diff --git a/src/handler.py b/src/handler.py index 70dcea9..f3e2cc3 100644 --- a/src/handler.py +++ b/src/handler.py @@ -65,7 +65,7 @@ llm.engine._log_system_stats = lambda x, y: vllm_log_system_stats( def concurrency_controller() -> int: global TOTAL_RUNNING_JOBS - return min(0, MAX_CONCURRENCY - TOTAL_RUNNING_JOBS) + return max(0, MAX_CONCURRENCY - TOTAL_RUNNING_JOBS) def prepare_metrics() -> dict: