Skip to content

pytorch fails signature tests with odd error message #159

Description

@asmeurer

This was discussed on slack. There are errors like

FAILED array_api_tests/test_signatures.py::test_func_signature[max] - hypothesis.errors.InvalidArgument: Could not create full array of dtype=torch.uint8 with fill value 0
FAILED array_api_tests/test_signatures.py::test_func_signature[mean] - hypothesis.errors.InvalidArgument: Could not create full array of dtype=torch.float32 with fill value 0.0
FAILED array_api_tests/test_signatures.py::test_func_signature[min] - hypothesis.errors.InvalidArgument: Could not create full array of dtype=torch.uint8 with fill value 0
FAILED array_api_tests/test_signatures.py::test_func_signature[prod] - hypothesis.errors.InvalidArgument: Could not create full array of dtype=torch.uint8 with fill value 0
FAILED array_api_tests/test_signatures.py::test_func_signature[std] - hypothesis.errors.InvalidArgument: Could not create full array of dtype=torch.float32 with fill value 0.0
FAILED array_api_tests/test_signatures.py::test_func_signature[sum] - hypothesis.errors.InvalidArgument: Could not create full array of dtype=torch.uint8 with fill value 0
FAILED array_api_tests/test_signatures.py::test_func_signature[var] - hypothesis.errors.InvalidArgument: Could not create full array of dtype=torch.float32 with fill value 0.0

which are not easy to read

The problem is that hypothesis tries to pass an integer to full, but torch only supports a tuple. I think a simple fix would be if hypothesis only tried passing a tuple to full to avoid this issue.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions