| ### model | |
| model_name_or_path: xlangai/Jedi-3B-1080p | |
| image_max_pixels: 211680000000 | |
| trust_remote_code: true | |
| ### method | |
| stage: sft | |
| do_train: true | |
| finetuning_type: full | |
| freeze_vision_tower: true | |
| freeze_multi_modal_projector: false | |
| freeze_language_model: false | |
| deepspeed: examples/deepspeed/ds_z2_config.json | |
| ### dataset | |
| dataset: train_data_jedi_original, train_data_lzy | |
| template: qwen2_vl | |
| cutoff_len: 8000 | |
| max_samples: null | |
| overwrite_cache: true | |
| preprocessing_num_workers: 30 | |
| dataloader_num_workers: 30 | |
| preprocessing_batch_size: 1 | |
| ### output | |
| output_dir: outputs/jedi-7b | |
| logging_steps: 10 | |
| save_strategy: epoch | |
| plot_loss: true | |
| overwrite_output_dir: true | |
| save_only_model: true | |
| report_to: wandb # choices: [none, wandb, tensorboard, swanlab, mlflow] | |
| ### train | |
| per_device_train_batch_size: 8 | |
| gradient_accumulation_steps: 2 | |
| learning_rate: 1.0e-5 | |
| num_train_epochs: 2.0 | |
| lr_scheduler_type: cosine | |
| warmup_ratio: 0.1 | |
| bf16: true | |
| ddp_timeout: 180000000 | |
| resume_from_checkpoint: null | |
| ### eval | |
| # val_size: 0.1 | |
| # per_device_eval_batch_size: 1 | |
| # eval_strategy: steps | |
| # eval_steps: 500 | |