Facing an error in Bert NLP "At most 4 tokens in tensor([ 2, 2, 2, 2, 44763, 44763, 2, 44763]) can be equal to eos_token_id: 2. Make sure tensor([ 2, 2, 2, 2, 44763, 44763, 2, 44763]) are corrected."

Models
1

I was having trouble with TensorFlow after training a Bert model in a Jupyter notebook.
It works fine on a Jupyter notebook but does not on another Ubuntu server where I am using trained model files. The error I’m facing is below:

At most 4 tokens in tensor([ 2, 2, 2, 2, 44763, 44763, 2, 44763]) can be equal to eos_token_id: 2. Make sure tensor([ 2, 2, 2, 2, 44763, 44763, 2, 44763]) are corrected.

We have finetuned our BERT model for text2text generation. It is working fine on the Jupyter notebook. But when I use the same trained model on another server of Ubuntu, then it shows the issue. This is my first post, so please bear with me.
My output generation code is:

from simpletransformers.seq2seq import Seq2SeqModel
#logging.basicConfig(level=logging.INFO)
#transformers_logger = logging.getLogger("transformers")
#transformers_logger.setLevel(logging.ERROR)
model = Seq2SeqModel(
    encoder_decoder_type="bart", encoder_decoder_name="PATHOFMODEL",use_cuda=False,
)
while True:
    original = input("Enter text to paraphrase: ")
    to_predict = [original]

    preds = model.predict(to_predict)

    print("---------------------------------------------------------")
    print(original)

    print()
    print("Predictions >>>")
    for pred in preds[0]:
        print(pred)

    print("---------------------------------------------------------")
    print()

Here’s the complete issue on StackOverflow.
https://stackoverflow.com/q/67195582
Someone told me to change the TensorFlow version. I tried it. It worked for one day. After that, I faced the same problem again.

1 Like