Sumanth Hegde
Software Engineer at Anyscale
San Francisco, California, United States
Summary
🤩
Rockstar🎓
Top SchoolSumanth Hegde is a software engineer at Anyscale focused on advancing fine-tuning and deployment workflows for large language models, with eight years of experience across ML research and production systems. He holds a master's in computer science from UC San Diego and has blended academic research in computer vision with hands-on ML engineering at startups and enterprises like C3 AI, Hakimo, and HyperVerge. Sumanth has contributed core IA3 functionality to Hugging Face’s widely used PEFT library and improved Accelerate tooling, reflecting a strong grasp of parameter-efficient fine-tuning and model adaptation across architectures. At Anyscale he sped up DPO training through prefix sharing, broadened model source and chat-template support, and led building an SDK for models trained on the platform. His undergrad work delivered a dramatic speedup in lensless camera reconstruction and an ECCV Workshop publication, showing a knack for physics-aware neural solutions and computational efficiency. Based in San Francisco, he combines deep research instincts with pragmatic engineering to move state-of-the-art ML techniques into production.
7 years of coding experience
2 years of employment as a software developer
Indian Institute of Technology MadrasUniversity of California, San Diego
English, Hindi, Kannada
Stackoverflow
1reputation
0reached
1answer
0questions
Github Skills (10)
transformers10
pytorch10
python10
adapter10
ml9
mle9
testing8
nlp7
macos6
mongodb6
Programming languages (5)
Github contributions (5)
huggingface/peft
Apr 2023 - Mar 2025
🤗 PEFT: State-of-the-art Parameter-Efficient Fine-Tuning.
Role in this project:
ML Engineer
ML EngineerContributions:37 reviews, 13 PRs, 68 comments in 1 year 11 months
Contributions summary:Sumanth primarily contributed to the implementation and improvement of Parameter-Efficient Fine-Tuning (PEFT) techniques, specifically IA3 (Infused Adapter). They added core IA3 functionality, including initial code, feedforward layer implementations, merge/unmerge capabilities, and support for saving/loading models. Their work involved addressing merge conflicts, fixing module mapping issues, and adding support for various model architectures, demonstrating a strong understanding of adapting PEFT methods to diverse transformer models. They also improved example notebooks and test coverage, including adding tests for module name matching.
parametermachine-learningfineparameter-efficientfine-tuning
SumanthRH/text-to-meme
Nov 2022 - Dec 2022
Contributions:66 commits, 33 pushes, 2 branches in 1 month
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.