diff --git "a/hw1/results_span/checkpoint-10855/trainer_state.json" "b/hw1/results_span/checkpoint-10855/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/hw1/results_span/checkpoint-10855/trainer_state.json"
@@ -0,0 +1,7656 @@
+{
+  "best_metric": 1.1905220746994019,
+  "best_model_checkpoint": "./results_span/checkpoint-10855",
+  "epoch": 4.998848722081511,
+  "eval_steps": 500,
+  "global_step": 10855,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": 7.3980183601379395,
+      "learning_rate": 1.998157531091663e-05,
+      "loss": 5.988,
+      "step": 10
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 8.050003051757812,
+      "learning_rate": 1.996315062183326e-05,
+      "loss": 5.2365,
+      "step": 20
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 9.016550064086914,
+      "learning_rate": 1.9944725932749887e-05,
+      "loss": 4.5939,
+      "step": 30
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 13.813095092773438,
+      "learning_rate": 1.9926301243666515e-05,
+      "loss": 4.2055,
+      "step": 40
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 22.46185874938965,
+      "learning_rate": 1.9907876554583144e-05,
+      "loss": 3.6101,
+      "step": 50
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 30.716407775878906,
+      "learning_rate": 1.9889451865499772e-05,
+      "loss": 2.8891,
+      "step": 60
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 25.901784896850586,
+      "learning_rate": 1.98710271764164e-05,
+      "loss": 2.4288,
+      "step": 70
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 25.005104064941406,
+      "learning_rate": 1.985260248733303e-05,
+      "loss": 2.2151,
+      "step": 80
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 30.323572158813477,
+      "learning_rate": 1.9834177798249658e-05,
+      "loss": 1.6689,
+      "step": 90
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 26.94261932373047,
+      "learning_rate": 1.9815753109166286e-05,
+      "loss": 1.8511,
+      "step": 100
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 23.10163116455078,
+      "learning_rate": 1.979732842008291e-05,
+      "loss": 1.4227,
+      "step": 110
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 38.42366027832031,
+      "learning_rate": 1.977890373099954e-05,
+      "loss": 1.5425,
+      "step": 120
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 23.266407012939453,
+      "learning_rate": 1.9760479041916168e-05,
+      "loss": 1.6761,
+      "step": 130
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 27.59915542602539,
+      "learning_rate": 1.9742054352832797e-05,
+      "loss": 1.3461,
+      "step": 140
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 19.75716781616211,
+      "learning_rate": 1.9723629663749425e-05,
+      "loss": 1.39,
+      "step": 150
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 29.772775650024414,
+      "learning_rate": 1.9705204974666054e-05,
+      "loss": 1.2671,
+      "step": 160
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 27.20684051513672,
+      "learning_rate": 1.9686780285582682e-05,
+      "loss": 1.0025,
+      "step": 170
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 32.70050048828125,
+      "learning_rate": 1.966835559649931e-05,
+      "loss": 1.1758,
+      "step": 180
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 14.448714256286621,
+      "learning_rate": 1.964993090741594e-05,
+      "loss": 1.329,
+      "step": 190
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 15.557060241699219,
+      "learning_rate": 1.9631506218332567e-05,
+      "loss": 1.1116,
+      "step": 200
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 22.65195655822754,
+      "learning_rate": 1.9613081529249196e-05,
+      "loss": 0.9291,
+      "step": 210
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 27.135047912597656,
+      "learning_rate": 1.9594656840165824e-05,
+      "loss": 1.0085,
+      "step": 220
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 25.942720413208008,
+      "learning_rate": 1.9576232151082453e-05,
+      "loss": 1.1094,
+      "step": 230
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 74.64520263671875,
+      "learning_rate": 1.955780746199908e-05,
+      "loss": 1.043,
+      "step": 240
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 37.2715950012207,
+      "learning_rate": 1.953938277291571e-05,
+      "loss": 0.9087,
+      "step": 250
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 28.86468505859375,
+      "learning_rate": 1.9520958083832338e-05,
+      "loss": 1.0978,
+      "step": 260
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 37.25987243652344,
+      "learning_rate": 1.9502533394748967e-05,
+      "loss": 1.1423,
+      "step": 270
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 22.999296188354492,
+      "learning_rate": 1.9484108705665595e-05,
+      "loss": 1.0373,
+      "step": 280
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 36.98142623901367,
+      "learning_rate": 1.9465684016582224e-05,
+      "loss": 1.1377,
+      "step": 290
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 29.248615264892578,
+      "learning_rate": 1.9447259327498852e-05,
+      "loss": 0.8616,
+      "step": 300
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 26.504417419433594,
+      "learning_rate": 1.942883463841548e-05,
+      "loss": 1.1013,
+      "step": 310
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 18.09845542907715,
+      "learning_rate": 1.941040994933211e-05,
+      "loss": 1.1981,
+      "step": 320
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 17.694475173950195,
+      "learning_rate": 1.9391985260248734e-05,
+      "loss": 1.038,
+      "step": 330
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 14.167038917541504,
+      "learning_rate": 1.9373560571165363e-05,
+      "loss": 0.7784,
+      "step": 340
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 29.2645263671875,
+      "learning_rate": 1.935513588208199e-05,
+      "loss": 0.7634,
+      "step": 350
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 43.838340759277344,
+      "learning_rate": 1.933671119299862e-05,
+      "loss": 0.7375,
+      "step": 360
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 14.70349407196045,
+      "learning_rate": 1.9318286503915248e-05,
+      "loss": 0.8513,
+      "step": 370
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 37.90299987792969,
+      "learning_rate": 1.9299861814831876e-05,
+      "loss": 1.1504,
+      "step": 380
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 21.050031661987305,
+      "learning_rate": 1.9281437125748505e-05,
+      "loss": 0.9603,
+      "step": 390
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 15.888511657714844,
+      "learning_rate": 1.9263012436665133e-05,
+      "loss": 0.8174,
+      "step": 400
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 20.397958755493164,
+      "learning_rate": 1.9244587747581762e-05,
+      "loss": 0.6461,
+      "step": 410
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 29.643152236938477,
+      "learning_rate": 1.922616305849839e-05,
+      "loss": 0.8521,
+      "step": 420
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 18.462970733642578,
+      "learning_rate": 1.920773836941502e-05,
+      "loss": 0.7945,
+      "step": 430
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 43.42022705078125,
+      "learning_rate": 1.9189313680331644e-05,
+      "loss": 0.8641,
+      "step": 440
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 18.571048736572266,
+      "learning_rate": 1.9170888991248272e-05,
+      "loss": 0.9883,
+      "step": 450
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 34.589630126953125,
+      "learning_rate": 1.91524643021649e-05,
+      "loss": 0.9357,
+      "step": 460
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 15.218515396118164,
+      "learning_rate": 1.913403961308153e-05,
+      "loss": 0.8711,
+      "step": 470
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 28.13692283630371,
+      "learning_rate": 1.9115614923998158e-05,
+      "loss": 0.7938,
+      "step": 480
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 35.69783401489258,
+      "learning_rate": 1.9097190234914786e-05,
+      "loss": 0.8474,
+      "step": 490
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 28.221805572509766,
+      "learning_rate": 1.9078765545831415e-05,
+      "loss": 0.7715,
+      "step": 500
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 17.21616554260254,
+      "learning_rate": 1.9060340856748043e-05,
+      "loss": 0.629,
+      "step": 510
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 24.84185028076172,
+      "learning_rate": 1.904191616766467e-05,
+      "loss": 1.2697,
+      "step": 520
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 24.61993408203125,
+      "learning_rate": 1.90234914785813e-05,
+      "loss": 0.7819,
+      "step": 530
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 17.489444732666016,
+      "learning_rate": 1.900506678949793e-05,
+      "loss": 0.8998,
+      "step": 540
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 30.639545440673828,
+      "learning_rate": 1.8986642100414557e-05,
+      "loss": 0.8763,
+      "step": 550
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 24.281757354736328,
+      "learning_rate": 1.8968217411331185e-05,
+      "loss": 0.7043,
+      "step": 560
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 26.09946060180664,
+      "learning_rate": 1.8949792722247814e-05,
+      "loss": 0.6647,
+      "step": 570
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 22.440410614013672,
+      "learning_rate": 1.8931368033164442e-05,
+      "loss": 0.9207,
+      "step": 580
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 14.211318969726562,
+      "learning_rate": 1.891294334408107e-05,
+      "loss": 0.6342,
+      "step": 590
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 31.617176055908203,
+      "learning_rate": 1.88945186549977e-05,
+      "loss": 0.6375,
+      "step": 600
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 14.567243576049805,
+      "learning_rate": 1.8876093965914328e-05,
+      "loss": 0.7227,
+      "step": 610
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 10.469161987304688,
+      "learning_rate": 1.8857669276830953e-05,
+      "loss": 0.9001,
+      "step": 620
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 22.509355545043945,
+      "learning_rate": 1.883924458774758e-05,
+      "loss": 0.7522,
+      "step": 630
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 40.5599365234375,
+      "learning_rate": 1.882081989866421e-05,
+      "loss": 0.9282,
+      "step": 640
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 20.42763900756836,
+      "learning_rate": 1.8802395209580838e-05,
+      "loss": 0.6526,
+      "step": 650
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 31.25125503540039,
+      "learning_rate": 1.8783970520497467e-05,
+      "loss": 0.6877,
+      "step": 660
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 15.839018821716309,
+      "learning_rate": 1.8765545831414095e-05,
+      "loss": 0.9643,
+      "step": 670
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 25.6279239654541,
+      "learning_rate": 1.8747121142330723e-05,
+      "loss": 0.714,
+      "step": 680
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 34.624786376953125,
+      "learning_rate": 1.8728696453247352e-05,
+      "loss": 0.9777,
+      "step": 690
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 28.802194595336914,
+      "learning_rate": 1.871027176416398e-05,
+      "loss": 0.7333,
+      "step": 700
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 26.19139289855957,
+      "learning_rate": 1.869184707508061e-05,
+      "loss": 0.6572,
+      "step": 710
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 14.55501651763916,
+      "learning_rate": 1.8673422385997237e-05,
+      "loss": 0.8561,
+      "step": 720
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 29.19109535217285,
+      "learning_rate": 1.8654997696913866e-05,
+      "loss": 0.5666,
+      "step": 730
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 27.53074836730957,
+      "learning_rate": 1.8636573007830494e-05,
+      "loss": 0.6896,
+      "step": 740
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 8.353479385375977,
+      "learning_rate": 1.8618148318747123e-05,
+      "loss": 1.1492,
+      "step": 750
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 6.831701278686523,
+      "learning_rate": 1.859972362966375e-05,
+      "loss": 0.7601,
+      "step": 760
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 15.854606628417969,
+      "learning_rate": 1.858129894058038e-05,
+      "loss": 0.8741,
+      "step": 770
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 21.185258865356445,
+      "learning_rate": 1.8562874251497008e-05,
+      "loss": 0.6322,
+      "step": 780
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 5.492204189300537,
+      "learning_rate": 1.8544449562413637e-05,
+      "loss": 0.5673,
+      "step": 790
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 35.550899505615234,
+      "learning_rate": 1.8526024873330265e-05,
+      "loss": 0.9037,
+      "step": 800
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 25.448638916015625,
+      "learning_rate": 1.8507600184246893e-05,
+      "loss": 0.7244,
+      "step": 810
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 25.769161224365234,
+      "learning_rate": 1.8489175495163522e-05,
+      "loss": 0.6508,
+      "step": 820
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 17.432018280029297,
+      "learning_rate": 1.847075080608015e-05,
+      "loss": 0.65,
+      "step": 830
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 2.552358388900757,
+      "learning_rate": 1.8452326116996775e-05,
+      "loss": 0.6029,
+      "step": 840
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 17.973438262939453,
+      "learning_rate": 1.8433901427913404e-05,
+      "loss": 0.6655,
+      "step": 850
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 29.16142463684082,
+      "learning_rate": 1.8415476738830032e-05,
+      "loss": 0.7864,
+      "step": 860
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 11.735014915466309,
+      "learning_rate": 1.839705204974666e-05,
+      "loss": 0.7559,
+      "step": 870
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 19.215181350708008,
+      "learning_rate": 1.837862736066329e-05,
+      "loss": 0.6155,
+      "step": 880
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 43.778419494628906,
+      "learning_rate": 1.8360202671579918e-05,
+      "loss": 0.5346,
+      "step": 890
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 17.800586700439453,
+      "learning_rate": 1.8341777982496546e-05,
+      "loss": 0.6885,
+      "step": 900
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 15.15772819519043,
+      "learning_rate": 1.8323353293413175e-05,
+      "loss": 0.7724,
+      "step": 910
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 15.694869041442871,
+      "learning_rate": 1.8304928604329803e-05,
+      "loss": 0.6258,
+      "step": 920
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 9.422131538391113,
+      "learning_rate": 1.828650391524643e-05,
+      "loss": 0.7137,
+      "step": 930
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 27.350553512573242,
+      "learning_rate": 1.826807922616306e-05,
+      "loss": 1.0469,
+      "step": 940
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 23.706172943115234,
+      "learning_rate": 1.824965453707969e-05,
+      "loss": 0.6155,
+      "step": 950
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 12.49008560180664,
+      "learning_rate": 1.8231229847996317e-05,
+      "loss": 0.6605,
+      "step": 960
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 14.3965482711792,
+      "learning_rate": 1.8212805158912945e-05,
+      "loss": 0.7649,
+      "step": 970
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 12.646517753601074,
+      "learning_rate": 1.8194380469829574e-05,
+      "loss": 0.486,
+      "step": 980
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 18.23296356201172,
+      "learning_rate": 1.8175955780746202e-05,
+      "loss": 0.8072,
+      "step": 990
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 6.323495388031006,
+      "learning_rate": 1.815753109166283e-05,
+      "loss": 0.5596,
+      "step": 1000
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 29.722766876220703,
+      "learning_rate": 1.813910640257946e-05,
+      "loss": 0.7949,
+      "step": 1010
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 28.09901237487793,
+      "learning_rate": 1.8120681713496088e-05,
+      "loss": 0.862,
+      "step": 1020
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 25.29299545288086,
+      "learning_rate": 1.8102257024412716e-05,
+      "loss": 0.7598,
+      "step": 1030
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 11.913007736206055,
+      "learning_rate": 1.8083832335329345e-05,
+      "loss": 0.5738,
+      "step": 1040
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 17.679841995239258,
+      "learning_rate": 1.8065407646245973e-05,
+      "loss": 0.693,
+      "step": 1050
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 16.19197654724121,
+      "learning_rate": 1.8046982957162598e-05,
+      "loss": 0.7047,
+      "step": 1060
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 29.931699752807617,
+      "learning_rate": 1.8028558268079227e-05,
+      "loss": 0.6355,
+      "step": 1070
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 27.28641700744629,
+      "learning_rate": 1.8010133578995855e-05,
+      "loss": 0.6272,
+      "step": 1080
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 16.989234924316406,
+      "learning_rate": 1.7991708889912484e-05,
+      "loss": 0.7087,
+      "step": 1090
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 18.663606643676758,
+      "learning_rate": 1.7973284200829112e-05,
+      "loss": 0.9253,
+      "step": 1100
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 41.59134292602539,
+      "learning_rate": 1.795485951174574e-05,
+      "loss": 0.7659,
+      "step": 1110
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 15.159369468688965,
+      "learning_rate": 1.793643482266237e-05,
+      "loss": 0.6151,
+      "step": 1120
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 44.6723518371582,
+      "learning_rate": 1.7918010133578997e-05,
+      "loss": 0.6237,
+      "step": 1130
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 14.44416332244873,
+      "learning_rate": 1.7899585444495626e-05,
+      "loss": 0.778,
+      "step": 1140
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 5.309757709503174,
+      "learning_rate": 1.7881160755412254e-05,
+      "loss": 0.6173,
+      "step": 1150
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 9.527763366699219,
+      "learning_rate": 1.7862736066328883e-05,
+      "loss": 0.8138,
+      "step": 1160
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 31.333438873291016,
+      "learning_rate": 1.784431137724551e-05,
+      "loss": 0.6644,
+      "step": 1170
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 25.758668899536133,
+      "learning_rate": 1.782588668816214e-05,
+      "loss": 0.7843,
+      "step": 1180
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 25.688190460205078,
+      "learning_rate": 1.7807461999078768e-05,
+      "loss": 0.7243,
+      "step": 1190
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 7.755898952484131,
+      "learning_rate": 1.7789037309995397e-05,
+      "loss": 0.538,
+      "step": 1200
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 26.402544021606445,
+      "learning_rate": 1.7770612620912025e-05,
+      "loss": 0.7287,
+      "step": 1210
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 21.852500915527344,
+      "learning_rate": 1.7752187931828654e-05,
+      "loss": 0.9519,
+      "step": 1220
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 29.12841033935547,
+      "learning_rate": 1.7733763242745282e-05,
+      "loss": 0.727,
+      "step": 1230
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 22.194992065429688,
+      "learning_rate": 1.771533855366191e-05,
+      "loss": 0.7272,
+      "step": 1240
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 17.878643035888672,
+      "learning_rate": 1.769691386457854e-05,
+      "loss": 0.8852,
+      "step": 1250
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 12.078960418701172,
+      "learning_rate": 1.7678489175495167e-05,
+      "loss": 0.6179,
+      "step": 1260
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 31.681859970092773,
+      "learning_rate": 1.7660064486411796e-05,
+      "loss": 0.6725,
+      "step": 1270
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 45.999324798583984,
+      "learning_rate": 1.764163979732842e-05,
+      "loss": 0.5542,
+      "step": 1280
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 22.470754623413086,
+      "learning_rate": 1.762321510824505e-05,
+      "loss": 0.6549,
+      "step": 1290
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 8.734942436218262,
+      "learning_rate": 1.7604790419161678e-05,
+      "loss": 0.5177,
+      "step": 1300
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 8.816222190856934,
+      "learning_rate": 1.7586365730078306e-05,
+      "loss": 0.833,
+      "step": 1310
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 19.542083740234375,
+      "learning_rate": 1.7567941040994935e-05,
+      "loss": 0.5955,
+      "step": 1320
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 15.15994930267334,
+      "learning_rate": 1.7549516351911563e-05,
+      "loss": 0.6391,
+      "step": 1330
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 10.065278053283691,
+      "learning_rate": 1.7531091662828192e-05,
+      "loss": 0.7856,
+      "step": 1340
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 26.093286514282227,
+      "learning_rate": 1.7512666973744817e-05,
+      "loss": 0.6607,
+      "step": 1350
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 22.391468048095703,
+      "learning_rate": 1.7494242284661445e-05,
+      "loss": 0.78,
