ParlAI Documentation

ParlAI is a one-stop-shop for dialog research.

Tutorials

• ParlAI Quick-start
• What is ParlAI?
• Tasks and Datasets in ParlAI
• Data Handling, Batching, and Hogwild
• Creating an Agent
• Using Torch Ranker Agent
• Tips and Tricks
• Using Mechanical Turk
• Using Facebook Messenger
• Using tensorboard for metric tracking

Tasks & Model Zoo

• Tasks
  • QA Tasks
  • Cloze Tasks
  • Goal Tasks
  • ChitChat Tasks
  • Negotiation Tasks
  • Visual Tasks
  • decanlp Tasks
• Model Zoo
  • Pretrained Embeddings
  • BERT
  • Kvmemnn Convai2 Model
  • Seq2Seq Convai2 Model
  • Seq2Seq Twitter Model
  • Drqa Squad Model
  • Wikipedia Retriever (Used For Open Squad)
  • Wikipedia Retriever (Used For Wizard Of Wikipedia)
  • Wizard Of Wikipedia (End To End Generator)
  • Wizard Of Wikipedia (Full Dialogue Retrieval Model)
  • Light Bert-Biranker Dialogue Model
  • Twitter Conversational Model
  • Controllable Dialogue Pretrained Models
  • Personality-Captions Pretrained Model

Core Library

• observations
• core.agents
• core.torch_agent
• core.build_data
• core.dict
• core.metrics
• core.params
• core.teachers
• core.utils
• core.thread_utils
• core.worlds

Reference Models

• agents.repeat_label
• agents.unigram
• agents.example_seq2seq

Scripts

• Command Line Usage
  • eval_ppl
  • build_pytorch_data
  • display_data
  • profile_train
  • interactive
  • build_dict
  • detect_offensive_language
  • eval_model
  • interactive_rank
  • data_stats
  • verify_data
  • display_model
  • extract_image_feature
  • train_model
  • eval_wordstat
  • interactive_web

Indices and tables

• Index

• Module Index