diff --git "a/checkpoint-9500/trainer_state.json" "b/checkpoint-9500/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/checkpoint-9500/trainer_state.json"
@@ -0,0 +1,11624 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 15.2,
+  "eval_steps": 100,
+  "global_step": 9500,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 0.5915625745430588,
+      "epoch": 0.016,
+      "grad_norm": 5.71875,
+      "learning_rate": 1.8e-07,
+      "loss": 0.8466,
+      "mean_token_accuracy": 0.8017937403172255,
+      "num_tokens": 340696.0,
+      "step": 10
+    },
+    {
+      "entropy": 0.5926362803205848,
+      "epoch": 0.032,
+      "grad_norm": 5.59375,
+      "learning_rate": 3.8e-07,
+      "loss": 0.8487,
+      "mean_token_accuracy": 0.8023959375917912,
+      "num_tokens": 675930.0,
+      "step": 20
+    },
+    {
+      "entropy": 0.591606929153204,
+      "epoch": 0.048,
+      "grad_norm": 4.96875,
+      "learning_rate": 5.800000000000001e-07,
+      "loss": 0.8302,
+      "mean_token_accuracy": 0.8044086921960115,
+      "num_tokens": 1007956.0,
+      "step": 30
+    },
+    {
+      "entropy": 0.6204556789249182,
+      "epoch": 0.064,
+      "grad_norm": 4.9375,
+      "learning_rate": 7.8e-07,
+      "loss": 0.8457,
+      "mean_token_accuracy": 0.7998181059956551,
+      "num_tokens": 1340560.0,
+      "step": 40
+    },
+    {
+      "entropy": 0.6393004363402724,
+      "epoch": 0.08,
+      "grad_norm": 4.3125,
+      "learning_rate": 9.800000000000001e-07,
+      "loss": 0.8343,
+      "mean_token_accuracy": 0.8025688324123621,
+      "num_tokens": 1679432.0,
+      "step": 50
+    },
+    {
+      "entropy": 0.6593567343428731,
+      "epoch": 0.096,
+      "grad_norm": 2.5625,
+      "learning_rate": 1.1800000000000001e-06,
+      "loss": 0.8046,
+      "mean_token_accuracy": 0.8078926544636488,
+      "num_tokens": 2021570.0,
+      "step": 60
+    },
+    {
+      "entropy": 0.7051705561578274,
+      "epoch": 0.112,
+      "grad_norm": 2.8125,
+      "learning_rate": 1.3800000000000001e-06,
+      "loss": 0.8246,
+      "mean_token_accuracy": 0.8029463469982148,
+      "num_tokens": 2359297.0,
+      "step": 70
+    },
+    {
+      "entropy": 0.754444376192987,
+      "epoch": 0.128,
+      "grad_norm": 2.4375,
+      "learning_rate": 1.5800000000000001e-06,
+      "loss": 0.8435,
+      "mean_token_accuracy": 0.79781529083848,
+      "num_tokens": 2683168.0,
+      "step": 80
+    },
+    {
+      "entropy": 0.7250820267945528,
+      "epoch": 0.144,
+      "grad_norm": 1.90625,
+      "learning_rate": 1.7800000000000001e-06,
+      "loss": 0.8195,
+      "mean_token_accuracy": 0.8042717296630144,
+      "num_tokens": 3019426.0,
+      "step": 90
+    },
+    {
+      "entropy": 0.7252413807436824,
+      "epoch": 0.16,
+      "grad_norm": 2.109375,
+      "learning_rate": 1.98e-06,
+      "loss": 0.8097,
+      "mean_token_accuracy": 0.8047077178955078,
+      "num_tokens": 3354390.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.16,
+      "eval_chemistry_entropy": 0.7109885907173157,
+      "eval_chemistry_loss": 0.804060161113739,
+      "eval_chemistry_mean_token_accuracy": 0.8113441648483276,
+      "eval_chemistry_num_tokens": 3354390.0,
+      "eval_chemistry_runtime": 34.8657,
+      "eval_chemistry_samples_per_second": 14.341,
+      "eval_chemistry_steps_per_second": 3.585,
+      "step": 100
+    },
+    {
+      "epoch": 0.16,
+      "eval_physics_entropy": 0.7321163377761841,
+      "eval_physics_loss": 0.8287621140480042,
+      "eval_physics_mean_token_accuracy": 0.8033607420921326,
+      "eval_physics_num_tokens": 3354390.0,
+      "eval_physics_runtime": 41.3727,
+      "eval_physics_samples_per_second": 12.085,
+      "eval_physics_steps_per_second": 3.021,
+      "step": 100
+    },
+    {
+      "entropy": 0.7267085557803512,
+      "epoch": 0.176,
+      "grad_norm": 2.234375,
+      "learning_rate": 2.1800000000000003e-06,
+      "loss": 0.8058,
+      "mean_token_accuracy": 0.8033798363059759,
+      "num_tokens": 3684374.0,
+      "step": 110
+    },
+    {
+      "entropy": 0.7212623069062829,
+      "epoch": 0.192,
+      "grad_norm": 1.828125,
+      "learning_rate": 2.38e-06,
+      "loss": 0.7989,
+      "mean_token_accuracy": 0.8043727930635214,
+      "num_tokens": 4023915.0,
+      "step": 120
+    },
+    {
+      "entropy": 0.7022871913388371,
+      "epoch": 0.208,
+      "grad_norm": 2.1875,
+      "learning_rate": 2.5800000000000003e-06,
+      "loss": 0.7758,
+      "mean_token_accuracy": 0.8102051325142383,
+      "num_tokens": 4352045.0,
+      "step": 130
+    },
+    {
+      "entropy": 0.6965231081470847,
+      "epoch": 0.224,
+      "grad_norm": 1.8203125,
+      "learning_rate": 2.7800000000000005e-06,
+      "loss": 0.7726,
+      "mean_token_accuracy": 0.8089764468371868,
+      "num_tokens": 4681371.0,
+      "step": 140
+    },
+    {
+      "entropy": 0.6956626677885651,
+      "epoch": 0.24,
+      "grad_norm": 1.6640625,
+      "learning_rate": 2.9800000000000003e-06,
+      "loss": 0.7608,
+      "mean_token_accuracy": 0.8104452796280384,
+      "num_tokens": 5021784.0,
+      "step": 150
+    },
+    {
+      "entropy": 0.708347842283547,
+      "epoch": 0.256,
+      "grad_norm": 1.9453125,
+      "learning_rate": 3.1800000000000005e-06,
+      "loss": 0.7636,
+      "mean_token_accuracy": 0.809263763204217,
+      "num_tokens": 5367308.0,
+      "step": 160
+    },
+    {
+      "entropy": 0.7138571232557297,
+      "epoch": 0.272,
+      "grad_norm": 2.03125,
+      "learning_rate": 3.3800000000000007e-06,
+      "loss": 0.7721,
+      "mean_token_accuracy": 0.8050564888864755,
+      "num_tokens": 5699101.0,
+      "step": 170
+    },
+    {
+      "entropy": 0.7222439678385854,
+      "epoch": 0.288,
+      "grad_norm": 1.7109375,
+      "learning_rate": 3.58e-06,
+      "loss": 0.7766,
+      "mean_token_accuracy": 0.8063390862196684,
+      "num_tokens": 6022837.0,
+      "step": 180
+    },
+    {
+      "entropy": 0.7117253359407186,
+      "epoch": 0.304,
+      "grad_norm": 2.265625,
+      "learning_rate": 3.7800000000000002e-06,
+      "loss": 0.7535,
+      "mean_token_accuracy": 0.809041840955615,
+      "num_tokens": 6343826.0,
+      "step": 190
+    },
+    {
+      "entropy": 0.7200546801090241,
+      "epoch": 0.32,
+      "grad_norm": 1.7265625,
+      "learning_rate": 3.980000000000001e-06,
+      "loss": 0.7524,
+      "mean_token_accuracy": 0.8073045868426562,
+      "num_tokens": 6670164.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.32,
+      "eval_chemistry_entropy": 0.711975371837616,
+      "eval_chemistry_loss": 0.7488377690315247,
+      "eval_chemistry_mean_token_accuracy": 0.8110983457565307,
+      "eval_chemistry_num_tokens": 6670164.0,
+      "eval_chemistry_runtime": 34.8588,
+      "eval_chemistry_samples_per_second": 14.344,
+      "eval_chemistry_steps_per_second": 3.586,
+      "step": 200
+    },
+    {
+      "epoch": 0.32,
+      "eval_physics_entropy": 0.7178810501098633,
+      "eval_physics_loss": 0.7585415244102478,
+      "eval_physics_mean_token_accuracy": 0.8081966361999512,
+      "eval_physics_num_tokens": 6670164.0,
+      "eval_physics_runtime": 41.7087,
+      "eval_physics_samples_per_second": 11.988,
+      "eval_physics_steps_per_second": 2.997,
+      "step": 200
+    },
+    {
+      "entropy": 0.7111945543438196,
+      "epoch": 0.336,
+      "grad_norm": 1.578125,
+      "learning_rate": 4.18e-06,
+      "loss": 0.7313,
+      "mean_token_accuracy": 0.8092666126787662,
+      "num_tokens": 7003404.0,
+      "step": 210
+    },
+    {
+      "entropy": 0.6962820990011096,
+      "epoch": 0.352,
+      "grad_norm": 1.8828125,
+      "learning_rate": 4.38e-06,
+      "loss": 0.712,
+      "mean_token_accuracy": 0.8133333619683981,
+      "num_tokens": 7342593.0,
+      "step": 220
+    },
+    {
+      "entropy": 0.6933605512604117,
+      "epoch": 0.368,
+      "grad_norm": 1.6484375,
+      "learning_rate": 4.58e-06,
+      "loss": 0.6986,
+      "mean_token_accuracy": 0.8163042381405831,
+      "num_tokens": 7675569.0,
+      "step": 230
+    },
+    {
+      "entropy": 0.7312716225162148,
+      "epoch": 0.384,
+      "grad_norm": 1.9296875,
+      "learning_rate": 4.78e-06,
+      "loss": 0.7319,
+      "mean_token_accuracy": 0.8038436233997345,
+      "num_tokens": 8012218.0,
+      "step": 240
+    },
+    {
+      "entropy": 0.6996616214513779,
+      "epoch": 0.4,
+      "grad_norm": 1.796875,
+      "learning_rate": 4.980000000000001e-06,
+      "loss": 0.6917,
+      "mean_token_accuracy": 0.8134940605610609,
+      "num_tokens": 8350120.0,
+      "step": 250
+    },
+    {
+      "entropy": 0.6945684082806111,
+      "epoch": 0.416,
+      "grad_norm": 1.6953125,
+      "learning_rate": 5.18e-06,
+      "loss": 0.678,
+      "mean_token_accuracy": 0.8151624854654074,
+      "num_tokens": 8688773.0,
+      "step": 260
+    },
+    {
+      "entropy": 0.7035746525973081,
+      "epoch": 0.432,
+      "grad_norm": 1.8359375,
+      "learning_rate": 5.380000000000001e-06,
+      "loss": 0.6888,
+      "mean_token_accuracy": 0.8118865180760622,
+      "num_tokens": 9022094.0,
+      "step": 270
+    },
+    {
+      "entropy": 0.6850203935056924,
+      "epoch": 0.448,
+      "grad_norm": 1.6640625,
+      "learning_rate": 5.580000000000001e-06,
+      "loss": 0.6727,
+      "mean_token_accuracy": 0.8186012674123049,
+      "num_tokens": 9360608.0,
+      "step": 280
+    },
+    {
+      "entropy": 0.6806540634483099,
+      "epoch": 0.464,
+      "grad_norm": 1.875,
+      "learning_rate": 5.78e-06,
+      "loss": 0.6592,
+      "mean_token_accuracy": 0.8200464811176061,
+      "num_tokens": 9688896.0,
+      "step": 290
+    },
+    {
+      "entropy": 0.6852828785777092,
+      "epoch": 0.48,
+      "grad_norm": 1.828125,
+      "learning_rate": 5.98e-06,
+      "loss": 0.6672,
+      "mean_token_accuracy": 0.8180346965789795,
+      "num_tokens": 10025864.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.48,
+      "eval_chemistry_entropy": 0.7274510788917542,
+      "eval_chemistry_loss": 0.7009354829788208,
+      "eval_chemistry_mean_token_accuracy": 0.8133535690307617,
+      "eval_chemistry_num_tokens": 10025864.0,
+      "eval_chemistry_runtime": 34.811,
+      "eval_chemistry_samples_per_second": 14.363,
+      "eval_chemistry_steps_per_second": 3.591,
+      "step": 300
+    },
+    {
+      "epoch": 0.48,
+      "eval_physics_entropy": 0.7100022671222687,
+      "eval_physics_loss": 0.7013525366783142,
+      "eval_physics_mean_token_accuracy": 0.813646514415741,
+      "eval_physics_num_tokens": 10025864.0,
+      "eval_physics_runtime": 41.4407,
+      "eval_physics_samples_per_second": 12.065,
+      "eval_physics_steps_per_second": 3.016,
+      "step": 300
+    },
+    {
+      "entropy": 0.7103753052651882,
+      "epoch": 0.496,
+      "grad_norm": 1.6171875,
+      "learning_rate": 6.18e-06,
+      "loss": 0.6847,
+      "mean_token_accuracy": 0.812765546515584,
+      "num_tokens": 10362573.0,
+      "step": 310
+    },
+    {
+      "entropy": 0.6982777336612344,
+      "epoch": 0.512,
+      "grad_norm": 1.8125,
+      "learning_rate": 6.380000000000001e-06,
+      "loss": 0.6716,
+      "mean_token_accuracy": 0.8169939771294594,
+      "num_tokens": 10694440.0,
+      "step": 320
+    },
+    {
+      "entropy": 0.708159969188273,
+      "epoch": 0.528,
+      "grad_norm": 1.7421875,
+      "learning_rate": 6.5800000000000005e-06,
+      "loss": 0.693,
+      "mean_token_accuracy": 0.8125004164874554,
+      "num_tokens": 11022357.0,
+      "step": 330
+    },
+    {
+      "entropy": 0.699538960121572,
+      "epoch": 0.544,
+      "grad_norm": 1.828125,
+      "learning_rate": 6.780000000000001e-06,
+      "loss": 0.6711,
+      "mean_token_accuracy": 0.8153361644595861,
+      "num_tokens": 11345911.0,
+      "step": 340
+    },
+    {
+      "entropy": 0.6910958087071777,
+      "epoch": 0.56,
+      "grad_norm": 1.5078125,
+      "learning_rate": 6.98e-06,
+      "loss": 0.6659,
+      "mean_token_accuracy": 0.8177970543503761,
+      "num_tokens": 11687225.0,
+      "step": 350
+    },
+    {
+      "entropy": 0.6805725138634443,
+      "epoch": 0.576,
+      "grad_norm": 1.6484375,
+      "learning_rate": 7.180000000000001e-06,
+      "loss": 0.6565,
+      "mean_token_accuracy": 0.8207161612808704,
+      "num_tokens": 12015720.0,
+      "step": 360
+    },
+    {
+      "entropy": 0.6865102933719754,
+      "epoch": 0.592,
+      "grad_norm": 1.84375,
+      "learning_rate": 7.3800000000000005e-06,
+      "loss": 0.6638,
+      "mean_token_accuracy": 0.8175321508198976,
+      "num_tokens": 12343393.0,
+      "step": 370
+    },
+    {
+      "entropy": 0.6841516578570008,
+      "epoch": 0.608,
+      "grad_norm": 1.6328125,
+      "learning_rate": 7.58e-06,
+      "loss": 0.6582,
+      "mean_token_accuracy": 0.8192304316908121,
+      "num_tokens": 12671474.0,
+      "step": 380
+    },
+    {
+      "entropy": 0.7005494212731719,
+      "epoch": 0.624,
+      "grad_norm": 1.6015625,
+      "learning_rate": 7.78e-06,
+      "loss": 0.6674,
+      "mean_token_accuracy": 0.8156571265310049,
+      "num_tokens": 12996077.0,
+      "step": 390
+    },
+    {
+      "entropy": 0.6937378702685237,
+      "epoch": 0.64,
+      "grad_norm": 1.6171875,
+      "learning_rate": 7.980000000000002e-06,
+      "loss": 0.6701,
+      "mean_token_accuracy": 0.8157013915479183,
+      "num_tokens": 13325291.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.64,
+      "eval_chemistry_entropy": 0.7147737612724304,
+      "eval_chemistry_loss": 0.692807137966156,
+      "eval_chemistry_mean_token_accuracy": 0.8137030324935913,
+      "eval_chemistry_num_tokens": 13325291.0,
+      "eval_chemistry_runtime": 34.8141,
+      "eval_chemistry_samples_per_second": 14.362,
+      "eval_chemistry_steps_per_second": 3.591,
+      "step": 400
+    },
+    {
+      "epoch": 0.64,
+      "eval_physics_entropy": 0.684398841381073,
+      "eval_physics_loss": 0.6776748895645142,
+      "eval_physics_mean_token_accuracy": 0.8169920797348023,
+      "eval_physics_num_tokens": 13325291.0,
+      "eval_physics_runtime": 41.4558,
+      "eval_physics_samples_per_second": 12.061,
+      "eval_physics_steps_per_second": 3.015,
+      "step": 400
+    },
+    {
+      "entropy": 0.6792764134705067,
+      "epoch": 0.656,
+      "grad_norm": 1.625,
+      "learning_rate": 8.18e-06,
+      "loss": 0.6538,
+      "mean_token_accuracy": 0.8188512332737445,
+      "num_tokens": 13671434.0,
+      "step": 410
+    },
+    {
+      "entropy": 0.6741176716983318,
+      "epoch": 0.672,
+      "grad_norm": 1.703125,
+      "learning_rate": 8.380000000000001e-06,
+      "loss": 0.6479,
+      "mean_token_accuracy": 0.8206427529454231,
+      "num_tokens": 13993668.0,
+      "step": 420
+    },
+    {
+      "entropy": 0.6748962674289942,
+      "epoch": 0.688,
+      "grad_norm": 1.5625,
+      "learning_rate": 8.580000000000001e-06,
+      "loss": 0.6489,
+      "mean_token_accuracy": 0.821071095019579,
+      "num_tokens": 14334907.0,
+      "step": 430
+    },
+    {
+      "entropy": 0.6882562624290586,
+      "epoch": 0.704,
+      "grad_norm": 1.765625,
+      "learning_rate": 8.78e-06,
+      "loss": 0.6631,
+      "mean_token_accuracy": 0.8166982632130384,
+      "num_tokens": 14670639.0,
+      "step": 440
+    },
+    {
+      "entropy": 0.6710160901769996,
+      "epoch": 0.72,
+      "grad_norm": 1.828125,
+      "learning_rate": 8.98e-06,
+      "loss": 0.6451,
+      "mean_token_accuracy": 0.8207047082483768,
+      "num_tokens": 14987995.0,
+      "step": 450
+    },
+    {
+      "entropy": 0.6616904048249126,
+      "epoch": 0.736,
+      "grad_norm": 1.65625,
+      "learning_rate": 9.180000000000002e-06,
+      "loss": 0.6374,
+      "mean_token_accuracy": 0.8214268710464239,
+      "num_tokens": 15324659.0,
+      "step": 460
+    },
+    {
+      "entropy": 0.6631792033091187,
+      "epoch": 0.752,
+      "grad_norm": 1.6171875,
+      "learning_rate": 9.38e-06,
+      "loss": 0.6379,
+      "mean_token_accuracy": 0.8226557578891516,
+      "num_tokens": 15657342.0,
+      "step": 470
+    },
+    {
+      "entropy": 0.6816092301160097,
+      "epoch": 0.768,
+      "grad_norm": 1.640625,
+      "learning_rate": 9.58e-06,
+      "loss": 0.6582,
+      "mean_token_accuracy": 0.8162681456655264,
+      "num_tokens": 15988993.0,
+      "step": 480
+    },
+    {
+      "entropy": 0.6584033869206906,
+      "epoch": 0.784,
+      "grad_norm": 1.6484375,
+      "learning_rate": 9.780000000000001e-06,
+      "loss": 0.6307,
+      "mean_token_accuracy": 0.8244805630296469,
+      "num_tokens": 16333857.0,
+      "step": 490
+    },
+    {
+      "entropy": 0.6675900464877487,
+      "epoch": 0.8,
+      "grad_norm": 1.625,
+      "learning_rate": 9.980000000000001e-06,
+      "loss": 0.6444,
+      "mean_token_accuracy": 0.8208535581827163,
+      "num_tokens": 16664690.0,
+      "step": 500
+    },
+    {
+      "epoch": 0.8,
+      "eval_chemistry_entropy": 0.7186227707862854,
+      "eval_chemistry_loss": 0.6891804337501526,
+      "eval_chemistry_mean_token_accuracy": 0.812873167514801,
+      "eval_chemistry_num_tokens": 16664690.0,
+      "eval_chemistry_runtime": 34.7308,
+      "eval_chemistry_samples_per_second": 14.396,
+      "eval_chemistry_steps_per_second": 3.599,
+      "step": 500
+    },
+    {
+      "epoch": 0.8,
+      "eval_physics_entropy": 0.6725036988258362,
+      "eval_physics_loss": 0.6597840785980225,
+      "eval_physics_mean_token_accuracy": 0.8193980932235718,
+      "eval_physics_num_tokens": 16664690.0,
+      "eval_physics_runtime": 41.0989,
+      "eval_physics_samples_per_second": 12.166,
+      "eval_physics_steps_per_second": 3.041,
+      "step": 500
+    },
+    {
+      "entropy": 0.6641527567058801,
+      "epoch": 0.816,
+      "grad_norm": 1.5625,
+      "learning_rate": 1.018e-05,
+      "loss": 0.6389,
+      "mean_token_accuracy": 0.8205956391990185,
+      "num_tokens": 16989307.0,
+      "step": 510
+    },
+    {
+      "entropy": 0.6729596307501197,
+      "epoch": 0.832,
+      "grad_norm": 1.5859375,
+      "learning_rate": 1.038e-05,
+      "loss": 0.6418,
+      "mean_token_accuracy": 0.8202375840395689,
+      "num_tokens": 17317524.0,
+      "step": 520
+    },
+    {
+      "entropy": 0.6599426086992025,
+      "epoch": 0.848,
+      "grad_norm": 1.40625,
+      "learning_rate": 1.0580000000000002e-05,
+      "loss": 0.6326,
+      "mean_token_accuracy": 0.8230770550668239,
+      "num_tokens": 17650424.0,
+      "step": 530
+    },
+    {
+      "entropy": 0.66338025983423,
+      "epoch": 0.864,
+      "grad_norm": 1.4765625,
+      "learning_rate": 1.0780000000000002e-05,
+      "loss": 0.6412,
+      "mean_token_accuracy": 0.8213306616991758,
+      "num_tokens": 17985547.0,
+      "step": 540
+    },
+    {
+      "entropy": 0.6687672117725014,
+      "epoch": 0.88,
+      "grad_norm": 1.4921875,
+      "learning_rate": 1.0980000000000002e-05,
+      "loss": 0.6426,
+      "mean_token_accuracy": 0.8197343621402979,
+      "num_tokens": 18315392.0,
+      "step": 550
+    },
+    {
+      "entropy": 0.6246686290949583,
+      "epoch": 0.896,
+      "grad_norm": 1.375,
+      "learning_rate": 1.1180000000000001e-05,
+      "loss": 0.6014,
+      "mean_token_accuracy": 0.8294213119894266,
+      "num_tokens": 18667124.0,
+      "step": 560
+    },
+    {
+      "entropy": 0.6640157140791416,
+      "epoch": 0.912,
+      "grad_norm": 1.5,
+      "learning_rate": 1.138e-05,
+      "loss": 0.6371,
+      "mean_token_accuracy": 0.820187458768487,
+      "num_tokens": 19002186.0,
+      "step": 570
+    },
+    {
+      "entropy": 0.6525983370840549,
+      "epoch": 0.928,
+      "grad_norm": 1.6640625,
+      "learning_rate": 1.1580000000000001e-05,
+      "loss": 0.6304,
+      "mean_token_accuracy": 0.8224839419126511,
+      "num_tokens": 19335013.0,
+      "step": 580
+    },
+    {
+      "entropy": 0.6459019472822547,
+      "epoch": 0.944,
+      "grad_norm": 1.6171875,
+      "learning_rate": 1.178e-05,
+      "loss": 0.6208,
+      "mean_token_accuracy": 0.8251514233648777,
+      "num_tokens": 19662733.0,
+      "step": 590
+    },
+    {
+      "entropy": 0.6452493501827121,
+      "epoch": 0.96,
+      "grad_norm": 1.5625,
+      "learning_rate": 1.198e-05,
+      "loss": 0.6276,
+      "mean_token_accuracy": 0.8237052172422409,
+      "num_tokens": 19996789.0,
+      "step": 600
+    },
+    {
+      "epoch": 0.96,
+      "eval_chemistry_entropy": 0.7299397034645081,
+      "eval_chemistry_loss": 0.6870628595352173,
+      "eval_chemistry_mean_token_accuracy": 0.8117328572273255,
+      "eval_chemistry_num_tokens": 19996789.0,
+      "eval_chemistry_runtime": 34.7851,
+      "eval_chemistry_samples_per_second": 14.374,
+      "eval_chemistry_steps_per_second": 3.593,
+      "step": 600
+    },
+    {
+      "epoch": 0.96,
+      "eval_physics_entropy": 0.6775157516002654,
+      "eval_physics_loss": 0.646020770072937,
+      "eval_physics_mean_token_accuracy": 0.8218063368797303,
+      "eval_physics_num_tokens": 19996789.0,
+      "eval_physics_runtime": 41.4104,
+      "eval_physics_samples_per_second": 12.074,
+      "eval_physics_steps_per_second": 3.019,
+      "step": 600
+    },
+    {
+      "entropy": 0.6451076414436102,
+      "epoch": 0.976,
+      "grad_norm": 1.5859375,
+      "learning_rate": 1.218e-05,
+      "loss": 0.6221,
+      "mean_token_accuracy": 0.825892923027277,
+      "num_tokens": 20332636.0,
+      "step": 610
+    },
+    {
+      "entropy": 0.6509507959708571,
+      "epoch": 0.992,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.2380000000000002e-05,
+      "loss": 0.6214,
+      "mean_token_accuracy": 0.8233620677143335,
+      "num_tokens": 20662209.0,
+      "step": 620
+    },
+    {
+      "entropy": 0.6204982601106167,
+      "epoch": 1.008,
+      "grad_norm": 1.46875,
+      "learning_rate": 1.2580000000000002e-05,
+      "loss": 0.5988,
+      "mean_token_accuracy": 0.8286421094089746,
+      "num_tokens": 21005731.0,
+      "step": 630
+    },
+    {
+      "entropy": 0.6242268454283476,
+      "epoch": 1.024,
+      "grad_norm": 1.4921875,
+      "learning_rate": 1.2780000000000001e-05,
+      "loss": 0.6046,
+      "mean_token_accuracy": 0.82848214879632,
+      "num_tokens": 21335407.0,
+      "step": 640
+    },
+    {
+      "entropy": 0.6163413025438785,
+      "epoch": 1.04,
+      "grad_norm": 1.5078125,
+      "learning_rate": 1.2980000000000001e-05,
+      "loss": 0.598,
+      "mean_token_accuracy": 0.8299466107040644,
+      "num_tokens": 21668490.0,
+      "step": 650
+    },
+    {
+      "entropy": 0.6118087394163012,
+      "epoch": 1.056,
+      "grad_norm": 1.4453125,
+      "learning_rate": 1.3180000000000001e-05,
+      "loss": 0.5883,
+      "mean_token_accuracy": 0.8311309419572354,
+      "num_tokens": 22006039.0,
+      "step": 660
+    },
+    {
+      "entropy": 0.6269845755770802,
+      "epoch": 1.072,
+      "grad_norm": 1.4453125,
+      "learning_rate": 1.3380000000000002e-05,
+      "loss": 0.6043,
+      "mean_token_accuracy": 0.8276471871882677,
+      "num_tokens": 22348206.0,
+      "step": 670
+    },
+    {
+      "entropy": 0.6239592241123318,
+      "epoch": 1.088,
+      "grad_norm": 1.5234375,
+      "learning_rate": 1.3580000000000002e-05,
+      "loss": 0.5947,
+      "mean_token_accuracy": 0.8277118816971779,
+      "num_tokens": 22676163.0,
+      "step": 680
+    },
+    {
+      "entropy": 0.6154211649671197,
+      "epoch": 1.104,
+      "grad_norm": 1.3046875,
+      "learning_rate": 1.378e-05,
+      "loss": 0.5951,
+      "mean_token_accuracy": 0.8299165233969689,
+      "num_tokens": 23009279.0,
+      "step": 690
+    },
+    {
+      "entropy": 0.6186880182474852,
+      "epoch": 1.12,
+      "grad_norm": 1.4765625,
+      "learning_rate": 1.398e-05,
+      "loss": 0.5987,
+      "mean_token_accuracy": 0.827697791159153,
+      "num_tokens": 23346828.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.12,
+      "eval_chemistry_entropy": 0.6806538386344909,
+      "eval_chemistry_loss": 0.6874776482582092,
+      "eval_chemistry_mean_token_accuracy": 0.8116622562408448,
+      "eval_chemistry_num_tokens": 23346828.0,
+      "eval_chemistry_runtime": 35.1659,
+      "eval_chemistry_samples_per_second": 14.218,
+      "eval_chemistry_steps_per_second": 3.555,
+      "step": 700
+    },
+    {
+      "epoch": 1.12,
+      "eval_physics_entropy": 0.6262999680042267,
+      "eval_physics_loss": 0.6376790404319763,
+      "eval_physics_mean_token_accuracy": 0.8227445411682129,
+      "eval_physics_num_tokens": 23346828.0,
+      "eval_physics_runtime": 41.445,
+      "eval_physics_samples_per_second": 12.064,
+      "eval_physics_steps_per_second": 3.016,
+      "step": 700
+    },
+    {
+      "entropy": 0.5965470356866718,
+      "epoch": 1.1360000000000001,
+      "grad_norm": 1.359375,
+      "learning_rate": 1.418e-05,
+      "loss": 0.5793,
+      "mean_token_accuracy": 0.8342778768390418,
+      "num_tokens": 23689161.0,
+      "step": 710
+    },
+    {
+      "entropy": 0.6170177994295954,
+      "epoch": 1.152,
+      "grad_norm": 1.6015625,
+      "learning_rate": 1.4380000000000001e-05,
+      "loss": 0.5984,
+      "mean_token_accuracy": 0.8301392119377852,
+      "num_tokens": 24016439.0,
+      "step": 720
+    },
+    {
+      "entropy": 0.6231384797021746,
+      "epoch": 1.168,
+      "grad_norm": 1.640625,
+      "learning_rate": 1.4580000000000001e-05,
+      "loss": 0.6024,
+      "mean_token_accuracy": 0.8265216089785099,
+      "num_tokens": 24344431.0,
+      "step": 730
+    },
+    {
+      "entropy": 0.6168253228068352,
+      "epoch": 1.184,
+      "grad_norm": 1.734375,
+      "learning_rate": 1.478e-05,
+      "loss": 0.5912,
+      "mean_token_accuracy": 0.8295746680349112,
+      "num_tokens": 24668087.0,
+      "step": 740
+    },
+    {
+      "entropy": 0.6063494171947241,
+      "epoch": 1.2,
+      "grad_norm": 1.3828125,
+      "learning_rate": 1.498e-05,
+      "loss": 0.5805,
+      "mean_token_accuracy": 0.8319725230336189,
+      "num_tokens": 25000629.0,
+      "step": 750
+    },
+    {
+      "entropy": 0.6178384765982627,
+      "epoch": 1.216,
+      "grad_norm": 1.6796875,
+      "learning_rate": 1.5180000000000002e-05,
+      "loss": 0.5998,
+      "mean_token_accuracy": 0.8278374817222357,
+      "num_tokens": 25324324.0,
+      "step": 760
+    },
+    {
+      "entropy": 0.6116716200485826,
+      "epoch": 1.232,
+      "grad_norm": 1.4453125,
+      "learning_rate": 1.5380000000000002e-05,
+      "loss": 0.5838,
+      "mean_token_accuracy": 0.8310758478939533,
+      "num_tokens": 25658795.0,
+      "step": 770
+    },
+    {
+      "entropy": 0.6036085536703467,
+      "epoch": 1.248,
+      "grad_norm": 1.4609375,
+      "learning_rate": 1.5580000000000003e-05,
+      "loss": 0.5835,
+      "mean_token_accuracy": 0.8323761031031609,
+      "num_tokens": 25993177.0,
+      "step": 780
+    },
+    {
+      "entropy": 0.6078255159780384,
+      "epoch": 1.264,
+      "grad_norm": 1.4921875,
+      "learning_rate": 1.578e-05,
+      "loss": 0.5819,
+      "mean_token_accuracy": 0.8314513519406319,
+      "num_tokens": 26329726.0,
+      "step": 790
+    },
+    {
+      "entropy": 0.6271371256560088,
+      "epoch": 1.28,
+      "grad_norm": 1.4609375,
+      "learning_rate": 1.5980000000000003e-05,
+      "loss": 0.6049,
+      "mean_token_accuracy": 0.8257384181022644,
+      "num_tokens": 26664093.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.28,
+      "eval_chemistry_entropy": 0.6834847524166107,
+      "eval_chemistry_loss": 0.6882474422454834,
+      "eval_chemistry_mean_token_accuracy": 0.8116786751747131,
+      "eval_chemistry_num_tokens": 26664093.0,
+      "eval_chemistry_runtime": 34.7916,
+      "eval_chemistry_samples_per_second": 14.371,
+      "eval_chemistry_steps_per_second": 3.593,
+      "step": 800
+    },
+    {
+      "epoch": 1.28,
+      "eval_physics_entropy": 0.6253828897476196,
+      "eval_physics_loss": 0.6271702647209167,
+      "eval_physics_mean_token_accuracy": 0.8248588938713074,
+      "eval_physics_num_tokens": 26664093.0,
+      "eval_physics_runtime": 41.5612,
+      "eval_physics_samples_per_second": 12.03,
+      "eval_physics_steps_per_second": 3.008,
+      "step": 800
+    },
+    {
+      "entropy": 0.5998634301126003,
+      "epoch": 1.296,
+      "grad_norm": 1.5625,
+      "learning_rate": 1.618e-05,
+      "loss": 0.5787,
+      "mean_token_accuracy": 0.8333883356302977,
+      "num_tokens": 26994877.0,
+      "step": 810
+    },
+    {
+      "entropy": 0.6259857648983598,
+      "epoch": 1.312,
+      "grad_norm": 1.5703125,
+      "learning_rate": 1.638e-05,
+      "loss": 0.6016,
+      "mean_token_accuracy": 0.8277184132486581,
+      "num_tokens": 27328198.0,
+      "step": 820
+    },
+    {
+      "entropy": 0.605760240368545,
+      "epoch": 1.328,
+      "grad_norm": 1.5546875,
+      "learning_rate": 1.658e-05,
+      "loss": 0.5891,
+      "mean_token_accuracy": 0.830071647465229,
+      "num_tokens": 27658818.0,
+      "step": 830
+    },
+    {
+      "entropy": 0.6093621645122766,
+      "epoch": 1.3439999999999999,
+      "grad_norm": 1.546875,
+      "learning_rate": 1.6780000000000002e-05,
+      "loss": 0.5882,
+      "mean_token_accuracy": 0.8291759472340345,
+      "num_tokens": 27991108.0,
+      "step": 840
+    },
+    {
+      "entropy": 0.6125000966712832,
+      "epoch": 1.3599999999999999,
+      "grad_norm": 1.3125,
+      "learning_rate": 1.698e-05,
+      "loss": 0.582,
+      "mean_token_accuracy": 0.8292897183448076,
+      "num_tokens": 28327775.0,
+      "step": 850
+    },
+    {
+      "entropy": 0.6071364052593708,
+      "epoch": 1.376,
+      "grad_norm": 1.671875,
+      "learning_rate": 1.718e-05,
+      "loss": 0.5823,
+      "mean_token_accuracy": 0.8310487173497677,
+      "num_tokens": 28666116.0,
+      "step": 860
+    },
+    {
+      "entropy": 0.5939128069207072,
+      "epoch": 1.392,
+      "grad_norm": 1.4375,
+      "learning_rate": 1.7380000000000003e-05,
+      "loss": 0.5723,
+      "mean_token_accuracy": 0.8350477192550898,
+      "num_tokens": 29004293.0,
+      "step": 870
+    },
+    {
+      "entropy": 0.6024157986044884,
+      "epoch": 1.408,
+      "grad_norm": 1.421875,
+      "learning_rate": 1.758e-05,
+      "loss": 0.5827,
+      "mean_token_accuracy": 0.8312420520931483,
+      "num_tokens": 29330131.0,
+      "step": 880
+    },
+    {
+      "entropy": 0.6242644298821688,
+      "epoch": 1.424,
+      "grad_norm": 1.2890625,
+      "learning_rate": 1.7780000000000003e-05,
+      "loss": 0.6039,
+      "mean_token_accuracy": 0.8263261545449495,
+      "num_tokens": 29661850.0,
+      "step": 890
+    },
+    {
+      "entropy": 0.5897390013560653,
+      "epoch": 1.44,
+      "grad_norm": 1.2890625,
+      "learning_rate": 1.798e-05,
+      "loss": 0.5687,
+      "mean_token_accuracy": 0.8350921101868153,
+      "num_tokens": 29999392.0,
+      "step": 900
+    },
+    {
+      "epoch": 1.44,
+      "eval_chemistry_entropy": 0.6641502072811126,
+      "eval_chemistry_loss": 0.6906805634498596,
+      "eval_chemistry_mean_token_accuracy": 0.8092845277786255,
+      "eval_chemistry_num_tokens": 29999392.0,
+      "eval_chemistry_runtime": 34.7482,
+      "eval_chemistry_samples_per_second": 14.389,
+      "eval_chemistry_steps_per_second": 3.597,
+      "step": 900
+    },
+    {
+      "epoch": 1.44,
+      "eval_physics_entropy": 0.6082331819534301,
+      "eval_physics_loss": 0.6185559034347534,
+      "eval_physics_mean_token_accuracy": 0.8265736656188964,
+      "eval_physics_num_tokens": 29999392.0,
+      "eval_physics_runtime": 41.4301,
+      "eval_physics_samples_per_second": 12.069,
+      "eval_physics_steps_per_second": 3.017,
+      "step": 900
+    },
+    {
+      "entropy": 0.5889361893758178,
+      "epoch": 1.456,
+      "grad_norm": 1.2265625,
+      "learning_rate": 1.8180000000000002e-05,
+      "loss": 0.5722,
+      "mean_token_accuracy": 0.8360798362642526,
+      "num_tokens": 30341004.0,
+      "step": 910
+    },
+    {
+      "entropy": 0.6223346112295985,
+      "epoch": 1.472,
+      "grad_norm": 1.484375,
+      "learning_rate": 1.8380000000000004e-05,
+      "loss": 0.599,
+      "mean_token_accuracy": 0.8281814876943827,
+      "num_tokens": 30667689.0,
+      "step": 920
+    },
+    {
+      "entropy": 0.6086510250344872,
+      "epoch": 1.488,
+      "grad_norm": 1.25,
+      "learning_rate": 1.858e-05,
+      "loss": 0.5865,
+      "mean_token_accuracy": 0.8305441569536924,
+      "num_tokens": 30999697.0,
+      "step": 930
+    },
+    {
+      "entropy": 0.5955386120826006,
+      "epoch": 1.504,
+      "grad_norm": 1.234375,
+      "learning_rate": 1.878e-05,
+      "loss": 0.5699,
+      "mean_token_accuracy": 0.8333926767110824,
+      "num_tokens": 31332775.0,
+      "step": 940
+    },
+    {
+      "entropy": 0.6133549332618713,
+      "epoch": 1.52,
+      "grad_norm": 1.4609375,
+      "learning_rate": 1.898e-05,
+      "loss": 0.5915,
+      "mean_token_accuracy": 0.8288876831531524,
+      "num_tokens": 31666521.0,
+      "step": 950
+    },
+    {
+      "entropy": 0.6097657412290574,
+      "epoch": 1.536,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.918e-05,
+      "loss": 0.5866,
+      "mean_token_accuracy": 0.8288511786609888,
+      "num_tokens": 31998232.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.5852528993040323,
+      "epoch": 1.552,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.938e-05,
+      "loss": 0.5684,
+      "mean_token_accuracy": 0.8353338003158569,
+      "num_tokens": 32328085.0,
+      "step": 970
+    },
+    {
+      "entropy": 0.5982964089140296,
+      "epoch": 1.568,
+      "grad_norm": 1.375,
+      "learning_rate": 1.9580000000000002e-05,
+      "loss": 0.5753,
+      "mean_token_accuracy": 0.832823446765542,
+      "num_tokens": 32656603.0,
+      "step": 980
+    },
+    {
+      "entropy": 0.6145529938861728,
+      "epoch": 1.584,
+      "grad_norm": 1.3828125,
+      "learning_rate": 1.978e-05,
+      "loss": 0.5882,
+      "mean_token_accuracy": 0.8275394171476365,
+      "num_tokens": 32983769.0,
+      "step": 990
+    },
+    {
+      "entropy": 0.6043317306786775,
+      "epoch": 1.6,
+      "grad_norm": 1.3125,
+      "learning_rate": 1.9980000000000002e-05,
+      "loss": 0.5823,
+      "mean_token_accuracy": 0.8310152795165777,
+      "num_tokens": 33309282.0,
+      "step": 1000
+    },
+    {
+      "epoch": 1.6,
+      "eval_chemistry_entropy": 0.6794673941135406,
+      "eval_chemistry_loss": 0.6896947026252747,
+      "eval_chemistry_mean_token_accuracy": 0.8101202611923217,
+      "eval_chemistry_num_tokens": 33309282.0,
+      "eval_chemistry_runtime": 34.8102,
+      "eval_chemistry_samples_per_second": 14.364,
+      "eval_chemistry_steps_per_second": 3.591,
+      "step": 1000
+    },
+    {
+      "epoch": 1.6,
+      "eval_physics_entropy": 0.607559362411499,
+      "eval_physics_loss": 0.6121883392333984,
+      "eval_physics_mean_token_accuracy": 0.8272243642807007,
+      "eval_physics_num_tokens": 33309282.0,
+      "eval_physics_runtime": 41.0884,
+      "eval_physics_samples_per_second": 12.169,
+      "eval_physics_steps_per_second": 3.042,
+      "step": 1000
+    },
+    {
+      "entropy": 0.5851200748234987,
+      "epoch": 1.616,
+      "grad_norm": 1.328125,
+      "learning_rate": 1.9980000000000002e-05,
+      "loss": 0.5675,
+      "mean_token_accuracy": 0.8361915960907936,
+      "num_tokens": 33647616.0,
+      "step": 1010
+    },
+    {
+      "entropy": 0.5935754748061299,
+      "epoch": 1.6320000000000001,
+      "grad_norm": 1.3125,
+      "learning_rate": 1.995777777777778e-05,
+      "loss": 0.5753,
+      "mean_token_accuracy": 0.8332207947969437,
+      "num_tokens": 33977586.0,
+      "step": 1020
+    },
+    {
+      "entropy": 0.5731598405167461,
+      "epoch": 1.6480000000000001,
+      "grad_norm": 1.390625,
+      "learning_rate": 1.9935555555555557e-05,
+      "loss": 0.5536,
+      "mean_token_accuracy": 0.8382600821554661,
+      "num_tokens": 34312436.0,
+      "step": 1030
+    },
+    {
+      "entropy": 0.5990395007655025,
+      "epoch": 1.6640000000000001,
+      "grad_norm": 1.5390625,
+      "learning_rate": 1.9913333333333335e-05,
+      "loss": 0.5747,
+      "mean_token_accuracy": 0.8332929655909538,
+      "num_tokens": 34641549.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.589907999895513,
+      "epoch": 1.6800000000000002,
+      "grad_norm": 1.1953125,
+      "learning_rate": 1.9891111111111112e-05,
+      "loss": 0.5685,
+      "mean_token_accuracy": 0.8342956956475973,
+      "num_tokens": 34977031.0,
+      "step": 1050
+    },
+    {
+      "entropy": 0.5947070028632879,
+      "epoch": 1.696,
+      "grad_norm": 1.3046875,
+      "learning_rate": 1.986888888888889e-05,
+      "loss": 0.5736,
+      "mean_token_accuracy": 0.8329529393464327,
+      "num_tokens": 35300962.0,
+      "step": 1060
+    },
+    {
+      "entropy": 0.595324145630002,
+      "epoch": 1.712,
+      "grad_norm": 1.3359375,
+      "learning_rate": 1.9846666666666668e-05,
+      "loss": 0.574,
+      "mean_token_accuracy": 0.831612591072917,
+      "num_tokens": 35642662.0,
+      "step": 1070
+    },
+    {
+      "entropy": 0.581332896463573,
+      "epoch": 1.728,
+      "grad_norm": 1.3046875,
+      "learning_rate": 1.9824444444444445e-05,
+      "loss": 0.5648,
+      "mean_token_accuracy": 0.8354741148650646,
+      "num_tokens": 35980993.0,
+      "step": 1080
+    },
+    {
+      "entropy": 0.5898414205759763,
+      "epoch": 1.744,
+      "grad_norm": 1.2890625,
+      "learning_rate": 1.9802222222222226e-05,
+      "loss": 0.568,
+      "mean_token_accuracy": 0.834344669431448,
+      "num_tokens": 36326797.0,
+      "step": 1090
+    },
+    {
+      "entropy": 0.5798839410766959,
+      "epoch": 1.76,
+      "grad_norm": 1.3828125,
+      "learning_rate": 1.978e-05,
+      "loss": 0.5617,
+      "mean_token_accuracy": 0.8352703254669904,
+      "num_tokens": 36654797.0,
+      "step": 1100
+    },
+    {
+      "epoch": 1.76,
+      "eval_chemistry_entropy": 0.662034518957138,
+      "eval_chemistry_loss": 0.6911905407905579,
+      "eval_chemistry_mean_token_accuracy": 0.8086481132507324,
+      "eval_chemistry_num_tokens": 36654797.0,
+      "eval_chemistry_runtime": 34.786,
+      "eval_chemistry_samples_per_second": 14.374,
+      "eval_chemistry_steps_per_second": 3.593,
+      "step": 1100
+    },
+    {
+      "epoch": 1.76,
+      "eval_physics_entropy": 0.5805578632354736,
+      "eval_physics_loss": 0.6039645671844482,
+      "eval_physics_mean_token_accuracy": 0.8294073853492737,
+      "eval_physics_num_tokens": 36654797.0,
+      "eval_physics_runtime": 41.4109,
+      "eval_physics_samples_per_second": 12.074,
+      "eval_physics_steps_per_second": 3.019,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6007018189877271,
+      "epoch": 1.776,
+      "grad_norm": 1.2890625,
+      "learning_rate": 1.975777777777778e-05,
+      "loss": 0.5839,
+      "mean_token_accuracy": 0.8299636762589216,
+      "num_tokens": 36989310.0,
+      "step": 1110
+    },
+    {
+      "entropy": 0.5859053635969758,
+      "epoch": 1.792,
+      "grad_norm": 1.3828125,
+      "learning_rate": 1.9735555555555556e-05,
+      "loss": 0.5609,
+      "mean_token_accuracy": 0.8349329460412264,
+      "num_tokens": 37318535.0,
+      "step": 1120
+    },
+    {
+      "entropy": 0.5717526819556952,
+      "epoch": 1.808,
+      "grad_norm": 1.2109375,
+      "learning_rate": 1.9713333333333337e-05,
+      "loss": 0.5516,
+      "mean_token_accuracy": 0.837704149633646,
+      "num_tokens": 37658889.0,
+      "step": 1130
+    },
+    {
+      "entropy": 0.5855952553451061,
+      "epoch": 1.8239999999999998,
+      "grad_norm": 1.25,
+      "learning_rate": 1.969111111111111e-05,
+      "loss": 0.5609,
+      "mean_token_accuracy": 0.8358533851802349,
+      "num_tokens": 37995198.0,
+      "step": 1140
+    },
+    {
+      "entropy": 0.576733535528183,
+      "epoch": 1.8399999999999999,
+      "grad_norm": 1.2421875,
+      "learning_rate": 1.9668888888888892e-05,
+      "loss": 0.5602,
+      "mean_token_accuracy": 0.8364770855754614,
+      "num_tokens": 38325542.0,
+      "step": 1150
+    },
+    {
+      "entropy": 0.5805189371109009,
+      "epoch": 1.8559999999999999,
+      "grad_norm": 1.234375,
+      "learning_rate": 1.9646666666666666e-05,
+      "loss": 0.5562,
+      "mean_token_accuracy": 0.8367651741951704,
+      "num_tokens": 38668148.0,
+      "step": 1160
+    },
+    {
+      "entropy": 0.5755893625319004,
+      "epoch": 1.8719999999999999,
+      "grad_norm": 1.234375,
+      "learning_rate": 1.9624444444444447e-05,
+      "loss": 0.5567,
+      "mean_token_accuracy": 0.8351109046489,
+      "num_tokens": 39001652.0,
+      "step": 1170
+    },
+    {
+      "entropy": 0.593781272508204,
+      "epoch": 1.888,
+      "grad_norm": 1.375,
+      "learning_rate": 1.9602222222222225e-05,
+      "loss": 0.5698,
+      "mean_token_accuracy": 0.8326846692711115,
+      "num_tokens": 39333474.0,
+      "step": 1180
+    },
+    {
+      "entropy": 0.5997770514339209,
+      "epoch": 1.904,
+      "grad_norm": 1.25,
+      "learning_rate": 1.9580000000000002e-05,
+      "loss": 0.5793,
+      "mean_token_accuracy": 0.8314935185015202,
+      "num_tokens": 39663055.0,
+      "step": 1190
+    },
+    {
+      "entropy": 0.5944939412176609,
+      "epoch": 1.92,
+      "grad_norm": 1.2265625,
+      "learning_rate": 1.955777777777778e-05,
+      "loss": 0.5733,
+      "mean_token_accuracy": 0.8316057369112968,
+      "num_tokens": 39989824.0,
+      "step": 1200
+    },
+    {
+      "epoch": 1.92,
+      "eval_chemistry_entropy": 0.6759118571281433,
+      "eval_chemistry_loss": 0.6883504986763,
+      "eval_chemistry_mean_token_accuracy": 0.8099100780487061,
+      "eval_chemistry_num_tokens": 39989824.0,
+      "eval_chemistry_runtime": 35.0777,
+      "eval_chemistry_samples_per_second": 14.254,
+      "eval_chemistry_steps_per_second": 3.564,
+      "step": 1200
+    },
+    {
+      "epoch": 1.92,
+      "eval_physics_entropy": 0.5913804955482483,
+      "eval_physics_loss": 0.5959852337837219,
+      "eval_physics_mean_token_accuracy": 0.8307065281867981,
+      "eval_physics_num_tokens": 39989824.0,
+      "eval_physics_runtime": 41.4833,
+      "eval_physics_samples_per_second": 12.053,
+      "eval_physics_steps_per_second": 3.013,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5832644954323769,
+      "epoch": 1.936,
+      "grad_norm": 1.1953125,
+      "learning_rate": 1.9535555555555557e-05,
+      "loss": 0.5641,
+      "mean_token_accuracy": 0.8353390946984292,
+      "num_tokens": 40325070.0,
