init

app.py

@@ -54,6 +54,7 @@ def run_omni(audio_path: str, instruction: str, max_tokens: int = 512) -> str:
     inputs = inputs.to(model.device)
 
     output_ids = model.generate(**inputs, max_new_tokens=int(max_tokens))
+    output_ids = output_ids[:, inputs["input_ids"].shape[1]:]
     response = processor.batch_decode(
         output_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False
     )[0]