init

app.py

@@ -12,7 +12,7 @@ model = Qwen2AudioForConditionalGeneration.from_pretrained(MODEL_ID, device_map=
 
 
 @spaces.GPU
-def run_qwen2audio(audio_path: str, instruction: str, max_tokens: int = 256) -> str:
+def run_qwen2audio(audio_path: str, instruction: str) -> str:
     if not audio_path:
         return "Please upload an audio file."
 
@@ -36,7 +36,7 @@ def run_qwen2audio(audio_path: str, instruction: str, max_tokens: int = 256) ->
     inputs = processor(text=text, audio=audios, return_tensors="pt", padding=True)
     inputs = inputs.to(model.device)
 
-    output_ids = model.generate(**inputs, max_length=int(max_tokens))
+    output_ids = model.generate(**inputs, max_new_tokens=4096)
     output_ids = output_ids[:, inputs.input_ids.size(1):]
     response = processor.batch_decode(output_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
     return response
@@ -50,12 +50,11 @@ with gr.Blocks(title="Qwen2-Audio Demo") as demo:
         with gr.Column():
             audio_input = gr.Audio(type="filepath", label="Upload Audio")
             instruction = gr.Textbox(label="Instruction", value="Transcribe the audio.")
-            max_tokens = gr.Slider(64, 1024, value=256, step=32, label="Max Tokens")
             submit_btn = gr.Button("Run", variant="primary")
         with gr.Column():
             output_text = gr.Textbox(label="Response", lines=12)
 
-    submit_btn.click(run_qwen2audio, [audio_input, instruction, max_tokens], output_text)
+    submit_btn.click(run_qwen2audio, [audio_input, instruction], output_text)
 
 
 if __name__ == "__main__":