Upload folder using huggingface_hub

.gitattributes

@@ -34,3 +34,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
 200M/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+400M/tokenizer.json filter=lfs diff=lfs merge=lfs -text

400M/chat_template.jinja

@@ -0,0 +1 @@
+{% if messages[0]['role'] == 'system' %}{% set loop_messages = messages[1:] %}{% set system_message = messages[0]['content'] %}{% else %}{% set loop_messages = messages %}{% endif %}{% if system_message is defined %}{{ system_message }}{% endif %}{% for message in loop_messages %}{% set content = message['content'] %}{% if message['role'] == 'user' %}{{ content }}{% elif message['role'] == 'assistant' %}{{ content }}{% endif %}{% endfor %}

400M/config.json

@@ -0,0 +1,170 @@
+{
+  "architectures": [
+    "Gemma4UnifiedForConditionalGeneration"
+  ],
+  "audio_config": {
+    "_name_or_path": "",
+    "architectures": null,
+    "audio_embed_dim": 640,
+    "chunk_size_feed_forward": 0,
+    "dtype": "bfloat16",
+    "id2label": {
+      "0": "LABEL_0",
+      "1": "LABEL_1"
+    },
+    "initializer_range": 0.02,
+    "is_encoder_decoder": false,
+    "label2id": {
+      "LABEL_0": 0,
+      "LABEL_1": 1
+    },
+    "model_type": "gemma4_unified_audio",
+    "output_attentions": false,
+    "output_hidden_states": false,
+    "problem_type": null,
+    "return_dict": true,
+    "rms_norm_eps": 1e-06
+  },
+  "audio_token_id": 258881,
+  "boa_token_id": 256000,
+  "boi_token_id": 255999,
+  "bos_token_id": 2,
+  "dtype": "bfloat16",
+  "eoa_token_index": 258883,
+  "eoi_token_id": 258882,
+  "eos_token_id": 1,
+  "hidden_size": 3840,
+  "image_token_id": 258880,
+  "initializer_range": 0.02,
+  "model_type": "gemma4_unified",
+  "pad_token_id": 0,
+  "text_config": {
+    "attention_bias": false,
+    "attention_dropout": 0.0,
+    "attention_k_eq_v": true,
+    "bos_token_id": 2,
+    "dtype": "bfloat16",
+    "enable_moe_block": false,
+    "eos_token_id": 1,
+    "final_logit_softcapping": 30.0,
+    "global_head_dim": 512,
+    "head_dim": 256,
+    "hidden_activation": "gelu_pytorch_tanh",
+    "hidden_size": 3840,
+    "hidden_size_per_layer_input": 0,
+    "initializer_range": 0.02,
+    "intermediate_size": 15360,
+    "layer_types": [
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "full_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "full_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "full_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "full_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "full_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "full_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "full_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "full_attention"
+    ],
+    "max_position_embeddings": 262144,
+    "model_type": "gemma4_unified_text",
+    "moe_intermediate_size": null,
+    "num_attention_heads": 16,
+    "num_experts": null,
+    "num_global_key_value_heads": 1,
+    "num_hidden_layers": 48,
+    "num_key_value_heads": 8,
+    "num_kv_shared_layers": 0,
+    "pad_token_id": 0,
+    "rms_norm_eps": 1e-06,
+    "rope_parameters": {
+      "full_attention": {
+        "partial_rotary_factor": 0.25,
+        "rope_theta": 1000000.0,
+        "rope_type": "proportional"
+      },
+      "sliding_attention": {
+        "rope_theta": 10000.0,
+        "rope_type": "default"
+      }
+    },
+    "sliding_window": 1024,
+    "tie_word_embeddings": true,
+    "top_k_experts": null,
+    "use_bidirectional_attention": "vision",
+    "use_cache": false,
+    "use_double_wide_mlp": false,
+    "vocab_size": 262144,
+    "vocab_size_per_layer_input": 262144
+  },
+  "tie_word_embeddings": true,
+  "transformers_version": "5.10.0",
+  "use_cache": false,
+  "video_token_id": 258884,
+  "vision_config": {
+    "_name_or_path": "",
+    "architectures": null,
+    "chunk_size_feed_forward": 0,
+    "dtype": "bfloat16",
+    "id2label": {
+      "0": "LABEL_0",
+      "1": "LABEL_1"
+    },
+    "initializer_range": 0.02,
+    "is_encoder_decoder": false,
+    "label2id": {
+      "LABEL_0": 0,
+      "LABEL_1": 1
+    },
+    "mm_embed_dim": 3840,
+    "mm_posemb_size": 1120,
+    "model_type": "gemma4_unified_vision",
+    "num_soft_tokens": 280,
+    "output_attentions": false,
+    "output_hidden_states": false,
+    "output_proj_dims": 3840,
+    "patch_size": 16,
+    "pooling_kernel_size": 3,
+    "problem_type": null,
+    "return_dict": true,
+    "rms_norm_eps": 1e-06
+  }
+}

400M/generation_config.json

@@ -0,0 +1,17 @@
+{
+  "bos_token_id": 2,
+  "do_sample": true,
+  "eos_token_id": [
+    1,
+    1
+  ],
+  "pad_token_id": 0,
+  "suppress_tokens": [
+    258883,
+    258882
+  ],
+  "temperature": 1.0,
+  "top_k": 64,
+  "top_p": 0.95,
+  "transformers_version": "5.10.0"
+}

400M/latest

@@ -0,0 +1 @@
+global_step4070

400M/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:033b9e23da8a64cea86117d9944acd05f08b8b3189af4fa20d1bdc1a84706594
+size 23919549408

400M/processor_config.json

@@ -0,0 +1,62 @@
+{
+  "audio_ms_per_token": 40,
+  "audio_seq_length": 750,
+  "feature_extractor": {
+    "audio_samples_per_token": 640,
+    "feature_extractor_type": "Gemma4UnifiedAudioFeatureExtractor",
+    "feature_size": 640,
+    "padding_side": "right",
+    "padding_value": 0.0,
+    "return_attention_mask": true,
+    "sampling_rate": 16000
+  },
+  "image_processor": {
+    "do_convert_rgb": true,
+    "do_normalize": false,
+    "do_rescale": true,
+    "do_resize": true,
+    "image_mean": [
+      0.0,
+      0.0,
+      0.0
+    ],
+    "image_processor_type": "Gemma4UnifiedImageProcessor",
+    "image_std": [
+      1.0,
+      1.0,
+      1.0
+    ],
+    "max_soft_tokens": 280,
+    "patch_size": 16,
+    "pooling_kernel_size": 3,
+    "resample": 3,
+    "rescale_factor": 0.00392156862745098
+  },
+  "image_seq_length": 280,
+  "processor_class": "Gemma4UnifiedProcessor",
+  "video_processor": {
+    "do_convert_rgb": true,
+    "do_normalize": true,
+    "do_rescale": true,
+    "do_resize": true,
+    "do_sample_frames": true,
+    "image_mean": [
+      0.0,
+      0.0,
+      0.0
+    ],
+    "image_std": [
+      1.0,
+      1.0,
+      1.0
+    ],
+    "max_soft_tokens": 70,
+    "num_frames": 32,
+    "patch_size": 16,
+    "pooling_kernel_size": 3,
+    "resample": 3,
+    "rescale_factor": 0.00392156862745098,
+    "return_metadata": false,
+    "video_processor_type": "Gemma4UnifiedVideoProcessor"
+  }
+}

400M/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4d7dcf54a8dbd1fff243640ddc034149fbe7c90ed70b8a95a6bb093e989a7577
+size 14469

400M/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f2a2b0b9d330140afd742a57df45106276d425d38a78e53b0fd43dba9559f8d4
+size 14469

400M/rng_state_10.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2ce222f8f11fd62fd6cb82264500aefbc9e7ef16907ed3e4aa3a8a99c21d95d3
+size 14476

400M/rng_state_11.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d07144bfc4ed675dda70ab66923d17d6f9233560b2db09689cbe1f262c590c88
+size 14476

400M/rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f5db634f6ed5151d3c898fbe3c365f950516c404bed926de16b253b26e63c7cf
+size 14469

400M/rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ad774eade44525098fff2d5910e3d16565fca4c2af29ed0bc4d9315a8ac9607a
+size 14469

400M/rng_state_4.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ebdee1b37be2238764bcfeed9dad268e938248152b0fba60eaf89adda99eac0f
+size 14469

400M/rng_state_5.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6046157bd3d69a7cf6ee508b7359b30cb71488a11b11b07a614041b06813c553
+size 14469

400M/rng_state_6.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ebc9300f101a324d0361f01aec1d57e26b57a703b4e536f4fe982c8ba57a9144
+size 14469

400M/rng_state_7.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:357df7b609d6ffd7f22b998281f7856a711ee83505e5ed2835959554b0185df7
+size 14469

400M/rng_state_8.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4fb33667d752a94ade6841f62b8b8fcc22256b4712bd568b141bc690d4a42e11
+size 14469

400M/rng_state_9.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:354a7a38e97ba11b8ad29bf4d00d74f92295b32124cff0ef46baa862aa9b0f2f
+size 14469

400M/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c373bca4c9a6ef963a59ee82f3b16a9a90b89b812375170f487261fe4db9e20a
+size 1465

400M/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:12bac982b793c44b03d52a250a9f0d0b666813da566b910c24a6da0695fd11e6
+size 32170070

400M/tokenizer_config.json

@@ -0,0 +1,56 @@
+{
+  "audio_token": "<|audio|>",
+  "backend": "tokenizers",
+  "boa_token": "<|audio>",
+  "boi_token": "<|image>",
+  "bos_token": "<bos>",
+  "eoa_token": "<audio|>",
+  "eoc_token": "<channel|>",
+  "eoi_token": "<image|>",
+  "eos_token": "<eos>",
+  "eot_token": "<turn|>",
+  "escape_token": "<|\"|>",
+  "etc_token": "<tool_call|>",
+  "etd_token": "<tool|>",
+  "etr_token": "<tool_response|>",
+  "extra_special_tokens": [
+    "<|video|>"
+  ],
+  "image_token": "<|image|>",
+  "is_local": true,
+  "local_files_only": true,
+  "mask_token": "<mask>",
+  "model_max_length": 1000000000000000019884624838656,
+  "model_specific_special_tokens": {
+    "audio_token": "<|audio|>",
+    "boa_token": "<|audio>",
+    "boi_token": "<|image>",
+    "eoa_token": "<audio|>",
+    "eoc_token": "<channel|>",
+    "eoi_token": "<image|>",
+    "eot_token": "<turn|>",
+    "escape_token": "<|\"|>",
+    "etc_token": "<tool_call|>",
+    "etd_token": "<tool|>",
+    "etr_token": "<tool_response|>",
+    "image_token": "<|image|>",
+    "soc_token": "<|channel>",
+    "sot_token": "<|turn>",
+    "stc_token": "<|tool_call>",
+    "std_token": "<|tool>",
+    "str_token": "<|tool_response>",
+    "think_token": "<|think|>"
+  },
+  "pad_token": "<pad>",
+  "padding_side": "right",
+  "processor_class": "Gemma4UnifiedProcessor",
+  "soc_token": "<|channel>",
+  "sot_token": "<|turn>",
+  "split_special_tokens": false,
+  "stc_token": "<|tool_call>",
+  "std_token": "<|tool>",
+  "str_token": "<|tool_response>",
+  "think_token": "<|think|>",
+  "tokenizer_class": "GemmaTokenizer",
+  "unk_token": "<unk>"
+}

