jiant is a software toolkit for natural language processing research, designed to facilitate work on multitask learning and transfer learning for sentence understanding tasks.

A few things you might want to know about jiant:

• jiant is configuration-driven. You can run an enormous variety of experiments by simply writing configuration files. Of course, if you need to add any major new features, you can also easily edit or extend the code.
• jiant contains implementations of strong baselines for the GLUE and SuperGLUE benchmarks, and it's the recommended starting point for work on these benchmarks.
• jiant was developed at the 2018 JSALT Workshop by the General-Purpose Sentence Representation Learning team and is maintained by the NYU Machine Learning for Language Lab, with help from many outside collaborators (especially Google AI Language's Ian Tenney).
• jiant is built on PyTorch. It also uses many components from AllenNLP and the HuggingFace PyTorch implementations of BERT and GPT.
• The name jiant doesn't mean much. The 'j' stands for JSALT. That's all the acronym we have.

To find the setup instructions for using jiant and to run a simple example demo experiment using data from GLUE, follow this getting started tutorial!

Our official documentation is here: https://jiant.info/documentation#/

To run an experiment, make a config file similar to config/demo.conf with your model configuration. In addition, you can use the --overrides flag to override specific variables. For example:

python main.py --config_file config/demo.conf \
    --overrides "exp_name = my_exp, run_name = foobar, d_hid = 256"

will run the demo config, but output to $JIANT_PROJECT_PREFIX/my_exp/foobar. To run the demo config, you will have to set environment variables. The best way to achieve that is to follow the instructions in user_config_template.sh

• $JIANT_PROJECT_PREFIX: the where the outputs will be saved.
• $JIANT_DATA_DIR: location of the saved data. This is usually the location of the GLUE data in a simple default setup.
• $WORD_EMBS_FILE: location of any word embeddings you want to use (not necessary when using ELMo, GPT, or BERT). You can download GloVe (840B) here or fastText (2M) here. To have user_config.sh run automatically, follow instructions in scripts/export_from_bash.sh.

If you use jiant in academic work, please cite it directly:

@misc{wang2019jiant,
    author = {Alex Wang and Ian F. Tenney and Yada Pruksachatkun and Katherin Yu and Jan Hula and Patrick Xia and Raghu Pappagari and Shuning Jin and R. Thomas McCoy and Roma Patel and Yinghui Huang and Jason Phang and Edouard Grave and Najoung Kim and Phu Mon Htut and Thibault F'{e}vry and Berlin Chen and Nikita Nangia and Haokun Liu and and Anhad Mohananey and Shikha Bordia and Ellie Pavlick and Samuel R. Bowman},
    title = {{jiant} 1.0: A software toolkit for research on general-purpose text understanding models},
    howpublished = {\url{http://jiant.info/}},
    year = {2019}
}

jiant has been used in these four papers so far:

• Can You Tell Me How to Get Past Sesame Street? Sentence-Level Pretraining Beyond Language Modeling (formerly "Looking for ELMo's Friends")
• What do you learn from context? Probing for sentence structure in contextualized word representations ("edge probing")
• BERT Rediscovers the Classical NLP Pipeline
• Probing What Different NLP Tasks Teach Machines about Function Word Comprehension ("function word probing")

To exactly reproduce experiments from the ELMo's Friends paper use the jsalt-experiments branch. That will contain a snapshot of the code as of early August, potentially with updated documentation.

For the edge probing paper, see the probing/ directory.

For the function word probing paper, use this branch and refer to the instructions in the scripts/fwords/ directory.

Post an issue here on GitHub if you have any problems, and create a pull request if you make any improvements (substantial or cosmetic) to the code that you're willing to share.

We use the black coding style with a line limit of 100. After installing the requirements, simply running pre-commit install should ensure you comply with this in all your future commits. If you're adding features or fixing a bug, please also add the tests.

This package is released under the MIT License. The material in the allennlp_mods directory is based on AllenNLP, which was originally released under the Apache 2.0 license.

• Part of the development of jiant took at the 2018 Frederick Jelinek Memorial Summer Workshop on Speech and Language Technologies, and was supported by Johns Hopkins University with unrestricted gifts from Amazon, Facebook, Google, Microsoft and Mitsubishi Electric Research Laboratories.
• This work was made possible in part by a donation to NYU from Eric and Wendy Schmidt made by recommendation of the Schmidt Futures program.
• We gratefully acknowledge the support of NVIDIA Corporation with the donation of a Titan V GPU used at NYU in this work.
• Developer Alex Wang is supported by the National Science Foundation Graduate Research Fellowship Program under Grant No. DGE 1342536. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.
• Developer Yada Pruksachatkun is supported by the Moore-Sloan Data Science Environment as part of the NYU Data Science Services initiative.