During March 2026, Zihao An contributed to the jeejeelee/vllm repository by developing comprehensive documentation for Parallel Draft Models (PARD) in speculative decoding. Focusing on AI model optimization and leveraging machine learning expertise, Zihao provided detailed offline and online mode examples to illustrate how vLLM performance can be enhanced through speculative decoding techniques. The documentation, written in Python, included concrete usage guidance to streamline onboarding and benchmarking for new users. Zihao ensured the quality and governance of the work by following proper sign-off procedures and contributor attribution, delivering a thorough and practical resource for the vLLM developer community.