add tests for qwenimage modular.

Author: sayakpaul

src/diffusers/modular_pipelines/qwenimage/before_denoise.py

@@ -132,6 +132,7 @@ def expected_components(self) -> List[ComponentSpec]:
     @property
     def inputs(self) -> List[InputParam]:
         return [
+            InputParam("latents"),
             InputParam(name="height"),
             InputParam(name="width"),
             InputParam(name="num_images_per_prompt", default=1),
@@ -196,11 +197,11 @@ def __call__(self, components: QwenImageModularPipeline, state: PipelineState) -
                 f"You have passed a list of generators of length {len(block_state.generator)}, but requested an effective batch"
                 f" size of {batch_size}. Make sure the batch size matches the length of the generators."
             )
-
-        block_state.latents = randn_tensor(
-            shape, generator=block_state.generator, device=device, dtype=block_state.dtype
-        )
-        block_state.latents = components.pachifier.pack_latents(block_state.latents)
+        if block_state.latents is None:
+            block_state.latents = randn_tensor(
+                shape, generator=block_state.generator, device=device, dtype=block_state.dtype
+            )
+            block_state.latents = components.pachifier.pack_latents(block_state.latents)
 
         self.set_block_state(state, block_state)
         return components, state
@@ -549,7 +550,7 @@ def __call__(self, components: QwenImageModularPipeline, state: PipelineState) -
                     block_state.width // components.vae_scale_factor // 2,
                 )
             ]
-            * block_state.batch_size
+            for _ in range(block_state.batch_size)
         ]
         block_state.txt_seq_lens = (
             block_state.prompt_embeds_mask.sum(dim=1).tolist() if block_state.prompt_embeds_mask is not None else None

src/diffusers/modular_pipelines/qwenimage/decoders.py

@@ -74,8 +74,9 @@ def __call__(self, components: QwenImageModularPipeline, state: PipelineState) -
         block_state = self.get_block_state(state)
 
         # YiYi Notes: remove support for output_type = "latents', we can just skip decode/encode step in modular
+        vae_scale_factor = 2 ** len(components.vae.temperal_downsample)
         block_state.latents = components.pachifier.unpack_latents(
-            block_state.latents, block_state.height, block_state.width
+            block_state.latents, block_state.height, block_state.width, vae_scale_factor=vae_scale_factor
         )
         block_state.latents = block_state.latents.to(components.vae.dtype)
 

src/diffusers/modular_pipelines/qwenimage/encoders.py

@@ -503,6 +503,8 @@ def __call__(self, components: QwenImageModularPipeline, state: PipelineState):
         block_state.prompt_embeds = block_state.prompt_embeds[:, : block_state.max_sequence_length]
         block_state.prompt_embeds_mask = block_state.prompt_embeds_mask[:, : block_state.max_sequence_length]
 
+        block_state.negative_prompt_embeds = None
+        block_state.negative_prompt_embeds_mask = None
         if components.requires_unconditional_embeds:
             negative_prompt = block_state.negative_prompt or ""
             block_state.negative_prompt_embeds, block_state.negative_prompt_embeds_mask = get_qwen_prompt_embeds(

src/diffusers/modular_pipelines/qwenimage/modular_pipeline.py

@@ -26,10 +26,7 @@ class QwenImagePachifier(ConfigMixin):
     config_name = "config.json"
 
     @register_to_config
-    def __init__(
-        self,
-        patch_size: int = 2,
-    ):
+    def __init__(self, patch_size: int = 2):
         super().__init__()
 
     def pack_latents(self, latents):

tests/modular_pipelines/qwen/__init__.py

@@ -0,0 +1,85 @@
+# coding=utf-8
+# Copyright 2025 HuggingFace Inc.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import unittest
+
+import numpy as np
+import torch
+
+from diffusers import ClassifierFreeGuidance
+from diffusers.modular_pipelines import QwenImageAutoBlocks, QwenImageModularPipeline
+
+from ...testing_utils import torch_device
+from ..test_modular_pipelines_common import ModularPipelineTesterMixin
+
+
+class QwenImagexModularTests:
+    pipeline_class = QwenImageModularPipeline
+    pipeline_blocks_class = QwenImageAutoBlocks
+    repo = "hf-internal-testing/tiny-qwenimage-modular"
+
+    params = frozenset(["prompt", "height", "width", "negative_prompt", "attention_kwargs", "image", "mask_image"])
+    batch_params = frozenset(["prompt", "negative_prompt", "image", "mask_image"])
+
+    def get_pipeline(self, components_manager=None, torch_dtype=torch.float32):
+        pipeline = self.pipeline_blocks_class().init_pipeline(self.repo, components_manager=components_manager)
+        pipeline.load_components(torch_dtype=torch_dtype)
+        pipeline.set_progress_bar_config(disable=None)
+        return pipeline
+
+    def get_dummy_inputs(self, device, seed=0):
+        if str(device).startswith("mps"):
+            generator = torch.manual_seed(seed)
+        else:
+            generator = torch.Generator(device=device).manual_seed(seed)
+        inputs = {
+            "prompt": "dance monkey",
+            "negative_prompt": "bad quality",
+            "generator": generator,
+            "num_inference_steps": 2,
+            "height": 32,
+            "width": 32,
+            "max_sequence_length": 16,
+            "output_type": "np",
+        }
+        return inputs
+
+
+class QwenImageModularGuiderTests:
+    def test_guider_cfg(self):
+        pipe = self.get_pipeline()
+        pipe = pipe.to(torch_device)
+
+        guider = ClassifierFreeGuidance(guidance_scale=1.0)
+        pipe.update_components(guider=guider)
+
+        inputs = self.get_dummy_inputs(torch_device)
+        out_no_cfg = pipe(**inputs, output="images")
+
+        guider = ClassifierFreeGuidance(guidance_scale=7.5)
+        pipe.update_components(guider=guider)
+        inputs = self.get_dummy_inputs(torch_device)
+        out_cfg = pipe(**inputs, output="images")
+
+        assert out_cfg.shape == out_no_cfg.shape
+        max_diff = np.abs(out_cfg - out_no_cfg).max()
+        assert max_diff > 1e-2, "Output with CFG must be different from normal inference"
+
+
+class QwenImageModularPipelineFastTests(
+    QwenImagexModularTests, QwenImageModularGuiderTests, ModularPipelineTesterMixin, unittest.TestCase
+):
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)