+      "step": 1210
+    },
+    {
+      "entropy": 0.5566705971956253,
+      "epoch": 1.952,
+      "grad_norm": 1.28125,
+      "learning_rate": 1.9513333333333335e-05,
+      "loss": 0.5374,
+      "mean_token_accuracy": 0.8413328900933266,
+      "num_tokens": 40659614.0,
+      "step": 1220
+    },
+    {
+      "entropy": 0.5830600494518876,
+      "epoch": 1.968,
+      "grad_norm": 1.15625,
+      "learning_rate": 1.9491111111111113e-05,
+      "loss": 0.5625,
+      "mean_token_accuracy": 0.835172512754798,
+      "num_tokens": 40991508.0,
+      "step": 1230
+    },
+    {
+      "entropy": 0.581772717833519,
+      "epoch": 1.984,
+      "grad_norm": 1.25,
+      "learning_rate": 1.946888888888889e-05,
+      "loss": 0.5629,
+      "mean_token_accuracy": 0.8342908933758736,
+      "num_tokens": 41328779.0,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5848113460466265,
+      "epoch": 2.0,
+      "grad_norm": 1.171875,
+      "learning_rate": 1.9446666666666668e-05,
+      "loss": 0.5611,
+      "mean_token_accuracy": 0.8350322283804417,
+      "num_tokens": 41664296.0,
+      "step": 1250
+    },
+    {
+      "entropy": 0.48217232096940277,
+      "epoch": 2.016,
+      "grad_norm": 1.375,
+      "learning_rate": 1.9424444444444446e-05,
+      "loss": 0.4598,
+      "mean_token_accuracy": 0.8575068596750498,
+      "num_tokens": 42001193.0,
+      "step": 1260
+    },
+    {
+      "entropy": 0.47964414674788713,
+      "epoch": 2.032,
+      "grad_norm": 1.4921875,
+      "learning_rate": 1.9402222222222223e-05,
+      "loss": 0.4587,
+      "mean_token_accuracy": 0.8569075904786587,
+      "num_tokens": 42340757.0,
+      "step": 1270
+    },
+    {
+      "entropy": 0.49382132217288016,
+      "epoch": 2.048,
+      "grad_norm": 1.4921875,
+      "learning_rate": 1.938e-05,
+      "loss": 0.4713,
+      "mean_token_accuracy": 0.8529856160283089,
+      "num_tokens": 42670324.0,
+      "step": 1280
+    },
+    {
+      "entropy": 0.47823069952428343,
+      "epoch": 2.064,
+      "grad_norm": 1.3984375,
+      "learning_rate": 1.935777777777778e-05,
+      "loss": 0.4566,
+      "mean_token_accuracy": 0.8581176064908504,
+      "num_tokens": 43010231.0,
+      "step": 1290
+    },
+    {
+      "entropy": 0.48923970870673655,
+      "epoch": 2.08,
+      "grad_norm": 1.3828125,
+      "learning_rate": 1.9335555555555556e-05,
+      "loss": 0.4667,
+      "mean_token_accuracy": 0.8556209966540337,
+      "num_tokens": 43340227.0,
+      "step": 1300
+    },
+    {
+      "epoch": 2.08,
+      "eval_chemistry_entropy": 0.546965966463089,
+      "eval_chemistry_loss": 0.7304518818855286,
+      "eval_chemistry_mean_token_accuracy": 0.8063767719268798,
+      "eval_chemistry_num_tokens": 43340227.0,
+      "eval_chemistry_runtime": 34.7458,
+      "eval_chemistry_samples_per_second": 14.39,
+      "eval_chemistry_steps_per_second": 3.598,
+      "step": 1300
+    },
+    {
+      "epoch": 2.08,
+      "eval_physics_entropy": 0.49418199634552,
+      "eval_physics_loss": 0.6146922707557678,
+      "eval_physics_mean_token_accuracy": 0.8299191513061523,
+      "eval_physics_num_tokens": 43340227.0,
+      "eval_physics_runtime": 41.365,
+      "eval_physics_samples_per_second": 12.088,
+      "eval_physics_steps_per_second": 3.022,
+      "step": 1300
+    },
+    {
+      "entropy": 0.4770382607355714,
+      "epoch": 2.096,
+      "grad_norm": 1.3984375,
+      "learning_rate": 1.9313333333333334e-05,
+      "loss": 0.461,
+      "mean_token_accuracy": 0.8564476538449526,
+      "num_tokens": 43673099.0,
+      "step": 1310
+    },
+    {
+      "entropy": 0.4707287013530731,
+      "epoch": 2.112,
+      "grad_norm": 1.390625,
+      "learning_rate": 1.9291111111111115e-05,
+      "loss": 0.4509,
+      "mean_token_accuracy": 0.8583695068955421,
+      "num_tokens": 44005986.0,
+      "step": 1320
+    },
+    {
+      "entropy": 0.4803559150546789,
+      "epoch": 2.128,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.926888888888889e-05,
+      "loss": 0.4643,
+      "mean_token_accuracy": 0.8548851139843464,
+      "num_tokens": 44321465.0,
+      "step": 1330
+    },
+    {
+      "entropy": 0.48221837151795627,
+      "epoch": 2.144,
+      "grad_norm": 1.375,
+      "learning_rate": 1.924666666666667e-05,
+      "loss": 0.4619,
+      "mean_token_accuracy": 0.8559276178479195,
+      "num_tokens": 44661027.0,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5006665755994618,
+      "epoch": 2.16,
+      "grad_norm": 1.4140625,
+      "learning_rate": 1.9224444444444444e-05,
+      "loss": 0.4817,
+      "mean_token_accuracy": 0.8508365921676159,
+      "num_tokens": 44994236.0,
+      "step": 1350
+    },
+    {
+      "entropy": 0.4735172586515546,
+      "epoch": 2.176,
+      "grad_norm": 1.328125,
+      "learning_rate": 1.9202222222222225e-05,
+      "loss": 0.4583,
+      "mean_token_accuracy": 0.856869063898921,
+      "num_tokens": 45334280.0,
+      "step": 1360
+    },
+    {
+      "entropy": 0.47296210322529075,
+      "epoch": 2.192,
+      "grad_norm": 1.3984375,
+      "learning_rate": 1.918e-05,
+      "loss": 0.4593,
+      "mean_token_accuracy": 0.857900059223175,
+      "num_tokens": 45666954.0,
+      "step": 1370
+    },
+    {
+      "entropy": 0.47185815051198005,
+      "epoch": 2.208,
+      "grad_norm": 1.4609375,
+      "learning_rate": 1.915777777777778e-05,
+      "loss": 0.452,
+      "mean_token_accuracy": 0.858792957663536,
+      "num_tokens": 45999546.0,
+      "step": 1380
+    },
+    {
+      "entropy": 0.47457288652658464,
+      "epoch": 2.224,
+      "grad_norm": 1.328125,
+      "learning_rate": 1.9135555555555555e-05,
+      "loss": 0.4591,
+      "mean_token_accuracy": 0.8570701099932194,
+      "num_tokens": 46333809.0,
+      "step": 1390
+    },
+    {
+      "entropy": 0.4764716824516654,
+      "epoch": 2.24,
+      "grad_norm": 1.3671875,
+      "learning_rate": 1.9113333333333336e-05,
+      "loss": 0.4596,
+      "mean_token_accuracy": 0.856460390239954,
+      "num_tokens": 46673328.0,
+      "step": 1400
+    },
+    {
+      "epoch": 2.24,
+      "eval_chemistry_entropy": 0.5586697208881378,
+      "eval_chemistry_loss": 0.7306581735610962,
+      "eval_chemistry_mean_token_accuracy": 0.8062993106842041,
+      "eval_chemistry_num_tokens": 46673328.0,
+      "eval_chemistry_runtime": 34.6938,
+      "eval_chemistry_samples_per_second": 14.412,
+      "eval_chemistry_steps_per_second": 3.603,
+      "step": 1400
+    },
+    {
+      "epoch": 2.24,
+      "eval_physics_entropy": 0.5080966579914094,
+      "eval_physics_loss": 0.611297607421875,
+      "eval_physics_mean_token_accuracy": 0.830946072101593,
+      "eval_physics_num_tokens": 46673328.0,
+      "eval_physics_runtime": 41.3319,
+      "eval_physics_samples_per_second": 12.097,
+      "eval_physics_steps_per_second": 3.024,
+      "step": 1400
+    },
+    {
+      "entropy": 0.4922581292688847,
+      "epoch": 2.2560000000000002,
+      "grad_norm": 1.484375,
+      "learning_rate": 1.9091111111111113e-05,
+      "loss": 0.4743,
+      "mean_token_accuracy": 0.8529286984354257,
+      "num_tokens": 47001509.0,
+      "step": 1410
+    },
+    {
+      "entropy": 0.47681235875934364,
+      "epoch": 2.2720000000000002,
+      "grad_norm": 1.4921875,
+      "learning_rate": 1.906888888888889e-05,
+      "loss": 0.4583,
+      "mean_token_accuracy": 0.8574760638177394,
+      "num_tokens": 47340142.0,
+      "step": 1420
+    },
+    {
+      "entropy": 0.4867669759318233,
+      "epoch": 2.288,
+      "grad_norm": 1.4453125,
+      "learning_rate": 1.904666666666667e-05,
+      "loss": 0.4711,
+      "mean_token_accuracy": 0.853106002509594,
+      "num_tokens": 47677550.0,
+      "step": 1430
+    },
+    {
+      "entropy": 0.4855895210057497,
+      "epoch": 2.304,
+      "grad_norm": 1.5703125,
+      "learning_rate": 1.9024444444444446e-05,
+      "loss": 0.4641,
+      "mean_token_accuracy": 0.855696403235197,
+      "num_tokens": 47993851.0,
+      "step": 1440
+    },
+    {
+      "entropy": 0.48173528388142584,
+      "epoch": 2.32,
+      "grad_norm": 1.328125,
+      "learning_rate": 1.9002222222222224e-05,
+      "loss": 0.4624,
+      "mean_token_accuracy": 0.8548369336873293,
+      "num_tokens": 48332577.0,
+      "step": 1450
+    },
+    {
+      "entropy": 0.46310502383857965,
+      "epoch": 2.336,
+      "grad_norm": 1.3984375,
+      "learning_rate": 1.898e-05,
+      "loss": 0.4466,
+      "mean_token_accuracy": 0.8590828448534011,
+      "num_tokens": 48676623.0,
+      "step": 1460
+    },
+    {
+      "entropy": 0.46570080444216727,
+      "epoch": 2.352,
+      "grad_norm": 1.3828125,
+      "learning_rate": 1.895777777777778e-05,
+      "loss": 0.4477,
+      "mean_token_accuracy": 0.8588027618825436,
+      "num_tokens": 49007770.0,
+      "step": 1470
+    },
+    {
+      "entropy": 0.47828576155006886,
+      "epoch": 2.368,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.8935555555555556e-05,
+      "loss": 0.463,
+      "mean_token_accuracy": 0.8564012508839369,
+      "num_tokens": 49350016.0,
+      "step": 1480
+    },
+    {
+      "entropy": 0.48590457029640677,
+      "epoch": 2.384,
+      "grad_norm": 1.609375,
+      "learning_rate": 1.8913333333333334e-05,
+      "loss": 0.4677,
+      "mean_token_accuracy": 0.8538792595267296,
+      "num_tokens": 49679074.0,
+      "step": 1490
+    },
+    {
+      "entropy": 0.4809177804738283,
+      "epoch": 2.4,
+      "grad_norm": 1.4921875,
+      "learning_rate": 1.8891111111111115e-05,
+      "loss": 0.4631,
+      "mean_token_accuracy": 0.8553439110517502,
+      "num_tokens": 50020648.0,
+      "step": 1500
+    },
+    {
+      "epoch": 2.4,
+      "eval_chemistry_entropy": 0.5487088859081268,
+      "eval_chemistry_loss": 0.7314473986625671,
+      "eval_chemistry_mean_token_accuracy": 0.8061141548156738,
+      "eval_chemistry_num_tokens": 50020648.0,
+      "eval_chemistry_runtime": 34.6612,
+      "eval_chemistry_samples_per_second": 14.425,
+      "eval_chemistry_steps_per_second": 3.606,
+      "step": 1500
+    },
+    {
+      "epoch": 2.4,
+      "eval_physics_entropy": 0.4990242130756378,
+      "eval_physics_loss": 0.6052322387695312,
+      "eval_physics_mean_token_accuracy": 0.8326480937004089,
+      "eval_physics_num_tokens": 50020648.0,
+      "eval_physics_runtime": 40.9615,
+      "eval_physics_samples_per_second": 12.207,
+      "eval_physics_steps_per_second": 3.052,
+      "step": 1500
+    },
+    {
+      "entropy": 0.4788668820634484,
+      "epoch": 2.416,
+      "grad_norm": 1.3984375,
+      "learning_rate": 1.886888888888889e-05,
+      "loss": 0.4614,
+      "mean_token_accuracy": 0.8562805935740471,
+      "num_tokens": 50370135.0,
+      "step": 1510
+    },
+    {
+      "entropy": 0.4815561521798372,
+      "epoch": 2.432,
+      "grad_norm": 1.4375,
+      "learning_rate": 1.884666666666667e-05,
+      "loss": 0.465,
+      "mean_token_accuracy": 0.8552235502749681,
+      "num_tokens": 50695363.0,
+      "step": 1520
+    },
+    {
+      "entropy": 0.494814782217145,
+      "epoch": 2.448,
+      "grad_norm": 1.4453125,
+      "learning_rate": 1.8824444444444445e-05,
+      "loss": 0.4771,
+      "mean_token_accuracy": 0.851343784853816,
+      "num_tokens": 51023498.0,
+      "step": 1530
+    },
+    {
+      "entropy": 0.4887571394443512,
+      "epoch": 2.464,
+      "grad_norm": 1.5234375,
+      "learning_rate": 1.8802222222222226e-05,
+      "loss": 0.4677,
+      "mean_token_accuracy": 0.8553626071661711,
+      "num_tokens": 51349816.0,
+      "step": 1540
+    },
+    {
+      "entropy": 0.48406565077602864,
+      "epoch": 2.48,
+      "grad_norm": 1.5,
+      "learning_rate": 1.878e-05,
+      "loss": 0.4662,
+      "mean_token_accuracy": 0.8541974198073149,
+      "num_tokens": 51680507.0,
+      "step": 1550
+    },
+    {
+      "entropy": 0.4840725682675838,
+      "epoch": 2.496,
+      "grad_norm": 1.40625,
+      "learning_rate": 1.875777777777778e-05,
+      "loss": 0.4692,
+      "mean_token_accuracy": 0.8539839003235101,
+      "num_tokens": 52017752.0,
+      "step": 1560
+    },
+    {
+      "entropy": 0.47237055795267224,
+      "epoch": 2.512,
+      "grad_norm": 1.5390625,
+      "learning_rate": 1.873555555555556e-05,
+      "loss": 0.4547,
+      "mean_token_accuracy": 0.8578830562531948,
+      "num_tokens": 52345439.0,
+      "step": 1570
+    },
+    {
+      "entropy": 0.47300295867025854,
+      "epoch": 2.528,
+      "grad_norm": 1.3671875,
+      "learning_rate": 1.8713333333333336e-05,
+      "loss": 0.456,
+      "mean_token_accuracy": 0.8584080573171378,
+      "num_tokens": 52683886.0,
+      "step": 1580
+    },
+    {
+      "entropy": 0.48024273496121167,
+      "epoch": 2.544,
+      "grad_norm": 1.5546875,
+      "learning_rate": 1.8691111111111114e-05,
+      "loss": 0.4609,
+      "mean_token_accuracy": 0.8557684738188982,
+      "num_tokens": 53004529.0,
+      "step": 1590
+    },
+    {
+      "entropy": 0.47996121235191824,
+      "epoch": 2.56,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.866888888888889e-05,
+      "loss": 0.4592,
+      "mean_token_accuracy": 0.8556542955338955,
+      "num_tokens": 53336280.0,
+      "step": 1600
+    },
+    {
+      "epoch": 2.56,
+      "eval_chemistry_entropy": 0.554151120185852,
+      "eval_chemistry_loss": 0.7288383841514587,
+      "eval_chemistry_mean_token_accuracy": 0.8056904196739196,
+      "eval_chemistry_num_tokens": 53336280.0,
+      "eval_chemistry_runtime": 34.6332,
+      "eval_chemistry_samples_per_second": 14.437,
+      "eval_chemistry_steps_per_second": 3.609,
+      "step": 1600
+    },
+    {
+      "epoch": 2.56,
+      "eval_physics_entropy": 0.49764198398590087,
+      "eval_physics_loss": 0.6016162037849426,
+      "eval_physics_mean_token_accuracy": 0.8329321603775024,
+      "eval_physics_num_tokens": 53336280.0,
+      "eval_physics_runtime": 41.2433,
+      "eval_physics_samples_per_second": 12.123,
+      "eval_physics_steps_per_second": 3.031,
+      "step": 1600
+    },
+    {
+      "entropy": 0.46981679294258355,
+      "epoch": 2.576,
+      "grad_norm": 1.4453125,
+      "learning_rate": 1.864666666666667e-05,
+      "loss": 0.4564,
+      "mean_token_accuracy": 0.8582920409739018,
+      "num_tokens": 53667647.0,
+      "step": 1610
+    },
+    {
+      "entropy": 0.47870821300894023,
+      "epoch": 2.592,
+      "grad_norm": 1.3046875,
+      "learning_rate": 1.8624444444444446e-05,
+      "loss": 0.4629,
+      "mean_token_accuracy": 0.8552918780595065,
+      "num_tokens": 53997614.0,
+      "step": 1620
+    },
+    {
+      "entropy": 0.47518060319125655,
+      "epoch": 2.608,
+      "grad_norm": 1.3828125,
+      "learning_rate": 1.8602222222222224e-05,
+      "loss": 0.4513,
+      "mean_token_accuracy": 0.8582707166671752,
+      "num_tokens": 54330195.0,
+      "step": 1630
+    },
+    {
+      "entropy": 0.48298279475420713,
+      "epoch": 2.624,
+      "grad_norm": 1.3671875,
+      "learning_rate": 1.858e-05,
+      "loss": 0.4625,
+      "mean_token_accuracy": 0.8556858822703362,
+      "num_tokens": 54664342.0,
+      "step": 1640
+    },
+    {
+      "entropy": 0.4753486838191748,
+      "epoch": 2.64,
+      "grad_norm": 1.46875,
+      "learning_rate": 1.855777777777778e-05,
+      "loss": 0.4582,
+      "mean_token_accuracy": 0.8570550225675106,
+      "num_tokens": 54993626.0,
+      "step": 1650
+    },
+    {
+      "entropy": 0.49773360043764114,
+      "epoch": 2.656,
+      "grad_norm": 1.5703125,
+      "learning_rate": 1.8535555555555557e-05,
+      "loss": 0.4796,
+      "mean_token_accuracy": 0.850704287365079,
+      "num_tokens": 55339962.0,
+      "step": 1660
+    },
+    {
+      "entropy": 0.48465393427759407,
+      "epoch": 2.672,
+      "grad_norm": 1.4765625,
+      "learning_rate": 1.8513333333333335e-05,
+      "loss": 0.4659,
+      "mean_token_accuracy": 0.8545554764568806,
+      "num_tokens": 55670567.0,
+      "step": 1670
+    },
+    {
+      "entropy": 0.4666805924847722,
+      "epoch": 2.6879999999999997,
+      "grad_norm": 1.46875,
+      "learning_rate": 1.8491111111111112e-05,
+      "loss": 0.4524,
+      "mean_token_accuracy": 0.8596824321895837,
+      "num_tokens": 56003156.0,
+      "step": 1680
+    },
+    {
+      "entropy": 0.4853338483721018,
+      "epoch": 2.7039999999999997,
+      "grad_norm": 1.3984375,
+      "learning_rate": 1.846888888888889e-05,
+      "loss": 0.4651,
+      "mean_token_accuracy": 0.8554550468921661,
+      "num_tokens": 56342926.0,
+      "step": 1690
+    },
+    {
+      "entropy": 0.47395940907299516,
+      "epoch": 2.7199999999999998,
+      "grad_norm": 1.5625,
+      "learning_rate": 1.8446666666666667e-05,
+      "loss": 0.4578,
+      "mean_token_accuracy": 0.8566099297255277,
+      "num_tokens": 56677521.0,
+      "step": 1700
+    },
+    {
+      "epoch": 2.7199999999999998,
+      "eval_chemistry_entropy": 0.5565269174575805,
+      "eval_chemistry_loss": 0.7264519333839417,
+      "eval_chemistry_mean_token_accuracy": 0.8058576831817627,
+      "eval_chemistry_num_tokens": 56677521.0,
+      "eval_chemistry_runtime": 34.6286,
+      "eval_chemistry_samples_per_second": 14.439,
+      "eval_chemistry_steps_per_second": 3.61,
+      "step": 1700
+    },
+    {
+      "epoch": 2.7199999999999998,
+      "eval_physics_entropy": 0.5007153513431549,
+      "eval_physics_loss": 0.5944604277610779,
+      "eval_physics_mean_token_accuracy": 0.8342739996910096,
+      "eval_physics_num_tokens": 56677521.0,
+      "eval_physics_runtime": 41.2898,
+      "eval_physics_samples_per_second": 12.11,
+      "eval_physics_steps_per_second": 3.027,
+      "step": 1700
+    },
+    {
+      "entropy": 0.49342543426901103,
+      "epoch": 2.7359999999999998,
+      "grad_norm": 1.4921875,
+      "learning_rate": 1.842444444444445e-05,
+      "loss": 0.4713,
+      "mean_token_accuracy": 0.8526182346045971,
+      "num_tokens": 57003161.0,
+      "step": 1710
+    },
+    {
+      "entropy": 0.4808831615373492,
+      "epoch": 2.752,
+      "grad_norm": 1.515625,
+      "learning_rate": 1.8402222222222223e-05,
+      "loss": 0.4654,
+      "mean_token_accuracy": 0.8546452801674604,
+      "num_tokens": 57335850.0,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4690231055021286,
+      "epoch": 2.768,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.8380000000000004e-05,
+      "loss": 0.4528,
+      "mean_token_accuracy": 0.8586602933704853,
+      "num_tokens": 57672649.0,
+      "step": 1730
+    },
+    {
+      "entropy": 0.48197978772222994,
+      "epoch": 2.784,
+      "grad_norm": 1.5,
+      "learning_rate": 1.8357777777777778e-05,
+      "loss": 0.4629,
+      "mean_token_accuracy": 0.856096263974905,
+      "num_tokens": 58007431.0,
+      "step": 1740
+    },
+    {
+      "entropy": 0.5036183711141348,
+      "epoch": 2.8,
+      "grad_norm": 1.4921875,
+      "learning_rate": 1.833555555555556e-05,
+      "loss": 0.4836,
+      "mean_token_accuracy": 0.8493415039032698,
+      "num_tokens": 58332730.0,
+      "step": 1750
+    },
+    {
+      "entropy": 0.462801137752831,
+      "epoch": 2.816,
+      "grad_norm": 1.359375,
+      "learning_rate": 1.8313333333333333e-05,
+      "loss": 0.4433,
+      "mean_token_accuracy": 0.8603610061109066,
+      "num_tokens": 58672106.0,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4801277067512274,
+      "epoch": 2.832,
+      "grad_norm": 1.4140625,
+      "learning_rate": 1.8291111111111114e-05,
+      "loss": 0.4664,
+      "mean_token_accuracy": 0.8545713942497969,
+      "num_tokens": 58997592.0,
+      "step": 1770
+    },
+    {
+      "entropy": 0.4840684751048684,
+      "epoch": 2.848,
+      "grad_norm": 1.4140625,
+      "learning_rate": 1.8268888888888888e-05,
+      "loss": 0.4668,
+      "mean_token_accuracy": 0.8554341107606888,
+      "num_tokens": 59326336.0,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4763260716572404,
+      "epoch": 2.864,
+      "grad_norm": 1.296875,
+      "learning_rate": 1.824666666666667e-05,
+      "loss": 0.4575,
+      "mean_token_accuracy": 0.858063718304038,
+      "num_tokens": 59668586.0,
+      "step": 1790
+    },
+    {
+      "entropy": 0.47250860314816234,
+      "epoch": 2.88,
+      "grad_norm": 1.6015625,
+      "learning_rate": 1.8224444444444447e-05,
+      "loss": 0.4591,
+      "mean_token_accuracy": 0.8573315985500812,
+      "num_tokens": 59993572.0,
+      "step": 1800
+    },
+    {
+      "epoch": 2.88,
+      "eval_chemistry_entropy": 0.5572835853099823,
+      "eval_chemistry_loss": 0.7282317876815796,
+      "eval_chemistry_mean_token_accuracy": 0.8056740365028381,
+      "eval_chemistry_num_tokens": 59993572.0,
+      "eval_chemistry_runtime": 34.6241,
+      "eval_chemistry_samples_per_second": 14.441,
+      "eval_chemistry_steps_per_second": 3.61,
+      "step": 1800
+    },
+    {
+      "epoch": 2.88,
+      "eval_physics_entropy": 0.5005052845478057,
+      "eval_physics_loss": 0.5933815240859985,
+      "eval_physics_mean_token_accuracy": 0.834332944393158,
+      "eval_physics_num_tokens": 59993572.0,
+      "eval_physics_runtime": 41.2544,
+      "eval_physics_samples_per_second": 12.12,
+      "eval_physics_steps_per_second": 3.03,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4770482426509261,
+      "epoch": 2.896,
+      "grad_norm": 1.5,
+      "learning_rate": 1.8202222222222225e-05,
+      "loss": 0.4576,
+      "mean_token_accuracy": 0.8566633488982915,
+      "num_tokens": 60324128.0,
+      "step": 1810
+    },
+    {
+      "entropy": 0.472668781876564,
+      "epoch": 2.912,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.8180000000000002e-05,
+      "loss": 0.4549,
+      "mean_token_accuracy": 0.858355012908578,
+      "num_tokens": 60657399.0,
+      "step": 1820
+    },
+    {
+      "entropy": 0.4690807094797492,
+      "epoch": 2.928,
+      "grad_norm": 1.3984375,
+      "learning_rate": 1.815777777777778e-05,
+      "loss": 0.4515,
+      "mean_token_accuracy": 0.8586832124739885,
+      "num_tokens": 60984711.0,
+      "step": 1830
+    },
+    {
+      "entropy": 0.4677178554236889,
+      "epoch": 2.944,
+      "grad_norm": 1.6015625,
+      "learning_rate": 1.8135555555555557e-05,
+      "loss": 0.4533,
+      "mean_token_accuracy": 0.858997655287385,
+      "num_tokens": 61321359.0,
+      "step": 1840
+    },
+    {
+      "entropy": 0.47983004804700613,
+      "epoch": 2.96,
+      "grad_norm": 1.3828125,
+      "learning_rate": 1.8113333333333335e-05,
+      "loss": 0.4607,
+      "mean_token_accuracy": 0.8569474756717682,
+      "num_tokens": 61659042.0,
+      "step": 1850
+    },
+    {
+      "entropy": 0.4755082868039608,
+      "epoch": 2.976,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.8091111111111113e-05,
+      "loss": 0.455,
+      "mean_token_accuracy": 0.8571343049407005,
+      "num_tokens": 61993595.0,
+      "step": 1860
+    },
+    {
+      "entropy": 0.46073452215641736,
+      "epoch": 2.992,
+      "grad_norm": 1.328125,
+      "learning_rate": 1.806888888888889e-05,
+      "loss": 0.4449,
+      "mean_token_accuracy": 0.8599927302449941,
+      "num_tokens": 62334012.0,
+      "step": 1870
+    },
+    {
+      "entropy": 0.41872623963281513,
+      "epoch": 3.008,
+      "grad_norm": 1.734375,
+      "learning_rate": 1.8046666666666668e-05,
+      "loss": 0.3776,
+      "mean_token_accuracy": 0.879940589517355,
+      "num_tokens": 62660928.0,
+      "step": 1880
+    },
+    {
+      "entropy": 0.3062732330523431,
+      "epoch": 3.024,
+      "grad_norm": 1.796875,
+      "learning_rate": 1.8024444444444445e-05,
+      "loss": 0.298,
+      "mean_token_accuracy": 0.9008689176291227,
+      "num_tokens": 62992670.0,
+      "step": 1890
+    },
+    {
+      "entropy": 0.314318940974772,
+      "epoch": 3.04,
+      "grad_norm": 1.890625,
+      "learning_rate": 1.8002222222222223e-05,
+      "loss": 0.2908,
+      "mean_token_accuracy": 0.9024943716824054,
+      "num_tokens": 63335157.0,
+      "step": 1900
+    },
+    {
+      "epoch": 3.04,
+      "eval_chemistry_entropy": 0.4083097501993179,
+      "eval_chemistry_loss": 0.8877620697021484,
+      "eval_chemistry_mean_token_accuracy": 0.795852870464325,
+      "eval_chemistry_num_tokens": 63335157.0,
+      "eval_chemistry_runtime": 34.5811,
+      "eval_chemistry_samples_per_second": 14.459,
+      "eval_chemistry_steps_per_second": 3.615,
+      "step": 1900
+    },
+    {
+      "epoch": 3.04,
+      "eval_physics_entropy": 0.356642371058464,
+      "eval_physics_loss": 0.7040289044380188,
+      "eval_physics_mean_token_accuracy": 0.8281279788017273,
+      "eval_physics_num_tokens": 63335157.0,
+      "eval_physics_runtime": 41.2473,
+      "eval_physics_samples_per_second": 12.122,
+      "eval_physics_steps_per_second": 3.031,
+      "step": 1900
+    },
+    {
+      "entropy": 0.3101253783330321,
+      "epoch": 3.056,
+      "grad_norm": 1.8671875,
+      "learning_rate": 1.798e-05,
+      "loss": 0.2855,
+      "mean_token_accuracy": 0.9037388786673546,
+      "num_tokens": 63662314.0,
+      "step": 1910
+    },
+    {
+      "entropy": 0.3101832845248282,
+      "epoch": 3.072,
+      "grad_norm": 1.9921875,
+      "learning_rate": 1.7957777777777778e-05,
+      "loss": 0.2901,
+      "mean_token_accuracy": 0.9020112063735723,
+      "num_tokens": 63997780.0,
+      "step": 1920
+    },
+    {
+      "entropy": 0.3178480817005038,
+      "epoch": 3.088,
+      "grad_norm": 1.921875,
+      "learning_rate": 1.7935555555555556e-05,
+      "loss": 0.2974,
+      "mean_token_accuracy": 0.8999546930193901,
+      "num_tokens": 64325274.0,
+      "step": 1930
+    },
+    {
+      "entropy": 0.30261183911934497,
+      "epoch": 3.104,
+      "grad_norm": 1.890625,
+      "learning_rate": 1.7913333333333337e-05,
+      "loss": 0.282,
+      "mean_token_accuracy": 0.9047445867210626,
+      "num_tokens": 64659683.0,
+      "step": 1940
+    },
+    {
+      "entropy": 0.29912759382277726,
+      "epoch": 3.12,
+      "grad_norm": 1.796875,
+      "learning_rate": 1.789111111111111e-05,
+      "loss": 0.2805,
+      "mean_token_accuracy": 0.9053699217736721,
+      "num_tokens": 64998212.0,
+      "step": 1950
+    },
+    {
+      "entropy": 0.31789954053238034,
+      "epoch": 3.136,
+      "grad_norm": 1.8671875,
+      "learning_rate": 1.7868888888888892e-05,
+      "loss": 0.2999,
+      "mean_token_accuracy": 0.8997865796089173,
+      "num_tokens": 65328436.0,
+      "step": 1960
+    },
+    {
+      "entropy": 0.3016470263712108,
+      "epoch": 3.152,
+      "grad_norm": 1.9453125,
+      "learning_rate": 1.7846666666666666e-05,
+      "loss": 0.281,
+      "mean_token_accuracy": 0.9046830009669066,
+      "num_tokens": 65659346.0,
+      "step": 1970
+    },
+    {
+      "entropy": 0.313287281896919,
+      "epoch": 3.168,
+      "grad_norm": 1.8359375,
+      "learning_rate": 1.7824444444444447e-05,
+      "loss": 0.2917,
+      "mean_token_accuracy": 0.9017079468816519,
+      "num_tokens": 65986382.0,
+      "step": 1980
+    },
+    {
+      "entropy": 0.3030808721669018,
+      "epoch": 3.184,
+      "grad_norm": 2.0,
+      "learning_rate": 1.780222222222222e-05,
+      "loss": 0.2839,
+      "mean_token_accuracy": 0.9039736580103636,
+      "num_tokens": 66324308.0,
+      "step": 1990
+    },
+    {
+      "entropy": 0.30991311026737095,
+      "epoch": 3.2,
+      "grad_norm": 1.8984375,
+      "learning_rate": 1.7780000000000003e-05,
+      "loss": 0.2896,
+      "mean_token_accuracy": 0.9023264229297638,
+      "num_tokens": 66658712.0,
+      "step": 2000
+    },
+    {
+      "epoch": 3.2,
+      "eval_chemistry_entropy": 0.3913348898887634,
+      "eval_chemistry_loss": 0.9072752594947815,
+      "eval_chemistry_mean_token_accuracy": 0.7967498378753662,
+      "eval_chemistry_num_tokens": 66658712.0,
+      "eval_chemistry_runtime": 34.6017,
+      "eval_chemistry_samples_per_second": 14.45,
+      "eval_chemistry_steps_per_second": 3.613,
+      "step": 2000
+    },
+    {
+      "epoch": 3.2,
+      "eval_physics_entropy": 0.3560866928100586,
+      "eval_physics_loss": 0.7074852585792542,
+      "eval_physics_mean_token_accuracy": 0.8281491875648499,
+      "eval_physics_num_tokens": 66658712.0,
+      "eval_physics_runtime": 40.9937,
+      "eval_physics_samples_per_second": 12.197,
+      "eval_physics_steps_per_second": 3.049,
+      "step": 2000
+    },
+    {
+      "entropy": 0.30841554757207634,
+      "epoch": 3.216,
+      "grad_norm": 1.796875,
+      "learning_rate": 1.7757777777777777e-05,
+      "loss": 0.2871,
+      "mean_token_accuracy": 0.9021381173282862,
+      "num_tokens": 66990142.0,
+      "step": 2010
+    },
+    {
+      "entropy": 0.3216819787397981,
+      "epoch": 3.232,
+      "grad_norm": 1.984375,
+      "learning_rate": 1.7735555555555558e-05,
+      "loss": 0.3005,
+      "mean_token_accuracy": 0.8979394420981407,
+      "num_tokens": 67323231.0,
+      "step": 2020
+    },
+    {
+      "entropy": 0.3005070075392723,
+      "epoch": 3.248,
+      "grad_norm": 2.09375,
+      "learning_rate": 1.7713333333333335e-05,
+      "loss": 0.2844,
+      "mean_token_accuracy": 0.9041368763893842,
+      "num_tokens": 67664253.0,
+      "step": 2030
+    },
+    {
+      "entropy": 0.31158956326544285,
+      "epoch": 3.2640000000000002,
+      "grad_norm": 2.109375,
+      "learning_rate": 1.7691111111111113e-05,
+      "loss": 0.2891,
+      "mean_token_accuracy": 0.9025689493864775,
+      "num_tokens": 67991082.0,
+      "step": 2040
+    },
+    {
+      "entropy": 0.308028197940439,
+      "epoch": 3.2800000000000002,
+      "grad_norm": 1.875,
+      "learning_rate": 1.766888888888889e-05,
+      "loss": 0.2913,
+      "mean_token_accuracy": 0.9015441410243511,
+      "num_tokens": 68317737.0,
+      "step": 2050
+    },
+    {
+      "entropy": 0.31388334231451154,
+      "epoch": 3.296,
+      "grad_norm": 1.890625,
+      "learning_rate": 1.7646666666666668e-05,
+      "loss": 0.2952,
+      "mean_token_accuracy": 0.9001540582627058,
+      "num_tokens": 68653995.0,
+      "step": 2060
+    },
+    {
+      "entropy": 0.3085142597556114,
+      "epoch": 3.312,
+      "grad_norm": 2.09375,
+      "learning_rate": 1.7624444444444446e-05,
+      "loss": 0.2916,
+      "mean_token_accuracy": 0.9023820515722036,
+      "num_tokens": 68994300.0,
+      "step": 2070
+    },
+    {
+      "entropy": 0.322191693354398,
+      "epoch": 3.328,
+      "grad_norm": 2.078125,
+      "learning_rate": 1.7602222222222223e-05,
+      "loss": 0.3004,
+      "mean_token_accuracy": 0.899126460403204,
+      "num_tokens": 69323456.0,
+      "step": 2080
+    },
+    {
+      "entropy": 0.31341270450502634,
+      "epoch": 3.344,
+      "grad_norm": 2.171875,
+      "learning_rate": 1.758e-05,
+      "loss": 0.2938,
+      "mean_token_accuracy": 0.9000593461096287,
+      "num_tokens": 69659373.0,
+      "step": 2090
+    },
+    {
+      "entropy": 0.32349290549755094,
+      "epoch": 3.36,
+      "grad_norm": 1.96875,
+      "learning_rate": 1.755777777777778e-05,
+      "loss": 0.3026,
+      "mean_token_accuracy": 0.8979135323315859,
+      "num_tokens": 69998761.0,
+      "step": 2100
+    },
+    {
+      "epoch": 3.36,
+      "eval_chemistry_entropy": 0.3982364737987518,
+      "eval_chemistry_loss": 0.9002445340156555,
+      "eval_chemistry_mean_token_accuracy": 0.7955028491020203,
+      "eval_chemistry_num_tokens": 69998761.0,
+      "eval_chemistry_runtime": 34.6288,
+      "eval_chemistry_samples_per_second": 14.439,
+      "eval_chemistry_steps_per_second": 3.61,
+      "step": 2100
+    },
+    {
+      "epoch": 3.36,
+      "eval_physics_entropy": 0.3635141521692276,
+      "eval_physics_loss": 0.6997887492179871,
+      "eval_physics_mean_token_accuracy": 0.8300124206542969,
+      "eval_physics_num_tokens": 69998761.0,
+      "eval_physics_runtime": 41.2511,
+      "eval_physics_samples_per_second": 12.121,
+      "eval_physics_steps_per_second": 3.03,
+      "step": 2100
+    },
+    {
+      "entropy": 0.3002217338420451,
+      "epoch": 3.376,
+      "grad_norm": 1.9296875,
+      "learning_rate": 1.7535555555555556e-05,
+      "loss": 0.2835,
+      "mean_token_accuracy": 0.9043725613504648,
+      "num_tokens": 70344663.0,
+      "step": 2110
+    },
+    {
+      "entropy": 0.3226173784583807,
+      "epoch": 3.392,
+      "grad_norm": 1.9609375,
+      "learning_rate": 1.7513333333333334e-05,
+      "loss": 0.3007,
+      "mean_token_accuracy": 0.8972969483584166,
+      "num_tokens": 70671470.0,
+      "step": 2120
+    },
+    {
+      "entropy": 0.31106266314163805,
+      "epoch": 3.408,
+      "grad_norm": 2.03125,
+      "learning_rate": 1.749111111111111e-05,
+      "loss": 0.2935,
+      "mean_token_accuracy": 0.9007665246725083,
+      "num_tokens": 71003257.0,
+      "step": 2130
+    },
+    {
+      "entropy": 0.3126481889747083,
+      "epoch": 3.424,
+      "grad_norm": 2.046875,
+      "learning_rate": 1.746888888888889e-05,
+      "loss": 0.2925,
+      "mean_token_accuracy": 0.9006055317819118,
+      "num_tokens": 71333695.0,
+      "step": 2140
+    },
+    {
+      "entropy": 0.31662706546485425,
+      "epoch": 3.44,
+      "grad_norm": 2.046875,
+      "learning_rate": 1.7446666666666667e-05,
+      "loss": 0.3004,
+      "mean_token_accuracy": 0.8997602794319391,
+      "num_tokens": 71672438.0,
+      "step": 2150
+    },
+    {
+      "entropy": 0.31781021589413283,
+      "epoch": 3.456,
+      "grad_norm": 1.9609375,
+      "learning_rate": 1.7424444444444444e-05,
+      "loss": 0.3002,
+      "mean_token_accuracy": 0.8984184488654137,
+      "num_tokens": 72009302.0,
+      "step": 2160
+    },
+    {
+      "entropy": 0.3236368427053094,
+      "epoch": 3.472,
+      "grad_norm": 2.03125,
+      "learning_rate": 1.7402222222222222e-05,
+      "loss": 0.3046,
+      "mean_token_accuracy": 0.8978909350931644,
+      "num_tokens": 72334603.0,
+      "step": 2170
+    },
+    {
+      "entropy": 0.31103177955374123,
+      "epoch": 3.488,
+      "grad_norm": 2.03125,
+      "learning_rate": 1.7380000000000003e-05,
+      "loss": 0.2936,
+      "mean_token_accuracy": 0.9011136863380671,
+      "num_tokens": 72669209.0,
+      "step": 2180
+    },
+    {
+      "entropy": 0.324208745919168,
+      "epoch": 3.504,
+      "grad_norm": 2.203125,
+      "learning_rate": 1.735777777777778e-05,
+      "loss": 0.306,
+      "mean_token_accuracy": 0.8976800374686718,
+      "num_tokens": 73001765.0,
+      "step": 2190
+    },
+    {
+      "entropy": 0.3198518265038729,
+      "epoch": 3.52,
+      "grad_norm": 2.109375,
+      "learning_rate": 1.7335555555555558e-05,
+      "loss": 0.3019,
+      "mean_token_accuracy": 0.8987439151853323,
+      "num_tokens": 73326681.0,
+      "step": 2200
+    },
+    {
+      "epoch": 3.52,
+      "eval_chemistry_entropy": 0.3984639751911163,
+      "eval_chemistry_loss": 0.908758282661438,
+      "eval_chemistry_mean_token_accuracy": 0.7953772258758545,
+      "eval_chemistry_num_tokens": 73326681.0,
+      "eval_chemistry_runtime": 34.6394,
+      "eval_chemistry_samples_per_second": 14.434,
+      "eval_chemistry_steps_per_second": 3.609,
+      "step": 2200
+    },
+    {
+      "epoch": 3.52,
+      "eval_physics_entropy": 0.36362937366962433,
+      "eval_physics_loss": 0.7028279900550842,
+      "eval_physics_mean_token_accuracy": 0.8286777052879334,
+      "eval_physics_num_tokens": 73326681.0,
+      "eval_physics_runtime": 41.2632,
+      "eval_physics_samples_per_second": 12.117,
+      "eval_physics_steps_per_second": 3.029,
+      "step": 2200
+    },
+    {
+      "entropy": 0.3174999676644802,
+      "epoch": 3.536,
+      "grad_norm": 1.9296875,
+      "learning_rate": 1.7313333333333336e-05,
+      "loss": 0.2998,
+      "mean_token_accuracy": 0.8993683304637671,
+      "num_tokens": 73658724.0,
+      "step": 2210
+    },
+    {
+      "entropy": 0.3130149595439434,
+      "epoch": 3.552,
+      "grad_norm": 1.90625,
+      "learning_rate": 1.7291111111111113e-05,
+      "loss": 0.293,
+      "mean_token_accuracy": 0.9002668939530849,
+      "num_tokens": 73990446.0,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3138899549841881,
+      "epoch": 3.568,
+      "grad_norm": 2.09375,
+      "learning_rate": 1.726888888888889e-05,
+      "loss": 0.2919,
+      "mean_token_accuracy": 0.8998954400420189,
+      "num_tokens": 74332932.0,
+      "step": 2230
+    },
+    {
+      "entropy": 0.3273557879962027,
+      "epoch": 3.584,
+      "grad_norm": 2.0,
+      "learning_rate": 1.724666666666667e-05,
+      "loss": 0.3092,
+      "mean_token_accuracy": 0.8954586464911699,
+      "num_tokens": 74660938.0,
+      "step": 2240
+    },
+    {
+      "entropy": 0.3111777361482382,
+      "epoch": 3.6,
+      "grad_norm": 2.125,
+      "learning_rate": 1.7224444444444446e-05,
+      "loss": 0.2927,
+      "mean_token_accuracy": 0.9013317011296749,
+      "num_tokens": 75004640.0,
+      "step": 2250
+    },
+    {
+      "entropy": 0.3160340915434062,
+      "epoch": 3.616,
+      "grad_norm": 2.03125,
+      "learning_rate": 1.7202222222222224e-05,
+      "loss": 0.3,
+      "mean_token_accuracy": 0.8990191236138344,
+      "num_tokens": 75344044.0,
+      "step": 2260
+    },
+    {
+      "entropy": 0.3194020535796881,
+      "epoch": 3.632,
+      "grad_norm": 1.9765625,
+      "learning_rate": 1.718e-05,
+      "loss": 0.3009,
+      "mean_token_accuracy": 0.8986658614128828,
+      "num_tokens": 75676495.0,
+      "step": 2270
+    },
+    {
+      "entropy": 0.321523059066385,
+      "epoch": 3.648,
+      "grad_norm": 1.9296875,
+      "learning_rate": 1.715777777777778e-05,
+      "loss": 0.3036,
+      "mean_token_accuracy": 0.8987799856811762,
+      "num_tokens": 76005637.0,
+      "step": 2280
+    },
+    {
+      "entropy": 0.31546984519809484,
+      "epoch": 3.664,
+      "grad_norm": 1.984375,
+      "learning_rate": 1.7135555555555557e-05,
+      "loss": 0.2967,
+      "mean_token_accuracy": 0.898678494617343,
+      "num_tokens": 76344141.0,
+      "step": 2290
+    },
+    {
+      "entropy": 0.32511516408994795,
+      "epoch": 3.68,
+      "grad_norm": 2.03125,
+      "learning_rate": 1.7113333333333334e-05,
+      "loss": 0.3053,
+      "mean_token_accuracy": 0.897456269338727,
+      "num_tokens": 76666670.0,
+      "step": 2300
+    },
+    {
+      "epoch": 3.68,
+      "eval_chemistry_entropy": 0.3999490444660187,
+      "eval_chemistry_loss": 0.905655026435852,
+      "eval_chemistry_mean_token_accuracy": 0.7958552212715149,
+      "eval_chemistry_num_tokens": 76666670.0,
+      "eval_chemistry_runtime": 34.6212,
+      "eval_chemistry_samples_per_second": 14.442,
+      "eval_chemistry_steps_per_second": 3.611,
+      "step": 2300
+    },
+    {
+      "epoch": 3.68,
+      "eval_physics_entropy": 0.36464127004146574,
+      "eval_physics_loss": 0.700202226638794,
+      "eval_physics_mean_token_accuracy": 0.8299637427330017,
+      "eval_physics_num_tokens": 76666670.0,
+      "eval_physics_runtime": 41.2617,
+      "eval_physics_samples_per_second": 12.118,
+      "eval_physics_steps_per_second": 3.029,
+      "step": 2300
+    },
+    {
+      "entropy": 0.32625418398529293,
+      "epoch": 3.6959999999999997,
+      "grad_norm": 2.25,
+      "learning_rate": 1.7091111111111112e-05,
+      "loss": 0.3093,
+      "mean_token_accuracy": 0.8957523014396429,
+      "num_tokens": 76997576.0,
+      "step": 2310
+    },
+    {
+      "entropy": 0.3156011073850095,
+      "epoch": 3.7119999999999997,
+      "grad_norm": 2.171875,
+      "learning_rate": 1.706888888888889e-05,
+      "loss": 0.2954,
+      "mean_token_accuracy": 0.9003329947590828,
+      "num_tokens": 77333230.0,
+      "step": 2320
+    },
+    {
+      "entropy": 0.3184890853241086,
+      "epoch": 3.7279999999999998,
+      "grad_norm": 2.046875,
+      "learning_rate": 1.704666666666667e-05,
+      "loss": 0.3026,
+      "mean_token_accuracy": 0.8980983804911375,
+      "num_tokens": 77662989.0,
+      "step": 2330
+    },
+    {
+      "entropy": 0.3237913983874023,
+      "epoch": 3.7439999999999998,
+      "grad_norm": 2.109375,
+      "learning_rate": 1.7024444444444445e-05,
+      "loss": 0.3056,
+      "mean_token_accuracy": 0.897695067897439,
+      "num_tokens": 77990559.0,
+      "step": 2340
+    },
+    {
+      "entropy": 0.3197625041939318,
+      "epoch": 3.76,
+      "grad_norm": 2.03125,
+      "learning_rate": 1.7002222222222226e-05,
+      "loss": 0.302,
+      "mean_token_accuracy": 0.8969969056546688,
+      "num_tokens": 78328058.0,
+      "step": 2350
+    },
+    {
+      "entropy": 0.32039122832939027,
+      "epoch": 3.776,
+      "grad_norm": 2.078125,
+      "learning_rate": 1.698e-05,
+      "loss": 0.2999,
+      "mean_token_accuracy": 0.8986205138266087,
+      "num_tokens": 78656796.0,
+      "step": 2360
+    },
+    {
+      "entropy": 0.31185208084061744,
+      "epoch": 3.792,
+      "grad_norm": 1.953125,
+      "learning_rate": 1.695777777777778e-05,
+      "loss": 0.296,
+      "mean_token_accuracy": 0.9003968022763729,
+      "num_tokens": 79000001.0,
+      "step": 2370
+    },
+    {
+      "entropy": 0.32015116736292837,
+      "epoch": 3.808,
+      "grad_norm": 1.9453125,
+      "learning_rate": 1.6935555555555555e-05,
+      "loss": 0.2999,
+      "mean_token_accuracy": 0.8986766755580902,
+      "num_tokens": 79332103.0,
+      "step": 2380
+    },
+    {
+      "entropy": 0.3285049434751272,
+      "epoch": 3.824,
+      "grad_norm": 2.109375,
+      "learning_rate": 1.6913333333333336e-05,
+      "loss": 0.3084,
+      "mean_token_accuracy": 0.8958335161209107,
+      "num_tokens": 79658616.0,
+      "step": 2390
+    },
+    {
+      "entropy": 0.3201352638192475,
+      "epoch": 3.84,
+      "grad_norm": 2.03125,
+      "learning_rate": 1.689111111111111e-05,
+      "loss": 0.3008,
+      "mean_token_accuracy": 0.8984422065317631,
+      "num_tokens": 79994326.0,
+      "step": 2400
+    },
+    {
+      "epoch": 3.84,
+      "eval_chemistry_entropy": 0.39785524785518644,
+      "eval_chemistry_loss": 0.9046195149421692,
+      "eval_chemistry_mean_token_accuracy": 0.796384777545929,
+      "eval_chemistry_num_tokens": 79994326.0,
+      "eval_chemistry_runtime": 34.628,
+      "eval_chemistry_samples_per_second": 14.439,
+      "eval_chemistry_steps_per_second": 3.61,
+      "step": 2400
+    },
+    {
+      "epoch": 3.84,
+      "eval_physics_entropy": 0.3649995505809784,
+      "eval_physics_loss": 0.6967271566390991,
+      "eval_physics_mean_token_accuracy": 0.8300955877304077,
+      "eval_physics_num_tokens": 79994326.0,
+      "eval_physics_runtime": 41.2661,
+      "eval_physics_samples_per_second": 12.116,
+      "eval_physics_steps_per_second": 3.029,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3189585231244564,
