Meet error when convert lama with flexable input shape!!!

[wudijimao]

🐞Describing the bug

• Make sure to only create an issue here for bugs in the coremltools Python package. If this is a bug with the Core ML Framework or Xcode, please submit your bug here: https://developer.apple.com/bug-reporting/
• Provide a clear and consise description of the bug.

Stack Trace

• If applicable, please paste the complete stack trace. Converting model Converting PyTorch Frontend ==> MIL Ops: 6%| | 192/3332 [00:00<00:02, 1079.82 op Traceback (most recent call last): File "/Users/wudijimao/CoreMLaMa/convert_lama.py", line 50, in coreml_model = ct.convert( File "/Users/wudijimao/anaconda3/envs/lama2/lib/python3.10/site-packages/coremltools/converters/_converters_entry.py", line 574, in convert mlmodel = mil_convert( File "/Users/wudijimao/anaconda3/envs/lama2/lib/python3.10/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 "/Users/wudijimao/anaconda3/envs/lama2/lib/python3.10/site-packages/coremltools/converters/mil/converter.py", line 212, in _mil_convert proto, mil_program = mil_convert_to_proto( File "/Users/wudijimao/anaconda3/envs/lama2/lib/python3.10/site-packages/coremltools/converters/mil/converter.py", line 286, in mil_convert_to_proto prog = frontend_converter(model, **kwargs) File "/Users/wudijimao/anaconda3/envs/lama2/lib/python3.10/site-packages/coremltools/converters/mil/converter.py", line 108, in call return load(*args, **kwargs) File "/Users/wudijimao/anaconda3/envs/lama2/lib/python3.10/site-packages/coremltools/converters/mil/frontend/torch/load.py", line 80, in load return _perform_torch_convert(converter, debug) File "/Users/wudijimao/anaconda3/envs/lama2/lib/python3.10/site-packages/coremltools/converters/mil/frontend/torch/load.py", line 99, in _perform_torch_convert prog = converter.convert() File "/Users/wudijimao/anaconda3/envs/lama2/lib/python3.10/site-packages/coremltools/converters/mil/frontend/torch/converter.py", line 519, in convert convert_nodes(self.context, self.graph) File "/Users/wudijimao/anaconda3/envs/lama2/lib/python3.10/site-packages/coremltools/converters/mil/frontend/torch/ops.py", line 88, in convert_nodes add_op(context, node) File "/Users/wudijimao/anaconda3/envs/lama2/lib/python3.10/site-packages/coremltools/converters/mil/frontend/torch/ops.py", line 6091, in complex result = mb.complex(real_data=real_part, imag_data=imag_part) File "/Users/wudijimao/anaconda3/envs/lama2/lib/python3.10/site-packages/coremltools/converters/mil/mil/ops/registry.py", line 182, in add_op return cls._add_op(op_cls_to_add, **kwargs) File "/Users/wudijimao/anaconda3/envs/lama2/lib/python3.10/site-packages/coremltools/converters/mil/mil/builder.py", line 184, in _add_op new_op.type_value_inference() File "/Users/wudijimao/anaconda3/envs/lama2/lib/python3.10/site-packages/coremltools/converters/mil/mil/operation.py", line 257, in type_value_inference output_types = self.type_inference() File "/Users/wudijimao/anaconda3/envs/lama2/lib/python3.10/site-packages/coremltools/converters/mil/mil/ops/defs/complex_dialect_ops.py", line 162, in type_inference raise ValueError( ValueError: The shape of real_data ((1, 192, is26, is27)) and imag_data ((1, 192, is28, is29)) must match to construct complex data.

