update test case

Author: openvino-dev-samples

tests/openvino/test_diffusion.py

@@ -82,6 +82,7 @@ class OVPipelineForText2ImageTest(unittest.TestCase):
         "latent-consistency",
         "stable-diffusion-3",
         "flux",
+        "qwen-image",
         "sana",
     ]
     NEGATIVE_PROMPT_SUPPORT_ARCHITECTURES = [
@@ -121,7 +122,7 @@ def test_load_vanilla_model_which_is_not_supported(self):
     @parameterized.expand(SUPPORTED_ARCHITECTURES)
     @require_diffusers
     def test_ov_pipeline_class_dispatch(self, model_arch: str):
-        auto_cls = self.AUTOMODEL_CLASS if "sana" not in model_arch else DiffusionPipeline
+        auto_cls = self.AUTOMODEL_CLASS if "sana" not in model_arch and "qwen-image" not in model_arch else DiffusionPipeline
         auto_pipeline = auto_cls.from_pretrained(MODEL_NAMES[model_arch])
         ov_pipeline = self.OVMODEL_CLASS.from_pretrained(MODEL_NAMES[model_arch], device=OPENVINO_DEVICE)
 
@@ -135,6 +136,8 @@ def test_ov_pipeline_class_dispatch(self, model_arch: str):
     @parameterized.expand(SUPPORTED_ARCHITECTURES)
     @require_diffusers
     def test_num_images_per_prompt(self, model_arch: str):
+        if "qwen-image" in model_arch:
+            self.skipTest("QwenImage comparison test is not supported yet")
         pipeline = self.OVMODEL_CLASS.from_pretrained(MODEL_NAMES[model_arch], device=OPENVINO_DEVICE)
 
         for batch_size in [1, 3]:
@@ -153,7 +156,7 @@ def test_compare_to_diffusers_pipeline(self, model_arch: str):
         height, width, batch_size = 64, 64, 1
         inputs = self.generate_inputs(height=height, width=width, batch_size=batch_size, model_type=model_arch)
         ov_pipeline = self.OVMODEL_CLASS.from_pretrained(MODEL_NAMES[model_arch], device=OPENVINO_DEVICE)
-        auto_cls = self.AUTOMODEL_CLASS if "sana" not in model_arch else DiffusionPipeline
+        auto_cls = self.AUTOMODEL_CLASS if "sana" not in model_arch and "qwen-image" not in model_arch else DiffusionPipeline
         diffusers_pipeline = auto_cls.from_pretrained(MODEL_NAMES[model_arch])
 
         for output_type in ["latent", "np", "pt"]:
@@ -164,7 +167,9 @@ def test_compare_to_diffusers_pipeline(self, model_arch: str):
 
             ov_output = ov_pipeline(**inputs, generator=get_generator("pt", SEED)).images
             diffusers_output = diffusers_pipeline(**inputs, generator=get_generator("pt", SEED)).images
-            np.testing.assert_allclose(ov_output, diffusers_output, atol=6e-3, rtol=1e-2)
+            # QwenImage has higher numerical differences due to its architecture
+            atol, rtol = (0.1, 0.15) if "qwen-image" in model_arch else (6e-3, 1e-2)
+            np.testing.assert_allclose(ov_output, diffusers_output, atol=atol, rtol=rtol)
 
         # test on inputs nondivisible on 64
         height, width, batch_size = 96, 96, 1
@@ -178,7 +183,9 @@ def test_compare_to_diffusers_pipeline(self, model_arch: str):
             ov_output = ov_pipeline(**inputs, generator=get_generator("pt", SEED)).images
             diffusers_output = diffusers_pipeline(**inputs, generator=get_generator("pt", SEED)).images
 
-            np.testing.assert_allclose(ov_output, diffusers_output, atol=6e-3, rtol=1e-2)
+            # QwenImage has higher numerical differences due to its architecture
+            atol, rtol = (0.1, 0.15) if "qwen-image" in model_arch else (6e-3, 1e-2)
+            np.testing.assert_allclose(ov_output, diffusers_output, atol=atol, rtol=rtol)
 
     @parameterized.expand(CALLBACK_SUPPORT_ARCHITECTURES)
     @require_diffusers
@@ -230,7 +237,7 @@ def test_shape(self, model_arch: str):
             elif output_type == "pt":
                 self.assertEqual(outputs.shape, (batch_size, 3, height, width))
             else:
-                if model_arch != "flux":
+                if model_arch != "flux" and "qwen-image" not in model_arch:
                     out_channels = (
                         pipeline.unet.config.out_channels
                         if pipeline.unet is not None
@@ -245,7 +252,7 @@ def test_shape(self, model_arch: str):
                             width // pipeline.vae_scale_factor,
                         ),
                     )
-                else:
+                elif model_arch in ["flux", "qwen-image"]:
                     packed_height = height // pipeline.vae_scale_factor // 2
                     packed_width = width // pipeline.vae_scale_factor // 2
                     channels = pipeline.transformer.config.in_channels
@@ -397,6 +404,9 @@ def test_load_and_save_pipeline_with_safety_checker(self):
 
