Parameters calling issue in skills in agent

[Vibi-007]

This is the code:

import pandas as pd
from langchain_core.prompts import PromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_google_vertexai import VertexAI
from tqdm import tqdm
from pandasai import Agent
from pandasai.skills import skill

 

Function doc string to give more context to the model for use this skill

@Skill
def summarize_conversations(df, llm):
    """
    Summarizes telephonic conversations in a DataFrame and identifies the context and issue type.
    Args:
        df (pd.DataFrame): DataFrame containing telephonic conversations.
        llm: Language model for summarization.
    Returns:
        pd.DataFrame: DataFrame with added columns for summary, context, and issue type.
    """
    template = """The following is a telephonic conversation between a telecommunication agent and a customer.
    Summarize the given conversation in one paragraph which contains the context of the conversation and identifies 
    the issue that the customer has spoken about.
    Also identify the context and the issue for which the customer has called the agent, the issue should 
    strictly be of one or two words only..
    {question}

 

    The response should strictly be in the following format:
    Summary:
    Context:
    Issue Type:
    """
    prompt = PromptTemplate.from_template(template)
    summary_chain = prompt | llm | StrOutputParser()

 

    response_list = []
    error_convos = []
    for convo in tqdm(df['Conversation']):
        try:
            response = summary_chain.invoke({"question": convo})
            response_list.append(response)
        except:
            response_list.append(None)
            error_convos.append(convo)

 

    df['llm_summary'] = response_list
    df.dropna(subset=["llm_summary"], inplace=True)

 

    summary = []
    context = []
    issue_type = []
    for i in df['llm_summary']:
        try:
            splits = i.replace("\n\n", "\cd").split("\cd")
            cd = splits[2]
        except:
            splits = i.split("\n")
        summary.append(splits[0])
        context.append(splits[1])
        issue_type.append(splits[2])

 

    df['summary'] = summary
    df['context'] = context
    df['issue_type'] = issue_type

 

    return df

 

 

Example usage with Agent

Load the DataFrame and language model

df = pd.read_excel("transcripts_0416 (1) 1.xlsx")
df = df[0:1000]
llm = VertexAI(model="gemini-1.5-pro-preview-0409", top_k=5, top_p=0.9, temperature=0.2, max_output_tokens=2048)

 

Create agent and add the skill

agent = Agent([df], config= {"llm":llm},memory_size=10)
agent.add_skills(summarize_conversations)

 

Chat with the agent

response = agent.chat("Can you summarize these telephonic conversations?")

The error : name 'llm' is not defined