[smg][ci] Migrate Response API e2e tests to shared infrastructure (#16680)

Co-authored-by: Chang Su <chang.s.su@oracle.com>
Co-authored-by: key4ng <rukeyang@gmail.com>

sgl-model-gateway/e2e_test/infra/model_pool.py

@@ -494,6 +494,11 @@ class ModelPool:
             if ib_device:
                 cmd.extend(["--disaggregation-ib-device", ib_device])
 
+        # Additional worker args from model spec (e.g., --context-length)
+        worker_args = spec.get("worker_args", [])
+        if worker_args:
+            cmd.extend(worker_args)
+
         # Build key based on worker type (or use custom key)
         if instance_key:
             key = instance_key

sgl-model-gateway/e2e_test/infra/model_specs.py

@@ -46,6 +46,16 @@ MODEL_SPECS: dict[str, dict] = {
         "tp": 1,
         "features": ["chat", "streaming", "function_calling", "pythonic_tools"],
     },
+    # Function calling specialist (larger, for Response API tests)
+    "qwen-14b": {
+        "model": _resolve_model_path("Qwen/Qwen2.5-14B-Instruct"),
+        "memory_gb": 28,
+        "tp": 2,
+        "features": ["chat", "streaming", "function_calling", "pythonic_tools"],
+        "worker_args": [
+            "--context-length=1000"
+        ],  # Faster startup, prevents memory issues
+    },
     # Reasoning model
     "deepseek-7b": {
         "model": _resolve_model_path("deepseek-ai/DeepSeek-R1-Distill-Qwen-7B"),