mirror of
https://github.com/vllm-project/vllm.git
synced 2025-10-20 14:53:52 +08:00
Change size of single CUDA graph for CI to 4 (#26089)
Signed-off-by: Thomas Parnell <tpa@zurich.ibm.com>
This commit is contained in:
@ -732,7 +732,7 @@ class VllmRunner:
|
||||
set_default_torch_num_threads(default_torch_num_threads))
|
||||
|
||||
if not kwargs.get("compilation_config", None):
|
||||
kwargs["compilation_config"] = {"cudagraph_capture_sizes": [8]}
|
||||
kwargs["compilation_config"] = {"cudagraph_capture_sizes": [4]}
|
||||
|
||||
with init_ctx:
|
||||
self.llm = LLM(
|
||||
|
Reference in New Issue
Block a user