diff --git "a/checkpoint-7000/trainer_state.json" "b/checkpoint-7000/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/checkpoint-7000/trainer_state.json"
@@ -0,0 +1,5033 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.3832769453842437,
+  "eval_steps": 500,
+  "global_step": 7000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": 75.8236083984375,
+      "learning_rate": 4.0000000000000003e-07,
+      "loss": 3.7148,
+      "step": 10
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 52.23726272583008,
+      "learning_rate": 9e-07,
+      "loss": 3.2721,
+      "step": 20
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 35.391998291015625,
+      "learning_rate": 1.4000000000000001e-06,
+      "loss": 2.4262,
+      "step": 30
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 15.225934028625488,
+      "learning_rate": 1.9e-06,
+      "loss": 1.8828,
+      "step": 40
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 13.488654136657715,
+      "learning_rate": 2.4000000000000003e-06,
+      "loss": 1.5398,
+      "step": 50
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 8.773133277893066,
+      "learning_rate": 2.8500000000000002e-06,
+      "loss": 1.2772,
+      "step": 60
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.736795902252197,
+      "learning_rate": 3.3500000000000005e-06,
+      "loss": 1.1159,
+      "step": 70
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.9463019371032715,
+      "learning_rate": 3.85e-06,
+      "loss": 0.9883,
+      "step": 80
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.3851969242095947,
+      "learning_rate": 4.35e-06,
+      "loss": 0.902,
+      "step": 90
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.4642181396484375,
+      "learning_rate": 4.85e-06,
+      "loss": 0.8701,
+      "step": 100
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.6514687538146973,
+      "learning_rate": 5.3500000000000004e-06,
+      "loss": 0.8548,
+      "step": 110
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.4448617994785309,
+      "learning_rate": 5.850000000000001e-06,
+      "loss": 0.8038,
+      "step": 120
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.5425733327865601,
+      "learning_rate": 6.35e-06,
+      "loss": 0.825,
+      "step": 130
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.47352057695388794,
+      "learning_rate": 6.8500000000000005e-06,
+      "loss": 0.812,
+      "step": 140
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.40401744842529297,
+      "learning_rate": 7.35e-06,
+      "loss": 0.7912,
+      "step": 150
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.3112718164920807,
+      "learning_rate": 7.850000000000001e-06,
+      "loss": 0.8057,
+      "step": 160
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.45351892709732056,
+      "learning_rate": 8.350000000000001e-06,
+      "loss": 0.7912,
+      "step": 170
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.38080641627311707,
+      "learning_rate": 8.85e-06,
+      "loss": 0.7956,
+      "step": 180
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.22949431836605072,
+      "learning_rate": 9.35e-06,
+      "loss": 0.7785,
+      "step": 190
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.8331744074821472,
+      "learning_rate": 9.85e-06,
+      "loss": 0.7769,
+      "step": 200
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.30213871598243713,
+      "learning_rate": 1.035e-05,
+      "loss": 0.7797,
+      "step": 210
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.32015591859817505,
+      "learning_rate": 1.0850000000000001e-05,
+      "loss": 0.7941,
+      "step": 220
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.22447015345096588,
+      "learning_rate": 1.1350000000000001e-05,
+      "loss": 0.8014,
+      "step": 230
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.2546006143093109,
+      "learning_rate": 1.185e-05,
+      "loss": 0.771,
+      "step": 240
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.23833389580249786,
+      "learning_rate": 1.235e-05,
+      "loss": 0.7956,
+      "step": 250
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.1831669956445694,
+      "learning_rate": 1.285e-05,
+      "loss": 0.7833,
+      "step": 260
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.19860319793224335,
+      "learning_rate": 1.3350000000000001e-05,
+      "loss": 0.7956,
+      "step": 270
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.21220380067825317,
+      "learning_rate": 1.3850000000000001e-05,
+      "loss": 0.7942,
+      "step": 280
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.26025038957595825,
+      "learning_rate": 1.435e-05,
+      "loss": 0.7877,
+      "step": 290
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.2029470056295395,
+      "learning_rate": 1.485e-05,
+      "loss": 0.7637,
+      "step": 300
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.21639183163642883,
+      "learning_rate": 1.535e-05,
+      "loss": 0.7885,
+      "step": 310
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.23670387268066406,
+      "learning_rate": 1.5850000000000002e-05,
+      "loss": 0.7558,
+      "step": 320
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.2855335772037506,
+      "learning_rate": 1.635e-05,
+      "loss": 0.7854,
+      "step": 330
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.21515759825706482,
+      "learning_rate": 1.6850000000000003e-05,
+      "loss": 0.7577,
+      "step": 340
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.18981383740901947,
+      "learning_rate": 1.7349999999999998e-05,
+      "loss": 0.7802,
+      "step": 350
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.2674611508846283,
+      "learning_rate": 1.785e-05,
+      "loss": 0.7809,
+      "step": 360
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.24821774661540985,
+      "learning_rate": 1.8350000000000002e-05,
+      "loss": 0.7965,
+      "step": 370
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.3644929528236389,
+      "learning_rate": 1.885e-05,
+      "loss": 0.7957,
+      "step": 380
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.3568972647190094,
+      "learning_rate": 1.9350000000000003e-05,
+      "loss": 0.7882,
+      "step": 390
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.3236319422721863,
+      "learning_rate": 1.985e-05,
+      "loss": 0.7841,
+      "step": 400
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.24553723633289337,
+      "learning_rate": 2.035e-05,
+      "loss": 0.7824,
+      "step": 410
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.1855400949716568,
+      "learning_rate": 2.085e-05,
+      "loss": 0.7645,
+      "step": 420
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.26212435960769653,
+      "learning_rate": 2.135e-05,
+      "loss": 0.7824,
+      "step": 430
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.3167509138584137,
+      "learning_rate": 2.1850000000000003e-05,
+      "loss": 0.7908,
+      "step": 440
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.21845023334026337,
+      "learning_rate": 2.235e-05,
+      "loss": 0.7888,
+      "step": 450
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.30318892002105713,
+      "learning_rate": 2.2850000000000003e-05,
+      "loss": 0.7682,
+      "step": 460
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.28828638792037964,
+      "learning_rate": 2.3350000000000002e-05,
+      "loss": 0.7588,
+      "step": 470
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.2422240674495697,
+      "learning_rate": 2.385e-05,
+      "loss": 0.7787,
+      "step": 480
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.3579326272010803,
+      "learning_rate": 2.435e-05,
+      "loss": 0.7932,
+      "step": 490
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.3549540042877197,
+      "learning_rate": 2.485e-05,
+      "loss": 0.7634,
+      "step": 500
+    },
+    {
+      "epoch": 0.24,
+      "eval_loss": 0.7744565606117249,
+      "eval_runtime": 43.9626,
+      "eval_samples_per_second": 45.493,
+      "eval_steps_per_second": 0.364,
+      "step": 500
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.2931281328201294,
+      "learning_rate": 2.5350000000000003e-05,
+      "loss": 0.7763,
+      "step": 510
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.21228083968162537,
+      "learning_rate": 2.585e-05,
+      "loss": 0.7748,
+      "step": 520
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.26342302560806274,
+      "learning_rate": 2.6350000000000004e-05,
+      "loss": 0.789,
+      "step": 530
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.2840302288532257,
+      "learning_rate": 2.6850000000000002e-05,
+      "loss": 0.768,
+      "step": 540
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.273771196603775,
+      "learning_rate": 2.7350000000000004e-05,
+      "loss": 0.7707,
+      "step": 550
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.6022837162017822,
+      "learning_rate": 2.7850000000000003e-05,
+      "loss": 0.7705,
+      "step": 560
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.3082919716835022,
+      "learning_rate": 2.8349999999999998e-05,
+      "loss": 0.7849,
+      "step": 570
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.25341132283210754,
+      "learning_rate": 2.885e-05,
+      "loss": 0.7658,
+      "step": 580
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.2250708043575287,
+      "learning_rate": 2.935e-05,
+      "loss": 0.756,
+      "step": 590
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.23522531986236572,
+      "learning_rate": 2.985e-05,
+      "loss": 0.7692,
+      "step": 600
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.2448827624320984,
+      "learning_rate": 3.035e-05,
+      "loss": 0.7635,
+      "step": 610
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.19717586040496826,
+      "learning_rate": 3.0850000000000004e-05,
+      "loss": 0.7548,
+      "step": 620
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.24299098551273346,
+      "learning_rate": 3.135e-05,
+      "loss": 0.7673,
+      "step": 630
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.2555953562259674,
+      "learning_rate": 3.185e-05,
+      "loss": 0.7528,
+      "step": 640
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.2102127969264984,
+      "learning_rate": 3.235e-05,
+      "loss": 0.7678,
+      "step": 650
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.21767646074295044,
+      "learning_rate": 3.2850000000000006e-05,
+      "loss": 0.7564,
+      "step": 660
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.3582068085670471,
+      "learning_rate": 3.3350000000000004e-05,
+      "loss": 0.7645,
+      "step": 670
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.27804431319236755,
+      "learning_rate": 3.385e-05,
+      "loss": 0.7645,
+      "step": 680
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.2464047521352768,
+      "learning_rate": 3.435e-05,
+      "loss": 0.7325,
+      "step": 690
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.22323842346668243,
+      "learning_rate": 3.485e-05,
+      "loss": 0.7579,
+      "step": 700
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.28329703211784363,
+      "learning_rate": 3.535e-05,
+      "loss": 0.7399,
+      "step": 710
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.2471865713596344,
+      "learning_rate": 3.585e-05,
+      "loss": 0.765,
+      "step": 720
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.314908504486084,
+      "learning_rate": 3.635e-05,
+      "loss": 0.7529,
+      "step": 730
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.21466995775699615,
+      "learning_rate": 3.685e-05,
+      "loss": 0.7365,
+      "step": 740
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.2545580565929413,
+      "learning_rate": 3.735e-05,
+      "loss": 0.7458,
+      "step": 750
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.2651059329509735,
+      "learning_rate": 3.7850000000000005e-05,
+      "loss": 0.7578,
+      "step": 760
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.2408379763364792,
+      "learning_rate": 3.8350000000000004e-05,
+      "loss": 0.7283,
+      "step": 770
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.2575875520706177,
+      "learning_rate": 3.885e-05,
+      "loss": 0.7367,
+      "step": 780
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.28182050585746765,
+      "learning_rate": 3.935e-05,
+      "loss": 0.7784,
+      "step": 790
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.20579397678375244,
+      "learning_rate": 3.9850000000000006e-05,
+      "loss": 0.7533,
+      "step": 800
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.1906532198190689,
+      "learning_rate": 4.0350000000000005e-05,
+      "loss": 0.7544,
+      "step": 810
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.18311142921447754,
+      "learning_rate": 4.085e-05,
+      "loss": 0.75,
+      "step": 820
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.1910839080810547,
+      "learning_rate": 4.135e-05,
+      "loss": 0.7538,
+      "step": 830
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.2178005427122116,
+      "learning_rate": 4.185e-05,
+      "loss": 0.7456,
+      "step": 840
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.27642378211021423,
+      "learning_rate": 4.235e-05,
+      "loss": 0.744,
+      "step": 850
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.26685312390327454,
+      "learning_rate": 4.285e-05,
+      "loss": 0.7611,
+      "step": 860
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.2021845579147339,
+      "learning_rate": 4.335e-05,
+      "loss": 0.758,
+      "step": 870
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.20516808331012726,
+      "learning_rate": 4.385e-05,
+      "loss": 0.7404,
+      "step": 880
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.20619668066501617,
+      "learning_rate": 4.435e-05,
+      "loss": 0.7379,
+      "step": 890
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.22439967095851898,
+      "learning_rate": 4.4850000000000006e-05,
+      "loss": 0.7495,
+      "step": 900
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.2192295491695404,
+      "learning_rate": 4.5350000000000005e-05,
+      "loss": 0.7377,
+      "step": 910
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.26503369212150574,
+      "learning_rate": 4.585e-05,
+      "loss": 0.7481,
+      "step": 920
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.2249353975057602,
+      "learning_rate": 4.635e-05,
+      "loss": 0.742,
+      "step": 930
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.21413160860538483,
+      "learning_rate": 4.685000000000001e-05,
+      "loss": 0.7485,
+      "step": 940
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.20971478521823883,
+      "learning_rate": 4.735e-05,
+      "loss": 0.7339,
+      "step": 950
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.20295798778533936,
+      "learning_rate": 4.785e-05,
+      "loss": 0.722,
+      "step": 960
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.1957596242427826,
+      "learning_rate": 4.835e-05,
+      "loss": 0.7538,
+      "step": 970
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.18014560639858246,
+      "learning_rate": 4.885e-05,
+      "loss": 0.7251,
+      "step": 980
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.19334861636161804,
+      "learning_rate": 4.935e-05,
+      "loss": 0.7444,
+      "step": 990
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.1753440499305725,
+      "learning_rate": 4.9850000000000006e-05,
+      "loss": 0.7376,
+      "step": 1000
+    },
+    {
+      "epoch": 0.48,
+      "eval_loss": 0.73988938331604,
+      "eval_runtime": 42.9535,
+      "eval_samples_per_second": 46.562,
+      "eval_steps_per_second": 0.372,
+      "step": 1000
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.24455028772354126,
+      "learning_rate": 4.99999307775404e-05,
+      "loss": 0.7404,
+      "step": 1010
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.2149570733308792,
+      "learning_rate": 4.999959172968145e-05,
+      "loss": 0.7342,
+      "step": 1020
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.1931283175945282,
+      "learning_rate": 4.999897014592085e-05,
+      "loss": 0.735,
+      "step": 1030
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.2331906110048294,
+      "learning_rate": 4.999806603328352e-05,
+      "loss": 0.7145,
+      "step": 1040
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.21177324652671814,
+      "learning_rate": 4.999687940198738e-05,
+      "loss": 0.7308,
+      "step": 1050
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.2026621401309967,
+      "learning_rate": 4.9995410265443286e-05,
+      "loss": 0.7445,
+      "step": 1060
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.19811251759529114,
+      "learning_rate": 4.999365864025487e-05,
+      "loss": 0.738,
+      "step": 1070
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.1981566846370697,
+      "learning_rate": 4.9991624546218334e-05,
+      "loss": 0.7286,
+      "step": 1080
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.17122291028499603,
+      "learning_rate": 4.9989308006322235e-05,
+      "loss": 0.737,
+      "step": 1090
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.2182038575410843,
+      "learning_rate": 4.9986709046747225e-05,
+      "loss": 0.7354,
+      "step": 1100
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.1823730617761612,
+      "learning_rate": 4.9983827696865764e-05,
+      "loss": 0.725,
+      "step": 1110
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.2195768803358078,
+      "learning_rate": 4.998066398924177e-05,
+      "loss": 0.7173,
+      "step": 1120
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.18349726498126984,
+      "learning_rate": 4.997721795963028e-05,
+      "loss": 0.7159,
+      "step": 1130
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.2440386265516281,
+      "learning_rate": 4.997348964697699e-05,
+      "loss": 0.7171,
+      "step": 1140
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.20371629297733307,
+      "learning_rate": 4.9969479093417894e-05,
+      "loss": 0.7179,
+      "step": 1150
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.2982828617095947,
+      "learning_rate": 4.9965186344278746e-05,
+      "loss": 0.7222,
+      "step": 1160
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.2097097784280777,
+      "learning_rate": 4.9960611448074555e-05,
+      "loss": 0.7174,
+      "step": 1170
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.2264368236064911,
+      "learning_rate": 4.995575445650907e-05,
+      "loss": 0.7079,
+      "step": 1180
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.24737605452537537,
+      "learning_rate": 4.995061542447418e-05,
+      "loss": 0.7272,
+      "step": 1190
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.19530241191387177,
+      "learning_rate": 4.9945194410049254e-05,
+      "loss": 0.7216,
+      "step": 1200
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.2791181802749634,
+      "learning_rate": 4.993949147450055e-05,
+      "loss": 0.712,
+      "step": 1210
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.2095661610364914,
+      "learning_rate": 4.993350668228049e-05,
+      "loss": 0.7067,
+      "step": 1220
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.17129307985305786,
+      "learning_rate": 4.9927240101026915e-05,
+      "loss": 0.7087,
+      "step": 1230
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.2154540866613388,
+      "learning_rate": 4.992069180156235e-05,
+      "loss": 0.7411,
+      "step": 1240
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.22642803192138672,
+      "learning_rate": 4.991386185789319e-05,
+      "loss": 0.7266,
+      "step": 1250
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.19676081836223602,
+      "learning_rate": 4.9906750347208866e-05,
+      "loss": 0.7012,
+      "step": 1260
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.2095811814069748,
+      "learning_rate": 4.989935734988098e-05,
+      "loss": 0.7162,
+      "step": 1270
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.21455398201942444,
+      "learning_rate": 4.9891682949462385e-05,
+      "loss": 0.7177,
+      "step": 1280
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.20205985009670258,
+      "learning_rate": 4.988372723268623e-05,
+      "loss": 0.718,
+      "step": 1290
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.18780440092086792,
+      "learning_rate": 4.987549028946502e-05,
+      "loss": 0.7103,
+      "step": 1300
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.20362792909145355,
+      "learning_rate": 4.986697221288956e-05,
+      "loss": 0.7304,
+      "step": 1310
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.2820493280887604,
+      "learning_rate": 4.985817309922792e-05,
+      "loss": 0.7182,
+      "step": 1320
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.20787598192691803,
+      "learning_rate": 4.984909304792435e-05,
+      "loss": 0.7142,
+      "step": 1330
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.21624509990215302,
+      "learning_rate": 4.983973216159813e-05,
+      "loss": 0.7208,
+      "step": 1340
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.19343602657318115,
+      "learning_rate": 4.983009054604246e-05,
+      "loss": 0.7097,
+      "step": 1350
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.254402756690979,
+      "learning_rate": 4.9820168310223215e-05,
+      "loss": 0.7193,
+      "step": 1360
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.1940561830997467,
+      "learning_rate": 4.980996556627774e-05,
+      "loss": 0.6988,
+      "step": 1370
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.22593462467193604,
+      "learning_rate": 4.9799482429513576e-05,
+      "loss": 0.7054,
+      "step": 1380
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.22020135819911957,
+      "learning_rate": 4.978871901840716e-05,
+      "loss": 0.7221,
+      "step": 1390
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.22807644307613373,
+      "learning_rate": 4.977767545460248e-05,
+      "loss": 0.7213,
+      "step": 1400
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.1925562173128128,
+      "learning_rate": 4.9766351862909703e-05,
+      "loss": 0.7127,
+      "step": 1410
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.20845136046409607,
+      "learning_rate": 4.9754748371303775e-05,
+      "loss": 0.7207,
+      "step": 1420
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.18583528697490692,
+      "learning_rate": 4.974286511092294e-05,
+      "loss": 0.7098,
+      "step": 1430
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.2019319236278534,
+      "learning_rate": 4.973070221606732e-05,
+      "loss": 0.7029,
+      "step": 1440
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.1966458261013031,
+      "learning_rate": 4.971825982419731e-05,
+      "loss": 0.7241,
+      "step": 1450
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.20924623310565948,
+      "learning_rate": 4.970553807593214e-05,
+      "loss": 0.7126,
+      "step": 1460
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.20655803382396698,
+      "learning_rate": 4.969253711504814e-05,
+      "loss": 0.6903,
+      "step": 1470
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.2179209142923355,
+      "learning_rate": 4.967925708847727e-05,
+      "loss": 0.683,
+      "step": 1480
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.2198410928249359,
+      "learning_rate": 4.966569814630534e-05,
+      "loss": 0.7087,
+      "step": 1490
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.17970815300941467,
+      "learning_rate": 4.9651860441770374e-05,
+      "loss": 0.7006,
+      "step": 1500
+    },
+    {
+      "epoch": 0.72,
+      "eval_loss": 0.7137542366981506,
+      "eval_runtime": 43.0052,
+      "eval_samples_per_second": 46.506,
+      "eval_steps_per_second": 0.372,
+      "step": 1500
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.1931023746728897,
+      "learning_rate": 4.963774413126086e-05,
+      "loss": 0.7245,
+      "step": 1510
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.21812495589256287,
+      "learning_rate": 4.9623349374313973e-05,
+      "loss": 0.7119,
+      "step": 1520
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.20784322917461395,
+      "learning_rate": 4.960867633361381e-05,
+      "loss": 0.704,
+      "step": 1530
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.18697650730609894,
+      "learning_rate": 4.9593725174989507e-05,
+      "loss": 0.7129,
+      "step": 1540
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.25061744451522827,
+      "learning_rate": 4.957849606741338e-05,
+      "loss": 0.6816,
+      "step": 1550
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.1888163685798645,
+      "learning_rate": 4.956298918299903e-05,
+      "loss": 0.6905,
+      "step": 1560
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.1991235315799713,
+      "learning_rate": 4.954720469699938e-05,
+      "loss": 0.7258,
+      "step": 1570
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.18779252469539642,
+      "learning_rate": 4.953114278780471e-05,
+      "loss": 0.6918,
+      "step": 1580
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.21268677711486816,
+      "learning_rate": 4.951480363694061e-05,
+      "loss": 0.6995,
+      "step": 1590
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.20242677628993988,
+      "learning_rate": 4.949818742906597e-05,
+      "loss": 0.7132,
+      "step": 1600
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.2314455509185791,
+      "learning_rate": 4.9481294351970874e-05,
+      "loss": 0.6866,
+      "step": 1610
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.19638539850711823,
+      "learning_rate": 4.9464124596574465e-05,
+      "loss": 0.6993,
+      "step": 1620
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.185603529214859,
+      "learning_rate": 4.944667835692281e-05,
+      "loss": 0.6971,
+      "step": 1630
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.1594298928976059,
+      "learning_rate": 4.942895583018669e-05,
+      "loss": 0.7258,
+      "step": 1640
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.21844029426574707,
+      "learning_rate": 4.9410957216659374e-05,
+      "loss": 0.7022,
+      "step": 1650
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.19146642088890076,
+      "learning_rate": 4.9392682719754364e-05,
+      "loss": 0.6954,
+      "step": 1660
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.20112347602844238,
+      "learning_rate": 4.9374132546003096e-05,
+      "loss": 0.7017,
+      "step": 1670
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.22593769431114197,
+      "learning_rate": 4.935530690505259e-05,
+      "loss": 0.7397,
+      "step": 1680
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.2611504793167114,
+      "learning_rate": 4.933620600966311e-05,
+      "loss": 0.7048,
+      "step": 1690
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.23995471000671387,
+      "learning_rate": 4.931683007570571e-05,
+      "loss": 0.709,
+      "step": 1700
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.22278735041618347,
+      "learning_rate": 4.9297179322159856e-05,
+      "loss": 0.7116,
+      "step": 1710
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.2096502035856247,
+      "learning_rate": 4.927725397111093e-05,
+      "loss": 0.7061,
+      "step": 1720
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.22335517406463623,
+      "learning_rate": 4.9257054247747644e-05,
+      "loss": 0.6982,
+      "step": 1730
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.18597687780857086,
+      "learning_rate": 4.923658038035965e-05,
+      "loss": 0.6927,
+      "step": 1740
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.1971946507692337,
+      "learning_rate": 4.9215832600334816e-05,
+      "loss": 0.6768,
+      "step": 1750
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.2101408839225769,
+      "learning_rate": 4.9194811142156674e-05,
+      "loss": 0.6933,
+      "step": 1760
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.17352430522441864,
+      "learning_rate": 4.917351624340178e-05,
+      "loss": 0.6918,
+      "step": 1770
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.18568973243236542,
+      "learning_rate": 4.915194814473699e-05,
+      "loss": 0.6711,
+      "step": 1780
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.22991472482681274,
+      "learning_rate": 4.913010708991679e-05,
+      "loss": 0.6938,
+      "step": 1790
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.201872780919075,
+      "learning_rate": 4.910799332578048e-05,
+      "loss": 0.7051,
+      "step": 1800
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.2069864720106125,
