Skip to content

Commit c23d9cb

Browse files
veitveit
committed
📝 Update licensing
1 parent 3d5abc2 commit c23d9cb

File tree

1 file changed

+126
-39
lines changed

1 file changed

+126
-39
lines changed

‎docs/productive/licensing.rst

Lines changed: 126 additions & 39 deletions
Original file line numberDiff line numberDiff line change
@@ -89,9 +89,8 @@ the public domain of software.
8989
Non-software licences
9090
---------------------
9191

92-
Open source software licences can also be used for works that are not software.
93-
They are often also the best choice, especially if the works in question are
94-
edited and versioned as source code.
92+
Open source software licences are often used for works that are not software.
93+
However, they are often not the best choice.
9594

9695
Data, media, :abbr:`etc. (et cetera)`
9796
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
@@ -104,37 +103,49 @@ they are `not recommended for software
104103
<https://creativecommons.org/faq/#can-i-apply-a-creative-commons-license-to-software>`_.
105104

106105
.. tip::
107-
The `DEAL consortium <https://deal-konsortium.de/en/>`_ recommends the CC BY
108-
licence for Open Access publications, see also `Why CC BY is the Best Choice
109-
for Open Access Publishing <https://deal-konsortium.de/en/why-ccby>`_.
110-
111-
:abbr:`RADAR (Research Data Repository)`, a cross-disciplinary repository for
112-
archiving and publishing research data, recommends only one of the `CC
113-
licences
114-
<https://radar.products.fiz-karlsruhe.de/en/radarfeatures/lizenzen-fuer-forschungsdaten#cc-licenses>`_.
115-
116-
The `Open Knowledge Foundation <https://okfn.org/en/>`_ has also published a set
117-
of `Open Data Commons <https://opendatacommons.org>`_ licences for
118-
data/databases:
119-
120-
`Open Data Commons Open Database License (ODbL) v1.0 <https://opendatacommons.org/licenses/odbl/1-0/>`_
121-
Attribution and sharing under equal terms.
122-
`Open Data Commons Attribution License (ODC-By) v1.0 <https://opendatacommons.org/licenses/by/1-0/>`_
123-
Attribution.
124-
`Open Data Commons Public Domain Dedication and License (PDDL) v1.0 <https://opendatacommons.org/licenses/pddl/1-0/>`_
125-
The PDDL places the data in the public domain and waives all rights.
126-
127-
`GovData <https://www.govdata.de>`_ has submitted the *Data Licence Germany* in two variants:
128-
129-
* `Datenlizenz Deutschland – Namensnennung – Version 2.0
130-
<https://www.govdata.de/dl-de/by-2-0>`_
131-
* `Datenlizenz Deutschland – Zero – Version 2.0
132-
<https://www.govdata.de/dl-de/zero-2-0>`_
133-
134-
When using the `Community Data License Agreement – Permissive, Version 2.0 <https://cdla.dev/permissive-2-0/>`_ the copyright notices must be retained.
135-
136-
Another possible licence for artistic works is the `Free Art License 1.3
137-
<https://artlibre.org/licence/lal/en/>`_.
106+
* The `DEAL consortium <https://deal-konsortium.de/en/>`_ recommends the CC
107+
BY licence for Open Access publications of research results, see also `Why
108+
CC BY is the Best Choice for Open Access Publishing
109+
<https://deal-konsortium.de/en/why-ccby>`_.
110+
111+
* :abbr:`RADAR (Research Data Repository)`, a cross-disciplinary repository
112+
for archiving and publishing research data, recommends only one of the `CC
113+
licences
114+
<https://radar.products.fiz-karlsruhe.de/en/radarfeatures/lizenzen-fuer-forschungsdaten#cc-licenses>`_.
115+
116+
* The `Open Knowledge Foundation <https://okfn.org/en/>`_ has also published a
117+
set of `Open Data Commons <https://opendatacommons.org>`_ licences for
118+
data/databases:
119+
120+
`Open Data Commons Open Database License (ODbL) v1.0 <https://opendatacommons.org/licenses/odbl/1-0/>`_
121+
Attribution and sharing under equal terms.
122+
`Open Data Commons Attribution License (ODC-By) v1.0 <https://opendatacommons.org/licenses/by/1-0/>`_
123+
Attribution.
124+
`Open Data Commons Public Domain Dedication and License (PDDL) v1.0 <https://opendatacommons.org/licenses/pddl/1-0/>`_
125+
The PDDL places the data in the public domain and waives all rights.
126+
127+
* `GovData <https://www.govdata.de>`_ has submitted the *Data Licence Germany*
128+
in two variants:
129+
130+
* `Datenlizenz Deutschland – Namensnennung – Version 2.0
131+
<https://www.govdata.de/dl-de/by-2-0>`_
132+
* `Datenlizenz Deutschland – Zero – Version 2.0
133+
<https://www.govdata.de/dl-de/zero-2-0>`_
134+
135+
* The Community Data License Agreement <https://cdla.dev>_ can be used in four
136+
different ways:
137+
138+
* `Community Data License Agreement – Permissive, Version 2.0
139+
<https://cdla.dev/permissive-2-0/>`_
140+
* `Community Data License Agreement – Sharing, Version 1.0
141+
<https://cdla.dev/sharing-1-0/>`_
142+
* `Open Use of Data Agreement, Version 1.0
143+
<https://cdla.dev/open-use-of-data-agreement-v1-0/>`_
144+
* `Computational Use of Data Agreement, Version 1.0
145+
<https://cdla.dev/computational-use-of-data-agreement-v1-0/>`_
146+
147+
* Another possible licence for artistic works is the `Free Art License 1.3
148+
<https://artlibre.org/licence/lal/en/>`_.
138149

