PyTorch now "supports" higher precision unsigned int dtypes like uint16, uint32, and uint64, but the support is very poor, to the point where every test for a two-argument function will fail.
>>> torch.equal(torch.tensor(0, dtype=torch.int32), torch.tensor(0, dtype=torch.uint16))
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
RuntimeError: Promotion for uint16, uint32, uint64 types is not supported, attempted to promote Int and UInt16
See data-apis/array-api-compat#138
I propose adding an environment variable ARRAY_API_TESTS_SKIP_DTYPES which could be set to something like uint16,uint32,uint64 to manually skip those dtypes (the same behavior as if they weren't present on the namespace). Skipping required dtypes would still not be supported.
PyTorch now "supports" higher precision unsigned int dtypes like uint16, uint32, and uint64, but the support is very poor, to the point where every test for a two-argument function will fail.
See data-apis/array-api-compat#138
I propose adding an environment variable
ARRAY_API_TESTS_SKIP_DTYPESwhich could be set to something likeuint16,uint32,uint64to manually skip those dtypes (the same behavior as if they weren't present on the namespace). Skipping required dtypes would still not be supported.