mquake-question-generation-model

Summary

The question-generation step in the MQuAKE pipeline uses GPT-3.5-turbo-instruct to produce exactly three varied multi-hop questions from each fact chain, which sets a fixed output contract that downstream stages (like filtering or evaluation) can rely on. In practice, this means the system's question volume and diversity are bounded by this model's behavior, so any change to the prompt or model would ripple through the entire pipeline.