NvidiaLaunchedChatQAIt is said that the performance of the model can be compared to the BiaoGPT-4, using efficient training methods such as two-stage instruction tuning and improved contextual retrieval.
ChatQA is a set of conversational Question and Answer (QA) models that can achieve GPT-4 level accuracy. Specifically, the development team proposes a two-stage instruction tuning approach that significantly improves zero-sample conversational QA results for large language models (LLMs).
In order to handle retrieval in conversational QA, a dense searcher was fine-tuned on a multi-round QA dataset, which provides a different approach than using theFirstinto the query rewriting model equivalent results, while significantly reducing deployment costs. Notably, ChatQA-70B outperforms GPT-4 in terms of average scores on 10 conversational QA datasets (54.14 vs. 53.90) without relying on any synthetic data from OpenAI GPT models.