from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch
import streamlit as st
device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")

model_name = "MoritzLaurer/mDeBERTa-v3-base-mnli-xnli"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)

with open('articles_list.pkl', 'rb') as file:
    articles_list = pickle.load(file)
    
label_names = []
for i in articles_list[0:20]:
    label_names.append(i[0:15])

def classify(text):
    input = tokenizer(text, truncation=True, return_tensors="pt")
    output = model(input["input_ids"].to(device))  # device = "cuda:0" or "cpu"
    prediction = torch.softmax(output["logits"][0], -1).tolist()
    prediction = {name: round(float(pred) * 100, 1) for pred, name in zip(prediction, label_names)}
    return prediction


text = st.text_input('Enter some text:')  # Input field for new text
if text:
    st.text(classify(text))