+      "learning_rate": 4.908560710224943e-05,
+      "loss": 0.6983,
+      "step": 1810
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.2075955718755722,
+      "learning_rate": 4.906294867232426e-05,
+      "loss": 0.6996,
+      "step": 1820
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.19524317979812622,
+      "learning_rate": 4.904001829208194e-05,
+      "loss": 0.691,
+      "step": 1830
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.1967206597328186,
+      "learning_rate": 4.901681622067293e-05,
+      "loss": 0.6964,
+      "step": 1840
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.2195650041103363,
+      "learning_rate": 4.899334272031825e-05,
+      "loss": 0.6982,
+      "step": 1850
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.23072399199008942,
+      "learning_rate": 4.8969598056306475e-05,
+      "loss": 0.6838,
+      "step": 1860
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.1950792670249939,
+      "learning_rate": 4.89455824969908e-05,
+      "loss": 0.6984,
+      "step": 1870
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.2289438396692276,
+      "learning_rate": 4.892129631378597e-05,
+      "loss": 0.7012,
+      "step": 1880
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.22316494584083557,
+      "learning_rate": 4.8896739781165215e-05,
+      "loss": 0.6904,
+      "step": 1890
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.24066634476184845,
+      "learning_rate": 4.8871913176657135e-05,
+      "loss": 0.6974,
+      "step": 1900
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.18922199308872223,
+      "learning_rate": 4.8846816780842606e-05,
+      "loss": 0.686,
+      "step": 1910
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.2425553798675537,
+      "learning_rate": 4.882145087735158e-05,
+      "loss": 0.715,
+      "step": 1920
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.2267230898141861,
+      "learning_rate": 4.879581575285988e-05,
+      "loss": 0.698,
+      "step": 1930
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.22119645774364471,
+      "learning_rate": 4.876991169708592e-05,
+      "loss": 0.6942,
+      "step": 1940
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.22854161262512207,
+      "learning_rate": 4.874373900278756e-05,
+      "loss": 0.6912,
+      "step": 1950
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.1927136778831482,
+      "learning_rate": 4.871729796575863e-05,
+      "loss": 0.6829,
+      "step": 1960
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.2238394170999527,
+      "learning_rate": 4.869058888482571e-05,
+      "loss": 0.6831,
+      "step": 1970
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.17643775045871735,
+      "learning_rate": 4.86636120618447e-05,
+      "loss": 0.6875,
+      "step": 1980
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.2161882072687149,
+      "learning_rate": 4.863636780169742e-05,
+      "loss": 0.6893,
+      "step": 1990
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.20689886808395386,
+      "learning_rate": 4.860885641228815e-05,
+      "loss": 0.6721,
+      "step": 2000
+    },
+    {
+      "epoch": 0.97,
+      "eval_loss": 0.701475203037262,
+      "eval_runtime": 43.016,
+      "eval_samples_per_second": 46.494,
+      "eval_steps_per_second": 0.372,
+      "step": 2000
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.18866655230522156,
+      "learning_rate": 4.8581078204540196e-05,
+      "loss": 0.6941,
+      "step": 2010
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.20633092522621155,
+      "learning_rate": 4.855303349239231e-05,
+      "loss": 0.6943,
+      "step": 2020
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.2335944026708603,
+      "learning_rate": 4.85247225927952e-05,
+      "loss": 0.6986,
+      "step": 2030
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.18950659036636353,
+      "learning_rate": 4.8496145825707905e-05,
+      "loss": 0.6894,
+      "step": 2040
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.22117875516414642,
+      "learning_rate": 4.8467303514094204e-05,
+      "loss": 0.6956,
+      "step": 2050
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.23478396236896515,
+      "learning_rate": 4.8438195983918967e-05,
+      "loss": 0.6919,
+      "step": 2060
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.24754443764686584,
+      "learning_rate": 4.8408823564144454e-05,
+      "loss": 0.6904,
+      "step": 2070
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 0.19141903519630432,
+      "learning_rate": 4.837918658672661e-05,
+      "loss": 0.6914,
+      "step": 2080
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 0.21387813985347748,
+      "learning_rate": 4.834928538661131e-05,
+      "loss": 0.6965,
+      "step": 2090
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 0.20495550334453583,
+      "learning_rate": 4.831912030173058e-05,
+      "loss": 0.6949,
+      "step": 2100
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.35174649953842163,
+      "learning_rate": 4.8288691672998765e-05,
+      "loss": 0.6813,
+      "step": 2110
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.2461618334054947,
+      "learning_rate": 4.825799984430867e-05,
+      "loss": 0.6687,
+      "step": 2120
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 0.20039792358875275,
+      "learning_rate": 4.8227045162527714e-05,
+      "loss": 0.6737,
+      "step": 2130
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 0.19774436950683594,
+      "learning_rate": 4.8195827977493955e-05,
+      "loss": 0.6815,
+      "step": 2140
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.21472251415252686,
+      "learning_rate": 4.8164348642012194e-05,
+      "loss": 0.6957,
+      "step": 2150
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.1990274041891098,
+      "learning_rate": 4.813260751184992e-05,
+      "loss": 0.6916,
+      "step": 2160
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 0.20387563109397888,
+      "learning_rate": 4.810060494573335e-05,
+      "loss": 0.6814,
+      "step": 2170
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 0.1792985051870346,
+      "learning_rate": 4.806834130534336e-05,
+      "loss": 0.698,
+      "step": 2180
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 0.22255714237689972,
+      "learning_rate": 4.803581695531134e-05,
+      "loss": 0.6765,
+      "step": 2190
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 0.19549421966075897,
+      "learning_rate": 4.8003032263215185e-05,
+      "loss": 0.6867,
+      "step": 2200
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.1929677277803421,
+      "learning_rate": 4.796998759957504e-05,
+      "loss": 0.6935,
+      "step": 2210
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.22705228626728058,
+      "learning_rate": 4.793668333784915e-05,
+      "loss": 0.6897,
+      "step": 2220
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.21798522770404816,
+      "learning_rate": 4.790311985442966e-05,
+      "loss": 0.6694,
+      "step": 2230
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.19095788896083832,
+      "learning_rate": 4.7869297528638315e-05,
+      "loss": 0.6699,
+      "step": 2240
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 0.23002374172210693,
+      "learning_rate": 4.7835216742722225e-05,
+      "loss": 0.6807,
+      "step": 2250
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 0.2107260823249817,
+      "learning_rate": 4.780087788184947e-05,
+      "loss": 0.6855,
+      "step": 2260
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 0.23853424191474915,
+      "learning_rate": 4.776628133410487e-05,
+      "loss": 0.6946,
+      "step": 2270
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 0.25372380018234253,
+      "learning_rate": 4.7731427490485455e-05,
+      "loss": 0.6724,
+      "step": 2280
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 0.182255819439888,
+      "learning_rate": 4.7696316744896146e-05,
+      "loss": 0.6976,
+      "step": 2290
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 0.20137862861156464,
+      "learning_rate": 4.7660949494145276e-05,
+      "loss": 0.6853,
+      "step": 2300
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 0.19899927079677582,
+      "learning_rate": 4.7625326137940106e-05,
+      "loss": 0.6696,
+      "step": 2310
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 0.24147948622703552,
+      "learning_rate": 4.758944707888228e-05,
+      "loss": 0.6615,
+      "step": 2320
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 0.19519510865211487,
+      "learning_rate": 4.7553312722463305e-05,
+      "loss": 0.685,
+      "step": 2330
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 0.20128916203975677,
+      "learning_rate": 4.751692347705998e-05,
+      "loss": 0.6829,
+      "step": 2340
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 0.16112156212329865,
+      "learning_rate": 4.748027975392976e-05,
+      "loss": 0.6714,
+      "step": 2350
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 0.2659178674221039,
+      "learning_rate": 4.744338196720608e-05,
+      "loss": 0.6834,
+      "step": 2360
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 0.20477250218391418,
+      "learning_rate": 4.740623053389374e-05,
+      "loss": 0.6664,
+      "step": 2370
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 0.18069259822368622,
+      "learning_rate": 4.7368825873864154e-05,
+      "loss": 0.669,
+      "step": 2380
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.2146090418100357,
+      "learning_rate": 4.733116840985058e-05,
+      "loss": 0.6836,
+      "step": 2390
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.20656266808509827,
+      "learning_rate": 4.729325856744341e-05,
+      "loss": 0.6818,
+      "step": 2400
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.1764717549085617,
+      "learning_rate": 4.725509677508528e-05,
+      "loss": 0.7012,
+      "step": 2410
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 0.19059453904628754,
+      "learning_rate": 4.721668346406631e-05,
+      "loss": 0.6885,
+      "step": 2420
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 0.1867195963859558,
+      "learning_rate": 4.7178019068519165e-05,
+      "loss": 0.6704,
+      "step": 2430
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 0.19844096899032593,
+      "learning_rate": 4.713910402541416e-05,
+      "loss": 0.6746,
+      "step": 2440
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 0.21426193416118622,
+      "learning_rate": 4.709993877455436e-05,
+      "loss": 0.6627,
+      "step": 2450
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 0.19326262176036835,
+      "learning_rate": 4.706052375857058e-05,
+      "loss": 0.6652,
+      "step": 2460
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 0.229305237531662,
+      "learning_rate": 4.7020859422916365e-05,
+      "loss": 0.6746,
+      "step": 2470
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 0.20060321688652039,
+      "learning_rate": 4.698094621586299e-05,
+      "loss": 0.6658,
+      "step": 2480
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 0.21455462276935577,
+      "learning_rate": 4.694078458849438e-05,
+      "loss": 0.6966,
+      "step": 2490
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 0.18379837274551392,
+      "learning_rate": 4.690037499470202e-05,
+      "loss": 0.6753,
+      "step": 2500
+    },
+    {
+      "epoch": 1.21,
+      "eval_loss": 0.6940956711769104,
+      "eval_runtime": 42.7889,
+      "eval_samples_per_second": 46.741,
+      "eval_steps_per_second": 0.374,
+      "step": 2500
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 0.19967205822467804,
+      "learning_rate": 4.68597178911798e-05,
+      "loss": 0.697,
+      "step": 2510
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 0.1688542366027832,
+      "learning_rate": 4.681881373741888e-05,
+      "loss": 0.681,
+      "step": 2520
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 0.19124309718608856,
+      "learning_rate": 4.67776629957025e-05,
+      "loss": 0.6855,
+      "step": 2530
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 0.22720417380332947,
+      "learning_rate": 4.6736266131100706e-05,
+      "loss": 0.6747,
+      "step": 2540
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 0.21285779774188995,
+      "learning_rate": 4.6694623611465185e-05,
+      "loss": 0.687,
+      "step": 2550
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 0.1932476907968521,
+      "learning_rate": 4.6652735907423886e-05,
+      "loss": 0.6745,
+      "step": 2560
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 0.2050684541463852,
+      "learning_rate": 4.661060349237574e-05,
+      "loss": 0.6865,
+      "step": 2570
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 0.20403894782066345,
+      "learning_rate": 4.656822684248533e-05,
+      "loss": 0.6897,
+      "step": 2580
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 0.20080254971981049,
+      "learning_rate": 4.652560643667747e-05,
+      "loss": 0.6768,
+      "step": 2590
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 0.20147019624710083,
+      "learning_rate": 4.648274275663183e-05,
+      "loss": 0.6911,
+      "step": 2600
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 0.17825543880462646,
+      "learning_rate": 4.643963628677743e-05,
+      "loss": 0.6635,
+      "step": 2610
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 0.20095126330852509,
+      "learning_rate": 4.6396287514287275e-05,
+      "loss": 0.664,
+      "step": 2620
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 0.20084308087825775,
+      "learning_rate": 4.6352696929072727e-05,
+      "loss": 0.6653,
+      "step": 2630
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.21178288757801056,
+      "learning_rate": 4.630886502377805e-05,
+      "loss": 0.6981,
+      "step": 2640
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.1963459551334381,
+      "learning_rate": 4.62647922937748e-05,
+      "loss": 0.6666,
+      "step": 2650
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 0.21702295541763306,
+      "learning_rate": 4.6220479237156254e-05,
+      "loss": 0.6868,
+      "step": 2660
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 0.20558403432369232,
+      "learning_rate": 4.6175926354731785e-05,
+      "loss": 0.6794,
+      "step": 2670
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 0.20398521423339844,
+      "learning_rate": 4.613113415002115e-05,
