search

dianna-ai / dianna

Deep Insight And Neural Network Analysis
https://dianna.readthedocs.io
Apache License 2.0
48 stars 13 forks source link

Updating READMEs (issues 815 and 863) #866

Closed elboyran closed 4 weeks ago

elboyran commented 1 month ago

Solves issues #815 and #863.

elboyran commented 1 month ago

Oh, as a Draft it passed all checks.... Looking at the fails now....

loostrum commented 1 month ago

Oh, as a Draft it passed all checks.... Looking at the fails now....

That may be because the downloader test doesn't run on draft PRs. We disabled that to avoid many downloads while we don't change the downloader itself that much.

elboyran commented 1 month ago

@loostrum Today I saw you didn't review my PR yet, so I quickly grabbed the chance ;-) to edit a bit more the READMEs and while all checks were green yesterday, now (after Laura's PR?) the EU law notebook fails on Ubuntu. @cwmeijer any idea why?

cwmeijer commented 1 month ago

The error is: "RuntimeError: operator torchvision::nms does not exist" while importing from torchvision. However, all of the systems have the same torchvision version (0.18.1). Pytorch-lightning versions are the same for all 3 as well. Same goes for transformers, which is the package bringing in torchvision as a dependency, all have version 4.46. No clue what is causing this.

elboyran commented 1 month ago

@loostrum would you have time to give it a go (review and hopefully error fix)?

cwmeijer commented 1 month ago

or @SarahAlidoost for that matter as she managed to fix a possibly similar issue a few weeks back. Whoever of you has time for this.

loostrum commented 1 month ago

I'm looking into it now. I do see that on ubuntu, the last passing run had an older version of transformers (4.45.2 vs 4.46.0). the 4.46.0 release of transformers was removed from pypi because of some issue with python 3.8, but that's not the version we're using. Just to be sure I've restarted the workflow, which is now using transformers 4.46.1. If that doesn't work, I'll try pinning to <4.46.0

loostrum commented 1 month ago

Pinning to <4.46 worked 🎉