← All repos

microsoft/LLMLingua

View on GitHub ↗ ★ 6,816 Python

[EMNLP'23, ACL'24] To speed up LLMs' inference and enhance LLM's perceive of key information, compress the prompt and KV-Cache, which achieves up to 20x compression with minimal performance loss.

3 mentions on Hacker News by 3 people: 2 stories, 1 comment · first 2023-12-18 · latest 2025-08-26