limit gen max token

run_demo_multi_models.py

@@ -22,6 +22,7 @@ MODEL_NAMES = [
     "bhuang/whisper-medium-cv11-french-case-punctuation",
 ]
 CHUNK_LENGTH_S = 30
+MAX_NEW_TOKENS = 225
 
 logging.basicConfig(
     format="%(asctime)s [%(levelname)s] [%(name)s] %(message)s",
@@ -48,6 +49,8 @@ def maybe_load_cached_pipeline(model_name):
         )
         # set forced_decoder_ids
         pipe.model.config.forced_decoder_ids = pipe.tokenizer.get_decoder_prompt_ids(language="fr", task="transcribe")
+        # limit genneration max length
+        pipe.model.config.max_length = MAX_NEW_TOKENS + 1
 
         logger.info(f"`{model_name}` pipeline has been initialized")