CWM: An Open-Weights LLM for Research on Code Generation with World Models 2025
FAIR CodeGen, J. Copet, Q. Carbonneaux, G. Cohen, J. Gehring, J. Kahn, J. Kossen, F. Kreuk, V. Seeker, et al.
arXiv - arXiv preprint
About Me
I'm a founding engineer at Axiom, an AI startup working toward mathematical superintelligence. I work on building ML systems for training, fine-tuning, and inference of large language models on mathematical datasets.
Before startups, I spent two years at Meta's Fundamental AI Research (FAIR) team in Menlo Park, and before that held research and teaching positions at the University of Edinburgh, where I worked at the intersection of machine learning and systems optimization.
Originally from Berlin, I've spent the last few years in the Bay Area working across academia, big tech research, and early-stage startups.
What I Work On
I enjoy using AI to solve hard problems and building ML systems to help do just that — data, training, inference, and infrastructure.
At Axiom
Full-stack ML for mathematical reasoning. Data pipelines for dataset generation and inference, training and inference infrastructure, fine-tuning and RL systems, and verification and evaluation frameworks for formal mathematics.
At FAIR
Large language models and embedding models for compiler optimization. Trained, fine-tuned and evaluated models for code generation and optimization. Built data pipelines for compiler IR and performance data.
Core Expertise
I work across the ML stack—from data engineering and model training to building training and inference infrastructure. My background spans both research and production systems, and I'm comfortable moving between designing experiments, writing training code, and architecting systems for training and inference.
Background
Since Jun. '25
Founding Engineer
Axiom AI
San Francisco, California, USA
Apr. '23 - Jun. '25
Research Engineer
Meta AI Research (FAIR)
Menlo Park, California, USA
Apr. '20 - Apr. '23
Research Associate
University of Edinburgh
Edinburgh, Scotland
Aug. '17 - Apr. '20
University Teacher
University of Edinburgh
Edinburgh, Scotland
May '16 - Jun. '17
Research Associate
Lancaster University
Lancaster, England
Nov. '11 - Dec. '16
PhD Computer Science
University of Edinburgh
Edinburgh, Scotland
Oct. '05 - May '11
Dipl.-Ing. Computer Engineering
TU Berlin
Berlin, Germany
Publications
A selection of my publications. For a complete list, see my Google Scholar profile.
Google ScholarDon't Transform the Code, Code the Transforms: Towards Precise Code Rewriting using LLMs 2024
C. Cummins, V. Seeker, J. Armengol-Estapé, A.H. Markosyan, G. Synnaeve, H. Leather
arXiv - arXiv preprint
Meta Large Language Model Compiler: Foundation Models of Compiler Optimization 2024
C. Cummins, V. Seeker, D. Grubisic, B. Roziere, J. Gehring, G. Synnaeve, H. Leather
arXiv - arXiv preprint
Distinguished Paper Award
Revealing Compiler Heuristics through Automated Discovery and Optimization 2024
V. Seeker, C. Cummins, M. Cole, B. Franke, K. Hazelwood, H. Leather
CGO - IEEE/ACM International Symposium on Code Generation and Optimization
Large Language Models for Compiler Optimization 2023
C. Cummins, V. Seeker, D. Grubisic, M. Elhoushi, Y. Liang, B. Roziere, J. Gehring, et al.
arXiv - arXiv preprint
Best Paper Award
Measuring QoE of Interactive Workloads and Characterising Frequency Governors on Mobile Devices 2014
V. Seeker, P. Petoumenos, H. Leather, B. Franke
IISWC - IEEE International Symposium on Workload Characterization
Scalable Multi-Core Simulation Using Parallel Dynamic Binary Translation 2011
O. Almer, I. Böhm, T.J.K. Edler von Koch, B. Franke, S. Kyle, V. Seeker, C. Thompson, N. Topham
IC-SAMOS - International Conference on Embedded Computer Systems: Architectures, Modelling, and Simulation
Fun Projects
Zeit²
Brightside Games · Published by Ubisoft · 2010
Before ML, I made games. During my undergrad at TU Berlin, I co-founded a small game development studio called Brightside Games. We built Zeit², a time-manipulation puzzle platformer that got published by Ubisoft and released on Steam and Xbox Live Arcade.
Turns out building a game studio and shipping a commercial game is excellent training for startup life—tight deadlines, constrained resources, and figuring out how to actually finish things. Eventually left to pursue a PhD at Edinburgh, but the experience of taking something from concept to shipped product stuck with me.
The game's still on Steam if you're into indie platformers with time-bending mechanics.