Unverified50% confidenceFactExact time
LLaMA-2 70B采用80层、64个注意力头的架构，并使用GQA（8个KV头共享64个Q头）
1
Sources
50%
Confidence
Long-term
Relevance
7/19/2026
First Seen
Sources
Related Claims
UnverifiedLlama-3 70B采用80层Transformer、每层64个注意力头、每头128维配置，并采用GQA设计77% similarUnverifiedLLaMA-2 70B 模型共有 80 个 Transformer 层，每层包含自注意力机制和前馈神经网络两大模块69% similarUnverifiedLFM2.5 has 24 layers total, with 18 using Liquid AI's proprietary LIV (Liquid) convolution and only 6 retaining traditional attention mechanisms65% similarUnverifiedGQA被Llama 2 70B和Llama 3系列采用，MQA被Falcon系列采用，通过多个查询头共享同一组KV头减少缓存的KV向量数量63% similarUnverifiedP106-100基于Pascal架构（GP106芯片），与GTX 1060共享相同的计算核心61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/561354API
curl https://kongchang.com/api/v1/knowledge/claims/561354MCP
get_claim(id=561354)