update

Author: a-r-r-o-w

src/diffusers/hooks/enhance_a_video.py

@@ -76,7 +76,7 @@ def __torch_function__(self, func, types, args=(), kwargs=None):
             query = kwargs.get("query", None) or args[0]
             key = kwargs.get("key", None) or args[1]
             self.query_key_save_callback(query, key)
-        return func(*args, **(kwargs or {}))
+        return func(*args, **kwargs)
 
 
 class EnhanceAVideoSDPAHook(ModelHook):

src/diffusers/models/modeling_utils.py

@@ -34,7 +34,6 @@
 from typing_extensions import Self
 
 from .. import __version__
-from ..hooks import apply_group_offloading, apply_layerwise_casting
 from ..quantizers import DiffusersAutoQuantizer, DiffusersQuantizer
 from ..quantizers.quantization_config import QuantizationMethod
 from ..utils import (
@@ -414,6 +413,7 @@ def enable_layerwise_casting(
             non_blocking (`bool`, *optional*, defaults to `False`):
                 If `True`, the weight casting operations are non-blocking.
         """
+        from ..hooks import apply_layerwise_casting
 
         user_provided_patterns = True
         if skip_modules_pattern is None:
@@ -479,6 +479,8 @@ def enable_group_offload(
             ... )
             ```
         """
+        from ..hooks import apply_group_offloading
+
         if getattr(self, "enable_tiling", None) is not None and getattr(self, "use_tiling", False) and use_stream:
             msg = (
                 "Applying group offloading on autoencoders, with CUDA streams, may not work as expected if the first "