use vllm cli directly

Author: chibu

src/automation/tasks/scripts/arenahard_generate_script.py

@@ -73,7 +73,8 @@ def clean_hocon_value(v):
 
     # Start vLLM server
     server_process, server_initialized, server_log = start_vllm_server(
-        vllm_args,
+        #vllm_args,
+        "vllm",
         model_id,
         arenahard_generate_args["target"],
         args["Args"]["server_wait_time"],