Mengshiun Yu
Technical Manager at MediaTek
Pittsburgh, Pennsylvania, Taiwan
Summary
🤩
Rockstar🎓
Top SchoolMengshiun Yu is a technical manager and PhD candidate in computer science with deep expertise in compiler optimization and performance tuning for neural network accelerators and computer vision applications. He brings hands-on experience building compilers, JITs, and executable formats for NPUs from his Kneron work and has contributed backend and hardware-specific optimizations to high-profile open-source projects like Apache TVM and MLC-LLM. As a visiting scholar at CMU he focused on optimizing small language models for edge devices, and he now leads technical efforts at MediaTek from Pittsburgh. Comfortable across C/C++, Python, embedded Linux and firmware, he blends research rigor with production-grade engineering and has a track record of shipping silicon-focused toolchains and runtime improvements. A less obvious strength is his cross-domain fluency—spanning low-level compiler internals to mobile/Android app integration—which lets him bridge chip, compiler and application layers effectively.
2 years of coding experience
12 years of employment as a software developer
Master's degree, Electrical and Electronics Engineering, Master's degree, Electrical and Electronics Engineering at National Chung Cheng UniversityBachelor's degree, Electrical and Electronics Engineering, Bachelor's degree, Electrical and Electronics Engineering at National Chin-Yi University of TechnologyDoctor of Philosophy (Ph.D.), Computer Science, Doctor of Philosophy (Ph.D.), Computer Science at National Tsing Hua University
Github Skills (22)
opencl10
tvm10
performance-monitor10
python10
cpu10
deeplearning-ai10
compiler-compiler10
performance-analysis10
deep-learning10
gpu10
compiler10
c-language9
pytorch9
machine-learning9
tensorflow9
Programming languages (3)
Github contributions (5)
mlc-ai/mlc-llm
Mar 2024 - Mar 2025
Universal LLM Deployment Engine with ML Compilation
Role in this project:
ML Engineer
ML EngineerContributions:4 reviews, 32 PRs, 5 pushes in 11 months
Contributions summary:Mengshiun's contributions primarily involve integrating new LLM models, specifically the Phi-3 series, into the MLC-LLM deployment engine. They introduced the microsoft/Phi-3 model, including its configuration and model implementation. The user also added support for Phi-3 vision, encompassing image preprocessing and the integration of the CLIP vision model. Furthermore, they made several bug fixes and enhancements to the existing Android application, switching the application to use MLCEngine and updating the default model list.
language-modelllmmachine-learning-compilationtvm
apache/tvm
Apr 2024 - Feb 2025
Open deep learning compiler stack for cpu, gpu and specialized accelerators
Role in this project:Back-end & Performance Engineer
Back-end & Performance EngineerContributions:5 reviews, 7 PRs, 7 comments in 10 months
Contributions summary:Mengshiun made significant contributions to the TVM compiler stack, focusing on adding support for specific hardware backends and optimizing existing functionalities. This involved adding an OpenCL device for target detection and introducing an NNAPI backend for BYOC, enabling execution on custom accelerators. Further, they refactored the `_attention_sequence_prefill` function and added KV Cache support for the CPU runtime. These changes involved modifying code in multiple languages like C++ and Python and optimizing for specific hardware architectures.
metalvulkancompilertensoropencl
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.