+      "step": 1360
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 33.6534538269043,
+      "learning_rate": 1.7475817595578074e-05,
+      "loss": 0.8747,
+      "step": 1370
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 39.28668975830078,
+      "learning_rate": 1.7457392906494702e-05,
+      "loss": 0.8342,
+      "step": 1380
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 29.750381469726562,
+      "learning_rate": 1.743896821741133e-05,
+      "loss": 0.7435,
+      "step": 1390
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 7.738365650177002,
+      "learning_rate": 1.742054352832796e-05,
+      "loss": 0.8094,
+      "step": 1400
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 30.06288719177246,
+      "learning_rate": 1.7402118839244588e-05,
+      "loss": 0.543,
+      "step": 1410
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 11.933544158935547,
+      "learning_rate": 1.7383694150161216e-05,
+      "loss": 0.6415,
+      "step": 1420
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 24.376806259155273,
+      "learning_rate": 1.7365269461077845e-05,
+      "loss": 0.9398,
+      "step": 1430
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 26.24981689453125,
+      "learning_rate": 1.7346844771994473e-05,
+      "loss": 0.6427,
+      "step": 1440
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 11.006143569946289,
+      "learning_rate": 1.73284200829111e-05,
+      "loss": 0.5115,
+      "step": 1450
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 12.366308212280273,
+      "learning_rate": 1.730999539382773e-05,
+      "loss": 0.4628,
+      "step": 1460
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 31.493104934692383,
+      "learning_rate": 1.729157070474436e-05,
+      "loss": 0.6848,
+      "step": 1470
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 5.7979559898376465,
+      "learning_rate": 1.7273146015660987e-05,
+      "loss": 0.5398,
+      "step": 1480
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 23.676733016967773,
+      "learning_rate": 1.7254721326577615e-05,
+      "loss": 0.5111,
+      "step": 1490
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 37.613582611083984,
+      "learning_rate": 1.7236296637494244e-05,
+      "loss": 0.7151,
+      "step": 1500
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 44.08135223388672,
+      "learning_rate": 1.7217871948410872e-05,
+      "loss": 0.5468,
+      "step": 1510
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 21.82039451599121,
+      "learning_rate": 1.71994472593275e-05,
+      "loss": 0.4688,
+      "step": 1520
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 23.483362197875977,
+      "learning_rate": 1.718102257024413e-05,
+      "loss": 0.6726,
+      "step": 1530
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 14.748257637023926,
+      "learning_rate": 1.7162597881160758e-05,
+      "loss": 0.4928,
+      "step": 1540
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 13.1818265914917,
+      "learning_rate": 1.7144173192077386e-05,
+      "loss": 0.7384,
+      "step": 1550
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 16.110506057739258,
+      "learning_rate": 1.7125748502994015e-05,
+      "loss": 0.7959,
+      "step": 1560
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 38.41685485839844,
+      "learning_rate": 1.710732381391064e-05,
+      "loss": 0.4775,
+      "step": 1570
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 34.30717086791992,
+      "learning_rate": 1.7088899124827268e-05,
+      "loss": 0.6529,
+      "step": 1580
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 18.27637481689453,
+      "learning_rate": 1.7070474435743897e-05,
+      "loss": 0.5476,
+      "step": 1590
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 35.768043518066406,
+      "learning_rate": 1.7052049746660525e-05,
+      "loss": 0.5432,
+      "step": 1600
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 6.374083042144775,
+      "learning_rate": 1.7033625057577153e-05,
+      "loss": 0.4312,
+      "step": 1610
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 16.26154136657715,
+      "learning_rate": 1.7015200368493782e-05,
+      "loss": 0.6649,
+      "step": 1620
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 19.94081687927246,
+      "learning_rate": 1.699677567941041e-05,
+      "loss": 0.447,
+      "step": 1630
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 24.889982223510742,
+      "learning_rate": 1.697835099032704e-05,
+      "loss": 0.5842,
+      "step": 1640
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 21.78171157836914,
+      "learning_rate": 1.6959926301243667e-05,
+      "loss": 0.5753,
+      "step": 1650
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 10.867707252502441,
+      "learning_rate": 1.6941501612160296e-05,
+      "loss": 0.7014,
+      "step": 1660
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 9.909357070922852,
+      "learning_rate": 1.6923076923076924e-05,
+      "loss": 0.7236,
+      "step": 1670
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 25.049203872680664,
+      "learning_rate": 1.6904652233993553e-05,
+      "loss": 0.7398,
+      "step": 1680
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 15.808841705322266,
+      "learning_rate": 1.688622754491018e-05,
+      "loss": 0.5656,
+      "step": 1690
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 12.406062126159668,
+      "learning_rate": 1.686780285582681e-05,
+      "loss": 0.5076,
+      "step": 1700
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 38.173828125,
+      "learning_rate": 1.6849378166743438e-05,
+      "loss": 0.6941,
+      "step": 1710
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 10.674052238464355,
+      "learning_rate": 1.6830953477660067e-05,
+      "loss": 0.5075,
+      "step": 1720
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 23.41109275817871,
+      "learning_rate": 1.6812528788576695e-05,
+      "loss": 0.7147,
+      "step": 1730
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 41.00368118286133,
+      "learning_rate": 1.6794104099493323e-05,
+      "loss": 0.6755,
+      "step": 1740
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 33.01752853393555,
+      "learning_rate": 1.6775679410409952e-05,
+      "loss": 0.6007,
+      "step": 1750
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 12.027016639709473,
+      "learning_rate": 1.675725472132658e-05,
+      "loss": 0.5876,
+      "step": 1760
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 12.914687156677246,
+      "learning_rate": 1.673883003224321e-05,
+      "loss": 0.7495,
+      "step": 1770
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 8.09559440612793,
+      "learning_rate": 1.6720405343159837e-05,
+      "loss": 0.5843,
+      "step": 1780
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 19.67277717590332,
+      "learning_rate": 1.6701980654076462e-05,
+      "loss": 0.7059,
+      "step": 1790
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 8.93681812286377,
+      "learning_rate": 1.668355596499309e-05,
+      "loss": 0.6573,
+      "step": 1800
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 34.15007781982422,
+      "learning_rate": 1.666513127590972e-05,
+      "loss": 0.6506,
+      "step": 1810
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 36.53522872924805,
+      "learning_rate": 1.6646706586826348e-05,
+      "loss": 0.5368,
+      "step": 1820
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 36.164466857910156,
+      "learning_rate": 1.6628281897742976e-05,
+      "loss": 0.6444,
+      "step": 1830
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 21.803714752197266,
+      "learning_rate": 1.6609857208659605e-05,
+      "loss": 0.6708,
+      "step": 1840
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 35.939292907714844,
+      "learning_rate": 1.6591432519576233e-05,
+      "loss": 0.8718,
+      "step": 1850
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 27.743404388427734,
+      "learning_rate": 1.657300783049286e-05,
+      "loss": 0.4574,
+      "step": 1860
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 9.182666778564453,
+      "learning_rate": 1.655458314140949e-05,
+      "loss": 0.6118,
+      "step": 1870
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 7.767799377441406,
+      "learning_rate": 1.653615845232612e-05,
+      "loss": 0.7109,
+      "step": 1880
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 9.985921859741211,
+      "learning_rate": 1.6517733763242747e-05,
+      "loss": 0.8876,
+      "step": 1890
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 29.76683807373047,
+      "learning_rate": 1.6499309074159375e-05,
+      "loss": 0.5554,
+      "step": 1900
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 37.71890640258789,
+      "learning_rate": 1.6480884385076004e-05,
+      "loss": 0.5083,
+      "step": 1910
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 11.626338958740234,
+      "learning_rate": 1.6462459695992632e-05,
+      "loss": 0.7916,
+      "step": 1920
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 23.077415466308594,
+      "learning_rate": 1.644403500690926e-05,
+      "loss": 0.6777,
+      "step": 1930
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 26.880935668945312,
+      "learning_rate": 1.642561031782589e-05,
+      "loss": 0.8292,
+      "step": 1940
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 9.294807434082031,
+      "learning_rate": 1.6407185628742518e-05,
+      "loss": 0.5059,
+      "step": 1950
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 13.783108711242676,
+      "learning_rate": 1.6388760939659146e-05,
+      "loss": 0.4911,
+      "step": 1960
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 27.606826782226562,
+      "learning_rate": 1.6370336250575775e-05,
+      "loss": 0.5173,
+      "step": 1970
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 13.165992736816406,
+      "learning_rate": 1.6351911561492403e-05,
+      "loss": 0.5361,
+      "step": 1980
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 25.736326217651367,
+      "learning_rate": 1.633348687240903e-05,
+      "loss": 0.4461,
+      "step": 1990
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 26.531147003173828,
+      "learning_rate": 1.631506218332566e-05,
+      "loss": 0.7049,
+      "step": 2000
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 19.434396743774414,
+      "learning_rate": 1.6296637494242285e-05,
+      "loss": 0.5887,
+      "step": 2010
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 16.657024383544922,
+      "learning_rate": 1.6278212805158914e-05,
+      "loss": 0.6068,
+      "step": 2020
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 29.762439727783203,
+      "learning_rate": 1.6259788116075542e-05,
+      "loss": 0.5592,
+      "step": 2030
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 40.036014556884766,
+      "learning_rate": 1.624136342699217e-05,
+      "loss": 0.5974,
+      "step": 2040
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 11.771231651306152,
+      "learning_rate": 1.62229387379088e-05,
+      "loss": 0.4743,
+      "step": 2050
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 27.17993927001953,
+      "learning_rate": 1.6204514048825427e-05,
+      "loss": 0.4261,
+      "step": 2060
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 30.872079849243164,
+      "learning_rate": 1.6186089359742056e-05,
+      "loss": 0.7783,
+      "step": 2070
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 10.339905738830566,
+      "learning_rate": 1.6167664670658684e-05,
+      "loss": 0.6504,
+      "step": 2080
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 10.363712310791016,
+      "learning_rate": 1.6149239981575313e-05,
+      "loss": 0.6144,
+      "step": 2090
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 22.501558303833008,
+      "learning_rate": 1.613081529249194e-05,
+      "loss": 0.6199,
+      "step": 2100
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 13.937215805053711,
+      "learning_rate": 1.611239060340857e-05,
+      "loss": 0.435,
+      "step": 2110
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 20.487646102905273,
+      "learning_rate": 1.6093965914325198e-05,
+      "loss": 0.5326,
+      "step": 2120
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 9.610267639160156,
+      "learning_rate": 1.6075541225241827e-05,
+      "loss": 0.4396,
+      "step": 2130
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 21.75743293762207,
+      "learning_rate": 1.6057116536158455e-05,
+      "loss": 0.5542,
+      "step": 2140
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 18.315757751464844,
+      "learning_rate": 1.6038691847075084e-05,
+      "loss": 0.6032,
+      "step": 2150
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 10.121980667114258,
+      "learning_rate": 1.6020267157991712e-05,
+      "loss": 0.5152,
+      "step": 2160
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 32.00328826904297,
+      "learning_rate": 1.600184246890834e-05,
+      "loss": 0.6403,
+      "step": 2170
+    },
+    {
+      "epoch": 1.0,
+      "eval_loss": 0.5622488856315613,
+      "eval_runtime": 23.6834,
+      "eval_samples_per_second": 127.051,
+      "eval_steps_per_second": 25.419,
+      "step": 2171
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 5.5356974601745605,
+      "learning_rate": 1.5983417779824966e-05,
+      "loss": 0.428,
+      "step": 2180
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 18.359163284301758,
+      "learning_rate": 1.5964993090741594e-05,
+      "loss": 0.2648,
+      "step": 2190
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 13.797196388244629,
+      "learning_rate": 1.5946568401658222e-05,
+      "loss": 0.2308,
+      "step": 2200
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 17.327110290527344,
+      "learning_rate": 1.592814371257485e-05,
+      "loss": 0.449,
+      "step": 2210
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 4.044314861297607,
+      "learning_rate": 1.590971902349148e-05,
+      "loss": 0.3927,
+      "step": 2220
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 18.582021713256836,
+      "learning_rate": 1.5891294334408108e-05,
+      "loss": 0.4003,
+      "step": 2230
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 20.895755767822266,
+      "learning_rate": 1.5872869645324736e-05,
+      "loss": 0.4102,
+      "step": 2240
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 19.490272521972656,
+      "learning_rate": 1.5854444956241365e-05,
+      "loss": 0.2202,
+      "step": 2250
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 17.711881637573242,
+      "learning_rate": 1.5836020267157993e-05,
+      "loss": 0.4689,
+      "step": 2260
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 11.649581909179688,
+      "learning_rate": 1.5817595578074622e-05,
+      "loss": 0.3405,
+      "step": 2270
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 13.362852096557617,
+      "learning_rate": 1.579917088899125e-05,
+      "loss": 0.3441,
+      "step": 2280
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 10.317503929138184,
+      "learning_rate": 1.578074619990788e-05,
+      "loss": 0.3964,
+      "step": 2290
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 20.8016357421875,
+      "learning_rate": 1.5762321510824504e-05,
+      "loss": 0.3718,
+      "step": 2300
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 7.092526912689209,
+      "learning_rate": 1.5743896821741132e-05,
+      "loss": 0.3324,
+      "step": 2310
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 9.475189208984375,
+      "learning_rate": 1.572547213265776e-05,
+      "loss": 0.4358,
+      "step": 2320
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 3.881099224090576,
+      "learning_rate": 1.570704744357439e-05,
+      "loss": 0.3627,
+      "step": 2330
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 5.3999104499816895,
+      "learning_rate": 1.5688622754491018e-05,
+      "loss": 0.4825,
+      "step": 2340
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 12.368169784545898,
+      "learning_rate": 1.5670198065407646e-05,
+      "loss": 0.2853,
+      "step": 2350
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 9.623372077941895,
+      "learning_rate": 1.5651773376324274e-05,
+      "loss": 0.312,
+      "step": 2360
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 25.741796493530273,
+      "learning_rate": 1.5633348687240903e-05,
+      "loss": 0.3367,
+      "step": 2370
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 16.55097007751465,
+      "learning_rate": 1.561492399815753e-05,
+      "loss": 0.2756,
+      "step": 2380
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 3.9125235080718994,
+      "learning_rate": 1.559649930907416e-05,
+      "loss": 0.3351,
+      "step": 2390
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 14.783336639404297,
+      "learning_rate": 1.557807461999079e-05,
+      "loss": 0.3814,
+      "step": 2400
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 13.630842208862305,
+      "learning_rate": 1.5559649930907417e-05,
+      "loss": 0.4304,
+      "step": 2410
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 18.931623458862305,
+      "learning_rate": 1.5541225241824045e-05,
+      "loss": 0.4827,
+      "step": 2420
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 7.653223991394043,
+      "learning_rate": 1.5522800552740674e-05,
+      "loss": 0.2914,
+      "step": 2430
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 30.541412353515625,
+      "learning_rate": 1.5504375863657302e-05,
+      "loss": 0.3019,
+      "step": 2440
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 18.25397491455078,
+      "learning_rate": 1.548595117457393e-05,
+      "loss": 0.4305,
+      "step": 2450
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 2.0572638511657715,
+      "learning_rate": 1.546752648549056e-05,
+      "loss": 0.2248,
+      "step": 2460
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 2.9597907066345215,
+      "learning_rate": 1.5449101796407188e-05,
+      "loss": 0.261,
+      "step": 2470
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 13.447916984558105,
+      "learning_rate": 1.5430677107323816e-05,
+      "loss": 0.3268,
+      "step": 2480
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 36.22604751586914,
+      "learning_rate": 1.5412252418240444e-05,
+      "loss": 0.4511,
+      "step": 2490
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 3.375164270401001,
+      "learning_rate": 1.5393827729157073e-05,
+      "loss": 0.2653,
+      "step": 2500
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 18.604650497436523,
+      "learning_rate": 1.53754030400737e-05,
+      "loss": 0.2925,
+      "step": 2510
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 50.66264343261719,
+      "learning_rate": 1.5356978350990326e-05,
+      "loss": 0.6112,
+      "step": 2520
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 15.180381774902344,
+      "learning_rate": 1.5338553661906955e-05,
+      "loss": 0.4578,
+      "step": 2530
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 7.913496017456055,
+      "learning_rate": 1.5320128972823583e-05,
+      "loss": 0.3743,
+      "step": 2540
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 9.254782676696777,
+      "learning_rate": 1.5301704283740212e-05,
+      "loss": 0.4518,
+      "step": 2550
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 31.521804809570312,
+      "learning_rate": 1.528327959465684e-05,
+      "loss": 0.3756,
+      "step": 2560
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 22.045595169067383,
+      "learning_rate": 1.526485490557347e-05,
+      "loss": 0.5432,
+      "step": 2570
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 12.028554916381836,
+      "learning_rate": 1.5246430216490097e-05,
+      "loss": 0.342,
+      "step": 2580
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 22.071338653564453,
+      "learning_rate": 1.5228005527406726e-05,
+      "loss": 0.3806,
+      "step": 2590
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 13.867335319519043,
+      "learning_rate": 1.5209580838323354e-05,
+      "loss": 0.3113,
+      "step": 2600
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 15.92098617553711,
+      "learning_rate": 1.5191156149239983e-05,
+      "loss": 0.6011,
+      "step": 2610
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 28.975446701049805,
+      "learning_rate": 1.5172731460156611e-05,
+      "loss": 0.4929,
+      "step": 2620
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 7.716545104980469,
+      "learning_rate": 1.515430677107324e-05,
+      "loss": 0.3718,
+      "step": 2630
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 27.41436004638672,
+      "learning_rate": 1.5135882081989868e-05,
+      "loss": 0.335,
+      "step": 2640
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 20.99591827392578,
+      "learning_rate": 1.5117457392906496e-05,
+      "loss": 0.4311,
+      "step": 2650
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 4.289469242095947,
+      "learning_rate": 1.5099032703823125e-05,
+      "loss": 0.4694,
+      "step": 2660
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 9.69908618927002,
+      "learning_rate": 1.5080608014739753e-05,
+      "loss": 0.2392,
+      "step": 2670
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 33.77755355834961,
+      "learning_rate": 1.506218332565638e-05,
+      "loss": 0.4988,
+      "step": 2680
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 25.595121383666992,
+      "learning_rate": 1.5043758636573009e-05,
+      "loss": 0.5265,
+      "step": 2690
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 9.969319343566895,
+      "learning_rate": 1.5025333947489637e-05,
+      "loss": 0.2933,
+      "step": 2700
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 12.514206886291504,
+      "learning_rate": 1.5006909258406266e-05,
+      "loss": 0.1739,
+      "step": 2710
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 12.35896110534668,
+      "learning_rate": 1.4988484569322894e-05,
+      "loss": 0.2632,
+      "step": 2720