+      "epoch": 3.856,
+      "grad_norm": 1.953125,
+      "learning_rate": 1.686888888888889e-05,
+      "loss": 0.2981,
+      "mean_token_accuracy": 0.8985602565109729,
+      "num_tokens": 80332596.0,
+      "step": 2410
+    },
+    {
+      "entropy": 0.3192179761826992,
+      "epoch": 3.872,
+      "grad_norm": 2.15625,
+      "learning_rate": 1.684666666666667e-05,
+      "loss": 0.3008,
+      "mean_token_accuracy": 0.8992702454328537,
+      "num_tokens": 80662674.0,
+      "step": 2420
+    },
+    {
+      "entropy": 0.32309232754632833,
+      "epoch": 3.888,
+      "grad_norm": 2.140625,
+      "learning_rate": 1.6824444444444447e-05,
+      "loss": 0.3064,
+      "mean_token_accuracy": 0.8973185170441866,
+      "num_tokens": 80995198.0,
+      "step": 2430
+    },
+    {
+      "entropy": 0.3297643651254475,
+      "epoch": 3.904,
+      "grad_norm": 2.0625,
+      "learning_rate": 1.6802222222222224e-05,
+      "loss": 0.3095,
+      "mean_token_accuracy": 0.8950018011033535,
+      "num_tokens": 81331457.0,
+      "step": 2440
+    },
+    {
+      "entropy": 0.32107119625434277,
+      "epoch": 3.92,
+      "grad_norm": 2.09375,
+      "learning_rate": 1.6780000000000002e-05,
+      "loss": 0.3046,
+      "mean_token_accuracy": 0.8978569827973842,
+      "num_tokens": 81668354.0,
+      "step": 2450
+    },
+    {
+      "entropy": 0.31930080968886615,
+      "epoch": 3.936,
+      "grad_norm": 2.03125,
+      "learning_rate": 1.675777777777778e-05,
+      "loss": 0.3,
+      "mean_token_accuracy": 0.8981735832989216,
+      "num_tokens": 81994993.0,
+      "step": 2460
+    },
+    {
+      "entropy": 0.32174809416756034,
+      "epoch": 3.952,
+      "grad_norm": 2.171875,
+      "learning_rate": 1.6735555555555557e-05,
+      "loss": 0.303,
+      "mean_token_accuracy": 0.8970499508082866,
+      "num_tokens": 82327299.0,
+      "step": 2470
+    },
+    {
+      "entropy": 0.31509073711931707,
+      "epoch": 3.968,
+      "grad_norm": 2.21875,
+      "learning_rate": 1.6713333333333335e-05,
+      "loss": 0.296,
+      "mean_token_accuracy": 0.8995682530105114,
+      "num_tokens": 82653218.0,
+      "step": 2480
+    },
+    {
+      "entropy": 0.31415768414735795,
+      "epoch": 3.984,
+      "grad_norm": 2.046875,
+      "learning_rate": 1.6691111111111112e-05,
+      "loss": 0.2954,
+      "mean_token_accuracy": 0.9006588537245989,
+      "num_tokens": 82997429.0,
+      "step": 2490
+    },
+    {
+      "entropy": 0.3128982843831182,
+      "epoch": 4.0,
+      "grad_norm": 2.09375,
+      "learning_rate": 1.666888888888889e-05,
+      "loss": 0.2974,
+      "mean_token_accuracy": 0.8994481891393662,
+      "num_tokens": 83328592.0,
+      "step": 2500
+    },
+    {
+      "epoch": 4.0,
+      "eval_chemistry_entropy": 0.39885407757759095,
+      "eval_chemistry_loss": 0.9091149568557739,
+      "eval_chemistry_mean_token_accuracy": 0.795774167060852,
+      "eval_chemistry_num_tokens": 83328592.0,
+      "eval_chemistry_runtime": 34.6043,
+      "eval_chemistry_samples_per_second": 14.449,
+      "eval_chemistry_steps_per_second": 3.612,
+      "step": 2500
+    },
+    {
+      "epoch": 4.0,
+      "eval_physics_entropy": 0.36530122339725496,
+      "eval_physics_loss": 0.6973084211349487,
+      "eval_physics_mean_token_accuracy": 0.8300330719947815,
+      "eval_physics_num_tokens": 83328592.0,
+      "eval_physics_runtime": 40.9427,
+      "eval_physics_samples_per_second": 12.212,
+      "eval_physics_steps_per_second": 3.053,
+      "step": 2500
+    },
+    {
+      "entropy": 0.21543310643173755,
+      "epoch": 4.016,
+      "grad_norm": 2.34375,
+      "learning_rate": 1.6646666666666668e-05,
+      "loss": 0.162,
+      "mean_token_accuracy": 0.9484033465385437,
+      "num_tokens": 83655133.0,
+      "step": 2510
+    },
+    {
+      "entropy": 0.19175719134509564,
+      "epoch": 4.032,
+      "grad_norm": 1.9296875,
+      "learning_rate": 1.6624444444444445e-05,
+      "loss": 0.1519,
+      "mean_token_accuracy": 0.9507987700402737,
+      "num_tokens": 83995704.0,
+      "step": 2520
+    },
+    {
+      "entropy": 0.19031685492955147,
+      "epoch": 4.048,
+      "grad_norm": 1.921875,
+      "learning_rate": 1.6602222222222223e-05,
+      "loss": 0.1504,
+      "mean_token_accuracy": 0.9515091743320226,
+      "num_tokens": 84323657.0,
+      "step": 2530
+    },
+    {
+      "entropy": 0.18541114097461103,
+      "epoch": 4.064,
+      "grad_norm": 1.921875,
+      "learning_rate": 1.658e-05,
+      "loss": 0.1469,
+      "mean_token_accuracy": 0.9520056396722794,
+      "num_tokens": 84662055.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.18697019712999463,
+      "epoch": 4.08,
+      "grad_norm": 1.984375,
+      "learning_rate": 1.6557777777777778e-05,
+      "loss": 0.1469,
+      "mean_token_accuracy": 0.9523114804178476,
+      "num_tokens": 84990884.0,
+      "step": 2550
+    },
+    {
+      "entropy": 0.18290056772530078,
+      "epoch": 4.096,
+      "grad_norm": 2.03125,
+      "learning_rate": 1.6535555555555556e-05,
+      "loss": 0.1474,
+      "mean_token_accuracy": 0.9523672234266997,
+      "num_tokens": 85334346.0,
+      "step": 2560
+    },
+    {
+      "entropy": 0.18544434658251702,
+      "epoch": 4.112,
+      "grad_norm": 2.125,
+      "learning_rate": 1.6513333333333333e-05,
+      "loss": 0.1467,
+      "mean_token_accuracy": 0.9518146891146898,
+      "num_tokens": 85654431.0,
+      "step": 2570
+    },
+    {
+      "entropy": 0.18685742230154573,
+      "epoch": 4.128,
+      "grad_norm": 2.125,
+      "learning_rate": 1.6491111111111114e-05,
+      "loss": 0.148,
+      "mean_token_accuracy": 0.9512113470584154,
+      "num_tokens": 85984868.0,
+      "step": 2580
+    },
+    {
+      "entropy": 0.19103994900360705,
+      "epoch": 4.144,
+      "grad_norm": 1.9296875,
+      "learning_rate": 1.646888888888889e-05,
+      "loss": 0.1533,
+      "mean_token_accuracy": 0.9495586734265089,
+      "num_tokens": 86313610.0,
+      "step": 2590
+    },
+    {
+      "entropy": 0.1833164133131504,
+      "epoch": 4.16,
+      "grad_norm": 1.96875,
+      "learning_rate": 1.644666666666667e-05,
+      "loss": 0.1491,
+      "mean_token_accuracy": 0.950814313814044,
+      "num_tokens": 86648308.0,
+      "step": 2600
+    },
+    {
+      "epoch": 4.16,
+      "eval_chemistry_entropy": 0.3191299349069595,
+      "eval_chemistry_loss": 1.0713955163955688,
+      "eval_chemistry_mean_token_accuracy": 0.7893559455871582,
+      "eval_chemistry_num_tokens": 86648308.0,
+      "eval_chemistry_runtime": 34.5895,
+      "eval_chemistry_samples_per_second": 14.455,
+      "eval_chemistry_steps_per_second": 3.614,
+      "step": 2600
+    },
+    {
+      "epoch": 4.16,
+      "eval_physics_entropy": 0.26984936237335205,
+      "eval_physics_loss": 0.8553917407989502,
+      "eval_physics_mean_token_accuracy": 0.8235324311256409,
+      "eval_physics_num_tokens": 86648308.0,
+      "eval_physics_runtime": 41.2124,
+      "eval_physics_samples_per_second": 12.132,
+      "eval_physics_steps_per_second": 3.033,
+      "step": 2600
+    },
+    {
+      "entropy": 0.1851897793356329,
+      "epoch": 4.176,
+      "grad_norm": 2.125,
+      "learning_rate": 1.6424444444444444e-05,
+      "loss": 0.1499,
+      "mean_token_accuracy": 0.9512480426579714,
+      "num_tokens": 86979720.0,
+      "step": 2610
+    },
+    {
+      "entropy": 0.1831663493067026,
+      "epoch": 4.192,
+      "grad_norm": 1.984375,
+      "learning_rate": 1.6402222222222225e-05,
+      "loss": 0.1477,
+      "mean_token_accuracy": 0.9515604998916387,
+      "num_tokens": 87322940.0,
+      "step": 2620
+    },
+    {
+      "entropy": 0.18741083294153213,
+      "epoch": 4.208,
+      "grad_norm": 2.0,
+      "learning_rate": 1.638e-05,
+      "loss": 0.149,
+      "mean_token_accuracy": 0.9508396230638028,
+      "num_tokens": 87654476.0,
+      "step": 2630
+    },
+    {
+      "entropy": 0.19511286839842795,
+      "epoch": 4.224,
+      "grad_norm": 2.09375,
+      "learning_rate": 1.635777777777778e-05,
+      "loss": 0.1587,
+      "mean_token_accuracy": 0.9478027503937483,
+      "num_tokens": 87986180.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18864089637063444,
+      "epoch": 4.24,
+      "grad_norm": 2.09375,
+      "learning_rate": 1.6335555555555558e-05,
+      "loss": 0.153,
+      "mean_token_accuracy": 0.9495376665145159,
+      "num_tokens": 88322431.0,
+      "step": 2650
+    },
+    {
+      "entropy": 0.18742439663037658,
+      "epoch": 4.256,
+      "grad_norm": 2.0625,
+      "learning_rate": 1.6313333333333335e-05,
+      "loss": 0.1525,
+      "mean_token_accuracy": 0.9499348193407059,
+      "num_tokens": 88650585.0,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19175748783163726,
+      "epoch": 4.272,
+      "grad_norm": 2.390625,
+      "learning_rate": 1.6291111111111113e-05,
+      "loss": 0.1569,
+      "mean_token_accuracy": 0.9484076362103224,
+      "num_tokens": 88983258.0,
+      "step": 2670
+    },
+    {
+      "entropy": 0.1896573031321168,
+      "epoch": 4.288,
+      "grad_norm": 2.234375,
+      "learning_rate": 1.626888888888889e-05,
+      "loss": 0.1533,
+      "mean_token_accuracy": 0.949416034668684,
+      "num_tokens": 89306176.0,
+      "step": 2680
+    },
+    {
+      "entropy": 0.1905643475241959,
+      "epoch": 4.304,
+      "grad_norm": 2.203125,
+      "learning_rate": 1.6246666666666668e-05,
+      "loss": 0.1542,
+      "mean_token_accuracy": 0.949146980792284,
+      "num_tokens": 89635604.0,
+      "step": 2690
+    },
+    {
+      "entropy": 0.19005935504101218,
+      "epoch": 4.32,
+      "grad_norm": 2.140625,
+      "learning_rate": 1.6224444444444446e-05,
+      "loss": 0.1539,
+      "mean_token_accuracy": 0.9488909468054771,
+      "num_tokens": 89961895.0,
+      "step": 2700
+    },
+    {
+      "epoch": 4.32,
+      "eval_chemistry_entropy": 0.3175645558834076,
+      "eval_chemistry_loss": 1.0784482955932617,
+      "eval_chemistry_mean_token_accuracy": 0.789765251159668,
+      "eval_chemistry_num_tokens": 89961895.0,
+      "eval_chemistry_runtime": 34.6063,
+      "eval_chemistry_samples_per_second": 14.448,
+      "eval_chemistry_steps_per_second": 3.612,
+      "step": 2700
+    },
+    {
+      "epoch": 4.32,
+      "eval_physics_entropy": 0.2694563043117523,
+      "eval_physics_loss": 0.8580674529075623,
+      "eval_physics_mean_token_accuracy": 0.8243379302024841,
+      "eval_physics_num_tokens": 89961895.0,
+      "eval_physics_runtime": 41.2089,
+      "eval_physics_samples_per_second": 12.133,
+      "eval_physics_steps_per_second": 3.033,
+      "step": 2700
+    },
+    {
+      "entropy": 0.18589448984712362,
+      "epoch": 4.336,
+      "grad_norm": 1.9765625,
+      "learning_rate": 1.6202222222222223e-05,
+      "loss": 0.1502,
+      "mean_token_accuracy": 0.9505696080625057,
+      "num_tokens": 90293218.0,
+      "step": 2710
+    },
+    {
+      "entropy": 0.18639519005082547,
+      "epoch": 4.352,
+      "grad_norm": 2.03125,
+      "learning_rate": 1.618e-05,
+      "loss": 0.1536,
+      "mean_token_accuracy": 0.9487312618643046,
+      "num_tokens": 90631006.0,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19433019501157106,
+      "epoch": 4.368,
+      "grad_norm": 2.109375,
+      "learning_rate": 1.615777777777778e-05,
+      "loss": 0.1593,
+      "mean_token_accuracy": 0.9474144730716944,
+      "num_tokens": 90951742.0,
+      "step": 2730
+    },
+    {
+      "entropy": 0.1920853616669774,
+      "epoch": 4.384,
+      "grad_norm": 2.03125,
+      "learning_rate": 1.6135555555555556e-05,
+      "loss": 0.1574,
+      "mean_token_accuracy": 0.9479308255016804,
+      "num_tokens": 91290391.0,
+      "step": 2740
+    },
+    {
+      "entropy": 0.18902287022210656,
+      "epoch": 4.4,
+      "grad_norm": 2.015625,
+      "learning_rate": 1.6113333333333334e-05,
+      "loss": 0.1541,
+      "mean_token_accuracy": 0.9490096058696509,
+      "num_tokens": 91623212.0,
+      "step": 2750
+    },
+    {
+      "entropy": 0.1853318156208843,
+      "epoch": 4.416,
+      "grad_norm": 2.171875,
+      "learning_rate": 1.609111111111111e-05,
+      "loss": 0.1516,
+      "mean_token_accuracy": 0.9501545939594507,
+      "num_tokens": 91963726.0,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18632075604982673,
+      "epoch": 4.432,
+      "grad_norm": 2.046875,
+      "learning_rate": 1.606888888888889e-05,
+      "loss": 0.1533,
+      "mean_token_accuracy": 0.9498950578272343,
+      "num_tokens": 92295130.0,
+      "step": 2770
+    },
+    {
+      "entropy": 0.19441989213228225,
+      "epoch": 4.448,
+      "grad_norm": 2.1875,
+      "learning_rate": 1.6046666666666667e-05,
+      "loss": 0.1587,
+      "mean_token_accuracy": 0.9473429799079895,
+      "num_tokens": 92619914.0,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18935320070013403,
+      "epoch": 4.464,
+      "grad_norm": 2.109375,
+      "learning_rate": 1.6024444444444444e-05,
+      "loss": 0.1563,
+      "mean_token_accuracy": 0.9483810655772686,
+      "num_tokens": 92954334.0,
+      "step": 2790
+    },
+    {
+      "entropy": 0.1866440698504448,
+      "epoch": 4.48,
+      "grad_norm": 2.09375,
+      "learning_rate": 1.6002222222222222e-05,
+      "loss": 0.1545,
+      "mean_token_accuracy": 0.9490117151290178,
+      "num_tokens": 93288025.0,
+      "step": 2800
+    },
+    {
+      "epoch": 4.48,
+      "eval_chemistry_entropy": 0.32049827003479003,
+      "eval_chemistry_loss": 1.0742449760437012,
+      "eval_chemistry_mean_token_accuracy": 0.7895661110877991,
+      "eval_chemistry_num_tokens": 93288025.0,
+      "eval_chemistry_runtime": 34.6079,
+      "eval_chemistry_samples_per_second": 14.448,
+      "eval_chemistry_steps_per_second": 3.612,
+      "step": 2800
+    },
+    {
+      "epoch": 4.48,
+      "eval_physics_entropy": 0.272767622590065,
+      "eval_physics_loss": 0.8561501502990723,
+      "eval_physics_mean_token_accuracy": 0.8236871433258056,
+      "eval_physics_num_tokens": 93288025.0,
+      "eval_physics_runtime": 41.2289,
+      "eval_physics_samples_per_second": 12.127,
+      "eval_physics_steps_per_second": 3.032,
+      "step": 2800
+    },
+    {
+      "entropy": 0.19370509777218103,
+      "epoch": 4.496,
+      "grad_norm": 2.03125,
+      "learning_rate": 1.5980000000000003e-05,
+      "loss": 0.1597,
+      "mean_token_accuracy": 0.9468937519937753,
+      "num_tokens": 93621454.0,
+      "step": 2810
+    },
+    {
+      "entropy": 0.19159375540912152,
+      "epoch": 4.5120000000000005,
+      "grad_norm": 2.0625,
+      "learning_rate": 1.5957777777777777e-05,
+      "loss": 0.1566,
+      "mean_token_accuracy": 0.9483846995979548,
+      "num_tokens": 93957847.0,
+      "step": 2820
+    },
+    {
+      "entropy": 0.19177836943417786,
+      "epoch": 4.5280000000000005,
+      "grad_norm": 2.078125,
+      "learning_rate": 1.5935555555555558e-05,
+      "loss": 0.1595,
+      "mean_token_accuracy": 0.9476821962743998,
+      "num_tokens": 94287674.0,
+      "step": 2830
+    },
+    {
+      "entropy": 0.19004529607482254,
+      "epoch": 4.5440000000000005,
+      "grad_norm": 2.109375,
+      "learning_rate": 1.5913333333333332e-05,
+      "loss": 0.1565,
+      "mean_token_accuracy": 0.9476745925843716,
+      "num_tokens": 94613456.0,
+      "step": 2840
+    },
+    {
+      "entropy": 0.19111841106787325,
+      "epoch": 4.5600000000000005,
+      "grad_norm": 2.234375,
+      "learning_rate": 1.5891111111111113e-05,
+      "loss": 0.1586,
+      "mean_token_accuracy": 0.9473724879324437,
+      "num_tokens": 94941684.0,
+      "step": 2850
+    },
+    {
+      "entropy": 0.19352899109944702,
+      "epoch": 4.576,
+      "grad_norm": 2.234375,
+      "learning_rate": 1.5868888888888888e-05,
+      "loss": 0.1578,
+      "mean_token_accuracy": 0.9478868711739779,
+      "num_tokens": 95278037.0,
+      "step": 2860
+    },
+    {
+      "entropy": 0.1918213707394898,
+      "epoch": 4.592,
+      "grad_norm": 2.046875,
+      "learning_rate": 1.584666666666667e-05,
+      "loss": 0.1574,
+      "mean_token_accuracy": 0.9480335034430027,
+      "num_tokens": 95611182.0,
+      "step": 2870
+    },
+    {
+      "entropy": 0.19304955611005425,
+      "epoch": 4.608,
+      "grad_norm": 2.046875,
+      "learning_rate": 1.5824444444444446e-05,
+      "loss": 0.16,
+      "mean_token_accuracy": 0.9472247928380966,
+      "num_tokens": 95952986.0,
+      "step": 2880
+    },
+    {
+      "entropy": 0.18729796851985156,
+      "epoch": 4.624,
+      "grad_norm": 2.015625,
+      "learning_rate": 1.5802222222222224e-05,
+      "loss": 0.1564,
+      "mean_token_accuracy": 0.9481069557368755,
+      "num_tokens": 96297175.0,
+      "step": 2890
+    },
+    {
+      "entropy": 0.1974953768774867,
+      "epoch": 4.64,
+      "grad_norm": 2.15625,
+      "learning_rate": 1.578e-05,
+      "loss": 0.1634,
+      "mean_token_accuracy": 0.9463304281234741,
+      "num_tokens": 96630392.0,
+      "step": 2900
+    },
+    {
+      "epoch": 4.64,
+      "eval_chemistry_entropy": 0.319516619682312,
+      "eval_chemistry_loss": 1.0775052309036255,
+      "eval_chemistry_mean_token_accuracy": 0.7886076564788819,
+      "eval_chemistry_num_tokens": 96630392.0,
+      "eval_chemistry_runtime": 34.5969,
+      "eval_chemistry_samples_per_second": 14.452,
+      "eval_chemistry_steps_per_second": 3.613,
+      "step": 2900
+    },
+    {
+      "epoch": 4.64,
+      "eval_physics_entropy": 0.27287652349472047,
+      "eval_physics_loss": 0.8562257289886475,
+      "eval_physics_mean_token_accuracy": 0.8237788276672363,
+      "eval_physics_num_tokens": 96630392.0,
+      "eval_physics_runtime": 41.247,
+      "eval_physics_samples_per_second": 12.122,
+      "eval_physics_steps_per_second": 3.031,
+      "step": 2900
+    },
+    {
+      "entropy": 0.18674168265424668,
+      "epoch": 4.656,
+      "grad_norm": 2.078125,
+      "learning_rate": 1.575777777777778e-05,
+      "loss": 0.1539,
+      "mean_token_accuracy": 0.9486437644809484,
+      "num_tokens": 96978935.0,
+      "step": 2910
+    },
+    {
+      "entropy": 0.1915013926103711,
+      "epoch": 4.672,
+      "grad_norm": 2.03125,
+      "learning_rate": 1.5735555555555557e-05,
+      "loss": 0.1604,
+      "mean_token_accuracy": 0.947072671726346,
+      "num_tokens": 97316431.0,
+      "step": 2920
+    },
+    {
+      "entropy": 0.19037255314178764,
+      "epoch": 4.688,
+      "grad_norm": 2.125,
+      "learning_rate": 1.5713333333333334e-05,
+      "loss": 0.158,
+      "mean_token_accuracy": 0.9471198979765176,
+      "num_tokens": 97665534.0,
+      "step": 2930
+    },
+    {
+      "entropy": 0.19228512230329214,
+      "epoch": 4.704,
+      "grad_norm": 2.265625,
+      "learning_rate": 1.5691111111111112e-05,
+      "loss": 0.1588,
+      "mean_token_accuracy": 0.9474438637495041,
+      "num_tokens": 98004052.0,
+      "step": 2940
+    },
+    {
+      "entropy": 0.18908405187539756,
+      "epoch": 4.72,
+      "grad_norm": 2.234375,
+      "learning_rate": 1.5668888888888893e-05,
+      "loss": 0.1562,
+      "mean_token_accuracy": 0.9483963634818793,
+      "num_tokens": 98325785.0,
+      "step": 2950
+    },
+    {
+      "entropy": 0.19748488427139818,
+      "epoch": 4.736,
+      "grad_norm": 2.203125,
+      "learning_rate": 1.5646666666666667e-05,
+      "loss": 0.161,
+      "mean_token_accuracy": 0.9457533366978168,
+      "num_tokens": 98649969.0,
+      "step": 2960
+    },
+    {
+      "entropy": 0.18734354763291777,
+      "epoch": 4.752,
+      "grad_norm": 2.03125,
+      "learning_rate": 1.5624444444444448e-05,
+      "loss": 0.1563,
+      "mean_token_accuracy": 0.947586378827691,
+      "num_tokens": 98987145.0,
+      "step": 2970
+    },
+    {
+      "entropy": 0.19386245338246227,
+      "epoch": 4.768,
+      "grad_norm": 2.140625,
+      "learning_rate": 1.5602222222222222e-05,
+      "loss": 0.1603,
+      "mean_token_accuracy": 0.9473159741610289,
+      "num_tokens": 99309770.0,
+      "step": 2980
+    },
+    {
+      "entropy": 0.19166214000433685,
+      "epoch": 4.784,
+      "grad_norm": 2.140625,
+      "learning_rate": 1.5580000000000003e-05,
+      "loss": 0.1588,
+      "mean_token_accuracy": 0.9476859755814075,
+      "num_tokens": 99643583.0,
+      "step": 2990
+    },
+    {
+      "entropy": 0.19860149454325438,
+      "epoch": 4.8,
+      "grad_norm": 2.15625,
+      "learning_rate": 1.5557777777777778e-05,
+      "loss": 0.1616,
+      "mean_token_accuracy": 0.946180134639144,
+      "num_tokens": 99972302.0,
+      "step": 3000
+    },
+    {
+      "epoch": 4.8,
+      "eval_chemistry_entropy": 0.3164873479604721,
+      "eval_chemistry_loss": 1.07830011844635,
+      "eval_chemistry_mean_token_accuracy": 0.7890521488189697,
+      "eval_chemistry_num_tokens": 99972302.0,
+      "eval_chemistry_runtime": 34.601,
+      "eval_chemistry_samples_per_second": 14.45,
+      "eval_chemistry_steps_per_second": 3.613,
+      "step": 3000
+    },
+    {
+      "epoch": 4.8,
+      "eval_physics_entropy": 0.27059548449516296,
+      "eval_physics_loss": 0.8571956753730774,
+      "eval_physics_mean_token_accuracy": 0.824473412513733,
+      "eval_physics_num_tokens": 99972302.0,
+      "eval_physics_runtime": 40.9799,
+      "eval_physics_samples_per_second": 12.201,
+      "eval_physics_steps_per_second": 3.05,
+      "step": 3000
+    },
+    {
+      "entropy": 0.19665487641468643,
+      "epoch": 4.816,
+      "grad_norm": 2.34375,
+      "learning_rate": 1.553555555555556e-05,
+      "loss": 0.1624,
+      "mean_token_accuracy": 0.9465024966746569,
+      "num_tokens": 100290023.0,
+      "step": 3010
+    },
+    {
+      "entropy": 0.1841772203333676,
+      "epoch": 4.832,
+      "grad_norm": 2.140625,
+      "learning_rate": 1.5513333333333333e-05,
+      "loss": 0.1531,
+      "mean_token_accuracy": 0.9493068099021912,
+      "num_tokens": 100643312.0,
+      "step": 3020
+    },
+    {
+      "entropy": 0.1933573042973876,
+      "epoch": 4.848,
+      "grad_norm": 2.171875,
+      "learning_rate": 1.5491111111111114e-05,
+      "loss": 0.1605,
+      "mean_token_accuracy": 0.9469866078346968,
+      "num_tokens": 100977034.0,
+      "step": 3030
+    },
+    {
+      "entropy": 0.19200214128941298,
+      "epoch": 4.864,
+      "grad_norm": 2.171875,
+      "learning_rate": 1.546888888888889e-05,
+      "loss": 0.1614,
+      "mean_token_accuracy": 0.9468019433319569,
+      "num_tokens": 101309531.0,
+      "step": 3040
+    },
+    {
+      "entropy": 0.19307670569978655,
+      "epoch": 4.88,
+      "grad_norm": 2.1875,
+      "learning_rate": 1.544666666666667e-05,
+      "loss": 0.16,
+      "mean_token_accuracy": 0.9469937805086375,
+      "num_tokens": 101646955.0,
+      "step": 3050
+    },
+    {
+      "entropy": 0.19330994882620872,
+      "epoch": 4.896,
+      "grad_norm": 2.25,
+      "learning_rate": 1.5424444444444447e-05,
+      "loss": 0.1587,
+      "mean_token_accuracy": 0.9467738643288612,
+      "num_tokens": 101981516.0,
+      "step": 3060
+    },
+    {
+      "entropy": 0.19191964222118257,
+      "epoch": 4.912,
+      "grad_norm": 2.078125,
+      "learning_rate": 1.5402222222222224e-05,
+      "loss": 0.1593,
+      "mean_token_accuracy": 0.947215560078621,
+      "num_tokens": 102312681.0,
+      "step": 3070
+    },
+    {
+      "entropy": 0.19578345152549445,
+      "epoch": 4.928,
+      "grad_norm": 2.0625,
+      "learning_rate": 1.5380000000000002e-05,
+      "loss": 0.1605,
+      "mean_token_accuracy": 0.9471450641751289,
+      "num_tokens": 102647889.0,
+      "step": 3080
+    },
+    {
+      "entropy": 0.19374837134964765,
+      "epoch": 4.944,
+      "grad_norm": 2.234375,
+      "learning_rate": 1.535777777777778e-05,
+      "loss": 0.1607,
+      "mean_token_accuracy": 0.9465171311050653,
+      "num_tokens": 102985809.0,
+      "step": 3090
+    },
+    {
+      "entropy": 0.19707705564796923,
+      "epoch": 4.96,
+      "grad_norm": 2.171875,
+      "learning_rate": 1.5335555555555557e-05,
+      "loss": 0.1624,
+      "mean_token_accuracy": 0.945984173566103,
+      "num_tokens": 103320682.0,
+      "step": 3100
+    },
+    {
+      "epoch": 4.96,
+      "eval_chemistry_entropy": 0.3143026298284531,
+      "eval_chemistry_loss": 1.082382082939148,
+      "eval_chemistry_mean_token_accuracy": 0.7889351572990417,
+      "eval_chemistry_num_tokens": 103320682.0,
+      "eval_chemistry_runtime": 34.6247,
+      "eval_chemistry_samples_per_second": 14.441,
+      "eval_chemistry_steps_per_second": 3.61,
+      "step": 3100
+    },
+    {
+      "epoch": 4.96,
+      "eval_physics_entropy": 0.26906219792366026,
+      "eval_physics_loss": 0.8594797849655151,
+      "eval_physics_mean_token_accuracy": 0.8249925918579102,
+      "eval_physics_num_tokens": 103320682.0,
+      "eval_physics_runtime": 41.2605,
+      "eval_physics_samples_per_second": 12.118,
+      "eval_physics_steps_per_second": 3.03,
+      "step": 3100
+    },
+    {
+      "entropy": 0.19229637561365961,
+      "epoch": 4.976,
+      "grad_norm": 2.34375,
+      "learning_rate": 1.5313333333333335e-05,
+      "loss": 0.1597,
+      "mean_token_accuracy": 0.9467990398406982,
+      "num_tokens": 103656297.0,
+      "step": 3110
+    },
+    {
+      "entropy": 0.19407273214310408,
+      "epoch": 4.992,
+      "grad_norm": 2.1875,
+      "learning_rate": 1.5291111111111112e-05,
+      "loss": 0.1596,
+      "mean_token_accuracy": 0.9464813090860844,
+      "num_tokens": 103984906.0,
+      "step": 3120
+    },
+    {
+      "entropy": 0.1618450829293579,
+      "epoch": 5.008,
+      "grad_norm": 1.5,
+      "learning_rate": 1.526888888888889e-05,
+      "loss": 0.1172,
+      "mean_token_accuracy": 0.9638370562344789,
+      "num_tokens": 104329988.0,
+      "step": 3130
+    },
+    {
+      "entropy": 0.12690922296606005,
+      "epoch": 5.024,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.5246666666666668e-05,
+      "loss": 0.0774,
+      "mean_token_accuracy": 0.9784610513597727,
+      "num_tokens": 104670082.0,
+      "step": 3140
+    },
+    {
+      "entropy": 0.12201016019098461,
+      "epoch": 5.04,
+      "grad_norm": 1.5859375,
+      "learning_rate": 1.5224444444444447e-05,
+      "loss": 0.0739,
+      "mean_token_accuracy": 0.9795386526733637,
+      "num_tokens": 105012648.0,
+      "step": 3150
+    },
+    {
+      "entropy": 0.12360574728809297,
+      "epoch": 5.056,
+      "grad_norm": 1.5390625,
+      "learning_rate": 1.5202222222222223e-05,
+      "loss": 0.0748,
+      "mean_token_accuracy": 0.9788818847388029,
+      "num_tokens": 105340074.0,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12035191748291255,
+      "epoch": 5.072,
+      "grad_norm": 1.515625,
+      "learning_rate": 1.5180000000000002e-05,
+      "loss": 0.0751,
+      "mean_token_accuracy": 0.9789559379220009,
+      "num_tokens": 105674220.0,
+      "step": 3170
+    },
+    {
+      "entropy": 0.1201790071092546,
+      "epoch": 5.088,
+      "grad_norm": 1.5078125,
+      "learning_rate": 1.5157777777777778e-05,
+      "loss": 0.0737,
+      "mean_token_accuracy": 0.9795673862099648,
+      "num_tokens": 106008744.0,
+      "step": 3180
+    },
+    {
+      "entropy": 0.12344955923035741,
+      "epoch": 5.104,
+      "grad_norm": 1.6328125,
+      "learning_rate": 1.5135555555555557e-05,
+      "loss": 0.0774,
+      "mean_token_accuracy": 0.9782973855733872,
+      "num_tokens": 106332594.0,
+      "step": 3190
+    },
+    {
+      "entropy": 0.11950491964817048,
+      "epoch": 5.12,
+      "grad_norm": 1.359375,
+      "learning_rate": 1.5113333333333335e-05,
+      "loss": 0.0748,
+      "mean_token_accuracy": 0.9788820918649435,
+      "num_tokens": 106670394.0,
+      "step": 3200
+    },
+    {
+      "epoch": 5.12,
+      "eval_chemistry_entropy": 0.28638320648670196,
+      "eval_chemistry_loss": 1.1629633903503418,
+      "eval_chemistry_mean_token_accuracy": 0.7877100954055786,
+      "eval_chemistry_num_tokens": 106670394.0,
+      "eval_chemistry_runtime": 34.6316,
+      "eval_chemistry_samples_per_second": 14.438,
+      "eval_chemistry_steps_per_second": 3.609,
+      "step": 3200
+    },
+    {
+      "epoch": 5.12,
+      "eval_physics_entropy": 0.23361974203586577,
+      "eval_physics_loss": 0.9455986618995667,
+      "eval_physics_mean_token_accuracy": 0.8229552741050721,
+      "eval_physics_num_tokens": 106670394.0,
+      "eval_physics_runtime": 41.2677,
+      "eval_physics_samples_per_second": 12.116,
+      "eval_physics_steps_per_second": 3.029,
+      "step": 3200
+    },
+    {
+      "entropy": 0.1212450547143817,
+      "epoch": 5.136,
+      "grad_norm": 1.53125,
+      "learning_rate": 1.5091111111111113e-05,
+      "loss": 0.077,
+      "mean_token_accuracy": 0.9779245402663946,
+      "num_tokens": 107010814.0,
+      "step": 3210
+    },
+    {
+      "entropy": 0.12022993490099906,
+      "epoch": 5.152,
+      "grad_norm": 1.5703125,
+      "learning_rate": 1.506888888888889e-05,
+      "loss": 0.0769,
+      "mean_token_accuracy": 0.9780712764710188,
+      "num_tokens": 107353217.0,
+      "step": 3220
+    },
+    {
+      "entropy": 0.1167789971921593,
+      "epoch": 5.168,
+      "grad_norm": 1.5,
+      "learning_rate": 1.5046666666666668e-05,
+      "loss": 0.0738,
+      "mean_token_accuracy": 0.9790678706020117,
+      "num_tokens": 107704886.0,
+      "step": 3230
+    },
+    {
+      "entropy": 0.12345399712212383,
+      "epoch": 5.184,
+      "grad_norm": 1.609375,
+      "learning_rate": 1.5024444444444445e-05,
+      "loss": 0.0784,
+      "mean_token_accuracy": 0.97759250998497,
+      "num_tokens": 108033564.0,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12121607293374836,
+      "epoch": 5.2,
+      "grad_norm": 1.6953125,
+      "learning_rate": 1.5002222222222223e-05,
+      "loss": 0.0767,
+      "mean_token_accuracy": 0.9779817793518305,
+      "num_tokens": 108368498.0,
+      "step": 3250
+    },
+    {
+      "entropy": 0.12202885267324745,
+      "epoch": 5.216,
+      "grad_norm": 1.671875,
+      "learning_rate": 1.498e-05,
+      "loss": 0.0772,
+      "mean_token_accuracy": 0.9780106049031019,
+      "num_tokens": 108699224.0,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1182090365793556,
+      "epoch": 5.232,
+      "grad_norm": 1.6328125,
+      "learning_rate": 1.495777777777778e-05,
+      "loss": 0.0751,
+      "mean_token_accuracy": 0.9786998964846134,
+      "num_tokens": 109037108.0,
+      "step": 3270
+    },
+    {
+      "entropy": 0.12363715153187513,
+      "epoch": 5.248,
+      "grad_norm": 1.609375,
+      "learning_rate": 1.4935555555555556e-05,
+      "loss": 0.078,
+      "mean_token_accuracy": 0.9779166419059038,
+      "num_tokens": 109368040.0,
+      "step": 3280
+    },
+    {
+      "entropy": 0.12303452021442354,
+      "epoch": 5.264,
+      "grad_norm": 1.640625,
+      "learning_rate": 1.4913333333333335e-05,
+      "loss": 0.0791,
+      "mean_token_accuracy": 0.9772373173385859,
+      "num_tokens": 109691145.0,
+      "step": 3290
+    },
+    {
+      "entropy": 0.12271357667632402,
+      "epoch": 5.28,
+      "grad_norm": 1.609375,
+      "learning_rate": 1.4891111111111111e-05,
+      "loss": 0.0796,
+      "mean_token_accuracy": 0.9776995267719031,
+      "num_tokens": 110019099.0,
+      "step": 3300
+    },
+    {
+      "epoch": 5.28,
+      "eval_chemistry_entropy": 0.2859676060676575,
+      "eval_chemistry_loss": 1.1696257591247559,
+      "eval_chemistry_mean_token_accuracy": 0.7871607184410095,
+      "eval_chemistry_num_tokens": 110019099.0,
+      "eval_chemistry_runtime": 34.6219,
+      "eval_chemistry_samples_per_second": 14.442,
+      "eval_chemistry_steps_per_second": 3.61,
+      "step": 3300
+    },
+    {
+      "epoch": 5.28,
+      "eval_physics_entropy": 0.2324518562555313,
+      "eval_physics_loss": 0.951405942440033,
+      "eval_physics_mean_token_accuracy": 0.8229812092781067,
+      "eval_physics_num_tokens": 110019099.0,
+      "eval_physics_runtime": 41.2492,
+      "eval_physics_samples_per_second": 12.121,
+      "eval_physics_steps_per_second": 3.03,
+      "step": 3300
+    },
+    {
+      "entropy": 0.12270643962547183,
+      "epoch": 5.296,
+      "grad_norm": 1.5625,
+      "learning_rate": 1.486888888888889e-05,
+      "loss": 0.0785,
+      "mean_token_accuracy": 0.9774255253374576,
+      "num_tokens": 110349034.0,
+      "step": 3310
+    },
+    {
+      "entropy": 0.12224719040095806,
+      "epoch": 5.312,
+      "grad_norm": 1.5859375,
+      "learning_rate": 1.4846666666666666e-05,
+      "loss": 0.0786,
+      "mean_token_accuracy": 0.9772947859019041,
+      "num_tokens": 110678547.0,
+      "step": 3320
+    },
+    {
+      "entropy": 0.12140276823192835,
+      "epoch": 5.328,
+      "grad_norm": 1.703125,
+      "learning_rate": 1.4824444444444446e-05,
+      "loss": 0.0788,
+      "mean_token_accuracy": 0.9770833935588599,
+      "num_tokens": 111013689.0,
+      "step": 3330
+    },
+    {
+      "entropy": 0.12011926304548978,
+      "epoch": 5.344,
+      "grad_norm": 1.7109375,
+      "learning_rate": 1.4802222222222225e-05,
+      "loss": 0.0782,
+      "mean_token_accuracy": 0.9773753527551889,
+      "num_tokens": 111344270.0,
+      "step": 3340
+    },
+    {
+      "entropy": 0.12195792109705508,
+      "epoch": 5.36,
+      "grad_norm": 1.703125,
+      "learning_rate": 1.478e-05,
+      "loss": 0.078,
+      "mean_token_accuracy": 0.9773913998156786,
+      "num_tokens": 111686051.0,
+      "step": 3350
+    },
+    {
+      "entropy": 0.12005897578783334,
+      "epoch": 5.376,
+      "grad_norm": 1.6015625,
+      "learning_rate": 1.475777777777778e-05,
+      "loss": 0.0768,
+      "mean_token_accuracy": 0.9782460037618875,
+      "num_tokens": 112023874.0,
+      "step": 3360
+    },
+    {
+      "entropy": 0.1225117139518261,
+      "epoch": 5.392,
+      "grad_norm": 1.6328125,
+      "learning_rate": 1.4735555555555556e-05,
+      "loss": 0.0798,
+      "mean_token_accuracy": 0.9767647817730903,
+      "num_tokens": 112358202.0,
+      "step": 3370
+    },
+    {
+      "entropy": 0.12196989636868238,
+      "epoch": 5.408,
+      "grad_norm": 1.5703125,
+      "learning_rate": 1.4713333333333335e-05,
+      "loss": 0.0788,
+      "mean_token_accuracy": 0.977450056746602,
+      "num_tokens": 112692972.0,
+      "step": 3380
+    },
+    {
+      "entropy": 0.12126540918834508,
+      "epoch": 5.424,
+      "grad_norm": 1.65625,
+      "learning_rate": 1.4691111111111111e-05,
+      "loss": 0.0785,
+      "mean_token_accuracy": 0.9775171469897032,
+      "num_tokens": 113025416.0,
+      "step": 3390
+    },
+    {
+      "entropy": 0.12269472479820251,
+      "epoch": 5.44,
+      "grad_norm": 1.65625,
+      "learning_rate": 1.466888888888889e-05,
+      "loss": 0.0804,
+      "mean_token_accuracy": 0.9770185109227896,
+      "num_tokens": 113360632.0,
+      "step": 3400
+    },
+    {
+      "epoch": 5.44,
+      "eval_chemistry_entropy": 0.2852749445438385,
+      "eval_chemistry_loss": 1.1744590997695923,
+      "eval_chemistry_mean_token_accuracy": 0.7866754922866821,
+      "eval_chemistry_num_tokens": 113360632.0,
+      "eval_chemistry_runtime": 34.6138,
+      "eval_chemistry_samples_per_second": 14.445,
+      "eval_chemistry_steps_per_second": 3.611,
+      "step": 3400
+    },
+    {
+      "epoch": 5.44,
+      "eval_physics_entropy": 0.23213770461082459,
+      "eval_physics_loss": 0.9554653167724609,
+      "eval_physics_mean_token_accuracy": 0.8229207797050476,
+      "eval_physics_num_tokens": 113360632.0,
+      "eval_physics_runtime": 41.2537,
+      "eval_physics_samples_per_second": 12.12,
+      "eval_physics_steps_per_second": 3.03,
+      "step": 3400
+    },
+    {
+      "entropy": 0.12320682564750314,
+      "epoch": 5.456,
+      "grad_norm": 1.75,
+      "learning_rate": 1.4646666666666666e-05,
+      "loss": 0.0804,
+      "mean_token_accuracy": 0.9772241070866585,
+      "num_tokens": 113687612.0,
+      "step": 3410
+    },
+    {
+      "entropy": 0.12352340053766966,
+      "epoch": 5.4719999999999995,
+      "grad_norm": 1.6171875,
+      "learning_rate": 1.4624444444444446e-05,
+      "loss": 0.0805,
+      "mean_token_accuracy": 0.9766743324697018,
+      "num_tokens": 114012825.0,
+      "step": 3420
+    },
+    {
+      "entropy": 0.11838670265860855,
+      "epoch": 5.4879999999999995,
+      "grad_norm": 1.609375,
+      "learning_rate": 1.4602222222222225e-05,
+      "loss": 0.0772,
+      "mean_token_accuracy": 0.9773839592933655,
+      "num_tokens": 114352454.0,
+      "step": 3430
+    },
+    {
+      "entropy": 0.1236205083783716,
+      "epoch": 5.504,
+      "grad_norm": 1.609375,
+      "learning_rate": 1.4580000000000001e-05,
+      "loss": 0.08,
+      "mean_token_accuracy": 0.9772874113172293,
+      "num_tokens": 114676433.0,
+      "step": 3440
+    },
+    {
+      "entropy": 0.12238946189172566,
+      "epoch": 5.52,
+      "grad_norm": 1.5859375,
+      "learning_rate": 1.455777777777778e-05,
+      "loss": 0.0806,
+      "mean_token_accuracy": 0.9762930776923895,
+      "num_tokens": 115016974.0,
+      "step": 3450
+    },
+    {
+      "entropy": 0.12248858972452581,
+      "epoch": 5.536,
+      "grad_norm": 1.6328125,
+      "learning_rate": 1.4535555555555556e-05,
+      "loss": 0.0797,
+      "mean_token_accuracy": 0.9767976652830839,
+      "num_tokens": 115349329.0,
+      "step": 3460
+    },
+    {
+      "entropy": 0.11857948149554431,
+      "epoch": 5.552,
+      "grad_norm": 1.78125,
+      "learning_rate": 1.4513333333333336e-05,
+      "loss": 0.0777,
+      "mean_token_accuracy": 0.9774250064045191,
+      "num_tokens": 115685578.0,
+      "step": 3470
+    },
+    {
+      "entropy": 0.12281590728089213,
+      "epoch": 5.568,
+      "grad_norm": 1.6796875,
+      "learning_rate": 1.4491111111111111e-05,
+      "loss": 0.0804,
+      "mean_token_accuracy": 0.9769137635827064,
+      "num_tokens": 116015903.0,
+      "step": 3480
+    },
+    {
+      "entropy": 0.1202359089627862,
+      "epoch": 5.584,
+      "grad_norm": 1.578125,
+      "learning_rate": 1.446888888888889e-05,
+      "loss": 0.0789,
+      "mean_token_accuracy": 0.9770346198230981,
+      "num_tokens": 116353789.0,
+      "step": 3490
+    },
+    {
+      "entropy": 0.12375429016537964,
+      "epoch": 5.6,
+      "grad_norm": 1.65625,
+      "learning_rate": 1.4446666666666668e-05,
+      "loss": 0.0804,
+      "mean_token_accuracy": 0.9767476823180914,
+      "num_tokens": 116691024.0,
+      "step": 3500
+    },
+    {
+      "epoch": 5.6,
+      "eval_chemistry_entropy": 0.2811563332080841,
+      "eval_chemistry_loss": 1.1783264875411987,
+      "eval_chemistry_mean_token_accuracy": 0.7872345504760743,
+      "eval_chemistry_num_tokens": 116691024.0,
+      "eval_chemistry_runtime": 34.6135,
+      "eval_chemistry_samples_per_second": 14.445,
+      "eval_chemistry_steps_per_second": 3.611,
+      "step": 3500
+    },
+    {
+      "epoch": 5.6,
+      "eval_physics_entropy": 0.23010396003723144,
+      "eval_physics_loss": 0.9547271132469177,
+      "eval_physics_mean_token_accuracy": 0.8229746632575988,
+      "eval_physics_num_tokens": 116691024.0,
+      "eval_physics_runtime": 40.9813,
+      "eval_physics_samples_per_second": 12.201,
+      "eval_physics_steps_per_second": 3.05,
+      "step": 3500
+    },
+    {
+      "entropy": 0.12420583721250296,
+      "epoch": 5.616,
+      "grad_norm": 1.625,
+      "learning_rate": 1.4424444444444446e-05,
+      "loss": 0.0812,
+      "mean_token_accuracy": 0.9763885572552681,
+      "num_tokens": 117018997.0,
+      "step": 3510
+    },
+    {
+      "entropy": 0.12365589709952474,
+      "epoch": 5.632,
+      "grad_norm": 1.6796875,
+      "learning_rate": 1.4402222222222224e-05,
+      "loss": 0.0809,
+      "mean_token_accuracy": 0.9765663549304009,
+      "num_tokens": 117343698.0,
+      "step": 3520
+    },
+    {
+      "entropy": 0.12242367826402187,
+      "epoch": 5.648,
+      "grad_norm": 1.7265625,
+      "learning_rate": 1.4380000000000001e-05,
+      "loss": 0.0815,
+      "mean_token_accuracy": 0.9766273528337479,
+      "num_tokens": 117674808.0,
+      "step": 3530
+    },
+    {
+      "entropy": 0.12430465552024543,
+      "epoch": 5.664,
+      "grad_norm": 1.65625,
+      "learning_rate": 1.4357777777777779e-05,
+      "loss": 0.0812,
+      "mean_token_accuracy": 0.9766393750905991,
+      "num_tokens": 118006899.0,
+      "step": 3540
+    },
+    {
+      "entropy": 0.12397917686030269,
+      "epoch": 5.68,
+      "grad_norm": 1.703125,
+      "learning_rate": 1.4335555555555556e-05,
+      "loss": 0.0808,
+      "mean_token_accuracy": 0.9764606077224016,
+      "num_tokens": 118343335.0,
+      "step": 3550
+    },
+    {
+      "entropy": 0.11925184228457511,
+      "epoch": 5.696,
+      "grad_norm": 1.6484375,
+      "learning_rate": 1.4313333333333334e-05,
+      "loss": 0.0799,
+      "mean_token_accuracy": 0.9763355746865272,
+      "num_tokens": 118685884.0,
+      "step": 3560
+    },
+    {
+      "entropy": 0.12273379270918668,
+      "epoch": 5.712,
+      "grad_norm": 2.03125,
+      "learning_rate": 1.4291111111111113e-05,
+      "loss": 0.0812,
+      "mean_token_accuracy": 0.9767820600420236,
+      "num_tokens": 119021524.0,
+      "step": 3570
+    },
+    {
+      "entropy": 0.11689370246604085,
+      "epoch": 5.728,
+      "grad_norm": 1.625,
+      "learning_rate": 1.426888888888889e-05,
+      "loss": 0.0788,
+      "mean_token_accuracy": 0.9771351911127567,
+      "num_tokens": 119366016.0,
+      "step": 3580
+    },
+    {
+      "entropy": 0.12286252113990485,
+      "epoch": 5.744,
+      "grad_norm": 1.7421875,
+      "learning_rate": 1.4246666666666669e-05,
+      "loss": 0.081,
+      "mean_token_accuracy": 0.9763384524732828,
+      "num_tokens": 119704334.0,
+      "step": 3590
+    },
+    {
+      "entropy": 0.12401787112466992,
+      "epoch": 5.76,
+      "grad_norm": 1.625,
+      "learning_rate": 1.4224444444444445e-05,
+      "loss": 0.0825,
+      "mean_token_accuracy": 0.9759017691016197,
+      "num_tokens": 120034852.0,
+      "step": 3600
+    },
+    {
+      "epoch": 5.76,
+      "eval_chemistry_entropy": 0.2826125292778015,
+      "eval_chemistry_loss": 1.1768851280212402,
+      "eval_chemistry_mean_token_accuracy": 0.7866820750236512,
+      "eval_chemistry_num_tokens": 120034852.0,
+      "eval_chemistry_runtime": 34.6255,
+      "eval_chemistry_samples_per_second": 14.44,
+      "eval_chemistry_steps_per_second": 3.61,
+      "step": 3600
+    },
+    {
+      "epoch": 5.76,
+      "eval_physics_entropy": 0.23052174055576324,
+      "eval_physics_loss": 0.9576201438903809,
+      "eval_physics_mean_token_accuracy": 0.822891321182251,
+      "eval_physics_num_tokens": 120034852.0,