400M/trainer_state.json

@@ -0,0 +1,2883 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.19622968998601803,
+  "eval_steps": 500,
+  "global_step": 4070,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0004821368304324767,
+      "grad_norm": 84.9173377406223,
+      "learning_rate": 2.910112359550562e-06,
+      "loss": 1.5154205322265626,
+      "step": 10
+    },
+    {
+      "epoch": 0.0009642736608649534,
+      "grad_norm": 57.560206229549436,
+      "learning_rate": 3.921348314606742e-06,
+      "loss": 0.9048759460449218,
+      "step": 20
+    },
+    {
+      "epoch": 0.0014464104912974302,
+      "grad_norm": 241.93706017750705,
+      "learning_rate": 4.932584269662922e-06,
+      "loss": 0.7844394207000732,
+      "step": 30
+    },
+    {
+      "epoch": 0.001928547321729907,
+      "grad_norm": 635.4867043743025,
+      "learning_rate": 5.943820224719102e-06,
+      "loss": 0.777440071105957,
+      "step": 40
+    },
+    {
+      "epoch": 0.0024106841521623836,
+      "grad_norm": 16.70216416699672,
+      "learning_rate": 6.955056179775282e-06,
+      "loss": 0.7813506126403809,
+      "step": 50
+    },
+    {
+      "epoch": 0.0028928209825948605,
+      "grad_norm": 104.65219520703347,
+      "learning_rate": 7.966292134831462e-06,
+      "loss": 0.7643991470336914,
+      "step": 60
+    },
+    {
+      "epoch": 0.0033749578130273373,
+      "grad_norm": 15.464766046614894,
+      "learning_rate": 8.977528089887641e-06,
+      "loss": 0.7663109779357911,
+      "step": 70
+    },
+    {
+      "epoch": 0.003857094643459814,
+      "grad_norm": 11.140861449353931,
+      "learning_rate": 9.988764044943822e-06,
+      "loss": 0.7194852352142334,
+      "step": 80
+    },
+    {
+      "epoch": 0.004339231473892291,
+      "grad_norm": 29.54062487849426,
+      "learning_rate": 1.1000000000000001e-05,
+      "loss": 0.7484091758728028,
+      "step": 90
+    },
+    {
+      "epoch": 0.004821368304324767,
+      "grad_norm": 623.2770461417654,
+      "learning_rate": 1.2011235955056182e-05,
+      "loss": 0.7809892177581788,
+      "step": 100
+    },
+    {
+      "epoch": 0.0053035051347572445,
+      "grad_norm": 12.453456400945432,
+      "learning_rate": 1.3022471910112362e-05,
+      "loss": 0.7879758358001709,
+      "step": 110
+    },
+    {
+      "epoch": 0.005785641965189721,
+      "grad_norm": 15.266718921768598,
+      "learning_rate": 1.403370786516854e-05,
+      "loss": 0.7485491752624511,
+      "step": 120
+    },
+    {
+      "epoch": 0.006267778795622197,
+      "grad_norm": 7.765117181857264,
+      "learning_rate": 1.504494382022472e-05,
+      "loss": 0.7564555168151855,
+      "step": 130
+    },
+    {
+      "epoch": 0.006749915626054675,
+      "grad_norm": 12.596599218999913,
+      "learning_rate": 1.60561797752809e-05,
+      "loss": 0.7486157894134522,
+      "step": 140
+    },
+    {
+      "epoch": 0.007232052456487151,
+      "grad_norm": 4.964019715699265,
+      "learning_rate": 1.706741573033708e-05,
+      "loss": 0.6995878219604492,
+      "step": 150
+    },
+    {
+      "epoch": 0.007714189286919628,
+      "grad_norm": 5.34953585002889,
+      "learning_rate": 1.8078651685393256e-05,
+      "loss": 0.7818653106689453,
+      "step": 160
+    },
+    {
+      "epoch": 0.008196326117352104,
+      "grad_norm": 13.677128436034314,
+      "learning_rate": 1.908988764044944e-05,
+      "loss": 0.7409422874450684,
+      "step": 170
+    },
+    {
+      "epoch": 0.008678462947784581,
+      "grad_norm": 11.253318513799217,
+      "learning_rate": 2e-05,
+      "loss": 0.7045968532562256,
+      "step": 180
+    },
+    {
+      "epoch": 0.009160599778217059,
+      "grad_norm": 3.7993902470656638,
+      "learning_rate": 2e-05,
+      "loss": 0.7415530204772949,
+      "step": 190
+    },
+    {
+      "epoch": 0.009642736608649534,
+      "grad_norm": 9.932013471995777,
+      "learning_rate": 2e-05,
+      "loss": 0.7458738327026367,
+      "step": 200
+    },
+    {
+      "epoch": 0.010124873439082012,
+      "grad_norm": 4.001842841007721,
+      "learning_rate": 2e-05,
+      "loss": 0.7419017791748047,
+      "step": 210
+    },
+    {
+      "epoch": 0.010607010269514489,
+      "grad_norm": 5.8081003010500405,
+      "learning_rate": 2e-05,
+      "loss": 0.727513599395752,
+      "step": 220
+    },
+    {
+      "epoch": 0.011089147099946964,
+      "grad_norm": 24.899690781901665,
+      "learning_rate": 2e-05,
+      "loss": 0.7438565731048584,
+      "step": 230
+    },
+    {
+      "epoch": 0.011571283930379442,
+      "grad_norm": 9.431012313223512,
+      "learning_rate": 2e-05,
+      "loss": 0.7601432800292969,
+      "step": 240
+    },
+    {
+      "epoch": 0.012053420760811919,
+      "grad_norm": 9.9069654154098,
+      "learning_rate": 2e-05,
+      "loss": 0.7237838268280029,
+      "step": 250
+    },
+    {
+      "epoch": 0.012535557591244395,
+      "grad_norm": 6.065436761140742,
+      "learning_rate": 2e-05,
+      "loss": 0.7469722747802734,
+      "step": 260
+    },
+    {
+      "epoch": 0.013017694421676872,
+      "grad_norm": 47.854432068582106,
+      "learning_rate": 2e-05,
+      "loss": 0.7887792587280273,
+      "step": 270
+    },
+    {
+      "epoch": 0.01349983125210935,
+      "grad_norm": 7.739139876114809,
+      "learning_rate": 2e-05,
+      "loss": 0.8068546295166016,
+      "step": 280
+    },
+    {
+      "epoch": 0.013981968082541825,
+      "grad_norm": 6.596407728810894,
+      "learning_rate": 2e-05,
+      "loss": 0.7717693328857422,
+      "step": 290
+    },
+    {
+      "epoch": 0.014464104912974302,
+      "grad_norm": 2.821767608514682,
+      "learning_rate": 2e-05,
+      "loss": 0.762939453125,
+      "step": 300
+    },
+    {
+      "epoch": 0.01494624174340678,
+      "grad_norm": 12.92389489442801,
+      "learning_rate": 2e-05,
+      "loss": 0.7905154705047608,
+      "step": 310
+    },
+    {
+      "epoch": 0.015428378573839255,
+      "grad_norm": 4.45893615790758,
+      "learning_rate": 2e-05,
+      "loss": 0.8191452026367188,
+      "step": 320
+    },
+    {
+      "epoch": 0.01591051540427173,
+      "grad_norm": 2.719882613980345,
+      "learning_rate": 2e-05,
+      "loss": 0.7672964096069336,
+      "step": 330
+    },
+    {
+      "epoch": 0.016392652234704208,
+      "grad_norm": 9.433039678311145,
+      "learning_rate": 2e-05,
+      "loss": 0.7420453548431396,
+      "step": 340
+    },
+    {
+      "epoch": 0.016874789065136685,
+      "grad_norm": 4.012690253492805,
+      "learning_rate": 2e-05,
+      "loss": 0.7563004493713379,
+      "step": 350
+    },
+    {
+      "epoch": 0.017356925895569163,
+      "grad_norm": 11.624966196988954,
+      "learning_rate": 2e-05,
+      "loss": 0.7373766899108887,
+      "step": 360
+    },
+    {
+      "epoch": 0.01783906272600164,
+      "grad_norm": 2.7344222258766258,
+      "learning_rate": 2e-05,
+      "loss": 0.7051375389099122,
+      "step": 370
+    },
+    {
+      "epoch": 0.018321199556434117,
+      "grad_norm": 4.539182866411412,
+      "learning_rate": 2e-05,
+      "loss": 0.7277828216552734,
+      "step": 380
+    },
+    {
+      "epoch": 0.01880333638686659,
+      "grad_norm": 4.87684609900965,
+      "learning_rate": 2e-05,
+      "loss": 0.7499900817871094,
+      "step": 390
+    },
+    {
+      "epoch": 0.01928547321729907,
+      "grad_norm": 2.305121309654471,
+      "learning_rate": 2e-05,
+      "loss": 0.7218913078308106,
+      "step": 400
+    },
+    {
+      "epoch": 0.019767610047731546,
+      "grad_norm": 2.637730882882061,
+      "learning_rate": 2e-05,
+      "loss": 0.7323745250701904,
+      "step": 410
+    },
+    {
+      "epoch": 0.020249746878164023,
+      "grad_norm": 2.9314732121215092,
+      "learning_rate": 2e-05,
+      "loss": 0.7070964336395263,
+      "step": 420
+    },
+    {
+      "epoch": 0.0207318837085965,
+      "grad_norm": 4.051033736245248,
+      "learning_rate": 2e-05,
+      "loss": 0.7526719570159912,
+      "step": 430
+    },
+    {
+      "epoch": 0.021214020539028978,
+      "grad_norm": 2.737987761268551,
+      "learning_rate": 2e-05,
+      "loss": 0.7400177001953125,
+      "step": 440
+    },
+    {
+      "epoch": 0.02169615736946145,
+      "grad_norm": 2.002819925510059,
+      "learning_rate": 2e-05,
+      "loss": 0.7192610740661621,
+      "step": 450
+    },
+    {
+      "epoch": 0.02217829419989393,
+      "grad_norm": 3.7426192848860755,
+      "learning_rate": 2e-05,
+      "loss": 0.7366347312927246,
+      "step": 460
+    },
+    {
+      "epoch": 0.022660431030326406,
+      "grad_norm": 1.1926626716478357,
+      "learning_rate": 2e-05,
+      "loss": 0.7299184322357177,
+      "step": 470
+    },
+    {
+      "epoch": 0.023142567860758884,
+      "grad_norm": 3.367060431673956,
+      "learning_rate": 2e-05,
+      "loss": 0.7061259269714355,
+      "step": 480
+    },
+    {
+      "epoch": 0.02362470469119136,
+      "grad_norm": 1.0695774993776388,
+      "learning_rate": 2e-05,
+      "loss": 0.7048762321472168,
+      "step": 490
+    },
+    {
+      "epoch": 0.024106841521623838,
+      "grad_norm": 1.9484852417237744,
+      "learning_rate": 2e-05,
+      "loss": 0.7045268058776856,
+      "step": 500
+    },
+    {
+      "epoch": 0.024588978352056312,
+      "grad_norm": 1.7161008623305876,
+      "learning_rate": 2e-05,
+      "loss": 0.7366765022277832,
+      "step": 510
+    },
+    {
+      "epoch": 0.02507111518248879,
+      "grad_norm": 2.7468813266571335,
+      "learning_rate": 2e-05,
+      "loss": 0.7303312301635743,
+      "step": 520
+    },
+    {
+      "epoch": 0.025553252012921267,
+      "grad_norm": 8.640342170047546,
+      "learning_rate": 2e-05,
+      "loss": 0.76505126953125,
+      "step": 530
+    },
+    {
+      "epoch": 0.026035388843353744,
+      "grad_norm": 14.403207471659123,
+      "learning_rate": 2e-05,
+      "loss": 0.7315347194671631,
+      "step": 540
+    },
+    {
+      "epoch": 0.02651752567378622,
+      "grad_norm": 4.9187469514726025,
+      "learning_rate": 2e-05,
+      "loss": 0.7313735961914063,
+      "step": 550
+    },
+    {
+      "epoch": 0.0269996625042187,
+      "grad_norm": 21.09742751886116,
+      "learning_rate": 2e-05,
+      "loss": 0.7265950679779053,
+      "step": 560
+    },
+    {
+      "epoch": 0.027481799334651173,
