Diffusers adds JAX/Flax support

·2022.10.13 09:00

Key point

Hugging Face's Diffusers library now supports Flax, enabling fast inference of Stable Diffusion on TPU environments.

Details

Starting with version 0.5.1, Hugging Face's Diffusers library supports Flax. This enables very fast inference of Stable Diffusion on Google TPU environments (Colab, Kaggle, GCP, etc.).

JAX can utilize all 8 accelerators on a TPU server in parallel, which is advantageous for maximizing hardware performance.

Key features include:

Model loading via FlaxStableDiffusionPipeline
bfloat16 data type support for TPU optimization
Model weights provided via the Hugging Face Hub, in compliance with the CreativeML OpenRAIL-M license

huggingface tpu jax diffusers stable-diffusion flax

Read originalShare on Threads

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.