Current onnx parsing is not in line with the specification that states activation function number should be 1/2 for RNN and 3/6 for LSTM. Instead there are helper functions that fill in missing functions in a slightly arbitrary way, instead of reporting errors (handled that way in onnx runtime). This expands to tests that also have lower or higher number of activation functions than needed.
Current onnx parsing is not in line with the specification that states activation function number should be 1/2 for RNN and 3/6 for LSTM. Instead there are helper functions that fill in missing functions in a slightly arbitrary way, instead of reporting errors (handled that way in onnx runtime). This expands to tests that also have lower or higher number of activation functions than needed.
Cleanup done for the GRU operator during review process as part of https://github.com/ROCm/AMDMIGraphX/pull/2415