[CI] Restructure vLLM-Omni Test Layout, Fixture Scope, and Support Modules (#2620)

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

.buildkite/test-nightly.yml

@@ -552,7 +552,7 @@ steps:
       - label: ":full_moon: Diffusion X2V · Accuracy Test"
         timeout_in_minutes: 180
         commands:
-          - pytest -s -v tests/e2e/accuracy/wan22_i2v/test_wan22_i2v_video_similarity.py --run-level advanced_model
+          - pytest -s -v tests/e2e/accuracy/wan22_i2v/test_wan22_i2v_video_similarity.py -m advanced_model --run-level advanced_model
         agents:
           queue: "mithril-h100-pool"
         plugins:

docs/contributing/ci/test_examples/l4_functionality_tests.inc.md

@@ -40,7 +40,7 @@ Currently all the features are available in online serving mode. Hence, only nee
 - Test marks: always add `advanced_model` and `diffusion`. Add GPU-related marks if needed. Ref: [Markers for Tests](https://docs.vllm.ai/projects/vllm-omni/en/latest/contributing/ci/tests_markers/).
 - To maximize code reuse, you may refer to
     - `tests/conftest.py` for `omni_server` (running server in subprocess) and `openai_client` fixtures (sending requests and validating output), `generate_synthetic_image` and `assert_XXX_valid` helper.
-    - `tests/utils.py` for `@hardware_test(...)` and `hardware_marks`.
+    - `tests/helpers/mark.py` for `@hardware_test(...)` and `hardware_marks`.
     - [Parametrizing tests (pytest doc)](https://docs.pytest.org/en/stable/example/parametrize.html) to reuse test function implementation for different cases.
 - Doc: add a concise docstring for each test function.
 - Reference L4 test implementation: [tests/e2e/online_serving/test_qwen_image_edit_expansion.py](https://github.com/vllm-project/vllm-omni/blob/main/tests/e2e/online_serving/test_qwen_image_edit_expansion.py).

docs/contributing/ci/tests_markers.md

@@ -38,7 +38,7 @@ Defined in `pyproject.toml`:
 ### Example usage for markers
 
 ```python
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
 
 @pytest.mark.core_model
 @pytest.mark.omni
@@ -53,7 +53,7 @@ def test_video_to_audio()
 
 ### Decorator: `@hardware_test`
 
-This decorator is intended to make hardware-aware, cross-platform test authoring easier and more robust for CI/CD environments. The `hardware_test` decorator in `vllm-omni/tests/utils.py` performs the following actions:
+This decorator is intended to make hardware-aware, cross-platform test authoring easier and more robust for CI/CD environments. The `hardware_test` decorator in `vllm-omni/tests/helpers/mark.py` performs the following actions:
 
 1. **Applies platform and resource markers**  
    Adds the appropriate pytest markers for each specified hardware platform (e.g., `cuda`, `rocm`, `xpu`, `npu`) and resource type (e.g., `L4`, `H100`, `MI325`, `B60`, `A2`, `A3`).
@@ -105,7 +105,7 @@ This decorator is intended to make hardware-aware, cross-platform test authoring
 `hardware_marks` returns a list of pytest mark objects with the same signature as `@hardware_test`. Use it when you need more flexibility, such as attaching hardware marks to individual `pytest.param` entries rather than an entire test function.
 
 ```python
-from tests.utils import hardware_marks
+from tests.helpers.mark import hardware_marks
 
 MULTI_CARD_MARKS = hardware_marks(
     res={"cuda": "H100", "rocm": "MI325", "npu": "A2"}, num_cards=2
@@ -133,9 +133,9 @@ If you want to add support for a new platform (e.g., "tpu" for a new accelerator
        "distributed_tpu: Tests that require multiple TPU devices",
    ]
    ```
-2. **Implement a marker construction function for your platform** in `vllm-omni/tests/utils.py`:
+2. **Implement a marker construction function for your platform** in `vllm-omni/tests/helpers/mark.py`:
    ```python
-   # In vllm-omni/tests/utils.py
+   # In vllm-omni/tests/helpers/mark.py
 
    def tpu_marks(*, res: str, num_cards: int):
        test_platform = pytest.mark.tpu
@@ -175,4 +175,4 @@ If you want to add support for a new platform (e.g., "tpu" for a new accelerator
 - Plug into `hardware_marks`  
 - You're done: tests using `@hardware_test` or `hardware_marks` with your platform now automatically get the correct markers, distribution, and isolation!
 
-See code in `vllm-omni/tests/utils.py` for existing examples (`cuda_marks`, `rocm_marks`, `npu_marks`).
+See code in `vllm-omni/tests/helpers/mark.py` for existing examples (`cuda_marks`, `rocm_marks`, `npu_marks`).

docs/contributing/ci/tests_style.md

@@ -221,15 +221,13 @@ from pathlib import Path
 import openai
 import pytest
 
-from tests.conftest import (
-    OmniServer,
-    convert_audio_to_text,
+from tests.helpers.media import (
+    convert_audio_bytes_to_text,
     cosine_similarity_text,
-    dummy_messages_from_mix_data,
     generate_synthetic_video,
-    merge_base64_and_convert_to_text,
 )
-from tests.utils import get_deploy_config_path
+from tests.helpers.runtime import OmniServer, dummy_messages_from_mix_data
+from tests.helpers.stage_config import get_deploy_config_path, modify_stage_config
 from vllm_omni.platforms import current_omni_platform
 
 # Edit: model name and stage config path
@@ -406,7 +404,7 @@ def test_mix_to_text_audio_001(client: openai.OpenAI, omni_server, request) -> N
     # PURPOSE: Verify text and audio outputs convey the same information
     # CUSTOMIZATION: Adjust similarity threshold (0.9) based on accuracy requirements
     assert audio_data is not None, "No audio output is generated"
-    audio_content = merge_base64_and_convert_to_text(audio_data)
+    audio_content = convert_audio_bytes_to_text(audio_data)
     print(f"text content is: {text_content}")
     print(f"audio content is: {audio_content}")
     similarity = cosine_similarity_text(audio_content.lower(), text_content.lower())
@@ -429,7 +427,7 @@ from pathlib import Path
 import pytest
 from vllm.assets.video import VideoAsset
 
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
 from ..multi_stages.conftest import OmniRunner
 
 # Optional: set process start method for workers

tests/core/test_prefix_cache.py

@@ -1,5 +1,3 @@
-from unittest.mock import Mock, patch
-
 import pytest
 import torch
 
@@ -19,10 +17,14 @@ class MockInputBatch:
         self.req_ids = ["req1", "req2"]
         self.req_id_to_index = {req_id: i for i, req_id in enumerate(self.req_ids)}
         self.num_computed_tokens_cpu = num_computed_tokens_cpu
+
         # Block table is only mocked for validation of length;
         # we don't actually need to add valid values here since
         # we patch the table when testing.
-        self.block_table = Mock()
+        class _DummyBlockTable:
+            pass
+
+        self.block_table = _DummyBlockTable()
         self.block_table.block_tables = [None]
 
 
@@ -186,7 +188,7 @@ def fake_get_cached_block_ids(self, req_idx, *args, **kwargs):
 
 
 @pytest.mark.parametrize("num_tokens_padded", [None, 16])
-def test_get_merged_hidden_states(num_tokens_padded):
+def test_get_merged_hidden_states(num_tokens_padded, mocker):
     """Ensure that hidden states are merged correctly."""
     cache = get_omni_pcache()
 
@@ -221,16 +223,16 @@ def test_get_merged_hidden_states(num_tokens_padded):
 
     input_batch = MockInputBatch(num_computed_tokens_cpu=torch.Tensor([orig_num_tokens_unpadded, 0]))
 
-    with patch(
+    mocker.patch(
         "vllm_omni.core.prefix_cache.OmniTensorPrefixCache._get_cached_block_ids",
         new=fake_get_cached_block_ids,
-    ):
-        merged_states = cache.get_merged_hidden_states(
-            query_start_loc=[0, num_new_toks_req1],
-            input_batch=input_batch,
-            hidden_states=new_hidden_states,
-            num_scheduled_tokens=num_scheduled_tokens,
-        )
+    )
+    merged_states = cache.get_merged_hidden_states(
+        query_start_loc=[0, num_new_toks_req1],
+        input_batch=input_batch,
+        hidden_states=new_hidden_states,
+        num_scheduled_tokens=num_scheduled_tokens,
+    )
 
     assert "req1" in merged_states and "req2" in merged_states
     req1_merged_states = merged_states["req1"]
@@ -255,7 +257,7 @@ def test_get_merged_hidden_states(num_tokens_padded):
         {"foo": 100, "bar": 50, "baz": 10},
     ],
 )
-def test_get_merged_multimodal_outputs(feat_dims, num_tokens_padded):
+def test_get_merged_multimodal_outputs(feat_dims, num_tokens_padded, mocker):
     cache = get_omni_pcache_with_mm_tensors(feat_dims, seq_len=DEFAULT_SEQ_LEN)
 
     orig_num_tokens_unpadded = 8
@@ -298,16 +300,16 @@ def test_get_merged_multimodal_outputs(feat_dims, num_tokens_padded):
 
     input_batch = MockInputBatch(num_computed_tokens_cpu=torch.Tensor([orig_num_tokens_unpadded, 0]))
 
