| # AlphaViT | |
| This repository provides the model weights for AlphaViT, AlphaViD, AlphaVDA, and Alphazero. | |
| ## Models | |
| This repository contains the weights for the following configurations: | |
| - **Single-Task Models**: Models trained individually for specific games such as Connect4, Gomoku, and Othello. | |
| - **Multi-Task Models**: Unified models capable of handling multiple games with varying board sizes. | |
| Each filename includes digits indicating the training iteration. For example, checkpoint_1000.model represents the model saved after 1000 training iterations. | |
| ## training_dataset_for_the_first_iteration/ | |
| Initial training datasets for various games, organized by game type. Files include: | |
| - Board states (*.boards.npy) | |
| - Move probabilities (*.probs.npy) | |
| - Value outputs (*.vs.npy) | |
| ### GitHub Repository | |
| The full source code for AlphaViT, including training scripts and detailed implementation, is available on GitHub: | |
| [https://github.com/KazuhisaFujita/AlphaViT](https://github.com/KazuhisaFujita/AlphaViT) | |
| ### Related Paper | |
| For detailed information on the methodology and experiments, refer to the research paper: | |
| ["Flexible Game-Playing AI with AlphaViT: Adapting to Multiple Games and Board Sizes"](https://arxiv.org/abs/2408.13871). | |