量化Quantization
Quantization
解决什么问题What problem it solves
把模型权重从高精度压成低精度，省显存、提速，代价是质量下降。Compress model weights from high to low precision, saving VRAM and speeding things up at the cost of quality.
这是什么What it is
用更少的位数表示每个权重数值。常见单位是 bpw（每个权重平均占多少比特）。
位数越低越省，但质量损失越大。
Representing each weight value with fewer bits. The common unit is bpw — average bits per weight.
Fewer bits saves more but loses more quality.
为什么需要知道Why it matters
这是本地部署最核心的技术手段。同样一个模型，选对量化档能决定能不能跑起来。The core technique in local deployment. For the same model, choosing the right quantisation tier often decides whether it runs at all.