-    with patch(
+    mocker.patch(
         "vllm_omni.core.prefix_cache.OmniTensorPrefixCache._get_cached_block_ids",
         new=fake_get_cached_block_ids,
-    ):
-        merged_mm_outputs = cache.get_merged_multimodal_states(
-            query_start_loc=[0, num_new_toks_req1],
-            input_batch=input_batch,
-            multimodal_outputs=new_mm_outputs,
-            num_scheduled_tokens=num_scheduled_tokens,
-        )
+    )
+    merged_mm_outputs = cache.get_merged_multimodal_states(
+        query_start_loc=[0, num_new_toks_req1],
+        input_batch=input_batch,
+        multimodal_outputs=new_mm_outputs,
+        num_scheduled_tokens=num_scheduled_tokens,
+    )
 
     # Ensure the passthrough data wasn't dropped
     assert "passthrough_data" in merged_mm_outputs

tests/dfx/conftest.py

@@ -4,7 +4,7 @@ from typing import Any
 
 import pytest
 
-from tests.conftest import modify_stage_config
+from tests.helpers.stage_config import modify_stage_config
 
 
 def load_configs(config_path: str) -> list[dict[str, Any]]:

tests/dfx/perf/scripts/run_benchmark.py

@@ -8,7 +8,6 @@ from typing import Any
 
 import pytest
 
-from tests.conftest import OmniServer
 from tests.dfx.conftest import (
     create_benchmark_indices,
     create_test_parameter_mapping,
@@ -16,6 +15,7 @@ from tests.dfx.conftest import (
     get_benchmark_params_for_server,
     load_configs,
 )
+from tests.helpers.runtime import OmniServer
 
 os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
 os.environ["VLLM_TEST_CLEAN_GPU_MEMORY"] = "0"

tests/dfx/stability/conftest.py

@@ -1,125 +1,17 @@
-"""
-Stability-specific conftest: when pytest is executed under this directory,
-resource monitoring is started before each test and finalized after each test,
-so each stability test case gets its own HTML report (one report per case).
-No need to wrap pytest with `bash resource_monitor.sh run -- pytest ...`.
-"""
+"""Stability pytest hooks and fixtures."""
 
-import os
 import subprocess
 import sys
 import threading
-import time
-from pathlib import Path
 
 import pytest
 
-STABILITY_DIR = Path(__file__).resolve().parent
-RESOURCE_MONITOR_SCRIPT = STABILITY_DIR / "scripts" / "resource_monitor.sh"
-REPO_ROOT = STABILITY_DIR.parent.parent.parent
-
-
-def _start_resource_monitor():
-    """Start `resource_monitor.sh start` in the background and return `Popen` or `None`."""
-    if not RESOURCE_MONITOR_SCRIPT.is_file():
-        return None
-    try:
-        proc = subprocess.Popen(
-            ["bash", str(RESOURCE_MONITOR_SCRIPT), "start", "--backend", "gpu"],
-            cwd=str(REPO_ROOT),
-            stdout=subprocess.DEVNULL,
-            stderr=subprocess.PIPE,
-            start_new_session=True,
-        )
-        try:
-            proc.wait(timeout=2)
-            if proc.returncode != 0:
-                stderr = proc.stderr.read().decode("utf-8", errors="ignore") if proc.stderr else ""
-                if stderr.strip():
-                    sys.stderr.write(f"[Stability] Resource monitor failed to start: {stderr.strip()}\n")
-                return None
-        except subprocess.TimeoutExpired:
-            pass
-        return proc
-    except (FileNotFoundError, OSError):
-        return None
-
-
-def _get_monitor_data_root() -> Path:
-    data_root = os.environ.get("RESOURCE_MONITOR_DATA_ROOT") or os.environ.get("GPU_MONITOR_DATA_ROOT")
-    if data_root:
-        return Path(data_root)
-    return STABILITY_DIR / "gpu_monitor_data"
-
-
-def _wait_for_run_dir(timeout_sec: int = 10) -> Path | None:
-    data_root = _get_monitor_data_root()
-    run_id_file = data_root / "current_run_id"
-    deadline = time.time() + timeout_sec
-    while time.time() < deadline:
-        if run_id_file.is_file():
-            run_id = run_id_file.read_text(encoding="utf-8").strip()
-            if run_id:
-                run_dir = data_root / run_id
-                if run_dir.is_dir():
-                    return run_dir
-        time.sleep(0.5)
-    return None
-
-
-def _report_latest_gpu_samples(stop_event: threading.Event) -> None:
-    """Periodically print the latest sampled GPU line."""
-    log_interval = int(
-        os.environ.get("RESOURCE_MONITOR_LOG_INTERVAL") or os.environ.get("GPU_MONITOR_LOG_INTERVAL") or "15"
-    )
-    log_interval = max(log_interval, 1)
-    last_line = ""
-
-    time.sleep(min(log_interval, 5))
-    while not stop_event.wait(log_interval):
-        run_dir = _wait_for_run_dir(timeout_sec=1)
-        if run_dir is None:
-            continue
-        csv_file = run_dir / "gpu_metrics.csv"
-        if not csv_file.is_file():
-            continue
-        try:
-            lines = csv_file.read_text(encoding="utf-8").splitlines()
-        except OSError:
-            continue
-        if len(lines) <= 1:
-            continue
-        latest = lines[-1].strip()
-        if latest and latest != last_line:
-            last_line = latest
-            sys.stderr.write(f"[GPU] {latest}\n")
-
-
-def _finalize_resource_monitor() -> str | None:
-    """
-    Run `resource_monitor.sh finalize` for the current run and generate the report.
-    Returns the bundle dir path (for this test case's report) if successful, else None.
-    """
-    if not RESOURCE_MONITOR_SCRIPT.is_file():
-        return None
-    try:
-        result = subprocess.run(
-            ["bash", str(RESOURCE_MONITOR_SCRIPT), "finalize", "--backend", "gpu"],
-            cwd=str(REPO_ROOT),
-            capture_output=True,
-            text=True,
-            timeout=60,
-            check=False,
-        )
-        if result.returncode != 0:
-            return None
-        for line in (result.stdout or "").splitlines():
-            if line.startswith("GPU_MONITOR_BUNDLE_DIR=") or line.startswith("RESOURCE_MONITOR_BUNDLE_DIR="):
-                _, _, value = line.partition("=")
-                return value.strip() if value else None
-        return None
-    except (FileNotFoundError, OSError, subprocess.TimeoutExpired):
-        return None
+from tests.dfx.stability.helpers import (
+    finalize_resource_monitor,
+    report_latest_gpu_samples,
+    start_resource_monitor,
+    wait_for_run_dir,
+)
 
 
 @pytest.fixture(autouse=True)
@@ -128,19 +20,19 @@ def stability_resource_monitor_per_test(request: pytest.FixtureRequest):
     For each test under this directory: start GPU monitor before the test,
     then finalize after the test so this case gets its own report.html.
     """
-    proc = _start_resource_monitor()
+    proc = start_resource_monitor()
     stop_event = threading.Event()
     reporter: threading.Thread | None = None
 
     if proc is not None:
         reporter = threading.Thread(
-            target=_report_latest_gpu_samples,
+            target=report_latest_gpu_samples,
             args=(stop_event,),
             name="stability-resource-monitor-reporter",
             daemon=True,
         )
         reporter.start()
-        run_dir = _wait_for_run_dir(timeout_sec=5)
+        run_dir = wait_for_run_dir(timeout_sec=5)
         node_name = request.node.name
         if run_dir is not None:
             sys.stderr.write(f"[Stability] Resource monitor started for test: {node_name} | run dir: {run_dir}\n")
@@ -161,7 +53,7 @@ def stability_resource_monitor_per_test(request: pytest.FixtureRequest):
             except subprocess.TimeoutExpired:
                 proc.kill()
                 proc.wait()
-        bundle_dir = _finalize_resource_monitor()
+        bundle_dir = finalize_resource_monitor()
         node_name = request.node.name
         if bundle_dir:
             sys.stderr.write(f"[Stability] Report for test «{node_name}»: {bundle_dir}/report.html\n")

