from pyexpat import model from transformers import GPT2LMHeadModel, GPT2Tokenizer from streamlit_lottie import st_lottie import json import pandas as pd import requests import torch import tensorflow as tf import streamlit as st from streamlit_option_menu import option_menu logo = "https://www.google.com/url?sa=i&url=https%3A%2F%2Ffr.depositphotos.com%2Fvector-images%2Frobot-logo.html&psig=AOvVaw14rAtmwJQVSpRFXFY6us7z&ust=1647274982461000&source=images&cd=vfe&ved=0CAsQjRxqFwoTCPjhzdO_w_YCFQAAAAAdAAAAABAD" st.set_page_config(page_icon = logo, page_title ="Bonsoir !", layout = "wide") @st.cache(allow_output_mutation=True) def load_tokenizer(): tokenizer = GPT2Tokenizer.from_pretrained("gpt2-large") model = GPT2LMHeadModel.from_pretrained("gpt2-large", pad_token_id=tokenizer.eos_token_id) return tokenizer @st.cache(allow_output_mutation=True) def load_model(): model = GPT2LMHeadModel.from_pretrained("gpt2-large", pad_token_id=tokenizer.eos_token_id) return model tokenizer =load_tokenizer() model = load_model() def reponse(question, temp=0.5, long=40): input_ids = tokenizer.encode(question, return_tensors='pt') output = model.generate(input_ids, max_length=long, temperature =temp, num_beams=5, no_repeat_ngram_size=2, early_stopping=True) rep = tokenizer.decode(output[0], skip_special_tokens=True) return rep def load_animation(url: str): r = requests.get(url) if r.status_code != 200 : return None return r.json() url = "https://assets10.lottiefiles.com/packages/lf20_96bovdur.json" robot = load_animation(url) def contact_message(): st.header(":mailbox: Let's Get In Touch !") name, message = st.columns((1,2)) with name: contact_form = """
""" st.markdown(contact_form, unsafe_allow_html=True) with message : contact_form2 = """
""" st.markdown(contact_form2, unsafe_allow_html=True) with open("style2.txt") as f: st.markdown(f"", unsafe_allow_html=True) def robot(): robot = load_animation(url) col1, col2, col3 = st.columns((5,1,5)) with col1: st.subheader("Choose the length of my answer") long = st.number_input("Be aware that long answers require more time to think !", min_value=10, max_value=250, step =10) st.subheader("Ask me something") question = st.text_input('Be aware that I speak only english for the moment !',max_chars = 60) question = str(question) ok = st.button('Ask me') with col3: st_lottie(robot, speed=1, loop=True, quality = "low",height =300, width = 300) if ok: rep = reponse(question, long = long) rep_style = f'

{rep}

' st.markdown(rep_style, unsafe_allow_html=True) def main(): st.title("Shall we chat ? Ask me a question") with st.sidebar: choice = option_menu( menu_title = "Ask Me", options = ["Question", "Envoie Moi Un Message"], icons=["chat","envelope"], menu_icon="robot" ) if choice == "Envoie Moi Un Message": contact_message() elif choice == "Question": robot() st.sidebar.subheader(":notebook_with_decorative_cover: Par Maxime Le Tutour") st.sidebar.write(" :blue_book: [**Mon LinkedIn**](https://www.linkedin.com/in/maxime-le-tutour-95994795/)", unsafe_allow_html =True) print("ok") if __name__ == '__main__': main()