Unverified50% confidenceFactExact time

MobileNetV3 采用深度可分离卷积，计算量约为标准卷积的 1/K²+1/C_out，当 K=3 且 C_out=256 时约为原来的 1/8 到 1/9

1

Sources

50%

Confidence

Long-term

Relevance

8/28/2026

First Seen

Sources

Training a Production-Grade Image Classifier at Home: Feasibility and Practical Roadmap

Related Entities

MobileNetV3model

Related Claims

Unverified双编码器通常将语义压缩为768维向量，该数字来源于BERT-base的隐藏层大小，工业部署中常压缩到128或256维62% similar Unverified将1536维向量分解为16个96维子空间、每个子空间使用256个码字，则整个向量仅需16字节表示，压缩比高达384:1，每次距离计算只需16次查表操作60% similar Unverified使用IQ4_XS配合-fa 1 -ctk q8_0 -ctv q8_0对KV缓存进行q8_0量化后，可加载完整32k上下文，常驻内存约16.6GB59% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/810815

API

curl https://kongchang.com/api/v1/knowledge/claims/810815

MCP

get_claim(id=810815)