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:c333b24a4826f07d6bac7c0b110dc0770737ca84095516acfe966c3c123a9d4f
+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": 0.0794099857569313,
+  "eval_steps": 500,
+  "global_step": 4070,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0001951105301153103,
+      "grad_norm": 1290.1151800966732,
+      "learning_rate": 2.910112359550562e-06,
+      "loss": 1.6666240692138672,
+      "step": 10
+    },
+    {
+      "epoch": 0.0003902210602306206,
+      "grad_norm": 2094.709006858091,
+      "learning_rate": 3.921348314606742e-06,
+      "loss": 0.9677549362182617,
+      "step": 20
+    },
+    {
+      "epoch": 0.000585331590345931,
+      "grad_norm": 30.714148238594614,
+      "learning_rate": 4.932584269662922e-06,
+      "loss": 0.7858523368835449,
+      "step": 30
+    },
+    {
+      "epoch": 0.0007804421204612412,
+      "grad_norm": 497.9186820845375,
+      "learning_rate": 5.943820224719102e-06,
+      "loss": 0.7518716812133789,
+      "step": 40
+    },
+    {
+      "epoch": 0.0009755526505765516,
+      "grad_norm": 40.68707475383215,
+      "learning_rate": 6.955056179775282e-06,
+      "loss": 0.7165522575378418,
+      "step": 50
+    },
+    {
+      "epoch": 0.001170663180691862,
+      "grad_norm": 49.16345817267902,
+      "learning_rate": 7.966292134831462e-06,
+      "loss": 0.6980477333068847,
+      "step": 60
+    },
+    {
+      "epoch": 0.0013657737108071723,
+      "grad_norm": 110.36540709084935,
+      "learning_rate": 8.977528089887641e-06,
+      "loss": 0.6940877437591553,
+      "step": 70
+    },
+    {
+      "epoch": 0.0015608842409224825,
+      "grad_norm": 45.22696804420163,
+      "learning_rate": 9.988764044943822e-06,
+      "loss": 0.7629052639007569,
+      "step": 80
+    },
+    {
+      "epoch": 0.001755994771037793,
+      "grad_norm": 8.589551842197837,
+      "learning_rate": 1.1000000000000001e-05,
+      "loss": 0.7278446197509766,
+      "step": 90
+    },
+    {
+      "epoch": 0.0019511053011531032,
+      "grad_norm": 22.256653345446292,
+      "learning_rate": 1.2011235955056182e-05,
+      "loss": 0.7335631370544433,
+      "step": 100
+    },
+    {
+      "epoch": 0.0021462158312684134,
+      "grad_norm": 37.8295200972987,
+      "learning_rate": 1.3022471910112362e-05,
+      "loss": 0.7461274623870849,
+      "step": 110
+    },
+    {
+      "epoch": 0.002341326361383724,
+      "grad_norm": 15.77833948739335,
+      "learning_rate": 1.403370786516854e-05,
+      "loss": 0.7302251338958741,
+      "step": 120
+    },
+    {
+      "epoch": 0.0025364368914990343,
+      "grad_norm": 662.5019348463974,
+      "learning_rate": 1.504494382022472e-05,
+      "loss": 0.7099417686462403,
+      "step": 130
+    },
+    {
+      "epoch": 0.0027315474216143445,
+      "grad_norm": 8.07873176220526,
+      "learning_rate": 1.60561797752809e-05,
+      "loss": 0.8300569534301758,
+      "step": 140
+    },
+    {
+      "epoch": 0.0029266579517296548,
+      "grad_norm": 108.63732872651582,
+      "learning_rate": 1.706741573033708e-05,
+      "loss": 0.7591472148895264,
+      "step": 150
+    },
+    {
+      "epoch": 0.003121768481844965,
+      "grad_norm": 12.742460468210709,
+      "learning_rate": 1.8078651685393256e-05,
+      "loss": 0.7846505641937256,
+      "step": 160
+    },
+    {
+      "epoch": 0.0033168790119602757,
+      "grad_norm": 19.46685500550541,
+      "learning_rate": 1.908988764044944e-05,
+      "loss": 0.7010777473449707,
+      "step": 170
+    },
+    {
+      "epoch": 0.003511989542075586,
+      "grad_norm": 3.79492579495788,
+      "learning_rate": 2e-05,
+      "loss": 0.7221960067749024,
+      "step": 180
+    },
+    {
+      "epoch": 0.003707100072190896,
+      "grad_norm": 3.761910629826001,
+      "learning_rate": 2e-05,
+      "loss": 0.7448129653930664,
+      "step": 190
+    },
+    {
+      "epoch": 0.0039022106023062064,
+      "grad_norm": 4.471289857736554,
+      "learning_rate": 2e-05,
+      "loss": 0.7120450496673584,
+      "step": 200
+    },
+    {
+      "epoch": 0.004097321132421517,
+      "grad_norm": 7.832558161000839,
+      "learning_rate": 2e-05,
+      "loss": 0.7076560974121093,
+      "step": 210
+    },
+    {
+      "epoch": 0.004292431662536827,
+      "grad_norm": 24.625069340034873,
+      "learning_rate": 2e-05,
+      "loss": 0.7173271179199219,
+      "step": 220
+    },
+    {
+      "epoch": 0.004487542192652137,
+      "grad_norm": 10.643240144364263,
+      "learning_rate": 2e-05,
+      "loss": 0.8566647529602051,
+      "step": 230
+    },
+    {
+      "epoch": 0.004682652722767448,
+      "grad_norm": 17.191035538928677,
+      "learning_rate": 2e-05,
+      "loss": 0.7157638549804688,
+      "step": 240
+    },
+    {
+      "epoch": 0.004877763252882758,
+      "grad_norm": 13.627404303033932,
+      "learning_rate": 2e-05,
+      "loss": 0.7801154136657715,
+      "step": 250
+    },
+    {
+      "epoch": 0.005072873782998069,
+      "grad_norm": 24.99914666002505,
+      "learning_rate": 2e-05,
+      "loss": 0.7654953002929688,
+      "step": 260
+    },
+    {
+      "epoch": 0.005267984313113379,
+      "grad_norm": 147.25366832545367,
+      "learning_rate": 2e-05,
+      "loss": 0.7404791831970214,
+      "step": 270
+    },
+    {
+      "epoch": 0.005463094843228689,
+      "grad_norm": 6.625240044249601,
+      "learning_rate": 2e-05,
+      "loss": 0.7907636165618896,
+      "step": 280
+    },
+    {
+      "epoch": 0.005658205373343999,
+      "grad_norm": 11.741773313405414,
+      "learning_rate": 2e-05,
+      "loss": 0.7048698425292969,
+      "step": 290
+    },
+    {
+      "epoch": 0.0058533159034593095,
+      "grad_norm": 5.519224801929464,
+      "learning_rate": 2e-05,
+      "loss": 0.7155918121337891,
+      "step": 300
+    },
+    {
+      "epoch": 0.00604842643357462,
+      "grad_norm": 11.598372302819296,
+      "learning_rate": 2e-05,
+      "loss": 0.7381957054138184,
+      "step": 310
+    },
+    {
+      "epoch": 0.00624353696368993,
+      "grad_norm": 9.444939061369409,
+      "learning_rate": 2e-05,
+      "loss": 0.7207625865936279,
+      "step": 320
+    },
+    {
+      "epoch": 0.006438647493805241,
+      "grad_norm": 7.801173531822594,
+      "learning_rate": 2e-05,
+      "loss": 0.723687744140625,
+      "step": 330
+    },
+    {
+      "epoch": 0.006633758023920551,
+      "grad_norm": 1.9811128361432186,
+      "learning_rate": 2e-05,
+      "loss": 0.7044608116149902,
+      "step": 340
+    },
+    {
+      "epoch": 0.0068288685540358616,
+      "grad_norm": 5.818968297777697,
+      "learning_rate": 2e-05,
+      "loss": 0.7668921947479248,
+      "step": 350
+    },
+    {
+      "epoch": 0.007023979084151172,
+      "grad_norm": 3.944741659070438,
+      "learning_rate": 2e-05,
+      "loss": 0.7118455410003662,
+      "step": 360
+    },
+    {
+      "epoch": 0.007219089614266482,
+      "grad_norm": 4.311433759305494,
+      "learning_rate": 2e-05,
+      "loss": 0.7700147151947021,
+      "step": 370
+    },
+    {
+      "epoch": 0.007414200144381792,
+      "grad_norm": 30.425032251061474,
+      "learning_rate": 2e-05,
+      "loss": 0.7222751140594482,
+      "step": 380
+    },
+    {
+      "epoch": 0.0076093106744971025,
+      "grad_norm": 2.782539467868993,
+      "learning_rate": 2e-05,
+      "loss": 0.6802415370941162,
+      "step": 390
+    },
+    {
+      "epoch": 0.007804421204612413,
+      "grad_norm": 6.499874836624361,
+      "learning_rate": 2e-05,
+      "loss": 0.6992295742034912,
+      "step": 400
+    },
+    {
+      "epoch": 0.007999531734727724,
+      "grad_norm": 5.632464299732364,
+      "learning_rate": 2e-05,
+      "loss": 0.7049863815307618,
+      "step": 410
+    },
+    {
+      "epoch": 0.008194642264843033,
+      "grad_norm": 11.183347233038614,
+      "learning_rate": 2e-05,
+      "loss": 0.7416215419769288,
+      "step": 420
+    },
+    {
+      "epoch": 0.008389752794958344,
+      "grad_norm": 7.528401125738135,
+      "learning_rate": 2e-05,
+      "loss": 0.7592568397521973,
+      "step": 430
+    },
+    {
+      "epoch": 0.008584863325073654,
+      "grad_norm": 5.002885634433326,
+      "learning_rate": 2e-05,
+      "loss": 0.756533145904541,
+      "step": 440
+    },
+    {
+      "epoch": 0.008779973855188965,
+      "grad_norm": 5.66603343994339,
+      "learning_rate": 2e-05,
+      "loss": 0.7430764198303222,
+      "step": 450
+    },
+    {
+      "epoch": 0.008975084385304274,
+      "grad_norm": 22.314858498949487,
+      "learning_rate": 2e-05,
+      "loss": 0.7967084884643555,
+      "step": 460
+    },
+    {
+      "epoch": 0.009170194915419585,
+      "grad_norm": 5.642840192700939,
+      "learning_rate": 2e-05,
+      "loss": 0.7398099422454834,
+      "step": 470
+    },
+    {
+      "epoch": 0.009365305445534896,
+      "grad_norm": 2.0656284529162243,
+      "learning_rate": 2e-05,
+      "loss": 0.7248655319213867,
+      "step": 480
+    },
+    {
+      "epoch": 0.009560415975650206,