tests/dfx/stability/helpers.py

@@ -0,0 +1,117 @@
+"""Stability resource monitor helpers."""
+
+from __future__ import annotations
+
+import os
+import subprocess
+import sys
+import threading
+import time
+from pathlib import Path
+
+STABILITY_DIR = Path(__file__).resolve().parent
+RESOURCE_MONITOR_SCRIPT = STABILITY_DIR / "scripts" / "resource_monitor.sh"
+REPO_ROOT = STABILITY_DIR.parent.parent.parent
+
+
+def start_resource_monitor():
+    """Start `resource_monitor.sh start` in the background and return `Popen` or `None`."""
+    if not RESOURCE_MONITOR_SCRIPT.is_file():
+        return None
+    try:
+        proc = subprocess.Popen(
+            ["bash", str(RESOURCE_MONITOR_SCRIPT), "start", "--backend", "gpu"],
+            cwd=str(REPO_ROOT),
+            stdout=subprocess.DEVNULL,
+            stderr=subprocess.PIPE,
+            start_new_session=True,
+        )
+        try:
+            proc.wait(timeout=2)
+            if proc.returncode != 0:
+                stderr = proc.stderr.read().decode("utf-8", errors="ignore") if proc.stderr else ""
+                if stderr.strip():
+                    sys.stderr.write(f"[Stability] Resource monitor failed to start: {stderr.strip()}\n")
+                return None
+        except subprocess.TimeoutExpired:
+            pass
+        return proc
+    except (FileNotFoundError, OSError):
+        return None
+
+
+def get_monitor_data_root() -> Path:
+    data_root = os.environ.get("RESOURCE_MONITOR_DATA_ROOT") or os.environ.get("GPU_MONITOR_DATA_ROOT")
+    if data_root:
+        return Path(data_root)
+    return STABILITY_DIR / "gpu_monitor_data"
+
+
+def wait_for_run_dir(timeout_sec: int = 10) -> Path | None:
+    data_root = get_monitor_data_root()
+    run_id_file = data_root / "current_run_id"
+    deadline = time.time() + timeout_sec
+    while time.time() < deadline:
+        if run_id_file.is_file():
+            run_id = run_id_file.read_text(encoding="utf-8").strip()
+            if run_id:
+                run_dir = data_root / run_id
+                if run_dir.is_dir():
+                    return run_dir
+        time.sleep(0.5)
+    return None
+
+
+def report_latest_gpu_samples(stop_event: threading.Event) -> None:
+    """Periodically print the latest sampled GPU line."""
+    log_interval = int(
+        os.environ.get("RESOURCE_MONITOR_LOG_INTERVAL") or os.environ.get("GPU_MONITOR_LOG_INTERVAL") or "15"
+    )
+    log_interval = max(log_interval, 1)
+    last_line = ""
+
+    time.sleep(min(log_interval, 5))
+    while not stop_event.wait(log_interval):
+        run_dir = wait_for_run_dir(timeout_sec=1)
+        if run_dir is None:
+            continue
+        csv_file = run_dir / "gpu_metrics.csv"
+        if not csv_file.is_file():
+            continue
+        try:
+            lines = csv_file.read_text(encoding="utf-8").splitlines()
+        except OSError:
+            continue
+        if len(lines) <= 1:
+            continue
+        latest = lines[-1].strip()
+        if latest and latest != last_line:
+            last_line = latest
+            sys.stderr.write(f"[GPU] {latest}\n")
+
+
+def finalize_resource_monitor() -> str | None:
+    """
+    Run `resource_monitor.sh finalize` for the current run and generate the report.
+    Returns the bundle dir path (for this test case's report) if successful, else None.
+    """
+    if not RESOURCE_MONITOR_SCRIPT.is_file():
+        return None
+    try:
+        result = subprocess.run(
+            ["bash", str(RESOURCE_MONITOR_SCRIPT), "finalize", "--backend", "gpu"],
+            cwd=str(REPO_ROOT),
+            capture_output=True,
+            text=True,
+            timeout=60,
+            check=False,
+        )
+        if result.returncode != 0:
+            return None
+        for line in (result.stdout or "").splitlines():
+            if line.startswith("GPU_MONITOR_BUNDLE_DIR=") or line.startswith("RESOURCE_MONITOR_BUNDLE_DIR="):
+                _, _, value = line.partition("=")
+                return value.strip() if value else None
+        return None
+    except (FileNotFoundError, OSError, subprocess.TimeoutExpired):
+        return None

tests/dfx/stability/scripts/test_benchmark_stability.py

@@ -24,7 +24,6 @@ from typing import Any
 
 import pytest
 
-from tests.conftest import OmniServer
 from tests.dfx.conftest import (
     create_benchmark_indices,
     create_test_parameter_mapping,
@@ -33,6 +32,7 @@ from tests.dfx.conftest import (
     load_configs,
 )
 from tests.dfx.perf.scripts.run_benchmark import run_benchmark
+from tests.helpers.runtime import OmniServer
 
 STABILITY_DIR = Path(__file__).resolve().parent.parent
 DEPLOY_CONFIGS_DIR = STABILITY_DIR / "deploy"

tests/diffusion/cache/test_teacache_extractors.py

@@ -21,7 +21,7 @@ from unittest.mock import MagicMock, Mock, patch
 import pytest
 import torch
 
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
 from vllm_omni.diffusion.cache.teacache.extractors import extract_flux2_context, extract_flux2_klein_context
 from vllm_omni.diffusion.models.flux2_klein.flux2_klein_transformer import (
     Flux2Transformer2DModel,

tests/diffusion/distributed/test_ulysses_uaa_perf.py

@@ -17,7 +17,7 @@ import pytest
 import torch
 import torch.distributed as dist
 
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
 from vllm_omni.diffusion.attention.parallel.ulysses import (
     _all_gather_int,
     _ulysses_all_to_all_any_o,

tests/diffusion/lora/test_lora_manager.py

@@ -8,7 +8,7 @@ import torch
 from vllm.lora.lora_weights import LoRALayerWeights
 from vllm.lora.utils import get_supported_lora_modules
 
-from tests.diffusion.lora.conftest import (
+from tests.diffusion.lora.helpers import (
     DummyBaseLayerWithLoRA,
     FakeLinearBase,
     fake_replace_submodule,

tests/diffusion/models/bagel/test_bagel_lora.py

@@ -11,7 +11,7 @@ import pytest
 import torch
 from safetensors.torch import save_file
 
-from tests.diffusion.lora.conftest import (
+from tests.diffusion.lora.helpers import (
     DummyBaseLayerWithLoRA,
     FakeLinearBase,
     fake_replace_submodule,

tests/diffusion/models/flux2/test_flux2_transformer_tp.py

@@ -2,7 +2,7 @@ import pytest
 import torch
 from pytest_mock import MockerFixture
 
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
 from vllm_omni.diffusion.models.flux2.flux2_transformer import (
     Flux2PosEmbed,
     Flux2Transformer2DModel,

tests/diffusion/models/glm_image/test_glm_image_sp.py

@@ -2,27 +2,26 @@
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 """Tests for GLM-Image Sequence Parallelism support."""
 
-from unittest.mock import MagicMock, patch
-
 import pytest
 
 from vllm_omni.diffusion.data import DiffusionParallelConfig
 
 
 @pytest.fixture(scope="function", autouse=True)
-def setup_sp_groups():
+def setup_sp_groups(mocker):
     """Set up SP and TP groups for each test function."""
-    with patch("vllm_omni.diffusion.distributed.parallel_state.get_sp_group") as mock_get_sp_group:
-        with patch("vllm.model_executor.layers.linear.get_tensor_model_parallel_world_size", return_value=1):
-            with patch("vllm.distributed.parallel_state.get_tp_group") as mock_get_tp_group:
-                mock_sp_group = MagicMock()
-                mock_sp_group.world_size = 4
-                mock_get_sp_group.return_value = mock_sp_group
+    mock_get_sp_group = mocker.patch("vllm_omni.diffusion.distributed.parallel_state.get_sp_group")
+    mocker.patch("vllm.model_executor.layers.linear.get_tensor_model_parallel_world_size", return_value=1)
+    mock_get_tp_group = mocker.patch("vllm.distributed.parallel_state.get_tp_group")
 
-                mock_tp_group = MagicMock()
-                mock_tp_group.world_size = 1
-                mock_get_tp_group.return_value = mock_tp_group
-                yield
+    mock_sp_group = mocker.MagicMock()
+    mock_sp_group.world_size = 4
+    mock_get_sp_group.return_value = mock_sp_group
+
+    mock_tp_group = mocker.MagicMock()
+    mock_tp_group.world_size = 1
+    mock_get_tp_group.return_value = mock_tp_group
+    yield
 
 
 pytestmark = [pytest.mark.core_model, pytest.mark.cpu]

tests/diffusion/quantization/test_quantization_quality.py

@@ -32,7 +32,7 @@ import numpy as np
 import pytest
 import torch
 
-from tests.utils import hardware_marks
+from tests.helpers.mark import hardware_marks
 
 # ---------------------------------------------------------------------------
 # Configuration — add new quantization methods / models here

tests/diffusion/test_diffusion_model_runner.py

@@ -8,7 +8,7 @@ import pytest
 import torch
 
 import vllm_omni.diffusion.worker.diffusion_model_runner as model_runner_module
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
 from vllm_omni.diffusion.worker.diffusion_model_runner import DiffusionModelRunner
 
 pytestmark = [pytest.mark.diffusion]

tests/diffusion/test_diffusion_step_pipeline.py

@@ -13,7 +13,7 @@ import torch
 from pytest_mock import MockerFixture
 
 import vllm_omni.diffusion.worker.diffusion_model_runner as model_runner_module
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
 from vllm_omni.diffusion.data import DiffusionOutput
 from vllm_omni.diffusion.diffusion_engine import DiffusionEngine
 from vllm_omni.diffusion.distributed.cfg_parallel import CFGParallelMixin

tests/e2e/accuracy/conftest.py

@@ -1,7 +1,6 @@
 from __future__ import annotations
 
 import os
-import shutil
 import subprocess
 from contextlib import contextmanager
 from dataclasses import dataclass
@@ -13,7 +12,7 @@ import requests
 import torch
 from PIL import Image
 
