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 14, 05:30 UTC
Change size of single CUDA graph for CI to 4 (#26089)
Signed-off-by: Thomas Parnell <tpa@zurich.ibm.com>
Thomas Parnell committed
Oct 2, 2025 at 16:14 UTC
be8921fbba97472e97fd91230e7d7d15c93a0d23
1 file changed
+1
-1
tests/conftest.py
+1
-1
@@ -732,7 +732,7 @@ class VllmRunner:
732
set_default_torch_num_threads(default_torch_num_threads))
733
734
if not kwargs.get("compilation_config", None):
735
- kwargs["compilation_config"] = {"cudagraph_capture_sizes": [8]}
735
+ kwargs["compilation_config"] = {"cudagraph_capture_sizes": [4]}
736
737
with init_ctx:
738
self.llm = LLM(