Lucas Pasqualin
Staff ML Engineer (TLM), Pytorch MSL at Meta
New York, New York, United States
Summary
🤩
Rockstar🎓
Top SchoolLucas Pasqualin is a Staff ML Engineer at Meta with nine years of experience building and hardening GPU-accelerated infrastructure for large-scale ML workloads. He specializes in distributed checkpointing, fault tolerance, and performance engineering that keep training runs resilient and efficient across clusters. A key contributor to PyTorch, he implemented functional collectives, improved FSDP/HSDP tests, and helped add 3D attention mask support—work that directly influences one of the most-used deep learning frameworks. Lucas is practical in crisis-mode firefighting yet deliberate in tool-building, evidenced by his open-source project torchstore for scalable storage and checkpointing. Based in New York, he blends systems software roots from robotics and simulation with production ML at Meta, favoring low-level fixes that prevent high-impact outages. He’s equally comfortable designing data pipelines as diving into async save and partial-load edge cases that others rarely touch.
9 years of coding experience
9 years of employment as a software developer
Bachelor of Engineering - BE, Computer Engineering, Bachelor of Engineering - BE, Computer Engineering at University of Central Florida
Github Skills (10)
pytorch10
fs10
python10
testing10
automation9
automations9
machine-learning9
deep-learning9
c-language4
cprogramming-language4
Programming languages (5)
Github contributions (5)
pytorch/pytorch
Apr 2023 - Mar 2025
Tensors and Dynamic neural networks in Python with strong GPU acceleration
Role in this project:
Back-end Developer & Automation Engineer
Back-end Developer & Automation EngineerContributions:128 reviews, 79 PRs, 420 pushes in 1 year 11 months
Contributions summary:Lucas contributed significantly to the PyTorch codebase by implementing and testing new functionalities related to distributed checkpointing and functional collectives. Their work included adding support for 3D attention masks in multihead attention, adding end-to-end tests for FSDP, HSDP, and FSDP+TP, and implementing broadcast functionality in functional collectives. They also focused on improving the comparison of state dicts for checkpoint E2E tests and fixing issues related to partial loading and async save.
pythongpu-accelerationdeep-learninggpunumpy
LucasLLC/nanoGPT
Jan 2023 - Mar 2023
The simplest, fastest repository for training/finetuning medium-sized GPTs. (Fork for performance patch)
Contributions:1 review, 2 PRs, 7 pushes in 1 month
pytorchpatchfinetuningdeep-learningsimplest
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.