fix: sdpa

app.py

@@ -1,3 +1,4 @@
+import spaces
 import gradio as gr
 import torch
 import torchaudio
@@ -149,6 +150,7 @@ def initialize_model():
 # Initialize model when starting the application
 initialize_model()
 
+@spaces.GPU
 def process_single_audio_generation(
     text_input: str,
     audio_mode: str,

generation_utils.py

@@ -15,7 +15,7 @@ SILENCE_DURATION = 5.0  # Fixed silence duration: 5 seconds
 def load_model(model_path, spt_config_path, spt_checkpoint_path):
     tokenizer = AutoTokenizer.from_pretrained(model_path)
     
-    model = AsteroidTTSInstruct.from_pretrained(model_path, torch_dtype=torch.bfloat16, attn_implementation="flash_attention_2")
+    model = AsteroidTTSInstruct.from_pretrained(model_path, torch_dtype=torch.bfloat16, attn_implementation="sdpa")
 
     spt = XY_Tokenizer.load_from_checkpoint(config_path=spt_config_path, ckpt_path=spt_checkpoint_path)