graph_nemi_sft_e2: SFT from scene_nemi_rl_e2_final on graph-stitched near-miss demos of the RL e2 rollouts; eval pass@8 11.1%

.gitattributes

@@ -58,3 +58,4 @@ scene_nemi_sft_e1/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 scene_nemi_rl_e2_final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 scene_nemi_rl_e2_best/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 nemi_sft_e2/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+graph_nemi_sft_e2/tokenizer.json filter=lfs diff=lfs merge=lfs -text

graph_nemi_sft_e2/all_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 3.6690647482014387,
+    "total_flos": 55035980414976.0,
+    "train_loss": 0.16481673653219261,
+    "train_runtime": 1297.8966,
+    "train_samples_per_second": 6.287,
+    "train_steps_per_second": 0.393
+}

graph_nemi_sft_e2/chat_template.jinja

@@ -0,0 +1,120 @@
+{%- if tools %}
+    {{- '<|im_start|>system\n' }}
+    {%- if messages[0].role == 'system' %}
+        {%- if messages[0].content is string %}
+            {{- messages[0].content }}
+        {%- else %}
+            {%- for content in messages[0].content %}
+                {%- if 'text' in content %}
+                    {{- content.text }}
+                {%- endif %}
+            {%- endfor %}
+        {%- endif %}
+        {{- '\n\n' }}
+    {%- endif %}
+    {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {{- '<|im_start|>system\n' }}
+        {%- if messages[0].content is string %}
+            {{- messages[0].content }}
+        {%- else %}
+            {%- for content in messages[0].content %}
+                {%- if 'text' in content %}
+                    {{- content.text }}
+                {%- endif %}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- for message in messages %}
+    {%- if message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' }}
+        {%- if message.content is string %}
+            {{- message.content }}
+        {%- else %}
+            {%- for content in message.content %}
+                {%- if content.type == 'image' or 'image' in content or 'image_url' in content %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                    {%- if add_vision_id %}Picture {{ image_count.value }}: {% endif -%}
+                    <|vision_start|><|image_pad|><|vision_end|>
+                {%- elif content.type == 'video' or 'video' in content %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                    {%- if add_vision_id %}Video {{ video_count.value }}: {% endif -%}
+                    <|vision_start|><|video_pad|><|vision_end|>
+                {%- elif 'text' in content %}
+                    {{- content.text }}
+                {%- endif %}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "assistant" %}
+        {{- '<|im_start|>' + message.role + '\n' }}
+        {%- if message.content is string %}
+            {{- message.content }}
+        {%- else %}
+            {%- for content_item in message.content %}
+                {%- if 'text' in content_item %}
+                    {{- content_item.text }}
+                {%- endif %}
+            {%- endfor %}
+        {%- endif %}
+        {%- if message.tool_calls %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if (loop.first and message.content) or (not loop.first) %}
+                    {{- '\n' }}
+                {%- endif %}
+                {%- if tool_call.function %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {{- '<tool_call>\n{"name": "' }}
+                {{- tool_call.name }}
+                {{- '", "arguments": ' }}
+                {%- if tool_call.arguments is string %}
+                    {{- tool_call.arguments }}
+                {%- else %}
+                    {{- tool_call.arguments | tojson }}
+                {%- endif %}
+                {{- '}\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {%- if message.content is string %}
+            {{- message.content }}
+        {%- else %}
+            {%- for content in message.content %}
+                {%- if content.type == 'image' or 'image' in content or 'image_url' in content %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                    {%- if add_vision_id %}Picture {{ image_count.value }}: {% endif -%}
+                    <|vision_start|><|image_pad|><|vision_end|>
+                {%- elif content.type == 'video' or 'video' in content %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                    {%- if add_vision_id %}Video {{ video_count.value }}: {% endif -%}
+                    <|vision_start|><|video_pad|><|vision_end|>
+                {%- elif 'text' in content %}
+                    {{- content.text }}
+                {%- endif %}
+            {%- endfor %}
+        {%- endif %}
+        {{- '\n</tool_response>' }}
+        {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+{%- endif %}

