RAT+ pretrains a dense recurrent-augmented attention model once and enables flexible switching to dilated or hybrid sparse attention at inference after short adaptation, with small accuracy loss at high dilation factors.
D† =1 means the attention operator or block without the recurrence
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
RAT+: Train Dense, Infer Sparse -- Recurrence Augmented Attention for Dilated Inference
RAT+ pretrains a dense recurrent-augmented attention model once and enables flexible switching to dilated or hybrid sparse attention at inference after short adaptation, with small accuracy loss at high dilation factors.