+      "eval_physics_runtime": 41.2187,
+      "eval_physics_samples_per_second": 12.13,
+      "eval_physics_steps_per_second": 3.033,
+      "step": 3600
+    },
+    {
+      "entropy": 0.1236722735222429,
+      "epoch": 5.776,
+      "grad_norm": 1.6015625,
+      "learning_rate": 1.4202222222222224e-05,
+      "loss": 0.081,
+      "mean_token_accuracy": 0.9761451706290245,
+      "num_tokens": 120364063.0,
+      "step": 3610
+    },
+    {
+      "entropy": 0.12318001296371221,
+      "epoch": 5.792,
+      "grad_norm": 1.640625,
+      "learning_rate": 1.418e-05,
+      "loss": 0.0817,
+      "mean_token_accuracy": 0.9760611515492201,
+      "num_tokens": 120702845.0,
+      "step": 3620
+    },
+    {
+      "entropy": 0.12329662996344268,
+      "epoch": 5.808,
+      "grad_norm": 1.640625,
+      "learning_rate": 1.4157777777777779e-05,
+      "loss": 0.081,
+      "mean_token_accuracy": 0.9769329752773046,
+      "num_tokens": 121030256.0,
+      "step": 3630
+    },
+    {
+      "entropy": 0.12317805914208294,
+      "epoch": 5.824,
+      "grad_norm": 1.6328125,
+      "learning_rate": 1.4135555555555555e-05,
+      "loss": 0.081,
+      "mean_token_accuracy": 0.9766381312161684,
+      "num_tokens": 121363411.0,
+      "step": 3640
+    },
+    {
+      "entropy": 0.1255562113597989,
+      "epoch": 5.84,
+      "grad_norm": 1.7109375,
+      "learning_rate": 1.4113333333333334e-05,
+      "loss": 0.0826,
+      "mean_token_accuracy": 0.9760429706424475,
+      "num_tokens": 121683493.0,
+      "step": 3650
+    },
+    {
+      "entropy": 0.12333199586719275,
+      "epoch": 5.856,
+      "grad_norm": 1.609375,
+      "learning_rate": 1.4091111111111114e-05,
+      "loss": 0.0807,
+      "mean_token_accuracy": 0.9761915192008018,
+      "num_tokens": 122013880.0,
+      "step": 3660
+    },
+    {
+      "entropy": 0.1253392260055989,
+      "epoch": 5.872,
+      "grad_norm": 1.75,
+      "learning_rate": 1.406888888888889e-05,
+      "loss": 0.0829,
+      "mean_token_accuracy": 0.9759009908884764,
+      "num_tokens": 122347894.0,
+      "step": 3670
+    },
+    {
+      "entropy": 0.12091331197880209,
+      "epoch": 5.888,
+      "grad_norm": 1.6953125,
+      "learning_rate": 1.4046666666666669e-05,
+      "loss": 0.0806,
+      "mean_token_accuracy": 0.976499255746603,
+      "num_tokens": 122684941.0,
+      "step": 3680
+    },
+    {
+      "entropy": 0.12405687246937305,
+      "epoch": 5.904,
+      "grad_norm": 4.28125,
+      "learning_rate": 1.4024444444444445e-05,
+      "loss": 0.0825,
+      "mean_token_accuracy": 0.9768069617450237,
+      "num_tokens": 123020602.0,
+      "step": 3690
+    },
+    {
+      "entropy": 0.12095803911797702,
+      "epoch": 5.92,
+      "grad_norm": 1.6796875,
+      "learning_rate": 1.4002222222222224e-05,
+      "loss": 0.08,
+      "mean_token_accuracy": 0.9767821855843067,
+      "num_tokens": 123357158.0,
+      "step": 3700
+    },
+    {
+      "epoch": 5.92,
+      "eval_chemistry_entropy": 0.2803421527147293,
+      "eval_chemistry_loss": 1.1812224388122559,
+      "eval_chemistry_mean_token_accuracy": 0.7871488614082336,
+      "eval_chemistry_num_tokens": 123357158.0,
+      "eval_chemistry_runtime": 34.5957,
+      "eval_chemistry_samples_per_second": 14.453,
+      "eval_chemistry_steps_per_second": 3.613,
+      "step": 3700
+    },
+    {
+      "epoch": 5.92,
+      "eval_physics_entropy": 0.22845887267589568,
+      "eval_physics_loss": 0.9604062438011169,
+      "eval_physics_mean_token_accuracy": 0.8229416389465332,
+      "eval_physics_num_tokens": 123357158.0,
+      "eval_physics_runtime": 41.2178,
+      "eval_physics_samples_per_second": 12.131,
+      "eval_physics_steps_per_second": 3.033,
+      "step": 3700
+    },
+    {
+      "entropy": 0.12068014587275684,
+      "epoch": 5.936,
+      "grad_norm": 1.625,
+      "learning_rate": 1.398e-05,
+      "loss": 0.0805,
+      "mean_token_accuracy": 0.9766017343848944,
+      "num_tokens": 123695748.0,
+      "step": 3710
+    },
+    {
+      "entropy": 0.12223677332513035,
+      "epoch": 5.952,
+      "grad_norm": 1.6328125,
+      "learning_rate": 1.395777777777778e-05,
+      "loss": 0.0796,
+      "mean_token_accuracy": 0.9768834766000509,
+      "num_tokens": 124026940.0,
+      "step": 3720
+    },
+    {
+      "entropy": 0.12707342663779855,
+      "epoch": 5.968,
+      "grad_norm": 1.734375,
+      "learning_rate": 1.3935555555555557e-05,
+      "loss": 0.083,
+      "mean_token_accuracy": 0.9756046898663044,
+      "num_tokens": 124350349.0,
+      "step": 3730
+    },
+    {
+      "entropy": 0.12378147379495204,
+      "epoch": 5.984,
+      "grad_norm": 1.734375,
+      "learning_rate": 1.3913333333333335e-05,
+      "loss": 0.0822,
+      "mean_token_accuracy": 0.9760129202157259,
+      "num_tokens": 124678890.0,
+      "step": 3740
+    },
+    {
+      "entropy": 0.12643266683444382,
+      "epoch": 6.0,
+      "grad_norm": 1.7265625,
+      "learning_rate": 1.3891111111111114e-05,
+      "loss": 0.0837,
+      "mean_token_accuracy": 0.9758120037615299,
+      "num_tokens": 124992888.0,
+      "step": 3750
+    },
+    {
+      "entropy": 0.08800205194856972,
+      "epoch": 6.016,
+      "grad_norm": 1.40625,
+      "learning_rate": 1.386888888888889e-05,
+      "loss": 0.0465,
+      "mean_token_accuracy": 0.9884943414479495,
+      "num_tokens": 125323982.0,
+      "step": 3760
+    },
+    {
+      "entropy": 0.08312874312978238,
+      "epoch": 6.032,
+      "grad_norm": 1.1953125,
+      "learning_rate": 1.3846666666666669e-05,
+      "loss": 0.0453,
+      "mean_token_accuracy": 0.9885493285953999,
+      "num_tokens": 125666740.0,
+      "step": 3770
+    },
+    {
+      "entropy": 0.0828782632946968,
+      "epoch": 6.048,
+      "grad_norm": 1.125,
+      "learning_rate": 1.3824444444444445e-05,
+      "loss": 0.0459,
+      "mean_token_accuracy": 0.988551089912653,
+      "num_tokens": 125993258.0,
+      "step": 3780
+    },
+    {
+      "entropy": 0.08486181979533285,
+      "epoch": 6.064,
+      "grad_norm": 0.96875,
+      "learning_rate": 1.3802222222222224e-05,
+      "loss": 0.0458,
+      "mean_token_accuracy": 0.9883942436426878,
+      "num_tokens": 126311534.0,
+      "step": 3790
+    },
+    {
+      "entropy": 0.08183146067894995,
+      "epoch": 6.08,
+      "grad_norm": 1.09375,
+      "learning_rate": 1.378e-05,
+      "loss": 0.0462,
+      "mean_token_accuracy": 0.9883392293006181,
+      "num_tokens": 126642761.0,
+      "step": 3800
+    },
+    {
+      "epoch": 6.08,
+      "eval_chemistry_entropy": 0.2633392469882965,
+      "eval_chemistry_loss": 1.2417505979537964,
+      "eval_chemistry_mean_token_accuracy": 0.7880106172561645,
+      "eval_chemistry_num_tokens": 126642761.0,
+      "eval_chemistry_runtime": 34.6658,
+      "eval_chemistry_samples_per_second": 14.423,
+      "eval_chemistry_steps_per_second": 3.606,
+      "step": 3800
+    },
+    {
+      "epoch": 6.08,
+      "eval_physics_entropy": 0.20876979553699493,
+      "eval_physics_loss": 1.0224370956420898,
+      "eval_physics_mean_token_accuracy": 0.8232405352592468,
+      "eval_physics_num_tokens": 126642761.0,
+      "eval_physics_runtime": 41.3258,
+      "eval_physics_samples_per_second": 12.099,
+      "eval_physics_steps_per_second": 3.025,
+      "step": 3800
+    },
+    {
+      "entropy": 0.08224272588267922,
+      "epoch": 6.096,
+      "grad_norm": 1.03125,
+      "learning_rate": 1.375777777777778e-05,
+      "loss": 0.0457,
+      "mean_token_accuracy": 0.9883229065686464,
+      "num_tokens": 126976948.0,
+      "step": 3810
+    },
+    {
+      "entropy": 0.08190394197590649,
+      "epoch": 6.112,
+      "grad_norm": 1.1796875,
+      "learning_rate": 1.3735555555555557e-05,
+      "loss": 0.0467,
+      "mean_token_accuracy": 0.9879281643778086,
+      "num_tokens": 127314551.0,
+      "step": 3820
+    },
+    {
+      "entropy": 0.08483500238507986,
+      "epoch": 6.128,
+      "grad_norm": 1.234375,
+      "learning_rate": 1.3713333333333335e-05,
+      "loss": 0.0474,
+      "mean_token_accuracy": 0.9877087824046612,
+      "num_tokens": 127635496.0,
+      "step": 3830
+    },
+    {
+      "entropy": 0.08074533776380122,
+      "epoch": 6.144,
+      "grad_norm": 1.140625,
+      "learning_rate": 1.3691111111111112e-05,
+      "loss": 0.0455,
+      "mean_token_accuracy": 0.9884730204939842,
+      "num_tokens": 127974017.0,
+      "step": 3840
+    },
+    {
+      "entropy": 0.08340199454687536,
+      "epoch": 6.16,
+      "grad_norm": 1.1015625,
+      "learning_rate": 1.366888888888889e-05,
+      "loss": 0.0471,
+      "mean_token_accuracy": 0.9877915620803833,
+      "num_tokens": 128303651.0,
+      "step": 3850
+    },
+    {
+      "entropy": 0.08018718773964792,
+      "epoch": 6.176,
+      "grad_norm": 1.1953125,
+      "learning_rate": 1.3646666666666668e-05,
+      "loss": 0.0462,
+      "mean_token_accuracy": 0.9882023938000202,
+      "num_tokens": 128649197.0,
+      "step": 3860
+    },
+    {
+      "entropy": 0.08327068700455129,
+      "epoch": 6.192,
+      "grad_norm": 1.171875,
+      "learning_rate": 1.3624444444444445e-05,
+      "loss": 0.0472,
+      "mean_token_accuracy": 0.988257608562708,
+      "num_tokens": 128986105.0,
+      "step": 3870
+    },
+    {
+      "entropy": 0.08418330932036042,
+      "epoch": 6.208,
+      "grad_norm": 1.21875,
+      "learning_rate": 1.3602222222222223e-05,
+      "loss": 0.0484,
+      "mean_token_accuracy": 0.9875238493084908,
+      "num_tokens": 129315037.0,
+      "step": 3880
+    },
+    {
+      "entropy": 0.08170841236133128,
+      "epoch": 6.224,
+      "grad_norm": 1.0703125,
+      "learning_rate": 1.3580000000000002e-05,
+      "loss": 0.0464,
+      "mean_token_accuracy": 0.9880779847502709,
+      "num_tokens": 129648083.0,
+      "step": 3890
+    },
+    {
+      "entropy": 0.0832998547470197,
+      "epoch": 6.24,
+      "grad_norm": 1.1640625,
+      "learning_rate": 1.3557777777777778e-05,
+      "loss": 0.0474,
+      "mean_token_accuracy": 0.9878678418695926,
+      "num_tokens": 129981253.0,
+      "step": 3900
+    },
+    {
+      "epoch": 6.24,
+      "eval_chemistry_entropy": 0.26424161648750305,
+      "eval_chemistry_loss": 1.2465729713439941,
+      "eval_chemistry_mean_token_accuracy": 0.786607096195221,
+      "eval_chemistry_num_tokens": 129981253.0,
+      "eval_chemistry_runtime": 34.6941,
+      "eval_chemistry_samples_per_second": 14.412,
+      "eval_chemistry_steps_per_second": 3.603,
+      "step": 3900
+    },
+    {
+      "epoch": 6.24,
+      "eval_physics_entropy": 0.20829280495643615,
+      "eval_physics_loss": 1.0281598567962646,
+      "eval_physics_mean_token_accuracy": 0.8229806046485901,
+      "eval_physics_num_tokens": 129981253.0,
+      "eval_physics_runtime": 41.7122,
+      "eval_physics_samples_per_second": 11.987,
+      "eval_physics_steps_per_second": 2.997,
+      "step": 3900
+    },
+    {
+      "entropy": 0.08336418468970805,
+      "epoch": 6.256,
+      "grad_norm": 1.2109375,
+      "learning_rate": 1.3535555555555557e-05,
+      "loss": 0.0482,
+      "mean_token_accuracy": 0.9876391071826219,
+      "num_tokens": 130318318.0,
+      "step": 3910
+    },
+    {
+      "entropy": 0.0824885551352054,
+      "epoch": 6.272,
+      "grad_norm": 1.15625,
+      "learning_rate": 1.3513333333333333e-05,
+      "loss": 0.0467,
+      "mean_token_accuracy": 0.987917210161686,
+      "num_tokens": 130653933.0,
+      "step": 3920
+    },
+    {
+      "entropy": 0.08328301263973117,
+      "epoch": 6.288,
+      "grad_norm": 1.1953125,
+      "learning_rate": 1.3491111111111113e-05,
+      "loss": 0.0477,
+      "mean_token_accuracy": 0.9875371098518372,
+      "num_tokens": 130990699.0,
+      "step": 3930
+    },
+    {
+      "entropy": 0.08321071907412261,
+      "epoch": 6.304,
+      "grad_norm": 1.109375,
+      "learning_rate": 1.3468888888888888e-05,
+      "loss": 0.048,
+      "mean_token_accuracy": 0.9875328954309225,
+      "num_tokens": 131322534.0,
+      "step": 3940
+    },
+    {
+      "entropy": 0.08432359651196748,
+      "epoch": 6.32,
+      "grad_norm": 1.1640625,
+      "learning_rate": 1.3446666666666668e-05,
+      "loss": 0.0484,
+      "mean_token_accuracy": 0.9873499438166619,
+      "num_tokens": 131650063.0,
+      "step": 3950
+    },
+    {
+      "entropy": 0.0830828903010115,
+      "epoch": 6.336,
+      "grad_norm": 1.28125,
+      "learning_rate": 1.3424444444444447e-05,
+      "loss": 0.0477,
+      "mean_token_accuracy": 0.9876048494130373,
+      "num_tokens": 131985745.0,
+      "step": 3960
+    },
+    {
+      "entropy": 0.08442043093964458,
+      "epoch": 6.352,
+      "grad_norm": 1.1015625,
+      "learning_rate": 1.3402222222222223e-05,
+      "loss": 0.0486,
+      "mean_token_accuracy": 0.9874293334782124,
+      "num_tokens": 132319846.0,
+      "step": 3970
+    },
+    {
+      "entropy": 0.08410298686940224,
+      "epoch": 6.368,
+      "grad_norm": 1.2578125,
+      "learning_rate": 1.3380000000000002e-05,
+      "loss": 0.0482,
+      "mean_token_accuracy": 0.987572344392538,
+      "num_tokens": 132656725.0,
+      "step": 3980
+    },
+    {
+      "entropy": 0.08180661161895841,
+      "epoch": 6.384,
+      "grad_norm": 1.0234375,
+      "learning_rate": 1.3357777777777778e-05,
+      "loss": 0.0466,
+      "mean_token_accuracy": 0.9879112608730793,
+      "num_tokens": 132995278.0,
+      "step": 3990
+    },
+    {
+      "entropy": 0.08174596836324781,
+      "epoch": 6.4,
+      "grad_norm": 1.0078125,
+      "learning_rate": 1.3335555555555558e-05,
+      "loss": 0.0481,
+      "mean_token_accuracy": 0.9874971948564053,
+      "num_tokens": 133334112.0,
+      "step": 4000
+    },
+    {
+      "epoch": 6.4,
+      "eval_chemistry_entropy": 0.2629435861110687,
+      "eval_chemistry_loss": 1.2468456029891968,
+      "eval_chemistry_mean_token_accuracy": 0.7867292041778564,
+      "eval_chemistry_num_tokens": 133334112.0,
+      "eval_chemistry_runtime": 34.6344,
+      "eval_chemistry_samples_per_second": 14.436,
+      "eval_chemistry_steps_per_second": 3.609,
+      "step": 4000
+    },
+    {
+      "epoch": 6.4,
+      "eval_physics_entropy": 0.20778464794158935,
+      "eval_physics_loss": 1.0285807847976685,
+      "eval_physics_mean_token_accuracy": 0.8228671875,
+      "eval_physics_num_tokens": 133334112.0,
+      "eval_physics_runtime": 41.022,
+      "eval_physics_samples_per_second": 12.189,
+      "eval_physics_steps_per_second": 3.047,
+      "step": 4000
+    },
+    {
+      "entropy": 0.08247077025007457,
+      "epoch": 6.416,
+      "grad_norm": 1.1953125,
+      "learning_rate": 1.3313333333333333e-05,
+      "loss": 0.0479,
+      "mean_token_accuracy": 0.9875911708921195,
+      "num_tokens": 133674943.0,
+      "step": 4010
+    },
+    {
+      "entropy": 0.0861233185045421,
+      "epoch": 6.432,
+      "grad_norm": 1.4375,
+      "learning_rate": 1.3291111111111113e-05,
+      "loss": 0.0493,
+      "mean_token_accuracy": 0.9870279934257269,
+      "num_tokens": 134002453.0,
+      "step": 4020
+    },
+    {
+      "entropy": 0.08455703966319561,
+      "epoch": 6.448,
+      "grad_norm": 1.1640625,
+      "learning_rate": 1.3268888888888889e-05,
+      "loss": 0.0482,
+      "mean_token_accuracy": 0.9875376716256141,
+      "num_tokens": 134323746.0,
+      "step": 4030
+    },
+    {
+      "entropy": 0.0815651074750349,
+      "epoch": 6.464,
+      "grad_norm": 1.140625,
+      "learning_rate": 1.3246666666666668e-05,
+      "loss": 0.0478,
+      "mean_token_accuracy": 0.987864351272583,
+      "num_tokens": 134661738.0,
+      "step": 4040
+    },
+    {
+      "entropy": 0.08226230149157346,
+      "epoch": 6.48,
+      "grad_norm": 1.1953125,
+      "learning_rate": 1.3224444444444446e-05,
+      "loss": 0.0477,
+      "mean_token_accuracy": 0.9879491850733757,
+      "num_tokens": 135000741.0,
+      "step": 4050
+    },
+    {
+      "entropy": 0.08494914614129812,
+      "epoch": 6.496,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.3202222222222223e-05,
+      "loss": 0.0502,
+      "mean_token_accuracy": 0.9867266416549683,
+      "num_tokens": 135326552.0,
+      "step": 4060
+    },
+    {
+      "entropy": 0.08172757881693542,
+      "epoch": 6.5120000000000005,
+      "grad_norm": 1.359375,
+      "learning_rate": 1.3180000000000001e-05,
+      "loss": 0.0471,
+      "mean_token_accuracy": 0.987838089466095,
+      "num_tokens": 135671893.0,
+      "step": 4070
+    },
+    {
+      "entropy": 0.08116255044005811,
+      "epoch": 6.5280000000000005,
+      "grad_norm": 1.234375,
+      "learning_rate": 1.3157777777777778e-05,
+      "loss": 0.0479,
+      "mean_token_accuracy": 0.9872550044208765,
+      "num_tokens": 136011918.0,
+      "step": 4080
+    },
+    {
+      "entropy": 0.08455753971356898,
+      "epoch": 6.5440000000000005,
+      "grad_norm": 1.21875,
+      "learning_rate": 1.3135555555555558e-05,
+      "loss": 0.0493,
+      "mean_token_accuracy": 0.9871864940971136,
+      "num_tokens": 136346081.0,
+      "step": 4090
+    },
+    {
+      "entropy": 0.08448477580677718,
+      "epoch": 6.5600000000000005,
+      "grad_norm": 1.0703125,
+      "learning_rate": 1.3113333333333334e-05,
+      "loss": 0.0489,
+      "mean_token_accuracy": 0.987289821729064,
+      "num_tokens": 136678871.0,
+      "step": 4100
+    },
+    {
+      "epoch": 6.5600000000000005,
+      "eval_chemistry_entropy": 0.2614769343137741,
+      "eval_chemistry_loss": 1.2504053115844727,
+      "eval_chemistry_mean_token_accuracy": 0.7877317543029785,
+      "eval_chemistry_num_tokens": 136678871.0,
+      "eval_chemistry_runtime": 34.8317,
+      "eval_chemistry_samples_per_second": 14.355,
+      "eval_chemistry_steps_per_second": 3.589,
+      "step": 4100
+    },
+    {
+      "epoch": 6.5600000000000005,
+      "eval_physics_entropy": 0.20628420758247376,
+      "eval_physics_loss": 1.033398151397705,
+      "eval_physics_mean_token_accuracy": 0.822932026386261,
+      "eval_physics_num_tokens": 136678871.0,
+      "eval_physics_runtime": 41.5382,
+      "eval_physics_samples_per_second": 12.037,
+      "eval_physics_steps_per_second": 3.009,
+      "step": 4100
+    },
+    {
+      "entropy": 0.08356979857198894,
+      "epoch": 6.576,
+      "grad_norm": 1.1640625,
+      "learning_rate": 1.3091111111111113e-05,
+      "loss": 0.0487,
+      "mean_token_accuracy": 0.9873802177608013,
+      "num_tokens": 137009120.0,
+      "step": 4110
+    },
+    {
+      "entropy": 0.08221683418378234,
+      "epoch": 6.592,
+      "grad_norm": 1.1875,
+      "learning_rate": 1.306888888888889e-05,
+      "loss": 0.0473,
+      "mean_token_accuracy": 0.9874389294534922,
+      "num_tokens": 137349002.0,
+      "step": 4120
+    },
+    {
+      "entropy": 0.08223750218749046,
+      "epoch": 6.608,
+      "grad_norm": 1.15625,
+      "learning_rate": 1.3046666666666668e-05,
+      "loss": 0.0491,
+      "mean_token_accuracy": 0.987232893705368,
+      "num_tokens": 137681271.0,
+      "step": 4130
+    },
+    {
+      "entropy": 0.08213731991127134,
+      "epoch": 6.624,
+      "grad_norm": 1.03125,
+      "learning_rate": 1.3024444444444446e-05,
+      "loss": 0.048,
+      "mean_token_accuracy": 0.9874076224863529,
+      "num_tokens": 138017964.0,
+      "step": 4140
+    },
+    {
+      "entropy": 0.08445660735014826,
+      "epoch": 6.64,
+      "grad_norm": 1.171875,
+      "learning_rate": 1.3002222222222223e-05,
+      "loss": 0.0489,
+      "mean_token_accuracy": 0.9871912717819213,
+      "num_tokens": 138344865.0,
+      "step": 4150
+    },
+    {
+      "entropy": 0.08164793641772121,
+      "epoch": 6.656,
+      "grad_norm": 1.15625,
+      "learning_rate": 1.2980000000000001e-05,
+      "loss": 0.0476,
+      "mean_token_accuracy": 0.9875594973564148,
+      "num_tokens": 138686067.0,
+      "step": 4160
+    },
+    {
+      "entropy": 0.08325545103289186,
+      "epoch": 6.672,
+      "grad_norm": 1.15625,
+      "learning_rate": 1.2957777777777779e-05,
+      "loss": 0.0472,
+      "mean_token_accuracy": 0.9876235947012901,
+      "num_tokens": 139020281.0,
+      "step": 4170
+    },
+    {
+      "entropy": 0.08389222861733288,
+      "epoch": 6.688,
+      "grad_norm": 1.2578125,
+      "learning_rate": 1.2935555555555556e-05,
+      "loss": 0.0486,
+      "mean_token_accuracy": 0.9872424896806479,
+      "num_tokens": 139358427.0,
+      "step": 4180
+    },
+    {
+      "entropy": 0.08402099246159196,
+      "epoch": 6.704,
+      "grad_norm": 1.1640625,
+      "learning_rate": 1.2913333333333336e-05,
+      "loss": 0.0488,
+      "mean_token_accuracy": 0.9872086331248283,
+      "num_tokens": 139691142.0,
+      "step": 4190
+    },
+    {
+      "entropy": 0.08593915849924087,
+      "epoch": 6.72,
+      "grad_norm": 1.3046875,
+      "learning_rate": 1.2891111111111112e-05,
+      "loss": 0.0504,
+      "mean_token_accuracy": 0.986945329606533,
+      "num_tokens": 140011774.0,
+      "step": 4200
+    },
+    {
+      "epoch": 6.72,
+      "eval_chemistry_entropy": 0.25900897359848024,
+      "eval_chemistry_loss": 1.254063606262207,
+      "eval_chemistry_mean_token_accuracy": 0.7872775292396545,
+      "eval_chemistry_num_tokens": 140011774.0,
+      "eval_chemistry_runtime": 34.8636,
+      "eval_chemistry_samples_per_second": 14.342,
+      "eval_chemistry_steps_per_second": 3.585,
+      "step": 4200
+    },
+    {
+      "epoch": 6.72,
+      "eval_physics_entropy": 0.20440801638364792,
+      "eval_physics_loss": 1.0369616746902466,
+      "eval_physics_mean_token_accuracy": 0.8233732528686524,
+      "eval_physics_num_tokens": 140011774.0,
+      "eval_physics_runtime": 41.5995,
+      "eval_physics_samples_per_second": 12.019,
+      "eval_physics_steps_per_second": 3.005,
+      "step": 4200
+    },
+    {
+      "entropy": 0.08310960887465627,
+      "epoch": 6.736,
+      "grad_norm": 1.2109375,
+      "learning_rate": 1.2868888888888891e-05,
+      "loss": 0.0487,
+      "mean_token_accuracy": 0.9873503148555756,
+      "num_tokens": 140349276.0,
+      "step": 4210
+    },
+    {
+      "entropy": 0.08384478732477874,
+      "epoch": 6.752,
+      "grad_norm": 1.1484375,
+      "learning_rate": 1.2846666666666667e-05,
+      "loss": 0.049,
+      "mean_token_accuracy": 0.9872010689228773,
+      "num_tokens": 140674986.0,
+      "step": 4220
+    },
+    {
+      "entropy": 0.08250078009441494,
+      "epoch": 6.768,
+      "grad_norm": 1.2265625,
+      "learning_rate": 1.2824444444444446e-05,
+      "loss": 0.0484,
+      "mean_token_accuracy": 0.9873097315430641,
+      "num_tokens": 141023119.0,
+      "step": 4230
+    },
+    {
+      "entropy": 0.08327699287328869,
+      "epoch": 6.784,
+      "grad_norm": 1.0859375,
+      "learning_rate": 1.2802222222222222e-05,
+      "loss": 0.0489,
+      "mean_token_accuracy": 0.9871655073016882,
+      "num_tokens": 141350499.0,
+      "step": 4240
+    },
+    {
+      "entropy": 0.08524548292625696,
+      "epoch": 6.8,
+      "grad_norm": 1.203125,
+      "learning_rate": 1.2780000000000001e-05,
+      "loss": 0.05,
+      "mean_token_accuracy": 0.9869431726634502,
+      "num_tokens": 141690904.0,
+      "step": 4250
+    },
+    {
+      "entropy": 0.08612931903917342,
+      "epoch": 6.816,
+      "grad_norm": 1.0546875,
+      "learning_rate": 1.2757777777777777e-05,
+      "loss": 0.0503,
+      "mean_token_accuracy": 0.9868447929620743,
+      "num_tokens": 142014404.0,
+      "step": 4260
+    },
+    {
+      "entropy": 0.08538674390874804,
+      "epoch": 6.832,
+      "grad_norm": 1.2109375,
+      "learning_rate": 1.2735555555555557e-05,
+      "loss": 0.0483,
+      "mean_token_accuracy": 0.9872184365987777,
+      "num_tokens": 142342717.0,
+      "step": 4270
+    },
+    {
+      "entropy": 0.08172557696234435,
+      "epoch": 6.848,
+      "grad_norm": 1.3203125,
+      "learning_rate": 1.2713333333333336e-05,
+      "loss": 0.0486,
+      "mean_token_accuracy": 0.9877215001732111,
+      "num_tokens": 142679601.0,
+      "step": 4280
+    },
+    {
+      "entropy": 0.08404393044766038,
+      "epoch": 6.864,
+      "grad_norm": 1.1640625,
+      "learning_rate": 1.2691111111111112e-05,
+      "loss": 0.0484,
+      "mean_token_accuracy": 0.9874385319650173,
+      "num_tokens": 143012034.0,
+      "step": 4290
+    },
+    {
+      "entropy": 0.08339044626336545,
+      "epoch": 6.88,
+      "grad_norm": 1.2109375,
+      "learning_rate": 1.2668888888888891e-05,
+      "loss": 0.0493,
+      "mean_token_accuracy": 0.9870673533529043,
+      "num_tokens": 143345287.0,
+      "step": 4300
+    },
+    {
+      "epoch": 6.88,
+      "eval_chemistry_entropy": 0.2598011150360107,
+      "eval_chemistry_loss": 1.2587450742721558,
+      "eval_chemistry_mean_token_accuracy": 0.786902666091919,
+      "eval_chemistry_num_tokens": 143345287.0,
+      "eval_chemistry_runtime": 34.9182,
+      "eval_chemistry_samples_per_second": 14.319,
+      "eval_chemistry_steps_per_second": 3.58,
+      "step": 4300
+    },
+    {
+      "epoch": 6.88,
+      "eval_physics_entropy": 0.20457589453458785,
+      "eval_physics_loss": 1.0422227382659912,
+      "eval_physics_mean_token_accuracy": 0.822963632106781,
+      "eval_physics_num_tokens": 143345287.0,
+      "eval_physics_runtime": 41.5177,
+      "eval_physics_samples_per_second": 12.043,
+      "eval_physics_steps_per_second": 3.011,
+      "step": 4300
+    },
+    {
+      "entropy": 0.08436821920331568,
+      "epoch": 6.896,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.2646666666666667e-05,
+      "loss": 0.0504,
+      "mean_token_accuracy": 0.9868172734975815,
+      "num_tokens": 143667749.0,
+      "step": 4310
+    },
+    {
+      "entropy": 0.08363941961433738,
+      "epoch": 6.912,
+      "grad_norm": 1.1796875,
+      "learning_rate": 1.2624444444444446e-05,
+      "loss": 0.049,
+      "mean_token_accuracy": 0.9870785433799029,
+      "num_tokens": 144000847.0,
+      "step": 4320
+    },
+    {
+      "entropy": 0.08455635379068553,
+      "epoch": 6.928,
+      "grad_norm": 1.28125,
+      "learning_rate": 1.2602222222222222e-05,
+      "loss": 0.0488,
+      "mean_token_accuracy": 0.9873150497674942,
+      "num_tokens": 144333903.0,
+      "step": 4330
+    },
+    {
+      "entropy": 0.08151172450743616,
+      "epoch": 6.944,
+      "grad_norm": 1.1328125,
+      "learning_rate": 1.2580000000000002e-05,
+      "loss": 0.048,
+      "mean_token_accuracy": 0.9873838592320681,
+      "num_tokens": 144671987.0,
+      "step": 4340
+    },
+    {
+      "entropy": 0.08456669892184436,
+      "epoch": 6.96,
+      "grad_norm": 1.1953125,
+      "learning_rate": 1.2557777777777779e-05,
+      "loss": 0.0489,
+      "mean_token_accuracy": 0.9870229866355658,
+      "num_tokens": 145001479.0,
+      "step": 4350
+    },
+    {
+      "entropy": 0.08347726857755333,
+      "epoch": 6.976,
+      "grad_norm": 1.2421875,
+      "learning_rate": 1.2535555555555557e-05,
+      "loss": 0.0495,
+      "mean_token_accuracy": 0.9868554197251797,
+      "num_tokens": 145330342.0,
+      "step": 4360
+    },
+    {
+      "entropy": 0.08411271646618843,
+      "epoch": 6.992,
+      "grad_norm": 1.328125,
+      "learning_rate": 1.2513333333333334e-05,
+      "loss": 0.0501,
+      "mean_token_accuracy": 0.9865896072238683,
+      "num_tokens": 145655640.0,
+      "step": 4370
+    },
+    {
+      "entropy": 0.07735087708570063,
+      "epoch": 7.008,
+      "grad_norm": 0.96484375,
+      "learning_rate": 1.2491111111111112e-05,
+      "loss": 0.043,
+      "mean_token_accuracy": 0.9888126287609339,
+      "num_tokens": 145979339.0,
+      "step": 4380
+    },
+    {
+      "entropy": 0.06566564028616995,
+      "epoch": 7.024,
+      "grad_norm": 0.8203125,
+      "learning_rate": 1.246888888888889e-05,
+      "loss": 0.0368,
+      "mean_token_accuracy": 0.9905497398227453,
+      "num_tokens": 146311273.0,
+      "step": 4390
+    },
+    {
+      "entropy": 0.06366870710626245,
+      "epoch": 7.04,
+      "grad_norm": 0.72265625,
+      "learning_rate": 1.2446666666666667e-05,
+      "loss": 0.0358,
+      "mean_token_accuracy": 0.9906815171241761,
+      "num_tokens": 146636386.0,
+      "step": 4400
+    },
+    {
+      "epoch": 7.04,
+      "eval_chemistry_entropy": 0.24919855165481566,
+      "eval_chemistry_loss": 1.2937006950378418,
+      "eval_chemistry_mean_token_accuracy": 0.7869456210136414,
+      "eval_chemistry_num_tokens": 146636386.0,
+      "eval_chemistry_runtime": 34.8891,
+      "eval_chemistry_samples_per_second": 14.331,
+      "eval_chemistry_steps_per_second": 3.583,
+      "step": 4400
+    },
+    {
+      "epoch": 7.04,
+      "eval_physics_entropy": 0.19420701318979264,
+      "eval_physics_loss": 1.0773091316223145,
+      "eval_physics_mean_token_accuracy": 0.8231866316795349,
+      "eval_physics_num_tokens": 146636386.0,
+      "eval_physics_runtime": 41.4788,
+      "eval_physics_samples_per_second": 12.054,
+      "eval_physics_steps_per_second": 3.014,
+      "step": 4400
+    },
+    {
+      "entropy": 0.06386563028208911,
+      "epoch": 7.056,
+      "grad_norm": 0.8515625,
+      "learning_rate": 1.2424444444444445e-05,
+      "loss": 0.0351,
+      "mean_token_accuracy": 0.9907928127795458,
+      "num_tokens": 146967650.0,
+      "step": 4410
+    },
+    {
+      "entropy": 0.06075377077795565,
+      "epoch": 7.072,
+      "grad_norm": 0.7890625,
+      "learning_rate": 1.2402222222222222e-05,
+      "loss": 0.0342,
+      "mean_token_accuracy": 0.9910480942577123,
+      "num_tokens": 147304637.0,
+      "step": 4420
+    },
+    {
+      "entropy": 0.06137628590222448,
+      "epoch": 7.088,
+      "grad_norm": 0.7578125,
+      "learning_rate": 1.2380000000000002e-05,
+      "loss": 0.0343,
+      "mean_token_accuracy": 0.9908504582941532,
+      "num_tokens": 147644140.0,
+      "step": 4430
+    },
+    {
+      "entropy": 0.06214636459480971,
+      "epoch": 7.104,
+      "grad_norm": 0.75,
+      "learning_rate": 1.235777777777778e-05,
+      "loss": 0.0354,
+      "mean_token_accuracy": 0.9906146645545959,
+      "num_tokens": 147976110.0,
+      "step": 4440
+    },
+    {
+      "entropy": 0.061552948271855715,
+      "epoch": 7.12,
+      "grad_norm": 0.87109375,
+      "learning_rate": 1.2335555555555557e-05,
+      "loss": 0.0348,
+      "mean_token_accuracy": 0.9908938217908144,
+      "num_tokens": 148310082.0,
+      "step": 4450
+    },
+    {
+      "entropy": 0.06064079294446856,
+      "epoch": 7.136,
+      "grad_norm": 0.80078125,
+      "learning_rate": 1.2313333333333335e-05,
+      "loss": 0.0342,
+      "mean_token_accuracy": 0.9908621162176132,
+      "num_tokens": 148651071.0,
+      "step": 4460
+    },
+    {
+      "entropy": 0.06230100248940289,
+      "epoch": 7.152,
+      "grad_norm": 0.796875,
+      "learning_rate": 1.2291111111111112e-05,
+      "loss": 0.0355,
+      "mean_token_accuracy": 0.9905878521502018,
+      "num_tokens": 148980540.0,
+      "step": 4470
+    },
+    {
+      "entropy": 0.061447409261018035,
+      "epoch": 7.168,
+      "grad_norm": 0.82421875,
+      "learning_rate": 1.226888888888889e-05,
+      "loss": 0.0345,
+      "mean_token_accuracy": 0.9908699303865433,
+      "num_tokens": 149318823.0,
+      "step": 4480
+    },
+    {
+      "entropy": 0.0620643375441432,
+      "epoch": 7.184,
+      "grad_norm": 0.890625,
+      "learning_rate": 1.2246666666666667e-05,
+      "loss": 0.035,
+      "mean_token_accuracy": 0.9907150626182556,
+      "num_tokens": 149650890.0,
+      "step": 4490
+    },
+    {
+      "entropy": 0.061626524198800325,
+      "epoch": 7.2,
+      "grad_norm": 0.72265625,
+      "learning_rate": 1.2224444444444445e-05,
+      "loss": 0.0349,
+      "mean_token_accuracy": 0.9907727047801018,
+      "num_tokens": 149978963.0,
+      "step": 4500
+    },
+    {
+      "epoch": 7.2,
+      "eval_chemistry_entropy": 0.24628682976961136,
+      "eval_chemistry_loss": 1.3082973957061768,
+      "eval_chemistry_mean_token_accuracy": 0.7866316366195679,
+      "eval_chemistry_num_tokens": 149978963.0,
+      "eval_chemistry_runtime": 34.8147,
+      "eval_chemistry_samples_per_second": 14.362,
+      "eval_chemistry_steps_per_second": 3.59,
+      "step": 4500
+    },
+    {
+      "epoch": 7.2,
+      "eval_physics_entropy": 0.19074303054809572,
+      "eval_physics_loss": 1.09184730052948,
+      "eval_physics_mean_token_accuracy": 0.823555606842041,
+      "eval_physics_num_tokens": 149978963.0,
+      "eval_physics_runtime": 41.2335,
+      "eval_physics_samples_per_second": 12.126,
+      "eval_physics_steps_per_second": 3.032,
+      "step": 4500
+    },
+    {
+      "entropy": 0.06277301630470902,
+      "epoch": 7.216,
+      "grad_norm": 0.70703125,
+      "learning_rate": 1.2202222222222224e-05,
+      "loss": 0.0356,
+      "mean_token_accuracy": 0.9906882908195257,
+      "num_tokens": 150308803.0,
+      "step": 4510
+    },
+    {
+      "entropy": 0.06299484143964947,
+      "epoch": 7.232,
+      "grad_norm": 1.0,
+      "learning_rate": 1.218e-05,
+      "loss": 0.0354,
+      "mean_token_accuracy": 0.9908652398735285,
+      "num_tokens": 150640693.0,
+      "step": 4520
+    },
+    {
+      "entropy": 0.06334666730836033,
+      "epoch": 7.248,
+      "grad_norm": 0.68359375,
+      "learning_rate": 1.215777777777778e-05,
+      "loss": 0.0357,
+      "mean_token_accuracy": 0.9905265558511018,
+      "num_tokens": 150969935.0,
+      "step": 4530
+    },
+    {
+      "entropy": 0.06278704162687063,
+      "epoch": 7.264,
+      "grad_norm": 0.89453125,
+      "learning_rate": 1.2135555555555556e-05,
+      "loss": 0.0354,
+      "mean_token_accuracy": 0.9906135443598032,
+      "num_tokens": 151300990.0,
+      "step": 4540
+    },
+    {
+      "entropy": 0.06298413707409053,
+      "epoch": 7.28,
+      "grad_norm": 0.84765625,
+      "learning_rate": 1.2113333333333335e-05,
+      "loss": 0.0359,
+      "mean_token_accuracy": 0.990505576133728,
+      "num_tokens": 151630968.0,
+      "step": 4550
+    },
+    {
+      "entropy": 0.061677922517992555,
+      "epoch": 7.296,
+      "grad_norm": 0.77734375,
+      "learning_rate": 1.209111111111111e-05,
+      "loss": 0.0352,
+      "mean_token_accuracy": 0.9906646896153688,
+      "num_tokens": 151972512.0,
+      "step": 4560
+    },
+    {
+      "entropy": 0.06136521329171955,
+      "epoch": 7.312,
+      "grad_norm": 1.0234375,
+      "learning_rate": 1.206888888888889e-05,
+      "loss": 0.0349,
+      "mean_token_accuracy": 0.9907136838883162,
+      "num_tokens": 152309262.0,
+      "step": 4570
+    },
+    {
+      "entropy": 0.06150672952644527,
+      "epoch": 7.328,
+      "grad_norm": 0.75390625,
+      "learning_rate": 1.204666666666667e-05,
+      "loss": 0.0353,
+      "mean_token_accuracy": 0.9906399689614773,
+      "num_tokens": 152641115.0,
+      "step": 4580
+    },
+    {
+      "entropy": 0.06296003204770387,
+      "epoch": 7.344,
+      "grad_norm": 0.79296875,
+      "learning_rate": 1.2024444444444445e-05,
+      "loss": 0.0359,
+      "mean_token_accuracy": 0.9904595598578453,
+      "num_tokens": 152969903.0,
+      "step": 4590
+    },
+    {
+      "entropy": 0.060705285519361496,
+      "epoch": 7.36,
+      "grad_norm": 0.88671875,
+      "learning_rate": 1.2002222222222225e-05,
+      "loss": 0.0351,
+      "mean_token_accuracy": 0.99081862680614,
+      "num_tokens": 153310394.0,
+      "step": 4600
+    },
+    {
+      "epoch": 7.36,
+      "eval_chemistry_entropy": 0.2472569968700409,
+      "eval_chemistry_loss": 1.3085415363311768,
+      "eval_chemistry_mean_token_accuracy": 0.7877958555221558,
+      "eval_chemistry_num_tokens": 153310394.0,
+      "eval_chemistry_runtime": 34.8838,
+      "eval_chemistry_samples_per_second": 14.333,
+      "eval_chemistry_steps_per_second": 3.583,
+      "step": 4600
+    },
+    {
+      "epoch": 7.36,
+      "eval_physics_entropy": 0.19158041775226592,
+      "eval_physics_loss": 1.0906758308410645,
+      "eval_physics_mean_token_accuracy": 0.8236755423545837,
+      "eval_physics_num_tokens": 153310394.0,
+      "eval_physics_runtime": 41.4951,
+      "eval_physics_samples_per_second": 12.05,
+      "eval_physics_steps_per_second": 3.012,
+      "step": 4600
+    },
+    {
+      "entropy": 0.061911187367513776,
+      "epoch": 7.376,
+      "grad_norm": 0.73046875,
+      "learning_rate": 1.198e-05,
+      "loss": 0.0349,
+      "mean_token_accuracy": 0.9907222799956799,
+      "num_tokens": 153641823.0,
+      "step": 4610
+    },
+    {
+      "entropy": 0.061987010017037394,
+      "epoch": 7.392,
+      "grad_norm": 1.1015625,
+      "learning_rate": 1.195777777777778e-05,
+      "loss": 0.0353,
+      "mean_token_accuracy": 0.9906540635973216,
+      "num_tokens": 153979365.0,
+      "step": 4620
+    },
+    {
+      "entropy": 0.061248681391589344,
+      "epoch": 7.408,
+      "grad_norm": 0.62890625,
+      "learning_rate": 1.1935555555555556e-05,
+      "loss": 0.0354,
+      "mean_token_accuracy": 0.9906076569110155,
+      "num_tokens": 154306629.0,
+      "step": 4630
+    },
+    {
+      "entropy": 0.0623122047400102,
+      "epoch": 7.424,
+      "grad_norm": 0.7890625,
+      "learning_rate": 1.1913333333333335e-05,
+      "loss": 0.0348,
+      "mean_token_accuracy": 0.9906488709151745,
+      "num_tokens": 154641261.0,
+      "step": 4640
+    },
+    {
+      "entropy": 0.06312244785949588,
+      "epoch": 7.44,
+      "grad_norm": 0.76953125,
+      "learning_rate": 1.1891111111111111e-05,
+      "loss": 0.0355,
+      "mean_token_accuracy": 0.9905753880739212,
+      "num_tokens": 154970740.0,
+      "step": 4650
+    },
+    {
+      "entropy": 0.06212042509578168,
+      "epoch": 7.456,
+      "grad_norm": 0.80859375,
+      "learning_rate": 1.186888888888889e-05,
+      "loss": 0.0353,
+      "mean_token_accuracy": 0.9903509549796581,
+      "num_tokens": 155305774.0,
+      "step": 4660
+    },
+    {
+      "entropy": 0.060513887810520826,
+      "epoch": 7.4719999999999995,
+      "grad_norm": 0.76953125,
+      "learning_rate": 1.1846666666666668e-05,
+      "loss": 0.035,
+      "mean_token_accuracy": 0.990780594944954,
+      "num_tokens": 155642312.0,
+      "step": 4670
+    },
+    {
+      "entropy": 0.06330470081884414,
+      "epoch": 7.4879999999999995,
+      "grad_norm": 0.6484375,
+      "learning_rate": 1.1824444444444445e-05,
+      "loss": 0.036,
+      "mean_token_accuracy": 0.9902980923652649,
+      "num_tokens": 155972296.0,
+      "step": 4680
+    },
+    {
+      "entropy": 0.06154982123989612,
+      "epoch": 7.504,
+      "grad_norm": 0.82421875,
+      "learning_rate": 1.1802222222222223e-05,
+      "loss": 0.0352,
+      "mean_token_accuracy": 0.990791617333889,
+      "num_tokens": 156313067.0,
+      "step": 4690
+    },
+    {
+      "entropy": 0.062418047641403976,
+      "epoch": 7.52,
+      "grad_norm": 0.82421875,
+      "learning_rate": 1.178e-05,
+      "loss": 0.0356,
+      "mean_token_accuracy": 0.9906047355383635,
+      "num_tokens": 156647045.0,
+      "step": 4700
+    },
+    {
+      "epoch": 7.52,
+      "eval_chemistry_entropy": 0.24599135047197343,
+      "eval_chemistry_loss": 1.3103718757629395,
+      "eval_chemistry_mean_token_accuracy": 0.787072063446045,
+      "eval_chemistry_num_tokens": 156647045.0,
+      "eval_chemistry_runtime": 34.7892,
+      "eval_chemistry_samples_per_second": 14.372,
+      "eval_chemistry_steps_per_second": 3.593,
+      "step": 4700
+    },
+    {
+      "epoch": 7.52,
+      "eval_physics_entropy": 0.1908007864356041,
+      "eval_physics_loss": 1.0921863317489624,
+      "eval_physics_mean_token_accuracy": 0.8236469144821167,
+      "eval_physics_num_tokens": 156647045.0,
+      "eval_physics_runtime": 41.44,
+      "eval_physics_samples_per_second": 12.066,
+      "eval_physics_steps_per_second": 3.016,
+      "step": 4700
+    },
+    {
+      "entropy": 0.06150080761872232,
+      "epoch": 7.536,
+      "grad_norm": 0.7578125,
+      "learning_rate": 1.1757777777777778e-05,
+      "loss": 0.0349,
+      "mean_token_accuracy": 0.9907149720937014,
+      "num_tokens": 156983291.0,
+      "step": 4710
+    },
+    {
+      "entropy": 0.06284960370976478,
+      "epoch": 7.552,
+      "grad_norm": 0.72265625,
+      "learning_rate": 1.1735555555555556e-05,
+      "loss": 0.0357,
+      "mean_token_accuracy": 0.9905340883880853,
+      "num_tokens": 157315799.0,
+      "step": 4720
+    },
+    {
+      "entropy": 0.06063510156236589,
+      "epoch": 7.568,
+      "grad_norm": 0.796875,
+      "learning_rate": 1.1713333333333334e-05,
+      "loss": 0.034,
+      "mean_token_accuracy": 0.990770636498928,
+      "num_tokens": 157661992.0,
+      "step": 4730
+    },
+    {
+      "entropy": 0.06052122050896287,
+      "epoch": 7.584,
+      "grad_norm": 0.8359375,
+      "learning_rate": 1.1691111111111113e-05,
+      "loss": 0.0352,
+      "mean_token_accuracy": 0.9905507165938616,
+      "num_tokens": 157994634.0,
+      "step": 4740
+    },
+    {
+      "entropy": 0.06159715664107353,
+      "epoch": 7.6,
+      "grad_norm": 0.78515625,
+      "learning_rate": 1.1668888888888889e-05,
+      "loss": 0.0353,
+      "mean_token_accuracy": 0.9906620573252439,
+      "num_tokens": 158331884.0,
+      "step": 4750
+    },
+    {
+      "entropy": 0.06264028470031917,
+      "epoch": 7.616,
+      "grad_norm": 0.64453125,
+      "learning_rate": 1.1646666666666668e-05,
+      "loss": 0.0357,
+      "mean_token_accuracy": 0.9904555916786194,
+      "num_tokens": 158659879.0,
+      "step": 4760
+    },
+    {
+      "entropy": 0.06209464946296066,
+      "epoch": 7.632,
+      "grad_norm": 0.77734375,
+      "learning_rate": 1.1624444444444446e-05,
+      "loss": 0.036,
+      "mean_token_accuracy": 0.99032150991261,
+      "num_tokens": 158984913.0,
+      "step": 4770
+    },
+    {
+      "entropy": 0.06313899885863065,
+      "epoch": 7.648,
+      "grad_norm": 0.96484375,
+      "learning_rate": 1.1602222222222223e-05,
+      "loss": 0.0361,
+      "mean_token_accuracy": 0.9904160272330046,
+      "num_tokens": 159312277.0,
+      "step": 4780
+    },
+    {
+      "entropy": 0.06148210281971842,
+      "epoch": 7.664,
+      "grad_norm": 0.8828125,
+      "learning_rate": 1.1580000000000001e-05,
+      "loss": 0.035,
+      "mean_token_accuracy": 0.9906214911490678,
+      "num_tokens": 159649435.0,
+      "step": 4790
+    },
+    {
+      "entropy": 0.061378013179637494,
+      "epoch": 7.68,
+      "grad_norm": 0.8359375,
+      "learning_rate": 1.1557777777777779e-05,
+      "loss": 0.0349,
+      "mean_token_accuracy": 0.9906430255621672,
+      "num_tokens": 159989472.0,
+      "step": 4800
+    },
+    {
+      "epoch": 7.68,
+      "eval_chemistry_entropy": 0.24605670869350432,
+      "eval_chemistry_loss": 1.313415765762329,
