Can LLMs Maintain Fundamental Abilities under KV Cache Compression?

Re-assign community