update

Author: DN6

src/diffusers/models/attention.py

@@ -12,19 +12,15 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+from typing import Callable, Dict, Optional, Tuple, Union
 
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
 
-from ..utils import deprecate, logging
+from ..utils import logging
 from ..utils.import_utils import is_torch_npu_available, is_torch_xla_available, is_xformers_available
-from ..utils.torch_utils import maybe_allow_in_graph
-from .activations import GEGLU, GELU, ApproximateGELU, FP32SiLU, LinearActivation, SwiGLU
-from .attention_processor import Attention, AttentionProcessor, JointAttnProcessor2_0
-from .embeddings import SinusoidalPositionalEmbedding
-from .normalization import AdaLayerNorm, AdaLayerNormContinuous, AdaLayerNormZero, RMSNorm, SD35AdaLayerNormZeroX
+from .attention_processor import AttentionProcessor
 
 
 if is_xformers_available():
@@ -511,78 +507,130 @@ def norm_encoder_hidden_states(self, encoder_hidden_states: torch.Tensor) -> tor
 
 def _chunked_feed_forward(*args, **kwargs):
     """Backward compatibility stub. Use transformers.modeling_common._chunked_feed_forward instead."""
+    logger.warning(
+        "Importing `_chunked_feed_forward` from `diffusers.models.attention` is deprecated and will be removed in a future version. "
+        "Please use `from diffusers.models.transformers.modeling_common import _chunked_feed_forward` instead."
+    )
     from .transformers.modeling_common import _chunked_feed_forward as _actual_chunked_feed_forward
+
     return _actual_chunked_feed_forward(*args, **kwargs)
 
 
 class GatedSelfAttentionDense:
     r"""
     Backward compatibility stub. Use transformers.modeling_common.GatedSelfAttentionDense instead.
     """
+
     def __new__(cls, *args, **kwargs):
+        logger.warning(
+            "Importing `GatedSelfAttentionDense` from `diffusers.models.attention` is deprecated and will be removed in a future version. "
+            "Please use `from diffusers.models.transformers.modeling_common import GatedSelfAttentionDense` instead."
+        )
         from .transformers.modeling_common import GatedSelfAttentionDense
+
         return GatedSelfAttentionDense(*args, **kwargs)
 
 
 class JointTransformerBlock:
     r"""
     Backward compatibility stub. Use transformers.modeling_common.JointTransformerBlock instead.
     """
+
     def __new__(cls, *args, **kwargs):
+        logger.warning(
+            "Importing `JointTransformerBlock` from `diffusers.models.attention` is deprecated and will be removed in a future version. "
+            "Please use `from diffusers.models.transformers.modeling_common import JointTransformerBlock` instead."
+        )
         from .transformers.modeling_common import JointTransformerBlock
+
         return JointTransformerBlock(*args, **kwargs)
 
 
 class BasicTransformerBlock:
     r"""
     Backward compatibility stub. Use transformers.modeling_common.BasicTransformerBlock instead.
     """
+
     def __new__(cls, *args, **kwargs):
+        logger.warning(
+            "Importing `BasicTransformerBlock` from `diffusers.models.attention` is deprecated and will be removed in a future version. "
+            "Please use `from diffusers.models.transformers.modeling_common import BasicTransformerBlock` instead."
+        )
         from .transformers.modeling_common import BasicTransformerBlock
+
         return BasicTransformerBlock(*args, **kwargs)
 
 
 class LuminaFeedForward:
     r"""
     Backward compatibility stub. Use transformers.modeling_common.LuminaFeedForward instead.
     """
+
     def __new__(cls, *args, **kwargs):
+        logger.warning(
+            "Importing `LuminaFeedForward` from `diffusers.models.attention` is deprecated and will be removed in a future version. "
+            "Please use `from diffusers.models.transformers.modeling_common import LuminaFeedForward` instead."
+        )
         from .transformers.modeling_common import LuminaFeedForward
+
         return LuminaFeedForward(*args, **kwargs)
 
 
 class TemporalBasicTransformerBlock:
     r"""
     Backward compatibility stub. Use transformers.modeling_common.TemporalBasicTransformerBlock instead.
     """
+
     def __new__(cls, *args, **kwargs):
+        logger.warning(
+            "Importing `TemporalBasicTransformerBlock` from `diffusers.models.attention` is deprecated and will be removed in a future version. "
+            "Please use `from diffusers.models.transformers.modeling_common import TemporalBasicTransformerBlock` instead."
+        )
         from .transformers.modeling_common import TemporalBasicTransformerBlock
+
         return TemporalBasicTransformerBlock(*args, **kwargs)
 
 
 class SkipFFTransformerBlock:
     r"""
     Backward compatibility stub. Use transformers.modeling_common.SkipFFTransformerBlock instead.
     """
+
     def __new__(cls, *args, **kwargs):
+        logger.warning(
+            "Importing `SkipFFTransformerBlock` from `diffusers.models.attention` is deprecated and will be removed in a future version. "
+            "Please use `from diffusers.models.transformers.modeling_common import SkipFFTransformerBlock` instead."
+        )
         from .transformers.modeling_common import SkipFFTransformerBlock
+
         return SkipFFTransformerBlock(*args, **kwargs)
 
 
 class FreeNoiseTransformerBlock:
     r"""
     Backward compatibility stub. Use transformers.modeling_common.FreeNoiseTransformerBlock instead.
     """
+
     def __new__(cls, *args, **kwargs):
+        logger.warning(
+            "Importing `FreeNoiseTransformerBlock` from `diffusers.models.attention` is deprecated and will be removed in a future version. "
+            "Please use `from diffusers.models.transformers.modeling_common import FreeNoiseTransformerBlock` instead."
+        )
         from .transformers.modeling_common import FreeNoiseTransformerBlock
-        return FreeNoiseTransformerBlock(*args, **kwargs)
 
+        return FreeNoiseTransformerBlock(*args, **kwargs)
 
 
 class FeedForward:
     r"""
     Backward compatibility stub. Use transformers.modeling_common.FeedForward instead.
     """
+
     def __new__(cls, *args, **kwargs):
+        logger.warning(
+            "Importing `FeedForward` from `diffusers.models.attention` is deprecated and will be removed in a future version. "
+            "Please use `from diffusers.models.transformers.modeling_common import FeedForward` instead."
+        )
         from .transformers.modeling_common import FeedForward
+
         return FeedForward(*args, **kwargs)

src/diffusers/models/transformers/dit_transformer_2d.py

@@ -19,10 +19,10 @@
 
 from ...configuration_utils import ConfigMixin, register_to_config
 from ...utils import logging
-from .modeling_common import BasicTransformerBlock
 from ..embeddings import PatchEmbed
 from ..modeling_outputs import Transformer2DModelOutput
 from ..modeling_utils import ModelMixin
+from .modeling_common import BasicTransformerBlock
 
 
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name

src/diffusers/models/transformers/modeling_common.py

@@ -1255,4 +1255,4 @@ def forward(self, hidden_states: torch.Tensor, *args, **kwargs) -> torch.Tensor:
             deprecate("scale", "1.0.0", deprecation_message)
         for module in self.net:
             hidden_states = module(hidden_states)
-        return hidden_states
+        return hidden_states

src/diffusers/models/transformers/transformer_2d.py

@@ -19,11 +19,11 @@
 
 from ...configuration_utils import LegacyConfigMixin, register_to_config
 from ...utils import deprecate, logging
-from .modeling_common import BasicTransformerBlock
 from ..embeddings import ImagePositionalEmbeddings, PatchEmbed, PixArtAlphaTextProjection
 from ..modeling_outputs import Transformer2DModelOutput
 from ..modeling_utils import LegacyModelMixin
 from ..normalization import AdaLayerNormSingle
+from .modeling_common import BasicTransformerBlock
 
 
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name

src/diffusers/models/transformers/transformer_sd3.py

@@ -20,7 +20,6 @@
 from ...loaders import FromOriginalModelMixin, PeftAdapterMixin, SD3Transformer2DLoadersMixin
 from ...utils import USE_PEFT_BACKEND, logging, scale_lora_layers, unscale_lora_layers
 from ...utils.torch_utils import maybe_allow_in_graph
-from .modeling_common import FeedForward, JointTransformerBlock
 from ..attention_processor import (
     Attention,
     AttentionProcessor,
@@ -31,6 +30,7 @@
 from ..modeling_outputs import Transformer2DModelOutput
 from ..modeling_utils import ModelMixin
 from ..normalization import AdaLayerNormContinuous, AdaLayerNormZero
+from .modeling_common import FeedForward, JointTransformerBlock
 
 
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name