search

lucidrains / DALLE-pytorch

Implementation / replication of DALL-E, OpenAI's Text to Image Transformer, in Pytorch
MIT License
5.55k stars 643 forks source link

CogView Released! (But it requires a chinese phone number to download the weights) #306

Closed afiaka87 closed 3 years ago

afiaka87 commented 3 years ago

https://github.com/THUDM/CogView

robvanvolt commented 3 years ago

I have the feeling that the CogView release is not the end of the journey... but just the beginning :D

afiaka87 commented 3 years ago

Definitely! Collaboration is fundamental to good science and all this means for me is that we now have another team who has explored the problem fairly well and actually documented and released their weights (unlike OpenAI).

The knowledge we've accrued over the past several months will be very useful for the replication study which I believe intends to explore not just DALL-E; but advancements made since DALLE as well such as the VQGAN and differing architectures discovered by CogView.

This is indeed the beginning. Something like this was quite unfathomable just a few years ago and annoyingly concealed by OpenAI. This work will democratize those ideas for any researcher to use; not just those with enough VC funding.

afiaka87 commented 3 years ago

https://github.com/THUDM/CogView/issues/11

Perhaps this isn't as open of a release as I had thought.

johnpaulbin commented 3 years ago

THUDM/CogView#11

Perhaps this isn't as open of a release as I had thought.

Fixed, weights are now available on a torrent.

YukiSakuma commented 3 years ago

Anyone could store the model in a mirror link like google drive?

johnpaulbin commented 3 years ago

Anyone could store the model in a mirror link like google drive?

Here you go @YukiSakuma , you can simply wget: https://the-eye.eu/public/AI/CogView/cogview-base.tar

My Colab featuring the link is here: https://colab.research.google.com/drive/1Bi2TnSUp2vNiSUhamsNuC4HqkZ2J4WwZ