bos_token_id is equals to eos_token_id
#3
by mnwato - opened
After fine-tuning the mGPT-13B model, I am facing a problem generating a sentence as long as max_length because the model does not stop itself. I suspect that this is because the model cannot detect eos_token during fine-tuning.
Upon checking the config.json file, I found that "bos_token_id": 50256 is equal to "bos_token_id": 50256.
Any help would be appreciated.
mnwato changed discussion title from why bos_token_id equals to eos_token_id to bos_token_id is equals to eos_token_id