| # Neo Model Specification | |
| Neo is an English-only System One decision model. | |
| - Bidirectional Transformer encoder. | |
| - 41,391,654 trainable parameters (approximately 41.4M). | |
| - 8 layers, hidden size 512, 8 attention heads. | |
| - Intermediate size 2048. | |
| - Maximum context 1,024 tokens. | |
| - Tokenizer: `bert-base-uncased`. | |
| - Decision heads: `choice`, `score`, and `noul`. | |
| - Choice head: up to 32 options. | |
| - Score head: up to 5 ordered levels. | |
| - Noul head: binary probability. | |
| The model is trained from scratch; the tokenizer is standard pretrained | |
| English vocabulary, not a custom tokenizer. | |
| Training data and configuration are documented in the source repository: | |
| https://github.com/tokenaii/Neo/blob/main/docs/MODEL_SPECIFICATION.md | |