nexusquant-kv
Near-lossless 3-bit/2-bit KV cache quantization via E8 lattice VQ. Training-free, calibration-free, 7 architectures validated. Optional token eviction for higher compression.
- Latest release
- 9h ago
- Releases
- 5
- Known CVEs
- 0
- First release
- Apr 07, 2026
- License
- Apache-2.0
Repository
Source
- Stars
- —
- Forks
- —
- Open issues
- —
Security score
No OpenSSF Scorecard available for this repository.
Packages from this repo
No other tracked packages from this repository.
Insights
Activity
- Total releases
- 5
- Last 12 months
- 5
- Cadence
- ~daily
- Dependencies
- 6
Releases per month
last 12 monthsRelease mix
- minor 1
- patch 3
5
releases
Dependencies
Depends on
0.5.3-
accelerate
>=0.26
-
kvpress
>=0.5.3
-
numpy
>=1.24
-
torch
>=2.0
-
transformers
>=4.46
-
zstandard
>=0.22
Used by
Nothing tracked depends on this yet.
Releases
| Version | Released | |
|---|---|---|
0.5.3
patch
| ||
0.5.2
patch
| ||
0.5.1
patch
| ||
0.5.0
minor
| ||
0.4.0
initial
|