accuracy": 0.7855305273085833,
+      "num_tokens": 61228692.0,
+      "step": 1850
+    },
+    {
+      "entropy": 0.839332346059382,
+      "epoch": 2.976,
+      "grad_norm": 18.375,
+      "learning_rate": 8.451851851851852e-06,
+      "loss": 13.4263,
+      "mean_token_accuracy": 0.7843829181045294,
+      "num_tokens": 61560852.0,
+      "step": 1860
+    },
+    {
+      "entropy": 0.8094001712277532,
+      "epoch": 2.992,
+      "grad_norm": 18.625,
+      "learning_rate": 8.377777777777779e-06,
+      "loss": 13.0269,
+      "mean_token_accuracy": 0.7902657438069582,
+      "num_tokens": 61898951.0,
+      "step": 1870
+    },
+    {
+      "entropy": 0.8142060209065676,
+      "epoch": 3.008,
+      "grad_norm": 17.625,
+      "learning_rate": 8.303703703703705e-06,
+      "loss": 13.047,
+      "mean_token_accuracy": 0.7903710417449474,
+      "num_tokens": 62223875.0,
+      "step": 1880
+    },
+    {
+      "entropy": 0.8257684953510761,
+      "epoch": 3.024,
+      "grad_norm": 18.5,
+      "learning_rate": 8.229629629629632e-06,
+      "loss": 13.2072,
+      "mean_token_accuracy": 0.7877598956227303,
+      "num_tokens": 62553166.0,
+      "step": 1890
+    },
+    {
+      "entropy": 0.8063547797501087,
+      "epoch": 3.04,
+      "grad_norm": 19.375,
+      "learning_rate": 8.155555555555556e-06,
+      "loss": 12.9977,
+      "mean_token_accuracy": 0.7906218387186528,
+      "num_tokens": 62893113.0,
+      "step": 1900
+    },
+    {
+      "epoch": 3.04,
+      "eval_biology_entropy": 2.6121346549987794,
+      "eval_biology_loss": 3.140951633453369,
+      "eval_biology_mean_token_accuracy": 0.4791584732532501,
+      "eval_biology_num_tokens": 62893113.0,
+      "eval_biology_runtime": 50.0258,
+      "eval_biology_samples_per_second": 9.995,
+      "eval_biology_steps_per_second": 2.499,
+      "step": 1900
+    },
+    {
+      "epoch": 3.04,
+      "eval_chemistry_entropy": 1.2029799346923828,
+      "eval_chemistry_loss": 1.3304651975631714,
+      "eval_chemistry_mean_token_accuracy": 0.7021719708442687,
+      "eval_chemistry_num_tokens": 62893113.0,
+      "eval_chemistry_runtime": 61.9254,
+      "eval_chemistry_samples_per_second": 8.074,
+      "eval_chemistry_steps_per_second": 2.019,
+      "step": 1900
+    },
+    {
+      "epoch": 3.04,
+      "eval_math_entropy": 0.9358730678558349,
+      "eval_math_loss": 1.189276099205017,
+      "eval_math_mean_token_accuracy": 0.7338773818016052,
+      "eval_math_num_tokens": 62893113.0,
+      "eval_math_runtime": 63.449,
+      "eval_math_samples_per_second": 7.88,
+      "eval_math_steps_per_second": 1.97,
+      "step": 1900
+    },
+    {
+      "epoch": 3.04,
+      "eval_physics_entropy": 0.8415746116638183,
+      "eval_physics_loss": 0.88661128282547,
+      "eval_physics_mean_token_accuracy": 0.7766357960700989,
+      "eval_physics_num_tokens": 62893113.0,
+      "eval_physics_runtime": 72.4553,
+      "eval_physics_samples_per_second": 6.901,
+      "eval_physics_steps_per_second": 1.725,
+      "step": 1900
+    },
+    {
+      "entropy": 0.8349037554115057,
+      "epoch": 3.056,
+      "grad_norm": 20.25,
+      "learning_rate": 8.081481481481483e-06,
+      "loss": 13.4095,
+      "mean_token_accuracy": 0.7856102827936411,
+      "num_tokens": 63218088.0,
+      "step": 1910
+    },
+    {
+      "entropy": 0.8310254758223892,
+      "epoch": 3.072,
+      "grad_norm": 20.625,
+      "learning_rate": 8.007407407407408e-06,
+      "loss": 13.3177,
+      "mean_token_accuracy": 0.7858017560094595,
+      "num_tokens": 63551201.0,
+      "step": 1920
+    },
+    {
+      "entropy": 0.8434441756457091,
