Merge branch 'main' into main

Author: yiyixuxu

.github/workflows/pr_style_bot.yml

@@ -9,12 +9,33 @@ permissions:
   pull-requests: write
 
 jobs:
-  run-style-bot:
+  check-permissions:
     if: >
       contains(github.event.comment.body, '@bot /style') &&
       github.event.issue.pull_request != null
     runs-on: ubuntu-latest
+    outputs:
+      is_authorized: ${{ steps.check_user_permission.outputs.has_permission }}
+    steps:
+      - name: Check user permission
+        id: check_user_permission
+        uses: actions/github-script@v6
+        with:
+          script: |
+            const comment_user = context.payload.comment.user.login;
+            const { data: permission } = await github.rest.repos.getCollaboratorPermissionLevel({
+              owner: context.repo.owner,
+              repo: context.repo.repo,
+              username: comment_user
+            });
+            const authorized = permission.permission === 'admin';
+            console.log(`User ${comment_user} has permission level: ${permission.permission}, authorized: ${authorized} (only admins allowed)`);
+            core.setOutput('has_permission', authorized);
 
+  run-style-bot:
+    needs: check-permissions
+    if: needs.check-permissions.outputs.is_authorized == 'true'
+    runs-on: ubuntu-latest
     steps:
       - name: Extract PR details
         id: pr_info
@@ -64,18 +85,38 @@ jobs:
         run: |
           pip install .[quality]
 
-      - name: Download Makefile from main branch
+      - name: Download necessary files from main branch of Diffusers
         run: |
           curl -o main_Makefile https://raw.githubusercontent.com/huggingface/diffusers/main/Makefile
+          curl -o main_setup.py https://raw.githubusercontent.com/huggingface/diffusers/refs/heads/main/setup.py
+          curl -o main_check_doc_toc.py https://raw.githubusercontent.com/huggingface/diffusers/refs/heads/main/utils/check_doc_toc.py
         
-      - name: Compare Makefiles
+      - name: Compare the files and raise error if needed
         run: |
+          diff_failed=0
+
           if ! diff -q main_Makefile Makefile; then
             echo "Error: The Makefile has changed. Please ensure it matches the main branch."
+            diff_failed=1
+          fi
+
+          if ! diff -q main_setup.py setup.py; then
+            echo "Error: The setup.py has changed. Please ensure it matches the main branch."
+            diff_failed=1
+          fi
+
+          if ! diff -q main_check_doc_toc.py utils/check_doc_toc.py; then
+            echo "Error: The utils/check_doc_toc.py has changed. Please ensure it matches the main branch."
+            diff_failed=1
+          fi
+
+          if [ $diff_failed -eq 1 ]; then
+            echo "❌ Error happened as we detected changes in the files that should not be changed ❌"
             exit 1
           fi
-          echo "No changes in Makefile. Proceeding..."
-          rm -rf main_Makefile
+
+          echo "No changes in the files. Proceeding..."
+          rm -rf main_Makefile main_setup.py main_check_doc_toc.py
 
       - name: Run make style and make quality
         run: |

.github/workflows/pr_tests_gpu.yml

@@ -11,6 +11,8 @@ on:
       - "src/diffusers/loaders/lora_base.py"
       - "src/diffusers/loaders/lora_pipeline.py"
       - "src/diffusers/loaders/peft.py"
+      - "tests/pipelines/test_pipelines_common.py"
+      - "tests/models/test_modeling_common.py"
   workflow_dispatch:
 
 concurrency:
@@ -104,11 +106,18 @@ jobs:
           # https://pytorch.org/docs/stable/notes/randomness.html#avoiding-nondeterministic-algorithms
           CUBLAS_WORKSPACE_CONFIG: :16:8
         run: |
-          pattern=$(cat ${{ steps.extract_tests.outputs.pattern_file }})
-          python -m pytest -n 1 --max-worker-restart=0 --dist=loadfile \
-            -s -v -k "not Flax and not Onnx and $pattern" \
-            --make-reports=tests_pipeline_${{ matrix.module }}_cuda \
-            tests/pipelines/${{ matrix.module }}
+          if [ "${{ matrix.module }}" = "ip_adapters" ]; then 
+              python -m pytest -n 1 --max-worker-restart=0 --dist=loadfile \
+              -s -v -k "not Flax and not Onnx" \
+              --make-reports=tests_pipeline_${{ matrix.module }}_cuda \
+              tests/pipelines/${{ matrix.module }}
+          else 
+              pattern=$(cat ${{ steps.extract_tests.outputs.pattern_file }})
+              python -m pytest -n 1 --max-worker-restart=0 --dist=loadfile \
+              -s -v -k "not Flax and not Onnx and $pattern" \
+              --make-reports=tests_pipeline_${{ matrix.module }}_cuda \
+              tests/pipelines/${{ matrix.module }}
+          fi 
 
