ai-text-adventure

Until I get a proper training environment (since graphic card prices are through the roof) all development is paused.

A text adventure game example based on custom pretrained GPT2. 한국어 매뉴얼은 README_KR.md를 참조해 주세요.

This program is developed actively, and may not be stable.

Running this game

Pre-requisites : Working python install

Recommended : python 3.8.5 and a nvidia graphics card(supports CUDA 10.1) with more than 6GB VRAM

1. clone this repo
2. python3 -m venv aita
3. source aita/bin/activate
4. pip install -r requirements.txt
5. (only for macOS 12.0+) Install tensorflow from here.
6. python3 main.py

Using the framework

Edit the following files at your leisure -

• actions.py for custom action calls
• constants.py for project-wide constants (i.e. API Endpoint)
• customclass.py for python classes used in-game (i.e. characters)
• errors.py for custom error handling
• generator.py for text generation (HuggingFace transformers & KoGPT compatible)
• interface.py for user interface (terminal)
• translation.py for translations
• utils.py for utility functions (i.e. custom sort)

Using translation.py

In translation.py, Each language has class of string constants for translation. Ideally, you copy the template and change the word, word order, etc. To make a custom translation, edit the initialize_translation() of translation.py to return the language class when a standard language code (en/ko/...) is input.

When adding another language, make sure the appropriate model is added in main.py! The program will not load properly without it!

Adding custom model

Modify the below template for your model.

if flags.LANG == LANG_CODE:
        flags.model_path = os.path.join(SCRIPT_PATH,'model',LANG_CODE)
        if os.path.exists(os.path.join(SCRIPT_PATH,'model',LANG_CODE,YOUR_MODEL_FILENAME)):
            flags.model_type = 'pt' # If model is trained using HF transformers
            NO_MODEL = False
        elif os.path.exists(os.path.join(SCRIPT_PATH,'model',LANG_CODE,YOUR_MODEL_FILENAME)):
            flags.model_type = 'tf' # If model is trained using the KoGPT2-train script(https://github.com/ksjae/KoGPT2-train)
            NO_MODEL = False

LICENSE

Any use of resources in this repository is possible, as long as it's for non-commercial purposes. This does not override other licenses set to parts of this repository such as corpora or initial story generator contents.

LICENSES by parts

The Korean GPT2: Public Domain

Initial Story Generator(EN): Non-commercial

Initial Story Generator (Korean Translation): Non-commercial, Attribution