8 ms·
LISA: Layerwise Importance Sampling for Memory-Efficient LLM Fine-Tuning
- convexstrictly 2y agoAn author claims better performance than LoRA in 50% of the time. https://twitter.com/Rui45898440/status/1772996453557997924 https://twitter.com/Rui45898440/status/1772996453557997924