Hongzhen Liu

资深技术专家 @ 理想汽车

About

大模型/自动驾驶/搜推广/高性能计算/异构计算/AI编译优化。 • AI Engineering for LLM & VLM. • AI Engineering for Automatic Drive. • AI Engineering for Recommendation System. • AI Compiler & General Computing Library. • Heterogeneous & High Performance Computing. • AI Infrastructure for DL/ML Inference & Training.

Country

China

City

Pudong

Industry

Internet

Skill

LLVM, Deep Learning, Parallel Computing, Software pipelining, Performance optimization, performance quantification, computer system architecture, OpenMP, Winograd algorithm, C++, Python, Linux, Android, Hardware Design

Experience

Director & Senior Staff Engineer

斑马网络技术有限公司

2021-12 - 2025-10 · 3 yrs 11 mos

* LLM Infra（Framework & Library） * AI Compiler (CPU/GPU/CDSP/VDSP) * Heterogenous Scheduling Framework * Heterogenous Computing Library

Staff Engineer

Ant Group

2020-12 - 2021-11 · 1 yr

Team leader of DL/ML high performance computing team. (PyTorch & Tensorflow) Created a Deep Learning optimization framework from scratch, accumulatively saved 6W+ resources (~30M RMB/Year) for Alipay online business in 2021, promoted four recommendation service upgrades.

Deep Learning Software Engineer

Intel

2017-12 - 2020-12 · 3 yrs 1 mo

Shanghai City, China

* Initiator of Intel Deep Learning Performance team Led MLPerf v0.7 training (2020-06) & MLPerf v0.7 inference (2020-09). * PyTorch CPU & Intel PyTorch Extension Tech Lead Led team to develop and optimize PyTorch community for Xeon processors. co-led team to create IPEX from scratch for high performance and product landing. * Deep Learning Inference Library Designed and implemented a high performance DNN library from scratch, featuring data level concurrency programming、cache locality modeling and algorithm level innovations. Contributed MLPerf benchmark and commercial product deployment. * Intel iDeep Designed a flexible library to maximize Intel oneDNN superiority and boost DL framework performance on Intel Xeon series processors. Has been adopted by serveral mainstream deep learning frameworks.(e.g. PyTorch、Caffe2、Chainer).

Senior Software Engineer

Intel

2015-11 - 2017-12 · 2 yrs 2 mos

Shanghai City, China

Intel Houdini Project: Core Developer & Module Maintainer. Dynamic Binary Translator, which vastly expands ARM-based Android applications to run natively on Intel abundant products from mobile/tablet/Vehicle-mounted intelligent device to PC and cloud gaming.