Training in progress, step 4000, checkpoint

checkpoint-4000/config.json

@@ -0,0 +1,26 @@
+{
+  "architectures": [
+    "Olmo2ForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "eos_token_id": 100257,
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "initializer_range": 0.02,
+  "intermediate_size": 8192,
+  "max_position_embeddings": 4096,
+  "model_type": "olmo2",
+  "num_attention_heads": 16,
+  "num_hidden_layers": 16,
+  "num_key_value_heads": 16,
+  "pad_token_id": 100277,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 500000,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.51.3",
+  "use_cache": false,
+  "vocab_size": 100352
+}

checkpoint-4000/generation_config.json

@@ -0,0 +1,6 @@
+{
+  "_from_model_config": true,
+  "eos_token_id": 100257,
+  "pad_token_id": 100277,
+  "transformers_version": "4.51.3"
+}

checkpoint-4000/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c05d135790b5b155202f1e7bf8fb8acb2e36e3e36c00eef27fb6b17f0e9917af
+size 2969854224

checkpoint-4000/optimizer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:073e320017d27dfdce66781b30670f5718787528a4f716a019ecc57533c60031
+size 4500154

checkpoint-4000/rng_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9b7e369537d6998664af0e30cd3cf69e4696679f42cc29aab323e2158145457e
+size 14180

checkpoint-4000/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:80e1fce10a735babf9f4952eeada7a13e0485757ef5f8b91359536833936c83d
+size 1064

checkpoint-4000/trainer_state.json

@@ -0,0 +1,2834 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.0,
+  "eval_steps": 500,
+  "global_step": 4000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0025,
+      "grad_norm": 116.5,
+      "learning_rate": 4.5000000000000003e-07,
+      "loss": 14.3625,
+      "step": 10
+    },
+    {
+      "epoch": 0.005,
+      "grad_norm": 114.5,
+      "learning_rate": 9.500000000000001e-07,
+      "loss": 14.7063,
+      "step": 20
+    },
+    {
+      "epoch": 0.0075,
+      "grad_norm": 93.0,
+      "learning_rate": 1.45e-06,
+      "loss": 14.3812,
+      "step": 30
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 101.0,
+      "learning_rate": 1.9500000000000004e-06,
+      "loss": 14.3,
+      "step": 40
+    },
+    {
+      "epoch": 0.0125,
+      "grad_norm": 112.5,
+      "learning_rate": 2.4500000000000003e-06,
+      "loss": 14.4625,
+      "step": 50
+    },
+    {
+      "epoch": 0.015,
+      "grad_norm": 83.0,
+      "learning_rate": 2.95e-06,
+      "loss": 14.5437,
+      "step": 60
+    },
+    {
+      "epoch": 0.0175,
+      "grad_norm": 79.5,
+      "learning_rate": 3.45e-06,
+      "loss": 14.4625,
+      "step": 70
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 83.5,
+      "learning_rate": 3.95e-06,
+      "loss": 14.1188,
+      "step": 80
+    },
+    {
+      "epoch": 0.0225,
+      "grad_norm": 114.0,
+      "learning_rate": 4.450000000000001e-06,
+      "loss": 14.25,
+      "step": 90
+    },
+    {
+      "epoch": 0.025,
+      "grad_norm": 82.0,
+      "learning_rate": 4.95e-06,
+      "loss": 14.0687,
+      "step": 100
+    },
+    {
+      "epoch": 0.0275,
+      "grad_norm": 151.0,
+      "learning_rate": 5.450000000000001e-06,
+      "loss": 14.2937,
+      "step": 110
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 98.0,
+      "learning_rate": 5.950000000000001e-06,
+      "loss": 13.6813,
+      "step": 120
+    },
+    {
+      "epoch": 0.0325,
+      "grad_norm": 76.5,
+      "learning_rate": 6.450000000000001e-06,
+      "loss": 13.5437,
+      "step": 130
+    },
+    {
+      "epoch": 0.035,
+      "grad_norm": 117.5,
+      "learning_rate": 6.95e-06,
+      "loss": 13.0813,
+      "step": 140
+    },
+    {
+      "epoch": 0.0375,
+      "grad_norm": 112.5,
+      "learning_rate": 7.450000000000001e-06,
+      "loss": 12.9938,
+      "step": 150
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 112.5,
+      "learning_rate": 7.950000000000002e-06,
+      "loss": 12.5375,
+      "step": 160
+    },
+    {
+      "epoch": 0.0425,
+      "grad_norm": 123.5,
+      "learning_rate": 8.45e-06,
+      "loss": 12.375,
+      "step": 170
+    },
+    {
+      "epoch": 0.045,
+      "grad_norm": 115.0,
+      "learning_rate": 8.95e-06,
+      "loss": 11.7625,
+      "step": 180
+    },
+    {
+      "epoch": 0.0475,
+      "grad_norm": 119.5,
+      "learning_rate": 9.450000000000001e-06,
+      "loss": 11.1062,
+      "step": 190
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 124.0,
+      "learning_rate": 9.950000000000001e-06,
+      "loss": 10.9187,
+      "step": 200
+    },
+    {
+      "epoch": 0.0525,
+      "grad_norm": 187.0,
+      "learning_rate": 1.045e-05,
+      "loss": 9.9125,
+      "step": 210
+    },
+    {
+      "epoch": 0.055,
+      "grad_norm": 104.0,
+      "learning_rate": 1.095e-05,
+      "loss": 9.5,
+      "step": 220
+    },
+    {
+      "epoch": 0.0575,
+      "grad_norm": 86.5,
+      "learning_rate": 1.145e-05,
+      "loss": 8.7781,
+      "step": 230
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 116.0,
+      "learning_rate": 1.195e-05,
+      "loss": 8.3656,
+      "step": 240
+    },
+    {
+      "epoch": 0.0625,
+      "grad_norm": 154.0,
+      "learning_rate": 1.2450000000000003e-05,
+      "loss": 7.7125,
+      "step": 250
+    },
+    {
+      "epoch": 0.065,
+      "grad_norm": 191.0,
+      "learning_rate": 1.295e-05,
+      "loss": 6.9594,
+      "step": 260
+    },
+    {
+      "epoch": 0.0675,
+      "grad_norm": 95.5,
+      "learning_rate": 1.3450000000000002e-05,
+      "loss": 6.7469,
+      "step": 270
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 104.5,
+      "learning_rate": 1.3950000000000002e-05,
+      "loss": 5.9094,
+      "step": 280
+    },
+    {
+      "epoch": 0.0725,
+      "grad_norm": 62.25,
+      "learning_rate": 1.4450000000000002e-05,
+      "loss": 5.4375,
+      "step": 290
+    },
+    {
+      "epoch": 0.075,
+      "grad_norm": 61.0,
+      "learning_rate": 1.4950000000000003e-05,
+      "loss": 4.9281,
+      "step": 300
+    },
+    {
+      "epoch": 0.0775,
+      "grad_norm": 45.25,
+      "learning_rate": 1.545e-05,
+      "loss": 3.9266,
+      "step": 310
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 35.25,
+      "learning_rate": 1.595e-05,
+      "loss": 3.8203,
+      "step": 320
+    },
+    {
+      "epoch": 0.0825,
+      "grad_norm": 30.5,
+      "learning_rate": 1.645e-05,
+      "loss": 3.7531,
+      "step": 330
+    },
+    {
+      "epoch": 0.085,
+      "grad_norm": 30.5,
+      "learning_rate": 1.6950000000000002e-05,
+      "loss": 3.5484,
+      "step": 340
+    },
+    {
+      "epoch": 0.0875,
+      "grad_norm": 14.1875,
+      "learning_rate": 1.7450000000000004e-05,
+      "loss": 3.4813,
+      "step": 350
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 14.5625,
+      "learning_rate": 1.795e-05,
+      "loss": 2.9406,
+      "step": 360
+    },
+    {
+      "epoch": 0.0925,
+      "grad_norm": 15.375,
+      "learning_rate": 1.845e-05,
+      "loss": 3.0297,
+      "step": 370
+    },
+    {
+      "epoch": 0.095,
+      "grad_norm": 14.8125,
+      "learning_rate": 1.8950000000000003e-05,
+      "loss": 3.025,
+      "step": 380
+    },
+    {
+      "epoch": 0.0975,
+      "grad_norm": 26.5,
+      "learning_rate": 1.9450000000000002e-05,
+      "loss": 2.8016,
+      "step": 390
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 11.0625,
+      "learning_rate": 1.9950000000000004e-05,
+      "loss": 2.8125,
+      "step": 400
+    },
+    {
+      "epoch": 0.1025,
+      "grad_norm": 11.5,
+      "learning_rate": 1.99996915764479e-05,
+      "loss": 2.5844,
+      "step": 410
+    },
+    {
+      "epoch": 0.105,
+      "grad_norm": 12.875,
+      "learning_rate": 1.9998625445384374e-05,
+      "loss": 2.6328,
+      "step": 420
+    },
+    {
+      "epoch": 0.1075,
+      "grad_norm": 9.125,
+      "learning_rate": 1.9996797880281935e-05,
+      "loss": 2.5555,
+      "step": 430
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 11.8125,
