app.py

import gradio as gr
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
import torch
import numpy as np
import cv2
from PIL import Image
import pytesseract as tess
from sentence_transformers import SentenceTransformer, util
import io

# save_directory = "spaces/Garvitj/grader"

# # Load the tokenizer from the saved directory
# tokenizer = AutoTokenizer.from_pretrained(save_directory)

# # Load the model from the saved directory
# model = AutoModelForCausalLM.from_pretrained(
#     save_directory,
#     torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
#     device_map="auto" if torch.cuda.is_available() else None
# )

# # Move model to the appropriate device (CPU or CUDA)
# device = "cuda" if torch.cuda.is_available() else "cpu"
# model.to(device)

# print(f"Model and tokenizer loaded from {save_directory}")

tess.pytesseract.tesseract_cmd = r"/app/tesseract.exe"

# Use a pipeline as a high-level helper
pipe = pipeline("text-generation", model="eachadea/vicuna-7b-1.1")

def get_embedding(text):
    return model1.encode(text, convert_to_tensor=True)

def calculate_similarity(text1, text2):
    embedding1 = get_embedding(text1)
    embedding2 = get_embedding(text2)
    similarity = util.pytorch_cos_sim(embedding1, embedding2)
    return similarity.item()

def get_grade(similarity_score):
    if similarity_score >= 0.9:
        return 5
    elif similarity_score >= 0.8:
        return 4
    elif similarity_score >= 0.7:
        return 3
    elif similarity_score >= 0.6:
        return 2
    else:
        return 1

def extract_text_from_image(image):
    # Convert PIL image to RGB format
    image = image.convert('RGB')
    # Use pytesseract to extract text from the image
    text = tess.image_to_string(image)
    return text.strip()

def evaluate_answer(image):
    student_answer = extract_text_from_image(image)
    model_answer = "The process of photosynthesis helps plants produce glucose using sunlight."
    similarity_score = calculate_similarity(student_answer, model_answer)
    grade = get_grade(similarity_score)
    feedback = generate_response("Student's answer: {student_answer}\nTeacher's answer: {model_answer}")
    return grade, similarity_score * 100, feedback

def generate_response(prompt):
    # Generate response from the new model using the pipeline
    response = pipe(prompt, max_length=150, temperature=0.7)
    return response[0]['generated_text']

def gradio_interface(image, prompt):
    grade, similarity_score, feedback = evaluate_answer(image)
    response = generate_response(prompt)
    return grade, similarity_score, feedback, response

# Define Gradio interface
interface = gr.Interface(
    fn=gradio_interface,
    inputs=[gr.Image(type="pil"), gr.Textbox(lines=2, placeholder="Enter your prompt here")],
    outputs=[gr.Label(), gr.Label(), gr.Textbox(), gr.Textbox()],
    live=True
)

if __name__ == "__main__":
    interface.launch()