+      "grad_norm": 4.657301334726203,
+      "learning_rate": 2e-05,
+      "loss": 0.7701823711395264,
+      "step": 490
+    },
+    {
+      "epoch": 0.009755526505765517,
+      "grad_norm": 4.995971591382155,
+      "learning_rate": 2e-05,
+      "loss": 0.733971357345581,
+      "step": 500
+    },
+    {
+      "epoch": 0.009950637035880826,
+      "grad_norm": 11.657412731021536,
+      "learning_rate": 2e-05,
+      "loss": 0.802979564666748,
+      "step": 510
+    },
+    {
+      "epoch": 0.010145747565996137,
+      "grad_norm": 3.161728049215805,
+      "learning_rate": 2e-05,
+      "loss": 0.753895378112793,
+      "step": 520
+    },
+    {
+      "epoch": 0.010340858096111447,
+      "grad_norm": 24.0896519055277,
+      "learning_rate": 2e-05,
+      "loss": 0.7964776992797852,
+      "step": 530
+    },
+    {
+      "epoch": 0.010535968626226758,
+      "grad_norm": 8.361986893202518,
+      "learning_rate": 2e-05,
+      "loss": 0.7574877738952637,
+      "step": 540
+    },
+    {
+      "epoch": 0.010731079156342067,
+      "grad_norm": 10.53310762606766,
+      "learning_rate": 2e-05,
+      "loss": 0.7854220867156982,
+      "step": 550
+    },
+    {
+      "epoch": 0.010926189686457378,
+      "grad_norm": 96.65480630904233,
+      "learning_rate": 2e-05,
+      "loss": 0.7968769073486328,
+      "step": 560
+    },
+    {
+      "epoch": 0.01112130021657269,
+      "grad_norm": 1.8376056131220815,
+      "learning_rate": 2e-05,
+      "loss": 0.7898524284362793,
+      "step": 570
+    },
+    {
+      "epoch": 0.011316410746687999,
+      "grad_norm": 1.446906475407358,
+      "learning_rate": 2e-05,
+      "loss": 0.7536652565002442,
+      "step": 580
+    },
+    {
+      "epoch": 0.01151152127680331,
+      "grad_norm": 2.63100640718635,
+      "learning_rate": 2e-05,
+      "loss": 0.7614954948425293,
+      "step": 590
+    },
+    {
+      "epoch": 0.011706631806918619,
+      "grad_norm": 1.7292424310123347,
+      "learning_rate": 2e-05,
+      "loss": 0.6910699844360352,
+      "step": 600
+    },
+    {
+      "epoch": 0.01190174233703393,
+      "grad_norm": 1.6185308860071268,
+      "learning_rate": 2e-05,
+      "loss": 0.7510101318359375,
+      "step": 610
+    },
+    {
+      "epoch": 0.01209685286714924,
+      "grad_norm": 9.938109528270555,
+      "learning_rate": 2e-05,
+      "loss": 0.7183078765869141,
+      "step": 620
+    },
+    {
+      "epoch": 0.01229196339726455,
+      "grad_norm": 3.1451531777998762,
+      "learning_rate": 2e-05,
+      "loss": 0.6925305843353271,
+      "step": 630
+    },
+    {
+      "epoch": 0.01248707392737986,
+      "grad_norm": 11.976814414360483,
+      "learning_rate": 2e-05,
+      "loss": 0.6874044418334961,
+      "step": 640
+    },
+    {
+      "epoch": 0.012682184457495171,
+      "grad_norm": 2.512055129296439,
+      "learning_rate": 2e-05,
+      "loss": 0.7078936576843262,
+      "step": 650
+    },
+    {
+      "epoch": 0.012877294987610482,
+      "grad_norm": 21.321440055444356,
+      "learning_rate": 2e-05,
+      "loss": 0.7755936622619629,
+      "step": 660
+    },
+    {
+      "epoch": 0.013072405517725792,
+      "grad_norm": 6.027137898134079,
+      "learning_rate": 2e-05,
+      "loss": 0.6783116817474365,
+      "step": 670
+    },
+    {
+      "epoch": 0.013267516047841103,
+      "grad_norm": 1.6349976601766352,
+      "learning_rate": 2e-05,
+      "loss": 0.7179695129394531,
+      "step": 680
+    },
+    {
+      "epoch": 0.013462626577956412,
+      "grad_norm": 2.696759389003869,
+      "learning_rate": 2e-05,
+      "loss": 0.6627010345458985,
+      "step": 690
+    },
+    {
+      "epoch": 0.013657737108071723,
+      "grad_norm": 3.094531262148124,
+      "learning_rate": 2e-05,
+      "loss": 0.691745662689209,
+      "step": 700
+    },
+    {
+      "epoch": 0.013852847638187032,
+      "grad_norm": 1.3167065811312815,
+      "learning_rate": 2e-05,
+      "loss": 0.6898958206176757,
+      "step": 710
+    },
+    {
+      "epoch": 0.014047958168302344,
+      "grad_norm": 3.096688041091074,
+      "learning_rate": 2e-05,
+      "loss": 0.6983632564544677,
+      "step": 720
+    },
+    {
+      "epoch": 0.014243068698417653,
+      "grad_norm": 1.5732201392921932,
+      "learning_rate": 2e-05,
+      "loss": 0.6881768703460693,
+      "step": 730
+    },
+    {
+      "epoch": 0.014438179228532964,
+      "grad_norm": 4.179444784177794,
+      "learning_rate": 2e-05,
+      "loss": 0.6782029628753662,
+      "step": 740
+    },
+    {
+      "epoch": 0.014633289758648273,
+      "grad_norm": 13.022740345020917,
+      "learning_rate": 2e-05,
+      "loss": 0.7157995223999023,
+      "step": 750
+    },
+    {
+      "epoch": 0.014828400288763584,
+      "grad_norm": 2.227508854051503,
+      "learning_rate": 2e-05,
+      "loss": 0.7245219230651856,
+      "step": 760
+    },
+    {
+      "epoch": 0.015023510818878896,
+      "grad_norm": 3.4074041057703774,
+      "learning_rate": 2e-05,
+      "loss": 0.7473674297332764,
+      "step": 770
+    },
+    {
+      "epoch": 0.015218621348994205,
+      "grad_norm": 4.053610258594328,
+      "learning_rate": 2e-05,
+      "loss": 0.664447021484375,
+      "step": 780
+    },
+    {
+      "epoch": 0.015413731879109516,
+      "grad_norm": 1.6609484317874577,
+      "learning_rate": 2e-05,
+      "loss": 0.6886637687683106,
+      "step": 790
+    },
+    {
+      "epoch": 0.015608842409224825,
+      "grad_norm": 1.5444399305626002,
+      "learning_rate": 2e-05,
+      "loss": 0.7088738918304444,
+      "step": 800
+    },
+    {
+      "epoch": 0.015803952939340137,
+      "grad_norm": 3.1762223604517237,
+      "learning_rate": 2e-05,
+      "loss": 0.7194277763366699,
+      "step": 810
+    },
+    {
+      "epoch": 0.015999063469455448,
+      "grad_norm": 3.8350609994341527,
+      "learning_rate": 2e-05,
+      "loss": 0.7160842895507813,
+      "step": 820
+    },
+    {
+      "epoch": 0.016194173999570755,
+      "grad_norm": 2.315457264069663,
+      "learning_rate": 2e-05,
+      "loss": 0.6816580772399903,
+      "step": 830
+    },
+    {
+      "epoch": 0.016389284529686066,
+      "grad_norm": 3.1340139053276896,
+      "learning_rate": 2e-05,
+      "loss": 0.6640499114990235,
+      "step": 840
+    },
+    {
+      "epoch": 0.016584395059801377,
+      "grad_norm": 2.103288119861107,
+      "learning_rate": 2e-05,
+      "loss": 0.6783723354339599,
+      "step": 850
+    },
+    {
+      "epoch": 0.01677950558991669,
+      "grad_norm": 3.7306310450926383,
+      "learning_rate": 2e-05,
+      "loss": 0.6990603446960449,
+      "step": 860
+    },
+    {
+      "epoch": 0.016974616120032,
+      "grad_norm": 2.3246559374347955,
+      "learning_rate": 2e-05,
+      "loss": 0.6684437751770019,
+      "step": 870
+    },
+    {
+      "epoch": 0.017169726650147307,
+      "grad_norm": 2.4616580742100016,
+      "learning_rate": 2e-05,
+      "loss": 0.6759738922119141,
+      "step": 880
+    },
+    {
+      "epoch": 0.01736483718026262,
+      "grad_norm": 1.4211835502848627,
+      "learning_rate": 2e-05,
+      "loss": 0.7110819816589355,
+      "step": 890
+    },
+    {
+      "epoch": 0.01755994771037793,
+      "grad_norm": 3.113362922049139,
+      "learning_rate": 2e-05,
+      "loss": 0.7206232070922851,
+      "step": 900
+    },
+    {
+      "epoch": 0.01775505824049324,
+      "grad_norm": 1.5202175869081638,
+      "learning_rate": 2e-05,
+      "loss": 0.6578310012817383,
+      "step": 910
+    },
+    {
+      "epoch": 0.017950168770608548,
+      "grad_norm": 1.610151123997545,
+      "learning_rate": 2e-05,
+      "loss": 0.7033986568450927,
+      "step": 920
+    },
+    {
+      "epoch": 0.01814527930072386,
+      "grad_norm": 2.213950402852014,
+      "learning_rate": 2e-05,
+      "loss": 0.7181561946868896,
+      "step": 930
+    },
+    {
+      "epoch": 0.01834038983083917,
+      "grad_norm": 1.0525256653678736,
+      "learning_rate": 2e-05,
+      "loss": 0.64858980178833,
+      "step": 940
+    },
+    {
+      "epoch": 0.01853550036095448,
+      "grad_norm": 1.1535760888310356,
+      "learning_rate": 2e-05,
+      "loss": 0.6768205642700196,
+      "step": 950
+    },
+    {
+      "epoch": 0.018730610891069793,
+      "grad_norm": 2.5687267328441172,
+      "learning_rate": 2e-05,
+      "loss": 0.7187290191650391,
+      "step": 960
+    },
+    {
+      "epoch": 0.0189257214211851,
+      "grad_norm": 2.363644647745971,
+      "learning_rate": 2e-05,
+      "loss": 0.6914881706237793,
+      "step": 970
+    },
+    {
+      "epoch": 0.01912083195130041,
+      "grad_norm": 1.6586735289269663,
+      "learning_rate": 2e-05,
+      "loss": 0.7053565979003906,
+      "step": 980
+    },
+    {
+      "epoch": 0.019315942481415722,
+      "grad_norm": 1.3154384327675313,
+      "learning_rate": 2e-05,
+      "loss": 0.6894444942474365,
+      "step": 990
+    },
+    {
+      "epoch": 0.019511053011531033,
+      "grad_norm": 1.4829582302075024,
+      "learning_rate": 2e-05,
+      "loss": 0.6964797019958496,