+      "learning_rate": 1.999420902031673e-05,
+      "loss": 2.7078,
+      "step": 440
+    },
+    {
+      "epoch": 0.1125,
+      "grad_norm": 8.625,
+      "learning_rate": 1.9990859062640478e-05,
+      "loss": 2.6945,
+      "step": 450
+    },
+    {
+      "epoch": 0.115,
+      "grad_norm": 16.375,
+      "learning_rate": 1.998674826236542e-05,
+      "loss": 2.6281,
+      "step": 460
+    },
+    {
+      "epoch": 0.1175,
+      "grad_norm": 10.375,
+      "learning_rate": 1.9981876932544918e-05,
+      "loss": 2.2406,
+      "step": 470
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 8.1875,
+      "learning_rate": 1.997624544414959e-05,
+      "loss": 2.4,
+      "step": 480
+    },
+    {
+      "epoch": 0.1225,
+      "grad_norm": 10.0625,
+      "learning_rate": 1.9969854226039088e-05,
+      "loss": 2.182,
+      "step": 490
+    },
+    {
+      "epoch": 0.125,
+      "grad_norm": 10.25,
+      "learning_rate": 1.9962703764929413e-05,
+      "loss": 2.2102,
+      "step": 500
+    },
+    {
+      "epoch": 0.1275,
+      "grad_norm": 5.875,
+      "learning_rate": 1.995479460535586e-05,
+      "loss": 2.0398,
+      "step": 510
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 14.9375,
+      "learning_rate": 1.9946127349631564e-05,
+      "loss": 2.1414,
+      "step": 520
+    },
+    {
+      "epoch": 0.1325,
+      "grad_norm": 6.1875,
+      "learning_rate": 1.9936702657801586e-05,
+      "loss": 1.7063,
+      "step": 530
+    },
+    {
+      "epoch": 0.135,
+      "grad_norm": 8.25,
+      "learning_rate": 1.992652124759271e-05,
+      "loss": 1.9187,
+      "step": 540
+    },
+    {
+      "epoch": 0.1375,
+      "grad_norm": 6.4375,
+      "learning_rate": 1.9915583894358744e-05,
+      "loss": 1.7375,
+      "step": 550
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 8.75,
+      "learning_rate": 1.9903891431021477e-05,
+      "loss": 1.9266,
+      "step": 560
+    },
+    {
+      "epoch": 0.1425,
+      "grad_norm": 8.0625,
+      "learning_rate": 1.989144474800726e-05,
+      "loss": 1.6797,
+      "step": 570
+    },
+    {
+      "epoch": 0.145,
+      "grad_norm": 6.25,
+      "learning_rate": 1.9878244793179198e-05,
+      "loss": 1.7086,
+      "step": 580
+    },
+    {
+      "epoch": 0.1475,
+      "grad_norm": 9.5625,
+      "learning_rate": 1.9864292571764956e-05,
+      "loss": 1.6016,
+      "step": 590
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 3.546875,
+      "learning_rate": 1.9849589146280212e-05,
+      "loss": 1.7102,
+      "step": 600
+    },
+    {
+      "epoch": 0.1525,
+      "grad_norm": 4.875,
+      "learning_rate": 1.9834135636447745e-05,
+      "loss": 1.5586,
+      "step": 610
+    },
+    {
+      "epoch": 0.155,
+      "grad_norm": 5.03125,
+      "learning_rate": 1.981793321911216e-05,
+      "loss": 1.6461,
+      "step": 620
+    },
+    {
+      "epoch": 0.1575,
+      "grad_norm": 3.640625,
+      "learning_rate": 1.9800983128150263e-05,
+      "loss": 1.5922,
+      "step": 630
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.65625,
+      "learning_rate": 1.978328665437711e-05,
+      "loss": 1.607,
+      "step": 640
+    },
+    {
+      "epoch": 0.1625,
+      "grad_norm": 5.15625,
+      "learning_rate": 1.9764845145447687e-05,
+      "loss": 1.5422,
+      "step": 650
+    },
+    {
+      "epoch": 0.165,
+      "grad_norm": 6.3125,
+      "learning_rate": 1.974566000575431e-05,
+      "loss": 1.4977,
+      "step": 660
+    },
+    {
+      "epoch": 0.1675,
+      "grad_norm": 5.5625,
+      "learning_rate": 1.9725732696319634e-05,
+      "loss": 1.5188,
+      "step": 670
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 7.125,
+      "learning_rate": 1.9705064734685425e-05,
+      "loss": 1.4133,
+      "step": 680
+    },
+    {
+      "epoch": 0.1725,
+      "grad_norm": 3.90625,
+      "learning_rate": 1.9683657694796988e-05,
+      "loss": 1.4055,
+      "step": 690
+    },
+    {
+      "epoch": 0.175,
+      "grad_norm": 3.078125,
+      "learning_rate": 1.9661513206883288e-05,
+      "loss": 1.3,
+      "step": 700
+    },
+    {
+      "epoch": 0.1775,
+      "grad_norm": 3.46875,
+      "learning_rate": 1.963863295733281e-05,
+      "loss": 1.3258,
+      "step": 710
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 3.5625,
+      "learning_rate": 1.961501868856515e-05,
+      "loss": 1.323,
+      "step": 720
+    },
+    {
+      "epoch": 0.1825,
+      "grad_norm": 3.21875,
+      "learning_rate": 1.9590672198898297e-05,
+      "loss": 1.3035,
+      "step": 730
+    },
+    {
+      "epoch": 0.185,
+      "grad_norm": 3.109375,
+      "learning_rate": 1.95655953424117e-05,
+      "loss": 1.3477,
+      "step": 740
+    },
+    {
+      "epoch": 0.1875,
+      "grad_norm": 3.1875,
+      "learning_rate": 1.953979002880507e-05,
+      "loss": 1.368,
+      "step": 750
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 3.609375,
+      "learning_rate": 1.951325822325295e-05,
+      "loss": 1.2703,
+      "step": 760
+    },
+    {
+      "epoch": 0.1925,
+      "grad_norm": 2.515625,
+      "learning_rate": 1.9486001946255046e-05,
+      "loss": 1.1863,
+      "step": 770
+    },
+    {
+      "epoch": 0.195,
+      "grad_norm": 2.484375,
+      "learning_rate": 1.945802327348239e-05,
+      "loss": 1.2457,
+      "step": 780
+    },
+    {
+      "epoch": 0.1975,
+      "grad_norm": 2.65625,
+      "learning_rate": 1.9429324335619234e-05,
+      "loss": 1.2875,
+      "step": 790
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 4.125,
+      "learning_rate": 1.93999073182008e-05,
+      "loss": 1.2703,
+      "step": 800
+    },
+    {
+      "epoch": 0.2025,
+      "grad_norm": 2.9375,
+      "learning_rate": 1.936977446144687e-05,
+      "loss": 1.2633,
+      "step": 810
+    },
+    {
+      "epoch": 0.205,
+      "grad_norm": 3.578125,
+      "learning_rate": 1.9338928060091145e-05,
+      "loss": 1.1969,
+      "step": 820
+    },
+    {
+      "epoch": 0.2075,
+      "grad_norm": 3.421875,
+      "learning_rate": 1.9307370463206512e-05,
+      "loss": 1.202,
+      "step": 830
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 2.390625,
+      "learning_rate": 1.9275104074026152e-05,
+      "loss": 1.2328,
+      "step": 840
+    },
+    {
+      "epoch": 0.2125,
+      "grad_norm": 2.71875,
+      "learning_rate": 1.9242131349760536e-05,
+      "loss": 1.1777,
+      "step": 850
+    },
+    {
+      "epoch": 0.215,
+      "grad_norm": 2.84375,
+      "learning_rate": 1.9208454801410267e-05,
+      "loss": 1.2652,
+      "step": 860
+    },
+    {
+      "epoch": 0.2175,
+      "grad_norm": 2.84375,
+      "learning_rate": 1.9174076993574883e-05,
+      "loss": 1.2937,
+      "step": 870
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 2.65625,
+      "learning_rate": 1.913900054425756e-05,
+      "loss": 1.302,
+      "step": 880
+    },
+    {
+      "epoch": 0.2225,
+      "grad_norm": 3.25,
+      "learning_rate": 1.9103228124665713e-05,
+      "loss": 1.2574,
+      "step": 890
+    },
+    {
+      "epoch": 0.225,
+      "grad_norm": 2.53125,
+      "learning_rate": 1.906676245900759e-05,
+      "loss": 1.2055,
+      "step": 900
+    },
+    {
+      "epoch": 0.2275,
+      "grad_norm": 2.671875,
+      "learning_rate": 1.9029606324284814e-05,
+      "loss": 1.2168,
+      "step": 910
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 3.34375,
+      "learning_rate": 1.8991762550080905e-05,
+      "loss": 1.1273,
+      "step": 920
+    },
+    {
+      "epoch": 0.2325,
+      "grad_norm": 2.578125,
+      "learning_rate": 1.895323401834578e-05,
+      "loss": 1.1535,
+      "step": 930
+    },
+    {
+      "epoch": 0.235,
+      "grad_norm": 2.734375,
+      "learning_rate": 1.8914023663176306e-05,
+      "loss": 1.1539,
+      "step": 940
+    },
+    {
+      "epoch": 0.2375,
+      "grad_norm": 2.21875,
+      "learning_rate": 1.8874134470592836e-05,
+      "loss": 1.225,
+      "step": 950