+      "loss": 0.69,
+      "step": 2680
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 0.2166917473077774,
+      "learning_rate": 4.6086103129248846e-05,
+      "loss": 0.6604,
+      "step": 2690
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 0.2121194452047348,
+      "learning_rate": 4.604083380133841e-05,
+      "loss": 0.6653,
+      "step": 2700
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 0.252469927072525,
+      "learning_rate": 4.5995326677906605e-05,
+      "loss": 0.6678,
+      "step": 2710
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 0.1864558607339859,
+      "learning_rate": 4.5949582273257656e-05,
+      "loss": 0.6806,
+      "step": 2720
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 0.19438904523849487,
+      "learning_rate": 4.59036011043775e-05,
+      "loss": 0.6832,
+      "step": 2730
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 0.1954311728477478,
+      "learning_rate": 4.5857383690927844e-05,
+      "loss": 0.6772,
+      "step": 2740
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 0.38669708371162415,
+      "learning_rate": 4.5810930555240364e-05,
+      "loss": 0.6663,
+      "step": 2750
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 0.5850951075553894,
+      "learning_rate": 4.576424222231078e-05,
+      "loss": 0.6578,
+      "step": 2760
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 0.22347845137119293,
+      "learning_rate": 4.571731921979292e-05,
+      "loss": 0.6735,
+      "step": 2770
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 0.22810417413711548,
+      "learning_rate": 4.567016207799276e-05,
+      "loss": 0.6628,
+      "step": 2780
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 0.18823856115341187,
+      "learning_rate": 4.562277132986241e-05,
+      "loss": 0.6803,
+      "step": 2790
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 0.219554141163826,
+      "learning_rate": 4.557514751099415e-05,
+      "loss": 0.6701,
+      "step": 2800
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 0.1830976903438568,
+      "learning_rate": 4.55272911596143e-05,
+      "loss": 0.6864,
+      "step": 2810
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 0.18995565176010132,
+      "learning_rate": 4.5479202816577195e-05,
+      "loss": 0.6683,
+      "step": 2820
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 0.19047056138515472,
+      "learning_rate": 4.543088302535903e-05,
+      "loss": 0.6855,
+      "step": 2830
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 0.18707235157489777,
+      "learning_rate": 4.538233233205177e-05,
+      "loss": 0.6786,
+      "step": 2840
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 0.19992856681346893,
+      "learning_rate": 4.533355128535693e-05,
+      "loss": 0.6676,
+      "step": 2850
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 0.20344853401184082,
+      "learning_rate": 4.5284540436579395e-05,
+      "loss": 0.6938,
+      "step": 2860
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 0.1867518275976181,
+      "learning_rate": 4.5235300339621164e-05,
+      "loss": 0.6736,
+      "step": 2870
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 0.23014211654663086,
+      "learning_rate": 4.518583155097517e-05,
+      "loss": 0.674,
+      "step": 2880
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 0.1905904859304428,
+      "learning_rate": 4.5136134629718855e-05,
+      "loss": 0.6704,
+      "step": 2890
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 0.2236865758895874,
+      "learning_rate": 4.5086210137508e-05,
+      "loss": 0.6633,
+      "step": 2900
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 0.19701044261455536,
+      "learning_rate": 4.5036058638570264e-05,
+      "loss": 0.6626,
+      "step": 2910
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 0.19464708864688873,
+      "learning_rate": 4.4985680699698855e-05,
+      "loss": 0.6595,
+      "step": 2920
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 0.19034789502620697,
+      "learning_rate": 4.493507689024614e-05,
+      "loss": 0.6688,
+      "step": 2930
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 0.19750766456127167,
+      "learning_rate": 4.488424778211717e-05,
+      "loss": 0.6521,
+      "step": 2940
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 0.20544354617595673,
+      "learning_rate": 4.483319394976325e-05,
+      "loss": 0.6947,
+      "step": 2950
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 0.22409361600875854,
+      "learning_rate": 4.478191597017541e-05,
+      "loss": 0.674,
+      "step": 2960
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 0.19409336149692535,
+      "learning_rate": 4.473041442287793e-05,
+      "loss": 0.6978,
+      "step": 2970
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 0.20303764939308167,
+      "learning_rate": 4.4678689889921755e-05,
+      "loss": 0.6771,
+      "step": 2980
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 0.193495512008667,
+      "learning_rate": 4.462674295587794e-05,
+      "loss": 0.6693,
+      "step": 2990
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 0.18171487748622894,
+      "learning_rate": 4.457457420783103e-05,
+      "loss": 0.6716,
+      "step": 3000
+    },
+    {
+      "epoch": 1.45,
+      "eval_loss": 0.6894034147262573,
+      "eval_runtime": 43.0285,
+      "eval_samples_per_second": 46.481,
+      "eval_steps_per_second": 0.372,
+      "step": 3000
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 0.22247907519340515,
+      "learning_rate": 4.452218423537241e-05,
+      "loss": 0.6753,
+      "step": 3010
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 0.2469698041677475,
+      "learning_rate": 4.4469573630593686e-05,
+      "loss": 0.6715,
+      "step": 3020
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 0.18804265558719635,
+      "learning_rate": 4.4416742988079945e-05,
+      "loss": 0.6546,
+      "step": 3030
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 0.19240662455558777,
+      "learning_rate": 4.436369290490307e-05,
+      "loss": 0.6857,
+      "step": 3040
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 0.22342869639396667,
+      "learning_rate": 4.431042398061499e-05,
+      "loss": 0.6938,
+      "step": 3050
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 0.20553608238697052,
+      "learning_rate": 4.425693681724086e-05,
+      "loss": 0.6687,
+      "step": 3060
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 0.19881366193294525,
+      "learning_rate": 4.420323201927231e-05,
+      "loss": 0.6644,
+      "step": 3070
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 0.18151846528053284,
+      "learning_rate": 4.41493101936606e-05,
+      "loss": 0.6667,
+      "step": 3080
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 0.1906978040933609,
+      "learning_rate": 4.409517194980974e-05,
+      "loss": 0.6578,
+      "step": 3090
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 0.18123480677604675,
+      "learning_rate": 4.4040817899569644e-05,
+      "loss": 0.681,
+      "step": 3100
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 0.18974031507968903,
+      "learning_rate": 4.3986248657229134e-05,
+      "loss": 0.6645,
+      "step": 3110
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 0.2028564214706421,
+      "learning_rate": 4.3931464839509105e-05,
+      "loss": 0.6913,
+      "step": 3120
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 0.17814776301383972,
+      "learning_rate": 4.387646706555548e-05,
+      "loss": 0.6949,
+      "step": 3130
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 0.21407385170459747,
+      "learning_rate": 4.382125595693224e-05,
+      "loss": 0.6656,
+      "step": 3140
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 0.21102961897850037,
+      "learning_rate": 4.376583213761438e-05,
+      "loss": 0.6612,
+      "step": 3150
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 0.19947603344917297,
+      "learning_rate": 4.371019623398088e-05,
+      "loss": 0.6894,
+      "step": 3160
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 0.20861291885375977,
+      "learning_rate": 4.365434887480763e-05,
+      "loss": 0.6739,
+      "step": 3170
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 0.2098263055086136,
+      "learning_rate": 4.359829069126028e-05,
+      "loss": 0.6739,
+      "step": 3180
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 0.19079270958900452,
+      "learning_rate": 4.3542022316887166e-05,
+      "loss": 0.6739,
+      "step": 3190
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 0.21104255318641663,
+      "learning_rate": 4.34855443876121e-05,
+      "loss": 0.652,
+      "step": 3200
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 0.19791875779628754,
+      "learning_rate": 4.342885754172721e-05,
+      "loss": 0.6551,
+      "step": 3210
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 0.22481437027454376,
+      "learning_rate": 4.337196241988573e-05,
+      "loss": 0.6649,
+      "step": 3220
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 0.19483722746372223,
+      "learning_rate": 4.3314859665094745e-05,
+      "loss": 0.6762,
+      "step": 3230
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 0.18499700725078583,
+      "learning_rate": 4.3257549922707926e-05,
+      "loss": 0.6754,
+      "step": 3240
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 0.21884921193122864,
+      "learning_rate": 4.320003384041823e-05,
+      "loss": 0.6757,
+      "step": 3250
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 0.20222313702106476,
+      "learning_rate": 4.314231206825061e-05,
+      "loss": 0.6527,
+      "step": 3260
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 0.18140584230422974,
+      "learning_rate": 4.3084385258554635e-05,
+      "loss": 0.6731,
+      "step": 3270
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 0.21039867401123047,
+      "learning_rate": 4.302625406599713e-05,
+      "loss": 0.6722,
+      "step": 3280
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 0.18162797391414642,
+      "learning_rate": 4.296791914755478e-05,
+      "loss": 0.6563,
+      "step": 3290
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 0.178864985704422,
+      "learning_rate": 4.29093811625067e-05,
+      "loss": 0.6666,
+      "step": 3300
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 0.21808697283267975,
+      "learning_rate": 4.285064077242699e-05,
+      "loss": 0.6689,
+      "step": 3310
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 0.20632699131965637,
+      "learning_rate": 4.279169864117727e-05,
+      "loss": 0.6664,
+      "step": 3320
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 0.20482878386974335,
+      "learning_rate": 4.273255543489912e-05,
+      "loss": 0.6407,
+      "step": 3330
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 0.18613804876804352,
+      "learning_rate": 4.267321182200664e-05,
+      "loss": 0.666,
+      "step": 3340
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 0.23219124972820282,
+      "learning_rate": 4.2613668473178836e-05,
+      "loss": 0.6802,
+      "step": 3350
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 0.17973752319812775,
+      "learning_rate": 4.255392606135202e-05,
+      "loss": 0.6571,
+      "step": 3360
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 0.21650458872318268,
+      "learning_rate": 4.2493985261712285e-05,
+      "loss": 0.6727,
+      "step": 3370
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 0.19476266205310822,
+      "learning_rate": 4.2433846751687815e-05,
+      "loss": 0.6836,
+      "step": 3380
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 0.20223549008369446,
+      "learning_rate": 4.237351121094121e-05,
+      "loss": 0.6666,
+      "step": 3390
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 0.2078579217195511,
+      "learning_rate": 4.231297932136189e-05,
+      "loss": 0.673,
+      "step": 3400
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 0.193430095911026,
+      "learning_rate": 4.225225176705829e-05,
+      "loss": 0.6718,
+      "step": 3410
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 0.18017272651195526,
+      "learning_rate": 4.2191329234350194e-05,
+      "loss": 0.6806,
+      "step": 3420
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 0.20289075374603271,
+      "learning_rate": 4.213021241176096e-05,
+      "loss": 0.667,
+      "step": 3430
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 0.2014080286026001,
+      "learning_rate": 4.2068901990009726e-05,
+      "loss": 0.6441,
+      "step": 3440
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 0.17385190725326538,
+      "learning_rate": 4.200739866200363e-05,
+      "loss": 0.6564,
+      "step": 3450
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 0.20427776873111725,
+      "learning_rate": 4.1945703122829924e-05,
+      "loss": 0.6509,
+      "step": 3460
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 0.2093549370765686,
+      "learning_rate": 4.1883816069748214e-05,
+      "loss": 0.6741,
+      "step": 3470
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 0.21946479380130768,
+      "learning_rate": 4.182173820218249e-05,
+      "loss": 0.6778,
+      "step": 3480
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 0.2244766503572464,
+      "learning_rate": 4.175947022171326e-05,
+      "loss": 0.681,
+      "step": 3490
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 0.22939437627792358,
+      "learning_rate": 4.169701283206961e-05,
+      "loss": 0.6595,
+      "step": 3500
+    },
+    {
+      "epoch": 1.69,
+      "eval_loss": 0.6864724159240723,
+      "eval_runtime": 43.0148,
+      "eval_samples_per_second": 46.496,
+      "eval_steps_per_second": 0.372,
+      "step": 3500
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 0.2838324010372162,
+      "learning_rate": 4.163436673912127e-05,
+      "loss": 0.666,
+      "step": 3510
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 0.1862134337425232,
+      "learning_rate": 4.1571532650870614e-05,
+      "loss": 0.6671,
+      "step": 3520
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 0.2170250117778778,
+      "learning_rate": 4.1508511277444674e-05,
+      "loss": 0.6566,
+      "step": 3530
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 0.17831502854824066,
+      "learning_rate": 4.1445303331087096e-05,
+      "loss": 0.6776,
+      "step": 3540
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 0.17921501398086548,
