A fork of AIDungeon2 with numerous improvements. Now supporting GPT-Neo.

Also take a look at AIDungeonPastes for some drawn gameplay examples.

• GPT-Neo support
• Complete rewrite of the user interface
  • Colored text
  • Suggested actions
  • Console bell dings when the AI finishes
  • Much improved prompt selection
  • Ability to save custom prompts
• Half precision floating point using significantly less GPU memory
• Repetition Penalty to reduce AI looping behavior
• A much larger library of fan made starting prompts
• Challenge added with your actions not always succeeding
• A simple config file
• Lots of changes to story history sampling/truncation to hopefully stay on track with longer games
• Eventually hope to improve the AI itself, but this will take some time

Officially we only support local installs. We encourage and recommend installing and running the game locally. However since the beginning most people have been playing it for free on Google's servers through their Colab platform. Allegedly it requires no effort to get started. Try this link and go to the 4chan threads for help and info.

To play with GPU acceleration, you need an Nvidia GPU, although some anons claim to have gotten it working with an AMD with ROCm in Linux. The original "XL" 1558M parameter model requires at least 4GB of VRAM. Smaller models may consume much less. On CPU, response times vary from about a minute on the XL GPT-2 1558M model, which is slow but usable, to about 6 minutes on GPT-Neo.

After any of the following install steps, you must get one of the models.

1. Download this repo. Github has a green download button to the top right that looks like: [⤓ Code]. Click it, select "Download Zip", then unzip it to a folder. Or you can use the git command git clone --depth=1 "https://github.com/cloveranon/Clover-Edition/" if you have git installed.
2. Go to your Clover Edition folder.
3. Run install.bat and follow the on-screen instructions.

• Users have reported that some anti-virus (specifically Kaspersky) isn't happy with the install.bat script. Please whitelist or temporarily disable anti-virus when installing.
• You can partially uninstall by deleting the venv/ folder, and fully uninstall by just deleting the entire Clover Edition folder.

1. Visit the repo for the Clover Edition installer project and download the latest executable.
2. Navigate to your folder for your downloads
3. Run the executable (Clover.Edition.Installer.exe) and follow on-screen prompts.

1. Install Python or use your package manager (e.g. sudo apt-get install python3 or sudo yum install python3 or something).
2. Download this repo. Github has a green download button to the top right that looks like: [⤓ Code]. Click it, select "Download Zip", then unzip it to a folder. Or you can use the git command git clone --depth=1 "https://github.com/cloveranon/Clover-Edition/" if you have git installed.
3. Go to your Clover Edition folder.
4. Run sh install.sh and follow the on-screen instructions.

1. Install Python. The installer should install pip and it should add it to your PATH automatically. Make sure you have the relevant options selected if the installer gives you any options.
2. Download this repo. Github has a green download button to the top right that looks like: [⤓ Code]. Click it, select "Download Zip", then unzip it to a folder. Or you can use the git command git clone --depth=1 "https://github.com/cloveranon/Clover-Edition/" if you have git installed.
3. Open a command line or terminal and go to your Clover Edition folder.
4. Run: pip --no-cache-dir install -r requirements/requirements.txt
5. Install PyTorch (you must do one of the following):
   • If you're using an Nvidia GPU with CUDA, run: pip install -r requirements/cuda_requirements.txt
   • If you're planning on just using your CPU, run: pip install -r requirements/cpu_requirements.txt

You can have multiple models installed, but you need at least one.

* For EleutherAI's GPT-Neo-2.7B, Download only pytorch_model.bin and make sure it's named that, put it into a new folder named gpt-neo-2.7b (see below for the structure), then copy config.json, merges.txt, and vocab.json from one of finetuneanon's models and put them in the same folder.

16-Bit models are half the size, and load 20 times faster (probably because they run out of RAM and hit the swap space). I found the original AID2 model (pytorch-gpt2-xl-aid2-v5) took 12.6 minutes to load, while the 16 bit version took 37 seconds.

Once downloaded, your model folder should look like this:

    ./models
    └── <MODEL-NAME>
        ├── config.json
        ├── merges.txt
        ├── pytorch_model.bin
        └── vocab.json

• Windows 10: double-click play.bat
• Windows 7+: python launch.py
• Linux: bash play.sh
• OS-agnostic install: python launch.py

Install Windows Terminal (recommended), cmder, or hyper, and use that as your terminal. If you installed Clover Edition with install.bat, you might already have Windows Terminal installed.

I threw together a quick page of some tips here. I plan to throw any links to interesting datasets or guides for training and fine-tuning the AI there. Please send me anything interesting.

Fine-tuning is not currently a push button thing and requires some minimal technical ability. Most people are using the program gpt-simple. You may have more luck with the much more advanced Huggingface-Transformers program that we use to power Clover-Edition. This seems to be their documentation on fine-tuning.

I have made the convert_gpt2_model.py script an idiot proof simple way of quickly converting tensorflow models to PyTorch models. Just run it on the folder containing a tensorflow model and you will get a PyTorch model. You can use the --full flag to get a full 32bit model, but do try 16bit models as they will be potentially half the size for the same accuracy.

See the test-models.py script to test the accuracy of 16 bit mode if you doubt the chad 16BIT models. My tests were well within expectations:

See that github issues page? Post any questions, requests, or problems there if you are willing to create a github account. Unless MicroAndSoft deletes us. Otherwise see:

• Website: 4chan Discussion
• Email: [email protected]

Contributions are more than welcome. You can fork the thing and send a pull request from your fork.

GPT-Neo, finetuneanon's horni and horni-ln:

magnet:?xt=urn:btih:31d956ff4a248dcf914b1b7e474cbac02d70d6a4&dn=gtp-neo-horni

GPT-2, AIDungeon2 Clover Edition (16-bit):

magnet:?xt=urn:btih:9c774c51bac64e5bffd481187b86e834839b4141&dn=model%5Fv5%5Fpytorch%5F16bit&tr=udp%3A%2F%2Ftracker.coppersurfer.tk%3A6969%2Fannounce&tr=udp%3A%2F%2Ftracker.leechers-paradise.org%3A6969%2Fannounce&tr=udp%3A%2F%2Ftracker.opentrackr.org%3A1337%2Fannounce&tr=udp%3A%2F%2Ftracker.pomf.se%3A80%2Fannounce&tr=udp%3A%2F%2Ftracker.openbittorrent.com%3A80%2Fannounce&tr=udp%3A%2F%2Fp4p.arenabg.com%3A1337%2Fannounce&tr=udp%3A%2F%2F9.rarbg.me%3A2710%2Fannounce&tr=udp%3A%2F%2F9.rarbg.to%3A2710%2Fannounce

GPT-2, AIDungeon2 Clover Edition (32-bit):

magnet:?xt=urn:btih:17dcfe3d12849db04a3f64070489e6ff5fc6f63f&dn=model_v5_pytorch&tr=udp%3a%2f%2ftracker.opentrackr.org%3a1337%2fannounce&tr=udp%3a%2f%2fopen.stealth.si%3a80%2fannounce&tr=udp%3a%2f%2fp4p.arenabg.com%3a1337%2fannounce&tr=udp%3a%2f%2ftracker.coppersurfer.tk%3a6969%2fannounce&tr=udp%3a%2f%2ftracker.cyberia.is%3a6969%2fannounce&tr=udp%3a%2f%2ftracker.moeking.me%3a6969%2fannounce&tr=udp%3a%2f%2f9.rarbg.me%3a2710%2fannounce&tr=udp%3a%2f%2ftracker3.itzmx.com%3a6961%2fannounce