+      "grad_norm": 9.120234530490755,
+      "learning_rate": 2e-05,
+      "loss": 0.752630615234375,
+      "step": 570
+    },
+    {
+      "epoch": 0.02796393616508365,
+      "grad_norm": 39.217881534027256,
+      "learning_rate": 2e-05,
+      "loss": 0.7522613048553467,
+      "step": 580
+    },
+    {
+      "epoch": 0.028446072995516127,
+      "grad_norm": 15.15469395513839,
+      "learning_rate": 2e-05,
+      "loss": 0.7306804656982422,
+      "step": 590
+    },
+    {
+      "epoch": 0.028928209825948605,
+      "grad_norm": 2.229860272443176,
+      "learning_rate": 2e-05,
+      "loss": 0.7174323081970215,
+      "step": 600
+    },
+    {
+      "epoch": 0.029410346656381082,
+      "grad_norm": 28.87034725965148,
+      "learning_rate": 2e-05,
+      "loss": 0.7655416488647461,
+      "step": 610
+    },
+    {
+      "epoch": 0.02989248348681356,
+      "grad_norm": 2.85440650717933,
+      "learning_rate": 2e-05,
+      "loss": 0.7155353546142578,
+      "step": 620
+    },
+    {
+      "epoch": 0.030374620317246033,
+      "grad_norm": 4.824295325677904,
+      "learning_rate": 2e-05,
+      "loss": 0.7334116935729981,
+      "step": 630
+    },
+    {
+      "epoch": 0.03085675714767851,
+      "grad_norm": 6.533365855916318,
+      "learning_rate": 2e-05,
+      "loss": 0.7016949653625488,
+      "step": 640
+    },
+    {
+      "epoch": 0.03133889397811099,
+      "grad_norm": 16.877036489930155,
+      "learning_rate": 2e-05,
+      "loss": 0.7247958660125733,
+      "step": 650
+    },
+    {
+      "epoch": 0.03182103080854346,
+      "grad_norm": 329.2790796078927,
+      "learning_rate": 2e-05,
+      "loss": 0.7393420219421387,
+      "step": 660
+    },
+    {
+      "epoch": 0.03230316763897594,
+      "grad_norm": 3.076950012733951,
+      "learning_rate": 2e-05,
+      "loss": 0.762078857421875,
+      "step": 670
+    },
+    {
+      "epoch": 0.032785304469408416,
+      "grad_norm": 4.030194347977115,
+      "learning_rate": 2e-05,
+      "loss": 0.7703542232513427,
+      "step": 680
+    },
+    {
+      "epoch": 0.0332674412998409,
+      "grad_norm": 1.813412222853007,
+      "learning_rate": 2e-05,
+      "loss": 0.7255609035491943,
+      "step": 690
+    },
+    {
+      "epoch": 0.03374957813027337,
+      "grad_norm": 3.3819806682901143,
+      "learning_rate": 2e-05,
+      "loss": 0.7121047973632812,
+      "step": 700
+    },
+    {
+      "epoch": 0.03423171496070585,
+      "grad_norm": 2.3872601241148304,
+      "learning_rate": 2e-05,
+      "loss": 0.706441068649292,
+      "step": 710
+    },
+    {
+      "epoch": 0.034713851791138325,
+      "grad_norm": 3.5426833842605157,
+      "learning_rate": 2e-05,
+      "loss": 0.7012114524841309,
+      "step": 720
+    },
+    {
+      "epoch": 0.0351959886215708,
+      "grad_norm": 4.564258584135792,
+      "learning_rate": 2e-05,
+      "loss": 0.723219633102417,
+      "step": 730
+    },
+    {
+      "epoch": 0.03567812545200328,
+      "grad_norm": 3.4441825412487086,
+      "learning_rate": 2e-05,
+      "loss": 0.6977805137634278,
+      "step": 740
+    },
+    {
+      "epoch": 0.036160262282435754,
+      "grad_norm": 3.8546226650995714,
+      "learning_rate": 2e-05,
+      "loss": 0.7184113979339599,
+      "step": 750
+    },
+    {
+      "epoch": 0.036642399112868235,
+      "grad_norm": 16.213966783350067,
+      "learning_rate": 2e-05,
+      "loss": 0.7428165435791015,
+      "step": 760
+    },
+    {
+      "epoch": 0.03712453594330071,
+      "grad_norm": 2.0741210758941913,
+      "learning_rate": 2e-05,
+      "loss": 0.7358497619628906,
+      "step": 770
+    },
+    {
+      "epoch": 0.03760667277373318,
+      "grad_norm": 2.5202301046172417,
+      "learning_rate": 2e-05,
+      "loss": 0.7103074550628662,
+      "step": 780
+    },
+    {
+      "epoch": 0.03808880960416566,
+      "grad_norm": 4.106110045969379,
+      "learning_rate": 2e-05,
+      "loss": 0.713107967376709,
+      "step": 790
+    },
+    {
+      "epoch": 0.03857094643459814,
+      "grad_norm": 3.088072216074999,
+      "learning_rate": 2e-05,
+      "loss": 0.7298454284667969,
+      "step": 800
+    },
+    {
+      "epoch": 0.03905308326503062,
+      "grad_norm": 2.171527717000864,
+      "learning_rate": 2e-05,
+      "loss": 0.7251861095428467,
+      "step": 810
+    },
+    {
+      "epoch": 0.03953522009546309,
+      "grad_norm": 2.0872404223379406,
+      "learning_rate": 2e-05,
+      "loss": 0.7178170680999756,
+      "step": 820
+    },
+    {
+      "epoch": 0.04001735692589557,
+      "grad_norm": 2.085406019030795,
+      "learning_rate": 2e-05,
+      "loss": 0.6847468376159668,
+      "step": 830
+    },
+    {
+      "epoch": 0.040499493756328046,
+      "grad_norm": 6.295654776698936,
+      "learning_rate": 2e-05,
+      "loss": 0.7038663387298584,
+      "step": 840
+    },
+    {
+      "epoch": 0.04098163058676052,
+      "grad_norm": 2.454918336179734,
+      "learning_rate": 2e-05,
+      "loss": 0.7551318645477295,
+      "step": 850
+    },
+    {
+      "epoch": 0.041463767417193,
+      "grad_norm": 2.5362213989256044,
+      "learning_rate": 2e-05,
+      "loss": 0.7268366813659668,
+      "step": 860
+    },
+    {
+      "epoch": 0.041945904247625475,
+      "grad_norm": 2.0309978712081254,
+      "learning_rate": 2e-05,
+      "loss": 0.725460433959961,
+      "step": 870
+    },
+    {
+      "epoch": 0.042428041078057956,
+      "grad_norm": 2.726392751612776,
+      "learning_rate": 2e-05,
+      "loss": 0.7183332443237305,
+      "step": 880
+    },
+    {
+      "epoch": 0.04291017790849043,
+      "grad_norm": 2.2408894596393836,
+      "learning_rate": 2e-05,
+      "loss": 0.7466308116912842,
+      "step": 890
+    },
+    {
+      "epoch": 0.0433923147389229,
+      "grad_norm": 8.581913026001844,
+      "learning_rate": 2e-05,
+      "loss": 0.67788667678833,
+      "step": 900
+    },
+    {
+      "epoch": 0.043874451569355384,
+      "grad_norm": 14.329361671424175,
+      "learning_rate": 2e-05,
+      "loss": 0.6972944259643554,
+      "step": 910
+    },
+    {
+      "epoch": 0.04435658839978786,
+      "grad_norm": 3.9088010842539327,
+      "learning_rate": 2e-05,
+      "loss": 0.699748182296753,
+      "step": 920
+    },
+    {
+      "epoch": 0.04483872523022034,
+      "grad_norm": 2.221525537997574,
+      "learning_rate": 2e-05,
+      "loss": 0.7268280982971191,
+      "step": 930
+    },
+    {
+      "epoch": 0.04532086206065281,
+      "grad_norm": 2.273319630076157,
+      "learning_rate": 2e-05,
+      "loss": 0.7030668258666992,
+      "step": 940
+    },
+    {
+      "epoch": 0.04580299889108529,
+      "grad_norm": 2.8492569998023027,
+      "learning_rate": 2e-05,
+      "loss": 0.7284522533416748,
+      "step": 950
+    },
+    {
+      "epoch": 0.04628513572151777,
+      "grad_norm": 3.6240853365585473,
+      "learning_rate": 2e-05,
+      "loss": 0.6911632061004639,
+      "step": 960
+    },
+    {
+      "epoch": 0.04676727255195024,
+      "grad_norm": 4.035089048393918,
+      "learning_rate": 2e-05,
+      "loss": 0.7237444877624511,
+      "step": 970
+    },
+    {
+      "epoch": 0.04724940938238272,
+      "grad_norm": 1.7484668255186857,
+      "learning_rate": 2e-05,
+      "loss": 0.7165475368499756,
+      "step": 980
+    },
+    {
+      "epoch": 0.047731546212815196,
+      "grad_norm": 3.6565231516157466,
+      "learning_rate": 2e-05,
+      "loss": 0.7023948669433594,
+      "step": 990
+    },
+    {
+      "epoch": 0.048213683043247677,
+      "grad_norm": 8.83591745574243,
+      "learning_rate": 2e-05,
+      "loss": 0.714229679107666,
+      "step": 1000
+    },
+    {
+      "epoch": 0.04869581987368015,
+      "grad_norm": 1.5477438520033022,
+      "learning_rate": 2e-05,
+      "loss": 0.6836585998535156,
+      "step": 1010
+    },
+    {
+      "epoch": 0.049177956704112624,
+      "grad_norm": 1.7108650020940415,
+      "learning_rate": 2e-05,
+      "loss": 0.6963685512542724,
+      "step": 1020
+    },
+    {
+      "epoch": 0.049660093534545105,
+      "grad_norm": 3.8025775202775516,
+      "learning_rate": 2e-05,
+      "loss": 0.7289498329162598,
+      "step": 1030
+    },
+    {
+      "epoch": 0.05014223036497758,
+      "grad_norm": 1.6346781298258524,
+      "learning_rate": 2e-05,
+      "loss": 0.7262353897094727,
+      "step": 1040
+    },
+    {
+      "epoch": 0.05062436719541006,
+      "grad_norm": 3.386679850071031,
+      "learning_rate": 2e-05,
+      "loss": 0.6808836936950684,
+      "step": 1050
+    },
+    {
+      "epoch": 0.051106504025842533,
+      "grad_norm": 2.3953301062602854,
+      "learning_rate": 2e-05,
+      "loss": 0.7089850425720214,
+      "step": 1060
+    },
+    {
+      "epoch": 0.051588640856275014,
+      "grad_norm": 2.5139615513318763,
+      "learning_rate": 2e-05,
+      "loss": 0.7181517601013183,
+      "step": 1070
+    },
+    {
+      "epoch": 0.05207077768670749,
+      "grad_norm": 2.8534620949787053,
+      "learning_rate": 2e-05,
+      "loss": 0.6958971977233886,
+      "step": 1080
+    },
+    {
+      "epoch": 0.05255291451713996,
+      "grad_norm": 1.9564028862046592,
+      "learning_rate": 2e-05,
+      "loss": 0.7039243698120117,
+      "step": 1090
+    },
+    {
+      "epoch": 0.05303505134757244,
+      "grad_norm": 1.557897571396239,
+      "learning_rate": 2e-05,
+      "loss": 0.7114754676818847,
+      "step": 1100
+    },
+    {
+      "epoch": 0.05351718817800492,
+      "grad_norm": 2.930498345415913,
+      "learning_rate": 2e-05,
+      "loss": 0.7031359672546387,
+      "step": 1110
+    },
+    {
+      "epoch": 0.0539993250084374,
+      "grad_norm": 1.4602419678677463,
+      "learning_rate": 2e-05,
+      "loss": 0.7077861785888672,
+      "step": 1120
+    },
+    {
+      "epoch": 0.05448146183886987,
+      "grad_norm": 1.7319226933333385,
+      "learning_rate": 2e-05,
+      "loss": 0.7047904968261719,
+      "step": 1130
+    },
+    {
+      "epoch": 0.054963598669302345,
+      "grad_norm": 2.520611756871299,
+      "learning_rate": 2e-05,
+      "loss": 0.7063097953796387,
+      "step": 1140
+    },
+    {
+      "epoch": 0.055445735499734826,
+      "grad_norm": 3.214756094588082,
+      "learning_rate": 2e-05,
+      "loss": 0.6928376197814942,
+      "step": 1150
+    },
+    {
+      "epoch": 0.0559278723301673,
