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 3, 10:30 UTC
Fix `--convert` passed without `--runner` on causal models (#42935)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Harry Mellor committed
May 19, 2026 at 00:56 UTC
b12745e4f31ffacf401cc20a97c592d6a49f3269
1 file changed
+8
-2
vllm/config/model.py
+8
-2
@@ -546,7 +546,9 @@ class ModelConfig:
546
is_generative_model = registry.is_text_generation_model(architectures, self)
547
is_pooling_model = registry.is_pooling_model(architectures, self)
548
549
- self.runner_type = self._get_runner_type(architectures, self.runner)
549
+ self.runner_type = self._get_runner_type(
550
+ architectures, self.runner, self.convert
551
+ )
552
self.convert_type = self._get_convert_type(
553
architectures, self.runner_type, self.convert
554
)
@@ -884,11 +886,15 @@ class ModelConfig:
886
self,
887
architectures: list[str],
888
runner: RunnerOption,
889
+ convert: ConvertOption,
890
) -> RunnerType:
891
if runner != "auto":
892
return runner
893
891
- runner_type = self._get_default_runner_type(architectures)
894
+ if convert in {"auto", "none"}:
895
+ runner_type = self._get_default_runner_type(architectures)
896
+ else:
897
+ runner_type = "pooling"
898
899
# Don't log the most common case
900
if runner_type != "generate":