Load a component's state dict from checkpoint and handle DDP prefix if present. Args: checkpoint_path: Path to the checkpoint file

Checks whether a model path points to a model on Hugging Face Hub.

Converts relative gripper actions (+1 for closing, -1 for opening) to absolute actions (0 = closed; 1 = open). Assumes that the first relati

Add bounding box visualization to the scene. Returns: Updated scene with bounding box

Assemble the state/action vector from the arm and base.

Create a SharedNDArray directly from a SharedMemoryManager Parameters ---------- mem_mgr SharedMemoryManager insta

This method forwards all its arguments to Qwen2TokenizerFast's [`~PreTrainedTokenizer.decode`]. Please refer to the docstring of this

All the function interfaces below are just examples You can modify them according to your implementation But we strongly recommend keepin

Find a specific checkpoint file matching a pattern. Args: pretrained_checkpoint: Path to the checkpoint directory file_patte

Generate six views from the GLB file, with the orthogonal camera framework rotated by a random angle, and return a combined image as a single

Get a training sample at a random timestep. Args: index (int, optional): the index of the episode. If not provide

获取 v2 在 v1 上的投影向量

Compute and return the statistics of the vectors processed so far. Returns: dict: A dictionary containing the computed s

Enable gradient checkpointing for memory optimization.

Initializes a module, optionally loads checkpoint, moves to device, and wraps with DDP. Args: module_class (Type[nn.Module]): Class

Loads the model weights. Args: params: Parameters of the model. This is a nested structure of array-like objects that

Pad a nested dictionary or list or tuple of sequence tensors in the time dimension (dimension 1). Args: seq (dict or list or tuple):

Pad an array to the target dimension with zeros along the specified axis.

Core function for VLA inference; maps input image and task instruction to continuous action (de-tokenizes). @param image: PIL Image

Append transforms to the group and return a new group. Args: inputs: Appended to the *end* of the current input transforms.

Register a new conversation template.

Reshape selected dimensions in a tensor to a target dimension. Args: x (torch.Tensor): tensor to reshape begin_axis (int): b

Reshape selected dimensions in a tensor to a target dimension. Args: x (torch.Tensor): tensor to reshape begin_axis (int): b

Run the task in simulation to evaluate success rate. Args: TASK_ENV (object): Task environment instance. args (dict): Task c