How to support Python build-in functions like len()?

[karpov3]

❓How to support Python build-in functions like len()?

System Information

• pytorch: 1.9.1
• coremltools: 1.5.1

I am trying to convert my model in Core ML with Coremltools. I had some issue that I replicated in this simple example.

import torch
import torch.nn as nn
import coremltools

class SimpleTest(nn.Module):
    def forward(self, x):
        value = len(x)
        return torch.tensor(value)

simple_model = SimpleTest()

scripted_model = torch.jit.script(simple_model)
mlmodel = coremltools.converters.convert(scripted_model, inputs=[coremltools.TensorType(shape=(400,))], debug=True)

I have this error.

debug=True info

the following model ops are IMPLEMENTED:
  constant
  tensor
the following model ops are MISSING:
  len

How can I use Python len() funcion inside the model?

[TobyRoseman]

coremltools.converters.mil.mil.ops.defs.control_flow.list_length is probably the right mil op to use here.

[JRGit4UE]

<newbie> .. how would I add the mil op to the TorchScript side ??

#  coremltools 5.1.0
from coremltools.converters.mil.mil.ops.defs.control_flow import list_length
list_length([ ]) #TypeError: __init__() takes 1 positional argument but 2 were given

</newbie>

[karpov3]

@TobyRoseman thank you for your suggestion. I'm wondering if there is some example how to use it? In my case i have this error that i don't know how to manage.