Unverified50% confidenceTradeoffExact time
q4_0使用对称量化（零点固定为0），q4_1使用非对称量化（零点可浮动），后者在数值分布不对称时更精确但需额外存储零点参数
1
Sources
50%
Confidence
Long-term
Relevance
8/8/2026
First Seen
Sources
Related Claims
UnverifiedQ4_K_M has a perplexity approximately 0.1–0.3 lower than pure Q4.76% 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75% similarUnverified与INT4纯整数量化不同，浮点格式能更好地表示接近零的小数值70% similarUnverifiedTernary quantization means each weight has only three possible value states, typically -1, 0, and +1.69% similarUnverified4-bit量化只能表示16个离散值，1-bit只有2个，三元量化（-1/0/+1）只有3个可能值69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/707998API
curl https://kongchang.com/api/v1/knowledge/claims/707998MCP
get_claim(id=707998)