Thomas Parnell
Principal Research Scientist at IBM
Zurich, Zurich, Switzerland
Summary
🤩
Rockstar🎓
Top SchoolThomas Parnell is a Principal Research Scientist based in Zurich with eight years driving research at the intersection of AI/ML, systems and cloud computing, currently leading a global team at IBM focused on optimizing LLM inference on modern hardware. He has a strong track record translating research into production: as technical lead for Snap ML he helped create a widely used accelerated ML library now embedded across IBM products, and he contributes performance improvements to the high-throughput vllm LLM inference engine. His background spans signal processing and hardware-aware algorithm design from early startup and CTO roles through to patented and published research at IBM. With a PhD in Mathematics from Warwick, he combines rigorous theoretical grounding with practical systems engineering and a knack for squeezing latency and memory gains out of large models.
8 years of coding experience
12 years of employment as a software developer
PhD, Mathematics, PhD, Mathematics at University of Warwick
Github Skills (12)
llm10
cuda10
pytorch10
transformer10
mlops10
inference10
sampling10
python10
gpt9
deepspeed9
cprogramming-language7
c-language7
Programming languages (5)
Github contributions (5)
vllm-project/vllm
Jan 2024 - Mar 2025
A high-throughput and memory-efficient inference and serving engine for LLMs
Role in this project:
ML Engineer
ML EngineerContributions:76 reviews, 33 PRs, 201 comments in 1 year 2 months
Contributions summary:Thomas primarily focused on improving the performance and functionality of the VLLM inference engine for large language models. Their commits included implementing a dynamic scheduler delay to enhance ITL (Iteration-to-Latency) performance, adding options to truncate prompt tokens in the completion API, and addressing bug fixes related to sampling and repetition penalties. They also contributed to the MLPSpeculator, including enabling loading FP8 checkpoints, and adding support for tie_weights and input scaling.
amdcudadeepseekgpthpu
IBM/snapml-docs
Jan 2021 - Oct 2024
IBM Snap ML Documentation
Contributions:9 releases, 7 reviews, 39 PRs in 3 years 9 months
data-sciencemachine-learningsnapibmidris
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.