William Unsworth

I think more options are definitely nice, as per @tntmod54321, but that definitely seems like more of an end goal thing than a first implementation. Personally, I'd vote for a...

Actually, I have no idea why this hasn't been brought up yet. I'll see what I can get going.

Depends how old/what your Radeon GPU is; I use a Radeon RX 590 in Marble Marcher, so it does work on AMD GPUs that are new enough. (see [the troubleshooting...

I've been trying to get this working with EditThis. mwclient only supports MediaWiki 1.16 and upwards in latest, while EditThis runs MediaWiki 1.15. So, since the last mwclient that supports...

Simply put, it doesn't use GPT-2 Simple to load the model. It uses Tensorflow with custom code, _not_ GPT-2 Simple. It used GPT-2 Simple to _train_ the model initially, but...

> If it's smaller than the pretrained 1.5B model then does that mean it doesn't use the pretrained model? It's the same size as the pretrained 1.5B model, and uses...

> > It used GPT-2 Simple to _train_ the model initially, > > Wait, what? I was under the impression that GPT-2 Simple couldn't train 1558M on Colaboratory, or am...

Use Python 3.6 and TF1.X.

> Thought I'd comment on this, since I've gotten it working: > > I've got a fairly beefy machine with 32 gigs of ram. As long as I use tensorflow...

Works in a P100, please remove the inability to finetune. At the very most, use a warning.