graph_nemi_sft_e2/config.json

@@ -0,0 +1,70 @@
+{
+  "architectures": [
+    "Qwen3VLForConditionalGeneration"
+  ],
+  "bos_token_id": null,
+  "dtype": "bfloat16",
+  "eos_token_id": 151645,
+  "hidden_size": 4096,
+  "image_token_id": 151655,
+  "model_type": "qwen3_vl",
+  "pad_token_id": 151643,
+  "text_config": {
+    "attention_bias": false,
+    "attention_dropout": 0.0,
+    "bos_token_id": 151643,
+    "dtype": "bfloat16",
+    "eos_token_id": 151645,
+    "head_dim": 128,
+    "hidden_act": "silu",
+    "hidden_size": 4096,
+    "initializer_range": 0.02,
+    "intermediate_size": 12288,
+    "max_position_embeddings": 262144,
+    "model_type": "qwen3_vl_text",
+    "num_attention_heads": 32,
+    "num_hidden_layers": 36,
+    "num_key_value_heads": 8,
+    "pad_token_id": null,
+    "rms_norm_eps": 1e-06,
+    "rope_parameters": {
+      "mrope_interleaved": true,
+      "mrope_section": [
+        24,
+        20,
+        20
+      ],
+      "rope_theta": 5000000,
+      "rope_type": "default"
+    },
+    "use_cache": false,
+    "vocab_size": 151936
+  },
+  "tie_word_embeddings": false,
+  "transformers_version": "5.8.0",
+  "use_cache": false,
+  "video_token_id": 151656,
+  "vision_config": {
+    "deepstack_visual_indexes": [
+      8,
+      16,
+      24
+    ],
+    "depth": 27,
+    "dtype": "bfloat16",
+    "hidden_act": "gelu_pytorch_tanh",
+    "hidden_size": 1152,
+    "in_channels": 3,
+    "initializer_range": 0.02,
+    "intermediate_size": 4304,
+    "model_type": "qwen3_vl_vision",
+    "num_heads": 16,
+    "num_position_embeddings": 2304,
+    "out_hidden_size": 4096,
+    "patch_size": 16,
+    "spatial_merge_size": 2,
+    "temporal_patch_size": 2
+  },
+  "vision_end_token_id": 151653,
+  "vision_start_token_id": 151652
+}

graph_nemi_sft_e2/generation_config.json

@@ -0,0 +1,14 @@
+{
+  "do_sample": true,
+  "eos_token_id": [
+    151645,
+    151645,
+    151643
+  ],
+  "pad_token_id": 151643,
+  "repetition_penalty": 1.0,
+  "temperature": 0.7,
+  "top_k": 20,
+  "top_p": 0.8,
+  "transformers_version": "5.8.0"
+}

graph_nemi_sft_e2/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8ee3dcda6dbc36b5f1d1d6dcc65d6031eb3bd7cd216233576774aacdb082cbf6
+size 17534340584

graph_nemi_sft_e2/processor_config.json

@@ -0,0 +1,60 @@
+{
+  "image_processor": {
+    "do_convert_rgb": true,
+    "do_normalize": true,
+    "do_rescale": true,
+    "do_resize": true,
+    "image_mean": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "image_processor_type": "Qwen2VLImageProcessor",
+    "image_std": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "merge_size": 2,
+    "patch_size": 16,
+    "resample": 3,
+    "rescale_factor": 0.00392156862745098,
+    "size": {
+      "longest_edge": 16777216,
+      "shortest_edge": 65536
+    },
+    "temporal_patch_size": 2
+  },
+  "processor_class": "Qwen3VLProcessor",
+  "video_processor": {
+    "do_convert_rgb": true,
+    "do_normalize": true,
+    "do_rescale": true,
+    "do_resize": true,
+    "do_sample_frames": true,
+    "fps": 2,
+    "image_mean": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "image_std": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "max_frames": 768,
+    "merge_size": 2,
+    "min_frames": 4,
+    "patch_size": 16,
+    "resample": 3,
+    "rescale_factor": 0.00392156862745098,
+    "return_metadata": false,
+    "size": {
+      "longest_edge": 25165824,
+      "shortest_edge": 4096
+    },
+    "temporal_patch_size": 2,
+    "video_processor_type": "Qwen3VLVideoProcessor"
+  }
+}

