Quentin Anthony
Member Of Technical Staff at Zyphra
Columbus, Ohio, United States
Summary
🤩
Rockstar🎓
Top SchoolQuentin Anthony is a Member of Technical Staff and distributed deep learning efficiency researcher with eight years of experience bridging quantum photonics and large-scale neural network systems. Based in Columbus, Ohio, he has led model training efforts at Zyphra building multimodal agent inference optimized for on-device AI, while pursuing a PhD in Computer Science at Ohio State focused on high-performance deep learning. His open-source contributions to flagship projects like DeepSpeed and GPT-NeoX include sparse attention extensions, ZeRO/loss-scaling fixes, monitoring/communication tooling, and bitsandbytes optimizer support—work that meaningfully improves distributed training throughput for autoregressive models. Energetic and curious, he combines hands-on engineering in model-parallel training with a research background in quantum photonic networks, giving him a rare cross-disciplinary perspective on efficient AI at scale.
8 years of coding experience
6 years of employment as a software developer
Home SchoolDoctor of Philosophy - PhD Computer Science, Doctor of Philosophy - PhD Computer Science at The Ohio State University
Stackoverflow
1reputation
0reached
0answers
0questions
Github Skills (17)
transformers10
pytorch10
language-model10
parallelization10
python10
optimizers10
optim10
machine-learning10
optimizer10
deep-learning10
gpu10
adam10
compression10
gpt10
zeroth9
Programming languages (7)
Github contributions (5)
EleutherAI/gpt-neox
Oct 2021 - Jan 2023
An implementation of model parallel autoregressive transformers on GPUs, based on the Megatron and DeepSpeed libraries
Role in this project:
ML Engineer
ML EngineerContributions:2 releases, 321 reviews, 29 commits in 1 year 2 months
Contributions summary:Quentin contributed to the `gpt-neox` repository by adding support for the `bitsandbytes` Adam optimizer, improving the efficiency of model training. They also implemented functionality to return logits during text generation. Additional commits included merging code from the main branch, improving text generation utils, and general code cleanup. The user further demonstrated their understanding of model parallel training and checkpointing.
pytorchlanguage-modeltransformersdeepspeedmodel-parallel
deepspeedai/DeepSpeed
May 2022 - Dec 2022
DeepSpeed is a deep learning optimization library that makes distributed training and inference easy, efficient, and effective.
Role in this project:ML Engineer
ML EngineerContributions:54 reviews, 67 commits, 23 PRs in 6 months
Contributions summary:Quentin made several contributions focused on improving and extending the DeepSpeed library's functionality, particularly in the realm of sparse attention mechanisms. They added a new attention type ("unidirectional") to the existing BigBird and BSLongformer sparse attention configurations, enabling support for autoregressive models. Additionally, the user addressed issues related to loss scaling and ZeRO optimizer, demonstrating a solid understanding of deep learning optimization techniques. The user also contributed to adding and enhancing the DeepSpeed monitoring infrastructure.
billion-parametersfinetuningtrainingmixture-of-expertszero
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.