app.py

import streamlit as st
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import numpy as np
import torch

TOP_N = 5

def main():
    
    st.set_page_config(  # Alternate names: setup_page, page, layout
        layout="centered",  # Can be "centered" or "wide". In the future also "dashboard", etc.
        initial_sidebar_state="auto",  # Can be "auto", "expanded", "collapsed"
        page_title="Emoji-motion!",  # String or None. Strings get appended with "• Streamlit".
        page_icon=None,  # String, anything supported by st.image, or None.
    )
    
    st.title('Emoji-motion!')
    
    example_prompts = [
        "it's pretty depressing when u hit pan on ur favourite highlighter",
        "After what just happened. In need to smoke.",
        "I've never been happier. I'm laying awake as I watch @user sleep. Thanks for making me happy again, babe.",
        "@user is the man",
        "Поприветствуем моего нового читателя @user",
        "сегодня у одной крутой бичи день рождения! @user поздравляю тебя с днем рождения! будь самой-самой счастливой,красота:* море любви тебе",
        "Никогда не явствовала себя ужаснее, чем сейчас:( я просто раздавленна",
        "Самое ужасное - это ожидание результатов",
        "печально что заряд одинаково фигово держится(",
    ]
    

    example = st.selectbox("Choose an example", example_prompts)
    
    # Take the message which needs to be processed
    message = st.text_area("...or paste some text to see the model's predictions", example)
    # st.title(message)
    st.text('')
    models_to_choose = [
        "amazon-sagemaker-community/xlm-roberta-en-ru-emoji-v2",
        "AlekseyDorkin/xlm-roberta-en-ru-emoji"
    ]
    
    BASE_MODEL = st.selectbox("Choose a model", models_to_choose)
    TOP_N = 5


    def preprocess(text):
        new_text = []
        for t in text.split(" "):
            t = '@user' if t.startswith('@') and len(t) > 1 else t
            t = 'http' if t.startswith('http') else t
            new_text.append(t)
        return " ".join(new_text)

    @st.cache(allow_output_mutation=True, suppress_st_warning=True, show_spinner=True)
    def load_model():
        tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL)
        model = AutoModelForSequenceClassification.from_pretrained(BASE_MODEL)
        return model, tokenizer

    def get_top_emojis(text, top_n=TOP_N):
        model, tokenizer = load_model()
        preprocessed = preprocess(text)
        inputs = tokenizer(preprocessed, return_tensors="pt")
        preds = model(**inputs).logits
        scores = torch.nn.functional.softmax(preds, dim=-1).detach().numpy()
        ranking = np.argsort(scores)
        ranking = ranking.squeeze()[::-1][:top_n]
        emojis = [model.config.id2label[i] for i in ranking]
        return ', '.join(map(str, emojis))
        
    
    # Define function to run when submit is clicked
    def submit(message):
        if len(message) > 0:
            emoji = get_top_emojis(message)
            html_str = f"""
            <style>
            p.a {{
            font: 26px Courier;
            }}
            </style>
            <p class="a">{emoji}</p>
            """

            st.markdown(html_str, unsafe_allow_html=True)
            # st.markdown(emoji)
        else:
            st.error("The text can't be empty")
    
    # Run algo when submit button is clicked
    if st.button('Submit'):
        submit(message)
    
    st.text('')
    st.markdown(
    '''<span style="color:blue; font-size:10px">App created by [@AlekseyDorkin](https://huggingface.co/AlekseyDorkin)
     and [@akshay7](https://huggingface.co/akshay7)</span>''', 
     unsafe_allow_html=True,
    )
    
if __name__ == "__main__":
    main()