[CI][Test] Tier diffusion offloader CI and reduce gpu_4_queue merge usage and fix #5182 (#5185)

Signed-off-by: wangyu <410167048@qq.com>

.buildkite/test-merge.yml

@@ -60,6 +60,23 @@ steps:
               volumes:
                 - "/fsx/hf_cache:/fsx/hf_cache"
 
+      - label: "Diffusion · Offloader Test"
+        timeout_in_minutes: 40
+        commands:
+          - timeout 40m pytest -s -v tests/diffusion/offloader -m "advanced_model and cuda"
+        agents:
+          queue: "gpu_1_queue" # g6.4xlarge instance on AWS, has 1 L4 GPU
+        plugins:
+          - docker#v5.2.0:
+              image: public.ecr.aws/q9t5s3a7/vllm-ci-test-repo:$BUILDKITE_COMMIT
+              always-pull: true
+              propagate-environment: true
+              environment:
+                - "HF_HOME=/fsx/hf_cache"
+                - "HF_TOKEN"
+              volumes:
+                - "/fsx/hf_cache:/fsx/hf_cache"
+
   - label: "Distributed Test"
     depends_on: upload-merge-pipeline
     timeout_in_minutes: 60
@@ -150,7 +167,7 @@ steps:
         commands:
           - pytest -s -v tests/entrypoints/ -m "advanced_model and cuda and L4" --run-level "advanced_model"
         agents:
-          queue: "gpu_4_queue" # g6.12xlarge instance on AWS, has 4 L4 GPU
+          queue: "gpu_1_queue" # g6.12xlarge instance on AWS, has 4 L4 GPU
         plugins:
           - docker#v5.2.0:
               image: public.ecr.aws/q9t5s3a7/vllm-ci-test-repo:$BUILDKITE_COMMIT
@@ -247,42 +264,6 @@ steps:
               volumes:
                 - "/fsx/hf_cache:/fsx/hf_cache"
 
-      - label: "TTS · IndexTTS2 Test"
-        source_file_dependencies:
-          - tests/e2e/online_serving/test_indextts2.py
-          - tests/e2e/offline_inference/test_indextts2.py
-          - tests/model_executor/models/indextts2/
-          - tests/model_executor/stage_input_processors/test_indextts2.py
-          - tests/assets/indextts2/
-          - vllm_omni/model_executor/models/indextts2/
-          - vllm_omni/model_executor/stage_input_processors/indextts2.py
-          - vllm_omni/deploy/indextts2.yaml
-          - vllm_omni/deploy/indextts2_low_latency.yaml
-          - vllm_omni/entrypoints/openai/serving_speech.py
-          - vllm_omni/entrypoints/openai/tts_adapters/indextts2.py
-          - vllm_omni/config/pipeline_registry.py
-          - vllm_omni/model_executor/models/registry.py
-        timeout_in_minutes: 60
-        commands:
-          - |
-            timeout 45m bash -c "
-              export VLLM_LOGGING_LEVEL=DEBUG
-              pytest -s -v tests/e2e/online_serving/test_indextts2.py tests/e2e/offline_inference/test_indextts2.py -m 'advanced_model and cuda' --run-level 'advanced_model'
-            "
-        agents:
-          queue: "gpu_4_queue"
-        plugins:
-          - docker#v5.2.0:
-              image: public.ecr.aws/q9t5s3a7/vllm-ci-test-repo:$BUILDKITE_COMMIT
-              always-pull: true
-              propagate-environment: true
-              environment:
-                - "HF_HOME=/fsx/hf_cache"
-                - "HF_TOKEN"
-              gpus: all
-              volumes:
-                - "/fsx/hf_cache:/fsx/hf_cache"
-
       - label: "Omni · Qwen3-Omni Test"
         source_file_dependencies:
           - tests/e2e/offline_inference/test_qwen3_omni.py

.buildkite/test-nightly.yml

@@ -1529,11 +1529,28 @@ steps:
                       path: /mnt/hf-cache
                       type: DirectoryOrCreate
 
-  - group: ":card_index_dividers: Diffusion · Quantization Test"
+  - group: ":card_index_dividers: Diffusion Test"
     depends_on: upload-nightly-pipeline
     if: build.env("NIGHTLY") == "1" || build.pull_request.labels includes "nightly-test" || build.pull_request.labels includes "diffusion-x2iat-test"
     steps:
-      - label: ":full_moon: Quantization Test with H100"
+      - label: ":full_moon: Diffusion · Offloader Test"
+        timeout_in_minutes: 40
+        commands:
+          - timeout 40m pytest -s -v tests/diffusion/offloader -m "full_model and cuda"
+        agents:
+          queue: "gpu_1_queue" # g6.4xlarge instance on AWS, has 1 L4 GPU
+        plugins:
+          - docker#v5.2.0:
+              image: public.ecr.aws/q9t5s3a7/vllm-ci-test-repo:$BUILDKITE_COMMIT
+              always-pull: true
+              propagate-environment: true
+              environment:
+                - "HF_HOME=/fsx/hf_cache"
+                - "HF_TOKEN"
+              volumes:
+                - "/fsx/hf_cache:/fsx/hf_cache"
+
+      - label: ":full_moon: Diffusion · Quantization Test with H100"
         commands:
           - timeout 60m pytest -sv tests/diffusion/quantization -m 'full_model and cuda and H100' --run-level "full_model"
         agents:
@@ -1570,7 +1587,7 @@ steps:
                       path: /mnt/hf-cache
                       type: DirectoryOrCreate
 
