LC - Cookbook - Routing by semantic similarity

https://python.langchain.com/docs/expression_language/cookbook/embedding_router

Routing by semantic similarity | 🦜️🔗 Langchain

 

 

Routing by semantic similarity

 

With LCEL you can easily add custom routing logic to your chain to dynamically determine the chain logic based on user input. All you need to do is define a function that given an input returns a Runnable.

 

LCEL을 사용하면 사용자 정의 라우팅 논리를 체인에 쉽게 추가하여 사용자 입력을 기반으로 체인 논리를 동적으로 결정할 수 있습니다. 당신이 해야 할 일은 주어진 입력이 Runnable을 반환하는 함수를 정의하는 것뿐입니다.

 

One especially useful technique is to use embeddings to route a query to the most relevant prompt. Here's a very simple example.

 

특히 유용한 기술 중 하나는 임베딩을 사용하여 쿼리를 가장 관련성이 높은 프롬프트로 라우팅하는 것입니다. 다음은 매우 간단한 예입니다.

 

from langchain.chat_models import ChatOpenAI
from langchain.embeddings import OpenAIEmbeddings
from langchain.prompts import PromptTemplate
from langchain.schema.output_parser import StrOutputParser
from langchain.schema.runnable import RunnableLambda, RunnablePassthrough
from langchain.utils.math import cosine_similarity


physics_template = """You are a very smart physics professor. \
You are great at answering questions about physics in a concise and easy to understand manner. \
When you don't know the answer to a question you admit that you don't know.

Here is a question:
{query}"""

math_template = """You are a very good mathematician. You are great at answering math questions. \
You are so good because you are able to break down hard problems into their component parts, \
answer the component parts, and then put them together to answer the broader question.

Here is a question:
{query}"""

embeddings = OpenAIEmbeddings()
prompt_templates = [physics_template, math_template]
prompt_embeddings = embeddings.embed_documents(prompt_templates)


def prompt_router(input):
    query_embedding = embeddings.embed_query(input["query"])
    similarity = cosine_similarity([query_embedding], prompt_embeddings)[0]
    most_similar = prompt_templates[similarity.argmax()]
    print("Using MATH" if most_similar == math_template else "Using PHYSICS")
    return PromptTemplate.from_template(most_similar)


chain = (
    {"query": RunnablePassthrough()}
    | RunnableLambda(prompt_router)
    | ChatOpenAI()
    | StrOutputParser()
)

 

 

이 코드는 LangChain에서 사용자의 질문을 처리하고 적절한 모델 및 프롬프트를 선택하는 프로세스를 구현합니다.

1. physics_template 및 math_template은 각각 물리학 교수와 수학자 역할을 하는 모델의 프롬프트 템플릿입니다. 이러한 템플릿은 해당 주제에 대한 역할과 동작 방식을 정의합니다.
2. embeddings 객체는 OpenAIEmbeddings를 사용하여 프롬프트 템플릿을 텍스트 임베딩으로 변환합니다. 이러한 임베딩을 사용하여 사용자의 질문과 프롬프트 템플릿 간의 유사성을 계산할 수 있습니다.
3. prompt_router 함수는 사용자의 입력 질문과 프롬프트 템플릿 간의 유사성을 계산하여 가장 유사한 템플릿을 선택합니다. 유사성은 코사인 유사도를 기반으로 계산됩니다. 그런 다음, 가장 유사한 프롬프트를 선택하고 해당 프롬프트를 반환합니다.
4. chain은 모델 선택 및 프롬프트 라우팅을 처리하는 체인입니다. 사용자의 질문이 체인을 통과하면 prompt_router 함수를 사용하여 가장 유사한 프롬프트를 선택하고 해당 프롬프트를 사용하여 ChatOpenAI 모델을 호출합니다. 마지막으로, StrOutputParser를 사용하여 모델의 출력을 처리하고 반환합니다.

이 프로세스를 통해 사용자의 질문에 대해 물리학 또는 수학과 관련된 가장 적절한 역할 및 프롬프트를 선택하고, 해당 주제에 대한 답변을 생성합니다.

 

print(chain.invoke("What's a black hole"))

 

    Using PHYSICS
    A black hole is a region in space where gravity is extremely strong, so strong that nothing, not even light, can escape its gravitational pull. It is formed when a massive star collapses under its own gravity during a supernova explosion. The collapse causes an incredibly dense mass to be concentrated in a small volume, creating a gravitational field that is so intense that it warps space and time. Black holes have a boundary called the event horizon, which marks the point of no return for anything that gets too close. Beyond the event horizon, the gravitational pull is so strong that even light cannot escape, hence the name "black hole." While we have a good understanding of black holes, there is still much to learn, especially about what happens inside them.

 

print(chain.invoke("What's a path integral"))

 

    Using MATH
    Thank you for your kind words! I will do my best to break down the concept of a path integral for you.
    
    In mathematics and physics, a path integral is a mathematical tool used to calculate the probability amplitude or wave function of a particle or system of particles. It was introduced by Richard Feynman and is an integral over all possible paths that a particle can take to go from an initial state to a final state.
    
    To understand the concept better, let's consider an example. Suppose we have a particle moving from point A to point B in space. Classically, we would describe this particle's motion using a definite trajectory, but in quantum mechanics, particles can simultaneously take multiple paths from A to B.
    
    The path integral formalism considers all possible paths that the particle could take and assigns a probability amplitude to each path. These probability amplitudes are then added up, taking into account the interference effects between different paths.
    
    To calculate a path integral, we need to define an action, which is a mathematical function that describes the behavior of the system. The action is usually expressed in terms of the particle's position, velocity, and time.
    
    Once we have the action, we can write down the path integral as an integral over all possible paths. Each path is weighted by a factor determined by the action and the principle of least action, which states that a particle takes a path that minimizes the action.
    
    Mathematically, the path integral is expressed as:
    
    ∫ e^(iS/ħ) D[x(t)]
    
    Here, S is the action, ħ is the reduced Planck's constant, and D[x(t)] represents the integration over all possible paths x(t) of the particle.
    
    By evaluating this integral, we can obtain the probability amplitude for the particle to go from the initial state to the final state. The absolute square of this amplitude gives us the probability of finding the particle in a particular state.
    
    Path integrals have proven to be a powerful tool in various areas of physics, including quantum mechanics, quantum field theory, and statistical mechanics. They allow us to study complex systems and calculate probabilities that are difficult to obtain using other methods.
    
    I hope this explanation helps you understand the concept of a path integral. If you have any further questions, feel free to ask!

 

이 예제는 사전에 물리학자와 수학자 둘을 각각 설정해 놓고 질문을 그것을 OpenAI embedding을 통해서 임베딩 값을 받아 놓는 일 부터 시작합니다. 그 다음 질문을 던지면 그 질문도 임베딩 값을 받아서 이 질문이 수학과 관련이 있는지 아니면 물리학과 관련이 있는지를 판단하고 좀 더 가까운 쪽 설정을 가져와서 OpenAI에 던질 프롬프트를 만들어서 전달하고 거기에 대한 답을 받는 일을 합니다.

 

질문에 어떤 설정들을 세팅해서 프롬프트를 만들어야 할 때 유용하게 사용할 수 있는 기능인 것 같습니다.

로컬에서 따로 실행한 결과는 아래와 같습니다.