How to use custom prompt when using openai LLM

[sachin-kumar56]

Hi guys,
I have searched everywhere on how to use custom prompts in pandasAI and all the methods mentioned seems to be not working can anyone pls help me with a simple example on how to use custom prompts with agents
Thanks

[dosubot[bot]]

Hey @sachin-kumar56! I'm here to help you with any bugs or questions you have. I can assist you with solving issues, answering questions, and becoming a contributor.

To use custom prompts with OpenAI LLM in pandasAI, you need to create a custom prompt class that inherits from BasePrompt and then use it with the LocalLLM class. Here is a simple example:

1. Create a Custom Prompt Class:

   from pandasai.prompts.base import BasePrompt
   
   class CustomPrompt(BasePrompt):
       def __init__(self, prompt_text: str):
           self.prompt_text = prompt_text
   
       def to_string(self) -> str:
           return self.prompt_text

2. Use the Custom Prompt with LocalLLM:

   from pandasai.llm.local_llm import LocalLLM
   from pandasai.pipelines.pipeline_context import PipelineContext
   
   # Initialize the LocalLLM with your OpenAI API details
   local_llm = LocalLLM(api_base="https://api.openai.com/v1", model="gpt-3.5-turbo", api_key="your_api_key")
   
   # Create an instance of your custom prompt
   custom_prompt = CustomPrompt("Your custom prompt text here")
   
   # Optionally, create a PipelineContext if you need to use memory
   context = PipelineContext(memory=None)  # or provide a Memory instance
   
   # Call the LLM with your custom prompt
   response = local_llm.call(custom_prompt, context)
   
   print(response)

In this example, CustomPrompt is a custom prompt class that you can use to define your own prompt text. You then create an instance of LocalLLM and use the call method with your custom prompt to get a response from the OpenAI LLM ^[1].

To continue talking to Dosu, mention @dosu.

[sachin-kumar56]

Hi looks like this is not the correct solution I want to use the pandasAI agent with custom prompt, thanks

[dosubot[bot]]

To use a custom prompt with the pandasAI agent, you can create a custom prompt class by inheriting from BasePrompt and then use it with the SemanticAgent class. Below is a simple example demonstrating how to achieve this:

1. Create a Custom Prompt Class:

   from pandasai.prompts.base import BasePrompt
   from jinja2 import Environment, FileSystemLoader
   from pathlib import Path
   
   class CustomPrompt(BasePrompt):
       template_path = "custom_prompt_template.tmpl"
   
       def __init__(self, **kwargs):
           self.props = kwargs
   
           if self.template:
               env = Environment()
               self.prompt = env.from_string(self.template)
           elif self.template_path:
               current_dir_path = Path(__file__).parent
               path_to_template = current_dir_path / "templates"
               env = Environment(loader=FileSystemLoader(path_to_template))
               self.prompt = env.get_template(self.template_path)
   
           self._resolved_prompt = None
   
       def build(self):
           return self.prompt.render(self.props)

2. Use the Custom Prompt with the SemanticAgent:

   import pandas as pd
   from pandasai.agent.semantic_agent import SemanticAgent
   from pandasai.llm.bamboo_llm import BambooLLM
   
   # Sample DataFrame
   df = pd.DataFrame({
       'A': [1, 2, 3],
       'B': [4, 5, 6]
   })
   
   # Initialize the agent
   agent = SemanticAgent(dfs=df, config={'llm': BambooLLM()})
   
   # Create an instance of the custom prompt
   custom_prompt = CustomPrompt(context=agent.context, custom_message="Analyze the data")
   
   # Call the LLM with the custom prompt
   result = agent.call_llm_with_prompt(custom_prompt)
   
   print(result)

In this example:

• We define a CustomPrompt class that inherits from BasePrompt.
• We override the build method to render a custom message using a Jinja2 template.
• We initialize the SemanticAgent with a sample DataFrame and a configuration that includes an LLM.
• We create an instance of the CustomPrompt and use it with the call_llm_with_prompt method of the SemanticAgent.

This approach allows you to customize the prompt message and use it within the pandasAI agent framework. Additionally, you can refer to the JudgeAgentPrompt class for more advanced usage, such as converting the prompt to JSON:

class JudgeAgentPrompt(BasePrompt):
    template_path = "judge_agent_prompt.tmpl"

    def __init__(self, **kwargs):
        self.props = kwargs

        if self.template:
            env = Environment()
            self.prompt = env.from_string(self.template)
        elif self.template_path:
            current_dir_path = Path(__file__).parent
            path_to_template = current_dir_path / "templates"
            env = Environment(loader=FileSystemLoader(path_to_template))
            self.prompt = env.get_template(self.template_path)

        self._resolved_prompt = None

    def to_json(self):
        context = self.props["context"]
        memory = context.memory
        conversations = memory.to_json()
        system_prompt = memory.get_system_prompt()
        return {
            "conversation": conversations,
            "system_prompt": system_prompt,
            "prompt": self.to_string(),
        }

This class demonstrates how to include additional context and memory in the prompt, which can be useful for more complex scenarios ^[1][2][3][4].

To continue talking to Dosu, mention @dosu.

[sachin-kumar56]

Hi this still is not working

[dosubot[bot]]

To troubleshoot why your custom prompt is not working with the pandasAI agent, here are a few steps you can follow:

1. Ensure Template File Exists: Verify that the template file (e.g., custom_prompt_template.tmpl) exists in the templates directory relative to the script's location.

2. Check Template Path: Ensure that the path to the template file is correctly set. The code constructs the path using current_dir_path / "templates" or os.path.join(current_dir_path, "templates"), so make sure this path is correct.

3. Verify Context Properties: Ensure that the context property in self.props contains the necessary data, especially memory.

4. Template Rendering: Ensure that the template is correctly rendered. You can add debug statements to check if self.prompt is correctly initialized.

Here is a sample code snippet to help you debug and fix the issue:

from pathlib import Path
from jinja2 import Environment, FileSystemLoader
from pandasai.prompts.base import BasePrompt

class CustomPrompt(BasePrompt):
    template_path = "custom_prompt_template.tmpl"

    def __init__(self, **kwargs):
        self.props = kwargs

        if self.template:
            env = Environment()
            self.prompt = env.from_string(self.template)
        elif self.template_path:
            # find path to template file
            current_dir_path = Path(__file__).parent

            path_to_template = current_dir_path / "templates"
            env = Environment(loader=FileSystemLoader(path_to_template))
            self.prompt = env.get_template(self.template_path)

        self._resolved_prompt = None

    def to_json(self):
        context = self.props.get("context")
        if not context:
            raise ValueError("Context is missing in the properties.")
        
        memory = context.memory
        conversations = memory.to_json()
        system_prompt = memory.get_system_prompt()
        return {
            "conversation": conversations,
            "system_prompt": system_prompt,
            "prompt": self.to_string(),
        }

    def to_string(self):
        if not self._resolved_prompt:
            self._resolved_prompt = self.prompt.render(self.props)
        return self._resolved_prompt

Make sure to:

• Check if the template file exists and is accessible.
• Verify that the context property is correctly passed and contains the necessary memory attribute.
• Render the template correctly using the to_string method.

Additionally, ensure that the BaseAgent class in pandasAI is correctly instantiated and that the context property is properly set up. The BaseAgent class initializes the context with a Memory object, which is crucial for the prompt to function correctly ^[1][2][3][4].

To continue talking to Dosu, mention @dosu.