Beta spec-overlap for EAGLE (#11398)

Co-authored-by: Lianmin Zheng <15100009+merrymercy@users.noreply.github.com>
Co-authored-by: Hanming Lu <69857889+hanming-lu@users.noreply.github.com>

python/sglang/srt/server_args.py

@@ -312,6 +312,7 @@ class ServerArgs:
     nsa_decode: str = "fa3"
 
     # Speculative decoding
+    enable_beta_spec: bool = False
     speculative_algorithm: Optional[str] = None
     speculative_draft_model_path: Optional[str] = None
     speculative_draft_model_revision: Optional[str] = None
@@ -1103,11 +1104,19 @@ class ServerArgs:
                 )
             if self.max_running_requests is None:
                 self.max_running_requests = 48
-            self.disable_overlap_schedule = True
-            logger.warning(
-                "Overlap scheduler is disabled because of using "
-                "eagle speculative decoding."
-            )
+
+            if self.speculative_algorithm == "EAGLE" and self.enable_beta_spec:
+                self.disable_overlap_schedule = False
+                logger.warning(
+                    "Beta spec is enabled for eagle speculative decoding and overlap schedule is turned on."
+                )
+
+            if not self.enable_beta_spec:
+                self.disable_overlap_schedule = True
+                logger.warning(
+                    "Overlap scheduler is disabled because of using eagle3 and standalone speculative decoding."
+                )
+
             if self.enable_mixed_chunk:
                 self.enable_mixed_chunk = False
                 logger.warning(
@@ -2127,6 +2136,7 @@ class ServerArgs:
         )
 
         # Speculative decoding
+        parser.add_argument("--enable-beta-spec", action="store_true")
         parser.add_argument(
             "--speculative-algorithm",
             type=str,