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 13.057723999023438,
+      "learning_rate": 1.4970059880239522e-05,
+      "loss": 0.2683,
+      "step": 2730
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 6.722895622253418,
+      "learning_rate": 1.4951635191156151e-05,
+      "loss": 0.441,
+      "step": 2740
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 3.201895236968994,
+      "learning_rate": 1.493321050207278e-05,
+      "loss": 0.3406,
+      "step": 2750
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 6.1700119972229,
+      "learning_rate": 1.4914785812989408e-05,
+      "loss": 0.3289,
+      "step": 2760
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 10.6145658493042,
+      "learning_rate": 1.4896361123906036e-05,
+      "loss": 0.4623,
+      "step": 2770
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 4.5302414894104,
+      "learning_rate": 1.4877936434822665e-05,
+      "loss": 0.2322,
+      "step": 2780
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 13.829955101013184,
+      "learning_rate": 1.4859511745739292e-05,
+      "loss": 0.4583,
+      "step": 2790
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 20.07440948486328,
+      "learning_rate": 1.484108705665592e-05,
+      "loss": 0.4343,
+      "step": 2800
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 16.297746658325195,
+      "learning_rate": 1.4822662367572548e-05,
+      "loss": 0.3804,
+      "step": 2810
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 7.148355007171631,
+      "learning_rate": 1.4804237678489177e-05,
+      "loss": 0.2734,
+      "step": 2820
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 15.609381675720215,
+      "learning_rate": 1.4785812989405805e-05,
+      "loss": 0.3811,
+      "step": 2830
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 26.84148597717285,
+      "learning_rate": 1.4767388300322434e-05,
+      "loss": 0.3585,
+      "step": 2840
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 27.43167495727539,
+      "learning_rate": 1.4748963611239062e-05,
+      "loss": 0.4556,
+      "step": 2850
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 19.240774154663086,
+      "learning_rate": 1.473053892215569e-05,
+      "loss": 0.3848,
+      "step": 2860
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 17.895036697387695,
+      "learning_rate": 1.471211423307232e-05,
+      "loss": 0.2408,
+      "step": 2870
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 12.280651092529297,
+      "learning_rate": 1.4693689543988948e-05,
+      "loss": 0.2191,
+      "step": 2880
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 13.397547721862793,
+      "learning_rate": 1.4675264854905576e-05,
+      "loss": 0.289,
+      "step": 2890
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 46.60029983520508,
+      "learning_rate": 1.4656840165822203e-05,
+      "loss": 0.3065,
+      "step": 2900
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 10.402423858642578,
+      "learning_rate": 1.4638415476738831e-05,
+      "loss": 0.2395,
+      "step": 2910
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 35.154842376708984,
+      "learning_rate": 1.461999078765546e-05,
+      "loss": 0.4574,
+      "step": 2920
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 19.88078498840332,
+      "learning_rate": 1.4601566098572088e-05,
+      "loss": 0.3113,
+      "step": 2930
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 27.43903350830078,
+      "learning_rate": 1.4583141409488717e-05,
+      "loss": 0.2438,
+      "step": 2940
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 23.36299705505371,
+      "learning_rate": 1.4564716720405345e-05,
+      "loss": 0.478,
+      "step": 2950
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 41.50914764404297,
+      "learning_rate": 1.4546292031321974e-05,
+      "loss": 0.375,
+      "step": 2960
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 44.59122848510742,
+      "learning_rate": 1.4527867342238602e-05,
+      "loss": 0.4543,
+      "step": 2970
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 46.586090087890625,
+      "learning_rate": 1.450944265315523e-05,
+      "loss": 0.3459,
+      "step": 2980
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 7.839626312255859,
+      "learning_rate": 1.4491017964071859e-05,
+      "loss": 0.3679,
+      "step": 2990
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 11.148602485656738,
+      "learning_rate": 1.4472593274988488e-05,
+      "loss": 0.4075,
+      "step": 3000
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 11.891318321228027,
+      "learning_rate": 1.4454168585905114e-05,
+      "loss": 0.3311,
+      "step": 3010
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 14.594648361206055,
+      "learning_rate": 1.4435743896821743e-05,
+      "loss": 0.2487,
+      "step": 3020
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 17.1004581451416,
+      "learning_rate": 1.4417319207738371e-05,
+      "loss": 0.4306,
+      "step": 3030
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 25.303600311279297,
+      "learning_rate": 1.4398894518654998e-05,
+      "loss": 0.2019,
+      "step": 3040
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 35.09557342529297,
+      "learning_rate": 1.4380469829571626e-05,
+      "loss": 0.4554,
+      "step": 3050
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 13.950404167175293,
+      "learning_rate": 1.4362045140488255e-05,
+      "loss": 0.3527,
+      "step": 3060
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 2.2372782230377197,
+      "learning_rate": 1.4343620451404883e-05,
+      "loss": 0.2494,
+      "step": 3070
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 48.524356842041016,
+      "learning_rate": 1.432519576232151e-05,
+      "loss": 0.6453,
+      "step": 3080
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 10.225868225097656,
+      "learning_rate": 1.4306771073238139e-05,
+      "loss": 0.3723,
+      "step": 3090
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 6.619189739227295,
+      "learning_rate": 1.4288346384154767e-05,
+      "loss": 0.4052,
+      "step": 3100
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 2.025775671005249,
+      "learning_rate": 1.4269921695071396e-05,
+      "loss": 0.2853,
+      "step": 3110
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 20.987529754638672,
+      "learning_rate": 1.4251497005988024e-05,
+      "loss": 0.2654,
+      "step": 3120
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 20.760652542114258,
+      "learning_rate": 1.4233072316904652e-05,
+      "loss": 0.4554,
+      "step": 3130
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 19.99015998840332,
+      "learning_rate": 1.4214647627821281e-05,
+      "loss": 0.2278,
+      "step": 3140
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 20.269723892211914,
+      "learning_rate": 1.419622293873791e-05,
+      "loss": 0.2913,
+      "step": 3150
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 26.287797927856445,
+      "learning_rate": 1.4177798249654538e-05,
+      "loss": 0.3953,
+      "step": 3160
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 8.766551971435547,
+      "learning_rate": 1.4159373560571166e-05,
+      "loss": 0.3804,
+      "step": 3170
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 12.3949556350708,
+      "learning_rate": 1.4140948871487795e-05,
+      "loss": 0.3197,
+      "step": 3180
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 33.85487747192383,
+      "learning_rate": 1.4122524182404422e-05,
+      "loss": 0.5943,
+      "step": 3190
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 32.4565315246582,
+      "learning_rate": 1.410409949332105e-05,
+      "loss": 0.3646,
+      "step": 3200
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 1.4143786430358887,
+      "learning_rate": 1.4085674804237678e-05,
+      "loss": 0.2804,
+      "step": 3210
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 11.36699390411377,
+      "learning_rate": 1.4067250115154307e-05,
+      "loss": 0.188,
+      "step": 3220
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 27.610795974731445,
+      "learning_rate": 1.4048825426070935e-05,
+      "loss": 0.2977,
+      "step": 3230
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 35.81709671020508,
+      "learning_rate": 1.4030400736987564e-05,
+      "loss": 0.4969,
+      "step": 3240
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 24.171161651611328,
+      "learning_rate": 1.4011976047904192e-05,
+      "loss": 0.3745,
+      "step": 3250
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 12.814797401428223,
+      "learning_rate": 1.399355135882082e-05,
+      "loss": 0.3305,
+      "step": 3260
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 1.07875657081604,
+      "learning_rate": 1.397512666973745e-05,
+      "loss": 0.3106,
+      "step": 3270
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 34.80621337890625,
+      "learning_rate": 1.3956701980654078e-05,
+      "loss": 0.5024,
+      "step": 3280
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 7.723507881164551,
+      "learning_rate": 1.3938277291570706e-05,
+      "loss": 0.3319,
+      "step": 3290
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 4.0220818519592285,
+      "learning_rate": 1.3919852602487333e-05,
+      "loss": 0.4593,
+      "step": 3300
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 34.51366424560547,
+      "learning_rate": 1.3901427913403961e-05,
+      "loss": 0.4468,
+      "step": 3310
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 19.171512603759766,
+      "learning_rate": 1.388300322432059e-05,
+      "loss": 0.3524,
+      "step": 3320
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 19.771678924560547,
+      "learning_rate": 1.3864578535237218e-05,
+      "loss": 0.4495,
+      "step": 3330
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 21.620101928710938,
+      "learning_rate": 1.3846153846153847e-05,
+      "loss": 0.3728,
+      "step": 3340
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 25.341487884521484,
+      "learning_rate": 1.3827729157070475e-05,
+      "loss": 0.3742,
+      "step": 3350
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 18.12294578552246,
+      "learning_rate": 1.3809304467987104e-05,
+      "loss": 0.3013,
+      "step": 3360
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 23.133394241333008,
+      "learning_rate": 1.3790879778903732e-05,
+      "loss": 0.3663,
+      "step": 3370
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 5.638241767883301,
+      "learning_rate": 1.377245508982036e-05,
+      "loss": 0.4776,
+      "step": 3380
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 17.469148635864258,
+      "learning_rate": 1.3754030400736989e-05,
+      "loss": 0.2727,
+      "step": 3390
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 2.431560516357422,
+      "learning_rate": 1.3735605711653618e-05,
+      "loss": 0.2623,
+      "step": 3400
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 8.694042205810547,
+      "learning_rate": 1.3717181022570244e-05,
+      "loss": 0.4844,
+      "step": 3410
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 7.8139753341674805,
+      "learning_rate": 1.3698756333486873e-05,
+      "loss": 0.204,
+      "step": 3420
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 3.7767226696014404,
+      "learning_rate": 1.3680331644403501e-05,
+      "loss": 0.2069,
+      "step": 3430
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 1.8119522333145142,
+      "learning_rate": 1.366190695532013e-05,
+      "loss": 0.4118,
+      "step": 3440
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 11.229525566101074,
+      "learning_rate": 1.3643482266236758e-05,
+      "loss": 0.4325,
+      "step": 3450
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 20.97142791748047,
+      "learning_rate": 1.3625057577153387e-05,
+      "loss": 0.2036,
+      "step": 3460
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 19.750877380371094,
+      "learning_rate": 1.3606632888070015e-05,
+      "loss": 0.3987,
+      "step": 3470
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 16.39549446105957,
+      "learning_rate": 1.3588208198986644e-05,
+      "loss": 0.2813,
+      "step": 3480
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 8.93962574005127,
+      "learning_rate": 1.3569783509903272e-05,
+      "loss": 0.3093,
+      "step": 3490
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 21.79482078552246,
+      "learning_rate": 1.35513588208199e-05,
+      "loss": 0.4513,
+      "step": 3500
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 10.98326587677002,
+      "learning_rate": 1.3532934131736529e-05,
+      "loss": 0.4562,
+      "step": 3510
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 13.73314094543457,
+      "learning_rate": 1.3514509442653156e-05,
+      "loss": 0.2003,
+      "step": 3520
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 6.683830261230469,
+      "learning_rate": 1.3496084753569784e-05,
+      "loss": 0.3162,
+      "step": 3530
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 14.548783302307129,
+      "learning_rate": 1.3477660064486413e-05,
+      "loss": 0.2785,
+      "step": 3540
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 0.2270669937133789,
+      "learning_rate": 1.3459235375403041e-05,
+      "loss": 0.4105,
+      "step": 3550
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 35.95677185058594,
+      "learning_rate": 1.344081068631967e-05,
+      "loss": 0.5008,
+      "step": 3560
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 12.25314712524414,
+      "learning_rate": 1.3422385997236298e-05,
+      "loss": 0.5097,
+      "step": 3570
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 4.997071266174316,
+      "learning_rate": 1.3403961308152926e-05,
+      "loss": 0.2987,
+      "step": 3580
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 4.993253707885742,
+      "learning_rate": 1.3385536619069555e-05,
+      "loss": 0.263,
+      "step": 3590
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 39.53264236450195,
+      "learning_rate": 1.3367111929986183e-05,
+      "loss": 0.4393,
+      "step": 3600
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 1.3203924894332886,
+      "learning_rate": 1.3348687240902812e-05,
+      "loss": 0.3499,
+      "step": 3610
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 38.170654296875,
+      "learning_rate": 1.333026255181944e-05,
+      "loss": 0.3342,
+      "step": 3620
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 3.518298387527466,
+      "learning_rate": 1.3311837862736067e-05,
+      "loss": 0.276,
+      "step": 3630
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 17.09138298034668,
+      "learning_rate": 1.3293413173652696e-05,
+      "loss": 0.4308,
+      "step": 3640
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 14.144304275512695,
+      "learning_rate": 1.3274988484569324e-05,
+      "loss": 0.5266,
+      "step": 3650
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 27.279146194458008,
+      "learning_rate": 1.3256563795485952e-05,
+      "loss": 0.4574,
+      "step": 3660
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 19.015138626098633,
+      "learning_rate": 1.3238139106402581e-05,
+      "loss": 0.3105,
+      "step": 3670
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 5.5302042961120605,
+      "learning_rate": 1.321971441731921e-05,
+      "loss": 0.2391,
+      "step": 3680
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 11.003339767456055,
+      "learning_rate": 1.3201289728235838e-05,
+      "loss": 0.4945,
+      "step": 3690
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 32.00448989868164,
+      "learning_rate": 1.3182865039152466e-05,
+      "loss": 0.3856,
+      "step": 3700
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 13.497784614562988,
+      "learning_rate": 1.3164440350069095e-05,
+      "loss": 0.2155,
+      "step": 3710
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 33.536006927490234,
+      "learning_rate": 1.3146015660985723e-05,
+      "loss": 0.3988,
+      "step": 3720
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 12.460599899291992,
+      "learning_rate": 1.3127590971902352e-05,
+      "loss": 0.1925,
+      "step": 3730
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 14.162660598754883,
+      "learning_rate": 1.3109166282818978e-05,
+      "loss": 0.2625,
+      "step": 3740
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 6.549794673919678,
+      "learning_rate": 1.3090741593735607e-05,
+      "loss": 0.1675,
+      "step": 3750
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 14.990144729614258,
+      "learning_rate": 1.3072316904652235e-05,
+      "loss": 0.3263,
+      "step": 3760
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 33.55649185180664,
+      "learning_rate": 1.3053892215568864e-05,
+      "loss": 0.2603,
+      "step": 3770
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 37.89139938354492,
+      "learning_rate": 1.3035467526485492e-05,
+      "loss": 0.2337,
+      "step": 3780
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 12.947864532470703,
+      "learning_rate": 1.301704283740212e-05,
+      "loss": 0.402,
+      "step": 3790
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 13.592373847961426,
+      "learning_rate": 1.299861814831875e-05,
+      "loss": 0.4879,
+      "step": 3800
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 8.494244575500488,
+      "learning_rate": 1.2980193459235378e-05,
+      "loss": 0.475,
+      "step": 3810
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 12.907691955566406,
+      "learning_rate": 1.2961768770152006e-05,
+      "loss": 0.377,
+      "step": 3820
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 11.92149543762207,
+      "learning_rate": 1.2943344081068635e-05,
+      "loss": 0.4144,
+      "step": 3830
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 33.31961441040039,
+      "learning_rate": 1.2924919391985263e-05,
+      "loss": 0.3986,
+      "step": 3840
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 7.538962364196777,
+      "learning_rate": 1.290649470290189e-05,
+      "loss": 0.2785,
+      "step": 3850
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 15.976167678833008,
+      "learning_rate": 1.2888070013818518e-05,
+      "loss": 0.3334,
+      "step": 3860
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 2.0210375785827637,
+      "learning_rate": 1.2869645324735147e-05,
+      "loss": 0.3353,
+      "step": 3870
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 27.50685691833496,
+      "learning_rate": 1.2851220635651775e-05,
+      "loss": 0.352,
+      "step": 3880
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 32.18745422363281,
+      "learning_rate": 1.2832795946568404e-05,
+      "loss": 0.6015,
+      "step": 3890
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 11.02371883392334,
+      "learning_rate": 1.2814371257485032e-05,
+      "loss": 0.2861,
+      "step": 3900
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 57.82242202758789,
+      "learning_rate": 1.2795946568401659e-05,
+      "loss": 0.368,
+      "step": 3910
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 19.825525283813477,
+      "learning_rate": 1.2777521879318286e-05,
+      "loss": 0.5488,
+      "step": 3920
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 21.79176902770996,
+      "learning_rate": 1.2759097190234914e-05,
+      "loss": 0.282,
+      "step": 3930
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 25.75213050842285,
+      "learning_rate": 1.2740672501151543e-05,
+      "loss": 0.4539,
+      "step": 3940
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 25.91878890991211,
+      "learning_rate": 1.2722247812068171e-05,
+      "loss": 0.3447,
+      "step": 3950
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 2.9834635257720947,
+      "learning_rate": 1.27038231229848e-05,
+      "loss": 0.3003,
+      "step": 3960
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 11.78036880493164,
+      "learning_rate": 1.2685398433901428e-05,
+      "loss": 0.2046,
+      "step": 3970
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 4.300836086273193,
+      "learning_rate": 1.2666973744818056e-05,
+      "loss": 0.4602,
+      "step": 3980
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 9.129778861999512,
+      "learning_rate": 1.2648549055734685e-05,
+      "loss": 0.4708,
+      "step": 3990
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 26.086679458618164,
+      "learning_rate": 1.2630124366651313e-05,
+      "loss": 0.2967,
+      "step": 4000
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 12.554874420166016,
+      "learning_rate": 1.2611699677567942e-05,
+      "loss": 0.409,
+      "step": 4010
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 18.155248641967773,
+      "learning_rate": 1.259327498848457e-05,
+      "loss": 0.5015,
+      "step": 4020
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 20.72637176513672,
+      "learning_rate": 1.2574850299401197e-05,
+      "loss": 0.3255,
+      "step": 4030
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 21.971538543701172,
+      "learning_rate": 1.2556425610317826e-05,
+      "loss": 0.1606,
+      "step": 4040
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 1.8396971225738525,
+      "learning_rate": 1.2538000921234454e-05,
+      "loss": 0.2972,
+      "step": 4050
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 10.291600227355957,
+      "learning_rate": 1.2519576232151082e-05,
+      "loss": 0.2915,
+      "step": 4060
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 31.712594985961914,
+      "learning_rate": 1.2501151543067711e-05,
+      "loss": 0.3489,
+      "step": 4070
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 20.02220344543457,
+      "learning_rate": 1.248272685398434e-05,
+      "loss": 0.4446,
+      "step": 4080
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 8.200404167175293,
+      "learning_rate": 1.2464302164900968e-05,
+      "loss": 0.3657,
+      "step": 4090
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 17.109661102294922,
+      "learning_rate": 1.2445877475817596e-05,
+      "loss": 0.3581,
