Update all no_trainer with skip_first_batches (#23664)

examples/pytorch/image-classification/run_image_classification_no_trainer.py

@@ -451,22 +451,26 @@ def main():
         if "epoch" in training_difference:
             starting_epoch = int(training_difference.replace("epoch_", "")) + 1
             resume_step = None
+            completed_steps = starting_epoch * num_update_steps_per_epoch
         else:
             resume_step = int(training_difference.replace("step_", ""))
             starting_epoch = resume_step // len(train_dataloader)
             resume_step -= starting_epoch * len(train_dataloader)
+            completed_steps = resume_step
+
+    # update the progress_bar if load from checkpoint
+    progress_bar.update(completed_steps)
 
     for epoch in range(starting_epoch, args.num_train_epochs):
         model.train()
         if args.with_tracking:
             total_loss = 0
-        for step, batch in enumerate(train_dataloader):
-            # We need to skip steps until we reach the resumed step
-            if args.resume_from_checkpoint and epoch == starting_epoch:
-                if resume_step is not None and step < resume_step:
-                    completed_steps += 1
-                    continue
-
+        if args.resume_from_checkpoint and epoch == starting_epoch and resume_step is not None:
+            # We skip the first `n` batches in the dataloader when resuming from a checkpoint
+            active_dataloader = accelerator.skip_first_batches(train_dataloader, resume_step)
+        else:
+            active_dataloader = train_dataloader
+        for step, batch in enumerate(active_dataloader):
             with accelerator.accumulate(model):
                 outputs = model(**batch)
                 loss = outputs.loss

examples/pytorch/image-pretraining/run_mim_no_trainer.py

@@ -660,29 +660,27 @@ def main():
         if "epoch" in training_difference:
             starting_epoch = int(training_difference.replace("epoch_", "")) + 1
             resume_step = None
+            completed_steps = starting_epoch * num_update_steps_per_epoch
         else:
             # need to multiply `gradient_accumulation_steps` to reflect real steps
             resume_step = int(training_difference.replace("step_", "")) * args.gradient_accumulation_steps
             starting_epoch = resume_step // len(train_dataloader)
             resume_step -= starting_epoch * len(train_dataloader)
+            completed_steps = resume_step
 
     # update the progress_bar if load from checkpoint
-    progress_bar.update(starting_epoch * num_update_steps_per_epoch)
-    completed_steps = starting_epoch * num_update_steps_per_epoch
+    progress_bar.update(completed_steps)
 
     for epoch in range(starting_epoch, args.num_train_epochs):
         model.train()
         if args.with_tracking:
             total_loss = 0
-        for step, batch in enumerate(train_dataloader):
-            # We need to skip steps until we reach the resumed step
-            if args.resume_from_checkpoint and epoch == starting_epoch:
-                if resume_step is not None and step < resume_step:
-                    if step % args.gradient_accumulation_steps == 0:
-                        progress_bar.update(1)
-                        completed_steps += 1
-                    continue
-
+        if args.resume_from_checkpoint and epoch == starting_epoch and resume_step is not None:
+            # We skip the first `n` batches in the dataloader when resuming from a checkpoint
+            active_dataloader = accelerator.skip_first_batches(train_dataloader, resume_step)
+        else:
+            active_dataloader = train_dataloader
+        for step, batch in enumerate(active_dataloader):
             with accelerator.accumulate(model):
                 outputs = model(**batch)
                 loss = outputs.loss

examples/pytorch/language-modeling/run_clm_no_trainer.py

@@ -566,29 +566,27 @@ def main():
         if "epoch" in training_difference:
             starting_epoch = int(training_difference.replace("epoch_", "")) + 1
             resume_step = None
+            completed_steps = starting_epoch * num_update_steps_per_epoch
         else:
             # need to multiply `gradient_accumulation_steps` to reflect real steps
             resume_step = int(training_difference.replace("step_", "")) * args.gradient_accumulation_steps
             starting_epoch = resume_step // len(train_dataloader)
             resume_step -= starting_epoch * len(train_dataloader)
+            completed_steps = resume_step
 
