LLMs are overparameterized for text embedding tasks and can be easily pruned arxiv.org 2 points by hdvr 4 hours ago