'Input to ChatPromptTemplate is missing variables {\'\\n "subject"\'}. Expected: [\'\\n "subject"\', \'agent_scratchpad\', \'input\'] Received: [\'input\', \'intermediate_steps\', \'agent_scratchpad\']'

[nourbidewe]

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

description_history = []
prompt_3 = f'''You are an assistant that extracts subjects from emails as well as a short description of the email and a one-sentence summary of the email.
1. Subject:
    - Make the subject short and consistent with predefined categories such as Payment Issue, Maintenance Request, etc.
    - Stick to the following subjects to maintain consistency with historical data: Payment Issue, Renewal Request, Maintenance Issue, Neighbors Complaint, Customer Service Complaint, Lease Termination, Parking Issue, App Issue, FAQs, Property Search, Refund Issues, Job Application, Early Termination.
    - If you do not find a suitable subject, output "Other" in the field of the JSON file.
2.  Description:
    - check the historical descriptions first {description_history}. If the email's description matches one from the history, use the historical description as is. If it does not match, output "Other". Never generate a description that is not in the description list!
3. Summary:
    - The summary is one sentence that summarizes the email. It can be detailed but always one sentence.
    - Below are some samples of summaries:
        - Enquired for the availability of 2 bedrooms in wasl 51.
        - When is my renewal due?
        - Following termination procedures, why is there 545 AED pending on my account balance?
        - The quality of work done on my AC is not good.
        - What is the status of my NOC ticket?
        - Asking for the availability of 3 bedrooms in wasl 51.
4. Output:
    - Output the results as a JSON file with keys: subject, description, and summary.
    - If you are unable to detect any of the deliverables, output a JSON with all the keys but with value NAN.
    - The output should always be JSON, nothing else.
    - Always provide a JSON file with the 3 keys specified above!
    - Never output a code or a sentence before the JSON file!
Examples:
Example 1:
Input: I am trying to make a payment but it is not working for some reason. I validated that my bank card is valid 100 percent please reply asap.
Output: {{
  "subject": "Payment Issue",
  "description": "Problems with payment issue for the customer",
  "summary": "The customer is trying to make a payment but it's not working for some reason"
}}
'''
@tool 
def manage_description_history(description):
    """Check if a similar description exists in the history"""
    for saved_description in description_history:
        if description in saved_description:
            return saved_description
    
    description_history.append(description)
    return description

class json_format(BaseModel):
    subject: str = Field(description="topic of the email")
    description: str = Field(description="description of the email")
    summary: str = Field(description="small summary of the email")

def model_function(input):
    # parser = JsonOutputParser(pydantic_object=json_format)
    parser = PydanticOutputParser(pydantic_object=json_format)
    format_instructions = parser.get_format_instructions()
    prompt = ChatPromptTemplate.from_messages(
    [
        ("system", prompt_3,
            
        ),
        # MessagesPlaceholder(variable_name="chat_history", optional=True),
        ("user", "{input}"),
        MessagesPlaceholder(variable_name="agent_scratchpad"),
        ]
    )
    
    prompt.partial_variables = {"format_instructions": format_instructions}

    model = "gpt-3.5-turbo-1106"
    tools = [manage_description_history]
    llm = ChatOpenAI(temperature=0, model=model, streaming=True)
  
    agent= create_tool_calling_agent(llm, tools, prompt)
    
    agent_executor = AgentExecutor(
        agent=agent, 
        tools=tools,
        handle_parsing_errors=True, 
        verbose=True, 
        return_intermediate_steps=True
    )

    answer = agent_executor.invoke({"input": input})
    result_json = json.loads(answer['output'])
    
    return answer['output']
ex_1 = 'what is the policy at your company when it comes to lease termination?'

result1= model_function(ex_1)
result1

Description

I am trying to use langchain agent to the run these requirement. I want to extract subject, description and summary from an email. For the description, I want the agent to use the tool in order to check if a similar description exist already in the list description history then use it else add it.
I am facing the below error when I am trying to run the code

KeyError: 'Input to ChatPromptTemplate is missing variables {'\n "subject"'}. Expected: ['\n "subject"', 'agent_scratchpad', 'input'] Received: ['input', 'intermediate_steps', 'agent_scratchpad']'

System Info

I am installing the latest version of langchain

[dosubot[bot]]

Hey @nourbidewe! I'm here to help you with any bugs, questions, or contributions you might have. Let's solve this together!

