app.py

from pptx import Presentation
import re
import gradio as gr
from transformers import AutoModelForSequenceClassification, AutoTokenizer
import torch
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")
device = torch.device("cpu")
model = model.to(device)  # Move the model to the CPU

# 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):
    try:
        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")
        input_encoding = {key: val.to(device) for key, val in input_encoding.items()}  # Move input tensor to CPU

        # 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

    except Exception as e:
        # Log the error details
        print(f"Error in predict_pptx_content: {e}")
        return {"error": str(e)}

# 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)