Unverified50% confidenceFactExact time
LLaMA-3 70B权重约140GB，推理时每生成一个Token都需要完整遍历一次权重
1
Sources
50%
Confidence
Long-term
Relevance
7/19/2026
First Seen
Sources
Related Claims
UnverifiedLLaMA-2-70B在128K上下文长度下单请求KV Cache可占用超过160GB显存77% similarUnverified以LLaMA-2 70B为例，处理4096 token上下文、批量大小32时，KV Cache可消耗超过80GB显存76% similarUnverified对于 LLaMA-2 70B 模型，处理 4096 token 的上下文时，FP16 精度下 KV Cache 约需 20GB 系统内存72% similarUnverifiedAn M4 Max with 128GB unified memory can directly hold large model weights72% similarUnverifiedMeta的LLaMA 2 70B模型即便使用4-bit量化仍需约35GB显存才能完整加载72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/559601API
curl https://kongchang.com/api/v1/knowledge/claims/559601MCP
get_claim(id=559601)