Unpin diffusers and pass onnx exporters tests

.github/workflows/test_export_onnx.yml

@@ -15,7 +15,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        python-version: ['3.9']
+        python-version: ["3.9"]
         os: [ubuntu-20.04]
 
     runs-on: ${{ matrix.os }}
@@ -27,13 +27,14 @@ jobs:
           python-version: ${{ matrix.python-version }}
       - name: Install dependencies for pytorch export
         run: |
-          pip install .[tests,exporters]
+          pip install .[tests,exporters,diffusers]
       - name: Test with unittest
         working-directory: tests
         run: |
           pytest exporters/onnx/test_onnx_*.py -s -n auto -m "not tensorflow_test and not timm_test" --durations=0
       - name: Install dependencies for tensorflow export
         run: |
+          pip uninstall diffusers -y
           pip install .[tests,exporters-tf]
       - name: Test with unittest
         working-directory: tests

.github/workflows/test_export_onnx_cli.yml

@@ -15,7 +15,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        python-version: [3.9]
+        python-version: ["3.9"]
         os: [ubuntu-20.04]
 
     runs-on: ${{ matrix.os }}
@@ -27,7 +27,7 @@ jobs:
           python-version: ${{ matrix.python-version }}
       - name: Install dependencies for pytorch export
         run: |
-          pip install .[tests,exporters]
+          pip install .[tests,exporters,diffusers]
       - name: Test with unittest
         working-directory: tests
         run: |

.github/workflows/test_exporters_slow.yml

@@ -14,26 +14,27 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        python-version: ['3.9']
+        python-version: ["3.9"]
 
     runs-on: ubuntu-20.04
     steps:
-    - uses: actions/checkout@v2
-    - name: Setup Python ${{ matrix.python-version }}
-      uses: actions/setup-python@v2
-      with:
-        python-version: ${{ matrix.python-version }}
-    - name: Install dependencies for pytorch export
-      run: |
-        pip install .[tests,exporters]
-    - name: Test with unittest
-      working-directory: tests
-      run: |
-        RUN_SLOW=1 pytest exporters -s -m "not tensorflow_test and run_slow" --durations=0
-    - name: Install dependencies for tensorflow export
-      run: |
-        pip install .[tests,exporters-tf]
-    - name: Test with unittest
-      working-directory: tests
-      run: |
-        RUN_SLOW=1 pytest exporters -s -m "tensorflow_test and run_slow" --durations=0
+      - uses: actions/checkout@v2
+      - name: Setup Python ${{ matrix.python-version }}
+        uses: actions/setup-python@v2
+        with:
+          python-version: ${{ matrix.python-version }}
+      - name: Install dependencies for pytorch export
+        run: |
+          pip install .[tests,exporters,diffusers]
+      - name: Test with unittest
+        working-directory: tests
+        run: |
+          RUN_SLOW=1 pytest exporters -s -m "not tensorflow_test and run_slow" --durations=0
+      - name: Install dependencies for tensorflow export
+        run: |
+          pip uninstall diffusers -y
+          pip install .[tests,exporters-tf]
+      - name: Test with unittest
+        working-directory: tests
+        run: |
+          RUN_SLOW=1 pytest exporters -s -m "tensorflow_test and run_slow" --durations=0

.github/workflows/test_onnx.yml

@@ -27,7 +27,7 @@ jobs:
         python-version: ${{ matrix.python-version }}
     - name: Install dependencies
       run: |
-        pip install .[tests,exporters]
+        pip install .[tests,exporters,diffusers]
     - name: Test with unittest
       working-directory: tests
       run: |

.github/workflows/test_onnxruntime.yml

@@ -38,17 +38,23 @@ jobs:
       - name: Setup Python
         uses: actions/setup-python@v5
         with:
-          python-version: '3.9'
+          python-version: "3.9"
 
       - name: Install dependencies
         run: |
           pip install --upgrade pip
           pip install --no-cache-dir torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu
-          pip install .[tests,onnxruntime]
+          pip install .[tests,onnxruntime,diffusers]
 
       - name: Install transformers ${{ matrix.transformers-version }}
         if: ${{ matrix.transformers-version != 'latest' }}
-        run: pip install transformers==${{ matrix.transformers-version }}
+        run: |
+          pip install "transformers==${{ matrix.transformers-version }}"
+
+      - name: Downgrade diffusers
+        if: matrix.transformers-version == '4.36.*'
+        run: |
+          pip install "diffusers<0.32.0"
 
       - name: Test with pytest (in series)
         run: |

.github/workflows/test_onnxruntime_slow.yml

@@ -14,7 +14,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        python-version: ['3.9']
+        python-version: ["3.9"]
         os: [ubuntu-20.04]
 
     runs-on: ${{ matrix.os }}
@@ -26,7 +26,7 @@ jobs:
         python-version: ${{ matrix.python-version }}
     - name: Install dependencies for export
       run: |
-        pip install .[tests,onnxruntime]
+        pip install .[tests,onnxruntime,diffusers]
     - name: Test with unittest
       working-directory: tests
       run: |

optimum/exporters/onnx/model_configs.py

@@ -181,8 +181,9 @@ def inputs(self) -> Dict[str, Dict[int, str]]:
         return {"input_ids": dynamic_axis, "attention_mask": dynamic_axis}
 
