Fix result saving errors of pytorch examples (#20276)

examples/pytorch/image-classification/run_image_classification_no_trainer.py

@@ -571,9 +571,9 @@ def main():
             if args.push_to_hub:
                 repo.push_to_hub(commit_message="End of training", auto_lfs_prune=True)
 
-    if args.output_dir is not None:
-        with open(os.path.join(args.output_dir, "all_results.json"), "w") as f:
-            json.dump({"eval_accuracy": eval_metric["accuracy"]}, f)
+            all_results = {f"eval_{k}": v for k, v in eval_metric.items()}
+            with open(os.path.join(args.output_dir, "all_results.json"), "w") as f:
+                json.dump(all_results, f)
 
 
 if __name__ == "__main__":

examples/pytorch/language-modeling/run_clm_no_trainer.py

@@ -666,8 +666,8 @@ def main():
             if args.push_to_hub:
                 repo.push_to_hub(commit_message="End of training", auto_lfs_prune=True)
 
-        with open(os.path.join(args.output_dir, "all_results.json"), "w") as f:
-            json.dump({"perplexity": perplexity}, f)
+            with open(os.path.join(args.output_dir, "all_results.json"), "w") as f:
+                json.dump({"perplexity": perplexity}, f)
 
 
 if __name__ == "__main__":

examples/pytorch/language-modeling/run_mlm_no_trainer.py

@@ -711,8 +711,8 @@ def main():
             if args.push_to_hub:
                 repo.push_to_hub(commit_message="End of training", auto_lfs_prune=True)
 
-        with open(os.path.join(args.output_dir, "all_results.json"), "w") as f:
-            json.dump({"perplexity": perplexity}, f)
+            with open(os.path.join(args.output_dir, "all_results.json"), "w") as f:
+                json.dump({"perplexity": perplexity}, f)
 
 
 if __name__ == "__main__":

examples/pytorch/multiple-choice/run_swag_no_trainer.py

@@ -85,7 +85,7 @@ def parse_args():
         "--validation_file", type=str, default=None, help="A csv or a json file containing the validation data."
     )
     parser.add_argument(
-        "--max_length",
+        "--max_seq_length",
         type=int,
         default=128,
         help=(
@@ -424,7 +424,7 @@ def main():
         tokenized_examples = tokenizer(
             first_sentences,
             second_sentences,
-            max_length=args.max_length,
+            max_length=args.max_seq_length,
             padding=padding,
             truncation=True,
         )
@@ -654,8 +654,10 @@ def main():
             tokenizer.save_pretrained(args.output_dir)
             if args.push_to_hub:
                 repo.push_to_hub(commit_message="End of training", auto_lfs_prune=True)
-        with open(os.path.join(args.output_dir, "all_results.json"), "w") as f:
-            json.dump({"eval_accuracy": eval_metric["accuracy"]}, f)
+
+            all_results = {f"eval_{k}": v for k, v in eval_metric.items()}
+            with open(os.path.join(args.output_dir, "all_results.json"), "w") as f:
+                json.dump(all_results, f)
 
 
 if __name__ == "__main__":

examples/pytorch/semantic-segmentation/run_semantic_segmentation_no_trainer.py

@@ -681,8 +681,9 @@ def main():
             if args.push_to_hub:
                 repo.push_to_hub(commit_message="End of training", auto_lfs_prune=True)
 
+            all_results = {f"eval_{k}": v for k, v in eval_metrics.items()}
             with open(os.path.join(args.output_dir, "all_results.json"), "w") as f:
-                json.dump({"eval_overall_accuracy": eval_metrics["overall_accuracy"]}, f)
+                json.dump(all_results, f)
 
 
 if __name__ == "__main__":

examples/pytorch/summarization/run_summarization_no_trainer.py

@@ -747,16 +747,10 @@ def main():
             tokenizer.save_pretrained(args.output_dir)
             if args.push_to_hub:
                 repo.push_to_hub(commit_message="End of training", auto_lfs_prune=True)
-        with open(os.path.join(args.output_dir, "all_results.json"), "w") as f:
-            json.dump(
-                {
-                    "eval_rouge1": result["rouge1"],
-                    "eval_rouge2": result["rouge2"],
-                    "eval_rougeL": result["rougeL"],
-                    "eval_rougeLsum": result["rougeLsum"],
-                },
-                f,
-            )
+
+            all_results = {f"eval_{k}": v for k, v in result.items()}
+            with open(os.path.join(args.output_dir, "all_results.json"), "w") as f:
+                json.dump(all_results, f)
 
 
 if __name__ == "__main__":

examples/pytorch/text-classification/run_glue_no_trainer.py

@@ -625,8 +625,9 @@ def main():
         logger.info(f"mnli-mm: {eval_metric}")
 
     if args.output_dir is not None:
+        all_results = {f"eval_{k}": v for k, v in eval_metric.items()}
         with open(os.path.join(args.output_dir, "all_results.json"), "w") as f:
-            json.dump({"eval_accuracy": eval_metric["accuracy"]}, f)
+            json.dump(all_results, f)
 
 
 if __name__ == "__main__":

examples/pytorch/token-classification/run_ner_no_trainer.py

@@ -766,10 +766,11 @@ def main():
             if args.push_to_hub:
                 repo.push_to_hub(commit_message="End of training", auto_lfs_prune=True)
 
-        with open(os.path.join(args.output_dir, "all_results.json"), "w") as f:
-            json.dump(
-                {"eval_accuracy": eval_metric["accuracy"], "train_loss": total_loss.item() / len(train_dataloader)}, f
-            )
+            all_results = {f"eval_{k}": v for k, v in eval_metric.items()}
+            if args.with_tracking:
+                all_results.update({"train_loss": total_loss.item() / len(train_dataloader)})
+            with open(os.path.join(args.output_dir, "all_results.json"), "w") as f:
+                json.dump(all_results, f)
 
 
 if __name__ == "__main__":