Make torch compile node use wrapper instead of object_patch for the entire diffusion_models object, allowing key assotiations on diffusion_models to not break (loras, getting attributes, etc.)

comfy_extras/nodes_torch_compile.py

@@ -1,5 +1,40 @@
+from __future__ import annotations
 import torch
 
+from comfy.patcher_extension import WrappersMP, WrapperExecutor
+from typing import TYPE_CHECKING, Callable, Optional
+if TYPE_CHECKING:
+    from comfy.model_patcher import ModelPatcher
+
+
+COMPILE_KEY = "torch.compile"
+
+
+def apply_torch_compile_factory(compiled_diffusion_model: Callable) -> Callable:
+    def apply_torch_compile_wrapper(executor: WrapperExecutor, *args, **kwargs):
+        try:
+            orig_diffusion_model = executor.class_obj.diffusion_model
+            executor.class_obj.diffusion_model = compiled_diffusion_model
+            return executor(*args, **kwargs)
+        finally:
+            executor.class_obj.diffusion_model = orig_diffusion_model
+    return apply_torch_compile_wrapper
+
+
+def set_torch_compile_wrapper(model: ModelPatcher, backend: str, options: Optional[dict[str,str]]=None, *args, **kwargs):
+    # clear out any other torch.compile wrappers
+    model.remove_wrappers_with_key(WrappersMP.APPLY_MODEL, COMPILE_KEY)
+    # add torch.compile wrapper
+    wrapper_func = apply_torch_compile_factory(
+        torch.compile(
+                model=model.get_model_object("diffusion_model"),
+                backend=backend,
+                options=options,
+            ),
+        )
+    model.add_wrapper_with_key(WrappersMP.APPLY_MODEL, COMPILE_KEY, wrapper_func)
+
+
 class TorchCompileModel:
     @classmethod
     def INPUT_TYPES(s):
@@ -14,7 +49,7 @@ class TorchCompileModel:
 
     def patch(self, model, backend):
         m = model.clone()
-        m.add_object_patch("diffusion_model", torch.compile(model=m.get_model_object("diffusion_model"), backend=backend))
+        set_torch_compile_wrapper(model=m, backend=backend)
         return (m, )
 
 NODE_CLASS_MAPPINGS = {