+      "step": 1000
+    },
+    {
+      "epoch": 0.01970616354164634,
+      "grad_norm": 11.246853734884391,
+      "learning_rate": 2e-05,
+      "loss": 0.6903947830200196,
+      "step": 1010
+    },
+    {
+      "epoch": 0.019901274071761652,
+      "grad_norm": 2.130270630304158,
+      "learning_rate": 2e-05,
+      "loss": 0.7131201744079589,
+      "step": 1020
+    },
+    {
+      "epoch": 0.020096384601876963,
+      "grad_norm": 2.3843191059461684,
+      "learning_rate": 2e-05,
+      "loss": 0.7245532989501953,
+      "step": 1030
+    },
+    {
+      "epoch": 0.020291495131992274,
+      "grad_norm": 1.8036658748418921,
+      "learning_rate": 2e-05,
+      "loss": 0.688844633102417,
+      "step": 1040
+    },
+    {
+      "epoch": 0.020486605662107586,
+      "grad_norm": 4.142854989409903,
+      "learning_rate": 2e-05,
+      "loss": 0.6744081020355225,
+      "step": 1050
+    },
+    {
+      "epoch": 0.020681716192222893,
+      "grad_norm": 1.472431946736185,
+      "learning_rate": 2e-05,
+      "loss": 0.6328347682952881,
+      "step": 1060
+    },
+    {
+      "epoch": 0.020876826722338204,
+      "grad_norm": 2.390747409072109,
+      "learning_rate": 2e-05,
+      "loss": 0.688608455657959,
+      "step": 1070
+    },
+    {
+      "epoch": 0.021071937252453515,
+      "grad_norm": 2.538662996092993,
+      "learning_rate": 2e-05,
+      "loss": 0.7029477119445801,
+      "step": 1080
+    },
+    {
+      "epoch": 0.021267047782568826,
+      "grad_norm": 2.4656407868003436,
+      "learning_rate": 2e-05,
+      "loss": 0.6541564464569092,
+      "step": 1090
+    },
+    {
+      "epoch": 0.021462158312684134,
+      "grad_norm": 1.4483652415733586,
+      "learning_rate": 2e-05,
+      "loss": 0.6665634155273438,
+      "step": 1100
+    },
+    {
+      "epoch": 0.021657268842799445,
+      "grad_norm": 3.7600001493285813,
+      "learning_rate": 2e-05,
+      "loss": 0.6685405731201172,
+      "step": 1110
+    },
+    {
+      "epoch": 0.021852379372914756,
+      "grad_norm": 4.4698831654409785,
+      "learning_rate": 2e-05,
+      "loss": 0.6686572074890137,
+      "step": 1120
+    },
+    {
+      "epoch": 0.022047489903030067,
+      "grad_norm": 1.2416499167615647,
+      "learning_rate": 2e-05,
+      "loss": 0.6792525291442871,
+      "step": 1130
+    },
+    {
+      "epoch": 0.02224260043314538,
+      "grad_norm": 1.261439766703017,
+      "learning_rate": 2e-05,
+      "loss": 0.6553254127502441,
+      "step": 1140
+    },
+    {
+      "epoch": 0.022437710963260686,
+      "grad_norm": 3.0580767819341967,
+      "learning_rate": 2e-05,
+      "loss": 0.670100736618042,
+      "step": 1150
+    },
+    {
+      "epoch": 0.022632821493375997,
+      "grad_norm": 1.9982728200546733,
+      "learning_rate": 2e-05,
+      "loss": 0.6652833461761475,
+      "step": 1160
+    },
+    {
+      "epoch": 0.02282793202349131,
+      "grad_norm": 1.4478470019999148,
+      "learning_rate": 2e-05,
+      "loss": 0.6802016258239746,
+      "step": 1170
+    },
+    {
+      "epoch": 0.02302304255360662,
+      "grad_norm": 1.390262972826178,
+      "learning_rate": 2e-05,
+      "loss": 0.6520106792449951,
+      "step": 1180
+    },
+    {
+      "epoch": 0.023218153083721927,
+      "grad_norm": 1.2517531765951326,
+      "learning_rate": 2e-05,
+      "loss": 0.6596714973449707,
+      "step": 1190
+    },
+    {
+      "epoch": 0.023413263613837238,
+      "grad_norm": 1.66285797208597,
+      "learning_rate": 2e-05,
+      "loss": 0.6890499114990234,
+      "step": 1200
+    },
+    {
+      "epoch": 0.02360837414395255,
+      "grad_norm": 1.4023367573944592,
+      "learning_rate": 2e-05,
+      "loss": 0.6909699440002441,
+      "step": 1210
+    },
+    {
+      "epoch": 0.02380348467406786,
+      "grad_norm": 2.103600383565939,
+      "learning_rate": 2e-05,
+      "loss": 0.6514087677001953,
+      "step": 1220
+    },
+    {
+      "epoch": 0.02399859520418317,
+      "grad_norm": 1.3538205014153204,
+      "learning_rate": 2e-05,
+      "loss": 0.7010597229003906,
+      "step": 1230
+    },
+    {
+      "epoch": 0.02419370573429848,
+      "grad_norm": 1.131100957434826,
+      "learning_rate": 2e-05,
+      "loss": 0.6904762268066407,
+      "step": 1240
+    },
+    {
+      "epoch": 0.02438881626441379,
+      "grad_norm": 0.8521222417760329,
+      "learning_rate": 2e-05,
+      "loss": 0.6021539688110351,
+      "step": 1250
+    },
+    {
+      "epoch": 0.0245839267945291,
+      "grad_norm": 0.9985116086981741,
+      "learning_rate": 2e-05,
+      "loss": 0.6618970394134521,
+      "step": 1260
+    },
+    {
+      "epoch": 0.024779037324644412,
+      "grad_norm": 1.0359703256994235,
+      "learning_rate": 2e-05,
+      "loss": 0.6823967933654785,
+      "step": 1270
+    },
+    {
+      "epoch": 0.02497414785475972,
+      "grad_norm": 1.2887090486112476,
+      "learning_rate": 2e-05,
+      "loss": 0.6628634452819824,
+      "step": 1280
+    },
+    {
+      "epoch": 0.02516925838487503,
+      "grad_norm": 3.432834498196539,
+      "learning_rate": 2e-05,
+      "loss": 0.7419169425964356,
+      "step": 1290
+    },
+    {
+      "epoch": 0.025364368914990342,
+      "grad_norm": 23.557815006813602,
+      "learning_rate": 2e-05,
+      "loss": 0.7083474636077881,
+      "step": 1300
+    },
+    {
+      "epoch": 0.025559479445105653,
+      "grad_norm": 1.3779699418257583,
+      "learning_rate": 2e-05,
+      "loss": 0.6835464477539063,
+      "step": 1310
+    },
+    {
+      "epoch": 0.025754589975220964,
+      "grad_norm": 1.0971280641281729,
+      "learning_rate": 2e-05,
+      "loss": 0.6729559898376465,
+      "step": 1320
+    },
+    {
+      "epoch": 0.025949700505336272,
+      "grad_norm": 86.50430124250552,
+      "learning_rate": 2e-05,
+      "loss": 0.7425766944885254,
+      "step": 1330
+    },
+    {
+      "epoch": 0.026144811035451583,
+      "grad_norm": 1.9060666983098362,
+      "learning_rate": 2e-05,
+      "loss": 0.6884101867675781,
+      "step": 1340
+    },
+    {
+      "epoch": 0.026339921565566894,
+      "grad_norm": 4.952310768605313,
+      "learning_rate": 2e-05,
+      "loss": 0.6712810516357421,
+      "step": 1350
+    },
+    {
+      "epoch": 0.026535032095682205,
+      "grad_norm": 2.1836902499589055,
+      "learning_rate": 2e-05,
+      "loss": 0.6641899108886719,
+      "step": 1360
+    },
+    {
+      "epoch": 0.026730142625797513,
+      "grad_norm": 0.9745030121353373,
+      "learning_rate": 2e-05,
+      "loss": 0.6548121929168701,
+      "step": 1370
+    },
+    {
+      "epoch": 0.026925253155912824,
+      "grad_norm": 1.0085691573367601,
+      "learning_rate": 2e-05,
+      "loss": 0.6872680187225342,
+      "step": 1380
+    },
+    {
+      "epoch": 0.027120363686028135,
+      "grad_norm": 1.0328562743257255,
+      "learning_rate": 2e-05,
+      "loss": 0.6593247413635254,
+      "step": 1390
+    },
+    {
+      "epoch": 0.027315474216143446,
+      "grad_norm": 1.7706110276938303,
+      "learning_rate": 2e-05,
+      "loss": 0.6563483715057373,
+      "step": 1400
+    },
+    {
+      "epoch": 0.027510584746258757,
+      "grad_norm": 1.1475044260212954,
+      "learning_rate": 2e-05,
+      "loss": 0.6485116004943847,
+      "step": 1410
+    },
+    {
+      "epoch": 0.027705695276374065,
+      "grad_norm": 2.1343320521467724,
+      "learning_rate": 2e-05,
+      "loss": 0.6518092155456543,
+      "step": 1420
+    },
+    {
+      "epoch": 0.027900805806489376,
+      "grad_norm": 19.40190752512825,
+      "learning_rate": 2e-05,
+      "loss": 0.6708407402038574,
+      "step": 1430
+    },
+    {
+      "epoch": 0.028095916336604687,
+      "grad_norm": 1.0429828458514099,
+      "learning_rate": 2e-05,
+      "loss": 0.6531154632568359,
+      "step": 1440
+    },
+    {
+      "epoch": 0.028291026866719998,
+      "grad_norm": 3.854687583398717,
+      "learning_rate": 2e-05,
+      "loss": 0.6318797588348388,
+      "step": 1450
+    },
+    {
+      "epoch": 0.028486137396835306,
+      "grad_norm": 1.2803239369157802,
+      "learning_rate": 2e-05,
+      "loss": 0.6483430385589599,
+      "step": 1460
+    },
+    {
+      "epoch": 0.028681247926950617,
+      "grad_norm": 1.7844005133176821,
+      "learning_rate": 2e-05,
+      "loss": 0.670600414276123,
+      "step": 1470
+    },
+    {
+      "epoch": 0.028876358457065928,
+      "grad_norm": 2.638360353786163,
+      "learning_rate": 2e-05,
+      "loss": 0.6612592220306397,
+      "step": 1480
+    },
+    {
+      "epoch": 0.02907146898718124,
+      "grad_norm": 1.2239072376785438,
+      "learning_rate": 2e-05,
+      "loss": 0.6572507381439209,
+      "step": 1490
+    },
+    {
+      "epoch": 0.029266579517296547,
+      "grad_norm": 3.9817061848469457,
+      "learning_rate": 2e-05,
+      "loss": 0.631273603439331,
+      "step": 1500
+    },
+    {
+      "epoch": 0.029461690047411858,
+      "grad_norm": 1.2083656296744134,
+      "learning_rate": 2e-05,
+      "loss": 0.6754149913787841,
+      "step": 1510