+      "grad_norm": 2.1913033139614546,
+      "learning_rate": 2e-05,
+      "loss": 0.7101710319519043,
+      "step": 1160
+    },
+    {
+      "epoch": 0.05641000916059978,
+      "grad_norm": 5.747581126680708,
+      "learning_rate": 2e-05,
+      "loss": 0.7223403453826904,
+      "step": 1170
+    },
+    {
+      "epoch": 0.056892145991032254,
+      "grad_norm": 1.478178600197385,
+      "learning_rate": 2e-05,
+      "loss": 0.7083209991455078,
+      "step": 1180
+    },
+    {
+      "epoch": 0.05737428282146473,
+      "grad_norm": 3.1153048058866166,
+      "learning_rate": 2e-05,
+      "loss": 0.7255118370056153,
+      "step": 1190
+    },
+    {
+      "epoch": 0.05785641965189721,
+      "grad_norm": 6.6985613434510975,
+      "learning_rate": 2e-05,
+      "loss": 0.726648473739624,
+      "step": 1200
+    },
+    {
+      "epoch": 0.05833855648232968,
+      "grad_norm": 5.377966618368178,
+      "learning_rate": 2e-05,
+      "loss": 0.7175424575805665,
+      "step": 1210
+    },
+    {
+      "epoch": 0.058820693312762164,
+      "grad_norm": 3.838988528518626,
+      "learning_rate": 2e-05,
+      "loss": 0.7453009605407714,
+      "step": 1220
+    },
+    {
+      "epoch": 0.05930283014319464,
+      "grad_norm": 2.6228691909450834,
+      "learning_rate": 2e-05,
+      "loss": 0.7184647560119629,
+      "step": 1230
+    },
+    {
+      "epoch": 0.05978496697362712,
+      "grad_norm": 2.6612776295017615,
+      "learning_rate": 2e-05,
+      "loss": 0.7309318542480469,
+      "step": 1240
+    },
+    {
+      "epoch": 0.06026710380405959,
+      "grad_norm": 1.8845329907929567,
+      "learning_rate": 2e-05,
+      "loss": 0.725009822845459,
+      "step": 1250
+    },
+    {
+      "epoch": 0.060749240634492066,
+      "grad_norm": 10.28021708941531,
+      "learning_rate": 2e-05,
+      "loss": 0.7213550567626953,
+      "step": 1260
+    },
+    {
+      "epoch": 0.06123137746492455,
+      "grad_norm": 1.7751708481039612,
+      "learning_rate": 2e-05,
+      "loss": 0.6989108562469483,
+      "step": 1270
+    },
+    {
+      "epoch": 0.06171351429535702,
+      "grad_norm": 2.058733549275171,
+      "learning_rate": 2e-05,
+      "loss": 0.6917882442474366,
+      "step": 1280
+    },
+    {
+      "epoch": 0.0621956511257895,
+      "grad_norm": 5.491929938751169,
+      "learning_rate": 2e-05,
+      "loss": 0.6999297142028809,
+      "step": 1290
+    },
+    {
+      "epoch": 0.06267778795622198,
+      "grad_norm": 6.688351091581926,
+      "learning_rate": 2e-05,
+      "loss": 0.7235936164855957,
+      "step": 1300
+    },
+    {
+      "epoch": 0.06315992478665446,
+      "grad_norm": 4.899979463669896,
+      "learning_rate": 2e-05,
+      "loss": 0.686549711227417,
+      "step": 1310
+    },
+    {
+      "epoch": 0.06364206161708692,
+      "grad_norm": 2.952614374772642,
+      "learning_rate": 2e-05,
+      "loss": 0.7102313041687012,
+      "step": 1320
+    },
+    {
+      "epoch": 0.0641241984475194,
+      "grad_norm": 10.566954823585363,
+      "learning_rate": 2e-05,
+      "loss": 0.7124320983886718,
+      "step": 1330
+    },
+    {
+      "epoch": 0.06460633527795188,
+      "grad_norm": 8.873763489301671,
+      "learning_rate": 2e-05,
+      "loss": 0.7242676734924316,
+      "step": 1340
+    },
+    {
+      "epoch": 0.06508847210838437,
+      "grad_norm": 3.170822217707811,
+      "learning_rate": 2e-05,
+      "loss": 0.709017276763916,
+      "step": 1350
+    },
+    {
+      "epoch": 0.06557060893881683,
+      "grad_norm": 4.133385298753556,
+      "learning_rate": 2e-05,
+      "loss": 0.7052340507507324,
+      "step": 1360
+    },
+    {
+      "epoch": 0.06605274576924931,
+      "grad_norm": 2.3980143582956575,
+      "learning_rate": 2e-05,
+      "loss": 0.7251494884490967,
+      "step": 1370
+    },
+    {
+      "epoch": 0.0665348825996818,
+      "grad_norm": 1.8149128248978232,
+      "learning_rate": 2e-05,
+      "loss": 0.6977263450622558,
+      "step": 1380
+    },
+    {
+      "epoch": 0.06701701943011426,
+      "grad_norm": 3.216048638536296,
+      "learning_rate": 2e-05,
+      "loss": 0.7132225036621094,
+      "step": 1390
+    },
+    {
+      "epoch": 0.06749915626054674,
+      "grad_norm": 2.759251898652736,
+      "learning_rate": 2e-05,
+      "loss": 0.6781229972839355,
+      "step": 1400
+    },
+    {
+      "epoch": 0.06798129309097922,
+      "grad_norm": 2.1615844816494523,
+      "learning_rate": 2e-05,
+      "loss": 0.6893566131591797,
+      "step": 1410
+    },
+    {
+      "epoch": 0.0684634299214117,
+      "grad_norm": 1.269324197852931,
+      "learning_rate": 2e-05,
+      "loss": 0.7151976585388183,
+      "step": 1420
+    },
+    {
+      "epoch": 0.06894556675184417,
+      "grad_norm": 2.475140950996921,
+      "learning_rate": 2e-05,
+      "loss": 0.7208847045898438,
+      "step": 1430
+    },
+    {
+      "epoch": 0.06942770358227665,
+      "grad_norm": 1.8637431609519701,
+      "learning_rate": 2e-05,
+      "loss": 0.6920453548431397,
+      "step": 1440
+    },
+    {
+      "epoch": 0.06990984041270913,
+      "grad_norm": 4.388822465185116,
+      "learning_rate": 2e-05,
+      "loss": 0.7050337791442871,
+      "step": 1450
+    },
+    {
+      "epoch": 0.0703919772431416,
+      "grad_norm": 1.7686798808274666,
+      "learning_rate": 2e-05,
+      "loss": 0.6877009391784668,
+      "step": 1460
+    },
+    {
+      "epoch": 0.07087411407357408,
+      "grad_norm": 2.1589166054135314,
+      "learning_rate": 2e-05,
+      "loss": 0.6840052604675293,
+      "step": 1470
+    },
+    {
+      "epoch": 0.07135625090400656,
+      "grad_norm": 3.4176787553601677,
+      "learning_rate": 2e-05,
+      "loss": 0.7044760704040527,
+      "step": 1480
+    },
+    {
+      "epoch": 0.07183838773443903,
+      "grad_norm": 5.058389900187091,
+      "learning_rate": 2e-05,
+      "loss": 0.7252198219299316,
+      "step": 1490
+    },
+    {
+      "epoch": 0.07232052456487151,
+      "grad_norm": 3.4902103862821328,
+      "learning_rate": 2e-05,
+      "loss": 0.7132100582122802,
+      "step": 1500
+    },
+    {
+      "epoch": 0.07280266139530399,
+      "grad_norm": 6.018184255339297,
+      "learning_rate": 2e-05,
+      "loss": 0.70932936668396,
+      "step": 1510
+    },
+    {
+      "epoch": 0.07328479822573647,
+      "grad_norm": 3.5144758180231968,
+      "learning_rate": 2e-05,
+      "loss": 0.7115926742553711,
+      "step": 1520
+    },
+    {
+      "epoch": 0.07376693505616894,
+      "grad_norm": 6.3900031796310985,
+      "learning_rate": 2e-05,
+      "loss": 0.7138595581054688,
+      "step": 1530
+    },
+    {
+      "epoch": 0.07424907188660142,
+      "grad_norm": 1.830421739776605,
+      "learning_rate": 2e-05,
+      "loss": 0.7291291236877442,
+      "step": 1540
+    },
+    {
+      "epoch": 0.0747312087170339,
+      "grad_norm": 4.409354344559001,
+      "learning_rate": 2e-05,
+      "loss": 0.7407273292541504,
+      "step": 1550
+    },
+    {
+      "epoch": 0.07521334554746636,
+      "grad_norm": 3.4473627534438513,
+      "learning_rate": 2e-05,
+      "loss": 0.7394618511199951,
+      "step": 1560
+    },
+    {
+      "epoch": 0.07569548237789885,
+      "grad_norm": 2.6382177594854754,
+      "learning_rate": 2e-05,
+      "loss": 0.6957945823669434,
+      "step": 1570
+    },
+    {
+      "epoch": 0.07617761920833133,
+      "grad_norm": 3.999056912619744,
+      "learning_rate": 2e-05,
+      "loss": 0.6612532615661622,
+      "step": 1580
+    },
+    {
+      "epoch": 0.07665975603876381,
+      "grad_norm": 3.7247803590530237,
+      "learning_rate": 2e-05,
+      "loss": 0.7023423194885254,
+      "step": 1590
+    },
+    {
+      "epoch": 0.07714189286919627,
+      "grad_norm": 1.8337543544271353,
+      "learning_rate": 2e-05,
+      "loss": 0.7308431148529053,
+      "step": 1600
+    },
+    {
+      "epoch": 0.07762402969962875,
+      "grad_norm": 4.832144300014454,
+      "learning_rate": 2e-05,
+      "loss": 0.7019494533538818,
+      "step": 1610
+    },
+    {
+      "epoch": 0.07810616653006124,
+      "grad_norm": 1.2557192725218083,
+      "learning_rate": 2e-05,
+      "loss": 0.7272754669189453,
+      "step": 1620
+    },
+    {
+      "epoch": 0.0785883033604937,
+      "grad_norm": 1.259086888724499,
+      "learning_rate": 2e-05,
+      "loss": 0.7088966369628906,
+      "step": 1630
+    },
+    {
+      "epoch": 0.07907044019092618,
+      "grad_norm": 1.4497993722505094,
+      "learning_rate": 2e-05,
+      "loss": 0.7055980682373046,
+      "step": 1640
+    },
+    {
+      "epoch": 0.07955257702135866,
+      "grad_norm": 1.9410198139904098,
+      "learning_rate": 2e-05,
+      "loss": 0.7067210674285889,
+      "step": 1650
+    },
+    {
+      "epoch": 0.08003471385179114,
+      "grad_norm": 1.8227117633516707,
+      "learning_rate": 2e-05,
+      "loss": 0.6959988594055175,
+      "step": 1660
+    },
+    {
+      "epoch": 0.08051685068222361,
+      "grad_norm": 8.864878739103949,
+      "learning_rate": 2e-05,
+      "loss": 0.7231966018676758,
+      "step": 1670
+    },
+    {
+      "epoch": 0.08099898751265609,
+      "grad_norm": 5.91600501179535,
+      "learning_rate": 2e-05,
+      "loss": 0.6998160362243653,
+      "step": 1680
+    },
+    {
+      "epoch": 0.08148112434308857,
+      "grad_norm": 2.7073467229023582,
+      "learning_rate": 2e-05,
+      "loss": 0.7311582088470459,
+      "step": 1690
+    },
+    {
+      "epoch": 0.08196326117352104,
+      "grad_norm": 2.6882475083061217,
+      "learning_rate": 2e-05,
+      "loss": 0.7005412101745605,
+      "step": 1700
+    },
+    {
+      "epoch": 0.08244539800395352,
+      "grad_norm": 1.3644811855655996,
+      "learning_rate": 2e-05,
+      "loss": 0.7040036678314209,
+      "step": 1710
+    },
+    {
+      "epoch": 0.082927534834386,
+      "grad_norm": 1.707401344240697,
+      "learning_rate": 2e-05,
+      "loss": 0.7216098308563232,
+      "step": 1720
+    },
+    {
+      "epoch": 0.08340967166481847,
+      "grad_norm": 2.2473714034468366,
+      "learning_rate": 2e-05,
+      "loss": 0.7012883186340332,
+      "step": 1730
+    },
+    {
+      "epoch": 0.08389180849525095,
+      "grad_norm": 3.8729069145758053,
+      "learning_rate": 2e-05,
+      "loss": 0.7171290397644043,
+      "step": 1740
+    },