+      "step": 4100
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 20.908864974975586,
+      "learning_rate": 1.2427452786734225e-05,
+      "loss": 0.3323,
+      "step": 4110
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 4.634097576141357,
+      "learning_rate": 1.2409028097650853e-05,
+      "loss": 0.3671,
+      "step": 4120
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 23.352479934692383,
+      "learning_rate": 1.2390603408567482e-05,
+      "loss": 0.6141,
+      "step": 4130
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 6.355703830718994,
+      "learning_rate": 1.2372178719484108e-05,
+      "loss": 0.2408,
+      "step": 4140
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 8.148225784301758,
+      "learning_rate": 1.2353754030400737e-05,
+      "loss": 0.2823,
+      "step": 4150
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 15.769392967224121,
+      "learning_rate": 1.2335329341317365e-05,
+      "loss": 0.2177,
+      "step": 4160
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 3.9665164947509766,
+      "learning_rate": 1.2316904652233994e-05,
+      "loss": 0.3426,
+      "step": 4170
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 4.84577751159668,
+      "learning_rate": 1.2298479963150622e-05,
+      "loss": 0.3412,
+      "step": 4180
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 10.677656173706055,
+      "learning_rate": 1.228005527406725e-05,
+      "loss": 0.5501,
+      "step": 4190
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 9.393346786499023,
+      "learning_rate": 1.226163058498388e-05,
+      "loss": 0.3983,
+      "step": 4200
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 8.390653610229492,
+      "learning_rate": 1.2243205895900508e-05,
+      "loss": 0.364,
+      "step": 4210
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 7.9062113761901855,
+      "learning_rate": 1.2224781206817136e-05,
+      "loss": 0.2105,
+      "step": 4220
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 15.656996726989746,
+      "learning_rate": 1.2206356517733765e-05,
+      "loss": 0.2599,
+      "step": 4230
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 24.69849967956543,
+      "learning_rate": 1.2187931828650393e-05,
+      "loss": 0.406,
+      "step": 4240
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 21.215103149414062,
+      "learning_rate": 1.216950713956702e-05,
+      "loss": 0.5532,
+      "step": 4250
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 12.31495189666748,
+      "learning_rate": 1.2151082450483648e-05,
+      "loss": 0.4519,
+      "step": 4260
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 13.373255729675293,
+      "learning_rate": 1.2132657761400277e-05,
+      "loss": 0.3162,
+      "step": 4270
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 13.948062896728516,
+      "learning_rate": 1.2114233072316905e-05,
+      "loss": 0.4039,
+      "step": 4280
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 23.39304542541504,
+      "learning_rate": 1.2095808383233534e-05,
+      "loss": 0.3122,
+      "step": 4290
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 9.909829139709473,
+      "learning_rate": 1.2077383694150162e-05,
+      "loss": 0.2963,
+      "step": 4300
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 15.296496391296387,
+      "learning_rate": 1.205895900506679e-05,
+      "loss": 0.3642,
+      "step": 4310
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 18.060197830200195,
+      "learning_rate": 1.2040534315983419e-05,
+      "loss": 0.4413,
+      "step": 4320
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 5.844440460205078,
+      "learning_rate": 1.2022109626900048e-05,
+      "loss": 0.2176,
+      "step": 4330
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 13.600260734558105,
+      "learning_rate": 1.2003684937816676e-05,
+      "loss": 0.2546,
+      "step": 4340
+    },
+    {
+      "epoch": 2.0,
+      "eval_loss": 0.6398792862892151,
+      "eval_runtime": 23.7287,
+      "eval_samples_per_second": 126.809,
+      "eval_steps_per_second": 25.37,
+      "step": 4343
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 7.213687896728516,
+      "learning_rate": 1.1985260248733304e-05,
+      "loss": 0.252,
+      "step": 4350
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 4.554001808166504,
+      "learning_rate": 1.1966835559649931e-05,
+      "loss": 0.0998,
+      "step": 4360
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 20.272781372070312,
+      "learning_rate": 1.194841087056656e-05,
+      "loss": 0.2188,
+      "step": 4370
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 8.232932090759277,
+      "learning_rate": 1.1929986181483188e-05,
+      "loss": 0.1305,
+      "step": 4380
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 33.73552703857422,
+      "learning_rate": 1.1911561492399817e-05,
+      "loss": 0.1914,
+      "step": 4390
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 13.19289779663086,
+      "learning_rate": 1.1893136803316445e-05,
+      "loss": 0.1011,
+      "step": 4400
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 14.15924072265625,
+      "learning_rate": 1.1874712114233074e-05,
+      "loss": 0.1913,
+      "step": 4410
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 16.715362548828125,
+      "learning_rate": 1.1856287425149702e-05,
+      "loss": 0.2317,
+      "step": 4420
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 3.293945550918579,
+      "learning_rate": 1.183786273606633e-05,
+      "loss": 0.1343,
+      "step": 4430
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 1.0425257682800293,
+      "learning_rate": 1.1819438046982959e-05,
+      "loss": 0.1822,
+      "step": 4440
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 5.2661967277526855,
+      "learning_rate": 1.1801013357899587e-05,
+      "loss": 0.3626,
+      "step": 4450
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 39.07841873168945,
+      "learning_rate": 1.1782588668816216e-05,
+      "loss": 0.0918,
+      "step": 4460
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 44.20453643798828,
+      "learning_rate": 1.1764163979732843e-05,
+      "loss": 0.1253,
+      "step": 4470
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 5.607463836669922,
+      "learning_rate": 1.1745739290649471e-05,
+      "loss": 0.1724,
+      "step": 4480
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 16.19831085205078,
+      "learning_rate": 1.17273146015661e-05,
+      "loss": 0.204,
+      "step": 4490
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 20.322694778442383,
+      "learning_rate": 1.1708889912482728e-05,
+      "loss": 0.2602,
+      "step": 4500
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 17.705106735229492,
+      "learning_rate": 1.1690465223399356e-05,
+      "loss": 0.0931,
+      "step": 4510
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 16.315465927124023,
+      "learning_rate": 1.1672040534315985e-05,
+      "loss": 0.2365,
+      "step": 4520
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 16.05750846862793,
+      "learning_rate": 1.1653615845232613e-05,
+      "loss": 0.2414,
+      "step": 4530
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 14.238204956054688,
+      "learning_rate": 1.1635191156149242e-05,
+      "loss": 0.1696,
+      "step": 4540
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 11.063755989074707,
+      "learning_rate": 1.161676646706587e-05,
+      "loss": 0.2321,
+      "step": 4550
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 14.908792495727539,
+      "learning_rate": 1.1598341777982499e-05,
+      "loss": 0.3676,
+      "step": 4560
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 5.8421220779418945,
+      "learning_rate": 1.1579917088899127e-05,
+      "loss": 0.193,
+      "step": 4570
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 13.632174491882324,
+      "learning_rate": 1.1561492399815754e-05,
+      "loss": 0.2137,
+      "step": 4580
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 49.80872344970703,
+      "learning_rate": 1.1543067710732382e-05,
+      "loss": 0.2256,
+      "step": 4590
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 21.89471435546875,
+      "learning_rate": 1.1524643021649011e-05,
+      "loss": 0.1075,
+      "step": 4600
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 4.6218085289001465,
+      "learning_rate": 1.150621833256564e-05,
+      "loss": 0.126,
+      "step": 4610
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 2.799926519393921,
+      "learning_rate": 1.1487793643482268e-05,
+      "loss": 0.1879,
+      "step": 4620
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 8.904982566833496,
+      "learning_rate": 1.1469368954398896e-05,
+      "loss": 0.2626,
+      "step": 4630
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 11.062409400939941,
+      "learning_rate": 1.1450944265315525e-05,
+      "loss": 0.1969,
+      "step": 4640
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 31.971511840820312,
+      "learning_rate": 1.1432519576232153e-05,
+      "loss": 0.1052,
+      "step": 4650
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 12.948674201965332,
+      "learning_rate": 1.1414094887148782e-05,
+      "loss": 0.441,
+      "step": 4660
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 0.9967204928398132,
+      "learning_rate": 1.139567019806541e-05,
+      "loss": 0.0794,
+      "step": 4670
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 2.283189535140991,
+      "learning_rate": 1.1377245508982037e-05,
+      "loss": 0.1084,
+      "step": 4680
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 14.36933708190918,
+      "learning_rate": 1.1358820819898665e-05,
+      "loss": 0.1824,
+      "step": 4690
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 48.689048767089844,
+      "learning_rate": 1.1340396130815294e-05,
+      "loss": 0.2762,
+      "step": 4700
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 13.062122344970703,
+      "learning_rate": 1.1321971441731922e-05,
+      "loss": 0.0942,
+      "step": 4710
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 1.64537513256073,
+      "learning_rate": 1.130354675264855e-05,
+      "loss": 0.1895,
+      "step": 4720
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 28.516090393066406,
+      "learning_rate": 1.128512206356518e-05,
+      "loss": 0.1992,
+      "step": 4730
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 25.513099670410156,
+      "learning_rate": 1.1266697374481808e-05,
+      "loss": 0.1466,
+      "step": 4740
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 5.271493911743164,
+      "learning_rate": 1.1248272685398436e-05,
+      "loss": 0.1687,
+      "step": 4750
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 24.21908950805664,
+      "learning_rate": 1.1229847996315065e-05,
+      "loss": 0.2033,
+      "step": 4760
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 1.7391105890274048,
+      "learning_rate": 1.1211423307231693e-05,
+      "loss": 0.1802,
+      "step": 4770
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 8.517230987548828,
+      "learning_rate": 1.1192998618148318e-05,
+      "loss": 0.1055,
+      "step": 4780
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 11.656946182250977,
+      "learning_rate": 1.1174573929064947e-05,
+      "loss": 0.1321,
+      "step": 4790
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 28.23984718322754,
+      "learning_rate": 1.1156149239981575e-05,
+      "loss": 0.1266,
+      "step": 4800
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 7.646342754364014,
+      "learning_rate": 1.1137724550898203e-05,
+      "loss": 0.1028,
+      "step": 4810
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 17.78961753845215,
+      "learning_rate": 1.1119299861814832e-05,
+      "loss": 0.1591,
+      "step": 4820
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 21.866825103759766,
+      "learning_rate": 1.110087517273146e-05,
+      "loss": 0.2235,
+      "step": 4830
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 0.8799836039543152,
+      "learning_rate": 1.1082450483648089e-05,
+      "loss": 0.2083,
+      "step": 4840
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 68.44153594970703,
+      "learning_rate": 1.1064025794564717e-05,
+      "loss": 0.2201,
+      "step": 4850
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 18.948060989379883,
+      "learning_rate": 1.1045601105481346e-05,
+      "loss": 0.1774,
+      "step": 4860
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 6.963264465332031,
+      "learning_rate": 1.1027176416397973e-05,
+      "loss": 0.1431,
+      "step": 4870
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 11.203207969665527,
+      "learning_rate": 1.1008751727314601e-05,
+      "loss": 0.1496,
+      "step": 4880
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 32.26466751098633,
+      "learning_rate": 1.099032703823123e-05,
+      "loss": 0.2773,
+      "step": 4890
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 9.825488090515137,
+      "learning_rate": 1.0971902349147858e-05,
+      "loss": 0.1353,
+      "step": 4900
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 17.938066482543945,
+      "learning_rate": 1.0953477660064486e-05,
+      "loss": 0.1447,
+      "step": 4910
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 4.524292945861816,
+      "learning_rate": 1.0935052970981115e-05,
+      "loss": 0.217,
+      "step": 4920
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 22.901792526245117,
+      "learning_rate": 1.0916628281897743e-05,
+      "loss": 0.4043,
+      "step": 4930
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 10.1048002243042,
+      "learning_rate": 1.0898203592814372e-05,
+      "loss": 0.1562,
+      "step": 4940
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 0.9836794137954712,
+      "learning_rate": 1.0879778903731e-05,
+      "loss": 0.1543,
+      "step": 4950
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 32.6209716796875,
+      "learning_rate": 1.0861354214647629e-05,
+      "loss": 0.2352,
+      "step": 4960
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 10.768982887268066,
+      "learning_rate": 1.0842929525564257e-05,
+      "loss": 0.1467,
+      "step": 4970
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 19.96872329711914,
+      "learning_rate": 1.0824504836480884e-05,
+      "loss": 0.12,
+      "step": 4980
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 60.0262451171875,
+      "learning_rate": 1.0806080147397512e-05,
+      "loss": 0.1321,
+      "step": 4990
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 2.7073252201080322,
+      "learning_rate": 1.0787655458314141e-05,
+      "loss": 0.2883,
+      "step": 5000
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 1.2513631582260132,
+      "learning_rate": 1.076923076923077e-05,
+      "loss": 0.1217,
+      "step": 5010
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 12.62414836883545,
+      "learning_rate": 1.0750806080147398e-05,
+      "loss": 0.1545,
+      "step": 5020
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 0.22555853426456451,
+      "learning_rate": 1.0732381391064026e-05,
+      "loss": 0.2319,
+      "step": 5030
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 4.9280476570129395,
+      "learning_rate": 1.0713956701980655e-05,
+      "loss": 0.0732,
+      "step": 5040
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 7.7146992683410645,
+      "learning_rate": 1.0695532012897283e-05,
+      "loss": 0.1643,
+      "step": 5050
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 14.150092124938965,
+      "learning_rate": 1.0677107323813912e-05,
+      "loss": 0.369,
+      "step": 5060
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 10.058494567871094,
+      "learning_rate": 1.065868263473054e-05,
+      "loss": 0.3114,
+      "step": 5070
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 11.912763595581055,
+      "learning_rate": 1.0640257945647169e-05,
+      "loss": 0.2257,
+      "step": 5080
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 0.9462236762046814,
+      "learning_rate": 1.0621833256563795e-05,
+      "loss": 0.2581,
+      "step": 5090
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 9.272418022155762,
+      "learning_rate": 1.0603408567480424e-05,
+      "loss": 0.0992,
+      "step": 5100
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 32.22952651977539,
+      "learning_rate": 1.0584983878397052e-05,
+      "loss": 0.1881,
+      "step": 5110
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 24.73603630065918,
+      "learning_rate": 1.056655918931368e-05,
+      "loss": 0.1555,
+      "step": 5120
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 17.692718505859375,
+      "learning_rate": 1.054813450023031e-05,
+      "loss": 0.2112,
+      "step": 5130
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 1.0601367950439453,
+      "learning_rate": 1.0529709811146938e-05,
+      "loss": 0.1373,
+      "step": 5140
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 19.26949119567871,
+      "learning_rate": 1.0511285122063566e-05,
+      "loss": 0.2945,
+      "step": 5150
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 0.31602057814598083,
+      "learning_rate": 1.0492860432980195e-05,
+      "loss": 0.1104,
+      "step": 5160
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 8.280445098876953,
+      "learning_rate": 1.0474435743896823e-05,
+      "loss": 0.1823,
+      "step": 5170
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 5.646317005157471,
+      "learning_rate": 1.0456011054813451e-05,
+      "loss": 0.0914,
+      "step": 5180
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 12.474950790405273,
+      "learning_rate": 1.043758636573008e-05,
+      "loss": 0.1658,
+      "step": 5190
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 30.270292282104492,
+      "learning_rate": 1.0419161676646707e-05,
+      "loss": 0.1898,
+      "step": 5200
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 23.049955368041992,
+      "learning_rate": 1.0400736987563335e-05,
+      "loss": 0.2372,
+      "step": 5210
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 16.111623764038086,
+      "learning_rate": 1.0382312298479964e-05,
+      "loss": 0.175,
+      "step": 5220
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 10.922385215759277,
+      "learning_rate": 1.0363887609396592e-05,
+      "loss": 0.2643,
+      "step": 5230
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 46.063232421875,
+      "learning_rate": 1.034546292031322e-05,
+      "loss": 0.243,
+      "step": 5240
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 13.100521087646484,
+      "learning_rate": 1.0327038231229849e-05,
+      "loss": 0.1361,
+      "step": 5250
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 15.185001373291016,
+      "learning_rate": 1.0308613542146477e-05,
+      "loss": 0.1992,
+      "step": 5260
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 22.965177536010742,
+      "learning_rate": 1.0290188853063106e-05,
+      "loss": 0.2327,
+      "step": 5270
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 20.243452072143555,
+      "learning_rate": 1.0271764163979734e-05,
+      "loss": 0.1212,
+      "step": 5280
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 12.384261131286621,
+      "learning_rate": 1.0253339474896363e-05,
+      "loss": 0.1982,
+      "step": 5290
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 7.480688095092773,
+      "learning_rate": 1.023491478581299e-05,
+      "loss": 0.2122,
+      "step": 5300
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 45.340660095214844,
+      "learning_rate": 1.0216490096729618e-05,
+      "loss": 0.121,
+      "step": 5310
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 13.358603477478027,
+      "learning_rate": 1.0198065407646247e-05,
+      "loss": 0.3049,
+      "step": 5320
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 37.51937484741211,
+      "learning_rate": 1.0179640718562875e-05,
+      "loss": 0.3271,
+      "step": 5330
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 46.40944290161133,
+      "learning_rate": 1.0161216029479503e-05,
+      "loss": 0.2185,
+      "step": 5340
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 23.897319793701172,
+      "learning_rate": 1.0142791340396132e-05,
+      "loss": 0.3567,
+      "step": 5350
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 3.7537150382995605,
+      "learning_rate": 1.012436665131276e-05,
+      "loss": 0.2138,
+      "step": 5360
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 28.463285446166992,
+      "learning_rate": 1.0105941962229389e-05,
+      "loss": 0.3334,
+      "step": 5370
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 14.563559532165527,
+      "learning_rate": 1.0087517273146017e-05,
+      "loss": 0.2001,
+      "step": 5380
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 43.06362533569336,
+      "learning_rate": 1.0069092584062646e-05,
+      "loss": 0.1986,
+      "step": 5390
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 0.45750346779823303,
+      "learning_rate": 1.0050667894979274e-05,
+      "loss": 0.1073,
+      "step": 5400
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 17.50289535522461,
+      "learning_rate": 1.0032243205895901e-05,
+      "loss": 0.4194,
+      "step": 5410
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 11.53402328491211,
+      "learning_rate": 1.001381851681253e-05,
+      "loss": 0.1093,
+      "step": 5420
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 17.4569149017334,
+      "learning_rate": 9.995393827729158e-06,
+      "loss": 0.147,
+      "step": 5430
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 1.9539880752563477,
+      "learning_rate": 9.976969138645786e-06,
+      "loss": 0.11,
+      "step": 5440
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 4.218153953552246,
+      "learning_rate": 9.958544449562413e-06,
+      "loss": 0.2474,
+      "step": 5450
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 41.902076721191406,
+      "learning_rate": 9.940119760479042e-06,
+      "loss": 0.2012,
+      "step": 5460
