LISA: Layerwise Importance Sampling for Memory-Efficient LLM Fine-Tuning

by convexstrictlyon 3/27/2024, 5:17 PMwith 1 comments

by convexstrictlyon 3/27/2024, 5:19 PM

An author claims better performance than LoRA in 50% of the time.

https://twitter.com/Rui45898440/status/1772996453557997924