File size: 1,266 Bytes
ae1b277 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 |
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: extracted_preference_1.5k_washed.json
train_datasets: /aifs4su/yaodong/datasets/aaa_dataset/TV2T-preference/extracted
train_optional_args: []
train_size: null
train_split: train
train_subset: null
train_template: NExTQA_preference
logger_cfgs:
cache_dir: null
log_project: align-anything
log_run_name: dpo
log_type: wandb
output_dir: ../outputs/dpo_tiv2t_1.5k_base
save_interval: 100000
model_cfgs:
model_max_length: 4096
model_name_or_path: /aifs4su/yaodong/models/Qwen2-VL-7B-Instruct
trust_remote_code: true
special_tokens: null
train_cfgs:
adam_betas:
- 0.9
- 0.95
bf16: true
ds_cfgs: ds_z3_config.json
epochs: 3
eval_interval: 10
eval_strategy: epoch
fp16: false
freeze_language_model: false
freeze_mm_proj: false
freeze_vision_tower: true
gradient_accumulation_steps: 1.0
gradient_checkpointing: true
learning_rate: 1.0e-06
lr_scheduler_type: cosine
lr_warmup_ratio: 0.01
per_device_eval_batch_size: 1.0
per_device_train_batch_size: 1.0
regularization: 0.001
scale_coeff: 0.1
seed: 42
weight_decay: 0.0
|