@@ -113,6 +113,10 @@ def __init__(self, hparams, **kwargs):
113113 self .eval_max_length = self .hparams .eval_max_gen_length
114114 else :
115115 self .eval_max_length = self .model .config .max_length
116+ if self .hparams .eval_min_gen_length is not None :
117+ self .eval_min_length = self .hparams .eval_min_gen_length
118+ else :
119+ self .eval_min_length = self .model .config .min_length
116120 self .val_metric = self .default_val_metric if self .hparams .val_metric is None else self .hparams .val_metric
117121
118122 def save_readable_batch (self , batch : Dict [str , torch .Tensor ]) -> Dict [str , List [str ]]:
@@ -219,6 +223,7 @@ def _generative_step(self, batch: dict) -> dict:
219223 decoder_start_token_id = self .decoder_start_token_id ,
220224 num_beams = self .eval_beams ,
221225 max_length = self .eval_max_length ,
226+ min_length = self .eval_min_length ,
222227 )
223228 gen_time = (time .time () - t0 ) / batch ["input_ids" ].shape [0 ]
224229 preds : List [str ] = self .ids_to_clean_text (generated_ids )
@@ -346,6 +351,7 @@ def add_model_specific_args(parser, root_dir):
346351 "--val_metric" , type = str , default = None , required = False , choices = ["bleu" , "rouge2" , "loss" , None ]
347352 )
348353 parser .add_argument ("--eval_max_gen_length" , type = int , default = None , help = "never generate more than n tokens" )
354+ parser .add_argument ("--eval_min_gen_length" , type = int , default = None , help = "never generate shorter than n tokens" )
349355 parser .add_argument ("--save_top_k" , type = int , default = 1 , required = False , help = "How many checkpoints to save" )
350356 parser .add_argument (
351357 "--early_stopping_patience" ,
0 commit comments