app.py

import gradio as gr
import openai
import pandas as pd 
import numpy as np
import csv
import os
from datasets import load_dataset
openai.api_key= os.environ.get("openai.api_key")
from openai.embeddings_utils import get_embedding
from openai.embeddings_utils import cosine_similarity
import requests
model_id = "sentence-transformers/all-MiniLM-L6-v2"
import json
hf_token = os.environ.get("hf_token")
import re
from sklearn.metrics.pairwise import cosine_similarity

def generate_embeddings(texts, model_id, hf_token):
    api_url = f"https://api-inference.huggingface.co/pipeline/feature-extraction/{model_id}"
    headers = {"Authorization": f"Bearer {hf_token}"}
    response = requests.post(api_url, headers=headers, json={"inputs": texts, "options":{"wait_for_model":True}})
    embeddings = response.json()
    return embeddings
AP_Bio = load_dataset('vjain/biology_AP_embeddings')
df1 = pd.DataFrame(AP_Bio['train'])
df1["similarity"] = 0

AP_Physics = load_dataset('vjain/AP_physics_embeddings')
df2 = pd.DataFrame(AP_Physics['train'])
df2["similarity"] = 0

dataframes = {
    "AP_Bio": df1,
    "AP_Physics": df2
}

#df = pd.read_csv("TA_embeddings.csv")
#df["embedding"]=df["embedding"].apply(eval).apply(np.array)
def reply(input, dataset_name):
    try:
        if dataset_name not in dataframes:
            return "Invalid dataset selected. Please select a valid dataset."
        if not input:
            return "Please Enter a Question to get an Answer"
        df = dataframes[dataset_name]
        input = input
        input_vector = generate_embeddings(input, model_id,hf_token)
        df["similarities"]=df["embedding"].apply(lambda x: cosine_similarity([x],[input_vector])[0][0])
        data = df.sort_values("similarities", ascending=False).head(10)
        data.to_csv("sorted.csv")
        context = []
        for i, row in data.iterrows():
            context.append(row['text'])
        context
        text = "\n".join(context)
        context = text
        prompt = f"""
                Answer the following question using the context given below.If you don't know the answer for certain, say I don't know.
                Context: {context}
                Q: {input}
                """      
        response= openai.Completion.create(
                    prompt=prompt,
                    temperature=1,
                    max_tokens=500,
                    top_p=1,
                    frequency_penalty=0,
                    presence_penalty=0,
                    model="text-davinci-003"
                )["choices"][0]["text"].strip(" \n")
        return response
    except Exception as e:
        return f"An error occurred: {e}"

csv_dropdown = gr.inputs.Dropdown(
    label="Select the Book",
    choices=["AP_Bio", "AP_Physics"],
    default="AP_Bio"
            
)
input_text = gr.inputs.Textbox(
    label="Enter your questions here",
    placeholder="E.g. What is DNA?",
    lines=3
    
)
text_output = gr.outputs.Textbox(label="Answer")

description = "Scholar Bot is a question answering system designed to provide accurate and relevant answers to questions from this book hosted by OpenStax https://openstax.org/details/books/biology-ap-courses. Simply enter your question in the text box above and Scholar Bot will use advanced natural language processing algorithms to search a large corpus of biology text to find the best answer for you. Scholar Bot uses the Sentence Transformers model to generate embeddings of text, and OpenAI's GPT-3 language model to provide answers to your questions."

ui = gr.Interface(fn=reply,
                  inputs=[input_text, csv_dropdown],
                  outputs=[text_output],
                  title="Scholar Bot",
                  description=description,
                  theme="light",
                  layout="vertical",
                  allow_flagging=False,
                  examples=[["What is the function of DNA polymerase?", "AP_Bio"]]
                )

ui.launch()