grail_v1_traj_b64_sft_e3_ck300

.gitattributes

@@ -84,3 +84,4 @@ grail_v2op_fcur_sr_b64_sft_e3_ck640/tokenizer.json filter=lfs diff=lfs merge=lfs
 grail_v2op_fcurlong_b64_sft_e3_ck640/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 grail_v2op_fcur_frombase_b64_sft_e0_ck640/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 grail_v2op_fcur_lora64_b64_sft_e3_merged/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+grail_v1_traj_b64_sft_e3_ck300/tokenizer.json filter=lfs diff=lfs merge=lfs -text

grail_v1_traj_b64_sft_e3_ck300/all_results.json

@@ -0,0 +1,12 @@
+{
+    "epoch": 5.264900662251655,
+    "eval_loss": 0.0813404768705368,
+    "eval_runtime": 14.1229,
+    "eval_samples_per_second": 13.524,
+    "eval_steps_per_second": 3.399,
+    "total_flos": 125011789873152.0,
+    "train_loss": 0.06632403369992972,
+    "train_runtime": 3433.8695,
+    "train_samples_per_second": 5.591,
+    "train_steps_per_second": 0.087
+}

grail_v1_traj_b64_sft_e3_ck300/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 %}

grail_v1_traj_b64_sft_e3_ck300/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
+}

grail_v1_traj_b64_sft_e3_ck300/eval_results.json

@@ -0,0 +1,7 @@
+{
+    "epoch": 5.264900662251655,
+    "eval_loss": 0.0813404768705368,
+    "eval_runtime": 14.1229,
+    "eval_samples_per_second": 13.524,
+    "eval_steps_per_second": 3.399
+}

grail_v1_traj_b64_sft_e3_ck300/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"
+}

grail_v1_traj_b64_sft_e3_ck300/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b0fcbe0dd75292b697a0f54ec8e9048270d58a75120292036235b276eb07b5d0
+size 17534340584

grail_v1_traj_b64_sft_e3_ck300/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"
+  }
+}

grail_v1_traj_b64_sft_e3_ck300/sft.yaml

@@ -0,0 +1,58 @@
+# 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: grail_v1_traj_sft_e3, exps/sft/grail_v1_traj_sft_e3_k1536, /scratch/kwc5354/viewagent/viewagent_private/exps/sft_runs/grail_v1_traj_b64_sft_e3.
+### model
+model_name_or_path: /scratch/kwc5354/viewagent/viewagent_private/exps/models/graph_nemi_sft_e2
+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: /scratch/kwc5354/viewagent/viewagent_private/LLaMA-Factory/examples/deepspeed/ds_z3_config.json
+
+### dataset
+dataset: grail_v1_traj_sft_e3
+dataset_dir: exps/sft/grail_v1_traj_sft_e3_k1536
+media_dir: /scratch/kwc5354/viewagent/viewagent_private
+cutoff_len: 8192
+overwrite_cache: true
+preprocessing_num_workers: 16
+dataloader_num_workers: 4
+
+### output
+output_dir: /scratch/kwc5354/viewagent/viewagent_private/exps/sft_runs/grail_v1_traj_b64_sft_e3
+logging_steps: 10
+save_strategy: steps
+save_steps: 300
+save_total_limit: 100
+save_only_model: true
+plot_loss: true
+overwrite_output_dir: true
+report_to: wandb
+
+### train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 16
+max_steps: 300
+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
+val_size: 0.05
+eval_strategy: steps
+eval_steps: 300
+per_device_eval_batch_size: 1
+run_name: grail_v1_traj_b64_sft_e3

grail_v1_traj_b64_sft_e3_ck300/tokenizer.json

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

grail_v1_traj_b64_sft_e3_ck300/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
+}

grail_v1_traj_b64_sft_e3_ck300/train_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 5.264900662251655,
+    "total_flos": 125011789873152.0,
+    "train_loss": 0.06632403369992972,
+    "train_runtime": 3433.8695,
+    "train_samples_per_second": 5.591,
+    "train_steps_per_second": 0.087
+}

grail_v1_traj_b64_sft_e3_ck300/trainer_state.json

@@ -0,0 +1,261 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 5.264900662251655,
+  "eval_steps": 300,
+  "global_step": 300,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.17660044150110377,
+      "grad_norm": 3.5069694604352004,
+      "learning_rate": 1e-05,
+      "loss": 0.8057702064514161,
+      "step": 10
+    },
+    {
+      "epoch": 0.35320088300220753,
+      "grad_norm": 0.6100709602569099,
+      "learning_rate": 9.970890689061622e-06,
+      "loss": 0.15153673887252808,
+      "step": 20
+    },
+    {
+      "epoch": 0.5298013245033113,
+      "grad_norm": 0.591385306998155,
+      "learning_rate": 9.883901697039809e-06,
+      "loss": 0.10313214063644409,
+      "step": 30
+    },
+    {
+      "epoch": 0.7064017660044151,
+      "grad_norm": 0.4694260719229303,
+      "learning_rate": 9.740045899781353e-06,
+      "loss": 0.09645296335220337,
+      "step": 40
+    },
+    {
+      "epoch": 0.8830022075055187,
+      "grad_norm": 0.6004214577482926,
+      "learning_rate": 9.540998314539327e-06,
+      "loss": 0.08442718386650086,
+      "step": 50
+    },
+    {
+      "epoch": 1.0529801324503312,
+      "grad_norm": 0.387708909325198,
+      "learning_rate": 9.289076596533873e-06,
+      "loss": 0.08105285167694092,
+      "step": 60
+    },
+    {
+      "epoch": 1.229580573951435,
+      "grad_norm": 0.4124167670666316,
+      "learning_rate": 8.987214052813605e-06,
+      "loss": 0.06465960144996644,
+      "step": 70
+    },
+    {
+      "epoch": 1.4061810154525387,
+      "grad_norm": 0.4585467137156774,
+      "learning_rate": 8.638925487636847e-06,
+      "loss": 0.060906398296356204,
+      "step": 80
+    },
+    {
+      "epoch": 1.5827814569536423,
+      "grad_norm": 0.4590351271658073,
+      "learning_rate": 8.248266277059607e-06,
+      "loss": 0.0629310965538025,
+      "step": 90
+    },
+    {
+      "epoch": 1.759381898454746,
+      "grad_norm": 0.47660127511967754,
+      "learning_rate": 7.819785149254534e-06,
+      "loss": 0.058675193786621095,
+      "step": 100
+    },
+    {
+      "epoch": 1.9359823399558498,
+      "grad_norm": 0.42678640887570546,
+      "learning_rate": 7.358471220373831e-06,
+      "loss": 0.05494329333305359,
+      "step": 110
+    },
+    {
+      "epoch": 2.1059602649006623,
+      "grad_norm": 0.3813516526907754,
+      "learning_rate": 6.869695902655898e-06,
+      "loss": 0.04111650884151459,
+      "step": 120
+    },
+    {
+      "epoch": 2.282560706401766,
+      "grad_norm": 0.7079484197633896,
+      "learning_rate": 6.3591503611817155e-06,
+      "loss": 0.03635471761226654,
+      "step": 130
+    },
+    {
+      "epoch": 2.45916114790287,
+      "grad_norm": 0.5720760097832721,
+      "learning_rate": 5.832779247517273e-06,
+      "loss": 0.035574516654014586,
+      "step": 140
+    },
+    {
+      "epoch": 2.6357615894039736,
+      "grad_norm": 0.511587358354466,
+      "learning_rate": 5.296711481829227e-06,
+      "loss": 0.0343822181224823,
+      "step": 150
+    },
+    {
+      "epoch": 2.8123620309050774,
+      "grad_norm": 0.5052088083589364,
+      "learning_rate": 4.757188889427761e-06,
+      "loss": 0.03578503727912903,
+      "step": 160
+    },
+    {
+      "epoch": 2.988962472406181,
+      "grad_norm": 0.46208800027866187,
