JORA: Jax library for distributed LLM fine-tuning #20405
Unanswered
aniquetahir
asked this question in
Show and tell
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
We created this library for distributed fine-tuning of Llama-based LLMs (more planned to be added). We reduce the lines of code for simple fine-tuning significantly compared to HuggingFace and we also observe 12x improvement in performance compared to HuggingFace/Deepspeed. Memory consumption per GPU is also reduced. Check it out here:
https://github.com/aniquetahir/JORA
Beta Was this translation helpful? Give feedback.
All reactions