Arash Bakhtiari
Member Of Technical Staff at Microsoft AI
Redmond, Washington, United States
Summary
🤩
Rockstar🎓
Top SchoolArash Bakhtiari is a research-focused software engineer with 13 years of experience in high-performance computing and deep learning, now serving as a Member of Technical Staff at Microsoft AI in Redmond. He holds a PhD in High Performance Computing and Parallel Algorithms from the Institute for Advanced Study at TUM and is fluent in C++, Python, CUDA/OpenCL, MPI/OpenMP and low-level ARM/assembly optimizations. His background spans edge AI and production inference—he led development of Larq Compute Engine to accelerate binarized neural networks by orders of magnitude and brought up and quantized major vision models for Blaize’s graph-native GSP. He’s also an active open-source contributor to DeepSpeed, adding OPT model support, custom MLP activations and kernel improvements that bridged research models into robust distributed training and inference. Arash combines deep systems-level performance tuning with practical deployment experience across CPUs, GPUs and specialized edge processors.
13 years of coding experience
13 years of employment as a software developer
Technical University of Munich, Institute for Advanced Study
German, Persian, English
Stackoverflow
1reputation
0reached
0answers
0questions
Github Skills (11)
transformers10
pytorch10
machine-learning10
deep-learning10
inference10
gpu10
parallelization10
python10
compression9
zeroth7
zero-code7
Programming languages (2)
Github contributions (5)
deepspeedai/DeepSpeed
Aug 2022 - Oct 2022
DeepSpeed is a deep learning optimization library that makes distributed training and inference easy, efficient, and effective.
Role in this project:
ML Engineer
ML EngineerContributions:11 reviews, 30 commits, 17 PRs in 2 months
Contributions summary:Arash primarily contributed to adding and refactoring support for OPT models within the DeepSpeed library. Their work involved implementing custom MLP activation functions, modifying inference APIs, and adding unit tests for OPT models. Additionally, the user made improvements to the residual add and bias kernels, and address compatibility issues with older versions of the transformers library.
billion-parametersfinetuningtrainingmixture-of-expertszero
arashb/turbulent_lbm_multigpu
Apr 2013 - Mar 2020
Contributions:214 commits, 1 push in 6 years 11 months
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.