Closed
Description
Test are starting to fail on test_prototype_builtin_datasets.py on 24 June 2022
Here are an example:
https://github.com/pytorch/vision/runs/7041792653?check_suite_focus=true
__________________ TestCommon.test_transformable[mnist-train] __________________
Traceback (most recent call last):
File "/home/runner/work/vision/vision/test/test_prototype_builtin_datasets.py", line 102, in test_transformable
next(iter(dataset.map(transforms.Identity())))
File "/opt/hostedtoolcache/Python/3.7.13/x64/lib/python3.7/site-packages/torch/utils/data/datapipes/datapipe.py", line 135, in class_function
result_pipe = cls(source_dp, *args, **kwargs)
File "/opt/hostedtoolcache/Python/3.7.13/x64/lib/python3.7/site-packages/torch/utils/data/datapipes/iter/callable.py", line 80, in __init__
validate_input_col(fn, input_col)
File "/opt/hostedtoolcache/Python/3.7.13/x64/lib/python3.7/site-packages/torch/utils/data/datapipes/utils/common.py", line 43, in validate_input_col
f"The function {fn.__name__} takes {len(non_default_params)} "
File "/opt/hostedtoolcache/Python/3.7.13/x64/lib/python3.7/site-packages/torch/nn/modules/module.py", line 1225, in __getattr__
type(self).__name__, name))
AttributeError: 'Identity' object has no attribute '__name__'
As mentioned by @datumbox , if we check the error trace it indicate the call on validate_input_col(fn, input_col) and this function is introduced on pytorch/pytorch#79344 , so for now this PR is the likely reason for the breakage.
cc @seemethere @bjuncek @pmeier @NicolasHug since this is about prototype dataset, can you also take a look on this?
Update:
- There is a fix on pytorch: Raise warning for unpickable local function pytorch#80140 (will confirm on next nightlies)
- It is now fixed (2022-06-27)