Generate the json dict and state for a given episode.

Generate the json dict and state for a given episode.

Generate the json dict and state for an episode in the dataset without state. If not state, we use the last action as current state.

Get bins per dimension according to the Lookup-Free Quantization paper (2 bins per dimension)

Initializes and returns the LIBERO environment, along with the task description.

Initializes and returns the LIBERO environment, along with the task description.

Handle syncing of files between current directory and checkpoint. Creates backups if files exist but differ, and copies current versions to

Initialize KV cache

Initialize KV cache

Copy of timm.models.vision_transformer.VisionTransformer._intermediate_layers() with modifications to take in language embeddings as

Check if the url is an OpenPI S3 bucket url.

Simple k-means clustering algorithm adapted from Karpathy's minGPT library https://github.com/karpathy/minGPT/blob/master/play_image.

Load dataset statistics used during training for action normalization. Args: vla: The VLA model checkpoint_path: Path to the

Loads and validates the weights. Returns a loaded subset of the weights.

Loads and validates the weights. Returns a loaded subset of the weights.

Patch all LayerScale modules to be compatible with HF's parameter naming. HF Transformers overwrites parameters with names containin

Prepares input for action prediction by adding necessary tokens

Creates labels tensor for action prediction if not provided

Copied from robosuite: https://github.com/ARISE-Initiative/robosuite/blob/eafb81f54ffc104f905ee48a16bb15f059176ad3/robosuite/utils/transform_util

Copied from robosuite: https://github.com/ARISE-Initiative/robosuite/blob/eafb81f54ffc104f905ee48a16bb15f059176ad3/robosuite/utils/transform_util

Receive response with numpy array reconstruction

Run L1 regression-based continuous action prediction or discrete action tokens prediction.

Set to position mode and do position resets: first open then close. Then change back to PWM mode

Set to position mode and do position resets: first open then close. Then change back to PWM mode

Set to position mode and do position resets: first open then close. Then change back to PWM mode

Set to position mode and do position resets: first open then close. Then change back to PWM mode

Replicates tf.image.resize_with_pad for one image using PIL. Resizes an image to a target height and width without distortion by padding with zero

Replicates tf.image.resize_with_pad for one image using PIL. Resizes an image to a target height and width without distortion by padding with zero

Run diffusion-based action prediction

Runs a single episode.

Runs a single episode.

Send request and receive response with numpy array support

Set all files to be read & writable, if it is a script, keep it as a script.

Set all files to be read & writable, if it is a script, keep it as a script.

Invalidate the cache if it is expired. Return True if the cache was invalidated.

Invalidate the cache if it is expired. Return True if the cache was invalidated.

Define data splits.

Define data splits.

Define data splits.

Define data splits.

Define data splits.

Define data splits.

Returns torch.sqrt(torch.max(0, x)) but with a zero subgradient where x is 0.

A single step of the runtime loop.

A single step of the runtime loop.

Unformat the unified action vector into the joint action to be executed. Args: action (torch.Tensor): The unified action

Unnormalize actions using dataset statistics

Update KV cache with new values

Update KV cache with new values