How to run pi05_full_droid_finetune with pytorch?

[flyyyyer]

When I run "python scripts/train_pytorch.py pi05_full_droid_finetune", I got the error of "PyTorch RLDS data loader is not supported yet"

TrainConfig(
    # This config is for fine-tuning pi05 on the *full* DROID dataset.
    # We use RLDS data loading to make training on this large dataset tractable.
    # For fine-tuning on your own DROID dataset, see below.
    name="pi05_full_droid_finetune",
    model=pi0_config.Pi0Config(
        pi05=True,
        action_dim=32,
        action_horizon=16,
    ),
    data=RLDSDroidDataConfig(
        repo_id="droid",
        # Set this to the path to your DROID RLDS dataset (the parent directory of the `droid` directory).
        rlds_data_dir="./data/datasets/DROID/",
        action_space=droid_rlds_dataset.DroidActionSpace.JOINT_POSITION,
        assets=AssetsConfig(
            assets_dir="gs://openpi-assets/checkpoints/pi05_base/assets/",
            asset_id="droid",
        ),
    ),
    weight_loader=weight_loaders.CheckpointWeightLoader("gs://openpi-assets/checkpoints/pi05_base/params"),
    lr_schedule=_optimizer.CosineDecaySchedule(
        warmup_steps=1_000,
        peak_lr=5e-5,
        decay_steps=1_000_000,
        decay_lr=5e-5,
    ),
    num_train_steps=100_000,
    batch_size=256,
    log_interval=100,
    save_interval=5000,
    keep_period=10_000,
    num_workers=0,  # Important: RLDS DataLoader requires num_workers=0, handles multi-processing internally
),

[YanYaoyuan]

HI, is it sloved?

[tahsinkose]

Here is a fix for that: #920