Revert "Optimize fp8 linalg_ext.attention by rework Q@K scaling" …
#26
| Job | Run time |
|---|---|
| 5s | |
| 1s | |
| 6s |
fp8 linalg_ext.attention by rework Q@K scaling" …
#26
| Job | Run time |
|---|---|
| 5s | |
| 1s | |
| 6s |