Prompt - ChatPromptTemplate

 Unlike prompts to llms that are strings, prompts to chat models must be a list of messages. We will assume the model to be considered as OpenAIs chat completion API and proceed.

Types of Messages

Messages can be associated with 3 types of roles. The roles are

1. AI assistant

 - this is the response for each prompt from the user

2. System

- this sets the overall environment for the conversation. Eg. You are a salesperson

3. Human

- this is the prompt given by the user

With each role a prompt may be associated and is referred to as "content".

Using Chat Models without templating - BaseMessage

It is possible to use chat models without templating as follows:

!pip install langchain-openai langchain

from langchain_openai import ChatOpenAI

chat = ChatOpenAI(
    temperature=0,
    openai_api_key=openai_api_key
)

from langchain_core.messages import SystemMessage, HumanMessage

messages = [
    SystemMessage(
        content="You are a helpful assistant that translates English to French."
    ),
    HumanMessage(
        content="Translate this sentence from English to French. Hello"
    ),
]

response = chat.invoke(messages)

ChatPromptTemplate

This is class used to template prompts for chat models. The 2 basic steps to follow are

1.  Create a ChatPromptTemplate object using from_messages() and pass the prompt associated with each role.

2. Format the messages as a list of messages to be passed to the chat model using format_messages().

Using from_messages()

1. List of 2-tuple

In the first type we pass a list [] of tuples () with each tuple having the pair (role,content) where both role and content are strings.

from langchain_core.prompts import ChatPromptTemplate

chat_template = ChatPromptTemplate.from_messages(
    [
        ("system","You are a bot created to act as customer support"),
        ("human", "Hello, I need some help"),
        ("ai","Hi, How can I help you?")
    ]
)

2. Using from_template()

from langchain_core.prompts import ChatPromptTemplate, HumanMessagePromptTemplate, SystemMessagePromptTemplate
from langchain_core.messages import SystemMessage

chat_template = ChatPromptTemplate.from_messages(
    [
      SystemMessagePromptTemplate.from_template("You write technical blogs"),
      HumanMessagePromptTemplate.from_template("Give some suggestions for latest in {topic}")
    ]
)

3. Using BaseMessage

from langchain_core.prompts import ChatPromptTemplate
from langchain_core.messages import SystemMessage, HumanMessage

chat_template = ChatPromptTemplate.from_messages(
    [
      SystemMessage(content = "You write technical blogs"), # String content of message
      HumanMessage(content="Give some suggestions for latest in {topic}")
    ]
)
prompt = chat_template.format_messages(topic="GAI")
print(prompt)

4. Combining BaseMessage and MessagePromptTemplate

from langchain_core.prompts import ChatPromptTemplate, HumanMessagePromptTemplate
from langchain_core.messages import SystemMessage

chat_template = ChatPromptTemplate.from_messages(
    [
      SystemMessage(content = "You write technical blogs"), # String content of message
      HumanMessagePromptTemplate.from_template("Give some suggestions for
                                                    latest in {topic}")
    ]
)
prompt = chat_template.format_messages(topic="GAI")
print(prompt)


Using format_messages()

This function is called with an object of ChatPromptTemplate and the input variables are passed as arguments.

chat_prompt = chat_template.format_messages(topic="GAI", content_type = "technical")
print(chat_prompt)

Using invoke()

PromptTemplate  and ChatPromptTemplate implement the Runnable interface. This means we can call invoke() and other LCEL methods. 

The input parameter for ChatPromptTemplate is a dictionary of the input variables and the output is ChatPromptValue.

The following code shows how to get the final response with and without using ChatPromptValue.


from langchain_core.prompts import ChatPromptTemplate
from langchain_core.messages import SystemMessage, HumanMessage

chat_template = ChatPromptTemplate.from_messages(
    [
      SystemMessage(content = "You write {content_type} blogs"), # String content of message
      HumanMessage(content="Give some suggestions for latest in {topic}")
    ]
)
chat_prompt = chat_template.format_messages(topic="GAI", content_type = "technical")
print(chat_prompt)

chat_prompt_val = chat_template.invoke({"content_type": "technical", "topic": "GAI"})

from langchain_openai import ChatOpenAI

chat = ChatOpenAI(
    temperature=0,
    openai_api_key=openai_api_key,
    max_tokens = 20
)

# Invoke with prompt messages directly
chat_response = chat.invoke(chat_prompt)
print(chat_response)

# Invoke with ChatPromptValue
chat_response = chat.invoke(chat_prompt_val)
print(chat_response)

Comments

Post a Comment

Popular posts from this blog

Retrieval 1 - Steps to Perform RAG

Image
As we seen the below given link, LLMs are trained with large amount of data.  https://gaillms.blogspot.com/2024/01/training-llm-dataset.html But this data does not include proprietary data and data not available for free on the internet. When we give prompts to an LLM, it replies based on the data in which it was trained. The problem arises when we want answers for data that the LLM was not trained on. Two options are generally considered for this purpose. 1. Model Fine Tuning 2. Retrieval Augmented Generation(RAG) Based on our requirement we can decide between the 2. Our topic of discussion is RAG. Retrieval It is to retrieve external data like pdf, html, videos etc. Generation It refers to generating output with the external data retrieved. The steps used in RAG are as follows: Why we need RAG? Consider a pdf document with 1000 pages about my company. This is the context within which we pose questions to the llm. That is the prompt expects answers based on the document.  For...
Read more

Case Study : Chatbot with Streamlit

Streamlit It is an open-source Python library to change python scripts(written with streamlit library) into web apps using machine learning and data science. Install streamlit pip install streamlit Note : Write as .py file Import streamlit import streamlit as st Maintain Conversation State The following statement helps to keep track of the messages exchanged during a session. from langchain_community.chat_message_histories                                                        import StreamlitChatMessageHistory msgs = StreamlitChatMessageHistory() print(msgs) Output: {'_messages': []} Step 1 : Add the First AI Message to chat history # Give initial ai message from Chatbot if  len (msgs.messages) ==  0 :      msgs.add_ai_message( "How can I help you?" )      print (msgs) Output {'_mess...
Read more