tensorflow_tts 中的fastspeech2 模型在转为onnx之后再转为mnn出现了 Failed: Loop don't support optional M and cond input

[dshwei]

平台(如果交叉编译请再附上交叉编译目标平台):

Platform(Include target platform as well if cross-compiling):

Github版本:

Github Version:

2.4.0

编译方式:

Compiling Method

cmake and make
https://mnn-docs.readthedocs.io/en/latest/compile/tools.html

cmake .. -DMNN_BUILD_CONVERTER=ON -DMNN_BUILD_TORCH=ON

编译日志:

Build Log:

./MNNConvert -f ONNX --modelFile fastspeech2/fastspeech2_onnx_model.onnx --MNNModel fastspeech2_onnx_model.mnn --bizCode biz

2023-03-09 16:43:03.632028609 [W:onnxruntime:, graph.cc:3490 CleanUnusedInitializersAndNodeArgs] Removing initializer 'zero_const__60'. It is not used by any node and should be removed from the model.
initial inputs value, input type : tensor(int32)
input name input_1 type is int32
initial inputs value, input type : tensor(int32)
input name input_2 type is int32
initial inputs value, input type : tensor(float)
input name input_3 type is float32
initial inputs value, input type : tensor(float)
input name input_4 type is float32
initial inputs value, input type : tensor(float)
input name input_5 type is float32
model graph output name :  ['output_1', 'output_2', 'output_3', 'output_4', 'output_5']
inputs: [{'name': 'input_1', 'shape': [1, 10]}, {'name': 'input_2', 'shape': [1]}, {'name': 'input_3', 'shape': [1]}, {'name': 'input_4', 'shape': [1]}, {'name': 'input_5', 'shape': [1]}]
【input key and value write into file 】
inputs file name :onnx/input_1.txt 
inputs file name :onnx/input_2.txt 
inputs file name :onnx/input_3.txt 
inputs file name :onnx/input_4.txt 
inputs file name :onnx/input_5.txt 
outputs: ['output_1', 'output_2', 'output_3', 'output_4', 'output_5']
output file name :onnx/output_1.txt , output shape (1, 64, 80),output type float32
output file name :onnx/output_2.txt , output shape (1, 64, 80),output type float32
output file name :onnx/output_3.txt , output shape (1, 10),output type int32
output file name :onnx/output_4.txt , output shape (1, 10),output type float32
output file name :onnx/output_5.txt , output shape (1, 10),output type float32
-----------------test TTS_tf2/Fastspeech2/model_dir/fastspeech2/fastspeech2_onnx_model.onnx onnx model done -------------
Segmentation fault (core dumped)
【convert to mnn model result 】: The device support dot:0, support fp16:0, support i8mm: 0
Start to Convert Other Model Format To MNN Model...
[16:43:04] MNN-2.4.0/tools/converter/source/onnx/onnxConverter.cpp:40: ONNX Model ir version: 7
[16:43:04] MNN-2.4.0/tools/converter/source/onnx/onnxConverter.cpp:108: Check it out ==> savable_tf_fast_speech2/length_regulator/while_loop:0 has empty input, the index is 0
Failed: Loop don't support optional M and cond input
Start to Optimize the MNN Net...

[wangzhaode]

能否提供对应的ONNX模型？

[jxt1234]

2.4.2 修正