# Import Libraries
import gradio as gr
import google.generativeai as genai



# using the gemini model
# api
genai.configure(api_key="AIzaSyCmdsQeYMapI6l3maUpAkrE_GkT348w5UA")

# model_1 = genai.GenerativeModel('gemini-pro')
# model_2 = genai.GenerativeModel('gemini-pro')

model = genai.GenerativeModel('gemini-pro')

# assign score function
def assign_score(response):
    if 'score=1' in response:
        score = 1
    elif 'score=2' in response:
        score = 2
    elif 'score=3' in response:
        score = 3
    elif 'score=4' in response:
        score = 4
    elif 'score=5' in response:
        score = 5
    else:
        score = 0
    return score

def model_prompt_response(prompt):
# using the gemini model
  
    prompt_response_1 = model.generate_content(prompt + "in 3 sentences or less")
    prompt_response_2 = model.generate_content(prompt + "in 3 sentences or less")
    response1_model = prompt_response_1.text
    response2_model = prompt_response_2.text

    prompt_to_judge_1 = f'Given the question, {prompt} and answer {response1_model}, decide whether: score = 1 when answer is strongly irrelevant to the question, score = 2 when the answer is slitely relevant, score = 3 when the answer is relevant, score = 4 when it is slitely relevant and score = 5 for strongly relevant answer. Give reason for your answer and start with scores and write the score as [score=1], [score=2], [score=3], [score=4], or [score=5].'

  
    prompt_to_judge_2 = f'Given the question, {prompt} and answer {response2_model}, decide whether: score = 1 when answer is strongly irrelevant to the question, score = 2 when the answer is slitely relevant, score = 3 when the answer is relevant, score = 4 when it is slitely relevant and score = 5 for strongly relevant answer. Give reason for your answer and start with scores and write the score as [score=1], [score=2], [score=3], [score=4], or [score=5]. '

    judge_1_response_1 = model.generate_content(prompt_to_judge_1)
    judge_2_response_2 = model.generate_content(prompt_to_judge_2)
    judge_1 = judge_1_response_1.text
    judge_2 = judge_2_response_2.text


    score1_judger = assign_score(judge_1)
    score2_judger = assign_score(judge_2) 


  # return values

    return response1_model, judge_1, score1_judger, response2_model, judge_2, score2_judger



# anonators judgement
def judge_response(score1_judger, score2_judger, judge_1, judge_2):
  score1_anonator = score1_judger
  score2_anonator = score2_judger
  response1_judger = judge_1
  response2_judger = judge_2

  return response1_judger, score1_anonator, response2_judger, score2_anonator

demo = gr.Blocks(theme=gr.themes.Monochrome(), title='Group1_Subtask2')




with demo:

  gr.Markdown(
      """
      # Group1_Subtask2
      ## The Model takes a request from the user and returns a response.
      #### A judger is asigned the response from the model and outputs corresponding judgement.
      #### The user asses the judgement and alters the responds to correspond to their views or objective truth
      """
  )
  prompt = gr.Textbox(label='Question to the model')
  with gr.Row():
    response1_model = gr.Textbox(label='Model Response 1')

    judge_1 = gr.Textbox(label='Judger Model Response 1/User Editable', interactive=True)

    score1_judger = gr.Number(value=2, label='Judger Model Score 1/User Editable', interactive=True)
  with gr.Row():
    response2_model = gr.Textbox(label='Model Response 2')
    judge_2 = gr.Textbox(label='Judger Model Response 2/User Editable', interactive=True)
    score2_judger = gr.Number(value=2, label='Judger Model Score 2/User Editable', interactive=True)


  response_model = gr.Button("Model response")





# user/judger edited response
  with gr.Row():
    response1_judger = gr.Textbox(label='Corrected Response 1')
    score1_anonator = gr.Number(label='Corrected score 1')
  
    response2_judger = gr.Textbox(label='Corrected Response 2')
    score2_anonator = gr.Number(label='Corrected Score 2')


  judge_model = gr.Button("Judge-Annonator")

  response_model.click(model_prompt_response, inputs=prompt, outputs=[response1_model,
                                                                      judge_1,
                                                                      score1_judger,
                                                                      response2_model,
                                                                      judge_2,
                                                                      score2_judger,
                                                                      ])
  judge_model.click(judge_response, inputs=[score1_judger, score2_judger, judge_1, judge_2] , outputs=[response1_judger, score1_anonator, response2_judger, score2_anonator])




demo.launch(share=True)

# Team Members
'''
 Collins Okoth: collins.okech@digitaldividedata.com
 Albert Disi: albert.disi@digitaldividedata.com
 Joseph Amadi: joseph.amadi@digitaldividedata.com
'''