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 2.890625,
+      "learning_rate": 1.8833569478311818e-05,
+      "loss": 1.1313,
+      "step": 960
+    },
+    {
+      "epoch": 0.2425,
+      "grad_norm": 3.953125,
+      "learning_rate": 1.879233177551447e-05,
+      "loss": 1.1656,
+      "step": 970
+    },
+    {
+      "epoch": 0.245,
+      "grad_norm": 2.875,
+      "learning_rate": 1.8750424502611528e-05,
+      "loss": 1.2047,
+      "step": 980
+    },
+    {
+      "epoch": 0.2475,
+      "grad_norm": 2.90625,
+      "learning_rate": 1.870785085100406e-05,
+      "loss": 1.1211,
+      "step": 990
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 3.140625,
+      "learning_rate": 1.8664614062840472e-05,
+      "loss": 1.25,
+      "step": 1000
+    },
+    {
+      "epoch": 0.2525,
+      "grad_norm": 2.5625,
+      "learning_rate": 1.8620717430769586e-05,
+      "loss": 1.2066,
+      "step": 1010
+    },
+    {
+      "epoch": 0.255,
+      "grad_norm": 1.984375,
+      "learning_rate": 1.8576164297689877e-05,
+      "loss": 1.1,
+      "step": 1020
+    },
+    {
+      "epoch": 0.2575,
+      "grad_norm": 2.53125,
+      "learning_rate": 1.8530958056494934e-05,
+      "loss": 1.0852,
+      "step": 1030
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 2.328125,
+      "learning_rate": 1.8485102149815038e-05,
+      "loss": 1.1875,
+      "step": 1040
+    },
+    {
+      "epoch": 0.2625,
+      "grad_norm": 2.515625,
+      "learning_rate": 1.8438600069755027e-05,
+      "loss": 1.198,
+      "step": 1050
+    },
+    {
+      "epoch": 0.265,
+      "grad_norm": 2.640625,
+      "learning_rate": 1.8391455357628334e-05,
+      "loss": 1.1926,
+      "step": 1060
+    },
+    {
+      "epoch": 0.2675,
+      "grad_norm": 2.21875,
+      "learning_rate": 1.834367160368732e-05,
+      "loss": 1.0836,
+      "step": 1070
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 2.65625,
+      "learning_rate": 1.8295252446849842e-05,
+      "loss": 1.0746,
+      "step": 1080
+    },
+    {
+      "epoch": 0.2725,
+      "grad_norm": 2.984375,
+      "learning_rate": 1.8246201574422164e-05,
+      "loss": 1.1871,
+      "step": 1090
+    },
+    {
+      "epoch": 0.275,
+      "grad_norm": 2.6875,
+      "learning_rate": 1.8196522721818128e-05,
+      "loss": 1.0598,
+      "step": 1100
+    },
+    {
+      "epoch": 0.2775,
+      "grad_norm": 2.28125,
+      "learning_rate": 1.8146219672274693e-05,
+      "loss": 1.1668,
+      "step": 1110
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 2.1875,
+      "learning_rate": 1.8095296256563845e-05,
+      "loss": 1.1094,
+      "step": 1120
+    },
+    {
+      "epoch": 0.2825,
+      "grad_norm": 2.5,
+      "learning_rate": 1.8043756352700844e-05,
+      "loss": 1.1344,
+      "step": 1130
+    },
+    {
+      "epoch": 0.285,
+      "grad_norm": 2.046875,
+      "learning_rate": 1.799160388564892e-05,
+      "loss": 1.0363,
+      "step": 1140
+    },
+    {
+      "epoch": 0.2875,
+      "grad_norm": 2.046875,
+      "learning_rate": 1.7938842827020347e-05,
+      "loss": 1.1289,
+      "step": 1150
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.78125,
+      "learning_rate": 1.788547719477402e-05,
+      "loss": 1.1938,
+      "step": 1160
+    },
+    {
+      "epoch": 0.2925,
+      "grad_norm": 3.765625,
+      "learning_rate": 1.7831511052909442e-05,
+      "loss": 1.1738,
+      "step": 1170
+    },
+    {
+      "epoch": 0.295,
+      "grad_norm": 2.546875,
+      "learning_rate": 1.777694851115726e-05,
+      "loss": 1.1242,
+      "step": 1180
+    },
+    {
+      "epoch": 0.2975,
+      "grad_norm": 2.234375,
+      "learning_rate": 1.772179372466627e-05,
+      "loss": 1.1371,
+      "step": 1190
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 2.125,
+      "learning_rate": 1.7666050893687007e-05,
+      "loss": 1.0867,
+      "step": 1200
+    },
+    {
+      "epoch": 0.3025,
+      "grad_norm": 3.5625,
+      "learning_rate": 1.760972426325187e-05,
+      "loss": 1.1574,
+      "step": 1210
+    },
+    {
+      "epoch": 0.305,
+      "grad_norm": 2.234375,
+      "learning_rate": 1.7552818122851838e-05,
+      "loss": 1.134,
+      "step": 1220
+    },
+    {
+      "epoch": 0.3075,
+      "grad_norm": 2.5625,
+      "learning_rate": 1.7495336806109828e-05,
+      "loss": 1.0508,
+      "step": 1230
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 2.234375,
+      "learning_rate": 1.7437284690450656e-05,
+      "loss": 1.0469,
+      "step": 1240
+    },
+    {
+      "epoch": 0.3125,
+      "grad_norm": 2.265625,
+      "learning_rate": 1.7378666196767685e-05,
+      "loss": 0.9703,
+      "step": 1250
+    },
+    {
+      "epoch": 0.315,
+      "grad_norm": 2.796875,
+      "learning_rate": 1.7319485789086164e-05,
+      "loss": 1.0586,
+      "step": 1260
+    },
+    {
+      "epoch": 0.3175,
+      "grad_norm": 3.28125,
+      "learning_rate": 1.7259747974223264e-05,
+      "loss": 1.1391,
+      "step": 1270
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 3.03125,
+      "learning_rate": 1.719945730144487e-05,
+      "loss": 1.0309,
+      "step": 1280
+    },
+    {
+      "epoch": 0.3225,
+      "grad_norm": 2.03125,
+      "learning_rate": 1.7138618362119136e-05,
+      "loss": 1.0547,
+      "step": 1290
+    },
+    {
+      "epoch": 0.325,
+      "grad_norm": 2.34375,
+      "learning_rate": 1.7077235789366845e-05,
+      "loss": 1.0324,
+      "step": 1300
+    },
+    {
+      "epoch": 0.3275,
+      "grad_norm": 2.4375,
+      "learning_rate": 1.701531425770856e-05,
+      "loss": 1.1266,
+      "step": 1310
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 2.140625,
+      "learning_rate": 1.6952858482708657e-05,
+      "loss": 1.0305,
+      "step": 1320
+    },
+    {
+      "epoch": 0.3325,
+      "grad_norm": 2.0625,
+      "learning_rate": 1.6889873220616208e-05,
+      "loss": 1.0449,
+      "step": 1330
+    },
+    {
+      "epoch": 0.335,
+      "grad_norm": 2.59375,
+      "learning_rate": 1.682636326800278e-05,
+      "loss": 1.0914,
+      "step": 1340
+    },
+    {
+      "epoch": 0.3375,
+      "grad_norm": 2.296875,
+      "learning_rate": 1.6762333461397157e-05,
+      "loss": 1.1625,
+      "step": 1350
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 2.171875,
+      "learning_rate": 1.6697788676917007e-05,
+      "loss": 1.009,
+      "step": 1360
+    },
+    {
+      "epoch": 0.3425,
+      "grad_norm": 1.796875,
+      "learning_rate": 1.6632733829897567e-05,
+      "loss": 1.0727,
+      "step": 1370
+    },
+    {
+      "epoch": 0.345,
+      "grad_norm": 2.40625,
+      "learning_rate": 1.656717387451731e-05,
+      "loss": 1.0738,
+      "step": 1380
+    },
+    {
+      "epoch": 0.3475,
+      "grad_norm": 2.140625,
+      "learning_rate": 1.650111380342066e-05,
+      "loss": 1.0074,
+      "step": 1390
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 2.125,
+      "learning_rate": 1.643455864733779e-05,
+      "loss": 1.0281,
+      "step": 1400
+    },
+    {
+      "epoch": 0.3525,
+      "grad_norm": 1.703125,
+      "learning_rate": 1.636751347470152e-05,
+      "loss": 0.9555,
+      "step": 1410
+    },
+    {
+      "epoch": 0.355,
+      "grad_norm": 2.53125,
+      "learning_rate": 1.6299983391261323e-05,
+      "loss": 1.0359,
+      "step": 1420
+    },
+    {
+      "epoch": 0.3575,
+      "grad_norm": 2.546875,
+      "learning_rate": 1.6231973539694506e-05,
+      "loss": 0.9559,
+      "step": 1430
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 2.546875,
+      "learning_rate": 1.616348909921457e-05,
+      "loss": 0.991,
+      "step": 1440
+    },
+    {
+      "epoch": 0.3625,
+      "grad_norm": 2.078125,
+      "learning_rate": 1.6094535285176812e-05,
+      "loss": 0.9996,
+      "step": 1450
+    },
+    {
+      "epoch": 0.365,
+      "grad_norm": 1.90625,
+      "learning_rate": 1.6025117348681133e-05,
+      "loss": 1.0793,
+      "step": 1460
+    },
+    {
+      "epoch": 0.3675,
+      "grad_norm": 1.7890625,