+      "eval_chemistry_mean_token_accuracy": 0.7872930493354797,
+      "eval_chemistry_num_tokens": 159989472.0,
+      "eval_chemistry_runtime": 34.7673,
+      "eval_chemistry_samples_per_second": 14.381,
+      "eval_chemistry_steps_per_second": 3.595,
+      "step": 4800
+    },
+    {
+      "epoch": 7.68,
+      "eval_physics_entropy": 0.19055269342660905,
+      "eval_physics_loss": 1.0957832336425781,
+      "eval_physics_mean_token_accuracy": 0.8236758513450623,
+      "eval_physics_num_tokens": 159989472.0,
+      "eval_physics_runtime": 41.4559,
+      "eval_physics_samples_per_second": 12.061,
+      "eval_physics_steps_per_second": 3.015,
+      "step": 4800
+    },
+    {
+      "entropy": 0.06214101852383465,
+      "epoch": 7.696,
+      "grad_norm": 0.8125,
+      "learning_rate": 1.1535555555555556e-05,
+      "loss": 0.0359,
+      "mean_token_accuracy": 0.9905785106122493,
+      "num_tokens": 160320528.0,
+      "step": 4810
+    },
+    {
+      "entropy": 0.06170573707204312,
+      "epoch": 7.712,
+      "grad_norm": 0.640625,
+      "learning_rate": 1.1513333333333334e-05,
+      "loss": 0.0349,
+      "mean_token_accuracy": 0.9908230375498533,
+      "num_tokens": 160658923.0,
+      "step": 4820
+    },
+    {
+      "entropy": 0.06164445518516004,
+      "epoch": 7.728,
+      "grad_norm": 0.65234375,
+      "learning_rate": 1.1491111111111113e-05,
+      "loss": 0.0351,
+      "mean_token_accuracy": 0.9907540172338486,
+      "num_tokens": 160996486.0,
+      "step": 4830
+    },
+    {
+      "entropy": 0.06168614816851914,
+      "epoch": 7.744,
+      "grad_norm": 0.828125,
+      "learning_rate": 1.1468888888888889e-05,
+      "loss": 0.0354,
+      "mean_token_accuracy": 0.9904147505760192,
+      "num_tokens": 161331359.0,
+      "step": 4840
+    },
+    {
+      "entropy": 0.06083308178931475,
+      "epoch": 7.76,
+      "grad_norm": 0.796875,
+      "learning_rate": 1.1446666666666668e-05,
+      "loss": 0.0352,
+      "mean_token_accuracy": 0.9906918577849865,
+      "num_tokens": 161665129.0,
+      "step": 4850
+    },
+    {
+      "entropy": 0.06237182246986776,
+      "epoch": 7.776,
+      "grad_norm": 0.83984375,
+      "learning_rate": 1.1424444444444444e-05,
+      "loss": 0.0356,
+      "mean_token_accuracy": 0.9904262360185385,
+      "num_tokens": 161994034.0,
+      "step": 4860
+    },
+    {
+      "entropy": 0.0600957038346678,
+      "epoch": 7.792,
+      "grad_norm": 0.76171875,
+      "learning_rate": 1.1402222222222224e-05,
+      "loss": 0.0344,
+      "mean_token_accuracy": 0.9907166257500648,
+      "num_tokens": 162337364.0,
+      "step": 4870
+    },
+    {
+      "entropy": 0.06367817365098745,
+      "epoch": 7.808,
+      "grad_norm": 0.85546875,
+      "learning_rate": 1.138e-05,
+      "loss": 0.0367,
+      "mean_token_accuracy": 0.9902267444878816,
+      "num_tokens": 162661168.0,
+      "step": 4880
+    },
+    {
+      "entropy": 0.06245433308649808,
+      "epoch": 7.824,
+      "grad_norm": 0.63671875,
+      "learning_rate": 1.1357777777777779e-05,
+      "loss": 0.0356,
+      "mean_token_accuracy": 0.9906036704778671,
+      "num_tokens": 162995603.0,
+      "step": 4890
+    },
+    {
+      "entropy": 0.0599086759146303,
+      "epoch": 7.84,
+      "grad_norm": 0.8671875,
+      "learning_rate": 1.1335555555555558e-05,
+      "loss": 0.034,
+      "mean_token_accuracy": 0.990909768268466,
+      "num_tokens": 163342363.0,
+      "step": 4900
+    },
+    {
+      "epoch": 7.84,
+      "eval_chemistry_entropy": 0.24625909847021102,
+      "eval_chemistry_loss": 1.3171546459197998,
+      "eval_chemistry_mean_token_accuracy": 0.7877407064437866,
+      "eval_chemistry_num_tokens": 163342363.0,
+      "eval_chemistry_runtime": 34.7118,
+      "eval_chemistry_samples_per_second": 14.404,
+      "eval_chemistry_steps_per_second": 3.601,
+      "step": 4900
+    },
+    {
+      "epoch": 7.84,
+      "eval_physics_entropy": 0.19048372292518614,
+      "eval_physics_loss": 1.099347710609436,
+      "eval_physics_mean_token_accuracy": 0.8234868450164795,
+      "eval_physics_num_tokens": 163342363.0,
+      "eval_physics_runtime": 41.3398,
+      "eval_physics_samples_per_second": 12.095,
+      "eval_physics_steps_per_second": 3.024,
+      "step": 4900
+    },
+    {
+      "entropy": 0.06133901767898351,
+      "epoch": 7.856,
+      "grad_norm": 0.75,
+      "learning_rate": 1.1313333333333334e-05,
+      "loss": 0.0351,
+      "mean_token_accuracy": 0.9907282523810863,
+      "num_tokens": 163678125.0,
+      "step": 4910
+    },
+    {
+      "entropy": 0.06085761564318091,
+      "epoch": 7.872,
+      "grad_norm": 0.94140625,
+      "learning_rate": 1.1291111111111113e-05,
+      "loss": 0.0355,
+      "mean_token_accuracy": 0.9904114957898855,
+      "num_tokens": 164008552.0,
+      "step": 4920
+    },
+    {
+      "entropy": 0.06117377581540495,
+      "epoch": 7.888,
+      "grad_norm": 0.71484375,
+      "learning_rate": 1.126888888888889e-05,
+      "loss": 0.035,
+      "mean_token_accuracy": 0.9905702136456966,
+      "num_tokens": 164343940.0,
+      "step": 4930
+    },
+    {
+      "entropy": 0.06296905796043575,
+      "epoch": 7.904,
+      "grad_norm": 0.75,
+      "learning_rate": 1.1246666666666669e-05,
+      "loss": 0.0359,
+      "mean_token_accuracy": 0.9903348471969367,
+      "num_tokens": 164669676.0,
+      "step": 4940
+    },
+    {
+      "entropy": 0.06158479375299066,
+      "epoch": 7.92,
+      "grad_norm": 0.91015625,
+      "learning_rate": 1.1224444444444444e-05,
+      "loss": 0.0356,
+      "mean_token_accuracy": 0.9904952511191368,
+      "num_tokens": 165003527.0,
+      "step": 4950
+    },
+    {
+      "entropy": 0.06204054574482143,
+      "epoch": 7.936,
+      "grad_norm": 0.77734375,
+      "learning_rate": 1.1202222222222224e-05,
+      "loss": 0.0359,
+      "mean_token_accuracy": 0.9905475355684757,
+      "num_tokens": 165329882.0,
+      "step": 4960
+    },
+    {
+      "entropy": 0.06273722697515041,
+      "epoch": 7.952,
+      "grad_norm": 0.6640625,
+      "learning_rate": 1.1180000000000001e-05,
+      "loss": 0.0364,
+      "mean_token_accuracy": 0.99009695276618,
+      "num_tokens": 165656028.0,
+      "step": 4970
+    },
+    {
+      "entropy": 0.061676447722129525,
+      "epoch": 7.968,
+      "grad_norm": 0.72265625,
+      "learning_rate": 1.1157777777777779e-05,
+      "loss": 0.0352,
+      "mean_token_accuracy": 0.9905911479145288,
+      "num_tokens": 165996204.0,
+      "step": 4980
+    },
+    {
+      "entropy": 0.06292406492866576,
+      "epoch": 7.984,
+      "grad_norm": 0.83203125,
+      "learning_rate": 1.1135555555555557e-05,
+      "loss": 0.0368,
+      "mean_token_accuracy": 0.9901123847812414,
+      "num_tokens": 166319440.0,
+      "step": 4990
+    },
+    {
+      "entropy": 0.061689905286766586,
+      "epoch": 8.0,
+      "grad_norm": 0.84765625,
+      "learning_rate": 1.1113333333333334e-05,
+      "loss": 0.0356,
+      "mean_token_accuracy": 0.9905704878270626,
+      "num_tokens": 166657184.0,
+      "step": 5000
+    },
+    {
+      "epoch": 8.0,
+      "eval_chemistry_entropy": 0.24496503752470017,
+      "eval_chemistry_loss": 1.3180774450302124,
+      "eval_chemistry_mean_token_accuracy": 0.787153443813324,
+      "eval_chemistry_num_tokens": 166657184.0,
+      "eval_chemistry_runtime": 34.7032,
+      "eval_chemistry_samples_per_second": 14.408,
+      "eval_chemistry_steps_per_second": 3.602,
+      "step": 5000
+    },
+    {
+      "epoch": 8.0,
+      "eval_physics_entropy": 0.1892443243265152,
+      "eval_physics_loss": 1.1011184453964233,
+      "eval_physics_mean_token_accuracy": 0.8235866379737854,
+      "eval_physics_num_tokens": 166657184.0,
+      "eval_physics_runtime": 41.4443,
+      "eval_physics_samples_per_second": 12.064,
+      "eval_physics_steps_per_second": 3.016,
+      "step": 5000
+    },
+    {
+      "entropy": 0.05658954493701458,
+      "epoch": 8.016,
+      "grad_norm": 0.458984375,
+      "learning_rate": 1.1091111111111112e-05,
+      "loss": 0.0306,
+      "mean_token_accuracy": 0.9916762802749872,
+      "num_tokens": 166984162.0,
+      "step": 5010
+    },
+    {
+      "entropy": 0.051856494625099,
+      "epoch": 8.032,
+      "grad_norm": 0.5234375,
+      "learning_rate": 1.106888888888889e-05,
+      "loss": 0.029,
+      "mean_token_accuracy": 0.9921161111444234,
+      "num_tokens": 167328157.0,
+      "step": 5020
+    },
+    {
+      "entropy": 0.05294681282248348,
+      "epoch": 8.048,
+      "grad_norm": 0.6015625,
+      "learning_rate": 1.1046666666666667e-05,
+      "loss": 0.03,
+      "mean_token_accuracy": 0.9917699113488198,
+      "num_tokens": 167660520.0,
+      "step": 5030
+    },
+    {
+      "entropy": 0.052667643525637686,
+      "epoch": 8.064,
+      "grad_norm": 0.45703125,
+      "learning_rate": 1.1024444444444445e-05,
+      "loss": 0.0294,
+      "mean_token_accuracy": 0.9919691778719425,
+      "num_tokens": 167998794.0,
+      "step": 5040
+    },
+    {
+      "entropy": 0.0552608639234677,
+      "epoch": 8.08,
+      "grad_norm": 0.5,
+      "learning_rate": 1.1002222222222222e-05,
+      "loss": 0.0307,
+      "mean_token_accuracy": 0.9914518911391497,
+      "num_tokens": 168323234.0,
+      "step": 5050
+    },
+    {
+      "entropy": 0.052295126882381734,
+      "epoch": 8.096,
+      "grad_norm": 0.48828125,
+      "learning_rate": 1.0980000000000002e-05,
+      "loss": 0.0298,
+      "mean_token_accuracy": 0.9917847782373428,
+      "num_tokens": 168656296.0,
+      "step": 5060
+    },
+    {
+      "entropy": 0.05391629205550998,
+      "epoch": 8.112,
+      "grad_norm": 0.50390625,
+      "learning_rate": 1.0957777777777778e-05,
+      "loss": 0.0299,
+      "mean_token_accuracy": 0.9919437572360039,
+      "num_tokens": 168990171.0,
+      "step": 5070
+    },
+    {
+      "entropy": 0.05125883354339748,
+      "epoch": 8.128,
+      "grad_norm": 0.45703125,
+      "learning_rate": 1.0935555555555557e-05,
+      "loss": 0.029,
+      "mean_token_accuracy": 0.9921612374484539,
+      "num_tokens": 169331697.0,
+      "step": 5080
+    },
+    {
+      "entropy": 0.05395408240146935,
+      "epoch": 8.144,
+      "grad_norm": 0.5546875,
+      "learning_rate": 1.0913333333333333e-05,
+      "loss": 0.0304,
+      "mean_token_accuracy": 0.9914871077984572,
+      "num_tokens": 169653974.0,
+      "step": 5090
+    },
+    {
+      "entropy": 0.053103095712140204,
+      "epoch": 8.16,
+      "grad_norm": 0.5546875,
+      "learning_rate": 1.0891111111111112e-05,
+      "loss": 0.0298,
+      "mean_token_accuracy": 0.9918887354433537,
+      "num_tokens": 169988271.0,
+      "step": 5100
+    },
+    {
+      "epoch": 8.16,
+      "eval_chemistry_entropy": 0.2404644394516945,
+      "eval_chemistry_loss": 1.3491953611373901,
+      "eval_chemistry_mean_token_accuracy": 0.7866144371032715,
+      "eval_chemistry_num_tokens": 169988271.0,
+      "eval_chemistry_runtime": 34.7372,
+      "eval_chemistry_samples_per_second": 14.394,
+      "eval_chemistry_steps_per_second": 3.598,
+      "step": 5100
+    },
+    {
+      "epoch": 8.16,
+      "eval_physics_entropy": 0.18387471383810045,
+      "eval_physics_loss": 1.1317907571792603,
+      "eval_physics_mean_token_accuracy": 0.8234862685203552,
+      "eval_physics_num_tokens": 169988271.0,
+      "eval_physics_runtime": 41.33,
+      "eval_physics_samples_per_second": 12.098,
+      "eval_physics_steps_per_second": 3.024,
+      "step": 5100
+    },
+    {
+      "entropy": 0.05399836541619152,
+      "epoch": 8.176,
+      "grad_norm": 0.69921875,
+      "learning_rate": 1.0868888888888888e-05,
+      "loss": 0.0302,
+      "mean_token_accuracy": 0.9915740687400103,
+      "num_tokens": 170318532.0,
+      "step": 5110
+    },
+    {
+      "entropy": 0.0519285038812086,
+      "epoch": 8.192,
+      "grad_norm": 0.8359375,
+      "learning_rate": 1.0846666666666667e-05,
+      "loss": 0.0295,
+      "mean_token_accuracy": 0.9918641287833452,
+      "num_tokens": 170654115.0,
+      "step": 5120
+    },
+    {
+      "entropy": 0.05231076295021921,
+      "epoch": 8.208,
+      "grad_norm": 0.5859375,
+      "learning_rate": 1.0824444444444447e-05,
+      "loss": 0.0297,
+      "mean_token_accuracy": 0.9916045140475035,
+      "num_tokens": 170992624.0,
+      "step": 5130
+    },
+    {
+      "entropy": 0.05148919146740809,
+      "epoch": 8.224,
+      "grad_norm": 0.54296875,
+      "learning_rate": 1.0802222222222223e-05,
+      "loss": 0.0295,
+      "mean_token_accuracy": 0.9919477555900812,
+      "num_tokens": 171330400.0,
+      "step": 5140
+    },
+    {
+      "entropy": 0.05204525606241077,
+      "epoch": 8.24,
+      "grad_norm": 0.578125,
+      "learning_rate": 1.0780000000000002e-05,
+      "loss": 0.0297,
+      "mean_token_accuracy": 0.9916645113378764,
+      "num_tokens": 171665654.0,
+      "step": 5150
+    },
+    {
+      "entropy": 0.0524903288576752,
+      "epoch": 8.256,
+      "grad_norm": 0.46875,
+      "learning_rate": 1.0757777777777778e-05,
+      "loss": 0.03,
+      "mean_token_accuracy": 0.9917620074003934,
+      "num_tokens": 171999055.0,
+      "step": 5160
+    },
+    {
+      "entropy": 0.051752155739814044,
+      "epoch": 8.272,
+      "grad_norm": 0.56640625,
+      "learning_rate": 1.0735555555555557e-05,
+      "loss": 0.0291,
+      "mean_token_accuracy": 0.9917046301066875,
+      "num_tokens": 172340397.0,
+      "step": 5170
+    },
+    {
+      "entropy": 0.05228375820443034,
+      "epoch": 8.288,
+      "grad_norm": 0.4453125,
+      "learning_rate": 1.0713333333333333e-05,
+      "loss": 0.0302,
+      "mean_token_accuracy": 0.9917192559689283,
+      "num_tokens": 172669133.0,
+      "step": 5180
+    },
+    {
+      "entropy": 0.05201203734613955,
+      "epoch": 8.304,
+      "grad_norm": 0.4921875,
+      "learning_rate": 1.0691111111111112e-05,
+      "loss": 0.0296,
+      "mean_token_accuracy": 0.9918897423893214,
+      "num_tokens": 173008902.0,
+      "step": 5190
+    },
+    {
+      "entropy": 0.052794103138148786,
+      "epoch": 8.32,
+      "grad_norm": 0.52734375,
+      "learning_rate": 1.0668888888888892e-05,
+      "loss": 0.0304,
+      "mean_token_accuracy": 0.9916907247155905,
+      "num_tokens": 173340185.0,
+      "step": 5200
+    },
+    {
+      "epoch": 8.32,
+      "eval_chemistry_entropy": 0.2388631991147995,
+      "eval_chemistry_loss": 1.3544237613677979,
+      "eval_chemistry_mean_token_accuracy": 0.7868574333190917,
+      "eval_chemistry_num_tokens": 173340185.0,
+      "eval_chemistry_runtime": 34.6629,
+      "eval_chemistry_samples_per_second": 14.425,
+      "eval_chemistry_steps_per_second": 3.606,
+      "step": 5200
+    },
+    {
+      "epoch": 8.32,
+      "eval_physics_entropy": 0.18251503622531892,
+      "eval_physics_loss": 1.1378728151321411,
+      "eval_physics_mean_token_accuracy": 0.8236250295639038,
+      "eval_physics_num_tokens": 173340185.0,
+      "eval_physics_runtime": 41.3432,
+      "eval_physics_samples_per_second": 12.094,
+      "eval_physics_steps_per_second": 3.023,
+      "step": 5200
+    },
+    {
+      "entropy": 0.0528043998638168,
+      "epoch": 8.336,
+      "grad_norm": 0.53515625,
+      "learning_rate": 1.0646666666666668e-05,
+      "loss": 0.0292,
+      "mean_token_accuracy": 0.9917436767369509,
+      "num_tokens": 173683875.0,
+      "step": 5210
+    },
+    {
+      "entropy": 0.051676605548709634,
+      "epoch": 8.352,
+      "grad_norm": 0.53515625,
+      "learning_rate": 1.0624444444444447e-05,
+      "loss": 0.0294,
+      "mean_token_accuracy": 0.9919365547597409,
+      "num_tokens": 174024460.0,
+      "step": 5220
+    },
+    {
+      "entropy": 0.053714934166055175,
+      "epoch": 8.368,
+      "grad_norm": 0.640625,
+      "learning_rate": 1.0602222222222223e-05,
+      "loss": 0.031,
+      "mean_token_accuracy": 0.9913693454116583,
+      "num_tokens": 174349489.0,
+      "step": 5230
+    },
+    {
+      "entropy": 0.052696572593413295,
+      "epoch": 8.384,
+      "grad_norm": 0.462890625,
+      "learning_rate": 1.0580000000000002e-05,
+      "loss": 0.0301,
+      "mean_token_accuracy": 0.9918276183307171,
+      "num_tokens": 174682644.0,
+      "step": 5240
+    },
+    {
+      "entropy": 0.05284402418183163,
+      "epoch": 8.4,
+      "grad_norm": 0.478515625,
+      "learning_rate": 1.0557777777777778e-05,
+      "loss": 0.0298,
+      "mean_token_accuracy": 0.9916664816439151,
+      "num_tokens": 175019376.0,
+      "step": 5250
+    },
+    {
+      "entropy": 0.05288535028230399,
+      "epoch": 8.416,
+      "grad_norm": 0.57421875,
+      "learning_rate": 1.0535555555555557e-05,
+      "loss": 0.0296,
+      "mean_token_accuracy": 0.9916816521435976,
+      "num_tokens": 175357102.0,
+      "step": 5260
+    },
+    {
+      "entropy": 0.05305887761060148,
+      "epoch": 8.432,
+      "grad_norm": 0.498046875,
+      "learning_rate": 1.0513333333333333e-05,
+      "loss": 0.0305,
+      "mean_token_accuracy": 0.9916395671665669,
+      "num_tokens": 175687673.0,
+      "step": 5270
+    },
+    {
+      "entropy": 0.0521154714981094,
+      "epoch": 8.448,
+      "grad_norm": 0.5390625,
+      "learning_rate": 1.0491111111111112e-05,
+      "loss": 0.03,
+      "mean_token_accuracy": 0.9917172603309155,
+      "num_tokens": 176023153.0,
+      "step": 5280
+    },
+    {
+      "entropy": 0.053183228278066964,
+      "epoch": 8.464,
+      "grad_norm": 0.6015625,
+      "learning_rate": 1.046888888888889e-05,
+      "loss": 0.0307,
+      "mean_token_accuracy": 0.9912721127271652,
+      "num_tokens": 176353603.0,
+      "step": 5290
+    },
+    {
+      "entropy": 0.05332763209007681,
+      "epoch": 8.48,
+      "grad_norm": 0.451171875,
+      "learning_rate": 1.0446666666666668e-05,
+      "loss": 0.0303,
+      "mean_token_accuracy": 0.9914166141301394,
+      "num_tokens": 176682767.0,
+      "step": 5300
+    },
+    {
+      "epoch": 8.48,
+      "eval_chemistry_entropy": 0.23878271359205247,
+      "eval_chemistry_loss": 1.353937029838562,
+      "eval_chemistry_mean_token_accuracy": 0.786796718120575,
+      "eval_chemistry_num_tokens": 176682767.0,
+      "eval_chemistry_runtime": 34.5972,
+      "eval_chemistry_samples_per_second": 14.452,
+      "eval_chemistry_steps_per_second": 3.613,
+      "step": 5300
+    },
+    {
+      "epoch": 8.48,
+      "eval_physics_entropy": 0.18226520359516143,
+      "eval_physics_loss": 1.1384316682815552,
+      "eval_physics_mean_token_accuracy": 0.8236245818138123,
+      "eval_physics_num_tokens": 176682767.0,
+      "eval_physics_runtime": 41.2286,
+      "eval_physics_samples_per_second": 12.127,
+      "eval_physics_steps_per_second": 3.032,
+      "step": 5300
+    },
+    {
+      "entropy": 0.05248126632068306,
+      "epoch": 8.496,
+      "grad_norm": 0.86328125,
+      "learning_rate": 1.0424444444444445e-05,
+      "loss": 0.0301,
+      "mean_token_accuracy": 0.9916587069630622,
+      "num_tokens": 177016406.0,
+      "step": 5310
+    },
+    {
+      "entropy": 0.052710394049063324,
+      "epoch": 8.512,
+      "grad_norm": 0.419921875,
+      "learning_rate": 1.0402222222222223e-05,
+      "loss": 0.0302,
+      "mean_token_accuracy": 0.991531053185463,
+      "num_tokens": 177351834.0,
+      "step": 5320
+    },
+    {
+      "entropy": 0.05350536028854549,
+      "epoch": 8.528,
+      "grad_norm": 0.78125,
+      "learning_rate": 1.038e-05,
+      "loss": 0.0307,
+      "mean_token_accuracy": 0.9914602119475603,
+      "num_tokens": 177681702.0,
+      "step": 5330
+    },
+    {
+      "entropy": 0.05140132904052734,
+      "epoch": 8.544,
+      "grad_norm": 0.486328125,
+      "learning_rate": 1.0357777777777778e-05,
+      "loss": 0.0294,
+      "mean_token_accuracy": 0.9917555499821902,
+      "num_tokens": 178024037.0,
+      "step": 5340
+    },
+    {
+      "entropy": 0.05117007324006408,
+      "epoch": 8.56,
+      "grad_norm": 0.484375,
+      "learning_rate": 1.0335555555555556e-05,
+      "loss": 0.0293,
+      "mean_token_accuracy": 0.9919215857982635,
+      "num_tokens": 178367229.0,
+      "step": 5350
+    },
+    {
+      "entropy": 0.05305402360390872,
+      "epoch": 8.576,
+      "grad_norm": 0.58203125,
+      "learning_rate": 1.0313333333333335e-05,
+      "loss": 0.0304,
+      "mean_token_accuracy": 0.9916315987706185,
+      "num_tokens": 178696766.0,
+      "step": 5360
+    },
+    {
+      "entropy": 0.05206144831608981,
+      "epoch": 8.592,
+      "grad_norm": 0.61328125,
+      "learning_rate": 1.0291111111111111e-05,
+      "loss": 0.0301,
+      "mean_token_accuracy": 0.9917347569018602,
+      "num_tokens": 179032286.0,
+      "step": 5370
+    },
+    {
+      "entropy": 0.05293563305167481,
+      "epoch": 8.608,
+      "grad_norm": 0.51953125,
+      "learning_rate": 1.026888888888889e-05,
+      "loss": 0.0298,
+      "mean_token_accuracy": 0.9915457583963871,
+      "num_tokens": 179368970.0,
+      "step": 5380
+    },
+    {
+      "entropy": 0.05167751736007631,
+      "epoch": 8.624,
+      "grad_norm": 0.57421875,
+      "learning_rate": 1.0246666666666666e-05,
+      "loss": 0.0298,
+      "mean_token_accuracy": 0.9917129289358855,
+      "num_tokens": 179708379.0,
+      "step": 5390
+    },
+    {
+      "entropy": 0.0511523469700478,
+      "epoch": 8.64,
+      "grad_norm": 0.46484375,
+      "learning_rate": 1.0224444444444446e-05,
+      "loss": 0.0294,
+      "mean_token_accuracy": 0.9919034983962774,
+      "num_tokens": 180052267.0,
+      "step": 5400
+    },
+    {
+      "epoch": 8.64,
+      "eval_chemistry_entropy": 0.23853224849700927,
+      "eval_chemistry_loss": 1.3539738655090332,
+      "eval_chemistry_mean_token_accuracy": 0.7867292647361755,
+      "eval_chemistry_num_tokens": 180052267.0,
+      "eval_chemistry_runtime": 34.5931,
+      "eval_chemistry_samples_per_second": 14.454,
+      "eval_chemistry_steps_per_second": 3.613,
+      "step": 5400
+    },
+    {
+      "epoch": 8.64,
+      "eval_physics_entropy": 0.18208700734376906,
+      "eval_physics_loss": 1.1386479139328003,
+      "eval_physics_mean_token_accuracy": 0.823716799736023,
+      "eval_physics_num_tokens": 180052267.0,
+      "eval_physics_runtime": 41.2348,
+      "eval_physics_samples_per_second": 12.126,
+      "eval_physics_steps_per_second": 3.031,
+      "step": 5400
+    },
+    {
+      "entropy": 0.05319649162702263,
+      "epoch": 8.656,
+      "grad_norm": 0.484375,
+      "learning_rate": 1.0202222222222221e-05,
+      "loss": 0.0305,
+      "mean_token_accuracy": 0.9915910948067903,
+      "num_tokens": 180384758.0,
+      "step": 5410
+    },
+    {
+      "entropy": 0.05291957095032558,
+      "epoch": 8.672,
+      "grad_norm": 0.5546875,
+      "learning_rate": 1.018e-05,
+      "loss": 0.0298,
+      "mean_token_accuracy": 0.9915907006710768,
+      "num_tokens": 180720042.0,
+      "step": 5420
+    },
+    {
+      "entropy": 0.052251234836876395,
+      "epoch": 8.688,
+      "grad_norm": 0.44140625,
+      "learning_rate": 1.0157777777777777e-05,
+      "loss": 0.0298,
+      "mean_token_accuracy": 0.9916108876466752,
+      "num_tokens": 181056448.0,
+      "step": 5430
+    },
+    {
+      "entropy": 0.053831005631946025,
+      "epoch": 8.704,
+      "grad_norm": 0.53125,
+      "learning_rate": 1.0135555555555556e-05,
+      "loss": 0.0306,
+      "mean_token_accuracy": 0.9913984097540378,
+      "num_tokens": 181385655.0,
+      "step": 5440
+    },
+    {
+      "entropy": 0.056271852529607716,
+      "epoch": 8.72,
+      "grad_norm": 0.578125,
+      "learning_rate": 1.0113333333333335e-05,
+      "loss": 0.0326,
+      "mean_token_accuracy": 0.9909847188740969,
+      "num_tokens": 181693788.0,
+      "step": 5450
+    },
+    {
+      "entropy": 0.05248148359823972,
+      "epoch": 8.736,
+      "grad_norm": 0.474609375,
+      "learning_rate": 1.0091111111111111e-05,
+      "loss": 0.0302,
+      "mean_token_accuracy": 0.9916448578238487,
+      "num_tokens": 182029600.0,
+      "step": 5460
+    },
+    {
+      "entropy": 0.05381674263626337,
+      "epoch": 8.752,
+      "grad_norm": 1.8125,
+      "learning_rate": 1.006888888888889e-05,
+      "loss": 0.0311,
+      "mean_token_accuracy": 0.9913179915398359,
+      "num_tokens": 182364272.0,
+      "step": 5470
+    },
+    {
+      "entropy": 0.05453286885749549,
+      "epoch": 8.768,
+      "grad_norm": 0.54296875,
+      "learning_rate": 1.0046666666666666e-05,
+      "loss": 0.0315,
+      "mean_token_accuracy": 0.9912139128893613,
+      "num_tokens": 182683058.0,
+      "step": 5480
+    },
+    {
+      "entropy": 0.053682068991474804,
+      "epoch": 8.784,
+      "grad_norm": 0.50390625,
+      "learning_rate": 1.0024444444444446e-05,
+      "loss": 0.0305,
+      "mean_token_accuracy": 0.9916103724390268,
+      "num_tokens": 183015151.0,
+      "step": 5490
+    },
+    {
+      "entropy": 0.0537078810390085,
+      "epoch": 8.8,
+      "grad_norm": 0.53125,
+      "learning_rate": 1.0002222222222222e-05,
+      "loss": 0.0303,
+      "mean_token_accuracy": 0.9913626555353403,
+      "num_tokens": 183347615.0,
+      "step": 5500
+    },
+    {
+      "epoch": 8.8,
+      "eval_chemistry_entropy": 0.23869588297605515,
+      "eval_chemistry_loss": 1.3565541505813599,
+      "eval_chemistry_mean_token_accuracy": 0.7877261233329773,
+      "eval_chemistry_num_tokens": 183347615.0,
+      "eval_chemistry_runtime": 34.5834,
+      "eval_chemistry_samples_per_second": 14.458,
+      "eval_chemistry_steps_per_second": 3.614,
+      "step": 5500
+    },
+    {
+      "epoch": 8.8,
+      "eval_physics_entropy": 0.18198584145307542,
+      "eval_physics_loss": 1.1420674324035645,
+      "eval_physics_mean_token_accuracy": 0.8237350630760193,
+      "eval_physics_num_tokens": 183347615.0,
+      "eval_physics_runtime": 40.9614,
+      "eval_physics_samples_per_second": 12.207,
+      "eval_physics_steps_per_second": 3.052,
+      "step": 5500
+    },
+    {
+      "entropy": 0.05384293962270022,
+      "epoch": 8.816,
+      "grad_norm": 0.4765625,
+      "learning_rate": 9.980000000000001e-06,
+      "loss": 0.0312,
+      "mean_token_accuracy": 0.9912165228277445,
+      "num_tokens": 183672390.0,
+      "step": 5510
+    },
+    {
+      "entropy": 0.0520067551638931,
+      "epoch": 8.832,
+      "grad_norm": 0.416015625,
+      "learning_rate": 9.957777777777779e-06,
+      "loss": 0.0302,
+      "mean_token_accuracy": 0.9915386583656073,
+      "num_tokens": 184006898.0,
+      "step": 5520
+    },
+    {
+      "entropy": 0.05241824253462255,
+      "epoch": 8.848,
+      "grad_norm": 0.490234375,
+      "learning_rate": 9.935555555555556e-06,
+      "loss": 0.0304,
+      "mean_token_accuracy": 0.9914692666381597,
+      "num_tokens": 184339825.0,
+      "step": 5530
+    },
+    {
+      "entropy": 0.052429745125118644,
+      "epoch": 8.864,
+      "grad_norm": 0.439453125,
+      "learning_rate": 9.913333333333334e-06,
+      "loss": 0.0305,
+      "mean_token_accuracy": 0.991571880504489,
+      "num_tokens": 184672114.0,
+      "step": 5540
+    },
+    {
+      "entropy": 0.05228077264036983,
+      "epoch": 8.88,
+      "grad_norm": 0.45703125,
+      "learning_rate": 9.891111111111113e-06,
+      "loss": 0.0298,
+      "mean_token_accuracy": 0.9916410472244024,
+      "num_tokens": 185011915.0,
+      "step": 5550
+    },
+    {
+      "entropy": 0.05289017081959173,
+      "epoch": 8.896,
+      "grad_norm": 0.62890625,
+      "learning_rate": 9.86888888888889e-06,
+      "loss": 0.0306,
+      "mean_token_accuracy": 0.9916071906685829,
+      "num_tokens": 185341169.0,
+      "step": 5560
+    },
+    {
+      "entropy": 0.052123815612867476,
+      "epoch": 8.912,
+      "grad_norm": 0.462890625,
+      "learning_rate": 9.846666666666668e-06,
+      "loss": 0.0309,
+      "mean_token_accuracy": 0.9914463710039854,
+      "num_tokens": 185671338.0,
+      "step": 5570
+    },
+    {
+      "entropy": 0.053904051275458185,
+      "epoch": 8.928,
+      "grad_norm": 0.51953125,
+      "learning_rate": 9.824444444444446e-06,
+      "loss": 0.0311,
+      "mean_token_accuracy": 0.9912458606064319,
+      "num_tokens": 185998244.0,
+      "step": 5580
+    },
+    {
+      "entropy": 0.05394856226630509,
+      "epoch": 8.943999999999999,
+      "grad_norm": 0.8828125,
+      "learning_rate": 9.802222222222224e-06,
+      "loss": 0.0312,
+      "mean_token_accuracy": 0.9912892255932093,
+      "num_tokens": 186329294.0,
+      "step": 5590
+    },
+    {
+      "entropy": 0.0512017457280308,
+      "epoch": 8.96,
+      "grad_norm": 0.470703125,
+      "learning_rate": 9.780000000000001e-06,
+      "loss": 0.0298,
+      "mean_token_accuracy": 0.991724006086588,
+      "num_tokens": 186668349.0,
+      "step": 5600
+    },
+    {
+      "epoch": 8.96,
+      "eval_chemistry_entropy": 0.23852831137180328,
+      "eval_chemistry_loss": 1.356632113456726,
+      "eval_chemistry_mean_token_accuracy": 0.7870347862243652,
+      "eval_chemistry_num_tokens": 186668349.0,
+      "eval_chemistry_runtime": 34.6058,
+      "eval_chemistry_samples_per_second": 14.448,
+      "eval_chemistry_steps_per_second": 3.612,
+      "step": 5600
+    },
+    {
+      "epoch": 8.96,
+      "eval_physics_entropy": 0.1819812126159668,
+      "eval_physics_loss": 1.141103982925415,
+      "eval_physics_mean_token_accuracy": 0.8235724921226502,
+      "eval_physics_num_tokens": 186668349.0,
+      "eval_physics_runtime": 41.2233,
+      "eval_physics_samples_per_second": 12.129,
+      "eval_physics_steps_per_second": 3.032,
+      "step": 5600
+    },
+    {
+      "entropy": 0.053236753307282926,
+      "epoch": 8.975999999999999,
+      "grad_norm": 0.5625,
+      "learning_rate": 9.757777777777779e-06,
+      "loss": 0.0307,
+      "mean_token_accuracy": 0.9913825329393149,
+      "num_tokens": 187004547.0,
+      "step": 5610
+    },
+    {
+      "entropy": 0.053915074467658995,
+      "epoch": 8.992,
+      "grad_norm": 0.435546875,
+      "learning_rate": 9.735555555555556e-06,
+      "loss": 0.0311,
+      "mean_token_accuracy": 0.9912134103477002,
+      "num_tokens": 187329380.0,
+      "step": 5620
+    },
+    {
+      "entropy": 0.05150243304669857,
+      "epoch": 9.008,
+      "grad_norm": 0.3046875,
+      "learning_rate": 9.713333333333334e-06,
+      "loss": 0.0294,
+      "mean_token_accuracy": 0.9918423641473055,
+      "num_tokens": 187662920.0,
+      "step": 5630
+    },
+    {
+      "entropy": 0.049462697317358105,
+      "epoch": 9.024,
+      "grad_norm": 0.451171875,
+      "learning_rate": 9.691111111111112e-06,
+      "loss": 0.0272,
+      "mean_token_accuracy": 0.9923757087439299,
+      "num_tokens": 188005792.0,
+      "step": 5640
+    },
+    {
+      "entropy": 0.04971582887228578,
+      "epoch": 9.04,
+      "grad_norm": 0.390625,
+      "learning_rate": 9.66888888888889e-06,
+      "loss": 0.028,
+      "mean_token_accuracy": 0.9921501968055964,
+      "num_tokens": 188339916.0,
+      "step": 5650
+    },
+    {
+      "entropy": 0.047379994322545825,
+      "epoch": 9.056,
+      "grad_norm": 0.380859375,
+      "learning_rate": 9.646666666666667e-06,
+      "loss": 0.0271,
+      "mean_token_accuracy": 0.9924122154712677,
+      "num_tokens": 188683490.0,
+      "step": 5660
+    },
+    {
+      "entropy": 0.048656794452108444,
+      "epoch": 9.072,
+      "grad_norm": 0.41796875,
+      "learning_rate": 9.624444444444445e-06,
+      "loss": 0.0274,
+      "mean_token_accuracy": 0.9923512354493141,
+      "num_tokens": 189026923.0,
+      "step": 5670
+    },
+    {
+      "entropy": 0.04907445493154228,
+      "epoch": 9.088,
+      "grad_norm": 0.57421875,
+      "learning_rate": 9.602222222222222e-06,
+      "loss": 0.0278,
+      "mean_token_accuracy": 0.9921332910656929,
+      "num_tokens": 189365025.0,
+      "step": 5680
+    },
+    {
+      "entropy": 0.049528668611310424,
+      "epoch": 9.104,
+      "grad_norm": 0.419921875,
+      "learning_rate": 9.58e-06,
+      "loss": 0.0284,
+      "mean_token_accuracy": 0.9921406161040067,
+      "num_tokens": 189692757.0,
+      "step": 5690
+    },
+    {
+      "entropy": 0.049729556345846505,
+      "epoch": 9.12,
+      "grad_norm": 0.4140625,
+      "learning_rate": 9.557777777777777e-06,
+      "loss": 0.0277,
+      "mean_token_accuracy": 0.9921250183135271,
+      "num_tokens": 190034454.0,
+      "step": 5700
+    },
+    {
+      "epoch": 9.12,
+      "eval_chemistry_entropy": 0.23608072745800018,
+      "eval_chemistry_loss": 1.3717834949493408,
+      "eval_chemistry_mean_token_accuracy": 0.7867323513031006,
+      "eval_chemistry_num_tokens": 190034454.0,
+      "eval_chemistry_runtime": 34.6055,
+      "eval_chemistry_samples_per_second": 14.449,
+      "eval_chemistry_steps_per_second": 3.612,
+      "step": 5700
+    },
+    {
+      "epoch": 9.12,
+      "eval_physics_entropy": 0.17964074355363846,
+      "eval_physics_loss": 1.1558970212936401,
+      "eval_physics_mean_token_accuracy": 0.8236214294433594,
+      "eval_physics_num_tokens": 190034454.0,
+      "eval_physics_runtime": 41.2472,
+      "eval_physics_samples_per_second": 12.122,
+      "eval_physics_steps_per_second": 3.031,
+      "step": 5700
+    },
+    {
+      "entropy": 0.05052087816875428,
+      "epoch": 9.136,
+      "grad_norm": 0.37890625,
+      "learning_rate": 9.535555555555557e-06,
+      "loss": 0.0286,
+      "mean_token_accuracy": 0.9919232524931431,
+      "num_tokens": 190365404.0,
+      "step": 5710
+    },
+    {
+      "entropy": 0.04877695455215871,
+      "epoch": 9.152,
+      "grad_norm": 0.375,
+      "learning_rate": 9.513333333333334e-06,
+      "loss": 0.0276,
+      "mean_token_accuracy": 0.9921815037727356,
+      "num_tokens": 190708172.0,
+      "step": 5720
+    },
+    {
+      "entropy": 0.049999333673622456,
+      "epoch": 9.168,
+      "grad_norm": 0.412109375,
+      "learning_rate": 9.491111111111112e-06,
+      "loss": 0.0283,
+      "mean_token_accuracy": 0.9920229602605104,
+      "num_tokens": 191041730.0,
+      "step": 5730
+    },
+    {
+      "entropy": 0.049637012812308964,
+      "epoch": 9.184,
+      "grad_norm": 0.427734375,
+      "learning_rate": 9.46888888888889e-06,
+      "loss": 0.0289,
+      "mean_token_accuracy": 0.9919045221060514,
+      "num_tokens": 191368269.0,
+      "step": 5740
+    },
+    {
+      "entropy": 0.04960813039215282,
+      "epoch": 9.2,
+      "grad_norm": 0.400390625,
+      "learning_rate": 9.446666666666667e-06,
+      "loss": 0.0284,
+      "mean_token_accuracy": 0.9920790564268828,
+      "num_tokens": 191702437.0,
+      "step": 5750
+    },
+    {
+      "entropy": 0.051444963028188795,
+      "epoch": 9.216,
+      "grad_norm": 0.43359375,
+      "learning_rate": 9.424444444444445e-06,
+      "loss": 0.029,
+      "mean_token_accuracy": 0.9917234398424626,
+      "num_tokens": 192027360.0,
+      "step": 5760
+    },
+    {
+      "entropy": 0.04831572223920375,
+      "epoch": 9.232,
+      "grad_norm": 0.400390625,
+      "learning_rate": 9.402222222222222e-06,
+      "loss": 0.0273,
+      "mean_token_accuracy": 0.9922237403690815,
+      "num_tokens": 192374053.0,
+      "step": 5770
+    },
+    {
+      "entropy": 0.04941471564816311,
+      "epoch": 9.248,
+      "grad_norm": 0.388671875,
+      "learning_rate": 9.38e-06,
+      "loss": 0.0283,
+      "mean_token_accuracy": 0.9920372307300568,
+      "num_tokens": 192708001.0,
+      "step": 5780
+    },
+    {
+      "entropy": 0.05107759142993018,
+      "epoch": 9.264,
+      "grad_norm": 0.458984375,
+      "learning_rate": 9.35777777777778e-06,
+      "loss": 0.0291,
+      "mean_token_accuracy": 0.9917293023318052,
+      "num_tokens": 193033414.0,
+      "step": 5790
+    },
+    {
+      "entropy": 0.0500594254815951,
+      "epoch": 9.28,
+      "grad_norm": 0.44140625,
+      "learning_rate": 9.335555555555557e-06,
+      "loss": 0.0287,
+      "mean_token_accuracy": 0.9919759843498468,
+      "num_tokens": 193359971.0,
+      "step": 5800
+    },
+    {
+      "epoch": 9.28,
+      "eval_chemistry_entropy": 0.235800461769104,
+      "eval_chemistry_loss": 1.3743587732315063,
+      "eval_chemistry_mean_token_accuracy": 0.7867599420547485,
+      "eval_chemistry_num_tokens": 193359971.0,
+      "eval_chemistry_runtime": 34.7169,
+      "eval_chemistry_samples_per_second": 14.402,
+      "eval_chemistry_steps_per_second": 3.601,
+      "step": 5800
+    },
+    {
+      "epoch": 9.28,
+      "eval_physics_entropy": 0.17898882150650025,
+      "eval_physics_loss": 1.1594042778015137,
+      "eval_physics_mean_token_accuracy": 0.8237328319549561,
+      "eval_physics_num_tokens": 193359971.0,
+      "eval_physics_runtime": 41.581,
+      "eval_physics_samples_per_second": 12.025,
+      "eval_physics_steps_per_second": 3.006,
+      "step": 5800
+    },
+    {
+      "entropy": 0.04959448673762381,
+      "epoch": 9.296,
+      "grad_norm": 0.37109375,
+      "learning_rate": 9.313333333333335e-06,
+      "loss": 0.0282,
+      "mean_token_accuracy": 0.9920639682561159,
+      "num_tokens": 193696382.0,
+      "step": 5810
+    },
+    {
+      "entropy": 0.050598076928872616,
+      "epoch": 9.312,
+      "grad_norm": 0.388671875,
+      "learning_rate": 9.291111111111112e-06,
+      "loss": 0.0285,
+      "mean_token_accuracy": 0.991863951832056,
+      "num_tokens": 194028242.0,
+      "step": 5820
+    },
+    {
+      "entropy": 0.04999363364186138,
+      "epoch": 9.328,
+      "grad_norm": 0.419921875,
+      "learning_rate": 9.26888888888889e-06,
+      "loss": 0.029,
+      "mean_token_accuracy": 0.9917803280055523,
+      "num_tokens": 194354796.0,
+      "step": 5830
+    },
+    {
+      "entropy": 0.0499969640863128,
+      "epoch": 9.344,
+      "grad_norm": 0.4609375,
+      "learning_rate": 9.246666666666667e-06,
+      "loss": 0.0286,
+      "mean_token_accuracy": 0.9919028174132108,
+      "num_tokens": 194685345.0,
+      "step": 5840
+    },
+    {
+      "entropy": 0.05002880855463445,
+      "epoch": 9.36,
+      "grad_norm": 0.455078125,
+      "learning_rate": 9.224444444444445e-06,
+      "loss": 0.029,
+      "mean_token_accuracy": 0.9917697183787823,
+      "num_tokens": 195013073.0,
+      "step": 5850
+    },
+    {
+      "entropy": 0.05089859606232494,
+      "epoch": 9.376,
+      "grad_norm": 0.453125,
+      "learning_rate": 9.202222222222224e-06,
+      "loss": 0.0291,
+      "mean_token_accuracy": 0.9918681979179382,
+      "num_tokens": 195337490.0,
+      "step": 5860
+    },
+    {
+      "entropy": 0.0501430586562492,
+      "epoch": 9.392,
+      "grad_norm": 0.3671875,
+      "learning_rate": 9.180000000000002e-06,
+      "loss": 0.0286,
+      "mean_token_accuracy": 0.9919176645576954,
+      "num_tokens": 195670838.0,
+      "step": 5870
+    },
+    {
+      "entropy": 0.050096994230989365,
+      "epoch": 9.408,
+      "grad_norm": 0.435546875,
+      "learning_rate": 9.15777777777778e-06,
+      "loss": 0.0286,
+      "mean_token_accuracy": 0.991869680210948,
+      "num_tokens": 196003346.0,
+      "step": 5880
+    },
+    {
+      "entropy": 0.0507117866538465,
+      "epoch": 9.424,
+      "grad_norm": 0.388671875,
+      "learning_rate": 9.135555555555557e-06,
+      "loss": 0.0292,
+      "mean_token_accuracy": 0.9917945537716151,
+      "num_tokens": 196329761.0,
+      "step": 5890
+    },
+    {
+      "entropy": 0.05021356453653425,
+      "epoch": 9.44,
+      "grad_norm": 0.390625,
+      "learning_rate": 9.113333333333335e-06,
+      "loss": 0.0285,
+      "mean_token_accuracy": 0.9919662341475487,
+      "num_tokens": 196662622.0,
+      "step": 5900
+    },
+    {
+      "epoch": 9.44,
+      "eval_chemistry_entropy": 0.2353560220003128,
+      "eval_chemistry_loss": 1.375707745552063,
+      "eval_chemistry_mean_token_accuracy": 0.7867635736465454,
+      "eval_chemistry_num_tokens": 196662622.0,
+      "eval_chemistry_runtime": 34.683,
+      "eval_chemistry_samples_per_second": 14.416,
+      "eval_chemistry_steps_per_second": 3.604,
+      "step": 5900
+    },
+    {
+      "epoch": 9.44,
+      "eval_physics_entropy": 0.17858504605293274,
+      "eval_physics_loss": 1.1612944602966309,
+      "eval_physics_mean_token_accuracy": 0.8237157382965088,
+      "eval_physics_num_tokens": 196662622.0,
+      "eval_physics_runtime": 41.4342,
+      "eval_physics_samples_per_second": 12.067,
+      "eval_physics_steps_per_second": 3.017,
+      "step": 5900
+    },
+    {
+      "entropy": 0.04900284027680755,
+      "epoch": 9.456,
+      "grad_norm": 0.451171875,
+      "learning_rate": 9.091111111111112e-06,
+      "loss": 0.0283,
+      "mean_token_accuracy": 0.9920027501881122,
+      "num_tokens": 196998557.0,
+      "step": 5910
+    },
+    {
+      "entropy": 0.050040658959187564,
+      "epoch": 9.472,
+      "grad_norm": 0.419921875,
+      "learning_rate": 9.06888888888889e-06,
+      "loss": 0.0294,
+      "mean_token_accuracy": 0.9916663836687803,
+      "num_tokens": 197322082.0,
+      "step": 5920
+    },
+    {
+      "entropy": 0.048204546328634024,
+      "epoch": 9.488,
+      "grad_norm": 0.47265625,
+      "learning_rate": 9.046666666666668e-06,
+      "loss": 0.0282,
+      "mean_token_accuracy": 0.9922185614705086,
+      "num_tokens": 197658039.0,
+      "step": 5930
+    },
+    {
+      "entropy": 0.050434952031355354,
+      "epoch": 9.504,
+      "grad_norm": 0.400390625,
+      "learning_rate": 9.024444444444445e-06,
+      "loss": 0.0291,
+      "mean_token_accuracy": 0.9918316569179296,
+      "num_tokens": 197984085.0,
+      "step": 5940
+    },
+    {
+      "entropy": 0.050668040954042225,
+      "epoch": 9.52,
+      "grad_norm": 0.4296875,
+      "learning_rate": 9.002222222222223e-06,
+      "loss": 0.0289,
+      "mean_token_accuracy": 0.9917506299912929,
+      "num_tokens": 198315173.0,
+      "step": 5950
+    },
+    {
+      "entropy": 0.049574420391581955,
+      "epoch": 9.536,
+      "grad_norm": 0.380859375,
+      "learning_rate": 8.98e-06,
+      "loss": 0.0287,
+      "mean_token_accuracy": 0.9918869096785784,
+      "num_tokens": 198646459.0,
+      "step": 5960
+    },
+    {
+      "entropy": 0.04931255161063745,
+      "epoch": 9.552,
+      "grad_norm": 0.3828125,
+      "learning_rate": 8.957777777777778e-06,
+      "loss": 0.0282,
+      "mean_token_accuracy": 0.9918757170438767,
+      "num_tokens": 198985080.0,
+      "step": 5970
+    },
+    {
+      "entropy": 0.051441012998111546,
+      "epoch": 9.568,
+      "grad_norm": 0.4453125,
+      "learning_rate": 8.935555555555556e-06,
+      "loss": 0.0296,
+      "mean_token_accuracy": 0.9914629425853491,
+      "num_tokens": 199306200.0,
+      "step": 5980
+    },
+    {
+      "entropy": 0.04893468508962542,
+      "epoch": 9.584,
+      "grad_norm": 0.423828125,
+      "learning_rate": 8.913333333333333e-06,
