52
tq-kv
v0.4.0 ExperimentalTurboQuant: Extreme KV Cache Compression for LLMs — ICLR 2026 Pure Rust implementation. Lloyd-Max codebook, fused attention, SIMD (AVX2), candle KvCache drop-in. 2-4 bit, up to 15x compression. Supports Qwen2, Llama, Mistral, Phi-3, Gemma2.
MIT OR Apache-2.0 Edition 2021
Quick Verdict
- ✓Actively maintained (updated 0d ago)
- !Pre-1.0: API may have breaking changes
- ✓Permissive license (MIT OR Apache-2.0)
Security
Checking security advisories...
Downloads
44
Dependents
0
Releases
4
Size
45KB
Deep Insights
📊
Download activity
44 downloads in the last 30 days (1/day avg).
📐
Compact crate
At 44KB, tq-kv is lightweight. Small crate size correlates with focused, well-scoped functionality.
Health Breakdown
Maintenance 19/25
Recency, release consistency, active ratio
Quality 15/25
Yanked ratio, deps, size, maturity, features
Community 6/20
Reverse deps, ownership, ecosystem
Popularity 2/15
Downloads, momentum, growth trend
Documentation 10/15
Docs, repo, license, metadata
Download Trend
Daily downloads · last 90 days
0/day avg
Version Adoption
v0.3.0
27%
v0.1.0
25%
v0.2.0
25%
v0.4.0
23%
Release Timeline
4 releasessince 2026
J
F
M
A
M
J
J
A
S
O
N
D
2026
4
LessMore
Feature Flags
default =["std"]
ffistd*candle
README
Loading README...
Maintainers
Dependencies
5
direct dependencies
Dependents
0
crates depend on tq-kv