Creating a RAG chain

Now to bring all the components together in your RAG workflow! You've prepared the documents and ingested them into a Chroma database for retrieval. You created a prompt template to include the retrieved chunks from the academic paper and answer questions.

The prompt template you created in the previous exercise is available as prompt_template, an OpenAI model has been initialized as llm, and the code to recreate your retriever has be included in the script.

This exercise is part of the course

Developing LLM Applications with LangChain

Exercise instructions

Create an LCEL chain to link retriever, prompt_template, and llm so the model can retrieve the documents.
Invoke the chain on the 'question' provided.

Hands-on interactive exercise

Have a go at this exercise by completing this sample code.

vectorstore = Chroma.from_documents(
    docs,
    embedding=OpenAIEmbeddings(api_key='', model='text-embedding-3-small'),
    persist_directory=os.getcwd()
)

retriever = vectorstore.as_retriever(
    search_type="similarity",
    search_kwargs={"k": 3}
)

# Create a chain to link retriever, prompt_template, and llm
rag_chain = ({"context": ____, "question": ____}
            | ____
            | ____)

# Invoke the chain
response = ____("Which popular LLMs were considered in the paper?")
print(response.content)

Edit and Run Code

This exercise is part of the course

Developing LLM Applications with LangChain

IntermediateSkill Level

4.8+

Start Course for Free

Welcome to the LangChain framework for building applications on LLMs! You'll learn about the main components of LangChain, including models, chains, agents, prompts, and parsers. You'll create chatbots using both open-source models from Hugging Face and proprietary models from OpenAI, create prompt templates, and integrate different chatbot memory strategies to manage context and resources during conversations.

Exercise 1: The LangChain ecosystem Exercise 2: OpenAI models in LangChain!Exercise 3: Hugging Face models in LangChain!Exercise 4: Prompt templates Exercise 5: Prompt templates and chaining Exercise 6: Chat prompt templates Exercise 7: Few-shot prompting Exercise 8: Creating the few-shot example set Exercise 9: Building the few-shot prompt template Exercise 10: Implementing few-shot prompting

Time to level up your LangChain chains! You'll learn to use the LangChain Expression Language (LCEL) for defining chains with greater flexibility. You'll create sequential chains, where inputs are passed between components to create more advanced applications. You'll also begin to integrate agents, which use LLMs for decision-making.

Exercise 1: Sequential chains Exercise 2: Building prompts for sequential chains Exercise 3: Sequential chains with LCEL Exercise 4: Introduction to LangChain agents Exercise 5: What's an agent?Exercise 6: ReAct agents Exercise 7: Custom tools for agents Exercise 8: Defining a function for tool use Exercise 9: Creating custom tools Exercise 10: Integrating custom tools with agents

One limitation of LLMs is that they have a knowledge cut-off due to being trained on data up to a certain point. In this chapter, you'll learn to create applications that use Retrieval Augmented Generation (RAG) to integrate external data with LLMs. The RAG workflow contains a few different processes, including splitting data, creating and storing the embeddings using a vector database, and retrieving the most relevant information for use in the application. You'll learn to master the entire workflow!

Exercise 1: Integrating document loaders Exercise 2: PDF document loaders Exercise 3: CSV document loaders Exercise 4: HTML document loaders Exercise 5: Splitting external data for retrieval Exercise 6: Splitting by character Exercise 7: Recursively splitting by character Exercise 8: Splitting HTML Exercise 9: RAG storage and retrieval using vector databases Exercise 10: Preparing the documents and vector database Exercise 11: Building a retrieval prompt template Exercise 12: Creating a RAG chain

Current Exercise

Exercise 13: Wrap-up!