Upload folder using huggingface_hub

checkpoints/pytorch_model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3a82a5ce5494ce85a3c5dca1f381195520f0f60824aafbbbe56ccf8d39f21f33
+size 18982885927

config.yaml

@@ -0,0 +1,95 @@
+seed: 42
+trackers:
+- jsonl
+- wandb
+wandb_entity: zbdz
+wandb_project: vla_jiang
+is_debug: false
+framework:
+  name: QwenFM
+  qwenvl:
+    base_vlm: /root/Unifolm-VLM-0
+    attn_implementation: flash_attention_2
+    vl_hidden_dim: 2048
+    model_type: qwen2_5_vl
+  action_model:
+    input_embedding_dim: 1536
+    hidden_size: 1024
+    add_pos_embed: true
+    max_seq_len: 1024
+    action_dim: 7
+    state_dim: 7
+    future_action_window_size: 15
+    action_horizon: 16
+    past_action_window_size: 0
+    repeated_diffusion_steps: 8
+    noise_beta_alpha: 1.5
+    noise_beta_beta: 1.0
+    noise_s: 0.999
+    num_timestep_buckets: 1000
+    num_inference_timesteps: 4
+    num_target_vision_tokens: 32
+    diffusion_model_cfg:
+      cross_attention_dim: 2048
+      attention_head_dim: 48
+      num_attention_heads: 32
+      dropout: 0.2
+      final_dropout: true
+      interleave_self_attention: true
+      norm_type: ada_norm
+      num_layers: 16
+      output_dim: 1024
+      positional_embeddings: null
+    use_relative_action: false
+  framework_py: unifolm_vla
+datasets:
+  vla_data:
+    data_root_dir: /DATA/disk0/rlds_format_datasets/g1
+    data_mix: Unitree_all_task
+    per_device_batch_size: 16
+    load_all_data_for_training: true
+    image_size:
+    - 224
+    - 224
+    window_size: 1
+trainer:
+  epochs: 100
+  max_train_steps: 300000
+  num_warmup_steps: 5000
+  save_interval: 10000
+  eval_interval: 500
+  learning_rate:
+    base: 4.0e-05
+    qwen_vl_interface: 1.0e-05
+    action_model: 0.0004
+  lr_scheduler_type: cosine_with_min_lr
+  scheduler_specific_kwargs:
+    min_lr: 5.0e-07
+  freeze_modules: null
+  loss_scale:
+    vla: 1.0
+    vlm: 0.1
+  repeated_diffusion_steps: 4
+  max_grad_norm: 1.0
+  warmup_ratio: 0.1
+  weight_decay: 0.0
+  logging_frequency: 500
+  gradient_clipping: 1.0
+  gradient_accumulation_steps: 1
+  optimizer:
+    name: AdamW
+    betas:
+    - 0.9
+    - 0.95
+    eps: 1.0e-08
+    weight_decay: 1.0e-08
+  is_resume: false
+  resume_epoch: null
+  resume_step: null
+  enable_gradient_checkpointing: true
+  enable_mixed_precision_training: true
+  shuffle_buffer_size: 10000
+  use_wrist_image: true
+  use_proprio: true
+  action_queries: 0
+