Returns the stride of a JAX array at a given axis. To calculate all strides, use get_strides. Args: array: JAX array or shape-d

Save the current parameters for restoring later. Args: parameters: Iterable of `torch.nn.Parameter`; the parameters to be

Recursively apply functions to a nested dictionary or list or tuple, given a dictionary of {data_type: function_to_apply}. Args:

Converts a JAX dtype to a Triton dtype. Args: dtype: JAX dtype. Returns: Triton dtype.

Resets the context of the model.

Converts a denoiser output to a Karras ODE derivative.

Updates the step count.

Update currently maintained parameters. Call this every time the parameters are updated, such as the result of the `optimizer

Calculates the noise level (sigma_down) to step down to and the amount of noise to add (sigma_up) when doing an ancestral sampling step.

Used to set all or multiple parameters of the benchmark at once.

Apply function @func to torch.Tensor objects in a nested dictionary or list or tuple. Args: x (dict or list or tuple): a possibl

Load a CLIP model Parameters ---------- name : str A model name listed by `clip.available_models()`, or the path to a model check

Create plot and save or return it.

Appends dimensions to the end of a tensor until it has target_dims dimensions.

Compute the required embeddings for the visual ones and the latent goal.

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

Compile the provided function with CUDA graphs. Args: fn: The function to compile. Should take no arguments. warmups: The nu

Compute the scalings for the denoising process. Args: sigma: The input sigma. Returns: The computed

Runs the benchmark and returns the runtime in milliseconds.

Soft caps a tensor to a value. Performs a tanh operation on the logits and scales the result to the cap value. Common technique in attention

Simple block wrapping a mixer class with LayerNorm/RMSNorm and residual connection" This Block has a slightly different structure co

Decodes latent samples into images. Parameters ---------- latent_samples : torch.autograd.Variable Samples from l

By default use the sum of the output as loss.

Return the corresponding traversal range in absolute terms.