[VLM] Optimize async mm data process mechanism (#12066)

Co-authored-by: luoyuan.luo <luoyuan.luo@antgroup.com>

python/sglang/srt/server_args.py

@@ -542,6 +542,10 @@ class ServerArgs:
     pdmux_config_path: Optional[str] = None
     sm_group_num: int = 8
 
+    # For Multi-Modal
+    mm_max_concurrent_calls: int = 32
+    mm_per_request_timeout: float = 10.0
+
     def __post_init__(self):
         """
         Orchestrates the handling of various server arguments, ensuring proper configuration and validation.
@@ -3519,6 +3523,20 @@ class ServerArgs:
             help="Read CLI options from a config file. Must be a YAML file with configuration options.",
         )
 
+        # For Multi-Modal
+        parser.add_argument(
+            "--mm-max-concurrent-calls",
+            type=int,
+            default=ServerArgs.mm_max_concurrent_calls,
+            help="The max concurrent calls for async mm data processing.",
+        )
+        parser.add_argument(
+            "--mm-per-request-timeout",
+            type=int,
+            default=ServerArgs.mm_per_request_timeout,
+            help="The timeout for each multi-modal request in seconds.",
+        )
+
     @classmethod
     def from_cli_args(cls, args: argparse.Namespace):
         args.tp_size = args.tensor_parallel_size