There is an operation in the original model that uses a tensor which is also an output tensor of the graph:
example:
slimmed:
The input B is expected to be:
something like: "_normalization_2d_axis0_expanded_function_Mean2D" but the output is identified as "Mean"
The connection between these two is not made in the optimized graph.
There is an operation in the original model that uses a tensor which is also an output tensor of the graph:
example:
slimmed:
The input B is expected to be: something like: "_normalization_2d_axis0_expanded_function_Mean2D" but the output is identified as "Mean" The connection between these two is not made in the optimized graph.
The model can be found here: https://github.com/onnx/onnx/tree/main/onnx/backend/test/data/node/test_layer_normalization_2d_axis0_expanded_ver18
best regards and thanks for your great tool!