[mxfp8 moe training] parallelize along col blocks in scale blocked format kernel for groups along K #3229
| Job | Run time |
|---|---|
| 7s | |
| 15m 16s | |
| 3m 10s | |
| 15m 0s | |
| 16m 2s | |
| 17m 3s | |
| 21m 11s | |
| 18m 9s | |
| 6m 13s | |
| 0s | |
| 20s | |
| 25s | |
| 19s | |
| 27s | |
| 27s | |
| 21s | |
| 18s | |
| 18s | |
| 0s | |
| 1h 55m 6s |
| Job | Run time |
|---|---|
| 7s | |
| 15m 16s | |
| 3m 10s | |
| 15m 0s | |
| 16m 2s | |
| 17m 3s | |
| 21m 11s | |
| 18m 9s | |
| 6m 13s | |
| 0s | |
| 20s | |
| 25s | |
| 19s | |
| 27s | |
| 27s | |
| 21s | |
| 18s | |
| 18s | |
| 0s | |
| 1h 55m 6s |