+      "epoch": 3.088,
+      "grad_norm": 19.5,
+      "learning_rate": 7.933333333333334e-06,
+      "loss": 13.5925,
+      "mean_token_accuracy": 0.7826492808759212,
+      "num_tokens": 63876362.0,
+      "step": 1930
+    },
+    {
+      "entropy": 0.8093680031597614,
+      "epoch": 3.104,
+      "grad_norm": 18.75,
+      "learning_rate": 7.859259259259259e-06,
+      "loss": 12.9195,
+      "mean_token_accuracy": 0.7914323821663857,
+      "num_tokens": 64208171.0,
+      "step": 1940
+    },
+    {
+      "entropy": 0.8109165525063873,
+      "epoch": 3.12,
+      "grad_norm": 18.625,
+      "learning_rate": 7.785185185185185e-06,
+      "loss": 12.965,
+      "mean_token_accuracy": 0.7911074563860894,
+      "num_tokens": 64544493.0,
+      "step": 1950
+    },
+    {
+      "entropy": 0.8407998465001583,
+      "epoch": 3.136,
+      "grad_norm": 19.375,
+      "learning_rate": 7.711111111111112e-06,
+      "loss": 13.5943,
+      "mean_token_accuracy": 0.7827663112431764,
+      "num_tokens": 64872488.0,
+      "step": 1960
+    },
+    {
+      "entropy": 0.8074584787711501,
+      "epoch": 3.152,
+      "grad_norm": 18.875,
+      "learning_rate": 7.637037037037037e-06,
+      "loss": 12.897,
+      "mean_token_accuracy": 0.7921464066952467,
+      "num_tokens": 65201124.0,
+      "step": 1970
+    },
+    {
+      "entropy": 0.8256604298949242,
+      "epoch": 3.168,
+      "grad_norm": 17.5,
+      "learning_rate": 7.562962962962963e-06,
+      "loss": 13.3492,
+      "mean_token_accuracy": 0.7878177661448718,
+      "num_tokens": 65525868.0,
+      "step": 1980
+    },
+    {
+      "entropy": 0.8148544320836664,
+      "epoch": 3.184,
+      "grad_norm": 19.5,
+      "learning_rate": 7.48888888888889e-06,
+      "loss": 13.0185,
+      "mean_token_accuracy": 0.7923109978437424,
+      "num_tokens": 65861603.0,
+      "step": 1990
+    },
+    {
+      "entropy": 0.825370734743774,
+      "epoch": 3.2,
+      "grad_norm": 18.25,
+      "learning_rate": 7.4148148148148155e-06,
+      "loss": 13.3044,
+      "mean_token_accuracy": 0.7872026544064283,
+      "num_tokens": 66193675.0,
+      "step": 2000
+    },
+    {
+      "epoch": 3.2,
+      "eval_biology_entropy": 2.592471206665039,
+      "eval_biology_loss": 3.1430797576904297,
+      "eval_biology_mean_token_accuracy": 0.47978607106208804,
+      "eval_biology_num_tokens": 66193675.0,
+      "eval_biology_runtime": 49.8434,
+      "eval_biology_samples_per_second": 10.031,
+      "eval_biology_steps_per_second": 2.508,
+      "step": 2000
+    },
+    {
+      "epoch": 3.2,
+      "eval_chemistry_entropy": 1.1954125108718872,
+      "eval_chemistry_loss": 1.3291075229644775,
+      "eval_chemistry_mean_token_accuracy": 0.7032857236862182,
+      "eval_chemistry_num_tokens": 66193675.0,
+      "eval_chemistry_runtime": 61.4502,
+      "eval_chemistry_samples_per_second": 8.137,
+      "eval_chemistry_steps_per_second": 2.034,
+      "step": 2000
+    },
+    {
+      "epoch": 3.2,
+      "eval_math_entropy": 0.932641517162323,
+      "eval_math_loss": 1.1883680820465088,
+      "eval_math_mean_token_accuracy": 0.7337643942832947,
+      "eval_math_num_tokens": 66193675.0,
+      "eval_math_runtime": 63.2406,
+      "eval_math_samples_per_second": 7.906,
+      "eval_math_steps_per_second": 1.977,
+      "step": 2000
+    },
+    {
+      "epoch": 3.2,
+      "eval_physics_entropy": 0.8384064409732819,
+      "eval_physics_loss": 0.885689377784729,
+      "eval_physics_mean_token_accuracy": 0.7765848093032837,
+      "eval_physics_num_tokens": 66193675.0,
+      "eval_physics_runtime": 72.1348,
+      "eval_physics_samples_per_second": 6.931,
