init

app.py

@@ -24,7 +24,7 @@ def _load_audio_mono_16k(file_path: str) -> torch.Tensor:
     return wav
 
 @spaces.GPU
-def process_audio(audio_path: str, instruction: str, max_tokens: int = 200) -> str:
+def process_audio(audio_path: str, instruction: str) -> str:
     if not audio_path:
         return "Please upload an audio file."
 
@@ -47,7 +47,7 @@ def process_audio(audio_path: str, instruction: str, max_tokens: int = 200) -> s
     model_inputs = processor(prompt, wav, device=device, return_tensors="pt").to(device)
     outputs = model.generate(
         **model_inputs,
-        max_new_tokens=int(max_tokens),
+        max_new_tokens=4096,
         do_sample=False,
         num_beams=1,
     )
@@ -69,12 +69,11 @@ with gr.Blocks(title="Granite Speech Demo") as demo:
                 label="Instruction",
                 value="can you transcribe the speech into a written format?",
             )
-            max_tokens = gr.Slider(50, 1000, value=200, step=50, label="Max Output Tokens")
             submit_btn = gr.Button("Transcribe", variant="primary")
         with gr.Column():
             output_text = gr.Textbox(label="Output", lines=12)
 
-    submit_btn.click(process_audio, [audio_input, instruction, max_tokens], output_text)
+    submit_btn.click(process_audio, [audio_input, instruction], output_text)
 
 
 if __name__ == "__main__":