app.py

import gradio as gr
import random
import numpy as np
import torch
from transformers import GPT2LMHeadModel, GPT2Tokenizer

gpt2_model = GPT2LMHeadModel.from_pretrained("gpt2-xl")
gpt2_tokenizer = GPT2Tokenizer.from_pretrained("gpt2-xl")


seed = random.randint(0, 13)
np.random.seed(seed)
torch.random.manual_seed(seed)
torch.cuda.manual_seed(seed)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

text = """All of this is right here, ready to be used
in your favorite pizza recipes."""

def get_story(text):
    print(text)
    input_ids = torch.tensor(gpt2_tokenizer.encode(text, add_special_tokens=True)).unsqueeze(0)  # bs=1

    gpt2_model.to(device)
    gpt2_model.eval()

    outputs = gpt2_model.generate(
        input_ids.to(device),
        max_length=500,
        do_sample=True,
        top_k=20,
        temperature=0.7
    )
    print(f'outputs: {gpt2_tokenizer.decode(outputs[0], skip_special_tokens=True)}')
    return(gpt2_tokenizer.decode(outputs[0], skip_special_tokens=True))

# print(gpt2_tokenizer.decode(outputs[0], skip_special_tokens=True))
# outputs.shape,outputs[0].shape # (torch.Size([1, 500]), torch.Size([500]))
# In the dark night he effortlessly climbed into the spacecraft and closed the ramped door.

input = gr.Textbox(lines=2, placeholder="Start your story here...", label='Story starter')
output = gr.Textbox(label='The Big Story', lines=300)

iface = gr.Interface(fn=get_story,
                      inputs=input,
                      outputs=output,
                      title='The Complete Story',
                      description='Enter the beginning of your story and we will finish it for you.',
                      sample_inputs='In the dark night he effortlessly climbed into the spacecraft and closed the ramped door.'
)
iface.launch()