config.yaml

resume: false
device: cuda
use_amp: false
seed: 100000
dataset_repo_id: lerobot/pusht
training:
  offline_steps: 800000
  online_steps: 0
  online_steps_between_rollouts: 1
  online_sampling_ratio: 0.5
  online_env_seed: ???
  eval_freq: 20000
  save_freq: 20000
  log_freq: 250
  save_checkpoint: true
  num_workers: 4
  batch_size: 64
  grad_clip_norm: 10
  lr: 0.0001
  lr_scheduler: cosine
  lr_warmup_steps: 500
  adam_betas:
  - 0.95
  - 0.999
  adam_eps: 1.0e-08
  adam_weight_decay: 1.0e-06
  vqvae_lr: 0.001
  n_vqvae_training_steps: 20000
  bet_weight_decay: 0.0002
  bet_learning_rate: 5.5e-05
  bet_betas:
  - 0.9
  - 0.999
  delta_timestamps:
    observation.image:
    - -0.4
    - -0.3
    - -0.2
    - -0.1
    - 0.0
    observation.state:
    - -0.4
    - -0.3
    - -0.2
    - -0.1
    - 0.0
    action:
    - -0.4
    - -0.3
    - -0.2
    - -0.1
    - 0.0
    - 0.1
    - 0.2
    - 0.3
    - 0.4
    - 0.5
    - 0.6
    - 0.7
    - 0.8
    - 0.9
    - 1.0
eval:
  n_episodes: 500
  batch_size: 50
  use_async_envs: false
wandb:
  enable: true
  disable_artifact: false
  project: lerobot
  notes: ''
fps: 10
env:
  name: pusht
  task: PushT-v0
  image_size: 96
  state_dim: 2
  action_dim: 2
  fps: ${fps}
  episode_length: 300
  gym:
    obs_type: pixels_agent_pos
    render_mode: rgb_array
    visualization_width: 384
    visualization_height: 384
override_dataset_stats:
  observation.image:
    mean:
    - - - 0.5
    - - - 0.5
    - - - 0.5
    std:
    - - - 0.5
    - - - 0.5
    - - - 0.5
  observation.state:
    min:
    - 13.456424
    - 32.938293
    max:
    - 496.14618
    - 510.9579
  action:
    min:
    - 12.0
    - 25.0
    max:
    - 511.0
    - 511.0
policy:
  name: vqbet
  n_obs_steps: 5
  n_action_pred_token: 7
  action_chunk_size: 5
  input_shapes:
    observation.image:
    - 3
    - 96
    - 96
    observation.state:
    - ${env.state_dim}
  output_shapes:
    action:
    - ${env.action_dim}
  input_normalization_modes:
    observation.image: mean_std
    observation.state: min_max
  output_normalization_modes:
    action: min_max
  vision_backbone: resnet18
  crop_shape:
  - 84
  - 84
  crop_is_random: true
  pretrained_backbone_weights: null
  use_group_norm: true
  spatial_softmax_num_keypoints: 32
  discretize_step: ${training.n_vqvae_training_steps}
  vqvae_n_embed: 16
  vqvae_embedding_dim: 256
  vqvae_enc_hidden_dim: 128
  gpt_block_size: 500
  gpt_input_dim: 512
  gpt_output_dim: 512
  gpt_n_layer: 8
  gpt_n_head: 8
  gpt_hidden_dim: 512
  dropout: 0.1
  mlp_hidden_dim: 1024
  offset_loss_weight: 10000.0
  primary_code_loss_weight: 5.0
  secondary_code_loss_weight: 0.5
  bet_softmax_temperature: 0.01
  sequentially_select: false