onnxruntime: unclear error message: INVALID_ARGUMENT : Unexpected input data type. Actual: (N11onnxruntime17PrimitiveDataTypeIlEE) , expected: (N11onnxruntime17PrimitiveDataTypeIiEE)

Hi,

I have a fine-tuned BERT model that has the following shape.

graph_name: tf_bert_for_multi_classification
domain: onnxmltools
description: 
input 0: "attention_mask" ["N", 7] Int32
input 1: "input_ids" ["N", 7] Int32
input 2: "token_type_ids" ["N", 7] Int32
output 0: "output_1" ["N", 4404, 1] Float

when I try to run the model with

results = session.run(None, inputs_onnx)

where inputs_onnx is

{'input_ids': array([ 101,  146, 1169, 1631, 1103, 3974,  117, 1169, 1128,  136,  102]),
 'token_type_ids': array([0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]),
 'attention_mask': array([1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1])}

I got the following error.

>>> results = session.run(None, inputs_onnx)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/nix/store/xws61xnjc03fjiwfh7ci5cwgg1chmp3l-python3.7-onnxruntime-1.4.0/lib/python3.7/site-packages/onnxruntime/capi/session.py", line 110, in run
    return self._sess.run(output_names, input_feed, run_options)
onnxruntime.capi.onnxruntime_pybind11_state.InvalidArgument: [ONNXRuntimeError] : 2 : INVALID_ARGUMENT : Unexpected input data type. Actual: (N11onnxruntime17PrimitiveDataTypeIlEE) , expected: (N11onnxruntime17PrimitiveDataTypeIiEE)

It is really hard for me to debug because it is not clear from the error message what’s wrong. I can’t find source code covering N11onnxruntime17PrimitiveDataTypeIiEE or N11onnxruntime17PrimitiveDataTypeIlEE

About this issue

Original URL
State: closed
Created 4 years ago
Comments: 16 (4 by maintainers)

Most upvoted comments

Well - you are right - symbolic dimension could be treated as 1 - and hence the shape of the input will then be [1, 7]. Shape = [1, 7] -> Rank = 2.

What you are feeding in is - shape = [7] -> Rank = 1. That is the complaint from ORT.

In Python, the input would look like this -> [[1, 1, 1, 1, 1, 1, 1]] (rank 2) as opposed to [1, 1, 1, 1, 1, 1, 1,] (rank 1).

Hope this helps

hariharans29 on Aug 17, 2020

I’m running into similar problems.

for me, the problem was a bug in the transformers library exporting TF BERT models to ONNX. I had to export the model in a different way.

@Zhen-hao Can you explain, which different way you exported your model to ONNX, though?

I didn’t figure out a way to export it to ONNX. I had to export it as a TF model.

Zhen-hao on Dec 16, 2020