re-add unet refactor PR (#7044)

* add

* remove copied from

---------
Co-authored-by: ultranity <1095429904@qq.com>
Co-authored-by: yiyixuxu <yixu310@gmail,com>

src/diffusers/models/unets/unet_2d_blocks.py

@@ -249,6 +249,81 @@ def get_down_block(
     raise ValueError(f"{down_block_type} does not exist.")
 
 
+def get_mid_block(
+    mid_block_type: str,
+    temb_channels: int,
+    in_channels: int,
+    resnet_eps: float,
+    resnet_act_fn: str,
+    resnet_groups: int,
+    output_scale_factor: float = 1.0,
+    transformer_layers_per_block: int = 1,
+    num_attention_heads: Optional[int] = None,
+    cross_attention_dim: Optional[int] = None,
+    dual_cross_attention: bool = False,
+    use_linear_projection: bool = False,
+    mid_block_only_cross_attention: bool = False,
+    upcast_attention: bool = False,
+    resnet_time_scale_shift: str = "default",
+    attention_type: str = "default",
+    resnet_skip_time_act: bool = False,
+    cross_attention_norm: Optional[str] = None,
+    attention_head_dim: Optional[int] = 1,
+    dropout: float = 0.0,
+):
+    if mid_block_type == "UNetMidBlock2DCrossAttn":
+        return UNetMidBlock2DCrossAttn(
+            transformer_layers_per_block=transformer_layers_per_block,
+            in_channels=in_channels,
+            temb_channels=temb_channels,
+            dropout=dropout,
+            resnet_eps=resnet_eps,
+            resnet_act_fn=resnet_act_fn,
+            output_scale_factor=output_scale_factor,
+            resnet_time_scale_shift=resnet_time_scale_shift,
+            cross_attention_dim=cross_attention_dim,
+            num_attention_heads=num_attention_heads,
+            resnet_groups=resnet_groups,
+            dual_cross_attention=dual_cross_attention,
+            use_linear_projection=use_linear_projection,
+            upcast_attention=upcast_attention,
+            attention_type=attention_type,
+        )
+    elif mid_block_type == "UNetMidBlock2DSimpleCrossAttn":
+        return UNetMidBlock2DSimpleCrossAttn(
+            in_channels=in_channels,
+            temb_channels=temb_channels,
+            dropout=dropout,
+            resnet_eps=resnet_eps,
+            resnet_act_fn=resnet_act_fn,
+            output_scale_factor=output_scale_factor,
+            cross_attention_dim=cross_attention_dim,
+            attention_head_dim=attention_head_dim,
+            resnet_groups=resnet_groups,
+            resnet_time_scale_shift=resnet_time_scale_shift,
+            skip_time_act=resnet_skip_time_act,
+            only_cross_attention=mid_block_only_cross_attention,
+            cross_attention_norm=cross_attention_norm,
+        )
+    elif mid_block_type == "UNetMidBlock2D":
+        return UNetMidBlock2D(
+            in_channels=in_channels,
+            temb_channels=temb_channels,
+            dropout=dropout,
+            num_layers=0,
+            resnet_eps=resnet_eps,
+            resnet_act_fn=resnet_act_fn,
+            output_scale_factor=output_scale_factor,
+            resnet_groups=resnet_groups,
+            resnet_time_scale_shift=resnet_time_scale_shift,
+            add_attention=False,
+        )
+    elif mid_block_type is None:
+        return None
+    else:
+        raise ValueError(f"unknown mid_block_type : {mid_block_type}")
+
+
 def get_up_block(
     up_block_type: str,
     num_layers: int,

src/diffusers/pipelines/deprecated/versatile_diffusion/modeling_text_unet.py

@@ -268,7 +268,6 @@ class GLIGENTextBoundingboxProjection(nn.Module):
         return objs
 
 
-# Copied from diffusers.models.unets.unet_2d_condition.UNet2DConditionModel with UNet2DConditionModel->UNetFlatConditionModel, nn.Conv2d->LinearMultiDim, Block2D->BlockFlat
 class UNetFlatConditionModel(ModelMixin, ConfigMixin):
     r"""
     A conditional 2D UNet model that takes a noisy sample, conditional state, and a timestep and returns a sample