ara-diac-news-1.0

Arabic news-register specialist (run-029): byt5-base seq2seq, register-pure training — 900K QCRI-silver units (~4.5M words of Wikipedia news-domain labels, published with EMNLP 2025) + WikiNews gold upsampling, NO classical tashkeela dilution, 2 epochs, r7-init. IMF v1 int8, 399 MB.

Metrics

benchmark	this model	ara-diac-2.0 (generalist)
WikiNews-2024 multiref WER/DER	10.13 / 8.98	17.38 / 11.83
SadeedDiac-25 Total DER	5.5008	2.2864

Use this model for news-domain text; use ara-diac-2.0 for classical/MSA. Two families, never averaged (RESULTS.md: the dose-response curve and the output-space oracle both close generalist blending).

Provenance

• Training: TODO.sota/25 (run-029, HF Jobs a100-large).
• sha256: e5a52b1530dfa6d6c7ef8970443206ce33af865d0337a58d145945adab33484e.
• License: BSD-3-Clause (code + weights).

Mirror provenance

Canonical channel: GitHub Releases (interscript/interscript-models), sha256 e5a52b1530dfa6d6c7ef8970443206ce33af865d0337a58d145945adab33484e.