[diffusion] feat: add performance mode server args (#24491)

test/manual/kv_transfer/test_mooncake_transfer_engine_init.py

@@ -59,12 +59,15 @@ def test_mooncake_te_condition(server_args: ServerArgs) -> bool:
             ib_device=ib_device,
         )
 
-    with patch(
-        "sglang.srt.distributed.device_communicators.mooncake_transfer_engine.init_mooncake_transfer_engine",
-        side_effect=_fake_init_mooncake_transfer_engine,
-    ), patch(
-        "sglang.srt.model_executor.model_runner.get_local_ip_auto",
-        return_value="127.0.0.1",
+    with (
+        patch(
+            "sglang.srt.distributed.device_communicators.mooncake_transfer_engine.init_mooncake_transfer_engine",
+            side_effect=_fake_init_mooncake_transfer_engine,
+        ),
+        patch(
+            "sglang.srt.model_executor.model_runner.get_local_ip_auto",
+            return_value="127.0.0.1",
+        ),
     ):
         ModelRunner.init_shared_mooncake_transfer_engine(dummy_runner)
 

test/manual/test_tokenizer_batch_encode.py

@@ -30,11 +30,13 @@ class TestTokenizerBatchEncode(unittest.TestCase):
         )
         self.port_args = PortArgs.init_new(self.server_args)
 
-        with patch("zmq.asyncio.Context"), patch(
-            "sglang.srt.utils.get_zmq_socket"
-        ), patch(
-            "sglang.srt.utils.hf_transformers_utils.get_tokenizer"
-        ) as mock_tokenizer:
+        with (
+            patch("zmq.asyncio.Context"),
+            patch("sglang.srt.utils.get_zmq_socket"),
+            patch(
+                "sglang.srt.utils.hf_transformers_utils.get_tokenizer"
+            ) as mock_tokenizer,
+        ):
 
             mock_tokenizer.return_value = Mock(vocab_size=32000)
             self.tokenizer_manager = TokenizerManager(self.server_args, self.port_args)

test/manual/test_tokenizer_manager.py

@@ -37,11 +37,13 @@ class TestInputFormatDetection(unittest.TestCase):
             self.server_args = ServerArgs(model_path=DEFAULT_SMALL_MODEL_NAME_FOR_TEST)
             self.port_args = PortArgs.init_new(self.server_args)
 
-        with patch("zmq.asyncio.Context"), patch(
-            "sglang.srt.utils.network.get_zmq_socket"
-        ), patch(
-            "sglang.srt.utils.hf_transformers_utils.get_tokenizer"
-        ) as mock_tokenizer:
+        with (
+            patch("zmq.asyncio.Context"),
+            patch("sglang.srt.utils.network.get_zmq_socket"),
+            patch(
+                "sglang.srt.utils.hf_transformers_utils.get_tokenizer"
+            ) as mock_tokenizer,
+        ):
             mock_tokenizer.return_value = Mock(vocab_size=32000)
             self.tokenizer_manager = TokenizerManager(self.server_args, self.port_args)
 
@@ -133,11 +135,13 @@ class TestTokenizerInputPreparation(unittest.TestCase):
             self.server_args = ServerArgs(model_path=DEFAULT_SMALL_MODEL_NAME_FOR_TEST)
             self.port_args = PortArgs.init_new(self.server_args)
 
-        with patch("zmq.asyncio.Context"), patch(
-            "sglang.srt.utils.network.get_zmq_socket"
-        ), patch(
-            "sglang.srt.utils.hf_transformers_utils.get_tokenizer"
-        ) as mock_tokenizer:
+        with (
+            patch("zmq.asyncio.Context"),
+            patch("sglang.srt.utils.network.get_zmq_socket"),
+            patch(
+                "sglang.srt.utils.hf_transformers_utils.get_tokenizer"
+            ) as mock_tokenizer,
+        ):
             mock_tokenizer.return_value = Mock(vocab_size=32000)
             self.tokenizer_manager = TokenizerManager(self.server_args, self.port_args)
 
@@ -191,11 +195,13 @@ class TestTokenizerResultExtraction(unittest.TestCase):
             self.server_args = ServerArgs(model_path=DEFAULT_SMALL_MODEL_NAME_FOR_TEST)
             self.port_args = PortArgs.init_new(self.server_args)
 
-        with patch("zmq.asyncio.Context"), patch(
-            "sglang.srt.utils.network.get_zmq_socket"
-        ), patch(
-            "sglang.srt.utils.hf_transformers_utils.get_tokenizer"
-        ) as mock_tokenizer:
+        with (
+            patch("zmq.asyncio.Context"),
+            patch("sglang.srt.utils.network.get_zmq_socket"),
+            patch(
+                "sglang.srt.utils.hf_transformers_utils.get_tokenizer"
+            ) as mock_tokenizer,
+        ):
             mock_tokenizer.return_value = Mock(vocab_size=32000)
             self.tokenizer_manager = TokenizerManager(self.server_args, self.port_args)
 
@@ -313,11 +319,13 @@ class TestTokenizerManagerIntegration(unittest.TestCase):
             self.server_args = ServerArgs(model_path=DEFAULT_SMALL_MODEL_NAME_FOR_TEST)
             self.port_args = PortArgs.init_new(self.server_args)
 
-        with patch("zmq.asyncio.Context"), patch(
-            "sglang.srt.utils.network.get_zmq_socket"
-        ), patch(
-            "sglang.srt.utils.hf_transformers_utils.get_tokenizer"
-        ) as mock_tokenizer:
+        with (
+            patch("zmq.asyncio.Context"),
+            patch("sglang.srt.utils.network.get_zmq_socket"),
+            patch(
+                "sglang.srt.utils.hf_transformers_utils.get_tokenizer"
+            ) as mock_tokenizer,
+        ):
             mock_tokenizer.return_value = Mock(vocab_size=32000)
             self.tokenizer_manager = TokenizerManager(self.server_args, self.port_args)