CUDA: faster dequantize kernels for Q4_0 and Q4_1 (#4938) #108
| Job | Run time |
|---|---|
| 1m 21s | |
| 3m 39s | |
| 1m 11s | |
| 1m 12s | |
| 4m 28s | |
| 2m 44s | |
| 1m 35s | |
| 2m 25s | |
| 2m 33s | |
| 1m 43s | |
| 1m 50s | |
| 1m 56s | |
| 6m 55s | |
| 4m 30s | |
| 17m 45s | |
| 3m 14s | |
| 6m 56s | |
| 3m 30s | |
| 6m 10s | |
| 4m 6s | |
| 4m 10s | |
| 17m 46s | |
| 7m 51s | |
| 17m 44s | |
| 2m 19s | |
| 3m 20s | |
| 46s | |
| 2h 13m 39s |