Deprecate model_path in Trainer.train (#9854)

examples/language-modeling/run_clm.py

@@ -362,12 +362,12 @@ def main():
     # Training
     if training_args.do_train:
         if last_checkpoint is not None:
-            model_path = last_checkpoint
+            checkpoint = last_checkpoint
         elif model_args.model_name_or_path is not None and os.path.isdir(model_args.model_name_or_path):
-            model_path = model_args.model_name_or_path
+            checkpoint = model_args.model_name_or_path
         else:
-            model_path = None
-        train_result = trainer.train(model_path=model_path)
+            checkpoint = None
+        train_result = trainer.train(resume_from_checkpoint=checkpoint)
         trainer.save_model()  # Saves the tokenizer too for easy upload
 
         output_train_file = os.path.join(training_args.output_dir, "train_results.txt")

examples/language-modeling/run_mlm.py

@@ -403,12 +403,12 @@ def main():
     # Training
     if training_args.do_train:
         if last_checkpoint is not None:
-            model_path = last_checkpoint
+            checkpoint = last_checkpoint
         elif model_args.model_name_or_path is not None and os.path.isdir(model_args.model_name_or_path):
-            model_path = model_args.model_name_or_path
+            checkpoint = model_args.model_name_or_path
         else:
-            model_path = None
-        train_result = trainer.train(model_path=model_path)
+            checkpoint = None
+        train_result = trainer.train(resume_from_checkpoint=checkpoint)
         trainer.save_model()  # Saves the tokenizer too for easy upload
 
         output_train_file = os.path.join(training_args.output_dir, "train_results.txt")

examples/language-modeling/run_mlm_wwm.py

@@ -355,12 +355,12 @@ def main():
     # Training
     if training_args.do_train:
         if last_checkpoint is not None:
-            model_path = last_checkpoint
+            checkpoint = last_checkpoint
         elif model_args.model_name_or_path is not None and os.path.isdir(model_args.model_name_or_path):
-            model_path = model_args.model_name_or_path
+            checkpoint = model_args.model_name_or_path
         else:
-            model_path = None
-        train_result = trainer.train(model_path=model_path)
+            checkpoint = None
+        train_result = trainer.train(resume_from_checkpoint=checkpoint)
         trainer.save_model()  # Saves the tokenizer too for easy upload
 
         output_train_file = os.path.join(training_args.output_dir, "train_results.txt")

examples/language-modeling/run_plm.py

@@ -384,12 +384,12 @@ def main():
     # Training
     if training_args.do_train:
         if last_checkpoint is not None:
-            model_path = last_checkpoint
+            checkpoint = last_checkpoint
         elif model_args.model_name_or_path is not None and os.path.isdir(model_args.model_name_or_path):
-            model_path = model_args.model_name_or_path
+            checkpoint = model_args.model_name_or_path
         else:
-            model_path = None
-        train_result = trainer.train(model_path=model_path)
+            checkpoint = None
+        train_result = trainer.train(resume_from_checkpoint=checkpoint)
         trainer.save_model()  # Saves the tokenizer too for easy upload
 
         output_train_file = os.path.join(training_args.output_dir, "train_results.txt")

examples/multiple-choice/run_swag.py

@@ -342,12 +342,12 @@ def main():
     # Training
     if training_args.do_train:
         if last_checkpoint is not None:
-            model_path = last_checkpoint
+            checkpoint = last_checkpoint
         elif os.path.isdir(model_args.model_name_or_path):
-            model_path = model_args.model_name_or_path
+            checkpoint = model_args.model_name_or_path
         else:
-            model_path = None
-        train_result = trainer.train(model_path=model_path)
+            checkpoint = None
+        train_result = trainer.train(resume_from_checkpoint=checkpoint)
         trainer.save_model()  # Saves the tokenizer too for easy upload
 
         output_train_file = os.path.join(training_args.output_dir, "train_results.txt")

examples/question-answering/run_qa.py

@@ -463,12 +463,12 @@ def main():
     # Training
     if training_args.do_train:
         if last_checkpoint is not None:
-            model_path = last_checkpoint
+            checkpoint = last_checkpoint
         elif os.path.isdir(model_args.model_name_or_path):
-            model_path = model_args.model_name_or_path
+            checkpoint = model_args.model_name_or_path
         else:
-            model_path = None
-        train_result = trainer.train(model_path=model_path)
+            checkpoint = None
+        train_result = trainer.train(resume_from_checkpoint=checkpoint)
         trainer.save_model()  # Saves the tokenizer too for easy upload
 
         output_train_file = os.path.join(training_args.output_dir, "train_results.txt")

examples/question-answering/run_qa_beam_search.py

@@ -502,12 +502,12 @@ def main():
     # Training
     if training_args.do_train:
         if last_checkpoint is not None:
-            model_path = last_checkpoint
+            checkpoint = last_checkpoint
         elif os.path.isdir(model_args.model_name_or_path):
-            model_path = model_args.model_name_or_path
+            checkpoint = model_args.model_name_or_path
         else:
-            model_path = None
-        train_result = trainer.train(model_path=model_path)
+            checkpoint = None
+        train_result = trainer.train(resume_from_checkpoint=checkpoint)
         trainer.save_model()  # Saves the tokenizer too for easy upload
 
         output_train_file = os.path.join(training_args.output_dir, "train_results.txt")

examples/seq2seq/run_seq2seq.py

@@ -491,12 +491,12 @@ def main():
     # Training
     if training_args.do_train:
         if last_checkpoint is not None:
-            model_path = last_checkpoint
+            checkpoint = last_checkpoint
         elif os.path.isdir(model_args.model_name_or_path):
-            model_path = model_args.model_name_or_path
+            checkpoint = model_args.model_name_or_path
         else:
-            model_path = None
-        train_result = trainer.train(model_path=model_path)
+            checkpoint = None
+        train_result = trainer.train(resume_from_checkpoint=checkpoint)
         trainer.save_model()  # Saves the tokenizer too for easy upload
 
         output_train_file = os.path.join(training_args.output_dir, "train_results.txt")

examples/text-classification/run_glue.py

@@ -399,12 +399,12 @@ def main():
     # Training
     if training_args.do_train:
         if last_checkpoint is not None:
-            model_path = last_checkpoint
+            checkpoint = last_checkpoint
         elif os.path.isdir(model_args.model_name_or_path):
-            model_path = model_args.model_name_or_path
+            checkpoint = model_args.model_name_or_path
         else:
-            model_path = None
-        train_result = trainer.train(model_path=model_path)
+            checkpoint = None
+        train_result = trainer.train(resume_from_checkpoint=checkpoint)
         metrics = train_result.metrics
 
         trainer.save_model()  # Saves the tokenizer too for easy upload

examples/token-classification/run_ner.py

@@ -380,12 +380,12 @@ def main():
     # Training
     if training_args.do_train:
         if last_checkpoint is not None:
-            model_path = last_checkpoint
+            checkpoint = last_checkpoint
         elif os.path.isdir(model_args.model_name_or_path):
-            model_path = model_args.model_name_or_path
+            checkpoint = model_args.model_name_or_path
         else:
-            model_path = None
-        train_result = trainer.train(model_path=model_path)
+            checkpoint = None
+        train_result = trainer.train(resume_from_checkpoint=checkpoint)
         trainer.save_model()  # Saves the tokenizer too for easy upload
 
         output_train_file = os.path.join(training_args.output_dir, "train_results.txt")