fix decoding in seq2seq

src/train_sft.py

@@ -80,15 +80,19 @@ def main():
     # Evaluation
     if training_args.do_eval:
         metrics = trainer.evaluate(metric_key_prefix="eval", **gen_kwargs)
+        if training_args.predict_with_generate: # eval_loss will be wrong if predict_with_generate is enabled
+            metrics.pop("eval_loss", None)
         trainer.log_metrics("eval", metrics)
         trainer.save_metrics("eval", metrics)
 
     # Predict
     if training_args.do_predict:
         predict_results = trainer.predict(dataset, metric_key_prefix="predict", **gen_kwargs)
+        if training_args.predict_with_generate: # predict_loss will be wrong if predict_with_generate is enabled
+            predict_results.metrics.pop("predict_loss", None)
         trainer.log_metrics("predict", predict_results.metrics)
         trainer.save_metrics("predict", predict_results.metrics)
-        trainer.save_predictions(predict_results, tokenizer)
+        trainer.save_predictions(predict_results)
 
 
 def _mp_fn(index):

src/utils/seq2seq.py

@@ -32,18 +32,12 @@ class ComputeMetrics:
         Uses the model predictions to compute metrics.
         """
         preds, labels = eval_preds
-
-        if isinstance(preds, tuple):
-            preds = preds[0]
-
-        # Replace IGNORE_INDEX in the labels with pad_token_id as we cannot decode them.
-        preds = np.where(preds != IGNORE_INDEX, preds, self.tokenizer.pad_token_id)
-        labels = np.where(labels != IGNORE_INDEX, labels, self.tokenizer.pad_token_id)
-
-        preds = preds[:, labels.shape[1]:] # remove prompts
         score_dict = {"rouge-1": [], "rouge-2": [], "rouge-l": [], "bleu-4": []}
 
         for pred, label in zip(preds, labels):
+            pred = pred[len(label) - np.sum(label == IGNORE_INDEX) : len(pred) - np.sum(pred == IGNORE_INDEX)] # remove prompts
+            label = label[:len(label) - np.sum(label == IGNORE_INDEX)]
+
             hypothesis = list(jieba.cut(self.tokenizer.decode(pred, skip_special_tokens=True)))
             reference = list(jieba.cut(self.tokenizer.decode(label, skip_special_tokens=True)))
 
@@ -70,8 +64,7 @@ class Seq2SeqPeftTrainer(PeftTrainer):
 
     def save_predictions(
             self,
-            predict_results: PredictionOutput,
+            predict_results: PredictionOutput
-            tokenizer: PreTrainedTokenizer
     ) -> None:
         r"""
         Saves model predictions to `output_dir`.
@@ -81,17 +74,17 @@ class Seq2SeqPeftTrainer(PeftTrainer):
         if not self.is_world_process_zero():
             return
 
-        preds = np.where(predict_results.predictions != IGNORE_INDEX, predict_results.predictions, self.tokenizer.pad_token_id)
-        labels = np.where(predict_results.label_ids != IGNORE_INDEX, predict_results.label_ids, self.tokenizer.pad_token_id)
-
-        preds = preds[:, labels.shape[1]:] # remove prompts
-        preds = [tokenizer.decode(pred, skip_special_tokens=True).strip() for pred in preds]
-        labels = [tokenizer.decode(label, skip_special_tokens=True).strip() for label in labels]
-
         output_prediction_file = os.path.join(self.args.output_dir, "generated_predictions.jsonl")
         logger.info(f"Saving prediction results to {output_prediction_file}")
         with open(output_prediction_file, "w", encoding="utf-8") as writer:
             res: List[str] = []
-            for pred, label in zip(preds, labels):
+            for pred, label in zip(predict_results.predictions, predict_results.label_ids):
+                pred = pred[len(label) - np.sum(label == IGNORE_INDEX) : len(pred) - np.sum(pred == IGNORE_INDEX)] # remove prompts
+                label = label[:len(label) - np.sum(label == IGNORE_INDEX)]
+
+                pred = self.tokenizer.decode(pred, skip_special_tokens=True)
+                label = self.tokenizer.decode(label, skip_special_tokens=True)
+
                 res.append(json.dumps({"label": label, "predict": pred}, ensure_ascii=False))
+
             writer.write("\n".join(res))