app.py

import pickle
import json
import torch
import pandas as pd
import gradio as gr
import spaces
from mario_gpt import MarioLM

# Load trained Random Forest model
with open("model.pkl", "rb") as f:
    rf_model = pickle.load(f)

# Load YejinGPT model from Hugging Face
BASE = "shyamsn97/Mario-GPT2-700-context-length"
YEJIN_MODEL_PATH = "YejinJ/LLM"

mario_lm = MarioLM(
    lm_path=YEJIN_MODEL_PATH,
    tokenizer_path=BASE
)

# This function is called by Unity and the Gradio UI
@spaces.GPU
def gradio_generate(jump_count, damage_count, death_count, playtime, coin_count, item_count):
    try:
        X = pd.DataFrame([{
            "jump_count": int(jump_count),
            "damage_count": int(damage_count),
            "death_count": int(death_count),
            "playtime": float(playtime),
            "coin_count": int(coin_count),
            "item_count": int(item_count)
        }])

        prediction = rf_model.predict(X)[0]

        pipes = prediction[0]
        enemies = prediction[1]
        blocks = prediction[2]
        elevation = prediction[3]

        prompt = f"{pipes} pipes, {enemies} enemies, {blocks} blocks, {elevation} elevation"

        device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
        mario_lm.to(device)

        generated_level = mario_lm.sample(
            prompts=[prompt],
            num_steps=1400,
            temperature=1.0,
            use_tqdm=False
        )

        level_data = generated_level.level

        if isinstance(level_data, list):
            level_string = "\n".join(level_data)
        else:
            level_string = str(level_data)

        return prompt, level_string

    except Exception as e:
        return "ERROR", str(e)


demo = gr.Interface(
    fn=gradio_generate,
    inputs=[
        gr.Number(label="jump_count", value=73),
        gr.Number(label="damage_count", value=5),
        gr.Number(label="death_count", value=2),
        gr.Number(label="playtime", value=91),
        gr.Number(label="coin_count", value=13),
        gr.Number(label="item_count", value=3),
    ],
    outputs=[
        gr.Textbox(label="Generated Prompt"),
        gr.Textbox(label="Generated Level String", lines=15),
    ],
    title="YejinGPT Level Generator",
    api_name="generate"
)

demo.queue()
demo.launch(server_name="0.0.0.0", server_port=7860)