I could very well be I missed it, but it seems there are no real instructions for running this on a TPU-VM combo, V2 or V3-8. yet there are quite a few speedbumps, which sadly the Colab TPU instructions don't seem to fix.
I've been trying to avoid using the TPU (so CPU only) while running the demos and trying all different kinds of XLA / Pytorch setups, but no luck yet. Is there any issue or docs that I'm missing?
If I can get this to work I could love to contribute some GPT-J based models to generate material to be trained into aitextgen models. Not sure if there's an official channel or list, but if there's a wish-list of models or any suggestions; would love to hear them. Helping with bringing AI to "regular folks" can have my compute anyday 🤗
p.s. Love you, most amazing repo I've encountered since Huggingface's and EleutherAI's. Was loving the blog already, just sad I didn't catch this earlier! Hope your blog will continue, despite the pressure being off by taking down your Patreon a little, doesn't have to be a big post, just some random snippets will at least keep me entertained for hours ;)
I could very well be I missed it, but it seems there are no real instructions for running this on a TPU-VM combo, V2 or V3-8. yet there are quite a few speedbumps, which sadly the Colab TPU instructions don't seem to fix.
I've been trying to avoid using the TPU (so CPU only) while running the demos and trying all different kinds of XLA / Pytorch setups, but no luck yet. Is there any issue or docs that I'm missing?
If I can get this to work I could love to contribute some GPT-J based models to generate material to be trained into aitextgen models. Not sure if there's an official channel or list, but if there's a wish-list of models or any suggestions; would love to hear them. Helping with bringing AI to "regular folks" can have my compute anyday 🤗
p.s. Love you, most amazing repo I've encountered since Huggingface's and EleutherAI's. Was loving the blog already, just sad I didn't catch this earlier! Hope your blog will continue, despite the pressure being off by taking down your Patreon a little, doesn't have to be a big post, just some random snippets will at least keep me entertained for hours ;)