diff --git a/Matrix-Game-3/wan/distributed/ulysses.py b/Matrix-Game-3/wan/distributed/ulysses.py index e9cd669..c7ee3b4 100644 --- a/Matrix-Game-3/wan/distributed/ulysses.py +++ b/Matrix-Game-3/wan/distributed/ulysses.py @@ -66,7 +66,7 @@ def distributed_attention( v, k_lens=seq_lens, window_size=window_size, - fa_version=fa_version, + version=fa_version, ) # scatter q/k/v sequence