appStore/rag.py

import os
import numpy as np
import pandas as pd
import openai
from haystack.schema import Document
import streamlit as st
from tenacity import retry, stop_after_attempt, wait_random_exponential
from huggingface_hub import InferenceClient


# Get openai API key
hf_token = os.environ["HF_API_KEY"]

# define a special function for putting the prompt together (as we can't use haystack)
def get_prompt(context, label):
  base_prompt="Summarize the following context efficiently in bullet points, the less the better - but keep concrete goals. \
  Summarize only elements of the context that address vulnerability of "+label+" to climate change. \
  If there is no mention of "+label+" in the context, return: 'No clear references to vulnerability of "+label+" found'.  \
  Do not include an introduction sentence, just the bullet points as per below. \
  Formatting example: \
    - Bullet point 1 \
    - Bullet point 2 \
  "

  prompt = base_prompt+"; Context: "+context+"; Answer:"

  return prompt


# # exception handling for issuing multiple API calls to openai (exponential backoff)
# @retry(wait=wait_random_exponential(min=1, max=60), stop=stop_after_attempt(6))
# def completion_with_backoff(**kwargs):
#     return openai.ChatCompletion.create(**kwargs)

class ChatCompletionResult:
    def __init__(self):
        self.content = ""

    def add_content(self, text):
        self.content += text

    def get_full_content(self):
        return self.content.strip()

def run_query(context, label, model_sel_name):
    '''
    Summarize provided test
    '''
    chatbot_role = """You are an analyst specializing in climate change impact assessments and producing insights from policy documents."""

    ### adding new template which doesnt need 'system' role [Further investigation needed if its applicable to all mmodels or not]
    messages = [{"role": "user", "content": chatbot_role + get_prompt(context, label)}]
    

    ## old template
    #messages = [{"role": "system", "content": chatbot_role},{"role": "user", "content": get_prompt(context, label)}]
    
    # Initialize the client, pointing it to one of the available models
    client = InferenceClient(model_sel_name, token=hf_token)
    
    # # Instantiate ChatCompletion as a generator object (stream is set to True)
    # chat_completion = client.chat.completions.create(
    #     messages=messages,
    #     stream=True
    # )

    # # Create an object to store the full chat completion
    # completion_result = ChatCompletionResult()
    # res_box = st.empty()

    # # Iterate through the streamed output
    # for chunk in chat_completion:
    #     # Extract the object containing the text
    #     if chunk.choices is not None:
    #         chunk_message = chunk.choices[0].delta
    #         if 'content' in chunk_message:
    #             completion_result.add_content(chunk_message['content'])  # Store the message
    #             # Add the latest text and merge it with all previous
    #             result = completion_result.get_full_content()
    #             res_box.success(result)  # Output to response text box

    
    # Use streaming text generation
    response_stream = client.text_generation(prompt, stream=True, max_new_tokens=512)

    completion_result = ChatCompletionResult()
    res_box = st.empty()

    for chunk in response_stream:
        completion_result.add_content(chunk)
        result = completion_result.get_full_content()
        res_box.success(result)

    return completion_result

    # # Return the stored chat completion object for later use
    # return completion_result