app.py

import streamlit as st
from transformers import pipeline, AutoTokenizer, AutoModelForSequenceClassification
from transformers import DistilBertTokenizerFast, DistilBertForSequenceClassification
import torch
from torch.utils.data import DataLoader
import pandas as pd

#for all other models
def sentiment_analysis(text, model_name):
    model = AutoModelForSequenceClassification.from_pretrained(model_name)
    tokenizer = AutoTokenizer.from_pretrained(model_name)

    classifier = pipeline('sentiment-analysis', model=model, tokenizer=tokenizer)
    result = classifier(text)

    return result[0]['label'], result[0]['score']
    
#for my finetuned model
def finetune(text):
    #defining tokenizer and loading model
    tokenizer = DistilBertTokenizerFast.from_pretrained("psychedelicbunny/bertfinetuned")
    model = DistilBertForSequenceClassification.from_pretrained("psychedelicbunny/bertfinetuned")
    
    model.eval()  # switch model to evaluation mode
    label_names = {
        0: 'toxic',
        1: 'severe_toxic',
        2: 'obscene',
        3: 'threat',
        4: 'insult',
        5: 'identity_hate'
    }
    #creating encodings from input text
    encoding = tokenizer(text, truncation = True, padding = True, max_length = 128)
    input_ids = torch.tensor(encoding["input_ids"]).unsqueeze(0)
    attention_mask = torch.tensor(encoding["attention_mask"]).unsqueeze(0)
    
    with torch.no_grad():
        #running the model
        outputs = model(input_ids=input_ids, attention_mask=attention_mask)
        logits = outputs.logits
        probabilities = torch.softmax(logits, dim=1)

        #score and label for first predicted label - usually 'toxic'
        score = max(probabilities[0]).item()
        predicted_class_idx = torch.argmax(probabilities, dim=1)
        predicted_label = label_names[predicted_class_idx.item()]
        #score and label for secondary predicted label
        confidence = max(probabilities[0][2:6]).item()
        other_index = torch.where(probabilities[0] == confidence)
        other_label = label_names[other_index[0].item()]

        #creating table
        df = pd.DataFrame({
            'Text': text,
            'Main Label': predicted_label,
            'Score1': score,
            'Second Label': other_label,
            'Score2': confidence
        }, index = [0])
        return df
        st.table(df)

#experiment/backup prediction function       
def predict(text):
    inputs = tokenizer(text, padding=True, truncation=True, return_tensors='pt')
    outputs = model(**inputs)
    _, prediction = torch.max(outputs.logits, dim=1)
    return prediction.item()
        

def main():
    st.title("Sentiment Analysis App")
    text = st.text_input("Enter text:",  value = "You're great!")
    model_name = st.selectbox("Select a pretrained model", ["psychedelicbunny/bertfinetuned", "bert-base-uncased", "finiteautomata/bertweet-base-sentiment-analysis", "roberta-base"])

    if st.button("Analyze"):
        with st.spinner('Analyzing...'):
            if model_name == "psychedelicbunny/bertfinetuned":
                sentiment = finetune(text)
            else:
                sentiment = sentiment_analysis(text, model_name)
            st.write("Sentiment:", sentiment)

if __name__ == '__main__':
    main()