fix bug

Author: LokiWager

src/easevoice/soundstorm/auto_reg/models/t2s_lightning_module.py

@@ -32,8 +32,8 @@ def __init__(self, config, output_dir, is_train=True):
         if is_train:
             self.automatic_optimization = False
             self.save_hyperparameters()
-            self.eval_dir.mkdir(parents=True, exist_ok=True)
             self.eval_dir = output_dir / "eval"
+            self.eval_dir.mkdir(parents=True, exist_ok=True)
 
     def training_step(self, batch: Dict, batch_idx: int):  # pyright: ignore
         opt: Any = self.optimizers()