Python报错：NOT_IMPLEMENTED : Could not find an implementation for Div(1) node with name ‘Block386‘

从github上下载了一个mnist-1.onnx模型，下载地址：models/validated/vision/classification/mnist/model at main · onnx/models · GitHub

运行以下的代码之后报错

import onnx
import onnxruntime
model_path = 'my path'
session = onnxruntime.InferenceSession(model_path)

报错信息：

NOT_IMPLEMENTED : Could not find an implementation for Div(1) node with name 'Block386'

报错原因：新版的onnxruntime不支持一些旧的opset，存在类似问题的onnx模型还有

rcnn-ilsvrc13-6.onnx
squeezenet1.0-3.onnx
resnet50-caffe2-v1-3.onnx
....

 解决方式：换新版的onnx模型，改用mnist-12.onnx后不报错

参考：Deprecated old models may need to be updated · Issue #439 · onnx/models · GitHub