Unverified50% confidenceFactExact time
4-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
1
Sources
50%
Confidence
Long-term
Relevance
7/20/2026
First Seen
Sources
Related Claims
Unverified在4-bit量化下，纯参数理论最小占用为：7B约3.26GiB，13B约6.05GiB，32B约14.90GiB，70B约32.60GiB75% similarUnverifiedTernary quantization means each weight has only three possible value states, typically -1, 0, and +1.72% similarUnverifiedThe FP4 format has only 4 bits (1 sign bit, 2 exponent bits, 1 mantissa bit) and can precisely represent only approximately 16 discrete values.72% similarUnverified1-bit量化（如BitNet b1.58）将每个参数用-1、0、+1三值表示，目前主要停留在研究阶段71% similarUnverifiedq4_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69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/571226API
curl https://kongchang.com/api/v1/knowledge/claims/571226MCP
get_claim(id=571226)