Unverified50% confidenceFactExact time
Meta Llama 3的405B参数版本以BF16精度加载需约810GB内存，量化至INT4精度需约200GB
1
Sources
50%
Confidence
Long-term
Relevance
7/16/2026
First Seen
Sources
Related Claims
UnverifiedMeta的Llama 3.1 70B参数模型以FP16加载约需140GB，405B参数模型以FP16加载约需810GB，INT4量化后约202GB78% similarUnverified一个1T参数模型以BF16精度存储需要约2TB显存，INT4量化压缩也需500GB以上77% similarUnverifiedLlama 3 70B量化版（Q4精度）仍需约35GB内存76% similarVerifiedLlama-3 70B模型在FP16精度下仅权重就需要约140GB显存75% similarVerified参数量达1750亿的GPT-4级模型，即便以fp8精度存储也需要约175GB显存75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/530183API
curl https://kongchang.com/api/v1/knowledge/claims/530183MCP
get_claim(id=530183)