5 Commits

Author	SHA1	Message	Date
Myle Ott	7633129ba8	Merge internal changes (#283) ... Summary: Pull Request resolved: https://github.com/pytorch/translate/pull/283 Pull Request resolved: https://github.com/pytorch/fairseq/pull/428 Differential Revision: D13564190 Pulled By: myleott fbshipit-source-id: 3b62282d7069c288f5bdd1dd2c120788cee4abb5	2019-01-04 20:03:19 -08:00
Sergey Edunov	1082ba352c	Switch to DistributedDataParallelC10d and bump version 0.5.0 -> 0.6.0 ... - no more FP16Trainer, we just have an FP16Optimizer wrapper - most of the distributed code is moved to a new wrapper class called DistributedFairseqModel, which behaves like DistributedDataParallel and a FairseqModel at the same time - Trainer now requires an extra dummy_batch argument at initialization, which we do fwd/bwd on when there's an uneven number of batches per worker. We hide the gradients from these dummy batches by multiplying the loss by 0 - Trainer.train_step now takes a list of samples, which will allow cleaner --update-freq	2018-09-25 17:36:43 -04:00
Sergey Edunov	fe2d1581a4	Fix docs	2018-09-17 22:34:17 -07:00
Myle Ott	4a47b88992	Update documentation	2018-09-03 20:03:37 -04:00
Myle Ott	6381cc977f	Add documentation	2018-09-03 19:15:23 -04:00