Combining RAG with other LLM prompting techniques
We can enhance RAG by combining it with other prompting techniques, such as CoT (see Chapter 20) or few-shot learning. Here’s an example that combines RAG with CoT:
class RAGWithCoT:
def __init__(self, retriever, generator):
self.retriever = retriever
self.generator = generator
self.tokenizer = AutoTokenizer.from_pretrained(generator)
def retrieve_and_generate(self, query: str) -> str:
retrieved_docs = self.retriever.retrieve(query, k=3)
context = " ".join(retrieved_docs)
cot_prompt = f"""Context: {context}
Question: {query}
Let's approach this step-by-step:
1...
