MingfeiĀ Ma
Deep Learning SW Engineer at Intel Corporation
Minhang District, Shanghai, China
Summary
š¤©
Rockstarš
Top School
Top expert inArtificial Intelligence and Machine Learning Technologies
Mingfei Ma is a Deep Learning Software Engineer with 9 years of experience specializing in CPU performance optimization for PyTorch, where he serves as the CPU performance tech lead and module maintainer. He has driven concrete speedups across quantization, mixed precision (float16/bfloat16), channels-last memory formats, oneDNN/oneMKL integrations, sparse and advanced indexing ops, and multi-core sort/collective routines. His contributions to the flagship pytorch/pytorch repo include efficient memory-copy strategies and benchmark-backed improvements to sampled_addmm, scatter_add, gather and other core CPU kernels. Based in Shanghai and backed by an engineering pedigree from Harbin Institute of Technology and The University of Hong Kong, he pairs deep low-level systems knowledge with practical customer-facing optimization work. Notably, he routinely translates microarchitecture-aware techniques into open-source patches that measurably improve real-world AI workloads.
9 years of coding experience
3 years of employment as a software developer
The University of Hong Kong (HKU)Bachelor of Engineering (BEng), Automation, Bachelor of Engineering (BEng), Automation at Harbin Institute of Technology
English
Github Skills (12)
tensorrt10
pytorch10
c-language10
tensor10
tensorflow10
cprogramming-language10
performance-optimization10
operation10
back-end-development10
multithreading10
quantization9
machine-learning8
Programming languages (6)
Github contributions (5)
pytorch/pytorch
Mar 2018 - Jan 2023
Tensors and Dynamic neural networks in Python with strong GPU acceleration
Role in this project:
Back-end Developer & Performance Engineer
Back-end Developer & Performance EngineerContributions:435 reviews, 1370 commits, 212 PRs in 4 years 10 months
Contributions summary:Mingfei contributed to optimizing the performance of the PyTorch library, specifically focusing on CPU-based operations. Their work involved implementing efficient memory copy operations within the quantized cat function. They also improved sort multi-core performance by adjusting the grain size and addressed a bug related to weight norm backward calculations. Furthermore, they optimized sampled_addmm, scatter_add, gather and other memory operations, significantly improving overall performance, as demonstrated by benchmark data.
pythongpu-accelerationdeep-learninggpunumpy
mingfeima/op_bench-py
Oct 2019 - Feb 2021
Contributions:41 commits, 38 pushes, 1 branch in 1 year 3 months
Find and Hire Top DevelopersWeāve 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.