Upload config.yaml

config.yaml

@@ -0,0 +1,100 @@
+actor:
+  model:
+    model_type: openpi_pytorch
+    precision: fp32
+    is_lora: false
+    lora_rank: 32
+    num_action_chunks: 20
+    action_dim: 10
+    use_proprio: true
+    num_steps: 5
+    add_value_head: false
+    openpi:
+      task: sft
+      config_name: pi05_singlefranka_tcp_rot6d
+      num_images_in_input: 3
+      action_chunk: 20
+      action_env_dim: 10
+      model_action_dim: 32
+      paligemma_variant: gemma_2b
+      action_expert_variant: gemma_300m
+      max_token_len: 200
+      discrete_state_input: false
+      assets_dir: /mnt/workspace/embody-h2/users/zhaozihao/DAgger/assets/pi05_singlefranka_tcp_rot6d/umi_left_franka_tcp_rot6d
+      asset_id: stack_cubes_umi_1_franka
+    model_path: /mnt/workspace/embody-h2/users/zhaozihao/pi05_base_pytorch_new
+  fsdp_config:
+    strategy: fsdp
+    sharding_strategy: full_shard
+    gradient_checkpointing: true
+    cpu_offload: false
+    offload_pin_memory: false
+    reshard_after_forward: true
+    enable_gradient_accumulation: true
+    forward_prefetch: false
+    limit_all_gathers: false
+    backward_prefetch: null
+    use_orig_params: true
+    use_liger_kernel: false
+    mixed_precision:
+      param_dtype: bf16
+      reduce_dtype: fp32
+      buffer_dtype: fp32
+    amp_autocast:
+      enabled: false
+      precision: bf16
+    grad_scaler:
+      enabled: false
+      init_scale: null
+      growth_interval: null
+    gradient_checkpointing_use_reentrant: false
+  group_name: ActorGroup
+  training_backend: fsdp
+  micro_batch_size: 4
+  eval_batch_size: 1
+  global_batch_size: 32
+  seed: 0
+  optim:
+    lr: 2.5e-05
+    adam_beta1: 0.9
+    adam_beta2: 0.95
+    adam_eps: 1.0e-08
+    weight_decay: 1.0e-10
+    clip_grad: 1.0
+    total_training_steps: 30000
+    lr_warmup_steps: 1000
+    lr_scheduler: openpi_cosine
+    num_cycles: 0.5
+    min_lr: 2.5e-06
+cluster:
+  num_nodes: 1
+  component_placement:
+    actor: 0-7
+runner:
+  task_type: sft
+  logger:
+    log_path: /mnt/workspace/embody-h2/users/zhaozihao/DAgger/logs/stack_cubes_umi_1_franka-10
+    project_name: rlinf
+    experiment_name: umi_left_franka_tcp_rot6d_openpi_pytorch_sft
+    logger_backends:
+    - tensorboard
+  max_epochs: -1
+  max_steps: 30000
+  val_check_interval: -1
+  save_interval: 2000
+  resume_dir: /mnt/workspace/embody-h2/users/zhaozihao/checkpoint_archive/stack_cubes_umi_1_franka-8/umi_left_franka_tcp_rot6d_openpi_pytorch_sft/checkpoints/global_step_20000
+  checkpoint_offload_dir: /mnt/workspace/embody-h2/users/zhaozihao/checkpoint_archive
+  per_worker_log: false
+  per_worker_log_path: null
+data:
+  train_data_paths: dataset/stack_cubes_umi_1_franka
+  post_train: true
+  post_train_data_path: /mnt/workspace/embody-h2/users/zhaozihao/DAgger/egodagger_dataset/stack_cubes_umi-egodagger-franka
+  repo_id: ''
+  num_workers: 4
+algorithm:
+  adv_type: gae
+reward:
+  use_reward_model: false
+critic:
+  use_critic_model: false