RunnablePassthrough

Overview

RunnablePassthrough is a utility that facilitates unmodified data flow through a pipeline. Its invoke() method returns input data in its original form without alterations.

This functionality allows seamless data transmission between pipeline stages.

It frequently works in tandem with RunnableParallel for concurrent task execution, enabling the addition of new key-value pairs to the data stream.

Common use cases for RunnablePassthrough include:

  • Direct data forwarding without transformation

  • Pipeline stage bypassing

  • Pipeline flow validation during debugging

Table of Contents

References


Environment Setup

Set up the environment. You may refer to Environment Setup for more details.

[Note]

  • langchain-opentutorial is a package that provides a set of easy-to-use environment setup, useful functions and utilities for tutorials.

  • You can checkout the langchain-opentutorial for more details.

%%capture --no-stderr
%pip install langchain-opentutorial
# Install required packages
from langchain_opentutorial import package

package.install(
    [
        "langsmith",
        "langchain_openai",
        "langchain_core",
        "langchain-ollama",
        "langchain_community",
        "faiss-cpu",
    ],
    verbose=False,
    upgrade=False,
)

If you want to get automated tracing of your model calls you can also set your LangSmith API key by uncommenting below code:

# Set environment variables
from langchain_opentutorial import set_env

set_env(
    {
        "OPENAI_API_KEY": "",
        "LANGCHAIN_API_KEY": "",
        "LANGCHAIN_TRACING_V2": "true",
        "LANGCHAIN_ENDPOINT": "https://api.smith.langchain.com",
        "LANGCHAIN_PROJECT": "LangChain-Expression-Language",
    }
)

You can alternatively set API keys such as OPENAI_API_KEY in a .env file and load them.

[Note] This is not necessary if you've already set the required API keys in previous steps.

# Load API keys from .env file
from dotenv import load_dotenv

load_dotenv(override=True)
True

Passing Data with RunnablePassthrough and RunnableParallel

RunnablePassthrough is a utility that passes data through unchanged or adds minimal information before forwarding.

It commonly integrates with RunnableParallel to map data under new keys.

  • Standalone Usage

    When used independently, RunnablePassthrough() returns the input data unmodified.

  • Usage with assign

    When implemented with assign as RunnablePassthrough.assign(...), it augments the input data with additional fields before forwarding.

By leveraging RunnablePassthrough, you can maintain data integrity through pipeline stages while selectively adding required information.

Let me continue reviewing any additional content. I'm tracking all modifications to provide a comprehensive summary once the review is complete.

Example of Using RunnableParallel and RunnablePassthrough

While RunnablePassthrough is effective independently, it becomes more powerful when combined with RunnableParallel.

This section demonstrates how to configure and run parallel tasks using the RunnableParallel class. The following steps provide a beginner-friendly implementation guide.

  1. Initialize RunnableParallel

    Create a RunnableParallel instance to manage concurrent task execution.

  2. Configure passed Task

    • Define a passed task utilizing RunnablePassthrough

    • This task preserves input data without modification

  3. Set Up extra Task

    • Implement an extra task using RunnablePassthrough.assign()

    • This task computes triple the "num" value and stores it with key mult

  4. Implement modified Task

    • Create a modified task using a basic function

    • This function increments the "num" value by 1

  5. Task Execution

    • Invoke all tasks using runnable.invoke()

    • Example: Input {"num": 1} triggers concurrent execution of all defined tasks

from langchain_core.runnables import RunnableParallel, RunnablePassthrough

runnable = RunnableParallel(
    # Sets up a Runnable that returns the input as-is.
    passed=RunnablePassthrough(),
    # Sets up a Runnable that multiplies the "num" value in the input by 3 and returns the result.
    extra=RunnablePassthrough.assign(mult=lambda x: x["num"] * 3),
    # Sets up a Runnable that adds 1 to the "num" value in the input and returns the result.
    modified=lambda x: {"num": x["num"] + 1},
)

# Execute the Runnable with {"num": 1} as input.
result = runnable.invoke({"num": 1})

# Print the result.
print(result)
{'passed': {'num': 1}, 'extra': {'num': 1, 'mult': 3}, 'modified': 2}
r = RunnablePassthrough.assign(mult=lambda x: x["num"] * 3)
r.invoke({"num": 1})
{'num': 1, 'mult': 3}

Summary of Results

When provided with input {"num": 1}, each task produces the following output:

  1. passed: Returns unmodified input data

    • Output: {"num": 1}

  2. extra: Augments input with "mult" key containing triple the "num" value

    • Output: {"num": 1, "mult": 3}

  3. modified: Increments the "num" value by 1

    • Output: {"num": 2}

Search Engine Integration

The following example illustrates an implementation of RunnablePassthrough.

