Fixed typo and reverted removal of skip_layers in SD3Transformer2DModel

Author: guiyrt

src/diffusers/models/transformers/transformer_sd3.py

@@ -341,6 +341,7 @@ def forward(
         block_controlnet_hidden_states: List = None,
         joint_attention_kwargs: Optional[Dict[str, Any]] = None,
         return_dict: bool = True,
+        skip_layers: Optional[List[int]] = None,
     ) -> Union[torch.FloatTensor, Transformer2DModelOutput]:
         """
         The [`SD3Transformer2DModel`] forward method.
@@ -363,6 +364,8 @@ def forward(
             return_dict (`bool`, *optional*, defaults to `True`):
                 Whether or not to return a [`~models.transformer_2d.Transformer2DModelOutput`] instead of a plain
                 tuple.
+            skip_layers (`list` of `int`, *optional*):
+                A list of layer indices to skip during the forward pass.
 
         Returns:
             If `return_dict` is True, an [`~models.transformer_2d.Transformer2DModelOutput`] is returned, otherwise a
@@ -390,7 +393,10 @@ def forward(
         encoder_hidden_states = self.context_embedder(encoder_hidden_states)
 
         for index_block, block in enumerate(self.transformer_blocks):
-            if self.training and self.gradient_checkpointing:
+            # Skip specified layers
+            is_skip = True if skip_layers is not None and index_block in skip_layers else False
+
+            if torch.is_grad_enabled() and self.gradient_checkpointing and not is_skip:
 
                 def create_custom_forward(module, return_dict=None):
                     def custom_forward(*inputs):
@@ -410,8 +416,7 @@ def custom_forward(*inputs):
                     joint_attention_kwargs,
                     **ckpt_kwargs,
                 )
-
-            else:
+            elif not is_skip:
                 encoder_hidden_states, hidden_states = block(
                     hidden_states=hidden_states, encoder_hidden_states=encoder_hidden_states, temb=temb,
                     joint_attention_kwargs=joint_attention_kwargs,

src/diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3.py

@@ -855,7 +855,7 @@ def set_ip_adapter_scale(self, scale):
         only conditioned by the text prompt. Lowering this value encourages the model to produce more diverse images, but they 
         may not be as aligned with the image prompt.
         """
-        for attn_processor in self.transformes.attn_processors.values():
+        for attn_processor in self.transformer.attn_processors.values():
             if isinstance(attn_processor, IPAdapterJointAttnProcessor2_0):
                 attn_processor.scale = scale