Pinned Loading

1. turboquant turboquant Public

   Open-source implementation of Google's TurboQuant (ICLR 2026) — KV cache compression to 2.5–4 bits with near-zero quality loss. 3.8–5.7x memory reduction on Mistral-7B, no training required.

   Python 50 9