norm.* -> norm

Author: a-r-r-o-w

src/diffusers/models/modeling_utils.py

@@ -353,7 +353,7 @@ def enable_layerwise_upcasting(
 
             >>> # Or, enable layerwise upcasting with custom arguments via the `apply_layerwise_upcasting` function
             >>> apply_layerwise_upcasting(
-            ...     transformer, torch.float8_e4m3fn, torch.bfloat16, skip_modules_pattern=["patch_embed", "norm.*"]
+            ...     transformer, torch.float8_e4m3fn, torch.bfloat16, skip_modules_pattern=["patch_embed", "norm"]
             ... )
             ```
 

src/diffusers/models/transformers/auraflow_transformer_2d.py

@@ -276,7 +276,7 @@ class AuraFlowTransformer2DModel(ModelMixin, ConfigMixin, FromOriginalModelMixin
     """
 
     _no_split_modules = ["AuraFlowJointTransformerBlock", "AuraFlowSingleTransformerBlock", "AuraFlowPatchEmbed"]
-    _always_upcast_modules = ["pos_embed", "norm.*"]
+    _always_upcast_modules = ["pos_embed", "norm"]
     _supports_gradient_checkpointing = True
 
     @register_to_config

src/diffusers/models/transformers/cogvideox_transformer_3d.py

@@ -212,7 +212,7 @@ class CogVideoXTransformer3DModel(ModelMixin, ConfigMixin, PeftAdapterMixin):
             Scaling factor to apply in 3D positional embeddings across temporal dimensions.
     """
 
-    _always_upcast_modules = ["patch_embed", "norm.*"]
+    _always_upcast_modules = ["patch_embed", "norm"]
     _supports_gradient_checkpointing = True
     _no_split_modules = ["CogVideoXBlock", "CogVideoXPatchEmbed"]
 

src/diffusers/models/transformers/dit_transformer_2d.py

@@ -64,7 +64,7 @@ class DiTTransformer2DModel(ModelMixin, ConfigMixin):
             A small constant added to the denominator in normalization layers to prevent division by zero.
     """
 
-    _always_upcast_modules = ["pos_embed", "norm.*"]
+    _always_upcast_modules = ["pos_embed", "norm"]
     _supports_gradient_checkpointing = True
 
     @register_to_config

src/diffusers/models/transformers/hunyuan_transformer_2d.py

@@ -244,7 +244,7 @@ class HunyuanDiT2DModel(ModelMixin, ConfigMixin):
             Whether or not to use style condition and image meta size. True for version <=1.1, False for version >= 1.2
     """
 
-    _always_upcast_modules = ["pos_embed", "norm.*", "pooler"]
+    _always_upcast_modules = ["pos_embed", "norm", "pooler"]
 
     @register_to_config
     def __init__(

src/diffusers/models/transformers/latte_transformer_3d.py

@@ -65,7 +65,7 @@ class LatteTransformer3DModel(ModelMixin, ConfigMixin):
             The number of frames in the video-like data.
     """
 
-    _always_upcast_modules = ["pos_embed", "norm.*"]
+    _always_upcast_modules = ["pos_embed", "norm"]
 
     @register_to_config
     def __init__(

src/diffusers/models/transformers/lumina_nextdit2d.py

@@ -221,7 +221,7 @@ class LuminaNextDiT2DModel(ModelMixin, ConfigMixin):
             overall scale of the model's operations.
     """
 
-    _always_upcast_modules = ["patch_embedder", "norm.*", "ffn_norm.*"]
+    _always_upcast_modules = ["patch_embedder", "norm", "ffn_norm"]
 
     @register_to_config
     def __init__(

src/diffusers/models/transformers/pixart_transformer_2d.py

@@ -79,7 +79,7 @@ class PixArtTransformer2DModel(ModelMixin, ConfigMixin):
 
     _supports_gradient_checkpointing = True
     _no_split_modules = ["BasicTransformerBlock", "PatchEmbed"]
-    _always_upcast_modules = ["pos_embed", "norm.*", "adaln_single"]
+    _always_upcast_modules = ["pos_embed", "norm", "adaln_single"]
 
     @register_to_config
     def __init__(

src/diffusers/models/transformers/sana_transformer.py

@@ -222,7 +222,7 @@ class SanaTransformer2DModel(ModelMixin, ConfigMixin, PeftAdapterMixin):
 
     _supports_gradient_checkpointing = True
     _no_split_modules = ["SanaTransformerBlock", "PatchEmbed"]
-    _always_upcast_modules = ["patch_embed", "norm.*"]
+    _always_upcast_modules = ["patch_embed", "norm"]
 
     @register_to_config
     def __init__(

src/diffusers/models/transformers/transformer_2d.py

@@ -66,7 +66,7 @@ class Transformer2DModel(LegacyModelMixin, LegacyConfigMixin):
 
     _supports_gradient_checkpointing = True
     _no_split_modules = ["BasicTransformerBlock"]
-    _always_upcast_modules = ["latent_image_embedding", "norm.*"]
+    _always_upcast_modules = ["latent_image_embedding", "norm"]
 
     @register_to_config
     def __init__(