Unverified50% confidenceBenchmarkExact time
On GPT-2, the moment relaxation certificate is tighter than linear bound propagation on every attention head.
1
Sources
50%
Confidence
Long-term
Relevance
11/3/2026
First Seen
Sources
Related Entities
Related Claims
Unverified扩散语言模型天然支持双向注意力,每个位置可以同时看到左侧和右侧的上下文,而GPT类模型受因果掩码限制只能看到左侧67% similarUnverifiedDense attention layers require a complete KV Cache, whereas linear layers rely on hidden states.66% similarUnverifiedEntries with large attention weights do not necessarily contribute most to the output after matrix multiplication; weight and output contribution are two distinct quantities.66% similarUnverified标准Attention的显存消耗随序列长度平方增长,Flash Attention将其压缩到线性增长64% similarUnverifiedGPU擅长稠密矩阵乘法,对标准多头注意力友好;某些NPU或边缘芯片可能对稀疏注意力、线性注意力或状态空间模型层有更好的支持。64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1129313API
curl https://kongchang.com/api/v1/knowledge/claims/1129313MCP
get_claim(id=1129313)