Division to num_train_optimizer of global_step in lr_this_step is removed.

examples/lm_finetuning/finetune_on_pregenerated.py

@@ -315,8 +315,7 @@ def main():
                     if args.fp16:
                         # modify learning rate with special warm up BERT uses
                         # if args.fp16 is False, BertAdam is used that handles this automatically
-                        lr_this_step = args.learning_rate * warmup_linear.get_lr(global_step,
-                                                                                 args.warmup_proportion)
+                        lr_this_step = args.learning_rate * warmup_linear.get_lr(global_step, args.warmup_proportion)
                         for param_group in optimizer.param_groups:
                             param_group['lr'] = lr_this_step
                     optimizer.step()

examples/lm_finetuning/simple_lm_finetuning.py

@@ -604,8 +604,7 @@ def main():
                     if args.fp16:
                         # modify learning rate with special warm up BERT uses
                         # if args.fp16 is False, BertAdam is used that handles this automatically
-                        lr_this_step = args.learning_rate * warmup_linear.get_lr(global_step,
-                                                                                 args.warmup_proportion)
+                        lr_this_step = args.learning_rate * warmup_linear.get_lr(global_step, args.warmup_proportion)
                         for param_group in optimizer.param_groups:
                             param_group['lr'] = lr_this_step
                     optimizer.step()

examples/run_classifier.py

@@ -855,8 +855,7 @@ def main():
                     if args.fp16:
                         # modify learning rate with special warm up BERT uses
                         # if args.fp16 is False, BertAdam is used that handles this automatically
-                        lr_this_step = args.learning_rate * warmup_linear.get_lr(global_step,
-                                                                                 args.warmup_proportion)
+                        lr_this_step = args.learning_rate * warmup_linear.get_lr(global_step, args.warmup_proportion)
                         for param_group in optimizer.param_groups:
                             param_group['lr'] = lr_this_step
                     optimizer.step()

examples/run_squad.py

@@ -1016,8 +1016,7 @@ def main():
                     if args.fp16:
                         # modify learning rate with special warm up BERT uses
                         # if args.fp16 is False, BertAdam is used and handles this automatically
-                        lr_this_step = args.learning_rate * warmup_linear.get_lr(global_step,
-                                                                                 args.warmup_proportion)
+                        lr_this_step = args.learning_rate * warmup_linear.get_lr(global_step, args.warmup_proportion)
                         for param_group in optimizer.param_groups:
                             param_group['lr'] = lr_this_step
                     optimizer.step()

examples/run_swag.py

@@ -467,8 +467,7 @@ def main():
                     if args.fp16:
                         # modify learning rate with special warm up BERT uses
                         # if args.fp16 is False, BertAdam is used that handles this automatically
-                        lr_this_step = args.learning_rate * warmup_linear.get_lr(global_step,
-                                                                                 args.warmup_proportion)
+                        lr_this_step = args.learning_rate * warmup_linear.get_lr(global_step, args.warmup_proportion)
                         for param_group in optimizer.param_groups:
                             param_group['lr'] = lr_this_step
                     optimizer.step()