Overview
ParlAI is an open-source, Python-based platform that helps users train, configure, and test dialogue models for conversational AI. ParlAI integrates with existing chat services and provides various datasets and reference models to improve dialog AI research.
This instructor-led, live training (online or onsite) is aimed at researchers and developers who wish to install, configure, customize, and manage the ParlAI platform to develop their AI models.
By the end of this training, participants will be able to share, train, and evaluate AI models to build and develop conversational solutions across existing chat services.
Format of the Course
- Interactive lecture and discussion.
- Lots of exercises and practice.
- Hands-on implementation in a live-lab environment.
Course Customization Options
- To request a customized training for this course, please contact us to arrange.
Requirements
- Knowledge of Python or other programming languages
- General understanding of artificial intelligence (AI) concepts
Audience
- Researchers
- Developers
Course Outline
Introduction
Overview of ParlAI Features and Architecture
- ParlAI framework
- Key capabilities and goals
- Core concepts (agents, messages, teachers, and worlds)
Getting Started with ParlAI for Conversational AI
- Installation
- Adding a simple model
- Simple display data script
- Validation and testing
- Tasks
- Agent training and evaluation
- Interacting with models
Working with Tasks and Datasets in ParlAI
- Adding datasets
- Separating data into sets (train, valid, or test)
- Using JSON instead of a text file
- Creating and executing tasks
Exploring Worlds, Sharing, and Batching
- The concept of Worlds
- Agent sharing
- Implementing batching
- Dynamic batching
Using Torch Generator and Ranker Agents
- Torch generator agent
- Torch ranker agent
- Example models
- Creating models
- Training and evaluating models
Adding Built-In and Custom Metrics
- Standard metrics
- Adding custom metrics
- Teacher metrics
- Agent level metrics (global and local)
- List of metrics
Speeding up Training Runs in ParlAI
- Setting a baseline
- Skip generation command
- Dynamic batching training command
- Using FP16 and multiple GPUs
- Background preprocessing
Exploring Other ParlAI Topics
- Using and writing mutators
- Running crowdsourcing tasks
- Using existing chat services
- Swapping out transformer subcomponents
- Running and writing tests
- ParlAI tips and tricks
Troubleshooting
Summary and Conclusion