ValueError/tensorflow InvalidArgumentError while trying to run README demo

[boronj]

While trying to run the following line of code on Google Colab:

k = subprocess.run([sys.executable, "/content/voca/run_voca.py", "--tf_model_fname", "/content/voca/model/gstep_52280.model", "--ds_fname", "/content/voca/ds_graph/output_graph.pb", "--audio_fname", f"{output_path}/output{x}.wav", "--template_fname", "/content/voca/template/FLAME_sample.ply", "--condition_idx", "3", "--uv_template_fname", "/content/voca/template/texture_mesh.obj", "--texture_img_fname", "/content/voca/template/texture_mesh.png", "--out_path", f"/content/data/outputs/{DATASET}_{SPLIT}/"], capture_output=True)

I get the following traceback:

  File "/usr/local/lib/python3.12/dist-packages/tensorflow/python/framework/importer.py", line 523, in _import_graph_def_internal
    results = graph_import_graphdef(c_graph, graph_def_input, options)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
tensorflow.python.framework.errors_impl.InvalidArgumentError: Node 'VOCA/gradients/VOCA/SpeechEncoder/batch_norm_1/cond/FusedBatchNorm_1_grad/FusedBatchNormGrad' has an _output_shapes attribute inconsistent with the GraphDef for output #3: Dimension 0 in both shapes must be equal, but are 0 and 1. Shapes are [0] and [1].

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/content/voca/run_voca.py", line 59, in <module>
    inference(tf_model_fname, ds_fname, audio_fname, template_fname, condition_idx, out_path, str2bool(args.visualize), uv_template_fname=uv_template_fname, texture_img_fname=texture_img_fname)
  File "/content/voca/utils/inference.py", line 114, in inference
    saver = tf.train.import_meta_graph(tf_model_fname + '.meta')
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/tensorflow/python/training/saver.py", line 1587, in import_meta_graph
    return _import_meta_graph_with_return_elements(meta_graph_or_file,
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/tensorflow/python/training/saver.py", line 1608, in _import_meta_graph_with_return_elements
    meta_graph.import_scoped_meta_graph_with_return_elements(
  File "/usr/local/lib/python3.12/dist-packages/tensorflow/python/framework/meta_graph.py", line 785, in import_scoped_meta_graph_with_return_elements
    imported_return_elements = importer.import_graph_def(
                               ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/tensorflow/python/util/deprecation.py", line 588, in new_func
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/tensorflow/python/framework/importer.py", line 411, in import_graph_def
    return _import_graph_def_internal(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/tensorflow/python/framework/importer.py", line 527, in _import_graph_def_internal
    raise ValueError(str(e))
ValueError: Node 'VOCA/gradients/VOCA/SpeechEncoder/batch_norm_1/cond/FusedBatchNorm_1_grad/FusedBatchNormGrad' has an _output_shapes attribute inconsistent with the GraphDef for output #3: Dimension 0 in both shapes must be equal, but are 0 and 1. Shapes are [0] and [1].

I'm assuming this is an issue with DeepSpeech per #27, which is weird as I've installed deepspeech-0.1.0 through ./fetch_data.sh.