đ Google's T5 for Closed Book Question Answering
Google's T5 model is designed for Closed Book Question Answering, leveraging pre - training and fine - tuning on multiple datasets to provide effective question - answering capabilities.
đĻ Installation
Since the provided README doesn't have specific installation steps, this section is skipped.
⨠Features
- Multi - stage Training: The model was pre - trained using T5's denoising objective on C4, then additionally pre - trained using REALM's salient span masking objective on Wikipedia, and finally fine - tuned on Natural Questions (NQ).
- Fine - tuning Details: The model was fine - tuned on 100% of the train splits of Natural Questions (NQ) for 10k steps.
đ Documentation
đģ Usage Examples
Basic Usage
The model can be used as follows for closed book question answering:
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
t5_qa_model = AutoModelForSeq2SeqLM.from_pretrained("google/t5-small-ssm-nq")
t5_tok = AutoTokenizer.from_pretrained("google/t5-small-ssm-nq")
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))
đ Results on Natural Questions - Test Set
Id |
link |
Exact Match |
T5 - small |
https://huggingface.co/google/t5-small-ssm-nq |
25.5 |
T5 - large |
https://huggingface.co/google/t5-large-ssm-nq |
30.4 |
T5 - xl |
https://huggingface.co/google/t5-xl-ssm-nq |
35.6 |
T5 - xxl |
https://huggingface.co/google/t5-xxl-ssm-nq |
37.9 |
T5 - 3b |
https://huggingface.co/google/t5-3b-ssm-nq |
33.2 |
T5 - 11b |
https://huggingface.co/google/t5-11b-ssm-nq |
36.6 |
đ§ Technical Details
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.

đ License
The model is licensed under the apache - 2.0 license.
â ī¸ Important Note
The model was fine - tuned on 100% of the train splits of Natural Questions (NQ) for 10k steps.