Dao-AILab
diff --git a/‎flash_attn/cute/interface.py‎
Lines changed: 0 additions & 1 deletion b/‎flash_attn/cute/interface.py‎
Lines changed: 0 additions & 1 deletion
@@ -1367,7 +1367,6 @@ def _flash_attn_bwd(
         disable_2cta = (
             requested_disable_2cta
             or block_sparse_tensors is not None
-            or softcap != 0.0
         )
         cluster_size = 2 if head_dim >= 128 and not disable_2cta else 1
         use_2cta_instrs = cluster_size==2
Original file line number	Diff line number	Diff line change
`@@ -1367,7 +1367,6 @@ def _flash_attn_bwd(`
`1367`	`1367`	`disable_2cta = (`
`1368`	`1368`	`requested_disable_2cta`
`1369`	`1369`	`or block_sparse_tensors is not None`
`1370`		`- or softcap != 0.0`
`1371`	`1370`	`)`
`1372`	`1371`	`cluster_size = 2 if head_dim >= 128 and not disable_2cta else 1`
`1373`	`1372`	`use_2cta_instrs = cluster_size==2`