Fix a bug for SD35 control net training and improve control net block index (#10065)

* wip

---------
Co-authored-by: YiYi Xu <yixu310@gmail.com>
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>

src/diffusers/models/controlnets/controlnet_sd3.py

@@ -393,6 +393,7 @@ class SD3ControlNetModel(ModelMixin, ConfigMixin, PeftAdapterMixin, FromOriginal
                     return custom_forward
 
                 ckpt_kwargs: Dict[str, Any] = {"use_reentrant": False} if is_torch_version(">=", "1.11.0") else {}
+                if self.context_embedder is not None:
                     encoder_hidden_states, hidden_states = torch.utils.checkpoint.checkpoint(
                         create_custom_forward(block),
                         hidden_states,
@@ -400,6 +401,11 @@ class SD3ControlNetModel(ModelMixin, ConfigMixin, PeftAdapterMixin, FromOriginal
                         temb,
                         **ckpt_kwargs,
                     )
+                else:
+                    # SD3.5 8b controlnet use single transformer block, which does not use `encoder_hidden_states`
+                    hidden_states = torch.utils.checkpoint.checkpoint(
+                        create_custom_forward(block), hidden_states, temb, **ckpt_kwargs
+                    )
 
             else:
                 if self.context_embedder is not None:

src/diffusers/models/transformers/transformer_sd3.py

@@ -15,7 +15,6 @@
 
 from typing import Any, Dict, List, Optional, Tuple, Union
 
-import numpy as np
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
@@ -424,8 +423,7 @@ class SD3Transformer2DModel(ModelMixin, ConfigMixin, PeftAdapterMixin, FromOrigi
             # controlnet residual
             if block_controlnet_hidden_states is not None and block.context_pre_only is False:
                 interval_control = len(self.transformer_blocks) / len(block_controlnet_hidden_states)
-                interval_control = int(np.ceil(interval_control))
-                hidden_states = hidden_states + block_controlnet_hidden_states[index_block // interval_control]
+                hidden_states = hidden_states + block_controlnet_hidden_states[int(index_block / interval_control)]
 
         hidden_states = self.norm_out(hidden_states, temb)
         hidden_states = self.proj_out(hidden_states)