Design flow for LLM

Features and tasks

• Database for retrieval system
  • Faiss database
  • Graph search
• Online searching
  • DuckduckGo search
• LLM
  • OpenAI
  • Gemini
  • HuggingFace
• Flow
  • PAL flow
• Documents

How to install

pip install git+https://github.com/phuvinhnguyen/FlowDesign.git

Example use

This framework can be used to create new flow easily, for example:

from FlowDesign.processor import *

class PALFlow:
    def __init__(self, chatbot, working_dir='.'):
        nlp = TextProcessor(chatbot, 'Write a executable Python program to solve the problem: {query}', 'llm')
        parser = ParseCodeProcessor('{llm}')
        exec = PythonProcessor(working_dir=working_dir)
        self.flow = nlp*parser*exec # ask llm to gen code -> parse code -> exec code

    def chat(self, query: str) -> str:
        result = self.flow(IOBase({'query': query})).run
        return result['program_out'] if result['program_out'] != None else result['program_err']

How to use the built-in PAL flow with gemini

from FlowDesign.flow.pal import PALFlow
from FlowDesign.chatbot.base import GeminiBot

bot = GeminiBot('GEMINI_TOKEN')
flow = PALFlow(bot)
output1 = flow.chat('what is the result of (1231.23*3242.432^2)/7?+pi^(2^pi)-e^(pi^2)') # can be difficult even with GPT4
output2 = flow.chat('sort this list: [1,232,4,432,34,25,13,31,23,32,324,3,5,23,32,3,2,2,2,2,2,2,2,2,2,2,2,2]') # can be very difficult with the list is larger (1000 numbers for example) even with CoT, ToT, GoT