+      "learning_rate": 4.138190952615012e-05,
+      "loss": 0.6876,
+      "step": 3550
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 0.19673947989940643,
+      "learning_rate": 4.131833057908648e-05,
+      "loss": 0.6596,
+      "step": 3560
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 0.1829831749200821,
+      "learning_rate": 4.1254567208441296e-05,
+      "loss": 0.6712,
+      "step": 3570
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 0.3023044764995575,
+      "learning_rate": 4.119062013484402e-05,
+      "loss": 0.6747,
+      "step": 3580
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 0.21133002638816833,
+      "learning_rate": 4.11264900810002e-05,
+      "loss": 0.674,
+      "step": 3590
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 0.19868247210979462,
+      "learning_rate": 4.106217777168339e-05,
+      "loss": 0.68,
+      "step": 3600
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 0.21643079817295074,
+      "learning_rate": 4.0997683933726894e-05,
+      "loss": 0.6555,
+      "step": 3610
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 0.1953793615102768,
+      "learning_rate": 4.093300929601562e-05,
+      "loss": 0.6704,
+      "step": 3620
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 0.21795640885829926,
+      "learning_rate": 4.086815458947777e-05,
+      "loss": 0.6583,
+      "step": 3630
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 0.22441154718399048,
+      "learning_rate": 4.080312054707665e-05,
+      "loss": 0.6772,
+      "step": 3640
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 0.20651431381702423,
+      "learning_rate": 4.073790790380232e-05,
+      "loss": 0.6793,
+      "step": 3650
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 0.21724598109722137,
+      "learning_rate": 4.0672517396663354e-05,
+      "loss": 0.6742,
+      "step": 3660
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 0.27277424931526184,
+      "learning_rate": 4.0606949764678434e-05,
+      "loss": 0.6495,
+      "step": 3670
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 0.21489045023918152,
+      "learning_rate": 4.054120574886808e-05,
+      "loss": 0.6516,
+      "step": 3680
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 0.19597062468528748,
+      "learning_rate": 4.0475286092246215e-05,
+      "loss": 0.6498,
+      "step": 3690
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 0.20840667188167572,
+      "learning_rate": 4.0409191539811774e-05,
+      "loss": 0.6394,
+      "step": 3700
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 0.23823177814483643,
+      "learning_rate": 4.0342922838540334e-05,
+      "loss": 0.6798,
+      "step": 3710
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 0.20505908131599426,
+      "learning_rate": 4.0276480737375606e-05,
+      "loss": 0.6607,
+      "step": 3720
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 0.28554779291152954,
+      "learning_rate": 4.0209865987221014e-05,
+      "loss": 0.6467,
+      "step": 3730
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 0.24168336391448975,
+      "learning_rate": 4.014307934093119e-05,
+      "loss": 0.6625,
+      "step": 3740
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 0.21519850194454193,
+      "learning_rate": 4.007612155330348e-05,
+      "loss": 0.6431,
+      "step": 3750
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 0.1954280287027359,
+      "learning_rate": 4.000899338106939e-05,
+      "loss": 0.6669,
+      "step": 3760
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 0.22829805314540863,
+      "learning_rate": 3.9941695582886065e-05,
+      "loss": 0.6507,
+      "step": 3770
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 0.30064237117767334,
+      "learning_rate": 3.9874228919327685e-05,
+      "loss": 0.664,
+      "step": 3780
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 0.21391743421554565,
+      "learning_rate": 3.980659415287689e-05,
+      "loss": 0.6577,
+      "step": 3790
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 0.237900048494339,
+      "learning_rate": 3.9738792047916143e-05,
+      "loss": 0.6678,
+      "step": 3800
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 0.2035730481147766,
+      "learning_rate": 3.967082337071911e-05,
+      "loss": 0.6822,
+      "step": 3810
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 0.25883620977401733,
+      "learning_rate": 3.960268888944202e-05,
+      "loss": 0.6657,
+      "step": 3820
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 0.21861082315444946,
+      "learning_rate": 3.9534389374114905e-05,
+      "loss": 0.6608,
+      "step": 3830
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 0.2185363620519638,
+      "learning_rate": 3.946592559663299e-05,
+      "loss": 0.6763,
+      "step": 3840
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 0.2056983858346939,
+      "learning_rate": 3.9397298330747905e-05,
+      "loss": 0.6703,
+      "step": 3850
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 0.19903923571109772,
+      "learning_rate": 3.932850835205899e-05,
+      "loss": 0.6887,
+      "step": 3860
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 0.19114772975444794,
+      "learning_rate": 3.9259556438004476e-05,
+      "loss": 0.6563,
+      "step": 3870
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 0.20320531725883484,
+      "learning_rate": 3.9190443367852736e-05,
+      "loss": 0.6471,
+      "step": 3880
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 0.23587201535701752,
+      "learning_rate": 3.912116992269348e-05,
+      "loss": 0.6705,
+      "step": 3890
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 0.19107288122177124,
+      "learning_rate": 3.9051736885428886e-05,
+      "loss": 0.6696,
+      "step": 3900
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 0.20316600799560547,
+      "learning_rate": 3.898214504076482e-05,
+      "loss": 0.6724,
+      "step": 3910
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 0.2201535403728485,
+      "learning_rate": 3.89123951752019e-05,
+      "loss": 0.658,
+      "step": 3920
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 0.20604528486728668,
+      "learning_rate": 3.884248807702665e-05,
+      "loss": 0.6746,
+      "step": 3930
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 0.24204471707344055,
+      "learning_rate": 3.8772424536302564e-05,
+      "loss": 0.6737,
+      "step": 3940
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 0.20363375544548035,
+      "learning_rate": 3.870220534486121e-05,
+      "loss": 0.669,
+      "step": 3950
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 0.22277967631816864,
+      "learning_rate": 3.8631831296293246e-05,
+      "loss": 0.6676,
+      "step": 3960
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.18391238152980804,
+      "learning_rate": 3.856130318593947e-05,
+      "loss": 0.6651,
+      "step": 3970
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.185078427195549,
+      "learning_rate": 3.849062181088183e-05,
+      "loss": 0.6432,
+      "step": 3980
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 0.2277977466583252,
+      "learning_rate": 3.841978796993442e-05,
+      "loss": 0.6518,
+      "step": 3990
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 0.22232377529144287,
+      "learning_rate": 3.834880246363443e-05,
+      "loss": 0.6743,
+      "step": 4000
+    },
+    {
+      "epoch": 1.93,
+      "eval_loss": 0.6847629547119141,
+      "eval_runtime": 42.7658,
+      "eval_samples_per_second": 46.766,
+      "eval_steps_per_second": 0.374,
+      "step": 4000
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 0.20745177567005157,
+      "learning_rate": 3.8277666094233115e-05,
+      "loss": 0.6787,
+      "step": 4010
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 0.21468785405158997,
+      "learning_rate": 3.820637966568675e-05,
+      "loss": 0.6603,
+      "step": 4020
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 0.18202567100524902,
+      "learning_rate": 3.81349439836475e-05,
+      "loss": 0.6487,
+      "step": 4030
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 0.21424496173858643,
+      "learning_rate": 3.806335985545434e-05,
+      "loss": 0.6674,
+      "step": 4040
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 0.19061969220638275,
+      "learning_rate": 3.7991628090123933e-05,
+      "loss": 0.651,
+      "step": 4050
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 0.20192697644233704,
+      "learning_rate": 3.7919749498341477e-05,
+      "loss": 0.661,
+      "step": 4060
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 0.1942390650510788,
+      "learning_rate": 3.784772489245155e-05,
+      "loss": 0.6688,
+      "step": 4070
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 0.19686420261859894,
+      "learning_rate": 3.777555508644893e-05,
+      "loss": 0.6492,
+      "step": 4080
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 0.2590291202068329,
+      "learning_rate": 3.770324089596937e-05,
+      "loss": 0.6597,
+      "step": 4090
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 0.208131343126297,
+      "learning_rate": 3.763078313828043e-05,
+      "loss": 0.6695,
+      "step": 4100
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 0.215104341506958,
+      "learning_rate": 3.755818263227219e-05,
+      "loss": 0.6521,
+      "step": 4110
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 0.22334003448486328,
+      "learning_rate": 3.748544019844803e-05,
+      "loss": 0.6775,
+      "step": 4120
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 0.2614120841026306,
+      "learning_rate": 3.741255665891534e-05,
+      "loss": 0.6643,
+      "step": 4130
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 0.22348462045192719,
+      "learning_rate": 3.733953283737625e-05,
+      "loss": 0.6598,
+      "step": 4140
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 0.21082797646522522,
+      "learning_rate": 3.726636955911825e-05,
+      "loss": 0.6697,
+      "step": 4150
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 0.20834925770759583,
+      "learning_rate": 3.719306765100499e-05,
+      "loss": 0.6562,
+      "step": 4160
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 0.223398357629776,
+      "learning_rate": 3.711962794146682e-05,
+      "loss": 0.6403,
+      "step": 4170
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 0.1990179866552353,
+      "learning_rate": 3.704605126049147e-05,
+      "loss": 0.6662,
+      "step": 4180
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 0.20885860919952393,
+      "learning_rate": 3.6972338439614676e-05,
+      "loss": 0.6669,
+      "step": 4190
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 0.21816882491111755,
+      "learning_rate": 3.6898490311910774e-05,
+      "loss": 0.6378,
+      "step": 4200
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 0.21286895871162415,
+      "learning_rate": 3.6824507711983294e-05,
+      "loss": 0.633,
+      "step": 4210
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 0.20113909244537354,
+      "learning_rate": 3.6750391475955506e-05,
+      "loss": 0.6561,
+      "step": 4220
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 0.1805492788553238,
+      "learning_rate": 3.6676142441461e-05,
+      "loss": 0.6478,
+      "step": 4230
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 0.198676198720932,
+      "learning_rate": 3.66017614476342e-05,
+      "loss": 0.6477,
+      "step": 4240
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 0.18248610198497772,
+      "learning_rate": 3.6527249335100856e-05,
+      "loss": 0.6476,
+      "step": 4250
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 0.20229819416999817,
+      "learning_rate": 3.645260694596862e-05,
+      "loss": 0.6675,
+      "step": 4260
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 0.24611905217170715,
+      "learning_rate": 3.637783512381745e-05,
+      "loss": 0.6546,
+      "step": 4270
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 0.1821327954530716,
+      "learning_rate": 3.6302934713690114e-05,
+      "loss": 0.6482,
+      "step": 4280
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 0.19002540409564972,
+      "learning_rate": 3.622790656208263e-05,
+      "loss": 0.6468,
+      "step": 4290
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.20899777114391327,
+      "learning_rate": 3.615275151693471e-05,
+      "loss": 0.6649,
+      "step": 4300
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.1918490082025528,
+      "learning_rate": 3.607747042762016e-05,
+      "loss": 0.6554,
+      "step": 4310
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 0.2790839970111847,
+      "learning_rate": 3.600206414493728e-05,
+      "loss": 0.6607,
+      "step": 4320
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 0.21416689455509186,
+      "learning_rate": 3.592653352109929e-05,
+      "loss": 0.6537,
+      "step": 4330
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 0.19238290190696716,
+      "learning_rate": 3.5850879409724624e-05,
+      "loss": 0.6566,
+      "step": 4340
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 0.24673143029212952,
+      "learning_rate": 3.577510266582737e-05,
+      "loss": 0.6599,
+      "step": 4350
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 0.1863541454076767,
+      "learning_rate": 3.569920414580754e-05,
+      "loss": 0.6593,
+      "step": 4360
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 0.2196236401796341,
+      "learning_rate": 3.56231847074414e-05,
+      "loss": 0.6483,
+      "step": 4370
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 0.18300732970237732,
+      "learning_rate": 3.5547045209871806e-05,
+      "loss": 0.6359,
+      "step": 4380
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 0.18719661235809326,
+      "learning_rate": 3.5470786513598476e-05,
+      "loss": 0.6707,
+      "step": 4390
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 0.1748167872428894,
+      "learning_rate": 3.539440948046827e-05,
+      "loss": 0.6516,
+      "step": 4400
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 0.21993610262870789,
+      "learning_rate": 3.531791497366543e-05,
+      "loss": 0.6539,
+      "step": 4410
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 0.20704199373722076,
+      "learning_rate": 3.524130385770186e-05,
+      "loss": 0.6734,
+      "step": 4420
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 0.23367135226726532,
+      "learning_rate": 3.516457699840733e-05,
