SystemMessagePromptTemplate with dynamic few shot examples

[donatoaz]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

No example

Description

Hi!

I am wondering if there is a way to have a few shot prompt template as the system message in a chat model. I know I can use the FewShotChatMessagePromptTemplate however this seems to render examples as human/ai human/tool messages, and not a system message.

So, my question is how to have a system message that has a dynamic example selection few shot prompt template within.

example:

prefix = “You are a helpful yada yada yada… here are some examples”
…
example_prompt = PromptTemplate.from… (“objection: {objection}, strategy: {strategy}”)
…
system_few_shot_prompt_template = ???(example_selector=selector, prefix=prefix, example_prompt=example_prompt, …)

final_prompt = ChatPromptTemplate.from_messages([
   system_few_shot_prompt_template,
   MessagesPlaceholder…
])

System Info

Langchain core 0.2.6

[dosubot[bot]]

Hello, @donatoaz! I'm here to assist you with any questions or issues you may have. Let's work together to resolve this matter.

To create a system message that includes a dynamic example selection few-shot prompt template within a chat model using LangChain, you can follow these steps:

1. Import the required modules:

   from langchain_core.prompts import SemanticSimilarityExampleSelector
   from langchain_core.embeddings import OpenAIEmbeddings
   from langchain_core.vectorstores import Chroma
   from langchain_core import SystemMessage
   from langchain_core.prompts import HumanMessagePromptTemplate, AIMessagePromptTemplate, SystemMessagePromptTemplate
   from langchain_core.prompts.few_shot import FewShotChatMessagePromptTemplate

2. Prepare the examples and create the example selector:

   examples = [
       {"input": "2 🦜 2", "output": "4"},
       {"input": "2 🦜 3", "output": "5"},
       {"input": "2 🦜 4", "output": "6"},
       {"input": "What did the cow say to the moon?", "output": "nothing at all"},
       {
           "input": "Write me a poem about the moon",
           "output": "One for the moon, and one for me, who are we to talk about the moon?",
       },
   ]
   
   to_vectorize = [" ".join(example.values()) for example in examples]
   embeddings = OpenAIEmbeddings()
   vectorstore = Chroma.from_texts(to_vectorize, embeddings, metadatas=examples)
   example_selector = SemanticSimilarityExampleSelector(vectorstore=vectorstore, k=2)

3. Create the few-shot prompt template:

   few_shot_prompt = FewShotChatMessagePromptTemplate(
       input_variables=["input"],
       example_selector=example_selector,
       example_prompt=(
           HumanMessagePromptTemplate.from_template("{input}")
           + AIMessagePromptTemplate.from_template("{output}")
       ),
   )

4. Define the overall prompt:

   final_prompt = (
       SystemMessagePromptTemplate.from_template("You are a wondrous wizard of math.")
       + few_shot_prompt
       + HumanMessagePromptTemplate.from_template("{input}")
   )

5. Format the prompt with an input:

   formatted_prompt = final_prompt.format_messages(input="What's 3 🦜 3?")
   print(formatted_prompt)  # This will show the formatted prompt

6. Use the prompt within a language model:

   from langchain_openai import ChatOpenAI
   
   chain = final_prompt | ChatOpenAI(model="gpt-3.5-turbo-0125", temperature=0.0)
   response = chain.invoke({"input": "What's 3 🦜 3?"})
   print(response)

This code sets up a system message with a dynamic example selection few-shot prompt template and uses it within a chat model. The SemanticSimilarityExampleSelector dynamically selects examples based on the input, and the FewShotChatMessagePromptTemplate formats these examples into the prompt ^[1][2].

To continue talking to Dosu, mention @dosu.

[donatoaz]

This is not using the few shot prompt AS the system message.