KANama

Fusing Kolmogorov–Arnold Networks with Meta's Llama model for next-level AI performance and versatility.

Usage

Install via PyPi:

pip install kanama

Available modeels

• KANamav1: The basic Llama3.1 model mith a KAN model instead of a basic MLP.
• KANamav2: More optimized then v1.
• KANamav3: A Llama3.1 model mith a KAN and a dynaicaly adjusting Softmax Temperature.
• KANamav4: More optimized then v3.
• KANaMoEv1: v4 but with a MoE architecture.

Examples

For a good introduction, you can look into the example files.

Citing KANama

The KANama software suite was developed by Gökdeniz Gülmez. If you find KANama useful in your research and wish to cite it, please use the following BibTex entry:

@software{
  KANama,
  author = {Gökdeniz Gülmez},
  title = {{KANama}: Fusing Kolmogorov–Arnold Networks with Meta's Llama model for next-level AI performance and versatility.},
  url = {https://github.com/Goekdeniz-Guelmez/KANama},
  version = {2.5.6},
  year = {2024},
}