huawei-csl/KVarN
KVarN is a native vLLM KV-cache quantization backend for your agents: 3-5x more context, throughput above FP16, and FP16-level accuracy. Calibration-free, one flag.
510
Stars
37
Forks
49.3
Radar Score
Rising
Momentum
Scoring Breakdown
Impact52.5
Velocity46.9
Health47.4
Growth Metrics
7-Day Stars
+0
30-Day Stars
+0
Acceleration
x
Language
Python
Why it ranks
Steady growth — consistently collecting stars and contributors.
LLM Frameworks · quantizationTop 42% globally