-      - label: ":full_moon: Quantization Test with L4"
+      - label: ":full_moon: Diffusion · Quantization Test with L4"
         timeout_in_minutes: 60
         commands:
           - pytest -sv tests/diffusion/quantization -m 'full_model and cuda and L4' --run-level "full_model"

.buildkite/test-ready.yml

@@ -108,6 +108,12 @@ steps:
                 - "/fsx/hf_cache:/fsx/hf_cache"
 
       - label: "Diffusion · Distributed Test"
+        source_file_dependencies:
+          - tests/diffusion/distributed/
+          - vllm_omni/diffusion/distributed/
+          - vllm_omni/diffusion/attention/parallel/
+          - vllm_omni/diffusion/data.py
+          - vllm_omni/diffusion/models/
         commands:
           - timeout 20m pytest -sv tests/diffusion/distributed -m 'core_model and cuda' --run-level "core_model"
         agents:

tests/diffusion/models/cosmos3/test_cosmos3_pipeline.py

@@ -716,10 +716,9 @@ def test_distilled_pipeline_initializes_sde_scheduler(
     stub_real_pipeline_init,
     monkeypatch: pytest.MonkeyPatch,
 ) -> None:
-    from diffusers import FlowMatchEulerDiscreteScheduler
-
     from vllm_omni.diffusion.models.cosmos3.pipeline_cosmos3 import (
         Cosmos3OmniDiffusersPipeline,
+        FlowMatchEulerDiscreteScheduler,
     )
 
     t_list = [1.0, 0.9375, 0.8333333333333334, 0.625]

tests/diffusion/offloader/test_diffusion_cpu_offload.py

@@ -24,6 +24,13 @@ IMAGE_VIDEO_MODELS = {
 
 MODELS = {**AUDIO_MODEL, **IMAGE_VIDEO_MODELS}
 
+MODEL_MARKS = {
+    "riverclouds/qwen_image_random": pytest.mark.core_model,
+    "Tongyi-MAI/Z-Image-Turbo": pytest.mark.advanced_model,
+    "stabilityai/stable-audio-open-1.0": pytest.mark.full_model,
+    "OmniGen2/OmniGen2": pytest.mark.full_model,
+}
+
 _GATED_MODELS = {"stabilityai/stable-audio-open-1.0"}
 
 # Aliased for backward compatibility (imported by test_diffusion_layerwise_offload.py).
@@ -97,10 +104,12 @@ def check_audio_determinism(audio1, audio2, atol=1e-2):
     return True
 
 
-@pytest.mark.core_model
 @pytest.mark.diffusion
 @hardware_test(res={"cuda": "L4", "rocm": "MI325"})
-@pytest.mark.parametrize("model_name", list(MODELS.keys()))
+@pytest.mark.parametrize(
+    "model_name",
+    [pytest.param(name, marks=MODEL_MARKS[name]) for name in MODELS],
+)
 def test_cpu_offload_diffusion_model(model_name: str):
     if model_name == "OmniGen2/OmniGen2":
         pytest.skip("issue #4537")

tests/e2e/offline_inference/test_indextts2_expansion.py

@@ -33,7 +33,7 @@ ASSETS_DIR = Path(__file__).resolve().parents[2] / "assets" / "indextts2"
 REF_AUDIO = str(ASSETS_DIR / "ref_audio.wav")
 
 pytestmark = [
-    pytest.mark.advanced_model,
+    pytest.mark.slow,
     pytest.mark.tts,
     pytest.mark.parametrize("omni_runner", [_OMNI_RUNNER_PARAM], indirect=True),
 ]

tests/e2e/online_serving/test_indextts2_expansion.py

@@ -19,7 +19,7 @@ from tests.helpers.media import load_test_audio_data_url
 from tests.helpers.runtime import OmniServerParams
 from tests.helpers.stage_config import get_deploy_config_path
 
-pytestmark = [pytest.mark.advanced_model, pytest.mark.tts]
+pytestmark = [pytest.mark.slow, pytest.mark.tts]
 
 MODEL = "IndexTeam/IndexTTS-2"
 

tests/entrypoints/openai_api/test_qwen3_tts_websocket.py

@@ -102,7 +102,7 @@ async def _run_ws_session(host: str, port: int, model: str) -> dict:
 class TestQwen3TTSWebSocket:
     @pytest.mark.advanced_model
     @pytest.mark.tts
-    @hardware_test(res={"cuda": "L4"}, num_cards=4)
+    @hardware_test(res={"cuda": "L4"}, num_cards=1)
     @pytest.mark.parametrize("omni_server", tts_ws_server_params, indirect=True)
     def test_streaming_pcm_output(self, omni_server) -> None:
         result = asyncio.run(_run_ws_session(omni_server.host, omni_server.port, omni_server.model))

tests/entrypoints/openai_api/test_serving_speech_batch.py

@@ -88,7 +88,7 @@ class TestSpeechBatchE2E:
 
     @pytest.mark.advanced_model
     @pytest.mark.tts
-    @hardware_test(res={"cuda": "H100"}, num_cards=1)
+    @hardware_test(res={"cuda": "L4"}, num_cards=1)
     @pytest.mark.parametrize("omni_server", default_server_params, indirect=True)
     def test_batch_basic_two_items(self, omni_server) -> None:
         """Batch with two items returns two successful base64-encoded results."""