dosubot[bot]
commented
1 month ago Hey @vsneelesh! I'm here to help you with any bugs, questions, or contributions you have regarding PandasAI. Let's tackle this challenge together!
To set up and use PandasAI agents with a local LLM, follow these steps:
Integration Steps:
- Install Dependencies: Ensure you have the necessary dependencies installed, such as
openaiandpandasai. - Create a Custom LLM Class: Extend the
LLMbase class to create your custom LLM class. - Implement Required Methods: Implement methods for chat completion and calling the LLM.
Creating a Custom LLM Class:
Here's an example of how to create a custom LocalLLM class:
from __future__ import annotations
from typing import TYPE_CHECKING
from openai import OpenAI
from pandasai.helpers.memory import Memory
from pandasai.prompts.base import BasePrompt
from pandasai.llm.base import LLM
if TYPE_CHECKING:
from pandasai.pipelines.pipeline_context import PipelineContext
class LocalLLM(LLM):
def __init__(self, api_base: str, model: str = "", api_key: str = "", **kwargs):
if not api_key:
api_key = "dummy"
self.model = model
self.client = OpenAI(base_url=api_base, api_key=api_key).chat.completions
self._invocation_params = kwargs
def chat_completion(self, value: str, memory: Memory) -> str:
messages = memory.to_openai_messages() if memory else []
# adding current prompt as latest query message
messages.append(
{
"role": "user",
"content": value,
}
)
params = {"model": self.model, "messages": messages, **self._invocation_params}
response = self.client.create(**params)
return response.choices[0].message.content
def call(self, instruction: BasePrompt, context: PipelineContext = None) -> str:
self.last_prompt = instruction.to_string()
memory = context.memory if context else None
return self.chat_completion(self.last_prompt, memory)
@property
def type(self) -> str:
return "local"Using the Agent with DataFrames:
- Create DataFrames: Prepare your data in DataFrames.
- Initialize the Agent: Set up the
Agentwith the DataFrames and configure the memory size. - Interact with the Agent: Use chat, clarification questions, and training methods to interact with the agent.
Example:
import os
import pandas as pd
from pandasai import Agent
employees_data = {
"EmployeeID": [1, 2, 3, 4, 5],
"Name": ["John", "Emma", "Liam", "Olivia", "William"],
"Department": ["HR", "Sales", "IT", "Marketing", "Finance"],
}
salaries_data = {
"EmployeeID": [1, 2, 3, 4, 5],
"Salary": [5000, 6000, 4500, 7000, 5500],
}
employees_df = pd.DataFrame(employees_data)
salaries_df = pd.DataFrame(salaries_data)
# By default, unless you choose a different LLM, it will use BambooLLM.
# You can get your free API key signing up at https://pandabi.ai (you can also configure it in your .env file)
os.environ["PANDASAI_API_KEY"] = "your-api-key"
agent = Agent([employees_df, salaries_df], memory_size=10)
# Chat with the agent
response = agent.chat("Who gets paid the most?")
print(response)
# Get Clarification Questions
questions = agent.clarification_questions("Who gets paid the most?")
for question in questions:
print(question)
# Explain how the chat response is generated
response = agent.explain()
print(response)
# Train with data
queries = [
"Display the distribution of ages in the population.",
"Visualize the distribution of product ratings.",
"Show the distribution of household incomes in a region.",
]
codes = [
"display_age_distribution()",
"visualize_product_ratings_distribution()",
"show_household_incomes_distribution_in_region()",
]
agent.train(queries, codes)
print("Done")Best Practices:
- Ensure correct configuration of
api_base,model, andapi_key. - Handle memory efficiently to maintain context across interactions.
- Use appropriate prompts and context to get accurate responses from the LLM.
By following these steps and using the provided code, you can effectively set up and use PandasAI agents with a local LLM [1][2][3][4].
To continue talking to Dosu, mention @dosu.
vsneelesh
I'm currently working with PandasAI and I'm interested in utilizing agents for my projects. However, I'm facing some challenges in integrating a local LLM instead of relying on the default Bamboo LLM.
I would like to know how to effectively set up and use PandasAI agents with my local LLM. Specifically, I'm looking for guidance on:
The steps needed to integrate a local LLM with PandasAI. Any examples or code snippets that demonstrate how to create a custom LLM class for this purpose. Best practices for ensuring the local LLM interacts smoothly with the PandasAI agent.
Thanks in advance for your help!