app.py

# Reference:
# https://huggingface.co/spaces/Sagar23p/mistralAI_chatBoat

import gradio as gr
from paddleocr import PaddleOCR, draw_ocr
import asyncio
import requests
from huggingface_hub import InferenceClient
import os

API_TOKEN = os.environ.get('HUGGINGFACE_API_KEY')

API_URL = "https://api-inference.huggingface.co/models/meta-llama/Meta-Llama-3-8B-Instruct"
headers = {"Authorization": "Bearer " +API_TOKEN}


def query(question):
    client = InferenceClient("meta-llama/Meta-Llama-3-8B-Instruct", headers=headers)
    messages = [
        {
            "role": "system",
            "content": "You are a helpful and honest assistant. Please, respond concisely and truthfully.",
        },
        {
            "role": "user",
            "content": question,
        },
    ]
    output = client.chat_completion(messages, model="meta-llama/Meta-Llama-3-8B-Instruct", max_tokens=1000)  
    if output.choices[0].message['content'].find('Yes')>=0:
        messages+=[output.choices[0].message]
        messages+=[{"role": "user",
            "content": "What is the mistake and what is the correct sentence?"}]
        output = client.chat_completion(messages, model="meta-llama/Meta-Llama-3-8B-Instruct", max_tokens=1000)  
    return output.choices[0].message['content']

def image2Text(image:str, langChoice:str):
    ocr = PaddleOCR(use_angle_cls=True, lang=langChoice)  # need to run only once to download and load model into memory
    img_path = image
    result = ocr.ocr(img_path, cls=True)
    text = ""
    for idx in range(len(result)):
        res = result[idx]
        for line in res:
            import re
            # remove pinyin if it's Chinese
            if langChoice=="ch":
                #t = re.sub('[a-z0-9.]', '', line[1][0])
                t = re.sub('[a-z]', '', line[1][0])
                t = re.sub('[0-9]\.', '', t)
                t = t.replace(" ", "")
                t = t.replace("（）", "")
                t = t.replace("（)", "")
                t = t.replace("( )", "")
                t = t.replace("()", "")            
                if t!="":
                    text +=((t) + "\n")             
            else:
                print(line)
                t = line[1][0]
                t = re.sub('Term [0-9] Spelling', '', t)
                t = re.sub('Page [0-9]', '', t)
                if t!="":
                    text += (t + "\n")
    
    text = text.replace("\n"," ").replace(".",".\n")
                            
    return text

def text2PrevMistake(recognized_text, langChoice:str, current_line, session_data):
    if len(session_data) == 0 or session_data[0] == 0 or session_data[0] == 1:
        session_data = []
    else:
        session_data = [session_data[0]-2]

    return text2NextMistake(recognized_text, langChoice, current_line, session_data)
    
def text2NextMistake(recognized_text, langChoice:str, current_line, session_data):

    lines = recognized_text.split("\n")
    while 1:
        if len(lines) == 0:
            return current_line, "No mistake. Empty text.", session_data
        elif len(session_data) == 0:
            session_data = [0]
            current_line = lines[session_data[0]]
        elif session_data[0] + 1 >= len(lines):
            session_data = []
            return current_line, "No more mistake. End of text", session_data
        else:
            session_data = [session_data[0]+1]
            current_line = lines[session_data[0]]
            
        question = f"Only answer Yes or No. Is there grammatical or logical mistake in the sentence: {current_line}"
        correction_text = query(question)

        if correction_text.find("No") == 0:
            continue
        else:
            break
            
    return current_line, correction_text, session_data
    
with gr.Blocks() as demo:
    gr.HTML("""<h1 align="center">Composition Corrector</h1>""")
    session_data = gr.State([])

    with gr.Row():
        with gr.Column(scale=1):
            upload_image = gr.Image(height=400,width=400, value = "compo.jpg")
            langChoice = gr.Radio(["en", "ch"], value="en", label="Select lanaguage: 'ch' for Chinese, 'en' for English", info="")
        with gr.Column(scale=3):
            recognized_text = gr.Textbox(show_label=False, placeholder="composition", lines=15)
    toText = gr.Button("Convert image to text")

    current_line = gr.Textbox(show_label=False, placeholder="current line", lines=1)
    correction_text = gr.Textbox(show_label=False, placeholder="corrections...", lines=15)

    with gr.Row():
        with gr.Column(scale=1):
            toPrevMistake = gr.Button("Find prev mistake", variant="primary")
        with gr.Column(scale=1):
            toNextMistake = gr.Button("Find next mistake", variant="primary")
    

    
    toText.click(
        image2Text,
        [upload_image, langChoice],
        [recognized_text],
        #show_progress=True,

    )
    toNextMistake.click(text2NextMistake , [recognized_text, langChoice, current_line, session_data], [current_line, correction_text, session_data])
    toPrevMistake.click(text2PrevMistake , [recognized_text, langChoice, current_line, session_data], [current_line, correction_text, session_data])

demo.queue().launch(share=False, inbrowser=True)