News for hackers. Stuff that matters.

A comment on We fine-tuned Llama 405B on AMD GPUs. Parent comment

Re:We fine-tuned Llama 405B on AMD GPUs by logicchains on Tuesday September 24, 2024 @03:02AM (#41633824)
Did you consider using https://github.com/AI-Hypercomputer/maxtext ? It has a Jax llama implementation, and gets decent MFU on TPU and GPU (I've only tried it on NVidia GPU, not AMD).
[ Parent | on HN ]
Re:We fine-tuned Llama 405B on AMD GPUs by lostmsu on Tuesday September 24, 2024 @10:18AM (#41636713)
Could you share performance so we could compare?
[ Parent | on HN ]