Hongyi Jin
PhD Candidate
Pittsburgh, Pennsylvania, United States
Summary
🤩
Rockstar🎓
Top SchoolHongyi Jin is a Carnegie Mellon PhD candidate and software engineer with seven years of experience building high-performance ML systems and compilers. He contributes to flagship open-source projects like Apache TVM and MLC-LLM, focusing on tensor compilation, scheduling primitives, loop-fusion correctness, and WebGPU-backed LLM deployment. His work spans back-end compiler engineering and model-inference optimization, including operator fusion, quantization fixes, and weight compression to accelerate prefill and decode stages. Based in Pittsburgh, he blends academic rigor with practical engineering, shipping optimizations that bridge research prototypes to browser and accelerator deployments. Less obvious: he routinely tackles subtle correctness bugs in compiler passes, demonstrating deep understanding of dependencies that prevents hard-to-detect runtime regressions.
7 years of coding experience
Carnegie Mellon University
Github Skills (18)
tvm10
code-optimization10
language-model10
compilation10
python10
machine-learning10
llm10
deeplearning-ai10
compiler-compiler10
deep-learning10
webgpu10
compile10
compiler10
quantization10
cuda9
Programming languages (10)
Github contributions (5)
mlc-ai/mlc-llm
Apr 2023 - Jan 2025
Universal LLM Deployment Engine with ML Compilation
Role in this project:
ML Engineer
ML EngineerContributions:30 reviews, 35 PRs, 29 pushes in 1 year 9 months
Contributions summary:Hongyi primarily contributed to the optimization and development of the ML compilation engine for LLMs. Their work involved integrating WebGPU support by adding system libraries, optimizing the model's inference by fusing operations, and improving performance through weight compression techniques. The user also addressed quantization computations and improved the prefill and decode stages to optimize overall model performance. They also worked on improving the system's ability to define new model architectures.
language-modelllmmachine-learning-compilationtvm
mlc-ai/web-llm
Apr 2023 - May 2023
High-performance In-browser LLM Inference Engine
Role in this project:ML Engineer
ML EngineerContributions:9 reviews, 18 PRs, 52 pushes in 1 month
Contributions summary:Hongyi primarily focused on modifying the `web_llm/transform/dispatch_tir_operator.py` file, specifically adjusting vocabulary sizes to support different versions of the Vicuna language model. This included updating manual schedules and potentially optimizing the model's performance. The changes indicate a deep understanding of the model's architecture and the underlying TVM framework used for deployment in a web browser. They also contributed to the optimization of softmax operations.
chatgptdeep-learninglanguage-modelllmtvm
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.