Yiwei Li
Senior Member Of Technical Staff at AMD
Beijing, United States
Summary
👤
Senior🎓
Top SchoolYiwei Li is a Senior Member of Technical Staff with eight years of experience building deep learning frameworks and compilers for AI accelerators, currently advancing DL compiler work at AMD from Beijing. His background spans framework-to-hardware model bringup, MLIR-based IR design and lowering, kernel matching/configuration, and graph optimizations developed during multi-year roles at Cerebras and Huawei (MindSpore). He has hands-on experience enabling large transformer families (BERT, T5, GPT variants) on novel accelerators and contributed practical 3D perception model integration and inference tooling in the ICCV 2023 SurroundOcc repository. Comfortable bridging research and production, he combines PhD-level training with production compiler engineering to translate TF/PyTorch graphs into optimized accelerator kernels. Notably, his work emphasizes pragmatic graph rewriting and kernel mapping techniques that speed model bringup across diverse AI hardware.
8 years of coding experience
10 years of employment as a software developer
Bachelor's degree, Computer Engineering, Bachelor's degree, Computer Engineering at Anhui UniversityMaster's Degree, Computer Science, Master's Degree, Computer Science at Western University
Github Skills (7)
computer-vision10
pytorch10
data-pipeline10
data-pipelines10
python10
image-processing9
cuda8
Programming languages (2)
Github contributions (5)
weiyithu/SurroundOcc
Feb 2023 - Mar 2023
[ICCV 2023] SurroundOcc: Multi-camera 3D Occupancy Prediction for Autonomous Driving
Role in this project:
ML Engineer
ML EngineerContributions:34 commits, 47 pushes, 1 branch in 22 days
Contributions summary:Yiwei's contributions primarily focus on the development and integration of 3D occupancy prediction models within the SurroundOcc framework. This includes the implementation of data processing pipelines, such as image normalization and photometric distortion. Furthermore, the user added inference code for the model and made modifications to support semantic segmentation and visualization of the occupancy predictions. These changes suggest a focus on end-to-end model functionality, from data loading and preprocessing to model inference and result visualization within a 3D perception context.
3d-reconstruction3d-semantic-segmentationoccupancyoccupancy-prediction
weiyithu/SurroundDepth
Apr 2022 - Oct 2022
[CoRL 2022] SurroundDepth: Entangling Surrounding Views for Self-Supervised Multi-Camera Depth Estimation
Contributions:16 commits, 13 pushes, 1 branch in 6 months
kitticameraself-supervised-learningcorlstereo
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.