Hacker GitHub

← All repos

huawei-csl/KVarN

View on GitHub ↗ ★ 520 Python

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.

3 mentions on Hacker News by 3 people: 1 story, 2 comments · first 2026-06-04 · latest 2026-06-05

Mentions per year 3 mentions · busiest year 2026 (3)
20082026
AllStoriesComments NewestTopOldest
  1. This is awesome! Let's give them some stars: - https://github.com/huawei-csl/KVarN (original repo, vLLM implementation) - https://github.com/Anbeeld/beellama.cpp (llama.cpp implementation + awesome evals)
    comment by lukasc-ch · 2026-06-05 · in KVarN: Native vLLM backend for KV-cache quantization by Huawei
  2. …the PR. edit: It might not be clear that it is based on vLLM 0.22, which is the current version: https://github.com/huawei-csl/KVarN/commit/d6290e99098d7426d... . All you have to do is create a diff off it; it's fairly straightforward.
    comment by esafak · 2026-06-04 · in KVarN: Native vLLM backend for KV-cache quantization by Huawei