From 91f4522cbf85df0a65f619b25f6751edf2d5f0d6 Mon Sep 17 00:00:00 2001 From: William Lin Date: Wed, 21 Aug 2024 11:49:19 -0700 Subject: [PATCH] [multi-step] Raise error if not using async engine (#7703) --- vllm/engine/llm_engine.py | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/vllm/engine/llm_engine.py b/vllm/engine/llm_engine.py index 94aed6b8c..f72902c37 100644 --- a/vllm/engine/llm_engine.py +++ b/vllm/engine/llm_engine.py @@ -1302,6 +1302,11 @@ class LLMEngine: raise NotImplementedError( "Pipeline parallelism is only supported through AsyncLLMEngine " "as performance will be severely degraded otherwise.") + + if self.scheduler_config.num_scheduler_steps > 1: + raise NotImplementedError( + "Multiple scheduler steps (multi-step) are only supported " + "through AsyncLLMEngine. ") seq_group_metadata_list, scheduler_outputs = self.scheduler[ 0].schedule()