LUKE (Language Understanding with Knowledge-based Embeddings) is a new pretrained contextualized representation of words and entities based on transformer. It was proposed in our paper LUKE: Deep Contextualized Entity Representations with Entity-aware Self-attention. It achieves state-of-the-art results on important NLP benchmarks including SQuAD v1.1 (extractive question answering), CoNLL-2003 (named entity recognition), ReCoRD (cloze-style question answering), TACRED (relation classification), and Open Entity (entity typing).
This repository contains the source code to pretrain the model and fine-tune it to solve downstream tasks.
April 13, 2022: The mLUKE fine-tuning code is available
The example code is updated.
Now it is based on allennlp and transformers.
You can reproduce the experiments in the LUKE and mLUKE papers with this implementation.
For the details, please see README.md under each example directory.
The older code used in the LUKE paper has been moved to examples/legacy.
April 13, 2022: The detailed instructions for pretraining LUKE models are available
For those interested in pretraining LUKE models, we explain how to prepare datasets and run the pretraining code on pretraining.md.
November 24, 2021: Entity disambiguation example is available
The example code of entity disambiguation based on LUKE has been added to this repository. This model was originally proposed in our paper, and achieved state-of-the-art results on five standard entity disambiguation datasets: AIDA-CoNLL, MSNBC, AQUAINT, ACE2004, and WNED-WIKI.
For further details, please refer to
examples/entity_disambiguation.
August 3, 2021: New example code based on Hugging Face Transformers and AllenNLP is available
New fine-tuning examples of three downstream tasks, i.e., NER, relation classification, and entity typing, have been added to LUKE. These examples are developed based on Hugging Face Transformers and AllenNLP. The fine-tuning models are defined using simple AllenNLP's Jsonnet config files!
The example code is available in examples.
May 5, 2021: LUKE is added to Hugging Face Transformers
LUKE has been added to the master branch of the Hugging Face Transformers library. You can now solve entity-related tasks (e.g., named entity recognition, relation classification, entity typing) easily using this library.
For example, the LUKE-large model fine-tuned on the TACRED dataset can be used as follows:
from transformers import LukeTokenizer, LukeForEntityPairClassification
model = LukeForEntityPairClassification.from_pretrained("studio-ousia/luke-large-finetuned-tacred")
tokenizer = LukeTokenizer.from_pretrained("studio-ousia/luke-large-finetuned-tacred")
text = "Beyoncé lives in Los Angeles."
entity_spans = [(0, 7), (17, 28)] # character-based entity spans corresponding to "Beyoncé" and "Los Angeles"
inputs = tokenizer(text, entity_spans=entity_spans, return_tensors="pt")
outputs = model(**inputs)
logits = outputs.logits
predicted_class_idx = int(logits[0].argmax())
print("Predicted class:", model.config.id2label[predicted_class_idx])
# Predicted class: per:cities_of_residenceWe also provide the following three Colab notebooks that show how to reproduce our experimental results on CoNLL-2003, TACRED, and Open Entity datasets using the library:
- Reproducing experimental results of LUKE on CoNLL-2003 Using Hugging Face Transformers
- Reproducing experimental results of LUKE on TACRED Using Hugging Face Transformers
- Reproducing experimental results of LUKE on Open Entity Using Hugging Face Transformers
Please refer to the official documentation for further details.
November 5, 2021: LUKE-500K (base) model
We released LUKE-500K (base), a new pretrained LUKE model which is smaller than existing LUKE-500K (large). The experimental results of the LUKE-500K (base) and LUKE-500K (large) on SQuAD v1 and CoNLL-2003 are shown as follows:
| Task | Dataset | Metric | LUKE-500K (base) | LUKE-500K (large) |
|---|---|---|---|---|
| Extractive Question Answering | SQuAD v1.1 | EM/F1 | 86.1/92.3 | 90.2/95.4 |
| Named Entity Recognition | CoNLL-2003 | F1 | 93.3 | 94.3 |
We tuned only the batch size and learning rate in the experiments based on LUKE-500K (base).
