Skip to content

ChatQA

Conversational QA models tuned in two stages

Visit official site

What chatqa is

The models are a family of conversational question answering models. A two-stage instruction tuning method improves zero-shot conversational QA, with a dense retriever tuned for dialogue retrieval.

What you can do with it

  • Build an assistant with natural dialogue
  • Retrieve context for multi-turn questions
  • Compare results across QA datasets

Key features

  • Two-stage instruction tuning
  • Fine-tuned dense retrieval
  • Lower deployment cost for QA

Pros & cons

✓ What we like

  • Retrieval results are comparable to query rewrite models
  • No synthetic data from OpenAI GPT models is used

! What to watch out for

  • The two-stage method is described for conversational QA
  • A dense retriever is fine-tuned on multi-turn QA datasets

FAQ

What method is proposed?

A two-stage instruction tuning method.

Which variant is named with a score?

ChatQA-70B.

Last reviewed: 2026-11-15

More AI chatbot tools

View all →

How we review