Fix learning_rate precision in SGD GPU kernel to use MT instead of T #25944
| Job | Run time |
|---|---|
| 1m 4s | |
| 6s | |
| 4m 47s | |
| 2m 10s | |
| 6s | |
| 8s | |
| 23m 3s | |
| 2m 25s | |
| 7s | |
| 6s | |
| 8s | |
| 6s | |
| 6s | |
| 52m 10s | |
| 1h 3m 20s | |
| 41m 8s | |
| 33m 4s | |
| 40m 35s | |
| 29m 36s | |
| 6s | |
| 7m 33s | |
| 1h 33m 31s | |
| 6s | |
| 1m 4s | |
| 3h 39m 9s | |
| 6s | |
| 4m 47s | |
| 2m 10s | |
| 8s | |
| 8s | |
| 6s | |
| 52m 10s | |
| 23m 3s | |
| 41m 8s | |
| 33m 4s | |
| 1h 3m 20s | |
| 40m 35s | |
| 6s | |
| 6s | |
| 7s | |
| 1h 33m 31s | |
| 7m 33s | |
| 29m 36s | |
| 6s | |
| 16h 47m 34s |