Update app.py

app.py

@@ -60,6 +60,7 @@ def process_prompt(prompt, voice, tokenizer, device):
     return modified_input_ids.to(device), attention_mask.to(device)
 
 # Generate speech tokens
+@spaces.GPU()
 def generate_speech_tokens(input_ids, attention_mask, model, params):
     with torch.no_grad():
         generated_ids = model.generate(
@@ -130,7 +131,6 @@ def redistribute_codes(code_list, snac_model):
     return audio_hat.detach().squeeze().cpu().numpy()  # Always return CPU numpy array
 
 # Main generation function
-spaces.GPU()
 def generate_speech(text, voice, temperature, top_p, repetition_penalty, max_new_tokens, progress=gr.Progress()):
     if not text.strip():
         return None