wenet-online-decoder-onnx icon indicating copy to clipboard operation
wenet-online-decoder-onnx copied to clipboard

用onnxruntime-cpp推理您的onnx模型时出现了报错,想问您有没有见过或者有什么建议

Open Dsqds opened this issue 3 years ago • 9 comments

[E:onnxruntime:, sequential_executor.cc:346 Execute] Non-zero status code returned while running Slice node. Name:'Slice_38' Status Message: slice.cc:153 FillVectorsFromInput Starts must be a 1-D array terminate called after throwing an instance of 'Ort::Exception' what(): Non-zero status code returned while running Slice node. Name:'Slice_38' Status Message: slice.cc:153 FillVectorsFromInput Starts must be a 1-D array Aborted (core dumped)

Dsqds avatar Aug 26 '21 08:08 Dsqds

我没有使用过cpp跑onnx模型,不过根据 FillVectorsFromInput Starts must be a 1-D array,应该是传入的参数格式不对。我猜测是不是offset用的不是数组?( 虽然这很奇怪,但是在python代码中我传入的offset为np.array(1,dtype=int64),我不确定这时的offset类型是什么 ),可以对照着python代码细节看一下。或者是可以再检查一个各个传入的参数,在python代码中均有详细注释。

Mashiro009 avatar Aug 26 '21 09:08 Mashiro009

[E:onnxruntime:, sequential_executor.cc:346 Execute] Non-zero status code returned while running Slice node. Name:'Slice_38' Status Message: slice.cc:153 FillVectorsFromInput Starts must be a 1-D array terminate called after throwing an instance of 'Ort::Exception' what(): Non-zero status code returned while running Slice node. Name:'Slice_38' Status Message: slice.cc:153 FillVectorsFromInput Starts must be a 1-D array Aborted (core dumped)

老哥 ,请问你这个问题解决了吗,我现在在用c++来推理这个模型,要不交换一下联系方式,大家一起交流一下 @Dsqds

wyk201722 avatar Dec 03 '21 02:12 wyk201722

我也遇到同样的问题@wyk201722 @Dsqds

segmentationFaults avatar Dec 03 '21 03:12 segmentationFaults

@segmentationFaults 老哥你邮箱在多少啊,我发你我微信,一起看看代码讨论下?

wyk201722 avatar Dec 03 '21 04:12 wyk201722

将offset = torch.tensor(1,dtype=torch.int64)改为 offset = torch.LongTensor(1) offset[0]=1 就可以

lzhin avatar Dec 03 '21 04:12 lzhin

将offset = torch.tensor(1,dtype=torch.int64)改为 offset = torch.LongTensor(1) offset[0]=1 就可以

好的谢谢,我试试

wyk201722 avatar Dec 03 '21 06:12 wyk201722

将offset = torch.tensor(1,dtype=torch.int64)改为 offset = torch.LongTensor(1) offset[0]=1 就可以 这个是python的吧,有没有试过用C++推断啊

segmentationFaults avatar Dec 06 '21 08:12 segmentationFaults

就是在导出onnx模型的时候 改下offset这句 导出的模型在c++推理的时候就不会报错

lzhin avatar Dec 06 '21 08:12 lzhin

受益匪浅,我再更改源代码的时候也遇到了这个问题

nooly1998 avatar Jul 19 '22 02:07 nooly1998