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 12, 14:45 UTC
[Bugfix] Make deprecated `--task embedding` consistent with `--runner… (#29312)
Signed-off-by: Maryam Tahhan <mtahhan@redhat.com>
Maryam Tahhan committed
Nov 25, 2025 at 03:19 UTC
87185c88d54bd97c4c08f1fd3c5a8564e4924e2a
1 file changed
+20
-10
vllm/config/model.py
+20
-10
@@ -585,16 +585,26 @@ class ModelConfig:
585
else: # task == "auto"
586
pass
587
else:
588
- debug_info = {
589
- "architectures": architectures,
590
- "is_generative_model": is_generative_model,
591
- "is_pooling_model": is_pooling_model,
592
- }
593
- raise AssertionError(
594
- "The model should be a generative or "
595
- "pooling model when task is set to "
596
- f"{self.task!r}. Found: {debug_info}"
597
- )
588
+ # Neither generative nor pooling model - try to convert if possible
589
+ if is_pooling_task:
590
+ runner = "pooling"
591
+ convert = _task_to_convert(self.task)
592
+ msg_hint = (
593
+ "Please replace this option with `--runner pooling "
594
+ f"--convert {convert}` to continue using this model "
595
+ "as a pooling model."
596
+ )
597
+ else:
598
+ debug_info = {
599
+ "architectures": architectures,
600
+ "is_generative_model": is_generative_model,
601
+ "is_pooling_model": is_pooling_model,
602
+ }
603
+ raise AssertionError(
604
+ "The model should be a generative or "
605
+ "pooling model when task is set to "
606
+ f"{self.task!r}. Found: {debug_info}"
607
+ )
608
609
self.runner = runner
610
self.convert = convert