139150
Machine learning models
140151
~~~~~~~~~~~~~~~~~~~~~~~
@@ -164,19 +175,95 @@ been developed for a company or specific models:
164175
* `BigScience OpenRAIL-M (Responsible AI License
165176
<https://www.licenses.ai/blog/2022/8/26/bigscience-open-rail-m-license>`_
166177

167-
RAIL-D
178+
There are other `Responsible AI Licenses (RAIL) <https://www.licenses.ai>`_
179+
with various restrictions on use:
180+
181+
OpenRAIL-D
168182
contains usage restrictions that only apply to the data.
169-
RAIL-A
183+
OpenRAIL-A
170184
contains usage restrictions that only apply to the
171185
application/executability.
172-
RAIL-M
186+
OpenRAIL-M
173187
contains usage restrictions that only apply to the model.
174-
RAIL-S
188+
189+
.. seealso::
190+
`RAIL-M
191+
<https://www.licenses.ai/blog/2022/8/26/bigscience-open-rail-m-license>`_
192+
193+
OpenRAIL-S
175194
contains usage restrictions that only apply to the source code.
176195

196+
AI models that are licensed under an open source licence but whose training data
197+
and programmes have **not** been published are not compliant with the `Debian
198+
Free Software Guidelines (DFSG)
199+
<https://de.wikipedia.org/wiki/Debian_Free_Software_Guidelines>`_, see also
200+
`Interpretation of DFSG on Artificial Intelligence (AI) Models
201+
<https://www.debian.org/vote/2025/vote_002>`_.
202+
203+
.. _osaid:
204+
205+
For the `Open Source Initiative (OSI) <https://opensource.org/>`__, the
206+
definition of open source AI also goes far beyond the use of a model – it must
207+
also be understandable how the model was created, and the model must also be
208+
able to be modified and shared with others for any purpose. These four freedoms
209+
are fulfilled with
210+
211+
Open Data
212+
Sufficiently detailed information about the data used to train the system so
213+
that an essentially equivalent system can be built
214+
Open Code
215+
The complete source code used to train and operate the system under licences
216+
approved by the OSI
217+
Open Weights
218+
Model parameters, such as weights or other configuration settings under
219+
licences approved by the OSI
220+
221+
Accordingly, the OSI developed `OSAID 1.0
222+
<https://opensource.org/ai/open-source-ai-definition>`_, which applies to the
223+
following models:
224+
225+
* EleutherAI: `Pythia <https://github.com/EleutherAI/pythia>`_, `GPT-J
226+
<https://www.eleuther.ai/artifacts/gpt-j>`_
227+
* The Allen Institute for Artificial Intelligence: `OLMo 2
228+
<https://allenai.org/olmo>`_, `Molmo <https://allenai.org/blog/molmo>`_
229+
* LLM360: `K2
230+
<https://huggingface.co/collections/LLM360/k2-6622ae6911e3eb6219690039>`_,
231+
`Amber
232+
<https://huggingface.co/collections/LLM360/amber-65e7333ff73c7bbb014f2f2f>`_,
233+
`CrystalCoder <https://huggingface.co/LLM360/Crystal>`_
234+
* Google: `T5
235+
<https://github.com/google-research/text-to-text-transfer-transformer>`_
236+
237+
Presumably the following models would also fulfil the requirements if they were
238+
to change their legal conditions:
239+
240+
* BigScience: `Bloom <https://huggingface.co/bigscience/bloom>`_
241+
* BigCode: `StarCoder 2 <https://github.com/bigcode-project/starcoder2>`_
242+
* Technology Innovation Institute: `Falcon
243+
<https://huggingface.co/collections/tiiuae/falcon-h1-6819f2795bc406da60fab8df>`_
244+
245+
However, there are also some models that have been analysed and failed because
246+
they lack required components and/or legal agreements:
247+
248+
* Meta: Llama2
249+
* xAI: Grok
250+
* Microsoft: Phi-2
251+
* Mistral AI: Mixtral
252+
177253
.. seealso::
178254
* `Licensing Machine Learning models
179-
<https://book.the-turing-way.org/reproducible-research/licensing/licensing-ml>`_ by The Turing Way Community
255+
<https://book.the-turing-way.org/reproducible-research/licensing/licensing-ml>`_
256+
by The Turing Way Community
257+
* Alek Tarkowski, Open Future in partnership with the Open Source Initiative:
258+
`Data Governance in Open Source AI
259+
<https://opensource.org/wp-content/uploads/2025/02/2025-OSI-DataGovernanceOSAI-final-v5.pdf>`_
260+
261+
Databases
262+
~~~~~~~~~
263+
264+
One of the few licences for databases is the `Open Data Commons Open Database
265+
License (ODbL) v1.0 <https://opendatacommons.org/licenses/odbl/1-0/>`_, which is
266+
used, for example, by `OpenStreetMap (OSM) <https://www.openstreetmap.org>`_.
180267

181268
Documentation
182269
~~~~~~~~~~~~~

0 commit comments

Comments
 (0)