@@ -341,9 +341,10 @@ async def close(self):
341341
342342
343343def make_arg_parser (parser : FlexibleArgumentParser ) -> FlexibleArgumentParser :
344+ _is_multi_server = os .environ .get ("FD_ENABLE_MULTI_API_SERVER" ) == "1"
344345 parser .add_argument ("--port" , default = 8000 , type = int , help = "port to the http server" )
345346 parser .add_argument ("--host" , default = "0.0.0.0" , type = str , help = "host to the http server" )
346- parser .add_argument ("--workers" , default = 1 , type = int , help = "number of workers" )
347+ parser .add_argument ("--workers" , default = 1 if _is_multi_server else 4 , type = int , help = "number of workers" )
347348 parser .add_argument ("--metrics-port" , default = None , type = int , help = "port for metrics server" )
348349 parser .add_argument ("--controller-port" , default = - 1 , type = int , help = "port for controller server" )
349350 parser .add_argument (
@@ -352,7 +353,9 @@ def make_arg_parser(parser: FlexibleArgumentParser) -> FlexibleArgumentParser:
352353 type = int ,
353354 help = "max waiting time for connection, if set value -1 means no waiting time limit" ,
354355 )
355- parser .add_argument ("--max-concurrency" , default = 512 , type = int , help = "max concurrency" )
356+ parser .add_argument (
357+ "--max-concurrency" , default = 512 if _is_multi_server else 2048 , type = int , help = "max concurrency"
358+ )
356359
357360 parser .add_argument (
358361 "--enable-mm-output" , action = "store_true" , help = "Enable 'multimodal_content' field in response output. "
0 commit comments