Fix apply() API change in TF 2.9 for unit tests #3427
Conversation
TF 2.9 removed apply() API, use __call__() instead. Signed-off-by: Chongxiao Cao <chongxiaoc@uber.com>
chongxiaoc
force-pushed
the
fix_tf_2_9
branch
from
be8e22a
to
72572ce
Compare
Unit Test Results (with flaky tests) 877 files + 10 877 suites +10 9h 41m 8s ⏱️ - 17m 0s For more details on these failures, see this check. Results for commit 72572ce. ± Comparison against base commit 642a6b3. |
EnricoMi
changed the title
| bn_func = bn.apply(x, training=True) if LooseVersion( | ||
| tf.__version__) < LooseVersion('2.9.0') else bn(x, training=True) | ||
| sync_bn_func = sync_bn.apply(tf.expand_dims(x[hvd.rank()], 0), training=True) if LooseVersion( | ||
| tf.__version__) < LooseVersion('2.9.0') else sync_bn(tf.expand_dims(x[hvd.rank()], 0), training=True) |
There was a problem hiding this comment.
The code could be simplified as follows:
| bn_func = bn.apply(x, training=True) if LooseVersion( | |
| tf.__version__) < LooseVersion('2.9.0') else bn(x, training=True) | |
| sync_bn_func = sync_bn.apply(tf.expand_dims(x[hvd.rank()], 0), training=True) if LooseVersion( | |
| tf.__version__) < LooseVersion('2.9.0') else sync_bn(tf.expand_dims(x[hvd.rank()], 0), training=True) | |
| if LooseVersion(tf.__version__) < LooseVersion('2.9.0'): | |
| bn = bn.apply | |
| sync_bn = sync_bn.apply | |
| bn_func = bn(x, training=True) | |
| sync_bn_func = sync_bn(tf.expand_dims(x[hvd.rank()], 0), training=True) |
| sync_bn_func = sync_bn.apply(tf.expand_dims(x[hvd.rank()], 0), training=True) | ||
| # apply() API is removed in TF 2.9.0 | ||
| bn_func = bn.apply(x, training=True) if LooseVersion( | ||
| tf.__version__) < LooseVersion('2.9.0') else bn(x, training=True) |
There was a problem hiding this comment.
I don't think you need the version check since __ call __ is by default always there in keras.layers.
|
Didn't realize you were working on the same failure. I also opened one to fix all head failures. But feel free to merge this in after addressing the comments. |
|
@chongxiaoc in that case, I would prefer the call approach for any tensorflow version as in #3431. You can do this here and we merge #3431 afterwards. |
|
close due to duplication with #3431 |
TF 2.9 removed apply() API, use call() instead.
Signed-off-by: Chongxiao Cao chongxiaoc@uber.com
Checklist before submitting
Description
Fixes #3422 .
Review process to land