ggml: avoid rebuild of GGML graph for each token (#7456) #199
Job | Run time |
---|---|
3m 19s | |
8m 41s | |
14m 48s | |
10m 18s | |
2m 36s | |
1m 41s | |
2m 51s | |
2m 58s | |
2m 38s | |
3m 10s | |
3m 33s | |
6m 38s | |
2m 12s | |
5m 36s | |
2m 34s | |
12s | |
2m 50s | |
23s | |
2m 2s | |
11s | |
10m 15s | |
12s | |
10m 12s | |
1s | |
9m 15s | |
36m 42s | |
8m 36s | |
9m 31s | |
9m 24s | |
40m 22s | |
8m 42s | |
9m 37s | |
8m 59s | |
9m 22s | |
8m 34s | |
16m 18s | |
2m 50s | |
2m 22s | |
1m 33s | |
12m 13s | |
9m 57s | |
0s | |
5h 4m 8s |