from Hacker News

Craylm: Open-source unified LLM training and inference for R1

by gdiamos on 2/3/25, 8:49 PM with 2 comments

• by gdiamos on 2/3/25, 9:38 PM

  Some interesting notes about the project if folks want to dig deeper:

  There is a dev mode that runs on a laptop:

    - https://docs.cray-lm.com/deployment/laptop
  

  There is a unified dockerfile for building vLLM and the training stacks on four targets [ROCm, CUDA, ARM, x86]

    - https://github.com/cray-lm/cray-lm/blob/main/Dockerfile

• by gdiamos on 2/3/25, 9:24 PM

  Some of the authors are available to discuss the current code or the roadmap.