TypeError: argument 'ids': 'list' object cannot be interpreted as an integer when lora training

šŸ¤—Transformers
5

I tried debugging it a little. I found the cause, or rather the trigger, but Iā€™m not sure what the right thing to do isā€¦
The cause is that the Evaluate function is trying to decode something that the tokenizer canā€™t decode.

    # 将 token IDs č§£ē äøŗę–‡ęœ¬
    print("predictions:", predictions)
    print("labels:", labels)
    #predictions_str = tokenizer.batch_decode(predictions.tolist(), skip_special_tokens=True) # cannot be decoded... like <class 'numpy.ndarray'> "predictions: [[[1.87890625, 8.796875, 11.3359375, 8.6328125, 7.796875, 10.8125, 11.3046875, 13.625, 9.8046875, 11.984375, ..."
    labels_str = tokenizer.batch_decode(labels.tolist(), skip_special_tokens=True)