3 ms·Medusa: Framework for Accelerating LLM Generation with Multiple Decoding Heads2 points by azeirah 3y agohustwindmaple1 3y agoVery cool work