待验证60% 置信事实精确时间
LFM2.5 has 24 layers total, with 18 using Liquid AI's proprietary LIV (Liquid) convolution and only 6 retaining traditional attention mechanisms
2
来源数
60%
置信度
中期 (~90 天)
时效性
2026/7/2
首次发现
有效期至：2026/9/30
来源
LFM2.5 Local Deployment Hands-On: An 8B Parameter Model That Outperforms GPT-o3s in Tool Calling
bilibiliAGI_Ananas2026/6/1
相关事实
待验证LFM2.5 uses a Mixture of Experts (MOE) architecture combined with Liquid convolution69% 相似待验证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）65% 相似待验证Liquid AI first publicly introduced liquid convolution technology in LFM1.065% 相似待验证LLaMA-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59% 相似待验证LFM2.5-8B-A1B在工具调用（tool calling）场景中官方声称可达到4倍体量模型的效果56% 相似
引用此条事实
Stable URI
https://kongchang.com/claim/47923API
curl https://kongchang.com/api/v1/knowledge/claims/47923MCP
get_claim(id=47923)