{ "run_id": "25214426e4ea472c9a9559a6182b4202", "model": { "model_name_or_path": "models/Qwen3.5-0.8B", "revision": "2fc06364715b967f1860aea9cf38778875588b17", "dtype": "bfloat16", "attn_implementation": "sdpa", "workspace_tokens": 128, "loops": 3, "loop_scope": "middle_workspace", "loop_start_block": 1, "loop_end_block": 24, "detach_loop_state": false, "classification_loss_scope": "all_loops", "aux_weight": 0.3, "aux_dim": 256, "aux_heads": 8, "aux_max_tokens": 4096, "aux_chunk_size": 1024, "aux_response_chunk_size": 16, "freeze_vision": false, "gradient_checkpointing": true, "training_stage": 1, "stage1_objective": "cot_only" }, "train": { "train_file": "data/stage1-multi-response/train.jsonl", "validation_file": "data/stage1-multi-response/validation.jsonl", "output_dir": "runs/stage1-multi-response", "epochs": 1, "batch_size": 32, "gradient_accumulation": 1, "learning_rate": 8e-05, "lr_schedule": "warmup_cosine", "warmup_ratio": 0.0, "min_learning_rate": 8e-06, "weight_decay": 0.01, "max_grad_norm": 1.0, "max_length": 5248, "max_image_pixels": 262144, "seed": 42, "device": "cuda", "max_steps": null, "save_every": 100, "keep_checkpoints": 2, "num_workers": 4, "classifier_learning_rate": null, "init_checkpoint": null, "answer_max_tokens": 256 }, "source_hashes": { "train_file": "ab837eac6a1c676ce86beb9d4fa8f3e6690729309cb731ae28975d353ec49e60", "validation_file": "3695d4a84f0a67d6a636218d92b863ebd7e55ec8a2b8253deaee36c5395dbd6f", "train_images": "c9469e2bbfba6c8762345d355953098e5ee6441c647a0f640757157847d7dbc9", "validation_images": "0a528fd1d1d6482964e75947317a8b06af3cd917dcacb7400b77a2e72baafcb0" }, "learning_rate_plan": { "type": "warmup_cosine", "total_steps": 6347, "warmup_steps": 0, "initial_learning_rate": 8e-05, "min_learning_rate": 8e-06 }, "torch_version": "2.14.0+cu130", "transformers_version": "5.17.0", "parameter_dtype": "float32", "initialization": null }