+      "learning_rate": 4.220493522673067e-06,
+      "loss": 0.036393216252326964,
+      "step": 170
+    },
+    {
+      "epoch": 3.1589403973509933,
+      "grad_norm": 0.4338900573397273,
+      "learning_rate": 3.6928745144891733e-06,
+      "loss": 0.021083831787109375,
+      "step": 180
+    },
+    {
+      "epoch": 3.335540838852097,
+      "grad_norm": 0.48061582646518486,
+      "learning_rate": 3.180475315182563e-06,
+      "loss": 0.018366239964962006,
+      "step": 190
+    },
+    {
+      "epoch": 3.512141280353201,
+      "grad_norm": 0.5468318240903717,
+      "learning_rate": 2.6892621598001157e-06,
+      "loss": 0.017860329151153563,
+      "step": 200
+    },
+    {
+      "epoch": 3.6887417218543046,
+      "grad_norm": 0.4738815193899744,
+      "learning_rate": 2.2249545989326516e-06,
+      "loss": 0.016805607080459594,
+      "step": 210
+    },
+    {
+      "epoch": 3.8653421633554084,
+      "grad_norm": 0.45784115761852623,
+      "learning_rate": 1.7929589018443016e-06,
+      "loss": 0.016949009895324708,
+      "step": 220
+    },
+    {
+      "epoch": 4.035320088300221,
+      "grad_norm": 0.27482802131942474,
+      "learning_rate": 1.3983051073632996e-06,
+      "loss": 0.013842955231666565,
+      "step": 230
+    },
+    {
+      "epoch": 4.211920529801325,
+      "grad_norm": 0.33946611116439473,
+      "learning_rate": 1.0455884554962725e-06,
+      "loss": 0.007631171494722366,
+      "step": 240
+    },
+    {
+      "epoch": 4.3885209713024285,
+      "grad_norm": 0.3925489708914817,
+      "learning_rate": 7.389158817201541e-07,
+      "loss": 0.006717391312122345,
+      "step": 250
+    },
+    {
+      "epoch": 4.565121412803532,
+      "grad_norm": 0.29974629524867163,
+      "learning_rate": 4.818581969574743e-07,
+      "loss": 0.005422165244817733,
+      "step": 260
+    },
+    {
+      "epoch": 4.741721854304636,
+      "grad_norm": 0.41686917004595164,
+      "learning_rate": 2.774085100381735e-07,
+      "loss": 0.00565408430993557,
+      "step": 270
+    },
+    {
+      "epoch": 4.91832229580574,
+      "grad_norm": 0.5247556069039157,
+      "learning_rate": 1.2794737676536993e-07,
+      "loss": 0.006780520081520081,
+      "step": 280
+    },
+    {
+      "epoch": 5.088300220750552,
+      "grad_norm": 0.3037727332451298,
+      "learning_rate": 3.521508137971807e-08,
+      "loss": 0.004680966585874557,
+      "step": 290
+    },
+    {
+      "epoch": 5.264900662251655,
+      "grad_norm": 0.18915971389659425,
+      "learning_rate": 2.9137316938265826e-10,
+      "loss": 0.003832855820655823,
+      "step": 300
+    },
+    {
+      "epoch": 5.264900662251655,
+      "eval_loss": 0.0813404768705368,
+      "eval_runtime": 14.298,
+      "eval_samples_per_second": 13.359,
+      "eval_steps_per_second": 3.357,
+      "step": 300
+    },
+    {
+      "epoch": 5.264900662251655,
+      "step": 300,
+      "total_flos": 125011789873152.0,
+      "train_loss": 0.06632403369992972,
+      "train_runtime": 3433.8695,
+      "train_samples_per_second": 5.591,
+      "train_steps_per_second": 0.087
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 300,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 6,
+  "save_steps": 300,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 125011789873152.0,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}