Reduce the memory usage of logits from O(context_length) to O(1) (#4688) #3825
Job | Run time |
---|---|
5s | |
16m 14s | |
15m 11s | |
51m 55s | |
30m 16s | |
5m 23s | |
6m 2s | |
1m 29s | |
1m 18s | |
12m 50s | |
11m 54s | |
13m 20s | |
11m 50s | |
12m 37s | |
13m 12s | |
11m 41s | |
12m 21s | |
13m 0s | |
10m 12s | |
15m 27s | |
10m 3s | |
18m 15s | |
10m 59s | |
17m 48s | |
11m 29s | |
17m 1s | |
17m 16s | |
12m 14s | |
15m 18s | |
10m 26s | |
16m 19s | |
10m 27s | |
10m 44s | |
17m 38s | |
10m 23s | |
10m 36s | |
15m 34s | |
13m 25s | |
10m 29s | |
15m 14s | |
10m 19s | |
10m 22s | |
10m 21s | |
15m 43s | |
10m 29s | |
15m 49s | |
13m 58s | |
16m 45s | |
13m 4s | |
10m 43s | |
17m 55s | |
11m 19s | |
15m 29s | |
17m 23s | |
12m 49s | |
12h 20m 23s |