ddpm_200_v2/policy_config.yaml

_target_: workspaces.train_bfn_workspace.TrainBFNWorkspace
name: train_ddpm_pusht_xarm_200_v2
task_name: pusht_xarm_200_v2
exp_name: ddpm_top
shape_meta:
  obs:
    camera_0:
      shape:
      - 3
      - 224
      - 224
      type: rgb
    robot_eef_pose:
      shape:
      - 2
      type: low_dim
  action:
    shape:
    - 9
horizon: 8
n_obs_steps: 2
n_action_steps: 4
n_latency_steps: 0
dataset_obs_steps: 2
past_action_visible: false
obs_as_global_cond: true
policy:
  _target_: diffusion_policy.policy.diffusion_unet_hybrid_image_policy.DiffusionUnetHybridImagePolicy
  shape_meta: ${shape_meta}
  horizon: ${horizon}
  n_action_steps: ${n_action_steps}
  n_obs_steps: ${n_obs_steps}
  noise_scheduler:
    _target_: diffusers.schedulers.scheduling_ddpm.DDPMScheduler
    num_train_timesteps: 100
    beta_start: 0.0001
    beta_end: 0.02
    beta_schedule: squaredcos_cap_v2
    variance_type: fixed_small
    clip_sample: true
    prediction_type: epsilon
  num_inference_steps: 100
  obs_as_global_cond: ${obs_as_global_cond}
  obs_encoder_group_norm: true
  eval_fixed_crop: true
  crop_shape:
  - 216
  - 216
  diffusion_step_embed_dim: 128
  down_dims:
  - 256
  - 512
  - 1024
  kernel_size: 5
  n_groups: 8
  cond_predict_scale: true
task:
  dataset:
    _target_: dataset.pusht_xarm_dataset.PushTXArmDataset
    zarr_path: data/pusht_xarm_200/replay.zarr
    horizon: ${horizon}
    pad_before: 1
    pad_after: 3
    n_obs_steps: ${n_obs_steps}
    seed: 42
    val_ratio: 0.1
    cameras:
    - camera_0
    use_eef_pose: true
    action_mode: onehot
  env_runner: null
ema:
  _target_: diffusion_policy.model.diffusion.ema_model.EMAModel
  update_after_step: 0
  inv_gamma: 1.0
  power: 0.75
  min_value: 0.0
  max_value: 0.9999
optimizer:
  _target_: torch.optim.AdamW
  lr: 0.0001
  betas:
  - 0.95
  - 0.999
  eps: 1.0e-08
  weight_decay: 1.0e-06
training:
  device: cuda:0
  seed: 42
  debug: false
  resume: false
  lr_scheduler: cosine
  lr_warmup_steps: 500
  num_epochs: 200
  gradient_accumulate_every: 1
  use_ema: true
  rollout_every: 50
  checkpoint_every: 10
  val_every: 1
  sample_every: 5
  max_train_steps: null
  max_val_steps: null
  tqdm_interval_sec: 1.0
dataloader:
  batch_size: 32
  num_workers: 4
  shuffle: true
  pin_memory: true
  persistent_workers: false
val_dataloader:
  batch_size: 32
  num_workers: 2
  shuffle: false
  pin_memory: true
  persistent_workers: false
checkpoint:
  topk:
    monitor_key: val_loss
    mode: min
    k: 3
    format_str: epoch={epoch:04d}-train_loss={train_loss:.4f}.ckpt
  save_last_ckpt: true
  save_last_snapshot: false
logging:
  project: pusht_xarm_200_ddpm
  resume: true
  mode: offline
  name: ${now:%Y.%m.%d-%H.%M.%S}_${name}
  tags:
  - ddpm
  - pusht_xarm
  - onehot
  - top
  id: null
  group: null
multi_run:
  run_dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}
  wandb_name_base: ${now:%Y.%m.%d-%H.%M.%S}_${name}