Sync preprocesses before loading the processor at run_speech_recognition_ctc.py (#21926)

* Update run_speech_recognition_ctc.py

Make sure all processes wait until data is saved before loading the processor from the output_dit

* Make sure all processes wait until data is saved before loading the processor from the output_dit

* Update run_speech_recognition_ctc.py

* Update run_speech_recognition_seq2seq.py

examples/pytorch/speech-recognition/run_speech_recognition_ctc.py

@@ -673,6 +673,9 @@ def main():
         return metrics
 
     # Now save everything to be able to create a single processor later
+    # make sure all processes wait until data is saved
+    with training_args.main_process_first():
+        # only the main process saves them
         if is_main_process(training_args.local_rank):
             # save feature extractor, tokenizer and config
             feature_extractor.save_pretrained(training_args.output_dir)

examples/pytorch/speech-recognition/run_speech_recognition_seq2seq.py

@@ -506,6 +506,9 @@ def main():
         return {"wer": wer}
 
     # 9. Create a single speech processor
+    # make sure all processes wait until data is saved
+    with training_args.main_process_first():
+        # only the main process saves them
         if is_main_process(training_args.local_rank):
             # save feature extractor, tokenizer and config
             feature_extractor.save_pretrained(training_args.output_dir)