| data_cfgs: |
| eval_data_files: null |
| eval_datasets: null |
| eval_optional_args: [] |
| eval_size: null |
| eval_split: null |
| eval_subset: null |
| eval_template: null |
| train_data_files: 8713_tokenized_v2.pt |
| train_datasets: /aifs4su/yaodong/datasets/llava_annotated |
| train_optional_args: [] |
| train_size: null |
| train_split: train |
| train_subset: null |
| train_template: ANYTHING_TI2TI |
| logger_cfgs: |
| cache_dir: null |
| log_project: align-anything |
| log_run_name: sft |
| log_type: wandb |
| output_dir: ../outputs/sft_chameleon_0727_0802_v2.1_1e-4 |
| save_interval: 500.0 |
| model_cfgs: |
| model_max_length: 4096 |
| model_name_or_path: /aifs4su/yaodong/projects/hantao/anole/facilitating_image_generation/model/chameleon_hf_0727 |
| trust_remote_code: true |
| special_tokens: null |
| train_cfgs: |
| adam_betas: |
| - 0.9 |
| - 0.95 |
| adam_epsilon: 1.0e-08 |
| bf16: true |
| ds_cfgs: ds_z3_config.json |
| epochs: 3 |
| eval_interval: 1000 |
| eval_strategy: steps |
| fp16: false |
| freeze_language_model: false |
| gradient_accumulation_steps: 2.0 |
| gradient_checkpointing: true |
| learning_rate: 0.0001 |
| lr_scheduler_type: cosine |
| lr_warmup_ratio: 0.03 |
| max_grad_norm: 1.0 |
| per_device_eval_batch_size: 2.0 |
| per_device_train_batch_size: 2.0 |
| seed: 42 |
| weight_decay: 0.0 |
|
|
