Returns config for specified gemma variant.

Returns the freeze filter based on the model config.

Get parameters from model, handling DDP wrapper.

Get state dict from model, handling DDP wrapper.

r""" Returns: text_features (`torch.FloatTensor` of shape `(batch_size, output_dim`): The text embeddings obtained by

Disable gradient checkpointing.

Convenience method for initializing all parameters, necessary due to the quirks of linen.

Check if gradient checkpointing is enabled.

Load the normalization stats from a directory.

Load JAX model and optionally convert to PyTorch. Args: checkpoint_dir: Path to the JAX checkpoint directory output_path: Path to

Creates a random input example for the Aloha policy.

Make a boolean mask for the given dimensions. Example: make_bool_mask(2, -2, 2) == (True, True, False, False, True, True) make_bo

Creates a random input example for the Droid policy.

Creates a random input example for the Libero policy.

Download a file or directory from a remote filesystem to the local cache, and return the local path. If the local file already exists, it will be

The model type.

A higher-order function to JIT-compile `nnx.Module` methods, freezing the module's state in the process. Why not `nnx.jit`? For some reason, naiv

Preprocess the observations by performing image augmentations (if train=True), resizing (if necessary), and filling in a default image mask (if ne

Torch.compile-compatible version of preprocess_observation_pytorch with simplified type annotations. This function avoids complex type annotation

Reset the policy to its initial state.

Reset the environment to its initial state. This will be called once before starting each episode.

Replicates tf.image.resize_with_pad for multiple images using PIL. Resizes a batch of images to a target height. Args: images: A batch of

Replicates tf.image.resize_with_pad. Resizes an image to a target height and width without distortion by padding with black. If the image is float

PyTorch version of resize_with_pad. Resizes an image to a target height and width without distortion by padding with black. If the image is float3

Restores unstructured params PyTree from a checkpoint. This works with checkpoints saved with `save_state` during openpi training (see `training/

Reformat observation and action keys, sample language instruction.

Runs the runtime loop in a new thread.