From 6bab3486826d896fd76f15c6c08675a4d8002566 Mon Sep 17 00:00:00 2001
From: Chi Zhang <zhangchi.usc1992@bytedance.com>
Date: Mon, 25 Aug 2025 19:10:53 +0800
Subject: [PATCH] Revert "[rollout] feat: use dummy load_format when init
 AsyncServer (#3184)"

This reverts commit 2c7a9c5708c022ac6a39c3c155b09185c23a3d6e.
---
 verl/workers/rollout/vllm_rollout/vllm_async_server.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/verl/workers/rollout/vllm_rollout/vllm_async_server.py b/verl/workers/rollout/vllm_rollout/vllm_async_server.py
index 099b3f3b052..2a50a11a7b7 100644
--- a/verl/workers/rollout/vllm_rollout/vllm_async_server.py
+++ b/verl/workers/rollout/vllm_rollout/vllm_async_server.py
@@ -276,7 +276,7 @@ async def init_engine(self):
             skip_tokenizer_init=False,
             max_model_len=self.max_model_len,
             max_num_seqs=config.max_num_seqs,
-            load_format=config.load_format,
+            load_format="auto",
             disable_log_stats=config.disable_log_stats,
             max_num_batched_tokens=max_num_batched_tokens,
             enable_chunked_prefill=config.enable_chunked_prefill,