google/t5-3b-ssm-nqo

The model was fine-tuned on 90% of the train splits of Natural Questions (NQ) for 20k steps . It was pre-trained using T5's denoising objective on C4 . The model can be used as follows for closed book question answering: t5_qa_model = AutoModelForSeq2SeqLM, AutoTokenizer.from_pretrained("google/t5-3b-ssm-nqo")t5_tok = AutoTokenizers.(T5_ok) (T5-tok) for T5-xxl, T5_qo_qL (QL),