-from tests.conftest import OmniServer, OmniServerParams
+from tests.helpers.runtime import OmniServer, OmniServerParams
 
 
 def pytest_addoption(parser):
@@ -208,18 +207,6 @@ def rabbit_image(accuracy_artifact_root: Path) -> Image.Image:
     return image
 
 
-def reset_artifact_dir(path: Path) -> Path:
-    if path.exists():
-        shutil.rmtree(path)
-    path.mkdir(parents=True, exist_ok=True)
-    return path
-
-
-def infer_model_label(model: str) -> str:
-    label = Path(model.rstrip("/\\")).name or "model"
-    return "".join(char if char.isalnum() or char in {"-", "_"} else "_" for char in label)
-
-
 def _build_accuracy_server_config(
     *,
     generate_model: str,

tests/e2e/accuracy/helpers.py

@@ -1,5 +1,3 @@
-from __future__ import annotations
-
 from pathlib import Path
 
 import numpy as np
@@ -9,6 +7,20 @@ from PIL import Image
 from torchmetrics.image import PeakSignalNoiseRatio, StructuralSimilarityIndexMeasure
 
 
+def reset_artifact_dir(path: Path) -> Path:
+    import shutil
+
+    if path.exists():
+        shutil.rmtree(path)
+    path.mkdir(parents=True, exist_ok=True)
+    return path
+
+
+def infer_model_label(model: str) -> str:
+    label = Path(model.rstrip("/\\")).name or "model"
+    return "".join(char if char.isalnum() or char in {"-", "_"} else "_" for char in label)
+
+
 def model_output_dir(parent_dir: Path, model: str) -> Path:
     safe_model_name = model.split("/")[-1].replace(".", "_")
     path = parent_dir / safe_model_name

tests/e2e/accuracy/test_gebench_h100_smoke.py

@@ -6,8 +6,8 @@ from pathlib import Path
 import pytest
 
 from benchmarks.accuracy.text_to_image.gbench import main as gbench_main
-from tests.e2e.accuracy.conftest import infer_model_label, reset_artifact_dir
-from tests.utils import hardware_test
+from tests.e2e.accuracy.helpers import infer_model_label, reset_artifact_dir
+from tests.helpers.mark import hardware_test
 
 
 @pytest.mark.advanced_model

tests/e2e/accuracy/test_gedit_bench_h100_smoke.py

@@ -7,8 +7,8 @@ import pytest
 
 from benchmarks.accuracy.image_to_image.gedit_bench import GROUPS
 from benchmarks.accuracy.image_to_image.gedit_bench import main as gedit_main
-from tests.e2e.accuracy.conftest import infer_model_label, reset_artifact_dir
-from tests.utils import hardware_test
+from tests.e2e.accuracy.helpers import infer_model_label, reset_artifact_dir
+from tests.helpers.mark import hardware_test
 
 
 @pytest.mark.advanced_model

tests/e2e/accuracy/test_qwen_image.py

@@ -12,13 +12,10 @@ import torch
 from diffusers.pipelines.pipeline_utils import DiffusionPipeline
 from PIL import Image
 
-from tests.conftest import (
-    OmniServer,
-    _run_post_test_cleanup,
-    _run_pre_test_cleanup,
-)
-from tests.e2e.accuracy.utils import assert_similarity, model_output_dir
-from tests.utils import hardware_test
+from tests.e2e.accuracy.helpers import assert_similarity, model_output_dir
+from tests.helpers.env import run_post_test_cleanup, run_pre_test_cleanup
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniServer
 
 MODEL_ID = "Qwen/Qwen-Image"
 MODEL_ENV_VAR = "QWEN_IMAGE_MODEL"
@@ -70,7 +67,7 @@ def _run_vllm_omni_qwen_image(*, model: str, output_path: Path) -> Image.Image:
 
 
 def _run_diffusers_qwen_image(*, model: str, output_path: Path) -> Image.Image:
-    _run_pre_test_cleanup(enable_force=True)
+    run_pre_test_cleanup(enable_force=True)
     pipe: DiffusionPipeline | None = None
     try:
         pipe = DiffusionPipeline.from_pretrained(
@@ -99,7 +96,7 @@ def _run_diffusers_qwen_image(*, model: str, output_path: Path) -> Image.Image:
         gc.collect()
         if torch.cuda.is_available():
             torch.cuda.empty_cache()
-        _run_post_test_cleanup(enable_force=True)
+        run_post_test_cleanup(enable_force=True)
 
 
 @pytest.mark.advanced_model

tests/e2e/accuracy/test_qwen_image_edit.py

@@ -10,13 +10,10 @@ from diffusers import QwenImageEditPipeline, QwenImageEditPlusPipeline
 from PIL import Image
 
 from benchmarks.accuracy.common import decode_base64_image, pil_to_png_bytes
-from tests.conftest import (
-    OmniServer,
-    _run_post_test_cleanup,
-    _run_pre_test_cleanup,
-)
-from tests.e2e.accuracy.utils import assert_similarity, model_output_dir
-from tests.utils import hardware_test
+from tests.e2e.accuracy.helpers import assert_similarity, model_output_dir
+from tests.helpers.env import run_post_test_cleanup, run_pre_test_cleanup
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniServer
 
 SINGLE_MODEL = "Qwen/Qwen-Image-Edit"
 MULTIPLE_MODEL = "Qwen/Qwen-Image-Edit-2509"
@@ -77,7 +74,7 @@ def _run_diffusers_image_edit(
     input_images: list[Image.Image],
     output_path: Path,
 ) -> Image.Image:
-    _run_pre_test_cleanup(enable_force=True)
+    run_pre_test_cleanup(enable_force=True)
     pipe: QwenImageEditPipeline | QwenImageEditPlusPipeline | None = None
     device = torch.device("cuda:0")
     torch.cuda.set_device(device)
@@ -110,7 +107,7 @@ def _run_diffusers_image_edit(
         gc.collect()
         if torch.cuda.is_available():
             torch.cuda.empty_cache()
-        _run_post_test_cleanup(enable_force=True)
+        run_post_test_cleanup(enable_force=True)
 
 
 def _vllm_omni_output_single_image(

tests/e2e/accuracy/test_qwen_image_layered.py

@@ -12,13 +12,10 @@ import torch
 from diffusers.pipelines.pipeline_utils import DiffusionPipeline
 from PIL import Image
 
-from tests.conftest import (
-    OmniServer,
-    _run_post_test_cleanup,
-    _run_pre_test_cleanup,
-)
-from tests.e2e.accuracy.utils import assert_image_sequence_similarity, model_output_dir
-from tests.utils import hardware_test
+from tests.e2e.accuracy.helpers import assert_image_sequence_similarity, model_output_dir
+from tests.helpers.env import run_post_test_cleanup, run_pre_test_cleanup
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniServer
 
 MODEL_ID = "Qwen/Qwen-Image-Layered"
 MODEL_ENV_VAR = "QWEN_IMAGE_LAYERED_MODEL"
@@ -93,7 +90,7 @@ def _run_vllm_omni_qwen_image_layered(*, model: str, input_image: Image.Image, o
 
 
 def _run_diffusers_qwen_image_layered(*, model: str, input_image: Image.Image, output_dir: Path) -> list[Image.Image]:
-    _run_pre_test_cleanup(enable_force=True)
+    run_pre_test_cleanup(enable_force=True)
     pipe: DiffusionPipeline | None = None
     try:
         pipe = DiffusionPipeline.from_pretrained(
@@ -126,7 +123,7 @@ def _run_diffusers_qwen_image_layered(*, model: str, input_image: Image.Image, o
         gc.collect()
         if torch.cuda.is_available():
             torch.cuda.empty_cache()
-        _run_post_test_cleanup(enable_force=True)
+        run_post_test_cleanup(enable_force=True)
 
 
 @pytest.mark.advanced_model

tests/e2e/accuracy/wan22_i2v/test_wan22_i2v_video_similarity.py

@@ -22,7 +22,6 @@ import torch
 from diffusers import UniPCMultistepScheduler
 from PIL import Image
 
-from tests.conftest import OmniServerParams
 from tests.e2e.accuracy.wan22_i2v.run_wan22_i2v_diffusers_cp import (
     _configure_scheduler,
     _ensure_wan_ftfy_fallback,
@@ -48,7 +47,8 @@ from tests.e2e.accuracy.wan22_i2v.wan22_i2v_video_similarity_common import (
     SSIM_THRESHOLD,
     WIDTH,
 )
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniServerParams
 
 
 def test_parse_video_metadata_extracts_dimensions_and_fps() -> None:

tests/e2e/offline_inference/custom_pipeline/test_async_omni_collective_rpc.py

@@ -26,7 +26,7 @@ from contextlib import ExitStack
 
 import pytest
 
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
 from vllm_omni.entrypoints.async_omni import AsyncOmni
 from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 from vllm_omni.outputs import OmniRequestOutput

tests/e2e/offline_inference/custom_pipeline/test_async_omni_qwen_image_generate.py

@@ -19,7 +19,7 @@ import numpy as np
 import pytest
 from transformers import AutoTokenizer
 
