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 16, 05:45 UTC
[bugfix] fixed parameter “n” when set parameter “bestof” > 1 (#10854)
Signed-off-by: jianzheng <57654625+o2363286@users.noreply.github.com>
jianzheng committed
Dec 4, 2024 at 16:48 UTC
8db957ee3a8234574430d9e570e520501d8539e9
1 file changed
+3
-2
vllm/sampling_params.py
+3
-2
@@ -293,8 +293,9 @@ class SamplingParams(
293
raise ValueError(
294
f"best_of must be greater than or equal to n, "
295
f"got n={self.n} and best_of={self.best_of}.")
296
- self._real_n = self.n
297
- self.n = self.best_of
296
+ if not self._real_n:
297
+ self._real_n = self.n
298
+ self.n = self.best_of
299
300
if 0 < self.temperature < _MAX_TEMP:
301
logger.warning(