+      "learning_rate": 1.5955240576172165e-05,
+      "loss": 0.9895,
+      "step": 1470
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 2.234375,
+      "learning_rate": 1.588491028903667e-05,
+      "loss": 0.9875,
+      "step": 1480
+    },
+    {
+      "epoch": 0.3725,
+      "grad_norm": 1.9296875,
+      "learning_rate": 1.5814131843198307e-05,
+      "loss": 0.9437,
+      "step": 1490
+    },
+    {
+      "epoch": 0.375,
+      "grad_norm": 2.453125,
+      "learning_rate": 1.5742910628709757e-05,
+      "loss": 1.0055,
+      "step": 1500
+    },
+    {
+      "epoch": 0.3775,
+      "grad_norm": 2.109375,
+      "learning_rate": 1.567125206934225e-05,
+      "loss": 1.0035,
+      "step": 1510
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 2.484375,
+      "learning_rate": 1.5599161622172515e-05,
+      "loss": 0.9711,
+      "step": 1520
+    },
+    {
+      "epoch": 0.3825,
+      "grad_norm": 1.65625,
+      "learning_rate": 1.552664477716722e-05,
+      "loss": 1.025,
+      "step": 1530
+    },
+    {
+      "epoch": 0.385,
+      "grad_norm": 2.796875,
+      "learning_rate": 1.5453707056764865e-05,
+      "loss": 0.9699,
+      "step": 1540
+    },
+    {
+      "epoch": 0.3875,
+      "grad_norm": 1.8671875,
+      "learning_rate": 1.538035401545525e-05,
+      "loss": 0.9516,
+      "step": 1550
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 2.375,
+      "learning_rate": 1.5306591239356477e-05,
+      "loss": 0.9523,
+      "step": 1560
+    },
+    {
+      "epoch": 0.3925,
+      "grad_norm": 2.109375,
+      "learning_rate": 1.5232424345789517e-05,
+      "loss": 1.0004,
+      "step": 1570
+    },
+    {
+      "epoch": 0.395,
+      "grad_norm": 2.109375,
+      "learning_rate": 1.5157858982850475e-05,
+      "loss": 1.0707,
+      "step": 1580
+    },
+    {
+      "epoch": 0.3975,
+      "grad_norm": 2.265625,
+      "learning_rate": 1.5082900828980425e-05,
+      "loss": 1.0309,
+      "step": 1590
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 2.34375,
+      "learning_rate": 1.5007555592532996e-05,
+      "loss": 1.0695,
+      "step": 1600
+    },
+    {
+      "epoch": 0.4025,
+      "grad_norm": 2.109375,
+      "learning_rate": 1.4931829011339658e-05,
+      "loss": 1.0,
+      "step": 1610
+    },
+    {
+      "epoch": 0.405,
+      "grad_norm": 2.328125,
+      "learning_rate": 1.4855726852272754e-05,
+      "loss": 0.9523,
+      "step": 1620
+    },
+    {
+      "epoch": 0.4075,
+      "grad_norm": 1.8671875,
+      "learning_rate": 1.4779254910806336e-05,
+      "loss": 0.9441,
+      "step": 1630
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 2.296875,
+      "learning_rate": 1.4702419010574824e-05,
+      "loss": 0.9906,
+      "step": 1640
+    },
+    {
+      "epoch": 0.4125,
+      "grad_norm": 2.25,
+      "learning_rate": 1.4625225002929501e-05,
+      "loss": 0.9125,
+      "step": 1650
+    },
+    {
+      "epoch": 0.415,
+      "grad_norm": 2.15625,
+      "learning_rate": 1.4547678766492919e-05,
+      "loss": 1.0262,
+      "step": 1660
+    },
+    {
+      "epoch": 0.4175,
+      "grad_norm": 1.859375,
+      "learning_rate": 1.4469786206711213e-05,
+      "loss": 0.9012,
+      "step": 1670
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 2.359375,
+      "learning_rate": 1.4391553255404385e-05,
+      "loss": 0.9602,
+      "step": 1680
+    },
+    {
+      "epoch": 0.4225,
+      "grad_norm": 2.3125,
+      "learning_rate": 1.4312985870314568e-05,
+      "loss": 1.0035,
+      "step": 1690
+    },
+    {
+      "epoch": 0.425,
+      "grad_norm": 2.15625,
+      "learning_rate": 1.4234090034652324e-05,
+      "loss": 1.0129,
+      "step": 1700
+    },
+    {
+      "epoch": 0.4275,
+      "grad_norm": 2.21875,
+      "learning_rate": 1.4154871756640996e-05,
+      "loss": 0.9832,
+      "step": 1710
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 2.203125,
+      "learning_rate": 1.4075337069059159e-05,
+      "loss": 1.059,
+      "step": 1720
+    },
+    {
+      "epoch": 0.4325,
+      "grad_norm": 2.34375,
+      "learning_rate": 1.3995492028781202e-05,
+      "loss": 0.9574,
+      "step": 1730
+    },
+    {
+      "epoch": 0.435,
+      "grad_norm": 1.9296875,
+      "learning_rate": 1.3915342716316075e-05,
+      "loss": 0.8828,
+      "step": 1740
+    },
+    {
+      "epoch": 0.4375,
+      "grad_norm": 2.25,
+      "learning_rate": 1.3834895235344242e-05,
+      "loss": 1.0012,
+      "step": 1750
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.8515625,
+      "learning_rate": 1.3754155712252832e-05,
+      "loss": 1.0215,
+      "step": 1760
+    },
+    {
+      "epoch": 0.4425,
+      "grad_norm": 2.5,
+      "learning_rate": 1.367313029566913e-05,
+      "loss": 1.0137,
+      "step": 1770
+    },
+    {
+      "epoch": 0.445,
+      "grad_norm": 1.953125,
+      "learning_rate": 1.3591825155992311e-05,
+      "loss": 0.9488,
+      "step": 1780
+    },
+    {
+      "epoch": 0.4475,
+      "grad_norm": 2.59375,
+      "learning_rate": 1.3510246484923547e-05,
+      "loss": 0.8926,
+      "step": 1790
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 2.265625,
+      "learning_rate": 1.3428400494994484e-05,
+      "loss": 1.0133,
+      "step": 1800
+    },
+    {
+      "epoch": 0.4525,
+      "grad_norm": 1.8671875,
+      "learning_rate": 1.3346293419094134e-05,
+      "loss": 0.9164,
+      "step": 1810
+    },
+    {
+      "epoch": 0.455,
+      "grad_norm": 2.015625,
+      "learning_rate": 1.326393150999422e-05,
+      "loss": 1.0004,
+      "step": 1820
+    },
+    {
+      "epoch": 0.4575,
+      "grad_norm": 2.109375,
+      "learning_rate": 1.3181321039872995e-05,
+      "loss": 0.9195,
+      "step": 1830
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 1.9765625,
+      "learning_rate": 1.3098468299837599e-05,
+      "loss": 0.9422,
+      "step": 1840
+    },
+    {
+      "epoch": 0.4625,
+      "grad_norm": 2.0,
+      "learning_rate": 1.3015379599444958e-05,
+      "loss": 0.9629,
+      "step": 1850
+    },
+    {
+      "epoch": 0.465,
+      "grad_norm": 2.359375,
+      "learning_rate": 1.2932061266221304e-05,
+      "loss": 0.9598,
+      "step": 1860
+    },
+    {
+      "epoch": 0.4675,
+      "grad_norm": 2.890625,
+      "learning_rate": 1.2848519645180296e-05,
+      "loss": 1.0582,
+      "step": 1870
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 2.03125,
+      "learning_rate": 1.276476109833981e-05,
+      "loss": 0.9309,
+      "step": 1880
+    },
+    {
+      "epoch": 0.4725,
+      "grad_norm": 2.421875,
+      "learning_rate": 1.268079200423748e-05,
+      "loss": 1.009,
+      "step": 1890
+    },
+    {
+      "epoch": 0.475,
+      "grad_norm": 2.140625,
+      "learning_rate": 1.2596618757444918e-05,
+      "loss": 0.932,
+      "step": 1900
+    },
+    {
+      "epoch": 0.4775,
+      "grad_norm": 2.03125,
+      "learning_rate": 1.2512247768080756e-05,
+      "loss": 1.0012,
+      "step": 1910
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 2.484375,
+      "learning_rate": 1.2427685461322497e-05,
+      "loss": 0.9324,
+      "step": 1920
+    },
+    {
+      "epoch": 0.4825,
+      "grad_norm": 2.40625,
+      "learning_rate": 1.2342938276917187e-05,
+      "loss": 1.018,
+      "step": 1930
+    },
+    {
+      "epoch": 0.485,
+      "grad_norm": 2.234375,
+      "learning_rate": 1.2258012668691039e-05,
+      "loss": 0.8996,
+      "step": 1940
+    },
+    {
+      "epoch": 0.4875,
+      "grad_norm": 1.65625,
+      "learning_rate": 1.2172915104057919e-05,
+      "loss": 0.9336,
+      "step": 1950
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.78125,
+      "learning_rate": 1.2087652063526837e-05,
+      "loss": 0.8477,
+      "step": 1960
+    },
+    {
+      "epoch": 0.4925,
+      "grad_norm": 2.03125,
+      "learning_rate": 1.2002230040208446e-05,
+      "loss": 0.8961,
+      "step": 1970
+    },
+    {
+      "epoch": 0.495,
+      "grad_norm": 2.078125,
+      "learning_rate": 1.1916655539320547e-05,
+      "loss": 0.9328,
+      "step": 1980
+    },
+    {
+      "epoch": 0.4975,
+      "grad_norm": 2.0,
+      "learning_rate": 1.1830935077692696e-05,
+      "loss": 0.9344,
+      "step": 1990
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 2.09375,
+      "learning_rate": 1.174507518326992e-05,
+      "loss": 0.9539,
+      "step": 2000
+    },
+    {
+      "epoch": 0.5025,
+      "grad_norm": 2.859375,
+      "learning_rate": 1.1659082394615608e-05,
+      "loss": 0.9363,
+      "step": 2010
+    },
+    {
+      "epoch": 0.505,
+      "grad_norm": 1.921875,
+      "learning_rate": 1.1572963260413547e-05,
+      "loss": 0.9266,
+      "step": 2020
+    },
+    {
+      "epoch": 0.5075,
+      "grad_norm": 2.125,
+      "learning_rate": 1.1486724338969233e-05,
+      "loss": 0.9648,
+      "step": 2030
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 2.21875,
+      "learning_rate": 1.1400372197710414e-05,
+      "loss": 0.9266,
+      "step": 2040
+    },
+    {
+      "epoch": 0.5125,
+      "grad_norm": 1.6875,
+      "learning_rate": 1.131391341268698e-05,
+      "loss": 0.9023,
+      "step": 2050
+    },
+    {
+      "epoch": 0.515,
+      "grad_norm": 2.21875,
+      "learning_rate": 1.122735456807015e-05,
+      "loss": 1.0465,
+      "step": 2060
+    },
+    {
+      "epoch": 0.5175,
+      "grad_norm": 1.859375,
+      "learning_rate": 1.1140702255651064e-05,
+      "loss": 0.9465,
+      "step": 2070
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 2.46875,
+      "learning_rate": 1.1053963074338798e-05,
+      "loss": 0.9625,
+      "step": 2080
+    },
+    {
+      "epoch": 0.5225,
+      "grad_norm": 2.25,
+      "learning_rate": 1.0967143629657842e-05,
+      "loss": 0.9332,
+      "step": 2090
+    },
+    {
+      "epoch": 0.525,
+      "grad_norm": 2.359375,
+      "learning_rate": 1.0880250533245038e-05,
+      "loss": 0.9395,
+      "step": 2100
+    },
+    {
+      "epoch": 0.5275,
+      "grad_norm": 2.25,
+      "learning_rate": 1.0793290402346094e-05,
+      "loss": 0.8977,
+      "step": 2110
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 2.40625,
+      "learning_rate": 1.0706269859311669e-05,
+      "loss": 0.8789,
+      "step": 2120
+    },
+    {
+      "epoch": 0.5325,
+      "grad_norm": 2.09375,
+      "learning_rate": 1.0619195531093019e-05,
+      "loss": 0.8488,
+      "step": 2130
+    },
+    {
+      "epoch": 0.535,
+      "grad_norm": 1.921875,
+      "learning_rate": 1.0532074048737364e-05,
+      "loss": 0.8609,
+      "step": 2140
+    },
+    {
+      "epoch": 0.5375,
+      "grad_norm": 1.5859375,
+      "learning_rate": 1.0444912046882889e-05,
+      "loss": 0.9293,
+      "step": 2150
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 2.140625,
+      "learning_rate": 1.0357716163253498e-05,
+      "loss": 0.8918,
+      "step": 2160
+    },
+    {
+      "epoch": 0.5425,
+      "grad_norm": 1.875,
+      "learning_rate": 1.027049303815332e-05,
+      "loss": 0.9473,
+      "step": 2170
+    },
+    {
+      "epoch": 0.545,
+      "grad_norm": 1.8359375,
+      "learning_rate": 1.018324931396103e-05,
+      "loss": 0.8762,
+      "step": 2180
+    },
+    {
+      "epoch": 0.5475,
+      "grad_norm": 2.078125,
+      "learning_rate": 1.0095991634624001e-05,
+      "loss": 0.8715,
+      "step": 2190
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.9921875,
+      "learning_rate": 1.0008726645152354e-05,
+      "loss": 0.8809,
+      "step": 2200
+    },
+    {
+      "epoch": 0.5525,
+      "grad_norm": 1.8828125,
+      "learning_rate": 9.921460991112891e-06,
+      "loss": 0.8617,
+      "step": 2210
+    },
+    {
+      "epoch": 0.555,
+      "grad_norm": 1.9140625,
+      "learning_rate": 9.834201318123026e-06,
+      "loss": 0.8957,
+      "step": 2220
+    },
+    {
+      "epoch": 0.5575,
+      "grad_norm": 1.75,
+      "learning_rate": 9.746954271344703e-06,
+      "loss": 0.8918,
+      "step": 2230
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 2.03125,
+      "learning_rate": 9.659726494978325e-06,
+      "loss": 0.9281,
+      "step": 2240
+    },
+    {
+      "epoch": 0.5625,
+      "grad_norm": 2.078125,
+      "learning_rate": 9.572524631756779e-06,
+      "loss": 0.8645,
+      "step": 2250
+    },
+    {
+      "epoch": 0.565,
+      "grad_norm": 2.015625,
+      "learning_rate": 9.48535532243956e-06,
+      "loss": 0.9059,
+      "step": 2260
+    },
+    {
+      "epoch": 0.5675,
+      "grad_norm": 1.671875,
+      "learning_rate": 9.398225205307067e-06,
+      "loss": 0.8383,
+      "step": 2270
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 2.234375,
+      "learning_rate": 9.311140915655055e-06,
+      "loss": 0.8492,
+      "step": 2280
+    },
+    {
+      "epoch": 0.5725,
+      "grad_norm": 1.640625,
+      "learning_rate": 9.224109085289343e-06,
+      "loss": 1.0,
+      "step": 2290
+    },
+    {
+      "epoch": 0.575,
+      "grad_norm": 2.125,
+      "learning_rate": 9.137136342020768e-06,
+      "loss": 0.9285,
+      "step": 2300
+    },
+    {
+      "epoch": 0.5775,
+      "grad_norm": 2.109375,
+      "learning_rate": 9.050229309160462e-06,
+      "loss": 0.9578,
+      "step": 2310
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 2.15625,
+      "learning_rate": 8.963394605015453e-06,
+      "loss": 0.8965,
+      "step": 2320
+    },
+    {
+      "epoch": 0.5825,
+      "grad_norm": 1.9140625,
+      "learning_rate": 8.876638842384645e-06,
+      "loss": 0.8984,
+      "step": 2330
+    },
+    {
+      "epoch": 0.585,
+      "grad_norm": 2.125,
+      "learning_rate": 8.789968628055262e-06,
+      "loss": 0.884,
+      "step": 2340
+    },
+    {
+      "epoch": 0.5875,
+      "grad_norm": 2.234375,
+      "learning_rate": 8.703390562299683e-06,
+      "loss": 0.8582,
+      "step": 2350
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 2.15625,
+      "learning_rate": 8.616911238372812e-06,
+      "loss": 0.8559,
+      "step": 2360
+    },
+    {
+      "epoch": 0.5925,
+      "grad_norm": 1.8984375,
+      "learning_rate": 8.530537242009985e-06,
+      "loss": 0.893,
+      "step": 2370
+    },
+    {
+      "epoch": 0.595,
+      "grad_norm": 1.859375,
+      "learning_rate": 8.444275150925431e-06,
+      "loss": 0.8984,
+      "step": 2380
+    },
+    {
+      "epoch": 0.5975,
+      "grad_norm": 1.828125,
+      "learning_rate": 8.358131534311372e-06,
+      "loss": 0.8355,
+      "step": 2390
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 2.078125,
+      "learning_rate": 8.272112952337728e-06,
+      "loss": 0.873,
+      "step": 2400
+    },
+    {
+      "epoch": 0.6025,
+      "grad_norm": 2.203125,
+      "learning_rate": 8.186225955652547e-06,
+      "loss": 0.9348,
+      "step": 2410
+    },
+    {
+      "epoch": 0.605,
+      "grad_norm": 2.6875,
+      "learning_rate": 8.100477084883156e-06,
+      "loss": 0.9051,
+      "step": 2420
+    },
+    {
+      "epoch": 0.6075,
+      "grad_norm": 1.9609375,
+      "learning_rate": 8.014872870138055e-06,
+      "loss": 0.9023,
+      "step": 2430
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 1.8125,
+      "learning_rate": 7.929419830509619e-06,
+      "loss": 0.9203,
+      "step": 2440
+    },
+    {
+      "epoch": 0.6125,
+      "grad_norm": 1.765625,
+      "learning_rate": 7.844124473577672e-06,
+      "loss": 0.8008,
+      "step": 2450
+    },
+    {
+      "epoch": 0.615,
+      "grad_norm": 2.046875,
+      "learning_rate": 7.758993294913873e-06,
+      "loss": 0.9168,
+      "step": 2460
+    },
+    {
+      "epoch": 0.6175,
+      "grad_norm": 1.765625,
+      "learning_rate": 7.674032777587076e-06,
+      "loss": 0.9113,
+      "step": 2470
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 2.21875,
+      "learning_rate": 7.589249391669615e-06,
+      "loss": 0.9016,
+      "step": 2480
+    },
+    {
+      "epoch": 0.6225,
+      "grad_norm": 1.84375,
+      "learning_rate": 7.5046495937445774e-06,
+      "loss": 0.9246,
+      "step": 2490
+    },
+    {
