Mark Palatucci reposted this
Stable Diffusion XL with Core ML on Apple Silicon! The diffusion model grew 3x in size so we are releasing a new mixed-bit model compression technique that yields as low as 3 bit precision on average (5.3x compression vs 16-bits) with minimal output difference. Check out the Hugging Face blog post for details. Had lots of fun working with Pedro Cuenca on this! If this work is interesting to you, consider joining one of our teams that focus on efficient inference: Blog post: https://lnkd.in/gGNhX4Bc Code: https://lnkd.in/gtfvhJbA Open Position: https://t.co/OlCwZ10WAe #SDXL #stablediffusion #generativeai #genai #deeplearning #machinelearning #apple