With the latest update, PyTorch v1.1.0 now requires an additional argument (no longer optional) for some convolutional layer.
When an older model is loaded, the code breaks.
Simply running the code (and generating a new model) is expected to break as well.
Solution for both (but needs understanding / investigation) should be adding the missing argument in models.py. The environment requirements environment.yaml needs to be updated (lift restriction on former PyTorch v1.0.1).
Solution for both (but needs understanding / investigation) should be adding the missing argument in
models.py. The environment requirementsenvironment.yamlneeds to be updated (lift restriction on former PyTorch v1.0.1).