diff --git a/python/sglang/srt/managers/scheduler.py b/python/sglang/srt/managers/scheduler.py index 8e32640fc..e37924650 100644 --- a/python/sglang/srt/managers/scheduler.py +++ b/python/sglang/srt/managers/scheduler.py @@ -2641,7 +2641,9 @@ class Scheduler( self.chunked_req = adder.add_chunked_req(self.chunked_req) if self.enable_lora: - running_loras = {req.lora_id for req in self.running_batch.reqs} + running_loras = { + req.lora_id for req in self.running_batch.reqs if not req.finished() + } # Account for LoRAs that are already loaded in the adder, such as chunked requests running_loras.update(req.lora_id for req in adder.can_run_list)