Licence: weights CC BY-NC-SA 4.0 (non-commercial); code, design and training process MIT

LICENSE.md

@@ -0,0 +1,65 @@
+# 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](https://creativecommons.org/licenses/by-nc-sa/4.0/) (attribution, non-commercial, share-alike) |
+| Code files in this repository (modeling, configuration, loaders) | MIT (text below) (no code files are kept in this repository) |
+| The model design and the complete training process: the training library, recipes, configs and reports | MIT, at [github.com/AbstractEyes/alephllm](https://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](https://huggingface.co/AbstractPhil/alephllm-mini-beatrix-training)
+lists its own.
+
+| source | licence |
+|---|---|
+| [FineWeb-Edu](https://huggingface.co/datasets/HuggingFaceFW/fineweb-edu) | ODC-By 1.0 (and the Common Crawl terms of use) |
+| [WikiText-103](https://huggingface.co/datasets/Salesforce/wikitext) | CC BY-SA 3.0 / GFDL |
+| [Wikipedia, simple and English](https://huggingface.co/datasets/wikimedia/wikipedia) (2023-11-01) | CC BY-SA / GFDL |
+| CHILDES child-directed speech via [AO-CHILDES](https://huggingface.co/datasets/deven367/babylm-100M-aochildes) | CC BY-NC-SA 3.0 ([TalkBank](https://talkbank.org/0share/rules.html)) |
+| [TinyStories](https://huggingface.co/datasets/roneneldan/TinyStories) | CDLA-Sharing-1.0 |
+| [Cosmopedia v2](https://huggingface.co/datasets/HuggingFaceTB/smollm-corpus) | ODC-By |
+| [SODA](https://huggingface.co/datasets/allenai/soda) | CC BY 4.0 |
+| [Social IQa](https://huggingface.co/datasets/allenai/social_i_qa) | CC BY 4.0 |
+| [bAbI tasks v1.2](https://huggingface.co/datasets/facebook/babi_qa) | CC BY 3.0 |
+| [ProofWriter](https://allenai.org/data/proofwriter) | CC BY (as reported; the upstream terms are being confirmed) |
+| [ATOMIC 2020](https://github.com/allenai/comet-atomic-2020) | CC BY |
+| [OpenGloss v1.3 definitions](https://huggingface.co/datasets/mjbommar/opengloss-v1.3-definitions) | CC BY 4.0 |
+| [Project Gutenberg English](https://huggingface.co/datasets/sedthh/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.

README.md

@@ -1,5 +1,5 @@
 ---
-license: mit
+license: cc-by-nc-sa-4.0
 base_model: AbstractPhil/mini-beatrix-3
 library_name: geolip-alephllm
 tags:
@@ -104,3 +104,7 @@ The readout files carry every number; the results files carry every block of eve
 
 The bar for the gap was .85 at every served block from 12 on; the bar for silence .995 at every block beside +.012 bits per byte.
 The MSE form is the one carried forward.
+
+## Licence
+
+The weights are **non-commercial**: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/), because the training text includes CHILDES child-directed speech (CC BY-NC-SA, via TalkBank) and Wikipedia / WikiText (CC BY-SA). These add-on modules are trained on the Beatrix core and carry its terms; the code that trains and mounts them is MIT ([AbstractEyes/alephllm](https://github.com/AbstractEyes/alephllm)). The full terms and the list of training sources: [LICENSE.md](LICENSE.md).