+      "eval_physics_steps_per_second": 1.733,
+      "step": 2000
+    },
+    {
+      "entropy": 0.817348986491561,
+      "epoch": 3.216,
+      "grad_norm": 20.0,
+      "learning_rate": 7.340740740740742e-06,
+      "loss": 13.1509,
+      "mean_token_accuracy": 0.7882370110601187,
+      "num_tokens": 66522835.0,
+      "step": 2010
+    },
+    {
+      "entropy": 0.854836180433631,
+      "epoch": 3.232,
+      "grad_norm": 19.375,
+      "learning_rate": 7.266666666666668e-06,
+      "loss": 13.7487,
+      "mean_token_accuracy": 0.7814681313931942,
+      "num_tokens": 66853693.0,
+      "step": 2020
+    },
+    {
+      "entropy": 0.7906106799840927,
+      "epoch": 3.248,
+      "grad_norm": 20.625,
+      "learning_rate": 7.192592592592593e-06,
+      "loss": 12.6765,
+      "mean_token_accuracy": 0.7941000793129206,
+      "num_tokens": 67192299.0,
+      "step": 2030
+    },
+    {
+      "entropy": 0.814421945810318,
+      "epoch": 3.2640000000000002,
+      "grad_norm": 19.875,
+      "learning_rate": 7.118518518518519e-06,
+      "loss": 13.1479,
+      "mean_token_accuracy": 0.7905503310263157,
+      "num_tokens": 67517068.0,
+      "step": 2040
+    },
+    {
+      "entropy": 0.8242502415552735,
+      "epoch": 3.2800000000000002,
+      "grad_norm": 19.5,
+      "learning_rate": 7.044444444444445e-06,
+      "loss": 13.1118,
+      "mean_token_accuracy": 0.7895435027778148,
+      "num_tokens": 67841380.0,
+      "step": 2050
+    },
+    {
+      "entropy": 0.8079620378091932,
+      "epoch": 3.296,
+      "grad_norm": 18.125,
+      "learning_rate": 6.97037037037037e-06,
+      "loss": 13.0632,
+      "mean_token_accuracy": 0.7900959871709347,
+      "num_tokens": 68175287.0,
+      "step": 2060
+    },
+    {
+      "entropy": 0.821021930873394,
+      "epoch": 3.312,
+      "grad_norm": 19.5,
+      "learning_rate": 6.896296296296297e-06,
+      "loss": 13.1696,
+      "mean_token_accuracy": 0.7892089109867811,
+      "num_tokens": 68513030.0,
+      "step": 2070
+    },
+    {
+      "entropy": 0.8524089317768813,
+      "epoch": 3.328,
+      "grad_norm": 20.125,
+      "learning_rate": 6.8222222222222225e-06,
+      "loss": 13.7199,
+      "mean_token_accuracy": 0.7818126205354929,
+      "num_tokens": 68839883.0,
+      "step": 2080
+    },
+    {
+      "entropy": 0.8236784156411886,
+      "epoch": 3.344,
+      "grad_norm": 21.25,
+      "learning_rate": 6.748148148148149e-06,
+      "loss": 13.2044,
+      "mean_token_accuracy": 0.7871458046138287,
+      "num_tokens": 69173450.0,
+      "step": 2090
+    },
+    {
+      "entropy": 0.8218971570953727,
+      "epoch": 3.36,
+      "grad_norm": 19.125,
+      "learning_rate": 6.674074074074075e-06,
+      "loss": 13.258,
+      "mean_token_accuracy": 0.788434125483036,
+      "num_tokens": 69510512.0,
+      "step": 2100
+    },
+    {
+      "epoch": 3.36,
+      "eval_biology_entropy": 2.6040579748153685,
+      "eval_biology_loss": 3.151533365249634,
+      "eval_biology_mean_token_accuracy": 0.47923961186409,
+      "eval_biology_num_tokens": 69510512.0,
+      "eval_biology_runtime": 49.823,
+      "eval_biology_samples_per_second": 10.036,
+      "eval_biology_steps_per_second": 2.509,
+      "step": 2100
+    },
+    {
+      "epoch": 3.36,
+      "eval_chemistry_entropy": 1.204939450263977,
+      "eval_chemistry_loss": 1.3286526203155518,
+      "eval_chemistry_mean_token_accuracy": 0.7025578045845031,
+      "eval_chemistry_num_tokens": 69510512.0,
+      "eval_chemistry_runtime": 61.7492,
+      "eval_chemistry_samples_per_second": 8.097,
+      "eval_chemistry_steps_per_second": 2.024,
