Medusa: Accelerating Serverless LLM Inference with Materialization [ASPLOS'25] - View it on GitHub