| { |
| "model_name": "Qwen/Qwen3-VL-8B-Instruct", |
| "data_dir": "dataset", |
| "image_dir": "images", |
| "image_hf_repo_id": "mlai-dante/road-image", |
| "image_hf_filename": "images.tar.gz", |
| "auto_download_images": true, |
| "force_download_images": false, |
| "output_dir": "outputs/qwen3_vl_rtx_pro_6000", |
| "val_size": 0.15, |
| "seed": 42, |
| "limit_train": 0, |
| "epochs": 1.0, |
| "per_device_train_batch_size": 2, |
| "per_device_eval_batch_size": 1, |
| "gradient_accumulation_steps": 4, |
| "learning_rate": 0.0002, |
| "warmup_ratio": 0.03, |
| "weight_decay": 0.01, |
| "logging_steps": 10, |
| "save_steps": 500, |
| "max_new_tokens": 128, |
| "min_pixels": 100352, |
| "max_pixels": 802816, |
| "lora_r": 8, |
| "lora_alpha": 16, |
| "lora_dropout": 0.05, |
| "target_modules": "q_proj,k_proj,v_proj,o_proj,gate_proj,up_proj,down_proj", |
| "bf16": true, |
| "gradient_checkpointing": true, |
| "attn_implementation": "sdpa", |
| "optim": "adamw_torch_fused", |
| "dataloader_num_workers": 4, |
| "trust_remote_code": true, |
| "do_train": true, |
| "do_validate": true, |
| "do_submit": true, |
| "push_to_hub": true, |
| "hub_upload_artifacts": true, |
| "preflight_only": false, |
| "shutdown_command": "", |
| "shutdown_delay_seconds": 60, |
| "hub_repo_id": "mlai-dante/road-model", |
| "hub_private": false, |
| "hub_commit_message": "Upload ROAD Qwen3-VL adapter and submission" |
| } |
