PyPI Stats

Search

All packages
Top packages

Track packages

nexusquant-kv


PyPI page
Home page
Author: Joao Marques
License: Apache-2.0
Summary: Near-lossless 3-bit/2-bit KV cache quantization via E8 lattice VQ. Training-free, calibration-free, 9 architecture families validated. Optional token eviction for higher compression.
Latest version: 0.6.2
Required dependencies: numpy | torch | zstandard
Optional dependencies: accelerate | kvpress | torch | transformers

Downloads last day: 19
Downloads last week: 383
Downloads last month: 956