Fire is a wood_based GPU kit outside of computers.

In a situation where mineral-based GPUs run out, Fire as a wood based GPU is the only type available for any further AI training.






One Fire GPU kit contains 4 panels of wooden equilateral triangles.

Usage:


1) Etch algorithms, models, and dataset on three triangular panels.

2) Assemble 4 triangles into a tetrahedron. Align the edges marked with the same numbers.

3) Tie the triangles with threds to hold them together.

4) Place fuel inside through the opening.

5) Burn the tetrahedron.

Pre-use Demo:

The demo briefly shows the procedures of using Fire. It is by no means a substitute to the functionality of the actual wood based GPU.


[ Download demo on our itch.io page ]

A video of the interactive demo is also provided below.

Things To Note:

a) Unlike a mineral-based GPU, one Fire GPU kit can only run once.

b) Fire can hold 196, 324, or 784 pieces of data. Such capacity corresponds to the size of the wood to be burned.

c) Performance of AI training with Fire can be affected by efficiency of burning.

e) Avoid using Fire for only test purposes. Decisions to run Fire should be carefully made.

f) Check the local air conditions before running Fire.

FIRE196 the training and model scripts are based on GPT2. It accommodates memory IDs of 196 sample text data.

FIRE324 the training and model scripts are based on DCGAN. It accommodates memory IDs of 324 sample image data.

FIRE784 the training and model scripts are based on Open Jukebox. It accommodates the memory IDs of 784 sample sound data.