+      "epoch": 0.625,
+      "grad_norm": 2.03125,
+      "learning_rate": 7.4202398264141195e-06,
+      "loss": 0.8012,
+      "step": 2500
+    },
+    {
+      "epoch": 0.6275,
+      "grad_norm": 2.34375,
+      "learning_rate": 7.336026517808828e-06,
+      "loss": 0.9129,
+      "step": 2510
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 1.7421875,
+      "learning_rate": 7.252016081098195e-06,
+      "loss": 0.8715,
+      "step": 2520
+    },
+    {
+      "epoch": 0.6325,
+      "grad_norm": 2.203125,
+      "learning_rate": 7.168214914002243e-06,
+      "loss": 0.877,
+      "step": 2530
+    },
+    {
+      "epoch": 0.635,
+      "grad_norm": 2.265625,
+      "learning_rate": 7.08462939830429e-06,
+      "loss": 0.9047,
+      "step": 2540
+    },
+    {
+      "epoch": 0.6375,
+      "grad_norm": 2.140625,
+      "learning_rate": 7.001265899364968e-06,
+      "loss": 0.8498,
+      "step": 2550
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 2.21875,
+      "learning_rate": 6.918130765637485e-06,
+      "loss": 0.8734,
+      "step": 2560
+    },
+    {
+      "epoch": 0.6425,
+      "grad_norm": 1.8359375,
+      "learning_rate": 6.835230328184139e-06,
+      "loss": 0.8289,
+      "step": 2570
+    },
+    {
+      "epoch": 0.645,
+      "grad_norm": 2.625,
+      "learning_rate": 6.752570900194206e-06,
+      "loss": 0.9523,
+      "step": 2580
+    },
+    {
+      "epoch": 0.6475,
+      "grad_norm": 2.125,
+      "learning_rate": 6.670158776503158e-06,
+      "loss": 0.907,
+      "step": 2590
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 1.9609375,
+      "learning_rate": 6.588000233113282e-06,
+      "loss": 0.9313,
+      "step": 2600
+    },
+    {
+      "epoch": 0.6525,
+      "grad_norm": 1.7734375,
+      "learning_rate": 6.506101526715749e-06,
+      "loss": 0.8465,
+      "step": 2610
+    },
+    {
+      "epoch": 0.655,
+      "grad_norm": 1.8359375,
+      "learning_rate": 6.424468894214137e-06,
+      "loss": 0.9383,
+      "step": 2620
+    },
+    {
+      "epoch": 0.6575,
+      "grad_norm": 1.5546875,
+      "learning_rate": 6.343108552249457e-06,
+      "loss": 0.8391,
+      "step": 2630
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 1.984375,
+      "learning_rate": 6.262026696726759e-06,
+      "loss": 0.977,
+      "step": 2640
+    },
+    {
+      "epoch": 0.6625,
+      "grad_norm": 1.8203125,
+      "learning_rate": 6.181229502343256e-06,
+      "loss": 0.8105,
+      "step": 2650
+    },
+    {
+      "epoch": 0.665,
+      "grad_norm": 1.890625,
+      "learning_rate": 6.100723122118121e-06,
+      "loss": 0.907,
+      "step": 2660
+    },
+    {
+      "epoch": 0.6675,
+      "grad_norm": 1.765625,
+      "learning_rate": 6.020513686923897e-06,
+      "loss": 0.8016,
+      "step": 2670
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 1.6796875,
+      "learning_rate": 5.940607305019623e-06,
+      "loss": 0.8281,
+      "step": 2680
+    },
+    {
+      "epoch": 0.6725,
+      "grad_norm": 2.265625,
+      "learning_rate": 5.861010061585652e-06,
+      "loss": 0.8668,
+      "step": 2690
+    },
+    {
+      "epoch": 0.675,
+      "grad_norm": 1.5625,
+      "learning_rate": 5.781728018260242e-06,
+      "loss": 0.9086,
+      "step": 2700
+    },
+    {
+      "epoch": 0.6775,
+      "grad_norm": 1.6796875,
+      "learning_rate": 5.7027672126779465e-06,
+      "loss": 0.859,
+      "step": 2710
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 1.515625,
+      "learning_rate": 5.624133658009817e-06,
+      "loss": 0.8441,
+      "step": 2720
+    },
+    {
+      "epoch": 0.6825,
+      "grad_norm": 2.125,
+      "learning_rate": 5.5458333425054825e-06,
+      "loss": 0.8914,
+      "step": 2730
+    },
+    {
+      "epoch": 0.685,
+      "grad_norm": 2.46875,
+      "learning_rate": 5.467872229037128e-06,
+      "loss": 0.898,
+      "step": 2740
+    },
+    {
+      "epoch": 0.6875,
+      "grad_norm": 2.078125,
+      "learning_rate": 5.390256254645379e-06,
+      "loss": 0.843,
+      "step": 2750
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 2.03125,
+      "learning_rate": 5.3129913300871846e-06,
+      "loss": 0.898,
+      "step": 2760
+    },
+    {
+      "epoch": 0.6925,
+      "grad_norm": 2.140625,
+      "learning_rate": 5.236083339385698e-06,
+      "loss": 0.9121,
+      "step": 2770
+    },
+    {
+      "epoch": 0.695,
+      "grad_norm": 1.7734375,
+      "learning_rate": 5.159538139382168e-06,
+      "loss": 0.9285,
+      "step": 2780
+    },
+    {
+      "epoch": 0.6975,
+      "grad_norm": 1.8203125,
+      "learning_rate": 5.0833615592899355e-06,
+      "loss": 0.8777,
+      "step": 2790
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 1.890625,
+      "learning_rate": 5.0075594002505035e-06,
+      "loss": 0.8766,
+      "step": 2800
+    },
+    {
+      "epoch": 0.7025,
+      "grad_norm": 1.7265625,
+      "learning_rate": 4.932137434891759e-06,
+      "loss": 0.8754,
+      "step": 2810
+    },
+    {
+      "epoch": 0.705,
+      "grad_norm": 1.9375,
+      "learning_rate": 4.857101406888378e-06,
+      "loss": 0.8742,
+      "step": 2820
+    },
+    {
+      "epoch": 0.7075,
+      "grad_norm": 2.15625,
+      "learning_rate": 4.782457030524405e-06,
+      "loss": 0.8336,
+      "step": 2830
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 1.9609375,
+      "learning_rate": 4.708209990258096e-06,
+      "loss": 0.8004,
+      "step": 2840
+    },
+    {
+      "epoch": 0.7125,
+      "grad_norm": 1.5546875,
+      "learning_rate": 4.634365940289026e-06,
+      "loss": 0.7947,
+      "step": 2850
+    },
+    {
+      "epoch": 0.715,
+      "grad_norm": 2.015625,
+      "learning_rate": 4.560930504127501e-06,
+      "loss": 0.941,
+      "step": 2860
+    },
+    {
+      "epoch": 0.7175,
+      "grad_norm": 2.0625,
+      "learning_rate": 4.4879092741663e-06,
+      "loss": 0.8086,
+      "step": 2870
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 2.546875,
+      "learning_rate": 4.4153078112547955e-06,
+      "loss": 0.8824,
+      "step": 2880
+    },
+    {
+      "epoch": 0.7225,
+      "grad_norm": 1.75,
+      "learning_rate": 4.343131644275478e-06,
+      "loss": 0.8527,
+      "step": 2890
+    },
+    {
+      "epoch": 0.725,
+      "grad_norm": 1.8671875,
+      "learning_rate": 4.271386269722909e-06,
+      "loss": 0.8316,
+      "step": 2900
+    },
+    {
+      "epoch": 0.7275,
+      "grad_norm": 2.1875,
+      "learning_rate": 4.20007715128513e-06,
+      "loss": 0.8324,
+      "step": 2910
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 1.84375,
+      "learning_rate": 4.129209719427596e-06,
+      "loss": 0.8141,
+      "step": 2920
+    },
+    {
+      "epoch": 0.7325,
+      "grad_norm": 2.328125,
+      "learning_rate": 4.058789370979616e-06,
+      "loss": 0.8602,
+      "step": 2930
+    },
+    {
+      "epoch": 0.735,
+      "grad_norm": 1.9296875,
+      "learning_rate": 3.988821468723368e-06,
+      "loss": 0.8812,
+      "step": 2940
+    },
+    {
+      "epoch": 0.7375,
+      "grad_norm": 1.9921875,
+      "learning_rate": 3.919311340985501e-06,
+      "loss": 0.827,
+      "step": 2950
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 1.8203125,
+      "learning_rate": 3.8502642812313574e-06,
+      "loss": 0.8047,
+      "step": 2960
+    },
+    {
+      "epoch": 0.7425,
+      "grad_norm": 1.7421875,
+      "learning_rate": 3.781685547661872e-06,
+      "loss": 0.8371,
+      "step": 2970
+    },
+    {
+      "epoch": 0.745,
+      "grad_norm": 2.296875,
+      "learning_rate": 3.7135803628131152e-06,
+      "loss": 0.8727,
+      "step": 2980
+    },
+    {
+      "epoch": 0.7475,
+      "grad_norm": 1.7109375,
+      "learning_rate": 3.645953913158593e-06,
+      "loss": 0.7598,
+      "step": 2990
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 2.390625,
+      "learning_rate": 3.5788113487142707e-06,
+      "loss": 0.9648,
+      "step": 3000
+    },
+    {
+      "epoch": 0.7525,
+      "grad_norm": 1.921875,
