A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 11, 13:45 UTC
[Bug] Fix attention_backend arg string parsing (#30534)
Signed-off-by: mgoin <mgoin64@gmail.com>
Michael Goin committed
Dec 12, 2025 at 10:40 UTC
09ad3b76b320fffcb6b0214bd90851c3328581ea
1 file changed
+7
-1
vllm/engine/arg_utils.py
+7
-1
@@ -1649,7 +1649,13 @@ class EngineArgs:
1649
"attention_backend and attention_config.backend "
1650
"are mutually exclusive"
1651
)
1652
- attention_config.backend = self.attention_backend
1652
+ # Convert string to enum if needed (CLI parsing returns a string)
1653
+ if isinstance(self.attention_backend, str):
1654
+ attention_config.backend = AttentionBackendEnum[
1655
+ self.attention_backend.upper()
1656
+ ]
1657
+ else:
1658
+ attention_config.backend = self.attention_backend
1659
1660
load_config = self.create_load_config()
1661