Skip to content
Get packs

ParlAI

A Python framework for training and testing dialogue AI models on many public datasets.

GitHub Download 1.7.2
Stars
10.6k
Forks
2k
Licence
MIT
Language
Python
Last update
3 months ago
Type
Library

This repository is archived and no longer maintained.

What it is

ParlAI is a Python framework for building, sharing, training and testing dialogue models, from open-domain chitchat to task-oriented dialogue and visual question answering. It bundles 100+ public datasets behind one common API, plus reference models and a collection of pretrained models. Researchers use it to compare chatbot approaches on the same data.

  • Display examples from a dataset such as SQuAD with the parlai display_data command
  • Evaluate a baseline model on a validation set, for example PersonaChat
  • Train a small transformer ranker on PersonaChat using the train_model command
  • Connect an agent to humans through a chat interface, such as the Facebook Messenger integration

Great for

  • Researchers comparing dialogue models on standard datasets
  • Students learning how conversational AI systems are trained and evaluated
  • Developers who want a reference framework for building their own dialogue agents
  • Teams running human evaluation or crowdsourced data collection via Mechanical Turk

Not ideal for

  • Windows users, since the README says it is not supported and works best on Linux or macOS
  • People wanting a ready-made chatbot product with no coding, since it is a research framework
  • Anyone needing active maintenance, since the repository is archived and the README notes the install may fail with some PyTorch dependencies

How to get it

  1. 1

    Install with a package manager

    You'll need: Python 3.8+ · PyTorch 1.6 or higher · Linux or macOS recommended

    Create a virtual environment and install from PyPI

    python3.8 -m venv venv && venv/bin/pip install --upgrade pip setuptools wheel && venv/bin/pip install parlai

    Developer install: clone the repository

    git clone https://github.com/facebookresearch/ParlAI.git ~/ParlAI
  2. 2

    Run it

    • Install ParlAI into a virtual environment using the commands above
    • Run parlai display_data -t squad to see 10 random examples from SQuAD
    • Run parlai eval_model -m ir_baseline -t personachat -dt valid to evaluate a baseline model
    • Follow the Quick Start tutorial in the documentation for next steps
  3. 3

    Get the most out of it

    • Use the complete task list in parlai/tasks/task_list.py to find datasets
    • Explore the agents list and model zoo to compare pretrained and reference models
    • Use the crowdsourcing tutorial to run data collection or human evaluation tasks
    • Check NEWS.md for an archived history of updates

Downloads

Source code (.zip)main branch · straight from GitHubRelease 1.7.2Release notes and sourceOfficial website & docsparl.ai

Strengths

  • Many popular dialogue and QA datasets available through one consistent API
  • Includes reference models from simple retrieval baselines to Transformers, plus a pretrained model zoo
  • Documented tutorials and a Colab notebook for a quick start

Trade-offs

  • The repository is archived, so it may not receive updates or fixes
  • Windows is not supported and some installs need manual PyTorch setup
  • Requires solid Python and machine learning knowledge to use beyond the examples

How it compares

ProjectStarsTypeLevelUpdatedBest for
ParlAI (this)10.6kLibraryadvanced3 months agoResearchers comparing dialogue models on standard datasets
fairseq32.2kFrameworkadvanced1.1 years agoResearchers reproducing published sequence-modeling papers

Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.