3 ms·
Neat idea! Are the batches encoded as tokens into the input sequence? This is something I really like about the multi-modal PALM papers since it enables the mul
by Q6T46nT668w6i3m 3y ago
Neat idea! Are the batches encoded as tokens into the input sequence? This is something I really like about the multi-modal PALM papers since it enables the multi-modal tokens to be referenced.
- ekelsen 3y agoImage patches are projected directly into an embedding that goes into the decoder Transformer. The same thing could be done for audio.