+    },
+    {
+      "epoch": 0.02965680057752717,
+      "grad_norm": 1.6701538309486328,
+      "learning_rate": 2e-05,
+      "loss": 0.6359077453613281,
+      "step": 1520
+    },
+    {
+      "epoch": 0.02985191110764248,
+      "grad_norm": 1.1088864730268388,
+      "learning_rate": 2e-05,
+      "loss": 0.6407184600830078,
+      "step": 1530
+    },
+    {
+      "epoch": 0.03004702163775779,
+      "grad_norm": 2.855041461691364,
+      "learning_rate": 2e-05,
+      "loss": 0.6601726531982421,
+      "step": 1540
+    },
+    {
+      "epoch": 0.0302421321678731,
+      "grad_norm": 1.041676019536129,
+      "learning_rate": 2e-05,
+      "loss": 0.6693766593933106,
+      "step": 1550
+    },
+    {
+      "epoch": 0.03043724269798841,
+      "grad_norm": 1.7903036215917778,
+      "learning_rate": 2e-05,
+      "loss": 0.6842814445495605,
+      "step": 1560
+    },
+    {
+      "epoch": 0.03063235322810372,
+      "grad_norm": 1.2275140901739492,
+      "learning_rate": 2e-05,
+      "loss": 0.6563383102416992,
+      "step": 1570
+    },
+    {
+      "epoch": 0.030827463758219032,
+      "grad_norm": 1.4416525013205308,
+      "learning_rate": 2e-05,
+      "loss": 0.70062255859375,
+      "step": 1580
+    },
+    {
+      "epoch": 0.03102257428833434,
+      "grad_norm": 1.022694275409909,
+      "learning_rate": 2e-05,
+      "loss": 0.6751978874206543,
+      "step": 1590
+    },
+    {
+      "epoch": 0.03121768481844965,
+      "grad_norm": 1.5434339515015798,
+      "learning_rate": 2e-05,
+      "loss": 0.6534595489501953,
+      "step": 1600
+    },
+    {
+      "epoch": 0.03141279534856496,
+      "grad_norm": 4.482223058250825,
+      "learning_rate": 2e-05,
+      "loss": 0.679110336303711,
+      "step": 1610
+    },
+    {
+      "epoch": 0.03160790587868027,
+      "grad_norm": 1.269130777338052,
+      "learning_rate": 2e-05,
+      "loss": 0.6603041648864746,
+      "step": 1620
+    },
+    {
+      "epoch": 0.031803016408795584,
+      "grad_norm": 3.5729377498646477,
+      "learning_rate": 2e-05,
+      "loss": 0.6539816856384277,
+      "step": 1630
+    },
+    {
+      "epoch": 0.031998126938910895,
+      "grad_norm": 2.6023274014566247,
+      "learning_rate": 2e-05,
+      "loss": 0.7010405540466309,
+      "step": 1640
+    },
+    {
+      "epoch": 0.032193237469026206,
+      "grad_norm": 9.210714854982013,
+      "learning_rate": 2e-05,
+      "loss": 0.6686701774597168,
+      "step": 1650
+    },
+    {
+      "epoch": 0.03238834799914151,
+      "grad_norm": 5.165666638947227,
+      "learning_rate": 2e-05,
+      "loss": 0.6095908164978028,
+      "step": 1660
+    },
+    {
+      "epoch": 0.03258345852925682,
+      "grad_norm": 9.16141664239496,
+      "learning_rate": 2e-05,
+      "loss": 0.7108460903167725,
+      "step": 1670
+    },
+    {
+      "epoch": 0.03277856905937213,
+      "grad_norm": 3.161658879203255,
+      "learning_rate": 2e-05,
+      "loss": 0.6284823417663574,
+      "step": 1680
+    },
+    {
+      "epoch": 0.032973679589487444,
+      "grad_norm": 5.944141446716445,
+      "learning_rate": 2e-05,
+      "loss": 0.6405398845672607,
+      "step": 1690
+    },
+    {
+      "epoch": 0.033168790119602755,
+      "grad_norm": 1.7808423949894736,
+      "learning_rate": 2e-05,
+      "loss": 0.6491862297058105,
+      "step": 1700
+    },
+    {
+      "epoch": 0.033363900649718066,
+      "grad_norm": 1.726175898675113,
+      "learning_rate": 2e-05,
+      "loss": 0.6724005699157715,
+      "step": 1710
+    },
+    {
+      "epoch": 0.03355901117983338,
+      "grad_norm": 2.376245899157911,
+      "learning_rate": 2e-05,
+      "loss": 0.6525126934051514,
+      "step": 1720
+    },
+    {
+      "epoch": 0.03375412170994869,
+      "grad_norm": 1.6653879954838984,
+      "learning_rate": 2e-05,
+      "loss": 0.6367118835449219,
+      "step": 1730
+    },
+    {
+      "epoch": 0.033949232240064,
+      "grad_norm": 1.3883560501158865,
+      "learning_rate": 2e-05,
+      "loss": 0.6553079128265381,
+      "step": 1740
+    },
+    {
+      "epoch": 0.0341443427701793,
+      "grad_norm": 11.844497956339469,
+      "learning_rate": 2e-05,
+      "loss": 0.5972353935241699,
+      "step": 1750
+    },
+    {
+      "epoch": 0.034339453300294615,
+      "grad_norm": 1.5888196717274026,
+      "learning_rate": 2e-05,
+      "loss": 0.658538818359375,
+      "step": 1760
+    },
+    {
+      "epoch": 0.034534563830409926,
+      "grad_norm": 1.2587083645893846,
+      "learning_rate": 2e-05,
+      "loss": 0.6199719905853271,
+      "step": 1770
+    },
+    {
+      "epoch": 0.03472967436052524,
+      "grad_norm": 2.0204514597287266,
+      "learning_rate": 2e-05,
+      "loss": 0.6569772243499756,
+      "step": 1780
+    },
+    {
+      "epoch": 0.03492478489064055,
+      "grad_norm": 1.1323396524728921,
+      "learning_rate": 2e-05,
+      "loss": 0.6597754478454589,
+      "step": 1790
+    },
+    {
+      "epoch": 0.03511989542075586,
+      "grad_norm": 4.803084432276915,
+      "learning_rate": 2e-05,
+      "loss": 0.6832777976989746,
+      "step": 1800
+    },
+    {
+      "epoch": 0.03531500595087117,
+      "grad_norm": 1.2706249626426194,
+      "learning_rate": 2e-05,
+      "loss": 0.6612133026123047,
+      "step": 1810
+    },
+    {
+      "epoch": 0.03551011648098648,
+      "grad_norm": 2.888385667334179,
+      "learning_rate": 2e-05,
+      "loss": 0.6800965785980224,
+      "step": 1820
+    },
+    {
+      "epoch": 0.03570522701110179,
+      "grad_norm": 1.7994196905308342,
+      "learning_rate": 2e-05,
+      "loss": 0.6731705665588379,
+      "step": 1830
+    },
+    {
+      "epoch": 0.035900337541217096,
+      "grad_norm": 2.2229464990180463,
+      "learning_rate": 2e-05,
+      "loss": 0.6999088287353515,
+      "step": 1840
+    },
+    {
+      "epoch": 0.03609544807133241,
+      "grad_norm": 1.7473284581429838,
+      "learning_rate": 2e-05,
+      "loss": 0.665509557723999,
+      "step": 1850
+    },
+    {
+      "epoch": 0.03629055860144772,
+      "grad_norm": 2.2975536238797036,
+      "learning_rate": 2e-05,
+      "loss": 0.6647231101989746,
+      "step": 1860
+    },
+    {
+      "epoch": 0.03648566913156303,
+      "grad_norm": 2.6214877289682246,
+      "learning_rate": 2e-05,
+      "loss": 0.62739577293396,
+      "step": 1870
+    },
+    {
+      "epoch": 0.03668077966167834,
+      "grad_norm": 2.192373938003488,
+      "learning_rate": 2e-05,
+      "loss": 0.7087226390838623,
+      "step": 1880
+    },
+    {
+      "epoch": 0.03687589019179365,
+      "grad_norm": 6.62847066861488,
+      "learning_rate": 2e-05,
+      "loss": 0.6841737270355225,
+      "step": 1890
+    },
+    {
+      "epoch": 0.03707100072190896,
+      "grad_norm": 8.573387218725129,
+      "learning_rate": 2e-05,
+      "loss": 0.6550696849822998,
+      "step": 1900
+    },
+    {
+      "epoch": 0.037266111252024274,
+      "grad_norm": 21.81644893450406,
+      "learning_rate": 2e-05,
+      "loss": 0.6534019470214844,
+      "step": 1910
+    },
+    {
+      "epoch": 0.037461221782139585,
+      "grad_norm": 1.972315212149555,
+      "learning_rate": 2e-05,
+      "loss": 0.6699990272521973,
+      "step": 1920
+    },
+    {
+      "epoch": 0.03765633231225489,
+      "grad_norm": 1.572875023378223,
+      "learning_rate": 2e-05,
+      "loss": 0.6702420234680175,
+      "step": 1930
+    },
+    {
+      "epoch": 0.0378514428423702,
+      "grad_norm": 1.2077505710786207,
+      "learning_rate": 2e-05,
+      "loss": 0.6410272121429443,
+      "step": 1940
+    },
+    {
+      "epoch": 0.03804655337248551,
+      "grad_norm": 1.3936237416130755,
+      "learning_rate": 2e-05,
+      "loss": 0.6933211803436279,
+      "step": 1950
+    },
+    {
+      "epoch": 0.03824166390260082,
+      "grad_norm": 2.6969645830470323,
+      "learning_rate": 2e-05,
+      "loss": 0.6899444580078125,
+      "step": 1960
+    },
+    {
+      "epoch": 0.038436774432716134,
+      "grad_norm": 3.6883874679320745,
+      "learning_rate": 2e-05,
+      "loss": 0.661356258392334,
+      "step": 1970
+    },
+    {
+      "epoch": 0.038631884962831445,
+      "grad_norm": 4.1375939813313325,
+      "learning_rate": 2e-05,
+      "loss": 0.6723477840423584,
+      "step": 1980
+    },
+    {
+      "epoch": 0.038826995492946756,
+      "grad_norm": 8.712693611376435,
+      "learning_rate": 2e-05,
+      "loss": 0.6563708305358886,
+      "step": 1990
+    },
+    {
+      "epoch": 0.03902210602306207,
+      "grad_norm": 1.7546049733134041,
+      "learning_rate": 2e-05,
+      "loss": 0.6711672782897949,
+      "step": 2000
+    },
+    {
+      "epoch": 0.03921721655317738,
+      "grad_norm": 1.7813893766887805,
+      "learning_rate": 2e-05,
+      "loss": 0.6956442832946778,
+      "step": 2010
+    },
+    {
+      "epoch": 0.03941232708329268,
+      "grad_norm": 1.2906449372308841,
+      "learning_rate": 2e-05,
+      "loss": 0.6290652275085449,
+      "step": 2020
+    },
+    {
+      "epoch": 0.03960743761340799,
