Decoder-only mode, image transformer

[malbergo]

Is there a way to use your codebase for the Image Transformer, which is a transformer made from solely a decoder (no encoder)? This models p(x) like a normal autoregressive model, instead of the translation purpose of p(y |x)

https://arxiv.org/abs/1802.05751

Thanks for your guidance!
Michael

[phohenecker]

Hi Michael

I don't know the image transformer, but I'll check it out, and come back to you.

[malbergo]

Thanks !