Skip failing AlignModelTest::test_multi_gpu_data_parallel_forward

[ydshieh]

What does this PR do?

tests/models/align/test_modeling_align.py::AlignModelTest::test_multi_gpu_data_parallel_forward starts to fail after we switch to torch+cu118. If I install back with torch+cu117, it passes again.

This test uses torch.nn.DataParallel which is not recommended (despite not deprecated yet). The error is pure CUDA thing for which I have no knowledge. Combing all the above facts + the usage of this model, let's just skip this particular test for AlignModelTest.

(This failing test cause the other 18 tests to fail due to the CUDA is in a bad state)

E       RuntimeError: Caught RuntimeError in replica 0 on device 0.
E       Original Traceback (most recent call last):
E         File "/usr/local/lib/python3.8/dist-packages/torch/nn/parallel/parallel_apply.py", line 64, in _worker
E           output = module(*input, **kwargs)
E         File "/usr/local/lib/python3.8/dist-packages/torch/nn/modules/module.py", line 1501, in _call_impl
E           return forward_call(*args, **kwargs)
E         File "/transformers/src/transformers/models/align/modeling_align.py", line 1596, in forward
E           vision_outputs = self.vision_model(
E         File "/usr/local/lib/python3.8/dist-packages/torch/nn/modules/module.py", line 1501, in _call_impl
E           return forward_call(*args, **kwargs)
E         File "/transformers/src/transformers/models/align/modeling_align.py", line 1395, in forward
E           embedding_output = self.embeddings(pixel_values)
E         File "/usr/local/lib/python3.8/dist-packages/torch/nn/modules/module.py", line 1501, in _call_impl
E           return forward_call(*args, **kwargs)
E         File "/transformers/src/transformers/models/align/modeling_align.py", line 345, in forward
E           features = self.convolution(features)
E         File "/usr/local/lib/python3.8/dist-packages/torch/nn/modules/module.py", line 1501, in _call_impl
E           return forward_call(*args, **kwargs)
E         File "/usr/local/lib/python3.8/dist-packages/torch/nn/modules/conv.py", line 463, in forward
E           return self._conv_forward(input, self.weight, self.bias)
E         File "/usr/local/lib/python3.8/dist-packages/torch/nn/modules/conv.py", line 459, in _conv_forward
E           return F.conv2d(input, weight, bias, self.stride,
E       RuntimeError: GET was unable to find an engine to execute this computation

[amyeroberts]

Thanks for resolving!

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint.