Salman M
Research Engineer Scientist at Luma AI
London, England, United Kingdom
Summary
🤩
Rockstar🎓
Top SchoolSalman M is a research engineer and scientist based in London with 9 years of experience building and deploying ML systems across healthcare, startups, and open-source ML projects. He has led production federated learning deployments across NHS hospitals and developed early-detection deep learning models for cardiovascular disease from CT scans, while also founding and technically leading an AI startup focused on process improvement. More recently he contributed model builders and integration work for large Code-Llama2 variants in the PyTorch torchtune ecosystem and implemented reward-model-centric training and tokenization strategies in the popular axolotl open-source project. Comfortable spanning research and engineering, he combines hands-on model training, quantization and LoRA support with production deployment experience. An unusual strength is pairing clinical-scale ML deployments with low-level model optimization and open-source tooling contributions, making him effective at moving models from prototype to hospital-ready systems.
9 years of coding experience
6 years of employment as a software developer
Master of Science - MSc Computing Science and Psychology, Master of Science - MSc Computing Science and Psychology at University of Glasgow
Github Skills (10)
transformers10
model-building10
transformer-models10
pytorch10
machine-learning10
trainings10
python10
modeling10
lora9
quantization8
Programming languages (4)
Github contributions (5)
meta-pytorch/torchtune
Apr 2024 - Jul 2025
PyTorch native post-training library
Role in this project:
ML Engineer
ML EngineerContributions:779 reviews, 200 PRs, 70 pushes in 1 year 3 months
Contributions summary:Salman contributed to the `torchtune` library by adding model builders for Code-Llama2 models, including the 7B, 13B, and 70B parameter versions. These model builders facilitate the instantiation and configuration of Code-Llama2 models within the `torchtune` framework. The changes include defining the model architecture and configurations, specifically implementing the builders for both standard and LoRA-enabled variants, including quantization.
post-trainingpytorch
axolotl-ai-cloud/axolotl
Jan 2025 - Jan 2026
Go ahead and axolotl questions
Role in this project:ML Engineer
ML EngineerContributions:389 reviews, 41 PRs, 265 pushes in 1 year 1 month
Contributions summary:Salman primarily focused on integrating and refining reward models within the axolotl project. They implemented a stepwise supervised prompt tokenizing strategy, indicating involvement in training and fine-tuning models. The commits demonstrate work on integrating the model with the existing framework and modifying configurations. Moreover, they updated the loss functions and tests to align with the reward model's requirements.
gofine-tuningllm
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.