[Test] Fit Qwen2.5 online serving placement to its two-card allocation

Signed-off-by: tzhouam <tzhouam@connect.ust.hk>

tests/e2e/online_serving/test_qwen2_5_omni_expansion.py

@@ -21,7 +21,13 @@ models = ["Qwen/Qwen2.5-Omni-7B"]
 
 # Single CI deploy YAML; rocm/xpu deltas are picked automatically via the
 # platforms: section in vllm_omni/deploy/ci/qwen2_5_omni.yaml.
-stage_configs = [modify_stage_config(get_deploy_config_path("ci/qwen2_5_omni.yaml"))]
+stage_configs = [
+    modify_stage_config(
+        get_deploy_config_path("ci/qwen2_5_omni.yaml"),
+        # This suite requests two cards; Code2Wav shares the Talker's card.
+        updates={"stages": {2: {"devices": "1"}}},
+    )
+]
 
 # Create parameter combinations for model and stage config
 test_params = [

tests/engine/test_async_omni_engine_stage_init.py

@@ -268,21 +268,11 @@ def test_stage_engine_core_client_module_reload_keeps_forward_refs_deferred():
     """Regression test for forward references in make_async_mp_client."""
     import vllm_omni.engine.stage_engine_core_client as client_mod
 
-    # Snapshot the module namespace so a reload cannot strand other test
-    # modules that captured the pre-reload class objects (isinstance checks
-    # against the stale class would then fail). ``importlib.reload`` mutates
-    # the module dict in place; restoring it leaves the runtime exactly as it
-    # was before this test ran.
-    saved = dict(client_mod.__dict__)
-    try:
-        importlib.reload(client_mod)
+    importlib.reload(client_mod)
 
-        assert client_mod.StageEngineCoreClientBase.make_async_mp_client.__annotations__["return"] == (
-            "StageEngineCoreClient | DPLBStageEngineCoreClient"
-        )
-    finally:
-        client_mod.__dict__.clear()
-        client_mod.__dict__.update(saved)
+    assert client_mod.StageEngineCoreClientBase.make_async_mp_client.__annotations__["return"] == (
+        "StageEngineCoreClient | DPLBStageEngineCoreClient"
+    )
 
 
 def test_async_omni_engine_initialize_stages_passes_log_stats_and_client_config_to_runtime(monkeypatch):