Currently, all input tensors in generated ONNX models are labeled "data". Since ONNX supports multiple inputs, this creates conflicts. For example, the following graph would be exported as having two separate inputs:
g = taso.new_graph()
i = g.new_input(dims=(5, 5, 10, 10))
g.add(i, i)
This PR fixes this issue by giving all inputs a unique name.
Currently, all input tensors in generated ONNX models are labeled "data". Since ONNX supports multiple inputs, this creates conflicts. For example, the following graph would be exported as having two separate inputs:
This PR fixes this issue by giving all inputs a unique name.