[Question] Could the kashgari framework convert to onnx model? I try it but i find this bug

[Igoslow]

You must follow the issue template and provide as much information as possible. otherwise, this issue will be closed. 请按照 issue 模板要求填写信息。如果没有按照 issue 模板填写，将会忽略并关闭这个 issue

Environment

• OS [e.g. Mac OS, Linux]: windows
• Python Version: 3.6.8
• requirements.txt: onnx kashgari

[Paste requirements.txt file here]
kashgari                         1.0.0
kashgari-tf                      0.5.5
Keras                            2.3.1
Keras-Applications               1.0.8
keras-bert                       0.80.0
keras-embed-sim                  0.7.0
keras-gpt-2                      0.14.0
keras-layer-normalization        0.14.0
keras-multi-head                 0.22.0
keras-pos-embd                   0.11.0
keras-position-wise-feed-forward 0.6.0
Keras-Preprocessing              1.1.0
keras-self-attention             0.41.0
keras-transformer                0.31.0
keras2onnx                       1.6.0
kiwisolver                       1.1.0
Markdown                         3.1.1
MarkupSafe                       1.1.1
matplotlib                       3.1.2
mock                             3.0.5
modelarts                        1.1.3
netCDF4                          1.5.3
nltk                             3.4.5
numpy                            1.19.4
oauthlib                         3.1.0
onnx                             1.8.0
onnx-tf                          1.5.0
onnxconverter-common             1.7.0
onnxmltools                      1.7.0
onnxruntime                      1.4.0
onnxruntime-tools                1.4.0
onnxtk                           0.0.1
opencv-python                    4.2.0.32
opt-einsum                       3.1.0
packaging                        20.4
pandas                           0.25.3
Pillow                           7.0.0
pip                              20.1.1
prometheus-client                0.3.1
protobuf                         3.10.0
psutil                           5.4.6
py-cpuinfo                       7.0.0
py3nvml                          0.2.6
pyahocorasick                    1.4.0
pyasn1                           0.4.7
pyasn1-modules                   0.2.7
pycparser                        2.20
PyMySQL                          0.10.1
pyparsing                        2.4.6
pyreadline                       2.1
python-dateutil                  2.8.1
pytorch-pretrained-bert          0.6.1
pytz                             2019.3
PyYAML                           5.1.2
pyzmq                            20.0.0
regex                            2019.11.1
requests                         2.22.0
requests-oauthlib                1.2.0
rsa                              4.0
s3transfer                       0.2.1
sacremoses                       0.0.43
scikit-learn                     0.21.3
scipy                            1.4.1
sentencepiece                    0.1.91
seqeval                          0.0.10
setuptools                       41.6.0
six                              1.12.0
skl2onnx                         1.7.0
smart-open                       1.9.0
snownlp                          0.12.3
SQLAlchemy                       1.3.19
sqlparse                         0.3.1
tensorboard                      1.15.0
tensorflow                       1.15.0
tensorflow-addons                0.11.2
tensorflow-estimator             1.15.1
termcolor                        1.1.0
tf2onnx                          1.7.2
tokenizers                       0.8.1rc1
torch                            1.2.0+cpu
torchvision                      0.4.0+cpu
tqdm                             4.37.0
transformers                     2.0.0```

## Question
When i use onnx package, i find this bug like this ValueError: Unknown layer: TokenEmbedding.

[A clear and concise description of what you want to know.]

[Igoslow]

when i change the load model way, i find a bug in onnxmltools.convert_keras code as follows:

import onnxmltools from keras.models import load_model import kashgari input_keras_model = 'E:\KG\kashgari/train\gonghang_2020_12_10'

Change this path to the output name and path for the ONNX model output_onnx_model = 'kashgari_model.onnx'

Load your Keras model loaded_model = kashgari.utils.load_model(input_keras_model)

onnx_model = onnxmltools.convert_keras(loaded_model.tf_model)

Save as protobuf onnxmltools.utils.save_model(onnx_model, output_onnx_model)

This is a bug as follows: Traceback (most recent call last): File "E:/KG/kashgari/test_kashgari_onnx.py", line 26, in onnx_model = onnxmltools.convert_keras(loaded_model.tf_model) File "C:\Users\Learn\Envs\jiansuo1\lib\site-packages\onnxmltools\convert\main.py", line 33, in convert_keras return convert(model, name, doc_string, target_opset, channel_first_inputs) File "C:\Users\Learn\Envs\jiansuo1\lib\site-packages\keras2onnx\main.py", line 82, in convert_keras parse_graph(topology, tf_graph, target_opset, output_names) File "C:\Users\Learn\Envs\jiansuo1\lib\site-packages\keras2onnx\parser.py", line 828, in parse_graph return _parse_graph_core(graph, keras_layer_ts_map, topo, top_level, output_names) File "C:\Users\Learn\Envs\jiansuo1\lib\site-packages\keras2onnx\parser.py", line 746, in _parse_graph_core _on_parsing_tf_subgraph(graph, nodes, varset) File "C:\Users\Learn\Envs\jiansuo1\lib\site-packages\keras2onnx\parser.py", line 402, in _on_parsing_tf_subgraph subgraph, replacement = create_subgraph(graph, node_list, sess, operator.full_name) File "C:\Users\Learn\Envs\jiansuo1\lib\site-packages\keras2onnx\subgraph.py", line 170, in create_subgraph tf.import_graph_def(output_graph_def, name=im_scope) File "C:\Users\Learn\Envs\jiansuo1\lib\site-packages\tensorflow_core\python\util\deprecation.py", line 507, in new_func return func(*args, **kwargs) File "C:\Users\Learn\Envs\jiansuo1\lib\site-packages\tensorflow_core\python\framework\importer.py", line 405, in import_graph_def producer_op_list=producer_op_list) File "C:\Users\Learn\Envs\jiansuo1\lib\site-packages\tensorflow_core\python\framework\importer.py", line 505, in _import_graph_def_internal raise ValueError(str(e)) ValueError: Input 0 of node TFNodes50/Embedding-Token/embedding_lookup was passed float from TFNodes50/Embedding-Token/embeddings:0 incompatible with expected resource.

Process finished with exit code 1

could you help me?

[BrikerMan]

I haven't tried the onnx, Maybe try to convert it on colab, so you can rule out the environment issue first.

[Igoslow]

Thank you very much! I get it.

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.