implementation detail · filed under model architecture
Dynamic Attention Window Size Training
A training mechanism that varies attention window size to support inference caching and offline audio understanding.
- source
- 1
- model
- 1
- lab adopt it
- 1
- strongest
- used
How sources treat it
One count per evidence span, weakest treatment to strongest.
used 1
Documented in
Evidence
1 span quoted from the sources, strongest treatment first.
The dynamic attention window size training mechanism is adopted for guaranting balance performance of inference under real-time prefill caching and for the offline audio understanding tasks.
usedtraining objectivein Qwen3.5-OmniQwen
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 SerializationFixed Sliding Window AttentionPer-Head GatingPure Sliding Window AttentionSink-Augmented Sliding Window AttentionSWA-1024SWA-128