[bug] `quantize_dynamic` results in `AttributeError: 'NoneType' object has no attribute 'HasField'`

[xenova]

Describe the issue

Versions:

• optimum==1.17.1
• onnx==1.15.0

Attempting to quantize a new APISR model results in the following error:

WARNING:root:Please consider to run pre-processing before quantization. Refer to example: https://github.com/microsoft/onnxruntime-inference-examples/blob/main/quantization/image_classification/cpu/ReadMe.md 
WARNING:root:Inference failed or unsupported type to quantize for tensor '/Slice_output_0', type is tensor_type {
  elem_type: 7
  shape {
    dim {
      dim_param: "unk__3"
    }
    dim {
      dim_value: 2
    }
  }
}
.
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
[<ipython-input-29-cd2ec8e65af8>](https://localhost:8080/#) in <cell line: 6>()
      4 )
      5 
----> 6 quantize_dynamic(
      7     model_input='model.onnx',
      8     model_output='output',

4 frames
[/usr/local/lib/python3.10/dist-packages/onnxruntime/quantization/quantize.py](https://localhost:8080/#) in quantize_dynamic(model_input, model_output, op_types_to_quantize, per_channel, reduce_range, weight_type, nodes_to_quantize, nodes_to_exclude, use_external_data_format, extra_options)
    640     )
    641 
--> 642     quantizer.quantize_model()
    643     quantizer.model.save_model_to_file(model_output, use_external_data_format)
    644 

[/usr/local/lib/python3.10/dist-packages/onnxruntime/quantization/onnx_quantizer.py](https://localhost:8080/#) in quantize_model(self)
    401             number_of_existing_new_nodes = len(self.new_nodes)
    402             op_quantizer = CreateOpQuantizer(self, node)
--> 403             op_quantizer.quantize()
    404             for i in range(number_of_existing_new_nodes, len(self.new_nodes)):
    405                 for output_name in self.new_nodes[i].output:

[/usr/local/lib/python3.10/dist-packages/onnxruntime/quantization/operators/base_operator.py](https://localhost:8080/#) in quantize(self)
     19         """
     20         for _, node_input in enumerate(self.node.input):
---> 21             dequantize_node = self.quantizer._dequantize_value(node_input)
     22             if dequantize_node is not None:
     23                 self.quantizer.new_nodes.append(dequantize_node)

[/usr/local/lib/python3.10/dist-packages/onnxruntime/quantization/onnx_quantizer.py](https://localhost:8080/#) in _dequantize_value(self, value_name)
   1362             ):
   1363                 # axis is not specified so scale_init must be a scalar.
-> 1364                 assert onnx.numpy_helper.to_array(scale_init).size == 1
   1365 
   1366             dqlinear_name = value_name + "_DequantizeLinear"

[/usr/local/lib/python3.10/dist-packages/onnx/numpy_helper.py](https://localhost:8080/#) in to_array(tensor, base_dir)
    182         arr: the converted array.
    183     """
--> 184     if tensor.HasField("segment"):
    185         raise ValueError("Currently not supporting loading segments.")
    186     if tensor.data_type == TensorProto.UNDEFINED:

AttributeError: 'NoneType' object has no attribute 'HasField'

NOTE: Downgrading to onnxruntime<1.16.0 and onnx==1.13.1 seems to work.

To reproduce

1. Download this model:

   wget https://huggingface.co/Xenova/4x_APISR_GRL_GAN_generator-onnx/resolve/main/onnx/model.onnx
   

2. Attempt to quantize the model

   from onnxruntime.quantization import (
       quantize_dynamic,
       QuantType
   )
   
   quantize_dynamic(
       model_input='model.onnx',
       model_output='output',
       weight_type=QuantType.QUInt8,
       extra_options=dict(
           EnableSubgraph=True
       ),
       per_channel=True,
       reduce_range=True,
   )

Note: preprocessing doesn't fix this either:

python -m onnxruntime.quantization.preprocess --input model.onnx --output model-infer.onnx

produces

unsupported broadcast between floor(floor(Pad_50_o0__d2/8)*floor(Pad_50_o0__d3/8)*floor(Pad_50_o0__d0*floor(Pad_50_o0__d2/8 + Min(4, Pad_50_o0__d2)/8 - 1/2)*floor(Pad_50_o0__d3/8 + Min(4, Pad_50_o0__d3)/8 - 1/2)/(floor(Pad_50_o0__d2/8)*floor(Pad_50_o0__d3/8)))) Pad_50_o0__d0*floor(Pad_50_o0__d2/8 + Min(4, Pad_50_o0__d2)/8 - 1/2)*floor(Pad_50_o0__d3/8 + Min(4, Pad_50_o0__d3)/8 - 1/2)
Traceback (most recent call last):
  File "/usr/lib/python3.10/runpy.py", line 196, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/usr/lib/python3.10/runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "/usr/local/lib/python3.10/dist-packages/onnxruntime/quantization/preprocess.py", line 127, in <module>
    quant_pre_process(
  File "/usr/local/lib/python3.10/dist-packages/onnxruntime/quantization/shape_inference.py", line 71, in quant_pre_process
    model = SymbolicShapeInference.infer_shapes(
  File "/usr/local/lib/python3.10/dist-packages/onnxruntime/tools/symbolic_shape_infer.py", line 2859, in infer_shapes
    raise Exception("Incomplete symbolic shape inference")
Exception: Incomplete symbolic shape inference

Urgency

Prevents me from upgrading to onnxruntime v1.17.1 to convert transformers.js models.

Platform

Linux

OS Version

Ubuntu 22.04.3 LTS

ONNX Runtime Installation

Released Package

ONNX Runtime Version or Commit ID

1.17.1

ONNX Runtime API

Python

Architecture

X64

Execution Provider

Other / Unknown

Execution Provider Library Version

n/a

[github-actions]

This issue has been automatically marked as stale due to inactivity and will be closed in 30 days if no further activity occurs. If further support is needed, please provide an update and/or more details.

[xenova]

Bump

[vidalgp]

Not sure if it is the same case, but i bumped into this issue while using SAM. Then found here the solution: facebookresearch/segment-anything#699
so actually you should downgrade onnxruntime: pip3 install onnxruntime==1.16.0

[steven-vvv]

Any progress on this? I encountered this issue again while trying to export SAM with version 1.19.2. 😭

[xadupre]

I fixed one wrong assert but the example is working for me with the development version. I still get the warning but no exception. preprocess does not work either but it seems a, independant issue.

[yihonglyu]

@xenova It should be fixed.