[perf]Qwen3-Omni performance optimization (#3164)

Signed-off-by: amy-why-3459 <wuhaiyan17@huawei.com>

tests/dfx/conftest.py

@@ -70,17 +70,17 @@ def _build_serve_args(serve_args: Any) -> list[str]:
 def create_unique_server_params(
     configs: list[dict[str, Any]],
     stage_configs_dir: Path,
-) -> list[tuple[str, str, str | None, str | None, tuple[str, ...]]]:
-    """Return one row per unique server configuration (same 5-tuple shape as upstream).
+) -> list[tuple[str, str, str | None, str | None, tuple[str, ...], bool]]:
+    """Return one row per unique server configuration.
 
-    ``(test_name, model, deploy_yaml_path, stage_overrides_json, extra_cli_args)``.
+    ``(test_name, model, deploy_yaml_path, stage_overrides_json, extra_cli_args, use_omni)``.
 
     JSON ``server_params.serve_args`` (dict/list) is expanded via ``_build_serve_args``
     and **prepended** to ``extra_cli_args`` so perf / stability ``omni_server`` fixtures
     stay identical to main while still honoring ``serve_args`` in benchmark JSON.
     """
-    unique_params: list[tuple[str, str, str | None, str | None, tuple[str, ...]]] = []
-    seen: set[tuple[str, str, str | None, str | None, tuple[str, ...]]] = set()
+    unique_params: list[tuple[str, str, str | None, str | None, tuple[str, ...], bool]] = []
+    seen: set[tuple[str, str, str | None, str | None, tuple[str, ...], bool]] = set()
     for config in configs:
         test_name = config["test_name"]
         server_params = config["server_params"]
@@ -104,8 +104,16 @@ def create_unique_server_params(
         serve_flat = _build_serve_args(server_params.get("serve_args"))
         raw_extra = tuple(server_params.get("extra_cli_args") or ())
         extra_cli_args = tuple(serve_flat) + raw_extra
+        use_omni = bool(server_params.get("use_omni", True))
 
-        server_param = (test_name, model, stage_config_path, stage_overrides_json, extra_cli_args)
+        server_param = (
+            test_name,
+            model,
+            stage_config_path,
+            stage_overrides_json,
+            extra_cli_args,
+            use_omni,
+        )
         if server_param not in seen:
             seen.add(server_param)
             unique_params.append(server_param)

tests/dfx/perf/scripts/run_benchmark.py

@@ -65,7 +65,7 @@ def omni_server(request):
     Multi-stage initialization can take 10-20+ minutes.
     """
     with _omni_server_lock:
-        test_name, model, stage_config_path, stage_overrides, extra_cli_args = request.param
+        test_name, model, stage_config_path, stage_overrides, extra_cli_args, use_omni = request.param
 
         print(f"Starting OmniServer with test: {test_name}, model: {model}")
 
@@ -78,7 +78,7 @@ def omni_server(request):
             server_args = ["--stage-overrides", stage_overrides] + server_args
         if extra_cli_args:
             server_args = list(extra_cli_args) + server_args
-        with OmniServer(model, server_args) as server:
+        with OmniServer(model, server_args, use_omni=use_omni) as server:
             server.test_name = test_name
             print("OmniServer started successfully")
             yield server

tests/dfx/stability/conftest.py

@@ -35,10 +35,10 @@ def omni_server(request: pytest.FixtureRequest):
     """Start OmniServer for stability tests, with per-module timeout override."""
     timeout_args = getattr(request.module, "STABILITY_SERVER_TIMEOUT_ARGS", DEFAULT_STABILITY_SERVER_TIMEOUT_ARGS)
     with _omni_server_lock:
-        # Same 5-tuple and CLI composition as ``tests/dfx/perf/scripts/run_benchmark.py`` on main;
+        # Same tuple and CLI composition as ``tests/dfx/perf/scripts/run_benchmark.py``;
         # ``serve_args`` from JSON are folded into ``extra_cli_args`` inside
         # ``create_unique_server_params``.
-        test_name, model, deploy_path, stage_overrides, extra_cli_args = request.param
+        test_name, model, deploy_path, stage_overrides, extra_cli_args, use_omni = request.param
 
         print(f"Starting OmniServer with test: {test_name}, model: {model}")
         server_args = list(timeout_args)
@@ -48,7 +48,7 @@ def omni_server(request: pytest.FixtureRequest):
             server_args = ["--stage-overrides", stage_overrides] + server_args
         if extra_cli_args:
             server_args = list(extra_cli_args) + server_args
-        with OmniServer(model, server_args) as server:
+        with OmniServer(model, server_args, use_omni=use_omni) as server:
             server.test_name = test_name
             print("OmniServer started successfully")
             yield server