app.py

from pptx import Presentation
import re
import gradio as gr
from transformers import AutoModelForSequenceClassification, AutoTokenizer
import torch
import torch
torch.device("cpu")
import torch.nn.functional as F
from transformers import pipeline

# Load the pre-trained model and tokenizer
tokenizer = AutoTokenizer.from_pretrained("Ahmed235/roberta_classification")
model = AutoModelForSequenceClassification.from_pretrained("Ahmed235/roberta_classification")

# Create a summarization pipeline
summarizer = pipeline("summarization", model="Falconsai/text_summarization")

def extract_text_from_pptx(file_path):
    presentation = Presentation(file_path)
    text = []
    for slide_number, slide in enumerate(presentation.slides, start=1):
        for shape in slide.shapes:
            if hasattr(shape, "text"):
                text.append(shape.text)
    return "\n".join(text)

def predict_pptx_content(file_path):
    extracted_text = extract_text_from_pptx(file_path)
    cleaned_text = re.sub(r'\s+', ' ', extracted_text)

    # Tokenize and encode the cleaned text
    input_encoding = tokenizer(cleaned_text, truncation=True, padding=True, return_tensors="pt")

    # Perform inference
    with torch.no_grad():
        outputs = model(**input_encoding)
        logits = outputs.logits

    probabilities = F.softmax(logits, dim=1)

    predicted_label_id = torch.argmax(logits, dim=1).item()
    predicted_label = model.config.id2label[predicted_label_id]
    predicted_probability = probabilities[0][predicted_label_id].item()

    # Summarize the cleaned text
    summary = summarizer(cleaned_text, max_length=80, min_length=30, do_sample=False)[0]['summary_text']

    prediction = {
        "Predicted Label": predicted_label,
        "Evaluation": f"Evaluate the topic according to {predicted_label} is: {predicted_probability}",
        "Summary": summary
    }

    return prediction

# Define the Gradio interface
iface = gr.Interface(
    fn=predict_pptx_content,
    inputs=gr.File(type="filepath", label="Upload PowerPoint (.pptx) file"),
    outputs=["text", "text", "text"],  # Predicted Label, Evaluation, Summary
    live=False,  # Change to True for one-time analysis
    title="<h1 style='color: lightgreen; text-align: center;'>PPTX Analyzer</h1>",
)

# Deploy the Gradio interface
iface.launch(share=True)