+      "loss": 0.6855,
+      "step": 4430
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 0.1824340671300888,
+      "learning_rate": 3.50877352629197e-05,
+      "loss": 0.6575,
+      "step": 4440
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 0.22375932335853577,
+      "learning_rate": 3.50107795196751e-05,
+      "loss": 0.6487,
+      "step": 4450
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 0.23603610694408417,
+      "learning_rate": 3.4933710638398156e-05,
+      "loss": 0.6505,
+      "step": 4460
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 0.22663302719593048,
+      "learning_rate": 3.485652949009212e-05,
+      "loss": 0.6569,
+      "step": 4470
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 0.21095499396324158,
+      "learning_rate": 3.4779236947029055e-05,
+      "loss": 0.6657,
+      "step": 4480
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 0.2422657608985901,
+      "learning_rate": 3.470183388273995e-05,
+      "loss": 0.645,
+      "step": 4490
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 0.19883093237876892,
+      "learning_rate": 3.462432117200489e-05,
+      "loss": 0.6647,
+      "step": 4500
+    },
+    {
+      "epoch": 2.17,
+      "eval_loss": 0.6818509697914124,
+      "eval_runtime": 42.9789,
+      "eval_samples_per_second": 46.534,
+      "eval_steps_per_second": 0.372,
+      "step": 4500
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 0.23546861112117767,
+      "learning_rate": 3.454669969084312e-05,
+      "loss": 0.6594,
+      "step": 4510
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 0.20284776389598846,
+      "learning_rate": 3.446897031650316e-05,
+      "loss": 0.65,
+      "step": 4520
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 0.19946908950805664,
+      "learning_rate": 3.4391133927452925e-05,
+      "loss": 0.6573,
+      "step": 4530
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 0.2393037974834442,
+      "learning_rate": 3.431319140336975e-05,
+      "loss": 0.6502,
+      "step": 4540
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 0.2206243872642517,
+      "learning_rate": 3.423514362513048e-05,
+      "loss": 0.6704,
+      "step": 4550
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 0.20536164939403534,
+      "learning_rate": 3.415699147480149e-05,
+      "loss": 0.6629,
+      "step": 4560
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 0.22283408045768738,
+      "learning_rate": 3.407873583562873e-05,
+      "loss": 0.6789,
+      "step": 4570
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 0.21526776254177094,
+      "learning_rate": 3.4000377592027754e-05,
+      "loss": 0.6608,
+      "step": 4580
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 0.2566728889942169,
+      "learning_rate": 3.3921917629573695e-05,
+      "loss": 0.6624,
+      "step": 4590
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 0.21391907334327698,
+      "learning_rate": 3.384335683499129e-05,
+      "loss": 0.6536,
+      "step": 4600
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 0.21451455354690552,
+      "learning_rate": 3.376469609614484e-05,
+      "loss": 0.648,
+      "step": 4610
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 0.19541354477405548,
+      "learning_rate": 3.368593630202818e-05,
+      "loss": 0.657,
+      "step": 4620
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 0.2338065356016159,
+      "learning_rate": 3.360707834275459e-05,
+      "loss": 0.6686,
+      "step": 4630
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 0.1928481012582779,
+      "learning_rate": 3.3528123109546844e-05,
+      "loss": 0.6496,
+      "step": 4640
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 0.23110058903694153,
+      "learning_rate": 3.344907149472702e-05,
+      "loss": 0.6556,
+      "step": 4650
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 0.22316649556159973,
+      "learning_rate": 3.336992439170649e-05,
+      "loss": 0.663,
+      "step": 4660
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 0.1999579668045044,
+      "learning_rate": 3.3290682694975775e-05,
+      "loss": 0.6436,
+      "step": 4670
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 0.276755154132843,
+      "learning_rate": 3.321134730009446e-05,
+      "loss": 0.6355,
+      "step": 4680
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 0.24972842633724213,
+      "learning_rate": 3.313191910368111e-05,
+      "loss": 0.6454,
+      "step": 4690
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 0.2169172465801239,
+      "learning_rate": 3.3052399003403046e-05,
+      "loss": 0.6419,
+      "step": 4700
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 0.20602068305015564,
+      "learning_rate": 3.297278789796629e-05,
+      "loss": 0.6459,
+      "step": 4710
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 0.25669389963150024,
+      "learning_rate": 3.289308668710532e-05,
+      "loss": 0.6675,
+      "step": 4720
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 0.21008487045764923,
+      "learning_rate": 3.281329627157302e-05,
+      "loss": 0.642,
+      "step": 4730
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 0.21287119388580322,
+      "learning_rate": 3.2733417553130384e-05,
+      "loss": 0.6348,
+      "step": 4740
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 0.21862877905368805,
+      "learning_rate": 3.2653451434536394e-05,
+      "loss": 0.6579,
+      "step": 4750
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 0.19740484654903412,
+      "learning_rate": 3.2573398819537767e-05,
+      "loss": 0.6428,
+      "step": 4760
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 0.2695826292037964,
+      "learning_rate": 3.2493260612858803e-05,
+      "loss": 0.658,
+      "step": 4770
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 0.19435593485832214,
+      "learning_rate": 3.2413037720191096e-05,
+      "loss": 0.6615,
+      "step": 4780
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 0.21996192634105682,
+      "learning_rate": 3.233273104818337e-05,
+      "loss": 0.6588,
+      "step": 4790
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 0.26496437191963196,
+      "learning_rate": 3.225234150443114e-05,
+      "loss": 0.6548,
+      "step": 4800
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 0.2110251486301422,
+      "learning_rate": 3.217186999746654e-05,
+      "loss": 0.6618,
+      "step": 4810
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 0.19939462840557098,
+      "learning_rate": 3.209131743674803e-05,
+      "loss": 0.6531,
+      "step": 4820
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 0.19813968241214752,
+      "learning_rate": 3.201068473265007e-05,
+      "loss": 0.6484,
+      "step": 4830
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 0.2121964991092682,
+      "learning_rate": 3.192997279645291e-05,
+      "loss": 0.6556,
+      "step": 4840
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 0.19998233020305634,
+      "learning_rate": 3.1849182540332214e-05,
+      "loss": 0.6554,
+      "step": 4850
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 0.23478324711322784,
+      "learning_rate": 3.176831487734882e-05,
+      "loss": 0.637,
+      "step": 4860
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 0.19482719898223877,
+      "learning_rate": 3.168737072143838e-05,
+      "loss": 0.6464,
+      "step": 4870
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 0.21184729039669037,
+      "learning_rate": 3.160635098740103e-05,
+      "loss": 0.6584,
+      "step": 4880
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 0.27235257625579834,
+      "learning_rate": 3.152525659089106e-05,
+      "loss": 0.6597,
+      "step": 4890
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 0.22343479096889496,
+      "learning_rate": 3.1444088448406584e-05,
+      "loss": 0.6673,
+      "step": 4900
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 0.19164462387561798,
+      "learning_rate": 3.136284747727916e-05,
+      "loss": 0.6617,
+      "step": 4910
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 0.21662816405296326,
+      "learning_rate": 3.128153459566341e-05,
+      "loss": 0.6505,
+      "step": 4920
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 0.23168237507343292,
+      "learning_rate": 3.1200150722526697e-05,
+      "loss": 0.6479,
+      "step": 4930
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 0.19140580296516418,
+      "learning_rate": 3.111869677763865e-05,
+      "loss": 0.6413,
+      "step": 4940
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 0.25583407282829285,
+      "learning_rate": 3.1037173681560874e-05,
+      "loss": 0.6443,
+      "step": 4950
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 0.1979406476020813,
+      "learning_rate": 3.095558235563647e-05,
+      "loss": 0.6668,
+      "step": 4960
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 0.21006809175014496,
+      "learning_rate": 3.0873923721979645e-05,
+      "loss": 0.6505,
+      "step": 4970
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 0.2352473884820938,
+      "learning_rate": 3.07921987034653e-05,
+      "loss": 0.6552,
+      "step": 4980
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 0.22581614553928375,
+      "learning_rate": 3.07104082237186e-05,
+      "loss": 0.6489,
+      "step": 4990
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 0.21772272884845734,
+      "learning_rate": 3.06285532071045e-05,
+      "loss": 0.6721,
+      "step": 5000
+    },
+    {
+      "epoch": 2.42,
+      "eval_loss": 0.6797133684158325,
+      "eval_runtime": 43.0653,
+      "eval_samples_per_second": 46.441,
+      "eval_steps_per_second": 0.372,
+      "step": 5000
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 0.20226365327835083,
+      "learning_rate": 3.054663457871736e-05,
+      "loss": 0.6723,
+      "step": 5010
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 0.2123214453458786,
+      "learning_rate": 3.046465326437043e-05,
+      "loss": 0.6479,
+      "step": 5020
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 0.21145904064178467,
+      "learning_rate": 3.0382610190585435e-05,
+      "loss": 0.6508,
+      "step": 5030
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 0.2079765647649765,
+      "learning_rate": 3.030050628458206e-05,
+      "loss": 0.6494,
+      "step": 5040
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 0.24468085169792175,
+      "learning_rate": 3.0218342474267513e-05,
+      "loss": 0.6628,
+      "step": 5050
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 0.21137215197086334,
+      "learning_rate": 3.0136119688225996e-05,
+      "loss": 0.6463,
+      "step": 5060
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 0.20148161053657532,
+      "learning_rate": 3.0053838855708243e-05,
+      "loss": 0.6635,
+      "step": 5070
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 0.21683721244335175,
+      "learning_rate": 2.9971500906621027e-05,
+      "loss": 0.6585,
+      "step": 5080
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 0.1892833113670349,
+      "learning_rate": 2.988910677151659e-05,
+      "loss": 0.6621,
+      "step": 5090
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 0.3029578924179077,
+      "learning_rate": 2.980665738158221e-05,
+      "loss": 0.6742,
+      "step": 5100
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 0.2016330510377884,
+      "learning_rate": 2.972415366862959e-05,
+      "loss": 0.6686,
+      "step": 5110
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 0.20674502849578857,
+      "learning_rate": 2.964159656508441e-05,
+      "loss": 0.6577,
+      "step": 5120
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 0.20945307612419128,
+      "learning_rate": 2.9558987003975736e-05,
+      "loss": 0.6673,
+      "step": 5130
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 0.2278815060853958,
+      "learning_rate": 2.9476325918925485e-05,
+      "loss": 0.6585,
+      "step": 5140
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 0.21968629956245422,
+      "learning_rate": 2.9393614244137875e-05,
+      "loss": 0.6547,
+      "step": 5150
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 0.22052909433841705,
+      "learning_rate": 2.9310852914388875e-05,
+      "loss": 0.6706,
+      "step": 5160
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 0.24963468313217163,
+      "learning_rate": 2.9228042865015647e-05,
+      "loss": 0.635,
+      "step": 5170
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 0.21189726889133453,
+      "learning_rate": 2.914518503190595e-05,
+      "loss": 0.6529,
+      "step": 5180
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 0.19712771475315094,
+      "learning_rate": 2.9062280351487587e-05,
+      "loss": 0.6589,
+      "step": 5190
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 0.23704534769058228,
+      "learning_rate": 2.8979329760717788e-05,
+      "loss": 0.6459,
+      "step": 5200
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 0.3048954904079437,
+      "learning_rate": 2.8896334197072667e-05,
+      "loss": 0.6468,
+      "step": 5210
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 0.30929285287857056,
+      "learning_rate": 2.8813294598536606e-05,
+      "loss": 0.651,
+      "step": 5220
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 0.2547236382961273,
+      "learning_rate": 2.8730211903591636e-05,
+      "loss": 0.6374,
+      "step": 5230
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 0.23137128353118896,
+      "learning_rate": 2.8647087051206862e-05,
+      "loss": 0.6486,
+      "step": 5240
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 0.21269945800304413,
+      "learning_rate": 2.856392098082783e-05,
+      "loss": 0.6751,
+      "step": 5250
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 0.19768312573432922,
+      "learning_rate": 2.8480714632365906e-05,
+      "loss": 0.6511,
+      "step": 5260
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 0.2293408066034317,
+      "learning_rate": 2.83974689461877e-05,
+      "loss": 0.6337,
+      "step": 5270
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 0.22362858057022095,
+      "learning_rate": 2.8314184863104347e-05,
+      "loss": 0.6449,
+      "step": 5280
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 0.22398819029331207,
+      "learning_rate": 2.8230863324360977e-05,
+      "loss": 0.6437,
+      "step": 5290
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 0.22831834852695465,
