Hugging Face Transformers gains out-of-the-box AMD Instinct GPU support via Text Generation Inference
AI Impact Summary
AMD and Hugging Face have integrated out-of-the-box support for running Hugging Face Transformers models on AMD Instinct GPUs, with production-ready Text Generation Inference (TGI) support for MI210/MI250 and no code changes required. The release highlights performance parity with Nvidia in TGI workflows and demonstrates that a single MI250 card can expose two PyTorch devices, enabling effective tensor parallelism for large models like Llama 34B and 70B. This reduces the migration barrier for teams evaluating AMD hardware for large-scale inference and signals a broader AMD roadmap with MI300 availability.
Affected Systems
- Date
- Date not specified
- Change type
- capability
- Severity
- info