+      "grad_norm": 1.1252587686982205,
+      "learning_rate": 2e-05,
+      "loss": 0.6268731117248535,
+      "step": 2030
+    },
+    {
+      "epoch": 0.039802548143523304,
+      "grad_norm": 1.4021563501477514,
+      "learning_rate": 2e-05,
+      "loss": 0.6229870796203614,
+      "step": 2040
+    },
+    {
+      "epoch": 0.039997658673638616,
+      "grad_norm": 3.6802270890598137,
+      "learning_rate": 2e-05,
+      "loss": 0.6835996150970459,
+      "step": 2050
+    },
+    {
+      "epoch": 0.04019276920375393,
+      "grad_norm": 2.187332194734741,
+      "learning_rate": 2e-05,
+      "loss": 0.6701037406921386,
+      "step": 2060
+    },
+    {
+      "epoch": 0.04038787973386924,
+      "grad_norm": 1.3611685482292595,
+      "learning_rate": 2e-05,
+      "loss": 0.6770423412322998,
+      "step": 2070
+    },
+    {
+      "epoch": 0.04058299026398455,
+      "grad_norm": 1.3441815988847614,
+      "learning_rate": 2e-05,
+      "loss": 0.6499631881713868,
+      "step": 2080
+    },
+    {
+      "epoch": 0.04077810079409986,
+      "grad_norm": 9.12711526768921,
+      "learning_rate": 2e-05,
+      "loss": 0.683414077758789,
+      "step": 2090
+    },
+    {
+      "epoch": 0.04097321132421517,
+      "grad_norm": 1.4188932279527693,
+      "learning_rate": 2e-05,
+      "loss": 0.6575721263885498,
+      "step": 2100
+    },
+    {
+      "epoch": 0.041168321854330475,
+      "grad_norm": 3.713308065642342,
+      "learning_rate": 2e-05,
+      "loss": 0.6424132347106933,
+      "step": 2110
+    },
+    {
+      "epoch": 0.041363432384445786,
+      "grad_norm": 1.216789291881629,
+      "learning_rate": 2e-05,
+      "loss": 0.6082934379577637,
+      "step": 2120
+    },
+    {
+      "epoch": 0.0415585429145611,
+      "grad_norm": 5.887098747054894,
+      "learning_rate": 2e-05,
+      "loss": 0.6324620723724366,
+      "step": 2130
+    },
+    {
+      "epoch": 0.04175365344467641,
+      "grad_norm": 1.1070926483576238,
+      "learning_rate": 2e-05,
+      "loss": 0.6326879024505615,
+      "step": 2140
+    },
+    {
+      "epoch": 0.04194876397479172,
+      "grad_norm": 1.1785127557251094,
+      "learning_rate": 2e-05,
+      "loss": 0.6675579071044921,
+      "step": 2150
+    },
+    {
+      "epoch": 0.04214387450490703,
+      "grad_norm": 1.006559988082113,
+      "learning_rate": 2e-05,
+      "loss": 0.6432580471038818,
+      "step": 2160
+    },
+    {
+      "epoch": 0.04233898503502234,
+      "grad_norm": 2.716372651072617,
+      "learning_rate": 2e-05,
+      "loss": 0.7396750450134277,
+      "step": 2170
+    },
+    {
+      "epoch": 0.04253409556513765,
+      "grad_norm": 0.9718403133049374,
+      "learning_rate": 2e-05,
+      "loss": 0.6772015571594239,
+      "step": 2180
+    },
+    {
+      "epoch": 0.042729206095252964,
+      "grad_norm": 0.9005964057824478,
+      "learning_rate": 2e-05,
+      "loss": 0.6328301906585694,
+      "step": 2190
+    },
+    {
+      "epoch": 0.04292431662536827,
+      "grad_norm": 1.2963325273598403,
+      "learning_rate": 2e-05,
+      "loss": 0.6432890892028809,
+      "step": 2200
+    },
+    {
+      "epoch": 0.04311942715548358,
+      "grad_norm": 1.001725956679164,
+      "learning_rate": 2e-05,
+      "loss": 0.6372798919677735,
+      "step": 2210
+    },
+    {
+      "epoch": 0.04331453768559889,
+      "grad_norm": 3.1587572866453257,
+      "learning_rate": 2e-05,
+      "loss": 0.6135322093963623,
+      "step": 2220
+    },
+    {
+      "epoch": 0.0435096482157142,
+      "grad_norm": 1.2804751215716266,
+      "learning_rate": 2e-05,
+      "loss": 0.6305487632751465,
+      "step": 2230
+    },
+    {
+      "epoch": 0.04370475874582951,
+      "grad_norm": 1.1457186328331357,
+      "learning_rate": 2e-05,
+      "loss": 0.6331235885620117,
+      "step": 2240
+    },
+    {
+      "epoch": 0.043899869275944824,
+      "grad_norm": 1.854166740392985,
+      "learning_rate": 2e-05,
+      "loss": 0.6291882514953613,
+      "step": 2250
+    },
+    {
+      "epoch": 0.044094979806060135,
+      "grad_norm": 1.9004721677403358,
+      "learning_rate": 2e-05,
+      "loss": 0.6142114162445068,
+      "step": 2260
+    },
+    {
+      "epoch": 0.044290090336175446,
+      "grad_norm": 1.41540507101651,
+      "learning_rate": 2e-05,
+      "loss": 0.6601502895355225,
+      "step": 2270
+    },
+    {
+      "epoch": 0.04448520086629076,
+      "grad_norm": 3.7027882830176577,
+      "learning_rate": 2e-05,
+      "loss": 0.6777358531951905,
+      "step": 2280
+    },
+    {
+      "epoch": 0.04468031139640606,
+      "grad_norm": 2.2529442152725214,
+      "learning_rate": 2e-05,
+      "loss": 0.6659074306488038,
+      "step": 2290
+    },
+    {
+      "epoch": 0.04487542192652137,
+      "grad_norm": 1.175525517819129,
+      "learning_rate": 2e-05,
+      "loss": 0.6502988338470459,
+      "step": 2300
+    },
+    {
+      "epoch": 0.04507053245663668,
+      "grad_norm": 3.629353206067376,
+      "learning_rate": 2e-05,
+      "loss": 0.6306095123291016,
+      "step": 2310
+    },
+    {
+      "epoch": 0.045265642986751994,
+      "grad_norm": 1.1993125186745976,
+      "learning_rate": 2e-05,
+      "loss": 0.6154571056365967,
+      "step": 2320
+    },
+    {
+      "epoch": 0.045460753516867305,
+      "grad_norm": 1.637233138652609,
+      "learning_rate": 2e-05,
+      "loss": 0.72869873046875,
+      "step": 2330
+    },
+    {
+      "epoch": 0.04565586404698262,
+      "grad_norm": 1.9118168385502368,
+      "learning_rate": 2e-05,
+      "loss": 0.6948616027832031,
+      "step": 2340
+    },
+    {
+      "epoch": 0.04585097457709793,
+      "grad_norm": 34.84879966449921,
+      "learning_rate": 2e-05,
+      "loss": 0.6756387710571289,
+      "step": 2350
+    },
+    {
+      "epoch": 0.04604608510721324,
+      "grad_norm": 2.3342267082519963,
+      "learning_rate": 2e-05,
+      "loss": 0.6326287269592286,
+      "step": 2360
+    },
+    {
+      "epoch": 0.04624119563732855,
+      "grad_norm": 2.694491295690956,
+      "learning_rate": 2e-05,
+      "loss": 0.6674449920654297,
+      "step": 2370
+    },
+    {
+      "epoch": 0.046436306167443854,
+      "grad_norm": 1.2923163665194242,
+      "learning_rate": 2e-05,
+      "loss": 0.6214028835296631,
+      "step": 2380
+    },
+    {
+      "epoch": 0.046631416697559165,
+      "grad_norm": 3.0360798029514533,
+      "learning_rate": 2e-05,
+      "loss": 0.6473525047302247,
+      "step": 2390
+    },
+    {
+      "epoch": 0.046826527227674476,
+      "grad_norm": 1.8742837447680314,
+      "learning_rate": 2e-05,
+      "loss": 0.6427742004394531,
+      "step": 2400
+    },
+    {
+      "epoch": 0.04702163775778979,
+      "grad_norm": 1.217814352410143,
+      "learning_rate": 2e-05,
+      "loss": 0.6365565776824951,
+      "step": 2410
+    },
+    {
+      "epoch": 0.0472167482879051,
+      "grad_norm": 1.1631505432109965,
+      "learning_rate": 2e-05,
+      "loss": 0.6714714050292969,
+      "step": 2420
+    },
+    {
+      "epoch": 0.04741185881802041,
+      "grad_norm": 1.32159178985653,
+      "learning_rate": 2e-05,
+      "loss": 0.6817323207855225,
+      "step": 2430
+    },
+    {
+      "epoch": 0.04760696934813572,
+      "grad_norm": 1.1018026265069345,
+      "learning_rate": 2e-05,
+      "loss": 0.6815985679626465,
+      "step": 2440
+    },
+    {
+      "epoch": 0.04780207987825103,
+      "grad_norm": 22.847932763546503,
+      "learning_rate": 2e-05,
+      "loss": 0.5750728607177734,
+      "step": 2450
+    },
+    {
+      "epoch": 0.04799719040836634,
+      "grad_norm": 1.4795482431427542,
+      "learning_rate": 2e-05,
+      "loss": 0.6426702499389648,
+      "step": 2460
+    },
+    {
+      "epoch": 0.04819230093848165,
+      "grad_norm": 1.3254518586884037,
+      "learning_rate": 2e-05,
+      "loss": 0.676140308380127,
+      "step": 2470
+    },
+    {
+      "epoch": 0.04838741146859696,
+      "grad_norm": 1.0882754172795883,
+      "learning_rate": 2e-05,
+      "loss": 0.6622509002685547,
+      "step": 2480
+    },
+    {
+      "epoch": 0.04858252199871227,
+      "grad_norm": 1.3447696761930397,
+      "learning_rate": 2e-05,
+      "loss": 0.6921796321868896,
+      "step": 2490
+    },
+    {
+      "epoch": 0.04877763252882758,
+      "grad_norm": 1.936302231460338,
+      "learning_rate": 2e-05,
+      "loss": 0.6208174705505372,
+      "step": 2500
+    },
+    {
+      "epoch": 0.04897274305894289,
+      "grad_norm": 1.2550117269317078,
+      "learning_rate": 2e-05,
+      "loss": 0.6440116882324218,
+      "step": 2510
+    },
+    {
+      "epoch": 0.0491678535890582,
+      "grad_norm": 1.9278477196956771,
+      "learning_rate": 2e-05,
+      "loss": 0.6667954921722412,
+      "step": 2520
+    },
+    {
+      "epoch": 0.049362964119173514,
+      "grad_norm": 5.439724358924595,
+      "learning_rate": 2e-05,
+      "loss": 0.6475661754608154,
+      "step": 2530
+    },
+    {
+      "epoch": 0.049558074649288825,
+      "grad_norm": 8.062280693841178,
