Decision Workspace
ohms-adaptq vs tq-kv vs lean-ctx
Side-by-side comparison of Rust crates
45
ohms-adaptq
growingv2.0.3
NOVAQ: Normalized Outlier-Vector Additive Quantization - Democratic 93-100x LLM compression with 99%+ accuracy retention. No restrictions, no gatekeeping.
52
tq-kv
experimentalv0.4.0
TurboQuant: 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.
51
lean-ctx
experimentalv2.7.0
Context Intelligence Engine with CCP. 23 MCP tools, 90+ compression patterns, cross-session memory (CCP), persistent AI knowledge, multi-agent sharing, LITM-aware positioning. Reduces LLM token consumption by up to 99%.
Core Metrics
| ohms-adaptq | tq-kv | lean-ctx | |
|---|---|---|---|
| Health Score | 45 | 52 | 51 |
| Total Downloads | 4.8K | 44 | 703 |
| 30d Downloads | 17 | 44 | 706 |
| Dependents | 0 | 0 | 0 |
| Releases | 15 | 4 | 41 |
| Last Updated | 220d ago | today | today |
| Age | 7m | 1d | 4d |
Health Breakdown
ohms-adaptq
Maintenance
10
Quality
15
Community
6
Popularity
4
Documentation
10
tq-kv
Maintenance
19
Quality
15
Community
6
Popularity
2
Documentation
10
lean-ctx
Maintenance
19
Quality
13
Community
6
Popularity
3
Documentation
10
Technical Details
| ohms-adaptq | tq-kv | lean-ctx | |
|---|---|---|---|
| Version | 2.0.3 | 0.4.0 | 2.7.0 |
| Stable (≥1.0) | ✓ Yes | ✗ No | ✓ Yes |
| License | MIT | MIT OR Apache-2.0 | MIT |
| Dependencies | 43 | 5 | 34 |
| Crate Size | 104KB | 45KB | 218KB |
| Features | 3 | 4 | 2 |
| Yanked % | 0.0% | 0.0% | 0.0% |
| Edition | 2021 | 2021 | 2021 |
| MSRV | — | — | — |
| Owners | 1 | 1 | 1 |
Links
Quick Verdict
- •tq-kv leads with a health score of 52/100, but none of the options score above 80.
- •ohms-adaptq has the most downloads (4.8K), suggesting wider adoption.
- •tq-kv is pre-1.0 — API may change.