Token indices sequence length is longer than the specified maximum sequence length for this model (565 > 512). Running this sequence through the model will result in indexing errors

[Roberyan]

I am trying to retrain the LISA with 7B model, I encounter such warning:

Token indices sequence length is longer than the specified maximum sequence length for this model (565 > 512). Running this sequence through the model will result in indexing errors

I don't know whether this has to with training error cuda out of memory after few iterations.