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
 from vllm_omni.entrypoints.async_omni import AsyncOmni
 from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 from vllm_omni.outputs import OmniRequestOutput

tests/e2e/offline_inference/custom_pipeline/test_worker_extension.py

@@ -10,7 +10,7 @@ import pytest
 from tests.e2e.offline_inference.custom_pipeline.worker_extension import (
     vLLMOmniColocateWorkerExtensionForTest,
 )
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
 from vllm_omni.diffusion.worker.diffusion_worker import CustomPipelineWorkerExtension
 from vllm_omni.entrypoints.async_omni import AsyncOmni
 

tests/e2e/offline_inference/test_bagel_img2img.py

@@ -22,9 +22,10 @@ import pytest
 from PIL import Image
 from vllm.assets.image import ImageAsset
 
-from tests.conftest import OmniRunner, modify_stage_config
-from tests.utils import hardware_test
-from vllm_omni import Omni
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniRunner
+from tests.helpers.stage_config import modify_stage_config
+from vllm_omni.entrypoints.omni import Omni
 from vllm_omni.platforms import current_omni_platform
 
 # Reference pixel data extracted from the known-good output image

tests/e2e/offline_inference/test_bagel_lora.py

@@ -31,9 +31,10 @@ import torch
 from PIL import Image
 from safetensors.torch import save_file
 
-from tests.conftest import OmniRunner, modify_stage_config
-from tests.utils import hardware_test
-from vllm_omni import Omni
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniRunner
+from tests.helpers.stage_config import modify_stage_config
+from vllm_omni.entrypoints.omni import Omni
 from vllm_omni.lora.request import LoRARequest
 from vllm_omni.lora.utils import stable_lora_int_id
 

tests/e2e/offline_inference/test_bagel_text2img.py

@@ -27,9 +27,10 @@ from typing import Any
 import pytest
 from PIL import Image
 
-from tests.conftest import OmniRunner, modify_stage_config
-from tests.utils import hardware_test
-from vllm_omni import Omni
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniRunner
+from tests.helpers.stage_config import modify_stage_config
+from vllm_omni.entrypoints.omni import Omni
 from vllm_omni.platforms import current_omni_platform
 
 # Reference pixel data extracted from the known-good output image

tests/e2e/offline_inference/test_bagel_understanding.py

@@ -26,8 +26,9 @@ from pathlib import Path
 import pytest
 from vllm.assets.image import ImageAsset
 
-from tests.conftest import OmniRunner, modify_stage_config
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniRunner
+from tests.helpers.stage_config import modify_stage_config
 
 MODEL_NAME = "ByteDance-Seed/BAGEL-7B-MoT"
 STAGE_CONFIG = str(Path(__file__).parent / "stage_configs" / "bagel_sharedmemory_ci.yaml")

tests/e2e/offline_inference/test_cache_dit.py

@@ -11,8 +11,8 @@ It uses minimal settings to keep test time short for CI.
 import pytest
 import torch
 
-from tests.conftest import OmniRunner
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniRunner
 from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 from vllm_omni.outputs import OmniRequestOutput
 from vllm_omni.platforms import current_omni_platform

tests/e2e/offline_inference/test_cosyvoice3.py

@@ -26,8 +26,8 @@ import yaml
 from huggingface_hub import snapshot_download
 from vllm.sampling_params import SamplingParams
 
-from tests.conftest import OmniRunner
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniRunner
 from vllm_omni.model_executor.models.cosyvoice3.config import CosyVoice3Config
 from vllm_omni.model_executor.models.cosyvoice3.tokenizer import get_qwen_tokenizer
 

tests/e2e/offline_inference/test_diffusion_cpu_offload.py

@@ -4,8 +4,9 @@ import pytest
 import torch
 from vllm.distributed.parallel_state import cleanup_dist_env_and_memory
 
-from tests.conftest import OmniRunner
-from tests.utils import DeviceMemoryMonitor, hardware_test
+from tests.helpers.env import DeviceMemoryMonitor
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniRunner
 from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 from vllm_omni.platforms import current_omni_platform
 

tests/e2e/offline_inference/test_diffusion_layerwise_offload.py

@@ -2,8 +2,8 @@ import pytest
 import torch
 from vllm.distributed.parallel_state import cleanup_dist_env_and_memory
 
-from tests.conftest import OmniRunner
-from tests.utils import DeviceMemoryMonitor
+from tests.helpers.env import DeviceMemoryMonitor
+from tests.helpers.runtime import OmniRunner
 from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 from vllm_omni.platforms import current_omni_platform
 

tests/e2e/offline_inference/test_diffusion_lora.py

@@ -7,7 +7,7 @@ import pytest
 import torch
 from safetensors.torch import save_file
 
-from tests.conftest import OmniRunner
+from tests.helpers.runtime import OmniRunner
 from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 from vllm_omni.outputs import OmniRequestOutput
 from vllm_omni.platforms import current_omni_platform

tests/e2e/offline_inference/test_dynin_omni.py

@@ -18,7 +18,7 @@ import pytest
 import torch
 from transformers import AutoTokenizer
 
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
 
 os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
 os.environ["VLLM_TEST_CLEAN_GPU_MEMORY"] = "0"

tests/e2e/offline_inference/test_expert_parallel.py

@@ -18,8 +18,8 @@ import torch
 import torch.distributed as dist
 from PIL import Image
 
-from tests.conftest import OmniRunner
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniRunner
 from vllm_omni.diffusion.data import DiffusionParallelConfig
 from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 from vllm_omni.platforms import current_omni_platform

tests/e2e/offline_inference/test_flux_autoround_w4a16.py

@@ -14,8 +14,9 @@ import pytest
 import torch
 from vllm.distributed.parallel_state import cleanup_dist_env_and_memory
 
-from tests.conftest import OmniRunner
-from tests.utils import DeviceMemoryMonitor, hardware_test
+from tests.helpers.env import DeviceMemoryMonitor
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniRunner
 from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 from vllm_omni.outputs import OmniRequestOutput
 from vllm_omni.platforms import current_omni_platform

tests/e2e/offline_inference/test_flux_kontext.py

@@ -13,7 +13,7 @@ import pytest
 from PIL import Image
 from vllm.assets.image import ImageAsset
 
-from tests.conftest import OmniRunner
+from tests.helpers.runtime import OmniRunner
 from vllm_omni.diffusion.data import DiffusionParallelConfig
 from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 

tests/e2e/offline_inference/test_hunyuanimage3_text2img.py

@@ -8,7 +8,7 @@ import torch.nn.functional as F
 from PIL import Image
 from transformers import CLIPModel, CLIPProcessor
 
-from tests.conftest import OmniRunner
+from tests.helpers.runtime import OmniRunner
 from vllm_omni import Omni
 from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 from vllm_omni.platforms import current_omni_platform

tests/e2e/offline_inference/test_ltx2_cfg_parallel_parity.py

@@ -11,7 +11,7 @@ import numpy as np
 import pytest
 from PIL import Image
 
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
 
 REPO_ROOT = Path(__file__).resolve().parents[3]
 T2V_EXAMPLE = REPO_ROOT / "examples" / "offline_inference" / "text_to_video" / "text_to_video.py"

tests/e2e/offline_inference/test_magi_human.py

@@ -8,8 +8,8 @@ import av
 import numpy as np
 import pytest
 
-from tests.conftest import OmniRunner
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniRunner
 from vllm_omni.diffusion.utils.media_utils import mux_video_audio_bytes
 from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 

tests/e2e/offline_inference/test_mammoth_moda2.py

@@ -23,8 +23,10 @@ import pytest
 import torch
 from vllm.sampling_params import SamplingParams
 
-from tests.conftest import OmniRunner
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniRunner
+
+os.environ["VLLM_TEST_CLEAN_GPU_MEMORY"] = "1"
 
 # ---------------------------------------------------------------------------
 # Constants

tests/e2e/offline_inference/test_ming_flash_omni.py

@@ -10,13 +10,13 @@ from pathlib import Path
 
 import pytest
 
-from tests.conftest import (
+from tests.helpers.mark import hardware_test
+from tests.helpers.media import (
     generate_synthetic_audio,
     generate_synthetic_image,
     generate_synthetic_video,
-    modify_stage_config,
 )
-from tests.utils import hardware_test
+from tests.helpers.stage_config import modify_stage_config
 
 models = ["Jonathan1909/Ming-flash-omni-2.0"]
 

tests/e2e/offline_inference/test_omnivoice.py

@@ -16,8 +16,8 @@ from pathlib import Path
 import numpy as np
 import pytest
 
-from tests.conftest import OmniRunner
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniRunner
 
 MODEL = "k2-fsa/OmniVoice"
 

tests/e2e/offline_inference/test_ovis_image.py

@@ -16,7 +16,7 @@ import pytest
 import torch
 from pytest_mock import MockerFixture
 
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
 from vllm_omni.diffusion.data import OmniDiffusionConfig, TransformerConfig
 
