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 15, 09:45 UTC
[Misc] Avoid cuda graph log when sizes still match (#18202)
Signed-off-by: NickLucche <nlucches@redhat.com>
Nicolò Lucchesi committed
May 15, 2025 at 18:59 UTC
e3f3aee6f4206df0c338709614056ce2860ef039
1 file changed
+6
-5
vllm/config.py
+6
-5
@@ -3950,11 +3950,12 @@ class CompilationConfig:
3950
self.cudagraph_capture_sizes = cudagraph_capture_sizes
3951
else:
3952
# de-duplicate the sizes provided by the config
3953
- self.cudagraph_capture_sizes = list(
3954
- set(self.cudagraph_capture_sizes))
3955
- logger.info(("cudagraph sizes specified by model runner"
3956
- " %s is overridden by config %s"),
3957
- cudagraph_capture_sizes, self.cudagraph_capture_sizes)
3953
+ dedup_sizes = list(set(self.cudagraph_capture_sizes))
3954
+ if len(dedup_sizes) < len(self.cudagraph_capture_sizes):
3955
+ logger.info(("cudagraph sizes specified by model runner"
3956
+ " %s is overridden by config %s"),
3957
+ cudagraph_capture_sizes, dedup_sizes)
3958
+ self.cudagraph_capture_sizes = dedup_sizes
3959
3960
computed_compile_sizes = []
3961
if self.compile_sizes is not None: