| _wandb: | |
| value: | |
| cli_version: 0.18.5 | |
| m: [] | |
| python_version: 3.11.10 | |
| t: | |
| "1": | |
| - 1 | |
| - 11 | |
| - 41 | |
| - 49 | |
| - 51 | |
| - 55 | |
| - 71 | |
| - 83 | |
| - 98 | |
| - 105 | |
| "2": | |
| - 1 | |
| - 11 | |
| - 41 | |
| - 49 | |
| - 51 | |
| - 55 | |
| - 71 | |
| - 83 | |
| - 98 | |
| - 105 | |
| "3": | |
| - 2 | |
| - 13 | |
| - 16 | |
| - 23 | |
| - 55 | |
| - 61 | |
| "4": 3.11.10 | |
| "5": 0.18.5 | |
| "6": 4.46.0.dev0 | |
| "8": | |
| - 5 | |
| "12": 0.18.5 | |
| "13": linux-x86_64 | |
| data_cfgs: | |
| value: | |
| eval_data_files: null | |
| eval_datasets: /home/align-anything/datasets/40k_ti2t_filtered | |
| eval_name: null | |
| eval_optional_args: [] | |
| eval_size: null | |
| eval_split: train | |
| eval_subset: null | |
| eval_template: AA_TI2T | |
| train_data_files: null | |
| train_datasets: /data/align-anything/jiayi/ti2t/refine_gen_wo_res/1021_13b_2000 | |
| train_name: null | |
| train_optional_args: [] | |
| train_size: null | |
| train_split: train | |
| train_subset: null | |
| train_template: AA_TI2T_Critique | |
| logger_cfgs: | |
| value: | |
| cache_dir: null | |
| log_project: align-anything | |
| log_run_name: rm | |
| log_type: wandb | |
| output_dir: /data/align-anything/jiayi/ti2t/rm_10_22_ours_13b_2000 | |
| save_interval: 598 | |
| model_cfgs: | |
| value: | |
| model_max_length: 2048 | |
| model_name_or_path: /data/models/llava/llava-1.5-13b-hf | |
| trust_remote_code: true | |
| special_tokens: | |
| value: null | |
| train_cfgs: | |
| value: | |
| adam_betas: | |
| - 0.9 | |
| - 0.95 | |
| bf16: true | |
| ds_cfgs: ds_z3_config.json | |
| epochs: 5 | |
| eval_interval: 10 | |
| eval_strategy: epoch | |
| fp16: false | |
| freeze_language_model: false | |
| freeze_mm_proj: false | |
| freeze_vision_tower: true | |
| gradient_accumulation_steps: 1 | |
| gradient_checkpointing: true | |
| learning_rate: 3e-05 | |
| lr_scheduler_type: constant_with_warmup | |
| lr_warmup_ratio: 0.03 | |
| per_device_eval_batch_size: 4 | |
| per_device_train_batch_size: 4 | |
| regularization: 0.001 | |
| seed: 42 | |
| weight_decay: 0 | |