 # Mock the OvisImageTransformer2DModel to avoid complex init if needed,

tests/e2e/offline_inference/test_quantization_fp8.py

@@ -36,8 +36,8 @@ from typing import Any
 import pytest
 import torch
 
-from tests.conftest import OmniRunner
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniRunner
 from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 from vllm_omni.outputs import OmniRequestOutput
 from vllm_omni.platforms import current_omni_platform

tests/e2e/offline_inference/test_qwen2_5_omni.py

@@ -4,13 +4,13 @@ E2E tests for Qwen2.5-Omni model with mixed modality inputs, audio and text outp
 
 import pytest
 
-from tests.conftest import (
+from tests.helpers.mark import hardware_test
+from tests.helpers.media import (
     generate_synthetic_audio,
     generate_synthetic_image,
     generate_synthetic_video,
-    modify_stage_config,
 )
-from tests.utils import get_deploy_config_path, hardware_test
+from tests.helpers.stage_config import get_deploy_config_path, modify_stage_config
 from vllm_omni.platforms import current_omni_platform
 
 models = ["Qwen/Qwen2.5-Omni-7B"]

tests/e2e/offline_inference/test_qwen3_omni.py

@@ -9,11 +9,9 @@ os.environ["VLLM_TEST_CLEAN_GPU_MEMORY"] = "0"
 
 import pytest
 
-from tests.conftest import (
-    generate_synthetic_video,
-    modify_stage_config,
-)
-from tests.utils import get_deploy_config_path, hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.media import generate_synthetic_video
+from tests.helpers.stage_config import get_deploy_config_path, modify_stage_config
 from vllm_omni.platforms import current_omni_platform
 
 models = ["Qwen/Qwen3-Omni-30B-A3B-Instruct"]

tests/e2e/offline_inference/test_qwen3_tts_base.py

@@ -15,8 +15,8 @@ os.environ["VLLM_TEST_CLEAN_GPU_MEMORY"] = "0"
 
 import pytest
 
-from tests.conftest import modify_stage_config
-from tests.utils import get_deploy_config_path, hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.stage_config import get_deploy_config_path, modify_stage_config
 
 MODEL = "Qwen/Qwen3-TTS-12Hz-0.6B-Base"
 REF_AUDIO_URL = "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen3-TTS-Repo/clone_2.wav"

tests/e2e/offline_inference/test_qwen3_tts_customvoice.py

@@ -15,8 +15,8 @@ os.environ["VLLM_TEST_CLEAN_GPU_MEMORY"] = "0"
 
 import pytest
 
-from tests.conftest import modify_stage_config
-from tests.utils import get_deploy_config_path, hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.stage_config import get_deploy_config_path, modify_stage_config
 
 MODEL = "Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice"
 

tests/e2e/offline_inference/test_qwen_image_diffusion_batching.py

@@ -36,8 +36,8 @@ from pathlib import Path
 import pytest
 import torch
 
-from tests.conftest import OmniRunner
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniRunner
 from vllm_omni.entrypoints.async_omni import AsyncOmni
 from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 from vllm_omni.outputs import OmniRequestOutput

tests/e2e/offline_inference/test_sequence_parallel.py

@@ -20,8 +20,8 @@ import torch
 import torch.distributed as dist
 from PIL import Image
 
-from tests.conftest import OmniRunner
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniRunner
 from vllm_omni.diffusion.data import DiffusionParallelConfig
 from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 from vllm_omni.platforms import current_omni_platform

tests/e2e/offline_inference/test_stable_audio_expansion.py

@@ -15,8 +15,8 @@ import numpy as np
 import pytest
 import torch
 
-from tests.conftest import assert_audio_valid
-from tests.utils import hardware_test
+from tests.helpers.assertions import assert_audio_valid
+from tests.helpers.mark import hardware_test
 from vllm_omni import Omni
 from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 from vllm_omni.outputs import OmniRequestOutput

tests/e2e/offline_inference/test_t2i_model.py

@@ -1,7 +1,7 @@
 import pytest
 import torch
 
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
 from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 from vllm_omni.outputs import OmniRequestOutput
 from vllm_omni.platforms import current_omni_platform

tests/e2e/offline_inference/test_t2v_model.py

@@ -3,7 +3,7 @@ import os
 import pytest
 import torch
 
-from tests.conftest import OmniRunner
+from tests.helpers.runtime import OmniRunner
 from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 from vllm_omni.outputs import OmniRequestOutput
 from vllm_omni.platforms import current_omni_platform

tests/e2e/offline_inference/test_teacache.py

@@ -11,8 +11,8 @@ It uses minimal settings to keep test time short for CI.
 import pytest
 import torch
 
-from tests.conftest import OmniRunner
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniRunner
 from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 from vllm_omni.outputs import OmniRequestOutput
 from vllm_omni.platforms import current_omni_platform

tests/e2e/offline_inference/test_vae_decode_parallelism.py

@@ -18,7 +18,7 @@ if str(REPO_ROOT) not in sys.path:
 
 import time
 
-from tests.conftest import OmniRunner
+from tests.helpers.runtime import OmniRunner
 from vllm_omni.platforms import current_omni_platform
 
 # os.environ["VLLM_TEST_CLEAN_GPU_MEMORY"] = "1"

tests/e2e/offline_inference/test_voxcpm.py

@@ -12,9 +12,9 @@ import numpy as np
 import pytest
 import torch
 
-import tests.conftest as omni_test_conftest
-from tests.conftest import OmniRunner
-from tests.utils import hardware_test
+import tests.helpers.runtime as omni_runtime
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniRunner
 from vllm_omni.model_executor.models.voxcpm.voxcpm_runtime_utils import (
     prepare_voxcpm_hf_config_dir,
     resolve_voxcpm_model_dir,
@@ -30,7 +30,7 @@ SAMPLE_RATE = 24000
 @pytest.fixture(autouse=True)
 def _patch_npu_cleanup_for_voxcpm(monkeypatch: pytest.MonkeyPatch):
     """Limit the NPU cleanup workaround to this VoxCPM test module only."""
-    original_cleanup = omni_test_conftest.cleanup_dist_env_and_memory
+    original_cleanup = omni_runtime.cleanup_dist_env_and_memory
 
     def _safe_cleanup() -> None:
         try:
@@ -40,7 +40,7 @@ def _patch_npu_cleanup_for_voxcpm(monkeypatch: pytest.MonkeyPatch):
                 return
             raise
 
-    monkeypatch.setattr(omni_test_conftest, "cleanup_dist_env_and_memory", _safe_cleanup)
+    monkeypatch.setattr(omni_runtime, "cleanup_dist_env_and_memory", _safe_cleanup)
 
 
 def _build_prompt(text: str) -> dict[str, Any]:

tests/e2e/offline_inference/test_voxcpm2.py

@@ -5,8 +5,8 @@ import os
 import pytest
 import torch
 
-from tests.conftest import OmniRunner
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniRunner
 
 VOXCPM2_MODEL = "openbmb/VoxCPM2"
 STAGE_CONFIG = os.path.join(

tests/e2e/offline_inference/test_voxtral_tts.py

@@ -29,8 +29,9 @@ from mistral_common.protocol.speech.request import SpeechRequest
 from mistral_common.tokens.tokenizers.mistral import MistralTokenizer
 from vllm import SamplingParams
 
-from tests.conftest import OmniRunner, modify_stage_config
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniRunner
+from tests.helpers.stage_config import modify_stage_config
 from vllm_omni.entrypoints.async_omni import AsyncOmni
 
 MODEL = "mistralai/Voxtral-4B-TTS-2603"

tests/e2e/offline_inference/test_zimage_parallelism.py

@@ -20,8 +20,9 @@ import pytest
 import torch
 from PIL import Image
 
-from tests.conftest import OmniRunner
-from tests.utils import DeviceMemoryMonitor, hardware_test
+from tests.helpers.env import DeviceMemoryMonitor
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniRunner
 from vllm_omni.diffusion.data import DiffusionParallelConfig
 from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 from vllm_omni.outputs import OmniRequestOutput
@@ -213,7 +214,10 @@ def test_zimage_tensor_parallel_tp2(tmp_path: Path):
     )
 
 
-@pytest.mark.integration
+@pytest.mark.advanced_model
+@pytest.mark.diffusion
+@pytest.mark.parallel
+@hardware_test(res={"cuda": "L4", "rocm": "MI325"}, num_cards={"cuda": 4, "rocm": 2})
 def test_zimage_vae_patch_parallel_tp2(tmp_path: Path):
     if current_omni_platform.is_npu():
         pytest.skip("Z-Image VAE patch parallel e2e test is only supported on CUDA and ROCm for now.")

tests/e2e/online_serving/test_bagel_expansion.py

@@ -16,13 +16,8 @@ assert_diffusion_response validates successful generation and the expected
 
 import pytest
 
-from tests.conftest import (
-    OmniServer,
-    OmniServerParams,
-    OpenAIClientHandler,
-    dummy_messages_from_mix_data,
-)
-from tests.utils import hardware_marks
+from tests.helpers.mark import hardware_marks
+from tests.helpers.runtime import OmniServer, OmniServerParams, OpenAIClientHandler, dummy_messages_from_mix_data
 
 PROMPT = "A futuristic city skyline at twilight, cyberpunk style, ultra-detailed, high resolution."
 NEGATIVE_PROMPT = "low quality, blurry, distorted, deformed, watermark"
@@ -138,7 +133,7 @@ def test_bagel(
     - Ulysses-SP (degree=2)
     - Ring-Attention (degree=2)
 
-    Validation is delegated to assert_diffusion_response in tests.conftest,
+    Validation is delegated to assert_diffusion_response in tests/helpers/assertions.py,
     which checks output dimensions and basic correctness.
     """
 

tests/e2e/online_serving/test_bagel_online.py

@@ -28,8 +28,8 @@ from pathlib import Path
 import pytest
 from vllm.assets.image import ImageAsset
 
