Fix no_trainer examples to properly calculate the number of samples (#17046)

* Update all examples to properly calculate progress bar

examples/pytorch/image-classification/run_image_classification_no_trainer.py

@@ -359,6 +359,10 @@ def main():
         model, optimizer, train_dataloader, eval_dataloader, lr_scheduler
     )
 
+    # We need to recalculate our total training steps as the size of the training dataloader may have changed.
+    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
+    args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
+
     # Figure out how many steps we should save the Accelerator states
     if hasattr(args.checkpointing_steps, "isdigit"):
         checkpointing_steps = args.checkpointing_steps

examples/pytorch/language-modeling/run_clm_no_trainer.py

@@ -472,6 +472,10 @@ def main():
         model, optimizer, train_dataloader, eval_dataloader, lr_scheduler
     )
 
+    # We need to recalculate our total training steps as the size of the training dataloader may have changed.
+    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
+    args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
+
     # Figure out how many steps we should save the Accelerator states
     if hasattr(args.checkpointing_steps, "isdigit"):
         checkpointing_steps = args.checkpointing_steps

examples/pytorch/language-modeling/run_mlm_no_trainer.py

@@ -518,6 +518,10 @@ def main():
         model, optimizer, train_dataloader, eval_dataloader, lr_scheduler
     )
 
+    # We need to recalculate our total training steps as the size of the training dataloader may have changed.
+    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
+    args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
+
     # Figure out how many steps we should save the Accelerator states
     if hasattr(args.checkpointing_steps, "isdigit"):
         checkpointing_steps = args.checkpointing_steps

examples/pytorch/multiple-choice/run_swag_no_trainer.py

@@ -472,6 +472,10 @@ def main():
         model, optimizer, train_dataloader, eval_dataloader, lr_scheduler
     )
 
+    # We need to recalculate our total training steps as the size of the training dataloader may have changed.
+    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
+    args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
+
     # Figure out how many steps we should save the Accelerator states
     if hasattr(args.checkpointing_steps, "isdigit"):
         checkpointing_steps = args.checkpointing_steps

examples/pytorch/question-answering/run_qa_beam_search_no_trainer.py

@@ -733,6 +733,10 @@ def main():
         model, optimizer, train_dataloader, eval_dataloader, lr_scheduler
     )
 
+    # We need to recalculate our total training steps as the size of the training dataloader may have changed.
+    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
+    args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
+
     # Figure out how many steps we should save the Accelerator states
     if hasattr(args.checkpointing_steps, "isdigit"):
         checkpointing_steps = args.checkpointing_steps

examples/pytorch/question-answering/run_qa_no_trainer.py

@@ -739,6 +739,10 @@ def main():
         model, optimizer, train_dataloader, eval_dataloader, lr_scheduler
     )
 
+    # We need to recalculate our total training steps as the size of the training dataloader may have changed.
+    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
+    args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
+
     # Figure out how many steps we should save the Accelerator states
     if hasattr(args.checkpointing_steps, "isdigit"):
         checkpointing_steps = args.checkpointing_steps

examples/pytorch/semantic-segmentation/run_semantic_segmentation_no_trainer.py

@@ -475,6 +475,10 @@ def main():
         model, optimizer, train_dataloader, eval_dataloader, lr_scheduler
     )
 
+    # We need to recalculate our total training steps as the size of the training dataloader may have changed.
+    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
+    args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
+
     # Instantiate metric
     metric = load_metric("mean_iou")
 

examples/pytorch/summarization/run_summarization_no_trainer.py

@@ -535,6 +535,10 @@ def main():
         model, optimizer, train_dataloader, eval_dataloader, lr_scheduler
     )
 
+    # We need to recalculate our total training steps as the size of the training dataloader may have changed.
+    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
+    args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
+
     # Figure out how many steps we should save the Accelerator states
     if hasattr(args.checkpointing_steps, "isdigit"):
         checkpointing_steps = args.checkpointing_steps

examples/pytorch/text-classification/run_glue_no_trainer.py

@@ -418,6 +418,10 @@ def main():
         model, optimizer, train_dataloader, eval_dataloader, lr_scheduler
     )
 
+    # We need to recalculate our total training steps as the size of the training dataloader may have changed.
+    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
+    args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
+
     # Figure out how many steps we should save the Accelerator states
     if hasattr(args.checkpointing_steps, "isdigit"):
         checkpointing_steps = args.checkpointing_steps

examples/pytorch/token-classification/run_ner_no_trainer.py

@@ -532,6 +532,10 @@ def main():
         model, optimizer, train_dataloader, eval_dataloader, lr_scheduler
     )
 
+    # We need to recalculate our total training steps as the size of the training dataloader may have changed.
+    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
+    args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
+
     # Figure out how many steps we should save the Accelerator states
     if hasattr(args.checkpointing_steps, "isdigit"):
         checkpointing_steps = args.checkpointing_steps

examples/pytorch/translation/run_translation_no_trainer.py

@@ -513,6 +513,10 @@ def main():
         model, optimizer, train_dataloader, eval_dataloader, lr_scheduler
     )
 
+    # We need to recalculate our total training steps as the size of the training dataloader may have changed.
+    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
+    args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
+
     # Figure out how many steps we should save the Accelerator states
     if hasattr(args.checkpointing_steps, "isdigit"):
         checkpointing_steps = args.checkpointing_steps