x

app/engine.py

@@ -22,6 +22,9 @@ class InferenceEngine:
             revision=settings.revision,
         )
 
+    def close(self) -> None:
+        return None
+
     def generate(self, req: GenerateRequest) -> GenerateResponse:
         sampling_params = SamplingParams(
             temperature=req.temperature,

app/main.py

@@ -25,6 +25,8 @@ async def lifespan(_: FastAPI):
     settings = get_settings()
     engine = InferenceEngine(settings)
     yield
+    if engine is not None:
+        engine.close()
     engine = None
 
 

app/model_catalog.py

@@ -43,11 +43,12 @@ def _resolve_profile_model_path(profile: dict[str, Any], model_root: str, model_
         raise ValueError(f"model profile '{model_key}' must provide local_path")
     if "://" in local_path:
         raise ValueError(f"model profile '{model_key}' local_path must be local filesystem path")
-    if local_path.startswith("/"):
-        return local_path
-    if not model_root:
-        raise ValueError("config.json model_root cannot be empty when local_path is relative")
-    return _join_posix(model_root, local_path)
+    resolved = local_path
+    if not local_path.startswith("/"):
+        if not model_root:
+            raise ValueError("config.json model_root cannot be empty when local_path is relative")
+        resolved = _join_posix(model_root, local_path)
+    return resolved
 
 
 def load_catalog(catalog_path: str = "config.json") -> dict[str, Any]:

app/start_openai.py

@@ -18,9 +18,9 @@ def build_command() -> list[str]:
         os.environ[key] = value
     host = str(runtime["openai_host"])
     port = str(runtime["openai_port"])
+    api_key = runtime["api_key"] or ""
     dtype = str(runtime["dtype"])
     revision = runtime["revision"] or ""
-    api_key = runtime["api_key"] or ""
     cmd = [
         sys.executable,
         "-m",