Josh Levy-Kramer
Member Of Technical Staff at Anthropic
Stony Stratford, England, United Kingdom
Summary
🤩
Rockstar🎓
Top SchoolJosh Levy-Kramer is an AI research engineer with 11 years of experience building and deploying large-scale LLM systems and ML infrastructure, currently on Anthropic's multimodal pretraining team. He led technical development of LLM tooling at Graphcore—helping create PopXL, tensor model parallelism, ZeRO optimizer support and float8 training research presented at NeurIPS—and has practical DevOps experience contributing Kubernetes deployments and monitoring integrations to the widely used EleutherAI/gpt-neox project. His background spans applied NLP and production ML across healthcare, pharma, and startups, combining a physics MSci with hands-on engineering from microfluidics to cloud-native clusters. Colleagues describe him as a pragmatic engineer who bridges research and production, with a hobbyist blog (robotchinwag.com) that signals a curious, communicative approach to technical topics.
10 years of coding experience
11 years of employment as a software developer
Highgate Wood School Arts CollegeA-Level, Pysics, Maths, Further Maths, Computing, A-Level, Pysics, Maths, Further Maths, Computing at Woodhouse CollegeMaster in Science (MSci), Physics, Master in Science (MSci), Physics at University of Bristol
Github Skills (8)
kubernetes10
bash10
docker10
dockers10
cicd10
kubernetes-pods10
yaml9
python8
Programming languages (12)
Github contributions (5)
EleutherAI/gpt-neox
Jan 2021 - Mar 2021
An implementation of model parallel autoregressive transformers on GPUs, based on the Megatron and DeepSpeed libraries
Role in this project:
DevOps Engineer
DevOps EngineerContributions:17 reviews, 236 commits, 33 PRs in 1 month
Contributions summary:Josh's contributions primarily focus on improving the deployment process for the gpt-neox project, implementing features such as Kubernetes deployment using a statefulset. They created a deployment script for Kubernetes, implemented features to configure cluster with specific nodes and a shared mount for data, and incorporated the use of ssh keys for secure access. Furthermore, the user worked on adding and enabling tools like wandb for monitoring of the models.
pytorchlanguage-modeltransformersdeepspeedmodel-parallel
joshlk/many_requests
Nov 2020 - Mar 2021
Dead easy interface for executing many HTTP requests asynchronously. Also provides helper functions for executing embarrassingly parallel async coroutines.
Contributions:29 commits, 2 PRs, 26 pushes in 4 months
http-requestsasynchronousasyncioparallelasynchronously
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.