Unverified50% confidenceFactExact time
The core innovation of Unlimited OCR 2 is sparse reference sliding window attention (SR-SWA), which compresses text-vision attention interactions into a fixed-size reference set (for example, 512 tokens).
1
Sources
50%
Confidence
Long-term
Relevance
10/27/2026
First Seen
Sources
Related Entities
Related Claims
Unverified新内核为滑动窗口注意力（SWA）层提供双输入支持，并引入 attention sinks 机制73% similarUnverified滑动窗口注意力（SWA）由 Mistral AI 广泛采用，将注意力复杂度从 O(N²) 降至 O(N×W)72% similarUnverifiedSliding window attention is a local attention variant in the Transformer architecture, where each token only attends to tokens within a fixed-length window before and after it, reducing the attention computation complexity from O(n²) to O(n·w).70% similarVerifiedSliding window attention reduces the complexity of attention computation from O(n²) to O(n·w), linearizing GPU memory overhead for long texts.69% similarUnverifiedLinformer 将注意力矩阵低秩近似为 O(n) 复杂度，FlashAttention 通过重排计算顺序降低 GPU 内存带宽占用，Longformer 引入滑动窗口注意力68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1102782API
curl https://kongchang.com/api/v1/knowledge/claims/1102782MCP
get_claim(id=1102782)