PyDialogue
An open-source library that generates continual dialogue data.
The Library
The library can be used to generate large-scale dialogue data to evaluate and improve conversational models. By mixing and matching different dialogues within a single chat session, it can create unique and complex contexts previously unseen by the model.
Heidi-3 Benchmark
Dialogues designed to challenge LLMs. They test whether your model can do in-context continual learning, social learning, and multi-hop reasoning while it actively participates in the context.
Templates Gallery
Browse handcrafted templates from different domains. We currently have 5 templates from the reasoning domain.