Load state in else (#25318)

* Load else

* New approach

* Propagate

examples/pytorch/image-classification/run_image_classification_no_trainer.py

@@ -454,14 +454,18 @@ def main():
     # Potentially load in the weights and states from a previous save
     if args.resume_from_checkpoint:
         if args.resume_from_checkpoint is not None or args.resume_from_checkpoint != "":
-            accelerator.print(f"Resumed from checkpoint: {args.resume_from_checkpoint}")
-            accelerator.load_state(args.resume_from_checkpoint)
+            checkpoint_path = args.resume_from_checkpoint
             path = os.path.basename(args.resume_from_checkpoint)
         else:
             # Get the most recent checkpoint
             dirs = [f.name for f in os.scandir(os.getcwd()) if f.is_dir()]
             dirs.sort(key=os.path.getctime)
             path = dirs[-1]  # Sorts folders by date modified, most recent checkpoint is the last
+            checkpoint_path = path
+            path = os.path.basename(checkpoint_path)
+
+        accelerator.print(f"Resumed from checkpoint: {checkpoint_path}")
+        accelerator.load_state(path)
         # Extract `epoch_{i}` or `step_{i}`
         training_difference = os.path.splitext(path)[0]
 

examples/pytorch/image-pretraining/run_mim_no_trainer.py

@@ -664,14 +664,18 @@ def main():
     # Potentially load in the weights and states from a previous save
     if args.resume_from_checkpoint:
         if args.resume_from_checkpoint is not None or args.resume_from_checkpoint != "":
-            accelerator.print(f"Resumed from checkpoint: {args.resume_from_checkpoint}")
-            accelerator.load_state(args.resume_from_checkpoint)
+            checkpoint_path = args.resume_from_checkpoint
             path = os.path.basename(args.resume_from_checkpoint)
         else:
             # Get the most recent checkpoint
             dirs = [f.name for f in os.scandir(os.getcwd()) if f.is_dir()]
             dirs.sort(key=os.path.getctime)
             path = dirs[-1]  # Sorts folders by date modified, most recent checkpoint is the last
+            checkpoint_path = path
+            path = os.path.basename(checkpoint_path)
+
+        accelerator.print(f"Resumed from checkpoint: {checkpoint_path}")
+        accelerator.load_state(path)
         # Extract `epoch_{i}` or `step_{i}`
         training_difference = os.path.splitext(path)[0]
 

examples/pytorch/language-modeling/run_clm_no_trainer.py

@@ -575,14 +575,18 @@ def main():
     # Potentially load in the weights and states from a previous save
     if args.resume_from_checkpoint:
         if args.resume_from_checkpoint is not None or args.resume_from_checkpoint != "":
-            accelerator.print(f"Resumed from checkpoint: {args.resume_from_checkpoint}")
-            accelerator.load_state(args.resume_from_checkpoint)
+            checkpoint_path = args.resume_from_checkpoint
             path = os.path.basename(args.resume_from_checkpoint)
         else:
             # Get the most recent checkpoint
             dirs = [f.name for f in os.scandir(os.getcwd()) if f.is_dir()]
             dirs.sort(key=os.path.getctime)
             path = dirs[-1]  # Sorts folders by date modified, most recent checkpoint is the last
+            checkpoint_path = path
+            path = os.path.basename(checkpoint_path)
+
+        accelerator.print(f"Resumed from checkpoint: {checkpoint_path}")
+        accelerator.load_state(path)
         # Extract `epoch_{i}` or `step_{i}`
         training_difference = os.path.splitext(path)[0]
 

examples/pytorch/language-modeling/run_mlm_no_trainer.py

@@ -613,14 +613,18 @@ def main():
     # Potentially load in the weights and states from a previous save
     if args.resume_from_checkpoint:
         if args.resume_from_checkpoint is not None or args.resume_from_checkpoint != "":
-            accelerator.print(f"Resumed from checkpoint: {args.resume_from_checkpoint}")
-            accelerator.load_state(args.resume_from_checkpoint)
+            checkpoint_path = args.resume_from_checkpoint
             path = os.path.basename(args.resume_from_checkpoint)
         else:
             # Get the most recent checkpoint
             dirs = [f.name for f in os.scandir(os.getcwd()) if f.is_dir()]
             dirs.sort(key=os.path.getctime)
             path = dirs[-1]  # Sorts folders by date modified, most recent checkpoint is the last
+            checkpoint_path = path
+            path = os.path.basename(checkpoint_path)
+
+        accelerator.print(f"Resumed from checkpoint: {checkpoint_path}")
+        accelerator.load_state(path)
         # Extract `epoch_{i}` or `step_{i}`
         training_difference = os.path.splitext(path)[0]
 

