diff --git a/python/sglang/srt/managers/expert_location.py b/python/sglang/srt/managers/expert_location.py
index 7391dc8c2..97d439558 100644
--- a/python/sglang/srt/managers/expert_location.py
+++ b/python/sglang/srt/managers/expert_location.py
@@ -136,7 +136,7 @@ class ExpertLocationMetadata:
         num_physical_experts = common["num_physical_experts"]
 
         phase = server_args.disaggregation_mode
-        if phase == "null":
+        if phase == "null" or model_config_for_expert_location.num_groups is None:
             phase = "decode"
 
         physical_to_logical_map, logical_to_all_physical_map, expert_count = (
diff --git a/python/sglang/srt/model_executor/expert_location_updater.py b/python/sglang/srt/model_executor/expert_location_updater.py
index ca31ef740..13c4adc8d 100644
--- a/python/sglang/srt/model_executor/expert_location_updater.py
+++ b/python/sglang/srt/model_executor/expert_location_updater.py
@@ -12,6 +12,7 @@
 # limitations under the License.
 # ==============================================================================
 import logging
+from datetime import timedelta
 from typing import Dict, List, Tuple
 
 import torch
@@ -340,7 +341,7 @@ def update_expert_weights_single_layer(
         reqs = torch.distributed.batch_isend_irecv(p2p_ops)
         try:
             for req in reqs:
-                req.wait(timeout=30)
+                req.wait(timeout=timedelta(seconds=30))
         except RuntimeError:
             logger.error(
                 f"Context: {rank=} {old_physical_to_logical_map=} {new_physical_to_logical_map=} {num_local_physical_experts=} {num_gpu_per_node=}"