Unverified50% confidenceFactExact time
128K长度的序列所需的注意力计算量是4K长度序列的1024倍
1
Sources
50%
Confidence
Long-term
Relevance
8/12/2026
First Seen
Sources
Related Claims
UnverifiedAt a sequence length of 4096, the N×N attention matrix alone consumes approximately 64 MB of memory in FP16 precision.76% similarUnverified128K上下文窗口的注意力矩阵规模是32K窗口的16倍73% similarUnverifiedIn a 170k token scenario, the attention score matrix requires approximately 54 GB of VRAM in BF16 precision (170,000 × 170,000 × 2 bytes).73% similarUnverifiedIn video generation, the computational cost of self-attention is proportional to the square of the number of tokens; after VAE encoding and patching, a 4-second 720p 24fps video can yield tens of thousands to over a hundred thousand tokens in total.70% similarVerifiedTaking a 70B parameter model with FP16 precision and 32 attention head layers as an example, the KV Cache for a single 200K-length request requires approximately 80–120 GB of VRAM, far exceeding the 80 GB VRAM limit of a single A100 GPU.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/734748API
curl https://kongchang.com/api/v1/knowledge/claims/734748MCP
get_claim(id=734748)