A keras model containing a tf.tile op layer with a tensor in the `multiples` arg fails when saving to hdf5

[currivan]

Please make sure that this is a bug. As per our GitHub Policy, we only address code/doc bugs, performance issues, feature requests and build/installation issues on GitHub. tag:bug_template

System information

• Have I written custom code (as opposed to using a stock example script provided in TensorFlow): Yes
• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): MacOS 10.14.5
• Mobile device (e.g. iPhone 8, Pixel 2, Samsung Galaxy) if the issue happens on mobile device:
• TensorFlow installed from (source or binary): binary
• TensorFlow version (use command below): v2.0.0-rc0-101-gd2d2566eef 2.0.0-rc1
• Python version: Python 3.6.8 :: Anaconda, Inc.
• Bazel version (if compiling from source):
• GCC/Compiler version (if compiling from source):
• CUDA/cuDNN version:
• GPU model and memory:

You can collect some of this information using our environment capture
script
You can also obtain the TensorFlow version with: 1. TF 1.0: python -c "import tensorflow as tf; print(tf.GIT_VERSION, tf.VERSION)" 2. TF 2.0: python -c "import tensorflow as tf; print(tf.version.GIT_VERSION, tf.version.VERSION)"

Describe the current behavior

A keras model containing a tf.tile op layer with a tensor in the multiples arg throws an exception when saving to hdf5.

(I'm using tf.tile because RepeatVector(n) doesn't accept a tensor for n. The goal is to stack a 2d feature batch so it can be concatenated to a variable length 3d batch of sequence features.)

Describe the expected behavior

Model.save() should save the model.

Code to reproduce the issue

import numpy as np
import tensorflow as tf
from tensorflow.keras.layers import Input
from tensorflow.keras import Model

a = Input(shape=(10,))
out = tf.tile(a, (1, tf.shape(a)[0]))
model = Model(a, out)

x = np.zeros((50,10), dtype=np.float32)
print(model(x).numpy())

model.save('my_model.h5')

Other info / logs

---

ValueError Traceback (most recent call last)
in
11 print(model(x).numpy())
12
---> 13 model.save(model_dir + '/my_model.h5')

~/miniconda3/envs/tf20/lib/python3.6/site-packages/tensorflow_core/python/keras/engine/network.py in save(self, filepath, overwrite, include_optimizer, save_format, signatures, options)
1187 """
1188 saving.save_model(self, filepath, overwrite, include_optimizer, save_format,
-> 1189 signatures, options)
1190
1191 def save_weights(self, filepath, overwrite=True, save_format=None):

~/miniconda3/envs/tf20/lib/python3.6/site-packages/tensorflow_core/python/keras/saving/save.py in save_model(model, filepath, overwrite, include_optimizer, save_format, signatures, options)
110 'or using save_weights.')
111 hdf5_format.save_model_to_hdf5(
--> 112 model, filepath, overwrite, include_optimizer)
113 else:
114 saved_model_save.save(model, filepath, overwrite, include_optimizer,

~/miniconda3/envs/tf20/lib/python3.6/site-packages/tensorflow_core/python/keras/saving/hdf5_format.py in save_model_to_hdf5(model, filepath, overwrite, include_optimizer)
107 model_weights_group = f.create_group('model_weights')
108 model_layers = model.layers
--> 109 save_weights_to_hdf5_group(model_weights_group, model_layers)
110
111 # TODO(b/128683857): Add integration tests between tf.keras and external

~/miniconda3/envs/tf20/lib/python3.6/site-packages/tensorflow_core/python/keras/saving/hdf5_format.py in save_weights_to_hdf5_group(f, layers)
623
624 for layer in layers:
--> 625 g = f.create_group(layer.name)
626 weights = _legacy_weights(layer)
627 weight_values = K.batch_get_value(weights)

~/miniconda3/envs/tf20/lib/python3.6/site-packages/h5py/_hl/group.py in create_group(self, name, track_order)
66 name, lcpl = self._e(name, lcpl=True)
67 gcpl = Group._gcpl_crt_order if track_order else None
---> 68 gid = h5g.create(self.id, name, lcpl=lcpl, gcpl=gcpl)
69 return Group(gid)
70

h5py/_objects.pyx in h5py._objects.with_phil.wrapper()

h5py/_objects.pyx in h5py._objects.with_phil.wrapper()

h5py/h5g.pyx in h5py.h5g.create()

ValueError: Unable to create group (name already exists)

[oanush]

I was able to replicate the issue with given code for TF-2.0rc1, please find the gist of colab.Thanks!

[jvishnuvardhan]

@currivan I Don't see any issue if you remove the ".h5" from the model.save. I changed only last line from your code as follows.

import numpy as np
import tensorflow as tf
from tensorflow.keras.layers import Input
from tensorflow.keras import Model

a = Input(shape=(10,))
out = tf.tile(a, (1, tf.shape(a)[0]))
model = Model(a, out)

x = np.zeros((50,10), dtype=np.float32)
print(model(x).numpy())

model.save('./my_model')

Please check the gist here. Thanks!

I am closing this issue as it was resolved. Please feel free to open it if the issue persists again. Thanks!

[tensorflow-bot]

Are you satisfied with the resolution of your issue?
Yes
No

[currivan]

This should not have been closed. Throwing an exception here is a bug, and the solution is unacceptable. Saving this type of model with h5 takes a second, and the proposed way it takes over a minute. Loading is also unacceptably slow.