Test failures on aarch64-darwin
Closed this issue · 3 comments
samuela commented
I'm observing the following failures in the test suite on aarch64-darwin.
============================= test session starts ==============================
platform darwin -- Python 3.10.12, pytest-7.4.0, pluggy-1.2.0
rootdir: /private/tmp/nix-build-python3.10-ml-dtypes-0.3.1.drv-0/source
configfile: pytest.ini
collected 475 items
tests/custom_float_test.py ..............................F.............. [ 9%]
...F.................................................................... [ 24%]
........................................................................ [ 39%]
.........F.............................................................. [ 54%]
........................................................................ [ 70%]
......... [ 72%]
tests/finfo_test.py ...... [ 73%]
tests/iinfo_test.py .... [ 74%]
tests/int4_test.py ..................................................... [ 85%]
..................................................................... [ 99%]
tests/metadata_test.py . [100%]
=================================== FAILURES ===================================
_____________________ CustomFloatTest.testArgmax_bfloat16 ______________________
self = <custom_float_test.CustomFloatTest testMethod=testArgmax_bfloat16>
float_type = <class 'ml_dtypes.bfloat16'>
def testArgmax(self, float_type):
values_to_sort = np.float32(
> float_type(np.float32(FLOAT_VALUES[float_type]))
)
E RuntimeWarning: invalid value encountered in cast
tests/custom_float_test.py:452: RuntimeWarning
_____________________ CustomFloatTest.testArgmin_bfloat16 ______________________
self = <custom_float_test.CustomFloatTest testMethod=testArgmin_bfloat16>
float_type = <class 'ml_dtypes.bfloat16'>
def testArgmin(self, float_type):
values_to_sort = np.float32(
> float_type(np.float32(FLOAT_VALUES[float_type]))
)
E RuntimeWarning: invalid value encountered in cast
tests/custom_float_test.py:474: RuntimeWarning
________________ CustomFloatTest.testRoundTripToNumpy_bfloat16 _________________
self = <custom_float_test.CustomFloatTest testMethod=testRoundTripToNumpy_bfloat16>
float_type = <class 'ml_dtypes.bfloat16'>
@ignore_warning(category=RuntimeWarning, message="overflow encountered")
def testRoundTripToNumpy(self, float_type):
for dtype in [
float_type,
np.float16,
np.float32,
np.float64,
np.longdouble,
]:
with self.subTest(dtype.__name__):
for v in FLOAT_VALUES[float_type]:
np.testing.assert_equal(dtype(v), dtype(float_type(dtype(v))))
np.testing.assert_equal(dtype(v), dtype(float_type(dtype(v))))
np.testing.assert_equal(
dtype(v), dtype(float_type(np.array(v, dtype)))
)
if dtype != float_type:
np.testing.assert_equal(
np.array(FLOAT_VALUES[float_type], dtype),
> float_type(np.array(FLOAT_VALUES[float_type], dtype)).astype(
dtype
),
)
E RuntimeWarning: invalid value encountered in cast
tests/custom_float_test.py:238: RuntimeWarning
=========================== short test summary info ============================
FAILED tests/custom_float_test.py::CustomFloatTest::testArgmax_bfloat16 - RuntimeWarning: invalid value encountered in cast
FAILED tests/custom_float_test.py::CustomFloatTest::testArgmin_bfloat16 - RuntimeWarning: invalid value encountered in cast
FAILED tests/custom_float_test.py::CustomFloatTest::testRoundTripToNumpy_bfloat16 - RuntimeWarning: invalid value encountered in cast
======================== 3 failed, 472 passed in 1.00s =========================
I'm on ml-dtypes 0.3.1, and building from source.
jakevdp commented
Thanks for the report – these look like expected warnings that we need to add to our ignore list (FLOAT_TYPES deliberately includes invalid values like nan). We have an ignore_warning context manager that we already employ frequently in the test suite, and it looks like we'll have to add it to the problematic tests – would you be willing to put together a PR?
samuela commented
Unfortunately I don't have bandwidth to put together a PR atm, but I'd be happy to test one out if one comes online.