[smg][ci] rename 3rd models from cloud backend and delete dead code (#16692)

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

@@ -26,7 +26,7 @@ from .constants import (  # Enums; Convenience sets; Fixture parameters; Default
     Runtime,
     WorkerType,
 )
-from .gateway import Gateway, WorkerInfo
+from .gateway import Gateway, WorkerInfo, launch_cloud_gateway
 from .gpu_allocator import (
     GPUAllocator,
     GPUInfo,
@@ -54,6 +54,7 @@ from .model_specs import (  # Default model paths; Model groups
     FUNCTION_CALLING_MODELS,
     MODEL_SPECS,
     REASONING_MODELS,
+    THIRD_PARTY_MODELS,
 )
 from .process_utils import (
     detect_ib_device,
@@ -121,6 +122,7 @@ __all__ = [
     # Gateway
     "Gateway",
     "WorkerInfo",
+    "launch_cloud_gateway",
     # Default model paths
     "DEFAULT_MODEL_PATH",
     "DEFAULT_SMALL_MODEL_PATH",
@@ -135,6 +137,8 @@ __all__ = [
     "EMBEDDING_MODELS",
     "REASONING_MODELS",
     "FUNCTION_CALLING_MODELS",
+    # Third-party models
+    "THIRD_PARTY_MODELS",
     # Evaluation
     "run_eval",
 ]

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

@@ -554,3 +554,42 @@ class Gateway:
 
     def __exit__(self, exc_type, exc_val, exc_tb) -> None:
         self.shutdown()
+
+
+def launch_cloud_gateway(
+    runtime: str,  # "openai" or "xai"
+    *,
+    history_backend: str = "memory",
+    extra_args: list[str] | None = None,
+    timeout: float = 60,
+    show_output: bool | None = None,
+) -> Gateway:
+    """Launch gateway with cloud API runtime.
+
+    Args:
+        runtime: Cloud runtime ("openai" or "xai")
+        history_backend: History storage backend ("memory" or "oracle")
+        extra_args: Additional router arguments
+        timeout: Startup timeout in seconds
+        show_output: Show subprocess output
+
+    Returns:
+        Gateway instance with running router
+    """
+    from .model_specs import THIRD_PARTY_MODELS
+
+    if runtime not in THIRD_PARTY_MODELS:
+        raise ValueError(
+            f"Unknown cloud runtime: {runtime}. "
+            f"Available: {list(THIRD_PARTY_MODELS.keys())}"
+        )
+
+    gateway = Gateway()
+    gateway.start(
+        cloud_backend=runtime,
+        history_backend=history_backend,
+        timeout=timeout,
+        show_output=show_output,
+        extra_args=extra_args,
+    )
+    return gateway

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

@@ -131,3 +131,21 @@ DEFAULT_QWEN_FUNCTION_CALLING_MODEL_PATH = MODEL_SPECS["qwen-7b"]["model"]
 DEFAULT_MISTRAL_FUNCTION_CALLING_MODEL_PATH = MODEL_SPECS["mistral-7b"]["model"]
 DEFAULT_GPT_OSS_MODEL_PATH = MODEL_SPECS["gpt-oss"]["model"]
 DEFAULT_EMBEDDING_MODEL_PATH = MODEL_SPECS["embedding"]["model"]
+
+
+# =============================================================================
+# Third-party model configurations (cloud APIs)
+# =============================================================================
+
+THIRD_PARTY_MODELS: dict[str, dict] = {
+    "openai": {
+        "description": "OpenAI API",
+        "model": "gpt-5-nano",
+        "api_key_env": "OPENAI_API_KEY",
+    },
+    "xai": {
+        "description": "xAI API",
+        "model": "grok-4-fast",
+        "api_key_env": "XAI_API_KEY",
+    },
+}