Licence
The model weights are non-commercial. The code, the model design and the training process are MIT.
| what | licence |
|---|---|
| Trained weights in this repository: the core model and every add-on module ("arm"), in every precision and checkpoint | CC BY-NC-SA 4.0 (attribution, non-commercial, share-alike) |
| Code files in this repository (modeling, configuration, loaders) | MIT (text below); the amoe-lora code in the snapshots: Apache-2.0 (NOTICE) |
| The model design and the complete training process: the training library, recipes, configs and reports | MIT, at github.com/AbstractEyes/alephllm |
Why the weights are non-commercial
The Beatrix line is trained from scratch on openly licensed text. Some of that text carries terms that the trained weights keep:
- child-directed speech from CHILDES, via TalkBank, is licensed CC BY-NC-SA 3.0, and TalkBank's rules exclude its use in commercial models;
- Wikipedia and WikiText are licensed CC BY-SA (share-alike).
So the weights are released for non-commercial use, with attribution and share-alike. Anything you make with the weights under these terms must carry this licence file. The code, the design and the full recipe stay MIT: you can reproduce or adapt the process for any purpose, including training your own model on data you are licensed to use.
Training sources
The sources of mini-beatrix-3, the broadest mix of the line. Earlier generations trained on subsets of these; each generation's manifest in the training repository lists its own.
| source | licence |
|---|---|
| FineWeb-Edu | ODC-By 1.0 (and the Common Crawl terms of use) |
| WikiText-103 | CC BY-SA 3.0 / GFDL |
| Wikipedia, simple and English (2023-11-01) | CC BY-SA / GFDL |
| CHILDES child-directed speech via AO-CHILDES | CC BY-NC-SA 3.0 (TalkBank) |
| TinyStories | CDLA-Sharing-1.0 |
| Cosmopedia v2 | ODC-By |
| SODA | CC BY 4.0 |
| Social IQa | CC BY 4.0 |
| bAbI tasks v1.2 | CC BY 3.0 |
| ProofWriter | CC BY (as reported; the upstream terms are being confirmed) |
| ATOMIC 2020 | CC BY |
| OpenGloss v1.3 definitions | CC BY 4.0 |
| Project Gutenberg English | public-domain texts |
| Procedural generators and identity text from the training library | MIT |
Some of these sources are machine-generated by other providers' models (TinyStories, SODA, OpenGloss, Cosmopedia); their dataset cards describe how.
MIT licence (code, design, process)
Copyright (c) 2026 AbstractPhil
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.