Unverified50% confidenceBenchmarkExact time
In the Q4_K_XL scheme, MoE layers use MXFP4 and the remaining layers are compressed to Q8_0, achieving 96% same on top-1% prediction consistency
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
8/6/2026
First Seen
Valid until: 11/4/2026
Sources
Q8_K_XL与MXFP4命名之争：FP8≠Q8_0的量化真相
Related Claims
UnverifiedGGUF格式的Q4_K_M是最主流的平衡选择，困惑度损失通常低于1%；Q8_0几乎无精度损失但文件体积是Q4的两倍74% similarUnverifiedQ4_K_M采用分组量化策略，将权重张量划分为64或256维子块并独立计算缩放因子与零点，能将困惑度损失控制在1–3%以内72% similarUnverifiedQ4量化可将模型体积压缩至原来的约1/8，代价是约1-3%的基准测试精度损失72% similarVerifiedQ4量化将每个权重参数从16位压缩到4位，理论上可将模型体积缩小约75%72% similarUnverifiedQ4_K_M has a perplexity approximately 0.1–0.3 lower than pure Q4.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/697312API
curl https://kongchang.com/api/v1/knowledge/claims/697312MCP
get_claim(id=697312)