+      "step": 2100
+    },
+    {
+      "epoch": 3.36,
+      "eval_math_entropy": 0.938358835697174,
+      "eval_math_loss": 1.1868449449539185,
+      "eval_math_mean_token_accuracy": 0.733868212223053,
+      "eval_math_num_tokens": 69510512.0,
+      "eval_math_runtime": 63.3033,
+      "eval_math_samples_per_second": 7.898,
+      "eval_math_steps_per_second": 1.975,
+      "step": 2100
+    },
+    {
+      "epoch": 3.36,
+      "eval_physics_entropy": 0.8448640689849853,
+      "eval_physics_loss": 0.8845359086990356,
+      "eval_physics_mean_token_accuracy": 0.7768239378929138,
+      "eval_physics_num_tokens": 69510512.0,
+      "eval_physics_runtime": 72.1284,
+      "eval_physics_samples_per_second": 6.932,
+      "eval_physics_steps_per_second": 1.733,
+      "step": 2100
+    },
+    {
+      "entropy": 0.7963997656479478,
+      "epoch": 3.376,
+      "grad_norm": 17.625,
+      "learning_rate": 6.600000000000001e-06,
+      "loss": 12.7732,
+      "mean_token_accuracy": 0.7945189572870731,
+      "num_tokens": 69853992.0,
+      "step": 2110
+    },
+    {
+      "entropy": 0.8507452296093106,
+      "epoch": 3.392,
+      "grad_norm": 19.75,
+      "learning_rate": 6.525925925925927e-06,
+      "loss": 13.6131,
+      "mean_token_accuracy": 0.7810999616980553,
+      "num_tokens": 70178388.0,
+      "step": 2120
+    },
+    {
+      "entropy": 0.8109092302620411,
+      "epoch": 3.408,
+      "grad_norm": 19.625,
+      "learning_rate": 6.4518518518518525e-06,
+      "loss": 12.9974,
+      "mean_token_accuracy": 0.7911774169653654,
+      "num_tokens": 70507903.0,
+      "step": 2130
+    },
+    {
+      "entropy": 0.8097791658714414,
+      "epoch": 3.424,
+      "grad_norm": 20.125,
+      "learning_rate": 6.377777777777778e-06,
+      "loss": 12.9972,
+      "mean_token_accuracy": 0.7909738626331091,
+      "num_tokens": 70835982.0,
+      "step": 2140
+    },
+    {
+      "entropy": 0.8342178747057915,
+      "epoch": 3.44,
+      "grad_norm": 19.25,
+      "learning_rate": 6.303703703703704e-06,
+      "loss": 13.5081,
+      "mean_token_accuracy": 0.7840817723423242,
+      "num_tokens": 71172338.0,
+      "step": 2150
+    },
+    {
+      "entropy": 0.835791602730751,
+      "epoch": 3.456,
+      "grad_norm": 20.5,
+      "learning_rate": 6.2296296296296295e-06,
+      "loss": 13.3876,
+      "mean_token_accuracy": 0.7847528338432312,
+      "num_tokens": 71506904.0,
+      "step": 2160
+    },
+    {
+      "entropy": 0.8411149365827442,
+      "epoch": 3.472,
+      "grad_norm": 20.25,
+      "learning_rate": 6.155555555555556e-06,
+      "loss": 13.5361,
+      "mean_token_accuracy": 0.7834794349968434,
+      "num_tokens": 71830111.0,
+      "step": 2170
+    },
+    {
+      "entropy": 0.8058004789054394,
+      "epoch": 3.488,
+      "grad_norm": 19.125,
+      "learning_rate": 6.081481481481482e-06,
+      "loss": 12.8852,
+      "mean_token_accuracy": 0.7940779197961092,
+      "num_tokens": 72162307.0,
+      "step": 2180
+    },
+    {
+      "entropy": 0.8313451528549194,
+      "epoch": 3.504,
+      "grad_norm": 20.5,
+      "learning_rate": 6.007407407407407e-06,
+      "loss": 13.4177,
+      "mean_token_accuracy": 0.7863111417740584,
+      "num_tokens": 72492324.0,
+      "step": 2190
+    },
+    {
+      "entropy": 0.8189271073788404,
+      "epoch": 3.52,
+      "grad_norm": 20.0,
+      "learning_rate": 5.933333333333335e-06,
+      "loss": 13.2294,
+      "mean_token_accuracy": 0.7890881035476923,
+      "num_tokens": 72814868.0,
+      "step": 2200
+    },
+    {
+      "epoch": 3.52,
+      "eval_biology_entropy": 2.6062594680786133,
