galaxybrain 0.0.4

LLM extensions framework

GalaxyBrain

Turning LLMs into mighty shape rotators!

GalaxyBrain is a workflow framework for large language models (LLMs). With GalaxyBrain, developers can define prompt completion steps, prompt rules, result validators, and more. Additionally, the framework adds memory capabilities to LLM workflows making it easy to handle state between steps.

Goals

1. Reduce surprises when it comes to working with LLMs.
2. Focus on production use cases and CI/CD compatibility.
3. Avoid bloat and keep base primitives simple.

Getting Started

First, install the library:

pip install galaxybrain

Currently, GalaxyBrain only supports OpenAI APIs, so you'll need to get an API key and keep it on your path as OPENAI_API_KEY or pass it to the driver object manually. For example:

import os
from dotenv import load_dotenv
from galaxybrain.drivers import OpenAiDriver

load_dotenv()

api_key = os.environ.get('OPENAI_KEY')
driver = OpenAiDriver(api_key)

Here is an example of some of GalaxyBrain's functionality:

from galaxybrain.rules import Rule, Validator
from galaxybrain.workflows import CompletionStep, Workflow
from galaxybrain.drivers import OpenAiDriver
from galaxybrain.prompts import Prompt
import galaxybrain.rules as rules


chat_rules = [
    rules.json.return_valid_json(),
    rules.json.put_answer_in_field("Names"),
    rules.conversation.be_truthful(),
    rules.conversation.your_name_is("GalaxyGPT"),
    Rule("only use information from fantasy novels")
]

driver = OpenAiDriver(temperature=0.5, user="demo")
workflow = Workflow(rules=chat_rules, driver=driver)

workflow.add_step(
    CompletionStep(input=Prompt("Give me ideas for two names from the same universe"))
)

workflow.start()

# Output:
# {
#     "Names": ["Frodo Baggins", "Gandalf the Grey"]
# }

workflow.add_step(
    CompletionStep(input=Prompt("Give me 3 more from another universe"))
)

workflow.resume()

# Output:
# {
#     "Names": ["Dumbledore", "Luna Lovegood", "Harry Potter"]
# }

Use ComputeStep to delegate computational tasks to Python:

workflow.add_step(
    ComputeStep(input=Prompt(f"generate two random 3x3 matrices and multiply them"))
)

This will generate the following code that GalaxyBrain runs locally and returns it to the LLM in the next prompt:

print(np.matmul(np.random.rand(3,3), np.random.rand(3,3)))

You can ask it open-ended computational questions:

workflow.add_step(
    ComputeStep(input=Prompt(f"Sally is 5 feet tall, Jack is 14 inches taller than Sally. How tall is Jack?"))
)

You can also validate results against your rules:

validator = Validator(workflow.last_step().output, chat_rules)

if validator.validate():
    print("Rule validations passed")
else:
    print("Rule validations failed")
    print(validator.failed_rules())