Unverified50% confidenceFactExact time
LLaMA-2 70B 模型共有 80 个 Transformer 层，每层约 8.75 亿参数，FP16 精度下单层约占 1.75GB 显存
1
Sources
50%
Confidence
Long-term
Relevance
7/19/2026
First Seen
Sources
Related Claims
UnverifiedLLaMA-2 70B 总参数约 700 亿均匀分布于 80 层，每层约 8.75 亿参数，FP16 精度下单层约占 1.75GB 显存83% similarUnverifiedLlama-3.1-8B 等模型基于 Transformer 解码器堆叠构建，通常包含32个decoder层，单层参数量约2.5亿，FP16精度下单层约占500MB显存79% similarUnverified7900XT搭载Navi 31核心，拥有96个计算单元、20GB GDDR6显存，理论FP16算力超过80 TFLOPS67% similarVerified以70B参数模型、FP16精度、32层注意力头为例，单条200K长度请求的KV Cache约需80-120GB显存，远超单张A100 GPU的80GB显存上限66% similarUnverifiedFlux 2的9B参数模型在FP16精度下模型权重约占18GB显存，通常需要24GB以上的GPU显存66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/563155API
curl https://kongchang.com/api/v1/knowledge/claims/563155MCP
get_claim(id=563155)