Grounded Generation
Grounded generation is the final LLM call in a RAG pipeline that produces an answer conditioned on the retrieved context, ideally staying faithful to the provided documents and citing them.
What is Grounded Generation?
Grounded generation is the final LLM call in a RAG pipeline that produces an answer conditioned on the retrieved context, ideally staying faithful to the provided documents and citing them.
Grounded generation is the final LLM call in a RAG pipeline that produces an answer conditioned on the retrieved context, ideally staying faithful to the provided documents and citing them.
Where is it used?
Perplexity, ChatGPT file search, and enterprise RAG apps use grounded generation; guardrails like NeMo Guardrails and LlamaIndex response synthesizers enforce faithfulness.
How to build it
Call `llm.invoke(rag_prompt)` where `rag_prompt` includes retrieved context; add an instruction like `"Answer only using the context; if not found, say 'I don't know.'"` to reduce hallucination.