Pengxin Yuan
Graphic Software Engineer at Intel
Minhang District, Shanghai, China
Summary
👤
Senior🎓
Top SchoolPengxin Yuan is a Graphic Software Engineer at Intel with nine years of experience optimizing deep learning performance across CPUs and GPUs. He is a main contributor to Intel’s Low Precision Optimization Tool and has driven low-bit LLM quantization and model compression work in prominent open-source projects like intel/neural-compressor. His background blends systems-level media/video codec engineering and DL benchmarking—having improved Intel Media SDK decoders and built a 200+ model benchmarking suite across TensorFlow and PyTorch. Pengxin’s contributions often focus on practical performance fixes and framework integrations (MXNet, ONNX Runtime), making low-precision inference more reliable in production. Based in Shanghai, he holds a master’s in software engineering and pairs rigorous perf-analysis skills with hands-on implementation of quantization and sparsity techniques. An understated strength is his ability to bridge decoding/codec internals with ML inference optimization, yielding cross-domain performance gains.
9 years of coding experience
3 years of employment as a software developer
硕士, 软件工程, 硕士, 软件工程 at 华东师范大学学士, 电子信息工程, 学士, 电子信息工程 at 中北大学
Github Skills (13)
quantization10
mxnet10
intel-media-sdk10
c-language10
cprogramming-language10
large-language-models10
model-optimization10
decoding10
hevc10
mxf9
av1-codec9
auto-tuning8
sparse7
Programming languages (4)
Github contributions (5)
intel/neural-compressor
Jun 2020 - Jan 2021
SOTA low-bit LLM quantization (INT8/FP8/INT4/FP4/NF4) & sparsity; leading model compression techniques on TensorFlow, PyTorch, and ONNX Runtime
Role in this project:
ML Engineer
ML EngineerContributions:67 commits, 1 comment in 7 months
Contributions summary:Pengxin's contributions center on enhancing the `intel/neural-compressor` repository, which focuses on low-bit LLM quantization. Their commits primarily involve enabling and fixing the integration of the iLiT (Intel Low Precision Optimization Tool) with the MXNet framework, as well as addressing model-specific issues. These changes include fixing example code, enabling model tuning, and inspecting tensor with the dequantize API, showcasing a focus on improving model compression techniques and performance.
knowledge-distillationauto-tuningcompressorsparsityintel
Intel-Media-SDK/MediaSDK
Nov 2021 - Aug 2022
The Intel® Media SDK
Role in this project:Back-end Developer
Back-end DeveloperContributions:9 reviews, 8 commits, 9 PRs in 8 months
Contributions summary:Pengxin primarily contributed to the Intel Media SDK, focusing on video decoding functionalities. Their work involved fixing bugs related to VC1 and AV1 decoding, specifically addressing issues arising from small bitstream sizes and frame type handling. They also made enhancements to HEVC decoding, including parsing HDR SEI data and integrating frame rate parameters from VideoParamSet. Furthermore, the user addressed memory leaks in the simple_5_transcode application.
intel-media-sdkintelsdkvideomfx
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.