|
Download README.md from tokenaii/Neo-Ar: direct link, hf CLI and curl.
- Browser
- Download file 2.11 kB
-
https://huggingface.co/tokenaii/Neo-Ar/resolve/main/README.md
- Command line
-
hf download hf://tokenaii/Neo-Ar/README.md
-
curl -L -o README.md https://huggingface.co/tokenaii/Neo-Ar/resolve/main/README.md
2.11 kB
| license: other | |
| license_name: tokenai-neo-ar-model-license | |
| license_link: https://huggingface.co/tokenaii/Neo-Ar/blob/main/licenses/MODEL_LICENSE.md | |
| library_name: pytorch | |
| pipeline_tag: text-classification | |
| language: | |
| - ar | |
| tags: | |
| - system-one | |
| - decision-model | |
| - rlcd | |
| - tool-routing | |
| - arabic | |
| - arabic-only | |
| # Neo Ar — Arabic Decision Model | |
| <p align="center"><img src="assets/neo-ar-cover.png" alt="Neo Ar — Arabic Decision Model" width="360"></p> | |
| Neo Ar is an Arabic-only TokenAI System One decision model. It selects bounded | |
| actions from declared candidate sets and returns typed probabilities for | |
| `choice`, `score`, and `noul`. It does not generate free-form conversation. | |
| ## Language policy | |
| The model weights, tokenizer, training records, labels, questions, tool | |
| descriptions, and evaluation cases are Arabic-only. This model card and the | |
| repository documentation are English for accessibility; English is not a | |
| supported model language. | |
| ## Architecture target | |
| - 12-layer bidirectional Transformer encoder. | |
| - Hidden size 768; 12 attention heads. | |
| - Intermediate size 3,072; maximum context 1,024 tokens. | |
| - Up to 64 choice options and 7 ordered score levels. | |
| - Binary `noul` head and up to 12 questions per request. | |
| - 86,393,160 trainable parameters with the released 644-token vocabulary. | |
| ## Links | |
| - Dataset: https://huggingface.co/datasets/tokenaii/Neo-Ar-Dataset | |
| - Complete source and training repository: https://github.com/tokenaii/Neo-Ar | |
| The GitHub repository contains the complete source code, Arabic data generator, | |
| tokenizer construction, training code, evaluation, benchmark results, and | |
| training history. This model repository contains release artifacts and public | |
| documentation only. | |
| ## Status and limitations | |
| This is a research model until held-out Arabic benchmarks demonstrate reliable | |
| tool routing, calibration, option-order robustness, and unseen-tool behavior. | |
| Do not use it for unsupervised high-impact or irreversible decisions. | |
| TokenAI is a non-profit startup founded in 2025 by Assem Sabry, based in | |
| Alexandria, Egypt. Contact: info@tokenai.llc — https://tokenai.llc. | |