🦜️🔗 LangChain Integration

TruLens provides TruChain, a deep integration with LangChain that allows you to inspect and evaluate the internals of your LangChain-built applications. This integration provides automatic instrumentation of key LangChain classes, enabling detailed tracking and evaluation without manual setup.

To see a list of classes instrumented, see Appendix: Instrumented LangChain Classes and Methods.

Instrumenting LangChain apps

To demonstrate usage, we'll create a standard RAG defined with LangChain Expression Language (LCEL).

First, this requires loading data into a vector store.

Create a RAG with LCEL

import bs4
from langchain_community.document_loaders import WebBaseLoader
from langchain_community.vectorstores import FAISS
from langchain_openai import OpenAIEmbeddings
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain import hub
from langchain_openai import ChatOpenAI
from langchain_core.output_parsers import StrOutputParser
from langchain_core.runnables import RunnablePassthrough

loader = WebBaseLoader(
    web_paths=("https://lilianweng.github.io/posts/2023-06-23-agent/",),
    bs_kwargs=dict(
        parse_only=bs4.SoupStrainer(
            class_=("post-content", "post-title", "post-header")
        )
    ),
)
docs = loader.load()
embeddings = OpenAIEmbeddings()
text_splitter = RecursiveCharacterTextSplitter()
documents = text_splitter.split_documents(docs)
vectordstore = FAISS.from_documents(documents, embeddings)

retriever = vectorstore.as_retriever()

prompt = hub.pull("rlm/rag-prompt")
llm = ChatOpenAI(model_name="gpt-3.5-turbo", temperature=0)

def format_docs(docs):
    return "\n\n".join(doc.page_content for doc in docs)

rag_chain = (
    {"context": retriever | format_docs, "question": RunnablePassthrough()}
    | prompt
    | llm
    | StrOutputParser()
)

Instrument with TruChain

from trulens.apps.langchain import TruChain

# instrument with TruChain
tru_recorder = TruChain(rag_chain)

Evaluating LangChain Apps

To properly evaluate LLM apps, we often need to point our evaluation at an internal step of our application, such as the retrieved context.

Evaluating retrieved context in LangChain

import numpy as np
from trulens.core import Metric, Selector
from trulens.providers.openai import OpenAI

provider = OpenAI()

f_context_relevance = Metric(
    implementation=provider.context_relevance,
    name="Context Relevance",
    selectors={
        "question": Selector.select_record_input(),
        "context": Selector.select_context(collect_list=False),
    },
    agg=np.mean,
)

You can find the full quickstart available here: LangChain Quickstart

Async Support

TruChain also provides async support for LangChain through the ainvoke method. This allows you to track and evaluate async and streaming LangChain applications.

Create an async chain with LCEL

from langchain.callbacks import AsyncIteratorCallbackHandler
from langchain.chains import LLMChain
from langchain.prompts import PromptTemplate
from langchain_openai import ChatOpenAI
from trulens.apps.langchain import TruChain

# Set up an async callback.
callback = AsyncIteratorCallbackHandler()

# Setup a simple question/answer chain with streaming ChatOpenAI.
prompt = PromptTemplate.from_template(
    "Honestly answer this question: {question}."
)
llm = ChatOpenAI(
    temperature=0.0,
    streaming=True,  # important
    callbacks=[callback],
)
async_chain = LLMChain(llm=llm, prompt=prompt)

Once you have created the async LLM chain you can instrument it just as before.

Instrument async apps with TruChain

async_tc_recorder = TruChain(async_chain)

with async_tc_recorder as recording:
    await async_chain.ainvoke(
        input=dict(question="What is 1+2? Explain your answer.")
    )

For examples of using TruChain, check out the TruLens Cookbook

Appendix: Instrumented LangChain Classes and Methods

The modules, classes, and methods that TruLens instruments can be retrieved from the appropriate Instrument subclass.

Instrumenting other classes/methods

Additional classes and methods can be instrumented by use of the trulens.core.otel.instrument methods and decorators.