To resolve the KeyError related to missing variables in the ChatPromptTemplate, you need to ensure that all required input variables are provided when formatting the template. In your case, the ChatPromptTemplate expects the variables subject, description, and summary, but they are not being provided.

Here's how you can modify your code to include these variables:

1. Define the required variables in the ChatPromptTemplate.
2. Ensure that these variables are passed when invoking the agent.

Here is the updated code:

from langchain_core.prompts.chat import ChatPromptTemplate
from langchain_core.prompts import PromptTemplate
from langchain_core.messages import HumanMessagePromptTemplate, MessagesPlaceholder
from langchain_core.agents import create_tool_calling_agent, AgentExecutor
from langchain_core.llms import ChatOpenAI
from pydantic import BaseModel, Field
import json

description_history = []

prompt_3 = '''You are an assistant that extracts subjects from emails as well as a short description of the email and a one-sentence summary of the email.
1. Subject:
    - Make the subject short and consistent with predefined categories such as Payment Issue, Maintenance Request, etc.
    - Stick to the following subjects to maintain consistency with historical data: Payment Issue, Renewal Request, Maintenance Issue, Neighbors Complaint, Customer Service Complaint, Lease Termination, Parking Issue, App Issue, FAQs, Property Search, Refund Issues, Job Application, Early Termination.
    - If you do not find a suitable subject, output "Other" in the field of the JSON file.
2.  Description:
    - check the historical descriptions first {description_history}. If the email's description matches one from the history, use the historical description as is. If it does not match, output "Other". Never generate a description that is not in the description list!
3. Summary:
    - The summary is one sentence that summarizes the email. It can be detailed but always one sentence.
    - Below are some samples of summaries:
        - Enquired for the availability of 2 bedrooms in wasl 51.
        - When is my renewal due?
        - Following termination procedures, why is there 545 AED pending on my account balance?
        - The quality of work done on my AC is not good.
        - What is the status of my NOC ticket?
        - Asking for the availability of 3 bedrooms in wasl 51.
4. Output:
    - Output the results as a JSON file with keys: subject, description, and summary.
    - If you are unable to detect any of the deliverables, output a JSON with all the keys but with value NAN.
    - The output should always be JSON, nothing else.
    - Always provide a JSON file with the 3 keys specified above!
    - Never output a code or a sentence before the JSON file!
Examples:
Example 1:
Input: I am trying to make a payment but it is not working for some reason. I validated that my bank card is valid 100 percent please reply asap.
Output: {{
  "subject": "Payment Issue",
  "description": "Problems with payment issue for the customer",
  "summary": "The customer is trying to make a payment but it's not working for some reason"
}}
'''

@tool 
def manage_description_history(description):
    """Check if a similar description exists in the history"""
    for saved_description in description_history:
        if description in saved_description:
            return saved_description
    
    description_history.append(description)
    return description

class json_format(BaseModel):
    subject: str = Field(description="topic of the email")
    description: str = Field(description="description of the email")
    summary: str = Field(description="small summary of the email")

def model_function(input):
    parser = PydanticOutputParser(pydantic_object=json_format)
    format_instructions = parser.get_format_instructions()
    
    prompt = ChatPromptTemplate.from_messages(
        [
            ("system", prompt_3),
            ("user", "{input}"),
            MessagesPlaceholder(variable_name="agent_scratchpad"),
        ]
    )
    
