The cubic sandbox Minecraft provides many opportunities for self-expression. The author of the YouTube channel sammyuri decided to assemble a small language model.
The parameters of his "small" CraftGPT model are 1020 x 260 x 1656 blocks. He clarified that due to these dimensions, he was forced to use the Distant Horizons mod for filming, so some components may look strange (redstone is displayed with a lower level of detail).
To train the model, sammyuri used Python and took the "TinyChat" dataset, which contains "basic conversations in English."
The model "has a vocabulary of 1920 tokens and consists of 6 layers. The context window size is 64 tokens, which is enough for (very) short conversations."