disable fp16 test for poolWithSequenceAxis (#3810)
This commit is contained in:
Родитель
e1467a79dc
Коммит
e939648002
|
@ -426,8 +426,9 @@ def test_AveragePool(tmpdir, dtype, device_id):
|
|||
#AveragePool
|
||||
@pytest.mark.parametrize("dtype", DType_Config)
|
||||
def test_AveragePoolWithSequenceAxis(tmpdir, dtype, device_id):
|
||||
if device_id == -1 and dtype == np.float16:
|
||||
pytest.skip('Test is skipped on CPU with float16 data')
|
||||
if dtype == np.float16:
|
||||
# CI reporting "FP16 convolution is only supported via cuDNN." on GPU with float16 data
|
||||
pytest.skip('Test is skipped with float16 data')
|
||||
device = cntk_device(device_id)
|
||||
with C.default_options(dtype=dtype):
|
||||
img = np.reshape(np.arange(16, dtype = dtype), [1, 4, 4])
|
||||
|
@ -1342,8 +1343,9 @@ def test_MaxPool(tmpdir, dtype, device_id):
|
|||
#MaxPool
|
||||
@pytest.mark.parametrize("dtype", DType_Config)
|
||||
def test_MaxPoolWithSequenceAxis(tmpdir, dtype, device_id):
|
||||
if device_id == -1 and dtype == np.float16:
|
||||
pytest.skip('Test is skipped on CPU with float16 data')
|
||||
if dtype == np.float16:
|
||||
# CI reporting "FP16 convolution is only supported via cuDNN." on GPU with float16 data
|
||||
pytest.skip('Test is skipped with float16 data')
|
||||
device = cntk_device(device_id)
|
||||
with C.default_options(dtype=dtype):
|
||||
img = np.reshape(np.arange(16, dtype = dtype), [1, 4, 4])
|
||||
|
|
Загрузка…
Ссылка в новой задаче