Compile a CoreML model locally (coreml.compile_model)

Declaration

compiled_path, err = coreml.compile_model(model_path)

Parameters

• model_path
  String. Path to the model to compile. iOS 15 and later support .mlpackage; iOS 12, 13, and 14 support legacy .mlmodel only.

  YOLO11 to CoreML documentation

  macOS command examples for converting YOLO11 formats

  # Assume you already have Python 3.10. If not, create one (tested with Python 3.10 only; higher versions have not been verified)
  # Install ultralytics if needed
  pip install ultralytics
  
  # On iOS 13 or iOS 14, export the legacy .mlmodel format
  yolo export format=mlmodel nms=True model=best.pt
  
  # On iOS 15+, export the .mlpackage format
  yolo export format=coreml nms=True model=best.pt
  
  # nms=True is optional
  # Enable it if you want NMS baked into the exported model
  # Skip it if you plan to use coreml.nms / coreml.rotated_nms in XXTouch

Returns

• compiled_path
  String | nil. On success, returns the compiled model directory path; on failure, returns nil.

• err
  String | nil. On success, returns nil; on failure, returns the error message.

Notes

• Available since app version 1.3.8+.

• Not supported on iOS versions below 12.

• This function compiles a .mlpackage or .mlmodel into a device-loadable .mlmodelc bundle.

• A compiled .mlmodelc bundle is not guaranteed to be portable across different devices or iOS versions.

  YOLO11 training commands on an NVIDIA GPU PC

  # Install the Visual Studio C++ Desktop Development workload if needed
  # https://visualstudio.microsoft.com/
  # Install the CUDA 11.8 Toolkit if needed
  # https://developer.nvidia.com/cuda-11-8-0-download-archive
  
  # Assume Python 3.10 is available (tested with Python 3.10 only; higher versions have not been verified)
  pip install ultralytics
  
  # Reinstall torch builds compatible with CUDA 11.8
  pip uninstall torch torchvision torchaudio
  pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu118
  
  yolo train model=yolo11m.pt data=COCO.yml epochs=300 imgsz=640 device=0

  YOLO11 training commands on macOS

  # Assume Python 3.10 is available (tested with Python 3.10 only; higher versions have not been verified)
  pip install ultralytics
  
  # Install CPU/Metal torch builds
  pip install torch torchvision torchaudio
  
  yolo train model=yolo11m.pt data=COCO.yml epochs=300 imgsz=640 device=mps

Example

compiled_model_path = XXT_HOME_PATH..'/models/yolo11m.mlmodelc' -- Where to store the compiled model

file.remove(compiled_model_path) -- Always recompile during testing; comment this out once the model is stable

if not file.exists(compiled_model_path) then -- Compile if it does not exist yet
    local tmp_path, err = coreml.compile_model(XXT_HOME_PATH..'/models/yolo11m.mlpackage')
    if not tmp_path then
        error(err)
    end
    file.move(tmp_path, compiled_model_path, 'mo')
end