Unverified50% confidenceTradeoffExact time
Perceiver Resampler和Q-Former能将数百个视觉token压缩为几十个固定数量的查询token，减少计算负担但可能丢失空间细节
1
Sources
50%
Confidence
Long-term
Relevance
8/20/2026
First Seen
Sources
Related Claims
UnverifiedThe Multi-Query Attention technique allows multiple attention heads to share a single set of Key and Value projection matrices, compressing KV cache memory requirements to 1/H of the original, at the cost of a slight reduction in model expressiveness.67% 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65% similarUnverifiedDeformable DETR replaces global self-attention with deformable attention, sampling only about 4 reference points per head per layer, reducing the attention computation complexity from O(N²) to O(NK).64% similarUnverifiedThe computational complexity of standard self-attention is quadratic with respect to sequence length, and achieving a context of one million tokens requires introducing techniques such as sparse attention, sliding window attention, and KV Cache compression.64% similarUnverified法线贴图可以让仅有数百面的低模在视觉上呈现出数万面高模的细节层次63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/780094API
curl https://kongchang.com/api/v1/knowledge/claims/780094MCP
get_claim(id=780094)