[Auto Sync] Update scheduler.py, server_args.py (20251014) (#11623)

Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: Stefan He <hebiaobuaa@gmail.com>

python/sglang/srt/managers/scheduler.py

@@ -658,6 +658,12 @@ class Scheduler(
     def launch_draft_worker(
         self, gpu_id, tp_rank, moe_ep_rank, server_args, port_args, dp_rank
     ):
+        if server_args.speculative_draft_load_format is not None:
+            server_args.load_format = server_args.speculative_draft_load_format
+            logger.info(
+                f"Using draft model load_format: '{server_args.speculative_draft_load_format}'"
+            )
+
         if self.spec_algorithm.is_eagle():
             from sglang.srt.speculative.eagle_worker import EAGLEWorker
             from sglang.srt.speculative.eagle_worker_v2 import EAGLEWorkerV2

python/sglang/srt/server_args.py

@@ -325,6 +325,7 @@ class ServerArgs:
     speculative_algorithm: Optional[str] = None
     speculative_draft_model_path: Optional[str] = None
     speculative_draft_model_revision: Optional[str] = None
+    speculative_draft_load_format: Optional[str] = None
     speculative_num_steps: Optional[int] = None
     speculative_eagle_topk: Optional[int] = None
     speculative_num_draft_tokens: Optional[int] = None
@@ -2223,6 +2224,15 @@ class ServerArgs:
             "name, a tag name, or a commit id. If unspecified, will use "
             "the default version.",
         )
+        parser.add_argument(
+            "--speculative-draft-load-format",
+            type=str,
+            default=ServerArgs.speculative_draft_load_format,
+            choices=LOAD_FORMAT_CHOICES,
+            help="The format of the draft model weights to load. "
+            "If not specified, will use the same format as --load-format. "
+            "Use 'dummy' to initialize draft model weights with random values for profiling.",
+        )
         parser.add_argument(
             "--speculative-num-steps",
             type=int,