Upload config.yaml with huggingface_hub

config.yaml

@@ -0,0 +1,118 @@
+datasets:
+  vla_data:
+    chunk_len: 8
+    cot_root: /scratch/pawsey1357/sxu1/workspace/code/starDRA/playground/Datasets/LIBERO_COT
+    data_mix: libero_all_cot
+    data_root_dir: /scratch/pawsey1357/sxu1/workspace/code/starDRA/playground/Datasets/LEROBOT_LIBERO_DATA
+    dataset_py: libero_cot_world_datasets
+    emit_state: true
+    emit_subgoal_image: false
+    emit_target_future_frames: true
+    held_out_tasks: null
+    horizon_steps: 20
+    image_size:
+    - 256
+    - 256
+    include_wrist_image: true
+    max_subgoals: 10
+    num_workers: 1
+    output_schema: movement_only
+    per_device_batch_size: 8
+    split_seed: 0
+    suites:
+    - libero_goal
+    - libero_object
+    - libero_spatial
+    - libero_10
+    target_last_frame_only: false
+    use_context_demo: false
+    val_frac: 0.1
+    video_backend: pyav
+framework:
+  action_loss_weight: 1.0
+  action_model:
+    action_dim: 7
+    action_head_type: flowmatching
+    action_loss_type: mse
+    action_model_type: DiT-B
+    add_pos_embed: true
+    diffusion_model_cfg:
+      cross_attention_dim: 2560
+      dropout: 0.2
+      final_dropout: true
+      interleave_self_attention: true
+      norm_type: ada_norm
+      num_layers: 16
+      output_dim: 1024
+      positional_embeddings: null
+    enable_subgoal_latent: false
+    enable_world_latent: true
+    future_action_window_size: 7
+    hidden_size: 1024
+    inject_base_patches: false
+    latent_embedding_form: absolute
+    latent_loss_type: flowmatching
+    latent_n_patch: 64
+    latent_objslots: 0
+    latent_only: true
+    latent_predictor: shared
+    latent_proj_dim: 256
+    latent_stop_grad: false
+    latent_target_dim: 2560
+    latent_target_pool: patch
+    latent_target_source: frozen
+    max_seq_len: 1024
+    noise_beta_alpha: 1.5
+    noise_beta_beta: 1.0
+    num_inference_timesteps: 4
+    num_latent_slots: 8
+    num_target_vision_tokens: 32
+    num_timestep_buckets: 1000
+    past_action_window_size: 0
+    state_dim: 8
+  include_wrist_image: true
+  language_loss_weight: 0.3
+  latent_loss_weight: 1.0
+  max_decode_new_tokens: 512
+  name: QwenGR00TWorldOFTCoT
+  output_schema: movement_only
+  qwenvl:
+    base_vlm: /scratch/pawsey1357/sxu1/workspace/code/starDRA/playground/Pretrained_models/Qwen3-VL-4B-Instruct
+    max_seq_len: 1024
+    vlm_family: qwen3vl
+  sigreg_weight: 0
+  use_context_demo: false
+  viz_interval: 500
+  viz_num_samples: 4
+output_dir: /scratch/pawsey1357/sxu1/workspace/code/starDRA/results/Checkpoints/hicovala_qwengr00t_worldoftcot_libero_qwen4b_state_pawsey_2node_16gpu_bs8_ga1_v1oftcot_30k
+run_id: hicovala_qwengr00t_worldoftcot_libero_qwen4b_state_pawsey_2node_16gpu_bs8_ga1_v1oftcot_30k
+run_root_dir: /scratch/pawsey1357/sxu1/workspace/code/starDRA/results/Checkpoints
+seed: 42
+trainer:
+  enable_gradient_checkpointing: true
+  eval_interval: 100
+  freeze_modules: ''
+  gradient_accumulation_steps: 1
+  gradient_clipping: 1.0
+  is_resume: false
+  learning_rate:
+    action_model: 5.0e-05
+    base: 5.0e-06
+    mlp_action_head: 5.0e-05
+    qwen_vl_interface: 5.0e-06
+  logging_frequency: 50
+  lr_scheduler_type: cosine_with_min_lr
+  max_train_steps: 30000
+  num_warmup_steps: 500
+  optimizer:
+    betas:
+    - 0.9
+    - 0.95
+    eps: 1.0e-08
+    weight_decay: 1.0e-08
+  save_interval: 2000
+  scheduler_specific_kwargs:
+    min_lr: 1.0e-06
+wandb_entity: siyuhsu
+wandb_mode: offline
+wandb_project: starVLA_Libero