+      "learning_rate": 2e-05,
+      "loss": 0.6361377239227295,
+      "step": 2540
+    },
+    {
+      "epoch": 0.049753185179404136,
+      "grad_norm": 1.0181825352319607,
+      "learning_rate": 2e-05,
+      "loss": 0.6536860466003418,
+      "step": 2550
+    },
+    {
+      "epoch": 0.04994829570951944,
+      "grad_norm": 1.3197436745932614,
+      "learning_rate": 2e-05,
+      "loss": 0.6890059471130371,
+      "step": 2560
+    },
+    {
+      "epoch": 0.05014340623963475,
+      "grad_norm": 1.0963279480590244,
+      "learning_rate": 2e-05,
+      "loss": 0.6805807113647461,
+      "step": 2570
+    },
+    {
+      "epoch": 0.05033851676975006,
+      "grad_norm": 1.0657884115182137,
+      "learning_rate": 2e-05,
+      "loss": 0.6054826259613038,
+      "step": 2580
+    },
+    {
+      "epoch": 0.05053362729986537,
+      "grad_norm": 1.5912624935136637,
+      "learning_rate": 2e-05,
+      "loss": 0.6484633445739746,
+      "step": 2590
+    },
+    {
+      "epoch": 0.050728737829980684,
+      "grad_norm": 1.4154164611136124,
+      "learning_rate": 2e-05,
+      "loss": 0.5808984756469726,
+      "step": 2600
+    },
+    {
+      "epoch": 0.050923848360095995,
+      "grad_norm": 1.537336896367386,
+      "learning_rate": 2e-05,
+      "loss": 0.6322601318359375,
+      "step": 2610
+    },
+    {
+      "epoch": 0.051118958890211307,
+      "grad_norm": 1.286134699209086,
+      "learning_rate": 2e-05,
+      "loss": 0.6271740913391113,
+      "step": 2620
+    },
+    {
+      "epoch": 0.05131406942032662,
+      "grad_norm": 1.386894457293401,
+      "learning_rate": 2e-05,
+      "loss": 0.6155960083007812,
+      "step": 2630
+    },
+    {
+      "epoch": 0.05150917995044193,
+      "grad_norm": 1.0545351003990986,
+      "learning_rate": 2e-05,
+      "loss": 0.646420955657959,
+      "step": 2640
+    },
+    {
+      "epoch": 0.05170429048055723,
+      "grad_norm": 0.955341679609493,
+      "learning_rate": 2e-05,
+      "loss": 0.6254727840423584,
+      "step": 2650
+    },
+    {
+      "epoch": 0.051899401010672544,
+      "grad_norm": 1.0146807160006226,
+      "learning_rate": 2e-05,
+      "loss": 0.6190547943115234,
+      "step": 2660
+    },
+    {
+      "epoch": 0.052094511540787855,
+      "grad_norm": 2.1720374533376954,
+      "learning_rate": 2e-05,
+      "loss": 0.6083064556121827,
+      "step": 2670
+    },
+    {
+      "epoch": 0.052289622070903166,
+      "grad_norm": 1.641332076277408,
+      "learning_rate": 2e-05,
+      "loss": 0.6259395599365234,
+      "step": 2680
+    },
+    {
+      "epoch": 0.05248473260101848,
+      "grad_norm": 11.332025115437233,
+      "learning_rate": 2e-05,
+      "loss": 0.6529458999633789,
+      "step": 2690
+    },
+    {
+      "epoch": 0.05267984313113379,
+      "grad_norm": 5.334903106648695,
+      "learning_rate": 2e-05,
+      "loss": 0.6095157623291015,
+      "step": 2700
+    },
+    {
+      "epoch": 0.0528749536612491,
+      "grad_norm": 2.9393362751341945,
+      "learning_rate": 2e-05,
+      "loss": 0.682373332977295,
+      "step": 2710
+    },
+    {
+      "epoch": 0.05307006419136441,
+      "grad_norm": 1.7952603597241157,
+      "learning_rate": 2e-05,
+      "loss": 0.632810401916504,
+      "step": 2720
+    },
+    {
+      "epoch": 0.05326517472147972,
+      "grad_norm": 2.7370127591142355,
+      "learning_rate": 2e-05,
+      "loss": 0.6315185070037842,
+      "step": 2730
+    },
+    {
+      "epoch": 0.053460285251595026,
+      "grad_norm": 1.1313026162392894,
+      "learning_rate": 2e-05,
+      "loss": 0.6877236366271973,
+      "step": 2740
+    },
+    {
+      "epoch": 0.05365539578171034,
+      "grad_norm": 0.9788852928625921,
+      "learning_rate": 2e-05,
+      "loss": 0.6380230903625488,
+      "step": 2750
+    },
+    {
+      "epoch": 0.05385050631182565,
+      "grad_norm": 1.2295112399626302,
+      "learning_rate": 2e-05,
+      "loss": 0.6558066368103027,
+      "step": 2760
+    },
+    {
+      "epoch": 0.05404561684194096,
+      "grad_norm": 1.7300697415690638,
+      "learning_rate": 2e-05,
+      "loss": 0.657502555847168,
+      "step": 2770
+    },
+    {
+      "epoch": 0.05424072737205627,
+      "grad_norm": 4.1104395879084885,
+      "learning_rate": 2e-05,
+      "loss": 0.6370492458343506,
+      "step": 2780
+    },
+    {
+      "epoch": 0.05443583790217158,
+      "grad_norm": 4.0220491270995335,
+      "learning_rate": 2e-05,
+      "loss": 0.6445048332214356,
+      "step": 2790
+    },
+    {
+      "epoch": 0.05463094843228689,
+      "grad_norm": 1.5525273209021713,
+      "learning_rate": 2e-05,
+      "loss": 0.6258399963378907,
+      "step": 2800
+    },
+    {
+      "epoch": 0.054826058962402204,
+      "grad_norm": 0.9572531286068277,
+      "learning_rate": 2e-05,
+      "loss": 0.6665648937225341,
+      "step": 2810
+    },
+    {
+      "epoch": 0.055021169492517515,
+      "grad_norm": 4.546053152231271,
+      "learning_rate": 2e-05,
+      "loss": 0.7149829864501953,
+      "step": 2820
+    },
+    {
+      "epoch": 0.05521628002263282,
+      "grad_norm": 5.843608943988325,
+      "learning_rate": 2e-05,
+      "loss": 0.7470980644226074,
+      "step": 2830
+    },
+    {
+      "epoch": 0.05541139055274813,
+      "grad_norm": 4.01133864918651,
+      "learning_rate": 2e-05,
+      "loss": 0.6170453548431396,
+      "step": 2840
+    },
+    {
+      "epoch": 0.05560650108286344,
+      "grad_norm": 1.9500803252145231,
+      "learning_rate": 2e-05,
+      "loss": 0.6097970962524414,
+      "step": 2850
+    },
+    {
+      "epoch": 0.05580161161297875,
+      "grad_norm": 2.486174718243125,
+      "learning_rate": 2e-05,
+      "loss": 0.6476648330688477,
+      "step": 2860
+    },
+    {
+      "epoch": 0.05599672214309406,
+      "grad_norm": 5.605876481434666,
+      "learning_rate": 2e-05,
+      "loss": 0.5887874603271485,
+      "step": 2870
+    },
+    {
+      "epoch": 0.056191832673209374,
+      "grad_norm": 1.0642005234995409,
+      "learning_rate": 2e-05,
+      "loss": 0.6330161094665527,
+      "step": 2880
+    },
+    {
+      "epoch": 0.056386943203324685,
+      "grad_norm": 2.815991081457381,
+      "learning_rate": 2e-05,
+      "loss": 0.644951057434082,
+      "step": 2890
+    },
+    {
+      "epoch": 0.056582053733439996,
+      "grad_norm": 103.34264586482593,
+      "learning_rate": 2e-05,
+      "loss": 0.6404618740081787,
+      "step": 2900
+    },
+    {
+      "epoch": 0.05677716426355531,
+      "grad_norm": 79.43667317410264,
+      "learning_rate": 2e-05,
+      "loss": 0.6286736965179444,
+      "step": 2910
+    },
+    {
+      "epoch": 0.05697227479367061,
+      "grad_norm": 2.4007276032745564,
+      "learning_rate": 2e-05,
+      "loss": 0.6209447383880615,
+      "step": 2920
+    },
+    {
+      "epoch": 0.05716738532378592,
+      "grad_norm": 1.7303282258258368,
+      "learning_rate": 2e-05,
+      "loss": 0.6572585105895996,
+      "step": 2930
+    },
+    {
+      "epoch": 0.057362495853901234,
+      "grad_norm": 3.108786120405242,
+      "learning_rate": 2e-05,
+      "loss": 0.608528470993042,
+      "step": 2940
+    },
+    {
+      "epoch": 0.057557606384016545,
+      "grad_norm": 2.10301470688165,
+      "learning_rate": 2e-05,
+      "loss": 0.6276017665863037,
+      "step": 2950
+    },
+    {
+      "epoch": 0.057752716914131856,
+      "grad_norm": 21.6124226595325,
+      "learning_rate": 2e-05,
+      "loss": 0.6253087520599365,
+      "step": 2960
+    },
+    {
+      "epoch": 0.05794782744424717,
+      "grad_norm": 1.035566957444965,
+      "learning_rate": 2e-05,
+      "loss": 0.6322129726409912,
+      "step": 2970
+    },
+    {
+      "epoch": 0.05814293797436248,
+      "grad_norm": 3.117969109903932,
+      "learning_rate": 2e-05,
+      "loss": 0.6098252773284912,
+      "step": 2980
+    },
+    {
+      "epoch": 0.05833804850447779,
+      "grad_norm": 3.07291560071334,
+      "learning_rate": 2e-05,
+      "loss": 0.6672710418701172,
+      "step": 2990
+    },
+    {
+      "epoch": 0.058533159034593094,
+      "grad_norm": 4.963609006749492,
+      "learning_rate": 2e-05,
+      "loss": 0.6251094818115235,
+      "step": 3000
+    },
+    {
+      "epoch": 0.058728269564708405,
+      "grad_norm": 3.8392304337804326,
+      "learning_rate": 2e-05,
+      "loss": 0.6889659881591796,
+      "step": 3010
+    },
+    {
+      "epoch": 0.058923380094823716,
+      "grad_norm": 1.781608332358966,
+      "learning_rate": 2e-05,
+      "loss": 0.6498517990112305,
+      "step": 3020
+    },
+    {
+      "epoch": 0.05911849062493903,
+      "grad_norm": 1.8656303826960035,
+      "learning_rate": 2e-05,
+      "loss": 0.6888402938842774,
+      "step": 3030
+    },
+    {
+      "epoch": 0.05931360115505434,
+      "grad_norm": 2.532717771736894,
+      "learning_rate": 2e-05,
+      "loss": 0.6240044593811035,
+      "step": 3040
+    },
+    {
+      "epoch": 0.05950871168516965,
+      "grad_norm": 125.9145591351418,
+      "learning_rate": 2e-05,
+      "loss": 0.6656532287597656,
