ugh

vall_e/train.py

@@ -207,10 +207,10 @@ def run_eval(engines, eval_name, dl, args=None):
 				training=False,
 			)
 
-			if self.version >= 7:
+			if engine.hyper_config.version >= 7:
 				kwargs = base_kwargs | cfg.evaluation.kwargs
 				# sample for NAR demask
-				if random.random() < cfg.model.experimental.masking_train_p:
+				if random.random() < engine.hyper_config.experimental.masking_train_p:
 					kwargs["len_list"] = [ resp.shape[0] for resp in batch["resps"] ]
 				# inference
 				resps_list = engine( **kwargs )