CUDA: faster q2_K, q3_K MMQ + int8 tensor cores #12973
| Job | Run time |
|---|---|
| 2m 29s | |
| 12m 16s | |
| 11m 9s | |
| 2m 4s | |
| 1m 51s | |
| 2m 40s | |
| 1m 44s | |
| 2m 56s | |
| 2m 13s | |
| 3m 49s | |
| 18m 3s | |
| 4m 49s | |
| 8m 34s | |
| 5m 56s | |
| 3m 26s | |
| 3m 19s | |
| 2m 7s | |
| 6m 26s | |
| 1m 36s | |
| 1m 30s | |
| 1m 51s | |
| 15m 9s | |
| 11m 37s | |
| 5m 8s | |
| 6m 17s | |
| 19m 49s | |
| 7m 43s | |
| 19m 19s | |
| 6m 52s | |
| 17m 6s | |
| 7m 0s | |
| 13m 17s | |
| 17m 51s | |
| 13m 13s | |
| 6m 21s | |
| 8m 56s | |
| 6m 49s | |
| 8m 3s | |
| 6m 11s | |
| 3m 23s | |
| 2m 45s | |
| 0s | |
| 5h 3m 37s |