Error when running under PyTorch 2.1.0 - TORCH_LIBRARY single use

[kornilich]

Tested on examples/testnet

Error: -> RuntimeError: Only a single TORCH_LIBRARY can be used to register the namespace _inductor_test; please put all of your definitions in a single TORCH_LIBRARY block. If you were trying to specify implementations, consider using TORCH_LIBRARY_IMPL (which can be duplicated). If you really intended to define operators for a single namespace in a distributed way, you can use TORCH_LIBRARY_FRAGMENT to explicitly indicate this. Previous registration of TORCH_LIBRARY was registered at /dev/null:241; latest registration was registered at /dev/null:241 -> This error occurred on line 27 when processing entry_point.py.

However, it works on PyTorch 2.0.1

[johncalesp]

Hello, thank you for open this issue. Pytorch recently changed their version and that created some conflicts, temporarily we are compatible up until version 2.0.1. We will look into it with more detail. If you find any more issues or features that will be useful please let us know. Thank you.

[sascharo]

How does it look with the compatibility of PyTorch 2.2.1? It seems deepview-profile 0.13.3 is still limited to PyTorch <= 2.0.1.

[johncalesp]

Hi @sascharo , We are currently working on a new release that will be compatible with the latest version of pytorch.

[sascharo]

We are currently working on a new release that will be compatible with the latest version of pytorch.

Awesome! Do you have an ETA?

[johncalesp]

We are running some test before we release to pypi, I would say a couple of weeks from now

[johncalesp]

Hello @sascharo , We released the new version, now it works with the latest version of pytorch. Let us know what you think, thanks!.