+    {
+      "epoch": 0.08437394532568343,
+      "grad_norm": 2.2548793711676427,
+      "learning_rate": 2e-05,
+      "loss": 0.6937798976898193,
+      "step": 1750
+    },
+    {
+      "epoch": 0.08485608215611591,
+      "grad_norm": 3.363398202084773,
+      "learning_rate": 2e-05,
+      "loss": 0.7099102973937989,
+      "step": 1760
+    },
+    {
+      "epoch": 0.08533821898654838,
+      "grad_norm": 1.9683460905836172,
+      "learning_rate": 2e-05,
+      "loss": 0.7375545501708984,
+      "step": 1770
+    },
+    {
+      "epoch": 0.08582035581698086,
+      "grad_norm": 2.7563303203565104,
+      "learning_rate": 2e-05,
+      "loss": 0.693428373336792,
+      "step": 1780
+    },
+    {
+      "epoch": 0.08630249264741334,
+      "grad_norm": 5.488470914545985,
+      "learning_rate": 2e-05,
+      "loss": 0.6957249641418457,
+      "step": 1790
+    },
+    {
+      "epoch": 0.0867846294778458,
+      "grad_norm": 2.09336496768559,
+      "learning_rate": 2e-05,
+      "loss": 0.6892104148864746,
+      "step": 1800
+    },
+    {
+      "epoch": 0.08726676630827829,
+      "grad_norm": 1.5894667449541344,
+      "learning_rate": 2e-05,
+      "loss": 0.6863278388977051,
+      "step": 1810
+    },
+    {
+      "epoch": 0.08774890313871077,
+      "grad_norm": 2.351872874716584,
+      "learning_rate": 2e-05,
+      "loss": 0.6925557136535645,
+      "step": 1820
+    },
+    {
+      "epoch": 0.08823103996914325,
+      "grad_norm": 5.709628555644903,
+      "learning_rate": 2e-05,
+      "loss": 0.7161927223205566,
+      "step": 1830
+    },
+    {
+      "epoch": 0.08871317679957572,
+      "grad_norm": 2.3637395099075658,
+      "learning_rate": 2e-05,
+      "loss": 0.6833971977233887,
+      "step": 1840
+    },
+    {
+      "epoch": 0.0891953136300082,
+      "grad_norm": 3.0782660457463638,
+      "learning_rate": 2e-05,
+      "loss": 0.6962731838226318,
+      "step": 1850
+    },
+    {
+      "epoch": 0.08967745046044068,
+      "grad_norm": 3.0074368386337524,
+      "learning_rate": 2e-05,
+      "loss": 0.6978533744812012,
+      "step": 1860
+    },
+    {
+      "epoch": 0.09015958729087314,
+      "grad_norm": 2.887814419604526,
+      "learning_rate": 2e-05,
+      "loss": 0.7000442504882812,
+      "step": 1870
+    },
+    {
+      "epoch": 0.09064172412130563,
+      "grad_norm": 3.6280575625203513,
+      "learning_rate": 2e-05,
+      "loss": 0.7079222679138184,
+      "step": 1880
+    },
+    {
+      "epoch": 0.0911238609517381,
+      "grad_norm": 1.3557511666014126,
+      "learning_rate": 2e-05,
+      "loss": 0.6967864036560059,
+      "step": 1890
+    },
+    {
+      "epoch": 0.09160599778217059,
+      "grad_norm": 1.403629260906539,
+      "learning_rate": 2e-05,
+      "loss": 0.7130854606628418,
+      "step": 1900
+    },
+    {
+      "epoch": 0.09208813461260305,
+      "grad_norm": 1.8739359959248085,
+      "learning_rate": 2e-05,
+      "loss": 0.6871781826019288,
+      "step": 1910
+    },
+    {
+      "epoch": 0.09257027144303553,
+      "grad_norm": 2.824655544812341,
+      "learning_rate": 2e-05,
+      "loss": 0.6981027126312256,
+      "step": 1920
+    },
+    {
+      "epoch": 0.09305240827346802,
+      "grad_norm": 1.4828172680841498,
+      "learning_rate": 2e-05,
+      "loss": 0.6929860591888428,
+      "step": 1930
+    },
+    {
+      "epoch": 0.09353454510390048,
+      "grad_norm": 7.2089191719005425,
+      "learning_rate": 2e-05,
+      "loss": 0.7134679794311524,
+      "step": 1940
+    },
+    {
+      "epoch": 0.09401668193433296,
+      "grad_norm": 2.1889125436919574,
+      "learning_rate": 2e-05,
+      "loss": 0.7014190196990967,
+      "step": 1950
+    },
+    {
+      "epoch": 0.09449881876476544,
+      "grad_norm": 6.312627812926516,
+      "learning_rate": 2e-05,
+      "loss": 0.7476691722869873,
+      "step": 1960
+    },
+    {
+      "epoch": 0.09498095559519791,
+      "grad_norm": 1.2052636178685785,
+      "learning_rate": 2e-05,
+      "loss": 0.7070323944091796,
+      "step": 1970
+    },
+    {
+      "epoch": 0.09546309242563039,
+      "grad_norm": 1.1645089780675213,
+      "learning_rate": 2e-05,
+      "loss": 0.6702715873718261,
+      "step": 1980
+    },
+    {
+      "epoch": 0.09594522925606287,
+      "grad_norm": 3.612031603825751,
+      "learning_rate": 2e-05,
+      "loss": 0.6985786437988282,
+      "step": 1990
+    },
+    {
+      "epoch": 0.09642736608649535,
+      "grad_norm": 1.2061047374120728,
+      "learning_rate": 2e-05,
+      "loss": 0.6996285915374756,
+      "step": 2000
+    },
+    {
+      "epoch": 0.09690950291692782,
+      "grad_norm": 1.319276109312496,
+      "learning_rate": 2e-05,
+      "loss": 0.693095588684082,
+      "step": 2010
+    },
+    {
+      "epoch": 0.0973916397473603,
+      "grad_norm": 1.358389710976587,
+      "learning_rate": 2e-05,
+      "loss": 0.7233096122741699,
+      "step": 2020
+    },
+    {
+      "epoch": 0.09787377657779278,
+      "grad_norm": 2.29102697168753,
+      "learning_rate": 2e-05,
+      "loss": 0.699147367477417,
+      "step": 2030
+    },
+    {
+      "epoch": 0.09835591340822525,
+      "grad_norm": 1.1523175187527268,
+      "learning_rate": 2e-05,
+      "loss": 0.6790700435638428,
+      "step": 2040
+    },
+    {
+      "epoch": 0.09883805023865773,
+      "grad_norm": 1.1879796016842294,
+      "learning_rate": 2e-05,
+      "loss": 0.6621855735778809,
+      "step": 2050
+    },
+    {
+      "epoch": 0.09932018706909021,
+      "grad_norm": 4.574145683318952,
+      "learning_rate": 2e-05,
+      "loss": 0.7080859184265137,
+      "step": 2060
+    },
+    {
+      "epoch": 0.09980232389952269,
+      "grad_norm": 1.7275450667215668,
+      "learning_rate": 2e-05,
+      "loss": 0.675408935546875,
+      "step": 2070
+    },
+    {
+      "epoch": 0.10028446072995516,
+      "grad_norm": 4.908901932326397,
+      "learning_rate": 2e-05,
+      "loss": 0.6688334941864014,
+      "step": 2080
+    },
+    {
+      "epoch": 0.10076659756038764,
+      "grad_norm": 2.0224140803935433,
+      "learning_rate": 2e-05,
+      "loss": 0.6846134185791015,
+      "step": 2090
+    },
+    {
+      "epoch": 0.10124873439082012,
+      "grad_norm": 1.8372272183375968,
+      "learning_rate": 2e-05,
+      "loss": 0.6918974876403808,
+      "step": 2100
+    },
+    {
+      "epoch": 0.10173087122125259,
+      "grad_norm": 1.5531444555006406,
+      "learning_rate": 2e-05,
+      "loss": 0.7000730514526368,
+      "step": 2110
+    },
+    {
+      "epoch": 0.10221300805168507,
+      "grad_norm": 2.908690261918158,
+      "learning_rate": 2e-05,
+      "loss": 0.6698556423187256,
+      "step": 2120
+    },
+    {
+      "epoch": 0.10269514488211755,
+      "grad_norm": 2.548012328425614,
+      "learning_rate": 2e-05,
+      "loss": 0.6925012111663819,
+      "step": 2130
+    },
+    {
+      "epoch": 0.10317728171255003,
+      "grad_norm": 1.19308031783503,
+      "learning_rate": 2e-05,
+      "loss": 0.7279344081878663,
+      "step": 2140
+    },
+    {
+      "epoch": 0.1036594185429825,
+      "grad_norm": 1.1864109230709758,
+      "learning_rate": 2e-05,
+      "loss": 0.7339591979980469,
+      "step": 2150
+    },
+    {
+      "epoch": 0.10414155537341498,
+      "grad_norm": 1.1608556740092548,
+      "learning_rate": 2e-05,
+      "loss": 0.7135494232177735,
+      "step": 2160
+    },
+    {
+      "epoch": 0.10462369220384746,
+      "grad_norm": 1.4213454351648482,
+      "learning_rate": 2e-05,
+      "loss": 0.7160530090332031,
+      "step": 2170
+    },
+    {
+      "epoch": 0.10510582903427992,
+      "grad_norm": 2.418144401848944,
+      "learning_rate": 2e-05,
+      "loss": 0.6826236724853516,
+      "step": 2180
+    },
+    {
+      "epoch": 0.1055879658647124,
+      "grad_norm": 1.0983236668468084,
+      "learning_rate": 2e-05,
+      "loss": 0.69876127243042,
+      "step": 2190
+    },
+    {
+      "epoch": 0.10607010269514489,
+      "grad_norm": 1.2800062403826349,
+      "learning_rate": 2e-05,
+      "loss": 0.6858653068542481,
+      "step": 2200
+    },
+    {
+      "epoch": 0.10655223952557735,
+      "grad_norm": 1.9669738775612915,
+      "learning_rate": 2e-05,
+      "loss": 0.6780145645141602,
+      "step": 2210
+    },
+    {
+      "epoch": 0.10703437635600983,
+      "grad_norm": 1.076581580763509,
+      "learning_rate": 2e-05,
+      "loss": 0.6984048843383789,
+      "step": 2220
+    },
+    {
+      "epoch": 0.10751651318644231,
+      "grad_norm": 1.092267155276319,
+      "learning_rate": 2e-05,
+      "loss": 0.6883275508880615,
+      "step": 2230
+    },
+    {
+      "epoch": 0.1079986500168748,
+      "grad_norm": 2.3144361176260944,
+      "learning_rate": 2e-05,
+      "loss": 0.6878983974456787,
+      "step": 2240
+    },
+    {
+      "epoch": 0.10848078684730726,
+      "grad_norm": 1.6075806692003578,
+      "learning_rate": 2e-05,
+      "loss": 0.691591739654541,
+      "step": 2250
+    },
+    {
+      "epoch": 0.10896292367773974,
+      "grad_norm": 1.2732184058631342,
+      "learning_rate": 2e-05,
+      "loss": 0.6779979228973388,
+      "step": 2260
+    },
+    {
+      "epoch": 0.10944506050817222,
+      "grad_norm": 1.4415867892927186,
+      "learning_rate": 2e-05,
+      "loss": 0.6897299766540528,
+      "step": 2270
+    },
+    {
+      "epoch": 0.10992719733860469,
+      "grad_norm": 1.9698774452252525,
+      "learning_rate": 2e-05,
+      "loss": 0.6754415035247803,
+      "step": 2280
+    },
+    {
+      "epoch": 0.11040933416903717,
+      "grad_norm": 1.8988393314031458,
+      "learning_rate": 2e-05,
+      "loss": 0.6988449096679688,
+      "step": 2290
+    },
+    {
+      "epoch": 0.11089147099946965,
+      "grad_norm": 2.0229172924552636,
+      "learning_rate": 2e-05,
+      "loss": 0.6995786190032959,
+      "step": 2300
+    },
+    {
+      "epoch": 0.11137360782990213,
+      "grad_norm": 2.5016680796886446,
+      "learning_rate": 2e-05,
+      "loss": 0.683745288848877,
+      "step": 2310
+    },
+    {
+      "epoch": 0.1118557446603346,
+      "grad_norm": 2.795671113173564,
+      "learning_rate": 2e-05,
+      "loss": 0.7132812023162842,
+      "step": 2320
+    },
+    {
+      "epoch": 0.11233788149076708,
+      "grad_norm": 15.429266895654095,
+      "learning_rate": 2e-05,
+      "loss": 0.6710845947265625,
