Xingjian Shi

@ddelange i tried running the multilingual model (internally it is mdeberta-base) of autogluon.text and it will trigger an error in huggingface/transformers. I think it’s due to some internal change of...

I checked PT lightning and it has the version min-cap of torch: https://github.com/Lightning-AI/lightning/blob/176ca1fdccefbb99ad610ad422c74f0c59653a9c/requirements/base.txt#L2 We can rely on that for min-cap our own torch version.

Just for sanity check. Are you able to import pytorch? ```python import torch a = torch.ones(10, 5) print(a) ```

@damirpolat @cjj490168650 Would you try the following version combination? ```bash pip install autogluon==0.4.1 protobuf==3.18.1 ```

Actually, I'm unable to reproduce the error in my EC2 instance, even with protobuf==3.20.1.

@cjj490168650 @damirpolat Would you provide more details on which environments you are using when you meet this issue?

@damirpolat Is it possible for you to help test if the following versions of protobuf will work? This can help us nail down the version cap. ``` pip3 install -U...

@yinweisu @gradientsky Do you think if we should limit protobuf to be

@andreas12345 will you try to import pytorch_lightning ? Also, can you try to upgrade pip and try again? pip3 install -U pip pip3 install -U setuptools wheel

@andreas12345 see a similar issue from PyG: https://github.com/pyg-team/pytorch_geometric/issues/4419 . For M1, the solution is to use venv or anaconda.