Practical overview
When it fits
Consider Jiant when a research workflow benefits from a configuration-driven toolkit spanning NLP tasks, benchmarks and transfer-learning experiments, and its maintenance status and dependency support fit the project.
Who it is for
- NLP researchers comparing language models across tasks.
- Developers adapting or extending research workflows built around task data and pretrained models.
Documented uses
- Configure and run a language-understanding experiment across one or more tasks.
- Fine-tune a pretrained model on a documented dataset such as MRPC.
- Organize experiments against GLUE, SuperGLUE or XTREME benchmarks.
Requirements & limits
The repository is public and specifies the MIT License. The README documents installation from source or with pip.
- The README lists support for more than 50 natural-language understanding tasks and the GLUE, SuperGLUE and XTREME benchmarks.
- Workflows are configuration-file driven and integrate with PyTorch, Hugging Face Datasets and Transformers.
- Active maintenance ended on 17 October 2021; the README says no new models, tasks or features, or updates for new library support, were planned.
- The README distinguishes the current 2.x line from v1.3.2, which is maintained in a separate legacy repository.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Install Jiant from source
Prepare a local source checkout so Jiant can be imported for research workflows.
Before you start
- Python environment; the README recommends a virtual or conda environment.
- Git and pip.
Steps
Clone https://github.com/nyu-mll/jiant.git and change into the jiant directory.
Install dependencies with pip install -r requirements.txt.
Add the checkout path to PYTHONPATH in the shell profile.
Expected outcome
The source checkout and its listed requirements are prepared for importing Jiant.
Guide limitations
- The README reports that active maintenance ended on 17 October 2021, including no planned updates for newer library support.
- The steps describe installation, not execution of an experiment.
Fine-tune RoBERTa on MRPC
Use Jiant's documented quick-start workflow to download MRPC data and configure a RoBERTa fine-tuning run.
Before you start
- Jiant installed in a Python environment.
- An experiment directory for task data and run files.
- The documented pretrained model identifier roberta-base.
Steps
Download the MRPC task data into the experiment directory's tasks folder using Jiant's data downloader.
Configure the run with the experiment and data directories, model identifier roberta-base, task mrpc, batch size 16 and three training epochs.
Call run.run_simple(args) to start the documented Python workflow.
Expected outcome
A configured Jiant simple-API fine-tuning run for RoBERTa on MRPC.
Guide limitations
- The README presents the workflow as an example; it does not establish results for a particular environment.
- The project reports that active maintenance ended on 17 October 2021.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 17 October 2026
- Original website title
- nyu-mll/jiant · Official homepage title · retained in its source language