+      "learning_rate": 3.5121577826463894e-06,
+      "loss": 0.8895,
+      "step": 3010
+    },
+    {
+      "epoch": 0.755,
+      "grad_norm": 1.9140625,
+      "learning_rate": 3.445998290882062e-06,
+      "loss": 0.9469,
+      "step": 3020
+    },
+    {
+      "epoch": 0.7575,
+      "grad_norm": 2.0625,
+      "learning_rate": 3.380337911722731e-06,
+      "loss": 0.8621,
+      "step": 3030
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 2.0,
+      "learning_rate": 3.315181645460488e-06,
+      "loss": 0.8336,
+      "step": 3040
+    },
+    {
+      "epoch": 0.7625,
+      "grad_norm": 1.9921875,
+      "learning_rate": 3.2505344539972703e-06,
+      "loss": 0.8711,
+      "step": 3050
+    },
+    {
+      "epoch": 0.765,
+      "grad_norm": 1.7109375,
+      "learning_rate": 3.1864012604669993e-06,
+      "loss": 0.8117,
+      "step": 3060
+    },
+    {
+      "epoch": 0.7675,
+      "grad_norm": 1.8203125,
+      "learning_rate": 3.1227869488606643e-06,
+      "loss": 0.7992,
+      "step": 3070
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 1.9140625,
+      "learning_rate": 3.0596963636543865e-06,
+      "loss": 0.8594,
+      "step": 3080
+    },
+    {
+      "epoch": 0.7725,
+      "grad_norm": 1.8984375,
+      "learning_rate": 2.9971343094404935e-06,
+      "loss": 0.8871,
+      "step": 3090
+    },
+    {
+      "epoch": 0.775,
+      "grad_norm": 2.09375,
+      "learning_rate": 2.935105550561631e-06,
+      "loss": 0.8023,
+      "step": 3100
+    },
+    {
+      "epoch": 0.7775,
+      "grad_norm": 1.671875,
+      "learning_rate": 2.8736148107479467e-06,
+      "loss": 0.8344,
+      "step": 3110
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 1.625,
+      "learning_rate": 2.8126667727573444e-06,
+      "loss": 0.9066,
+      "step": 3120
+    },
+    {
+      "epoch": 0.7825,
+      "grad_norm": 2.046875,
+      "learning_rate": 2.7522660780188904e-06,
+      "loss": 0.8305,
+      "step": 3130
+    },
+    {
+      "epoch": 0.785,
+      "grad_norm": 2.0,
+      "learning_rate": 2.6924173262793386e-06,
+      "loss": 0.9117,
+      "step": 3140
+    },
+    {
+      "epoch": 0.7875,
+      "grad_norm": 1.7578125,
+      "learning_rate": 2.633125075252848e-06,
+      "loss": 0.9074,
+      "step": 3150
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 1.8828125,
+      "learning_rate": 2.5743938402738923e-06,
+      "loss": 0.7902,
+      "step": 3160
+    },
+    {
+      "epoch": 0.7925,
+      "grad_norm": 2.046875,
+      "learning_rate": 2.5162280939534024e-06,
+      "loss": 0.8387,
+      "step": 3170
+    },
+    {
+      "epoch": 0.795,
+      "grad_norm": 1.828125,
+      "learning_rate": 2.4586322658381566e-06,
+      "loss": 0.9055,
+      "step": 3180
+    },
+    {
+      "epoch": 0.7975,
+      "grad_norm": 1.7421875,
+      "learning_rate": 2.4016107420734625e-06,
+      "loss": 0.8629,
+      "step": 3190
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 2.15625,
+      "learning_rate": 2.3451678650691202e-06,
+      "loss": 0.9141,
+      "step": 3200
+    },
+    {
+      "epoch": 0.8025,
+      "grad_norm": 2.015625,
+      "learning_rate": 2.2893079331687397e-06,
+      "loss": 0.7875,
+      "step": 3210
+    },
+    {
+      "epoch": 0.805,
+      "grad_norm": 2.109375,
+      "learning_rate": 2.2340352003224065e-06,
+      "loss": 0.8426,
+      "step": 3220
+    },
+    {
+      "epoch": 0.8075,
+      "grad_norm": 2.046875,
+      "learning_rate": 2.1793538757627217e-06,
+      "loss": 0.9023,
+      "step": 3230
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 2.265625,
+      "learning_rate": 2.125268123684262e-06,
+      "loss": 0.8246,
+      "step": 3240
+    },
+    {
+      "epoch": 0.8125,
+      "grad_norm": 1.953125,
+      "learning_rate": 2.0717820629264496e-06,
+      "loss": 0.8844,
+      "step": 3250
+    },
+    {
+      "epoch": 0.815,
+      "grad_norm": 2.015625,
+      "learning_rate": 2.0188997666598853e-06,
+      "loss": 0.8379,
+      "step": 3260
+    },
+    {
+      "epoch": 0.8175,
+      "grad_norm": 1.7421875,
+      "learning_rate": 1.9666252620761782e-06,
+      "loss": 0.8316,
+      "step": 3270
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 2.171875,
+      "learning_rate": 1.9149625300812403e-06,
+      "loss": 0.8441,
+      "step": 3280
+    },
+    {
+      "epoch": 0.8225,
+      "grad_norm": 2.328125,
+      "learning_rate": 1.863915504992131e-06,
+      "loss": 0.9191,
+      "step": 3290
+    },
+    {
+      "epoch": 0.825,
+      "grad_norm": 2.03125,
+      "learning_rate": 1.8134880742374483e-06,
+      "loss": 0.891,
+      "step": 3300
+    },
+    {
+      "epoch": 0.8275,
+      "grad_norm": 2.078125,
+      "learning_rate": 1.763684078061283e-06,
+      "loss": 0.9062,
+      "step": 3310
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 1.8828125,
+      "learning_rate": 1.7145073092307663e-06,
+      "loss": 0.8121,
+      "step": 3320
+    },
+    {
+      "epoch": 0.8325,
+      "grad_norm": 2.09375,
+      "learning_rate": 1.6659615127472384e-06,
+      "loss": 0.9273,
+      "step": 3330
+    },
+    {
+      "epoch": 0.835,
+      "grad_norm": 1.9921875,
+      "learning_rate": 1.6180503855610563e-06,
+      "loss": 0.85,
+      "step": 3340
+    },
+    {
+      "epoch": 0.8375,
+      "grad_norm": 1.8828125,
+      "learning_rate": 1.5707775762900546e-06,
+      "loss": 0.9297,
+      "step": 3350
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 1.546875,
+      "learning_rate": 1.5241466849416797e-06,
+      "loss": 0.8484,
+      "step": 3360
+    },
+    {
+      "epoch": 0.8425,
+      "grad_norm": 1.6015625,
+      "learning_rate": 1.4781612626388475e-06,
+      "loss": 0.882,
+      "step": 3370
+    },
+    {
+      "epoch": 0.845,
+      "grad_norm": 2.453125,
+      "learning_rate": 1.4328248113495046e-06,
+      "loss": 0.7793,
+      "step": 3380
+    },
+    {
+      "epoch": 0.8475,
+      "grad_norm": 2.03125,
+      "learning_rate": 1.3881407836199446e-06,
+      "loss": 0.8543,
+      "step": 3390
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 2.078125,
+      "learning_rate": 1.3441125823118816e-06,
+      "loss": 0.8391,
+      "step": 3400
+    },
+    {
+      "epoch": 0.8525,
+      "grad_norm": 2.125,
+      "learning_rate": 1.3007435603433039e-06,
+      "loss": 0.8832,
+      "step": 3410
+    },
+    {
+      "epoch": 0.855,
+      "grad_norm": 2.0625,
+      "learning_rate": 1.2580370204331482e-06,
+      "loss": 0.8434,
+      "step": 3420
+    },
+    {
+      "epoch": 0.8575,
+      "grad_norm": 1.8984375,
+      "learning_rate": 1.2159962148497728e-06,
+      "loss": 0.8723,
+      "step": 3430
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 1.671875,
+      "learning_rate": 1.17462434516329e-06,
+      "loss": 0.8309,
+      "step": 3440
+    },
+    {
+      "epoch": 0.8625,
+      "grad_norm": 1.921875,
+      "learning_rate": 1.1339245620017525e-06,
+      "loss": 0.9281,
+      "step": 3450
+    },
+    {
+      "epoch": 0.865,
+      "grad_norm": 1.9609375,
+      "learning_rate": 1.093899964811227e-06,
+      "loss": 0.8457,
+      "step": 3460
+    },
+    {
+      "epoch": 0.8675,
+      "grad_norm": 2.78125,
+      "learning_rate": 1.05455360161975e-06,
+      "loss": 0.9082,
+      "step": 3470
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 2.484375,
+      "learning_rate": 1.0158884688052107e-06,
+      "loss": 0.877,
+      "step": 3480
+    },
+    {
+      "epoch": 0.8725,
+      "grad_norm": 1.984375,
+      "learning_rate": 9.779075108671766e-07,
+      "loss": 0.9555,
+      "step": 3490
+    },
+    {
+      "epoch": 0.875,
+      "grad_norm": 2.015625,
+      "learning_rate": 9.406136202026417e-07,
+      "loss": 0.8816,
+      "step": 3500
+    },
+    {
+      "epoch": 0.8775,
+      "grad_norm": 2.15625,
+      "learning_rate": 9.040096368857676e-07,
+      "loss": 0.8691,
+      "step": 3510
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 2.234375,