graph_nemi_sft_e2/sft.yaml

@@ -0,0 +1,54 @@
+# LLaMA-Factory full-parameter SFT of Qwen3-VL-8B-Instruct on one HER-final dataset.
+# Four L40S (46 GB), one run at a time: DeepSpeed ZeRO-3 with optimizer offload to host
+# RAM (~130 GB per run, so two runs do not fit in 372 GB); the vision tower and projector
+# stay frozen (LLaMA-Factory's default for VLMs), the LLM trains.
+# Filled per run by launch_sft.sh: graph_nemi_sft_e2, /root/sft_data/graph_nemi_sft_e2_k1024, /root/runs/graph_nemi_sft_e2.
+### model
+model_name_or_path: /root/viewagent_private/exps/rl/scene_near_miss_epoch_2/verl_checkpoints/global_step_71/actor/huggingface
+template: qwen3_vl_nothink
+trust_remote_code: true
+image_max_pixels: 65536          # frames are 256x256 already; never upscale
+image_min_pixels: 4096
+
+### method
+stage: sft
+do_train: true
+finetuning_type: full
+freeze_vision_tower: true
+freeze_multi_modal_projector: true
+freeze_language_model: false
+deepspeed: /root/LlamaFactory/examples/deepspeed/ds_z3_config.json
+
+### dataset
+dataset: graph_nemi_sft_e2
+dataset_dir: /root/sft_data/graph_nemi_sft_e2_k1024
+media_dir: /root/viewagent_private
+cutoff_len: 8192
+overwrite_cache: true
+preprocessing_num_workers: 16
+dataloader_num_workers: 4
+
+### output
+output_dir: /root/runs/graph_nemi_sft_e2
+logging_steps: 10
+save_strategy: steps
+save_steps: 170
+save_total_limit: 1              # only the latest checkpoint stays (resumable); the final model is saved separately
+save_only_model: true
+plot_loss: true
+overwrite_output_dir: true
+report_to: wandb
+
+### train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 2
+max_steps: 510                     # equal update budget for both runs (~1 pass over 8,173 rows)
+learning_rate: 1.0e-5
+lr_scheduler_type: cosine
+warmup_ratio: 0.03
+weight_decay: 0.01
+bf16: true
+gradient_checkpointing: true
+ddp_timeout: 180000000
+seed: 0
+run_name: graph_nemi_sft_e2

graph_nemi_sft_e2/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:be75606093db2094d7cd20f3c2f385c212750648bd6ea4fb2bf507a6a4c55506
+size 11422650

graph_nemi_sft_e2/tokenizer_config.json

@@ -0,0 +1,32 @@
+{
+  "add_prefix_space": false,
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "extra_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "is_local": true,
+  "local_files_only": false,
+  "model_max_length": 262144,
+  "pad_token": "<|endoftext|>",
+  "padding_side": "right",
+  "processor_class": "Qwen3VLProcessor",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}

graph_nemi_sft_e2/train_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 3.6690647482014387,
+    "total_flos": 55035980414976.0,
+    "train_loss": 0.16481673653219261,
+    "train_runtime": 1297.8966,
+    "train_samples_per_second": 6.287,
+    "train_steps_per_second": 0.393
+}

