From 7a94efa3dc6fb4a4a4256dc0efa7ac12ef506201 Mon Sep 17 00:00:00 2001 From: Charlene Yang <8636796+cyanguwa@users.noreply.github.com> Date: Thu, 2 Jul 2026 15:46:17 -0700 Subject: [PATCH 1/2] disable 9.23.0/.1 for mxfp8 attention Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com> --- .../pytorch/attention/dot_product_attention/utils.py | 6 ++++++ 1 file changed, 6 insertions(+) diff --git a/transformer_engine/pytorch/attention/dot_product_attention/utils.py b/transformer_engine/pytorch/attention/dot_product_attention/utils.py index 9913b78dfc..9b739bbf01 100644 --- a/transformer_engine/pytorch/attention/dot_product_attention/utils.py +++ b/transformer_engine/pytorch/attention/dot_product_attention/utils.py @@ -629,6 +629,12 @@ def _disable_all_flash_attention() -> None: if cudnn_version < (9, 21, 0): logger.debug("Disabling FusedAttention for MXFP8 with cuDNN < 9.21.0") use_fused_attention = False + elif cudnn_version in ((9, 23, 0), (9, 23, 1)): + # 9.23.0/9.23.1: known bugs with MXFP8 SDPA + logger.debug( + "Disabling FusedAttention for MXFP8 with cuDNN 9.23.0/9.23.1" + ) + use_fused_attention = False elif qkv_format == "thd": logger.debug("Disabling FusedAttention for MXFP8 with qkv_format = thd") use_fused_attention = False From ad4cbefe0881e177b60aa49af108ce139a01c483 Mon Sep 17 00:00:00 2001 From: "pre-commit-ci[bot]" <66853113+pre-commit-ci[bot]@users.noreply.github.com> Date: Thu, 2 Jul 2026 22:50:35 +0000 Subject: [PATCH 2/2] [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci --- .../pytorch/attention/dot_product_attention/utils.py | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-) diff --git a/transformer_engine/pytorch/attention/dot_product_attention/utils.py b/transformer_engine/pytorch/attention/dot_product_attention/utils.py index 9b739bbf01..7be94a6fa1 100644 --- a/transformer_engine/pytorch/attention/dot_product_attention/utils.py +++ b/transformer_engine/pytorch/attention/dot_product_attention/utils.py @@ -631,9 +631,7 @@ def _disable_all_flash_attention() -> None: use_fused_attention = False elif cudnn_version in ((9, 23, 0), (9, 23, 1)): # 9.23.0/9.23.1: known bugs with MXFP8 SDPA - logger.debug( - "Disabling FusedAttention for MXFP8 with cuDNN 9.23.0/9.23.1" - ) + logger.debug("Disabling FusedAttention for MXFP8 with cuDNN 9.23.0/9.23.1") use_fused_attention = False elif qkv_format == "thd": logger.debug("Disabling FusedAttention for MXFP8 with qkv_format = thd")