Update app.py

app.py

@@ -127,7 +127,6 @@ def _safe_load_lm(model_id: str, device: str) -> AutoModelForCausalLM:
     return m
 
 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
 
     logger.info(f"Using device: {device}")
@@ -157,7 +156,7 @@ def load_models(force_device: str | None = None):
 
 load_models()
 
-@torch.inference_mode()
+@spaces.GPU
 def inference_batch_transformers(
     lm: AutoModelForCausalLM,
     codec: XCodec2Model,