fix(conversion): load .mlpackage instead of unloadable .mlmodelc (#59)

* fix(conversion): load .mlpackage instead of unloadable .mlmodelc

The native ct.models.MLModel runtime added in the diffusers conversion
path cannot load compiled .mlmodelc directories (no Manifest.json), so
the converter's compiled output failed at load with "A valid manifest
does not exist". Both converters now return the .mlpackage directly and
the loader lists only .mlpackage. Removes the now-dead coremlcompiler
wrappers.

* chore(release): bump version to 2.0.1

coreml_suite/converter.py

@@ -1,6 +1,5 @@
 import gc
 import os
-import shutil
 import time
 
 import coremltools as ct
@@ -97,23 +96,6 @@ def get_out_path(submodule_name, model_name):
     return out_path
 
 
-def compile_coreml_model(source_model_path, output_dir, final_name):
-    """Compiles Core ML models using the coremlcompiler utility from Xcode toolchain"""
-    target_path = os.path.join(output_dir, f"{final_name}.mlmodelc")
-    if os.path.exists(target_path):
-        logger.warning(f"Found existing compiled model at {target_path}! Skipping..")
-        return target_path
-
-    logger.info(f"Compiling {source_model_path}")
-    source_model_name = os.path.basename(os.path.splitext(source_model_path)[0])
-
-    os.system(f"xcrun coremlcompiler compile {source_model_path} {output_dir}")
-    compiled_output = os.path.join(output_dir, f"{source_model_name}.mlmodelc")
-    shutil.move(compiled_output, target_path)
-
-    return target_path
-
-
 def get_sample_input(batch_size, encoder_hidden_states_shape, sample_shape):
     sample_unet_inputs = dict(
         [
@@ -338,14 +320,3 @@ def load_unet(ckpt_path, config_path):
         ckpt_path,
         original_config=config_path,
     )
-
-
-def compile_model(out_path, out_name, submodule_name):
-    from folder_paths import get_folder_paths
-
-    # Compile the model
-    target_path = compile_coreml_model(
-        out_path, get_folder_paths(submodule_name)[0], f"{out_name}_{submodule_name}"
-    )
-    logger.info(f"Compiled {out_path} to {target_path}")
-    return target_path

coreml_suite/lcm/converter.py

@@ -1,5 +1,4 @@
 import os
-import shutil
 import logging
 import time
 import gc
@@ -117,23 +116,6 @@ def get_out_path(submodule_name, model_name):
     return out_path
 
 
-def compile_coreml_model(source_model_path, output_dir, final_name):
-    """Compiles Core ML models using the coremlcompiler utility from Xcode toolchain"""
-    target_path = os.path.join(output_dir, f"{final_name}.mlmodelc")
-    if os.path.exists(target_path):
-        logger.warning(f"Found existing compiled model at {target_path}! Skipping..")
-        return target_path
-
-    logger.info(f"Compiling {source_model_path}")
-    source_model_name = os.path.basename(os.path.splitext(source_model_path)[0])
-
-    os.system(f"xcrun coremlcompiler compile {source_model_path} {output_dir}")
-    compiled_output = os.path.join(output_dir, f"{source_model_name}.mlmodelc")
-    shutil.move(compiled_output, target_path)
-
-    return target_path
-
-
 def get_sample_input(batch_size, encoder_hidden_states_shape, sample_shape, scheduler):
     sample_unet_inputs = dict(
         [
@@ -262,17 +244,6 @@ def convert(
     logger.info(f"Saved unet into {out_path}")
 
 
-def compile_model(out_path, out_name):
-    from folder_paths import get_folder_paths
-
-    # Compile the model
-    target_path = compile_coreml_model(
-        out_path, get_folder_paths("unet")[0], f"{out_name}_unet"
-    )
-    logger.info(f"Compiled {out_path} to {target_path}")
-    return target_path
-
-
 if __name__ == "__main__":
     h = 512
     w = 512
@@ -286,4 +257,3 @@ if __name__ == "__main__":
     out_path = get_out_path("unet", f"{out_name}")
     if not os.path.exists(out_path):
         convert(out_path=out_path, sample_size=sample_size, batch_size=batch_size)
-    compile_model(out_path=out_path, out_name=out_name)

coreml_suite/lcm/nodes.py

@@ -66,6 +66,5 @@ class COREML_CONVERT_LCM(COREML_NODE):
                 batch_size=batch_size,
                 controlnet_support=controlnet_support,
             )
-        target_path = lcm_converter.compile_model(out_path=out_path, out_name=out_name)
 
-        return (CoreMLModel(target_path, compute_unit),)
+        return (CoreMLModel(out_path, compute_unit),)

coreml_suite/nodes.py

@@ -167,7 +167,7 @@ class CoreMLLoader(COREML_NODE):
 
     @classmethod
     def coreml_filenames(cls):
-        extensions = (".mlmodelc", ".mlpackage")
+        extensions = (".mlpackage",)
         all_paths = folder_paths.get_filename_list_(cls.PACKAGE_DIRNAME)[1]
         coreml_paths = folder_paths.filter_files_extensions(all_paths, extensions)
 
@@ -337,11 +337,7 @@ class CoreMLConverter(COREML_NODE):
             config_path=config_path,
             quantize_nbits=quantize_nbits,
         )
-        unet_target_path = converter.compile_model(
-            out_path=unet_out_path, out_name=out_name, submodule_name="unet"
-        )
-
-        return (CoreMLModel(unet_target_path, compute_unit),)
+        return (CoreMLModel(unet_out_path, compute_unit),)
 
     @staticmethod
     def lora_path(lora_name):

pyproject.toml

@@ -1,7 +1,7 @@
 [project]
 name = "comfyui-coremlsuite"
 description = "This extension contains a set of custom nodes for ComfyUI that allow you to use Core ML models in your ComfyUI workflows."
-version = "2.0.0"
+version = "2.0.1"
 license = "MIT"
 requires-python = ">=3.12,<3.13"
 packages = [{ include = "coreml_suite" }]

uv.lock

@@ -206,7 +206,7 @@ wheels = [
 
 [[package]]
 name = "comfyui-coremlsuite"
-version = "2.0.0"
+version = "2.0.1"
 source = { virtual = "." }
 dependencies = [
     { name = "coremltools" },