+      "learning_rate": 2.8147505271626002e-05,
+      "loss": 0.6493,
+      "step": 5300
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 0.2500370740890503,
+      "learning_rate": 2.8064111646980524e-05,
+      "loss": 0.6451,
+      "step": 5310
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 0.21840785443782806,
+      "learning_rate": 2.7980683392907632e-05,
+      "loss": 0.6558,
+      "step": 5320
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 0.21673984825611115,
+      "learning_rate": 2.7897221452281813e-05,
+      "loss": 0.6493,
+      "step": 5330
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 0.22077977657318115,
+      "learning_rate": 2.7813726768358263e-05,
+      "loss": 0.6443,
+      "step": 5340
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 0.22384098172187805,
+      "learning_rate": 2.7730200284762215e-05,
+      "loss": 0.6525,
+      "step": 5350
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 0.2188693881034851,
+      "learning_rate": 2.7646642945478314e-05,
+      "loss": 0.649,
+      "step": 5360
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 0.21913747489452362,
+      "learning_rate": 2.7563055694839884e-05,
+      "loss": 0.6543,
+      "step": 5370
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 0.20839491486549377,
+      "learning_rate": 2.7479439477518344e-05,
+      "loss": 0.6462,
+      "step": 5380
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 0.21764138340950012,
+      "learning_rate": 2.7395795238512446e-05,
+      "loss": 0.6527,
+      "step": 5390
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 0.21299828588962555,
+      "learning_rate": 2.7312123923137667e-05,
+      "loss": 0.638,
+      "step": 5400
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 0.20629416406154633,
+      "learning_rate": 2.7228426477015447e-05,
+      "loss": 0.6592,
+      "step": 5410
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 0.23245516419410706,
+      "learning_rate": 2.714470384606258e-05,
+      "loss": 0.6424,
+      "step": 5420
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 0.2076595425605774,
+      "learning_rate": 2.706095697648048e-05,
+      "loss": 0.6335,
+      "step": 5430
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 0.19132575392723083,
+      "learning_rate": 2.6977186814744503e-05,
+      "loss": 0.649,
+      "step": 5440
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 0.21425828337669373,
+      "learning_rate": 2.6893394307593228e-05,
+      "loss": 0.6485,
+      "step": 5450
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 0.24599085748195648,
+      "learning_rate": 2.680958040201778e-05,
+      "loss": 0.6719,
+      "step": 5460
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 0.2181704044342041,
+      "learning_rate": 2.6725746045251126e-05,
+      "loss": 0.6403,
+      "step": 5470
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 0.2156282663345337,
+      "learning_rate": 2.6641892184757365e-05,
+      "loss": 0.6414,
+      "step": 5480
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 0.210015669465065,
+      "learning_rate": 2.655801976822102e-05,
+      "loss": 0.6616,
+      "step": 5490
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 0.22476404905319214,
+      "learning_rate": 2.6474129743536323e-05,
+      "loss": 0.6642,
+      "step": 5500
+    },
+    {
+      "epoch": 2.66,
+      "eval_loss": 0.6780422329902649,
+      "eval_runtime": 42.8508,
+      "eval_samples_per_second": 46.674,
+      "eval_steps_per_second": 0.373,
+      "step": 5500
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 0.2253371626138687,
+      "learning_rate": 2.6390223058796497e-05,
+      "loss": 0.6704,
+      "step": 5510
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 0.2072548270225525,
+      "learning_rate": 2.6306300662283073e-05,
+      "loss": 0.6319,
+      "step": 5520
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 0.1958586722612381,
+      "learning_rate": 2.6222363502455133e-05,
+      "loss": 0.6462,
+      "step": 5530
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 0.19193071126937866,
+      "learning_rate": 2.6138412527938617e-05,
+      "loss": 0.6408,
+      "step": 5540
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 0.235664963722229,
+      "learning_rate": 2.605444868751558e-05,
+      "loss": 0.6585,
+      "step": 5550
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 0.27757498621940613,
+      "learning_rate": 2.5970472930113515e-05,
+      "loss": 0.6437,
+      "step": 5560
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 0.19725674390792847,
+      "learning_rate": 2.588648620479455e-05,
+      "loss": 0.6533,
+      "step": 5570
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 0.20764590799808502,
+      "learning_rate": 2.5802489460744817e-05,
+      "loss": 0.6473,
+      "step": 5580
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 0.2171824425458908,
+      "learning_rate": 2.571848364726363e-05,
+      "loss": 0.6344,
+      "step": 5590
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 0.20881757140159607,
+      "learning_rate": 2.563446971375283e-05,
+      "loss": 0.6554,
+      "step": 5600
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 0.2502697706222534,
+      "learning_rate": 2.5550448609706035e-05,
+      "loss": 0.6767,
+      "step": 5610
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 0.26100224256515503,
+      "learning_rate": 2.546642128469787e-05,
+      "loss": 0.6651,
+      "step": 5620
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 0.22215646505355835,
+      "learning_rate": 2.5382388688373288e-05,
+      "loss": 0.6371,
+      "step": 5630
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 0.2387877255678177,
+      "learning_rate": 2.529835177043682e-05,
+      "loss": 0.6422,
+      "step": 5640
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 0.23950129747390747,
+      "learning_rate": 2.5214311480641823e-05,
+      "loss": 0.6602,
+      "step": 5650
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 0.2006569504737854,
+      "learning_rate": 2.513026876877978e-05,
+      "loss": 0.6513,
+      "step": 5660
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 0.1996818333864212,
+      "learning_rate": 2.5046224584669537e-05,
+      "loss": 0.6509,
+      "step": 5670
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 0.21665816009044647,
+      "learning_rate": 2.496217987814656e-05,
+      "loss": 0.6698,
+      "step": 5680
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 0.24070510268211365,
+      "learning_rate": 2.4878135599052265e-05,
+      "loss": 0.6668,
+      "step": 5690
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 0.19807234406471252,
+      "learning_rate": 2.47940926972232e-05,
+      "loss": 0.6494,
+      "step": 5700
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 0.2191707193851471,
+      "learning_rate": 2.4710052122480345e-05,
+      "loss": 0.6356,
+      "step": 5710
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 0.19569790363311768,
+      "learning_rate": 2.4626014824618415e-05,
+      "loss": 0.6587,
+      "step": 5720
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 0.2184823751449585,
+      "learning_rate": 2.4541981753395045e-05,
+      "loss": 0.6644,
+      "step": 5730
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 0.22695264220237732,
+      "learning_rate": 2.445795385852015e-05,
+      "loss": 0.6327,
+      "step": 5740
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 0.20581942796707153,
+      "learning_rate": 2.4373932089645117e-05,
+      "loss": 0.6506,
+      "step": 5750
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 0.21353425085544586,
+      "learning_rate": 2.428991739635208e-05,
+      "loss": 0.6466,
+      "step": 5760
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 0.252079039812088,
+      "learning_rate": 2.420591072814326e-05,
+      "loss": 0.6559,
+      "step": 5770
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 0.19642478227615356,
+      "learning_rate": 2.4121913034430123e-05,
+      "loss": 0.6441,
+      "step": 5780
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 0.19868357479572296,
+      "learning_rate": 2.4037925264522766e-05,
+      "loss": 0.6424,
+      "step": 5790
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 0.20668402314186096,
+      "learning_rate": 2.3953948367619085e-05,
+      "loss": 0.6698,
+      "step": 5800
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 0.21798357367515564,
+      "learning_rate": 2.38699832927941e-05,
+      "loss": 0.6346,
+      "step": 5810
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 0.23406192660331726,
+      "learning_rate": 2.3786030988989257e-05,
+      "loss": 0.6535,
+      "step": 5820
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 0.20458374917507172,
+      "learning_rate": 2.3702092405001627e-05,
+      "loss": 0.6426,
+      "step": 5830
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 0.2875809073448181,
+      "learning_rate": 2.3618168489473258e-05,
+      "loss": 0.6344,
+      "step": 5840
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 0.21651776134967804,
+      "learning_rate": 2.3534260190880396e-05,
+      "loss": 0.6403,
+      "step": 5850
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 0.22318010032176971,
+      "learning_rate": 2.3450368457522787e-05,
+      "loss": 0.6662,
+      "step": 5860
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 0.20754455029964447,
+      "learning_rate": 2.3366494237513e-05,
+      "loss": 0.64,
+      "step": 5870
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 0.20914232730865479,
+      "learning_rate": 2.3282638478765634e-05,
+      "loss": 0.6374,
+      "step": 5880
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 0.22443871200084686,
+      "learning_rate": 2.3198802128986673e-05,
+      "loss": 0.6576,
+      "step": 5890
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 0.2213885337114334,
+      "learning_rate": 2.3114986135662728e-05,
+      "loss": 0.6434,
+      "step": 5900
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 0.2547857463359833,
+      "learning_rate": 2.3031191446050348e-05,
+      "loss": 0.6519,
+      "step": 5910
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 0.2463701069355011,
+      "learning_rate": 2.2947419007165355e-05,
+      "loss": 0.642,
+      "step": 5920
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 0.24427345395088196,
+      "learning_rate": 2.286366976577205e-05,
+      "loss": 0.6661,
+      "step": 5930
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 0.20179685950279236,
+      "learning_rate": 2.2779944668372596e-05,
+      "loss": 0.6619,
+      "step": 5940
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 0.2376372367143631,
+      "learning_rate": 2.2696244661196285e-05,
+      "loss": 0.6364,
+      "step": 5950
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 0.22518636286258698,
+      "learning_rate": 2.2612570690188828e-05,
+      "loss": 0.6619,
+      "step": 5960
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 0.21611256897449493,
+      "learning_rate": 2.252892370100172e-05,
+      "loss": 0.665,
+      "step": 5970
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 0.19709035754203796,
+      "learning_rate": 2.2445304638981483e-05,
+      "loss": 0.6461,
+      "step": 5980
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 0.21152737736701965,
+      "learning_rate": 2.236171444915905e-05,
+      "loss": 0.6564,
+      "step": 5990
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 0.23283518850803375,
+      "learning_rate": 2.227815407623903e-05,
+      "loss": 0.6653,
+      "step": 6000
+    },
+    {
+      "epoch": 2.9,
+      "eval_loss": 0.6763524413108826,
+      "eval_runtime": 43.1507,
+      "eval_samples_per_second": 46.349,
+      "eval_steps_per_second": 0.371,
+      "step": 6000
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 0.24773156642913818,
+      "learning_rate": 2.2194624464589052e-05,
+      "loss": 0.6409,
+      "step": 6010
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 0.19031673669815063,
+      "learning_rate": 2.2111126558229102e-05,
+      "loss": 0.641,
+      "step": 6020
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 0.21194177865982056,
+      "learning_rate": 2.2027661300820833e-05,
+      "loss": 0.6411,
+      "step": 6030
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 0.22639860212802887,
+      "learning_rate": 2.1944229635656938e-05,
+      "loss": 0.6559,
+      "step": 6040
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 0.20527996122837067,
+      "learning_rate": 2.186083250565043e-05,
+      "loss": 0.6441,
+      "step": 6050
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 0.21662482619285583,
+      "learning_rate": 2.1777470853324045e-05,
+      "loss": 0.6517,
+      "step": 6060
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 0.21164235472679138,
+      "learning_rate": 2.1694145620799543e-05,
+      "loss": 0.634,
+      "step": 6070
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 0.20878757536411285,
+      "learning_rate": 2.161085774978709e-05,
+      "loss": 0.6676,
+      "step": 6080
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 0.19167086482048035,
+      "learning_rate": 2.1527608181574616e-05,
+      "loss": 0.6569,
+      "step": 6090
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 0.18406851589679718,
+      "learning_rate": 2.1444397857017154e-05,
+      "loss": 0.6455,
+      "step": 6100
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 0.21034060418605804,
+      "learning_rate": 2.1361227716526226e-05,
+      "loss": 0.6445,
+      "step": 6110
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 0.20496372878551483,
+      "learning_rate": 2.1278098700059212e-05,
+      "loss": 0.659,
+      "step": 6120
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 0.22188597917556763,
+      "learning_rate": 2.11950117471087e-05,
+      "loss": 0.636,
+      "step": 6130
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 0.21596181392669678,
+      "learning_rate": 2.1111967796691946e-05,
+      "loss": 0.6567,
+      "step": 6140
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 0.24058407545089722,
+      "learning_rate": 2.102896778734013e-05,
+      "loss": 0.6336,
+      "step": 6150
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 0.2177901268005371,
+      "learning_rate": 2.094601265708791e-05,
+      "loss": 0.666,
