Stas Bekman
Senior Software Engineer (AI ML) at StasoSphere.com
Nanaimo, British Columbia, Canada
Summary
🤩
Rockstar🎓
Top SchoolStas Bekman is a senior software engineer specializing in designing high-performance, highly scalable training and inference systems for large language models, with 24 years of engineering experience. He has led massive LLM trainings—most notably as lead engineer on BLOOM-176B and contributor to IDEFICS-80B—working across DeepSpeed, PyTorch and Hugging Face tooling to make multi-hundred-GPU projects practical. An active open-source contributor and author, he improves core ML infra (Transformers, DeepSpeed, accelerate, PyTorch) and publishes an open "Machine Learning Engineering" book that codifies his ops-to-research expertise. Beyond ML, his background spans long-term open-source leadership (mod_perl/Apache), technical authorship, entrepreneurship and angel investing, combining systems-level rigor with product and community instincts. Based in Nanaimo, BC, he consults on hard scaling/debug problems and still prefers email for outreach, reflecting a pragmatic, hands-on approach to complex ML challenges.
25 years of coding experience
25 years of employment as a software developer
B.Sc. Electrical Engineering, B.Sc. Electrical Engineering at Technion - Israel Institute of TechnologyLevel Master/Teacher Usui Reiki, Level Master/Teacher Usui Reiki at Reiki Master/TeacherFrench Proficiency Intensive french, French Proficiency Intensive french at McGill University
Stackoverflow
5,766reputation
1.3mreached
32answers
5questions
google-colaboratory
top-5%
machine-learning
top-5%
debugging
top-5%
github
top-5%
github-api
top-5%
python
top-5%
Github Skills (63)
pytorch10
javascript10
distributed-training10
integrate10
slurm10
python10
scripting10
css10
machine-learning10
bash10
build-system10
makefile10
integrations10
cicd10
transformer-models10
Programming languages (20)
Github contributions (5)
bigscience-workshop/bigscience
May 2021 - Aug 2022
Central place for the engineering/scaling WG: documentation, SLURM scripts and logs, compute environment and data.
Role in this project:
DevOps Engineer & ML Engineer
DevOps Engineer & ML EngineerContributions:37 reviews, 732 commits, 7 PRs in 1 year 4 months
Contributions summary:Stas's contributions primarily revolve around setting up and configuring the computational environment for large language model training and related benchmarking. They've implemented and tested bandwidth benchmarks, and they've integrated a pre-trained model from a different repository. The user has also contributed to setting up the continuous integration process through adding tools for hub syncing and email alerts related to job status on Jean Zay, as well as created tools to monitor the status of Slurm jobs.
slurmnlpmachine-learningtrainingmodels
bigscience-workshop/Megatron-DeepSpeed
May 2021 - Sep 2022
Ongoing research training transformer language models at scale, including: BERT & GPT-2
Role in this project:ML Engineer
ML EngineerContributions:291 reviews, 334 commits, 176 PRs in 1 year 4 months
Contributions summary:Stas primarily contributed to the development and refinement of training scripts and related functionalities for transformer language models. Their work includes the addition and modification of scripts for training on single nodes with DeepSpeed, demonstrating an understanding of distributed training. The user also addressed and reverted issues with training sample calculations, and incorporated a merging datasets capability and a kill switch, enhancing the project's robustness and flexibility.
bertgptlanguage-model
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.