Bartłomiej Wróblewski
Senior Software Engineer at Google DeepMind
Zurich, Zurich, Switzerland
Summary
👤
Senior🎓
Top SchoolBartłomiej Wróblewski is a Senior Software Engineer with eight years of experience optimizing deep learning performance across diverse accelerator architectures, currently working on AI hardware at Google DeepMind in Zurich. He has delivered substantial speedups on platforms including AMD RDNA/CDNA, Intel NNP‑T and GPUs, Graphcore IPU, Habana Gaudi and Huawei Ascend by designing high‑performance kernels and memory optimizations. His background blends practical firmware and back-end performance engineering with research in computer vision (semantic segmentation) and graph theory, and he has contributed noteworthy fixes and transforms to the widely used PyTorch Geometric library and performance primitives in oneDNN. Known for squeezing order‑of‑magnitude gains from kernels (e.g., 2x+ and 10x improvements in key cases), he combines low‑level C++/GPU expertise with a knack for API‑level usability that accelerates model deployment.
8 years of coding experience
6 years of employment as a software developer
Master of Science - MS Computer Science, Master of Science - MS Computer Science at Gdańsk University of Technology
English, Polish
Github Skills (20)
opencl10
pytorch10
c-language10
datatypes10
python10
gpu-programming10
vectorization10
numpy10
deeplearning-ai10
deep-learning10
performance-optimization10
pooling10
graph-neural-network10
cprogramming-language10
batch-normalization9
Programming languages (4)
Github contributions (5)
uxlfoundation/oneDNN
Sep 2020 - Sep 2021
oneAPI Deep Neural Network Library (oneDNN)
Role in this project:
Back-end Developer / Performance Engineer
Back-end Developer / Performance EngineerContributions:59 commits in 1 year
Contributions summary:Bartłomiej primarily focused on optimizing the oneDNN library's performance, particularly within the GPU domain. Their contributions include adding support for asymmetric data types in pooling operations and adding support for new data layouts for specific hardware architectures. They also implemented vectorization techniques for batch normalization calculations and fixed issues related to post-operations. Furthermore, the user introduced optimized kernels for reduction operations and global pooling.
bfloat16sse41avx512openmpopencl
pyg-team/pytorch_geometric
Jan 2023 - Jan 2023
Graph Neural Network Library for PyTorch
Role in this project:ML Engineer
ML EngineerContributions:4 reviews, 3 commits, 8 PRs in 14 days
Contributions summary:Bartłomiej contributed to the PyTorch Geometric library, focusing on enhancing its functionality and maintainability. Their work included refactoring code to address potential issues with mutable default arguments, ensuring compatibility with updated dependencies like NumPy, and adding a transform to pad node and edge features for consistent tensor shapes. The user also introduced the `AddRemainingSelfLoops` transform, improving graph manipulation capabilities within the library. Several commits were co-authored by Matthias Fey, indicating collaboration and code review.
pytorchgraph-convolutional-networksgeometric-deep-learningdeep-learningneural-graph
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.