Use concise Ekko PnC v2 CC BY 4.0 terms

MODEL_LICENSE.md

@@ -1,23 +1,11 @@
-# ekko-pnc-stage3a model terms
+# Ekko PnC v2 model license
 
-This is a modified model derived from `jonfd/electra-small-nordic`, licensed CC BY 4.0. RyeAI
-applies Creative Commons Attribution 4.0 International to the modifications it has
-the right to license. A copy is included at `LICENSES/CC-BY-4.0.txt`.
+Ekko PnC v2 is licensed under Creative Commons Attribution 4.0 International
+(CC BY 4.0). It is derived from `jonfd/electra-small-nordic`, also licensed
+under CC BY 4.0. The full license is in `LICENSES/CC-BY-4.0.txt`.
 
-Training material included text associated with CoRal. `LICENSES/CORAL-LICENSE`
-defines models trained on the data as Adapted Material and requires its use
-restrictions and licence copy to accompany distribution; those restrictions apply
-in addition to CC BY 4.0.
+Use and redistribution are also subject to the CoRal restrictions in section 4
+of `LICENSES/CORAL-LICENSE`, which are incorporated into these model terms.
 
-FT Speech material was also used; its source statement is at
-`LICENSES/FTSPEECH-LICENSE.md`. The effect of those terms on redistributed weights
-is not settled by this notice. The project owner has chosen to proceed with the
-source statement and Folketinget links attached; no separate legal opinion is
-claimed.
-
-NST text was also used. The publisher's
-[`alexandrainst/nst-da` dataset card](https://huggingface.co/datasets/alexandrainst/nst-da)
-identifies its license as CC0 1.0. The legal code is included at
-`LICENSES/CC0-1.0.txt`.
-
-This notice is not legal advice.
+Training-source notices and terms for FT Speech and NST are bundled in
+`NOTICE` and `LICENSES/`.

NOTICE

@@ -1,7 +1,7 @@
-ekko-pnc-stage3a is derived from jonfd/electra-small-nordic (CC BY 4.0).
+Ekko PnC v2 is derived from jonfd/electra-small-nordic (CC BY 4.0).
 
 Modifications by RyeAI: a fused Danish punctuation and capitalization head,
-a running-text fine-tune (stage 3a), a boundary-logit calibration folded into
+a running-text fine-tune, a boundary-logit calibration folded into
 the classifier bias, and ONNX exports. Training text included CoRal, FT Speech
 and NST; CoRal and FT Speech punctuation was restored by GPT-4o (Azure OpenAI).
 

SHA256SUMS

@@ -2,8 +2,8 @@
 a2010f343487d3f7618affe54f789f5487602331c0a8d03f49e9a7c547cf0499  LICENSES/CC0-1.0.txt
 ee93c98df9a894464d1c042b66c6039543776c463d06d1a6e93f176e08ca67bd  LICENSES/CORAL-LICENSE
 39708f5d77ae217b3a540363cb1b37fe079478e4ed1d3464f38d05d5a1c58b9a  LICENSES/FTSPEECH-LICENSE.md
-82fa0b5f0b8a2462939c621e1d56c34f02c6a719c6d7658022e5eeaf5a3cfb85  MODEL_LICENSE.md
-9ef67a5fc18e74cebda6268a8cb3b9205b5541fd2d90ea9331f3c792fb2a8a72  NOTICE
+590f820521961c89e7795bd3f46575794f4cbdcb02cc9db682e825c0afc874aa  MODEL_LICENSE.md
+bc86a9b47ff32a3ab29bb0ff8f414ae8b2ef1932ccf8bae2e4dacb5a6812ba8d  NOTICE
 a48af6ebefb6b003a855852289b4635ece42a8e8e40d47bd9341860e561e7bdf  README.md
 9268092acbdc6b1774144e75362976df583acf8035ecdefcfcdff85decb85fe0  calibration.json
 dc655c46f69eeb15dd43786ad58fb5862cc794b36cc9d2c426ab5bc4b01f23ca  config.json