single mention · model architecture
Block Max Pooling
Block Max Pooling compresses token-level attention scores into block-level scores.1 It sits in the index branch of sparse attention, producing scores that TopK uses to decide which KV blocks to keep.2
How this definition was written
Summarised by glm-5.3-flash from 1 source; every sentence is checked against the quote it cites.
Block Max Pool then compresses the token-level scores into block-level scores
— MiniMax Goes Sparse: Decoding M3's Attention from a Single DiagramFinally, TopK decides which KV blocks to keep for this layer and this GQA group
— MiniMax Goes Sparse: Decoding M3's Attention from a Single Diagram
Also called Block Max Pool.
Filed under model architecture › token mixer › sparse attention › block selection