[Feature] consider multimodal model when dummy run (#6045)

* add mm do profile

* updata code

* update code

* update code

* update code

* update test case

* update code

* update code

* fix xpu bug

* update code

* add mm do profile

* update test case

* update code

fastdeploy/worker/worker_process.py

@@ -1026,6 +1026,13 @@ def parse_args():
         help="Enable output of token-level entropy.",
     )
 
+    parser.add_argument(
+        "--mm_max_tokens_per_item",
+        type=json.loads,
+        default=None,
+        help="Maximum tokens per item in mm input.",
+    )
+
     parser.add_argument(
         "--num_cpu_blocks",
         type=int,