From 335f57baf86094907a14de7ddc9f3e791ae3519b Mon Sep 17 00:00:00 2001 From: thomwolf Date: Tue, 18 Jun 2019 14:03:46 +0200 Subject: [PATCH] only on main process --- examples/run_squad.py | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-) diff --git a/examples/run_squad.py b/examples/run_squad.py index 775e93e4db..e6afeab1c2 100644 --- a/examples/run_squad.py +++ b/examples/run_squad.py @@ -917,7 +917,8 @@ def main(): model = torch.nn.DataParallel(model) if args.do_train: - writer = SummaryWriter() + if args.local_rank in [-1, 0]: + writer = SummaryWriter() # Prepare data loader train_examples = read_squad_examples( input_file=args.train_file, is_training=True, version_2_with_negative=args.version_2_with_negative) @@ -1016,8 +1017,9 @@ def main(): else: loss.backward() if (step + 1) % args.gradient_accumulation_steps == 0: - writer.add_scalar('lr', optimizer.get_lr()[0], global_step) - writer.add_scalar('loss', loss.item(), global_step) + if args.local_rank in [-1, 0]: + writer.add_scalar('lr', optimizer.get_lr()[0], global_step) + writer.add_scalar('loss', loss.item(), global_step) if args.fp16: # modify learning rate with special warm up BERT uses # if args.fp16 is False, BertAdam is used and handles this automatically