From 9fc5511cd463263c09eebb8549a42d73642fc897 Mon Sep 17 00:00:00 2001 From: Happy Date: Wed, 24 Sep 2025 15:39:47 +0800 Subject: [PATCH] Update vllm.py --- xtuner/v1/ray/rollout/vllm.py | 2 ++ 1 file changed, 2 insertions(+) diff --git a/xtuner/v1/ray/rollout/vllm.py b/xtuner/v1/ray/rollout/vllm.py index 902f76033..605975dc9 100644 --- a/xtuner/v1/ray/rollout/vllm.py +++ b/xtuner/v1/ray/rollout/vllm.py @@ -122,6 +122,8 @@ def _transform_rollout_config_to_server_configs(self) -> Namespace: args.disable_log_requests = True args.disable_log_stats = True args.tensor_parallel_size = self.config.tensor_parallel_size + # https://github.com/vllm-project/vllm/blob/main/docs/serving/openai_compatible_server.md?plain=1#L38 + args.generation_config = 'vllm' if args.expert_parallel_size > 1: args.tensor_parallel_size = self.config.expert_parallel_size args.enable_expert_parallel = True