| { |
| "created_at": "2026-07-19T06:55:19", |
| "global_optimizer_step": 7344, |
| "git_commit": "e8ffda2aed348adb6b48591f2219b69582c4fc44", |
| "package_versions": { |
| "torch": "2.11.0+cu128", |
| "transformers": "4.46.2", |
| "accelerate": "0.33.0", |
| "bitsandbytes": "0.49.2", |
| "flash-attn": "2.6.3" |
| }, |
| "config_snapshot": { |
| "experiment": { |
| "name": "wad_internvl_pretrain_traj_cls", |
| "version": "1.0", |
| "description": "InternVL2-2B trajectory pretrain with cls_add fusion", |
| "tags": [ |
| "internvl2", |
| "pretrain", |
| "trajectory", |
| "cls_add" |
| ] |
| }, |
| "model": { |
| "architecture": "internvl", |
| "name": "OpenGVLab/InternVL2_5-2B", |
| "trust_remote_code": true, |
| "multimodal_max_length": 8192, |
| "qformer": { |
| "enabled": true, |
| "source_model": "Salesforce/instructblip-flan-t5-xl", |
| "cache_dir": "./qformer_cache", |
| "num_query_tokens": 32, |
| "freeze_qformer": true, |
| "freeze_mlp1": true, |
| "prompt_aware": true, |
| "max_text_length": 128, |
| "train_lora_llm": false, |
| "bridge_mode": "prompt_aware_preproj_mlp1" |
| }, |
| "lora": { |
| "enabled": false, |
| "r": 16, |
| "alpha": 32, |
| "dropout": 0.05, |
| "target_modules": [ |
| "wqkv", |
| "wo", |
| "w1", |
| "w2", |
| "w3" |
| ], |
| "bias": "none", |
| "task_type": "CAUSAL_LM" |
| }, |
| "quantization": { |
| "enabled": true, |
| "bits": 4, |
| "type": "nf4", |
| "double_quant": true, |
| "compute_dtype": "bfloat16" |
| }, |
| "vision": { |
| "freeze_encoder": true, |
| "min_pixels": 256, |
| "max_pixels": 1280, |
| "image_size": [ |
| 448, |
| 448 |
| ] |
| } |
| }, |
| "trajectory": { |
| "enabled": true, |
| "fusion_mode": "cls_add", |
| "source_file": "json", |
| "num_objects": 6, |
| "d_cat": 32, |
| "d_numeric_hidden": 64, |
| "d_traj": 384, |
| "num_heads": 4, |
| "num_layers": 4, |
| "ffn_dim": 768, |
| "dropout": 0.1 |
| }, |
| "pretrain": { |
| "question_train_file": "./json/question_train_split_train.jsonl", |
| "question_val_file": "./json/question_train_split_val.jsonl", |
| "question_test_file": "./json/question_train_split_test.jsonl", |
| "frame_index_file": "./wad_dataset/frame_index.pkl", |
| "train_split_ratio": 0.8, |
| "val_split_ratio": 0.1, |
| "test_split_ratio": 0.1, |
| "val_split_seed": 42, |
| "movement_enabled": true |
| }, |
| "data": { |
| "name": "minhdang0901/WAD_Images_All_Size", |
| "num_frames": 1, |
| "seed": 42, |
| "response_format": "direct_text" |
| }, |
| "training": { |
| "output_dir": "./outputs/pretrain_traj_cls", |
| "num_epochs": 50, |
| "batch_size": 24, |
| "gradient_accumulation_steps": 1, |
| "use_accelerate": true, |
| "learning_rate": "1e-5", |
| "trajectory_learning_rate": "3e-4", |
| "bridge_learning_rate": "5e-5", |
| "warmup_ratio": 0.1, |
| "warmup_min_steps": 20, |
| "warmup_max_steps": 100, |
| "weight_decay": 0.02, |
| "loss_mode": "label_smoothing", |
| "label_smoothing": 0.1, |
| "early_stopping_patience": 8, |
| "early_stopping_min_delta": 0.001, |
| "early_stopping_burn_in_epochs": 1, |
| "restore_best_checkpoint": true, |
| "bf16": true, |
| "fp16": false, |
| "gradient_checkpointing": false, |
| "optimizer": "adamw", |
| "lr_scheduler": "cosine", |
| "max_grad_norm": 1.0, |
| "logging_steps": 10, |
| "log_token_stats": false, |
| "token_log_batches": 0, |
| "log_trainable_parameter_names": false, |
| "profile_steps": 0, |
| "gradient_debug_steps": 3, |
| "gradient_health_steps": 0, |
| "batch_debug_steps": 0, |
| "memory_debug_steps": 0 |
| }, |
| "hardware": { |
| "device": "cuda", |
| "num_workers": 4, |
| "pin_memory": true, |
| "persistent_workers": true, |
| "prefetch_factor": 2, |
| "cuda_alloc_conf": "expandable_segments:True" |
| } |
| } |
| } |
