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.
Primary LanguagePythonMIT LicenseMIT
No one’s watching this repository yet.