• Description:

Long T5 benchmark

This dataset collection comprises the evaluation benchmark used in the paper: LongT5: Efficient Text-To-Text Transformer for Long Sequences

LongT5 is an extension of the T5 model that handles long sequence inputs more efficiently. LongT5 achieves state-of-the-art performance on several summarization benchmarks that required longer context or multi-document understanding.

• Homepage:

https://github.com/google-research/longt5

• Versions:

  • 1.0.0 (default): Initial release

• Datasets in the default version:

  • natural_questions: natural_questions/longt5:0.1.0
  • media_sum: media_sum:1.0.0

• Citation: