17 tests fail with unpickling errors

[yurivict]

All of them seem to fail with thie error:

E                           _pickle.UnpicklingError: Weights only load failed. This file can still be loaded, to do so you have two options, do those steps only if you trust the source of the checkpoint. 
E                               (1) In PyTorch 2.6, we changed the default value of the `weights_only` argument in `torch.load` from `False` to `True`. Re-running `torch.load` with `weights_only` set to `False` will likely succeed, but it can result in arbitrary code execution. Do it only if you got the file from a trusted source.
E                               (2) Alternatively, to load with `weights_only=True` please check the recommended steps in the following error message.
E                               WeightsUnpickler error: Unsupported global: GLOBAL numpy.core.multiarray._reconstruct was not an allowed global by default. Please use `torch.serialization.add_safe_globals([_reconstruct])` or the `torch.serialization.safe_globals([_reconstruct])` context manager to allowlist this global if you trust this class/function.
E                           
E                           Check the documentation of torch.load to learn more about types accepted by default with weights_only https://pytorch.org/docs/stable/generated/torch.load.html.

/usr/local/lib/python3.11/site-packages/torch/serialization.py:1470: UnpicklingError

Version: 0.4.4
Python-3.11
FreeBSD 14.2

[mjwen]

Might be some of data files used for tests are loaded using an old way.

@ipcamit Any idea?

[ipcamit]

@yurivict can you please tell your Torch version? Nevermind, it says 2.6 in the error. Let me try and fis this. @mjwen This happens because torch added this option in newer releases and gives error based on version you are using. For mitigation I made sure that it calls both versions and handle the exception, but here it raises UnpicklingError exception and not the one that it was raising in CI (I think it was ValueError) therefore the issue. Let me think of a better solution and push.

[ipcamit]

@yurivict Unfortunately I am unable to replicate it on my end. As torch.load in KLIFF now by default uses weights_only=False. So it should not be an issue. Can you give more verbose output as to where is this issue exactly getting triggered? I can try replicating it in docker image afterwards (there does not seem to be any authoritative FreeBSD docker image).

[mjwen]

Hi @ipcamit,

I think @yurivict saw the bug in Version: 0.4.4. Which should not have the weights_only=False fix.

@yurivict: We are in the process of transitioning from v0.x.x to v1.x.x. The problem you mentioned should go away (hopefully) in v1.x.x. We will try to release a v0.4.5 to fix the problem.

[yurivict]

The problem you mentioned should go away (hopefully) in v1.x.x. We will try to release a v0.4.5 to fix the problem.

Cool.

[yurivict]

pytorch-2.6.0