To Reproduce

• Please add a minimal code example that can reproduce the error when running it. base on https://github.com/mallman/CoreMLaMa

image_shape = ct.EnumeratedShapes(shapes=[
    [1, 4, 512, 512], [1, 4, 1024, 1024]],
    default=[1, 4, 512, 512])
coreml_model = ct.convert(
    jit_model,
    convert_to="mlprogram",
    inputs=[
        ct.TensorType(name="image",
                     shape=image_shape)
    ],
    outputs=[ct.TensorType(name="output")]
)

change CoreMLaMa.py to:

class CoreMLaMa(torch.nn.Module):
    def __init__(self, lama):
        super(CoreMLaMa, self).__init__()
        self.lama = lama

    def forward(self, image):
        selected_mask = image[:, 0:1, :, :]
        normalized_mask = ((selected_mask > 0) * 1).byte()
        lama_out = self.lama(image[:, 1:4, :, :], normalized_mask)
        return lama_out

• If the model conversion succeeds, but there is a numerical mismatch in predictions, please include the code used for comparisons. no, not success. Can success with fixed shape.

System environment (please complete the following information):

• coremltools version:coremltools 7.1
• OS (e.g. MacOS version or Linux type):MacOS 14.0 (23A344)
• Any other relevant version information (e.g. PyTorch or TensorFlow version):torch 2.0.1（seems old version also has problems)

Additional context

• Add anything else about the problem here that you want to share.

pip list Package Version

---

accelerate 0.24.1 aiofiles 23.2.1 altair 5.2.0 annotated-types 0.6.0 antlr4-python3-runtime 4.9.3 anyio 3.7.1 attrs 23.1.0 cattrs 23.2.3 certifi 2023.11.17 charset-normalizer 3.3.2 click 7.1.2 colorama 0.4.6 contourpy 1.2.0 coremltools 7.1 cycler 0.12.1 diffusers 0.14.0 exceptiongroup 1.2.0 fastapi 0.104.1 ffmpy 0.3.1 filelock 3.13.1 Flask 1.1.4 Flask-Cors 4.0.0 flaskwebgui 0.3.5 fonttools 4.45.1 fsspec 2023.10.0 gradio 4.7.1 gradio_client 0.7.0 h11 0.14.0 httpcore 1.0.2 httpx 0.25.2 huggingface-hub 0.19.4 idna 3.6 imageio 2.33.0 importlib-metadata 6.8.0 importlib-resources 6.1.1 iniconfig 2.0.0 itsdangerous 1.1.0 Jinja2 2.11.3 jsonschema 4.20.0 jsonschema-specifications 2023.11.1 kiwisolver 1.4.5 lama-cleaner 1.1.1 loguru 0.7.2 markdown-it-py 3.0.0 MarkupSafe 2.0.1 matplotlib 3.8.2 mdurl 0.1.2 mpmath 1.3.0 networkx 3.2.1 numpy 1.26.2 omegaconf 2.3.0 opencv-python 4.8.1.78 orjson 3.9.10 packaging 23.2 pandas 2.1.3 piexif 1.1.3 Pillow 10.1.0 pip 23.3.1 pluggy 1.3.0 protobuf 3.20.3 psutil 5.9.6 pyaml 23.9.7 pydantic 2.5.2 pydantic_core 2.14.5 pydub 0.25.1 Pygments 2.17.2 pyparsing 3.1.1 pytest 7.4.3 python-dateutil 2.8.2 python-multipart 0.0.6 pytz 2023.3.post1 PyWavelets 1.5.0 PyYAML 6.0.1 referencing 0.31.1 regex 2023.10.3 requests 2.31.0 rich 13.7.0 rpds-py 0.13.2 safetensors 0.4.1 scikit-image 0.19.3 scipy 1.11.4 semantic-version 2.10.0 setuptools 68.0.0 shellingham 1.5.4 six 1.16.0 sniffio 1.3.0 starlette 0.27.0 sympy 1.12 tifffile 2023.9.26 tokenizers 0.13.3 tomli 2.0.1 tomlkit 0.12.0 toolz 0.12.0 torch 2.0.1 tqdm 4.66.1 transformers 4.27.4 typer 0.9.0 typing_extensions 4.8.0 tzdata 2023.3 urllib3 2.1.0 uvicorn 0.24.0.post1 websockets 11.0.3 Werkzeug 1.0.1 wheel 0.41.2 whichcraft 0.6.1 yacs 0.1.8 zipp 3.17.0

[wudijimao]

any one?

[jakesabathia2]

This look like a bug in the complex op type inference.