LLMs are overparameterized for text embedding tasks and can be easily pruned arxiv.org 3 points by hdvr 10 hours ago