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

[TF:XLA] Increase tolerance testing XLA's AtrousConv2d gradient.

Browse Source 
Auto tuning on the GPU caused this error tolerance to be exceeded very often with XLA.

PiperOrigin-RevId: 231248158
This commit is contained in:
A. Unique TensorFlower 2019-01-28 10:46:54 -08:00 committed by TensorFlower Gardener
parent 54dc5baa26
commit 367f164359
1 changed files with 1 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
tensorflow/python/kernel_tests/atrous_conv2d_test.py
View File

@ -139,7 +139,6 @@ class AtrousConv2DTest(test.TestCase):
y1.eval(), self.evaluate(y2), rtol=1e-2, atol=1e-2) y1.eval(), self.evaluate(y2), rtol=1e-2, atol=1e-2)
@test_util.run_deprecated_v1 @test_util.run_deprecated_v1
@test_util.disable_xla("This test never passed for XLA") # larger error range
def testGradient(self): def testGradient(self):
with self.session(use_gpu=True): with self.session(use_gpu=True):
# Input: [batch, height, width, input_depth] # Input: [batch, height, width, input_depth]
@ -161,7 +160,7 @@ class AtrousConv2DTest(test.TestCase):
[x_shape, f_shape], [x_shape, f_shape],
output, y_shape) output, y_shape)
print("atrous_conv2d gradient err = %g " % err) print("atrous_conv2d gradient err = %g " % err)
err_tolerance = 1e-3 err_tolerance = 4e-3 if test_util.is_xla_enabled() else 1e-3
self.assertLess(err, err_tolerance) self.assertLess(err, err_tolerance)