Merge remote-tracking branch 'origin/main' into jthomson04/kvbm-trtll…

…m-disagg

components/src/dynamo/trtllm/main.py

@@ -117,10 +117,8 @@ def build_kv_connector_config(config: Config):
     return None
 
 
-@dynamo_worker(static=False)
-async def worker(runtime: DistributedRuntime):
-    # Set up signal handler for graceful shutdown
-    loop = asyncio.get_running_loop()
+async def worker():
+    config = cmd_line_args()
 
     loop = asyncio.get_running_loop()
     runtime = DistributedRuntime(loop, config.store_kv, config.request_plane)

components/src/dynamo/trtllm/utils/trtllm_utils.py

@@ -58,6 +58,8 @@ def __init__(self) -> None:
         self.tool_call_parser: Optional[str] = None
         self.dump_config_to: Optional[str] = None
         self.custom_jinja_template: Optional[str] = None
+        self.store_kv: str = ""
+        self.request_plane: str = ""
         self.connector: Optional[str] = None
 
     def __str__(self) -> str:
@@ -361,6 +363,8 @@ def cmd_line_args():
     config.reasoning_parser = args.dyn_reasoning_parser
     config.tool_call_parser = args.dyn_tool_call_parser
     config.dump_config_to = args.dump_config_to
+    config.store_kv = args.store_kv
+    config.request_plane = args.request_plane
     config.connector = args.connector
 
     # Handle custom jinja template path expansion (environment variables and home directory)

tests/kvbm_integration/test_determinism_disagg.py

@@ -509,9 +509,7 @@ def llm_server(request, runtime_services):
     elif importlib.util.find_spec("tensorrt_llm") is not None:
         server_type = ServerType.trtllm
     else:
-        raise Exception(
-            "Neither the vllm nor the tensorrt_llm module is available in the current environment."
-        )
+        pytest.skip("vllm module is not available in the current environment.")
 
     server_manager = LLMServerManager(
         port=port,