diff --git a/python/sglang/srt/dllm/mixin/scheduler.py b/python/sglang/srt/dllm/mixin/scheduler.py index 1ca55c222..656355e20 100644 --- a/python/sglang/srt/dllm/mixin/scheduler.py +++ b/python/sglang/srt/dllm/mixin/scheduler.py @@ -60,7 +60,7 @@ class SchedulerDllmMixin: def _fetch_waiting_reqs(self: Scheduler): # Calculate how many requests can be added to DLLM manager - max_dllm_capacity = self.server_args.max_running_requests - len( + max_dllm_capacity = self.dllm_config.max_running_requests - len( self.dllm_manager.waiting_queue ) num_requests_to_add = min(max_dllm_capacity, len(self.waiting_queue))