+      "learning_rate": 8.680983484516048e-07,
+      "loss": 0.8758,
+      "step": 3520
+    },
+    {
+      "epoch": 0.8825,
+      "grad_norm": 2.125,
+      "learning_rate": 8.328824896838006e-07,
+      "loss": 0.8598,
+      "step": 3530
+    },
+    {
+      "epoch": 0.885,
+      "grad_norm": 2.234375,
+      "learning_rate": 7.983647424063468e-07,
+      "loss": 0.8441,
+      "step": 3540
+    },
+    {
+      "epoch": 0.8875,
+      "grad_norm": 1.8125,
+      "learning_rate": 7.645477352793362e-07,
+      "loss": 0.8098,
+      "step": 3550
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 1.6328125,
+      "learning_rate": 7.314340435987921e-07,
+      "loss": 0.8215,
+      "step": 3560
+    },
+    {
+      "epoch": 0.8925,
+      "grad_norm": 2.109375,
+      "learning_rate": 6.990261891005401e-07,
+      "loss": 0.857,
+      "step": 3570
+    },
+    {
+      "epoch": 0.895,
+      "grad_norm": 1.859375,
+      "learning_rate": 6.673266397681732e-07,
+      "loss": 0.8238,
+      "step": 3580
+    },
+    {
+      "epoch": 0.8975,
+      "grad_norm": 2.015625,
+      "learning_rate": 6.363378096451011e-07,
+      "loss": 0.8742,
+      "step": 3590
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 2.140625,
+      "learning_rate": 6.060620586507183e-07,
+      "loss": 0.8793,
+      "step": 3600
+    },
+    {
+      "epoch": 0.9025,
+      "grad_norm": 2.140625,
+      "learning_rate": 5.765016924006784e-07,
+      "loss": 0.8543,
+      "step": 3610
+    },
+    {
+      "epoch": 0.905,
+      "grad_norm": 2.203125,
+      "learning_rate": 5.476589620313189e-07,
+      "loss": 0.791,
+      "step": 3620
+    },
+    {
+      "epoch": 0.9075,
+      "grad_norm": 2.109375,
+      "learning_rate": 5.195360640282254e-07,
+      "loss": 0.8219,
+      "step": 3630
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 2.0,
+      "learning_rate": 4.92135140058967e-07,
+      "loss": 0.8609,
+      "step": 3640
+    },
+    {
+      "epoch": 0.9125,
+      "grad_norm": 2.359375,
+      "learning_rate": 4.65458276809988e-07,
+      "loss": 0.8852,
+      "step": 3650
+    },
+    {
+      "epoch": 0.915,
+      "grad_norm": 2.015625,
+      "learning_rate": 4.3950750582770985e-07,
+      "loss": 0.8453,
+      "step": 3660
+    },
+    {
+      "epoch": 0.9175,
+      "grad_norm": 2.0,
+      "learning_rate": 4.142848033638136e-07,
+      "loss": 0.8266,
+      "step": 3670
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 1.9921875,
+      "learning_rate": 3.897920902247465e-07,
+      "loss": 0.8063,
+      "step": 3680
+    },
+    {
+      "epoch": 0.9225,
+      "grad_norm": 1.7734375,
+      "learning_rate": 3.6603123162544216e-07,
+      "loss": 0.8461,
+      "step": 3690
+    },
+    {
+      "epoch": 0.925,
+      "grad_norm": 2.046875,
+      "learning_rate": 3.4300403704727536e-07,
+      "loss": 0.827,
+      "step": 3700
+    },
+    {
+      "epoch": 0.9275,
+      "grad_norm": 2.34375,
+      "learning_rate": 3.207122601002677e-07,
+      "loss": 0.9016,
+      "step": 3710
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 2.25,
+      "learning_rate": 2.991575983895401e-07,
+      "loss": 0.9215,
+      "step": 3720
+    },
+    {
+      "epoch": 0.9325,
+      "grad_norm": 2.296875,
+      "learning_rate": 2.783416933860339e-07,
+      "loss": 0.9125,
+      "step": 3730
+    },
+    {
+      "epoch": 0.935,
+      "grad_norm": 2.09375,
+      "learning_rate": 2.582661303015066e-07,
+      "loss": 0.8855,
+      "step": 3740
+    },
+    {
+      "epoch": 0.9375,
+      "grad_norm": 1.9375,
+      "learning_rate": 2.3893243796781374e-07,
+      "loss": 0.8465,
+      "step": 3750
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 2.140625,
+      "learning_rate": 2.203420887204788e-07,
+      "loss": 0.8602,
+      "step": 3760
+    },
+    {
+      "epoch": 0.9425,
+      "grad_norm": 2.328125,
+      "learning_rate": 2.0249649828657292e-07,
+      "loss": 0.8262,
+      "step": 3770
+    },
+    {
+      "epoch": 0.945,
+      "grad_norm": 1.9609375,
+      "learning_rate": 1.8539702567689888e-07,
+      "loss": 0.8328,
+      "step": 3780
+    },
+    {
+      "epoch": 0.9475,
+      "grad_norm": 2.03125,
+      "learning_rate": 1.6904497308250056e-07,
+      "loss": 0.8355,
+      "step": 3790
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 1.5,
+      "learning_rate": 1.534415857754923e-07,
+      "loss": 0.8668,
+      "step": 3800
+    },
+    {
+      "epoch": 0.9525,
+      "grad_norm": 2.0625,
+      "learning_rate": 1.3858805201422798e-07,
+      "loss": 0.8094,
+      "step": 3810
+    },
+    {
+      "epoch": 0.955,
+      "grad_norm": 2.375,
+      "learning_rate": 1.2448550295281136e-07,
+      "loss": 0.8645,
+      "step": 3820
+    },
+    {
+      "epoch": 0.9575,
+      "grad_norm": 2.359375,
+      "learning_rate": 1.1113501255495485e-07,
+      "loss": 0.9066,
+      "step": 3830
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 2.3125,
+      "learning_rate": 9.853759751218828e-08,
+      "loss": 0.8148,
+      "step": 3840
+    },
+    {
+      "epoch": 0.9625,
+      "grad_norm": 2.46875,
+      "learning_rate": 8.66942171664431e-08,
+      "loss": 0.9203,
+      "step": 3850
+    },
+    {
+      "epoch": 0.965,
+      "grad_norm": 1.953125,
+      "learning_rate": 7.560577343698306e-08,
+      "loss": 0.9016,
+      "step": 3860
+    },
+    {
+      "epoch": 0.9675,
+      "grad_norm": 2.09375,
+      "learning_rate": 6.527311075173127e-08,
+      "loss": 0.8219,
+      "step": 3870
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 1.5625,
+      "learning_rate": 5.5697015982955074e-08,
+      "loss": 0.7977,
+      "step": 3880
+    },
+    {
+      "epoch": 0.9725,
+      "grad_norm": 2.15625,
+      "learning_rate": 4.687821838734618e-08,
+      "loss": 0.902,
+      "step": 3890
+    },
+    {
+      "epoch": 0.975,
+      "grad_norm": 1.9609375,
+      "learning_rate": 3.881738955048398e-08,
+      "loss": 0.8852,
+      "step": 3900
+    },
+    {
+      "epoch": 0.9775,
+      "grad_norm": 2.125,
+      "learning_rate": 3.1515143335695274e-08,
+      "loss": 0.9145,
+      "step": 3910
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 1.9921875,
+      "learning_rate": 2.4972035837299567e-08,
+      "loss": 0.8195,
+      "step": 3920
+    },
+    {
+      "epoch": 0.9825,
+      "grad_norm": 2.546875,
+      "learning_rate": 1.9188565338265073e-08,
+      "loss": 0.8227,
+      "step": 3930
+    },
+    {
+      "epoch": 0.985,
+      "grad_norm": 1.9765625,
+      "learning_rate": 1.416517227226355e-08,
+      "loss": 0.9078,
+      "step": 3940
+    },
+    {
+      "epoch": 0.9875,
+      "grad_norm": 2.953125,
+      "learning_rate": 9.902239190124896e-09,
+      "loss": 0.8301,
+      "step": 3950
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 1.7890625,
+      "learning_rate": 6.400090730708242e-09,
+      "loss": 0.8906,
+      "step": 3960
+    },
+    {
+      "epoch": 0.9925,
+      "grad_norm": 1.8359375,
+      "learning_rate": 3.6589935961772736e-09,
+      "loss": 0.8465,
+      "step": 3970
+    },
+    {
+      "epoch": 0.995,
+      "grad_norm": 1.9140625,
+      "learning_rate": 1.6791565316920388e-09,
+      "loss": 0.8723,
+      "step": 3980
+    },
+    {
+      "epoch": 0.9975,
+      "grad_norm": 1.6796875,
+      "learning_rate": 4.6073030951054866e-10,
+      "loss": 0.8043,
+      "step": 3990
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 2.015625,
+      "learning_rate": 3.807717505743824e-12,
+      "loss": 0.907,
+      "step": 4000
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 4000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 9223372036854775807,
+  "save_steps": 2000,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 5.0307891462144e+17,
+  "train_batch_size": 32,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-4000/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f78cefbcf499b12b200d26ae4003ea029a600cba52d0aaf5224decdf492db0ab
+size 5368