[Perf] Optimize Higgs Audio v3 serving (#4204)

Signed-off-by: Sy03 <1370724210@qq.com>
Signed-off-by: Yueqian Lin <linyueqian@outlook.com>
Co-authored-by: Yueqian Lin <linyueqian@outlook.com>
Co-authored-by: Hongsheng Liu <liuhongsheng4@huawei.com>

tests/distributed/omni_connectors/test_adapter_and_flow.py

@@ -145,7 +145,7 @@ def test_shm_connector_flow(mocker: MockerFixture):
     Using real SharedMemoryConnector (inline mode for simplicity).
     """
     # 1. Setup Connector
-    config = {"shm_threshold_bytes": 1024}  # Large threshold to use inline
+    config = {"shm_threshold_bytes": 1024, "inline_small_payloads": True}
     connector = SharedMemoryConnector(config)
     connectors_map = {("0", "1"): connector}
 

tests/distributed/omni_connectors/test_basic_connectors.py

@@ -75,18 +75,21 @@ def test_create_unknown_connector():
 
 @pytest.fixture
 def shm_connector():
-    config = {"shm_threshold_bytes": 100}  # Small threshold for testing
+    config = {"shm_threshold_bytes": 100, "inline_small_payloads": True}
     return SharedMemoryConnector(config)
 
 
 def test_put_get_inline(shm_connector):
-    """Test transfer for small data (currently always uses SHM)."""
+    """Test inline transfer for small data."""
     data = {"small": "data"}
 
     success, size, metadata = shm_connector.put("stage_0", "stage_1", "req_1", data)
     assert success is True
-    assert "shm" in metadata
+    assert "inline_bytes" in metadata
+    assert "shm" not in metadata
     assert "size" in metadata
+    assert shm_connector._metrics["inline_writes"] == 1
+    assert shm_connector._metrics["shm_writes"] == 0
 
     # Retrieve
     retrieved_data, ret_size = shm_connector.get("stage_0", "stage_1", "req_1", metadata)

tests/distributed/omni_connectors/test_shm_connector.py

@@ -13,7 +13,7 @@ pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
 
 @pytest.fixture()
 def connector():
-    c = SharedMemoryConnector({"shm_threshold_bytes": 64})
+    c = SharedMemoryConnector({"shm_threshold_bytes": 0})
     yield c
     c.close()