Update RopeFusion for qwen, gpt-j models to support new pattern SDPA to PA conversion #16278
Job | Run time |
---|---|
19s | |
44m 6s | |
2m 40s | |
10m 53s | |
13m 25s | |
42m 55s | |
5m 28s | |
4m 26s | |
5m 23s | |
8m 57s | |
6m 31s | |
3m 35s | |
7m 14s | |
23m 10s | |
38m 7s | |
1s | |
3h 37m 10s |
Job | Run time |
---|---|
19s | |
44m 6s | |
2m 40s | |
10m 53s | |
13m 25s | |
42m 55s | |
5m 28s | |
4m 26s | |
5m 23s | |
8m 57s | |
6m 31s | |
3m 35s | |
7m 14s | |
23m 10s | |
38m 7s | |
1s | |
3h 37m 10s |