specific method · filed under model architecture
Sink-Augmented Sliding Window Attention
Sliding-window attention augmented with sinks to enable information exchange across window boundaries over successive layers.
Also called sink-augmented SWA.
- source
- 1
- models
- 2
- lab adopt it
- 1
- strongest
- core
How sources treat it
One count per evidence span, weakest treatment to strongest.
core 1
Documented in
Evidence
1 span quoted from the sources, strongest treatment first.
It replaces fixed, non-overlapping window attention in MiMo-VL-7B with sink-augmented SWA, enabling information exchange across window boundaries over successive layers
coremodel architecturein MiMo-ViTXiaomi
Filed alongside
Other methods under model architecture :: token mixer :: softmax attention :: sliding window attention.
Sliding Window AttentionHybrid Sliding Window Attention512-token Sliding Window AttentionDecoder SWA Bounded ReplayAlternating Row-Major and Column-Major Token SerializationDynamic Attention Window Size TrainingFixed Sliding Window AttentionPer-Head GatingPure Sliding Window AttentionSWA-1024SWA-128