Nicolas Patry
Machine Learning Engineer at Genesis AI
Netherlands
Summary
🤩
Rockstar🎓
Top SchoolNicolas Patry is a Machine Learning Engineer with 11 years of experience building high-performance ML systems and libraries, currently based in the Netherlands. He blends production-grade backend engineering in Python and Rust with deep ML know-how across computer vision, NLP and PyTorch, having driven performance and stability improvements in widely used projects like Hugging Face's Transformers, Diffusers and Tokenizers. At Hugging Face he maintained inference infrastructure and contributed safetensors support and model-loading optimizations that directly improved API and demo performance, and earlier led latency-sensitive Rust work for a Gmail Smart Compose product as CTO at Ottomate. His open-source contributions span core algorithmic work (matmul/backprop in a Rust ML framework), dataset ingestion at scale, and cross-platform system tooling, showing strength across low-level performance and high-level model engineering. Known for spotting and fixing performance bottlenecks, he combines research-rooted rigor from École Centrale Paris with hands-on delivery in production ML systems.
11 years of coding experience
15 years of employment as a software developer
2011 Applied Maths, 2011 Applied Maths at Ecole Centrale ParisBac S Mention bien, Bac S Mention bien at EABJM
French, English, Italian, Spanish, Japanese
Github Skills (69)
transformers10
python10
operation10
diffusion-models10
datasets10
deep-learning10
lora10
optimization10
tokenize10
event-handling10
data-engineering10
diffusion-probabilistic10
tokenizer10
performance-optimization10
optimisation10
Programming languages (17)
Github contributions (5)
huggingface/tokenizers
Sep 2020 - Jan 2023
💥 Fast State-of-the-Art Tokenizers optimized for Research and Production
Role in this project:
Back-end Developer
Back-end DeveloperContributions:53 releases, 348 reviews, 122 commits in 2 years 5 months
Contributions summary:Nicolas focused on improving the training process of the tokenizer by optimizing file reading and introducing parallel processing. They implemented changes to the core `tokenizers/src/tokenizer/mod.rs` file, particularly in the `train` method. The user also addressed code performance bottlenecks and added features such as byte fallback, demonstrating a focus on performance and functionality.
nlptransformer-architecturelanguage-modelnatural-language-understandingtransformers
safetensors/safetensors
Sep 2022 - Jan 2023
Simple, safe way to store and distribute tensors
Role in this project:DevOps Engineer & Python Developer
DevOps Engineer & Python DeveloperContributions:51 releases, 165 reviews, 130 commits in 4 months
Contributions summary:Nicolas primarily focused on improving the build process and CI/CD pipeline of the project. They created and modified build scripts, particularly for Python wheel builds, source distributions, and Conda scripts. Furthermore, the user implemented a utility tool to convert models on the Hugging Face Hub to the safetensors format and added necessary modifications for better compatibility with models such as those used by Diffusers.
pythontensorflowdistributesafetensors
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.