LLM Bot

What is a Bot

In Dandy we want to make sure all the things you do have a distinct name to isolate them from your projects code. Bots should represent a distinct and singular thing you want to do with in your project.

Create a Basic LLM Bot

To create your own bot we are going to use the BaseLlmBot class from the dandy.llm module.

from dandy.llm import BaseLlmBot, Prompt, DefaultLlmIntel

class AssistantBot(BaseLlmBot):
    @classmethod
    def process(cls, user_prompt: Prompt | str) -> DefaultLlmIntel:
        default_intel = cls.process_prompt_to_intel(
            prompt=user_prompt,
            intel_class=DefaultLlmIntel
        )

        return default_intel

intel = AssistantBot.process('Can you give me an idea for a book?')

print(intel.text)

Sure! How about a story set in a world where everyone has the ability to manipulate one of the four elements: earth, air, fire, or water. The protagonist discovers they can control time and must use this unique power to prevent a catastrophic event that threatens their world.

Advanced LLM Bot

When you create a bot it uses all the defaults of the dandy_settings.py file.

Below is an example of how you can customize bots to make sure they work the way you want.

from dandy.intel import BaseIntel
from dandy.llm import BaseLlmBot, Prompt, LlmConfigOptions


class CandyIntel(BaseIntel):
    short_name: str
    long_name: str
    description: str


class CandyDesignBot(BaseLlmBot):
    config = 'LLAMA_3_2_3B'
    config_options = LlmConfigOptions(
        temperature=0.1,
        max_input_tokens=2000,
        max_output_tokens=2000,
        prompt_retry_count=3,
        randomize_seed=True
    )
    instructions_prompt = (
        Prompt()
        .text('You are a candy design bot and will be given a request to make a new type of candy.')
        .line_break()
        .heading('Rules')
        .list([
            'Make sure you response is sugar based not chocolate based.',
            'Do not include any chocolate based words or phrases in the response.',
            'Incorporate the theme of the request into the response.',
        ])
    )
    intel_class = CandyIntel

    @classmethod
    def process(cls, user_prompt: Prompt | str, candy_theme: str) -> CandyIntel:
        prompt = (
            Prompt()
            .heading('Request')
            .prompt(user_prompt)
            .line_break()
            .heading('Theme')
            .prompt(candy_theme)
        )

        return cls.process_prompt_to_intel(
            prompt=prompt,
        )


candy_intel = CandyDesignBot.process(
    user_prompt='Strawberries and Cookie Dough',
    candy_theme='Medieval Times'
)

print(candy_intel)

short_name='Strawberry Knight' long_name='Strawberry Knight Candy' description='A sweet treat fit for a medieval knight, our Strawberry Knight candy combines the sweetness of fresh strawberries with the creamy delight of cookie dough. The strawberry flavor is reminiscent of a lush castle garden, while the cookie dough adds a fun and playful touch, like a jestering jester in the royal court. The candy is shaped like a miniature sword, complete with a golden wrapper adorned with a regal crest.'