Update app.py

app.py

@@ -11,11 +11,12 @@ from transformers import pipeline
 import gradio as gr
 from fastapi.responses import RedirectResponse
 import numpy as np
+
 # Initialize FastAPI
 app = FastAPI()
 
-# Load AI Model for Question Answering
-qa_pipeline = pipeline("text2text-generation", model="google/flan-t5-large", tokenizer="google/flan-t5-large", use_fast=True)
+# Load AI Model for Question Answering (Summarization-based approach)
+qa_pipeline = pipeline("summarization", model="facebook/bart-large-cnn", tokenizer="facebook/bart-large-cnn")
 
 # Load Pretrained Object Detection Model (Torchvision)
 model = fasterrcnn_resnet50_fpn(pretrained=True)
@@ -60,7 +61,7 @@ def extract_text_from_excel(excel_file):
             text.append(" ".join(map(str, row)))
     return "\n".join(text)
 
-# Function to perform object detection using Torchvision
+# Function to extract text from image
 def extract_text_from_image(image_file):
     if isinstance(image_file, np.ndarray):  # Check if input is a NumPy array
         image = Image.fromarray(image_file)  # Convert NumPy array to PIL image
@@ -70,7 +71,8 @@ def extract_text_from_image(image_file):
     reader = easyocr.Reader(["en"])
     result = reader.readtext(np.array(image))  # Convert PIL image back to NumPy array
     return " ".join([res[1] for res in result])
-# Function to answer questions based on document content
+
+# Function to answer questions based on document content using BART summarization
 def answer_question_from_document(file, question):
     file_ext = file.name.split(".")[-1].lower()
     
@@ -89,10 +91,10 @@ def answer_question_from_document(file, question):
         return "No text extracted from the document."
     
     truncated_text = truncate_text(text)
-    input_text = f"Question: {question} Context: {truncated_text}"
-    response = qa_pipeline(input_text)
+    input_text = f"Context: {truncated_text} Question: {question}"
+    response = qa_pipeline(input_text, max_length=100, min_length=30, do_sample=False)
     
-    return response[0]["generated_text"]
+    return response[0]["summary_text"]
 
 # Function to answer questions based on image content
 def answer_question_from_image(image, question):
@@ -101,10 +103,10 @@ def answer_question_from_image(image, question):
         return "No meaningful content detected in the image."
     
     truncated_text = truncate_text(image_text)
-    input_text = f"Question: {question} Context: {truncated_text}"
-    response = qa_pipeline(input_text)
+    input_text = f"Context: {truncated_text} Question: {question}"
+    response = qa_pipeline(input_text, max_length=100, min_length=30, do_sample=False)
     
-    return response[0]["generated_text"]
+    return response[0]["summary_text"]
 
 # Gradio UI for Document & Image QA
 doc_interface = gr.Interface(