examples/pytorch/multiple-choice/run_swag_no_trainer.py

@@ -560,14 +560,18 @@ def main():
     # Potentially load in the weights and states from a previous save
     if args.resume_from_checkpoint:
         if args.resume_from_checkpoint is not None or args.resume_from_checkpoint != "":
-            accelerator.print(f"Resumed from checkpoint: {args.resume_from_checkpoint}")
-            accelerator.load_state(args.resume_from_checkpoint)
+            checkpoint_path = args.resume_from_checkpoint
             path = os.path.basename(args.resume_from_checkpoint)
         else:
             # Get the most recent checkpoint
             dirs = [f.name for f in os.scandir(os.getcwd()) if f.is_dir()]
             dirs.sort(key=os.path.getctime)
             path = dirs[-1]  # Sorts folders by date modified, most recent checkpoint is the last
+            checkpoint_path = path
+            path = os.path.basename(checkpoint_path)
+
+        accelerator.print(f"Resumed from checkpoint: {checkpoint_path}")
+        accelerator.load_state(path)
         # Extract `epoch_{i}` or `step_{i}`
         training_difference = os.path.splitext(path)[0]
 

examples/pytorch/question-answering/run_qa_beam_search_no_trainer.py

@@ -797,14 +797,18 @@ def main():
     # Potentially load in the weights and states from a previous save
     if args.resume_from_checkpoint:
         if args.resume_from_checkpoint is not None or args.resume_from_checkpoint != "":
-            accelerator.print(f"Resumed from checkpoint: {args.resume_from_checkpoint}")
-            accelerator.load_state(args.resume_from_checkpoint)
+            checkpoint_path = args.resume_from_checkpoint
             path = os.path.basename(args.resume_from_checkpoint)
         else:
             # Get the most recent checkpoint
             dirs = [f.name for f in os.scandir(os.getcwd()) if f.is_dir()]
             dirs.sort(key=os.path.getctime)
             path = dirs[-1]  # Sorts folders by date modified, most recent checkpoint is the last
+            checkpoint_path = path
+            path = os.path.basename(checkpoint_path)
+
+        accelerator.print(f"Resumed from checkpoint: {checkpoint_path}")
+        accelerator.load_state(path)
         # Extract `epoch_{i}` or `step_{i}`
         training_difference = os.path.splitext(path)[0]
 

examples/pytorch/question-answering/run_qa_no_trainer.py

@@ -828,14 +828,18 @@ def main():
     # Potentially load in the weights and states from a previous save
     if args.resume_from_checkpoint:
         if args.resume_from_checkpoint is not None or args.resume_from_checkpoint != "":
-            accelerator.print(f"Resumed from checkpoint: {args.resume_from_checkpoint}")
-            accelerator.load_state(args.resume_from_checkpoint)
+            checkpoint_path = args.resume_from_checkpoint
             path = os.path.basename(args.resume_from_checkpoint)
         else:
             # Get the most recent checkpoint
             dirs = [f.name for f in os.scandir(os.getcwd()) if f.is_dir()]
             dirs.sort(key=os.path.getctime)
             path = dirs[-1]  # Sorts folders by date modified, most recent checkpoint is the last
+            checkpoint_path = path
+            path = os.path.basename(checkpoint_path)
+
+        accelerator.print(f"Resumed from checkpoint: {checkpoint_path}")
+        accelerator.load_state(path)
         # Extract `epoch_{i}` or `step_{i}`
         training_difference = os.path.splitext(path)[0]
 

examples/pytorch/semantic-segmentation/run_semantic_segmentation_no_trainer.py

@@ -558,14 +558,18 @@ def main():
     # Potentially load in the weights and states from a previous save
     if args.resume_from_checkpoint:
         if args.resume_from_checkpoint is not None or args.resume_from_checkpoint != "":
-            accelerator.print(f"Resumed from checkpoint: {args.resume_from_checkpoint}")
-            accelerator.load_state(args.resume_from_checkpoint)
+            checkpoint_path = args.resume_from_checkpoint
             path = os.path.basename(args.resume_from_checkpoint)
         else:
             # Get the most recent checkpoint
             dirs = [f.name for f in os.scandir(os.getcwd()) if f.is_dir()]
             dirs.sort(key=os.path.getctime)
             path = dirs[-1]  # Sorts folders by date modified, most recent checkpoint is the last
+            checkpoint_path = path
+            path = os.path.basename(checkpoint_path)
+
+        accelerator.print(f"Resumed from checkpoint: {checkpoint_path}")
+        accelerator.load_state(path)
         # Extract `epoch_{i}` or `step_{i}`
         training_difference = os.path.splitext(path)[0]
 

