Publications
论文发表- "SD-HC: Speculative Decoding on a Heterogeneous Computing System." MLSys 2026, Young Professionals Symposium (YPS) Program.
- "Breaking the Draft Barrier." EMNLP 2026.
- "Clover: Regressive Lightweight Speculative Decoding with Sequential Knowledge." arXiv, 2024. [PDF] [Code]
- "EAGLE-3 Speculative Decoding on AMD Instinct GPUs." AMD ROCm Tech Blog. [Blog]
Experience
学术经历
Jun 2026 – Present
Deep Learning Software Engineer Intern, NVIDIA
Santa Clara · GPU kernel optimization, TensorRT
Sep 2025 – Dec 2026
M.Eng. in Computer Engineering, University of Washington
Seattle · ML systems, LLM inference
Sep 2024 – Sep 2025
Software Engineer, Model Compression Team, AMD
Beijing · Quantization, Quark Model Optimizer
Jun 2024 – Sep 2024
Software Engineer Intern, SenseTime
Beijing · HPC / AI systems, OpenPPL, CUDA operators
Sep 2023 – Mar 2024
Software Engineer Intern, Baichuan AI
Beijing · LLM inference infrastructure, speculative decoding
Sep 2020 – Jun 2024
B.S. in Applied Physics, Harbin Institute of Technology
Honors School
Projects
研究项目Distributed Runtime and Automated Optimization for Multimodal Serving
UW SyFI Lab, May 2026 – present. Multimodal serving runtime; automated optimization of quantization and speculative decoding.
SD-HC: Speculative Decoding on a Heterogeneous Computing System
AMD, Feb 2025 – Aug 2025. CPU + NPU + GPU heterogeneous inference.
EAGLE-3 Quantization on a Speculative-Decoding Framework
SGLang × University of Waterloo, Aug 2025 – Feb 2026. Quantization for speculative decoding.
Blog
博客Hello, World
2026-10-29 · First post.