IPU Diffusers

A few 🤗 diffusers have been adapted for use with IPUs. The available IPU diffusers are:

API reference

IPUStableDiffusionPipeline class

This class is based on the 🤗 StableDiffusionPipeline pipeline.

Run the following notebooks to see how this pipeline is used:

IPUStableDiffusionImg2ImgPipeline class

This class is based on the 🤗 StableDiffusionImg2ImgPipeline pipeline.

Run the Stable Diffusion Image-to-Image Generation on IPUs notebook for an example of how this pipeline is used.

class optimum.graphcore.diffusers.IPUStableDiffusionImg2ImgPipeline

< >

( vae text_encoder tokenizer unet scheduler safety_checker feature_extractor requires_safety_checker = True n_ipu = 4 num_prompts = 1 num_images_per_prompt = 1 unet_ipu_config = None text_encoder_ipu_config = None vae_ipu_config = None safety_checker_ipu_config = None common_ipu_config_kwargs = None )

IPUStableDiffusionInpaintPipeline class

Based on the 🤗 StableDiffusionInpaintPipeline pipeline.

Run the Text-guided Inpainting on IPUs using Stable Diffusion - Inference notebook for an example of how this pipeline is used.

class optimum.graphcore.diffusers.IPUStableDiffusionInpaintPipeline

< >

( vae text_encoder tokenizer unet scheduler safety_checker feature_extractor requires_safety_checker = True n_ipu = 4 num_prompts = 1 num_images_per_prompt = 1 unet_ipu_config = None text_encoder_ipu_config = None vae_ipu_config = None safety_checker_ipu_config = None common_ipu_config_kwargs = None )