Self-RAG
Model decides for itself when to retrieve and critiques what it retrieved.
Resumen
Self-RAG trains a model to emit reflection tokens: whether retrieval is needed, whether a passage is relevant, whether the answer is supported by it. It solves the problem that always retrieving wastes work on questions the model can answer and does not check whether what came back was useful. It exists because retrieval quality is a decision, not a fixed step.
Grafo de tecnologías
Cómo se conecta Self-RAG con el resto del stack.
Usa y depende de
- RAG technique Techniques Self-RAG based on RAG High ✓ source