Pokie26/Basic_bert_finetuning_practical_1

1import streamlit as st2from transformers import DistilBertTokenizerFast, DistilBertForSequenceClassification3import torch4import torch.nn.functional as F5 6import zipfile7import shutil8import os9 10def unzip_and_save(zip_file_path, extraction_path):11    # Create the extraction directory if it doesn't exist12    os.makedirs(extraction_path, exist_ok=True)13 14    with zipfile.ZipFile(zip_file_path, 'r') as zip_ref:15        folder_name = os.path.basename(zip_file_path).split('.')[0]16        zip_ref.extractall(extraction_path)17        source_path = os.path.join(extraction_path, folder_name)18        destination_path = os.path.join(extraction_path, folder_name)19        if os.path.exists(destination_path):20            print(f"Error: Destination path '{destination_path}' already exists")21        else:22            shutil.move(source_path, destination_path)23 24# Example usage:25# Path to your ZIP file which is your sentimetn analysis model zip26zip_file_path = 'bert_model_dir.zip'  27# Destination folder for extraction28extraction_path = 'bert_model_sentiment_v1'  29 30unzip_and_save(zip_file_path, extraction_path)31 32# Load the fine-tuned model and tokenizer33model_path = "bert_model_sentiment_v1/bert_model_dir"34tokenizer_path = "bert_model_sentiment_v1/bert_model_dir"35 36@st.cache_resource37def load_model():38    model = DistilBertForSequenceClassification.from_pretrained(model_path)39    tokenizer = DistilBertTokenizerFast.from_pretrained(tokenizer_path)40    return model, tokenizer41 42model, tokenizer = load_model()43 44def predict_sentiment(text):45    device = 'cuda' if torch.cuda.is_available() else 'cpu'46    model.to(device)47 48    tokenized = tokenizer(text, truncation=True, padding=True, return_tensors='pt').to(device)49    outputs = model(**tokenized)50 51    probs = F.softmax(outputs.logits, dim=-1)52    preds = torch.argmax(outputs.logits, dim=-1).item()53    probs_max = probs.max().detach().cpu().numpy()54 55    prediction = "Positive" if preds == 1 else "Negative"56    return prediction, probs_max * 10057 58st.title("Sentiment Analysis App")59text = st.text_area("Enter your text:")60 61if st.button("Predict Sentiment"):62    if text:63        sentiment, confidence = predict_sentiment(text)64        st.write(f"Sentiment: {sentiment}")65        st.write(f"Confidence: {confidence:.2f}%")66    else:67        st.write("Please enter some text.")