xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 08:04:58 +08:00
Code Issues Packages Projects Releases Wiki Activity

[MISC] Fix misleading batch_size_capture_list when cuda_graph_sizes < 4 (#25829)

Browse Source 
Signed-off-by: billishyahao <bill.he@amd.com>
Co-authored-by: Luka Govedic <ProExpertProg@users.noreply.github.com>
This commit is contained in:
billishyahao 2025-10-01 20:39:45 +08:00 committed by GitHub
parent a332b84578
commit 2518230d3e
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 6 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
vllm/config/vllm.py
View File

@ -580,9 +580,12 @@ class VllmConfig:
not self.model_config.enforce_eager:
cuda_graph_sizes = self.scheduler_config.cuda_graph_sizes
if len(cuda_graph_sizes) == 1:
batch_size_capture_list = [1, 2, 4] + [
i for i in range(8, cuda_graph_sizes[0] + 1, 8)
]
max_graph_size = cuda_graph_sizes[0]
assert max_graph_size >= 1, "Maximum cudagraph size should be" \
" greater than or equal to 1."
batch_size_capture_list = [
i for i in [1, 2, 4] if i <= max_graph_size
] + list(range(8, max_graph_size + 1, 8))
elif len(cuda_graph_sizes) > 1:
batch_size_capture_list = sorted(cuda_graph_sizes)
else: