ParlAI, is a unified platform, implemented in Python, for training and evaluating AI models on a variety of openly available dialog datasets using open-sourced learning agents.
Its goal is to provide researchers:
- A unified framework for sharing, training and testing dialog models
- Multi-task training over many datasets at once
- Seamless integration of Amazon Mechanical Turk for data collection and human evaluation
Over 20 tasks are currently supported, including popular datasets such as SQuAD, bAbI tasks, MSMARCO, MCTest, [WikiQA, WebQuestions, SimpleQuestions, WikiMovies, QACNN & QADailyMail, CBT, BookTest, bAbI Dialog tasks, Ubuntu Dialog, OpenSubtitles, Cornell Movie, VQA-COCO2014, VisDial and CLEVR. See here for the current complete task list.
Included are examples of training neural models with PyTorch and Lua Torch, with batch training on GPU or hogwild training on CPUs. Using Theano or Tensorflow instead is also straightforward.
Our aim is for the number of tasks and agents that train on them to grow in a community-based way.
ParlAI is described in the following paper: ParlAI: A Dialog Research Software Platform, arXiv:1705.06476.
We are in Beta. Expect some adventures and rough edges. See the news page for the latest additions & updates, and the website http://parl.ai for further docs.