The model was pre-trained using T5's denoising objective on C4, subsequently additionally pre-trained using REALM's salient span masking objective on Wikipedia, and finally fine-tuned on Natural Questions (NQ).

Note: The model was fine-tuned on 90% of the train splits of Natural Questions (NQ) for 20k steps and validated on the held-out 10% of the train split.

Other community Checkpoints: here

Paper: How Much Knowledge Can You Pack Into the Parameters of a Language Model?

Authors: Adam Roberts, Colin Raffel, Noam Shazeer

Results on Natural Questions - Test Set

Id	link	Exact Match
T5-large	https://huggingface.co/google/t5-large-ssm-nqo	29.0
T5-xxl	https://huggingface.co/google/t5-xxl-ssm-nqo	35.2
T5-3b	https://huggingface.co/google/t5-3b-ssm-nqo	31.7
T5-11b	https://huggingface.co/google/t5-11b-ssm-nqo	34.8

Usage

The model can be used as follows for closed book question answering:

python

from transformers import AutoModelForSeq2SeqLM, AutoTokenizer

t5_qa_model = AutoModelForSeq2SeqLM.from_pretrained("google/t5-large-ssm-nqo")
t5_tok = AutoTokenizer.from_pretrained("google/t5-large-ssm-nqo")

input_ids = t5_tok("When was Franklin D. Roosevelt born?", return_tensors="pt").input_ids
gen_output = t5_qa_model.generate(input_ids)[0]

print(t5_tok.decode(gen_output, skip_special_tokens=True))

Abstract

It has recently been observed that neural language models trained on unstructured text can implicitly store and retrieve knowledge using natural language queries. In this short paper, we measure the practical utility of this approach by fine-tuning pre-trained models to answer questions without access to any external context or knowledge. We show that this approach scales with model size and performs competitively with open-domain systems that explicitly retrieve answers from an external knowledge source when answering questions. To facilitate reproducibility and future work, we release our code and trained models at https://goo.gle/t5-cbqa.