Peter Yeh
Principal Member Of Technical Staff at AMD
Mountain View, California, United States
Summary
🤩
Rockstar🎓
Top SchoolPeter Yeh is a Principal Member of Technical Staff based in Mountain View with eight years focused on accelerating generative AI and LLM performance, currently optimizing PyTorch and end-to-end AI workloads at AMD. He has deep practical expertise shipping ROCm/MIOpen support across TVM and PyTorch—adding new AMD GPU architectures, fp16/FP8 datatype support, and performance fixes that directly improve GPU inference and training. Prior roles include leading AI accelerator development for autonomous systems and edge inference optimization at Argo AI, reflecting a consistent focus on hardware-software co-design for high-throughput ML. Peter blends low-level GPU compiler and backend engineering with product-minded system architecture, and his open-source contributions to widely used projects like TVM and PyTorch highlight impact beyond internal teams. He holds an MS in Electrical Engineering from USC and brings a track record of turning research-level capabilities into production-grade accelerator support.
8 years of coding experience
13 years of employment as a software developer
Master of Science - MS Electrical and Electronics Engineering, Master of Science - MS Electrical and Electronics Engineering at University of Southern CaliforniaBachelor of Science - BS Electrical Engineering, Bachelor of Science - BS Electrical Engineering at National Taiwan University
Github Skills (17)
pytorch10
python10
machine-learning10
deeplearning-ai10
roc10
compiler-compiler10
deep-learning10
gpu10
compiler10
amdgpu10
c-language9
opencl9
cprogramming-language9
tensor8
cuda7
Programming languages (5)
Github contributions (5)
apache/tvm
Aug 2019 - Dec 2019
Open deep learning compiler stack for cpu, gpu and specialized accelerators
Role in this project:
ML Engineer
ML EngineerContributions:11 commits, 14 PRs, 33 comments in 3 months
Contributions summary:Peter contributed to the ROCm and MIOpen backend of the TVM compiler, which focuses on optimizing deep learning workloads on AMD GPUs. They implemented support for new AMD GPU architectures (gfx906, gfx1010), enabled features like miopen transpose convolution and fp16 support, and added group convolution capabilities. Furthermore, they fixed compilation issues related to the ROCm backend and implemented testing.
metalvulkancompilertensoropencl
pytorch/pytorch
May 2018 - Jun 2018
Tensors and Dynamic neural networks in Python with strong GPU acceleration
Role in this project:ML Engineer
ML EngineerContributions:11 reviews, 2 commits, 37 PRs in 21 days
Contributions summary:Peter primarily focused on improving the ROCm (AMD GPU) support within the PyTorch framework. Their contributions included fixing issues related to int8 matrix multiplication (int_mm) integration with hipblasLT and enabling tests on ROCm. They also worked on enabling faster load/save operations for Fused_SGD and added OCP FP8 support for new GPUs, adding support for new architectures, and updating data type handling.
pythongpu-accelerationdeep-learninggpunumpy
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.