Rick Ho
Performance Expert at ByteDance
San Jose, California, United States
Summary
🤩
Rockstar🎓
Top SchoolRick Ho is a performance-focused software engineer and PhD candidate from Tsinghua University with 11 years of experience optimizing large-scale ML and systems workloads. Based in San Jose and currently a Performance Expert at ByteDance, he specializes in low-level CUDA optimization and high-throughput model infrastructure, notably implementing a fast Mixture-of-Experts layer for PyTorch with custom batched GEMM, scatter/gather kernels, and a global exchange mechanism. He blends rigorous academic training with production-oriented engineering to squeeze latency and throughput improvements out of GPU stacks. Known for building testing and benchmarking infrastructure around his optimizations, he brings a measurable, benchmark-driven approach to performance problems that often uncovers non-obvious system bottlenecks.
11 years of coding experience
Doctor of Philosophy - PhD, Doctor of Philosophy - PhD at Tsinghua University
Github Skills (8)
kernel10
matrix-multiplication10
cuda10
pytorch10
machine-learning10
deep-learning10
parallel-computing10
performance-optimization9
Programming languages (9)
Github contributions (5)
laekov/fastmoe
Dec 2020 - Dec 2022
A fast MoE impl for PyTorch
Role in this project:
ML Engineer
ML EngineerContributions:7 releases, 57 reviews, 300 commits in 2 years
Contributions summary:Rick's contributions primarily involve implementing and optimizing a fast MoE (Mixture of Experts) implementation for PyTorch. Their work includes creating CUDA kernels for batched matrix multiplication, designing a global exchange mechanism, and building kernels for optimized scatter and gather operations. They have added code for performance testing and incorporated a testing infrastructure within the repository to validate and benchmark their CUDA-based MoE layer.
pytorch
haoxizhong/TUOJ
Sep 2016 - Nov 2016
Let's discover a new world. — Edit
Contributions:142 commits, 3 PRs, 94 pushes in 2 months
discovernew-worldlet
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.