+      "step": 2330
+    },
+    {
+      "epoch": 0.11282001832119956,
+      "grad_norm": 10.28073301520565,
+      "learning_rate": 2e-05,
+      "loss": 0.7072024345397949,
+      "step": 2340
+    },
+    {
+      "epoch": 0.11330215515163203,
+      "grad_norm": 2.5529434317741475,
+      "learning_rate": 2e-05,
+      "loss": 0.7183896064758301,
+      "step": 2350
+    },
+    {
+      "epoch": 0.11378429198206451,
+      "grad_norm": 2.3176071939604768,
+      "learning_rate": 2e-05,
+      "loss": 0.7112489700317383,
+      "step": 2360
+    },
+    {
+      "epoch": 0.11426642881249699,
+      "grad_norm": 2.2340973277391556,
+      "learning_rate": 2e-05,
+      "loss": 0.6871528148651123,
+      "step": 2370
+    },
+    {
+      "epoch": 0.11474856564292946,
+      "grad_norm": 5.698121231020915,
+      "learning_rate": 2e-05,
+      "loss": 0.7194876670837402,
+      "step": 2380
+    },
+    {
+      "epoch": 0.11523070247336194,
+      "grad_norm": 1.2389861862090972,
+      "learning_rate": 2e-05,
+      "loss": 0.7029046535491943,
+      "step": 2390
+    },
+    {
+      "epoch": 0.11571283930379442,
+      "grad_norm": 1.0835085632703585,
+      "learning_rate": 2e-05,
+      "loss": 0.6844483852386475,
+      "step": 2400
+    },
+    {
+      "epoch": 0.1161949761342269,
+      "grad_norm": 1.5728645257042526,
+      "learning_rate": 2e-05,
+      "loss": 0.6933661937713623,
+      "step": 2410
+    },
+    {
+      "epoch": 0.11667711296465937,
+      "grad_norm": 1.6879257390220759,
+      "learning_rate": 2e-05,
+      "loss": 0.6991979122161865,
+      "step": 2420
+    },
+    {
+      "epoch": 0.11715924979509185,
+      "grad_norm": 1.86006325380769,
+      "learning_rate": 2e-05,
+      "loss": 0.7063355445861816,
+      "step": 2430
+    },
+    {
+      "epoch": 0.11764138662552433,
+      "grad_norm": 2.4323496874311745,
+      "learning_rate": 2e-05,
+      "loss": 0.6896146297454834,
+      "step": 2440
+    },
+    {
+      "epoch": 0.1181235234559568,
+      "grad_norm": 1.2764606849218354,
+      "learning_rate": 2e-05,
+      "loss": 0.6714626312255859,
+      "step": 2450
+    },
+    {
+      "epoch": 0.11860566028638928,
+      "grad_norm": 1.3536153362668832,
+      "learning_rate": 2e-05,
+      "loss": 0.6861135482788085,
+      "step": 2460
+    },
+    {
+      "epoch": 0.11908779711682176,
+      "grad_norm": 1.0698060532539524,
+      "learning_rate": 2e-05,
+      "loss": 0.6550200462341309,
+      "step": 2470
+    },
+    {
+      "epoch": 0.11956993394725424,
+      "grad_norm": 1.2301952555430697,
+      "learning_rate": 2e-05,
+      "loss": 0.6896649837493897,
+      "step": 2480
+    },
+    {
+      "epoch": 0.1200520707776867,
+      "grad_norm": 1.199823311171627,
+      "learning_rate": 2e-05,
+      "loss": 0.6958704948425293,
+      "step": 2490
+    },
+    {
+      "epoch": 0.12053420760811918,
+      "grad_norm": 1.4038590113770721,
+      "learning_rate": 2e-05,
+      "loss": 0.6850681304931641,
+      "step": 2500
+    },
+    {
+      "epoch": 0.12101634443855167,
+      "grad_norm": 2.9612679143825424,
+      "learning_rate": 2e-05,
+      "loss": 0.6791827201843261,
+      "step": 2510
+    },
+    {
+      "epoch": 0.12149848126898413,
+      "grad_norm": 5.83314529557438,
+      "learning_rate": 2e-05,
+      "loss": 0.7032989025115967,
+      "step": 2520
+    },
+    {
+      "epoch": 0.12198061809941661,
+      "grad_norm": 2.5477970924994966,
+      "learning_rate": 2e-05,
+      "loss": 0.6850883007049561,
+      "step": 2530
+    },
+    {
+      "epoch": 0.1224627549298491,
+      "grad_norm": 3.3401088169773034,
+      "learning_rate": 2e-05,
+      "loss": 0.6954312801361084,
+      "step": 2540
+    },
+    {
+      "epoch": 0.12294489176028157,
+      "grad_norm": 2.156136005924588,
+      "learning_rate": 2e-05,
+      "loss": 0.6937005519866943,
+      "step": 2550
+    },
+    {
+      "epoch": 0.12342702859071404,
+      "grad_norm": 0.9506305729021246,
+      "learning_rate": 2e-05,
+      "loss": 0.7065269470214843,
+      "step": 2560
+    },
+    {
+      "epoch": 0.12390916542114652,
+      "grad_norm": 1.0754149213576563,
+      "learning_rate": 2e-05,
+      "loss": 0.6961036682128906,
+      "step": 2570
+    },
+    {
+      "epoch": 0.124391302251579,
+      "grad_norm": 2.3768460001969407,
+      "learning_rate": 2e-05,
+      "loss": 0.664985466003418,
+      "step": 2580
+    },
+    {
+      "epoch": 0.12487343908201147,
+      "grad_norm": 1.3084725617767137,
+      "learning_rate": 2e-05,
+      "loss": 0.6766395568847656,
+      "step": 2590
+    },
+    {
+      "epoch": 0.12535557591244395,
+      "grad_norm": 1.0291527417088837,
+      "learning_rate": 2e-05,
+      "loss": 0.6760396480560302,
+      "step": 2600
+    },
+    {
+      "epoch": 0.12583771274287642,
+      "grad_norm": 0.941713441397077,
+      "learning_rate": 2e-05,
+      "loss": 0.6730161190032959,
+      "step": 2610
+    },
+    {
+      "epoch": 0.1263198495733089,
+      "grad_norm": 1.504391543980072,
+      "learning_rate": 2e-05,
+      "loss": 0.6974763870239258,
+      "step": 2620
+    },
+    {
+      "epoch": 0.12680198640374138,
+      "grad_norm": 1.1573050856193627,
+      "learning_rate": 2e-05,
+      "loss": 0.6692431449890137,
+      "step": 2630
+    },
+    {
+      "epoch": 0.12728412323417385,
+      "grad_norm": 1.05212506731337,
+      "learning_rate": 2e-05,
+      "loss": 0.708095932006836,
+      "step": 2640
+    },
+    {
+      "epoch": 0.12776626006460634,
+      "grad_norm": 1.2342820641380825,
+      "learning_rate": 2e-05,
+      "loss": 0.670927906036377,
+      "step": 2650
+    },
+    {
+      "epoch": 0.1282483968950388,
+      "grad_norm": 1.3953957212568666,
+      "learning_rate": 2e-05,
+      "loss": 0.7032140254974365,
+      "step": 2660
+    },
+    {
+      "epoch": 0.1287305337254713,
+      "grad_norm": 6.118740903384293,
+      "learning_rate": 2e-05,
+      "loss": 0.6814794063568115,
+      "step": 2670
+    },
+    {
+      "epoch": 0.12921267055590377,
+      "grad_norm": 1.0141506610694353,
+      "learning_rate": 2e-05,
+      "loss": 0.6771867275238037,
+      "step": 2680
+    },
+    {
+      "epoch": 0.12969480738633624,
+      "grad_norm": 5.236945389567668,
+      "learning_rate": 2e-05,
+      "loss": 0.6872392177581788,
+      "step": 2690
+    },
+    {
+      "epoch": 0.13017694421676873,
+      "grad_norm": 1.3224427182296696,
+      "learning_rate": 2e-05,
+      "loss": 0.6877142906188964,
+      "step": 2700
+    },
+    {
+      "epoch": 0.1306590810472012,
+      "grad_norm": 2.8930047020555665,
+      "learning_rate": 2e-05,
+      "loss": 0.6983075618743897,
+      "step": 2710
+    },
+    {
+      "epoch": 0.13114121787763366,
+      "grad_norm": 1.095684392846267,
+      "learning_rate": 2e-05,
+      "loss": 0.6783349990844727,
+      "step": 2720
+    },
+    {
+      "epoch": 0.13162335470806616,
+      "grad_norm": 4.47691483015509,
+      "learning_rate": 2e-05,
+      "loss": 0.6583886623382569,
+      "step": 2730
+    },
+    {
+      "epoch": 0.13210549153849863,
+      "grad_norm": 1.5187211298061205,
+      "learning_rate": 2e-05,
+      "loss": 0.6880706310272217,
+      "step": 2740
+    },
+    {
+      "epoch": 0.1325876283689311,
+      "grad_norm": 3.2659296740988983,
+      "learning_rate": 2e-05,
+      "loss": 0.6818893432617188,
+      "step": 2750
+    },
+    {
+      "epoch": 0.1330697651993636,
+      "grad_norm": 2.9770080325455517,
+      "learning_rate": 2e-05,
+      "loss": 0.6884412288665771,
+      "step": 2760
+    },
+    {
+      "epoch": 0.13355190202979605,
+      "grad_norm": 3.264974211891368,
+      "learning_rate": 2e-05,
+      "loss": 0.6844497203826905,
+      "step": 2770
+    },
+    {
+      "epoch": 0.13403403886022852,
+      "grad_norm": 1.4102066352693954,
+      "learning_rate": 2e-05,
+      "loss": 0.6756868362426758,
+      "step": 2780
+    },
+    {
+      "epoch": 0.13451617569066102,
+      "grad_norm": 3.487588341656881,
+      "learning_rate": 2e-05,
+      "loss": 0.6730589866638184,
+      "step": 2790
+    },
+    {
+      "epoch": 0.13499831252109348,
+      "grad_norm": 1.4561280668529009,
+      "learning_rate": 2e-05,
+      "loss": 0.6855874538421631,
+      "step": 2800
+    },
+    {
+      "epoch": 0.13548044935152595,
+      "grad_norm": 2.3566879942304677,
+      "learning_rate": 2e-05,
+      "loss": 0.6573580741882324,
+      "step": 2810
+    },
+    {
+      "epoch": 0.13596258618195844,
+      "grad_norm": 1.3005046968806246,
+      "learning_rate": 2e-05,
+      "loss": 0.6513078212738037,
+      "step": 2820
+    },
+    {
+      "epoch": 0.1364447230123909,
+      "grad_norm": 1.2183089382076557,
+      "learning_rate": 2e-05,
+      "loss": 0.6996438503265381,
+      "step": 2830
+    },
+    {
+      "epoch": 0.1369268598428234,
+      "grad_norm": 1.4063935661432136,
+      "learning_rate": 2e-05,
+      "loss": 0.655107069015503,
+      "step": 2840
+    },
+    {
+      "epoch": 0.13740899667325587,
+      "grad_norm": 1.2423309716184736,
+      "learning_rate": 2e-05,
+      "loss": 0.68148832321167,
+      "step": 2850
+    },
+    {
+      "epoch": 0.13789113350368834,
+      "grad_norm": 1.315583154853977,
+      "learning_rate": 2e-05,
+      "loss": 0.6690854549407959,
+      "step": 2860
+    },
+    {
+      "epoch": 0.13837327033412083,
+      "grad_norm": 3.2685693317604065,
+      "learning_rate": 2e-05,
+      "loss": 0.6860932350158692,
+      "step": 2870
+    },
+    {
+      "epoch": 0.1388554071645533,
+      "grad_norm": 1.6404916651968342,
+      "learning_rate": 2e-05,
+      "loss": 0.6692975997924805,
+      "step": 2880
+    },
+    {
+      "epoch": 0.13933754399498577,
+      "grad_norm": 1.0467271314758744,
+      "learning_rate": 2e-05,
+      "loss": 0.7087518692016601,
+      "step": 2890
+    },
+    {
+      "epoch": 0.13981968082541826,
+      "grad_norm": 1.6145047549036693,
+      "learning_rate": 2e-05,
+      "loss": 0.6947719573974609,
+      "step": 2900
+    },
+    {
+      "epoch": 0.14030181765585073,
+      "grad_norm": 1.701843802720987,
+      "learning_rate": 2e-05,
+      "loss": 0.6641913414001465,
+      "step": 2910
+    },
+    {
+      "epoch": 0.1407839544862832,
+      "grad_norm": 1.0188551579918583,
+      "learning_rate": 2e-05,
