Trainer中保存最佳模型存在bug

fastNLP/core/trainer.py

@@ -718,7 +718,7 @@ class Trainer(object):
                 self._save_model(self.model,
                                  "best_" + "_".join([self.model.__class__.__name__, self.metric_key, self.start_time]))
             elif self._load_best_model:
-                self._best_model_states = {name: param.cpu().clone() for name, param in self.model.state_dict()}
+                self._best_model_states = {name: param.cpu().clone() for name, param in self.model.state_dict().items()}
             self.best_dev_perf = res
             self.best_dev_epoch = epoch
             self.best_dev_step = step