I made some more tests, trying to create a very simple model: class SimpleModel(nn.Module): def __init__(self): super().__init__() self.linear1 = nn.Linear(3, 3) self.activation1 = nn.ReLU() self.linear2 = nn.Linear(3, 1) def forward(self, x): x = self.linear1(x) x = self.activation1(x) x = self.linear2(x) return x And I could convert it with no problem. But if I try to use a layer with different shape, the outputSchema error occurs: class SimpleModel(nn.Module): def __init__(self): super().__init__() self.linear1 = nn.Linear(3, 4) self.activation1 = nn.ReLU() self.linear2 = nn.Linear(4, 1) def forward(self, x): x = self.linear1(x) x = self.activation1(x) x = self.linear2(x) return x It seems to work (with same shape) independent of the number of hidden layers. I'm doing the conversion using this code: mlmodel = ct.convert( traced_model, inputs=[ct.TensorType(name="input", shape=input.shape)], ) pipeline_network = pipeline.Pipeline ( input_features = [("input",datatypes.Array(1,3))], output_features=[("linear_1",datatypes.Array(1,1))] ) pipeline_network.add_model(mlmodel) pipeline_spec = pipeline_network.spec ct.utils.convert_double_to_float_multiarray_type(pipeline_spec) ct.utils.save_spec(pipeline_spec, "Core.mlmodel")