+      "step": 3050
+    },
+    {
+      "epoch": 0.05970382221528496,
+      "grad_norm": 1.943201815002491,
+      "learning_rate": 2e-05,
+      "loss": 0.6358197212219239,
+      "step": 3060
+    },
+    {
+      "epoch": 0.05989893274540027,
+      "grad_norm": 2.1611113411071496,
+      "learning_rate": 2e-05,
+      "loss": 0.6162757873535156,
+      "step": 3070
+    },
+    {
+      "epoch": 0.06009404327551558,
+      "grad_norm": 8.505308423903141,
+      "learning_rate": 2e-05,
+      "loss": 0.653986120223999,
+      "step": 3080
+    },
+    {
+      "epoch": 0.060289153805630886,
+      "grad_norm": 1.6542601536413564,
+      "learning_rate": 2e-05,
+      "loss": 0.6210150718688965,
+      "step": 3090
+    },
+    {
+      "epoch": 0.0604842643357462,
+      "grad_norm": 1.7764769359485486,
+      "learning_rate": 2e-05,
+      "loss": 0.6315331935882569,
+      "step": 3100
+    },
+    {
+      "epoch": 0.06067937486586151,
+      "grad_norm": 4.836501282917753,
+      "learning_rate": 2e-05,
+      "loss": 0.6621027946472168,
+      "step": 3110
+    },
+    {
+      "epoch": 0.06087448539597682,
+      "grad_norm": 5.1501101327822845,
+      "learning_rate": 2e-05,
+      "loss": 0.6463577270507812,
+      "step": 3120
+    },
+    {
+      "epoch": 0.06106959592609213,
+      "grad_norm": 2.3159714005969705,
+      "learning_rate": 2e-05,
+      "loss": 0.6429004192352294,
+      "step": 3130
+    },
+    {
+      "epoch": 0.06126470645620744,
+      "grad_norm": 2.728195642075128,
+      "learning_rate": 2e-05,
+      "loss": 0.6332973003387451,
+      "step": 3140
+    },
+    {
+      "epoch": 0.06145981698632275,
+      "grad_norm": 1.2279562609381547,
+      "learning_rate": 2e-05,
+      "loss": 0.6325988292694091,
+      "step": 3150
+    },
+    {
+      "epoch": 0.061654927516438064,
+      "grad_norm": 1.9493092556172391,
+      "learning_rate": 2e-05,
+      "loss": 0.6550612926483155,
+      "step": 3160
+    },
+    {
+      "epoch": 0.061850038046553375,
+      "grad_norm": 2.019023304591436,
+      "learning_rate": 2e-05,
+      "loss": 0.6142223358154297,
+      "step": 3170
+    },
+    {
+      "epoch": 0.06204514857666868,
+      "grad_norm": 1.74420411133503,
+      "learning_rate": 2e-05,
+      "loss": 0.6268470764160157,
+      "step": 3180
+    },
+    {
+      "epoch": 0.06224025910678399,
+      "grad_norm": 1.6619355604873653,
+      "learning_rate": 2e-05,
+      "loss": 0.6191567420959473,
+      "step": 3190
+    },
+    {
+      "epoch": 0.0624353696368993,
+      "grad_norm": 4.140970560730074,
+      "learning_rate": 2e-05,
+      "loss": 0.6285354137420655,
+      "step": 3200
+    },
+    {
+      "epoch": 0.06263048016701461,
+      "grad_norm": 5.159480027508822,
+      "learning_rate": 2e-05,
+      "loss": 0.6047804832458497,
+      "step": 3210
+    },
+    {
+      "epoch": 0.06282559069712992,
+      "grad_norm": 1.8557966038129536,
+      "learning_rate": 2e-05,
+      "loss": 0.6683725357055664,
+      "step": 3220
+    },
+    {
+      "epoch": 0.06302070122724523,
+      "grad_norm": 1.9736528648225522,
+      "learning_rate": 2e-05,
+      "loss": 0.6185779571533203,
+      "step": 3230
+    },
+    {
+      "epoch": 0.06321581175736055,
+      "grad_norm": 18.34026324855907,
+      "learning_rate": 2e-05,
+      "loss": 0.6165416717529297,
+      "step": 3240
+    },
+    {
+      "epoch": 0.06341092228747586,
+      "grad_norm": 15.522868255555723,
+      "learning_rate": 2e-05,
+      "loss": 0.6219786643981934,
+      "step": 3250
+    },
+    {
+      "epoch": 0.06360603281759117,
+      "grad_norm": 1.202487696485921,
+      "learning_rate": 2e-05,
+      "loss": 0.6668967247009278,
+      "step": 3260
+    },
+    {
+      "epoch": 0.06380114334770648,
+      "grad_norm": 1.0722125159360762,
+      "learning_rate": 2e-05,
+      "loss": 0.622715950012207,
+      "step": 3270
+    },
+    {
+      "epoch": 0.06399625387782179,
+      "grad_norm": 0.9520172401507281,
+      "learning_rate": 2e-05,
+      "loss": 0.6406888484954834,
+      "step": 3280
+    },
+    {
+      "epoch": 0.0641913644079371,
+      "grad_norm": 1.5574105897988617,
+      "learning_rate": 2e-05,
+      "loss": 0.6616538047790528,
+      "step": 3290
+    },
+    {
+      "epoch": 0.06438647493805241,
+      "grad_norm": 0.9534801349175593,
+      "learning_rate": 2e-05,
+      "loss": 0.5777172088623047,
+      "step": 3300
+    },
+    {
+      "epoch": 0.06458158546816772,
+      "grad_norm": 2.490289099615107,
+      "learning_rate": 2e-05,
+      "loss": 0.651535940170288,
+      "step": 3310
+    },
+    {
+      "epoch": 0.06477669599828302,
+      "grad_norm": 1.1719574017262055,
+      "learning_rate": 2e-05,
+      "loss": 0.6687840938568115,
+      "step": 3320
+    },
+    {
+      "epoch": 0.06497180652839833,
+      "grad_norm": 1.7894175678356339,
+      "learning_rate": 2e-05,
+      "loss": 0.6507015228271484,
+      "step": 3330
+    },
+    {
+      "epoch": 0.06516691705851364,
+      "grad_norm": 0.9617747704284962,
+      "learning_rate": 2e-05,
+      "loss": 0.6277545928955078,
+      "step": 3340
+    },
+    {
+      "epoch": 0.06536202758862895,
+      "grad_norm": 1.024868673927139,
+      "learning_rate": 2e-05,
+      "loss": 0.6204674720764161,
+      "step": 3350
+    },
+    {
+      "epoch": 0.06555713811874427,
+      "grad_norm": 1.16100322238524,
+      "learning_rate": 2e-05,
+      "loss": 0.6270739555358886,
+      "step": 3360
+    },
+    {
+      "epoch": 0.06575224864885958,
+      "grad_norm": 1.05899895998444,
+      "learning_rate": 2e-05,
+      "loss": 0.6357701301574707,
+      "step": 3370
+    },
+    {
+      "epoch": 0.06594735917897489,
+      "grad_norm": 2.0518410013921167,
+      "learning_rate": 2e-05,
+      "loss": 0.6172528743743897,
+      "step": 3380
+    },
+    {
+      "epoch": 0.0661424697090902,
+      "grad_norm": 1.549139093474472,
+      "learning_rate": 2e-05,
+      "loss": 0.6199777126312256,
+      "step": 3390
+    },
+    {
+      "epoch": 0.06633758023920551,
+      "grad_norm": 1.5599766278608658,
+      "learning_rate": 2e-05,
+      "loss": 0.619226598739624,
+      "step": 3400
+    },
+    {
+      "epoch": 0.06653269076932082,
+      "grad_norm": 0.9607097459040843,
+      "learning_rate": 2e-05,
+      "loss": 0.6402338981628418,
+      "step": 3410
+    },
+    {
+      "epoch": 0.06672780129943613,
+      "grad_norm": 0.9380194617053622,
+      "learning_rate": 2e-05,
+      "loss": 0.6315077781677246,
+      "step": 3420
+    },
+    {
+      "epoch": 0.06692291182955144,
+      "grad_norm": 1.2360701977959478,
+      "learning_rate": 2e-05,
+      "loss": 0.5841626167297364,
+      "step": 3430
+    },
+    {
+      "epoch": 0.06711802235966675,
+      "grad_norm": 1.2198333844022478,
+      "learning_rate": 2e-05,
+      "loss": 0.6027783870697021,
+      "step": 3440
+    },
+    {
+      "epoch": 0.06731313288978207,
+      "grad_norm": 1.1844850338976578,
+      "learning_rate": 2e-05,
+      "loss": 0.5964821815490723,
+      "step": 3450
+    },
+    {
+      "epoch": 0.06750824341989738,
+      "grad_norm": 0.9930932636074196,
+      "learning_rate": 2e-05,
+      "loss": 0.6134414672851562,
+      "step": 3460
+    },
+    {
+      "epoch": 0.06770335395001269,
+      "grad_norm": 1.138524640736631,
+      "learning_rate": 2e-05,
+      "loss": 0.6110162734985352,
+      "step": 3470
+    },
+    {
+      "epoch": 0.067898464480128,
+      "grad_norm": 1.0047825982374765,
+      "learning_rate": 2e-05,
+      "loss": 0.637286376953125,
+      "step": 3480
+    },
+    {
+      "epoch": 0.06809357501024331,
+      "grad_norm": 1.260216801811752,
+      "learning_rate": 2e-05,
+      "loss": 0.6786258697509766,
+      "step": 3490
+    },
+    {
+      "epoch": 0.0682886855403586,
+      "grad_norm": 2.1246781599054554,
+      "learning_rate": 2e-05,
+      "loss": 0.663703441619873,
+      "step": 3500
+    },
+    {
+      "epoch": 0.06848379607047392,
+      "grad_norm": 1.948740029087723,
+      "learning_rate": 2e-05,
+      "loss": 0.6510276317596435,
+      "step": 3510
+    },
+    {
+      "epoch": 0.06867890660058923,
+      "grad_norm": 1.4026841352229737,
+      "learning_rate": 2e-05,
+      "loss": 0.6513165473937989,
+      "step": 3520
+    },
+    {
+      "epoch": 0.06887401713070454,
+      "grad_norm": 5.813095243655102,
+      "learning_rate": 2e-05,
+      "loss": 0.5855354309082031,
+      "step": 3530
+    },
+    {
+      "epoch": 0.06906912766081985,
+      "grad_norm": 2.3155682194928118,
+      "learning_rate": 2e-05,
+      "loss": 0.6335483074188233,
+      "step": 3540
+    },
+    {
+      "epoch": 0.06926423819093516,
+      "grad_norm": 1.5694715217411332,
+      "learning_rate": 2e-05,
+      "loss": 0.6248700618743896,
+      "step": 3550
+    },
+    {
+      "epoch": 0.06945934872105047,
+      "grad_norm": 1.161316501242528,
+      "learning_rate": 2e-05,
+      "loss": 0.604718017578125,
+      "step": 3560
+    },
+    {
+      "epoch": 0.06965445925116578,
