Set vllm multiproc method to spawn (#3310)

* Use spawn instead of fork method for vllm

* Fix lint

Author: mreso

ts/torch_handler/vllm_handler.py

@@ -1,5 +1,6 @@
 import asyncio
 import logging
+import os
 import pathlib
 import time
 from unittest.mock import MagicMock
@@ -41,6 +42,8 @@ def initialize(self, ctx):
             ctx.model_yaml_config.get("handler", {})
         )
 
+        os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
+
         self.vllm_engine = AsyncLLMEngine.from_engine_args(vllm_engine_config)
 
         self.adapters = ctx.model_yaml_config.get("handler", {}).get("adapters", {})