[go: nahoru, domu]
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

TFLite fails to convert when the Model has regex_split_with_offsets #51134

Closed
cpoohee opened this issue Aug 3, 2021 · 4 comments
Closed

TFLite fails to convert when the Model has regex_split_with_offsets #51134

cpoohee opened this issue Aug 3, 2021 · 4 comments
Assignees
@abattery @sushreebarsa
Labels
TFLiteConverter For issues related to TFLite converter type:support Support issues

Comments

@cpoohee
Copy link
cpoohee commented Aug 3, 2021

1. System information

  • OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Ubuntu 20.04
  • TensorFlow installation (pip package or built from source): pip package
  • TensorFlow library (version, if pip package or github SHA, if built from source): 2.5.0

2. Code

Provide code to help us reproduce your issues using one of the following options:

def tokenize(self, x):
    x = tf_text.case_fold_utf8(x)
    x = tf_text.normalize_utf8(x, "NFD")
    x = tf.strings.regex_replace(x, r"\p{Mn}", "")
    x = tf.strings.regex_replace(x, r"\p{Cc}|\p{Cf}", " ")
    x,_ ,_ = tf_text.regex_split_with_offsets(
        x, self._delim_regex_pattern, self._keep_delim_regex_pattern,
        "BasicTokenizer")
    return x

Testing with a simpler delim and keep_delim inputs will also result in the conversion error.
The full model is in the following colab gist.

Reference To [TensorFlow Model, and tflite conversion Colab] Colab gist

In the gist, the model is shown to work before and after saving to SavedModel.
The code attempts to convert to tflite from SavedModel.

3. Failure during conversion

TFLite Fails to convert when model contains regex_split_with_offsets ops.
tensorflow_text is imported,
Select Ops flex option is enabled. such as:
converter.target_spec.supported_ops = [ tf.lite.OpsSet.TFLITE_BUILTINS, tf.lite.OpsSet.SELECT_TF_OPS ]

In the example, there are other tf_text ops before problematic regex_split_with_offsets.
When commenting away the line with tf_text.regex_split_with_offsets, it will convert successfully.

4. (optional) Any other info / logs

The error prompt says the "Graph does not contain node: ". Full trace logs is available below and in the gist.

 ---------------------------------------------------------------------------
Exception                                 Traceback (most recent call last)
/usr/local/lib/python3.7/dist-packages/tensorflow/lite/python/convert.py in toco_convert_protos(model_flags_str, toco_flags_str, input_data_str, debug_info_str, enable_mlir_converter)
    293                                                  debug_info_str,
--> 294                                                  enable_mlir_converter)
    295       return model_str

4 frames
/usr/local/lib/python3.7/dist-packages/tensorflow/lite/python/wrap_toco.py in wrapped_toco_convert(model_flags_str, toco_flags_str, input_data_str, debug_info_str, enable_mlir_converter)
     37       debug_info_str,
---> 38       enable_mlir_converter)
     39 

Exception: Graph does not contain node: 

During handling of the above exception, another exception occurred:

ConverterError                            Traceback (most recent call last)
<ipython-input-82-51fa59e54032> in <module>()
      5   tf.lite.OpsSet.SELECT_TF_OPS # enable TensorFlow ops.
      6 ]
----> 7 tflite_model = converter.convert()
      8 with open(tfLite_filepath, 'wb') as f:
      9   f.write(tflite_model)

/usr/local/lib/python3.7/dist-packages/tensorflow/lite/python/lite.py in convert(self)
    911     converter_kwargs.update(quant_mode.converter_flags())
    912 
--> 913     result = _convert_saved_model(**converter_kwargs)
    914     if self.experimental_new_quantizer:
    915       calibrate_and_quantize, flags = quant_mode.quantizer_flags(

/usr/local/lib/python3.7/dist-packages/tensorflow/lite/python/convert.py in convert_saved_model(saved_model_dir, saved_model_version, saved_model_tags, saved_model_exported_names, **kwargs)
    725       None,  # input_data, unused
    726       None,  # debug_info_str, unused
--> 727       enable_mlir_converter=True)
    728   return data
    729 

/usr/local/lib/python3.7/dist-packages/tensorflow/lite/python/convert.py in toco_convert_protos(model_flags_str, toco_flags_str, input_data_str, debug_info_str, enable_mlir_converter)
    295       return model_str
    296     except Exception as e:
--> 297       raise ConverterError(str(e))
    298 
    299   if distutils.spawn.find_executable(_toco_from_proto_bin) is None:

ConverterError: Graph does not contain node:
@cpoohee cpoohee added the TFLiteConverter For issues related to TFLite converter label Aug 3, 2021
@abattery abattery self-assigned this Aug 3, 2021
@abattery abattery added the type:bug Bug label Aug 4, 2021
@abattery
Copy link
Contributor
abattery commented Aug 4, 2021

I can reproduce this issue. Thanks for the report!

@abattery
Copy link
Contributor
abattery commented Aug 4, 2021

Ragged Tensors are not a simple tensor type. Returning objects should be one of TensorFlow Lite types. Just returning a Ragged Tensor object is not supported.

For example,

   x,_ ,_ = tf_text.regex_split_with_offsets(
        x, self._delim_regex_pattern, self._keep_delim_regex_pattern,
        "BasicTokenizer")
    return {"values": x.values, "shapes": x.shapes}

@abattery abattery closed this as completed Aug 4, 2021
@google-ml-butler
Copy link

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

@abattery abattery added type:support Support issues and removed type:bug Bug labels Aug 4, 2021
@cpoohee
Copy link
Author
cpoohee commented Aug 4, 2021 •
edited
Loading

Thank you for your hints. For those interested, I am able to successfully convert to tflite after introducing the line after regex_split_with_offsets with,
x = x.to_tensor()

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees

@abattery abattery

@sushreebarsa sushreebarsa

Labels
TFLiteConverter For issues related to TFLite converter type:support Support issues
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
3 participants
@cpoohee @abattery @sushreebarsa