       - name: Failure short reports
         if: ${{ failure() }}

src/diffusers/models/controlnets/controlnet_union.py

@@ -605,12 +605,13 @@ def forward(
         controlnet_cond: List[torch.Tensor],
         control_type: torch.Tensor,
         control_type_idx: List[int],
-        conditioning_scale: float = 1.0,
+        conditioning_scale: Union[float, List[float]] = 1.0,
         class_labels: Optional[torch.Tensor] = None,
         timestep_cond: Optional[torch.Tensor] = None,
         attention_mask: Optional[torch.Tensor] = None,
         added_cond_kwargs: Optional[Dict[str, torch.Tensor]] = None,
         cross_attention_kwargs: Optional[Dict[str, Any]] = None,
+        from_multi: bool = False,
         guess_mode: bool = False,
         return_dict: bool = True,
     ) -> Union[ControlNetOutput, Tuple[Tuple[torch.Tensor, ...], torch.Tensor]]:
@@ -647,6 +648,8 @@ def forward(
                 Additional conditions for the Stable Diffusion XL UNet.
             cross_attention_kwargs (`dict[str]`, *optional*, defaults to `None`):
                 A kwargs dictionary that if specified is passed along to the `AttnProcessor`.
+            from_multi (`bool`, defaults to `False`):
+                Use standard scaling when called from `MultiControlNetUnionModel`.
             guess_mode (`bool`, defaults to `False`):
                 In this mode, the ControlNet encoder tries its best to recognize the input content of the input even if
                 you remove all prompts. A `guidance_scale` between 3.0 and 5.0 is recommended.
@@ -658,6 +661,9 @@ def forward(
                 If `return_dict` is `True`, a [`~models.controlnet.ControlNetOutput`] is returned, otherwise a tuple is
                 returned where the first element is the sample tensor.
         """
+        if isinstance(conditioning_scale, float):
+            conditioning_scale = [conditioning_scale] * len(controlnet_cond)
+
         # check channel order
         channel_order = self.config.controlnet_conditioning_channel_order
 
@@ -742,12 +748,16 @@ def forward(
         inputs = []
         condition_list = []
 
-        for cond, control_idx in zip(controlnet_cond, control_type_idx):
+        for cond, control_idx, scale in zip(controlnet_cond, control_type_idx, conditioning_scale):
             condition = self.controlnet_cond_embedding(cond)
             feat_seq = torch.mean(condition, dim=(2, 3))
             feat_seq = feat_seq + self.task_embedding[control_idx]
-            inputs.append(feat_seq.unsqueeze(1))
-            condition_list.append(condition)
+            if from_multi:
+                inputs.append(feat_seq.unsqueeze(1))
+                condition_list.append(condition)
+            else:
+                inputs.append(feat_seq.unsqueeze(1) * scale)
+                condition_list.append(condition * scale)
 
         condition = sample
         feat_seq = torch.mean(condition, dim=(2, 3))
@@ -759,10 +769,13 @@ def forward(
             x = layer(x)
 
         controlnet_cond_fuser = sample * 0.0
-        for idx, condition in enumerate(condition_list[:-1]):
+        for (idx, condition), scale in zip(enumerate(condition_list[:-1]), conditioning_scale):
             alpha = self.spatial_ch_projs(x[:, idx])
             alpha = alpha.unsqueeze(-1).unsqueeze(-1)
-            controlnet_cond_fuser += condition + alpha
+            if from_multi:
+                controlnet_cond_fuser += condition + alpha
+            else:
+                controlnet_cond_fuser += condition + alpha * scale
 
         sample = sample + controlnet_cond_fuser
 
@@ -806,12 +819,13 @@ def forward(
         # 6. scaling
         if guess_mode and not self.config.global_pool_conditions:
             scales = torch.logspace(-1, 0, len(down_block_res_samples) + 1, device=sample.device)  # 0.1 to 1.0
-            scales = scales * conditioning_scale
+            if from_multi:
+                scales = scales * conditioning_scale[0]
             down_block_res_samples = [sample * scale for sample, scale in zip(down_block_res_samples, scales)]
             mid_block_res_sample = mid_block_res_sample * scales[-1]  # last one
-        else:
-            down_block_res_samples = [sample * conditioning_scale for sample in down_block_res_samples]
-            mid_block_res_sample = mid_block_res_sample * conditioning_scale
+        elif from_multi:
+            down_block_res_samples = [sample * conditioning_scale[0] for sample in down_block_res_samples]
+            mid_block_res_sample = mid_block_res_sample * conditioning_scale[0]
 
