experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Increase tolerance for conv2d_backprop_filter_grad_test

Browse Source 
And enable the XLA GPU test.

PiperOrigin-RevId: 252146139
This commit is contained in:
Sanjoy Das 2019-06-07 17:00:00 -07:00 committed by TensorFlower Gardener
parent dfacd7618a
commit 30af0f2592
2 changed files with 2 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
tensorflow/python/kernel_tests/BUILD
View File

@ -2916,8 +2916,7 @@ cuda_py_test(
"no_rocm",
"optonly", # flaky timeouts unless optimized
],
# TODO(b/134677329): Re-enable this test.
xla_enable_strict_auto_jit = False,
xla_enable_strict_auto_jit = True,
)
cuda_py_test(

2
tensorflow/python/kernel_tests/conv2d_backprop_filter_grad_test.py
View File

@ -110,7 +110,7 @@ class Conv2DBackpropFilterGradTest(test.TestCase):
[in_val, out_backprop_val], [in_shape, out_backprop_shape],
output, filter_shape)
print("conv2d_backprop_filter gradient err = %g " % err)
err_tolerance = 2e-3
err_tolerance = 1e-2
self.assertLess(err, err_tolerance)