Hugging Face and AMD partner on accelerating transformer models

AI Impact Summary

Hugging Face and AMD are partnering to accelerate transformer model performance across CPU and GPU platforms, addressing concerns around compute-intensive workloads like pre-training and inference. This collaboration focuses initially on AMD’s Instinct MI2xx/MI3xx GPUs and Ryzen/EPYC CPUs, alongside the Alveo V70 accelerator, offering potential performance gains over competing hardware, particularly for models like BERT and GPT-2. This partnership will significantly broaden the hardware options available to Hugging Face users, driving down costs and improving performance.