Closed yaroslavvb closed 6 years ago
The following commit claims to overlap training and computation @bearpelican
https://github.com/pytorch/pytorch/commit/5b7951057d975a7e7ae0ea5e5a651d86b370993d
This is less than a month old, so I'm guessing it's not used in any of our experiments?
Wow great find! Ok I'll try to build from source again and see if we can use the latest and greatest
both c10d and Apex are in active development, maybe revisit this in a few weeks
The following commit claims to overlap training and computation @bearpelican
https://github.com/pytorch/pytorch/commit/5b7951057d975a7e7ae0ea5e5a651d86b370993d
This is less than a month old, so I'm guessing it's not used in any of our experiments?