Naman Goyal
Member Of Technical Staff at Thinking Machines Lab
New York, New York, United States
Summary
🤩
Rockstar🎓
Top SchoolNaman Goyal is a machine learning-focused software engineer with 11 years of experience building high-performance ML systems and production software, currently a Member of Technical Staff at Thinking Machines Lab in New York. He spent seven years as a Research Engineer at Facebook FAIR contributing to major open-source projects like fairseq and metaseq, where he added BF16 support, model resharding, and dataset/task improvements for challenging benchmarks such as Winogrande. His background spans applied ML, data engineering and systems development—from real-time aggregation engines at TIBCO to demand forecasting and optimization work during internships and academic research. Known for bridging research and engineering, he routinely transforms research models into optimized, deployable code and has a track record of shipping reproducible models and fine-tuning workflows. Trained at Georgia Tech (MS CS), he combines rigorous academic grounding with hands-on experience across industry research labs and production teams.
11 years of coding experience
11 years of employment as a software developer
Master of Science (MS) Computer Science, Master of Science (MS) Computer Science at Georgia Institute of TechnologyBachelor of Engineering (BE) Computer Engineering, Bachelor of Engineering (BE) Computer Engineering at Savitribai Phule Pune University
Github Skills (14)
transformer-models10
pytorch10
machine-learning10
artificial-intelligence10
nlp10
bart10
deep-learning10
python10
fairseq10
natural-language-processing10
model-optimization10
parallelization9
distributed-training9
cuda9
Programming languages (6)
Github contributions (5)
facebookresearch/metaseq
May 2022 - Jan 2023
Repo for external large-scale work
Role in this project:
ML Engineer
ML EngineerContributions:25 reviews, 23 commits, 24 PRs in 7 months
Contributions summary:Naman primarily focused on enhancing the codebase to support bfloat16 (BF16) precision, indicating a focus on optimizing for efficient model training and execution. Their contributions include modifying transformer layers and unit tests to incorporate BF16 support, likely improving the performance of large-scale models. Additional changes included adding features to disable bias and layer normalization, suggesting attempts to further optimize model architecture and training. Furthermore, they added code to reshard model parts and worked on interactive hosted models.
transformersdockerscalelarge-scalehuggingface
facebookresearch/fairseq
Feb 2019 - Apr 2021
Facebook AI Research Sequence-to-Sequence Toolkit written in Python.
Role in this project:ML Engineer
ML EngineerContributions:14 commits, 9 PRs, 24 pushes in 2 years 3 months
Contributions summary:Naman contributed significantly to the `fairseq` repository, focusing on tasks related to the Winogrande dataset, likely improving the performance of a language model. They added an efficient WSC task/criterion for Winogrande, and subsequently made code changes and released a model. The user also added instructions for fine-tuning BART on the CNN-DM dataset and releasing a Flores pretrained model.
aipythonsequence-to-sequencepytorchartificial-intelligence
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.