+      "step": 6160
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 0.25122541189193726,
+      "learning_rate": 2.086310334346268e-05,
+      "loss": 0.6363,
+      "step": 6170
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 0.21556386351585388,
+      "learning_rate": 2.0780240783474045e-05,
+      "loss": 0.6571,
+      "step": 6180
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 0.21746757626533508,
+      "learning_rate": 2.069742591360323e-05,
+      "loss": 0.6398,
+      "step": 6190
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 0.18603651225566864,
+      "learning_rate": 2.0614659669792467e-05,
+      "loss": 0.6534,
+      "step": 6200
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 0.2334885150194168,
+      "learning_rate": 2.053194298743446e-05,
+      "loss": 0.6707,
+      "step": 6210
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 0.24816961586475372,
+      "learning_rate": 2.0449276801361766e-05,
+      "loss": 0.6462,
+      "step": 6220
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 0.29403042793273926,
+      "learning_rate": 2.036666204583625e-05,
+      "loss": 0.6335,
+      "step": 6230
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 0.23217493295669556,
+      "learning_rate": 2.0284099654538556e-05,
+      "loss": 0.6326,
+      "step": 6240
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 0.20210884511470795,
+      "learning_rate": 2.0201590560557483e-05,
+      "loss": 0.6366,
+      "step": 6250
+    },
+    {
+      "epoch": 3.03,
+      "grad_norm": 0.22504980862140656,
+      "learning_rate": 2.0119135696379536e-05,
+      "loss": 0.65,
+      "step": 6260
+    },
+    {
+      "epoch": 3.03,
+      "grad_norm": 0.24354460835456848,
+      "learning_rate": 2.0036735993878296e-05,
+      "loss": 0.6626,
+      "step": 6270
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 0.21757520735263824,
+      "learning_rate": 1.9954392384303942e-05,
+      "loss": 0.6287,
+      "step": 6280
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 0.19983159005641937,
+      "learning_rate": 1.9872105798272712e-05,
+      "loss": 0.6422,
+      "step": 6290
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 0.2321237325668335,
+      "learning_rate": 1.9789877165756378e-05,
+      "loss": 0.6578,
+      "step": 6300
+    },
+    {
+      "epoch": 3.05,
+      "grad_norm": 0.20653708279132843,
+      "learning_rate": 1.9707707416071762e-05,
+      "loss": 0.6409,
+      "step": 6310
+    },
+    {
+      "epoch": 3.05,
+      "grad_norm": 0.2338685691356659,
+      "learning_rate": 1.96255974778702e-05,
+      "loss": 0.6484,
+      "step": 6320
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 0.24382874369621277,
+      "learning_rate": 1.9543548279127048e-05,
+      "loss": 0.6389,
+      "step": 6330
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 0.21047548949718475,
+      "learning_rate": 1.946156074713124e-05,
+      "loss": 0.6374,
+      "step": 6340
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 0.2392827272415161,
+      "learning_rate": 1.937963580847475e-05,
+      "loss": 0.6313,
+      "step": 6350
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 0.2402975857257843,
+      "learning_rate": 1.9297774389042164e-05,
+      "loss": 0.6673,
+      "step": 6360
+    },
+    {
+      "epoch": 3.08,
+      "grad_norm": 0.2071339339017868,
+      "learning_rate": 1.921597741400018e-05,
+      "loss": 0.6492,
+      "step": 6370
+    },
+    {
+      "epoch": 3.08,
+      "grad_norm": 0.22973790764808655,
+      "learning_rate": 1.9134245807787177e-05,
+      "loss": 0.6498,
+      "step": 6380
+    },
+    {
+      "epoch": 3.09,
+      "grad_norm": 0.2571215331554413,
+      "learning_rate": 1.905258049410277e-05,
+      "loss": 0.6189,
+      "step": 6390
+    },
+    {
+      "epoch": 3.09,
+      "grad_norm": 0.22487790882587433,
+      "learning_rate": 1.897098239589735e-05,
+      "loss": 0.6558,
+      "step": 6400
+    },
+    {
+      "epoch": 3.1,
+      "grad_norm": 0.2679091989994049,
+      "learning_rate": 1.888945243536167e-05,
+      "loss": 0.6487,
+      "step": 6410
+    },
+    {
+      "epoch": 3.1,
+      "grad_norm": 0.25580644607543945,
+      "learning_rate": 1.8807991533916395e-05,
+      "loss": 0.6338,
+      "step": 6420
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 0.2231510728597641,
+      "learning_rate": 1.8726600612201765e-05,
+      "loss": 0.6449,
+      "step": 6430
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 0.253778338432312,
+      "learning_rate": 1.8645280590067072e-05,
+      "loss": 0.632,
+      "step": 6440
+    },
+    {
+      "epoch": 3.12,
+      "grad_norm": 0.21836714446544647,
+      "learning_rate": 1.8564032386560367e-05,
+      "loss": 0.6444,
+      "step": 6450
+    },
+    {
+      "epoch": 3.12,
+      "grad_norm": 0.21077515184879303,
+      "learning_rate": 1.8482856919918017e-05,
+      "loss": 0.645,
+      "step": 6460
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 0.25506341457366943,
+      "learning_rate": 1.8401755107554344e-05,
+      "loss": 0.6618,
+      "step": 6470
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 0.20035551488399506,
+      "learning_rate": 1.8320727866051286e-05,
+      "loss": 0.6638,
+      "step": 6480
+    },
+    {
+      "epoch": 3.14,
+      "grad_norm": 0.21936412155628204,
+      "learning_rate": 1.823977611114797e-05,
+      "loss": 0.649,
+      "step": 6490
+    },
+    {
+      "epoch": 3.14,
+      "grad_norm": 0.21108798682689667,
+      "learning_rate": 1.8158900757730423e-05,
+      "loss": 0.643,
+      "step": 6500
+    },
+    {
+      "epoch": 3.14,
+      "eval_loss": 0.6756294965744019,
+      "eval_runtime": 43.1292,
+      "eval_samples_per_second": 46.372,
+      "eval_steps_per_second": 0.371,
+      "step": 6500
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 0.23786285519599915,
+      "learning_rate": 1.807810271982121e-05,
+      "loss": 0.6417,
+      "step": 6510
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 0.2198764979839325,
+      "learning_rate": 1.79973829105691e-05,
+      "loss": 0.6596,
+      "step": 6520
+    },
+    {
+      "epoch": 3.16,
+      "grad_norm": 0.201913520693779,
+      "learning_rate": 1.791674224223876e-05,
+      "loss": 0.6539,
+      "step": 6530
+    },
+    {
+      "epoch": 3.16,
+      "grad_norm": 0.18852323293685913,
+      "learning_rate": 1.7836181626200425e-05,
+      "loss": 0.6442,
+      "step": 6540
+    },
+    {
+      "epoch": 3.17,
+      "grad_norm": 0.2102418839931488,
+      "learning_rate": 1.7755701972919607e-05,
+      "loss": 0.6323,
+      "step": 6550
+    },
+    {
+      "epoch": 3.17,
+      "grad_norm": 0.20450511574745178,
+      "learning_rate": 1.7675304191946835e-05,
+      "loss": 0.6348,
+      "step": 6560
+    },
+    {
+      "epoch": 3.18,
+      "grad_norm": 0.21254278719425201,
+      "learning_rate": 1.7594989191907315e-05,
+      "loss": 0.6532,
+      "step": 6570
+    },
+    {
+      "epoch": 3.18,
+      "grad_norm": 0.22247746586799622,
+      "learning_rate": 1.751475788049072e-05,
+      "loss": 0.6574,
+      "step": 6580
+    },
+    {
+      "epoch": 3.19,
+      "grad_norm": 0.22997824847698212,
+      "learning_rate": 1.743461116444089e-05,
+      "loss": 0.6464,
+      "step": 6590
+    },
+    {
+      "epoch": 3.19,
+      "grad_norm": 0.22006092965602875,
+      "learning_rate": 1.7354549949545586e-05,
+      "loss": 0.625,
+      "step": 6600
+    },
+    {
+      "epoch": 3.19,
+      "grad_norm": 0.24055121839046478,
+      "learning_rate": 1.7274575140626318e-05,
+      "loss": 0.6348,
+      "step": 6610
+    },
+    {
+      "epoch": 3.2,
+      "grad_norm": 0.21221566200256348,
+      "learning_rate": 1.7194687641528013e-05,
+      "loss": 0.6571,
+      "step": 6620
+    },
+    {
+      "epoch": 3.2,
+      "grad_norm": 0.24489928781986237,
+      "learning_rate": 1.711488835510889e-05,
+      "loss": 0.6354,
+      "step": 6630
+    },
+    {
+      "epoch": 3.21,
+      "grad_norm": 0.21739131212234497,
+      "learning_rate": 1.703517818323021e-05,
+      "loss": 0.6563,
+      "step": 6640
+    },
+    {
+      "epoch": 3.21,
+      "grad_norm": 0.21341422200202942,
+      "learning_rate": 1.695555802674608e-05,
+      "loss": 0.6352,
+      "step": 6650
+    },
+    {
+      "epoch": 3.22,
+      "grad_norm": 0.24106374382972717,
+      "learning_rate": 1.687602878549333e-05,
+      "loss": 0.6569,
+      "step": 6660
+    },
+    {
+      "epoch": 3.22,
+      "grad_norm": 0.21376299858093262,
+      "learning_rate": 1.6796591358281255e-05,
+      "loss": 0.635,
+      "step": 6670
+    },
+    {
+      "epoch": 3.23,
+      "grad_norm": 0.24214418232440948,
+      "learning_rate": 1.671724664288153e-05,
+      "loss": 0.6589,
+      "step": 6680
+    },
+    {
+      "epoch": 3.23,
+      "grad_norm": 0.2144564688205719,
+      "learning_rate": 1.6637995536018023e-05,
+      "loss": 0.6443,
+      "step": 6690
+    },
+    {
+      "epoch": 3.24,
+      "grad_norm": 0.20162871479988098,
+      "learning_rate": 1.6558838933356676e-05,
+      "loss": 0.6273,
+      "step": 6700
+    },
+    {
+      "epoch": 3.24,
+      "grad_norm": 0.22555021941661835,
+      "learning_rate": 1.6479777729495387e-05,
+      "loss": 0.6482,
+      "step": 6710
+    },
+    {
+      "epoch": 3.25,
+      "grad_norm": 0.2198752909898758,
+      "learning_rate": 1.640081281795388e-05,
+      "loss": 0.6592,
+      "step": 6720
+    },
+    {
+      "epoch": 3.25,
+      "grad_norm": 0.20376479625701904,
+      "learning_rate": 1.6321945091163642e-05,
+      "loss": 0.6486,
+      "step": 6730
+    },
+    {
+      "epoch": 3.26,
+      "grad_norm": 0.20879609882831573,
+      "learning_rate": 1.62431754404578e-05,
+      "loss": 0.6352,
+      "step": 6740
+    },
+    {
+      "epoch": 3.26,
+      "grad_norm": 0.23657208681106567,
+      "learning_rate": 1.616450475606105e-05,
+      "loss": 0.6365,
+      "step": 6750
+    },
+    {
+      "epoch": 3.27,
+      "grad_norm": 0.22932332754135132,
+      "learning_rate": 1.608593392707964e-05,
+      "loss": 0.6257,
+      "step": 6760
+    },
+    {
+      "epoch": 3.27,
+      "grad_norm": 0.23238502442836761,
+      "learning_rate": 1.600746384149125e-05,
+      "loss": 0.6421,
+      "step": 6770
+    },
+    {
+      "epoch": 3.28,
+      "grad_norm": 0.22237473726272583,
+      "learning_rate": 1.5929095386135035e-05,
+      "loss": 0.646,
+      "step": 6780
+    },
+    {
+      "epoch": 3.28,
+      "grad_norm": 0.2505503296852112,
+      "learning_rate": 1.5850829446701548e-05,
+      "loss": 0.6564,
+      "step": 6790
+    },
+    {
+      "epoch": 3.29,
+      "grad_norm": 0.24202491343021393,
+      "learning_rate": 1.5772666907722732e-05,
+      "loss": 0.6293,
+      "step": 6800
+    },
+    {
+      "epoch": 3.29,
+      "grad_norm": 0.2263512760400772,
+      "learning_rate": 1.569460865256196e-05,
+      "loss": 0.6464,
+      "step": 6810
+    },
+    {
+      "epoch": 3.3,
+      "grad_norm": 0.20612984895706177,
+      "learning_rate": 1.5616655563404005e-05,
+      "loss": 0.6356,
+      "step": 6820
+    },
+    {
+      "epoch": 3.3,
+      "grad_norm": 0.2401859611272812,
+      "learning_rate": 1.5538808521245132e-05,
+      "loss": 0.6335,
+      "step": 6830
+    },
+    {
+      "epoch": 3.31,
+      "grad_norm": 0.23205476999282837,
+      "learning_rate": 1.546106840588306e-05,
+      "loss": 0.637,
+      "step": 6840
+    },
+    {
+      "epoch": 3.31,
+      "grad_norm": 0.21428649127483368,
+      "learning_rate": 1.538343609590709e-05,
+      "loss": 0.6445,
+      "step": 6850
+    },
+    {
+      "epoch": 3.32,
+      "grad_norm": 0.22657857835292816,
+      "learning_rate": 1.5305912468688132e-05,
+      "loss": 0.6505,
+      "step": 6860
+    },
+    {
+      "epoch": 3.32,
+      "grad_norm": 0.24213473498821259,
+      "learning_rate": 1.52284984003688e-05,
+      "loss": 0.621,
+      "step": 6870
+    },
+    {
+      "epoch": 3.33,
+      "grad_norm": 0.292008638381958,
+      "learning_rate": 1.5151194765853557e-05,
+      "loss": 0.6334,
+      "step": 6880
+    },
+    {
+      "epoch": 3.33,
+      "grad_norm": 0.23929265141487122,
+      "learning_rate": 1.5074002438798717e-05,
+      "loss": 0.6358,
+      "step": 6890
+    },
+    {
+      "epoch": 3.33,
+      "grad_norm": 0.2356884777545929,
+      "learning_rate": 1.4996922291602709e-05,
+      "loss": 0.6383,
+      "step": 6900
+    },
+    {
+      "epoch": 3.34,
+      "grad_norm": 0.27760398387908936,
+      "learning_rate": 1.4919955195396096e-05,
+      "loss": 0.6149,
+      "step": 6910
+    },
+    {
+      "epoch": 3.34,
+      "grad_norm": 0.24114906787872314,
+      "learning_rate": 1.4843102020031796e-05,
+      "loss": 0.6435,
+      "step": 6920
+    },
+    {
+      "epoch": 3.35,
+      "grad_norm": 0.22814461588859558,
+      "learning_rate": 1.476636363407526e-05,
+      "loss": 0.649,
+      "step": 6930
+    },
+    {
+      "epoch": 3.35,
+      "grad_norm": 0.23456543684005737,
+      "learning_rate": 1.4689740904794588e-05,
+      "loss": 0.6244,
+      "step": 6940
+    },
+    {
+      "epoch": 3.36,
+      "grad_norm": 0.24401907622814178,
+      "learning_rate": 1.4613234698150824e-05,
+      "loss": 0.6543,
+      "step": 6950
+    },
+    {
+      "epoch": 3.36,
+      "grad_norm": 0.23866045475006104,
+      "learning_rate": 1.4536845878788086e-05,
+      "loss": 0.6516,
+      "step": 6960
+    },
+    {
+      "epoch": 3.37,
+      "grad_norm": 0.2096366286277771,
+      "learning_rate": 1.4460575310023805e-05,
+      "loss": 0.6303,
+      "step": 6970
+    },
+    {
+      "epoch": 3.37,
+      "grad_norm": 0.22500860691070557,
+      "learning_rate": 1.4384423853839055e-05,
+      "loss": 0.6354,
+      "step": 6980
+    },
+    {
+      "epoch": 3.38,
+      "grad_norm": 0.20588767528533936,
+      "learning_rate": 1.4308392370868673e-05,
+      "loss": 0.6485,
+      "step": 6990
+    },
+    {
+      "epoch": 3.38,
+      "grad_norm": 0.24658453464508057,
+      "learning_rate": 1.4232481720391645e-05,
+      "loss": 0.6532,
+      "step": 7000
+    },
+    {
+      "epoch": 3.38,
+      "eval_loss": 0.6749144196510315,
+      "eval_runtime": 42.8253,
+      "eval_samples_per_second": 46.701,
+      "eval_steps_per_second": 0.374,
+      "step": 7000
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 10345,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 5,
+  "save_steps": 500,
+  "total_flos": 1.7722361512405762e+19,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}