graph_nemi_sft_e2/trainer_state.json

@@ -0,0 +1,400 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.6690647482014387,
+  "eval_steps": 500,
+  "global_step": 510,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.07194244604316546,
+      "grad_norm": 6.992695652389583,
+      "learning_rate": 5.625e-06,
+      "loss": 0.3966069221496582,
+      "step": 10
+    },
+    {
+      "epoch": 0.14388489208633093,
+      "grad_norm": 2.7827766218402834,
+      "learning_rate": 9.999090054896397e-06,
+      "loss": 0.3248250484466553,
+      "step": 20
+    },
+    {
+      "epoch": 0.2158273381294964,
+      "grad_norm": 2.638756241168688,
+      "learning_rate": 9.98292246503335e-06,
+      "loss": 0.3306441307067871,
+      "step": 30
+    },
+    {
+      "epoch": 0.28776978417266186,
+      "grad_norm": 2.973619950403627,
+      "learning_rate": 9.946609117610508e-06,
+      "loss": 0.3181817054748535,
+      "step": 40
+    },
+    {
+      "epoch": 0.3597122302158273,
+      "grad_norm": 2.538882332430301,
+      "learning_rate": 9.89029682604881e-06,
+      "loss": 0.33042106628417967,
+      "step": 50
+    },
+    {
+      "epoch": 0.4316546762589928,
+      "grad_norm": 2.5886006446840195,
+      "learning_rate": 9.814213258705813e-06,
+      "loss": 0.3458705902099609,
+      "step": 60
+    },
+    {
+      "epoch": 0.5035971223021583,
+      "grad_norm": 2.5327460685173553,
+      "learning_rate": 9.71866601842156e-06,
+      "loss": 0.33172221183776857,
+      "step": 70
+    },
+    {
+      "epoch": 0.5755395683453237,
+      "grad_norm": 2.406201521364234,
+      "learning_rate": 9.604041398892528e-06,
+      "loss": 0.3516120433807373,
+      "step": 80
+    },
+    {
+      "epoch": 0.6474820143884892,
+      "grad_norm": 2.2570367751597096,
+      "learning_rate": 9.470802822901558e-06,
+      "loss": 0.3278696537017822,
+      "step": 90
+    },
+    {
+      "epoch": 0.7194244604316546,
+      "grad_norm": 2.310411101825302,
+      "learning_rate": 9.31948896871795e-06,
+      "loss": 0.3334329605102539,
+      "step": 100
+    },
+    {
+      "epoch": 0.7913669064748201,
+      "grad_norm": 2.365311232554325,
+      "learning_rate": 9.150711592242627e-06,
+      "loss": 0.323340916633606,
+      "step": 110
+    },
+    {
+      "epoch": 0.8633093525179856,
+      "grad_norm": 2.2549879554890353,
+      "learning_rate": 8.965153053703325e-06,
+      "loss": 0.32022433280944823,
+      "step": 120
+    },
+    {
+      "epoch": 0.935251798561151,
+      "grad_norm": 2.23081737371207,
+      "learning_rate": 8.763563558899317e-06,
+      "loss": 0.32753605842590333,
+      "step": 130
+    },
+    {
+      "epoch": 1.0071942446043165,
+      "grad_norm": 1.5950412892390498,
+      "learning_rate": 8.546758126149148e-06,
+      "loss": 0.32393584251403806,
+      "step": 140
+    },
+    {
+      "epoch": 1.079136690647482,
+      "grad_norm": 1.579289460031863,
+      "learning_rate": 8.315613291203977e-06,
+      "loss": 0.17007782459259033,
+      "step": 150
+    },
+    {
+      "epoch": 1.1510791366906474,
+      "grad_norm": 2.430742321039002,
+      "learning_rate": 8.071063563448341e-06,
+      "loss": 0.16097400188446045,
+      "step": 160
+    },
+    {
+      "epoch": 1.223021582733813,
+      "grad_norm": 2.2881024642083445,
+      "learning_rate": 7.814097647715848e-06,
+      "loss": 0.18136271238327026,
+      "step": 170
+    },
+    {
+      "epoch": 1.2949640287769784,
+      "grad_norm": 2.1454132429025337,
