Halil Akın
Research Engineer at Meta
Menlo Park, California, United States
Summary
🤩
Rockstar🎓
Top SchoolHalil Akın is a research engineer in Menlo Park with 11 years of experience building scalable ML systems at the intersection of AI and protein science. Currently at Meta, he focuses on robust distributed training and memory-efficient model engineering, contributing notable fixes to high-profile open-source toolkits like fairseq and PyTorch's translate library. He brings startup DNA from co-founding a data-driven real estate platform—where he led product, growth, and a cross-functional engineering team that served millions of visits—and applies that product-minded rigor to research engineering. Comfortable across backend infrastructure, distributed GPU training, and production-ready tooling, he has a proven track record of solving OOM, synchronization, and data-loading challenges in large-scale NLP workloads.
11 years of coding experience
2 years of employment as a software developer
Bachelor, Computer Science and Engineering, Bachelor, Computer Science and Engineering at University of WashingtonMaster of Engineering (M.Eng.), Computer Science, Master of Engineering (M.Eng.), Computer Science at Cornell UniversityBachelor, Computer Science, High Honors Degree, Bachelor, Computer Science, High Honors Degree at Boğaziçi University
English, Turkish
Github Skills (12)
pytorch10
debug10
machine-learning10
artificial-intelligence10
distributed-training10
deep-learning10
python10
fairseq10
optimization9
nlp9
data-loading9
cuda8
Programming languages (8)
Github contributions (5)
pytorch/translate
Sep 2018 - May 2020
Translate - a PyTorch Language Library
Role in this project:
Back-end Developer
Back-end DeveloperContributions:80 commits, 54 PRs, 2 comments in 1 year 7 months
Contributions summary:Halil primarily focused on fixing bugs and improving the stability of the PyTorch Translate library. Their work included resolving test failures related to GPU training, pretrained word embeddings, and distributed syncing issues. Additionally, the user addressed OOM (Out of Memory) and overflow issues during training, as well as implemented improvements to checkpoint management and data loading for high-resource languages.
pytorchnlptranslationmachine-learningonnx
facebookresearch/fairseq
Oct 2018 - Apr 2020
Facebook AI Research Sequence-to-Sequence Toolkit written in Python.
Role in this project:ML Engineer
ML EngineerContributions:14 commits, 3 PRs, 1 comment in 1 year 6 months
Contributions summary:Halil primarily contributed to the improvement and robustness of the `fairseq` library, a sequence-to-sequence toolkit for AI research. Their work focused on resolving issues related to distributed training, particularly addressing out-of-memory (OOM) errors and synchronization problems across multiple GPUs. The user implemented fixes and improvements to the training loop and memory management, ensuring more stable and efficient model training. This included converting existing code to utilize quantizable nn.Linear modules for enhanced performance.
pytorchnlpsequencepythontransformer-architecture
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.