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 10.567264556884766,
+      "learning_rate": 9.92169507139567e-06,
+      "loss": 0.2148,
+      "step": 5470
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 15.393685340881348,
+      "learning_rate": 9.903270382312299e-06,
+      "loss": 0.4102,
+      "step": 5480
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 6.008276462554932,
+      "learning_rate": 9.884845693228927e-06,
+      "loss": 0.098,
+      "step": 5490
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 2.9895923137664795,
+      "learning_rate": 9.866421004145555e-06,
+      "loss": 0.1946,
+      "step": 5500
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 8.41813850402832,
+      "learning_rate": 9.847996315062184e-06,
+      "loss": 0.1283,
+      "step": 5510
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 17.439990997314453,
+      "learning_rate": 9.829571625978812e-06,
+      "loss": 0.1452,
+      "step": 5520
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 29.21497344970703,
+      "learning_rate": 9.811146936895441e-06,
+      "loss": 0.2062,
+      "step": 5530
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 12.464773178100586,
+      "learning_rate": 9.79272224781207e-06,
+      "loss": 0.1843,
+      "step": 5540
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 15.92199993133545,
+      "learning_rate": 9.774297558728698e-06,
+      "loss": 0.1459,
+      "step": 5550
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 22.772214889526367,
+      "learning_rate": 9.755872869645325e-06,
+      "loss": 0.2318,
+      "step": 5560
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 5.2987775802612305,
+      "learning_rate": 9.737448180561953e-06,
+      "loss": 0.272,
+      "step": 5570
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 5.54741907119751,
+      "learning_rate": 9.719023491478581e-06,
+      "loss": 0.1055,
+      "step": 5580
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 13.03636646270752,
+      "learning_rate": 9.70059880239521e-06,
+      "loss": 0.1955,
+      "step": 5590
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 11.713648796081543,
+      "learning_rate": 9.682174113311838e-06,
+      "loss": 0.1791,
+      "step": 5600
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 19.74137306213379,
+      "learning_rate": 9.663749424228467e-06,
+      "loss": 0.1576,
+      "step": 5610
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 4.425504207611084,
+      "learning_rate": 9.645324735145095e-06,
+      "loss": 0.1447,
+      "step": 5620
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 19.346540451049805,
+      "learning_rate": 9.626900046061724e-06,
+      "loss": 0.2814,
+      "step": 5630
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 0.6840003132820129,
+      "learning_rate": 9.608475356978352e-06,
+      "loss": 0.1342,
+      "step": 5640
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 20.5738582611084,
+      "learning_rate": 9.59005066789498e-06,
+      "loss": 0.2389,
+      "step": 5650
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 5.354895114898682,
+      "learning_rate": 9.57162597881161e-06,
+      "loss": 0.1684,
+      "step": 5660
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 0.07469089329242706,
+      "learning_rate": 9.553201289728236e-06,
+      "loss": 0.1014,
+      "step": 5670
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 16.641284942626953,
+      "learning_rate": 9.534776600644864e-06,
+      "loss": 0.2056,
+      "step": 5680
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 13.647951126098633,
+      "learning_rate": 9.516351911561493e-06,
+      "loss": 0.1436,
+      "step": 5690
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 18.261024475097656,
+      "learning_rate": 9.497927222478121e-06,
+      "loss": 0.253,
+      "step": 5700
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 19.334999084472656,
+      "learning_rate": 9.47950253339475e-06,
+      "loss": 0.2574,
+      "step": 5710
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 5.767329216003418,
+      "learning_rate": 9.461077844311378e-06,
+      "loss": 0.2443,
+      "step": 5720
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 18.11223030090332,
+      "learning_rate": 9.442653155228007e-06,
+      "loss": 0.1587,
+      "step": 5730
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 14.437811851501465,
+      "learning_rate": 9.424228466144635e-06,
+      "loss": 0.0824,
+      "step": 5740
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 35.331626892089844,
+      "learning_rate": 9.405803777061264e-06,
+      "loss": 0.1561,
+      "step": 5750
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 3.8839168548583984,
+      "learning_rate": 9.387379087977892e-06,
+      "loss": 0.1736,
+      "step": 5760
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 8.244375228881836,
+      "learning_rate": 9.36895439889452e-06,
+      "loss": 0.253,
+      "step": 5770
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 36.071163177490234,
+      "learning_rate": 9.350529709811147e-06,
+      "loss": 0.1798,
+      "step": 5780
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 3.5033106803894043,
+      "learning_rate": 9.332105020727776e-06,
+      "loss": 0.1834,
+      "step": 5790
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 21.806026458740234,
+      "learning_rate": 9.313680331644404e-06,
+      "loss": 0.1135,
+      "step": 5800
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 4.923533916473389,
+      "learning_rate": 9.295255642561033e-06,
+      "loss": 0.0966,
+      "step": 5810
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 13.423205375671387,
+      "learning_rate": 9.276830953477661e-06,
+      "loss": 0.1836,
+      "step": 5820
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 8.534943580627441,
+      "learning_rate": 9.25840626439429e-06,
+      "loss": 0.1505,
+      "step": 5830
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 0.9873027205467224,
+      "learning_rate": 9.239981575310918e-06,
+      "loss": 0.1614,
+      "step": 5840
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 19.601699829101562,
+      "learning_rate": 9.221556886227547e-06,
+      "loss": 0.192,
+      "step": 5850
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 0.9434264898300171,
+      "learning_rate": 9.203132197144175e-06,
+      "loss": 0.2691,
+      "step": 5860
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 4.060325622558594,
+      "learning_rate": 9.184707508060802e-06,
+      "loss": 0.142,
+      "step": 5870
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 4.938087463378906,
+      "learning_rate": 9.16628281897743e-06,
+      "loss": 0.1508,
+      "step": 5880
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 8.728765487670898,
+      "learning_rate": 9.147858129894059e-06,
+      "loss": 0.1206,
+      "step": 5890
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 8.366435050964355,
+      "learning_rate": 9.129433440810687e-06,
+      "loss": 0.211,
+      "step": 5900
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 6.929793357849121,
+      "learning_rate": 9.111008751727316e-06,
+      "loss": 0.1725,
+      "step": 5910
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 17.052446365356445,
+      "learning_rate": 9.092584062643942e-06,
+      "loss": 0.3627,
+      "step": 5920
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 6.324765682220459,
+      "learning_rate": 9.07415937356057e-06,
+      "loss": 0.0696,
+      "step": 5930
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 21.453115463256836,
+      "learning_rate": 9.0557346844772e-06,
+      "loss": 0.1068,
+      "step": 5940
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 26.758047103881836,
+      "learning_rate": 9.037309995393828e-06,
+      "loss": 0.1912,
+      "step": 5950
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 19.585784912109375,
+      "learning_rate": 9.018885306310456e-06,
+      "loss": 0.1188,
+      "step": 5960
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 9.23475456237793,
+      "learning_rate": 9.000460617227085e-06,
+      "loss": 0.1877,
+      "step": 5970
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 6.171353816986084,
+      "learning_rate": 8.982035928143713e-06,
+      "loss": 0.1149,
+      "step": 5980
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 39.07819747924805,
+      "learning_rate": 8.963611239060342e-06,
+      "loss": 0.2018,
+      "step": 5990
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 8.198976516723633,
+      "learning_rate": 8.94518654997697e-06,
+      "loss": 0.1268,
+      "step": 6000
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 14.360639572143555,
+      "learning_rate": 8.926761860893599e-06,
+      "loss": 0.2003,
+      "step": 6010
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 25.200519561767578,
+      "learning_rate": 8.908337171810227e-06,
+      "loss": 0.2817,
+      "step": 6020
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 27.658586502075195,
+      "learning_rate": 8.889912482726854e-06,
+      "loss": 0.2225,
+      "step": 6030
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 1.176766276359558,
+      "learning_rate": 8.871487793643482e-06,
+      "loss": 0.1794,
+      "step": 6040
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 21.037872314453125,
+      "learning_rate": 8.85306310456011e-06,
+      "loss": 0.241,
+      "step": 6050
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 9.630943298339844,
+      "learning_rate": 8.834638415476739e-06,
+      "loss": 0.1673,
+      "step": 6060
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 4.158888816833496,
+      "learning_rate": 8.816213726393368e-06,
+      "loss": 0.1396,
+      "step": 6070
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 13.107511520385742,
+      "learning_rate": 8.797789037309996e-06,
+      "loss": 0.1918,
+      "step": 6080
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 0.08003009855747223,
+      "learning_rate": 8.779364348226625e-06,
+      "loss": 0.2224,
+      "step": 6090
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 17.526002883911133,
+      "learning_rate": 8.760939659143253e-06,
+      "loss": 0.2616,
+      "step": 6100
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 12.158259391784668,
+      "learning_rate": 8.742514970059881e-06,
+      "loss": 0.2986,
+      "step": 6110
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 4.206471920013428,
+      "learning_rate": 8.72409028097651e-06,
+      "loss": 0.1147,
+      "step": 6120
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 39.77715301513672,
+      "learning_rate": 8.705665591893138e-06,
+      "loss": 0.3807,
+      "step": 6130
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 9.958640098571777,
+      "learning_rate": 8.687240902809765e-06,
+      "loss": 0.2831,
+      "step": 6140
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 0.9956091046333313,
+      "learning_rate": 8.668816213726394e-06,
+      "loss": 0.2365,
+      "step": 6150
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 2.713994026184082,
+      "learning_rate": 8.650391524643022e-06,
+      "loss": 0.185,
+      "step": 6160
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 21.708200454711914,
+      "learning_rate": 8.63196683555965e-06,
+      "loss": 0.3709,
+      "step": 6170
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 33.87235641479492,
+      "learning_rate": 8.613542146476279e-06,
+      "loss": 0.2614,
+      "step": 6180
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 11.753754615783691,
+      "learning_rate": 8.595117457392907e-06,
+      "loss": 0.1524,
+      "step": 6190
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 6.883747100830078,
+      "learning_rate": 8.576692768309536e-06,
+      "loss": 0.1187,
+      "step": 6200
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 25.431095123291016,
+      "learning_rate": 8.558268079226164e-06,
+      "loss": 0.3465,
+      "step": 6210
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 29.600175857543945,
+      "learning_rate": 8.539843390142793e-06,
+      "loss": 0.2435,
+      "step": 6220
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 0.04582909122109413,
+      "learning_rate": 8.521418701059421e-06,
+      "loss": 0.0884,
+      "step": 6230
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 10.439550399780273,
+      "learning_rate": 8.50299401197605e-06,
+      "loss": 0.1292,
+      "step": 6240
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 22.33614158630371,
+      "learning_rate": 8.484569322892677e-06,
+      "loss": 0.1548,
+      "step": 6250
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 22.231586456298828,
+      "learning_rate": 8.466144633809305e-06,
+      "loss": 0.2064,
+      "step": 6260
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 35.562808990478516,
+      "learning_rate": 8.447719944725933e-06,
+      "loss": 0.3442,
+      "step": 6270
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 19.240463256835938,
+      "learning_rate": 8.429295255642562e-06,
+      "loss": 0.2041,
+      "step": 6280
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 14.270471572875977,
+      "learning_rate": 8.41087056655919e-06,
+      "loss": 0.2951,
+      "step": 6290
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 26.90625,
+      "learning_rate": 8.392445877475817e-06,
+      "loss": 0.2049,
+      "step": 6300
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 23.390901565551758,
+      "learning_rate": 8.374021188392446e-06,
+      "loss": 0.296,
+      "step": 6310
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 15.102962493896484,
+      "learning_rate": 8.355596499309074e-06,
+      "loss": 0.2576,
+      "step": 6320
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 24.74591636657715,
+      "learning_rate": 8.337171810225703e-06,
+      "loss": 0.2296,
+      "step": 6330
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 37.43843078613281,
+      "learning_rate": 8.318747121142331e-06,
+      "loss": 0.1374,
+      "step": 6340
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 7.070268154144287,
+      "learning_rate": 8.30032243205896e-06,
+      "loss": 0.1804,
+      "step": 6350
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 22.491127014160156,
+      "learning_rate": 8.281897742975588e-06,
+      "loss": 0.0973,
+      "step": 6360
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 51.84587478637695,
+      "learning_rate": 8.263473053892216e-06,
+      "loss": 0.1538,
+      "step": 6370
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 4.979628562927246,
+      "learning_rate": 8.245048364808845e-06,
+      "loss": 0.1907,
+      "step": 6380
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 32.33693313598633,
+      "learning_rate": 8.226623675725473e-06,
+      "loss": 0.0928,
+      "step": 6390
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 9.570844650268555,
+      "learning_rate": 8.2081989866421e-06,
+      "loss": 0.1259,
+      "step": 6400
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 15.803118705749512,
+      "learning_rate": 8.189774297558729e-06,
+      "loss": 0.1201,
+      "step": 6410
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 31.298608779907227,
+      "learning_rate": 8.171349608475357e-06,
+      "loss": 0.2208,
+      "step": 6420
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 0.8448058366775513,
+      "learning_rate": 8.152924919391985e-06,
+      "loss": 0.1176,
+      "step": 6430
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 2.9713213443756104,
+      "learning_rate": 8.134500230308614e-06,
+      "loss": 0.2007,
+      "step": 6440
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 31.035558700561523,
+      "learning_rate": 8.116075541225242e-06,
+      "loss": 0.1685,
+      "step": 6450
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 0.1656145453453064,
+      "learning_rate": 8.09765085214187e-06,
+      "loss": 0.1816,
+      "step": 6460
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 3.5192034244537354,
+      "learning_rate": 8.0792261630585e-06,
+      "loss": 0.2072,
+      "step": 6470
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 1.1226110458374023,
+      "learning_rate": 8.060801473975128e-06,
+      "loss": 0.1254,
+      "step": 6480
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 27.589365005493164,
+      "learning_rate": 8.042376784891756e-06,
+      "loss": 0.0926,
+      "step": 6490
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 51.21137619018555,
+      "learning_rate": 8.023952095808385e-06,
+      "loss": 0.1978,
+      "step": 6500
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 19.613243103027344,
+      "learning_rate": 8.005527406725011e-06,
+      "loss": 0.1666,
+      "step": 6510
+    },
+    {
+      "epoch": 3.0,
+      "eval_loss": 0.8739933967590332,
+      "eval_runtime": 23.7281,
+      "eval_samples_per_second": 126.812,
+      "eval_steps_per_second": 25.371,
+      "step": 6514
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 14.523187637329102,
+      "learning_rate": 7.98710271764164e-06,
+      "loss": 0.1438,
+      "step": 6520
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 7.872727870941162,
+      "learning_rate": 7.968678028558268e-06,
+      "loss": 0.0433,
+      "step": 6530
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 17.458667755126953,
+      "learning_rate": 7.950253339474897e-06,
+      "loss": 0.0661,
+      "step": 6540
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 53.7939453125,
+      "learning_rate": 7.931828650391525e-06,
+      "loss": 0.0805,
+      "step": 6550
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 15.321584701538086,
+      "learning_rate": 7.913403961308154e-06,
+      "loss": 0.0927,
+      "step": 6560
+    },
+    {
+      "epoch": 3.03,
+      "grad_norm": 5.969335556030273,
+      "learning_rate": 7.894979272224782e-06,
+      "loss": 0.083,
+      "step": 6570
+    },
+    {
+      "epoch": 3.03,
+      "grad_norm": 50.377628326416016,
+      "learning_rate": 7.87655458314141e-06,
+      "loss": 0.1211,
+      "step": 6580
+    },
+    {
+      "epoch": 3.03,
+      "grad_norm": 4.179170608520508,
+      "learning_rate": 7.858129894058039e-06,
+      "loss": 0.1027,
+      "step": 6590
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 60.53282928466797,
+      "learning_rate": 7.839705204974668e-06,
+      "loss": 0.1632,
+      "step": 6600
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 0.3479783236980438,
+      "learning_rate": 7.821280515891296e-06,
+      "loss": 0.054,
+      "step": 6610
+    },
+    {
+      "epoch": 3.05,
+      "grad_norm": 1.1004326343536377,
+      "learning_rate": 7.802855826807923e-06,
+      "loss": 0.048,
+      "step": 6620
+    },
+    {
+      "epoch": 3.05,
+      "grad_norm": 1.2223320007324219,
+      "learning_rate": 7.784431137724551e-06,
+      "loss": 0.105,
+      "step": 6630
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 0.6669906377792358,
+      "learning_rate": 7.76600644864118e-06,
+      "loss": 0.3239,
+      "step": 6640
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 18.512128829956055,
+      "learning_rate": 7.747581759557808e-06,
+      "loss": 0.0961,
+      "step": 6650
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 0.915468156337738,
+      "learning_rate": 7.729157070474437e-06,
+      "loss": 0.0924,
+      "step": 6660
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 0.14565250277519226,
+      "learning_rate": 7.710732381391065e-06,
+      "loss": 0.0417,
+      "step": 6670
+    },
+    {
+      "epoch": 3.08,
+      "grad_norm": 2.098234176635742,
+      "learning_rate": 7.692307692307694e-06,
+      "loss": 0.0816,
+      "step": 6680
+    },
+    {
+      "epoch": 3.08,
+      "grad_norm": 34.24040985107422,
+      "learning_rate": 7.673883003224322e-06,
+      "loss": 0.1236,
+      "step": 6690
+    },
+    {
+      "epoch": 3.09,
+      "grad_norm": 12.831363677978516,
+      "learning_rate": 7.65545831414095e-06,
+      "loss": 0.1075,
+      "step": 6700
+    },
+    {
+      "epoch": 3.09,
+      "grad_norm": 11.374115943908691,
+      "learning_rate": 7.637033625057579e-06,
+      "loss": 0.0876,
+      "step": 6710
+    },
+    {
+      "epoch": 3.09,
+      "grad_norm": 7.911901473999023,
+      "learning_rate": 7.618608935974207e-06,
+      "loss": 0.0386,
+      "step": 6720
+    },
+    {
+      "epoch": 3.1,
+      "grad_norm": 35.1680793762207,
+      "learning_rate": 7.600184246890835e-06,
+      "loss": 0.067,
+      "step": 6730
+    },
+    {
+      "epoch": 3.1,
+      "grad_norm": 11.210906982421875,
+      "learning_rate": 7.581759557807462e-06,
+      "loss": 0.0747,
+      "step": 6740
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 40.5838737487793,
+      "learning_rate": 7.56333486872409e-06,
+      "loss": 0.0981,
+      "step": 6750
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 89.54202270507812,
+      "learning_rate": 7.544910179640719e-06,
+      "loss": 0.1013,
+      "step": 6760
+    },
+    {
+      "epoch": 3.12,
+      "grad_norm": 20.022287368774414,
+      "learning_rate": 7.526485490557347e-06,
+      "loss": 0.1613,
+      "step": 6770
+    },
+    {
+      "epoch": 3.12,
+      "grad_norm": 2.7612907886505127,
+      "learning_rate": 7.508060801473976e-06,
+      "loss": 0.0872,
+      "step": 6780
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 4.413137435913086,
+      "learning_rate": 7.489636112390603e-06,
+      "loss": 0.1292,
+      "step": 6790
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 12.015055656433105,
+      "learning_rate": 7.471211423307232e-06,
+      "loss": 0.0508,
+      "step": 6800
+    },
+    {
+      "epoch": 3.14,
+      "grad_norm": 0.2437933385372162,
+      "learning_rate": 7.45278673422386e-06,
+      "loss": 0.0501,
+      "step": 6810
+    },
+    {
+      "epoch": 3.14,
+      "grad_norm": 34.62786865234375,
+      "learning_rate": 7.434362045140489e-06,
+      "loss": 0.1001,
+      "step": 6820
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 30.18385124206543,
+      "learning_rate": 7.415937356057117e-06,
+      "loss": 0.047,
+      "step": 6830
