add lb default

Author: lilei199908

slime/backends/megatron_utils/model_provider.py

@@ -33,6 +33,8 @@ def __init__(
         self.sequence_parallel = config.sequence_parallel
         if self.sequence_parallel:
             self.weight.sequence_parallel = True
+            if bias:
+                self.bias.sequence_parallel = True
 
         self.weight.data.normal_(mean=0.0, std=0.02)
         if bias:

slime/backends/sglang_utils/arguments.py

@@ -25,18 +25,6 @@ def add_sglang_router_arguments(parser):
         default=14400,
         help="Timeout for requests to the SGLang router in seconds",
     )
-    parser.add_argument(
-        "--sglang-router-balance-abs-threshold",
-        type=int,
-        default=10,
-        help="Absolute threshold for balance in the SGLang router",
-    )
-    parser.add_argument(
-        "--sglang-router-balance-rel-threshold",
-        type=float,
-        default=1.2,
-        help="Relative threshold for balance in the SGLang router",
-    )
     return parser
 
 
@@ -45,6 +33,7 @@ def add_sglang_arguments(parser):
     Add arguments to the parser for the SGLang server.
     """
     parser = add_sglang_router_arguments(parser)
+    parser.set_defaults(router_balance_abs_threshold=10, router_balance_rel_threshold=1.2)
     parser.add_argument("--sglang-server-concurrency", type=int, default=512)
 
     old_add_argument = parser.add_argument

slime/ray/rollout.py

@@ -659,7 +659,7 @@ def _start_router(args):
             router_args.pd_disaggregation = True
 
         logger.info(f"Launch router with args: {router_args}")
-
+    print("!!!", router_args.balance_abs_threshold)
     process = multiprocessing.Process(
         target=run_router,
         args=(router_args,),