choose caption type according to caption_proportion, only choose available types. Guarantee: return a caption type that exists in inf

Get a numpy array of weights, one per diffusion step. The weights needn't be normalized, but must be positive.

Return a ChunkedSampler for the current worker in distributed training. Reverts to a simple ChunkedSampler if not running in distributed mode.

This method allows the object to be called like a function. It simply calls the forward method.

Dispatch to :meth:`forward_long` for cached AR inference, else :meth:`forward`. ``forward_long`` is selected iff the caller passes a non-``No

This method allows the object to be called like a function. It simply calls the forward method.

This method allows the object to be called like a function. It simply calls the forward method.

This method allows the object to be called like a function. It simply calls the forward method.

Center cropping implementation from ADM. https://github.com/openai/guided-diffusion/blob/8fb3ad9197f16bbc40620447b2742e13458d2831/gui

Base class for denoising loss. Input: - x: the clean data with shape [B, F, C, H, W] - x_pred: the predicted

Args: clip (torch.tensor, dtype=torch.uint8): Size is (T, C, H, W) Return: clip (torch.tensor, dtype=torch.fl

Args: clip (torch.tensor): Video clip to be cropped. Size is (T, C, H, W) Returns: torch.tensor: scale resize

Make the object callable by delegating to forward method.

Return whether the cache contains the given key.

Remove the given key from the cache.