+      "eval_biology_loss": 3.1397392749786377,
+      "eval_biology_mean_token_accuracy": 0.4786659035682678,
+      "eval_biology_num_tokens": 72814868.0,
+      "eval_biology_runtime": 49.8201,
+      "eval_biology_samples_per_second": 10.036,
+      "eval_biology_steps_per_second": 2.509,
+      "step": 2200
+    },
+    {
+      "epoch": 3.52,
+      "eval_chemistry_entropy": 1.2033264746665955,
+      "eval_chemistry_loss": 1.3279136419296265,
+      "eval_chemistry_mean_token_accuracy": 0.7024473180770874,
+      "eval_chemistry_num_tokens": 72814868.0,
+      "eval_chemistry_runtime": 61.7534,
+      "eval_chemistry_samples_per_second": 8.097,
+      "eval_chemistry_steps_per_second": 2.024,
+      "step": 2200
+    },
+    {
+      "epoch": 3.52,
+      "eval_math_entropy": 0.936860782623291,
+      "eval_math_loss": 1.1871042251586914,
+      "eval_math_mean_token_accuracy": 0.7336988892555237,
+      "eval_math_num_tokens": 72814868.0,
+      "eval_math_runtime": 63.3178,
+      "eval_math_samples_per_second": 7.897,
+      "eval_math_steps_per_second": 1.974,
+      "step": 2200
+    },
+    {
+      "epoch": 3.52,
+      "eval_physics_entropy": 0.8426528668403626,
+      "eval_physics_loss": 0.8837320804595947,
+      "eval_physics_mean_token_accuracy": 0.7770564141273498,
+      "eval_physics_num_tokens": 72814868.0,
+      "eval_physics_runtime": 72.1699,
+      "eval_physics_samples_per_second": 6.928,
+      "eval_physics_steps_per_second": 1.732,
+      "step": 2200
+    },
+    {
+      "entropy": 0.8303699800744653,
+      "epoch": 3.536,
+      "grad_norm": 17.875,
+      "learning_rate": 5.85925925925926e-06,
+      "loss": 13.3117,
+      "mean_token_accuracy": 0.7869497302919626,
+      "num_tokens": 73144676.0,
+      "step": 2210
+    },
+    {
+      "entropy": 0.8237640831619502,
+      "epoch": 3.552,
+      "grad_norm": 17.0,
+      "learning_rate": 5.785185185185186e-06,
+      "loss": 13.134,
+      "mean_token_accuracy": 0.7901033364236355,
+      "num_tokens": 73474150.0,
+      "step": 2220
+    },
+    {
+      "entropy": 0.8104285070672631,
+      "epoch": 3.568,
+      "grad_norm": 20.0,
+      "learning_rate": 5.711111111111112e-06,
+      "loss": 13.0371,
+      "mean_token_accuracy": 0.7900369968265295,
+      "num_tokens": 73814337.0,
+      "step": 2230
+    },
+    {
+      "entropy": 0.847630956210196,
+      "epoch": 3.584,
+      "grad_norm": 19.75,
+      "learning_rate": 5.637037037037037e-06,
+      "loss": 13.8021,
+      "mean_token_accuracy": 0.7800593625754118,
+      "num_tokens": 74139916.0,
+      "step": 2240
+    },
+    {
+      "entropy": 0.8181004714220762,
+      "epoch": 3.6,
+      "grad_norm": 20.625,
+      "learning_rate": 5.562962962962963e-06,
+      "loss": 12.9809,
+      "mean_token_accuracy": 0.7909910753369331,
+      "num_tokens": 74481410.0,
+      "step": 2250
+    },
+    {
+      "entropy": 0.8192324381321668,
+      "epoch": 3.616,
+      "grad_norm": 18.375,
+      "learning_rate": 5.4888888888888895e-06,
+      "loss": 13.1107,
+      "mean_token_accuracy": 0.7886204149574041,
+      "num_tokens": 74818399.0,
+      "step": 2260
+    },
+    {
+      "entropy": 0.818259296566248,
+      "epoch": 3.632,
+      "grad_norm": 19.0,
+      "learning_rate": 5.414814814814815e-06,
+      "loss": 13.1013,
+      "mean_token_accuracy": 0.7898475378751755,
+      "num_tokens": 75148445.0,
+      "step": 2270
+    },
+    {
+      "entropy": 0.8245074760168791,
+      "epoch": 3.648,
+      "grad_norm": 19.0,
+      "learning_rate": 5.340740740740741e-06,
+      "loss": 13.3386,
+      "mean_token_accuracy": 0.7864024586975574,
