Skip to content
/ DOOLY Public

🦕 A library that handles everything with 🤗 and supports batching to models in PORORO

License

Notifications You must be signed in to change notification settings

jinmang2/DOOLY

Repository files navigation

DOOLY 🦕

PORORO에는 아래 세 가지 단점이 존재합니다.

  • 일부 task의 batch화 불가능
  • 내부 tokenize 과정 및 모듈 구조 확인이 어려움
  • fairseq dependency

Dooly는 위 단점 세 가지를 개선한 라이브러리입니다.

  • 모든 task를 batch화하여 inference 가능
  • task별 tokenizer, model을 모듈로 분리하여 출력값 확인 가능
  • 모든 것을 huggingface transformers로 처리

How to use?

아래와 같이 간단하게 사용 가능합니다

  • install
$ pip install dooly
  • how to use
    • PORORO와 동일하게 사용할 수 있습니다.
from dooly import Dooly

ner = Dooly(task="ner", lang="ko")

Supported Tasks

  • Back Translation Data Augmentation
  • Dependency Parsing
  • Machine Reading Comprehension
  • Machine Translation
  • Named Entity Recognition
  • Natural Language Inference
  • Pos Tagging
  • Question Generation
  • Word Embedding
  • Word Sense Disambiguation
  • Zero Shot Topic Classification

Citations

@misc{pororo,
  author       = {Heo, Hoon and Ko, Hyunwoong and Kim, Soohwan and
                  Han, Gunsoo and Park, Jiwoo and Park, Kyubyong},
  title        = {PORORO: Platform Of neuRal mOdels for natuRal language prOcessing},
  howpublished = {\url{https://github.com/kakaobrain/pororo}},
  year         = {2021},
}

About

🦕 A library that handles everything with 🤗 and supports batching to models in PORORO

Resources

License

Stars

Watchers

Forks

Packages

No packages published