    prompt.partial_variables = {"format_instructions": format_instructions}

    model = "gpt-3.5-turbo-1106"
    tools = [manage_description_history]
    llm = ChatOpenAI(temperature=0, model=model, streaming=True)
  
    agent = create_tool_calling_agent(llm, tools, prompt)
    
    agent_executor = AgentExecutor(
        agent=agent, 
        tools=tools,
        handle_parsing_errors=True, 
        verbose=True, 
        return_intermediate_steps=True
    )

    answer = agent_executor.invoke({
        "input": input,
        "subject": "Payment Issue",  # Example subject
        "description": "Problems with payment issue for the customer",  # Example description
        "summary": "The customer is trying to make a payment but it's not working for some reason"  # Example summary
    })
    
    result_json = json.loads(answer['output'])
    
    return answer['output']

ex_1 = 'what is the policy at your company when it comes to lease termination?'
result1 = model_function(ex_1)
print(result1)

In this updated code, the invoke method of agent_executor is provided with the required variables subject, description, and summary to ensure that the ChatPromptTemplate has all the necessary inputs ^[1][2].

To continue talking to Dosu, mention @dosu.

[VangapanduLakshmi]

import streamlit as st
from pydantic import BaseModel, Field, ValidationError
from langchain_community.llms import HuggingFaceHub
from langchain_core.prompts import PromptTemplate
from langchain_core.output_parsers import PydanticOutputParser
import json
import os
from huggingface_hub import login

API_KEY = "hf_"
login(API_KEY)
os.environ["HUGGINGFACE_HUB_TOKEN"] = API_KEY

llm = HuggingFaceHub(
    repo_id="google/gemma-1.1-2b-it",
    model_kwargs={
        "temperature": 0.2, 
        "repetition_penalty": 1.1, 
        "max_new_tokens": 200
    },
    huggingfacehub_api_token=API_KEY
)


class ReviewAnalysis(BaseModel):
    summary: str = Field(description="A brief summary of the customer review with maximum 3 lines")
    positives: list = Field(description="Positives mentioned by the customer, max 3 points")
    negatives: list = Field(description="Negatives mentioned by the customer, max 3 points")
    sentiment: str = Field(description="Overall sentiment of the review (positive, negative, or neutral)")
    emotions: list = Field(description="3-5 emotions expressed based on the sentiment")
    email: str = Field(description="Detailed email to the customer based on the sentiment")

parser = PydanticOutputParser(pydantic_object=ReviewAnalysis)

prompt_text = """
Analyze the customer review below and provide a JSON response with the following structure:

{
  "summary": "A brief summary of the customer review with maximum 3 lines",
  "positives": ["Positives mentioned in the review, max 3 points"],
  "negatives": ["Negatives mentioned in the review, max 3 points"],
  "sentiment": "Overall sentiment of the review (positive, negative, or neutral)",
  "emotions": ["3-5 emotions expressed based on the sentiment"],
  "email": "Detailed email to the customer based on the sentiment"
}

Make sure to fill the fields correctly. Do not include any extra information or metadata. Provide only the required JSON content.
Customer Review: {review}
"""

simple_review = "The product is good, but the delivery was late."

# Adjust input variables to pass both review and summary to the prompt
prompt = PromptTemplate(
    template=prompt_text,
    input_variables=["review"],
    partial_variables={}  # No additional variables are required here
)

chain = (prompt | llm | parser)


try:
    response = chain.invoke({"review": simple_review})
   
    review_analysis = ReviewAnalysis(**response)

    st.write("Response:", review_analysis.dict())

except ValidationError as exc:

    st.error(f"Validation Error: {exc}")
except Exception as e:

    st.error(f"Error: {e}")
    
    
    
    

i got error like this
Error: 'Input to PromptTemplate is missing variables {'\n "summary"'}. Expected: ['\n "summary"', 'review'] Received: ['review']'