Make num_train_optimization_steps int

examples/run_classifier.py

@@ -777,7 +777,7 @@ def main():
             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
+        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()
 

examples/run_squad.py

@@ -946,7 +946,7 @@ def main():
         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
+        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()
 

examples/run_swag.py

@@ -393,7 +393,7 @@ def main():
             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
+        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()