+      "loss": 0.0286,
+      "mean_token_accuracy": 0.991978258267045,
+      "num_tokens": 199641017.0,
+      "step": 5990
+    },
+    {
+      "entropy": 0.049285993294324724,
+      "epoch": 9.6,
+      "grad_norm": 0.439453125,
+      "learning_rate": 8.891111111111111e-06,
+      "loss": 0.0282,
+      "mean_token_accuracy": 0.9919727381318808,
+      "num_tokens": 199977866.0,
+      "step": 6000
+    },
+    {
+      "epoch": 9.6,
+      "eval_chemistry_entropy": 0.23499409383535386,
+      "eval_chemistry_loss": 1.3759660720825195,
+      "eval_chemistry_mean_token_accuracy": 0.7869590911865234,
+      "eval_chemistry_num_tokens": 199977866.0,
+      "eval_chemistry_runtime": 34.7563,
+      "eval_chemistry_samples_per_second": 14.386,
+      "eval_chemistry_steps_per_second": 3.596,
+      "step": 6000
+    },
+    {
+      "epoch": 9.6,
+      "eval_physics_entropy": 0.1785295672416687,
+      "eval_physics_loss": 1.161868691444397,
+      "eval_physics_mean_token_accuracy": 0.8237499132156372,
+      "eval_physics_num_tokens": 199977866.0,
+      "eval_physics_runtime": 41.0689,
+      "eval_physics_samples_per_second": 12.175,
+      "eval_physics_steps_per_second": 3.044,
+      "step": 6000
+    },
+    {
+      "entropy": 0.04930304625304416,
+      "epoch": 9.616,
+      "grad_norm": 0.4296875,
+      "learning_rate": 8.868888888888888e-06,
+      "loss": 0.0285,
+      "mean_token_accuracy": 0.9919850662350654,
+      "num_tokens": 200316537.0,
+      "step": 6010
+    },
+    {
+      "entropy": 0.05006378333782777,
+      "epoch": 9.632,
+      "grad_norm": 0.42578125,
+      "learning_rate": 8.846666666666668e-06,
+      "loss": 0.0284,
+      "mean_token_accuracy": 0.9918942846357822,
+      "num_tokens": 200655952.0,
+      "step": 6020
+    },
+    {
+      "entropy": 0.05047245887108147,
+      "epoch": 9.648,
+      "grad_norm": 0.373046875,
+      "learning_rate": 8.824444444444445e-06,
+      "loss": 0.0292,
+      "mean_token_accuracy": 0.9916995711624622,
+      "num_tokens": 200984957.0,
+      "step": 6030
+    },
+    {
+      "entropy": 0.04898955838289112,
+      "epoch": 9.664,
+      "grad_norm": 0.431640625,
+      "learning_rate": 8.802222222222223e-06,
+      "loss": 0.0287,
+      "mean_token_accuracy": 0.9919974647462368,
+      "num_tokens": 201319278.0,
+      "step": 6040
+    },
+    {
+      "entropy": 0.0491935983998701,
+      "epoch": 9.68,
+      "grad_norm": 0.427734375,
+      "learning_rate": 8.78e-06,
+      "loss": 0.0285,
+      "mean_token_accuracy": 0.9919443521648645,
+      "num_tokens": 201655605.0,
+      "step": 6050
+    },
+    {
+      "entropy": 0.05007319065043703,
+      "epoch": 9.696,
+      "grad_norm": 0.392578125,
+      "learning_rate": 8.757777777777778e-06,
+      "loss": 0.0287,
+      "mean_token_accuracy": 0.9918694365769625,
+      "num_tokens": 201989059.0,
+      "step": 6060
+    },
+    {
+      "entropy": 0.050567583518568424,
+      "epoch": 9.712,
+      "grad_norm": 0.478515625,
+      "learning_rate": 8.735555555555556e-06,
+      "loss": 0.0292,
+      "mean_token_accuracy": 0.991656395047903,
+      "num_tokens": 202316614.0,
+      "step": 6070
+    },
+    {
+      "entropy": 0.05005731907440349,
+      "epoch": 9.728,
+      "grad_norm": 1.0078125,
+      "learning_rate": 8.713333333333333e-06,
+      "loss": 0.0292,
+      "mean_token_accuracy": 0.9916078060865402,
+      "num_tokens": 202649913.0,
+      "step": 6080
+    },
+    {
+      "entropy": 0.048602585680782795,
+      "epoch": 9.744,
+      "grad_norm": 0.48046875,
+      "learning_rate": 8.691111111111111e-06,
+      "loss": 0.0286,
+      "mean_token_accuracy": 0.9921061486005783,
+      "num_tokens": 202987761.0,
+      "step": 6090
+    },
+    {
+      "entropy": 0.05005403611576185,
+      "epoch": 9.76,
+      "grad_norm": 0.4375,
+      "learning_rate": 8.66888888888889e-06,
+      "loss": 0.0292,
+      "mean_token_accuracy": 0.9917492404580116,
+      "num_tokens": 203316789.0,
+      "step": 6100
+    },
+    {
+      "epoch": 9.76,
+      "eval_chemistry_entropy": 0.2352653083205223,
+      "eval_chemistry_loss": 1.37761652469635,
+      "eval_chemistry_mean_token_accuracy": 0.7867525358200074,
+      "eval_chemistry_num_tokens": 203316789.0,
+      "eval_chemistry_runtime": 35.0028,
+      "eval_chemistry_samples_per_second": 14.285,
+      "eval_chemistry_steps_per_second": 3.571,
+      "step": 6100
+    },
+    {
+      "epoch": 9.76,
+      "eval_physics_entropy": 0.17874816590547563,
+      "eval_physics_loss": 1.163141131401062,
+      "eval_physics_mean_token_accuracy": 0.8237716612815857,
+      "eval_physics_num_tokens": 203316789.0,
+      "eval_physics_runtime": 41.636,
+      "eval_physics_samples_per_second": 12.009,
+      "eval_physics_steps_per_second": 3.002,
+      "step": 6100
+    },
+    {
+      "entropy": 0.04992901057703421,
+      "epoch": 9.776,
+      "grad_norm": 0.412109375,
+      "learning_rate": 8.646666666666668e-06,
+      "loss": 0.0288,
+      "mean_token_accuracy": 0.9918312799185515,
+      "num_tokens": 203651436.0,
+      "step": 6110
+    },
+    {
+      "entropy": 0.04892389699816704,
+      "epoch": 9.792,
+      "grad_norm": 0.396484375,
+      "learning_rate": 8.624444444444446e-06,
+      "loss": 0.0279,
+      "mean_token_accuracy": 0.9919425990432501,
+      "num_tokens": 203994753.0,
+      "step": 6120
+    },
+    {
+      "entropy": 0.05120947604300454,
+      "epoch": 9.808,
+      "grad_norm": 0.4453125,
+      "learning_rate": 8.602222222222223e-06,
+      "loss": 0.0293,
+      "mean_token_accuracy": 0.9914714429527521,
+      "num_tokens": 204321524.0,
+      "step": 6130
+    },
+    {
+      "entropy": 0.04996306181419641,
+      "epoch": 9.824,
+      "grad_norm": 0.41015625,
+      "learning_rate": 8.580000000000001e-06,
+      "loss": 0.0287,
+      "mean_token_accuracy": 0.99155597910285,
+      "num_tokens": 204656255.0,
+      "step": 6140
+    },
+    {
+      "entropy": 0.04921742535661906,
+      "epoch": 9.84,
+      "grad_norm": 0.390625,
+      "learning_rate": 8.557777777777778e-06,
+      "loss": 0.0285,
+      "mean_token_accuracy": 0.9919517882168293,
+      "num_tokens": 204994423.0,
+      "step": 6150
+    },
+    {
+      "entropy": 0.0506559717701748,
+      "epoch": 9.856,
+      "grad_norm": 0.40625,
+      "learning_rate": 8.535555555555556e-06,
+      "loss": 0.0292,
+      "mean_token_accuracy": 0.9916590839624405,
+      "num_tokens": 205324125.0,
+      "step": 6160
+    },
+    {
+      "entropy": 0.05047166268341243,
+      "epoch": 9.872,
+      "grad_norm": 0.431640625,
+      "learning_rate": 8.513333333333335e-06,
+      "loss": 0.0292,
+      "mean_token_accuracy": 0.9916647884994745,
+      "num_tokens": 205652631.0,
+      "step": 6170
+    },
+    {
+      "entropy": 0.05160766703775153,
+      "epoch": 9.888,
+      "grad_norm": 0.4609375,
+      "learning_rate": 8.491111111111113e-06,
+      "loss": 0.0293,
+      "mean_token_accuracy": 0.9913864985108376,
+      "num_tokens": 205981306.0,
+      "step": 6180
+    },
+    {
+      "entropy": 0.049271152459550646,
+      "epoch": 9.904,
+      "grad_norm": 0.53125,
+      "learning_rate": 8.46888888888889e-06,
+      "loss": 0.0289,
+      "mean_token_accuracy": 0.9918338861316442,
+      "num_tokens": 206314946.0,
+      "step": 6190
+    },
+    {
+      "entropy": 0.04907179562142119,
+      "epoch": 9.92,
+      "grad_norm": 0.388671875,
+      "learning_rate": 8.446666666666668e-06,
+      "loss": 0.0286,
+      "mean_token_accuracy": 0.9919856276363135,
+      "num_tokens": 206651819.0,
+      "step": 6200
+    },
+    {
+      "epoch": 9.92,
+      "eval_chemistry_entropy": 0.23484134083986283,
+      "eval_chemistry_loss": 1.3781731128692627,
+      "eval_chemistry_mean_token_accuracy": 0.7877883834838867,
+      "eval_chemistry_num_tokens": 206651819.0,
+      "eval_chemistry_runtime": 34.8617,
+      "eval_chemistry_samples_per_second": 14.342,
+      "eval_chemistry_steps_per_second": 3.586,
+      "step": 6200
+    },
+    {
+      "epoch": 9.92,
+      "eval_physics_entropy": 0.17827929562330247,
+      "eval_physics_loss": 1.1649293899536133,
+      "eval_physics_mean_token_accuracy": 0.823787766456604,
+      "eval_physics_num_tokens": 206651819.0,
+      "eval_physics_runtime": 41.5685,
+      "eval_physics_samples_per_second": 12.028,
+      "eval_physics_steps_per_second": 3.007,
+      "step": 6200
+    },
+    {
+      "entropy": 0.05131993019022048,
+      "epoch": 9.936,
+      "grad_norm": 0.41796875,
+      "learning_rate": 8.424444444444446e-06,
+      "loss": 0.0295,
+      "mean_token_accuracy": 0.9914973482489586,
+      "num_tokens": 206978472.0,
+      "step": 6210
+    },
+    {
+      "entropy": 0.05075159383704886,
+      "epoch": 9.952,
+      "grad_norm": 0.380859375,
+      "learning_rate": 8.402222222222223e-06,
+      "loss": 0.0292,
+      "mean_token_accuracy": 0.9916805736720562,
+      "num_tokens": 207309818.0,
+      "step": 6220
+    },
+    {
+      "entropy": 0.050251115555875005,
+      "epoch": 9.968,
+      "grad_norm": 0.45703125,
+      "learning_rate": 8.380000000000001e-06,
+      "loss": 0.0289,
+      "mean_token_accuracy": 0.991663160175085,
+      "num_tokens": 207642686.0,
+      "step": 6230
+    },
+    {
+      "entropy": 0.0490652690641582,
+      "epoch": 9.984,
+      "grad_norm": 0.388671875,
+      "learning_rate": 8.357777777777779e-06,
+      "loss": 0.0285,
+      "mean_token_accuracy": 0.9917710069566965,
+      "num_tokens": 207980319.0,
+      "step": 6240
+    },
+    {
+      "entropy": 0.048726341675501314,
+      "epoch": 10.0,
+      "grad_norm": 0.435546875,
+      "learning_rate": 8.335555555555556e-06,
+      "loss": 0.0283,
+      "mean_token_accuracy": 0.9916945673525334,
+      "num_tokens": 208321480.0,
+      "step": 6250
+    },
+    {
+      "entropy": 0.047574253962375225,
+      "epoch": 10.016,
+      "grad_norm": 0.353515625,
+      "learning_rate": 8.313333333333334e-06,
+      "loss": 0.0266,
+      "mean_token_accuracy": 0.9924902655184269,
+      "num_tokens": 208667473.0,
+      "step": 6260
+    },
+    {
+      "entropy": 0.0487474333611317,
+      "epoch": 10.032,
+      "grad_norm": 0.37109375,
+      "learning_rate": 8.291111111111112e-06,
+      "loss": 0.0274,
+      "mean_token_accuracy": 0.9922765307128429,
+      "num_tokens": 209001064.0,
+      "step": 6270
+    },
+    {
+      "entropy": 0.04906477937474847,
+      "epoch": 10.048,
+      "grad_norm": 0.37109375,
+      "learning_rate": 8.268888888888889e-06,
+      "loss": 0.0279,
+      "mean_token_accuracy": 0.9921461876481772,
+      "num_tokens": 209331660.0,
+      "step": 6280
+    },
+    {
+      "entropy": 0.0483110407483764,
+      "epoch": 10.064,
+      "grad_norm": 0.345703125,
+      "learning_rate": 8.246666666666667e-06,
+      "loss": 0.0277,
+      "mean_token_accuracy": 0.9922177352011203,
+      "num_tokens": 209662084.0,
+      "step": 6290
+    },
+    {
+      "entropy": 0.04768453579163179,
+      "epoch": 10.08,
+      "grad_norm": 0.376953125,
+      "learning_rate": 8.224444444444444e-06,
+      "loss": 0.0271,
+      "mean_token_accuracy": 0.9922822460532188,
+      "num_tokens": 210001077.0,
+      "step": 6300
+    },
+    {
+      "epoch": 10.08,
+      "eval_chemistry_entropy": 0.23402770352363586,
+      "eval_chemistry_loss": 1.3871182203292847,
+      "eval_chemistry_mean_token_accuracy": 0.7865667424201965,
+      "eval_chemistry_num_tokens": 210001077.0,
+      "eval_chemistry_runtime": 34.9114,
+      "eval_chemistry_samples_per_second": 14.322,
+      "eval_chemistry_steps_per_second": 3.58,
+      "step": 6300
+    },
+    {
+      "epoch": 10.08,
+      "eval_physics_entropy": 0.17744062381982803,
+      "eval_physics_loss": 1.1722344160079956,
+      "eval_physics_mean_token_accuracy": 0.8235944089889526,
+      "eval_physics_num_tokens": 210001077.0,
+      "eval_physics_runtime": 41.6047,
+      "eval_physics_samples_per_second": 12.018,
+      "eval_physics_steps_per_second": 3.004,
+      "step": 6300
+    },
+    {
+      "entropy": 0.048164489946793766,
+      "epoch": 10.096,
+      "grad_norm": 0.404296875,
+      "learning_rate": 8.202222222222222e-06,
+      "loss": 0.027,
+      "mean_token_accuracy": 0.9921187374740839,
+      "num_tokens": 210341102.0,
+      "step": 6310
+    },
+    {
+      "entropy": 0.04952049327548593,
+      "epoch": 10.112,
+      "grad_norm": 0.400390625,
+      "learning_rate": 8.18e-06,
+      "loss": 0.028,
+      "mean_token_accuracy": 0.9920637808740139,
+      "num_tokens": 210667631.0,
+      "step": 6320
+    },
+    {
+      "entropy": 0.048261914739850906,
+      "epoch": 10.128,
+      "grad_norm": 0.373046875,
+      "learning_rate": 8.157777777777779e-06,
+      "loss": 0.0278,
+      "mean_token_accuracy": 0.9921109657734632,
+      "num_tokens": 210998457.0,
+      "step": 6330
+    },
+    {
+      "entropy": 0.046670902927871794,
+      "epoch": 10.144,
+      "grad_norm": 0.439453125,
+      "learning_rate": 8.135555555555557e-06,
+      "loss": 0.027,
+      "mean_token_accuracy": 0.9923853676766157,
+      "num_tokens": 211338733.0,
+      "step": 6340
+    },
+    {
+      "entropy": 0.048213460179977116,
+      "epoch": 10.16,
+      "grad_norm": 0.41796875,
+      "learning_rate": 8.113333333333334e-06,
+      "loss": 0.0275,
+      "mean_token_accuracy": 0.9921022199094296,
+      "num_tokens": 211674161.0,
+      "step": 6350
+    },
+    {
+      "entropy": 0.04750183164142072,
+      "epoch": 10.176,
+      "grad_norm": 0.39453125,
+      "learning_rate": 8.091111111111112e-06,
+      "loss": 0.0273,
+      "mean_token_accuracy": 0.9922580599784852,
+      "num_tokens": 212011147.0,
+      "step": 6360
+    },
+    {
+      "entropy": 0.048013800324406476,
+      "epoch": 10.192,
+      "grad_norm": 0.408203125,
+      "learning_rate": 8.06888888888889e-06,
+      "loss": 0.0277,
+      "mean_token_accuracy": 0.9921882249414921,
+      "num_tokens": 212344991.0,
+      "step": 6370
+    },
+    {
+      "entropy": 0.04942237173672766,
+      "epoch": 10.208,
+      "grad_norm": 0.423828125,
+      "learning_rate": 8.046666666666667e-06,
+      "loss": 0.0284,
+      "mean_token_accuracy": 0.9919225871562958,
+      "num_tokens": 212668941.0,
+      "step": 6380
+    },
+    {
+      "entropy": 0.04757546194596216,
+      "epoch": 10.224,
+      "grad_norm": 0.388671875,
+      "learning_rate": 8.024444444444445e-06,
+      "loss": 0.0271,
+      "mean_token_accuracy": 0.9922715984284878,
+      "num_tokens": 213008608.0,
+      "step": 6390
+    },
+    {
+      "entropy": 0.04933032592525706,
+      "epoch": 10.24,
+      "grad_norm": 0.41015625,
+      "learning_rate": 8.002222222222222e-06,
+      "loss": 0.0277,
+      "mean_token_accuracy": 0.9919833671301603,
+      "num_tokens": 213342593.0,
+      "step": 6400
+    },
+    {
+      "epoch": 10.24,
+      "eval_chemistry_entropy": 0.23305599641799926,
+      "eval_chemistry_loss": 1.3876862525939941,
+      "eval_chemistry_mean_token_accuracy": 0.786642315864563,
+      "eval_chemistry_num_tokens": 213342593.0,
+      "eval_chemistry_runtime": 34.9006,
+      "eval_chemistry_samples_per_second": 14.326,
+      "eval_chemistry_steps_per_second": 3.582,
+      "step": 6400
+    },
+    {
+      "epoch": 10.24,
+      "eval_physics_entropy": 0.17650638908147812,
+      "eval_physics_loss": 1.1735656261444092,
+      "eval_physics_mean_token_accuracy": 0.823695821762085,
+      "eval_physics_num_tokens": 213342593.0,
+      "eval_physics_runtime": 41.554,
+      "eval_physics_samples_per_second": 12.033,
+      "eval_physics_steps_per_second": 3.008,
+      "step": 6400
+    },
+    {
+      "entropy": 0.04932310256408527,
+      "epoch": 10.256,
+      "grad_norm": 0.392578125,
+      "learning_rate": 7.980000000000002e-06,
+      "loss": 0.0284,
+      "mean_token_accuracy": 0.9918336506932974,
+      "num_tokens": 213669887.0,
+      "step": 6410
+    },
+    {
+      "entropy": 0.04859801979037002,
+      "epoch": 10.272,
+      "grad_norm": 0.43359375,
+      "learning_rate": 7.957777777777779e-06,
+      "loss": 0.0277,
+      "mean_token_accuracy": 0.9920740131288767,
+      "num_tokens": 214005303.0,
+      "step": 6420
+    },
+    {
+      "entropy": 0.047787131369113924,
+      "epoch": 10.288,
+      "grad_norm": 0.375,
+      "learning_rate": 7.935555555555557e-06,
+      "loss": 0.0277,
+      "mean_token_accuracy": 0.9920490551739931,
+      "num_tokens": 214340621.0,
+      "step": 6430
+    },
+    {
+      "entropy": 0.048124253132846205,
+      "epoch": 10.304,
+      "grad_norm": 0.365234375,
+      "learning_rate": 7.913333333333334e-06,
+      "loss": 0.0277,
+      "mean_token_accuracy": 0.992217680066824,
+      "num_tokens": 214676174.0,
+      "step": 6440
+    },
+    {
+      "entropy": 0.049201241170521826,
+      "epoch": 10.32,
+      "grad_norm": 0.47265625,
+      "learning_rate": 7.891111111111112e-06,
+      "loss": 0.0283,
+      "mean_token_accuracy": 0.9918848153203725,
+      "num_tokens": 215002769.0,
+      "step": 6450
+    },
+    {
+      "entropy": 0.048331574397161604,
+      "epoch": 10.336,
+      "grad_norm": 0.41796875,
+      "learning_rate": 7.86888888888889e-06,
+      "loss": 0.0274,
+      "mean_token_accuracy": 0.9921278230845928,
+      "num_tokens": 215342450.0,
+      "step": 6460
+    },
+    {
+      "entropy": 0.05043000646401197,
+      "epoch": 10.352,
+      "grad_norm": 0.416015625,
+      "learning_rate": 7.846666666666667e-06,
+      "loss": 0.0287,
+      "mean_token_accuracy": 0.9915751740336418,
+      "num_tokens": 215667454.0,
+      "step": 6470
+    },
+    {
+      "entropy": 0.047733783698640765,
+      "epoch": 10.368,
+      "grad_norm": 0.375,
+      "learning_rate": 7.824444444444445e-06,
+      "loss": 0.0277,
+      "mean_token_accuracy": 0.9920804053544998,
+      "num_tokens": 216001463.0,
+      "step": 6480
+    },
+    {
+      "entropy": 0.04685619664378464,
+      "epoch": 10.384,
+      "grad_norm": 0.408203125,
+      "learning_rate": 7.802222222222222e-06,
+      "loss": 0.0273,
+      "mean_token_accuracy": 0.9921970557421446,
+      "num_tokens": 216345486.0,
+      "step": 6490
+    },
+    {
+      "entropy": 0.04740955668967217,
+      "epoch": 10.4,
+      "grad_norm": 0.5,
+      "learning_rate": 7.78e-06,
+      "loss": 0.0274,
+      "mean_token_accuracy": 0.9923183683305978,
+      "num_tokens": 216684431.0,
+      "step": 6500
+    },
+    {
+      "epoch": 10.4,
+      "eval_chemistry_entropy": 0.23370064491033554,
+      "eval_chemistry_loss": 1.3870038986206055,
+      "eval_chemistry_mean_token_accuracy": 0.786834554195404,
+      "eval_chemistry_num_tokens": 216684431.0,
+      "eval_chemistry_runtime": 34.8044,
+      "eval_chemistry_samples_per_second": 14.366,
+      "eval_chemistry_steps_per_second": 3.591,
+      "step": 6500
+    },
+    {
+      "epoch": 10.4,
+      "eval_physics_entropy": 0.1771286502480507,
+      "eval_physics_loss": 1.1730924844741821,
+      "eval_physics_mean_token_accuracy": 0.8238932604789734,
+      "eval_physics_num_tokens": 216684431.0,
+      "eval_physics_runtime": 41.1341,
+      "eval_physics_samples_per_second": 12.155,
+      "eval_physics_steps_per_second": 3.039,
+      "step": 6500
+    },
+    {
+      "entropy": 0.04747484175022691,
+      "epoch": 10.416,
+      "grad_norm": 0.421875,
+      "learning_rate": 7.75777777777778e-06,
+      "loss": 0.0271,
+      "mean_token_accuracy": 0.9921270228922368,
+      "num_tokens": 217028320.0,
+      "step": 6510
+    },
+    {
+      "entropy": 0.04852663362398744,
+      "epoch": 10.432,
+      "grad_norm": 0.416015625,
+      "learning_rate": 7.735555555555557e-06,
+      "loss": 0.028,
+      "mean_token_accuracy": 0.9919953543692828,
+      "num_tokens": 217360886.0,
+      "step": 6520
+    },
+    {
+      "entropy": 0.04761390700004995,
+      "epoch": 10.448,
+      "grad_norm": 0.40234375,
+      "learning_rate": 7.713333333333335e-06,
+      "loss": 0.0276,
+      "mean_token_accuracy": 0.9922554593533277,
+      "num_tokens": 217700465.0,
+      "step": 6530
+    },
+    {
+      "entropy": 0.049304455507081,
+      "epoch": 10.464,
+      "grad_norm": 0.451171875,
+      "learning_rate": 7.691111111111112e-06,
+      "loss": 0.0284,
+      "mean_token_accuracy": 0.9919312756508589,
+      "num_tokens": 218028690.0,
+      "step": 6540
+    },
+    {
+      "entropy": 0.05015104198828339,
+      "epoch": 10.48,
+      "grad_norm": 0.416015625,
+      "learning_rate": 7.66888888888889e-06,
+      "loss": 0.029,
+      "mean_token_accuracy": 0.9918051160871982,
+      "num_tokens": 218351668.0,
+      "step": 6550
+    },
+    {
+      "entropy": 0.04980442795203999,
+      "epoch": 10.496,
+      "grad_norm": 0.4375,
+      "learning_rate": 7.646666666666667e-06,
+      "loss": 0.0285,
+      "mean_token_accuracy": 0.9916744876652956,
+      "num_tokens": 218678029.0,
+      "step": 6560
+    },
+    {
+      "entropy": 0.04762532782042399,
+      "epoch": 10.512,
+      "grad_norm": 0.380859375,
+      "learning_rate": 7.624444444444445e-06,
+      "loss": 0.0273,
+      "mean_token_accuracy": 0.9920876882970333,
+      "num_tokens": 219018839.0,
+      "step": 6570
+    },
+    {
+      "entropy": 0.047857774724252525,
+      "epoch": 10.528,
+      "grad_norm": 0.439453125,
+      "learning_rate": 7.602222222222223e-06,
+      "loss": 0.0277,
+      "mean_token_accuracy": 0.9920621104538441,
+      "num_tokens": 219356804.0,
+      "step": 6580
+    },
+    {
+      "entropy": 0.0484519615303725,
+      "epoch": 10.544,
+      "grad_norm": 0.3984375,
+      "learning_rate": 7.58e-06,
+      "loss": 0.0282,
+      "mean_token_accuracy": 0.9919655386358499,
+      "num_tokens": 219688905.0,
+      "step": 6590
+    },
+    {
+      "entropy": 0.048742625454906374,
+      "epoch": 10.56,
+      "grad_norm": 0.423828125,
+      "learning_rate": 7.557777777777779e-06,
+      "loss": 0.0285,
+      "mean_token_accuracy": 0.9917002696543932,
+      "num_tokens": 220015274.0,
+      "step": 6600
+    },
+    {
+      "epoch": 10.56,
+      "eval_chemistry_entropy": 0.23333027350902558,
+      "eval_chemistry_loss": 1.3877805471420288,
+      "eval_chemistry_mean_token_accuracy": 0.7868291339874267,
+      "eval_chemistry_num_tokens": 220015274.0,
+      "eval_chemistry_runtime": 34.8436,
+      "eval_chemistry_samples_per_second": 14.35,
+      "eval_chemistry_steps_per_second": 3.587,
+      "step": 6600
+    },
+    {
+      "epoch": 10.56,
+      "eval_physics_entropy": 0.17668923830986022,
+      "eval_physics_loss": 1.17435622215271,
+      "eval_physics_mean_token_accuracy": 0.8237164525985717,
+      "eval_physics_num_tokens": 220015274.0,
+      "eval_physics_runtime": 41.4197,
+      "eval_physics_samples_per_second": 12.072,
+      "eval_physics_steps_per_second": 3.018,
+      "step": 6600
+    },
+    {
+      "entropy": 0.04757428133161738,
+      "epoch": 10.576,
+      "grad_norm": 0.400390625,
+      "learning_rate": 7.535555555555556e-06,
+      "loss": 0.0276,
+      "mean_token_accuracy": 0.9922905944287777,
+      "num_tokens": 220359247.0,
+      "step": 6610
+    },
+    {
+      "entropy": 0.04917768740560859,
+      "epoch": 10.592,
+      "grad_norm": 0.37109375,
+      "learning_rate": 7.513333333333334e-06,
+      "loss": 0.0286,
+      "mean_token_accuracy": 0.9916475094854832,
+      "num_tokens": 220685397.0,
+      "step": 6620
+    },
+    {
+      "entropy": 0.04764970927499235,
+      "epoch": 10.608,
+      "grad_norm": 0.39453125,
+      "learning_rate": 7.4911111111111116e-06,
+      "loss": 0.0278,
+      "mean_token_accuracy": 0.992089631780982,
+      "num_tokens": 221023357.0,
+      "step": 6630
+    },
+    {
+      "entropy": 0.05037611761363223,
+      "epoch": 10.624,
+      "grad_norm": 0.45703125,
+      "learning_rate": 7.46888888888889e-06,
+      "loss": 0.0285,
+      "mean_token_accuracy": 0.9917111974209547,
+      "num_tokens": 221353123.0,
+      "step": 6640
+    },
+    {
+      "entropy": 0.04916413738392293,
+      "epoch": 10.64,
+      "grad_norm": 0.435546875,
+      "learning_rate": 7.446666666666668e-06,
+      "loss": 0.0287,
+      "mean_token_accuracy": 0.9917195219546556,
+      "num_tokens": 221679310.0,
+      "step": 6650
+    },
+    {
+      "entropy": 0.0491714600357227,
+      "epoch": 10.656,
+      "grad_norm": 0.435546875,
+      "learning_rate": 7.424444444444445e-06,
+      "loss": 0.0285,
+      "mean_token_accuracy": 0.991625689715147,
+      "num_tokens": 222010042.0,
+      "step": 6660
+    },
+    {
+      "entropy": 0.04928614452946931,
+      "epoch": 10.672,
+      "grad_norm": 0.44140625,
+      "learning_rate": 7.402222222222223e-06,
+      "loss": 0.0286,
+      "mean_token_accuracy": 0.9918153759092092,
+      "num_tokens": 222337940.0,
+      "step": 6670
+    },
+    {
+      "entropy": 0.04848613451467827,
+      "epoch": 10.688,
+      "grad_norm": 0.427734375,
+      "learning_rate": 7.3800000000000005e-06,
+      "loss": 0.0283,
+      "mean_token_accuracy": 0.9919930532574653,
+      "num_tokens": 222670011.0,
+      "step": 6680
+    },
+    {
+      "entropy": 0.049349892209284006,
+      "epoch": 10.704,
+      "grad_norm": 0.42578125,
+      "learning_rate": 7.357777777777778e-06,
+      "loss": 0.0291,
+      "mean_token_accuracy": 0.9917138006538153,
+      "num_tokens": 222992837.0,
+      "step": 6690
+    },
+    {
+      "entropy": 0.049544220021925864,
+      "epoch": 10.72,
+      "grad_norm": 0.412109375,
+      "learning_rate": 7.335555555555556e-06,
+      "loss": 0.0283,
+      "mean_token_accuracy": 0.9916808977723122,
+      "num_tokens": 223325310.0,
+      "step": 6700
+    },
+    {
+      "epoch": 10.72,
+      "eval_chemistry_entropy": 0.2328893860578537,
+      "eval_chemistry_loss": 1.3881638050079346,
+      "eval_chemistry_mean_token_accuracy": 0.7868953666687012,
+      "eval_chemistry_num_tokens": 223325310.0,
+      "eval_chemistry_runtime": 34.7896,
+      "eval_chemistry_samples_per_second": 14.372,
+      "eval_chemistry_steps_per_second": 3.593,
+      "step": 6700
+    },
+    {
+      "epoch": 10.72,
+      "eval_physics_entropy": 0.1761813908815384,
+      "eval_physics_loss": 1.1754546165466309,
+      "eval_physics_mean_token_accuracy": 0.8236912369728089,
+      "eval_physics_num_tokens": 223325310.0,
+      "eval_physics_runtime": 41.4124,
+      "eval_physics_samples_per_second": 12.074,
+      "eval_physics_steps_per_second": 3.018,
+      "step": 6700
+    },
+    {
+      "entropy": 0.04828044916503131,
+      "epoch": 10.736,
+      "grad_norm": 0.408203125,
+      "learning_rate": 7.313333333333333e-06,
+      "loss": 0.0281,
+      "mean_token_accuracy": 0.991859270632267,
+      "num_tokens": 223659912.0,
+      "step": 6710
+    },
+    {
+      "entropy": 0.04862959766760468,
+      "epoch": 10.752,
+      "grad_norm": 0.39453125,
+      "learning_rate": 7.291111111111112e-06,
+      "loss": 0.028,
+      "mean_token_accuracy": 0.9918654780834913,
+      "num_tokens": 223993306.0,
+      "step": 6720
+    },
+    {
+      "entropy": 0.04931755895959213,
+      "epoch": 10.768,
+      "grad_norm": 0.4453125,
+      "learning_rate": 7.268888888888889e-06,
+      "loss": 0.0285,
+      "mean_token_accuracy": 0.9917810220271349,
+      "num_tokens": 224321417.0,
+      "step": 6730
+    },
+    {
+      "entropy": 0.04935128873912618,
+      "epoch": 10.784,
+      "grad_norm": 0.404296875,
+      "learning_rate": 7.246666666666667e-06,
+      "loss": 0.0286,
+      "mean_token_accuracy": 0.9917261451482773,
+      "num_tokens": 224648496.0,
+      "step": 6740
+    },
+    {
+      "entropy": 0.0497755874064751,
+      "epoch": 10.8,
+      "grad_norm": 0.458984375,
+      "learning_rate": 7.224444444444445e-06,
+      "loss": 0.0289,
+      "mean_token_accuracy": 0.9916196443140507,
+      "num_tokens": 224972337.0,
+      "step": 6750
+    },
+    {
+      "entropy": 0.04919935124926269,
+      "epoch": 10.816,
+      "grad_norm": 0.375,
+      "learning_rate": 7.202222222222222e-06,
+      "loss": 0.0282,
+      "mean_token_accuracy": 0.9917631778866053,
+      "num_tokens": 225305583.0,
+      "step": 6760
+    },
+    {
+      "entropy": 0.0485584985348396,
+      "epoch": 10.832,
+      "grad_norm": 0.408203125,
+      "learning_rate": 7.180000000000001e-06,
+      "loss": 0.0287,
+      "mean_token_accuracy": 0.991681256890297,
+      "num_tokens": 225631725.0,
+      "step": 6770
+    },
+    {
+      "entropy": 0.04775191367371008,
+      "epoch": 10.848,
+      "grad_norm": 0.376953125,
+      "learning_rate": 7.157777777777778e-06,
+      "loss": 0.0276,
+      "mean_token_accuracy": 0.9920791570097208,
+      "num_tokens": 225973253.0,
+      "step": 6780
+    },
+    {
+      "entropy": 0.04784941094694659,
+      "epoch": 10.864,
+      "grad_norm": 0.38671875,
+      "learning_rate": 7.135555555555557e-06,
+      "loss": 0.0278,
+      "mean_token_accuracy": 0.9919260185211897,
+      "num_tokens": 226311456.0,
+      "step": 6790
+    },
+    {
+      "entropy": 0.04948324863798916,
+      "epoch": 10.88,
+      "grad_norm": 0.40625,
+      "learning_rate": 7.113333333333334e-06,
+      "loss": 0.0288,
+      "mean_token_accuracy": 0.9917005077004433,
+      "num_tokens": 226639744.0,
+      "step": 6800
+    },
+    {
+      "epoch": 10.88,
+      "eval_chemistry_entropy": 0.23311283946037292,
+      "eval_chemistry_loss": 1.3892673254013062,
+      "eval_chemistry_mean_token_accuracy": 0.7867928977012635,
+      "eval_chemistry_num_tokens": 226639744.0,
+      "eval_chemistry_runtime": 34.8636,
+      "eval_chemistry_samples_per_second": 14.342,
+      "eval_chemistry_steps_per_second": 3.585,
+      "step": 6800
+    },
+    {
+      "epoch": 10.88,
+      "eval_physics_entropy": 0.17646411782503127,
+      "eval_physics_loss": 1.1763180494308472,
+      "eval_physics_mean_token_accuracy": 0.8237738695144653,
+      "eval_physics_num_tokens": 226639744.0,
+      "eval_physics_runtime": 41.4971,
+      "eval_physics_samples_per_second": 12.049,
+      "eval_physics_steps_per_second": 3.012,
+      "step": 6800
+    },
+    {
+      "entropy": 0.048017301899380985,
+      "epoch": 10.896,
+      "grad_norm": 0.3984375,
+      "learning_rate": 7.091111111111112e-06,
+      "loss": 0.0281,
+      "mean_token_accuracy": 0.991930341348052,
+      "num_tokens": 226974690.0,
+      "step": 6810
+    },
+    {
+      "entropy": 0.04839458940550685,
+      "epoch": 10.912,
+      "grad_norm": 0.46875,
+      "learning_rate": 7.06888888888889e-06,
+      "loss": 0.0276,
+      "mean_token_accuracy": 0.9919111553579569,
+      "num_tokens": 227315806.0,
+      "step": 6820
+    },
+    {
+      "entropy": 0.04956060533877462,
+      "epoch": 10.928,
+      "grad_norm": 0.3671875,
+      "learning_rate": 7.046666666666667e-06,
+      "loss": 0.0292,
+      "mean_token_accuracy": 0.991617638990283,
+      "num_tokens": 227639280.0,
+      "step": 6830
+    },
+    {
+      "entropy": 0.04979866745416075,
+      "epoch": 10.943999999999999,
+      "grad_norm": 0.447265625,
+      "learning_rate": 7.024444444444445e-06,
+      "loss": 0.0288,
+      "mean_token_accuracy": 0.9917906813323498,
+      "num_tokens": 227966410.0,
+      "step": 6840
+    },
+    {
+      "entropy": 0.04816381558775902,
+      "epoch": 10.96,
+      "grad_norm": 0.39453125,
+      "learning_rate": 7.0022222222222225e-06,
+      "loss": 0.028,
+      "mean_token_accuracy": 0.9919066082686185,
+      "num_tokens": 228302153.0,
+      "step": 6850
+    },
+    {
+      "entropy": 0.04712158420588821,
+      "epoch": 10.975999999999999,
+      "grad_norm": 0.41015625,
+      "learning_rate": 6.98e-06,
+      "loss": 0.0274,
+      "mean_token_accuracy": 0.9920823790133,
+      "num_tokens": 228647179.0,
+      "step": 6860
+    },
+    {
+      "entropy": 0.047671990853268655,
+      "epoch": 10.992,
+      "grad_norm": 0.73828125,
+      "learning_rate": 6.9577777777777785e-06,
+      "loss": 0.0277,
+      "mean_token_accuracy": 0.9920064345002174,
+      "num_tokens": 228986231.0,
+      "step": 6870
+    },
+    {
+      "entropy": 0.04898794995388016,
+      "epoch": 11.008,
+      "grad_norm": 0.380859375,
+      "learning_rate": 6.935555555555556e-06,
+      "loss": 0.0281,
+      "mean_token_accuracy": 0.9917954411357641,
+      "num_tokens": 229315234.0,
+      "step": 6880
+    },
+    {
+      "entropy": 0.04681897725677118,
+      "epoch": 11.024,
+      "grad_norm": 0.345703125,
+      "learning_rate": 6.913333333333334e-06,
+      "loss": 0.0267,
+      "mean_token_accuracy": 0.9924691669642925,
+      "num_tokens": 229653285.0,
+      "step": 6890
+    },
+    {
+      "entropy": 0.04728088501142338,
+      "epoch": 11.04,
+      "grad_norm": 0.369140625,
+      "learning_rate": 6.891111111111111e-06,
+      "loss": 0.0272,
+      "mean_token_accuracy": 0.9923715580254793,
+      "num_tokens": 229988790.0,
+      "step": 6900
+    },
+    {
+      "epoch": 11.04,
+      "eval_chemistry_entropy": 0.2325030166506767,
+      "eval_chemistry_loss": 1.3919543027877808,
+      "eval_chemistry_mean_token_accuracy": 0.7866416792869568,
+      "eval_chemistry_num_tokens": 229988790.0,
+      "eval_chemistry_runtime": 34.7521,
+      "eval_chemistry_samples_per_second": 14.388,
+      "eval_chemistry_steps_per_second": 3.597,
+      "step": 6900
+    },
+    {
+      "epoch": 11.04,
+      "eval_physics_entropy": 0.1759334578514099,
+      "eval_physics_loss": 1.1786656379699707,
+      "eval_physics_mean_token_accuracy": 0.8236969685554505,
+      "eval_physics_num_tokens": 229988790.0,
+      "eval_physics_runtime": 41.4322,
+      "eval_physics_samples_per_second": 12.068,
+      "eval_physics_steps_per_second": 3.017,
+      "step": 6900
+    },
+    {
+      "entropy": 0.04765494554303586,
+      "epoch": 11.056,
+      "grad_norm": 0.37890625,
+      "learning_rate": 6.868888888888889e-06,
+      "loss": 0.0275,
+      "mean_token_accuracy": 0.9922619007527829,
+      "num_tokens": 230319178.0,
+      "step": 6910
+    },
+    {
+      "entropy": 0.04712109939428046,
+      "epoch": 11.072,
+      "grad_norm": 0.384765625,
+      "learning_rate": 6.846666666666667e-06,
+      "loss": 0.0267,
+      "mean_token_accuracy": 0.9924171775579452,
+      "num_tokens": 230658698.0,
+      "step": 6920
+    },
+    {
+      "entropy": 0.04816797635285184,
+      "epoch": 11.088,
+      "grad_norm": 0.365234375,
+      "learning_rate": 6.824444444444444e-06,
+      "loss": 0.0268,
+      "mean_token_accuracy": 0.9922629605978728,
+      "num_tokens": 230996786.0,
+      "step": 6930
+    },
+    {
+      "entropy": 0.04790292409015819,
+      "epoch": 11.104,
+      "grad_norm": 0.416015625,
+      "learning_rate": 6.802222222222223e-06,
+      "loss": 0.0277,
+      "mean_token_accuracy": 0.9921916209161281,
+      "num_tokens": 231325868.0,
+      "step": 6940
+    },
+    {
+      "entropy": 0.04852128324564546,
+      "epoch": 11.12,
+      "grad_norm": 0.376953125,
+      "learning_rate": 6.780000000000001e-06,
+      "loss": 0.0277,
+      "mean_token_accuracy": 0.9921166706830263,
+      "num_tokens": 231654840.0,
+      "step": 6950
+    },
+    {
+      "entropy": 0.04831116185523569,
+      "epoch": 11.136,
+      "grad_norm": 0.4140625,
+      "learning_rate": 6.757777777777779e-06,
+      "loss": 0.0279,
+      "mean_token_accuracy": 0.9919036451727152,
+      "num_tokens": 231980610.0,
+      "step": 6960
+    },
+    {
+      "entropy": 0.048215917765628544,
+      "epoch": 11.152,
+      "grad_norm": 0.412109375,
+      "learning_rate": 6.735555555555556e-06,
+      "loss": 0.0274,
+      "mean_token_accuracy": 0.9919559702277183,
+      "num_tokens": 232314288.0,
+      "step": 6970
+    },
+    {
+      "entropy": 0.048762065800838174,
+      "epoch": 11.168,
+      "grad_norm": 0.421875,
+      "learning_rate": 6.713333333333334e-06,
+      "loss": 0.0281,
+      "mean_token_accuracy": 0.9919058971107007,
+      "num_tokens": 232637534.0,
+      "step": 6980
+    },
+    {
+      "entropy": 0.048856847430579366,
+      "epoch": 11.184,
+      "grad_norm": 0.396484375,
+      "learning_rate": 6.691111111111112e-06,
+      "loss": 0.028,
+      "mean_token_accuracy": 0.9918563816696405,
+      "num_tokens": 232962955.0,
+      "step": 6990
+    },
+    {
+      "entropy": 0.047482672077603635,
+      "epoch": 11.2,
+      "grad_norm": 0.38671875,
+      "learning_rate": 6.668888888888889e-06,
+      "loss": 0.0276,
+      "mean_token_accuracy": 0.9920215297490358,
+      "num_tokens": 233293008.0,
+      "step": 7000
+    },
+    {
+      "epoch": 11.2,
+      "eval_chemistry_entropy": 0.2325966637134552,
+      "eval_chemistry_loss": 1.3951178789138794,
+      "eval_chemistry_mean_token_accuracy": 0.7866007509231567,
+      "eval_chemistry_num_tokens": 233293008.0,
+      "eval_chemistry_runtime": 34.7519,
+      "eval_chemistry_samples_per_second": 14.388,
+      "eval_chemistry_steps_per_second": 3.597,
+      "step": 7000
+    },
+    {
+      "epoch": 11.2,
+      "eval_physics_entropy": 0.17592375284433365,
+      "eval_physics_loss": 1.1814687252044678,
+      "eval_physics_mean_token_accuracy": 0.8236216421127319,
+      "eval_physics_num_tokens": 233293008.0,
+      "eval_physics_runtime": 41.1205,
+      "eval_physics_samples_per_second": 12.159,
+      "eval_physics_steps_per_second": 3.04,
+      "step": 7000
+    },
+    {
+      "entropy": 0.046440727543085814,
+      "epoch": 11.216,
+      "grad_norm": 0.37109375,
+      "learning_rate": 6.646666666666667e-06,
+      "loss": 0.0269,
+      "mean_token_accuracy": 0.9923358991742134,
+      "num_tokens": 233633040.0,
+      "step": 7010
+    },
+    {
+      "entropy": 0.047706857812590896,
+      "epoch": 11.232,
+      "grad_norm": 0.40625,
+      "learning_rate": 6.6244444444444445e-06,
+      "loss": 0.0278,
+      "mean_token_accuracy": 0.9921537846326828,
+      "num_tokens": 233961892.0,
+      "step": 7020
+    },
+    {
+      "entropy": 0.04752079215832054,
+      "epoch": 11.248,
+      "grad_norm": 0.3828125,
+      "learning_rate": 6.602222222222223e-06,
+      "loss": 0.0277,
+      "mean_token_accuracy": 0.9921526420861483,
+      "num_tokens": 234294300.0,
+      "step": 7030
+    },
+    {
+      "entropy": 0.04974490953609347,
+      "epoch": 11.264,
+      "grad_norm": 0.412109375,
+      "learning_rate": 6.5800000000000005e-06,
+      "loss": 0.0283,
+      "mean_token_accuracy": 0.9918520383536815,
+      "num_tokens": 234619896.0,
+      "step": 7040
+    },
+    {
+      "entropy": 0.04697937336750328,
+      "epoch": 11.28,
+      "grad_norm": 0.42578125,
+      "learning_rate": 6.557777777777778e-06,
+      "loss": 0.0274,
+      "mean_token_accuracy": 0.9921463459730149,
+      "num_tokens": 234955038.0,
+      "step": 7050
+    },
+    {
+      "entropy": 0.047558799386024475,
+      "epoch": 11.296,
+      "grad_norm": 0.447265625,
+      "learning_rate": 6.535555555555556e-06,
+      "loss": 0.0276,
+      "mean_token_accuracy": 0.9920218735933304,
+      "num_tokens": 235285322.0,
+      "step": 7060
+    },
+    {
+      "entropy": 0.04819324740674347,
+      "epoch": 11.312,
+      "grad_norm": 0.435546875,
+      "learning_rate": 6.513333333333333e-06,
+      "loss": 0.0275,
+      "mean_token_accuracy": 0.9922151681035757,
+      "num_tokens": 235620514.0,
+      "step": 7070
+    },
+    {
+      "entropy": 0.04757968414342031,
+      "epoch": 11.328,
+      "grad_norm": 0.41796875,
+      "learning_rate": 6.491111111111111e-06,
+      "loss": 0.0277,
+      "mean_token_accuracy": 0.9919170897454024,
+      "num_tokens": 235949223.0,
+      "step": 7080
+    },
+    {
+      "entropy": 0.047577433427795766,
+      "epoch": 11.344,
+      "grad_norm": 0.40625,
+      "learning_rate": 6.468888888888889e-06,
+      "loss": 0.0275,
+      "mean_token_accuracy": 0.9921996362507344,
+      "num_tokens": 236283934.0,
+      "step": 7090
+    },
+    {
+      "entropy": 0.046396936383098365,
+      "epoch": 11.36,
+      "grad_norm": 0.39453125,
+      "learning_rate": 6.446666666666668e-06,
+      "loss": 0.0271,
+      "mean_token_accuracy": 0.9922611985355616,
+      "num_tokens": 236621013.0,
+      "step": 7100
+    },
+    {
+      "epoch": 11.36,
+      "eval_chemistry_entropy": 0.23291492408514022,
+      "eval_chemistry_loss": 1.3945815563201904,
+      "eval_chemistry_mean_token_accuracy": 0.7867159643173218,
+      "eval_chemistry_num_tokens": 236621013.0,
+      "eval_chemistry_runtime": 34.7519,
+      "eval_chemistry_samples_per_second": 14.388,
+      "eval_chemistry_steps_per_second": 3.597,
+      "step": 7100
+    },
+    {
+      "epoch": 11.36,
+      "eval_physics_entropy": 0.17613509422540666,
+      "eval_physics_loss": 1.1810358762741089,
+      "eval_physics_mean_token_accuracy": 0.8236798253059388,
+      "eval_physics_num_tokens": 236621013.0,
+      "eval_physics_runtime": 41.3979,
+      "eval_physics_samples_per_second": 12.078,
+      "eval_physics_steps_per_second": 3.019,
+      "step": 7100
+    },
+    {
+      "entropy": 0.04900334149133414,
+      "epoch": 11.376,
+      "grad_norm": 0.419921875,
+      "learning_rate": 6.4244444444444455e-06,
+      "loss": 0.0281,
+      "mean_token_accuracy": 0.991936006769538,
+      "num_tokens": 236948817.0,
+      "step": 7110
+    },
+    {
+      "entropy": 0.04684424267616123,
+      "epoch": 11.392,
+      "grad_norm": 0.40625,
+      "learning_rate": 6.402222222222223e-06,
+      "loss": 0.0273,
+      "mean_token_accuracy": 0.9921260058879853,
+      "num_tokens": 237283368.0,
+      "step": 7120
+    },
+    {
+      "entropy": 0.045224152388982475,
+      "epoch": 11.408,
+      "grad_norm": 0.46484375,
+      "learning_rate": 6.380000000000001e-06,
+      "loss": 0.0267,
+      "mean_token_accuracy": 0.9923562560230493,
+      "num_tokens": 237627096.0,
+      "step": 7130
+    },
+    {
+      "entropy": 0.04832157021155581,
+      "epoch": 11.424,
+      "grad_norm": 0.404296875,
+      "learning_rate": 6.357777777777778e-06,
+      "loss": 0.0279,
+      "mean_token_accuracy": 0.991941137239337,
+      "num_tokens": 237957137.0,
+      "step": 7140
+    },
+    {
+      "entropy": 0.0496531565557234,
+      "epoch": 11.44,
+      "grad_norm": 0.49609375,
+      "learning_rate": 6.335555555555556e-06,
+      "loss": 0.0285,
+      "mean_token_accuracy": 0.9916909664869309,
+      "num_tokens": 238281139.0,
+      "step": 7150
+    },
+    {
+      "entropy": 0.04692766816588119,
+      "epoch": 11.456,
+      "grad_norm": 0.4453125,
+      "learning_rate": 6.313333333333334e-06,
+      "loss": 0.0274,
+      "mean_token_accuracy": 0.9920871052891016,
+      "num_tokens": 238615826.0,
+      "step": 7160
+    },
+    {
+      "entropy": 0.04930061105405912,
