paddle.onnx. export ( layer, path, input_spec=None, opset_version=9, **configs ) [source]

Export Layer to ONNX format, which can use for inference via onnxruntime or other backends. For more details, Please refer to paddle2onnx .

  • layer (Layer) – The Layer to be exported.

  • path (str) – The path prefix to export model. The format is dirname/file_prefix or file_prefix , and the exported ONNX file suffix is .onnx .

  • input_spec (list[InputSpec|Tensor], optional) – Describes the input of the exported model’s forward method, which can be described by InputSpec or example Tensor. If None, all input variables of the original Layer’s forward method would be the inputs of the exported ONNX model. Default: None.

  • opset_version (int, optional) – Opset version of exported ONNX model. Now, stable supported opset version include 9, 10, 11. Default: 9.

  • **configs (dict, optional) – Other export configuration options for compatibility. We do not recommend using these configurations, they may be removed in the future. If not necessary, DO NOT use them. Default None. The following options are currently supported: (1) output_spec (list[Tensor]): Selects the output targets of the exported model. By default, all return variables of original Layer’s forward method are kept as the output of the exported model. If the provided output_spec list is not all output variables, the exported model will be pruned according to the given output_spec list.




import paddle

class LinearNet(paddle.nn.Layer):
    def __init__(self):
        super(LinearNet, self).__init__()
        self._linear = paddle.nn.Linear(128, 10)

    def forward(self, x):
        return self._linear(x)

# Export model with 'InputSpec' to support dynamic input shape.
def export_linear_net():
    model = LinearNet()
    x_spec = paddle.static.InputSpec(shape=[None, 128], dtype='float32')
    paddle.onnx.export(model, 'linear_net', input_spec=[x_spec])


class Logic(paddle.nn.Layer):
    def __init__(self):
        super(Logic, self).__init__()

    def forward(self, x, y, z):
        if z:
            return x
            return y

# Export model with 'Tensor' to support pruned model by set 'output_spec'.
def export_logic():
    model = Logic()
    x = paddle.to_tensor([1])
    y = paddle.to_tensor([2])
    # Static and run model.
    out = model(x, y, z=True)
    paddle.onnx.export(model, 'pruned', input_spec=[x], output_spec=[out])