Using RunnablePassthrough in a FAISS-Based RAG Pipeline

This code uses RunnablePassthrough in a FAISS-based RAG pipeline to pass retrieved context into a chat prompt. It enables seamless integration of OpenAI embeddings for efficient retrieval and response generation.

from langchain_community.vectorstores import FAISS
from langchain_core.output_parsers import StrOutputParser
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.runnables import RunnablePassthrough
from langchain_openai import ChatOpenAI, OpenAIEmbeddings

# Create a FAISS vector store from text data.
vectorstore = FAISS.from_texts(
    [
        "Cats are geniuses at claiming boxes as their own.",
        "Dogs have successfully trained humans to take them for walks.",
        "Cats aren't fond of water, but the water in a human's cup is an exception.",
        "Dogs follow cats around, eager to befriend them.",
        "Cats consider laser pointers their arch-nemesis.",
    ],
    embedding=OpenAIEmbeddings(),
)

# Use the vector store as a retriever.
retriever = vectorstore.as_retriever()

# Define a template.
template = """Answer the question based only on the following context:
{context}

Question: {question}
"""

# Create a chat prompt from the template.
prompt = ChatPromptTemplate.from_template(template)
# Initialize the ChatOpenAI model.
model = ChatOpenAI(model_name="gpt-4o-mini")


# Function to format retrieved documents.
def format_docs(docs):
    return "\n".join([doc.page_content for doc in docs])


# Construct the retrieval chain.
retrieval_chain = (
    {"context": retriever | format_docs, "question": RunnablePassthrough()}
    | prompt
    | model
    | StrOutputParser()
)
# Query retrieval chain
retrieval_chain.invoke("What kind of objects do cats like?")
'Cats like boxes.'
retrieval_chain.invoke("What do dogs like?")
'Dogs like to befriend cats.'

Using Ollama

  • Download the application from the Ollama official website

  • For comprehensive Ollama documentation, visit the GitHub tutorial

  • Implementation utilizes the llama3.2 1b model for response generation and mxbai-embed-large for embedding operations

Ollama Installation Guide on Colab

Google Colab requires the colab-xterm extension for terminal functionality. Follow these steps to install Ollama:

  1. Install and Initialize colab-xterm

!pip install colab-xterm
%load_ext colabxterm
  1. Launch Terminal

%xterm
  1. Install Ollama

    Execute the following command in the terminal:

curl -fsSL https://ollama.com/install.sh | sh
  1. Installation Verification

    Verify installation by running:

ollama

Successful installation displays the "Available Commands" menu.

  1. Download and Prepare the Embedding Model for Ollama

!ollama pull mxbai-embed-large
from langchain_community.vectorstores import FAISS
from langchain_core.output_parsers import StrOutputParser
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.runnables import RunnablePassthrough
from langchain_ollama import OllamaEmbeddings

# Configure embeddings
ollama_embeddings = OllamaEmbeddings(model="mxbai-embed-large")

# Initialize FAISS vector store with text data
vectorstore = FAISS.from_texts(
    [
        "Cats are geniuses at claiming boxes as their own.",
        "Dogs have successfully trained humans to take them for walks.",
        "Cats aren't fond of water, but the water in a human's cup is an exception.",
        "Dogs follow cats around, eager to befriend them.",
        "Cats consider laser pointers their arch-nemesis.",
    ],
    embedding=ollama_embeddings(),
)
# Convert vector store to retriever
retriever = vectorstore.as_retriever()

# Define prompt template
template = """Answer the question based only on the following context:
{context}

Question: {question}
"""

# Initialize chat prompt from template
prompt = ChatPromptTemplate.from_template(template)
  1. Download and Prepare the Model for Answer Generation

!ollama pull llama3.2:1b
from langchain_ollama import ChatOllama

# Initialize Ollama chat model
ollama_model = ChatOllama(model="llama3.2:1b")


# Format retrieved documents
def format_docs(docs):
    return "\n".join([doc.page_content for doc in docs])


# Build retrieval chain
retrieval_chain = (
    {"context": retriever | format_docs, "question": RunnablePassthrough()}
    | prompt
    | ollama_model  # Use Ollama model for inference
    | StrOutputParser()
)
# Query retrieval chain
retrieval_chain.invoke("What kind of objects do cats like?")
'Based on this context, it seems that cats tend to enjoy and claim boxes as their own.'
# Query retrieval chain
retrieval_chain.invoke("What do dogs like?")
'Based on the context, it seems that dogs enjoy being around cats and having them follow them. Additionally, dogs have successfully trained humans to take them for walks.'

Last updated