[router][ci] Quick Improvement to make CI more stable

sgl-router/py_test/e2e_response_api/backends/test_grpc_backend.py

@@ -25,11 +25,11 @@
 
 
 class TestGrpcBackend(StateManagementTests, MCPTests, StructuredOutputBaseTest):
-    """End to end tests for gRPC backend (Regular backend with Llama)."""
+    """End to end tests for gRPC backend (Regular backend with Qwen2.5)."""
 
     @classmethod
     def setUpClass(cls):
-        cls.model = "/home/ubuntu/models/meta-llama/Llama-3.1-8B-Instruct"
+        cls.model = "/home/ubuntu/models/Qwen/Qwen2.5-14B-Instruct"
         cls.base_url_port = "http://127.0.0.1:30030"
 
         cls.cluster = popen_launch_workers_and_router(
@@ -46,7 +46,7 @@ def setUpClass(cls):
                 "--history-backend",
                 "memory",
                 "--tool-call-parser",
-                "llama",
+                "qwen",
             ],
         )
 
@@ -152,7 +152,7 @@ def setUpClass(cls):
             cls.base_url_port,
             timeout=90,
             num_workers=1,
-            tp_size=2,
+            tp_size=4,
             policy="round_robin",
             worker_args=[
                 "--reasoning-parser=gpt-oss",

sgl-router/py_test/e2e_response_api/mixins/function_call.py

@@ -86,10 +86,16 @@ def test_basic_function_call(self):
                 },
             },
         ]
+        system_prompt = (
+            "You are a helpful assistant that can call functions. "
+            "When a user asks for horoscope information, call the function. "
+            "IMPORTANT: Don't reply directly to the user, only call the function. "
+        )
 
         # Create a running input list we will add to over time
         input_list = [
-            {"role": "user", "content": "What is my horoscope? I am an Aquarius."}
+            {"role": "system", "content": system_prompt},
+            {"role": "user", "content": "What is my horoscope? I am an Aquarius."},
         ]
 
         # 2. Prompt the model with tools defined