image-gpt

[bil-ash]

@karpathy Please add code for training of 125M image-gpt to this repository. May be, also extend the context length to 2k so that we can get 45x45 pixel image instead of 32x32 pixel and also make the color palette R4G4B4 instead of R3G3B3 for crispier image. Also, I guess a recent dataset like LAION could be used. Then we will a get a small text-to image generator which although much underpowered, will be useful in many cases.

[Aleksa Gordić]

you would just need to tokenize images and everything else remains pretty much the same?

we didn't have multimodal plans for this repo for the near future