+      "learning_rate": 7.5457544469948164e-06,
+      "loss": 0.16978039741516113,
+      "step": 180
+    },
+    {
+      "epoch": 1.3669064748201438,
+      "grad_norm": 2.156445681072055,
+      "learning_rate": 7.267118862184767e-06,
+      "loss": 0.17723881006240844,
+      "step": 190
+    },
+    {
+      "epoch": 1.4388489208633093,
+      "grad_norm": 1.662715192203829,
+      "learning_rate": 6.9793174058851805e-06,
+      "loss": 0.16350069046020507,
+      "step": 200
+    },
+    {
+      "epoch": 1.5107913669064748,
+      "grad_norm": 1.7614960653690992,
+      "learning_rate": 6.683513647949826e-06,
+      "loss": 0.16108211278915405,
+      "step": 210
+    },
+    {
+      "epoch": 1.5827338129496402,
+      "grad_norm": 2.4655428911283153,
+      "learning_rate": 6.38090351122006e-06,
+      "loss": 0.1663761854171753,
+      "step": 220
+    },
+    {
+      "epoch": 1.6546762589928057,
+      "grad_norm": 2.2352199198827987,
+      "learning_rate": 6.072710436456293e-06,
+      "loss": 0.1689173936843872,
+      "step": 230
+    },
+    {
+      "epoch": 1.7266187050359711,
+      "grad_norm": 2.138484166092592,
+      "learning_rate": 5.760180436015604e-06,
+      "loss": 0.16107372045516968,
+      "step": 240
+    },
+    {
+      "epoch": 1.7985611510791366,
+      "grad_norm": 1.9613032689723835,
+      "learning_rate": 5.444577056273284e-06,
+      "loss": 0.163911771774292,
+      "step": 250
+    },
+    {
+      "epoch": 1.870503597122302,
+      "grad_norm": 2.092867233484161,
+      "learning_rate": 5.1271762691550375e-06,
+      "loss": 0.15840039253234864,
+      "step": 260
+    },
+    {
+      "epoch": 1.9424460431654675,
+      "grad_norm": 2.208903817919513,
+      "learning_rate": 4.809261313433176e-06,
+      "loss": 0.1667783260345459,
+      "step": 270
+    },
+    {
+      "epoch": 2.014388489208633,
+      "grad_norm": 1.3012808430842233,
+      "learning_rate": 4.492117506643256e-06,
+      "loss": 0.1429366111755371,
+      "step": 280
+    },
+    {
+      "epoch": 2.0863309352517985,
+      "grad_norm": 1.7405033312236557,
+      "learning_rate": 4.17702704859633e-06,
+      "loss": 0.08609053492546082,
+      "step": 290
+    },
+    {
+      "epoch": 2.158273381294964,
+      "grad_norm": 1.629117929895658,
+      "learning_rate": 3.865263837496072e-06,
+      "loss": 0.08188690543174744,
+      "step": 300
+    },
+    {
+      "epoch": 2.2302158273381294,
+      "grad_norm": 2.044856116228938,
+      "learning_rate": 3.5580883196189265e-06,
+      "loss": 0.07971378564834594,
+      "step": 310
+    },
+    {
+      "epoch": 2.302158273381295,
+      "grad_norm": 1.4679168316153588,
+      "learning_rate": 3.256742393379909e-06,
+      "loss": 0.07733185887336731,
+      "step": 320
+    },
+    {
+      "epoch": 2.3741007194244603,
+      "grad_norm": 2.0748124395283836,
+      "learning_rate": 2.9624443883866403e-06,
+      "loss": 0.0740477204322815,
+      "step": 330
+    },
+    {
+      "epoch": 2.446043165467626,
+      "grad_norm": 1.976302404160449,
+      "learning_rate": 2.6763841397811576e-06,
+      "loss": 0.07251015901565552,
+      "step": 340
+    },
+    {
+      "epoch": 2.5179856115107913,
+      "grad_norm": 1.6931347161433883,
+      "learning_rate": 2.3997181777836955e-06,
+      "loss": 0.08174086809158325,
+      "step": 350
+    },
+    {
+      "epoch": 2.5899280575539567,
+      "grad_norm": 1.9804774832049175,
+      "learning_rate": 2.1335650518869555e-06,
