Fine-tuning Stable Diffusion models on Intel CPUs — leveraging Sapphire Rapids AMX

AI Impact Summary

Fine-tuning Stable Diffusion models on Intel CPUs using Sapphire Rapids processors with AMX hardware acceleration is now possible. This enables significantly faster training times compared to previous GPU-only approaches, leveraging the new Intel hardware extensions. The process involves utilizing the Diffusers library, oneCCL for communication, and IPEX to optimize the U-Net and VAE models for the AMX architecture, alongside a distributed training setup across a cluster of Intel Xeon Sapphire Rapids CPUs. This represents a key shift in accessibility for generative AI model development.