transformer_engine/pytorch/attention.py · 76187a5e0c38165d6fa2b732ae8f4700562321c4 · OpenDAS / TransformerEngine · GitLab

Find file Blame History Permalink

fix a sync race error of softmax_lse in CP+THD+P2P (#1624) · 76187a5e
Xiaowei Ren authored Mar 31, 2025
```
fix a race error softmax_lse
Signed-off-by: Xiaowei Ren <xren@nvidia.com>
```
76187a5e

attention.py 326 KB