examples/pytorch/summarization/run_summarization_no_trainer.py

@@ -629,14 +629,18 @@ def main():
     # Potentially load in the weights and states from a previous save
     if args.resume_from_checkpoint:
         if args.resume_from_checkpoint is not None or args.resume_from_checkpoint != "":
-            accelerator.print(f"Resumed from checkpoint: {args.resume_from_checkpoint}")
-            accelerator.load_state(args.resume_from_checkpoint)
+            checkpoint_path = args.resume_from_checkpoint
             path = os.path.basename(args.resume_from_checkpoint)
         else:
             # Get the most recent checkpoint
             dirs = [f.name for f in os.scandir(os.getcwd()) if f.is_dir()]
             dirs.sort(key=os.path.getctime)
             path = dirs[-1]  # Sorts folders by date modified, most recent checkpoint is the last
+            checkpoint_path = path
+            path = os.path.basename(checkpoint_path)
+
+        accelerator.print(f"Resumed from checkpoint: {checkpoint_path}")
+        accelerator.load_state(path)
         # Extract `epoch_{i}` or `step_{i}`
         training_difference = os.path.splitext(path)[0]
 

examples/pytorch/text-classification/run_glue_no_trainer.py

@@ -507,14 +507,18 @@ def main():
     # Potentially load in the weights and states from a previous save
     if args.resume_from_checkpoint:
         if args.resume_from_checkpoint is not None or args.resume_from_checkpoint != "":
-            accelerator.print(f"Resumed from checkpoint: {args.resume_from_checkpoint}")
-            accelerator.load_state(args.resume_from_checkpoint)
+            checkpoint_path = args.resume_from_checkpoint
             path = os.path.basename(args.resume_from_checkpoint)
         else:
             # Get the most recent checkpoint
             dirs = [f.name for f in os.scandir(os.getcwd()) if f.is_dir()]
             dirs.sort(key=os.path.getctime)
             path = dirs[-1]  # Sorts folders by date modified, most recent checkpoint is the last
+            checkpoint_path = path
+            path = os.path.basename(checkpoint_path)
+
+        accelerator.print(f"Resumed from checkpoint: {checkpoint_path}")
+        accelerator.load_state(path)
         # Extract `epoch_{i}` or `step_{i}`
         training_difference = os.path.splitext(path)[0]
 

examples/pytorch/token-classification/run_ner_no_trainer.py

@@ -667,14 +667,18 @@ def main():
     # Potentially load in the weights and states from a previous save
     if args.resume_from_checkpoint:
         if args.resume_from_checkpoint is not None or args.resume_from_checkpoint != "":
-            accelerator.print(f"Resumed from checkpoint: {args.resume_from_checkpoint}")
-            accelerator.load_state(args.resume_from_checkpoint)
+            checkpoint_path = args.resume_from_checkpoint
             path = os.path.basename(args.resume_from_checkpoint)
         else:
             # Get the most recent checkpoint
             dirs = [f.name for f in os.scandir(os.getcwd()) if f.is_dir()]
             dirs.sort(key=os.path.getctime)
             path = dirs[-1]  # Sorts folders by date modified, most recent checkpoint is the last
+            checkpoint_path = path
+            path = os.path.basename(checkpoint_path)
+
+        accelerator.print(f"Resumed from checkpoint: {checkpoint_path}")
+        accelerator.load_state(path)
         # Extract `epoch_{i}` or `step_{i}`
         training_difference = os.path.splitext(path)[0]
 

examples/pytorch/translation/run_translation_no_trainer.py

@@ -610,14 +610,18 @@ def main():
     # Potentially load in the weights and states from a previous save
     if args.resume_from_checkpoint:
         if args.resume_from_checkpoint is not None or args.resume_from_checkpoint != "":
-            accelerator.print(f"Resumed from checkpoint: {args.resume_from_checkpoint}")
-            accelerator.load_state(args.resume_from_checkpoint)
+            checkpoint_path = args.resume_from_checkpoint
             path = os.path.basename(args.resume_from_checkpoint)
         else:
             # Get the most recent checkpoint
             dirs = [f.name for f in os.scandir(os.getcwd()) if f.is_dir()]
             dirs.sort(key=os.path.getctime)
             path = dirs[-1]  # Sorts folders by date modified, most recent checkpoint is the last
+            checkpoint_path = path
+            path = os.path.basename(checkpoint_path)
+
+        accelerator.print(f"Resumed from checkpoint: {checkpoint_path}")
+        accelerator.load_state(path)
         # Extract `epoch_{i}` or `step_{i}`
         training_difference = os.path.splitext(path)[0]