[TRANSFORMATIONS] SDPAToPagedAttention transformation: support decompression case in the Qwen-7b-Chat pattern #16295
Job | Run time |
---|---|
24s | |
24s | |
42m 5s | |
42m 5s | |
2m 59s | |
2m 59s | |
9m 37s | |
9m 37s | |
14m 39s | |
14m 39s | |
41m 19s | |
41m 19s | |
5m 21s | |
5m 21s | |
4m 38s | |
4m 38s | |
6m 46s | |
6m 46s | |
5m 6s | |
5m 6s | |
8m 34s | |
8m 34s | |
26m 28s | |
26m 28s | |
21m 27s | |
21m 27s | |
7m 8s | |
7m 8s | |
45m 1s | |
42m 8s | |
1s | |
1s | |
8h 0m 13s |