LUKE outperforms the previous state-of-the-art methods on five important NLP tasks:
| Task | Dataset | Metric | LUKE-500K (large) | Previous SOTA |
|---|---|---|---|---|
| Extractive Question Answering | SQuAD v1.1 | EM/F1 | 90.2/95.4 | 89.9/95.1 (Yang et al., 2019) |
| Named Entity Recognition | CoNLL-2003 | F1 | 94.3 | 93.5 (Baevski et al., 2019) |
| Cloze-style Question Answering | ReCoRD | EM/F1 | 90.6/91.2 | 83.1/83.7 (Li et al., 2019) |
| Relation Classification | TACRED | F1 | 72.7 | 72.0 (Wang et al. , 2020) |
| Fine-grained Entity Typing | Open Entity | F1 | 78.2 | 77.6 (Wang et al. , 2020) |
These numbers are reported in our EMNLP 2020 paper.
LUKE can be installed using Poetry:
poetry install
# If you want to run pretraining for LUKE
poetry install --extras "pretraining opennlp"
# If you want to run pretraining for mLUKE
poetry install --extras "pretraining icu"The virtual environment automatically created by Poetry can be activated by
poetry shell.
A note on installing torch
The pytorch installed via poetry install does not necessarily match your hardware.
In such case, see the official site and reinstall the correct version with the pip command.
poetry run pip3 uninstall torch torchvision torchaudio
# Example for Linux with CUDA 11.3
poetry run pip3 install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu113Our pretrained models can be used with the transformers library. The model documentations can be found in the following links: LUKE and mLUKE.
Currently, the following models are available on the Hugging Face Model Hub.
| Name | model_name | Entity Vocab Size | Params |
|---|---|---|---|
| LUKE (base) | studio-ousia/luke-base | 500K | 253 M |
| LUKE (large) | studio-ousia/luke-large | 500K | 484 M |
| mLUKE (base) | studio-ousia/mluke-base | 1.2M | 586 M |
| mLUKE (large) | studio-ousia/mluke-large | 1.2M | 868 M |
The entity embeddings cause a large memory footprint as they contain all the Wikipedia entities that we used in pretraining.
However, in some downstream tasks (e.g., entity typing, named entity recognition, and relation classification), we only need special entity embeddings such as [MASK].
Also, you may want to only use the word representations.
With such use-cases in mind, to make our models easier to use, we have uploaded lite models only with special entity embeddings. These models perform exactly the same as the full models but have much fewer parameters, which enable fine-tuning the model with small GPUs.
| Name | model_name | Params |
|---|---|---|
| LUKE (base) | studio-ousia/luke-base-lite | 125 M |
| LUKE (large) | studio-ousia/luke-large-lite | 356 M |
| mLUKE (base) | studio-ousia/mluke-base-lite | 279 M |
| mLUKE (large) | studio-ousia/mluke-large-lite | 561 M |
We release the fine-tuning code based on allennlp and transformers under examples.
You can run fine-tuning experiments very easily with pre-defined config files and the allennlp train command.
For the details and example commands for each task, please see the task directory under examples.
The detailed instructions for pretraining luke models can be found on pretraining.md.
If you use LUKE in your work, please cite the original paper.
@inproceedings{yamada-etal-2020-luke,
title = "{LUKE}: Deep Contextualized Entity Representations with Entity-aware Self-attention",
author = "Yamada, Ikuya and
Asai, Akari and
Shindo, Hiroyuki and
Takeda, Hideaki and
Matsumoto, Yuji",
booktitle = "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)",
year = "2020",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2020.emnlp-main.523",
doi = "10.18653/v1/2020.emnlp-main.523",
}
The full reference information can also be found here.
For mLUKE, please cite this paper.
@inproceedings{ri-etal-2022-mluke,
title = "m{LUKE}: {T}he Power of Entity Representations in Multilingual Pretrained Language Models",
author = "Ri, Ryokan and
Yamada, Ikuya and
Tsuruoka, Yoshimasa",
booktitle = "ACL 2022 (to appear)",
year = "2022",
}