LISA: Layerwise Importance Sampling for Memory-Efficient LLM Fine-Tuningarxiv.org 3 pointsconvexstrictly2 years ago1 commentSaveHideCopy link On HNComments−convexstrictlyOP2yAn author claims better performance than LoRA in 50% of the time.https://twitter.com/Rui45898440/status/1772996453557997924
Comments
An author claims better performance than LoRA in 50% of the time.
https://twitter.com/Rui45898440/status/1772996453557997924