         if self.config.global_pool_conditions:
             down_block_res_samples = [

src/diffusers/models/controlnets/multicontrolnet_union.py

@@ -47,9 +47,12 @@ def forward(
         guess_mode: bool = False,
         return_dict: bool = True,
     ) -> Union[ControlNetOutput, Tuple]:
+        down_block_res_samples, mid_block_res_sample = None, None
         for i, (image, ctype, ctype_idx, scale, controlnet) in enumerate(
             zip(controlnet_cond, control_type, control_type_idx, conditioning_scale, self.nets)
         ):
+            if scale == 0.0:
+                continue
             down_samples, mid_sample = controlnet(
                 sample=sample,
                 timestep=timestep,
@@ -63,12 +66,13 @@ def forward(
                 attention_mask=attention_mask,
                 added_cond_kwargs=added_cond_kwargs,
                 cross_attention_kwargs=cross_attention_kwargs,
+                from_multi=True,
                 guess_mode=guess_mode,
                 return_dict=return_dict,
             )
 
             # merge samples
-            if i == 0:
+            if down_block_res_samples is None and mid_block_res_sample is None:
                 down_block_res_samples, mid_block_res_sample = down_samples, mid_sample
             else:
                 down_block_res_samples = [

src/diffusers/pipelines/controlnet/pipeline_controlnet_union_sd_xl.py

@@ -757,15 +757,9 @@ def check_inputs(
             for images_ in image:
                 for image_ in images_:
                     self.check_image(image_, prompt, prompt_embeds)
-        else:
-            assert False
 
         # Check `controlnet_conditioning_scale`
-        # TODO Update for https://github.com/huggingface/diffusers/pull/10723
-        if isinstance(controlnet, ControlNetUnionModel):
-            if not isinstance(controlnet_conditioning_scale, float):
-                raise TypeError("For single controlnet: `controlnet_conditioning_scale` must be type `float`.")
-        elif isinstance(controlnet, MultiControlNetUnionModel):
+        if isinstance(controlnet, MultiControlNetUnionModel):
             if isinstance(controlnet_conditioning_scale, list):
                 if any(isinstance(i, list) for i in controlnet_conditioning_scale):
                     raise ValueError("A single batch of multiple conditionings is not supported at the moment.")
@@ -776,8 +770,6 @@ def check_inputs(
                     "For multiple controlnets: When `controlnet_conditioning_scale` is specified as `list`, it must have"
                     " the same length as the number of controlnets"
                 )
-        else:
-            assert False
 
         if len(control_guidance_start) != len(control_guidance_end):
             raise ValueError(
@@ -808,8 +800,6 @@ def check_inputs(
             for _control_mode, _controlnet in zip(control_mode, self.controlnet.nets):
                 if max(_control_mode) >= _controlnet.config.num_control_type:
                     raise ValueError(f"control_mode: must be lower than {_controlnet.config.num_control_type}.")
-        else:
-            assert False
 
         # Equal number of `image` and `control_mode` elements
         if isinstance(controlnet, ControlNetUnionModel):
@@ -823,8 +813,6 @@ def check_inputs(
 
             elif sum(len(x) for x in image) != sum(len(x) for x in control_mode):
                 raise ValueError("Expected len(control_image) == len(control_mode)")
-        else:
-            assert False
 
         if ip_adapter_image is not None and ip_adapter_image_embeds is not None:
             raise ValueError(
@@ -1201,28 +1189,33 @@ def __call__(
 
         controlnet = self.controlnet._orig_mod if is_compiled_module(self.controlnet) else self.controlnet
 
