Improve the metrics for PD (#12580)

Co-authored-by: Kan Wu <wukanustc@gmail.com>
Co-authored-by: cctry <shiyang@x.ai>

python/sglang/srt/server_args.py

@@ -552,6 +552,10 @@ class ServerArgs:
     mm_max_concurrent_calls: int = 32
     mm_per_request_timeout: float = 10.0
 
+    # For checkpoint decryption
+    decrypted_config_file: Optional[str] = None
+    decrypted_draft_config_file: Optional[str] = None
+
     def __post_init__(self):
         """
         Orchestrates the handling of various server arguments, ensuring proper configuration and validation.
@@ -3551,6 +3555,20 @@ class ServerArgs:
             help="The timeout for each multi-modal request in seconds.",
         )
 
+        # For checkpoint decryption
+        parser.add_argument(
+            "--decrypted-config-file",
+            type=str,
+            default=ServerArgs.decrypted_config_file,
+            help="The path of the decrypted config file.",
+        )
+        parser.add_argument(
+            "--decrypted-draft-config-file",
+            type=str,
+            default=ServerArgs.decrypted_draft_config_file,
+            help="The path of the decrypted draft config file.",
+        )
+
     @classmethod
     def from_cli_args(cls, args: argparse.Namespace):
         args.tp_size = args.tensor_parallel_size