Set IP-Adapter scales per-transformer block. Input `scale` could be a single config or a list of configs for granular control over ea

r""" Args: Save the current parameters for restoring later. parameters: Iterable of `torch.nn.Parameter`; the parameters t

This does the final conversion step: take locally converted weights and apply a global renaming to them. It splits attention layers, and take

r""" Args: Restore the parameters stored with the `store` method. Useful to validate the model with EMA parameters without: af

Predict the sample from the previous timestep by reversing the SDE. This function propagates the sample with the linear multistep met

Predict the sample from the previous timestep by reversing the SDE. This function propagates the sample with the Runge-Kutta method.

This does the final conversion step: take locally converted weights and apply a global renaming to them. It splits attention layers, and take

r""" Enable long video generation using FreeNoise. Args: context_length (`int`, defaults to `16`, *optional*):

Sets the `lora_state_dict` into `text_encoder` coming from `transformers`. Args: lora_state_dict: The state dictionary to be set.

This does the final conversion step: take locally converted weights and apply a global renaming to them. It splits attention layers, and take

This does the final conversion step: take locally converted weights and apply a global renaming to them. It splits attention layers, and take

r""" Enable tiled VAE decoding. When this option is enabled, the VAE will split the input tensor into tiles to compute decoding and en

r""" Apply perturbed attention guidance to the noise prediction. Args: noise_pred (torch.Tensor): The noise prediction te

Enables the FreeInit mechanism as in https://arxiv.org/abs/2312.07537. This implementation has been adapted from the [official repository](ht

Enables fused QKV projections. For self-attention modules, all projection matrices (i.e., query, key, value) are fused. For cross-att

r""" Set the the self-attention layers to apply PAG. Raise ValueError if the input is invalid. Args: pag_applied_layers (

This does the final conversion step: take locally converted weights and apply a global renaming to them. It splits attention layers, and take

This does the final conversion step: take locally converted weights and apply a global renaming to them. It splits attention layers, and take

This does the final conversion step: take locally converted weights and apply a global renaming to them. It splits attention layers, and take

This does the final conversion step: take locally converted weights and apply a global renaming to them. It splits attention layers, and take

This does the final conversion step: take locally converted weights and apply a global renaming to them. It splits attention layers, and take

This does the final conversion step: take locally converted weights and apply a global renaming to them. It splits attention layers, and take

r""" Returns: `torch.device`: The torch device on which the pipeline is located.

r""" Encodes the prompt into text encoder hidden states. Args: prompt (`str` or `List[str]`, *optional*):

This matches the implementation in Denoising Diffusion Probabilistic Models: Create sinusoidal timestep embeddings. Args timesteps (

Disables the fused QKV projection if enabled. <Tip warning={true}> This API is 🧪 experimental. </Tip>

r""" Maybe convert a prompt into a "multi vector"-compatible prompt. If the prompt includes a token that corresponds to a multi-vector

Runs garbage collection. Then clears the cache of the available accelerator.

Updates paths inside resnets to the new naming scheme (local renaming)

Updates paths inside resnets to the new naming scheme (local renaming)

Updates paths inside resnets to the new naming scheme (local renaming)