+      "epoch": 11.472,
+      "grad_norm": 0.376953125,
+      "learning_rate": 6.291111111111111e-06,
+      "loss": 0.0285,
+      "mean_token_accuracy": 0.9916456084698438,
+      "num_tokens": 238937689.0,
+      "step": 7170
+    },
+    {
+      "entropy": 0.04741948866285384,
+      "epoch": 11.488,
+      "grad_norm": 0.4140625,
+      "learning_rate": 6.26888888888889e-06,
+      "loss": 0.0277,
+      "mean_token_accuracy": 0.9920615538954735,
+      "num_tokens": 239269756.0,
+      "step": 7180
+    },
+    {
+      "entropy": 0.046355318860150874,
+      "epoch": 11.504,
+      "grad_norm": 0.396484375,
+      "learning_rate": 6.246666666666667e-06,
+      "loss": 0.027,
+      "mean_token_accuracy": 0.9922457475215196,
+      "num_tokens": 239610878.0,
+      "step": 7190
+    },
+    {
+      "entropy": 0.0478028149693273,
+      "epoch": 11.52,
+      "grad_norm": 0.388671875,
+      "learning_rate": 6.224444444444445e-06,
+      "loss": 0.0281,
+      "mean_token_accuracy": 0.9920501060783863,
+      "num_tokens": 239939093.0,
+      "step": 7200
+    },
+    {
+      "epoch": 11.52,
+      "eval_chemistry_entropy": 0.23264170211553573,
+      "eval_chemistry_loss": 1.3950202465057373,
+      "eval_chemistry_mean_token_accuracy": 0.7867832498550416,
+      "eval_chemistry_num_tokens": 239939093.0,
+      "eval_chemistry_runtime": 34.7886,
+      "eval_chemistry_samples_per_second": 14.373,
+      "eval_chemistry_steps_per_second": 3.593,
+      "step": 7200
+    },
+    {
+      "epoch": 11.52,
+      "eval_physics_entropy": 0.17588690716028213,
+      "eval_physics_loss": 1.1816986799240112,
+      "eval_physics_mean_token_accuracy": 0.8236192560195923,
+      "eval_physics_num_tokens": 239939093.0,
+      "eval_physics_runtime": 41.4102,
+      "eval_physics_samples_per_second": 12.074,
+      "eval_physics_steps_per_second": 3.019,
+      "step": 7200
+    },
+    {
+      "entropy": 0.04801568454131484,
+      "epoch": 11.536,
+      "grad_norm": 0.439453125,
+      "learning_rate": 6.2022222222222225e-06,
+      "loss": 0.0277,
+      "mean_token_accuracy": 0.9920029643923044,
+      "num_tokens": 240272277.0,
+      "step": 7210
+    },
+    {
+      "entropy": 0.048277478280942886,
+      "epoch": 11.552,
+      "grad_norm": 0.421875,
+      "learning_rate": 6.18e-06,
+      "loss": 0.0279,
+      "mean_token_accuracy": 0.9920069571584463,
+      "num_tokens": 240601667.0,
+      "step": 7220
+    },
+    {
+      "entropy": 0.04607030840124935,
+      "epoch": 11.568,
+      "grad_norm": 0.384765625,
+      "learning_rate": 6.157777777777778e-06,
+      "loss": 0.0268,
+      "mean_token_accuracy": 0.99229018650949,
+      "num_tokens": 240945224.0,
+      "step": 7230
+    },
+    {
+      "entropy": 0.046531455463264135,
+      "epoch": 11.584,
+      "grad_norm": 0.349609375,
+      "learning_rate": 6.135555555555555e-06,
+      "loss": 0.0273,
+      "mean_token_accuracy": 0.9921449899673462,
+      "num_tokens": 241287731.0,
+      "step": 7240
+    },
+    {
+      "entropy": 0.04824015131453052,
+      "epoch": 11.6,
+      "grad_norm": 0.365234375,
+      "learning_rate": 6.113333333333333e-06,
+      "loss": 0.0276,
+      "mean_token_accuracy": 0.9918418738991022,
+      "num_tokens": 241622563.0,
+      "step": 7250
+    },
+    {
+      "entropy": 0.04845166703453287,
+      "epoch": 11.616,
+      "grad_norm": 0.41796875,
+      "learning_rate": 6.091111111111112e-06,
+      "loss": 0.0282,
+      "mean_token_accuracy": 0.9919390242546797,
+      "num_tokens": 241949450.0,
+      "step": 7260
+    },
+    {
+      "entropy": 0.04824669075896963,
+      "epoch": 11.632,
+      "grad_norm": 0.388671875,
+      "learning_rate": 6.06888888888889e-06,
+      "loss": 0.0274,
+      "mean_token_accuracy": 0.9917767606675625,
+      "num_tokens": 242285573.0,
+      "step": 7270
+    },
+    {
+      "entropy": 0.0490477797575295,
+      "epoch": 11.648,
+      "grad_norm": 0.419921875,
+      "learning_rate": 6.0466666666666675e-06,
+      "loss": 0.0284,
+      "mean_token_accuracy": 0.9917098835110665,
+      "num_tokens": 242611548.0,
+      "step": 7280
+    },
+    {
+      "entropy": 0.04796445817919448,
+      "epoch": 11.664,
+      "grad_norm": 0.427734375,
+      "learning_rate": 6.024444444444445e-06,
+      "loss": 0.0278,
+      "mean_token_accuracy": 0.9921103999018669,
+      "num_tokens": 242944709.0,
+      "step": 7290
+    },
+    {
+      "entropy": 0.04770536229480058,
+      "epoch": 11.68,
+      "grad_norm": 0.412109375,
+      "learning_rate": 6.002222222222223e-06,
+      "loss": 0.0278,
+      "mean_token_accuracy": 0.9919927552342415,
+      "num_tokens": 243278367.0,
+      "step": 7300
+    },
+    {
+      "epoch": 11.68,
+      "eval_chemistry_entropy": 0.23242050033807754,
+      "eval_chemistry_loss": 1.3952710628509521,
+      "eval_chemistry_mean_token_accuracy": 0.7865613141059875,
+      "eval_chemistry_num_tokens": 243278367.0,
+      "eval_chemistry_runtime": 34.6446,
+      "eval_chemistry_samples_per_second": 14.432,
+      "eval_chemistry_steps_per_second": 3.608,
+      "step": 7300
+    },
+    {
+      "epoch": 11.68,
+      "eval_physics_entropy": 0.17575073128938676,
+      "eval_physics_loss": 1.1822236776351929,
+      "eval_physics_mean_token_accuracy": 0.8235895795822143,
+      "eval_physics_num_tokens": 243278367.0,
+      "eval_physics_runtime": 41.2734,
+      "eval_physics_samples_per_second": 12.114,
+      "eval_physics_steps_per_second": 3.029,
+      "step": 7300
+    },
+    {
+      "entropy": 0.04782159347087145,
+      "epoch": 11.696,
+      "grad_norm": 0.416015625,
+      "learning_rate": 5.98e-06,
+      "loss": 0.0278,
+      "mean_token_accuracy": 0.991872738674283,
+      "num_tokens": 243612160.0,
+      "step": 7310
+    },
+    {
+      "entropy": 0.047998094698414207,
+      "epoch": 11.712,
+      "grad_norm": 0.470703125,
+      "learning_rate": 5.957777777777778e-06,
+      "loss": 0.0279,
+      "mean_token_accuracy": 0.9918551001697778,
+      "num_tokens": 243942664.0,
+      "step": 7320
+    },
+    {
+      "entropy": 0.04701628105249256,
+      "epoch": 11.728,
+      "grad_norm": 0.392578125,
+      "learning_rate": 5.9355555555555556e-06,
+      "loss": 0.0269,
+      "mean_token_accuracy": 0.9922316294163466,
+      "num_tokens": 244287354.0,
+      "step": 7330
+    },
+    {
+      "entropy": 0.0459823242505081,
+      "epoch": 11.744,
+      "grad_norm": 0.390625,
+      "learning_rate": 5.913333333333334e-06,
+      "loss": 0.0267,
+      "mean_token_accuracy": 0.9923303164541721,
+      "num_tokens": 244632442.0,
+      "step": 7340
+    },
+    {
+      "entropy": 0.04870257138973102,
+      "epoch": 11.76,
+      "grad_norm": 0.396484375,
+      "learning_rate": 5.891111111111112e-06,
+      "loss": 0.0282,
+      "mean_token_accuracy": 0.9917425870895386,
+      "num_tokens": 244960680.0,
+      "step": 7350
+    },
+    {
+      "entropy": 0.0474628541385755,
+      "epoch": 11.776,
+      "grad_norm": 0.37890625,
+      "learning_rate": 5.868888888888889e-06,
+      "loss": 0.0272,
+      "mean_token_accuracy": 0.9919994749128819,
+      "num_tokens": 245300920.0,
+      "step": 7360
+    },
+    {
+      "entropy": 0.04731073308503255,
+      "epoch": 11.792,
+      "grad_norm": 0.427734375,
+      "learning_rate": 5.846666666666667e-06,
+      "loss": 0.0278,
+      "mean_token_accuracy": 0.9919415000826121,
+      "num_tokens": 245633536.0,
+      "step": 7370
+    },
+    {
+      "entropy": 0.04833471442107111,
+      "epoch": 11.808,
+      "grad_norm": 0.44140625,
+      "learning_rate": 5.8244444444444445e-06,
+      "loss": 0.028,
+      "mean_token_accuracy": 0.991770077496767,
+      "num_tokens": 245964775.0,
+      "step": 7380
+    },
+    {
+      "entropy": 0.04759642768185586,
+      "epoch": 11.824,
+      "grad_norm": 0.41796875,
+      "learning_rate": 5.802222222222222e-06,
+      "loss": 0.0278,
+      "mean_token_accuracy": 0.9918633200228214,
+      "num_tokens": 246299419.0,
+      "step": 7390
+    },
+    {
+      "entropy": 0.04769166146870703,
+      "epoch": 11.84,
+      "grad_norm": 0.412109375,
+      "learning_rate": 5.78e-06,
+      "loss": 0.0278,
+      "mean_token_accuracy": 0.991833696514368,
+      "num_tokens": 246633322.0,
+      "step": 7400
+    },
+    {
+      "epoch": 11.84,
+      "eval_chemistry_entropy": 0.23273953205347062,
+      "eval_chemistry_loss": 1.395848274230957,
+      "eval_chemistry_mean_token_accuracy": 0.786546067237854,
+      "eval_chemistry_num_tokens": 246633322.0,
+      "eval_chemistry_runtime": 34.6363,
+      "eval_chemistry_samples_per_second": 14.436,
+      "eval_chemistry_steps_per_second": 3.609,
+      "step": 7400
+    },
+    {
+      "epoch": 11.84,
+      "eval_physics_entropy": 0.17592825770378112,
+      "eval_physics_loss": 1.1827595233917236,
+      "eval_physics_mean_token_accuracy": 0.8236251907348633,
+      "eval_physics_num_tokens": 246633322.0,
+      "eval_physics_runtime": 41.2892,
+      "eval_physics_samples_per_second": 12.11,
+      "eval_physics_steps_per_second": 3.027,
+      "step": 7400
+    },
+    {
+      "entropy": 0.04780968824634328,
+      "epoch": 11.856,
+      "grad_norm": 0.376953125,
+      "learning_rate": 5.757777777777779e-06,
+      "loss": 0.0275,
+      "mean_token_accuracy": 0.9920414347201586,
+      "num_tokens": 246970012.0,
+      "step": 7410
+    },
+    {
+      "entropy": 0.047446249541826545,
+      "epoch": 11.872,
+      "grad_norm": 0.423828125,
+      "learning_rate": 5.735555555555557e-06,
+      "loss": 0.0273,
+      "mean_token_accuracy": 0.9919114019721746,
+      "num_tokens": 247310002.0,
+      "step": 7420
+    },
+    {
+      "entropy": 0.04886368457227945,
+      "epoch": 11.888,
+      "grad_norm": 0.390625,
+      "learning_rate": 5.713333333333334e-06,
+      "loss": 0.0283,
+      "mean_token_accuracy": 0.9916328560560942,
+      "num_tokens": 247637841.0,
+      "step": 7430
+    },
+    {
+      "entropy": 0.04758717410732061,
+      "epoch": 11.904,
+      "grad_norm": 0.3828125,
+      "learning_rate": 5.691111111111112e-06,
+      "loss": 0.0275,
+      "mean_token_accuracy": 0.9919271893799305,
+      "num_tokens": 247975131.0,
+      "step": 7440
+    },
+    {
+      "entropy": 0.047614185302518305,
+      "epoch": 11.92,
+      "grad_norm": 0.3828125,
+      "learning_rate": 5.6688888888888895e-06,
+      "loss": 0.0277,
+      "mean_token_accuracy": 0.9919915556907654,
+      "num_tokens": 248309755.0,
+      "step": 7450
+    },
+    {
+      "entropy": 0.04707025629468262,
+      "epoch": 11.936,
+      "grad_norm": 0.41796875,
+      "learning_rate": 5.646666666666667e-06,
+      "loss": 0.0276,
+      "mean_token_accuracy": 0.9919815827161074,
+      "num_tokens": 248645564.0,
+      "step": 7460
+    },
+    {
+      "entropy": 0.04757521008141339,
+      "epoch": 11.952,
+      "grad_norm": 0.4140625,
+      "learning_rate": 5.624444444444445e-06,
+      "loss": 0.0277,
+      "mean_token_accuracy": 0.9918838202953338,
+      "num_tokens": 248981848.0,
+      "step": 7470
+    },
+    {
+      "entropy": 0.04723024608101696,
+      "epoch": 11.968,
+      "grad_norm": 0.365234375,
+      "learning_rate": 5.602222222222222e-06,
+      "loss": 0.0273,
+      "mean_token_accuracy": 0.9920327696949244,
+      "num_tokens": 249323006.0,
+      "step": 7480
+    },
+    {
+      "entropy": 0.04786972624715417,
+      "epoch": 11.984,
+      "grad_norm": 0.41015625,
+      "learning_rate": 5.580000000000001e-06,
+      "loss": 0.0279,
+      "mean_token_accuracy": 0.9917761087417603,
+      "num_tokens": 249654901.0,
+      "step": 7490
+    },
+    {
+      "entropy": 0.04861369674326852,
+      "epoch": 12.0,
+      "grad_norm": 0.419921875,
+      "learning_rate": 5.557777777777778e-06,
+      "loss": 0.0281,
+      "mean_token_accuracy": 0.9917664911597968,
+      "num_tokens": 249985776.0,
+      "step": 7500
+    },
+    {
+      "epoch": 12.0,
+      "eval_chemistry_entropy": 0.2323100711107254,
+      "eval_chemistry_loss": 1.395796537399292,
+      "eval_chemistry_mean_token_accuracy": 0.7867488856315613,
+      "eval_chemistry_num_tokens": 249985776.0,
+      "eval_chemistry_runtime": 34.6202,
+      "eval_chemistry_samples_per_second": 14.442,
+      "eval_chemistry_steps_per_second": 3.611,
+      "step": 7500
+    },
+    {
+      "epoch": 12.0,
+      "eval_physics_entropy": 0.17559167432785033,
+      "eval_physics_loss": 1.18276047706604,
+      "eval_physics_mean_token_accuracy": 0.8237361998558045,
+      "eval_physics_num_tokens": 249985776.0,
+      "eval_physics_runtime": 40.9505,
+      "eval_physics_samples_per_second": 12.21,
+      "eval_physics_steps_per_second": 3.052,
+      "step": 7500
+    },
+    {
+      "entropy": 0.04508789298124612,
+      "epoch": 12.016,
+      "grad_norm": 0.36328125,
+      "learning_rate": 5.535555555555556e-06,
+      "loss": 0.0258,
+      "mean_token_accuracy": 0.9927411187440157,
+      "num_tokens": 250334719.0,
+      "step": 7510
+    },
+    {
+      "entropy": 0.04760314099257812,
+      "epoch": 12.032,
+      "grad_norm": 0.3515625,
+      "learning_rate": 5.513333333333334e-06,
+      "loss": 0.0268,
+      "mean_token_accuracy": 0.9924074649810791,
+      "num_tokens": 250672098.0,
+      "step": 7520
+    },
+    {
+      "entropy": 0.04751529867062345,
+      "epoch": 12.048,
+      "grad_norm": 0.349609375,
+      "learning_rate": 5.491111111111111e-06,
+      "loss": 0.027,
+      "mean_token_accuracy": 0.9921338777989149,
+      "num_tokens": 251004102.0,
+      "step": 7530
+    },
+    {
+      "entropy": 0.04761073595145717,
+      "epoch": 12.064,
+      "grad_norm": 0.39453125,
+      "learning_rate": 5.468888888888889e-06,
+      "loss": 0.0273,
+      "mean_token_accuracy": 0.9923585925251246,
+      "num_tokens": 251333686.0,
+      "step": 7540
+    },
+    {
+      "entropy": 0.04869025647640228,
+      "epoch": 12.08,
+      "grad_norm": 0.431640625,
+      "learning_rate": 5.4466666666666665e-06,
+      "loss": 0.0283,
+      "mean_token_accuracy": 0.991908035799861,
+      "num_tokens": 251652413.0,
+      "step": 7550
+    },
+    {
+      "entropy": 0.04862467342754826,
+      "epoch": 12.096,
+      "grad_norm": 0.41015625,
+      "learning_rate": 5.424444444444444e-06,
+      "loss": 0.0274,
+      "mean_token_accuracy": 0.9920980051159859,
+      "num_tokens": 251980585.0,
+      "step": 7560
+    },
+    {
+      "entropy": 0.046205937664490195,
+      "epoch": 12.112,
+      "grad_norm": 0.388671875,
+      "learning_rate": 5.402222222222223e-06,
+      "loss": 0.0269,
+      "mean_token_accuracy": 0.9923609714955092,
+      "num_tokens": 252316579.0,
+      "step": 7570
+    },
+    {
+      "entropy": 0.04609726486960426,
+      "epoch": 12.128,
+      "grad_norm": 0.41796875,
+      "learning_rate": 5.380000000000001e-06,
+      "loss": 0.0267,
+      "mean_token_accuracy": 0.9923820272088051,
+      "num_tokens": 252654389.0,
+      "step": 7580
+    },
+    {
+      "entropy": 0.04772428745636716,
+      "epoch": 12.144,
+      "grad_norm": 0.439453125,
+      "learning_rate": 5.357777777777779e-06,
+      "loss": 0.0278,
+      "mean_token_accuracy": 0.9920915137976408,
+      "num_tokens": 252979873.0,
+      "step": 7590
+    },
+    {
+      "entropy": 0.046907130780164154,
+      "epoch": 12.16,
+      "grad_norm": 0.42578125,
+      "learning_rate": 5.335555555555556e-06,
+      "loss": 0.0269,
+      "mean_token_accuracy": 0.9923364825546741,
+      "num_tokens": 253316595.0,
+      "step": 7600
+    },
+    {
+      "epoch": 12.16,
+      "eval_chemistry_entropy": 0.2323877632021904,
+      "eval_chemistry_loss": 1.399549126625061,
+      "eval_chemistry_mean_token_accuracy": 0.7864844288825988,
+      "eval_chemistry_num_tokens": 253316595.0,
+      "eval_chemistry_runtime": 34.6168,
+      "eval_chemistry_samples_per_second": 14.444,
+      "eval_chemistry_steps_per_second": 3.611,
+      "step": 7600
+    },
+    {
+      "epoch": 12.16,
+      "eval_physics_entropy": 0.1755651298761368,
+      "eval_physics_loss": 1.1859861612319946,
+      "eval_physics_mean_token_accuracy": 0.8236834073066711,
+      "eval_physics_num_tokens": 253316595.0,
+      "eval_physics_runtime": 41.5923,
+      "eval_physics_samples_per_second": 12.021,
+      "eval_physics_steps_per_second": 3.005,
+      "step": 7600
+    },
+    {
+      "entropy": 0.04736769343726337,
+      "epoch": 12.176,
+      "grad_norm": 0.404296875,
+      "learning_rate": 5.313333333333334e-06,
+      "loss": 0.0269,
+      "mean_token_accuracy": 0.9922827742993832,
+      "num_tokens": 253653923.0,
+      "step": 7610
+    },
+    {
+      "entropy": 0.04715582146309316,
+      "epoch": 12.192,
+      "grad_norm": 0.384765625,
+      "learning_rate": 5.2911111111111115e-06,
+      "loss": 0.0273,
+      "mean_token_accuracy": 0.9921531699597835,
+      "num_tokens": 253984612.0,
+      "step": 7620
+    },
+    {
+      "entropy": 0.04785220557823777,
+      "epoch": 12.208,
+      "grad_norm": 0.408203125,
+      "learning_rate": 5.268888888888889e-06,
+      "loss": 0.028,
+      "mean_token_accuracy": 0.9918720994144679,
+      "num_tokens": 254306969.0,
+      "step": 7630
+    },
+    {
+      "entropy": 0.04722507435362786,
+      "epoch": 12.224,
+      "grad_norm": 0.3984375,
+      "learning_rate": 5.246666666666667e-06,
+      "loss": 0.0274,
+      "mean_token_accuracy": 0.992166755720973,
+      "num_tokens": 254637639.0,
+      "step": 7640
+    },
+    {
+      "entropy": 0.04629259455250576,
+      "epoch": 12.24,
+      "grad_norm": 0.451171875,
+      "learning_rate": 5.224444444444445e-06,
+      "loss": 0.027,
+      "mean_token_accuracy": 0.9923653081059456,
+      "num_tokens": 254974544.0,
+      "step": 7650
+    },
+    {
+      "entropy": 0.046991573600098493,
+      "epoch": 12.256,
+      "grad_norm": 0.408203125,
+      "learning_rate": 5.202222222222223e-06,
+      "loss": 0.0268,
+      "mean_token_accuracy": 0.9923080742359162,
+      "num_tokens": 255313630.0,
+      "step": 7660
+    },
+    {
+      "entropy": 0.047093414154369385,
+      "epoch": 12.272,
+      "grad_norm": 0.431640625,
+      "learning_rate": 5.18e-06,
+      "loss": 0.0271,
+      "mean_token_accuracy": 0.9921616919338703,
+      "num_tokens": 255649061.0,
+      "step": 7670
+    },
+    {
+      "entropy": 0.047323476884048435,
+      "epoch": 12.288,
+      "grad_norm": 0.39453125,
+      "learning_rate": 5.157777777777778e-06,
+      "loss": 0.0273,
+      "mean_token_accuracy": 0.9921791329979897,
+      "num_tokens": 255981823.0,
+      "step": 7680
+    },
+    {
+      "entropy": 0.04674618551507592,
+      "epoch": 12.304,
+      "grad_norm": 0.39453125,
+      "learning_rate": 5.135555555555556e-06,
+      "loss": 0.027,
+      "mean_token_accuracy": 0.992146011814475,
+      "num_tokens": 256317363.0,
+      "step": 7690
+    },
+    {
+      "entropy": 0.045185890479478984,
+      "epoch": 12.32,
+      "grad_norm": 0.38671875,
+      "learning_rate": 5.113333333333333e-06,
+      "loss": 0.0265,
+      "mean_token_accuracy": 0.992483926564455,
+      "num_tokens": 256660007.0,
+      "step": 7700
+    },
+    {
+      "epoch": 12.32,
+      "eval_chemistry_entropy": 0.23227442216873168,
+      "eval_chemistry_loss": 1.3991879224777222,
+      "eval_chemistry_mean_token_accuracy": 0.7866473956108093,
+      "eval_chemistry_num_tokens": 256660007.0,
+      "eval_chemistry_runtime": 34.6511,
+      "eval_chemistry_samples_per_second": 14.43,
+      "eval_chemistry_steps_per_second": 3.607,
+      "step": 7700
+    },
+    {
+      "epoch": 12.32,
+      "eval_physics_entropy": 0.17541469842195512,
+      "eval_physics_loss": 1.1861995458602905,
+      "eval_physics_mean_token_accuracy": 0.823622531414032,
+      "eval_physics_num_tokens": 256660007.0,
+      "eval_physics_runtime": 41.2534,
+      "eval_physics_samples_per_second": 12.12,
+      "eval_physics_steps_per_second": 3.03,
+      "step": 7700
+    },
+    {
+      "entropy": 0.04902239058865234,
+      "epoch": 12.336,
+      "grad_norm": 0.466796875,
+      "learning_rate": 5.091111111111111e-06,
+      "loss": 0.0283,
+      "mean_token_accuracy": 0.9917640380561352,
+      "num_tokens": 256982012.0,
+      "step": 7710
+    },
+    {
+      "entropy": 0.04705402229446918,
+      "epoch": 12.352,
+      "grad_norm": 0.44140625,
+      "learning_rate": 5.06888888888889e-06,
+      "loss": 0.0273,
+      "mean_token_accuracy": 0.9921987757086754,
+      "num_tokens": 257315080.0,
+      "step": 7720
+    },
+    {
+      "entropy": 0.0467334458604455,
+      "epoch": 12.368,
+      "grad_norm": 0.43359375,
+      "learning_rate": 5.046666666666668e-06,
+      "loss": 0.0268,
+      "mean_token_accuracy": 0.9923334632068872,
+      "num_tokens": 257654361.0,
+      "step": 7730
+    },
+    {
+      "entropy": 0.04817101656226441,
+      "epoch": 12.384,
+      "grad_norm": 0.396484375,
+      "learning_rate": 5.024444444444445e-06,
+      "loss": 0.028,
+      "mean_token_accuracy": 0.9918757636100054,
+      "num_tokens": 257978983.0,
+      "step": 7740
+    },
+    {
+      "entropy": 0.04686940559186041,
+      "epoch": 12.4,
+      "grad_norm": 0.43359375,
+      "learning_rate": 5.002222222222223e-06,
+      "loss": 0.0272,
+      "mean_token_accuracy": 0.9922051772475242,
+      "num_tokens": 258312899.0,
+      "step": 7750
+    },
+    {
+      "entropy": 0.04625321145867929,
+      "epoch": 12.416,
+      "grad_norm": 0.392578125,
+      "learning_rate": 4.980000000000001e-06,
+      "loss": 0.027,
+      "mean_token_accuracy": 0.9921592123806476,
+      "num_tokens": 258648951.0,
+      "step": 7760
+    },
+    {
+      "entropy": 0.04678296793717891,
+      "epoch": 12.432,
+      "grad_norm": 0.384765625,
+      "learning_rate": 4.957777777777778e-06,
+      "loss": 0.0265,
+      "mean_token_accuracy": 0.9923734743148088,
+      "num_tokens": 258992908.0,
+      "step": 7770
+    },
+    {
+      "entropy": 0.047582005138974634,
+      "epoch": 12.448,
+      "grad_norm": 0.42578125,
+      "learning_rate": 4.935555555555556e-06,
+      "loss": 0.0275,
+      "mean_token_accuracy": 0.9920337703078985,
+      "num_tokens": 259323671.0,
+      "step": 7780
+    },
+    {
+      "entropy": 0.047750978695694356,
+      "epoch": 12.464,
+      "grad_norm": 0.392578125,
+      "learning_rate": 4.9133333333333334e-06,
+      "loss": 0.0274,
+      "mean_token_accuracy": 0.9918526388704777,
+      "num_tokens": 259656145.0,
+      "step": 7790
+    },
+    {
+      "entropy": 0.046700223837979137,
+      "epoch": 12.48,
+      "grad_norm": 0.4375,
+      "learning_rate": 4.891111111111111e-06,
+      "loss": 0.0269,
+      "mean_token_accuracy": 0.9922252856194973,
+      "num_tokens": 259995949.0,
+      "step": 7800
+    },
+    {
+      "epoch": 12.48,
+      "eval_chemistry_entropy": 0.23217756485939026,
+      "eval_chemistry_loss": 1.3997950553894043,
+      "eval_chemistry_mean_token_accuracy": 0.7865884881019592,
+      "eval_chemistry_num_tokens": 259995949.0,
+      "eval_chemistry_runtime": 34.6092,
+      "eval_chemistry_samples_per_second": 14.447,
+      "eval_chemistry_steps_per_second": 3.612,
+      "step": 7800
+    },
+    {
+      "epoch": 12.48,
+      "eval_physics_entropy": 0.17545815068483353,
+      "eval_physics_loss": 1.1863703727722168,
+      "eval_physics_mean_token_accuracy": 0.823757504940033,
+      "eval_physics_num_tokens": 259995949.0,
+      "eval_physics_runtime": 41.2379,
+      "eval_physics_samples_per_second": 12.125,
+      "eval_physics_steps_per_second": 3.031,
+      "step": 7800
+    },
+    {
+      "entropy": 0.04697601619409397,
+      "epoch": 12.496,
+      "grad_norm": 0.40625,
+      "learning_rate": 4.8688888888888895e-06,
+      "loss": 0.0273,
+      "mean_token_accuracy": 0.9921011120080948,
+      "num_tokens": 260328029.0,
+      "step": 7810
+    },
+    {
+      "entropy": 0.04663161822827533,
+      "epoch": 12.512,
+      "grad_norm": 0.43359375,
+      "learning_rate": 4.846666666666667e-06,
+      "loss": 0.0273,
+      "mean_token_accuracy": 0.9920815654098988,
+      "num_tokens": 260660640.0,
+      "step": 7820
+    },
+    {
+      "entropy": 0.04710413708817214,
+      "epoch": 12.528,
+      "grad_norm": 0.392578125,
+      "learning_rate": 4.824444444444445e-06,
+      "loss": 0.027,
+      "mean_token_accuracy": 0.9922457631677389,
+      "num_tokens": 261000121.0,
+      "step": 7830
+    },
+    {
+      "entropy": 0.047257978632114825,
+      "epoch": 12.544,
+      "grad_norm": 0.43359375,
+      "learning_rate": 4.802222222222222e-06,
+      "loss": 0.0276,
+      "mean_token_accuracy": 0.9921400487422943,
+      "num_tokens": 261331319.0,
+      "step": 7840
+    },
+    {
+      "entropy": 0.04731051673879847,
+      "epoch": 12.56,
+      "grad_norm": 0.5,
+      "learning_rate": 4.78e-06,
+      "loss": 0.0275,
+      "mean_token_accuracy": 0.9919674444943667,
+      "num_tokens": 261662703.0,
+      "step": 7850
+    },
+    {
+      "entropy": 0.04867914315545931,
+      "epoch": 12.576,
+      "grad_norm": 0.4921875,
+      "learning_rate": 4.7577777777777784e-06,
+      "loss": 0.0278,
+      "mean_token_accuracy": 0.991700691357255,
+      "num_tokens": 261992231.0,
+      "step": 7860
+    },
+    {
+      "entropy": 0.048283949587494136,
+      "epoch": 12.592,
+      "grad_norm": 0.361328125,
+      "learning_rate": 4.735555555555556e-06,
+      "loss": 0.0279,
+      "mean_token_accuracy": 0.9917588155716658,
+      "num_tokens": 262317414.0,
+      "step": 7870
+    },
+    {
+      "entropy": 0.04621037285542116,
+      "epoch": 12.608,
+      "grad_norm": 0.4140625,
+      "learning_rate": 4.713333333333334e-06,
+      "loss": 0.027,
+      "mean_token_accuracy": 0.9922691266983747,
+      "num_tokens": 262656539.0,
+      "step": 7880
+    },
+    {
+      "entropy": 0.048271358234342185,
+      "epoch": 12.624,
+      "grad_norm": 0.3828125,
+      "learning_rate": 4.691111111111111e-06,
+      "loss": 0.0282,
+      "mean_token_accuracy": 0.9918258927762509,
+      "num_tokens": 262981547.0,
+      "step": 7890
+    },
+    {
+      "entropy": 0.046158150280825795,
+      "epoch": 12.64,
+      "grad_norm": 0.40234375,
+      "learning_rate": 4.66888888888889e-06,
+      "loss": 0.0272,
+      "mean_token_accuracy": 0.9921741422265768,
+      "num_tokens": 263316542.0,
+      "step": 7900
+    },
+    {
+      "epoch": 12.64,
+      "eval_chemistry_entropy": 0.2319939746260643,
+      "eval_chemistry_loss": 1.4000248908996582,
+      "eval_chemistry_mean_token_accuracy": 0.7867046823501587,
+      "eval_chemistry_num_tokens": 263316542.0,
+      "eval_chemistry_runtime": 34.5864,
+      "eval_chemistry_samples_per_second": 14.457,
+      "eval_chemistry_steps_per_second": 3.614,
+      "step": 7900
+    },
+    {
+      "epoch": 12.64,
+      "eval_physics_entropy": 0.1752476676106453,
+      "eval_physics_loss": 1.1869360208511353,
+      "eval_physics_mean_token_accuracy": 0.8236105089187622,
+      "eval_physics_num_tokens": 263316542.0,
+      "eval_physics_runtime": 41.2646,
+      "eval_physics_samples_per_second": 12.117,
+      "eval_physics_steps_per_second": 3.029,
+      "step": 7900
+    },
+    {
+      "entropy": 0.047235365922097115,
+      "epoch": 12.656,
+      "grad_norm": 0.427734375,
+      "learning_rate": 4.646666666666667e-06,
+      "loss": 0.0276,
+      "mean_token_accuracy": 0.9919094521552324,
+      "num_tokens": 263647560.0,
+      "step": 7910
+    },
+    {
+      "entropy": 0.04638051148504019,
+      "epoch": 12.672,
+      "grad_norm": 0.4375,
+      "learning_rate": 4.624444444444445e-06,
+      "loss": 0.0268,
+      "mean_token_accuracy": 0.9920998543500901,
+      "num_tokens": 263988717.0,
+      "step": 7920
+    },
+    {
+      "entropy": 0.0477219334570691,
+      "epoch": 12.688,
+      "grad_norm": 0.390625,
+      "learning_rate": 4.602222222222223e-06,
+      "loss": 0.0276,
+      "mean_token_accuracy": 0.9920312263071537,
+      "num_tokens": 264320100.0,
+      "step": 7930
+    },
+    {
+      "entropy": 0.04707860058406368,
+      "epoch": 12.704,
+      "grad_norm": 0.4375,
+      "learning_rate": 4.58e-06,
+      "loss": 0.0275,
+      "mean_token_accuracy": 0.9920841138809919,
+      "num_tokens": 264652034.0,
+      "step": 7940
+    },
+    {
+      "entropy": 0.046084119251463565,
+      "epoch": 12.72,
+      "grad_norm": 0.439453125,
+      "learning_rate": 4.557777777777778e-06,
+      "loss": 0.0268,
+      "mean_token_accuracy": 0.9923493307083845,
+      "num_tokens": 264994072.0,
+      "step": 7950
+    },
+    {
+      "entropy": 0.04904477837262675,
+      "epoch": 12.736,
+      "grad_norm": 0.416015625,
+      "learning_rate": 4.535555555555555e-06,
+      "loss": 0.0282,
+      "mean_token_accuracy": 0.9916485507041216,
+      "num_tokens": 265319570.0,
+      "step": 7960
+    },
+    {
+      "entropy": 0.04702546034241095,
+      "epoch": 12.752,
+      "grad_norm": 0.412109375,
+      "learning_rate": 4.513333333333333e-06,
+      "loss": 0.0275,
+      "mean_token_accuracy": 0.9920566383749246,
+      "num_tokens": 265652929.0,
+      "step": 7970
+    },
+    {
+      "entropy": 0.04862874624086544,
+      "epoch": 12.768,
+      "grad_norm": 0.40625,
+      "learning_rate": 4.4911111111111115e-06,
+      "loss": 0.0282,
+      "mean_token_accuracy": 0.9916845068335534,
+      "num_tokens": 265978681.0,
+      "step": 7980
+    },
+    {
+      "entropy": 0.04637602334842086,
+      "epoch": 12.784,
+      "grad_norm": 0.421875,
+      "learning_rate": 4.468888888888889e-06,
+      "loss": 0.0271,
+      "mean_token_accuracy": 0.9922208707779646,
+      "num_tokens": 266317134.0,
+      "step": 7990
+    },
+    {
+      "entropy": 0.046095350617542864,
+      "epoch": 12.8,
+      "grad_norm": 0.4453125,
+      "learning_rate": 4.446666666666667e-06,
+      "loss": 0.0271,
+      "mean_token_accuracy": 0.9920317839831114,
+      "num_tokens": 266653718.0,
+      "step": 8000
+    },
+    {
+      "epoch": 12.8,
+      "eval_chemistry_entropy": 0.23158643394708633,
+      "eval_chemistry_loss": 1.400480031967163,
+      "eval_chemistry_mean_token_accuracy": 0.7866065373420715,
+      "eval_chemistry_num_tokens": 266653718.0,
+      "eval_chemistry_runtime": 34.6225,
+      "eval_chemistry_samples_per_second": 14.441,
+      "eval_chemistry_steps_per_second": 3.61,
+      "step": 8000
+    },
+    {
+      "epoch": 12.8,
+      "eval_physics_entropy": 0.17500228744745255,
+      "eval_physics_loss": 1.1872649192810059,
+      "eval_physics_mean_token_accuracy": 0.8238505530357361,
+      "eval_physics_num_tokens": 266653718.0,
+      "eval_physics_runtime": 40.9781,
+      "eval_physics_samples_per_second": 12.202,
+      "eval_physics_steps_per_second": 3.05,
+      "step": 8000
+    },
+    {
+      "entropy": 0.0470819668378681,
+      "epoch": 12.816,
+      "grad_norm": 0.427734375,
+      "learning_rate": 4.424444444444444e-06,
+      "loss": 0.0275,
+      "mean_token_accuracy": 0.9919280797243119,
+      "num_tokens": 266986144.0,
+      "step": 8010
+    },
+    {
+      "entropy": 0.04731467426754534,
+      "epoch": 12.832,
+      "grad_norm": 0.380859375,
+      "learning_rate": 4.402222222222223e-06,
+      "loss": 0.0275,
+      "mean_token_accuracy": 0.9918080165982246,
+      "num_tokens": 267319553.0,
+      "step": 8020
+    },
+    {
+      "entropy": 0.046327706705778834,
+      "epoch": 12.848,
+      "grad_norm": 0.46484375,
+      "learning_rate": 4.38e-06,
+      "loss": 0.0271,
+      "mean_token_accuracy": 0.992161101475358,
+      "num_tokens": 267657145.0,
+      "step": 8030
+    },
+    {
+      "entropy": 0.046308819111436604,
+      "epoch": 12.864,
+      "grad_norm": 0.41796875,
+      "learning_rate": 4.357777777777778e-06,
+      "loss": 0.0269,
+      "mean_token_accuracy": 0.9922258976846934,
+      "num_tokens": 267997025.0,
+      "step": 8040
+    },
+    {
+      "entropy": 0.047848813293967396,
+      "epoch": 12.88,
+      "grad_norm": 0.41015625,
+      "learning_rate": 4.3355555555555565e-06,
+      "loss": 0.0276,
+      "mean_token_accuracy": 0.9919355485588313,
+      "num_tokens": 268330273.0,
+      "step": 8050
+    },
+    {
+      "entropy": 0.047563881450332704,
+      "epoch": 12.896,
+      "grad_norm": 0.453125,
+      "learning_rate": 4.313333333333334e-06,
+      "loss": 0.0272,
+      "mean_token_accuracy": 0.9920494124293328,
+      "num_tokens": 268668661.0,
+      "step": 8060
+    },
+    {
+      "entropy": 0.04701653203228488,
+      "epoch": 12.912,
+      "grad_norm": 0.46875,
+      "learning_rate": 4.291111111111112e-06,
+      "loss": 0.0276,
+      "mean_token_accuracy": 0.9920382276177406,
+      "num_tokens": 269000866.0,
+      "step": 8070
+    },
+    {
+      "entropy": 0.04945960993645713,
+      "epoch": 12.928,
+      "grad_norm": 0.447265625,
+      "learning_rate": 4.268888888888889e-06,
+      "loss": 0.0288,
+      "mean_token_accuracy": 0.9914790764451027,
+      "num_tokens": 269318126.0,
+      "step": 8080
+    },
+    {
+      "entropy": 0.04679766782792285,
+      "epoch": 12.943999999999999,
+      "grad_norm": 0.421875,
+      "learning_rate": 4.246666666666667e-06,
+      "loss": 0.0274,
+      "mean_token_accuracy": 0.9920704193413258,
+      "num_tokens": 269654539.0,
+      "step": 8090
+    },
+    {
+      "entropy": 0.04841487837256864,
+      "epoch": 12.96,
+      "grad_norm": 0.41796875,
+      "learning_rate": 4.2244444444444446e-06,
+      "loss": 0.0281,
+      "mean_token_accuracy": 0.9916097294539213,
+      "num_tokens": 269980402.0,
+      "step": 8100
+    },
+    {
+      "epoch": 12.96,
+      "eval_chemistry_entropy": 0.2326057094335556,
+      "eval_chemistry_loss": 1.4004255533218384,
+      "eval_chemistry_mean_token_accuracy": 0.7866143679618836,
+      "eval_chemistry_num_tokens": 269980402.0,
+      "eval_chemistry_runtime": 34.6156,
+      "eval_chemistry_samples_per_second": 14.444,
+      "eval_chemistry_steps_per_second": 3.611,
+      "step": 8100
+    },
+    {
+      "epoch": 12.96,
+      "eval_physics_entropy": 0.1757092592716217,
+      "eval_physics_loss": 1.1873486042022705,
+      "eval_physics_mean_token_accuracy": 0.8237286796569824,
+      "eval_physics_num_tokens": 269980402.0,
+      "eval_physics_runtime": 41.2651,
+      "eval_physics_samples_per_second": 12.117,
+      "eval_physics_steps_per_second": 3.029,
+      "step": 8100
+    },
+    {
+      "entropy": 0.047674737905617805,
+      "epoch": 12.975999999999999,
+      "grad_norm": 0.388671875,
+      "learning_rate": 4.202222222222222e-06,
+      "loss": 0.0274,
+      "mean_token_accuracy": 0.9920268796384335,
+      "num_tokens": 270314665.0,
+      "step": 8110
+    },
+    {
+      "entropy": 0.04646383442450315,
+      "epoch": 12.992,
+      "grad_norm": 0.451171875,
+      "learning_rate": 4.18e-06,
+      "loss": 0.0273,
+      "mean_token_accuracy": 0.9921383555978537,
+      "num_tokens": 270651844.0,
+      "step": 8120
+    },
+    {
+      "entropy": 0.048440729232970624,
+      "epoch": 13.008,
+      "grad_norm": 0.427734375,
+      "learning_rate": 4.157777777777778e-06,
+      "loss": 0.0279,
+      "mean_token_accuracy": 0.9918460164219141,
+      "num_tokens": 270976329.0,
+      "step": 8130
+    },
+    {
+      "entropy": 0.047949786437675355,
+      "epoch": 13.024,
+      "grad_norm": 0.43359375,
+      "learning_rate": 4.135555555555556e-06,
+      "loss": 0.0273,
+      "mean_token_accuracy": 0.9920937933027745,
+      "num_tokens": 271305651.0,
+      "step": 8140
+    },
+    {
+      "entropy": 0.04676522330846637,
+      "epoch": 13.04,
+      "grad_norm": 0.40234375,
+      "learning_rate": 4.1133333333333335e-06,
+      "loss": 0.0266,
+      "mean_token_accuracy": 0.992388304322958,
+      "num_tokens": 271641410.0,
+      "step": 8150
+    },
+    {
+      "entropy": 0.04760597330750897,
+      "epoch": 13.056,
+      "grad_norm": 0.38671875,
+      "learning_rate": 4.091111111111111e-06,
+      "loss": 0.027,
+      "mean_token_accuracy": 0.9923049241304398,
+      "num_tokens": 271977638.0,
+      "step": 8160
+    },
+    {
+      "entropy": 0.04567671064287424,
+      "epoch": 13.072,
+      "grad_norm": 0.349609375,
+      "learning_rate": 4.0688888888888896e-06,
+      "loss": 0.0267,
+      "mean_token_accuracy": 0.9922806091606617,
+      "num_tokens": 272313362.0,
+      "step": 8170
+    },
+    {
+      "entropy": 0.046953307010699066,
+      "epoch": 13.088,
+      "grad_norm": 0.40234375,
+      "learning_rate": 4.046666666666667e-06,
+      "loss": 0.0268,
+      "mean_token_accuracy": 0.9922934826463461,
+      "num_tokens": 272648063.0,
+      "step": 8180
+    },
+    {
+      "entropy": 0.04681742761749774,
+      "epoch": 13.104,
+      "grad_norm": 0.375,
+      "learning_rate": 4.024444444444445e-06,
+      "loss": 0.0268,
+      "mean_token_accuracy": 0.9923634801059962,
+      "num_tokens": 272983995.0,
+      "step": 8190
+    },
+    {
+      "entropy": 0.04719428322277963,
+      "epoch": 13.12,
+      "grad_norm": 0.3828125,
+      "learning_rate": 4.002222222222222e-06,
+      "loss": 0.0274,
+      "mean_token_accuracy": 0.9921347040683031,
+      "num_tokens": 273311463.0,
+      "step": 8200
+    },
+    {
+      "epoch": 13.12,
+      "eval_chemistry_entropy": 0.23189832872152327,
+      "eval_chemistry_loss": 1.4016609191894531,
+      "eval_chemistry_mean_token_accuracy": 0.7865863122940063,
+      "eval_chemistry_num_tokens": 273311463.0,
+      "eval_chemistry_runtime": 34.6375,
+      "eval_chemistry_samples_per_second": 14.435,
+      "eval_chemistry_steps_per_second": 3.609,
+      "step": 8200
+    },
+    {
+      "epoch": 13.12,
+      "eval_physics_entropy": 0.1751188644170761,
+      "eval_physics_loss": 1.1881003379821777,
+      "eval_physics_mean_token_accuracy": 0.8236429672241211,
+      "eval_physics_num_tokens": 273311463.0,
+      "eval_physics_runtime": 41.2665,
+      "eval_physics_samples_per_second": 12.116,
+      "eval_physics_steps_per_second": 3.029,
+      "step": 8200
+    },
+    {
+      "entropy": 0.046174107200931756,
+      "epoch": 13.136,
+      "grad_norm": 0.447265625,
+      "learning_rate": 3.980000000000001e-06,
+      "loss": 0.0269,
+      "mean_token_accuracy": 0.9923071503639221,
+      "num_tokens": 273646055.0,
+      "step": 8210
+    },
+    {
+      "entropy": 0.04800432809861377,
+      "epoch": 13.152,
+      "grad_norm": 0.41796875,
+      "learning_rate": 3.9577777777777785e-06,
+      "loss": 0.0275,
+      "mean_token_accuracy": 0.9922458719462156,
+      "num_tokens": 273973730.0,
+      "step": 8220
+    },
+    {
+      "entropy": 0.04695513305487111,
+      "epoch": 13.168,
+      "grad_norm": 0.40234375,
+      "learning_rate": 3.935555555555556e-06,
+      "loss": 0.027,
+      "mean_token_accuracy": 0.9922262959182262,
+      "num_tokens": 274308238.0,
+      "step": 8230
+    },
+    {
+      "entropy": 0.04677723310887814,
+      "epoch": 13.184,
+      "grad_norm": 0.37109375,
+      "learning_rate": 3.913333333333334e-06,
+      "loss": 0.0272,
+      "mean_token_accuracy": 0.9923283621668816,
+      "num_tokens": 274641138.0,
+      "step": 8240
+    },
+    {
+      "entropy": 0.0468893485260196,
+      "epoch": 13.2,
+      "grad_norm": 0.41796875,
+      "learning_rate": 3.891111111111111e-06,
+      "loss": 0.0271,
+      "mean_token_accuracy": 0.9922400940209627,
+      "num_tokens": 274973868.0,
+      "step": 8250
+    },
+    {
+      "entropy": 0.047284358250908556,
+      "epoch": 13.216,
+      "grad_norm": 0.396484375,
+      "learning_rate": 3.868888888888889e-06,
+      "loss": 0.0275,
+      "mean_token_accuracy": 0.9920473750680685,
+      "num_tokens": 275301395.0,
+      "step": 8260
+    },
+    {
+      "entropy": 0.04818462786497548,
+      "epoch": 13.232,
+      "grad_norm": 0.419921875,
+      "learning_rate": 3.8466666666666665e-06,
+      "loss": 0.0277,
+      "mean_token_accuracy": 0.9919338777661324,
+      "num_tokens": 275625514.0,
+      "step": 8270
+    },
+    {
+      "entropy": 0.047513406269717964,
+      "epoch": 13.248,
+      "grad_norm": 0.4296875,
+      "learning_rate": 3.824444444444444e-06,
+      "loss": 0.0274,
+      "mean_token_accuracy": 0.9919922914355993,
+      "num_tokens": 275954975.0,
+      "step": 8280
+    },
+    {
+      "entropy": 0.04754307991825044,
+      "epoch": 13.264,
+      "grad_norm": 0.4921875,
+      "learning_rate": 3.8022222222222226e-06,
+      "loss": 0.0277,
+      "mean_token_accuracy": 0.9921210341155529,
+      "num_tokens": 276280177.0,
+      "step": 8290
+    },
+    {
+      "entropy": 0.047081450535915795,
+      "epoch": 13.28,
+      "grad_norm": 0.3984375,
+      "learning_rate": 3.7800000000000002e-06,
+      "loss": 0.0273,
+      "mean_token_accuracy": 0.9922505389899016,
+      "num_tokens": 276610984.0,
+      "step": 8300
+    },
+    {
+      "epoch": 13.28,
+      "eval_chemistry_entropy": 0.23142269897460938,
+      "eval_chemistry_loss": 1.402401328086853,
+      "eval_chemistry_mean_token_accuracy": 0.7865597581863404,
+      "eval_chemistry_num_tokens": 276610984.0,
+      "eval_chemistry_runtime": 34.6201,
+      "eval_chemistry_samples_per_second": 14.442,
+      "eval_chemistry_steps_per_second": 3.611,
+      "step": 8300
+    },
+    {
+      "epoch": 13.28,
+      "eval_physics_entropy": 0.17478029161691666,
+      "eval_physics_loss": 1.1889208555221558,
+      "eval_physics_mean_token_accuracy": 0.823728690624237,
+      "eval_physics_num_tokens": 276610984.0,
+      "eval_physics_runtime": 41.2819,
+      "eval_physics_samples_per_second": 12.112,
+      "eval_physics_steps_per_second": 3.028,
+      "step": 8300
+    },
+    {
+      "entropy": 0.04560810077236965,
+      "epoch": 13.296,
+      "grad_norm": 0.38671875,
+      "learning_rate": 3.757777777777778e-06,
+      "loss": 0.0262,
+      "mean_token_accuracy": 0.9925641395151615,
+      "num_tokens": 276955876.0,
+      "step": 8310
+    },
+    {
+      "entropy": 0.045926834479905664,
+      "epoch": 13.312,
+      "grad_norm": 0.388671875,
+      "learning_rate": 3.7355555555555555e-06,
+      "loss": 0.0269,
+      "mean_token_accuracy": 0.9922940567135811,
+      "num_tokens": 277291484.0,
+      "step": 8320
+    },
+    {
+      "entropy": 0.0468043290427886,
+      "epoch": 13.328,
+      "grad_norm": 0.388671875,
+      "learning_rate": 3.713333333333334e-06,
+      "loss": 0.0271,
+      "mean_token_accuracy": 0.9922083973884582,
+      "num_tokens": 277626344.0,
+      "step": 8330
+    },
+    {
+      "entropy": 0.047565514396410435,
+      "epoch": 13.344,
+      "grad_norm": 0.41015625,
+      "learning_rate": 3.6911111111111115e-06,
+      "loss": 0.0272,
+      "mean_token_accuracy": 0.9921212323009968,
