Langchain summarize

Summarization¶

In this example, you will learn how to create a summarization app and evaluate + track it in TruLens

Import libraries¶

In [ ]:

# !pip install trulens trulens-apps-langchain trulens-providers-openai langchain==0.0.283 langchain_community

# !pip install trulens trulens-apps-langchain trulens-providers-openai langchain==0.0.283 langchain_community

In [ ]:

from langchain.chains.summarize import load_summarize_chain
from langchain.text_splitter import RecursiveCharacterTextSplitter
from trulens.apps.langchain import Feedback
from trulens.apps.langchain import FeedbackMode
from trulens.apps.langchain import Query
from trulens.apps.langchain import TruSession
from trulens.apps.langchain import TruChain
from trulens.providers.openai import OpenAI

session = TruSession()

from langchain.chains.summarize import load_summarize_chain from langchain.text_splitter import RecursiveCharacterTextSplitter from trulens.apps.langchain import Feedback from trulens.apps.langchain import FeedbackMode from trulens.apps.langchain import Query from trulens.apps.langchain import TruSession from trulens.apps.langchain import TruChain from trulens.providers.openai import OpenAI session = TruSession()

Set API Keys¶

For this example, we need API keys for the Huggingface and OpenAI

In [ ]:

import os

os.environ["OPENAI_API_KEY"] = "..."
os.environ["HUGGINGFACE_API_KEY"] = "..."

import os os.environ["OPENAI_API_KEY"] = "..." os.environ["HUGGINGFACE_API_KEY"] = "..."

In [ ]:

provider = OpenAI()

# Define a moderation feedback function using HuggingFace.
mod_not_hate = Feedback(provider.moderation_not_hate).on(
    text=Query.RecordInput[:].page_content
)


def wrap_chain_trulens(chain):
    return TruChain(
        chain,
        app_name="ChainOAI",
        feedbacks=[mod_not_hate],
        feedback_mode=FeedbackMode.WITH_APP,  # calls to TruChain will block until feedback is done evaluating
    )


def get_summary_model(text):
    """
    Produce summary chain, given input text.
    """

    llm = OpenAI(temperature=0, openai_api_key="")
    text_splitter = RecursiveCharacterTextSplitter(
        separators=["\n\n", "\n", " "], chunk_size=8000, chunk_overlap=350
    )
    docs = text_splitter.create_documents([text])
    print(f"You now have {len(docs)} docs instead of 1 piece of text.")

    return docs, load_summarize_chain(llm=llm, chain_type="map_reduce")

provider = OpenAI() # Define a moderation feedback function using HuggingFace. mod_not_hate = Feedback(provider.moderation_not_hate).on( text=Query.RecordInput[:].page_content ) def wrap_chain_trulens(chain): return TruChain( chain, app_name="ChainOAI", feedbacks=[mod_not_hate], feedback_mode=FeedbackMode.WITH_APP, # calls to TruChain will block until feedback is done evaluating ) def get_summary_model(text): """ Produce summary chain, given input text. """ llm = OpenAI(temperature=0, openai_api_key="") text_splitter = RecursiveCharacterTextSplitter( separators=["\n\n", "\n", " "], chunk_size=8000, chunk_overlap=350 ) docs = text_splitter.create_documents([text]) print(f"You now have {len(docs)} docs instead of 1 piece of text.") return docs, load_summarize_chain(llm=llm, chain_type="map_reduce")

In [ ]:

from datasets import load_dataset

billsum = load_dataset("billsum", split="ca_test")
text = billsum["text"][0]

docs, chain = get_summary_model(text)

# use wrapped chain as context manager
with wrap_chain_trulens(chain) as recording:
    chain(docs)

from datasets import load_dataset billsum = load_dataset("billsum", split="ca_test") text = billsum["text"][0] docs, chain = get_summary_model(text) # use wrapped chain as context manager with wrap_chain_trulens(chain) as recording: chain(docs)

Run the TruLens dashboard¶

In [ ]:

from trulens.dashboard import run_dashboard

run_dashboard(session)

from trulens.dashboard import run_dashboard run_dashboard(session)