AllenNLP Toolkit

An AllenNLP Toolkit is an NLP framework managed by AllenAI.

• Context:
  • It can include AllenNLP Models[1].
  • It can (typically) be built using PyTorch.
  • …
• Example(s):
  • AllenNLP v0.7 (~2018-10-05) [2].
  • …
• Counter-Example(s):
  • NLTK Toolkit.
  • PyTorch-NLP Library.
  • PyText.
• See: Allen Institute of AI.

---

---

References

• http://allennlp.org

2018

• (2018). “Writing Code for NLP Research." Presentation at EMNLP-2018.
  • QUOTE:

    ... under the covers, every piece of a model is a torch.nn.Module and every number is part of a torch.Tensor ...

    ... At the core of AllenNLP is the Model. A Model turns inputs into predictions + possibly a loss.

    The key things to notice are:

    • every Model is a torch.nn.Module.
    • it has a forward function that takes any number of inputs and returns a dict of Tensors (this is where our attempts to type hint everything break down)
      (torch.nn.Module is not a well-typed abstraction)
    • it also knows how to compute zero or more metrics (e.g. f1_score or categorical_accuracy)
    • it can be saved to a file and loaded from a file

2017

• https://github.com/allenai/allennlp
  • QUOTE: Built on PyTorch, AllenNLP makes it easy to design and evaluate new deep learning models for nearly any NLP problem, along with the infrastructure to easily run them in the cloud or on your laptop. AllenNLP was designed with the following principles:
    1. Hyper-modular and lightweight. Use the parts which you like seamlessly with PyTorch.
    2. Extensively tested and easy to extend. Test coverage is above 90% and the example models provide a template for contributions.
    3. Take padding and masking seriously, making it easy to implement correct models without the pain.
    4. Experiment friendly. Run reproducible experiments from a json specification with comprehensive logging.
  • AllenNLP includes reference implementations of high quality models for Semantic Role Labelling, Question and Answering (BiDAF), Entailment (decomposable attention), and more.

    AllenNLP is built and maintained by the Allen Institute for Artificial Intelligence, in close collaboration with researchers at the University of Washington and elsewhere. With a dedicated team of best-in-field researchers and software engineers, the AllenNLP project is uniquely positioned to provide state of the art models with high quality engineering.

---