Torch.onnx.export a fp16 model but get the output tensor fp32

1

Hi there，i’ve just upgraded my torch from version 1.12.1 to 2.0.1.
however, i get the problem as mentioned in the title, that the output tensor in the exported onnx become a fp32 one with the others fp16.
the code as follows,
torch.onnx.export(
deploy, opset_version=11, args=(dummy_a, dummy_b), f=‘export.onnx’,
input_names=[“a”, “b”], output_names=[“c”], verbose=False
)