+        if not isinstance(control_image, list):
+            control_image = [control_image]
+        else:
+            control_image = control_image.copy()
+
+        if not isinstance(control_mode, list):
+            control_mode = [control_mode]
+
+        if isinstance(controlnet, MultiControlNetUnionModel):
+            control_image = [[item] for item in control_image]
+            control_mode = [[item] for item in control_mode]
+
         # align format for control guidance
         if not isinstance(control_guidance_start, list) and isinstance(control_guidance_end, list):
             control_guidance_start = len(control_guidance_end) * [control_guidance_start]
         elif not isinstance(control_guidance_end, list) and isinstance(control_guidance_start, list):
             control_guidance_end = len(control_guidance_start) * [control_guidance_end]
         elif not isinstance(control_guidance_start, list) and not isinstance(control_guidance_end, list):
-            mult = len(controlnet.nets) if isinstance(controlnet, MultiControlNetUnionModel) else 1
+            mult = len(controlnet.nets) if isinstance(controlnet, MultiControlNetUnionModel) else len(control_mode)
             control_guidance_start, control_guidance_end = (
                 mult * [control_guidance_start],
                 mult * [control_guidance_end],
             )
 
-        if not isinstance(control_image, list):
-            control_image = [control_image]
-        else:
-            control_image = control_image.copy()
-
-        if not isinstance(control_mode, list):
-            control_mode = [control_mode]
-
-        if isinstance(controlnet, MultiControlNetUnionModel) and isinstance(controlnet_conditioning_scale, float):
-            controlnet_conditioning_scale = [controlnet_conditioning_scale] * len(controlnet.nets)
+        if isinstance(controlnet_conditioning_scale, float):
+            mult = len(controlnet.nets) if isinstance(controlnet, MultiControlNetUnionModel) else len(control_mode)
+            controlnet_conditioning_scale = [controlnet_conditioning_scale] * mult
 
         # 1. Check inputs
         self.check_inputs(
@@ -1357,9 +1350,6 @@ def __call__(
             control_image = control_images
             height, width = control_image[0][0].shape[-2:]
 
-        else:
-            assert False
-
         # 5. Prepare timesteps
         timesteps, num_inference_steps = retrieve_timesteps(
             self.scheduler, num_inference_steps, device, timesteps, sigmas
@@ -1397,7 +1387,7 @@ def __call__(
                 1.0 - float(i / len(timesteps) < s or (i + 1) / len(timesteps) > e)
                 for s, e in zip(control_guidance_start, control_guidance_end)
             ]
-            controlnet_keep.append(keeps[0] if isinstance(controlnet, ControlNetUnionModel) else keeps)
+            controlnet_keep.append(keeps)
 
         # 7.2 Prepare added time ids & embeddings
         original_size = original_size or (height, width)

tests/models/test_modeling_common.py

@@ -1173,17 +1173,16 @@ def test_disk_offload_without_safetensors(self):
         base_output = model(**inputs_dict)
 
         model_size = compute_module_sizes(model)[""]
+        max_size = int(self.model_split_percents[0] * model_size)
+        # Force disk offload by setting very small CPU memory
+        max_memory = {0: max_size, "cpu": int(0.1 * max_size)}
+
         with tempfile.TemporaryDirectory() as tmp_dir:
             model.cpu().save_pretrained(tmp_dir, safe_serialization=False)
-
             with self.assertRaises(ValueError):
-                max_size = int(self.model_split_percents[0] * model_size)
-                max_memory = {0: max_size, "cpu": max_size}
                 # This errors out because it's missing an offload folder
                 new_model = self.model_class.from_pretrained(tmp_dir, device_map="auto", max_memory=max_memory)
 
-            max_size = int(self.model_split_percents[0] * model_size)
-            max_memory = {0: max_size, "cpu": max_size}
             new_model = self.model_class.from_pretrained(
                 tmp_dir, device_map="auto", max_memory=max_memory, offload_folder=tmp_dir
             )

tests/models/transformers/test_models_transformer_omnigen.py

@@ -30,6 +30,7 @@ class OmniGenTransformerTests(ModelTesterMixin, unittest.TestCase):
     model_class = OmniGenTransformer2DModel
     main_input_name = "hidden_states"
     uses_custom_attn_processor = True
+    model_split_percents = [0.1, 0.1, 0.1]
 
     @property
     def dummy_input(self):
@@ -73,9 +74,9 @@ def prepare_init_args_and_inputs_for_common(self):
             "num_attention_heads": 4,
             "num_key_value_heads": 4,
             "intermediate_size": 32,
-            "num_layers": 1,
+            "num_layers": 20,
             "pad_token_id": 0,
-            "vocab_size": 100,
+            "vocab_size": 1000,
             "in_channels": 4,
             "time_step_dim": 4,
             "rope_scaling": {"long_factor": list(range(1, 3)), "short_factor": list(range(1, 3))},