Interface for a data loader.

A class that keeps all conversation history.

A file lock class.

Class for creating and manipulating bounding boxes around 3D models.

Conv1d --> GroupNorm --> Mish

Diffusion models with a Transformer backbone.

AdamW optimizer.

Determines the location of assets (e.g., norm stats) that will be used to set up the data pipeline. These assets will be replicated inside the ch

A class that manages prompt templates and keeps all conversation history.

Class for rendering 3D scenes to images.

Load a policy from a trained checkpoint.

Load a policy from a trained checkpoint.

Conv1d --> GroupNorm --> Mish

This class is used to sample episodes from the embododiment dataset stored in HDF5.

Creates model transforms for standard pi0 models.

Conv1d --> GroupNorm --> Mish

Focal Loss, as described in https://arxiv.org/abs/1708.02002. It is essentially an enhancement to cross entropy loss and is useful for classif

Dataset for supervised fine-tuning.

This class is used to sample episodes from the embododiment dataset.

Class for creating visual elements like arrows and markers for scene visualization.

Attention module.

Collate examples for supervised fine-tuning.

Wrapper for OpenVLA's vision backbone that implements feature-wise linear modulation (FiLM). Wraps the Vision Transformers in the vision bac

MLP with residual connection blocks.

Transformer MLP / feed-forward block.

Transformer MLP / feed-forward block.

Creates model transforms for standard pi0 models.

Base class for Qwen2VL causal language model (or autoregressive) outputs. Args: loss (`torch.FloatTensor` of shape `(1,)`, *optional