| # Neo Ar Model Specification | |
| Neo Ar is an Arabic-only System One decision model with a larger encoder than | |
| the English Neo model. | |
| - 12 Transformer layers. | |
| - Hidden size 768. | |
| - 12 attention heads. | |
| - Intermediate size 3,072. | |
| - Maximum context 1,024 tokens. | |
| - Up to 64 choice options. | |
| - Up to 7 score levels. | |
| - Binary `noul` decision. | |
| - Custom Arabic WordPiece tokenizer. | |
| The released architecture contains 86,393,160 trainable parameters with a | |
| 644-token Arabic vocabulary. | |
| The final parameter count is measured from the released checkpoint after the | |
| Arabic tokenizer vocabulary is finalized. The full source and training process | |
| are documented at https://github.com/tokenaii/Neo-Ar. | |