Unverified50% confidenceFactExact time
GPTQ（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
1
Sources
50%
Confidence
Long-term
Relevance
7/7/2026
First Seen
Sources
Related Claims
UnverifiedGPTQ was proposed in 2022 and extends the OBS framework from pruning to quantization, using the inverse of the Hessian matrix to optimally adjust unquantized weights to compensate for quantization error, with efficient updates via Cholesky decomposition.72% similarUnverifiedModern quantization schemes such as GPTQ, AWQ, and the grouped quantization of the GGUF format can reduce model size by more than 75% while retaining the vast majority of reasoning capability through layer-by-layer calibration and outlier protection.70% similarUnverifiedBundle adjustment (BA) exploits the sparse structure of the Hessian matrix via the Schur complement (marginalization) to decompose large-scale problems into smaller subproblems, thereby reducing computational complexity.68% similarUnverified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67% 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67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/187836API
curl https://kongchang.com/api/v1/knowledge/claims/187836MCP
get_claim(id=187836)