r""" Function invoked when calling the pipeline for generation. Args: prompt (`str` or `List[str]`, *optional*):

r""" Function invoked when calling the pipeline for generation. Args: prompt (`str` or `List[str]`): The

r""" Function invoked when calling the pipeline for generation. Args: prompt (`str`): The prompt to guid

r""" Function invoked when calling the pipeline for generation. Args: prompt (`str` or `List[str]`): The

Initialize an empty AttentionStore :param step_index: used to visualize only a specific step in the diffusion process

Initialize an empty AttentionStore :param step_index: used to visualize only a specific step in the diffusion process

Aggregates the attention for each token and computes the max activation value for each token to alter.

Aggregates the attention for each token and computes the max activation value for each token to alter.

Helper method to cast floating-point values of given parameter `PyTree` to given `dtype`.

r""" Combines a latent image img_vae of shape (B, C, H, W), a CLIP-embedded image img_clip of shape (B, L_img, clip_img_dim), and a te

Computes the attend-and-excite loss using the maximum attention value for each token.

generate a hook method for a specific block and facet.

Optionally removes offloading in case the pipeline has been already sequentially offloaded to CPU. Args: _pipeline (`Dif

Format refs and preds for sacrebleu corpus calculation. It is very particular

r""" Splits a flattened embedding x of shape (B, C * H * W + clip_img_dim) into two tensors of shape (B, C, H, W) and (B, 1, clip_img_

r""" Splits a flattened embedding x of shape (B, C * H * W + clip_img_dim + text_seq_len * text_dim] into (img_vae, img_clip, text) wh

Unwraps a model.

:returns: {str: [metric_score] -> float} A dictionary where keys are the names of submetrics and values are functions

Perform backward process given list of time steps Args: latents: Latents at time timesteps[0].

Perform backward process given list of time steps. Args: latents: Latents at time timesteps[0].

Batched version of the `step` function, to be able to reverse the SDE for multiple samples/timesteps at once. Also, does not add any

Create a mask. Args: image (`PIL.Image.Image`): The image input, should be a PIL image. Returns

Create a filtering pipeline.

Calculate patch indices and corresponding loss. Args: - kps_1, kps_2: Keypoints for the two images. - desc_1, desc_2: Descriptor

Checks if the LoRA layers are correctly set with peft

Context manager that checks out a given commit when entered, but gets back to the reference it was at on exit. Args: repo (`git.Repo

Cleans the table of content of the model documentation by removing duplicates and sorting models alphabetically.

Takes a state dict and a config, and returns a converted checkpoint.

Convert the model output to the corresponding type the data_prediction/noise_prediction algorithm needs. Noise_prediction is designed

r""" Converts a state dict to the PEFT format The state dict can be from previous diffusers format (`OLD_DIFFUSERS`), or new diffusers format

Maps the keys in a UniDiffuser transformer block (`Block`) to the keys in a diffusers transformer block (`UTransformerBlock`/`UniDiffuserBloc

:param model_type: a string specifying which model to load. [dino_vits8 | dino_vits16 | dino_vitb8 | dino_vitb16 |

Creates a config for the diffusers based on the config of the LDM model.

Args: Deletes the LoRA layers of `adapter_name` for the unet and text-encoder(s). adapter_names (`Union[List[str], str]`)

helper function to detect language(s) of prompt

r""" Disable all adapters attached to the model and fallback to inference with the base model only. If you are not familiar with adap

r""" Disable sliced attention computation. If `enable_attention_slicing` was previously called, attention is computed in one step.

Disables the FreeInit mechanism if enabled.

One step for the first-order DPMSolver (equivalent to DDIM). Args: model_output (`torch.Tensor`): The di

Sets the attention processor to use [feed forward chunking](https://huggingface.co/blog/reformer#2-chunked-feed-forward-layers).

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

Reverse the denoising step process to recover a noisy image from the generated image. Args: images (`List[PIL Image]`):

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