 
-class ModernBertOnnxConfig(DistilBertOnnxConfig):
-    pass
+# TODO: uncomment when transformers>=4.48.0
+# class ModernBertOnnxConfig(DistilBertOnnxConfig):
+#     pass
 
 
 class MPNetOnnxConfig(DistilBertOnnxConfig):

optimum/exporters/tasks.py

@@ -891,15 +891,15 @@ class TasksManager:
             "image-classification",
             onnx="MobileNetV2OnnxConfig",
         ),
-        "modernbert": supported_tasks_mapping(
-            "feature-extraction",
-            "fill-mask",
-            "text-classification",
-            "multiple-choice",
-            "token-classification",
-            "question-answering",
-            onnx="ModernBertOnnxConfig",
-        ),
+        # "modernbert": supported_tasks_mapping(
+        #     "feature-extraction",
+        #     "fill-mask",
+        #     "text-classification",
+        #     "multiple-choice",
+        #     "token-classification",
+        #     "question-answering",
+        #     onnx="ModernBertOnnxConfig",
+        # ),
         "mpnet": supported_tasks_mapping(
             "feature-extraction",
             "fill-mask",

setup.py

@@ -31,7 +31,6 @@
     "Pillow",
     "sacremoses",
     "torchvision",
-    "diffusers>=0.17.0,<0.32.0",
     "torchaudio",
     "einops",
     "timm",
@@ -89,7 +88,7 @@
         "executorch>=0.4.0",
         "transformers>=4.46",
     ],
-    "diffusers": ["diffusers<0.32.0"],
+    "diffusers": ["diffusers"],
     "intel": "optimum-intel>=1.18.0",
     "openvino": "optimum-intel[openvino]>=1.18.0",
     "nncf": "optimum-intel[nncf]>=1.18.0",

tests/exporters/exporters_utils.py

@@ -125,7 +125,7 @@
     "mobilenet-v2": "hf-internal-testing/tiny-random-MobileNetV2Model",
     "mobilenet-v1": "google/mobilenet_v1_0.75_192",
     "mobilevit": "hf-internal-testing/tiny-random-mobilevit",
-    "modernbert": "hf-internal-testing/tiny-random-ModernBertForMaskedLM",
+    # "modernbert": "hf-internal-testing/tiny-random-ModernBertForMaskedLM",
     "mpnet": "hf-internal-testing/tiny-random-MPNetModel",
     "mpt": "hf-internal-testing/tiny-random-MptForCausalLM",
     "mt5": "lewtun/tiny-random-mt5",
@@ -267,7 +267,7 @@
     # "mobilenet_v1": "google/mobilenet_v1_0.75_192",
     # "mobilenet_v2": "google/mobilenet_v2_0.35_96",
     "mobilevit": "apple/mobilevit-small",
-    "modernbert": "answerdotai/ModernBERT-base",
+    # "modernbert": "answerdotai/ModernBERT-base",
     "mpt": "mosaicml/mpt-7b",
     "mt5": "lewtun/tiny-random-mt5",  # Not using google/mt5-small because it takes too much time for testing.
     "musicgen": "facebook/musicgen-small",

tests/onnxruntime/test_diffusion.py

@@ -54,6 +54,7 @@ def _generate_prompts(batch_size=1):
         "guidance_scale": 7.5,
         "output_type": "np",
     }
+
     return inputs
 
 
@@ -105,8 +106,7 @@ class ORTPipelineForText2ImageTest(ORTModelTestMixin):
     def generate_inputs(self, height=128, width=128, batch_size=1):
         inputs = _generate_prompts(batch_size=batch_size)
 
-        inputs["height"] = height
-        inputs["width"] = width
+        inputs["height"], inputs["width"] = height, width
 
         return inputs
 
@@ -224,17 +224,19 @@ def test_shape(self, model_arch: str):
             elif output_type == "pt":
                 self.assertEqual(outputs.shape, (batch_size, 3, height, width))
             else:
-                expected_height = height // pipeline.vae_scale_factor
-                expected_width = width // pipeline.vae_scale_factor
-
                 if model_arch == "flux":
+                    expected_height = height // (pipeline.vae_scale_factor * 2)
+                    expected_width = width // (pipeline.vae_scale_factor * 2)
                     channels = pipeline.transformer.config.in_channels
                     expected_shape = (batch_size, expected_height * expected_width, channels)
-                elif model_arch == "stable-diffusion-3":
-                    out_channels = pipeline.transformer.config.out_channels
-                    expected_shape = (batch_size, out_channels, expected_height, expected_width)
                 else:
-                    out_channels = pipeline.unet.config.out_channels
+                    expected_height = height // pipeline.vae_scale_factor
+                    expected_width = width // pipeline.vae_scale_factor
+                    out_channels = (
+                        pipeline.unet.config.out_channels
+                        if getattr(pipeline, "unet", None) is not None
+                        else pipeline.transformer.config.out_channels
+                    )
                     expected_shape = (batch_size, out_channels, expected_height, expected_width)
 
                 self.assertEqual(outputs.shape, expected_shape)
@@ -363,6 +365,7 @@ def generate_inputs(self, height=128, width=128, batch_size=1, channel=3, input_
             height=height, width=width, batch_size=batch_size, channel=channel, input_type=input_type
         )
 
+        inputs["height"], inputs["width"] = height, width
         inputs["strength"] = 0.75
 
         return inputs
@@ -602,9 +605,8 @@ def generate_inputs(self, height=128, width=128, batch_size=1, channel=3, input_
             height=height, width=width, batch_size=batch_size, channel=1, input_type=input_type
         )
 
+        inputs["height"], inputs["width"] = height, width
         inputs["strength"] = 0.75
-        inputs["height"] = height
-        inputs["width"] = width
 
         return inputs