Skip to content

[PyTorch] Restore FlashAttention 2 head dim support on sm103 #3360

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
kalectory wants to merge 1 commit into
base: main
Choose a base branch
from
+2 −10
Open

[PyTorch] Restore FlashAttention 2 head dim support on sm103 #3360

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
12 changes: 2 additions & 10 deletions transformer_engine/pytorch/attention/dot_product_attention/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -913,19 +913,11 @@ def _disable_all_flash_attention() -> None:
if ( # pylint: disable=too-many-boolean-expressions
use_flash_attention_2
and FlashAttentionUtils.is_installed
and (
fa2_padded_head_dim > 256
or fa2_padded_head_dim % 8 != 0
or (
fa2_padded_head_dim > 192
and device_compute_capability not in ((8, 0), (9, 0), (10, 0), (12, 0))
)
)
and (fa2_padded_head_dim > 256 or fa2_padded_head_dim % 8 != 0)
):
logger.debug(
"Disabling FlashAttention 2 due to unsupported head_dim_qk and head_dim_v. "
"Supported after padding: padded head_dim %%8 = 0, padded head_dim <= 256 "
"(>192 requires sm80/90/100+). "
"Supported after padding: padded head_dim %%8 = 0, padded head_dim <= 256. "
"Found: head_dim_qk = %s, head_dim_v = %s, padded head_dim = %s, on sm%s.",
head_dim_qk,
head_dim_v,
Expand Down
Loading