AttributeError: 'list' object has no attribute 'val'

[deepakHonakeri05]

🐞Describing the bug

• When I try to convert torchscript model to core ML model using coremltools. I get the following bug:
• 'AttributeError: 'list' object has no attribute 'val'

Stack Trace

  File "/Users/deepak/startup/SnapTagger/ai/app.py", line 33, in <module>
    vision_model = ct.convert(
                   ^^^^^^^^^^^
  File "/opt/homebrew/lib/python3.11/site-packages/coremltools/converters/_converters_entry.py", line 574, in convert
    mlmodel = mil_convert(
              ^^^^^^^^^^^^
  File "/opt/homebrew/lib/python3.11/site-packages/coremltools/converters/mil/converter.py", line 188, in mil_convert
    return _mil_convert(model, convert_from, convert_to, ConverterRegistry, MLModel, compute_units, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/homebrew/lib/python3.11/site-packages/coremltools/converters/mil/converter.py", line 212, in _mil_convert
    proto, mil_program = mil_convert_to_proto(
                         ^^^^^^^^^^^^^^^^^^^^^
  File "/opt/homebrew/lib/python3.11/site-packages/coremltools/converters/mil/converter.py", line 286, in mil_convert_to_proto
    prog = frontend_converter(model, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/homebrew/lib/python3.11/site-packages/coremltools/converters/mil/converter.py", line 108, in __call__
    return load(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/opt/homebrew/lib/python3.11/site-packages/coremltools/converters/mil/frontend/torch/load.py", line 80, in load
    return _perform_torch_convert(converter, debug)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/homebrew/lib/python3.11/site-packages/coremltools/converters/mil/frontend/torch/load.py", line 99, in _perform_torch_convert
    prog = converter.convert()
           ^^^^^^^^^^^^^^^^^^^
  File "/opt/homebrew/lib/python3.11/site-packages/coremltools/converters/mil/frontend/torch/converter.py", line 519, in convert
    convert_nodes(self.context, self.graph)
  File "/opt/homebrew/lib/python3.11/site-packages/coremltools/converters/mil/frontend/torch/ops.py", line 88, in convert_nodes
    add_op(context, node)
  File "/opt/homebrew/lib/python3.11/site-packages/coremltools/converters/mil/frontend/torch/ops.py", line 1665, in pad
    if pad.val is not None:
       ^^^^^^^
AttributeError: 'list' object has no attribute 'val'

To Reproduce

from transformers import ViltProcessor, ViltForQuestionAnswering
from PIL import Image
import sys
import torch, os, psutil
import coremltools as ct

processor = ViltProcessor.from_pretrained("MariaK/vilt_finetuned_200")

image = Image.open(sys.argv[1])
question = "Where is the man?"

inputs = processor(image, question, return_tensors="pt")

model = ViltForQuestionAnswering.from_pretrained("MariaK/vilt_finetuned_200")

with torch.no_grad():
    outputs = model(**inputs)

logits = outputs.logits
idx = logits.argmax(-1).item()
print("Predicted answer:", model.config.id2label[idx])


traced_vision_model = torch.jit.trace(model, [inputs.input_ids, inputs.attention_mask, inputs.token_type_ids,  inputs.pixel_values, inputs.pixel_mask], strict= False)
torch.jit.save(traced_vision_model, "traced_vision_model.pt")
loaded_model = torch.jit.load("traced_vision_model.pt")
vision_model = ct.convert(
    loaded_model,
    inputs=[
        ct.TensorType(shape = inputs.input_ids.shape, name="input_ids"),
        ct.TensorType(shape = inputs.attention_mask.shape, name="attention_mask"),
        ct.TensorType(shape = inputs.token_type_ids.shape, name="token_type_ids"),
        ct.TensorType(shape = inputs.pixel_values.shape, name="pixel_values"),
        ct.TensorType(shape = inputs.pixel_mask.shape, name="pixel_mask"),
    ],
    minimum_deployment_target=ct.target.iOS17,
 )
vision_model.save("vilt.mlmodel")

System environment (please complete the following information):

• coremltools version:
• OS (e.g. MacOS version or Linux type): macOS Sonoma 14.3.1
• Any other relevant version information (e.g. PyTorch or TensorFlow version): PyTorch Version 2.1.0 and 2.2.1

[TobyRoseman]

Are you using the most recent version of coremltools?

Loading an untrusted PyTorch model is a security risk. So I'm not able to run your code.

If this is still an issue with the most recent version of coremltools, please give us self-contained code (i.e. doesn't load an external model) to reproduce the issue.

[deepakHonakeri05]

Yes I'm using coremltools v7.1

Ok I will try to provide a self-contained code

[w4-jihunlorenzopark]

Hello, I am facing the same error. In my case, pad function called inside for loop causes the error. I am not sure this is the same cause with the @deepakHonakeri05's case, though. hope this is the only cause. Here is the self-contained code. Could you check this out?

I am using coremltools==7.2

import torch
import torch.nn as nn
import coremltools as ct


@torch.jit.script
def pad(x):
    
    for i in range(3):
        x = torch.nn.functional.pad(x, (0, x.size(-1)))
    return x

# # It works without for loop
# @torch.jit.script
# def pad(x):
#     x = torch.nn.functional.pad(x, (0, x.size(-1)))
#     return x


class SimpleModule(nn.Module):
    def __init__(self):
        super().__init__()

    def forward(self, x):
        padded_tensor = pad(x)
        return padded_tensor

module = SimpleModule()
module.eval()

example_input = torch.rand(3, 4, 4)
traced_module = torch.jit.trace(module, example_input)
model = ct.convert(
    traced_module,
    convert_to="mlprogram",
    inputs=[ct.TensorType(shape=example_input.shape)]
 )

Here is the TorchScript graph.

print(pad.graph)

graph(%x.1 : Tensor):
  %18 : Function = prim::Constant[name="pad"]()
  %17 : NoneType = prim::Constant()
  %16 : str = prim::Constant[value="constant"]()
  %12 : int = prim::Constant[value=-1]() # /var/folders/c5/yknsh7rd1jz7schyq8qt4lp8000507/T/ipykernel_55333/923301787.py:9:50
  %6 : bool = prim::Constant[value=1]() # /var/folders/c5/yknsh7rd1jz7schyq8qt4lp8000507/T/ipykernel_55333/923301787.py:8:4
  %2 : int = prim::Constant[value=0]() # /var/folders/c5/yknsh7rd1jz7schyq8qt4lp8000507/T/ipykernel_55333/923301787.py:8:26
  %3 : int = aten::size(%x.1, %2) # /var/folders/c5/yknsh7rd1jz7schyq8qt4lp8000507/T/ipykernel_55333/923301787.py:8:19
  %x : Tensor = prim::Loop(%3, %6, %x.1) # /var/folders/c5/yknsh7rd1jz7schyq8qt4lp8000507/T/ipykernel_55333/923301787.py:8:4
    block0(%i : int, %x.15 : Tensor):
      %13 : int = aten::size(%x.15, %12) # /var/folders/c5/yknsh7rd1jz7schyq8qt4lp8000507/T/ipykernel_55333/923301787.py:9:43
      %15 : int[] = prim::ListConstruct(%2, %13)
      %x.9 : Tensor = prim::CallFunction(%18, %x.15, %15, %16, %17) # /var/folders/c5/yknsh7rd1jz7schyq8qt4lp8000507/T/ipykernel_55333/923301787.py:9:12
      -> (%6, %x.9)
  return (%x)

[TobyRoseman]

Thanks @w4-jihunlorenzopark - I can reproduce the bug using your code.