Add repetition_penalty

app.py

@@ -448,7 +448,7 @@ def infer(audio_path, youtube_url, prompt_text):
             return_dict=True,
         ).to(model.device)
 
-        gen_ids = model.generate(**batch, max_new_tokens=4096)
+        gen_ids = model.generate(**batch, max_new_tokens=4096, repetition_penalty=1.2)
         inp_len = batch["input_ids"].shape[1]
         new_tokens = gen_ids[:, inp_len:]
         texts = processor.batch_decode(new_tokens, skip_special_tokens=True, clean_up_tokenization_spaces=False)