diff --git a/vllm/engine/arg_utils.py b/vllm/engine/arg_utils.py index 757023e12d439..2867532756450 100644 --- a/vllm/engine/arg_utils.py +++ b/vllm/engine/arg_utils.py @@ -1649,7 +1649,13 @@ class EngineArgs: "attention_backend and attention_config.backend " "are mutually exclusive" ) - attention_config.backend = self.attention_backend + # Convert string to enum if needed (CLI parsing returns a string) + if isinstance(self.attention_backend, str): + attention_config.backend = AttentionBackendEnum[ + self.attention_backend.upper() + ] + else: + attention_config.backend = self.attention_backend load_config = self.create_load_config()