基于漂移的无标签混合精度文本嵌入量化
原标题：Quantize by Drift: Label-Free Mixed-Precision Post-Training Quantization for Text Embedders
AI 摘要
该论文提出一种无需标签的混合精度训练后量化方法，用于文本嵌入模型。方法通过量化单个模块后重新编码语料，测量输出嵌入相对全精度位置的漂移，以此作为模块敏感度信号，在硬性字节预算下进行加性分配。在五个开发用嵌入模型上，配置级漂移与留出检索质量达到宏观 Spearman 0.911，并在三个未触碰模型上验证了预注册假设，但主预算下漂移分配数值上低于同预算均匀精度。
正文节选
Quantize by Drift: Label-Free Mixed-Precision Post-Training Quantization for Text Embedders Abstract Mixed-precision post-training quantization needs a per-module sensitivity signal, and for a text embedder the obvious one — how much retrieval quality a module costs when quantized — requires relevance labels that deployments rarely have. We measure a label-free substitute: quantization-induced representation drift, obtained by quantizing one module, re-encoding the corpus, and recording how far