It looks like the way you are pruning the model in version 1 is fine; according to your error message, the resulting pruned model cannot be saved because it is not "trackable", which is a necessary condition for saving a model with tf.saved_model.save
. One way to make a trackable object is to inherit from the tf.Module
class, as described in the guides for using the SavedModel format and concrete functions. Below is an example of trying to save a tf.function
object (which fails because the object is not trackable), inheriting fromtf.module
, and saving the resulting object:
(Using Python version 3.7.6, TensorFlow version 2.1.0, and NumPy version 1.18.1)
import tensorflow as tf, numpy as np
# Define a random TensorFlow function and generate a reference output
conv_filter = tf.random.normal([1, 2, 4, 2], seed=1254)
@tf.function
def conv_model(x):
return tf.nn.conv2d(x, conv_filter, 1, "SAME")
input_tensor = tf.ones([1, 2, 3, 4])
output_tensor = conv_model(input_tensor)
print("Original model outputs:", output_tensor, sep="
")
# Try saving the model: it won't work because a tf.function is not trackable
export_dir = "./tmp/"
try: tf.saved_model.save(conv_model, export_dir)
except ValueError: print(
"Can't save {} object because it's not trackable".format(type(conv_model)))
# Now define a trackable object by inheriting from the tf.Module class
class MyModule(tf.Module):
@tf.function
def __call__(self, x): return conv_model(x)
# Instantiate the trackable object, and call once to trace-compile a graph
module_func = MyModule()
module_func(input_tensor)
tf.saved_model.save(module_func, export_dir)
# Restore the model and verify that the outputs are consistent
restored_model = tf.saved_model.load(export_dir)
restored_output_tensor = restored_model(input_tensor)
print("Restored model outputs:", restored_output_tensor, sep="
")
if np.array_equal(output_tensor.numpy(), restored_output_tensor.numpy()):
print("Outputs are consistent :)")
else: print("Outputs are NOT consistent :(")
Console output:
Original model outputs:
tf.Tensor(
[[[[-2.3629642 1.2904963 ]
[-2.3629642 1.2904963 ]
[-0.02110204 1.3400152 ]]
[[-2.3629642 1.2904963 ]
[-2.3629642 1.2904963 ]
[-0.02110204 1.3400152 ]]]], shape=(1, 2, 3, 2), dtype=float32)
Can't save <class 'tensorflow.python.eager.def_function.Function'> object
because it's not trackable
Restored model outputs:
tf.Tensor(
[[[[-2.3629642 1.2904963 ]
[-2.3629642 1.2904963 ]
[-0.02110204 1.3400152 ]]
[[-2.3629642 1.2904963 ]
[-2.3629642 1.2904963 ]
[-0.02110204 1.3400152 ]]]], shape=(1, 2, 3, 2), dtype=float32)
Outputs are consistent :)
Therefore you should try modifying your code as follows:
svmod = tf.saved_model.load(fn) #version 1
svmod2 = svmod.prune(feeds=['foo:0'], fetches=['bar:0'])
class Exportable(tf.Module):
@tf.function
def __call__(self, model_inputs): return svmod2(model_inputs)
svmod2_export = Exportable()
svmod2_export(typical_input) # call once with typical input to trace-compile
tf.saved_model.save(svmod2_export, '/tmp/saved_model/')
If you don't want to inherit from tf.Module
, you can alternatively just instantiate a tf.Module
object and add a tf.function
method/callable attribute by replacing that section of code as follows:
to_export = tf.Module()
to_export.call = tf.function(conv_model)
to_export.call(input_tensor)
tf.saved_model.save(to_export, export_dir)
restored_module = tf.saved_model.load(export_dir)
restored_func = restored_module.call