| --- |
| license: other |
| license_name: tokenai-neo-dataset-license |
| license_link: https://huggingface.co/datasets/tokenaii/Neo-dataset/blob/main/legal/DATASET_LICENSE.md |
| language: |
| - en |
| task_categories: |
| - text-classification |
| tags: |
| - system-one |
| - decision-model |
| - rlcd |
| - synthetic |
| - tool-routing |
| --- |
| |
| # NEO - Decision Model Dataset |
|
|
| <p align="center"><img src="assets/neo-cover.png" alt="NEO - Decision Model Dataset" width="360"></p> |
|
|
| TokenAI is a non-profit startup founded in 2025 by Assem Sabry, based in Alexandria, Egypt. The dataset, |
| model, source code, documentation, and related materials are owned by TokenAI. |
| Contact: info@tokenai.llc · https://tokenai.llc |
|
|
| Canonical reference: `tokenaii/Neo` — https://huggingface.co/tokenaii/Neo |
|
|
| Complete source repository: [github.com/tokenaii/Neo](https://github.com/tokenaii/Neo). |
| It contains the complete project source code, training code, synthetic-data |
| generation code, evaluation and benchmark scripts, configurations, examples, |
| documentation, licenses, and reproducibility materials. This repository |
| contains the dataset files, manifest, provenance information, and dataset legal |
| notices. |
|
|
| Synthetic, schema-driven decision records for training and evaluating Neo, a System One decision model. |
|
|
| The associated Neo checkpoint contains exactly 41,391,654 trainable parameters |
| (approximately 41.4 million), as verified from the published `model.safetensors` |
| metadata and the model architecture. |
|
|
| ## Contents |
|
|
| - 1,000,000 individual decisions. |
| - 400,000 state records. |
| - `choice`, `score`, and `noul` questions. |
| - English-only state text, questions, labels, and tool descriptions. |
| - Tool routing, urgency, human-review, ambiguity, distractor tools, option permutations, and unknowable cases. |
|
|
| ## Provenance |
|
|
| The records are generated by the `neo-synthetic` generator in the Neo repository. No rows from public datasets are copied into the generated corpus. Public datasets are used only for schema design, sanity checks, or separately documented evaluation. |
|
|
| Generator: `neo-english-synthetic`. |
| Generator seed: `20261003`. |
|
|
| The complete source file used by training has SHA-256 |
| `3eff0bcf1ced0e748698fefefcf8069d2cc997888d7568fddb5ccd6a1a651c50`. |
| The four files under `data/` are a byte-preserving concatenation of that |
| training source. |
|
|
| ## Splits |
|
|
| - `train`: 799,882 decisions |
| - `calibration`: 100,138 decisions |
| - `test`: 99,980 decisions |
|
|
| The calibration split is reserved for probability calibration. The test split |
| is reserved for final evaluation and must not be used for fitting. |
|
|
| ## Intended use and limitations |
|
|
| This dataset is a research corpus for typed decision models. It is not a substitute for domain-specific validation, human review, or safety testing. Synthetic distributions can contain template artifacts and must not be treated as real user behavior. |
|
|
| ## Identity and license |
|
|
| This dataset must remain identified as the **TokenAI Neo Decision Model Dataset** and must |
| retain the canonical reference `tokenaii/Neo`. Rebranding, renaming, |
| white-labeling, redistributing, rehosting, or publishing it under another |
| identity is prohibited. The dataset must not be used without preserving its |
| reference and provenance. See [the dataset license](legal/DATASET_LICENSE.md) |
| and [the ownership notice](legal/NOTICE.md). |
|
|
| Commercial, monetized, paid, sponsored, advertising-supported, production, |
| client, resale, or financially beneficial use is forbidden. Any model or |
| experiment trained with this dataset must clearly state that it used |
| `tokenaii/Neo-dataset`. |
|
|
| Third-party datasets referenced by the project are not included in this release |
| and retain their own terms. |
|
|
