From 7717d0838b2b74313184a462bb6e08711a5950c6 Mon Sep 17 00:00:00 2001
From: coolcloudcol <131243218+coolcloudcol@users.noreply.github.com>
Date: Tue, 4 Jul 2023 06:22:28 +0800
Subject: [PATCH] Fix an endless loop issue when engine_step throws a
 RuntimeError (#339)

---
 vllm/engine/async_llm_engine.py | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/vllm/engine/async_llm_engine.py b/vllm/engine/async_llm_engine.py
index 4f3af70576e3..6806dda52a0a 100644
--- a/vllm/engine/async_llm_engine.py
+++ b/vllm/engine/async_llm_engine.py
@@ -144,7 +144,11 @@ class AsyncLLMEngine:
 
             # Kick the engine if the engine is not running.
             if not self.is_engine_running:
-                await self.engine_step(request_id)
+                try:
+                    await self.engine_step(request_id)
+                except RuntimeError as e:
+                    await self.abort(request_id)
+                    raise e
 
             # Wait for new output. The group_event will be set in engine_step
             # when there is new output available for the sequence group.