LongT5 (local attention, base-sized model)

LongT5 model pre-trained on English language. The model was introduced in the paper LongT5: Efficient Text-To-Text Transformer for Long Sequences by Guo et al. and first released in the LongT5 repository. All the model architecture and configuration can be found in Flaxformer repository which uses another Google research project repository T5x.

Disclaimer: The team releasing LongT5 did not write a model card for this model so this model card has been written by the Hugging Face team.

Model description

LongT5 model is an encoder-decoder transformer pre-trained in a text-to-text denoising generative setting (Pegasus-like generation pre-training). LongT5 model is an extension of T5 model, and it enables using one of the two different efficient attention mechanisms - (1) Local attention, or (2) Transient-Global attention. The usage of attention sparsity patterns allows the model to efficiently handle input sequence.

LongT5 is particularly effective when fine-tuned for text generation (summarization, question answering) which requires handling long input sequences (up to 16,384 tokens).

Intended uses & limitations

The model is mostly meant to be fine-tuned on a supervised dataset. See the model hub to look for fine-tuned versions on a task that interests you.

How to use

from transformers import AutoTokenizer, LongT5Model

tokenizer = AutoTokenizer.from_pretrained("google/long-t5-local-base")
model = LongT5Model.from_pretrained("google/long-t5-local-base")

inputs = tokenizer("Hello, my dog is cute", return_tensors="pt")
outputs = model(**inputs)

last_hidden_states = outputs.last_hidden_state

BibTeX entry and citation info

@article{guo2021longt5,
  title={LongT5: Efficient Text-To-Text Transformer for Long Sequences},
  author={Guo, Mandy and Ainslie, Joshua and Uthus, David and Ontanon, Santiago and Ni, Jianmo and Sung, Yun-Hsuan and Yang, Yinfei},
  journal={arXiv preprint arXiv:2112.07916},
  year={2021}
}