07. Runnable

# .env Read a file and set it as an environment variable
from dotenv import load_dotenv

# load token information
load_dotenv()

True

# Set up LangSmith tracking. https://smith.langchain.com
# !pip install -qU langchain-teddynote
from langchain_teddynote import logging

# Enter a project name.
logging.langsmith("CH01-Basic")

Start tracking LangSmith.
[project name]
CH01-Basic

How to effectively convey data

• RunnablePassthrough can be passed without changing the input or by adding additional keys.

• RunnablePassthrough() When called alone, it simply takes input and passes it on as is.

• RunnablePassthrough.assign(...) When called in this way, it takes input and adds any additional arguments passed to the assign function.

RunnablePassthrough

from langchain_core.prompts import PromptTemplate
from langchain_openai import ChatOpenAI


# Generate prompt and llm.
prompt = PromptTemplate.from_template("{num} 의 10배는?")
llm = ChatOpenAI(temperature=0)

# Create a chain.
chain = prompt | llm

chain 을 invoke() When executing, the input data type must be a dictionary.

# run the chain.
chain.invoke({"num": 5})

AIMessage(content='50입니다.', response_metadata={'token_usage': {'completion_tokens': 3, 'prompt_tokens': 16, 'total_tokens': 19}, 'model_name': 'gpt-3.5-turbo', 'system_fingerprint': None, 'finish_reason': 'stop', 'logprobs': None}, id='run-29242a8b-01c0-41ad-8f5b-7613e6876dc5-0', usage_metadata={'input_tokens': 16, 'output_tokens': 3, 'total_tokens': 19})

However, with the update of the langchain library, it is also possible to pass only the value if the template contains only one variable.

# run the chain.
chain.invoke(5)

AIMessage(content='50', response_metadata={'token_usage': {'completion_tokens': 1, 'prompt_tokens': 16, 'total_tokens': 17}, 'model_name': 'gpt-3.5-turbo', 'system_fingerprint': None, 'finish_reason': 'stop', 'logprobs': None}, id='run-44f606db-2437-4d1f-9cee-ddfc69428745-0', usage_metadata={'input_tokens': 16, 'output_tokens': 1, 'total_tokens': 17})

Below is an example using RunnablePassthrough.w

RunnablePassthrough is a runnable object, and a runnable object can be executed separately using the invoke() method.

from langchain_core.runnables import RunnablePassthrough

# runnable
RunnablePassthrough().invoke({"num": 10})

{'num': 10}

Below is an example of constructing a chain with RunnablePassthrough.

runnable_chain = {"num": RunnablePassthrough()} | prompt | ChatOpenAI()

# dict 값이 RunnablePassthrough() has been changed to.
runnable_chain.invoke(10)

AIMessage(content='100 no see.', response_metadata={'token_usage': {'completion_tokens': 3, 'prompt_tokens': 16, 'total_tokens': 19}, 'model_name': 'gpt-3.5-turbo', 'system_fingerprint': None, 'finish_reason': 'stop', 'logprobs': None}, id='run-66270ca2-4d62-4c71-859b-de6318f29909-0', usage_metadata={'input_tokens': 16, 'output_tokens': 3, 'total_tokens': 19})

다음은 RunnablePassthrough.assign() The results are compared with when using .

RunnablePassthrough().invoke({"num": 1})

{'num': 1}

RunnablePassthrough.assign()

• Merges the key/value pairs of the values ​​passed as input values ​​with the newly assigned key/value pairs.

# Input key: num, Assign key: new_num
(RunnablePassthrough.assign(new_num=lambda x: x["num"] * 3)).invoke({"num": 1})

{'num': 1, 'new_num': 3}

RunnableParallel

from langchain_core.runnables import RunnableParallel

# Creates a RunnableParallel instance, which can run multiple Runnable instances in parallel.
runnable = RunnableParallel(
    # Pass a RunnablePassthrough instance as the 'passed' keyword argument. This will pass the input data through as is.
    passed=RunnablePassthrough(),
    # We use RunnablePassthrough.assign with the 'extra' keyword argument to assign the 'mult' lambda function, which will triple the value corresponding to the 'num' key in the input dictionary.
    extra=RunnablePassthrough.assign(mult=lambda x: x["num"] * 3),
    # Pass a lambda function as the 'modified' keyword argument. This function adds 1 to the value corresponding to the 'num' key in the input dictionary.
    modified=lambda x: x["num"] + 1,
)

