Reza Yazdani
Senior Software Engineer at Snowflake
Vancouver, British Columbia, Canada
Summary
🤩
Rockstar🎓
Top SchoolReza Yazdani is a Senior Software Engineer and PhD-trained computer architect specializing in AI performance engineering across GPU, CPU, FPGA and ASIC platforms. With six years of industry experience and a research foundation from Universitat Politècnica de Catalunya, he bridges low-level accelerator design (including Viterbi-beam search and LSTM hardware) with large-scale ML systems engineering. At Microsoft and Snowflake he focused on machine-learning optimization and research-grade SDE work, and he contributes to prominent open-source projects like DeepSpeed—adding transformer kernel optimizations, quantized inference support, and CPUAdam for zero-offload. He excels at squeezing efficiency from both algorithms and hardware, routinely tackling dynamic sequence lengths, mixed-precision training, and model-parallelism challenges. Based in Vancouver, he combines academic rigor with production impact, often revealing order-of-magnitude energy or size savings that aren’t obvious from high-level benchmarks.
6 years of coding experience
15 years of employment as a software developer
Bachelor of Computer hardware Engineering Computer hardware Engineering, Bachelor of Computer hardware Engineering Computer hardware Engineering at Sheikh Bahaei UniversityMaster of Computer Architecture Computer Architecture, Master of Computer Architecture Computer Architecture at University of TehranUPC Universitat Politècnica de Catalunya
English, Persian
Github Skills (13)
cuda10
transformer-models10
pytorch10
machine-learning10
deepspeed10
deep-learning10
inference10
gpu10
parallelization10
optimization9
optimisation9
quantization9
python8
Programming languages (2)
Github contributions (5)
deepspeedai/DeepSpeed
Jun 2020 - Jan 2023
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:215 reviews, 498 commits, 172 PRs in 2 years 7 months
Contributions summary:Reza's commits primarily focused on modifying and extending transformer kernels within the DeepSpeed library, specifically addressing issues related to configurable intermediate sizes, dynamic sequence length support, and support for larger hidden dimensions. They added support for the CPUAdam optimizer for zero-offload and worked on fixing convergence issues related to it. Furthermore, the user contributed to the development of quantized inference kernels and support for BigScience-BLOOM architecture.
billion-parametersfinetuningtrainingmixture-of-expertszero
deepspeedai/DeepSpeedExamples
May 2020 - Jan 2023
Example models using DeepSpeed
Role in this project:MLOps Engineer
MLOps EngineerContributions:17 reviews, 55 commits, 18 PRs in 2 years 8 months
Contributions summary:Reza primarily contributed to the integration and optimization of DeepSpeed, a deep learning optimization library, within the repository's example models. Their work involved modifying scripts and code to leverage DeepSpeed's features, including transformer kernel acceleration, mixed-precision training, and model parallelism for improved performance and efficiency. Key contributions include adding and configuring DeepSpeed for various tasks and model architectures, as well as fixing inference tests related to the latest DeepSpeed changes. The user's changes focused on integrating and utilizing DeepSpeed effectively in different training and inference scenarios.
deep-learningpytorchdeepspeed
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.