-from tests.conftest import OmniServerParams
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniServerParams
 
 os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
 os.environ["VLLM_TEST_CLEAN_GPU_MEMORY"] = "0"

tests/e2e/online_serving/test_cosyvoice3_tts.py

@@ -16,8 +16,8 @@ from pathlib import Path
 
 import pytest
 
-from tests.conftest import OmniServerParams
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniServerParams
 
 MODEL = "FunAudioLLM/Fun-CosyVoice3-0.5B-2512"
 

tests/e2e/online_serving/test_dynin_omni_expansion.py

@@ -15,9 +15,9 @@ import pytest
 import soundfile as sf
 from vllm.assets.image import ImageAsset
 
-from tests import conftest as tests_conftest
-from tests.conftest import OmniServerParams
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.media import convert_audio_bytes_to_text
+from tests.helpers.runtime import OmniServerParams
 
 os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
 os.environ["VLLM_TEST_CLEAN_GPU_MEMORY"] = "0"
@@ -87,7 +87,7 @@ def _convert_audio_bytes_to_text_without_ffmpeg(raw_bytes: bytes) -> str:
 @pytest.fixture
 def dynin_t2s_openai_client(openai_client, monkeypatch):
     monkeypatch.setattr(
-        tests_conftest,
+        convert_audio_bytes_to_text,
         "convert_audio_bytes_to_text",
         _convert_audio_bytes_to_text_without_ffmpeg,
     )

tests/e2e/online_serving/test_flux2_expansion.py

@@ -11,12 +11,8 @@ Coverage:
 
 import pytest
 
-from tests.conftest import (
-    OmniServer,
-    OmniServerParams,
-    OpenAIClientHandler,
-)
-from tests.utils import hardware_marks
+from tests.helpers.mark import hardware_marks
+from tests.helpers.runtime import OmniServer, OmniServerParams, OpenAIClientHandler
 
 FOUR_CARD_FEATURE_MARKS = hardware_marks(res={"cuda": "L4"}, num_cards=4)
 POSITIVE_PROMPT = "A cat sitting on a windowsill"

tests/e2e/online_serving/test_flux2_klein_inpaint_expansion.py

@@ -14,7 +14,7 @@ import httpx
 import pytest
 from PIL import Image, ImageDraw
 
-from tests.conftest import OmniServer, OmniServerParams
+from tests.helpers.runtime import OmniServer, OmniServerParams
 
 MODEL = "black-forest-labs/FLUX.2-klein-4B"
 

tests/e2e/online_serving/test_flux_2_dev_expansion.py

@@ -14,13 +14,8 @@ image resolution.
 
 import pytest
 
-from tests.conftest import (
-    OmniServer,
-    OmniServerParams,
-    OpenAIClientHandler,
-    dummy_messages_from_mix_data,
-)
-from tests.utils import hardware_marks
+from tests.helpers.mark import hardware_marks
+from tests.helpers.runtime import OmniServer, OmniServerParams, OpenAIClientHandler, dummy_messages_from_mix_data
 
 MODEL = "black-forest-labs/FLUX.2-dev"
 PROMPT = "A cinematic mountain landscape at sunrise, dramatic clouds, ultra-detailed, realistic photography."

tests/e2e/online_serving/test_flux_kontext_expansion.py

@@ -5,13 +5,8 @@ and are supported by the FluxKontext model.
 
 import pytest
 
-from tests.conftest import (
-    OmniServer,
-    OmniServerParams,
-    OpenAIClientHandler,
-    dummy_messages_from_mix_data,
-    generate_synthetic_image,
-)
+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."
 NEGATIVE_PROMPT = "blurry, low quality, modern, geometrist"

tests/e2e/online_serving/test_hunyuan_video_15_expansion.py

@@ -11,12 +11,8 @@ Coverage (H100, since model cannot fit L4):
 
 import pytest
 
-from tests.conftest import (
-    OmniServer,
-    OmniServerParams,
-    OpenAIClientHandler,
-)
-from tests.utils import hardware_marks
+from tests.helpers.mark import hardware_marks
+from tests.helpers.runtime import OmniServer, OmniServerParams, OpenAIClientHandler
 
 PROMPT = "A cat walking across a sunlit garden, cinematic lighting, slow motion."
 NEGATIVE_PROMPT = "low quality, blurry, distorted"

tests/e2e/online_serving/test_image_gen_edit.py

@@ -22,7 +22,7 @@ from PIL import Image
 from vllm.assets.image import ImageAsset
 from vllm.utils.network_utils import get_open_port
 
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
 
 os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
 # Increase timeout for downloading assets from S3 (default 5s is too short for CI)

tests/e2e/online_serving/test_images_generations_lora.py

@@ -22,8 +22,8 @@ import torch
 from PIL import Image
 from safetensors.torch import save_file
 
-from tests.conftest import OmniServer
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniServer
 
 os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
 

tests/e2e/online_serving/test_longcat_image_edit_expansion.py

@@ -13,14 +13,9 @@ This validates:
 
 import pytest
 
-from tests.conftest import (
-    OmniServer,
-    OmniServerParams,
-    OpenAIClientHandler,
-    dummy_messages_from_mix_data,
-    generate_synthetic_image,
-)
-from tests.utils import hardware_marks
+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 image into a cinematic animation style with vibrant colors and soft lighting."
 NEGATIVE_PROMPT = "blurry, low quality, distorted, oversaturated"

tests/e2e/online_serving/test_longcat_image_expansion.py

@@ -13,13 +13,8 @@ This validates:
 
 import pytest
 
-from tests.conftest import (
-    OmniServer,
-    OmniServerParams,
-    OpenAIClientHandler,
-    dummy_messages_from_mix_data,
-)
-from tests.utils import hardware_marks
+from tests.helpers.mark import hardware_marks
+from tests.helpers.runtime import OmniServer, OmniServerParams, OpenAIClientHandler, dummy_messages_from_mix_data
 
 TEXT_TO_IMAGE_PROMPT = (
     "A cinematic illustration of a cat typing on a silver laptop, soft window light, highly detailed."

tests/e2e/online_serving/test_mimo_audio.py

@@ -9,13 +9,10 @@ from pathlib import Path
 
 import pytest
 
-from tests.conftest import (
-    OmniServerParams,
-    dummy_messages_from_mix_data,
-    generate_synthetic_audio,
-    modify_stage_config,
-)
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.media import generate_synthetic_audio
+from tests.helpers.runtime import OmniServerParams, dummy_messages_from_mix_data
+from tests.helpers.stage_config import modify_stage_config
 from vllm_omni.model_executor.model_loader.weight_utils import (
     download_weights_from_hf_specific,
 )

tests/e2e/online_serving/test_ming_flash_omni.py

@@ -10,15 +10,14 @@ from pathlib import Path
 
 import pytest
 
-from tests.conftest import (
-    OmniServerParams,
-    dummy_messages_from_mix_data,
+from tests.helpers.mark import hardware_test
+from tests.helpers.media import (
     generate_synthetic_audio,
     generate_synthetic_image,
     generate_synthetic_video,
-    modify_stage_config,
 )
-from tests.utils import hardware_test
+from tests.helpers.runtime import OmniServerParams, dummy_messages_from_mix_data
+from tests.helpers.stage_config import modify_stage_config
 
 os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
 os.environ["VLLM_TEST_CLEAN_GPU_MEMORY"] = "0"

tests/e2e/online_serving/test_nextstep_expansion.py

@@ -6,13 +6,13 @@ import os
 
 import pytest
 
-from tests.conftest import (
+from tests.helpers.mark import hardware_marks
+from tests.helpers.runtime import (
     OmniServer,
     OmniServerParams,
     OpenAIClientHandler,
     dummy_messages_from_mix_data,
 )
-from tests.utils import hardware_marks
 
 # L4: 4 GPUs + TP=4; XPU B60: 2 cards (use num_cards={"cuda": 4, "xpu": 4} if needed)
 FOUR_CARD_MARKS = hardware_marks(

tests/e2e/online_serving/test_omnivoice.py

@@ -17,8 +17,9 @@ from pathlib import Path
 import httpx
 import pytest
 
-from tests.conftest import OmniServerParams, generate_synthetic_audio
-from tests.utils import hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.media import generate_synthetic_audio
+from tests.helpers.runtime import OmniServerParams
 
 try:
     from transformers import HiggsAudioV2TokenizerModel  # noqa: F401

tests/e2e/online_serving/test_qwen2_5_omni.py

@@ -6,15 +6,10 @@ import os
 
 import pytest
 
-from tests.conftest import (
-    OmniServerParams,
-    dummy_messages_from_mix_data,
-    generate_synthetic_audio,
-    generate_synthetic_image,
-    generate_synthetic_video,
-    modify_stage_config,
-)
-from tests.utils import get_deploy_config_path, hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.media import generate_synthetic_audio, generate_synthetic_image, generate_synthetic_video
+from tests.helpers.runtime import OmniServerParams, dummy_messages_from_mix_data
+from tests.helpers.stage_config import get_deploy_config_path, modify_stage_config
 
 os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
 os.environ["VLLM_TEST_CLEAN_GPU_MEMORY"] = "0"

