Enable autograd graph to propagate after multi-device syncing (for loss functions in ddp
)
#9717
Job | Run time |
---|---|
2m 19s | |
2m 25s | |
2m 26s | |
4m 0s | |
4m 35s | |
4m 23s | |
4m 35s | |
4m 18s | |
4m 54s | |
33m 55s |
ddp
)
#9717
Job | Run time |
---|---|
2m 19s | |
2m 25s | |
2m 26s | |
4m 0s | |
4m 35s | |
4m 23s | |
4m 35s | |
4m 18s | |
4m 54s | |
33m 55s |