langchain/docs/modules/chains/examples/llm_requests.ipynb

LLMRequestsChain

Using the request library to get HTML results from a URL and then an LLM to parse results

In [1]:

from langchain.llms import OpenAI
from langchain.chains import LLMRequestsChain, LLMChain

In [2]:

from langchain.prompts import PromptTemplate

template = """Between >>> and <<< are the raw search result text from google.
Extract the answer to the question '{query}' or say "not found" if the information is not contained.
Use the format
Extracted:<answer or "not found">
>>> {requests_result} <<<
Extracted:"""

PROMPT = PromptTemplate(
    input_variables=["query", "requests_result"],
    template=template,
)

In [3]:

chain = LLMRequestsChain(llm_chain=LLMChain(llm=OpenAI(temperature=0), prompt=PROMPT))

In [4]:

question = "What are the Three (3) biggest countries, and their respective sizes?"
inputs = {
    "query": question,
    "url": "https://www.google.com/search?q=" + question.replace(" ", "+"),
}

In [5]:

chain(inputs)

Out [5]:

{'query': 'What are the Three (3) biggest countries, and their respective sizes?',
 'url': 'https://www.google.com/search?q=What+are+the+Three+(3)+biggest+countries,+and+their+respective+sizes?',
 'output': ' Russia (17,098,242 km²), Canada (9,984,670 km²), United States (9,826,675 km²)'}

In [ ]: