Unverified50% confidenceBenchmarkExact time
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32.60GiB
1
Sources
50%
Confidence
Long-term
Relevance
8/12/2026
First Seen
Sources
Related Claims
Unverified4-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75% similarUnverifiedINT4量化将原本由65536个不同值（FP16）表示的权重压缩到仅16个离散级别68% similarVerifiedDeepSeek-V3 has 671B total parameters, but only approximately 37B parameters are activated per inference.67% similarUnverifiedMixtral 8×22B has 141B total parameters, but only approximately 39B are activated per inference.66% similarVerifiedThe larger the number of model parameters, the greater the robustness to quantization, allowing 70B-level models to maintain performance close to FP16 after 4-bit quantization.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/735389API
curl https://kongchang.com/api/v1/knowledge/claims/735389MCP
get_claim(id=735389)