# On a runnable instance {'num': 1} Passing a dictionary as input invoke Call a method.
runnable.invoke({"num": 1})

{'passed': {'num': 1}, 'extra': {'num': 1, 'mult': 3}, 'modified': 2}

Chain can also apply RunnableParallel.

chain1 = (
    {"country": RunnablePassthrough()}
    | PromptTemplate.from_template("{country} The capital of?")
    | ChatOpenAI()
)
chain2 = (
    {"country": RunnablePassthrough()}
    | PromptTemplate.from_template("{country} The area of?")
    | ChatOpenAI()
)

combined_chain = RunnableParallel(capital=chain1, area=chain2)
combined_chain.invoke("korea")

{'capital': AIMessage(content='It's Seoul.', response_metadata={'token_usage': {'completion_tokens': 5, 'prompt_tokens': 19, 'total_tokens': 24}, 'model_name': 'gpt-3.5-turbo', 'system_fingerprint': None, 'finish_reason': 'stop', 'logprobs': None}, id='run-d9324c24-9670-4430-97d6-1272f5dbe0f2-0', usage_metadata={'input_tokens': 19, 'output_tokens': 5, 'total_tokens': 24}), 'area': AIMessage(content='대한민국의 총 면적은 약 100,363 km²입니다.', response_metadata={'token_usage': {'completion_tokens': 24, 'prompt_tokens': 20, 'total_tokens': 44}, 'model_name': 'gpt-3.5-turbo', 'system_fingerprint': None, 'finish_reason': 'stop', 'logprobs': None}, id='run-f27442a3-fc9c-4d08-9fdf-189c1b4585c8-0', usage_metadata={'input_tokens': 20, 'output_tokens': 24, 'total_tokens': 44})}

RunnableLambda

You can map your own custom functions using RunnableLambda.

from langchain_core.output_parsers import StrOutputParser
from langchain_core.prompts import PromptTemplate
from langchain_openai import ChatOpenAI
from datetime import datetime


def get_today(a):
    # get today's date
    return datetime.today().strftime("%b-%d")


# print today's date
get_today(None)

'Jun-19'

from langchain_core.runnables import RunnableLambda, RunnablePassthrough

# Generate prompt and llm.
prompt = PromptTemplate.from_template(
    "{today} List {n} famous people whose birthdays are: Please indicate their dates of birth."
)
llm = ChatOpenAI(temperature=0, model_name="gpt-4o")

# Create a chain.
chain = (
    {"today": RunnableLambda(get_today), "n": RunnablePassthrough()}
    | prompt
    | llm
    | StrOutputParser()
)

# output of power
print(chain.invoke(3))

Here are some famous people whose birthday is on June 19th:

1. Paul Dano - June 19, 1984
2. Direk Novak Djokovic - June 19, 1987
3. Philippe Coutinho - 19 June 1992

They are active in various fields as actors, tennis players, and soccer players respectively.

Use itemgetter to extract a specific key.

from operator import itemgetter

from langchain_core.prompts import ChatPromptTemplate
from langchain_core.runnables import RunnableLambda
from langchain_openai import ChatOpenAI


# A function that returns the length of a sentence.
def length_function(text):
    return len(text)


# A function that returns the product of the lengths of two sentences.
def _multiple_length_function(text1, text2):
    return len(text1) * len(text2)


# This function returns the product of the lengths of two sentences using the _multiple_length_function function.
def multiple_length_function(_dict):
    return _multiple_length_function(_dict["text1"], _dict["text2"])


prompt = ChatPromptTemplate.from_template("{a} + {b} What is?")
model = ChatOpenAI()

chain1 = prompt | model

chain = (
    {
        "a": itemgetter("word1") | RunnableLambda(length_function),
        "b": {"text1": itemgetter("word1"), "text2": itemgetter("word2")}
        | RunnableLambda(multiple_length_function),
    }
    | prompt
    | model
)

chain.invoke({"word1": "hello", "word2": "world"})

AIMessage(content='5 + 25 = It's 30.', response_metadata={'token_usage': {'completion_tokens': 9, 'prompt_tokens': 22, 'total_tokens': 31}, 'model_name': 'gpt-3.5-turbo', 'system_fingerprint': None, 'finish_reason': 'stop', 'logprobs': None}, id='run-5db9b475-09ee-4edb-af9d-b37b320bee1e-0', usage_metadata={'input_tokens': 22, 'output_tokens': 9, 'total_tokens': 31})

Last edited by : Sept. 5, 2024, 9:18 a.m.