unamed fused node

[AtomicVar]

I wrote a deep learning framework (like MXNet/TensorFlow) and added the ONNX exporting functionality to it.

Using my own framework, I exported a model of VGG16 and tried to run the model using ONNX Runtime. The inference is successful with basic graph optimization. If I turn on extended graph optimization, the error message is:

Traceback (most recent call last):
  File "src/onnx/orttest.py", line 10, in <module>
    session = onnxruntime.InferenceSession('vgg16.onnx', sess_options=sess_options)
  File "/root/anaconda3/envs/pytorch_gpu/lib/python3.8/site-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 283, in __init__
    self._create_inference_session(providers, provider_options, disabled_optimizers)
  File "/root/anaconda3/envs/pytorch_gpu/lib/python3.8/site-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 321, in _create_inference_session
    sess.initialize_session(providers, provider_options, disabled_optimizers)
onnxruntime.capi.onnxruntime_pybind11_state.InvalidGraph: [ONNXRuntimeError] : 10 : INVALID_GRAPH : This is an invalid model. Error in Node:fused  : Node (fused ) has input size 2 not in range [min=3, max=3].

The important line is

INVALID_GRAPH : This is an invalid model. Error in Node:fused  : Node (fused ) has input size 2 not in range [min=3, max=3].

I don't know which part of my graph is invalid. I think there is something missing after fused . Can someone give me some insights?