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, 19:00 UTC
[ROCm][CI] Skip ROCm batch invalid-input test pending torch fix (#41572)
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Andreas Karatzas committed
May 13, 2026 at 05:50 UTC
d628a3c5cb333d3d3dd90c68bfde79ff5b012a27
1 file changed
+6
vllm/sampling_params.py
+6
@@ -535,6 +535,12 @@ class SamplingParams(
535
"stop strings are only supported when detokenize is True. "
536
"Set detokenize=True to use stop."
537
)
538
+ assert isinstance(self.bad_words, list)
539
+ if any(not bad_word for bad_word in self.bad_words):
540
+ raise ValueError(
541
+ f"bad_words cannot contain an empty string. "
542
+ f"Got bad_words={self.bad_words}"
543
+ )
544
545
def _verify_greedy_sampling(self) -> None:
546
if self.n > 1: