Upload starvla/v0320_v61_qwenOFT_finetune_v2/config.yaml with huggingface_hub

starvla/v0320_v61_qwenOFT_finetune_v2/config.yaml

@@ -0,0 +1,51 @@
+datasets:
+  vla_data:
+    data_mix: custom_v0320_v61
+    data_root_dir: playground/Datasets/Custom
+    dataset_py: lerobot_datasets
+    image_size:
+    - 224
+    - 224
+    per_device_batch_size: 8
+    video_backend: torchvision_av
+framework:
+  action_model:
+    action_dim: 14
+    action_hidden_dim: 2560
+    action_model_type: DiT-B
+    future_action_window_size: 15
+    past_action_window_size: 0
+  name: QwenOFT
+  qwenvl:
+    attn_implementation: sdpa
+    base_vlm: playground/Pretrained_models/Qwen3-VL-4B-Instruct
+output_dir: ./results/Checkpoints/v0320_v61_qwenOFT_finetune_v2
+run_id: v0320_v61_qwenOFT_finetune_v2
+run_root_dir: ./results/Checkpoints
+seed: 42
+trainer:
+  eval_interval: 1000
+  freeze_modules: null
+  gradient_accumulation_steps: 2
+  gradient_clipping: 1.0
+  is_resume: true
+  learning_rate:
+    action_model: 0.0001
+    base: 1.0e-05
+    qwen_vl_interface: 1.0e-05
+  logging_frequency: 100
+  lr_scheduler_type: cosine_with_min_lr
+  max_train_steps: 60000
+  num_warmup_steps: 5000
+  optimizer:
+    betas:
+    - 0.9
+    - 0.95
+    eps: 1.0e-08
+    weight_decay: 1.0e-08
+  pretrained_checkpoint: ./checkpoints/Qwen3-VL-OFT-RoboTwin2-All/checkpoints/steps_140000_pytorch_model.pt
+  save_interval: 60000
+  scheduler_specific_kwargs:
+    min_lr: 5.0e-07
+wandb_entity: hca
+wandb_project: starVLA_v61_finetune_v2