Unverified50% confidenceFactExact time
DepthGraft was experimented on dense models and MoE models ranging from 0.5B to 16B parameters, covering both GQA and MLA attention mechanisms.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/3/2026
First Seen
Valid until: 2/1/2027
Sources
Related Entities
Related Claims
UnverifiedAutoHorizon、DEHP、ChunkTrust、SplineWAM 等工作将执行视界 L 视为自适应决策变量,在运行时从注意力权重、学习分支、潜证据或 B 样条世界-动作模型中估计它。71% similarUnverifiedSenseTime has open-sourced the Loop-9B model, which adopts recurrent attention and parameter-sharing mechanisms, runs smoothly locally with just 16G of memory, and surpasses some larger-parameter models on complex tasks.70% similarUnverifiedIn GGUF quantization, a single model file uses different quantization schemes for different tensors, with attention layers, feed-forward layers, and embedding layers potentially each using different quantization bit-widths and types.68% similarVerified混合专家架构MoE通过稀疏激活机制,推理时只激活模型总参数的1/8至1/468% similarUnverifiedLLaVA-1.5 and its follow-up work introduced two-layer MLPs and even cross-attention mechanisms to enhance cross-modal information fusion.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1130764API
curl https://kongchang.com/api/v1/knowledge/claims/1130764MCP
get_claim(id=1130764)