+      "loss": 0.07251532077789306,
+      "step": 360
+    },
+    {
+      "epoch": 2.661870503597122,
+      "grad_norm": 1.9966968847358386,
+      "learning_rate": 1.8790008086049534e-06,
+      "loss": 0.07646512985229492,
+      "step": 370
+    },
+    {
+      "epoch": 2.7338129496402876,
+      "grad_norm": 1.9418478201006433,
+      "learning_rate": 1.6370546410597066e-06,
+      "loss": 0.07421975135803223,
+      "step": 380
+    },
+    {
+      "epoch": 2.805755395683453,
+      "grad_norm": 1.8618302510711722,
+      "learning_rate": 1.4087047279943267e-06,
+      "loss": 0.06676316857337952,
+      "step": 390
+    },
+    {
+      "epoch": 2.8776978417266186,
+      "grad_norm": 1.9410030413904114,
+      "learning_rate": 1.1948742790352342e-06,
+      "loss": 0.07537304162979126,
+      "step": 400
+    },
+    {
+      "epoch": 2.949640287769784,
+      "grad_norm": 1.701877525847746,
+      "learning_rate": 9.964278021923107e-07,
+      "loss": 0.07032604217529297,
+      "step": 410
+    },
+    {
+      "epoch": 3.0215827338129495,
+      "grad_norm": 0.8779122416193211,
+      "learning_rate": 8.141676086873574e-07,
+      "loss": 0.0627902865409851,
+      "step": 420
+    },
+    {
+      "epoch": 3.093525179856115,
+      "grad_norm": 0.9568877816955332,
+      "learning_rate": 6.488305692417074e-07,
+      "loss": 0.04383156299591064,
+      "step": 430
+    },
+    {
+      "epoch": 3.1654676258992804,
+      "grad_norm": 1.163317753657708,
+      "learning_rate": 5.010851349371704e-07,
+      "loss": 0.04115341901779175,
+      "step": 440
+    },
+    {
+      "epoch": 3.237410071942446,
+      "grad_norm": 1.1681180832595788,
+      "learning_rate": 3.715286346948671e-07,
+      "loss": 0.04058186709880829,
+      "step": 450
+    },
+    {
+      "epoch": 3.3093525179856114,
+      "grad_norm": 1.1661871807153057,
+      "learning_rate": 2.6068486029813154e-07,
+      "loss": 0.038700607419013974,
+      "step": 460
+    },
+    {
+      "epoch": 3.381294964028777,
+      "grad_norm": 1.1529922962894799,
+      "learning_rate": 1.69001948723142e-07,
+      "loss": 0.04098464548587799,
+      "step": 470
+    },
+    {
+      "epoch": 3.4532374100719423,
+      "grad_norm": 0.9515821172046874,
+      "learning_rate": 9.685057033892998e-08,
+      "loss": 0.03639445900917053,
+      "step": 480
+    },
+    {
+      "epoch": 3.5251798561151078,
+      "grad_norm": 1.2895908079687524,
+      "learning_rate": 4.45224303018027e-08,
+      "loss": 0.038283434510231015,
+      "step": 490
+    },
+    {
+      "epoch": 3.597122302158273,
+      "grad_norm": 0.9797918306377669,
+      "learning_rate": 1.2229089202987487e-08,
+      "loss": 0.036254468560218814,
+      "step": 500
+    },
+    {
+      "epoch": 3.6690647482014387,
+      "grad_norm": 1.0421802932934643,
+      "learning_rate": 1.011077375662195e-10,
+      "loss": 0.03906009197235107,
+      "step": 510
+    },
+    {
+      "epoch": 3.6690647482014387,
+      "step": 510,
+      "total_flos": 55035980414976.0,
+      "train_loss": 0.16481673653219261,
+      "train_runtime": 1297.8966,
+      "train_samples_per_second": 6.287,
+      "train_steps_per_second": 0.393
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 510,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 4,
+  "save_steps": 170,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 55035980414976.0,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}