app.py

import gradio as gr
from transformers import pipeline
import numpy as np
import pandas as pd
from sentence_transformers import SentenceTransformer, util
import nltk
from nltk import sent_tokenize
nltk.download("punkt")

# Loading in dataframes
krishnamurti_df = pd.read_json("krishnamurti_df.json")
stoic_df = pd.read_json("stoic_df.json")

# Loading in sentence_similarity model
sentence_similarity_model = "all-mpnet-base-v2"
model = SentenceTransformer(sentence_similarity_model) 

# Loading in text-generation models
stoic_generator = pipeline("text-generation", model="eliwill/stoic-generator-10e")
krishnamurti_generator =  pipeline("text-generation", model="distilgpt2")

# Creating philosopher dictionary
philosopher_dictionary = {
    "stoic": {
        "generator": stoic_generator,
        "dataframe": stoic_df
        },

    "krishnamurti": {
        "generator": krishnamurti_generator,
        "dataframe": krishnamurti_df
        }
}

############### DEFINING FUNCTIONS ###########################

def ask_philosopher(philosopher, question):
  """ Return first 5 sentences generated by question for the given philosopher model """

  generator = philosopher_dictionary[philosopher]['generator']
  answer = generator(question, min_length=100, max_length=120)[0]['generated_text'] # generate about 50 word tokens
  answer = " ".join(sent_tokenize(answer)[:6]) # Get the first five sentences
  return answer

def get_similar_quotes(philosopher, question):
  """ Return top 3 most similar quotes to the question from a philosopher's dataframe """
  df = philosopher_dictionary[philosopher]['dataframe']
  question_embedding = model.encode(question)
  sims = [util.dot_score(question_embedding, quote_embedding) for quote_embedding in df['Embedding']]
  ind = np.argpartition(sims, -3)[-3:]
  similar_sentences = [df['quote'][i] for i in ind]
  top3quotes = pd.DataFrame(data = similar_sentences, columns=["Quotes"], index=range(1,4))
  top3quotes['Quotes'] = top3quotes['Quotes'].str[:-1].str[:250] + "..."
  return top3quotes

def main(question, philosopher):
  out_image = "marcus-aurelius.jpg"
  return ask_philosopher(philosopher, question), get_similar_quotes(philosopher, question), out_image
  
with gr.Blocks(css=".gradio-container {background-image: url('file=mountains_resized.jpg')} #title {color: white}") as demo:
    gr.Markdown("""
    # Ask a Philsopher
    """, elem_id="title"
    )
    with gr.Row():
        with gr.Column():
          inp1 = gr.Textbox(placeholder="Place your question here...", label="Ask a question")
          inp2 = gr.Dropdown(choices=["stoic", "krishnamurti"], value="stoic", label="Choose a philosopher")
          out_image = gr.Image()

        with gr.Column():
            out1 = gr.Textbox(
                        lines=3, 
                        max_lines=10,
                        label="Answer"
                    )
            out2 = gr.DataFrame(
                        headers=["Quotes"],
                        max_rows=5,
                        interactive=False,
                        wrap=True)
    btn = gr.Button("Run")
    btn.click(fn=main, inputs=[inp1,inp2], outputs=[out1,out2,out_image])

demo.launch()