Most efficient way to fine-tune an LLM in 2024?

0
admin
Aug 24, 2024 03:55 PM 0 Answers Ask Question
Member Since Apr 2019
Subscribed Subscribe Not subscribe
Flag(0)

The most efficient way to fine-tune an LLM in 2024, given budget constraints, is to use QLoRA with Axolotl on services like runpod, using GPUs such as a single A100/H100 or multiple RTX3090s. Start with a smaller model to prove hypotheses, then use techniques like DPO or KTO to polish, and consider applying grammars for structured results.

0 Subscribers
Submit Answer
Please login to submit answer.
0 Answers
Sort By:

Share: