Description
Domain fine-tuning needs quality QA data that is slow and costly to label by hand. GraphGen builds a knowledge graph from your corpus, finds what the model knows poorly and synthesizes targeted QA data.
Output plugs straight into LLaMA-Factory, XTuner and other SFT frameworks, with a full cookbook.
Knowledge graph:Entities and relations.
Weak spots:Targets model gaps.
QA types:Atomic, aggregated and multi-hop.
Training ready:Mainstream frameworks.
Output plugs straight into LLaMA-Factory, XTuner and other SFT frameworks, with a full cookbook.
Features
Knowledge graph:Entities and relations.
Weak spots:Targets model gaps.
QA types:Atomic, aggregated and multi-hop.
Training ready:Mainstream frameworks.
