3 ms·
Does it beat them because it's a transformer, or because it's a much larger end-to-end model with higher quality multimodal training?
by orbital-decay 2y ago
Does it beat them because it's a transformer, or because it's a much larger end-to-end model with higher quality multimodal training?
- scratchyone 2y agoI wonder if it benefits because it can attend to individual tokens of the prompt while generating, compared to typical diffusion models that just get a static vector embedding of the prompt.