kv-cache-compression

Enables 8x KV cache compression with minimal training overhead to improve reasoning accuracy by allowing more token generation within computational budgets.

Independent GitHub snapshot · Automated classification · Original documentation governs