+      "num_tokens": 75475448.0,
+      "step": 2280
+    },
+    {
+      "entropy": 0.8122432397678494,
+      "epoch": 3.664,
+      "grad_norm": 18.25,
+      "learning_rate": 5.2666666666666665e-06,
+      "loss": 12.9921,
+      "mean_token_accuracy": 0.7909019440412521,
+      "num_tokens": 75811516.0,
+      "step": 2290
+    },
+    {
+      "entropy": 0.8354866733774543,
+      "epoch": 3.68,
+      "grad_norm": 18.625,
+      "learning_rate": 5.192592592592594e-06,
+      "loss": 13.5011,
+      "mean_token_accuracy": 0.7843391906470061,
+      "num_tokens": 76131750.0,
+      "step": 2300
+    },
+    {
+      "epoch": 3.68,
+      "eval_biology_entropy": 2.605925470352173,
+      "eval_biology_loss": 3.1485085487365723,
+      "eval_biology_mean_token_accuracy": 0.4793458366394043,
+      "eval_biology_num_tokens": 76131750.0,
+      "eval_biology_runtime": 49.8966,
+      "eval_biology_samples_per_second": 10.021,
+      "eval_biology_steps_per_second": 2.505,
+      "step": 2300
+    },
+    {
+      "epoch": 3.68,
+      "eval_chemistry_entropy": 1.2000784149169923,
+      "eval_chemistry_loss": 1.3273272514343262,
+      "eval_chemistry_mean_token_accuracy": 0.7031446852684021,
+      "eval_chemistry_num_tokens": 76131750.0,
+      "eval_chemistry_runtime": 61.8443,
+      "eval_chemistry_samples_per_second": 8.085,
+      "eval_chemistry_steps_per_second": 2.021,
+      "step": 2300
+    },
+    {
+      "epoch": 3.68,
+      "eval_math_entropy": 0.9317488698959351,
+      "eval_math_loss": 1.1878645420074463,
+      "eval_math_mean_token_accuracy": 0.7335646581649781,
+      "eval_math_num_tokens": 76131750.0,
+      "eval_math_runtime": 63.4072,
+      "eval_math_samples_per_second": 7.886,
+      "eval_math_steps_per_second": 1.971,
+      "step": 2300
+    },
+    {
+      "epoch": 3.68,
+      "eval_physics_entropy": 0.840427505493164,
+      "eval_physics_loss": 0.8833025097846985,
+      "eval_physics_mean_token_accuracy": 0.7773792266845703,
+      "eval_physics_num_tokens": 76131750.0,
+      "eval_physics_runtime": 72.28,
+      "eval_physics_samples_per_second": 6.918,
+      "eval_physics_steps_per_second": 1.729,
+      "step": 2300
+    },
+    {
+      "entropy": 0.8487831071019173,
+      "epoch": 3.6959999999999997,
+      "grad_norm": 19.5,
+      "learning_rate": 5.1185185185185195e-06,
+      "loss": 13.6905,
+      "mean_token_accuracy": 0.782705707848072,
+      "num_tokens": 76460225.0,
+      "step": 2310
+    },
+    {
+      "entropy": 0.816498682461679,
+      "epoch": 3.7119999999999997,
+      "grad_norm": 21.5,
+      "learning_rate": 5.044444444444445e-06,
+      "loss": 13.0929,
+      "mean_token_accuracy": 0.7900269690901041,
+      "num_tokens": 76793614.0,
+      "step": 2320
+    },
+    {
+      "entropy": 0.8413596417754888,
+      "epoch": 3.7279999999999998,
+      "grad_norm": 18.625,
+      "learning_rate": 4.970370370370371e-06,
+      "loss": 13.5577,
+      "mean_token_accuracy": 0.7830862667411566,
+      "num_tokens": 77121009.0,
+      "step": 2330
+    },
+    {
+      "entropy": 0.8246031574904918,
+      "epoch": 3.7439999999999998,
+      "grad_norm": 20.25,
+      "learning_rate": 4.8962962962962965e-06,
+      "loss": 13.2587,
+      "mean_token_accuracy": 0.7871674124151469,
+      "num_tokens": 77446311.0,
+      "step": 2340
+    },
+    {
+      "entropy": 0.8347618822008371,
+      "epoch": 3.76,
+      "grad_norm": 19.0,
+      "learning_rate": 4.822222222222222e-06,
+      "loss": 13.3709,
+      "mean_token_accuracy": 0.7857745707035064,
+      "num_tokens": 77781488.0,
+      "step": 2350
+    },
+    {
+      "entropy": 0.8304801799356938,
