[Tensorflow] module 'tensorflow_core._api' has no attribute 'v2'

Autonomous Machines Jetson & Embedded Systems Jetson Nano
1

Finally, I’m using tf_core and solved my problem, as follows:

import tensorflow_core as tf_core
tf.compat.v1.enable_eager_execution()
tf_core.python.saved_model.load('model/trt_int8', tags=[trt.tag_constants.SERVING])
  • Jetpack 4.4 DP
  • tensorflow-2.1.0+nv20.4-cp36-cp36m-linux_aarch64

When I was trying loading my tensor rt model, some error occur:

tf.saved_model.load('model/trt_int8', tags=[trt.tag_constants.SERVING])
>>>TypeError: load() missing 1 required positional argument: 'export_dir'

Then, I found load() func require “sess”:

tf.saved_model.load()
>>>TypeError: load() missing 3 required positional arguments: 'sess', 'tags', and 'export_dir'

Enable eager execution, nothing better:

tf.compat.v1.enable_eager_execution()
tf.saved_model.load()
>>>TypeError: load() missing 3 required positional arguments: 'sess', 'tags', and 'export_dir'

Finally, I’m trying use tensorflow_core api.

import tensorflow_core as tf_core
tf_core._api.v2.saved_model.load('model/trt_int8', tags=[trt.tag_constants.SERVING])
>>> AttributeError: module 'tensorflow_core._api' has no attribute 'v2'

How to solve this issue,
Thanks.

2

What kind of model are you trying to load? A HDF5 or a SaveModel? Also did you check the documentation here? I usually use tf.keras.models.load_model to load serialized models.

3

Saved model

This code will work in my host PC.

5

Hi,

Thanks for your report.
This issue is related to TensorFlow version and the source which is built with.

Please check this comment for information:

Thanks.

1 Like