[CI] GLM-Image testcases

.buildkite/test-amd-ready.yaml

@@ -121,6 +121,7 @@ steps:
   mirror_hardwares: [amdproduction]
   grade: Blocking
   commands:
+    - pip list|grep vllm
     - export VLLM_WORKER_MULTIPROC_METHOD=spawn
     - |
       timeout 15m bash -c '

tests/e2e/online_serving/test_glm_image_expansion.py

@@ -0,0 +1,77 @@
+"""
+Comprehensive tests of diffusion features that are available in online serving mode
+and are supported by the following models:
+- GLM-Image: single image & text input
+CFG-Parallel and Tensor-Parallel features are covered
+"""
+
+import random
+import shutil
+
+import pytest
+
+from tests.helpers.mark import hardware_marks
+from tests.helpers.media import generate_synthetic_image
+from tests.helpers.runtime import (
+    OmniServer,
+    OmniServerParams,
+    OpenAIClientHandler,
+    dummy_messages_from_mix_data,
+)
+
+EDIT_PROMPT = "Transform this modern, geometrist image into a Vincent van Gogh style impressionist painting."
+SINGLE_CARD_FEATURE_MARKS = hardware_marks(res={"cuda": "H100"})
+PARALLEL_FEATURE_MARKS = hardware_marks(res={"cuda": "H100"}, num_cards=2)
+
+
+# This test file targets two models, so I write a helper function.
+# If a similar test only involves one model, one can just define a global list variable.
+def _get_diffusion_feature_cases(model: str):
+    return [
+        pytest.param(
+            OmniServerParams(
+                model=model,
+                server_args=[],
+            ),
+            id="basic_001",
+            marks=SINGLE_CARD_FEATURE_MARKS,
+        ),
+    ]
+
+
+@pytest.mark.full_model
+@pytest.mark.diffusion
+@pytest.mark.parametrize(
+    "omni_server",
+    _get_diffusion_feature_cases("zai-org/GLM-Image"),
+    indirect=True,
+)
+def test_glm_image(omni_server: OmniServer, openai_client: OpenAIClientHandler):
+    """Test all diffusion features with GLM-Image in regular end-user scenarios."""
+    image_size = 1024
+    synthetic_image = generate_synthetic_image(image_size, image_size)
+    image_data_url = f"data:image/jpeg;base64,{synthetic_image['base64']}"
+
+    messages = dummy_messages_from_mix_data(image_data_url=image_data_url, content_text=EDIT_PROMPT)
+
+    # CFG parallel is only activated when a negative prompt and true_cfg_scale > 1.0 are both present
+    request_config = {
+        "model": omni_server.model,
+        "messages": messages,
+        "extra_body": {
+            "height": image_size,
+            "width": image_size,
+            "num_inference_steps": 2,
+            "guidance_scale": 1.5,
+            "true_cfg_scale": 4.0,
+            "seed": 42,
+        },
+    }
+
+    responses = openai_client.send_diffusion_request(request_config)
+
+    pair_id = random.randint(10**8, 10**9 - 1)
+    shutil.copyfile(synthetic_image["file_path"], f"in_{pair_id}.jpg")
+    for response in responses:
+        for i, image in enumerate(response.images) or ():
+            image.convert("RGB").save(f"out_{pair_id}_{i}.jpg", format="JPEG")

vllm_omni/model_executor/models/glm_image/glm_image_ar.py

@@ -210,6 +210,19 @@ def get_data_parser(self) -> GlmImageDataParser:
             expected_hidden_size=self._get_expected_hidden_size(),
         )
 
+    def parse_mm_data(
+        self,
+        mm_data: MultiModalDataDict,
+        *,
+        validate: bool = True,
+    ) -> MultiModalDataItems:
+        # Remap "img2img" → "image" *in-place* so that every downstream
+        # consumer (hashes, kwargs, prompt-updates, _validate_mm_uuids)
+        # sees a single, consistent modality key.
+        if "img2img" in mm_data and "image" not in mm_data:
+            mm_data["image"] = mm_data.pop("img2img")
+        return super().parse_mm_data(mm_data, validate=validate)
+
     def get_supported_mm_limits(self) -> Mapping[str, int | None]:
         # GLM-Image is an image GENERATION model that supports:
         # - Text-to-image (t2i): no multimodal input needed