LISA: Layerwise Importance Sampling for Memory-Efficient LLM Fine-Tuning | Dark Hacker News

Dark Hacker News

new|best|ask|show|jobs

Built with the HackerNews API. Developed by LJT.AI

Note: The HackerNews API provides up to 500 stories per category

LISA: Layerwise Importance Sampling for Memory-Efficient LLM Fine-Tuning | Dark Hacker News

LISA: Layerwise Importance Sampling for Memory-Efficient LLM Fine-Tuning(arxiv.org)

3 points by convexstrictly 2 years ago | 1 comment

convexstrictly 2 years ago |

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

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