Feat (nn/sdpa): quantization of scaled dot-product attention (#1090) #1130
Job | Run time |
---|---|
2m 11s | |
2m 2s | |
2m 17s | |
1m 30s | |
2m 12s | |
1m 37s | |
2m 27s | |
1m 37s | |
2m 12s | |
1m 32s | |
2m 30s | |
1m 42s | |
2m 20s | |
1m 41s | |
2m 18s | |
1m 32s | |
2m 19s | |
1m 40s | |
2m 13s | |
1m 33s | |
2m 12s | |
1m 31s | |
2m 31s | |
1m 37s | |
2m 26s | |
1m 38s | |
2m 45s | |
1m 43s | |
55m 48s |