app.py

import gradio as gr
import torch
import asyncio
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

# Load model and tokenizer
model_name = "hassaanik/grammar-correction-model"
tokenizer = AutoTokenizer.from_pretrained(model_name)

# Use GPU if available, otherwise fall back to CPU
device = "cuda" if torch.cuda.is_available() else "cpu"
model = AutoModelForSeq2SeqLM.from_pretrained(model_name).to(device)

# Use FP16 for faster inference on GPU
if torch.cuda.is_available():
    model.half()

# Async grammar correction function
async def correct_grammar_async(text):
    # Tokenize input and move it to the correct device (CPU/GPU)
    inputs = tokenizer.encode(text, return_tensors="pt", max_length=512, truncation=True).to(device)

    # Asynchronous operation to run grammar correction
    outputs = await asyncio.to_thread(model.generate, inputs, max_length=512, num_beams=5, early_stopping=True)
    
    # Decode output and return corrected text
    corrected_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return corrected_text

# Gradio interface function to handle input and output
def correct_grammar_interface(text):
    corrected_text = asyncio.run(correct_grammar_async(text))
    return corrected_text

# Create Gradio Interface
with gr.Blocks() as grammar_app:
    gr.Markdown("<h1>Async Grammar Correction App</h1>")
    
    with gr.Row():
        input_box = gr.Textbox(label="Input Text", placeholder="Enter text to be corrected", lines=4)
        output_box = gr.Textbox(label="Corrected Text", placeholder="Corrected text will appear here", lines=4)

    submit_button = gr.Button("Correct Grammar")
    
    # When the button is clicked, run the correction process
    submit_button.click(fn=correct_grammar_interface, inputs=input_box, outputs=output_box)

# Launch the app
if __name__ == "__main__":
    grammar_app.launch()