+      "epoch": 3.776,
+      "grad_norm": 20.375,
+      "learning_rate": 4.748148148148149e-06,
+      "loss": 13.3514,
+      "mean_token_accuracy": 0.7865095619112253,
+      "num_tokens": 78108138.0,
+      "step": 2360
+    },
+    {
+      "entropy": 0.808137602545321,
+      "epoch": 3.792,
+      "grad_norm": 18.75,
+      "learning_rate": 4.674074074074074e-06,
+      "loss": 12.9843,
+      "mean_token_accuracy": 0.7931205939501524,
+      "num_tokens": 78449177.0,
+      "step": 2370
+    },
+    {
+      "entropy": 0.8280873820185661,
+      "epoch": 3.808,
+      "grad_norm": 18.5,
+      "learning_rate": 4.600000000000001e-06,
+      "loss": 13.3025,
+      "mean_token_accuracy": 0.7860911298543215,
+      "num_tokens": 78778761.0,
+      "step": 2380
+    },
+    {
+      "entropy": 0.8410927448421717,
+      "epoch": 3.824,
+      "grad_norm": 19.25,
+      "learning_rate": 4.5259259259259265e-06,
+      "loss": 13.5029,
+      "mean_token_accuracy": 0.784429706633091,
+      "num_tokens": 79102912.0,
+      "step": 2390
+    },
+    {
+      "entropy": 0.821617573313415,
+      "epoch": 3.84,
+      "grad_norm": 20.125,
+      "learning_rate": 4.451851851851852e-06,
+      "loss": 13.2334,
+      "mean_token_accuracy": 0.7883923355489969,
+      "num_tokens": 79436236.0,
+      "step": 2400
+    },
+    {
+      "epoch": 3.84,
+      "eval_biology_entropy": 2.608323128700256,
+      "eval_biology_loss": 3.1368472576141357,
+      "eval_biology_mean_token_accuracy": 0.47951707887649536,
+      "eval_biology_num_tokens": 79436236.0,
+      "eval_biology_runtime": 49.8217,
+      "eval_biology_samples_per_second": 10.036,
+      "eval_biology_steps_per_second": 2.509,
+      "step": 2400
+    },
+    {
+      "epoch": 3.84,
+      "eval_chemistry_entropy": 1.2032200002670288,
+      "eval_chemistry_loss": 1.3268632888793945,
+      "eval_chemistry_mean_token_accuracy": 0.702893283367157,
+      "eval_chemistry_num_tokens": 79436236.0,
+      "eval_chemistry_runtime": 62.1592,
+      "eval_chemistry_samples_per_second": 8.044,
+      "eval_chemistry_steps_per_second": 2.011,
+      "step": 2400
+    },
+    {
+      "epoch": 3.84,
+      "eval_math_entropy": 0.9371942892074585,
+      "eval_math_loss": 1.1872590780258179,
+      "eval_math_mean_token_accuracy": 0.7340162506103516,
+      "eval_math_num_tokens": 79436236.0,
+      "eval_math_runtime": 63.6063,
+      "eval_math_samples_per_second": 7.861,
+      "eval_math_steps_per_second": 1.965,
+      "step": 2400
+    },
+    {
+      "epoch": 3.84,
+      "eval_physics_entropy": 0.8427950644493103,
+      "eval_physics_loss": 0.8829500675201416,
+      "eval_physics_mean_token_accuracy": 0.7773289399147034,
+      "eval_physics_num_tokens": 79436236.0,
+      "eval_physics_runtime": 72.2234,
+      "eval_physics_samples_per_second": 6.923,
+      "eval_physics_steps_per_second": 1.731,
+      "step": 2400
+    },
+    {
+      "entropy": 0.8030849374830723,
+      "epoch": 3.856,
+      "grad_norm": 18.375,
+      "learning_rate": 4.377777777777778e-06,
+      "loss": 12.8811,
+      "mean_token_accuracy": 0.7927878838032484,
+      "num_tokens": 79772163.0,
+      "step": 2410
+    },
+    {
+      "entropy": 0.8331086363643407,
+      "epoch": 3.872,
+      "grad_norm": 20.75,
+      "learning_rate": 4.3037037037037035e-06,
+      "loss": 13.2838,
+      "mean_token_accuracy": 0.7862316433340311,
+      "num_tokens": 80099967.0,
+      "step": 2420
+    },
+    {
+      "entropy": 0.8246728168800473,
+      "epoch": 3.888,
+      "grad_norm": 20.25,
+      "learning_rate": 4.22962962962963e-06,
+      "loss": 13.2966,
