Piotr Chmiel
C AI Software Engineer at Graphcore
Gdańsk, Pomeranian Voivodeship, Poland
Summary
🤩
Rockstar🎓
Top SchoolPiotr Chmiel is a C++ AI software engineer with 11 years of experience building high-performance systems across telecommunications and semiconductor industries, now focused on AI hardware and software at Graphcore. He has deep expertise in modern C++, embedded development, distributed and parallel programming, and squeezing performance from CPUs and accelerators using SIMD, JIT, and multithreading techniques. At Intel he helped design and optimize primitives for oneDNN and the Nervana NNP-T accelerator, while contributing backend improvements to the widely used PyTorch Geometric library to make graph operations more static-shape and torch.compile friendly. Comfortable across low-level system code and higher-level ML stacks, he advocates for modern C++ standards and pragmatic performance engineering. Based in Gdańsk, he combines production-grade engineering for silicon targets with open-source contributions that improve runtime compatibility and scalability.
11 years of coding experience
6 years of employment as a software developer
Bachelor's degree, Computer Science, 2016, Bachelor's degree, Computer Science, 2016 at Politechnika Wrocławska
English, Russian, białoruski, Polish
Stackoverflow
1reputation
0reached
0answers
0questions
Github Skills (16)
tensorrt10
geometric-deep-learning10
batch10
pytorch10
batch-updates10
tensor10
deep-learning10
tensorflow10
graph-neural-network10
bulk-operation10
batchfile10
python10
operation10
batch-processing10
optimization10
Programming languages (3)
Github contributions (5)
pyg-team/pytorch_geometric
Apr 2023 - Jun 2023
Graph Neural Network Library for PyTorch
Role in this project:
Back-end Developer
Back-end DeveloperContributions:3 reviews, 11 PRs, 5 comments in 1 month
Contributions summary:Piotr primarily contributed to the PyTorch Geometric library by adding new functionalities and improving existing ones. They added arguments like `batch_size` and `max_num_nodes` to various pooling and aggregation layers, such as `MemPooling` and `SortAggregation`, to optimize calculations. The user also addressed potential performance issues by modifying the `to_dense_batch` and `voxel_grid` functions to remove Python fallbacks and incorporate `repeat` for better compatibility with torch.compile, while also adding the use of an experimental flag to disable dynamic shapes for static devices.
pytorchgraph-convolutional-networksgeometric-deep-learningdeep-learningneural-graph
piotrchmiel/libfabric
Nov 2024 - Dec 2024
Open Fabric Interfaces
Contributions:6 pushes, 2 branches in 25 days
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.