Running ML models

[darinkishore]

Hi! Any source on the running ml models example? I’ve been trying to get that to work, always unsuccessfully, for the last couple days. Seems like the lack of ML python libraries are a big blocker.

[dave1010]

I've just used up my GPT-4 quota again but I think scikit-learn is what I was playing with.

I've got it to do linear regression and k nearest neighbours without installing anything new.

It also has a few datasets built in. I got it to build a model based on load_digits() in sklearn.datasets, then take in coordinates from an 8 by 8 pixel grid and classify a number I drew.

I haven't tried torch or tensorflow yet.

If a package is missing you should be able to download it and upload it to ChatGPT.

If you have any specific things you're trying to do then let me know.

[darinkishore]

I see! I’m trying to train a roberta model, but no transformers and no internet access means that’s harder than anticipated. Also, the model is too big to be uploaded. Sent from Proton Mail for iOS

…

On Tue, Jul 18, 2023 at 16:26, Dave Hulbert ***@***.***(mailto:On Tue, Jul 18, 2023 at 16:26, Dave Hulbert <<a href=)> wrote: I've just used up my GPT-4 quota again but I think scikit-learn is what I was playing with. I've got it to do linear regression and k nearest neighbours without installing anything new. It also has a few datasets built in. I got it to build a model based on load_digits() in sklearn.datasets, then take in coordinates from an 8 by 8 pixel grid and classify a number I drew. I haven't tried torch or tensorflow yet. If a package is missing you should be able to download it and upload it to ChatGPT. If you have any specific things you're trying to do then let me know. — Reply to this email directly, [view it on GitHub](#1 (comment)), or [unsubscribe](https://github.com/notifications/unsubscribe-auth/AUVJDX2IB6KKMMBMWF26XF3XQ3WQFANCNFSM6AAAAAA2NKOFTU). You are receiving this because you authored the thread.Message ID: ***@***.***>

[dave1010]

I think training a Transformer model might be too much for it. It might be possible (you can upload larger files by splitting them) but I think it's going to time out and reset the environment, then lose your data.

I think the easiest thing would be to run it in an environment dedicated to ML, getting ChatGPT and CI to help with the code. Google Collab might work.

If you have developer access to ChatGPT plugins then you can run https://github.com/dave1010/pandora locally