+      "loss": 0.6774641036987304,
+      "step": 2920
+    },
+    {
+      "epoch": 0.1412660913167157,
+      "grad_norm": 1.1481854023006681,
+      "learning_rate": 2e-05,
+      "loss": 0.6788041591644287,
+      "step": 2930
+    },
+    {
+      "epoch": 0.14174822814714816,
+      "grad_norm": 1.4560369115077942,
+      "learning_rate": 2e-05,
+      "loss": 0.681424617767334,
+      "step": 2940
+    },
+    {
+      "epoch": 0.14223036497758063,
+      "grad_norm": 2.5032736820411583,
+      "learning_rate": 2e-05,
+      "loss": 0.6913356781005859,
+      "step": 2950
+    },
+    {
+      "epoch": 0.14271250180801312,
+      "grad_norm": 1.151825196136896,
+      "learning_rate": 2e-05,
+      "loss": 0.6814065456390381,
+      "step": 2960
+    },
+    {
+      "epoch": 0.1431946386384456,
+      "grad_norm": 1.1185710273784648,
+      "learning_rate": 2e-05,
+      "loss": 0.6830487251281738,
+      "step": 2970
+    },
+    {
+      "epoch": 0.14367677546887805,
+      "grad_norm": 1.0947324337188409,
+      "learning_rate": 2e-05,
+      "loss": 0.6666187286376953,
+      "step": 2980
+    },
+    {
+      "epoch": 0.14415891229931055,
+      "grad_norm": 1.1444381139788962,
+      "learning_rate": 2e-05,
+      "loss": 0.7106166839599609,
+      "step": 2990
+    },
+    {
+      "epoch": 0.14464104912974302,
+      "grad_norm": 4.369241256060388,
+      "learning_rate": 2e-05,
+      "loss": 0.6684728145599366,
+      "step": 3000
+    },
+    {
+      "epoch": 0.1451231859601755,
+      "grad_norm": 1.65482387871921,
+      "learning_rate": 2e-05,
+      "loss": 0.655134916305542,
+      "step": 3010
+    },
+    {
+      "epoch": 0.14560532279060798,
+      "grad_norm": 3.8788031700801193,
+      "learning_rate": 2e-05,
+      "loss": 0.6682104110717774,
+      "step": 3020
+    },
+    {
+      "epoch": 0.14608745962104044,
+      "grad_norm": 6.673831148036922,
+      "learning_rate": 2e-05,
+      "loss": 0.6606300354003907,
+      "step": 3030
+    },
+    {
+      "epoch": 0.14656959645147294,
+      "grad_norm": 1.6037926296803937,
+      "learning_rate": 2e-05,
+      "loss": 0.7429388999938965,
+      "step": 3040
+    },
+    {
+      "epoch": 0.1470517332819054,
+      "grad_norm": 1.351928692493476,
+      "learning_rate": 2e-05,
+      "loss": 0.667429256439209,
+      "step": 3050
+    },
+    {
+      "epoch": 0.14753387011233787,
+      "grad_norm": 1.4552556136717532,
+      "learning_rate": 2e-05,
+      "loss": 0.6731912612915039,
+      "step": 3060
+    },
+    {
+      "epoch": 0.14801600694277037,
+      "grad_norm": 1.0539729227924972,
+      "learning_rate": 2e-05,
+      "loss": 0.6738934993743897,
+      "step": 3070
+    },
+    {
+      "epoch": 0.14849814377320283,
+      "grad_norm": 1.4456675023006782,
+      "learning_rate": 2e-05,
+      "loss": 0.676184606552124,
+      "step": 3080
+    },
+    {
+      "epoch": 0.1489802806036353,
+      "grad_norm": 1.077443568362399,
+      "learning_rate": 2e-05,
+      "loss": 0.6643596649169922,
+      "step": 3090
+    },
+    {
+      "epoch": 0.1494624174340678,
+      "grad_norm": 1.0222170857558752,
+      "learning_rate": 2e-05,
+      "loss": 0.6713498592376709,
+      "step": 3100
+    },
+    {
+      "epoch": 0.14994455426450026,
+      "grad_norm": 4.219194991350471,
+      "learning_rate": 2e-05,
+      "loss": 0.6660583019256592,
+      "step": 3110
+    },
+    {
+      "epoch": 0.15042669109493273,
+      "grad_norm": 2.8379852412475444,
+      "learning_rate": 2e-05,
+      "loss": 0.6860692977905274,
+      "step": 3120
+    },
+    {
+      "epoch": 0.15090882792536522,
+      "grad_norm": 1.4914961806386975,
+      "learning_rate": 2e-05,
+      "loss": 0.6793073177337646,
+      "step": 3130
+    },
+    {
+      "epoch": 0.1513909647557977,
+      "grad_norm": 0.9940149224219857,
+      "learning_rate": 2e-05,
+      "loss": 0.6694391250610352,
+      "step": 3140
+    },
+    {
+      "epoch": 0.15187310158623019,
+      "grad_norm": 1.7353980809287022,
+      "learning_rate": 2e-05,
+      "loss": 0.6642420768737793,
+      "step": 3150
+    },
+    {
+      "epoch": 0.15235523841666265,
+      "grad_norm": 1.7214847019768242,
+      "learning_rate": 2e-05,
+      "loss": 0.6977147579193115,
+      "step": 3160
+    },
+    {
+      "epoch": 0.15283737524709512,
+      "grad_norm": 5.352629935949013,
+      "learning_rate": 2e-05,
+      "loss": 0.654951810836792,
+      "step": 3170
+    },
+    {
+      "epoch": 0.15331951207752761,
+      "grad_norm": 3.29019939471364,
+      "learning_rate": 2e-05,
+      "loss": 0.6566079616546631,
+      "step": 3180
+    },
+    {
+      "epoch": 0.15380164890796008,
+      "grad_norm": 2.6884608924500295,
+      "learning_rate": 2e-05,
+      "loss": 0.6857626914978028,
+      "step": 3190
+    },
+    {
+      "epoch": 0.15428378573839255,
+      "grad_norm": 1.255094118183628,
+      "learning_rate": 2e-05,
+      "loss": 0.6701702117919922,
+      "step": 3200
+    },
+    {
+      "epoch": 0.15476592256882504,
+      "grad_norm": 1.4093512886672244,
+      "learning_rate": 2e-05,
+      "loss": 0.6666848659515381,
+      "step": 3210
+    },
+    {
+      "epoch": 0.1552480593992575,
+      "grad_norm": 1.1999858400364931,
+      "learning_rate": 2e-05,
+      "loss": 0.6728490829467774,
+      "step": 3220
+    },
+    {
+      "epoch": 0.15573019622968998,
+      "grad_norm": 0.9312290611940234,
+      "learning_rate": 2e-05,
+      "loss": 0.6722506999969482,
+      "step": 3230
+    },
+    {
+      "epoch": 0.15621233306012247,
+      "grad_norm": 2.553025777130132,
+      "learning_rate": 2e-05,
+      "loss": 0.6876326560974121,
+      "step": 3240
+    },
+    {
+      "epoch": 0.15669446989055494,
+      "grad_norm": 1.268898222189589,
+      "learning_rate": 2e-05,
+      "loss": 0.6727676391601562,
+      "step": 3250
+    },
+    {
+      "epoch": 0.1571766067209874,
+      "grad_norm": 0.9871861877915589,
+      "learning_rate": 2e-05,
+      "loss": 0.6898754119873047,
+      "step": 3260
+    },
+    {
+      "epoch": 0.1576587435514199,
+      "grad_norm": 1.3121034015976296,
+      "learning_rate": 2e-05,
+      "loss": 0.7000826835632324,
+      "step": 3270
+    },
+    {
+      "epoch": 0.15814088038185237,
+      "grad_norm": 1.573041076216053,
+      "learning_rate": 2e-05,
+      "loss": 0.6678867340087891,
+      "step": 3280
+    },
+    {
+      "epoch": 0.15862301721228483,
+      "grad_norm": 1.2468074927739778,
+      "learning_rate": 2e-05,
+      "loss": 0.6610896110534668,
+      "step": 3290
+    },
+    {
+      "epoch": 0.15910515404271733,
+      "grad_norm": 1.4460520138282018,
+      "learning_rate": 2e-05,
+      "loss": 0.6774022579193115,
+      "step": 3300
+    },
+    {
+      "epoch": 0.1595872908731498,
+      "grad_norm": 0.9181019937438198,
+      "learning_rate": 2e-05,
+      "loss": 0.6579627990722656,
+      "step": 3310
+    },
+    {
+      "epoch": 0.1600694277035823,
+      "grad_norm": 1.2301838291865153,
+      "learning_rate": 2e-05,
+      "loss": 0.6787763595581054,
+      "step": 3320
+    },
+    {
+      "epoch": 0.16055156453401476,
+      "grad_norm": 1.1662238353602683,
+      "learning_rate": 2e-05,
+      "loss": 0.6730850219726563,
+      "step": 3330
+    },
+    {
+      "epoch": 0.16103370136444722,
+      "grad_norm": 1.048654384841359,
+      "learning_rate": 2e-05,
+      "loss": 0.6611888408660889,
+      "step": 3340
+    },
+    {
+      "epoch": 0.16151583819487972,
+      "grad_norm": 1.1452674746253244,
+      "learning_rate": 2e-05,
+      "loss": 0.6924826622009277,
+      "step": 3350
+    },
+    {
+      "epoch": 0.16199797502531219,
+      "grad_norm": 0.8425856165639504,
+      "learning_rate": 2e-05,
+      "loss": 0.682227897644043,
+      "step": 3360
+    },
+    {
+      "epoch": 0.16248011185574465,
+      "grad_norm": 0.7641959163703967,
+      "learning_rate": 2e-05,
+      "loss": 0.650839376449585,
+      "step": 3370
+    },
+    {
+      "epoch": 0.16296224868617715,
+      "grad_norm": 0.9676006615740037,
+      "learning_rate": 2e-05,
+      "loss": 0.6620234966278076,
+      "step": 3380
+    },
+    {
+      "epoch": 0.1634443855166096,
+      "grad_norm": 1.0275477898065137,
+      "learning_rate": 2e-05,
+      "loss": 0.6789710044860839,
+      "step": 3390
+    },
+    {
+      "epoch": 0.16392652234704208,
+      "grad_norm": 0.9651639220774981,
+      "learning_rate": 2e-05,
+      "loss": 0.6729005336761474,
+      "step": 3400
+    },
+    {
+      "epoch": 0.16440865917747458,
+      "grad_norm": 1.0095576265042059,
+      "learning_rate": 2e-05,
+      "loss": 0.6622889995574951,
+      "step": 3410
+    },
+    {
+      "epoch": 0.16489079600790704,
+      "grad_norm": 1.4102750503502044,
+      "learning_rate": 2e-05,
+      "loss": 0.688470983505249,
+      "step": 3420
+    },
+    {
+      "epoch": 0.1653729328383395,
+      "grad_norm": 1.0104386911688485,
+      "learning_rate": 2e-05,
+      "loss": 0.6837211608886719,
+      "step": 3430
+    },
+    {
+      "epoch": 0.165855069668772,
+      "grad_norm": 0.8530483623781624,
+      "learning_rate": 2e-05,
+      "loss": 0.6495920181274414,
+      "step": 3440
+    },
+    {
+      "epoch": 0.16633720649920447,
+      "grad_norm": 0.8328789961389231,
+      "learning_rate": 2e-05,
+      "loss": 0.6881947994232178,
+      "step": 3450
+    },
+    {
+      "epoch": 0.16681934332963694,
+      "grad_norm": 0.8923267286524855,
+      "learning_rate": 2e-05,
+      "loss": 0.6668416023254394,
+      "step": 3460
+    },
+    {
+      "epoch": 0.16730148016006943,
+      "grad_norm": 0.9905631349362998,
+      "learning_rate": 2e-05,
+      "loss": 0.6595633506774903,
+      "step": 3470
+    },
+    {
+      "epoch": 0.1677836169905019,
+      "grad_norm": 1.1932777887680304,
+      "learning_rate": 2e-05,
+      "loss": 0.698433780670166,
+      "step": 3480
+    },
+    {
+      "epoch": 0.1682657538209344,
+      "grad_norm": 1.2144385972852307,
+      "learning_rate": 2e-05,
+      "loss": 0.6673844814300537,
+      "step": 3490
+    },
+    {
+      "epoch": 0.16874789065136686,
+      "grad_norm": 0.9355627133530535,
+      "learning_rate": 2e-05,
+      "loss": 0.6595808029174804,
+      "step": 3500
+    },
+    {
+      "epoch": 0.16923002748179933,
+      "grad_norm": 0.8881233185562102,
