Benjamin Warner
Benjamin Warner
Resolving this issue would require increasing fastai's minimum torch version from [1.7.0](https://pytorch.org/docs/1.7.0/generated/torch.div.html) to [1.8.0](https://pytorch.org/docs/1.8.0/generated/torch.div.html), which is the first version that deprecates [floor_divide](https://pytorch.org/docs/1.8.0/generated/torch.floor_divide.html).
This issue should be resolved by #3558.
This issue is resolved by #3558.
I believe this was resolved with #3510.
You need to upgrade the fastai install from Colab's default of fastai v1 to fastai v2: ``` !pip install fastcore fastai --upgrade ``` upgrading fastcore wouldn't hurt either.
The latest release of PyTorch doesn't support Python 3.12, so fastai currently does not support it either. Could you provide more details about the error? I am unable to replicate...
@steve-estes fastai doesn't directly require cymem, so it must be from an upstream requirement. You can see [fastai's requirements here](https://github.com/fastai/fastai/blob/8e268e3b501ff27bc9efac1cfa7f4a746271c8bb/settings.ini#L17). If you could figure out which package is pinning to...
@fxmarty @ArthurZucker I didn't touch Llava. The current "Tensor-likes are not close" error shouldn't have anything to do with this PR. It should be ready to go from my end.
@buxdehude2 The latest release of fastai supports PyTorch 2.1.
@Max1x1 PR #3884 resolves this issue, and the fix will be in the next release of fastai.