+      "num_tokens": 277958936.0,
+      "step": 8340
+    },
+    {
+      "entropy": 0.04647928281920031,
+      "epoch": 13.36,
+      "grad_norm": 0.376953125,
+      "learning_rate": 3.668888888888889e-06,
+      "loss": 0.027,
+      "mean_token_accuracy": 0.9922680966556072,
+      "num_tokens": 278293564.0,
+      "step": 8350
+    },
+    {
+      "entropy": 0.047966270649340004,
+      "epoch": 13.376,
+      "grad_norm": 0.4453125,
+      "learning_rate": 3.6466666666666668e-06,
+      "loss": 0.0275,
+      "mean_token_accuracy": 0.9920612812042237,
+      "num_tokens": 278622466.0,
+      "step": 8360
+    },
+    {
+      "entropy": 0.04638690529391169,
+      "epoch": 13.392,
+      "grad_norm": 0.4375,
+      "learning_rate": 3.624444444444445e-06,
+      "loss": 0.0271,
+      "mean_token_accuracy": 0.992247536033392,
+      "num_tokens": 278956369.0,
+      "step": 8370
+    },
+    {
+      "entropy": 0.04688066720264032,
+      "epoch": 13.408,
+      "grad_norm": 0.41015625,
+      "learning_rate": 3.6022222222222224e-06,
+      "loss": 0.0269,
+      "mean_token_accuracy": 0.9921065874397754,
+      "num_tokens": 279292109.0,
+      "step": 8380
+    },
+    {
+      "entropy": 0.046442077262327075,
+      "epoch": 13.424,
+      "grad_norm": 0.396484375,
+      "learning_rate": 3.58e-06,
+      "loss": 0.0264,
+      "mean_token_accuracy": 0.9922865692526102,
+      "num_tokens": 279633578.0,
+      "step": 8390
+    },
+    {
+      "entropy": 0.046268373262137175,
+      "epoch": 13.44,
+      "grad_norm": 0.390625,
+      "learning_rate": 3.5577777777777785e-06,
+      "loss": 0.027,
+      "mean_token_accuracy": 0.9923589698970318,
+      "num_tokens": 279968537.0,
+      "step": 8400
+    },
+    {
+      "epoch": 13.44,
+      "eval_chemistry_entropy": 0.23127510923147201,
+      "eval_chemistry_loss": 1.4031931161880493,
+      "eval_chemistry_mean_token_accuracy": 0.7864331722259521,
+      "eval_chemistry_num_tokens": 279968537.0,
+      "eval_chemistry_runtime": 34.5964,
+      "eval_chemistry_samples_per_second": 14.452,
+      "eval_chemistry_steps_per_second": 3.613,
+      "step": 8400
+    },
+    {
+      "epoch": 13.44,
+      "eval_physics_entropy": 0.17467521530389785,
+      "eval_physics_loss": 1.1895179748535156,
+      "eval_physics_mean_token_accuracy": 0.8235966181755066,
+      "eval_physics_num_tokens": 279968537.0,
+      "eval_physics_runtime": 41.2344,
+      "eval_physics_samples_per_second": 12.126,
+      "eval_physics_steps_per_second": 3.031,
+      "step": 8400
+    },
+    {
+      "entropy": 0.04732351736165583,
+      "epoch": 13.456,
+      "grad_norm": 0.4609375,
+      "learning_rate": 3.535555555555556e-06,
+      "loss": 0.0273,
+      "mean_token_accuracy": 0.9920300986617804,
+      "num_tokens": 280298759.0,
+      "step": 8410
+    },
+    {
+      "entropy": 0.04656473679933697,
+      "epoch": 13.472,
+      "grad_norm": 0.3828125,
+      "learning_rate": 3.5133333333333337e-06,
+      "loss": 0.0268,
+      "mean_token_accuracy": 0.992165508493781,
+      "num_tokens": 280635071.0,
+      "step": 8420
+    },
+    {
+      "entropy": 0.04573026488069445,
+      "epoch": 13.488,
+      "grad_norm": 0.38671875,
+      "learning_rate": 3.4911111111111113e-06,
+      "loss": 0.0264,
+      "mean_token_accuracy": 0.9924745764583349,
+      "num_tokens": 280977712.0,
+      "step": 8430
+    },
+    {
+      "entropy": 0.04696386993164196,
+      "epoch": 13.504,
+      "grad_norm": 0.42578125,
+      "learning_rate": 3.4688888888888894e-06,
+      "loss": 0.0268,
+      "mean_token_accuracy": 0.9922691740095615,
+      "num_tokens": 281315707.0,
+      "step": 8440
+    },
+    {
+      "entropy": 0.04728903893847018,
+      "epoch": 13.52,
+      "grad_norm": 0.423828125,
+      "learning_rate": 3.446666666666667e-06,
+      "loss": 0.0274,
+      "mean_token_accuracy": 0.9920141603797674,
+      "num_tokens": 281646175.0,
+      "step": 8450
+    },
+    {
+      "entropy": 0.0466050231247209,
+      "epoch": 13.536,
+      "grad_norm": 0.42578125,
+      "learning_rate": 3.4244444444444446e-06,
+      "loss": 0.0269,
+      "mean_token_accuracy": 0.9921470366418361,
+      "num_tokens": 281981824.0,
+      "step": 8460
+    },
+    {
+      "entropy": 0.046762042271438986,
+      "epoch": 13.552,
+      "grad_norm": 0.4453125,
+      "learning_rate": 3.4022222222222222e-06,
+      "loss": 0.027,
+      "mean_token_accuracy": 0.9922975119203329,
+      "num_tokens": 282317631.0,
+      "step": 8470
+    },
+    {
+      "entropy": 0.046874707494862375,
+      "epoch": 13.568,
+      "grad_norm": 0.3984375,
+      "learning_rate": 3.3800000000000007e-06,
+      "loss": 0.0274,
+      "mean_token_accuracy": 0.9921759441494942,
+      "num_tokens": 282646885.0,
+      "step": 8480
+    },
+    {
+      "entropy": 0.048530175734777005,
+      "epoch": 13.584,
+      "grad_norm": 0.462890625,
+      "learning_rate": 3.3577777777777783e-06,
+      "loss": 0.028,
+      "mean_token_accuracy": 0.9918554745614528,
+      "num_tokens": 282967621.0,
+      "step": 8490
+    },
+    {
+      "entropy": 0.046591929323039946,
+      "epoch": 13.6,
+      "grad_norm": 0.376953125,
+      "learning_rate": 3.335555555555556e-06,
+      "loss": 0.0271,
+      "mean_token_accuracy": 0.9921377964317799,
+      "num_tokens": 283300989.0,
+      "step": 8500
+    },
+    {
+      "epoch": 13.6,
+      "eval_chemistry_entropy": 0.2317947347164154,
+      "eval_chemistry_loss": 1.4032597541809082,
+      "eval_chemistry_mean_token_accuracy": 0.7864612159729004,
+      "eval_chemistry_num_tokens": 283300989.0,
+      "eval_chemistry_runtime": 34.5942,
+      "eval_chemistry_samples_per_second": 14.453,
+      "eval_chemistry_steps_per_second": 3.613,
+      "step": 8500
+    },
+    {
+      "epoch": 13.6,
+      "eval_physics_entropy": 0.17504956197738647,
+      "eval_physics_loss": 1.1898363828659058,
+      "eval_physics_mean_token_accuracy": 0.823625750541687,
+      "eval_physics_num_tokens": 283300989.0,
+      "eval_physics_runtime": 40.9425,
+      "eval_physics_samples_per_second": 12.212,
+      "eval_physics_steps_per_second": 3.053,
+      "step": 8500
+    },
+    {
+      "entropy": 0.04840943677118048,
+      "epoch": 13.616,
+      "grad_norm": 0.443359375,
+      "learning_rate": 3.3133333333333335e-06,
+      "loss": 0.0277,
+      "mean_token_accuracy": 0.9918745446950197,
+      "num_tokens": 283627913.0,
+      "step": 8510
+    },
+    {
+      "entropy": 0.04781493510818109,
+      "epoch": 13.632,
+      "grad_norm": 0.421875,
+      "learning_rate": 3.2911111111111116e-06,
+      "loss": 0.0274,
+      "mean_token_accuracy": 0.9918079081922769,
+      "num_tokens": 283958337.0,
+      "step": 8520
+    },
+    {
+      "entropy": 0.045751148799899964,
+      "epoch": 13.648,
+      "grad_norm": 0.453125,
+      "learning_rate": 3.268888888888889e-06,
+      "loss": 0.0268,
+      "mean_token_accuracy": 0.9921128552407026,
+      "num_tokens": 284295898.0,
+      "step": 8530
+    },
+    {
+      "entropy": 0.04728087403345853,
+      "epoch": 13.664,
+      "grad_norm": 0.439453125,
+      "learning_rate": 3.2466666666666668e-06,
+      "loss": 0.0277,
+      "mean_token_accuracy": 0.9919241178780794,
+      "num_tokens": 284622683.0,
+      "step": 8540
+    },
+    {
+      "entropy": 0.04760847404832021,
+      "epoch": 13.68,
+      "grad_norm": 0.48046875,
+      "learning_rate": 3.2244444444444444e-06,
+      "loss": 0.0272,
+      "mean_token_accuracy": 0.9920303765684366,
+      "num_tokens": 284954820.0,
+      "step": 8550
+    },
+    {
+      "entropy": 0.04592409890610725,
+      "epoch": 13.696,
+      "grad_norm": 0.462890625,
+      "learning_rate": 3.202222222222223e-06,
+      "loss": 0.027,
+      "mean_token_accuracy": 0.9922276195138693,
+      "num_tokens": 285290884.0,
+      "step": 8560
+    },
+    {
+      "entropy": 0.047011823602952064,
+      "epoch": 13.712,
+      "grad_norm": 0.390625,
+      "learning_rate": 3.1800000000000005e-06,
+      "loss": 0.0271,
+      "mean_token_accuracy": 0.9920003108680249,
+      "num_tokens": 285624637.0,
+      "step": 8570
+    },
+    {
+      "entropy": 0.0472241434501484,
+      "epoch": 13.728,
+      "grad_norm": 0.3984375,
+      "learning_rate": 3.157777777777778e-06,
+      "loss": 0.0269,
+      "mean_token_accuracy": 0.9919873371720314,
+      "num_tokens": 285960562.0,
+      "step": 8580
+    },
+    {
+      "entropy": 0.04670818275772035,
+      "epoch": 13.744,
+      "grad_norm": 0.423828125,
+      "learning_rate": 3.1355555555555557e-06,
+      "loss": 0.0271,
+      "mean_token_accuracy": 0.991954792663455,
+      "num_tokens": 286293490.0,
+      "step": 8590
+    },
+    {
+      "entropy": 0.046361166355200115,
+      "epoch": 13.76,
+      "grad_norm": 0.37890625,
+      "learning_rate": 3.1133333333333337e-06,
+      "loss": 0.0267,
+      "mean_token_accuracy": 0.9920958980917931,
+      "num_tokens": 286630530.0,
+      "step": 8600
+    },
+    {
+      "epoch": 13.76,
+      "eval_chemistry_entropy": 0.23179515421390534,
+      "eval_chemistry_loss": 1.4032371044158936,
+      "eval_chemistry_mean_token_accuracy": 0.7865589370727539,
+      "eval_chemistry_num_tokens": 286630530.0,
+      "eval_chemistry_runtime": 34.6172,
+      "eval_chemistry_samples_per_second": 14.444,
+      "eval_chemistry_steps_per_second": 3.611,
+      "step": 8600
+    },
+    {
+      "epoch": 13.76,
+      "eval_physics_entropy": 0.17493844276666642,
+      "eval_physics_loss": 1.1899559497833252,
+      "eval_physics_mean_token_accuracy": 0.8236779232025147,
+      "eval_physics_num_tokens": 286630530.0,
+      "eval_physics_runtime": 41.2312,
+      "eval_physics_samples_per_second": 12.127,
+      "eval_physics_steps_per_second": 3.032,
+      "step": 8600
+    },
+    {
+      "entropy": 0.04748656039591879,
+      "epoch": 13.776,
+      "grad_norm": 0.4453125,
+      "learning_rate": 3.0911111111111114e-06,
+      "loss": 0.0273,
+      "mean_token_accuracy": 0.9919046938419342,
+      "num_tokens": 286962647.0,
+      "step": 8610
+    },
+    {
+      "entropy": 0.045754980435594915,
+      "epoch": 13.792,
+      "grad_norm": 0.427734375,
+      "learning_rate": 3.068888888888889e-06,
+      "loss": 0.0268,
+      "mean_token_accuracy": 0.9922747708857059,
+      "num_tokens": 287300408.0,
+      "step": 8620
+    },
+    {
+      "entropy": 0.046366979146841916,
+      "epoch": 13.808,
+      "grad_norm": 0.451171875,
+      "learning_rate": 3.0466666666666666e-06,
+      "loss": 0.0271,
+      "mean_token_accuracy": 0.9920321416109801,
+      "num_tokens": 287633713.0,
+      "step": 8630
+    },
+    {
+      "entropy": 0.04801953637506813,
+      "epoch": 13.824,
+      "grad_norm": 0.41015625,
+      "learning_rate": 3.024444444444445e-06,
+      "loss": 0.0281,
+      "mean_token_accuracy": 0.9918234799057245,
+      "num_tokens": 287956523.0,
+      "step": 8640
+    },
+    {
+      "entropy": 0.04689986879238859,
+      "epoch": 13.84,
+      "grad_norm": 0.416015625,
+      "learning_rate": 3.0022222222222227e-06,
+      "loss": 0.0276,
+      "mean_token_accuracy": 0.9920868016779423,
+      "num_tokens": 288285017.0,
+      "step": 8650
+    },
+    {
+      "entropy": 0.04772832897724584,
+      "epoch": 13.856,
+      "grad_norm": 0.435546875,
+      "learning_rate": 2.9800000000000003e-06,
+      "loss": 0.0277,
+      "mean_token_accuracy": 0.9917003199458122,
+      "num_tokens": 288610143.0,
+      "step": 8660
+    },
+    {
+      "entropy": 0.04366730636684224,
+      "epoch": 13.872,
+      "grad_norm": 0.38671875,
+      "learning_rate": 2.957777777777778e-06,
+      "loss": 0.0253,
+      "mean_token_accuracy": 0.9928082872182131,
+      "num_tokens": 288969251.0,
+      "step": 8670
+    },
+    {
+      "entropy": 0.046243089984636755,
+      "epoch": 13.888,
+      "grad_norm": 0.453125,
+      "learning_rate": 2.935555555555556e-06,
+      "loss": 0.0264,
+      "mean_token_accuracy": 0.9922671038657427,
+      "num_tokens": 289312553.0,
+      "step": 8680
+    },
+    {
+      "entropy": 0.04785272516310215,
+      "epoch": 13.904,
+      "grad_norm": 0.41015625,
+      "learning_rate": 2.9133333333333335e-06,
+      "loss": 0.0273,
+      "mean_token_accuracy": 0.9920256718993187,
+      "num_tokens": 289644642.0,
+      "step": 8690
+    },
+    {
+      "entropy": 0.04700564269442111,
+      "epoch": 13.92,
+      "grad_norm": 0.427734375,
+      "learning_rate": 2.891111111111111e-06,
+      "loss": 0.027,
+      "mean_token_accuracy": 0.9920411620289087,
+      "num_tokens": 289980456.0,
+      "step": 8700
+    },
+    {
+      "epoch": 13.92,
+      "eval_chemistry_entropy": 0.23111346340179442,
+      "eval_chemistry_loss": 1.4026784896850586,
+      "eval_chemistry_mean_token_accuracy": 0.7865741834640503,
+      "eval_chemistry_num_tokens": 289980456.0,
+      "eval_chemistry_runtime": 34.6063,
+      "eval_chemistry_samples_per_second": 14.448,
+      "eval_chemistry_steps_per_second": 3.612,
+      "step": 8700
+    },
+    {
+      "epoch": 13.92,
+      "eval_physics_entropy": 0.1744995818734169,
+      "eval_physics_loss": 1.189294457435608,
+      "eval_physics_mean_token_accuracy": 0.8236797914505005,
+      "eval_physics_num_tokens": 289980456.0,
+      "eval_physics_runtime": 41.2501,
+      "eval_physics_samples_per_second": 12.121,
+      "eval_physics_steps_per_second": 3.03,
+      "step": 8700
+    },
+    {
+      "entropy": 0.04785891758510843,
+      "epoch": 13.936,
+      "grad_norm": 0.453125,
+      "learning_rate": 2.868888888888889e-06,
+      "loss": 0.0278,
+      "mean_token_accuracy": 0.9918517608195543,
+      "num_tokens": 290305589.0,
+      "step": 8710
+    },
+    {
+      "entropy": 0.046141260163858534,
+      "epoch": 13.952,
+      "grad_norm": 0.42578125,
+      "learning_rate": 2.8466666666666672e-06,
+      "loss": 0.0267,
+      "mean_token_accuracy": 0.9921846803277731,
+      "num_tokens": 290646288.0,
+      "step": 8720
+    },
+    {
+      "entropy": 0.04761830334318802,
+      "epoch": 13.968,
+      "grad_norm": 0.455078125,
+      "learning_rate": 2.824444444444445e-06,
+      "loss": 0.0276,
+      "mean_token_accuracy": 0.9919141631573438,
+      "num_tokens": 290975813.0,
+      "step": 8730
+    },
+    {
+      "entropy": 0.04500821526162326,
+      "epoch": 13.984,
+      "grad_norm": 0.58203125,
+      "learning_rate": 2.8022222222222225e-06,
+      "loss": 0.0264,
+      "mean_token_accuracy": 0.9924173794686795,
+      "num_tokens": 291322505.0,
+      "step": 8740
+    },
+    {
+      "entropy": 0.047368002007715404,
+      "epoch": 14.0,
+      "grad_norm": 0.45703125,
+      "learning_rate": 2.7800000000000005e-06,
+      "loss": 0.0276,
+      "mean_token_accuracy": 0.9918529242277145,
+      "num_tokens": 291650072.0,
+      "step": 8750
+    },
+    {
+      "entropy": 0.04701939327642322,
+      "epoch": 14.016,
+      "grad_norm": 0.39453125,
+      "learning_rate": 2.757777777777778e-06,
+      "loss": 0.0271,
+      "mean_token_accuracy": 0.992370355129242,
+      "num_tokens": 291981477.0,
+      "step": 8760
+    },
+    {
+      "entropy": 0.04676372940884903,
+      "epoch": 14.032,
+      "grad_norm": 0.392578125,
+      "learning_rate": 2.7355555555555557e-06,
+      "loss": 0.0268,
+      "mean_token_accuracy": 0.9923609357327223,
+      "num_tokens": 292315484.0,
+      "step": 8770
+    },
+    {
+      "entropy": 0.045877280598506334,
+      "epoch": 14.048,
+      "grad_norm": 0.44140625,
+      "learning_rate": 2.7133333333333333e-06,
+      "loss": 0.0265,
+      "mean_token_accuracy": 0.9922053273767233,
+      "num_tokens": 292652863.0,
+      "step": 8780
+    },
+    {
+      "entropy": 0.04796519298106432,
+      "epoch": 14.064,
+      "grad_norm": 0.427734375,
+      "learning_rate": 2.6911111111111114e-06,
+      "loss": 0.0274,
+      "mean_token_accuracy": 0.9919871296733618,
+      "num_tokens": 292979970.0,
+      "step": 8790
+    },
+    {
+      "entropy": 0.045646668074186894,
+      "epoch": 14.08,
+      "grad_norm": 0.38671875,
+      "learning_rate": 2.6688888888888894e-06,
+      "loss": 0.0266,
+      "mean_token_accuracy": 0.992378706112504,
+      "num_tokens": 293317333.0,
+      "step": 8800
+    },
+    {
+      "epoch": 14.08,
+      "eval_chemistry_entropy": 0.23110914415121078,
+      "eval_chemistry_loss": 1.4031108617782593,
+      "eval_chemistry_mean_token_accuracy": 0.7866686191558838,
+      "eval_chemistry_num_tokens": 293317333.0,
+      "eval_chemistry_runtime": 34.6093,
+      "eval_chemistry_samples_per_second": 14.447,
+      "eval_chemistry_steps_per_second": 3.612,
+      "step": 8800
+    },
+    {
+      "epoch": 14.08,
+      "eval_physics_entropy": 0.174433416724205,
+      "eval_physics_loss": 1.18995201587677,
+      "eval_physics_mean_token_accuracy": 0.8236737580299377,
+      "eval_physics_num_tokens": 293317333.0,
+      "eval_physics_runtime": 41.2302,
+      "eval_physics_samples_per_second": 12.127,
+      "eval_physics_steps_per_second": 3.032,
+      "step": 8800
+    },
+    {
+      "entropy": 0.04783440825995058,
+      "epoch": 14.096,
+      "grad_norm": 0.412109375,
+      "learning_rate": 2.646666666666667e-06,
+      "loss": 0.0274,
+      "mean_token_accuracy": 0.9921569172292948,
+      "num_tokens": 293645713.0,
+      "step": 8810
+    },
+    {
+      "entropy": 0.04583299480145797,
+      "epoch": 14.112,
+      "grad_norm": 0.37890625,
+      "learning_rate": 2.6244444444444446e-06,
+      "loss": 0.0265,
+      "mean_token_accuracy": 0.9924966610968113,
+      "num_tokens": 293983358.0,
+      "step": 8820
+    },
+    {
+      "entropy": 0.04600594906369224,
+      "epoch": 14.128,
+      "grad_norm": 0.376953125,
+      "learning_rate": 2.6022222222222227e-06,
+      "loss": 0.0263,
+      "mean_token_accuracy": 0.9923935204744339,
+      "num_tokens": 294323665.0,
+      "step": 8830
+    },
+    {
+      "entropy": 0.047379521862603725,
+      "epoch": 14.144,
+      "grad_norm": 0.416015625,
+      "learning_rate": 2.5800000000000003e-06,
+      "loss": 0.0275,
+      "mean_token_accuracy": 0.9921382214874029,
+      "num_tokens": 294649185.0,
+      "step": 8840
+    },
+    {
+      "entropy": 0.047153377044014635,
+      "epoch": 14.16,
+      "grad_norm": 0.41015625,
+      "learning_rate": 2.557777777777778e-06,
+      "loss": 0.0271,
+      "mean_token_accuracy": 0.9922716565430164,
+      "num_tokens": 294980633.0,
+      "step": 8850
+    },
+    {
+      "entropy": 0.04657066201325506,
+      "epoch": 14.176,
+      "grad_norm": 0.419921875,
+      "learning_rate": 2.5355555555555555e-06,
+      "loss": 0.0268,
+      "mean_token_accuracy": 0.9923305183649063,
+      "num_tokens": 295315635.0,
+      "step": 8860
+    },
+    {
+      "entropy": 0.04656577772693708,
+      "epoch": 14.192,
+      "grad_norm": 0.392578125,
+      "learning_rate": 2.5133333333333336e-06,
+      "loss": 0.0268,
+      "mean_token_accuracy": 0.9922851853072643,
+      "num_tokens": 295650654.0,
+      "step": 8870
+    },
+    {
+      "entropy": 0.04785572569817305,
+      "epoch": 14.208,
+      "grad_norm": 0.41796875,
+      "learning_rate": 2.491111111111111e-06,
+      "loss": 0.0276,
+      "mean_token_accuracy": 0.991978519782424,
+      "num_tokens": 295975765.0,
+      "step": 8880
+    },
+    {
+      "entropy": 0.04786907739471644,
+      "epoch": 14.224,
+      "grad_norm": 0.416015625,
+      "learning_rate": 2.468888888888889e-06,
+      "loss": 0.0277,
+      "mean_token_accuracy": 0.9921110771596432,
+      "num_tokens": 296299671.0,
+      "step": 8890
+    },
+    {
+      "entropy": 0.04755596778122708,
+      "epoch": 14.24,
+      "grad_norm": 0.419921875,
+      "learning_rate": 2.446666666666667e-06,
+      "loss": 0.0275,
+      "mean_token_accuracy": 0.9921339806169271,
+      "num_tokens": 296627350.0,
+      "step": 8900
+    },
+    {
+      "epoch": 14.24,
+      "eval_chemistry_entropy": 0.23117413228750228,
+      "eval_chemistry_loss": 1.403738260269165,
+      "eval_chemistry_mean_token_accuracy": 0.7866291980743408,
+      "eval_chemistry_num_tokens": 296627350.0,
+      "eval_chemistry_runtime": 34.6039,
+      "eval_chemistry_samples_per_second": 14.449,
+      "eval_chemistry_steps_per_second": 3.612,
+      "step": 8900
+    },
+    {
+      "epoch": 14.24,
+      "eval_physics_entropy": 0.17450073474645614,
+      "eval_physics_loss": 1.1902495622634888,
+      "eval_physics_mean_token_accuracy": 0.823687020778656,
+      "eval_physics_num_tokens": 296627350.0,
+      "eval_physics_runtime": 41.254,
+      "eval_physics_samples_per_second": 12.12,
+      "eval_physics_steps_per_second": 3.03,
+      "step": 8900
+    },
+    {
+      "entropy": 0.04534710723673925,
+      "epoch": 14.256,
+      "grad_norm": 0.396484375,
+      "learning_rate": 2.4244444444444444e-06,
+      "loss": 0.0259,
+      "mean_token_accuracy": 0.9924611084163188,
+      "num_tokens": 296974800.0,
+      "step": 8910
+    },
+    {
+      "entropy": 0.0470557852066122,
+      "epoch": 14.272,
+      "grad_norm": 0.392578125,
+      "learning_rate": 2.4022222222222225e-06,
+      "loss": 0.0274,
+      "mean_token_accuracy": 0.9922266885638237,
+      "num_tokens": 297303400.0,
+      "step": 8920
+    },
+    {
+      "entropy": 0.04700539183104411,
+      "epoch": 14.288,
+      "grad_norm": 0.375,
+      "learning_rate": 2.38e-06,
+      "loss": 0.0269,
+      "mean_token_accuracy": 0.992211253196001,
+      "num_tokens": 297637511.0,
+      "step": 8930
+    },
+    {
+      "entropy": 0.04955838636960834,
+      "epoch": 14.304,
+      "grad_norm": 0.447265625,
+      "learning_rate": 2.357777777777778e-06,
+      "loss": 0.0283,
+      "mean_token_accuracy": 0.9918044660240412,
+      "num_tokens": 297955882.0,
+      "step": 8940
+    },
+    {
+      "entropy": 0.04762960836524144,
+      "epoch": 14.32,
+      "grad_norm": 0.40234375,
+      "learning_rate": 2.3355555555555557e-06,
+      "loss": 0.027,
+      "mean_token_accuracy": 0.992082392051816,
+      "num_tokens": 298287601.0,
+      "step": 8950
+    },
+    {
+      "entropy": 0.046825525944586846,
+      "epoch": 14.336,
+      "grad_norm": 0.47265625,
+      "learning_rate": 2.3133333333333333e-06,
+      "loss": 0.0267,
+      "mean_token_accuracy": 0.9920855965465307,
+      "num_tokens": 298625218.0,
+      "step": 8960
+    },
+    {
+      "entropy": 0.047198918892536315,
+      "epoch": 14.352,
+      "grad_norm": 0.46484375,
+      "learning_rate": 2.2911111111111114e-06,
+      "loss": 0.0272,
+      "mean_token_accuracy": 0.9921624083071947,
+      "num_tokens": 298955473.0,
+      "step": 8970
+    },
+    {
+      "entropy": 0.047780354809947315,
+      "epoch": 14.368,
+      "grad_norm": 0.443359375,
+      "learning_rate": 2.268888888888889e-06,
+      "loss": 0.0277,
+      "mean_token_accuracy": 0.9920976348221302,
+      "num_tokens": 299280026.0,
+      "step": 8980
+    },
+    {
+      "entropy": 0.04666613774606958,
+      "epoch": 14.384,
+      "grad_norm": 0.4140625,
+      "learning_rate": 2.2466666666666666e-06,
+      "loss": 0.0271,
+      "mean_token_accuracy": 0.9921989075839519,
+      "num_tokens": 299612135.0,
+      "step": 8990
+    },
+    {
+      "entropy": 0.04387344506103545,
+      "epoch": 14.4,
+      "grad_norm": 0.390625,
+      "learning_rate": 2.2244444444444447e-06,
+      "loss": 0.0254,
+      "mean_token_accuracy": 0.9928333893418312,
+      "num_tokens": 299967668.0,
+      "step": 9000
+    },
+    {
+      "epoch": 14.4,
+      "eval_chemistry_entropy": 0.23094946533441543,
+      "eval_chemistry_loss": 1.4041998386383057,
+      "eval_chemistry_mean_token_accuracy": 0.7865245981216431,
+      "eval_chemistry_num_tokens": 299967668.0,
+      "eval_chemistry_runtime": 34.6044,
+      "eval_chemistry_samples_per_second": 14.449,
+      "eval_chemistry_steps_per_second": 3.612,
+      "step": 9000
+    },
+    {
+      "epoch": 14.4,
+      "eval_physics_entropy": 0.17434898179769515,
+      "eval_physics_loss": 1.190666913986206,
+      "eval_physics_mean_token_accuracy": 0.8236196632385254,
+      "eval_physics_num_tokens": 299967668.0,
+      "eval_physics_runtime": 40.9418,
+      "eval_physics_samples_per_second": 12.212,
+      "eval_physics_steps_per_second": 3.053,
+      "step": 9000
+    },
+    {
+      "entropy": 0.04618626976152882,
+      "epoch": 14.416,
+      "grad_norm": 0.40234375,
+      "learning_rate": 2.2022222222222227e-06,
+      "loss": 0.0266,
+      "mean_token_accuracy": 0.9922905556857586,
+      "num_tokens": 300305666.0,
+      "step": 9010
+    },
+    {
+      "entropy": 0.046593683981336655,
+      "epoch": 14.432,
+      "grad_norm": 0.419921875,
+      "learning_rate": 2.1800000000000003e-06,
+      "loss": 0.0271,
+      "mean_token_accuracy": 0.9923408150672912,
+      "num_tokens": 300637055.0,
+      "step": 9020
+    },
+    {
+      "entropy": 0.04682989446446299,
+      "epoch": 14.448,
+      "grad_norm": 0.43359375,
+      "learning_rate": 2.157777777777778e-06,
+      "loss": 0.0273,
+      "mean_token_accuracy": 0.9921644061803818,
+      "num_tokens": 300967148.0,
+      "step": 9030
+    },
+    {
+      "entropy": 0.04802712860982865,
+      "epoch": 14.464,
+      "grad_norm": 0.443359375,
+      "learning_rate": 2.1355555555555555e-06,
+      "loss": 0.0274,
+      "mean_token_accuracy": 0.9919784449040889,
+      "num_tokens": 301295976.0,
+      "step": 9040
+    },
+    {
+      "entropy": 0.047026652726344764,
+      "epoch": 14.48,
+      "grad_norm": 0.404296875,
+      "learning_rate": 2.1133333333333336e-06,
+      "loss": 0.0272,
+      "mean_token_accuracy": 0.9922332696616649,
+      "num_tokens": 301626993.0,
+      "step": 9050
+    },
+    {
+      "entropy": 0.046344331128057095,
+      "epoch": 14.496,
+      "grad_norm": 0.4375,
+      "learning_rate": 2.091111111111111e-06,
+      "loss": 0.0265,
+      "mean_token_accuracy": 0.9921721555292606,
+      "num_tokens": 301965972.0,
+      "step": 9060
+    },
+    {
+      "entropy": 0.0472197690512985,
+      "epoch": 14.512,
+      "grad_norm": 0.4296875,
+      "learning_rate": 2.0688888888888892e-06,
+      "loss": 0.0271,
+      "mean_token_accuracy": 0.9923027787357569,
+      "num_tokens": 302299045.0,
+      "step": 9070
+    },
+    {
+      "entropy": 0.04541057139867917,
+      "epoch": 14.528,
+      "grad_norm": 0.451171875,
+      "learning_rate": 2.046666666666667e-06,
+      "loss": 0.0263,
+      "mean_token_accuracy": 0.9924955647438765,
+      "num_tokens": 302641339.0,
+      "step": 9080
+    },
+    {
+      "entropy": 0.048486329230945556,
+      "epoch": 14.544,
+      "grad_norm": 0.427734375,
+      "learning_rate": 2.024444444444445e-06,
+      "loss": 0.0278,
+      "mean_token_accuracy": 0.9918663572520018,
+      "num_tokens": 302965245.0,
+      "step": 9090
+    },
+    {
+      "entropy": 0.04650490303756669,
+      "epoch": 14.56,
+      "grad_norm": 0.40625,
+      "learning_rate": 2.0022222222222225e-06,
+      "loss": 0.0271,
+      "mean_token_accuracy": 0.9922069352120161,
+      "num_tokens": 303296013.0,
+      "step": 9100
+    },
+    {
+      "epoch": 14.56,
+      "eval_chemistry_entropy": 0.2310245249271393,
+      "eval_chemistry_loss": 1.404402256011963,
+      "eval_chemistry_mean_token_accuracy": 0.7866413993835449,
+      "eval_chemistry_num_tokens": 303296013.0,
+      "eval_chemistry_runtime": 34.5897,
+      "eval_chemistry_samples_per_second": 14.455,
+      "eval_chemistry_steps_per_second": 3.614,
+      "step": 9100
+    },
+    {
+      "epoch": 14.56,
+      "eval_physics_entropy": 0.1744271154999733,
+      "eval_physics_loss": 1.1907479763031006,
+      "eval_physics_mean_token_accuracy": 0.8237457370758057,
+      "eval_physics_num_tokens": 303296013.0,
+      "eval_physics_runtime": 41.2146,
+      "eval_physics_samples_per_second": 12.132,
+      "eval_physics_steps_per_second": 3.033,
+      "step": 9100
+    },
+    {
+      "entropy": 0.04655418993206695,
+      "epoch": 14.576,
+      "grad_norm": 0.4609375,
+      "learning_rate": 1.98e-06,
+      "loss": 0.0271,
+      "mean_token_accuracy": 0.9922778684645891,
+      "num_tokens": 303627551.0,
+      "step": 9110
+    },
+    {
+      "entropy": 0.04656230836408213,
+      "epoch": 14.592,
+      "grad_norm": 0.423828125,
+      "learning_rate": 1.9577777777777777e-06,
+      "loss": 0.0273,
+      "mean_token_accuracy": 0.9922937303781509,
+      "num_tokens": 303957592.0,
+      "step": 9120
+    },
+    {
+      "entropy": 0.04746072359848767,
+      "epoch": 14.608,
+      "grad_norm": 0.419921875,
+      "learning_rate": 1.9355555555555558e-06,
+      "loss": 0.0274,
+      "mean_token_accuracy": 0.9921062760055065,
+      "num_tokens": 304286763.0,
+      "step": 9130
+    },
+    {
+      "entropy": 0.046136400720570236,
+      "epoch": 14.624,
+      "grad_norm": 0.40625,
+      "learning_rate": 1.9133333333333334e-06,
+      "loss": 0.0268,
+      "mean_token_accuracy": 0.9922884847968817,
+      "num_tokens": 304621198.0,
+      "step": 9140
+    },
+    {
+      "entropy": 0.046801235247403386,
+      "epoch": 14.64,
+      "grad_norm": 0.41015625,
+      "learning_rate": 1.8911111111111114e-06,
+      "loss": 0.0268,
+      "mean_token_accuracy": 0.9921584818512201,
+      "num_tokens": 304956952.0,
+      "step": 9150
+    },
+    {
+      "entropy": 0.047246238612569866,
+      "epoch": 14.656,
+      "grad_norm": 0.435546875,
+      "learning_rate": 1.868888888888889e-06,
+      "loss": 0.027,
+      "mean_token_accuracy": 0.9922087162733078,
+      "num_tokens": 305289637.0,
+      "step": 9160
+    },
+    {
+      "entropy": 0.04670299170538783,
+      "epoch": 14.672,
+      "grad_norm": 0.3984375,
+      "learning_rate": 1.8466666666666668e-06,
+      "loss": 0.027,
+      "mean_token_accuracy": 0.992400411888957,
+      "num_tokens": 305622714.0,
+      "step": 9170
+    },
+    {
+      "entropy": 0.046416766697075215,
+      "epoch": 14.688,
+      "grad_norm": 0.453125,
+      "learning_rate": 1.8244444444444445e-06,
+      "loss": 0.0263,
+      "mean_token_accuracy": 0.9922697741538287,
+      "num_tokens": 305964941.0,
+      "step": 9180
+    },
+    {
+      "entropy": 0.04615663911681622,
+      "epoch": 14.704,
+      "grad_norm": 0.45703125,
+      "learning_rate": 1.8022222222222225e-06,
+      "loss": 0.0268,
+      "mean_token_accuracy": 0.9923350751399994,
+      "num_tokens": 306300994.0,
+      "step": 9190
+    },
+    {
+      "entropy": 0.045685567730106416,
+      "epoch": 14.72,
+      "grad_norm": 0.3984375,
+      "learning_rate": 1.7800000000000001e-06,
+      "loss": 0.0261,
+      "mean_token_accuracy": 0.9925338890403509,
+      "num_tokens": 306645868.0,
+      "step": 9200
+    },
+    {
+      "epoch": 14.72,
+      "eval_chemistry_entropy": 0.230956188082695,
+      "eval_chemistry_loss": 1.4042856693267822,
+      "eval_chemistry_mean_token_accuracy": 0.7865281295776367,
+      "eval_chemistry_num_tokens": 306645868.0,
+      "eval_chemistry_runtime": 34.6033,
+      "eval_chemistry_samples_per_second": 14.449,
+      "eval_chemistry_steps_per_second": 3.612,
+      "step": 9200
+    },
+    {
+      "epoch": 14.72,
+      "eval_physics_entropy": 0.17433010345697403,
+      "eval_physics_loss": 1.190711498260498,
+      "eval_physics_mean_token_accuracy": 0.8236757984161377,
+      "eval_physics_num_tokens": 306645868.0,
+      "eval_physics_runtime": 41.6071,
+      "eval_physics_samples_per_second": 12.017,
+      "eval_physics_steps_per_second": 3.004,
+      "step": 9200
+    },
+    {
+      "entropy": 0.04692138609243557,
+      "epoch": 14.736,
+      "grad_norm": 0.455078125,
+      "learning_rate": 1.757777777777778e-06,
+      "loss": 0.027,
+      "mean_token_accuracy": 0.9920923925936223,
+      "num_tokens": 306977882.0,
+      "step": 9210
+    },
+    {
+      "entropy": 0.046423110109753904,
+      "epoch": 14.752,
+      "grad_norm": 0.419921875,
+      "learning_rate": 1.7355555555555555e-06,
+      "loss": 0.027,
+      "mean_token_accuracy": 0.9922549232840538,
+      "num_tokens": 307311785.0,
+      "step": 9220
+    },
+    {
+      "entropy": 0.04615975025808439,
+      "epoch": 14.768,
+      "grad_norm": 0.427734375,
+      "learning_rate": 1.7133333333333336e-06,
+      "loss": 0.0264,
+      "mean_token_accuracy": 0.9923492431640625,
+      "num_tokens": 307651731.0,
+      "step": 9230
+    },
+    {
+      "entropy": 0.04744623482692987,
+      "epoch": 14.784,
+      "grad_norm": 0.40234375,
+      "learning_rate": 1.6911111111111112e-06,
+      "loss": 0.0273,
+      "mean_token_accuracy": 0.9919729132205248,
+      "num_tokens": 307980649.0,
+      "step": 9240
+    },
+    {
+      "entropy": 0.04643636910477653,
+      "epoch": 14.8,
+      "grad_norm": 0.439453125,
+      "learning_rate": 1.668888888888889e-06,
+      "loss": 0.0269,
+      "mean_token_accuracy": 0.9921998590230942,
+      "num_tokens": 308314533.0,
+      "step": 9250
+    },
+    {
+      "entropy": 0.04688454223796725,
+      "epoch": 14.816,
+      "grad_norm": 0.408203125,
+      "learning_rate": 1.6466666666666666e-06,
+      "loss": 0.0272,
+      "mean_token_accuracy": 0.9922893583774567,
+      "num_tokens": 308646555.0,
+      "step": 9260
+    },
+    {
+      "entropy": 0.046201233624015,
+      "epoch": 14.832,
+      "grad_norm": 0.39453125,
+      "learning_rate": 1.6244444444444447e-06,
+      "loss": 0.027,
+      "mean_token_accuracy": 0.9922460056841373,
+      "num_tokens": 308980075.0,
+      "step": 9270
+    },
+    {
+      "entropy": 0.04652195667149499,
+      "epoch": 14.848,
+      "grad_norm": 0.482421875,
+      "learning_rate": 1.6022222222222223e-06,
+      "loss": 0.0271,
+      "mean_token_accuracy": 0.9921664882451295,
+      "num_tokens": 309311253.0,
+      "step": 9280
+    },
+    {
+      "entropy": 0.04623570635449141,
+      "epoch": 14.864,
+      "grad_norm": 0.38671875,
+      "learning_rate": 1.5800000000000001e-06,
+      "loss": 0.0267,
+      "mean_token_accuracy": 0.9922077853232623,
+      "num_tokens": 309648387.0,
+      "step": 9290
+    },
+    {
+      "entropy": 0.04711627794895321,
+      "epoch": 14.88,
+      "grad_norm": 0.388671875,
+      "learning_rate": 1.5577777777777777e-06,
+      "loss": 0.0269,
+      "mean_token_accuracy": 0.9920925341546536,
+      "num_tokens": 309983882.0,
+      "step": 9300
+    },
+    {
+      "epoch": 14.88,
+      "eval_chemistry_entropy": 0.231436703145504,
+      "eval_chemistry_loss": 1.404557228088379,
+      "eval_chemistry_mean_token_accuracy": 0.7867694778442382,
+      "eval_chemistry_num_tokens": 309983882.0,
+      "eval_chemistry_runtime": 34.5991,
+      "eval_chemistry_samples_per_second": 14.451,
+      "eval_chemistry_steps_per_second": 3.613,
+      "step": 9300
+    },
+    {
+      "epoch": 14.88,
+      "eval_physics_entropy": 0.17477411526441575,
+      "eval_physics_loss": 1.1908060312271118,
+      "eval_physics_mean_token_accuracy": 0.8237089943885804,
+      "eval_physics_num_tokens": 309983882.0,
+      "eval_physics_runtime": 41.2609,
+      "eval_physics_samples_per_second": 12.118,
+      "eval_physics_steps_per_second": 3.03,
+      "step": 9300
+    },
+    {
+      "entropy": 0.04863139950903132,
+      "epoch": 14.896,
+      "grad_norm": 0.390625,
+      "learning_rate": 1.5355555555555558e-06,
+      "loss": 0.0275,
+      "mean_token_accuracy": 0.9918009910732508,
+      "num_tokens": 310310979.0,
+      "step": 9310
+    },
+    {
+      "entropy": 0.046743931679520755,
+      "epoch": 14.912,
+      "grad_norm": 0.37890625,
+      "learning_rate": 1.5133333333333334e-06,
+      "loss": 0.027,
+      "mean_token_accuracy": 0.9921553559601307,
+      "num_tokens": 310643621.0,
+      "step": 9320
+    },
+    {
+      "entropy": 0.046375993068795654,
+      "epoch": 14.928,
+      "grad_norm": 0.396484375,
+      "learning_rate": 1.4911111111111112e-06,
+      "loss": 0.0266,
+      "mean_token_accuracy": 0.9924283813685179,
+      "num_tokens": 310982030.0,
+      "step": 9330
+    },
+    {
+      "entropy": 0.04733075440162793,
+      "epoch": 14.943999999999999,
+      "grad_norm": 0.41015625,
+      "learning_rate": 1.468888888888889e-06,
+      "loss": 0.0271,
+      "mean_token_accuracy": 0.9921328749507665,
+      "num_tokens": 311313876.0,
+      "step": 9340
+    },
+    {
+      "entropy": 0.04763160449219868,
+      "epoch": 14.96,
+      "grad_norm": 0.451171875,
+      "learning_rate": 1.4466666666666669e-06,
+      "loss": 0.0272,
+      "mean_token_accuracy": 0.992208581790328,
+      "num_tokens": 311645941.0,
+      "step": 9350
+    },
+    {
+      "entropy": 0.0466714451322332,
+      "epoch": 14.975999999999999,
+      "grad_norm": 0.435546875,
+      "learning_rate": 1.4244444444444447e-06,
+      "loss": 0.0271,
+      "mean_token_accuracy": 0.9921425424516201,
+      "num_tokens": 311978229.0,
+      "step": 9360
+    },
+    {
+      "entropy": 0.04626302561955527,
+      "epoch": 14.992,
+      "grad_norm": 0.42578125,
+      "learning_rate": 1.4022222222222223e-06,
+      "loss": 0.0267,
+      "mean_token_accuracy": 0.9922904260456562,
+      "num_tokens": 312316959.0,
+      "step": 9370
+    },
+    {
+      "entropy": 0.047080084250774236,
+      "epoch": 15.008,
+      "grad_norm": 0.408203125,
+      "learning_rate": 1.3800000000000001e-06,
+      "loss": 0.0269,
+      "mean_token_accuracy": 0.9922917228192091,
+      "num_tokens": 312652646.0,
+      "step": 9380
+    },
+    {
+      "entropy": 0.04765689440537244,
+      "epoch": 15.024,
+      "grad_norm": 0.421875,
+      "learning_rate": 1.357777777777778e-06,
+      "loss": 0.0277,
+      "mean_token_accuracy": 0.9920204062014818,
+      "num_tokens": 312975628.0,
+      "step": 9390
+    },
+    {
+      "entropy": 0.046637659368570894,
+      "epoch": 15.04,
+      "grad_norm": 0.427734375,
+      "learning_rate": 1.3355555555555558e-06,
+      "loss": 0.0272,
+      "mean_token_accuracy": 0.9923853181302548,
+      "num_tokens": 313307140.0,
+      "step": 9400
+    },
+    {
+      "epoch": 15.04,
+      "eval_chemistry_entropy": 0.23140740382671357,
+      "eval_chemistry_loss": 1.4044877290725708,
+      "eval_chemistry_mean_token_accuracy": 0.7865885219573975,
+      "eval_chemistry_num_tokens": 313307140.0,
+      "eval_chemistry_runtime": 34.5824,
+      "eval_chemistry_samples_per_second": 14.458,
+      "eval_chemistry_steps_per_second": 3.615,
+      "step": 9400
+    },
+    {
+      "epoch": 15.04,
+      "eval_physics_entropy": 0.17470828014612197,
+      "eval_physics_loss": 1.1908527612686157,
+      "eval_physics_mean_token_accuracy": 0.8236756443977356,
+      "eval_physics_num_tokens": 313307140.0,
+      "eval_physics_runtime": 41.22,
+      "eval_physics_samples_per_second": 12.13,
+      "eval_physics_steps_per_second": 3.033,
+      "step": 9400
+    },
+    {
+      "entropy": 0.04628250304376706,
+      "epoch": 15.056,
+      "grad_norm": 0.47265625,
+      "learning_rate": 1.3133333333333334e-06,
+      "loss": 0.0266,
+      "mean_token_accuracy": 0.9923945918679238,
+      "num_tokens": 313643562.0,
+      "step": 9410
+    },
+    {
+      "entropy": 0.04717752797296271,
+      "epoch": 15.072,
+      "grad_norm": 0.43359375,
+      "learning_rate": 1.2911111111111112e-06,
+      "loss": 0.0269,
+      "mean_token_accuracy": 0.992208905518055,
+      "num_tokens": 313978203.0,
+      "step": 9420
+    },
+    {
+      "entropy": 0.04706686328863725,
+      "epoch": 15.088,
+      "grad_norm": 0.416015625,
+      "learning_rate": 1.268888888888889e-06,
+      "loss": 0.0271,
+      "mean_token_accuracy": 0.9921652182936669,
+      "num_tokens": 314309111.0,
+      "step": 9430
+    },
+    {
+      "entropy": 0.04718129952671006,
+      "epoch": 15.104,
+      "grad_norm": 0.439453125,
+      "learning_rate": 1.2466666666666667e-06,
+      "loss": 0.0272,
+      "mean_token_accuracy": 0.992178899422288,
+      "num_tokens": 314640154.0,
+      "step": 9440
+    },
+    {
+      "entropy": 0.04655654202215374,
+      "epoch": 15.12,
+      "grad_norm": 0.39453125,
+      "learning_rate": 1.2244444444444445e-06,
+      "loss": 0.0266,
+      "mean_token_accuracy": 0.9923786576837301,
+      "num_tokens": 314976323.0,
+      "step": 9450
+    },
+    {
+      "entropy": 0.046580646361690015,
+      "epoch": 15.136,
+      "grad_norm": 0.41796875,
+      "learning_rate": 1.2022222222222223e-06,
+      "loss": 0.0267,
+      "mean_token_accuracy": 0.992363540828228,
+      "num_tokens": 315314587.0,
+      "step": 9460
+    },
+    {
+      "entropy": 0.04678839889820665,
+      "epoch": 15.152,
+      "grad_norm": 0.47265625,
+      "learning_rate": 1.1800000000000001e-06,
+      "loss": 0.027,
+      "mean_token_accuracy": 0.9923653259873391,
+      "num_tokens": 315647269.0,
+      "step": 9470
+    },
+    {
+      "entropy": 0.04674161525908858,
+      "epoch": 15.168,
+      "grad_norm": 0.421875,
+      "learning_rate": 1.1577777777777778e-06,
+      "loss": 0.0269,
+      "mean_token_accuracy": 0.9921786695718765,
+      "num_tokens": 315977897.0,
+      "step": 9480
+    },
+    {
+      "entropy": 0.04645551113644615,
+      "epoch": 15.184,
+      "grad_norm": 0.3828125,
+      "learning_rate": 1.1355555555555558e-06,
+      "loss": 0.0266,
+      "mean_token_accuracy": 0.9924337957054377,
+      "num_tokens": 316316256.0,
+      "step": 9490
+    },
+    {
+      "entropy": 0.04722816005814821,
+      "epoch": 15.2,
+      "grad_norm": 0.376953125,
+      "learning_rate": 1.1133333333333334e-06,
+      "loss": 0.0274,
+      "mean_token_accuracy": 0.9920893721282482,
+      "num_tokens": 316644438.0,
+      "step": 9500
+    },
+    {
+      "epoch": 15.2,
+      "eval_chemistry_entropy": 0.23101960027217866,
+      "eval_chemistry_loss": 1.4043958187103271,
+      "eval_chemistry_mean_token_accuracy": 0.7865079116821289,
+      "eval_chemistry_num_tokens": 316644438.0,
+      "eval_chemistry_runtime": 34.5886,
+      "eval_chemistry_samples_per_second": 14.456,
+      "eval_chemistry_steps_per_second": 3.614,
+      "step": 9500
+    },
+    {
+      "epoch": 15.2,
+      "eval_physics_entropy": 0.17439718753099442,
+      "eval_physics_loss": 1.191015362739563,
+      "eval_physics_mean_token_accuracy": 0.8236544723510743,
+      "eval_physics_num_tokens": 316644438.0,
+      "eval_physics_runtime": 40.9521,
+      "eval_physics_samples_per_second": 12.209,
+      "eval_physics_steps_per_second": 3.052,
+      "step": 9500
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 10000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 16,
+  "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": 2.3567728651726307e+19,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}