     # update the progress_bar if load from checkpoint
-    progress_bar.update(starting_epoch * num_update_steps_per_epoch)
-    completed_steps = starting_epoch * num_update_steps_per_epoch
+    progress_bar.update(completed_steps)
 
     for epoch in range(starting_epoch, args.num_train_epochs):
         model.train()
         if args.with_tracking:
             total_loss = 0
-        for step, batch in enumerate(train_dataloader):
-            # We need to skip steps until we reach the resumed step
-            if args.resume_from_checkpoint and epoch == starting_epoch:
-                if resume_step is not None and step < resume_step:
-                    if step % args.gradient_accumulation_steps == 0:
-                        progress_bar.update(1)
-                        completed_steps += 1
-                    continue
-
+        if args.resume_from_checkpoint and epoch == starting_epoch and resume_step is not None:
+            # We skip the first `n` batches in the dataloader when resuming from a checkpoint
+            active_dataloader = accelerator.skip_first_batches(train_dataloader, resume_step)
+        else:
+            active_dataloader = train_dataloader
+        for step, batch in enumerate(active_dataloader):
             with accelerator.accumulate(model):
                 outputs = model(**batch)
                 loss = outputs.loss

examples/pytorch/language-modeling/run_mlm_no_trainer.py

@@ -610,29 +610,27 @@ def main():
         if "epoch" in training_difference:
             starting_epoch = int(training_difference.replace("epoch_", "")) + 1
             resume_step = None
+            completed_steps = starting_epoch * num_update_steps_per_epoch
         else:
             # need to multiply `gradient_accumulation_steps` to reflect real steps
             resume_step = int(training_difference.replace("step_", "")) * args.gradient_accumulation_steps
             starting_epoch = resume_step // len(train_dataloader)
             resume_step -= starting_epoch * len(train_dataloader)
+            completed_steps = resume_step
 
     # update the progress_bar if load from checkpoint
-    progress_bar.update(starting_epoch * num_update_steps_per_epoch)
-    completed_steps = starting_epoch * num_update_steps_per_epoch
+    progress_bar.update(completed_steps)
 
     for epoch in range(starting_epoch, args.num_train_epochs):
         model.train()
         if args.with_tracking:
             total_loss = 0
-        for step, batch in enumerate(train_dataloader):
-            # We need to skip steps until we reach the resumed step
-            if args.resume_from_checkpoint and epoch == starting_epoch:
-                if resume_step is not None and step < resume_step:
-                    if step % args.gradient_accumulation_steps == 0:
-                        progress_bar.update(1)
-                        completed_steps += 1
-                    continue
-
+        if args.resume_from_checkpoint and epoch == starting_epoch and resume_step is not None:
+            # We skip the first `n` batches in the dataloader when resuming from a checkpoint
+            active_dataloader = accelerator.skip_first_batches(train_dataloader, resume_step)
+        else:
+            active_dataloader = train_dataloader
+        for step, batch in enumerate(active_dataloader):
             with accelerator.accumulate(model):
                 outputs = model(**batch)
                 loss = outputs.loss

examples/pytorch/multiple-choice/run_swag_no_trainer.py

@@ -557,22 +557,26 @@ def main():
         if "epoch" in training_difference:
             starting_epoch = int(training_difference.replace("epoch_", "")) + 1
             resume_step = None
+            completed_steps = starting_epoch * num_update_steps_per_epoch
         else:
             resume_step = int(training_difference.replace("step_", ""))
             starting_epoch = resume_step // len(train_dataloader)
             resume_step -= starting_epoch * len(train_dataloader)
+            completed_steps = resume_step
+
+    # update the progress_bar if load from checkpoint
+    progress_bar.update(completed_steps)
 
     for epoch in range(starting_epoch, args.num_train_epochs):
         model.train()
         if args.with_tracking:
             total_loss = 0
-        for step, batch in enumerate(train_dataloader):
-            # We need to skip steps until we reach the resumed step
-            if args.resume_from_checkpoint and epoch == starting_epoch:
-                if resume_step is not None and step < resume_step:
-                    completed_steps += 1
-                    continue
-
+        if args.resume_from_checkpoint and epoch == starting_epoch and resume_step is not None:
+            # We skip the first `n` batches in the dataloader when resuming from a checkpoint
+            active_dataloader = accelerator.skip_first_batches(train_dataloader, resume_step)
+        else:
+            active_dataloader = train_dataloader
+        for step, batch in enumerate(active_dataloader):
             with accelerator.accumulate(model):
                 outputs = model(**batch)
                 loss = outputs.loss

examples/pytorch/question-answering/run_qa_beam_search_no_trainer.py

@@ -809,22 +809,26 @@ def main():
         if "epoch" in training_difference:
             starting_epoch = int(training_difference.replace("epoch_", "")) + 1
             resume_step = None
+            completed_steps = starting_epoch * num_update_steps_per_epoch
         else:
             resume_step = int(training_difference.replace("step_", ""))
             starting_epoch = resume_step // len(train_dataloader)
             resume_step -= starting_epoch * len(train_dataloader)
+            completed_steps = resume_step
+
+    # update the progress_bar if load from checkpoint
+    progress_bar.update(completed_steps)
 
