distributed traing t_total

examples/run_squad.py

@@ -234,10 +234,11 @@ def main():
             train_sampler = RandomSampler(train_data)
         else:
             train_sampler = DistributedSampler(train_data)
+
         train_dataloader = DataLoader(train_data, sampler=train_sampler, batch_size=args.train_batch_size)
         num_train_optimization_steps = len(train_dataloader) // args.gradient_accumulation_steps * args.num_train_epochs
-        if args.local_rank != -1:
-            num_train_optimization_steps = num_train_optimization_steps // torch.distributed.get_world_size()
+        # if args.local_rank != -1:
+        #     num_train_optimization_steps = num_train_optimization_steps // torch.distributed.get_world_size()
 
         # Prepare optimizer
         param_optimizer = list(model.named_parameters())