tests/e2e/online_serving/test_qwen3_omni.py

@@ -6,15 +6,10 @@ import os
 
 import pytest
 
-from tests.conftest import (
-    OmniServerParams,
-    dummy_messages_from_mix_data,
-    generate_synthetic_audio,
-    generate_synthetic_image,
-    generate_synthetic_video,
-    modify_stage_config,
-)
-from tests.utils import get_deploy_config_path, hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.media import generate_synthetic_audio, generate_synthetic_image, generate_synthetic_video
+from tests.helpers.runtime import OmniServerParams, dummy_messages_from_mix_data
+from tests.helpers.stage_config import get_deploy_config_path, modify_stage_config
 from vllm_omni.platforms import current_omni_platform
 
 os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"

tests/e2e/online_serving/test_qwen3_omni_expansion.py

@@ -10,15 +10,10 @@ os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
 
 import pytest
 
-from tests.conftest import (
-    OmniServerParams,
-    dummy_messages_from_mix_data,
-    generate_synthetic_audio,
-    generate_synthetic_image,
-    generate_synthetic_video,
-    modify_stage_config,
-)
-from tests.utils import get_deploy_config_path, hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.media import generate_synthetic_audio, generate_synthetic_image, generate_synthetic_video
+from tests.helpers.runtime import OmniServerParams, dummy_messages_from_mix_data
+from tests.helpers.stage_config import get_deploy_config_path, modify_stage_config
 
 model = "Qwen/Qwen3-Omni-30B-A3B-Instruct"
 
@@ -455,7 +450,7 @@ def test_one_word_prompt_001(omni_server, openai_client) -> None:
         "key_words": {"text": ["london"]},
     }
 
-    # Retry only when assert_omni_response fails on text/audio cosine similarity (see tests/conftest.py).
+    # Retry only when assert_omni_response fails on text/audio cosine similarity (see tests/helpers/assertions.py).
     _similarity_assert_msg = "The audio content is not same as the text"
     _max_retries = 3
     for attempt in range(_max_retries):
@@ -520,7 +515,7 @@ def test_speaker_002(omni_server, openai_client) -> None:
         "key_words": {"text": ["beijing"]},
     }
 
-    # Retry only when assert_omni_response fails on preset voice gender (see tests/conftest.py).
+    # Retry only when assert_omni_response fails on preset voice gender (see tests/helpers/assertions.py).
     _gender_assert_substr = "estimated gender"
     _max_retries = 3
     for attempt in range(_max_retries):

tests/e2e/online_serving/test_qwen3_omni_realtime_websocket.py

@@ -16,14 +16,14 @@ import wave
 import pytest
 import websockets
 
-from tests.conftest import (
-    OmniServerParams,
+from tests.helpers.mark import hardware_test
+from tests.helpers.media import (
     convert_audio_bytes_to_text,
     cosine_similarity_text,
     generate_synthetic_audio,
-    modify_stage_config,
 )
-from tests.utils import get_deploy_config_path, hardware_test
+from tests.helpers.runtime import OmniServerParams
+from tests.helpers.stage_config import get_deploy_config_path, modify_stage_config
 
 os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
 

tests/e2e/online_serving/test_qwen3_tts_base.py

@@ -14,8 +14,9 @@ os.environ["VLLM_TEST_CLEAN_GPU_MEMORY"] = "0"
 
 import pytest
 
-from tests.conftest import OmniServerParams
-from tests.utils import get_deploy_config_path, hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniServerParams
+from tests.helpers.stage_config import get_deploy_config_path
 
 MODEL = "Qwen/Qwen3-TTS-12Hz-0.6B-Base"
 

tests/e2e/online_serving/test_qwen3_tts_base_expansion.py

@@ -14,8 +14,9 @@ os.environ["VLLM_TEST_CLEAN_GPU_MEMORY"] = "0"
 
 import pytest
 
-from tests.conftest import OmniServerParams
-from tests.utils import get_deploy_config_path, hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniServerParams
+from tests.helpers.stage_config import get_deploy_config_path
 
 MODEL = "Qwen/Qwen3-TTS-12Hz-0.6B-Base"
 

tests/e2e/online_serving/test_qwen3_tts_batch.py

@@ -22,12 +22,10 @@ import httpx
 import pytest
 import yaml
 
-from tests.conftest import (
-    OmniServer,
-    convert_audio_file_to_text,
-    cosine_similarity_text,
-)
-from tests.utils import get_deploy_config_path, hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.media import convert_audio_file_to_text, cosine_similarity_text
+from tests.helpers.runtime import OmniServer
+from tests.helpers.stage_config import get_deploy_config_path
 
 MODEL = "Qwen/Qwen3-TTS-12Hz-0.6B-CustomVoice"
 STAGE_INIT_TIMEOUT_S = 120

tests/e2e/online_serving/test_qwen3_tts_customvoice.py

@@ -14,8 +14,9 @@ os.environ["VLLM_TEST_CLEAN_GPU_MEMORY"] = "0"
 
 import pytest
 
-from tests.conftest import OmniServerParams
-from tests.utils import get_deploy_config_path, hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniServerParams
+from tests.helpers.stage_config import get_deploy_config_path
 
 MODEL = "Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice"
 

tests/e2e/online_serving/test_qwen3_tts_customvoice_expansion.py

@@ -14,8 +14,9 @@ os.environ["VLLM_TEST_CLEAN_GPU_MEMORY"] = "0"
 
 import pytest
 
-from tests.conftest import OmniServerParams
-from tests.utils import get_deploy_config_path, hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniServerParams
+from tests.helpers.stage_config import get_deploy_config_path
 
 MODEL = "Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice"
 

tests/e2e/online_serving/test_qwen3_tts_speaker_embedding.py

@@ -17,8 +17,9 @@ import struct
 import httpx
 import pytest
 
-from tests.conftest import OmniServer
-from tests.utils import get_deploy_config_path, hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniServer
+from tests.helpers.stage_config import get_deploy_config_path
 
 MODEL_BASE = "Qwen/Qwen3-TTS-12Hz-0.6B-Base"
 MODEL_BASE_1_7B = "Qwen/Qwen3-TTS-12Hz-1.7B-Base"

tests/e2e/online_serving/test_qwen3_tts_websocket.py

@@ -11,8 +11,9 @@ import os
 import pytest
 import websockets
 
-from tests.conftest import OmniServer
-from tests.utils import get_deploy_config_path, hardware_test
+from tests.helpers.mark import hardware_test
+from tests.helpers.runtime import OmniServer
+from tests.helpers.stage_config import get_deploy_config_path
 
 os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
 os.environ["VLLM_TEST_CLEAN_GPU_MEMORY"] = "0"

tests/e2e/online_serving/test_qwen_image_edit_expansion.py

@@ -7,14 +7,9 @@ and are supported by the following models:
 
 import pytest
 
-from tests.conftest import (
-    OmniServer,
-    OmniServerParams,
-    OpenAIClientHandler,
-    dummy_messages_from_mix_data,
-    generate_synthetic_image,
-)
-from tests.utils import hardware_marks
+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."
 MULTI_EDIT_PROMPT = (

tests/e2e/online_serving/test_qwen_image_expansion.py

@@ -12,13 +12,8 @@ See docs/user_guide/diffusion_acceleration.md.
 
 import pytest
 
-from tests.conftest import (
-    OmniServer,
-    OmniServerParams,
-    OpenAIClientHandler,
-    dummy_messages_from_mix_data,
-)
-from tests.utils import hardware_marks
+from tests.helpers.mark import hardware_marks
+from tests.helpers.runtime import OmniServer, OmniServerParams, OpenAIClientHandler, dummy_messages_from_mix_data
 
 T2I_PROMPT = "A photo of a cat sitting on a laptop keyboard, digital art style."
 NEGATIVE_PROMPT = "blurry, low quality"

tests/e2e/online_serving/test_qwen_image_layered_expansion.py

@@ -14,15 +14,9 @@ Total distinct features covered: cache_dit, Ulysses-SP, CFG-Parallel, layerwise-
 
 import pytest
 
-from tests.conftest import (
-    OmniServer,
-    OmniServerParams,
-    OpenAIClientHandler,
-    decode_b64_image,
-    dummy_messages_from_mix_data,
-    generate_synthetic_image,
-)
-from tests.utils import hardware_marks
+from tests.helpers.mark import hardware_marks
+from tests.helpers.media import decode_b64_image, generate_synthetic_image
+from tests.helpers.runtime import OmniServer, OmniServerParams, OpenAIClientHandler, dummy_messages_from_mix_data
 
 MODEL = "Qwen/Qwen-Image-Layered"
 EDIT_PROMPT = "Decompose this image into layers."

tests/e2e/online_serving/test_sd3_expansion.py

@@ -4,12 +4,8 @@ Tests for Stable Diffusion 3.5 medium model.
 
 import pytest
 
-from tests.conftest import (
-    OmniServer,
-    OmniServerParams,
-    OpenAIClientHandler,
-)
-from tests.utils import hardware_marks
+from tests.helpers.mark import hardware_marks
+from tests.helpers.runtime import OmniServer, OmniServerParams, OpenAIClientHandler
 
 FOUR_CARD_FEATURE_MARKS = hardware_marks(res={"cuda": "L4"}, num_cards=4)
 POSITIVE_PROMPT = "A serene mountain landscape at sunset"