Unverified50% confidenceFactExact time
分组量化将权重矩阵按固定大小（通常32或64个元素）分组，每组独立计算量化缩放因子以减少精度损失
1
Sources
50%
Confidence
Long-term
Relevance
7/11/2026
First Seen
Sources
28 Enterprise-Grade AI Agent Projects: A Complete Guide That Even Beginners Can Deploy
bilibili伴你流水-共桃夭7/9/2026
Related Claims
UnverifiedLow-rank decomposition and quantization compression are orthogonally complementary dimensions; quantizing the smaller matrices A and B separately results in smaller errors, and the combined compression ratio can reach several times that of either method alone73% similarUnverifiedK-quant 系列引入了分块量化策略，对权重矩阵中不同的 block 使用不同的缩放参数以减少量化误差71% similarUnverifiedSpectral pooling can reduce a feature map from any size to any smaller size by simply retaining the corresponding number of low-frequency DFT coefficients, whereas max pooling typically downsamples by integer factors.70% similarVerified量化技术通过将模型参数从32位浮点数降低到4位或8位整数表示，可大幅减少内存占用和计算量68% similarUnverifiedGPTQ（2022）引入逐层Hessian矩阵近似，通过最小化权重量化前后的输出重建误差优化量化参数67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/482650API
curl https://kongchang.com/api/v1/knowledge/claims/482650MCP
get_claim(id=482650)