Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Fixing GPU and CPU tf head failures #3431

Merged
merged 1 commit into from Feb 28, 2022
Merged

Fixing GPU and CPU tf head failures #3431

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Jump to
Jump to file
Failed to load files.
Diff view
Diff view
7 changes: 4 additions & 3 deletions horovod/tensorflow/mpi_ops.cc
View file
Open in desktop
Expand Up @@ -1137,13 +1137,14 @@ class HorovodJoinOp : public AsyncOpKernel {
}
};

#if HOROVOD_GPU_ALLREDUCE
REGISTER_KERNEL_BUILDER(Name("HorovodJoin")
.Device(DEVICE_CPU)
.Device(DEVICE_GPU)
.HostMemory("output"),
HorovodJoinOp);
#if HOROVOD_GPU_ALLREDUCE
#else
REGISTER_KERNEL_BUILDER(Name("HorovodJoin")
.Device(DEVICE_GPU)
.Device(DEVICE_CPU)
.HostMemory("output"),
HorovodJoinOp);
#endif
Expand Down
12 changes: 4 additions & 8 deletions test/parallel/test_tensorflow.py
View file
Open in desktop
Expand Up @@ -4051,8 +4051,6 @@ def test_horovod_join_allreduce(self):
self.assertSequenceEqual(ret_values, [ret] * size,
msg="hvd.join() did not return the same value on each rank")

@pytest.mark.skipif(LooseVersion(tf.__version__) >=
LooseVersion('2.9.0'), reason='https://github.com/horovod/horovod/issues/3422')
def test_horovod_syncbn_gpu(self):
"""Test that the SyncBatchNormalization implementation is correct on GPU."""
# Only do this test if there are GPUs available.
Expand Down Expand Up @@ -4085,8 +4083,8 @@ def test_horovod_syncbn_gpu(self):
for x in x_list:
bn = tf.keras.layers.BatchNormalization(axis=1, fused=False)
sync_bn = hvd.SyncBatchNormalization(axis=1)
bn_func = bn.apply(x, training=True)
chongxiaoc marked this conversation as resolved.
Show resolved Hide resolved
sync_bn_func = sync_bn.apply(tf.expand_dims(x[hvd.rank()], 0), training=True)
bn_func = bn(x, training=True)
sync_bn_func = sync_bn(tf.expand_dims(x[hvd.rank()], 0), training=True)

try:
init = tf.global_variables_initializer()
Expand All @@ -4100,8 +4098,6 @@ def test_horovod_syncbn_gpu(self):
self.assertAllClose(self.evaluate(sync_bn.moving_mean), self.evaluate(bn.moving_mean))
self.assertAllClose(self.evaluate(sync_bn.moving_variance), self.evaluate(bn.moving_variance))

@pytest.mark.skipif(LooseVersion(tf.__version__) >=
LooseVersion('2.9.0'), reason='https://github.com/horovod/horovod/issues/3422')
def test_horovod_syncbn_cpu(self):
"""Test that the SyncBatchNormalization implementation is correct on CPU."""

Expand Down Expand Up @@ -4131,8 +4127,8 @@ def test_horovod_syncbn_cpu(self):
for x in x_list:
bn = tf.keras.layers.BatchNormalization(axis=1, fused=False)
sync_bn = hvd.SyncBatchNormalization(axis=1)
bn_func = bn.apply(x, training=True)
sync_bn_func = sync_bn.apply(tf.expand_dims(x[hvd.rank()], 0), training=True)
bn_func = bn(x, training=True)
sync_bn_func = sync_bn(tf.expand_dims(x[hvd.rank()], 0), training=True)

try:
init = tf.global_variables_initializer()
Expand Down