Support dispatching logical to physical experts (#6385)

python/sglang/srt/server_args.py

@@ -170,6 +170,7 @@ class ServerArgs:
     enable_ep_moe: bool = False
     enable_deepep_moe: bool = False
     deepep_mode: Optional[Literal["auto", "normal", "low_latency"]] = "auto"
+    ep_dispatch_algorithm: Optional[Literal["static", "dynamic"]] = None
     init_expert_location: str = "trivial"
     expert_distribution_recorder_mode: Optional[
         Literal["stat", "per_pass", "per_token"]
@@ -1271,6 +1272,12 @@ class ServerArgs:
             default="auto",
             help="Select the mode when enable DeepEP MoE, could be `normal`, `low_latency` or `auto`. Default is `auto`, which means `low_latency` for decode batch and `normal` for prefill batch.",
         )
+        parser.add_argument(
+            "--ep-dispatch-algorithm",
+            type=str,
+            default=ServerArgs.ep_dispatch_algorithm,
+            help="The algorithm to choose ranks for redundant experts in expert parallel.",
+        )
         parser.add_argument(
             "--init-expert-location",
             type=str,