NCP-GENL

Practice session

A team must fine-tune a 70B-parameter base model on 8x A100 80GB GPUs without losing quality vs full fine-tuning, but cannot fit full fine-tuning in memory. Which approach is most appropriate first?

Loading your practice session…