[Backport] Allow setting FlashAttention's causal mask #6625
Job | Run time |
---|---|
55m 20s | |
0s | |
11m 53s | |
16m 53s | |
25m 38s | |
1h 11m 45s | |
9m 42s | |
46m 59s | |
14m 33s | |
50m 30s | |
16m 33s | |
1h 59m 14s | |
19m 16s | |
9m 37s | |
11m 39s | |
25m 1s | |
8h 24m 33s |
Job | Run time |
---|---|
55m 20s | |
0s | |
11m 53s | |
16m 53s | |
25m 38s | |
1h 11m 45s | |
9m 42s | |
46m 59s | |
14m 33s | |
50m 30s | |
16m 33s | |
1h 59m 14s | |
19m 16s | |
9m 37s | |
11m 39s | |
25m 1s | |
8h 24m 33s |