Problem in retrain with universal-sentence-encoder-multilingual-qa

[丁磊]

hi,

I'm new in this field.

I have a problem in retrain a text classification model based on universal-sentence-encoder-multilingual-qa.

I find an example on the Internet to do it in this way:

    description_embeddings = hub.text_embedding_column(

        "movie_descriptions",

        module_spec="https://tfhub.dev/google/universal-sentence-encoder/2",

        trainable=True

    )

But when I modify the "module_spec" parameter with a downloaded universal-sentence-encoder-multilingual-qa path, I got such error:

Traceback (most recent call last):

  File "use_tc2.py", line 32, in <module>

    trainable=True

  File "/usr/local/lib/python3.6/dist-packages/tensorflow_hub/feature_column.py", line 106, in text_embedding_column

    key=key, module_spec_path=module_spec, trainable=trainable)

  File "/usr/local/lib/python3.6/dist-packages/tensorflow_hub/feature_column.py", line 156, in __init__

    self.module_spec = module.as_module_spec(self.module_spec_path)

  File "/usr/local/lib/python3.6/dist-packages/tensorflow_hub/module.py", line 36, in as_module_spec

    return load_module_spec(spec)

  File "/usr/local/lib/python3.6/dist-packages/tensorflow_hub/module.py", line 64, in load_module_spec

    return registry.loader(path)

  File "/usr/local/lib/python3.6/dist-packages/tensorflow_hub/registry.py", line 45, in __call__

    self._name, args, kwargs))

RuntimeError: Missing implementation that supports: loader(*('./usemodel/universal-sentence-encoder-multilingual-qa_3',), **{})

I have no idea how to solve it.

Thank you for your help