linear convert error

[oneTaken]

I convert the .onnx model to coreml model, And I found pooling pading mismatch bug , #7 .
And then , I found the linear layer convert error, There is missing w, just b for a linear layer, So I manully add w, and then the return MLModel(builder.spec) raises

RuntimeError: Error compiling model: "Error reading protobuf spec. validator error: Layer '250' consumes a layer named '136' which is not present in this network.".

I try to solve it by myself, and I found pytorch.onnx convert linear into Transpose and Gemm, I dived into the pytorch export file, and I got stucked in the trace part. So, I dived into onnx_coreml.convert file, and I got stucked in the 'from ..libcoremlpython import _MLModelProxy', thr libcoremlpython is .so file.

It's really a pain to use onnx_coreml, please spend some time to fix this, or tell me how to solve this.

[oneTaken]

And this is part of the graph, maybe helpful.

{
  %248 : Float(1, 1024, 1, 1) = AveragePool[kernel_shape=[7, 7], pads=[0, 0], strides=[7, 7]](%247)
  %249 : Float(1, 1024) = Reshape[shape=[1, -1]](%248)
  %250 : Float(1024!, 1024!) = Transpose[perm=[1, 0]](%136)
  %252 : Float(1, 1024) = Gemm[alpha=1, beta=1, broadcast=1](%249, %250, %137)
  %253 : Float(1, 1024) = Relu(%252)
  %254 : Float(1, 1024), %255 : UNKNOWN_TYPE = Dropout[is_test=1, ratio=0.5](%253)
  %256 : Float(1024!, 17!) = Transpose[perm=[1, 0]](%138)
  %258 : Float(1, 17) = Gemm[alpha=1, beta=1, broadcast=1](%254, %256, %139)
  return (%258);
}

[bhushan23]

@oneTaken could you please give a try with latest beta?
https://github.com/onnx/onnx-coreml#new-beta-onnx-coreml-converter-with-core-ml-3

feel free to re-open if you still face this issue