xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-01-23 07:04:27 +08:00
Code Issues Packages Projects Releases Wiki Activity

[CI] Add comment about the single cudagraph capture size that is used (#26252)

Browse Source
This commit is contained in:
Thomas Parnell 2025-10-06 04:35:37 +02:00 committed by GitHub
parent f509a20846
commit d3c84297c3
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 4 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
tests/conftest.py
View File

@ -740,6 +740,10 @@ class VllmRunner:
)
if not kwargs.get("compilation_config", None):
# Note(@tdoublep): This is set to 4 because some tests (e.g., hybrid
# model tests) may set max_num_seqs=4. If min cudagraph_capture_size is
# set to larger than max_num_seqs, then it will lead to *no* graphs
# being captured which can trigger edge cases that we don't handle yet.
kwargs["compilation_config"] = {"cudagraph_capture_sizes": [4]}
with init_ctx: