[Kernel][Attention] Separate Attention.kv_scale
into k_scale
and v_scale
#15203
Job | Run time |
---|---|
1m 13s | |
1m 20s | |
1m 11s | |
50s | |
4m 34s |
Attention.kv_scale
into k_scale
and v_scale
#15203
Job | Run time |
---|---|
1m 13s | |
1m 20s | |
1m 11s | |
50s | |
4m 34s |