+      "learning_rate": 2e-05,
+      "loss": 0.6607702255249024,
+      "step": 3510
+    },
+    {
+      "epoch": 0.16971216431223182,
+      "grad_norm": 0.9692529681819855,
+      "learning_rate": 2e-05,
+      "loss": 0.6615869045257569,
+      "step": 3520
+    },
+    {
+      "epoch": 0.1701943011426643,
+      "grad_norm": 1.0731444302408328,
+      "learning_rate": 2e-05,
+      "loss": 0.6751644134521484,
+      "step": 3530
+    },
+    {
+      "epoch": 0.17067643797309676,
+      "grad_norm": 0.9907663950433366,
+      "learning_rate": 2e-05,
+      "loss": 0.6754093170166016,
+      "step": 3540
+    },
+    {
+      "epoch": 0.17115857480352925,
+      "grad_norm": 1.2987126031750855,
+      "learning_rate": 2e-05,
+      "loss": 0.6764405250549317,
+      "step": 3550
+    },
+    {
+      "epoch": 0.17164071163396172,
+      "grad_norm": 0.9802822317408426,
+      "learning_rate": 2e-05,
+      "loss": 0.6801197052001953,
+      "step": 3560
+    },
+    {
+      "epoch": 0.17212284846439418,
+      "grad_norm": 1.6789933700775186,
+      "learning_rate": 2e-05,
+      "loss": 0.6903232097625732,
+      "step": 3570
+    },
+    {
+      "epoch": 0.17260498529482668,
+      "grad_norm": 0.9001516666422892,
+      "learning_rate": 2e-05,
+      "loss": 0.690924596786499,
+      "step": 3580
+    },
+    {
+      "epoch": 0.17308712212525915,
+      "grad_norm": 0.8999927791689533,
+      "learning_rate": 2e-05,
+      "loss": 0.6645955085754395,
+      "step": 3590
+    },
+    {
+      "epoch": 0.1735692589556916,
+      "grad_norm": 0.8686685467224973,
+      "learning_rate": 2e-05,
+      "loss": 0.6666905879974365,
+      "step": 3600
+    },
+    {
+      "epoch": 0.1740513957861241,
+      "grad_norm": 0.9875623486780754,
+      "learning_rate": 2e-05,
+      "loss": 0.6690942764282226,
+      "step": 3610
+    },
+    {
+      "epoch": 0.17453353261655657,
+      "grad_norm": 1.044984026115174,
+      "learning_rate": 2e-05,
+      "loss": 0.6931906700134277,
+      "step": 3620
+    },
+    {
+      "epoch": 0.17501566944698904,
+      "grad_norm": 1.2290695131066534,
+      "learning_rate": 2e-05,
+      "loss": 0.6761062622070313,
+      "step": 3630
+    },
+    {
+      "epoch": 0.17549780627742154,
+      "grad_norm": 1.3006519828871421,
+      "learning_rate": 2e-05,
+      "loss": 0.6779094696044922,
+      "step": 3640
+    },
+    {
+      "epoch": 0.175979943107854,
+      "grad_norm": 1.0077266440890464,
+      "learning_rate": 2e-05,
+      "loss": 0.672411298751831,
+      "step": 3650
+    },
+    {
+      "epoch": 0.1764620799382865,
+      "grad_norm": 0.9829321808598609,
+      "learning_rate": 2e-05,
+      "loss": 0.6660247325897217,
+      "step": 3660
+    },
+    {
+      "epoch": 0.17694421676871896,
+      "grad_norm": 1.0961424512756348,
+      "learning_rate": 2e-05,
+      "loss": 0.6902739524841308,
+      "step": 3670
+    },
+    {
+      "epoch": 0.17742635359915143,
+      "grad_norm": 1.5075989932386027,
+      "learning_rate": 2e-05,
+      "loss": 0.6640830039978027,
+      "step": 3680
+    },
+    {
+      "epoch": 0.17790849042958393,
+      "grad_norm": 1.2149987426342592,
+      "learning_rate": 2e-05,
+      "loss": 0.6788613796234131,
+      "step": 3690
+    },
+    {
+      "epoch": 0.1783906272600164,
+      "grad_norm": 1.010951007032087,
+      "learning_rate": 2e-05,
+      "loss": 0.6687648296356201,
+      "step": 3700
+    },
+    {
+      "epoch": 0.17887276409044886,
+      "grad_norm": 1.6387930211518675,
+      "learning_rate": 2e-05,
+      "loss": 0.6767978191375732,
+      "step": 3710
+    },
+    {
+      "epoch": 0.17935490092088135,
+      "grad_norm": 0.8420795004909761,
+      "learning_rate": 2e-05,
+      "loss": 0.6681337356567383,
+      "step": 3720
+    },
+    {
+      "epoch": 0.17983703775131382,
+      "grad_norm": 1.3002487248232895,
+      "learning_rate": 2e-05,
+      "loss": 0.6419276237487793,
+      "step": 3730
+    },
+    {
+      "epoch": 0.1803191745817463,
+      "grad_norm": 0.9790965164949196,
+      "learning_rate": 2e-05,
+      "loss": 0.6518523693084717,
+      "step": 3740
+    },
+    {
+      "epoch": 0.18080131141217878,
+      "grad_norm": 3.3498979029584466,
+      "learning_rate": 2e-05,
+      "loss": 0.6688310623168945,
+      "step": 3750
+    },
+    {
+      "epoch": 0.18128344824261125,
+      "grad_norm": 1.0803828584507384,
+      "learning_rate": 2e-05,
+      "loss": 0.6670453548431396,
+      "step": 3760
+    },
+    {
+      "epoch": 0.18176558507304372,
+      "grad_norm": 0.9842895928426686,
+      "learning_rate": 2e-05,
+      "loss": 0.6521693229675293,
+      "step": 3770
+    },
+    {
+      "epoch": 0.1822477219034762,
+      "grad_norm": 1.625574902458031,
+      "learning_rate": 2e-05,
+      "loss": 0.6563037395477295,
+      "step": 3780
+    },
+    {
+      "epoch": 0.18272985873390868,
+      "grad_norm": 1.340884168638072,
+      "learning_rate": 2e-05,
+      "loss": 0.6562198638916016,
+      "step": 3790
+    },
+    {
+      "epoch": 0.18321199556434117,
+      "grad_norm": 1.1855364311851662,
+      "learning_rate": 2e-05,
+      "loss": 0.6470540523529053,
+      "step": 3800
+    },
+    {
+      "epoch": 0.18369413239477364,
+      "grad_norm": 1.3937241477733824,
+      "learning_rate": 2e-05,
+      "loss": 0.6616937160491944,
+      "step": 3810
+    },
+    {
+      "epoch": 0.1841762692252061,
+      "grad_norm": 2.499856969162339,
+      "learning_rate": 2e-05,
+      "loss": 0.6882741928100586,
+      "step": 3820
+    },
+    {
+      "epoch": 0.1846584060556386,
+      "grad_norm": 1.1811834436643003,
+      "learning_rate": 2e-05,
+      "loss": 0.6623140335083008,
+      "step": 3830
+    },
+    {
+      "epoch": 0.18514054288607107,
+      "grad_norm": 2.5103076355873792,
+      "learning_rate": 2e-05,
+      "loss": 0.6777871608734131,
+      "step": 3840
+    },
+    {
+      "epoch": 0.18562267971650354,
+      "grad_norm": 1.4938204779265354,
+      "learning_rate": 2e-05,
+      "loss": 0.6687375545501709,
+      "step": 3850
+    },
+    {
+      "epoch": 0.18610481654693603,
+      "grad_norm": 1.5918977985322353,
+      "learning_rate": 2e-05,
+      "loss": 0.652255630493164,
+      "step": 3860
+    },
+    {
+      "epoch": 0.1865869533773685,
+      "grad_norm": 1.0168121573216373,
+      "learning_rate": 2e-05,
+      "loss": 0.6587842464447021,
+      "step": 3870
+    },
+    {
+      "epoch": 0.18706909020780096,
+      "grad_norm": 1.6105453622752801,
+      "learning_rate": 2e-05,
+      "loss": 0.6472054958343506,
+      "step": 3880
+    },
+    {
+      "epoch": 0.18755122703823346,
+      "grad_norm": 1.3554003107577053,
+      "learning_rate": 2e-05,
+      "loss": 0.6695438385009765,
+      "step": 3890
+    },
+    {
+      "epoch": 0.18803336386866593,
+      "grad_norm": 1.2536203717348502,
+      "learning_rate": 2e-05,
+      "loss": 0.6728156566619873,
+      "step": 3900
+    },
+    {
+      "epoch": 0.1885155006990984,
+      "grad_norm": 1.1490320054865537,
+      "learning_rate": 2e-05,
+      "loss": 0.6469772338867188,
+      "step": 3910
+    },
+    {
+      "epoch": 0.1889976375295309,
+      "grad_norm": 0.864390985345404,
+      "learning_rate": 2e-05,
+      "loss": 0.6844843864440918,
+      "step": 3920
+    },
+    {
+      "epoch": 0.18947977435996335,
+      "grad_norm": 1.3376223021036016,
+      "learning_rate": 2e-05,
+      "loss": 0.6520900249481201,
+      "step": 3930
+    },
+    {
+      "epoch": 0.18996191119039582,
+      "grad_norm": 1.1185837603192583,
+      "learning_rate": 2e-05,
+      "loss": 0.6744009971618652,
+      "step": 3940
+    },
+    {
+      "epoch": 0.19044404802082832,
+      "grad_norm": 2.041335111707897,
+      "learning_rate": 2e-05,
+      "loss": 0.6582718849182129,
+      "step": 3950
+    },
+    {
+      "epoch": 0.19092618485126078,
+      "grad_norm": 8.089363808893046,
+      "learning_rate": 2e-05,
+      "loss": 0.6604873180389405,
+      "step": 3960
+    },
+    {
+      "epoch": 0.19140832168169328,
+      "grad_norm": 1.407317481179978,
+      "learning_rate": 2e-05,
+      "loss": 0.6561919212341308,
+      "step": 3970
+    },
+    {
+      "epoch": 0.19189045851212574,
+      "grad_norm": 1.2741877061498277,
+      "learning_rate": 2e-05,
+      "loss": 0.6372320175170898,
+      "step": 3980
+    },
+    {
+      "epoch": 0.1923725953425582,
+      "grad_norm": 1.0222292379660791,
+      "learning_rate": 2e-05,
+      "loss": 0.6546836376190186,
+      "step": 3990
+    },
+    {
+      "epoch": 0.1928547321729907,
+      "grad_norm": 1.2109651218513655,
+      "learning_rate": 2e-05,
+      "loss": 0.6561654567718506,
+      "step": 4000
+    },
+    {
+      "epoch": 0.19333686900342317,
+      "grad_norm": 1.3559963225620308,
+      "learning_rate": 2e-05,
+      "loss": 0.6688140869140625,
+      "step": 4010
+    },
+    {
+      "epoch": 0.19381900583385564,
+      "grad_norm": 5.991866731552668,
+      "learning_rate": 2e-05,
+      "loss": 0.6864636421203614,
+      "step": 4020
+    },
+    {
+      "epoch": 0.19430114266428813,
+      "grad_norm": 1.6425505570703132,
+      "learning_rate": 2e-05,
+      "loss": 0.6600022315979004,
+      "step": 4030
+    },
+    {
+      "epoch": 0.1947832794947206,
+      "grad_norm": 0.9471804168745132,
+      "learning_rate": 2e-05,
+      "loss": 0.6522530555725098,
+      "step": 4040
+    },
+    {
+      "epoch": 0.19526541632515307,
+      "grad_norm": 1.1117834906672652,
+      "learning_rate": 2e-05,
+      "loss": 0.6452474594116211,
+      "step": 4050
+    },
+    {
+      "epoch": 0.19574755315558556,
+      "grad_norm": 1.0043318760452347,
+      "learning_rate": 2e-05,
+      "loss": 0.6822322845458985,
+      "step": 4060
+    },
+    {
+      "epoch": 0.19622968998601803,
+      "grad_norm": 1.7208919388142667,
+      "learning_rate": 2e-05,
+      "loss": 0.6523431777954102,
+      "step": 4070
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 5935,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 2035,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1927226987642880.0,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

400M/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e04372c0377560216662710081ce91385f58bcf95862bebd29a7102741e0347e
+size 7313