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 38.06462097167969,
+      "learning_rate": 7.397512666973745e-06,
+      "loss": 0.0611,
+      "step": 6840
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 8.356391906738281,
+      "learning_rate": 7.379087977890373e-06,
+      "loss": 0.0809,
+      "step": 6850
+    },
+    {
+      "epoch": 3.16,
+      "grad_norm": 33.118499755859375,
+      "learning_rate": 7.360663288807002e-06,
+      "loss": 0.1588,
+      "step": 6860
+    },
+    {
+      "epoch": 3.16,
+      "grad_norm": 24.03145408630371,
+      "learning_rate": 7.34223859972363e-06,
+      "loss": 0.1688,
+      "step": 6870
+    },
+    {
+      "epoch": 3.17,
+      "grad_norm": 8.658808708190918,
+      "learning_rate": 7.323813910640259e-06,
+      "loss": 0.1418,
+      "step": 6880
+    },
+    {
+      "epoch": 3.17,
+      "grad_norm": 11.386858940124512,
+      "learning_rate": 7.305389221556887e-06,
+      "loss": 0.1646,
+      "step": 6890
+    },
+    {
+      "epoch": 3.18,
+      "grad_norm": 0.1153230145573616,
+      "learning_rate": 7.286964532473515e-06,
+      "loss": 0.0259,
+      "step": 6900
+    },
+    {
+      "epoch": 3.18,
+      "grad_norm": 0.0691702663898468,
+      "learning_rate": 7.268539843390143e-06,
+      "loss": 0.0416,
+      "step": 6910
+    },
+    {
+      "epoch": 3.19,
+      "grad_norm": 9.336770057678223,
+      "learning_rate": 7.250115154306772e-06,
+      "loss": 0.0919,
+      "step": 6920
+    },
+    {
+      "epoch": 3.19,
+      "grad_norm": 9.22982120513916,
+      "learning_rate": 7.2316904652234e-06,
+      "loss": 0.1417,
+      "step": 6930
+    },
+    {
+      "epoch": 3.2,
+      "grad_norm": 54.999488830566406,
+      "learning_rate": 7.2132657761400285e-06,
+      "loss": 0.2418,
+      "step": 6940
+    },
+    {
+      "epoch": 3.2,
+      "grad_norm": 4.923679828643799,
+      "learning_rate": 7.194841087056656e-06,
+      "loss": 0.1563,
+      "step": 6950
+    },
+    {
+      "epoch": 3.21,
+      "grad_norm": 1.46045982837677,
+      "learning_rate": 7.176416397973285e-06,
+      "loss": 0.1404,
+      "step": 6960
+    },
+    {
+      "epoch": 3.21,
+      "grad_norm": 13.783041000366211,
+      "learning_rate": 7.157991708889913e-06,
+      "loss": 0.1763,
+      "step": 6970
+    },
+    {
+      "epoch": 3.21,
+      "grad_norm": 30.49824333190918,
+      "learning_rate": 7.1395670198065415e-06,
+      "loss": 0.0685,
+      "step": 6980
+    },
+    {
+      "epoch": 3.22,
+      "grad_norm": 19.967546463012695,
+      "learning_rate": 7.12114233072317e-06,
+      "loss": 0.1665,
+      "step": 6990
+    },
+    {
+      "epoch": 3.22,
+      "grad_norm": 0.7595927119255066,
+      "learning_rate": 7.1027176416397984e-06,
+      "loss": 0.0148,
+      "step": 7000
+    },
+    {
+      "epoch": 3.23,
+      "grad_norm": 5.721044063568115,
+      "learning_rate": 7.084292952556426e-06,
+      "loss": 0.0647,
+      "step": 7010
+    },
+    {
+      "epoch": 3.23,
+      "grad_norm": 1.92232346534729,
+      "learning_rate": 7.0658682634730545e-06,
+      "loss": 0.0776,
+      "step": 7020
+    },
+    {
+      "epoch": 3.24,
+      "grad_norm": 2.41032075881958,
+      "learning_rate": 7.047443574389683e-06,
+      "loss": 0.2073,
+      "step": 7030
+    },
+    {
+      "epoch": 3.24,
+      "grad_norm": 5.554447174072266,
+      "learning_rate": 7.029018885306311e-06,
+      "loss": 0.1498,
+      "step": 7040
+    },
+    {
+      "epoch": 3.25,
+      "grad_norm": 2.6749956607818604,
+      "learning_rate": 7.01059419622294e-06,
+      "loss": 0.1356,
+      "step": 7050
+    },
+    {
+      "epoch": 3.25,
+      "grad_norm": 4.520447254180908,
+      "learning_rate": 6.9921695071395675e-06,
+      "loss": 0.1327,
+      "step": 7060
+    },
+    {
+      "epoch": 3.26,
+      "grad_norm": 19.17978858947754,
+      "learning_rate": 6.973744818056196e-06,
+      "loss": 0.089,
+      "step": 7070
+    },
+    {
+      "epoch": 3.26,
+      "grad_norm": 12.770272254943848,
+      "learning_rate": 6.955320128972824e-06,
+      "loss": 0.1528,
+      "step": 7080
+    },
+    {
+      "epoch": 3.27,
+      "grad_norm": 0.029514513909816742,
+      "learning_rate": 6.936895439889453e-06,
+      "loss": 0.0426,
+      "step": 7090
+    },
+    {
+      "epoch": 3.27,
+      "grad_norm": 22.827943801879883,
+      "learning_rate": 6.918470750806081e-06,
+      "loss": 0.096,
+      "step": 7100
+    },
+    {
+      "epoch": 3.27,
+      "grad_norm": 12.793851852416992,
+      "learning_rate": 6.90004606172271e-06,
+      "loss": 0.1109,
+      "step": 7110
+    },
+    {
+      "epoch": 3.28,
+      "grad_norm": 32.1894645690918,
+      "learning_rate": 6.881621372639337e-06,
+      "loss": 0.037,
+      "step": 7120
+    },
+    {
+      "epoch": 3.28,
+      "grad_norm": 4.462479591369629,
+      "learning_rate": 6.863196683555966e-06,
+      "loss": 0.0522,
+      "step": 7130
+    },
+    {
+      "epoch": 3.29,
+      "grad_norm": 16.41106605529785,
+      "learning_rate": 6.844771994472594e-06,
+      "loss": 0.0193,
+      "step": 7140
+    },
+    {
+      "epoch": 3.29,
+      "grad_norm": 22.32258415222168,
+      "learning_rate": 6.826347305389223e-06,
+      "loss": 0.1378,
+      "step": 7150
+    },
+    {
+      "epoch": 3.3,
+      "grad_norm": 0.635785698890686,
+      "learning_rate": 6.807922616305851e-06,
+      "loss": 0.1394,
+      "step": 7160
+    },
+    {
+      "epoch": 3.3,
+      "grad_norm": 1.9430464506149292,
+      "learning_rate": 6.789497927222478e-06,
+      "loss": 0.0729,
+      "step": 7170
+    },
+    {
+      "epoch": 3.31,
+      "grad_norm": 10.261595726013184,
+      "learning_rate": 6.7710732381391065e-06,
+      "loss": 0.0938,
+      "step": 7180
+    },
+    {
+      "epoch": 3.31,
+      "grad_norm": 3.868656635284424,
+      "learning_rate": 6.752648549055735e-06,
+      "loss": 0.0839,
+      "step": 7190
+    },
+    {
+      "epoch": 3.32,
+      "grad_norm": 0.43338310718536377,
+      "learning_rate": 6.734223859972363e-06,
+      "loss": 0.0231,
+      "step": 7200
+    },
+    {
+      "epoch": 3.32,
+      "grad_norm": 44.59185791015625,
+      "learning_rate": 6.715799170888991e-06,
+      "loss": 0.2106,
+      "step": 7210
+    },
+    {
+      "epoch": 3.32,
+      "grad_norm": 41.37869644165039,
+      "learning_rate": 6.6973744818056195e-06,
+      "loss": 0.1362,
+      "step": 7220
+    },
+    {
+      "epoch": 3.33,
+      "grad_norm": 23.846168518066406,
+      "learning_rate": 6.678949792722248e-06,
+      "loss": 0.1093,
+      "step": 7230
+    },
+    {
+      "epoch": 3.33,
+      "grad_norm": 0.23152123391628265,
+      "learning_rate": 6.660525103638876e-06,
+      "loss": 0.1206,
+      "step": 7240
+    },
+    {
+      "epoch": 3.34,
+      "grad_norm": 0.3266163766384125,
+      "learning_rate": 6.642100414555505e-06,
+      "loss": 0.0782,
+      "step": 7250
+    },
+    {
+      "epoch": 3.34,
+      "grad_norm": 12.91064453125,
+      "learning_rate": 6.6236757254721325e-06,
+      "loss": 0.1967,
+      "step": 7260
+    },
+    {
+      "epoch": 3.35,
+      "grad_norm": 15.349573135375977,
+      "learning_rate": 6.605251036388761e-06,
+      "loss": 0.0567,
+      "step": 7270
+    },
+    {
+      "epoch": 3.35,
+      "grad_norm": 1.3838051557540894,
+      "learning_rate": 6.586826347305389e-06,
+      "loss": 0.0309,
+      "step": 7280
+    },
+    {
+      "epoch": 3.36,
+      "grad_norm": 9.241544723510742,
+      "learning_rate": 6.568401658222018e-06,
+      "loss": 0.1491,
+      "step": 7290
+    },
+    {
+      "epoch": 3.36,
+      "grad_norm": 3.2800493240356445,
+      "learning_rate": 6.549976969138646e-06,
+      "loss": 0.0747,
+      "step": 7300
+    },
+    {
+      "epoch": 3.37,
+      "grad_norm": 24.178730010986328,
+      "learning_rate": 6.531552280055275e-06,
+      "loss": 0.1145,
+      "step": 7310
+    },
+    {
+      "epoch": 3.37,
+      "grad_norm": 2.277247905731201,
+      "learning_rate": 6.513127590971902e-06,
+      "loss": 0.1044,
+      "step": 7320
+    },
+    {
+      "epoch": 3.38,
+      "grad_norm": 26.602750778198242,
+      "learning_rate": 6.494702901888531e-06,
+      "loss": 0.0967,
+      "step": 7330
+    },
+    {
+      "epoch": 3.38,
+      "grad_norm": 23.51183319091797,
+      "learning_rate": 6.476278212805159e-06,
+      "loss": 0.1011,
+      "step": 7340
+    },
+    {
+      "epoch": 3.38,
+      "grad_norm": 2.549064874649048,
+      "learning_rate": 6.457853523721788e-06,
+      "loss": 0.1313,
+      "step": 7350
+    },
+    {
+      "epoch": 3.39,
+      "grad_norm": 26.325971603393555,
+      "learning_rate": 6.439428834638416e-06,
+      "loss": 0.1205,
+      "step": 7360
+    },
+    {
+      "epoch": 3.39,
+      "grad_norm": 24.40768051147461,
+      "learning_rate": 6.421004145555044e-06,
+      "loss": 0.0285,
+      "step": 7370
+    },
+    {
+      "epoch": 3.4,
+      "grad_norm": 31.17082405090332,
+      "learning_rate": 6.402579456471672e-06,
+      "loss": 0.1665,
+      "step": 7380
+    },
+    {
+      "epoch": 3.4,
+      "grad_norm": 4.351948261260986,
+      "learning_rate": 6.384154767388301e-06,
+      "loss": 0.1117,
+      "step": 7390
+    },
+    {
+      "epoch": 3.41,
+      "grad_norm": 14.400015830993652,
+      "learning_rate": 6.365730078304929e-06,
+      "loss": 0.1177,
+      "step": 7400
+    },
+    {
+      "epoch": 3.41,
+      "grad_norm": 0.44112566113471985,
+      "learning_rate": 6.347305389221558e-06,
+      "loss": 0.0811,
+      "step": 7410
+    },
+    {
+      "epoch": 3.42,
+      "grad_norm": 1.3765614032745361,
+      "learning_rate": 6.328880700138186e-06,
+      "loss": 0.0798,
+      "step": 7420
+    },
+    {
+      "epoch": 3.42,
+      "grad_norm": 20.30915069580078,
+      "learning_rate": 6.310456011054814e-06,
+      "loss": 0.1183,
+      "step": 7430
+    },
+    {
+      "epoch": 3.43,
+      "grad_norm": 13.63874626159668,
+      "learning_rate": 6.292031321971442e-06,
+      "loss": 0.1076,
+      "step": 7440
+    },
+    {
+      "epoch": 3.43,
+      "grad_norm": 5.400328636169434,
+      "learning_rate": 6.273606632888071e-06,
+      "loss": 0.0501,
+      "step": 7450
+    },
+    {
+      "epoch": 3.44,
+      "grad_norm": 1.0178033113479614,
+      "learning_rate": 6.255181943804699e-06,
+      "loss": 0.1069,
+      "step": 7460
+    },
+    {
+      "epoch": 3.44,
+      "grad_norm": 1.193809151649475,
+      "learning_rate": 6.236757254721328e-06,
+      "loss": 0.0553,
+      "step": 7470
+    },
+    {
+      "epoch": 3.44,
+      "grad_norm": 9.88496208190918,
+      "learning_rate": 6.218332565637955e-06,
+      "loss": 0.1172,
+      "step": 7480
+    },
+    {
+      "epoch": 3.45,
+      "grad_norm": 0.05173638463020325,
+      "learning_rate": 6.199907876554584e-06,
+      "loss": 0.1362,
+      "step": 7490
+    },
+    {
+      "epoch": 3.45,
+      "grad_norm": 8.535846710205078,
+      "learning_rate": 6.181483187471212e-06,
+      "loss": 0.0837,
+      "step": 7500
+    },
+    {
+      "epoch": 3.46,
+      "grad_norm": 7.838191032409668,
+      "learning_rate": 6.163058498387841e-06,
+      "loss": 0.0281,
+      "step": 7510
+    },
+    {
+      "epoch": 3.46,
+      "grad_norm": 32.58201599121094,
+      "learning_rate": 6.144633809304469e-06,
+      "loss": 0.0892,
+      "step": 7520
+    },
+    {
+      "epoch": 3.47,
+      "grad_norm": 12.18652057647705,
+      "learning_rate": 6.1262091202210976e-06,
+      "loss": 0.1055,
+      "step": 7530
+    },
+    {
+      "epoch": 3.47,
+      "grad_norm": 3.2115883827209473,
+      "learning_rate": 6.107784431137725e-06,
+      "loss": 0.0221,
+      "step": 7540
+    },
+    {
+      "epoch": 3.48,
+      "grad_norm": 3.6199662685394287,
+      "learning_rate": 6.089359742054354e-06,
+      "loss": 0.0957,
+      "step": 7550
+    },
+    {
+      "epoch": 3.48,
+      "grad_norm": 14.36513900756836,
+      "learning_rate": 6.070935052970982e-06,
+      "loss": 0.1835,
+      "step": 7560
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 35.61921310424805,
+      "learning_rate": 6.0525103638876106e-06,
+      "loss": 0.1147,
+      "step": 7570
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 8.08601188659668,
+      "learning_rate": 6.034085674804239e-06,
+      "loss": 0.1403,
+      "step": 7580
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 0.04037042334675789,
+      "learning_rate": 6.015660985720867e-06,
+      "loss": 0.0913,
+      "step": 7590
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 15.60925006866455,
+      "learning_rate": 5.997236296637494e-06,
+      "loss": 0.115,
+      "step": 7600
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 10.180851936340332,
+      "learning_rate": 5.978811607554123e-06,
+      "loss": 0.1069,
+      "step": 7610
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 11.0906400680542,
+      "learning_rate": 5.960386918470751e-06,
+      "loss": 0.1307,
+      "step": 7620
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 1.723253607749939,
+      "learning_rate": 5.941962229387379e-06,
+      "loss": 0.0649,
+      "step": 7630
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 29.21575355529785,
+      "learning_rate": 5.923537540304007e-06,
+      "loss": 0.0704,
+      "step": 7640
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 18.723695755004883,
+      "learning_rate": 5.905112851220636e-06,
+      "loss": 0.0318,
+      "step": 7650
+    },
+    {
+      "epoch": 3.53,
+      "grad_norm": 0.029926879331469536,
+      "learning_rate": 5.886688162137264e-06,
+      "loss": 0.1639,
+      "step": 7660
+    },
+    {
+      "epoch": 3.53,
+      "grad_norm": 38.5188102722168,
+      "learning_rate": 5.868263473053893e-06,
+      "loss": 0.0768,
+      "step": 7670
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 2.693557024002075,
+      "learning_rate": 5.84983878397052e-06,
+      "loss": 0.1178,
+      "step": 7680
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 4.7778520584106445,
+      "learning_rate": 5.831414094887149e-06,
+      "loss": 0.0673,
+      "step": 7690
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 0.5049407482147217,
+      "learning_rate": 5.812989405803777e-06,
+      "loss": 0.0676,
+      "step": 7700
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 2.3128838539123535,
+      "learning_rate": 5.794564716720406e-06,
+      "loss": 0.0563,
+      "step": 7710
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 0.5181612372398376,
+      "learning_rate": 5.776140027637034e-06,
+      "loss": 0.2807,
+      "step": 7720
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 36.27579879760742,
+      "learning_rate": 5.7577153385536626e-06,
+      "loss": 0.1216,
+      "step": 7730
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 1.0602400302886963,
+      "learning_rate": 5.73929064947029e-06,
+      "loss": 0.1645,
+      "step": 7740
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 12.10837459564209,
+      "learning_rate": 5.720865960386919e-06,
+      "loss": 0.0527,
+      "step": 7750
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 27.241737365722656,
+      "learning_rate": 5.702441271303547e-06,
+      "loss": 0.1339,
+      "step": 7760
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 3.2355740070343018,
+      "learning_rate": 5.6840165822201756e-06,
+      "loss": 0.0544,
+      "step": 7770
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 18.80722999572754,
+      "learning_rate": 5.665591893136804e-06,
+      "loss": 0.0666,
+      "step": 7780
+    },
+    {
+      "epoch": 3.59,
+      "grad_norm": 6.459070682525635,
+      "learning_rate": 5.647167204053432e-06,
+      "loss": 0.1226,
+      "step": 7790
+    },
+    {
+      "epoch": 3.59,
+      "grad_norm": 11.722634315490723,
+      "learning_rate": 5.62874251497006e-06,
+      "loss": 0.2263,
+      "step": 7800
+    },
+    {
+      "epoch": 3.6,
+      "grad_norm": 0.41630393266677856,
+      "learning_rate": 5.6103178258866886e-06,
+      "loss": 0.06,
+      "step": 7810
+    },
+    {
+      "epoch": 3.6,
+      "grad_norm": 19.83670997619629,
+      "learning_rate": 5.591893136803317e-06,
+      "loss": 0.1468,
+      "step": 7820
+    },
+    {
+      "epoch": 3.61,
+      "grad_norm": 4.748047828674316,
+      "learning_rate": 5.5734684477199455e-06,
+      "loss": 0.0682,
+      "step": 7830
+    },
+    {
+      "epoch": 3.61,
+      "grad_norm": 25.43903350830078,
+      "learning_rate": 5.555043758636574e-06,
+      "loss": 0.0673,
+      "step": 7840
+    },
+    {
+      "epoch": 3.62,
+      "grad_norm": 0.9281651377677917,
+      "learning_rate": 5.5366190695532016e-06,
+      "loss": 0.1553,
+      "step": 7850
+    },
+    {
+      "epoch": 3.62,
+      "grad_norm": 0.2185637652873993,
+      "learning_rate": 5.51819438046983e-06,
+      "loss": 0.0318,
+      "step": 7860
+    },
+    {
+      "epoch": 3.62,
+      "grad_norm": 1.79883873462677,
+      "learning_rate": 5.4997696913864585e-06,
+      "loss": 0.1946,
+      "step": 7870
+    },
+    {
+      "epoch": 3.63,
+      "grad_norm": 12.928549766540527,
+      "learning_rate": 5.481345002303087e-06,
+      "loss": 0.1787,
+      "step": 7880
+    },
+    {
+      "epoch": 3.63,
+      "grad_norm": 11.246813774108887,
+      "learning_rate": 5.462920313219715e-06,
+      "loss": 0.1512,
+      "step": 7890
+    },
+    {
+      "epoch": 3.64,
+      "grad_norm": 1.1604872941970825,
+      "learning_rate": 5.444495624136343e-06,
+      "loss": 0.1015,
+      "step": 7900
+    },
+    {
+      "epoch": 3.64,
+      "grad_norm": 3.489548921585083,
+      "learning_rate": 5.4260709350529715e-06,
+      "loss": 0.1165,
+      "step": 7910
+    },
+    {
+      "epoch": 3.65,
+      "grad_norm": 5.3134660720825195,
+      "learning_rate": 5.4076462459696e-06,
+      "loss": 0.0948,
+      "step": 7920
+    },
+    {
+      "epoch": 3.65,
+      "grad_norm": 4.447476863861084,
+      "learning_rate": 5.389221556886228e-06,
+      "loss": 0.1302,
+      "step": 7930
+    },
+    {
+      "epoch": 3.66,
+      "grad_norm": 0.5889253616333008,
+      "learning_rate": 5.370796867802857e-06,
+      "loss": 0.0508,
+      "step": 7940
+    },
+    {
+      "epoch": 3.66,
+      "grad_norm": 21.49410057067871,
+      "learning_rate": 5.3523721787194845e-06,
+      "loss": 0.2965,
+      "step": 7950
+    },
+    {
+      "epoch": 3.67,
+      "grad_norm": 3.1076550483703613,
+      "learning_rate": 5.333947489636113e-06,
+      "loss": 0.1659,
+      "step": 7960
+    },
+    {
+      "epoch": 3.67,
+      "grad_norm": 2.327097177505493,
+      "learning_rate": 5.315522800552741e-06,
+      "loss": 0.1339,
+      "step": 7970
+    },
+    {
+      "epoch": 3.67,
+      "grad_norm": 35.96466827392578,
+      "learning_rate": 5.29709811146937e-06,
+      "loss": 0.1133,
+      "step": 7980
+    },
+    {
+      "epoch": 3.68,
+      "grad_norm": 5.595996379852295,
+      "learning_rate": 5.278673422385998e-06,
+      "loss": 0.0986,
+      "step": 7990
+    },
+    {
+      "epoch": 3.68,
+      "grad_norm": 44.037532806396484,
+      "learning_rate": 5.260248733302627e-06,
+      "loss": 0.229,
+      "step": 8000
+    },
+    {
+      "epoch": 3.69,
+      "grad_norm": 54.138702392578125,
+      "learning_rate": 5.241824044219254e-06,
+      "loss": 0.1126,
+      "step": 8010
+    },
+    {
+      "epoch": 3.69,
+      "grad_norm": 1.4185423851013184,
+      "learning_rate": 5.223399355135883e-06,
+      "loss": 0.0984,
+      "step": 8020
+    },
+    {
+      "epoch": 3.7,
+      "grad_norm": 8.82161808013916,
+      "learning_rate": 5.204974666052511e-06,
+      "loss": 0.0318,
+      "step": 8030
+    },
+    {
+      "epoch": 3.7,
+      "grad_norm": 0.6462627649307251,
+      "learning_rate": 5.186549976969139e-06,
+      "loss": 0.0423,
+      "step": 8040
+    },
+    {
+      "epoch": 3.71,
+      "grad_norm": 1.0346858501434326,
+      "learning_rate": 5.1681252878857665e-06,
+      "loss": 0.0062,
+      "step": 8050
+    },
+    {
+      "epoch": 3.71,
+      "grad_norm": 40.59715270996094,
+      "learning_rate": 5.149700598802395e-06,
+      "loss": 0.1837,
+      "step": 8060
+    },
+    {
+      "epoch": 3.72,
+      "grad_norm": 8.906074523925781,
+      "learning_rate": 5.1312759097190235e-06,
+      "loss": 0.107,
+      "step": 8070
+    },
+    {
+      "epoch": 3.72,
+      "grad_norm": 4.6563310623168945,
+      "learning_rate": 5.112851220635652e-06,
+      "loss": 0.0679,
+      "step": 8080
+    },
+    {
+      "epoch": 3.73,
+      "grad_norm": 0.19132518768310547,
+      "learning_rate": 5.09442653155228e-06,
+      "loss": 0.0699,
+      "step": 8090
+    },
+    {
+      "epoch": 3.73,
+      "grad_norm": 15.865428924560547,
+      "learning_rate": 5.076001842468908e-06,
+      "loss": 0.1192,
+      "step": 8100
+    },
+    {
+      "epoch": 3.73,
+      "grad_norm": 2.1365442276000977,
+      "learning_rate": 5.0575771533855365e-06,
+      "loss": 0.0613,
+      "step": 8110
+    },
+    {
+      "epoch": 3.74,
+      "grad_norm": 0.031343813985586166,
+      "learning_rate": 5.039152464302165e-06,
+      "loss": 0.0621,
+      "step": 8120
+    },
+    {
+      "epoch": 3.74,
+      "grad_norm": 6.723030090332031,
+      "learning_rate": 5.020727775218793e-06,
+      "loss": 0.069,
+      "step": 8130
+    },
+    {
+      "epoch": 3.75,
+      "grad_norm": 18.560302734375,
+      "learning_rate": 5.002303086135422e-06,
+      "loss": 0.0563,
+      "step": 8140
+    },
+    {
+      "epoch": 3.75,
+      "grad_norm": 52.29302978515625,
+      "learning_rate": 4.98387839705205e-06,
+      "loss": 0.0968,
+      "step": 8150
+    },
+    {
+      "epoch": 3.76,
+      "grad_norm": 0.18237358331680298,
+      "learning_rate": 4.965453707968678e-06,
+      "loss": 0.1005,
+      "step": 8160
+    },
+    {
+      "epoch": 3.76,
+      "grad_norm": 36.439632415771484,
+      "learning_rate": 4.947029018885306e-06,
+      "loss": 0.2414,
+      "step": 8170
+    },
+    {
+      "epoch": 3.77,
+      "grad_norm": 8.807538986206055,
+      "learning_rate": 4.928604329801935e-06,
+      "loss": 0.112,
+      "step": 8180
+    },
+    {
+      "epoch": 3.77,
+      "grad_norm": 0.596357524394989,
+      "learning_rate": 4.910179640718563e-06,
+      "loss": 0.1056,
+      "step": 8190
+    },
+    {
