lerobot/profiling/model_profiling_specs.json

{
  "act": {
    "steps": 12,
    "train_args": [
      "--dataset.repo_id=lerobot/pusht",
      "--dataset.episodes=[0]",
      "--policy.type=act",
      "--policy.device=cuda",
      "--batch_size=4",
      "--cudnn_deterministic=true"
    ]
  },
  "diffusion": {
    "steps": 12,
    "train_args": [
      "--dataset.repo_id=lerobot/pusht",
      "--dataset.episodes=[0]",
      "--policy.type=diffusion",
      "--policy.device=cuda",
      "--batch_size=4",
      "--cudnn_deterministic=true"
    ]
  },
  "groot": {
    "steps": 12,
    "train_args": [
      "--dataset.repo_id=lerobot/libero_plus",
      "--dataset.episodes=[0]",
      "--policy.type=groot",
      "--policy.base_model_path=nvidia/GR00T-N1.5-3B",
      "--policy.tune_diffusion_model=true",
      "--policy.tune_projector=true",
      "--policy.tune_llm=false",
      "--policy.tune_visual=false",
      "--policy.use_bf16=true",
      "--policy.device=cuda",
      "--batch_size=1",
      "--rename_map={\"observation.images.image\": \"observation.images.camera1\", \"observation.images.image2\": \"observation.images.camera2\"}"
    ]
  },
  "multi_task_dit": {
    "steps": 12,
    "train_args": [
      "--dataset.repo_id=lerobot/pusht",
      "--dataset.episodes=[0]",
      "--policy.type=multi_task_dit",
      "--policy.device=cuda",
      "--policy.horizon=32",
      "--policy.n_action_steps=30",
      "--batch_size=4",
      "--cudnn_deterministic=true"
    ]
  },
  "pi0": {
    "steps": 12,
    "train_args": [
      "--dataset.repo_id=lerobot/libero_plus",
      "--dataset.episodes=[0]",
      "--policy.path=lerobot/pi0_base",
      "--policy.device=cuda",
      "--policy.dtype=bfloat16",
      "--policy.n_action_steps=30",
      "--policy.use_amp=true",
      "--policy.gradient_checkpointing=true",
      "--batch_size=1",
      "--use_policy_training_preset=false",
      "--optimizer.type=sgd",
      "--optimizer.lr=1e-5",
      "--optimizer.weight_decay=0",
      "--optimizer.grad_clip_norm=1.0",
      "--scheduler.type=cosine_decay_with_warmup",
      "--scheduler.peak_lr=1e-5",
      "--scheduler.decay_lr=1e-6",
      "--scheduler.num_warmup_steps=0",
      "--scheduler.num_decay_steps=12",
      "--rename_map={\"observation.images.front\": \"observation.images.base_0_rgb\", \"observation.images.wrist\": \"observation.images.left_wrist_0_rgb\"}"
    ]
  },
  "pi0_fast": {
    "steps": 12,
    "train_args": [
      "--dataset.repo_id=lerobot/libero_plus",
      "--dataset.episodes=[0]",
      "--policy.path=lerobot/pi0fast-base",
      "--policy.device=cuda",
      "--policy.dtype=bfloat16",
      "--policy.n_action_steps=30",
      "--policy.use_amp=true",
      "--policy.gradient_checkpointing=true",
      "--batch_size=1",
      "--use_policy_training_preset=false",
      "--optimizer.type=sgd",
      "--optimizer.lr=1e-5",
      "--optimizer.weight_decay=0",
      "--optimizer.grad_clip_norm=1.0",
      "--scheduler.type=cosine_decay_with_warmup",
      "--scheduler.peak_lr=1e-5",
      "--scheduler.decay_lr=1e-6",
      "--scheduler.num_warmup_steps=0",
      "--scheduler.num_decay_steps=12",
      "--rename_map={\"observation.images.front\": \"observation.images.base_0_rgb\", \"observation.images.wrist\": \"observation.images.left_wrist_0_rgb\"}"
    ]
  },
  "pi05": {
    "steps": 12,
    "train_args": [
      "--dataset.repo_id=lerobot/libero_plus",
      "--dataset.episodes=[0]",
      "--policy.path=lerobot/pi05_base",
      "--policy.device=cuda",
      "--policy.dtype=bfloat16",
      "--policy.n_action_steps=30",
      "--policy.use_amp=true",
      "--policy.gradient_checkpointing=true",
      "--batch_size=1",
      "--use_policy_training_preset=false",
      "--optimizer.type=sgd",
      "--optimizer.lr=1e-5",
      "--optimizer.weight_decay=0",
      "--optimizer.grad_clip_norm=1.0",
      "--scheduler.type=cosine_decay_with_warmup",
      "--scheduler.peak_lr=1e-5",
      "--scheduler.decay_lr=1e-6",
      "--scheduler.num_warmup_steps=0",
      "--scheduler.num_decay_steps=12",
      "--policy.normalization_mapping={\"ACTION\": \"MEAN_STD\", \"STATE\": \"MEAN_STD\", \"VISUAL\": \"IDENTITY\"}",
      "--rename_map={\"observation.images.front\": \"observation.images.base_0_rgb\", \"observation.images.wrist\": \"observation.images.left_wrist_0_rgb\"}"
    ]
  },
  "smolvla": {
    "steps": 12,
    "train_args": [
      "--dataset.repo_id=lerobot/libero_plus",
      "--dataset.episodes=[0]",
      "--policy.path=lerobot/smolvla_base",
      "--policy.load_vlm_weights=true",
      "--policy.freeze_vision_encoder=false",
      "--policy.train_expert_only=false",
      "--policy.empty_cameras=1",
      "--policy.device=cuda",
      "--batch_size=1",
      "--rename_map={\"observation.images.front\": \"observation.images.camera1\", \"observation.images.wrist\": \"observation.images.camera2\"}"
    ]
  },
  "wall_x": {
    "steps": 12,
    "train_args": [
      "--dataset.repo_id=lerobot/aloha_sim_insertion_human",
      "--dataset.episodes=[0]",
      "--policy.type=wall_x",
      "--policy.pretrained_name_or_path=x-square-robot/wall-oss-flow",
      "--policy.prediction_mode=diffusion",
      "--policy.attn_implementation=eager",
      "--policy.device=cuda",
      "--batch_size=1"
    ]
  },
  "xvla": {
    "steps": 12,
    "train_args": [
      "--dataset.repo_id=lerobot/libero_plus",
      "--dataset.episodes=[0]",
      "--policy.path=lerobot/xvla-widowx",
      "--policy.action_mode=auto",
      "--policy.empty_cameras=1",
      "--policy.device=cuda",
      "--batch_size=1",
      "--rename_map={\"observation.images.image\": \"observation.images.camera1\", \"observation.images.image2\": \"observation.images.camera2\"}"
    ]
  }
}