Alexandre Sablayrolles
Research Scientist at Mistral AI
Paris, Ile-de-France
Summary
🤩
Rockstar🎓
Top SchoolAlexandre Sablayrolles is a research scientist based in Paris with 11 years of experience building and optimizing machine learning systems. Trained at École Polytechnique and NYU, he blends strong engineering rigor with academic depth to tackle production-focused research problems. He contributes to prominent open-source ML tooling—most notably implementing a faster, memory-efficient DPLSTM and per-sample gradient accumulation support in PyTorch/Opacus for differential privacy use cases. His work reflects a practical focus on performance, correctness, and privacy-aware modeling, including careful handling of embeddings and evaluation-time activations. Comfortable moving between low-level optimization and high-level research, he accelerates privacy-preserving training without sacrificing model usability. Colleagues describe him as a pragmatic problem-solver who brings research-grade ideas into production-ready code.
11 years of coding experience
Master of Science (M.S.), Master of Science (M.S.) at New York UniversityBaccalauréat, Baccalauréat at Lycée Janson de SaillyDiplôme d'ingénieur, Diplôme d'ingénieur at Ecole polytechniqueLycée Sainte Geneviève
Github Skills (6)
neural-network10
pytorch10
machine-learning10
lstm10
deep-learning10
python9
Programming languages (4)
Github contributions (5)
pytorch/opacus
Nov 2020 - Dec 2022
Training PyTorch models with differential privacy
Role in this project:
ML Engineer
ML EngineerContributions:50 reviews, 84 commits, 43 PRs in 2 years 1 month
Contributions summary:Alexandre implemented a new version of the DPLSTM layer, optimizing it for speed and memory efficiency. This involved refactoring the code, adding a new AccumulateLinear layer for per-sample gradient accumulation, and optimizing the storage of activations. The user also made adjustments to support nn.EmbeddingBag and addressed issues with the correct handling of activations during model evaluation. Their work focused on improving the performance and functionality of the differential privacy-friendly LSTM layers.
pytorchpytorch-modelsprivacydeep-learningprivacy-preserving-machine-learning
facebookresearch/spreadingvectors
Feb 2019 - Jun 2021
Open source implementation of "Spreading Vectors for Similarity Search"
Contributions:2 commits, 1 push, 6 comments in 2 years 4 months
spreadingvectorssimilarity-searchsimilaritycosine-similarity
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.