Marek Wawrzos
Senior Deep Learning Algorithms Engineer at NVIDIA
Czechowice-Dziedzice, Silesian Voivodeship, Poland
Summary
🤩
Rockstar🎓
Top SchoolMarek Wawrzos is a Senior Deep Learning Algorithms Engineer with 11 years of experience building and optimizing production-grade ML systems, currently at NVIDIA in Poland. He specializes in accelerating state-of-the-art deep learning models and has practical impact on projects like DeepLearningExamples and the MLPerf benchmarks, where he contributed optimizations for training pipelines and an RNN-Transducer speech recognition benchmark. Marek combines low-level implementation skills (C++, performance tuning) with deep learning algorithm expertise, applying techniques such as epoch-based evaluation, warm-up, LR decay, initialization tweaks and gradient clipping to improve real-world training. His background includes software development roles at Nokia, Flowbox and IBM, and an MSc/BSc in Computer Science from AGH, reflecting a solid engineering foundation. Notably, he works at the intersection of research and engineering—translating academic models into efficient, reproducible benchmarks used by the wider ML community.
11 years of coding experience
5 years of employment as a software developer
BSc, Computer Science, 4.5, BSc, Computer Science, 4.5 at AGH University of KrakowAGH University
Polish, English
Stackoverflow
1reputation
0reached
0answers
0questions
Github Skills (12)
pytorch10
machine-learning10
python10
model-optimization10
benchmarking9
benchmark9
deep-learning9
rnn-model8
n8
dockers7
docker7
cuda6
Programming languages (6)
Github contributions (5)
mlcommons/training
Nov 2019 - Aug 2021
Reference implementations of MLPerf™ training benchmarks
Role in this project:
ML Engineer
ML EngineerContributions:15 reviews, 6 commits, 12 PRs in 1 year 9 months
Contributions summary:Marek focused on optimizing and extending the MLPerf training benchmarks within the repository. They implemented epoch-based evaluation, warm-up, and learning rate decay strategies for single-stage detectors. Furthermore, the user added and refined an RNN-Transducer speech recognition benchmark, including modifications to model architecture, data preprocessing, and evaluation scripts. The contributions included improving and updating the training pipeline, and refining specific aspects of model training such as initialization and gradient clipping.
implementationsbenchmarkingdeep-learningmlperfmachine-learning
mwawrzos/training
Sep 2019 - Jul 2021
Reference implementations of training benchmarks
Contributions:9 PRs, 49 pushes, 26 branches in 1 year 9 months
implementationsbenchmarkingdeep-learningmachine-learningtraining
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.