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 quotes dataset
df = pd.read_json("krishnamurti_df.json")

# Loading back in our sentence similarity and language model
model = SentenceTransformer("msmarco-roberta-base-v3") # best performing model

krishnamurti_generator = pipeline("text-generation", model="distilgpt2")

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

def ask_krishnamurti(question):
  answer = krishnamurti_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(question):
  question_embedding = model.encode(question)
  sims = [util.dot_score(question_embedding, quote_embedding) for quote_embedding in df['Embedding']]
  ind = np.argpartition(sims, -5)[-5:]
  similar_sentences = [df['Quotes'][i] for i in ind]
  top5quotes = pd.DataFrame(data = similar_sentences, columns=["Quotes"], index=range(1,6))
  top5quotes['Quotes'] = top5quotes['Quotes'].str[:-1].str[:250] + "..."
  return top5quotes
 
def main(question):
    return ask_krishnamurti(question), get_similar_quotes(question)

with gr.Blocks() as demo:
    gr.Markdown("""
    # Ask Krishanmurti
    """
    )
    with gr.Row():
        inp = gr.Textbox(placeholder="Place your question here...")
        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=inp, outputs=[out1,out2])

demo.launch()