Sentence-T5: Scalable Sentence Encoders from Pre-trained Text-to-Text Models

TL;DR

This paper explores methods to generate high-quality sentence embeddings from T5 models, introducing a new benchmark and demonstrating that scaled T5 models outperform existing approaches on various transfer tasks.

Contribution

It is the first to systematically extract sentence embeddings from T5, establish a new transfer benchmark, and show that scaled T5 models achieve state-of-the-art results.

Findings

Encoder-only T5 models outperform Sentence-BERT and SimCSE on SentEval and SentGLUE.

Scaling T5 from millions to billions of parameters improves embedding quality.

Encoder-decoder T5 achieves new state-of-the-art on STS benchmark.

Abstract

We provide the first exploration of sentence embeddings from text-to-text transformers (T5). Sentence embeddings are broadly useful for language processing tasks. While T5 achieves impressive performance on language tasks cast as sequence-to-sequence mapping problems, it is unclear how to produce sentence embeddings from encoder-decoder models. We investigate three methods for extracting T5 sentence embeddings: two utilize only the T5 encoder and one uses the full T5 encoder-decoder model. To support our investigation, we establish a new sentence representation transfer benchmark, SentGLUE, which extends the SentEval toolkit to nine tasks from the GLUE benchmark. Our encoder-only models outperforms Sentence-BERT and SimCSE sentence embeddings on both SentEval and SentGLUE transfer tasks, including semantic textual similarity (STS). Scaling up T5 from millions to billions of parameters is found to produce consistent further improvements. Finally, our encoder-decoder method achieves a new state-of-the-art on STS when using sentence embeddings. Our models are released at https://tfhub.dev/google/collections/sentence-t5/1.