app.py

import sentencepiece as spm
import numpy as np
import tensorflow as tf
from tensorflow.keras.preprocessing.sequence import pad_sequences
from valx import detect_profanity, detect_hate_speech
import gradio as gr

sp = spm.SentencePieceProcessor()
sp.Load("dungen_dev_preview.model")

model = tf.keras.models.load_model("dungen_dev_preview_model.keras")

max_seq_len = 25

def generate_text(seed_text, next_words=30, temperature=0.5):
    seed_text = seed_text.strip().lower()

    if "|" in seed_text:
        gr.Warning("The prompt should not contain the '|' character. Using default prompt.")
        seed_text = 'game name | '
    elif detect_profanity([seed_text], language='All'):
        gr.Warning("Profanity detected in the prompt, using the default prompt.")
        seed_text = 'game name | '
    elif (hate_speech_result := detect_hate_speech(seed_text)) and hate_speech_result[0] in ['Hate Speech', 'Offensive Speech']:
        gr.Warning('Harmful speech detected in the prompt, using default prompt.')
        seed_text = 'game name | '
    else:
        seed_text += ' | '

    generated_text = seed_text
    if generated_text != 'game name | ': # only generate if not the default prompt
        for _ in range(next_words):
            token_list = sp.encode_as_ids(generated_text)
            token_list = pad_sequences([token_list], maxlen=max_seq_len - 1, padding='pre')
            predicted = model.predict(token_list, verbose=0)[0]

            predicted = np.asarray(predicted).astype("float64")
            predicted = np.log(predicted + 1e-8) / temperature
            exp_preds = np.exp(predicted)
            predicted = exp_preds / np.sum(exp_preds)

            next_index = np.random.choice(len(predicted), p=predicted)
            next_token = sp.id_to_piece(next_index)
            generated_text += next_token

            if next_token.endswith('</s>') or next_token.endswith('<unk>'):
                break

    decoded = sp.decode_pieces(sp.encode_as_pieces(generated_text))
    decoded = decoded.replace("</s>", "").replace("<unk>", "").strip()

    if '|' in decoded:
        decoded = decoded.split('|', 1)[1].strip()

    if any(detect_profanity([decoded], language='All')) or (hate_speech_result := detect_hate_speech(decoded)) and hate_speech_result[0] in ['Hate Speech', 'Offensive Speech']:
        gr.Warning("Flagged potentially harmful output.")
        decoded = 'Flagged Output'

    return decoded

demo = gr.Interface(
    fn=generate_text,
    inputs=[
        gr.Textbox(label="Prompt", value="a female character name", max_lines=1),
        gr.Slider(1, 100, step=1, label='Next Words', value=30),
        gr.Slider(0.1, 1, value=0.5, label='Temperature', info='Controls randomness of generation, higher values = more creative, lower values = more probalistic')
    ],
    outputs=gr.Textbox(label="Generated Names"),
    title='Dungen Dev - Name Generator',
    description='A prompt-based name generator for game developers. Dungen Dev is an experimental model, and may produce outputs that are inappropriate, biased, or potentially harmful and inaccurate. Caution is advised.',
    examples=[
        ["a male character name", 30, 0.5],
        ["a futuristic city name", 30, 0.5],
        ["an item name", 30, 0.5],
        ["a dark and mysterious forest name", 30, 0.5],
        ["an evil character name", 30, 0.5]
    ]
)

demo.launch()