     for epoch in range(starting_epoch, args.num_train_epochs):
         model.train()
         if args.with_tracking:
             total_loss = 0
-        for step, batch in enumerate(train_dataloader):
-            # We need to skip steps until we reach the resumed step
-            if args.resume_from_checkpoint and epoch == starting_epoch:
-                if resume_step is not None and step < resume_step:
-                    completed_steps += 1
-                    continue
-
+        if args.resume_from_checkpoint and epoch == starting_epoch and resume_step is not None:
+            # We skip the first `n` batches in the dataloader when resuming from a checkpoint
+            active_dataloader = accelerator.skip_first_batches(train_dataloader, resume_step)
+        else:
+            active_dataloader = train_dataloader
+        for step, batch in enumerate(active_dataloader):
             with accelerator.accumulate(model):
                 outputs = model(**batch)
                 loss = outputs.loss

examples/pytorch/question-answering/run_qa_no_trainer.py

@@ -825,22 +825,26 @@ def main():
         if "epoch" in training_difference:
             starting_epoch = int(training_difference.replace("epoch_", "")) + 1
             resume_step = None
+            completed_steps = starting_epoch * num_update_steps_per_epoch
         else:
             resume_step = int(training_difference.replace("step_", ""))
             starting_epoch = resume_step // len(train_dataloader)
             resume_step -= starting_epoch * len(train_dataloader)
+            completed_steps = resume_step
+
+    # update the progress_bar if load from checkpoint
+    progress_bar.update(completed_steps)
 
     for epoch in range(starting_epoch, args.num_train_epochs):
         model.train()
         if args.with_tracking:
             total_loss = 0
-        for step, batch in enumerate(train_dataloader):
-            # We need to skip steps until we reach the resumed step
-            if args.resume_from_checkpoint and epoch == starting_epoch:
-                if resume_step is not None and step < resume_step:
-                    completed_steps += 1
-                    continue
-
+        if args.resume_from_checkpoint and epoch == starting_epoch and resume_step is not None:
+            # We skip the first `n` batches in the dataloader when resuming from a checkpoint
+            active_dataloader = accelerator.skip_first_batches(train_dataloader, resume_step)
+        else:
+            active_dataloader = train_dataloader
+        for step, batch in enumerate(active_dataloader):
             with accelerator.accumulate(model):
                 outputs = model(**batch)
                 loss = outputs.loss

examples/pytorch/semantic-segmentation/run_semantic_segmentation_no_trainer.py

@@ -554,22 +554,26 @@ def main():
         if "epoch" in training_difference:
             starting_epoch = int(training_difference.replace("epoch_", "")) + 1
             resume_step = None
+            completed_steps = starting_epoch * num_update_steps_per_epoch
         else:
             resume_step = int(training_difference.replace("step_", ""))
             starting_epoch = resume_step // len(train_dataloader)
             resume_step -= starting_epoch * len(train_dataloader)
+            completed_steps = resume_step
+
+    # update the progress_bar if load from checkpoint
+    progress_bar.update(completed_steps)
 
     for epoch in range(starting_epoch, args.num_train_epochs):
+        model.train()
         if args.with_tracking:
             total_loss = 0
-        model.train()
-        for step, batch in enumerate(train_dataloader):
-            # We need to skip steps until we reach the resumed step
-            if args.resume_from_checkpoint and epoch == starting_epoch:
-                if resume_step is not None and step < resume_step:
-                    completed_steps += 1
-                    continue
-
+        if args.resume_from_checkpoint and epoch == starting_epoch and resume_step is not None:
+            # We skip the first `n` batches in the dataloader when resuming from a checkpoint
+            active_dataloader = accelerator.skip_first_batches(train_dataloader, resume_step)
+        else:
+            active_dataloader = train_dataloader
+        for step, batch in enumerate(active_dataloader):
             with accelerator.accumulate(model):
                 outputs = model(**batch)
                 loss = outputs.loss

examples/pytorch/summarization/run_summarization_no_trainer.py

@@ -626,22 +626,26 @@ def main():
         if "epoch" in training_difference:
             starting_epoch = int(training_difference.replace("epoch_", "")) + 1
             resume_step = None
+            completed_steps = starting_epoch * num_update_steps_per_epoch
         else:
             resume_step = int(training_difference.replace("step_", ""))
             starting_epoch = resume_step // len(train_dataloader)
             resume_step -= starting_epoch * len(train_dataloader)
+            completed_steps = resume_step
+
+    # update the progress_bar if load from checkpoint
+    progress_bar.update(completed_steps)
 
     for epoch in range(starting_epoch, args.num_train_epochs):
         model.train()
         if args.with_tracking:
             total_loss = 0
-        for step, batch in enumerate(train_dataloader):
-            # We need to skip steps until we reach the resumed step
-            if args.resume_from_checkpoint and epoch == starting_epoch:
-                if resume_step is not None and step < resume_step:
-                    completed_steps += 1
-                    continue
-
+        if args.resume_from_checkpoint and epoch == starting_epoch and resume_step is not None:
+            # We skip the first `n` batches in the dataloader when resuming from a checkpoint
+            active_dataloader = accelerator.skip_first_batches(train_dataloader, resume_step)
+        else:
+            active_dataloader = train_dataloader
+        for step, batch in enumerate(active_dataloader):
             with accelerator.accumulate(model):
                 outputs = model(**batch)
                 loss = outputs.loss

examples/pytorch/text-classification/run_glue_no_trainer.py

@@ -510,12 +510,12 @@ def main():
         model.train()
         if args.with_tracking:
             total_loss = 0
-        for step, batch in enumerate(train_dataloader):
-            # We need to skip steps until we reach the resumed step
-            if args.resume_from_checkpoint and epoch == starting_epoch:
-                if resume_step is not None and step < resume_step:
-                    completed_steps += 1
-                    continue
+        if args.resume_from_checkpoint and epoch == starting_epoch and resume_step is not None:
+            # We skip the first `n` batches in the dataloader when resuming from a checkpoint
+            active_dataloader = accelerator.skip_first_batches(train_dataloader, resume_step)
+        else:
+            active_dataloader = train_dataloader
+        for step, batch in enumerate(active_dataloader):
             outputs = model(**batch)
             loss = outputs.loss
             # We keep track of the loss at each epoch

examples/pytorch/token-classification/run_ner_no_trainer.py

@@ -668,12 +668,12 @@ def main():
         model.train()
         if args.with_tracking:
             total_loss = 0
-        for step, batch in enumerate(train_dataloader):
-            # We need to skip steps until we reach the resumed step
-            if args.resume_from_checkpoint and epoch == starting_epoch:
-                if resume_step is not None and step < resume_step:
-                    completed_steps += 1
-                    continue
+        if args.resume_from_checkpoint and epoch == starting_epoch and resume_step is not None:
+            # We skip the first `n` batches in the dataloader when resuming from a checkpoint
+            active_dataloader = accelerator.skip_first_batches(train_dataloader, resume_step)
+        else:
+            active_dataloader = train_dataloader
+        for step, batch in enumerate(active_dataloader):
             outputs = model(**batch)
             loss = outputs.loss
             # We keep track of the loss at each epoch

examples/pytorch/translation/run_translation_no_trainer.py

@@ -607,28 +607,27 @@ def main():
         if "epoch" in training_difference:
             starting_epoch = int(training_difference.replace("epoch_", "")) + 1
             resume_step = None
+            completed_steps = starting_epoch * num_update_steps_per_epoch
         else:
             # need to multiply `gradient_accumulation_steps` to reflect real steps
             resume_step = int(training_difference.replace("step_", "")) * args.gradient_accumulation_steps
             starting_epoch = resume_step // len(train_dataloader)
             resume_step -= starting_epoch * len(train_dataloader)
+            completed_steps = resume_step
 
     # update the progress_bar if load from checkpoint
-    progress_bar.update(starting_epoch * num_update_steps_per_epoch)
-    completed_steps = starting_epoch * num_update_steps_per_epoch
+    progress_bar.update(completed_steps)
 
     for epoch in range(starting_epoch, args.num_train_epochs):
         model.train()
         if args.with_tracking:
             total_loss = 0
-        for step, batch in enumerate(train_dataloader):
-            # We need to skip steps until we reach the resumed step
-            if args.resume_from_checkpoint and epoch == starting_epoch:
-                if resume_step is not None and step < resume_step:
-                    if step % args.gradient_accumulation_steps == 0:
-                        progress_bar.update(1)
-                        completed_steps += 1
-                    continue
+        if args.resume_from_checkpoint and epoch == starting_epoch and resume_step is not None:
+            # We skip the first `n` batches in the dataloader when resuming from a checkpoint
+            active_dataloader = accelerator.skip_first_batches(train_dataloader, resume_step)
+        else:
+            active_dataloader = train_dataloader
+        for step, batch in enumerate(active_dataloader):
             outputs = model(**batch)
             loss = outputs.loss
             # We keep track of the loss at each epoch