I've kind of been toying around with the idea of parallelizing Lingmess computation using torch DistributedDataParallel, but don't see any clear way to shard all of the non-tensor components in the forward pass. Any ideas? Is this something you've looked at?
By the way, great package — better by far than any other coref resolution research out there!
I've kind of been toying around with the idea of parallelizing Lingmess computation using torch
DistributedDataParallel, but don't see any clear way to shard all of the non-tensor components in theforwardpass. Any ideas? Is this something you've looked at?By the way, great package — better by far than any other coref resolution research out there!