Skip to content

fix(lora): use block_n=32 on sm_90 to avoid lora_expand NaN #9

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Jackie2049 wants to merge 1 commit into
base: main
Choose a base branch
from
+7 −1
Open

fix(lora): use block_n=32 on sm_90 to avoid lora_expand NaN #9

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 7 additions & 1 deletion vllm/lora/ops/triton_ops/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -260,9 +260,15 @@ def get_lora_op_configs(
"split_k": 1,
}
else:
# On sm_90 (Hopper), BLOCK_N=128 causes NaN in lora_expand due to
# Triton OOB tile reads when N is not aligned to 128 (issue #48590).
# Use BLOCK_N=32 on sm_90 to avoid this.
default_block_n = 128
if current_platform.is_cuda() and current_platform.has_device_capability(90):
default_block_n = 32
default = {
"block_m": 64,
"block_n": 64 if num_slices > 1 else 128,
"block_n": 64 if num_slices > 1 else default_block_n,
"block_k": 32,
"num_warps": 4,
"num_ctas": 1,
Expand Down
Loading