跳到主要内容Skip to main content

bpwBits Per Weight

Bits Per Weight

  • 硬件与部署Hardware & deployment
  • 第五站 · 跑起来Station 5 · Running It
解决什么问题What problem it solves

量化档位的数值表示:平均每个权重占几个比特。The numeric form of a quantisation level: how many bits each weight averages.

这是什么What it is

衡量量化程度的指标。本站按 bpw 分成五档:极限压缩、长上下文优先、平衡档、保守档、近似无损。

数值越低,模型体积越小。

A measure of how aggressively a model is quantised. This site groups them into five tiers: extreme compression, long context first, balanced, conservative and near-lossless.

Lower numbers mean a smaller model.

为什么需要知道Why it matters

它把「Q4、Q8、Q3」这些各家不同的叫法统一成可比较的数字。这是选型时最实用的标尺。It unifies vendor-specific names like Q4, Q8 and Q3 into one comparable number — the most practical yardstick when choosing.

出处Sources
  • 量化档位横向对比(千问 3.8)Quant tiers side by side (Qwen3.8) ↗