+      "epoch": 3.78,
+      "grad_norm": 6.536358833312988,
+      "learning_rate": 4.891754951635192e-06,
+      "loss": 0.188,
+      "step": 8200
+    },
+    {
+      "epoch": 3.78,
+      "grad_norm": 1.2331651449203491,
+      "learning_rate": 4.873330262551819e-06,
+      "loss": 0.0398,
+      "step": 8210
+    },
+    {
+      "epoch": 3.79,
+      "grad_norm": 4.492482662200928,
+      "learning_rate": 4.854905573468448e-06,
+      "loss": 0.1442,
+      "step": 8220
+    },
+    {
+      "epoch": 3.79,
+      "grad_norm": 0.9302757978439331,
+      "learning_rate": 4.836480884385076e-06,
+      "loss": 0.0938,
+      "step": 8230
+    },
+    {
+      "epoch": 3.79,
+      "grad_norm": 21.692852020263672,
+      "learning_rate": 4.818056195301705e-06,
+      "loss": 0.1236,
+      "step": 8240
+    },
+    {
+      "epoch": 3.8,
+      "grad_norm": 8.847552299499512,
+      "learning_rate": 4.799631506218333e-06,
+      "loss": 0.121,
+      "step": 8250
+    },
+    {
+      "epoch": 3.8,
+      "grad_norm": 1.9204087257385254,
+      "learning_rate": 4.781206817134961e-06,
+      "loss": 0.1062,
+      "step": 8260
+    },
+    {
+      "epoch": 3.81,
+      "grad_norm": 49.04825210571289,
+      "learning_rate": 4.762782128051589e-06,
+      "loss": 0.063,
+      "step": 8270
+    },
+    {
+      "epoch": 3.81,
+      "grad_norm": 0.8194244503974915,
+      "learning_rate": 4.744357438968218e-06,
+      "loss": 0.0443,
+      "step": 8280
+    },
+    {
+      "epoch": 3.82,
+      "grad_norm": 52.35334396362305,
+      "learning_rate": 4.725932749884846e-06,
+      "loss": 0.2434,
+      "step": 8290
+    },
+    {
+      "epoch": 3.82,
+      "grad_norm": 11.662982940673828,
+      "learning_rate": 4.707508060801475e-06,
+      "loss": 0.138,
+      "step": 8300
+    },
+    {
+      "epoch": 3.83,
+      "grad_norm": 1.0320931673049927,
+      "learning_rate": 4.689083371718103e-06,
+      "loss": 0.1436,
+      "step": 8310
+    },
+    {
+      "epoch": 3.83,
+      "grad_norm": 1.6648021936416626,
+      "learning_rate": 4.670658682634731e-06,
+      "loss": 0.0397,
+      "step": 8320
+    },
+    {
+      "epoch": 3.84,
+      "grad_norm": 16.942092895507812,
+      "learning_rate": 4.652233993551359e-06,
+      "loss": 0.1269,
+      "step": 8330
+    },
+    {
+      "epoch": 3.84,
+      "grad_norm": 4.541980266571045,
+      "learning_rate": 4.633809304467988e-06,
+      "loss": 0.1143,
+      "step": 8340
+    },
+    {
+      "epoch": 3.85,
+      "grad_norm": 3.432004690170288,
+      "learning_rate": 4.615384615384616e-06,
+      "loss": 0.048,
+      "step": 8350
+    },
+    {
+      "epoch": 3.85,
+      "grad_norm": 32.53287124633789,
+      "learning_rate": 4.596959926301244e-06,
+      "loss": 0.0444,
+      "step": 8360
+    },
+    {
+      "epoch": 3.85,
+      "grad_norm": 2.8432607650756836,
+      "learning_rate": 4.578535237217872e-06,
+      "loss": 0.0721,
+      "step": 8370
+    },
+    {
+      "epoch": 3.86,
+      "grad_norm": 3.2313575744628906,
+      "learning_rate": 4.560110548134501e-06,
+      "loss": 0.0517,
+      "step": 8380
+    },
+    {
+      "epoch": 3.86,
+      "grad_norm": 40.1295280456543,
+      "learning_rate": 4.541685859051128e-06,
+      "loss": 0.0828,
+      "step": 8390
+    },
+    {
+      "epoch": 3.87,
+      "grad_norm": 1.1382088661193848,
+      "learning_rate": 4.523261169967757e-06,
+      "loss": 0.0692,
+      "step": 8400
+    },
+    {
+      "epoch": 3.87,
+      "grad_norm": 9.753777503967285,
+      "learning_rate": 4.504836480884385e-06,
+      "loss": 0.0555,
+      "step": 8410
+    },
+    {
+      "epoch": 3.88,
+      "grad_norm": 0.4970550537109375,
+      "learning_rate": 4.486411791801014e-06,
+      "loss": 0.1379,
+      "step": 8420
+    },
+    {
+      "epoch": 3.88,
+      "grad_norm": 0.49835485219955444,
+      "learning_rate": 4.467987102717642e-06,
+      "loss": 0.0165,
+      "step": 8430
+    },
+    {
+      "epoch": 3.89,
+      "grad_norm": 3.073554754257202,
+      "learning_rate": 4.449562413634271e-06,
+      "loss": 0.094,
+      "step": 8440
+    },
+    {
+      "epoch": 3.89,
+      "grad_norm": 0.8424185514450073,
+      "learning_rate": 4.431137724550898e-06,
+      "loss": 0.0556,
+      "step": 8450
+    },
+    {
+      "epoch": 3.9,
+      "grad_norm": 0.044442884624004364,
+      "learning_rate": 4.412713035467527e-06,
+      "loss": 0.0396,
+      "step": 8460
+    },
+    {
+      "epoch": 3.9,
+      "grad_norm": 5.324331283569336,
+      "learning_rate": 4.394288346384155e-06,
+      "loss": 0.0217,
+      "step": 8470
+    },
+    {
+      "epoch": 3.91,
+      "grad_norm": 5.435234546661377,
+      "learning_rate": 4.375863657300784e-06,
+      "loss": 0.0768,
+      "step": 8480
+    },
+    {
+      "epoch": 3.91,
+      "grad_norm": 18.99291229248047,
+      "learning_rate": 4.357438968217412e-06,
+      "loss": 0.0758,
+      "step": 8490
+    },
+    {
+      "epoch": 3.91,
+      "grad_norm": 9.785869598388672,
+      "learning_rate": 4.33901427913404e-06,
+      "loss": 0.134,
+      "step": 8500
+    },
+    {
+      "epoch": 3.92,
+      "grad_norm": 0.314177542924881,
+      "learning_rate": 4.320589590050668e-06,
+      "loss": 0.0289,
+      "step": 8510
+    },
+    {
+      "epoch": 3.92,
+      "grad_norm": 12.571952819824219,
+      "learning_rate": 4.302164900967297e-06,
+      "loss": 0.1721,
+      "step": 8520
+    },
+    {
+      "epoch": 3.93,
+      "grad_norm": 0.1812465935945511,
+      "learning_rate": 4.283740211883925e-06,
+      "loss": 0.0234,
+      "step": 8530
+    },
+    {
+      "epoch": 3.93,
+      "grad_norm": 7.266148090362549,
+      "learning_rate": 4.2653155228005535e-06,
+      "loss": 0.0535,
+      "step": 8540
+    },
+    {
+      "epoch": 3.94,
+      "grad_norm": 23.121519088745117,
+      "learning_rate": 4.246890833717182e-06,
+      "loss": 0.0468,
+      "step": 8550
+    },
+    {
+      "epoch": 3.94,
+      "grad_norm": 8.852749824523926,
+      "learning_rate": 4.22846614463381e-06,
+      "loss": 0.1127,
+      "step": 8560
+    },
+    {
+      "epoch": 3.95,
+      "grad_norm": 3.583594560623169,
+      "learning_rate": 4.210041455550438e-06,
+      "loss": 0.1857,
+      "step": 8570
+    },
+    {
+      "epoch": 3.95,
+      "grad_norm": 100.020263671875,
+      "learning_rate": 4.191616766467066e-06,
+      "loss": 0.1334,
+      "step": 8580
+    },
+    {
+      "epoch": 3.96,
+      "grad_norm": 1.3925378322601318,
+      "learning_rate": 4.173192077383694e-06,
+      "loss": 0.2389,
+      "step": 8590
+    },
+    {
+      "epoch": 3.96,
+      "grad_norm": 9.038330078125,
+      "learning_rate": 4.154767388300323e-06,
+      "loss": 0.0313,
+      "step": 8600
+    },
+    {
+      "epoch": 3.97,
+      "grad_norm": 6.1151227951049805,
+      "learning_rate": 4.136342699216951e-06,
+      "loss": 0.1669,
+      "step": 8610
+    },
+    {
+      "epoch": 3.97,
+      "grad_norm": 0.02052326872944832,
+      "learning_rate": 4.1179180101335795e-06,
+      "loss": 0.0402,
+      "step": 8620
+    },
+    {
+      "epoch": 3.97,
+      "grad_norm": 24.436038970947266,
+      "learning_rate": 4.099493321050207e-06,
+      "loss": 0.0638,
+      "step": 8630
+    },
+    {
+      "epoch": 3.98,
+      "grad_norm": 15.031303405761719,
+      "learning_rate": 4.081068631966836e-06,
+      "loss": 0.058,
+      "step": 8640
+    },
+    {
+      "epoch": 3.98,
+      "grad_norm": 17.125295639038086,
+      "learning_rate": 4.062643942883464e-06,
+      "loss": 0.1504,
+      "step": 8650
+    },
+    {
+      "epoch": 3.99,
+      "grad_norm": 0.5870439410209656,
+      "learning_rate": 4.0442192538000925e-06,
+      "loss": 0.0684,
+      "step": 8660
+    },
+    {
+      "epoch": 3.99,
+      "grad_norm": 21.024118423461914,
+      "learning_rate": 4.025794564716721e-06,
+      "loss": 0.2096,
+      "step": 8670
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 4.700799465179443,
+      "learning_rate": 4.007369875633349e-06,
+      "loss": 0.0516,
+      "step": 8680
+    },
+    {
+      "epoch": 4.0,
+      "eval_loss": 1.067345380783081,
+      "eval_runtime": 23.7341,
+      "eval_samples_per_second": 126.78,
+      "eval_steps_per_second": 25.364,
+      "step": 8686
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 28.95529556274414,
+      "learning_rate": 3.988945186549977e-06,
+      "loss": 0.2064,
+      "step": 8690
+    },
+    {
+      "epoch": 4.01,
+      "grad_norm": 2.2235357761383057,
+      "learning_rate": 3.9705204974666055e-06,
+      "loss": 0.0902,
+      "step": 8700
+    },
+    {
+      "epoch": 4.01,
+      "grad_norm": 0.3367871642112732,
+      "learning_rate": 3.952095808383234e-06,
+      "loss": 0.0176,
+      "step": 8710
+    },
+    {
+      "epoch": 4.02,
+      "grad_norm": 0.05267007276415825,
+      "learning_rate": 3.9336711192998624e-06,
+      "loss": 0.0434,
+      "step": 8720
+    },
+    {
+      "epoch": 4.02,
+      "grad_norm": 1.8301619291305542,
+      "learning_rate": 3.915246430216491e-06,
+      "loss": 0.0293,
+      "step": 8730
+    },
+    {
+      "epoch": 4.02,
+      "grad_norm": 41.90878677368164,
+      "learning_rate": 3.8968217411331185e-06,
+      "loss": 0.0868,
+      "step": 8740
+    },
+    {
+      "epoch": 4.03,
+      "grad_norm": 0.11788254231214523,
+      "learning_rate": 3.878397052049747e-06,
+      "loss": 0.0269,
+      "step": 8750
+    },
+    {
+      "epoch": 4.03,
+      "grad_norm": 2.0098631381988525,
+      "learning_rate": 3.8599723629663754e-06,
+      "loss": 0.0745,
+      "step": 8760
+    },
+    {
+      "epoch": 4.04,
+      "grad_norm": 12.796669960021973,
+      "learning_rate": 3.841547673883004e-06,
+      "loss": 0.202,
+      "step": 8770
+    },
+    {
+      "epoch": 4.04,
+      "grad_norm": 17.97233772277832,
+      "learning_rate": 3.823122984799632e-06,
+      "loss": 0.0605,
+      "step": 8780
+    },
+    {
+      "epoch": 4.05,
+      "grad_norm": 0.27959758043289185,
+      "learning_rate": 3.8046982957162604e-06,
+      "loss": 0.0456,
+      "step": 8790
+    },
+    {
+      "epoch": 4.05,
+      "grad_norm": 4.501171588897705,
+      "learning_rate": 3.786273606632888e-06,
+      "loss": 0.1102,
+      "step": 8800
+    },
+    {
+      "epoch": 4.06,
+      "grad_norm": 3.0515573024749756,
+      "learning_rate": 3.7678489175495165e-06,
+      "loss": 0.0093,
+      "step": 8810
+    },
+    {
+      "epoch": 4.06,
+      "grad_norm": 1.5558323860168457,
+      "learning_rate": 3.7494242284661445e-06,
+      "loss": 0.0217,
+      "step": 8820
+    },
+    {
+      "epoch": 4.07,
+      "grad_norm": 0.038831084966659546,
+      "learning_rate": 3.730999539382773e-06,
+      "loss": 0.0282,
+      "step": 8830
+    },
+    {
+      "epoch": 4.07,
+      "grad_norm": 0.040046580135822296,
+      "learning_rate": 3.7125748502994014e-06,
+      "loss": 0.047,
+      "step": 8840
+    },
+    {
+      "epoch": 4.08,
+      "grad_norm": 0.07264097034931183,
+      "learning_rate": 3.6941501612160295e-06,
+      "loss": 0.0433,
+      "step": 8850
+    },
+    {
+      "epoch": 4.08,
+      "grad_norm": 20.486833572387695,
+      "learning_rate": 3.675725472132658e-06,
+      "loss": 0.0602,
+      "step": 8860
+    },
+    {
+      "epoch": 4.08,
+      "grad_norm": 28.8565616607666,
+      "learning_rate": 3.6573007830492864e-06,
+      "loss": 0.052,
+      "step": 8870
+    },
+    {
+      "epoch": 4.09,
+      "grad_norm": 4.701373100280762,
+      "learning_rate": 3.6388760939659144e-06,
+      "loss": 0.0517,
+      "step": 8880
+    },
+    {
+      "epoch": 4.09,
+      "grad_norm": 0.03912418708205223,
+      "learning_rate": 3.620451404882543e-06,
+      "loss": 0.0755,
+      "step": 8890
+    },
+    {
+      "epoch": 4.1,
+      "grad_norm": 21.40830421447754,
+      "learning_rate": 3.6020267157991714e-06,
+      "loss": 0.092,
+      "step": 8900
+    },
+    {
+      "epoch": 4.1,
+      "grad_norm": 0.7065961956977844,
+      "learning_rate": 3.5836020267157994e-06,
+      "loss": 0.0725,
+      "step": 8910
+    },
+    {
+      "epoch": 4.11,
+      "grad_norm": 2.3827240467071533,
+      "learning_rate": 3.565177337632428e-06,
+      "loss": 0.0334,
+      "step": 8920
+    },
+    {
+      "epoch": 4.11,
+      "grad_norm": 0.14933879673480988,
+      "learning_rate": 3.546752648549056e-06,
+      "loss": 0.0246,
+      "step": 8930
+    },
+    {
+      "epoch": 4.12,
+      "grad_norm": 22.343624114990234,
+      "learning_rate": 3.5283279594656844e-06,
+      "loss": 0.0489,
+      "step": 8940
+    },
+    {
+      "epoch": 4.12,
+      "grad_norm": 18.39412498474121,
+      "learning_rate": 3.509903270382313e-06,
+      "loss": 0.0725,
+      "step": 8950
+    },
+    {
+      "epoch": 4.13,
+      "grad_norm": 1.3741297721862793,
+      "learning_rate": 3.491478581298941e-06,
+      "loss": 0.0536,
+      "step": 8960
+    },
+    {
+      "epoch": 4.13,
+      "grad_norm": 1.3275773525238037,
+      "learning_rate": 3.4730538922155693e-06,
+      "loss": 0.0317,
+      "step": 8970
+    },
+    {
+      "epoch": 4.14,
+      "grad_norm": 48.96095275878906,
+      "learning_rate": 3.4546292031321978e-06,
+      "loss": 0.0681,
+      "step": 8980
+    },
+    {
+      "epoch": 4.14,
+      "grad_norm": 15.314319610595703,
+      "learning_rate": 3.436204514048826e-06,
+      "loss": 0.0182,
+      "step": 8990
+    },
+    {
+      "epoch": 4.14,
+      "grad_norm": 5.03747034072876,
+      "learning_rate": 3.4177798249654543e-06,
+      "loss": 0.1368,
+      "step": 9000
+    },
+    {
+      "epoch": 4.15,
+      "grad_norm": 12.523550987243652,
+      "learning_rate": 3.399355135882082e-06,
+      "loss": 0.1895,
+      "step": 9010
+    },
+    {
+      "epoch": 4.15,
+      "grad_norm": 11.011829376220703,
+      "learning_rate": 3.3809304467987104e-06,
+      "loss": 0.0094,
+      "step": 9020
+    },
+    {
+      "epoch": 4.16,
+      "grad_norm": 14.987722396850586,
+      "learning_rate": 3.3625057577153384e-06,
+      "loss": 0.0409,
+      "step": 9030
+    },
+    {
+      "epoch": 4.16,
+      "grad_norm": 0.008163328282535076,
+      "learning_rate": 3.344081068631967e-06,
+      "loss": 0.0846,
+      "step": 9040
+    },
+    {
+      "epoch": 4.17,
+      "grad_norm": 55.82210159301758,
+      "learning_rate": 3.3256563795485953e-06,
+      "loss": 0.0572,
+      "step": 9050
+    },
+    {
+      "epoch": 4.17,
+      "grad_norm": 2.261767864227295,
+      "learning_rate": 3.3072316904652234e-06,
+      "loss": 0.0378,
+      "step": 9060
+    },
+    {
+      "epoch": 4.18,
+      "grad_norm": 10.426541328430176,
+      "learning_rate": 3.288807001381852e-06,
+      "loss": 0.0109,
+      "step": 9070
+    },
+    {
+      "epoch": 4.18,
+      "grad_norm": 0.27620577812194824,
+      "learning_rate": 3.2703823122984803e-06,
+      "loss": 0.0663,
+      "step": 9080
+    },
+    {
+      "epoch": 4.19,
+      "grad_norm": 12.906811714172363,
+      "learning_rate": 3.2519576232151083e-06,
+      "loss": 0.0683,
+      "step": 9090
+    },
+    {
+      "epoch": 4.19,
+      "grad_norm": 0.09020080417394638,
+      "learning_rate": 3.2335329341317368e-06,
+      "loss": 0.0194,
+      "step": 9100
+    },
+    {
+      "epoch": 4.2,
+      "grad_norm": 31.966867446899414,
+      "learning_rate": 3.2151082450483652e-06,
+      "loss": 0.056,
+      "step": 9110
+    },
+    {
+      "epoch": 4.2,
+      "grad_norm": 1.6081633567810059,
+      "learning_rate": 3.1966835559649933e-06,
+      "loss": 0.0566,
+      "step": 9120
+    },
+    {
+      "epoch": 4.2,
+      "grad_norm": 0.3287917673587799,
+      "learning_rate": 3.1782588668816217e-06,
+      "loss": 0.0112,
+      "step": 9130
+    },
+    {
+      "epoch": 4.21,
+      "grad_norm": 4.9299702644348145,
+      "learning_rate": 3.1598341777982498e-06,
+      "loss": 0.0241,
+      "step": 9140
+    },
+    {
+      "epoch": 4.21,
+      "grad_norm": 0.21217700839042664,
+      "learning_rate": 3.1414094887148782e-06,
+      "loss": 0.0816,
+      "step": 9150
+    },
+    {
+      "epoch": 4.22,
+      "grad_norm": 0.09667451679706573,
+      "learning_rate": 3.1229847996315067e-06,
+      "loss": 0.0672,
+      "step": 9160
+    },
+    {
+      "epoch": 4.22,
+      "grad_norm": 17.0867919921875,
+      "learning_rate": 3.1045601105481347e-06,
+      "loss": 0.1054,
+      "step": 9170
+    },
+    {
+      "epoch": 4.23,
+      "grad_norm": 2.1458334922790527,
+      "learning_rate": 3.086135421464763e-06,
+      "loss": 0.0137,
+      "step": 9180
+    },
+    {
+      "epoch": 4.23,
+      "grad_norm": 14.453232765197754,
+      "learning_rate": 3.0677107323813917e-06,
+      "loss": 0.055,
+      "step": 9190
+    },
+    {
+      "epoch": 4.24,
+      "grad_norm": 0.03414759412407875,
+      "learning_rate": 3.0492860432980197e-06,
+      "loss": 0.0648,
+      "step": 9200
+    },
+    {
+      "epoch": 4.24,
+      "grad_norm": 0.03962136059999466,
+      "learning_rate": 3.030861354214648e-06,
+      "loss": 0.0745,
+      "step": 9210
+    },
+    {
+      "epoch": 4.25,
+      "grad_norm": 10.716452598571777,
+      "learning_rate": 3.0124366651312766e-06,
+      "loss": 0.0588,
+      "step": 9220
+    },
+    {
+      "epoch": 4.25,
+      "grad_norm": 11.228912353515625,
+      "learning_rate": 2.9940119760479042e-06,
+      "loss": 0.0273,
+      "step": 9230
+    },
+    {
+      "epoch": 4.26,
+      "grad_norm": 1.6607997417449951,
+      "learning_rate": 2.9755872869645323e-06,
+      "loss": 0.003,
+      "step": 9240
+    },
+    {
+      "epoch": 4.26,
+      "grad_norm": 0.01613614894449711,
+      "learning_rate": 2.9571625978811607e-06,
+      "loss": 0.0093,
+      "step": 9250
+    },
+    {
+      "epoch": 4.26,
+      "grad_norm": 0.053402092307806015,
+      "learning_rate": 2.938737908797789e-06,
+      "loss": 0.0153,
+      "step": 9260
+    },
+    {
+      "epoch": 4.27,
+      "grad_norm": 31.77798080444336,
+      "learning_rate": 2.9203132197144172e-06,
+      "loss": 0.0495,
+      "step": 9270
+    },
+    {
+      "epoch": 4.27,
+      "grad_norm": 0.6066150069236755,
+      "learning_rate": 2.9018885306310457e-06,
+      "loss": 0.0112,
+      "step": 9280
+    },
+    {
+      "epoch": 4.28,
+      "grad_norm": 0.16345077753067017,
+      "learning_rate": 2.883463841547674e-06,
+      "loss": 0.0317,
+      "step": 9290
+    },
+    {
+      "epoch": 4.28,
+      "grad_norm": 31.696971893310547,
+      "learning_rate": 2.865039152464302e-06,
+      "loss": 0.2268,
+      "step": 9300
+    },
+    {
+      "epoch": 4.29,
+      "grad_norm": 27.824552536010742,
+      "learning_rate": 2.8466144633809307e-06,
+      "loss": 0.0143,
+      "step": 9310
+    },
+    {
+      "epoch": 4.29,
+      "grad_norm": 1.506197214126587,
+      "learning_rate": 2.828189774297559e-06,
+      "loss": 0.0288,
+      "step": 9320
+    },
+    {
+      "epoch": 4.3,
+      "grad_norm": 23.88454818725586,
+      "learning_rate": 2.809765085214187e-06,
+      "loss": 0.0382,
+      "step": 9330
+    },
+    {
+      "epoch": 4.3,
+      "grad_norm": 0.4401777982711792,
+      "learning_rate": 2.7913403961308156e-06,
+      "loss": 0.0468,
+      "step": 9340
+    },
+    {
+      "epoch": 4.31,
+      "grad_norm": 26.670501708984375,
+      "learning_rate": 2.7729157070474437e-06,
+      "loss": 0.0637,
+      "step": 9350
+    },
+    {
+      "epoch": 4.31,
+      "grad_norm": 0.05980992317199707,
+      "learning_rate": 2.754491017964072e-06,
+      "loss": 0.145,
+      "step": 9360
+    },
+    {
+      "epoch": 4.31,
+      "grad_norm": 22.419076919555664,
+      "learning_rate": 2.7360663288807006e-06,
+      "loss": 0.0209,
+      "step": 9370
+    },
+    {
+      "epoch": 4.32,
+      "grad_norm": 29.66434097290039,
+      "learning_rate": 2.7176416397973286e-06,
+      "loss": 0.0784,
+      "step": 9380
+    },
+    {
+      "epoch": 4.32,
+      "grad_norm": 0.5142760872840881,
+      "learning_rate": 2.699216950713957e-06,
+      "loss": 0.1512,
+      "step": 9390
+    },
+    {
+      "epoch": 4.33,
+      "grad_norm": 3.753553628921509,
+      "learning_rate": 2.6807922616305855e-06,
+      "loss": 0.027,
+      "step": 9400
+    },
+    {
+      "epoch": 4.33,
+      "grad_norm": 8.520071029663086,
+      "learning_rate": 2.6623675725472136e-06,
+      "loss": 0.0061,
+      "step": 9410
+    },
+    {
+      "epoch": 4.34,
+      "grad_norm": 144.6112823486328,
+      "learning_rate": 2.643942883463842e-06,
+      "loss": 0.1219,
+      "step": 9420
+    },
+    {
+      "epoch": 4.34,
+      "grad_norm": 0.030717788264155388,
+      "learning_rate": 2.62551819438047e-06,
+      "loss": 0.0388,
+      "step": 9430
+    },
+    {
+      "epoch": 4.35,
+      "grad_norm": 0.12438066303730011,
+      "learning_rate": 2.6070935052970985e-06,
+      "loss": 0.1555,
+      "step": 9440
+    },
+    {
+      "epoch": 4.35,
+      "grad_norm": 25.153072357177734,
+      "learning_rate": 2.588668816213726e-06,
+      "loss": 0.0447,
+      "step": 9450
+    },
+    {
+      "epoch": 4.36,
+      "grad_norm": 28.93138313293457,
+      "learning_rate": 2.5702441271303546e-06,
+      "loss": 0.045,
+      "step": 9460
+    },
+    {
+      "epoch": 4.36,
+      "grad_norm": 5.940371036529541,
+      "learning_rate": 2.551819438046983e-06,
+      "loss": 0.1211,
+      "step": 9470
+    },
+    {
+      "epoch": 4.37,
+      "grad_norm": 14.700384140014648,
+      "learning_rate": 2.533394748963611e-06,
+      "loss": 0.06,
+      "step": 9480
+    },
+    {
+      "epoch": 4.37,
+      "grad_norm": 0.2683044672012329,
+      "learning_rate": 2.5149700598802396e-06,
+      "loss": 0.0732,
+      "step": 9490
+    },
+    {
+      "epoch": 4.37,
+      "grad_norm": 4.030006408691406,
+      "learning_rate": 2.496545370796868e-06,
+      "loss": 0.0272,
+      "step": 9500
+    },
+    {
+      "epoch": 4.38,
+      "grad_norm": 43.98999786376953,
+      "learning_rate": 2.478120681713496e-06,
+      "loss": 0.0916,
+      "step": 9510
+    },
+    {
+      "epoch": 4.38,
+      "grad_norm": 1.069204568862915,
+      "learning_rate": 2.4596959926301245e-06,
+      "loss": 0.039,
+      "step": 9520
+    },
+    {
+      "epoch": 4.39,
+      "grad_norm": 0.019278841093182564,
+      "learning_rate": 2.441271303546753e-06,
+      "loss": 0.0116,
+      "step": 9530
+    },
+    {
+      "epoch": 4.39,