+      "grad_norm": 2.044224356570064,
+      "learning_rate": 2e-05,
+      "loss": 0.6216660499572754,
+      "step": 3570
+    },
+    {
+      "epoch": 0.0698495697812811,
+      "grad_norm": 1.6679796752893745,
+      "learning_rate": 2e-05,
+      "loss": 0.6200233459472656,
+      "step": 3580
+    },
+    {
+      "epoch": 0.0700446803113964,
+      "grad_norm": 1.7666150549131676,
+      "learning_rate": 2e-05,
+      "loss": 0.6485730648040772,
+      "step": 3590
+    },
+    {
+      "epoch": 0.07023979084151172,
+      "grad_norm": 2.326009432234743,
+      "learning_rate": 2e-05,
+      "loss": 0.618149709701538,
+      "step": 3600
+    },
+    {
+      "epoch": 0.07043490137162703,
+      "grad_norm": 4.154343961373888,
+      "learning_rate": 2e-05,
+      "loss": 0.6535586357116699,
+      "step": 3610
+    },
+    {
+      "epoch": 0.07063001190174234,
+      "grad_norm": 1.3209472948134615,
+      "learning_rate": 2e-05,
+      "loss": 0.6286499977111817,
+      "step": 3620
+    },
+    {
+      "epoch": 0.07082512243185765,
+      "grad_norm": 1.109490149817695,
+      "learning_rate": 2e-05,
+      "loss": 0.6330195903778076,
+      "step": 3630
+    },
+    {
+      "epoch": 0.07102023296197296,
+      "grad_norm": 1.310086522110561,
+      "learning_rate": 2e-05,
+      "loss": 0.6741226196289063,
+      "step": 3640
+    },
+    {
+      "epoch": 0.07121534349208827,
+      "grad_norm": 1.5645094510625,
+      "learning_rate": 2e-05,
+      "loss": 0.6502818107604981,
+      "step": 3650
+    },
+    {
+      "epoch": 0.07141045402220358,
+      "grad_norm": 1.0117104696448032,
+      "learning_rate": 2e-05,
+      "loss": 0.6151797294616699,
+      "step": 3660
+    },
+    {
+      "epoch": 0.0716055645523189,
+      "grad_norm": 1.8008334734207228,
+      "learning_rate": 2e-05,
+      "loss": 0.6476142406463623,
+      "step": 3670
+    },
+    {
+      "epoch": 0.07180067508243419,
+      "grad_norm": 1.327637666045341,
+      "learning_rate": 2e-05,
+      "loss": 0.6369030952453614,
+      "step": 3680
+    },
+    {
+      "epoch": 0.0719957856125495,
+      "grad_norm": 1.1491539820218042,
+      "learning_rate": 2e-05,
+      "loss": 0.6163894653320312,
+      "step": 3690
+    },
+    {
+      "epoch": 0.07219089614266481,
+      "grad_norm": 3.1965338595766424,
+      "learning_rate": 2e-05,
+      "loss": 0.5750933647155761,
+      "step": 3700
+    },
+    {
+      "epoch": 0.07238600667278013,
+      "grad_norm": 1.3632057599003788,
+      "learning_rate": 2e-05,
+      "loss": 0.6531260490417481,
+      "step": 3710
+    },
+    {
+      "epoch": 0.07258111720289544,
+      "grad_norm": 1.2512607073301945,
+      "learning_rate": 2e-05,
+      "loss": 0.6505274772644043,
+      "step": 3720
+    },
+    {
+      "epoch": 0.07277622773301075,
+      "grad_norm": 1.936600042650387,
+      "learning_rate": 2e-05,
+      "loss": 0.6244311332702637,
+      "step": 3730
+    },
+    {
+      "epoch": 0.07297133826312606,
+      "grad_norm": 1.737115862444927,
+      "learning_rate": 2e-05,
+      "loss": 0.6188953399658204,
+      "step": 3740
+    },
+    {
+      "epoch": 0.07316644879324137,
+      "grad_norm": 2.62098562991079,
+      "learning_rate": 2e-05,
+      "loss": 0.625301742553711,
+      "step": 3750
+    },
+    {
+      "epoch": 0.07336155932335668,
+      "grad_norm": 0.9956574198637514,
+      "learning_rate": 2e-05,
+      "loss": 0.6125325679779052,
+      "step": 3760
+    },
+    {
+      "epoch": 0.07355666985347199,
+      "grad_norm": 1.202861962831585,
+      "learning_rate": 2e-05,
+      "loss": 0.6161567687988281,
+      "step": 3770
+    },
+    {
+      "epoch": 0.0737517803835873,
+      "grad_norm": 2.3867228064920165,
+      "learning_rate": 2e-05,
+      "loss": 0.61738600730896,
+      "step": 3780
+    },
+    {
+      "epoch": 0.07394689091370261,
+      "grad_norm": 9.870304841199335,
+      "learning_rate": 2e-05,
+      "loss": 0.5734426498413085,
+      "step": 3790
+    },
+    {
+      "epoch": 0.07414200144381793,
+      "grad_norm": 1.2209044685058315,
+      "learning_rate": 2e-05,
+      "loss": 0.6289399623870849,
+      "step": 3800
+    },
+    {
+      "epoch": 0.07433711197393324,
+      "grad_norm": 1.1980012076194961,
+      "learning_rate": 2e-05,
+      "loss": 0.6223122596740722,
+      "step": 3810
+    },
+    {
+      "epoch": 0.07453222250404855,
+      "grad_norm": 2.5588839671427155,
+      "learning_rate": 2e-05,
+      "loss": 0.5936367988586426,
+      "step": 3820
+    },
+    {
+      "epoch": 0.07472733303416386,
+      "grad_norm": 1.417936010939305,
+      "learning_rate": 2e-05,
+      "loss": 0.5945808410644531,
+      "step": 3830
+    },
+    {
+      "epoch": 0.07492244356427917,
+      "grad_norm": 1.0267517787839726,
+      "learning_rate": 2e-05,
+      "loss": 0.609653377532959,
+      "step": 3840
+    },
+    {
+      "epoch": 0.07511755409439448,
+      "grad_norm": 2.032089595872863,
+      "learning_rate": 2e-05,
+      "loss": 0.5973081111907959,
+      "step": 3850
+    },
+    {
+      "epoch": 0.07531266462450978,
+      "grad_norm": 1.6419694096855528,
+      "learning_rate": 2e-05,
+      "loss": 0.5852180480957031,
+      "step": 3860
+    },
+    {
+      "epoch": 0.07550777515462509,
+      "grad_norm": 1.3533316887780575,
+      "learning_rate": 2e-05,
+      "loss": 0.5879265785217285,
+      "step": 3870
+    },
+    {
+      "epoch": 0.0757028856847404,
+      "grad_norm": 1.6493989950912267,
+      "learning_rate": 2e-05,
+      "loss": 0.6273761749267578,
+      "step": 3880
+    },
+    {
+      "epoch": 0.07589799621485571,
+      "grad_norm": 1.1923090187885024,
+      "learning_rate": 2e-05,
+      "loss": 0.5740846157073974,
+      "step": 3890
+    },
+    {
+      "epoch": 0.07609310674497102,
+      "grad_norm": 1.125262015612364,
+      "learning_rate": 2e-05,
+      "loss": 0.6070968627929687,
+      "step": 3900
+    },
+    {
+      "epoch": 0.07628821727508633,
+      "grad_norm": 1.8822115756900608,
+      "learning_rate": 2e-05,
+      "loss": 0.6380067348480225,
+      "step": 3910
+    },
+    {
+      "epoch": 0.07648332780520165,
+      "grad_norm": 2.6492104223189603,
+      "learning_rate": 2e-05,
+      "loss": 0.5887240409851074,
+      "step": 3920
+    },
+    {
+      "epoch": 0.07667843833531696,
+      "grad_norm": 1.0802032535218116,
+      "learning_rate": 2e-05,
+      "loss": 0.609073543548584,
+      "step": 3930
+    },
+    {
+      "epoch": 0.07687354886543227,
+      "grad_norm": 1.2766327415429273,
+      "learning_rate": 2e-05,
+      "loss": 0.60869140625,
+      "step": 3940
+    },
+    {
+      "epoch": 0.07706865939554758,
+      "grad_norm": 0.9799005241543788,
+      "learning_rate": 2e-05,
+      "loss": 0.646174144744873,
+      "step": 3950
+    },
+    {
+      "epoch": 0.07726376992566289,
+      "grad_norm": 13.964947510566653,
+      "learning_rate": 2e-05,
+      "loss": 0.6144445419311524,
+      "step": 3960
+    },
+    {
+      "epoch": 0.0774588804557782,
+      "grad_norm": 1.0004620492038792,
+      "learning_rate": 2e-05,
+      "loss": 0.626788854598999,
+      "step": 3970
+    },
+    {
+      "epoch": 0.07765399098589351,
+      "grad_norm": 1.1942672679545576,
+      "learning_rate": 2e-05,
+      "loss": 0.6477014064788819,
+      "step": 3980
+    },
+    {
+      "epoch": 0.07784910151600882,
+      "grad_norm": 1.5540049073495548,
+      "learning_rate": 2e-05,
+      "loss": 0.5892033576965332,
+      "step": 3990
+    },
+    {
+      "epoch": 0.07804421204612413,
+      "grad_norm": 1.7721602030135375,
+      "learning_rate": 2e-05,
+      "loss": 0.5847073554992676,
+      "step": 4000
+    },
+    {
+      "epoch": 0.07823932257623945,
+      "grad_norm": 1.0107626309700466,
+      "learning_rate": 2e-05,
+      "loss": 0.6158422470092774,
+      "step": 4010
+    },
+    {
+      "epoch": 0.07843443310635476,
+      "grad_norm": 3.3184566149453802,
+      "learning_rate": 2e-05,
+      "loss": 0.5711278438568115,
+      "step": 4020
+    },
+    {
+      "epoch": 0.07862954363647005,
+      "grad_norm": 1.337890934858505,
+      "learning_rate": 2e-05,
+      "loss": 0.5999331951141358,
+      "step": 4030
+    },
+    {
+      "epoch": 0.07882465416658536,
+      "grad_norm": 1.1433277357358822,
+      "learning_rate": 2e-05,
+      "loss": 0.6507415771484375,
+      "step": 4040
+    },
+    {
+      "epoch": 0.07901976469670068,
+      "grad_norm": 0.9185616693431335,
+      "learning_rate": 2e-05,
+      "loss": 0.567193603515625,
+      "step": 4050
+    },
+    {
+      "epoch": 0.07921487522681599,
+      "grad_norm": 1.999621417135845,
+      "learning_rate": 2e-05,
+      "loss": 0.5888802528381347,
+      "step": 4060
+    },
+    {
+      "epoch": 0.0794099857569313,
+      "grad_norm": 0.8840826484115074,
+      "learning_rate": 2e-05,
+      "loss": 0.5788058280944824,
+      "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 sha256:70e4e3435ce903e96f0f05ad12e0be3c1bc36afff496aa175c867be51fe32316
+size 7313