+      "mean_token_accuracy": 0.7876805573701858,
+      "num_tokens": 80430055.0,
+      "step": 2430
+    },
+    {
+      "entropy": 0.859740087762475,
+      "epoch": 3.904,
+      "grad_norm": 22.0,
+      "learning_rate": 4.155555555555556e-06,
+      "loss": 13.8673,
+      "mean_token_accuracy": 0.7792655471712351,
+      "num_tokens": 80763923.0,
+      "step": 2440
+    },
+    {
+      "entropy": 0.8212427636608481,
+      "epoch": 3.92,
+      "grad_norm": 20.75,
+      "learning_rate": 4.081481481481482e-06,
+      "loss": 13.1513,
+      "mean_token_accuracy": 0.7892008528113366,
+      "num_tokens": 81098568.0,
+      "step": 2450
+    },
+    {
+      "entropy": 0.832253485918045,
+      "epoch": 3.936,
+      "grad_norm": 19.0,
+      "learning_rate": 4.007407407407408e-06,
+      "loss": 13.3578,
+      "mean_token_accuracy": 0.7863949816673994,
+      "num_tokens": 81423109.0,
+      "step": 2460
+    },
+    {
+      "entropy": 0.8406696753576398,
+      "epoch": 3.952,
+      "grad_norm": 21.875,
+      "learning_rate": 3.9333333333333335e-06,
+      "loss": 13.5337,
+      "mean_token_accuracy": 0.7836084965616464,
+      "num_tokens": 81753214.0,
+      "step": 2470
+    },
+    {
+      "entropy": 0.8148748876526952,
+      "epoch": 3.968,
+      "grad_norm": 22.125,
+      "learning_rate": 3.85925925925926e-06,
+      "loss": 13.0495,
+      "mean_token_accuracy": 0.7895875804126262,
+      "num_tokens": 82076472.0,
+      "step": 2480
+    },
+    {
+      "entropy": 0.8054082557559014,
+      "epoch": 3.984,
+      "grad_norm": 18.5,
+      "learning_rate": 3.7851851851851857e-06,
+      "loss": 12.8509,
+      "mean_token_accuracy": 0.7923774540424346,
+      "num_tokens": 82418472.0,
+      "step": 2490
+    },
+    {
+      "entropy": 0.8139926699921489,
+      "epoch": 4.0,
+      "grad_norm": 21.125,
+      "learning_rate": 3.7111111111111113e-06,
+      "loss": 13.094,
+      "mean_token_accuracy": 0.789514297619462,
+      "num_tokens": 82747216.0,
+      "step": 2500
+    },
+    {
+      "epoch": 4.0,
+      "eval_biology_entropy": 2.596646837234497,
+      "eval_biology_loss": 3.1402840614318848,
+      "eval_biology_mean_token_accuracy": 0.47984995126724245,
+      "eval_biology_num_tokens": 82747216.0,
+      "eval_biology_runtime": 49.7911,
+      "eval_biology_samples_per_second": 10.042,
+      "eval_biology_steps_per_second": 2.51,
+      "step": 2500
+    },
+    {
+      "epoch": 4.0,
+      "eval_chemistry_entropy": 1.19823095703125,
+      "eval_chemistry_loss": 1.3272966146469116,
+      "eval_chemistry_mean_token_accuracy": 0.702995774269104,
+      "eval_chemistry_num_tokens": 82747216.0,
+      "eval_chemistry_runtime": 61.3566,
+      "eval_chemistry_samples_per_second": 8.149,
+      "eval_chemistry_steps_per_second": 2.037,
+      "step": 2500
+    },
+    {
+      "epoch": 4.0,
+      "eval_math_entropy": 0.9340950083732605,
+      "eval_math_loss": 1.187938928604126,
+      "eval_math_mean_token_accuracy": 0.733943380355835,
+      "eval_math_num_tokens": 82747216.0,
+      "eval_math_runtime": 63.2435,
+      "eval_math_samples_per_second": 7.906,
+      "eval_math_steps_per_second": 1.976,
+      "step": 2500
+    },
+    {
+      "epoch": 4.0,
+      "eval_physics_entropy": 0.8391828942298889,
+      "eval_physics_loss": 0.8828689455986023,
+      "eval_physics_mean_token_accuracy": 0.7774432940483094,
+      "eval_physics_num_tokens": 82747216.0,
+      "eval_physics_runtime": 72.1271,
+      "eval_physics_samples_per_second": 6.932,
+      "eval_physics_steps_per_second": 1.733,
+      "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
+}