None of the colab notebooks listed pytorch.org/hub work out of the box about hub HOT 9 CLOSED

@ailzhang - sorry missed this. @soumith and i connected last week and, as he mentioned, he followed up with the Colab folks to get the latest binaries supported.

Going forward, we have an item in the release checklist to account for this in future releases.

from hub.

recording that DEEPLABV3-RESNET101 has been verified fixed.

from hub.

Experiencing the exact same error as listed for DEEPLABV3-RESNET101 with the MobileNetV2 colab notebook, with the "AttributeError: module 'torch.jit' has no attribute 'unused'" when it tries to import DenseNet.

from hub.

There're a few todo out of this issue I think:

• We need to work with Colab to make sure they're on our latest public release.(1.2 vs 1.1) cc: @jspisak do you know whether we have a POC for it?
• (Ailing) Check whether we can enable GPU as default with the colab notebooks.
• The most important issue is we have a versioning problem here: for repos like torchvision which is under heavy development to keep up with Pytorch master, sometimes they require pytorch nightly to work.
  We could potentially tag each model with releases they work with, either manually or automatically.
  cc: @soumith @colesbury Do you have ideas how to do these properly?

from hub.

submitted googlecolab/colabtools#762 to upgrade pytorch to 1.2 on colab. forgot to do that after 1.2 release.

from hub.

Is there an easy fix for this at the moment? Not just on collab, to make it work on our own systems, what's the pytorch/torchvision combination needed to make things work?

Thanks in advance!

from hub.

@Spandan-Madan pytorch/pytorch#26608 (comment) might help if you are on stable release.
If you'd like some new features on torchvision that's only on master, try install torchvision from master and pytorch-nightly, that'll work with hub.load('pytorch/vision',...).

Let me know if you have followup questions.

from hub.

fixed the fairseq transformer issue via: pytorch/pytorch.github.io@b888765

from hub.

tracking the waveglow issue in #62

from hub.