+      "grad_norm": 0.8535725474357605,
+      "learning_rate": 2.422846614463381e-06,
+      "loss": 0.0393,
+      "step": 9540
+    },
+    {
+      "epoch": 4.4,
+      "grad_norm": 0.15458782017230988,
+      "learning_rate": 2.4044219253800095e-06,
+      "loss": 0.092,
+      "step": 9550
+    },
+    {
+      "epoch": 4.4,
+      "grad_norm": 0.5165956020355225,
+      "learning_rate": 2.3859972362966375e-06,
+      "loss": 0.0227,
+      "step": 9560
+    },
+    {
+      "epoch": 4.41,
+      "grad_norm": 5.305365085601807,
+      "learning_rate": 2.367572547213266e-06,
+      "loss": 0.0416,
+      "step": 9570
+    },
+    {
+      "epoch": 4.41,
+      "grad_norm": 13.265985488891602,
+      "learning_rate": 2.3491478581298945e-06,
+      "loss": 0.082,
+      "step": 9580
+    },
+    {
+      "epoch": 4.42,
+      "grad_norm": 0.006752397865056992,
+      "learning_rate": 2.3307231690465225e-06,
+      "loss": 0.1609,
+      "step": 9590
+    },
+    {
+      "epoch": 4.42,
+      "grad_norm": 0.02739543281495571,
+      "learning_rate": 2.312298479963151e-06,
+      "loss": 0.1303,
+      "step": 9600
+    },
+    {
+      "epoch": 4.43,
+      "grad_norm": 39.791893005371094,
+      "learning_rate": 2.293873790879779e-06,
+      "loss": 0.026,
+      "step": 9610
+    },
+    {
+      "epoch": 4.43,
+      "grad_norm": 59.16732406616211,
+      "learning_rate": 2.2754491017964075e-06,
+      "loss": 0.0425,
+      "step": 9620
+    },
+    {
+      "epoch": 4.43,
+      "grad_norm": 0.6513130068778992,
+      "learning_rate": 2.2570244127130355e-06,
+      "loss": 0.0512,
+      "step": 9630
+    },
+    {
+      "epoch": 4.44,
+      "grad_norm": 0.14722326397895813,
+      "learning_rate": 2.238599723629664e-06,
+      "loss": 0.022,
+      "step": 9640
+    },
+    {
+      "epoch": 4.44,
+      "grad_norm": 0.9835072159767151,
+      "learning_rate": 2.220175034546292e-06,
+      "loss": 0.0377,
+      "step": 9650
+    },
+    {
+      "epoch": 4.45,
+      "grad_norm": 11.475459098815918,
+      "learning_rate": 2.2017503454629205e-06,
+      "loss": 0.0595,
+      "step": 9660
+    },
+    {
+      "epoch": 4.45,
+      "grad_norm": 0.6809270977973938,
+      "learning_rate": 2.183325656379549e-06,
+      "loss": 0.015,
+      "step": 9670
+    },
+    {
+      "epoch": 4.46,
+      "grad_norm": 12.190295219421387,
+      "learning_rate": 2.164900967296177e-06,
+      "loss": 0.0682,
+      "step": 9680
+    },
+    {
+      "epoch": 4.46,
+      "grad_norm": 11.654221534729004,
+      "learning_rate": 2.1464762782128054e-06,
+      "loss": 0.0355,
+      "step": 9690
+    },
+    {
+      "epoch": 4.47,
+      "grad_norm": 12.240227699279785,
+      "learning_rate": 2.128051589129434e-06,
+      "loss": 0.0495,
+      "step": 9700
+    },
+    {
+      "epoch": 4.47,
+      "grad_norm": 3.1359102725982666,
+      "learning_rate": 2.109626900046062e-06,
+      "loss": 0.0177,
+      "step": 9710
+    },
+    {
+      "epoch": 4.48,
+      "grad_norm": 5.230315208435059,
+      "learning_rate": 2.09120221096269e-06,
+      "loss": 0.0591,
+      "step": 9720
+    },
+    {
+      "epoch": 4.48,
+      "grad_norm": 0.3116825520992279,
+      "learning_rate": 2.0727775218793184e-06,
+      "loss": 0.022,
+      "step": 9730
+    },
+    {
+      "epoch": 4.49,
+      "grad_norm": 17.061006546020508,
+      "learning_rate": 2.0543528327959465e-06,
+      "loss": 0.0964,
+      "step": 9740
+    },
+    {
+      "epoch": 4.49,
+      "grad_norm": 32.36555099487305,
+      "learning_rate": 2.035928143712575e-06,
+      "loss": 0.12,
+      "step": 9750
+    },
+    {
+      "epoch": 4.49,
+      "grad_norm": 8.813058853149414,
+      "learning_rate": 2.0175034546292034e-06,
+      "loss": 0.0642,
+      "step": 9760
+    },
+    {
+      "epoch": 4.5,
+      "grad_norm": 0.04783041030168533,
+      "learning_rate": 1.9990787655458314e-06,
+      "loss": 0.0166,
+      "step": 9770
+    },
+    {
+      "epoch": 4.5,
+      "grad_norm": 28.42287826538086,
+      "learning_rate": 1.98065407646246e-06,
+      "loss": 0.1366,
+      "step": 9780
+    },
+    {
+      "epoch": 4.51,
+      "grad_norm": 25.180908203125,
+      "learning_rate": 1.9622293873790883e-06,
+      "loss": 0.0703,
+      "step": 9790
+    },
+    {
+      "epoch": 4.51,
+      "grad_norm": 0.8870308995246887,
+      "learning_rate": 1.9438046982957164e-06,
+      "loss": 0.0086,
+      "step": 9800
+    },
+    {
+      "epoch": 4.52,
+      "grad_norm": 0.04639322683215141,
+      "learning_rate": 1.925380009212345e-06,
+      "loss": 0.0133,
+      "step": 9810
+    },
+    {
+      "epoch": 4.52,
+      "grad_norm": 0.004572540521621704,
+      "learning_rate": 1.906955320128973e-06,
+      "loss": 0.015,
+      "step": 9820
+    },
+    {
+      "epoch": 4.53,
+      "grad_norm": 0.0027952073141932487,
+      "learning_rate": 1.8885306310456011e-06,
+      "loss": 0.0582,
+      "step": 9830
+    },
+    {
+      "epoch": 4.53,
+      "grad_norm": 10.944767951965332,
+      "learning_rate": 1.8701059419622294e-06,
+      "loss": 0.0686,
+      "step": 9840
+    },
+    {
+      "epoch": 4.54,
+      "grad_norm": 0.010235718451440334,
+      "learning_rate": 1.8516812528788578e-06,
+      "loss": 0.0183,
+      "step": 9850
+    },
+    {
+      "epoch": 4.54,
+      "grad_norm": 0.39722442626953125,
+      "learning_rate": 1.833256563795486e-06,
+      "loss": 0.1005,
+      "step": 9860
+    },
+    {
+      "epoch": 4.55,
+      "grad_norm": 0.2935176491737366,
+      "learning_rate": 1.8148318747121143e-06,
+      "loss": 0.0204,
+      "step": 9870
+    },
+    {
+      "epoch": 4.55,
+      "grad_norm": 7.280217170715332,
+      "learning_rate": 1.7964071856287426e-06,
+      "loss": 0.035,
+      "step": 9880
+    },
+    {
+      "epoch": 4.55,
+      "grad_norm": 56.73902893066406,
+      "learning_rate": 1.777982496545371e-06,
+      "loss": 0.0887,
+      "step": 9890
+    },
+    {
+      "epoch": 4.56,
+      "grad_norm": 0.41625165939331055,
+      "learning_rate": 1.7595578074619993e-06,
+      "loss": 0.075,
+      "step": 9900
+    },
+    {
+      "epoch": 4.56,
+      "grad_norm": 14.798352241516113,
+      "learning_rate": 1.7411331183786275e-06,
+      "loss": 0.0148,
+      "step": 9910
+    },
+    {
+      "epoch": 4.57,
+      "grad_norm": 10.674389839172363,
+      "learning_rate": 1.7227084292952558e-06,
+      "loss": 0.0768,
+      "step": 9920
+    },
+    {
+      "epoch": 4.57,
+      "grad_norm": 4.959482192993164,
+      "learning_rate": 1.7042837402118843e-06,
+      "loss": 0.0485,
+      "step": 9930
+    },
+    {
+      "epoch": 4.58,
+      "grad_norm": 1.490976095199585,
+      "learning_rate": 1.6858590511285123e-06,
+      "loss": 0.0051,
+      "step": 9940
+    },
+    {
+      "epoch": 4.58,
+      "grad_norm": 11.091095924377441,
+      "learning_rate": 1.6674343620451405e-06,
+      "loss": 0.0243,
+      "step": 9950
+    },
+    {
+      "epoch": 4.59,
+      "grad_norm": 0.06007695570588112,
+      "learning_rate": 1.6490096729617688e-06,
+      "loss": 0.0294,
+      "step": 9960
+    },
+    {
+      "epoch": 4.59,
+      "grad_norm": 1.6352295875549316,
+      "learning_rate": 1.630584983878397e-06,
+      "loss": 0.0802,
+      "step": 9970
+    },
+    {
+      "epoch": 4.6,
+      "grad_norm": 11.841517448425293,
+      "learning_rate": 1.6121602947950255e-06,
+      "loss": 0.0354,
+      "step": 9980
+    },
+    {
+      "epoch": 4.6,
+      "grad_norm": 0.21027013659477234,
+      "learning_rate": 1.5937356057116538e-06,
+      "loss": 0.1059,
+      "step": 9990
+    },
+    {
+      "epoch": 4.61,
+      "grad_norm": 0.00931438896805048,
+      "learning_rate": 1.575310916628282e-06,
+      "loss": 0.0075,
+      "step": 10000
+    },
+    {
+      "epoch": 4.61,
+      "grad_norm": 52.3951416015625,
+      "learning_rate": 1.5568862275449105e-06,
+      "loss": 0.0485,
+      "step": 10010
+    },
+    {
+      "epoch": 4.61,
+      "grad_norm": 66.5012435913086,
+      "learning_rate": 1.5384615384615387e-06,
+      "loss": 0.036,
+      "step": 10020
+    },
+    {
+      "epoch": 4.62,
+      "grad_norm": 0.45424723625183105,
+      "learning_rate": 1.520036849378167e-06,
+      "loss": 0.0195,
+      "step": 10030
+    },
+    {
+      "epoch": 4.62,
+      "grad_norm": 29.831295013427734,
+      "learning_rate": 1.5016121602947952e-06,
+      "loss": 0.1666,
+      "step": 10040
+    },
+    {
+      "epoch": 4.63,
+      "grad_norm": 2.3781497478485107,
+      "learning_rate": 1.4831874712114233e-06,
+      "loss": 0.0374,
+      "step": 10050
+    },
+    {
+      "epoch": 4.63,
+      "grad_norm": 0.4033482074737549,
+      "learning_rate": 1.4647627821280517e-06,
+      "loss": 0.0337,
+      "step": 10060
+    },
+    {
+      "epoch": 4.64,
+      "grad_norm": 2.8482015132904053,
+      "learning_rate": 1.44633809304468e-06,
+      "loss": 0.1243,
+      "step": 10070
+    },
+    {
+      "epoch": 4.64,
+      "grad_norm": 0.23622600734233856,
+      "learning_rate": 1.4279134039613082e-06,
+      "loss": 0.0173,
+      "step": 10080
+    },
+    {
+      "epoch": 4.65,
+      "grad_norm": 3.973984956741333,
+      "learning_rate": 1.4094887148779365e-06,
+      "loss": 0.0235,
+      "step": 10090
+    },
+    {
+      "epoch": 4.65,
+      "grad_norm": 0.05539378896355629,
+      "learning_rate": 1.391064025794565e-06,
+      "loss": 0.0165,
+      "step": 10100
+    },
+    {
+      "epoch": 4.66,
+      "grad_norm": 23.146671295166016,
+      "learning_rate": 1.3726393367111932e-06,
+      "loss": 0.0469,
+      "step": 10110
+    },
+    {
+      "epoch": 4.66,
+      "grad_norm": 0.9822247624397278,
+      "learning_rate": 1.3542146476278214e-06,
+      "loss": 0.0024,
+      "step": 10120
+    },
+    {
+      "epoch": 4.66,
+      "grad_norm": 40.405921936035156,
+      "learning_rate": 1.3357899585444497e-06,
+      "loss": 0.086,
+      "step": 10130
+    },
+    {
+      "epoch": 4.67,
+      "grad_norm": 26.897912979125977,
+      "learning_rate": 1.3173652694610781e-06,
+      "loss": 0.0703,
+      "step": 10140
+    },
+    {
+      "epoch": 4.67,
+      "grad_norm": 17.91119956970215,
+      "learning_rate": 1.2989405803777062e-06,
+      "loss": 0.1031,
+      "step": 10150
+    },
+    {
+      "epoch": 4.68,
+      "grad_norm": 41.387184143066406,
+      "learning_rate": 1.2805158912943344e-06,
+      "loss": 0.0915,
+      "step": 10160
+    },
+    {
+      "epoch": 4.68,
+      "grad_norm": 0.24964046478271484,
+      "learning_rate": 1.2620912022109627e-06,
+      "loss": 0.0561,
+      "step": 10170
+    },
+    {
+      "epoch": 4.69,
+      "grad_norm": 83.06295013427734,
+      "learning_rate": 1.243666513127591e-06,
+      "loss": 0.1021,
+      "step": 10180
+    },
+    {
+      "epoch": 4.69,
+      "grad_norm": 0.8376673460006714,
+      "learning_rate": 1.2252418240442194e-06,
+      "loss": 0.0273,
+      "step": 10190
+    },
+    {
+      "epoch": 4.7,
+      "grad_norm": 16.103069305419922,
+      "learning_rate": 1.2068171349608476e-06,
+      "loss": 0.0228,
+      "step": 10200
+    },
+    {
+      "epoch": 4.7,
+      "grad_norm": 0.007931999862194061,
+      "learning_rate": 1.1883924458774759e-06,
+      "loss": 0.157,
+      "step": 10210
+    },
+    {
+      "epoch": 4.71,
+      "grad_norm": 19.681793212890625,
+      "learning_rate": 1.1699677567941041e-06,
+      "loss": 0.1459,
+      "step": 10220
+    },
+    {
+      "epoch": 4.71,
+      "grad_norm": 0.68458092212677,
+      "learning_rate": 1.1515430677107326e-06,
+      "loss": 0.0487,
+      "step": 10230
+    },
+    {
+      "epoch": 4.72,
+      "grad_norm": 60.22011947631836,
+      "learning_rate": 1.1331183786273606e-06,
+      "loss": 0.046,
+      "step": 10240
+    },
+    {
+      "epoch": 4.72,
+      "grad_norm": 2.8590993881225586,
+      "learning_rate": 1.114693689543989e-06,
+      "loss": 0.0426,
+      "step": 10250
+    },
+    {
+      "epoch": 4.72,
+      "grad_norm": 33.82815933227539,
+      "learning_rate": 1.0962690004606173e-06,
+      "loss": 0.1487,
+      "step": 10260
+    },
+    {
+      "epoch": 4.73,
+      "grad_norm": 0.24914075434207916,
+      "learning_rate": 1.0778443113772456e-06,
+      "loss": 0.0323,
+      "step": 10270
+    },
+    {
+      "epoch": 4.73,
+      "grad_norm": 0.33567020297050476,
+      "learning_rate": 1.0594196222938738e-06,
+      "loss": 0.0351,
+      "step": 10280
+    },
+    {
+      "epoch": 4.74,
+      "grad_norm": 3.9129273891448975,
+      "learning_rate": 1.040994933210502e-06,
+      "loss": 0.13,
+      "step": 10290
+    },
+    {
+      "epoch": 4.74,
+      "grad_norm": 83.58686065673828,
+      "learning_rate": 1.0225702441271303e-06,
+      "loss": 0.0715,
+      "step": 10300
+    },
+    {
+      "epoch": 4.75,
+      "grad_norm": 0.13932563364505768,
+      "learning_rate": 1.0041455550437588e-06,
+      "loss": 0.044,
+      "step": 10310
+    },
+    {
+      "epoch": 4.75,
+      "grad_norm": 5.902960300445557,
+      "learning_rate": 9.85720865960387e-07,
+      "loss": 0.0944,
+      "step": 10320
+    },
+    {
+      "epoch": 4.76,
+      "grad_norm": 17.72955322265625,
+      "learning_rate": 9.672961768770153e-07,
+      "loss": 0.0508,
+      "step": 10330
+    },
+    {
+      "epoch": 4.76,
+      "grad_norm": 0.004214110318571329,
+      "learning_rate": 9.488714877936434e-07,
+      "loss": 0.0074,
+      "step": 10340
+    },
+    {
+      "epoch": 4.77,
+      "grad_norm": 0.1862613707780838,
+      "learning_rate": 9.304467987102718e-07,
+      "loss": 0.0071,
+      "step": 10350
+    },
+    {
+      "epoch": 4.77,
+      "grad_norm": 0.09862124174833298,
+      "learning_rate": 9.120221096269001e-07,
+      "loss": 0.0245,
+      "step": 10360
+    },
+    {
+      "epoch": 4.78,
+      "grad_norm": 0.07321223616600037,
+      "learning_rate": 8.935974205435284e-07,
+      "loss": 0.0388,
+      "step": 10370
+    },
+    {
+      "epoch": 4.78,
+      "grad_norm": 0.15424856543540955,
+      "learning_rate": 8.751727314601568e-07,
+      "loss": 0.0206,
+      "step": 10380
+    },
+    {
+      "epoch": 4.78,
+      "grad_norm": 12.509618759155273,
+      "learning_rate": 8.56748042376785e-07,
+      "loss": 0.0903,
+      "step": 10390
+    },
+    {
+      "epoch": 4.79,
+      "grad_norm": 3.1824991703033447,
+      "learning_rate": 8.383233532934132e-07,
+      "loss": 0.0393,
+      "step": 10400
+    },
+    {
+      "epoch": 4.79,
+      "grad_norm": 0.0018185112858191133,
+      "learning_rate": 8.198986642100415e-07,
+      "loss": 0.0549,
+      "step": 10410
+    },
+    {
+      "epoch": 4.8,
+      "grad_norm": 20.757373809814453,
+      "learning_rate": 8.014739751266698e-07,
+      "loss": 0.0547,
+      "step": 10420
+    },
+    {
+      "epoch": 4.8,
+      "grad_norm": 3.057952404022217,
+      "learning_rate": 7.830492860432981e-07,
+      "loss": 0.0953,
+      "step": 10430
+    },
+    {
+      "epoch": 4.81,
+      "grad_norm": 2.5925347805023193,
+      "learning_rate": 7.646245969599264e-07,
+      "loss": 0.0536,
+      "step": 10440
+    },
+    {
+      "epoch": 4.81,
+      "grad_norm": 5.123808860778809,
+      "learning_rate": 7.461999078765546e-07,
+      "loss": 0.0964,
+      "step": 10450
+    },
+    {
+      "epoch": 4.82,
+      "grad_norm": 0.8164027333259583,
+      "learning_rate": 7.277752187931829e-07,
+      "loss": 0.0528,
+      "step": 10460
+    },
+    {
+      "epoch": 4.82,
+      "grad_norm": 0.057908181101083755,
+      "learning_rate": 7.093505297098112e-07,
+      "loss": 0.0943,
+      "step": 10470
+    },
+    {
+      "epoch": 4.83,
+      "grad_norm": 27.861305236816406,
+      "learning_rate": 6.909258406264395e-07,
+      "loss": 0.0868,
+      "step": 10480
+    },
+    {
+      "epoch": 4.83,
+      "grad_norm": 0.0010043345391750336,
+      "learning_rate": 6.725011515430678e-07,
+      "loss": 0.0421,
+      "step": 10490
+    },
+    {
+      "epoch": 4.84,
+      "grad_norm": 0.40924185514450073,
+      "learning_rate": 6.540764624596961e-07,
+      "loss": 0.0522,
+      "step": 10500
+    },
+    {
+      "epoch": 4.84,
+      "grad_norm": 4.352136135101318,
+      "learning_rate": 6.356517733763243e-07,
+      "loss": 0.0917,
+      "step": 10510
+    },
+    {
+      "epoch": 4.84,
+      "grad_norm": 0.10656201094388962,
+      "learning_rate": 6.172270842929526e-07,
+      "loss": 0.0299,
+      "step": 10520
+    },
+    {
+      "epoch": 4.85,
+      "grad_norm": 5.475188255310059,
+      "learning_rate": 5.988023952095809e-07,
+      "loss": 0.0084,
+      "step": 10530
+    },
+    {
+      "epoch": 4.85,
+      "grad_norm": 0.7824916243553162,
+      "learning_rate": 5.803777061262092e-07,
+      "loss": 0.093,
+      "step": 10540
+    },
+    {
+      "epoch": 4.86,
+      "grad_norm": 4.100082874298096,
+      "learning_rate": 5.619530170428374e-07,
+      "loss": 0.08,
+      "step": 10550
+    },
+    {
+      "epoch": 4.86,
+      "grad_norm": 0.15313991904258728,
+      "learning_rate": 5.435283279594658e-07,
+      "loss": 0.0856,
+      "step": 10560
+    },
+    {
+      "epoch": 4.87,
+      "grad_norm": 14.245123863220215,
+      "learning_rate": 5.25103638876094e-07,
+      "loss": 0.0179,
+      "step": 10570
+    },
+    {
+      "epoch": 4.87,
+      "grad_norm": 17.8038387298584,
+      "learning_rate": 5.066789497927223e-07,
+      "loss": 0.0163,
+      "step": 10580
+    },
+    {
+      "epoch": 4.88,
+      "grad_norm": 1.0783069133758545,
+      "learning_rate": 4.882542607093505e-07,
+      "loss": 0.0267,
+      "step": 10590
+    },
+    {
+      "epoch": 4.88,
+      "grad_norm": 26.675325393676758,
+      "learning_rate": 4.6982957162597884e-07,
+      "loss": 0.1119,
+      "step": 10600
+    },
+    {
+      "epoch": 4.89,
+      "grad_norm": 1.53312087059021,
+      "learning_rate": 4.5140488254260714e-07,
+      "loss": 0.142,
+      "step": 10610
+    },
+    {
+      "epoch": 4.89,
+      "grad_norm": 0.6964000463485718,
+      "learning_rate": 4.3298019345923544e-07,
+      "loss": 0.0527,
+      "step": 10620
+    },
+    {
+      "epoch": 4.9,
+      "grad_norm": 15.056397438049316,
+      "learning_rate": 4.145555043758637e-07,
+      "loss": 0.0305,
+      "step": 10630
+    },
+    {
+      "epoch": 4.9,
+      "grad_norm": 38.042449951171875,
+      "learning_rate": 3.96130815292492e-07,
+      "loss": 0.0686,
+      "step": 10640
+    },
+    {
+      "epoch": 4.9,
+      "grad_norm": 30.0996036529541,
+      "learning_rate": 3.777061262091203e-07,
+      "loss": 0.0472,
+      "step": 10650
+    },
+    {
+      "epoch": 4.91,
+      "grad_norm": 23.385677337646484,
+      "learning_rate": 3.592814371257485e-07,
+      "loss": 0.0504,
+      "step": 10660
+    },
+    {
+      "epoch": 4.91,
+      "grad_norm": 1.0652536153793335,
+      "learning_rate": 3.4085674804237685e-07,
+      "loss": 0.0028,
+      "step": 10670
+    },
+    {
+      "epoch": 4.92,
+      "grad_norm": 8.1002836227417,
+      "learning_rate": 3.2243205895900505e-07,
+      "loss": 0.094,
+      "step": 10680
+    },
+    {
+      "epoch": 4.92,
+      "grad_norm": 0.4647037088871002,
+      "learning_rate": 3.0400736987563335e-07,
+      "loss": 0.0179,
+      "step": 10690
+    },
+    {
+      "epoch": 4.93,
+      "grad_norm": 17.33069610595703,
+      "learning_rate": 2.8558268079226165e-07,
+      "loss": 0.06,
+      "step": 10700
+    },
+    {
+      "epoch": 4.93,
+      "grad_norm": 0.17706802487373352,
+      "learning_rate": 2.6715799170888996e-07,
+      "loss": 0.1556,
+      "step": 10710
+    },
+    {
+      "epoch": 4.94,
+      "grad_norm": 2.1443045139312744,
+      "learning_rate": 2.487333026255182e-07,
+      "loss": 0.0408,
+      "step": 10720
+    },
+    {
+      "epoch": 4.94,
+      "grad_norm": 14.372916221618652,
+      "learning_rate": 2.3030861354214648e-07,
+      "loss": 0.0528,
+      "step": 10730
+    },
+    {
+      "epoch": 4.95,
+      "grad_norm": 18.052717208862305,
+      "learning_rate": 2.1188392445877476e-07,
+      "loss": 0.0293,
+      "step": 10740
+    },
+    {
+      "epoch": 4.95,
+      "grad_norm": 0.1709592044353485,
+      "learning_rate": 1.9345923537540306e-07,
+      "loss": 0.0808,
+      "step": 10750
+    },
+    {
+      "epoch": 4.96,
+      "grad_norm": 1.1096876859664917,
+      "learning_rate": 1.7503454629203134e-07,
+      "loss": 0.0035,
+      "step": 10760
+    },
+    {
+      "epoch": 4.96,
+      "grad_norm": 16.798601150512695,
+      "learning_rate": 1.566098572086596e-07,
+      "loss": 0.0388,
+      "step": 10770
+    },
+    {
+      "epoch": 4.96,
+      "grad_norm": 26.02865219116211,
+      "learning_rate": 1.381851681252879e-07,
+      "loss": 0.134,
+      "step": 10780
+    },
+    {
+      "epoch": 4.97,
+      "grad_norm": 0.631191611289978,
+      "learning_rate": 1.197604790419162e-07,
+      "loss": 0.044,
+      "step": 10790
+    },
+    {
+      "epoch": 4.97,
+      "grad_norm": 0.2309337705373764,
+      "learning_rate": 1.0133578995854445e-07,
+      "loss": 0.0158,
+      "step": 10800
+    },
+    {
+      "epoch": 4.98,
+      "grad_norm": 28.822986602783203,
+      "learning_rate": 8.291110087517274e-08,
+      "loss": 0.076,
+      "step": 10810
+    },
+    {
+      "epoch": 4.98,
+      "grad_norm": 1.1942684650421143,
+      "learning_rate": 6.448641179180102e-08,
+      "loss": 0.0081,
+      "step": 10820
+    },
+    {
+      "epoch": 4.99,
+      "grad_norm": 21.137388229370117,
+      "learning_rate": 4.6061722708429296e-08,
+      "loss": 0.0414,
+      "step": 10830
+    },
+    {
+      "epoch": 4.99,
+      "grad_norm": 0.2688593864440918,
+      "learning_rate": 2.763703362505758e-08,
+      "loss": 0.1751,
+      "step": 10840
+    },
+    {
+      "epoch": 5.0,
+      "grad_norm": 0.653512716293335,
+      "learning_rate": 9.21234454168586e-09,
+      "loss": 0.0891,
+      "step": 10850
+    },
+    {
+      "epoch": 5.0,
+      "eval_loss": 1.1905220746994019,
+      "eval_runtime": 23.7372,
+      "eval_samples_per_second": 126.763,
+      "eval_steps_per_second": 25.361,
+      "step": 10855
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 10855,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 5,
+  "save_steps": 500,
+  "total_flos": 2.8362717756665856e+16,
+  "train_batch_size": 5,
+  "trial_name": null,
+  "trial_params": null
+}