add dataset-level distillation-style regularization

toolkit/config_modules.py

@@ -806,6 +806,7 @@ class DatasetConfig:
         self.buckets: bool = kwargs.get('buckets', True)
         self.bucket_tolerance: int = kwargs.get('bucket_tolerance', 64)
         self.is_reg: bool = kwargs.get('is_reg', False)
+        self.prior_reg: bool = kwargs.get('prior_reg', False)
         self.network_weight: float = float(kwargs.get('network_weight', 1.0))
         self.token_dropout_rate: float = float(kwargs.get('token_dropout_rate', 0.0))
         self.shuffle_tokens: bool = kwargs.get('shuffle_tokens', False)