app.py

import gradio as gr
from model import *

DEVICE = "cuda" if torch.cuda.is_available() else "cpu"

model = GPTLanguageModel().to(DEVICE)

model.load_state_dict(torch.load("mini-gpt.pth",map_location=DEVICE), strict=False)
model.eval()
answer = decode(model.generate(context, max_new_tokens=3000)[0].tolist())

def display(text,number):
    combined_text = text + answer[:number + 1]
    return combined_text

input_box = gr.Textbox()
input_slider = gr.Slider(minimum=500, maximum=2000, default=500, label="Select the maxium number of tokens/words:",step=100)
output_text = gr.Textbox() 
gr.Interface(fn=display, inputs=[input_box,input_slider], outputs=output_text).launch()