@tduncan / Nvidia-Optimization-Stack-VLLM Mirror · read-only / commits / ebb0a71ad0
Fork 0

A high-throughput and memory-efficient inference and serving engine for LLMs

[Bugfix] Reject out-of-range temperature values in SamplingParams (#44965)

Signed-off-by: Peter Pan <Peter.Pan@daocloud.io>

Peter Pan committed Jun 15, 2026 at 14:12 UTC ebb0a71ad0b2e2a09ed76b14338465f6b121ebfd
1 file changed +6
vllm/sampling_params.py
+6
@@ -526,6 +526,12 @@ class SamplingParams(
526 parameter="temperature",
527 value=self.temperature,
528 )
529 + if self.temperature > 2.0:
530 + raise VLLMValidationError(
531 + f"temperature must be in [0, 2], got {self.temperature}.",
532 + parameter="temperature",
533 + value=self.temperature,
534 + )
535 if not 0.0 < self.top_p <= 1.0:
536 raise VLLMValidationError(
537 f"top_p must be in (0, 1], got {self.top_p}.",