Return the best torch backend available.

Get a value from a nested dictionary using a path.

backwards compatible wrapper for huggingface's hf_hub_download. they changed the argument name from `use_auth_token` to `token`

Make a distorted copy of the data. The node number is the index in the node map.

Generate a random tensor with a given seed.

Given a region of interest, returns a square region of interest.

Create sinusoidal timestep embeddings. :param timesteps: a 1-D Tensor of N indices, one per batch element. These may be

Upsamples the input by the given scale factor. If upsample_factor is None, the input shape is taken from the context. It will throw an error

Controlnet is a Half-UNet that collects residuals from the UNet and uses them to condition the UNet. Input is a `batch 3 width height` tensor

Check if a module has weighted leaf modules that are not basic layers.

Execute a forward pass and store the output of specific sub-modules. - `module`: The module to trace. - `args`: The argument

Count the number of basic layers in a module.

Print the shapes of the sub-modules in `source_keys` and `target_keys`.

Load a diffusion model. Weights location may also be shortcut name, e.g. "SD-1.5"

Have the AI generate images. alias:imagine.

Check if a module type is a subclass of a torch basic layer.