From the source
Hugging Face announces first-class integration of
AMD Instinct MI300 GPU servers into the Hugging Face Platform, enabling seamless deployment of AI models on AMD hardware without code changes.
The collaboration includes CI/CD testing on MI300 GPUs for Transformers and text-generation-inference, and performance optimizations for inference workloads using techniques like Flash Attention v2, Paged Attention, GPTQ/AWQ compression, and AMD's TunableOp.
