RAG Rerank And Answer
A 3-step recipe with NVIDIA NIM. We can help you get access.
Recipe · 3 steps
RAG Rerank And Answer
with NVIDIA NIM · Artificial Intelligence Management
Embed a query, rerank candidate passages, and answer grounded in the best one.
- Generate a dense embedding vector for the query using a NeMo Retriever embedding model with the asymmetric query input type.
- Score every candidate passage against the query with a cross-encoder reranker so the most relevant passage can be selected.
- Compose a grounded answer by giving the chat model the highest-ranked passage as system context alongside the original query.
What you'll need to have ready
- apiKey: NVIDIA developer API key (nvapi-...) sent as a Bearer token.
- query: The user's natural-language question.
- passages: Candidate passage objects ({ text }) to rerank against the query.
- embeddingModel: NeMo Retriever embedding model id.
- rerankModel: NeMo Retriever reranker model id.
- chatModel: LLM model id used to compose the grounded answer.
For your developer
Arazzo workflow nvidia-nim:rag-rerank-answer, 3 operations: createEmbedding → rankPassages → createChatCompletion. Specs on APIs.io.