Medusa: Framework for Accelerating LLM Generation with Multiple Decoding Headssites.google.com2 points·azeirah··1 commentOpen articleSaveView on HN