Adrian Wälchli
Member Of Engineering at poolside
Bern, Bern, Switzerland
Summary
🤩
Rockstar🎓
Top SchoolAdrian Wälchli is a research-focused software engineer with 11 years of experience building scalable deep learning infrastructure from Bern, Switzerland. He has driven core improvements in PyTorch and played a leading role at Lightning AI/Studio making large-model training and fault-tolerant multi-node orchestration practical for production. Adrian co-designed Lightning Fabric and Apps and refactored trainer internals to enable FSDP, 2D-parallelism, distributed checkpoints and seamless HPC integrations like SLURM. His open-source contributions include significant refactors to PyTorch Lightning’s trainer and optimizer loops and improvements to PyTorch’s scheduler and dynamo state handling—work that materially eases scaling models across GPUs, TPUs and other accelerators. Comfortable across research, backend systems and cluster administration, he pairs academic depth (Structure from Motion and Optical Flow research) with hands-on production deployments. Colleagues rely on him for clean, fault-tolerant designs that reveal non-obvious edge cases in large-scale training.
11 years of coding experience
5 years of employment as a software developer
Master of Science - MS, Computer Science, Master of Science - MS, Computer Science at University of BernBachelor of Science - BS, Computer Science, Bachelor of Science - BS, Computer Science at Bern
Stackoverflow
816reputation
56kreached
13answers
17questions
Github Skills (26)
pytorch10
checkpoint10
python10
machine-learning10
checkpointing10
deep-learning10
neural-network10
evaluation9
back-end-development9
debug9
debugging9
trainings9
eval9
tensor9
modeling9
Programming languages (10)
Github contributions (5)
Lightning-AI/pytorch-lightning
Apr 2020 - Jan 2023
Pretrain, finetune ANY AI model of ANY size on multiple GPUs, TPUs with zero code changes.
Role in this project:
ML Engineer & Backend Developer
ML Engineer & Backend DeveloperContributions:35 releases, 9733 reviews, 1088 commits in 2 years 10 months
Contributions summary:Adrian made significant contributions to the trainer's internal logic refactoring in the PyTorch Lightning repository. They focused on removing deprecated code, specifically trainer hidden state related, and streamlining the optimizer loop logic for both manual and automatic optimization. Their work included improving the checkpointing system's ability to handle various setups, and enhancing the test and evaluation processes.
pythonheadachespytorch-modelsdata-sciencehandling
pytorch/pytorch
Dec 2021 - Dec 2022
Tensors and Dynamic neural networks in Python with strong GPU acceleration
Role in this project:Back-end Developer
Back-end DeveloperContributions:21 reviews, 2 commits, 13 PRs in 1 year
Contributions summary:Adrian primarily contributed to the PyTorch library, focusing on improving its core functionalities. Their work involved fixing bugs, enhancing the flexibility of learning rate schedulers by making them pickleable, and modifying the state dict handling of the dynamo optimized module. Additionally, they addressed documentation issues and refactored existing code to ensure proper functionalities. Their contributions spanned across multiple areas of the library and demonstrate a deep understanding of its internal workings.
pythongpu-accelerationdeep-learninggpunumpy
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.