ChihHanShen's picture
Upload folder using huggingface_hub
aa4e64f verified
datasets:
vla_data:
CoT_prompt: Your task is {instruction}. To identify the key objects for your task.
Locate their bounding boxes in [x1,y1,x2,y2] format.
data_mix: libero_90_task_2
data_root_dir: playground/Datasets/LEROBOT_LIBERO_DATA
dataset_py: lerobot_datasets
per_device_batch_size: 1
sequential_step_sampling: false
video_backend: torchvision_av
framework:
action_model:
action_dim: 7
future_action_window_size: 7
past_action_window_size: 0
name: QwenFast
qwenvl:
base_vlm: playground/Pretrained_models/Qwen2.5-VL-3B-Instruct-Action
output_dir: ./results/Checkpoints/finetune_task2_2000step
run_id: finetune_task2_2000step
run_root_dir: ./results/Checkpoints
seed: 42
trainer:
eval_interval: 100
freeze_modules: qwen_vl_interface.model.model.visual,dino_encoder
gradient_accumulation_steps: 1
gradient_clipping: 1.0
is_resume: true
learning_rate:
action_model: 0.0001
base: 2.5e-05
qwen_vl_interface: 1.0e-05
logging_frequency: 100
lr_scheduler_type: cosine_with_min_lr
max_train_steps: 2000
num_warmup_steps: 5000
optimizer:
betas:
- 0.9
- 0.95
eps: 1.0e-08
weight_decay: 1.0e-08
pretrained_checkpoint: /content/starVLA_r/results/Checkpoints/Qwen2.5-VL-FAST-LIBERO-4in1/checkpoints/steps_30000_pytorch_model.pt
save_interval: 500
scheduler_specific_kwargs:
min_lr: 1.0e-06
wandb_entity: michellelin9102-usc
wandb_project: starVLA_Libero