     @parameterized.expand(SUPPORTED_ARCHITECTURES)
     def test_height_width_properties(self, model_arch: str):
+        if "qwen-image" in model_arch:
+            self.skipTest("QwenImage comparison test is not supported yet")
+            
         batch_size, height, width, num_images_per_prompt = 2, 128, 64, 4
         ov_pipeline = self.OVMODEL_CLASS.from_pretrained(
             MODEL_NAMES[model_arch], export=True, compile=False, dynamic_shapes=True, device=OPENVINO_DEVICE
@@ -469,6 +479,7 @@ def test_load_custom_weight_variant(self):
     @parameterized.expand(SUPPORTED_ARCHITECTURES)
     @require_diffusers
     def test_static_shape_image_generation(self, model_arch):
+            
         pipeline = self.OVMODEL_CLASS.from_pretrained(MODEL_NAMES[model_arch], compile=False, device=OPENVINO_DEVICE)
         pipeline.reshape(batch_size=1, height=32, width=32)
         pipeline.compile()

tests/openvino/test_export.py

@@ -20,7 +20,11 @@
 from parameterized import parameterized
 from sentence_transformers import SentenceTransformer, models
 from transformers import AutoConfig, AutoTokenizer, GenerationConfig
-from utils_tests import MODEL_NAMES, OPENVINO_DEVICE
+from utils_tests import (
+    MODEL_NAMES,
+    OPENVINO_DEVICE,
+    REMOTE_CODE_MODELS,
+)
 
 from optimum.exporters.onnx.constants import SDPA_ARCHS_ONNX_EXPORT_NOT_SUPPORTED
 from optimum.exporters.onnx.model_configs import BertOnnxConfig
@@ -45,6 +49,7 @@
     OVModelForTokenClassification,
     OVModelForVisualCausalLM,
     OVModelForZeroShotImageClassification,
+    OVQwenImagePipeline,
     OVSamModel,
     OVStableDiffusion3Pipeline,
     OVStableDiffusionPipeline,
@@ -54,7 +59,7 @@
 from optimum.intel.openvino.modeling_base import OVBaseModel
 from optimum.intel.openvino.modeling_visual_language import MODEL_TYPE_TO_CLS_MAPPING
 from optimum.intel.openvino.utils import TemporaryDirectory
-from optimum.intel.utils.import_utils import _transformers_version, is_openvino_version, is_transformers_version
+from optimum.intel.utils.import_utils import _transformers_version, is_transformers_version
 from optimum.utils.save_utils import maybe_load_preprocessors
 
 
@@ -85,13 +90,20 @@ class ExportModelTest(unittest.TestCase):
         "stable-diffusion-3": OVStableDiffusion3Pipeline,
         "flux": OVFluxPipeline,
         "ltx-video": OVLTXPipeline,
+        "qwen-image": OVQwenImagePipeline,
     }
 
     if is_transformers_version(">=", "4.49"):
         SUPPORTED_ARCHITECTURES.update({"zamba2": OVModelForCausalLM})
 
-    if is_transformers_version(">=", "4.54.0") and is_openvino_version(">=", "2025.4.0"):
-        SUPPORTED_ARCHITECTURES.update({"lfm2": OVModelForCausalLM})
+    if is_transformers_version(">=", "4.53.0"):
+        SUPPORTED_ARCHITECTURES.update({"granite-moe-hybrid": OVModelForCausalLM})
+
+    if is_transformers_version(">=", "4.54"):
+        SUPPORTED_ARCHITECTURES.update({"exaone4": OVModelForCausalLM, "lfm2": OVModelForCausalLM})
+
+    if is_transformers_version(">=", "4.55.0") and is_transformers_version("<", "4.58.0"):
+        SUPPORTED_ARCHITECTURES.update({"afmoe": OVModelForCausalLM})
 
     EXPECTED_DIFFUSERS_SCALE_FACTORS = {
         "stable-diffusion-xl": {"vae_encoder": "128.0", "vae_decoder": "128.0"},
@@ -119,6 +131,9 @@ def _openvino_export(
         library_name = TasksManager.infer_library_from_model(model_name)
         loading_kwargs = {"attn_implementation": "eager"} if model_type in SDPA_ARCHS_ONNX_EXPORT_NOT_SUPPORTED else {}
 
+        if model_type in REMOTE_CODE_MODELS:
+            loading_kwargs["trust_remote_code"] = True
+
         if library_name == "timm":
             model_class = TasksManager.get_model_class_for_task(task, library=library_name)
             model = model_class(f"hf_hub:{model_name}", pretrained=True, exportable=True)
@@ -148,7 +163,9 @@ def _openvino_export(
                 )
 
                 use_cache = supported_task.endswith("-with-past")
-                ov_model = auto_model.from_pretrained(tmpdirname, use_cache=use_cache)
+                ov_model = auto_model.from_pretrained(
+                    tmpdirname, use_cache=use_cache, trust_remote_code=model_type in REMOTE_CODE_MODELS
+                )
                 self.assertIsInstance(ov_model, OVBaseModel)
 
                 if "text-generation" in task: