4 ms·
The transformer in transformer (TnT) model looks promising - you can set up multiple overlapping domains of attention, at arbitrary scales over the input.
by robbedpeter 5y ago
The transformer in transformer (TnT) model looks promising - you can set up multiple overlapping domains of attention, at arbitrary scales over the input.