Update app.py

app.py

@@ -126,7 +126,6 @@ def _safe_load_lm(model_id: str, device: str) -> AutoModelForCausalLM:
     m.eval().to(device)
     return m
 
-@spaces.GPU
 def load_models(force_device: str | None = None):
     """本地：加载并缓存模型（无 spaces/ZeroGPU）"""
     global model, codec_model, asr_model, tokenizer
@@ -156,6 +155,8 @@ def load_models(force_device: str | None = None):
 
     logger.info("✅ All models loaded.")
 
+load_models()
+
 @torch.inference_mode()
 def inference_batch_transformers(
     lm: AutoModelForCausalLM,
@@ -254,7 +255,6 @@ def build_control_tags(age, gender, pitch, pitch_var, volume, speed, emo):
 
 @spaces.GPU
 def inference_select_best3(refined_text, instruct_text, age, gender, pitch, pitch_var, volume, speed, emo):
-    load_models()
     
     control_tags = build_control_tags(age, gender, pitch, pitch_var, volume, speed, emo)