From 48bb138a25b20d92451415d01194c25c2fcf1cba Mon Sep 17 00:00:00 2001 From: Ronald Tse Date: Sun, 30 Aug 2026 18:36:43 +0200 Subject: [PATCH 1/2] release: ara-diac-2.0 (int8, parity cer_delta 0.0947pp on 600 samples) --- models.yaml | 18 ++++++------------ models/ara-diac/ara-diac-2.0.metadata.yaml | 10 +++++----- 2 files changed, 11 insertions(+), 17 deletions(-) diff --git a/models.yaml b/models.yaml index d4ef59a..9cd9324 100644 --- a/models.yaml +++ b/models.yaml @@ -259,21 +259,15 @@ models: ara-diac-2.0: task: diacritization scripts: [Arab, Arab] - precision: fp32 - filename: ara-diac-2.0-fp32.zip - parts: - - url: https://github.com/interscript/interscript-ml/releases/download/ara-diac-2.0/ara-diac-2.0-fp32.zip.part-00 - sha256: 28c272b5ce58031c356e75823eb581f12351a1675a52d8e4000f851c80d5e14d - size: 1500000000 - - url: https://github.com/interscript/interscript-ml/releases/download/ara-diac-2.0/ara-diac-2.0-fp32.zip.part-01 - sha256: 86016da35e14a8fc05d4fbf3b1ad9ef24c29f9f7893f1716e40bef4b491effa2 - size: 1280299051 - sha256: 872c655ec6750bf44e5986f958de26767eafa13f8e0a7c64340bc35fd8a4550e - size: 2780299051 + precision: int8 + filename: ara-diac-2.0-int8.zip + url: https://github.com/interscript/interscript-ml/releases/download/ara-diac-2.0/ara-diac-2.0-int8.zip + sha256: 7f1d55d43f97564ad62fc14a7e0cabe52c01d8298b747677a75900de748bb6c6 + size: 519207340 metrics: - {name: der_total_greedy, value: 2.2864, source: rababa/docs/RESULTS.md#r7-verdict-table-sadeeddiac-25-2026-08-28} - {name: der_morph_greedy, value: 1.3343, source: rababa/docs/RESULTS.md#r7-verdict-table-sadeeddiac-25-2026-08-28} - {name: wer_wikinews_multiref, value: 17.3794, source: rababa/docs/RESULTS.md#r7-ood-verdict-table-wikinews-2024-multiref-2026-08-28} - {name: der_wikinews_multiref, value: 11.8273, source: rababa/docs/RESULTS.md#r7-ood-verdict-table-wikinews-2024-multiref-2026-08-28} - parity: {samples: 600, cer_delta: 0.0903} + parity: {samples: 600, cer_delta: 0.0947} license: BSD-3-Clause diff --git a/models/ara-diac/ara-diac-2.0.metadata.yaml b/models/ara-diac/ara-diac-2.0.metadata.yaml index 26d434a..2e3f026 100644 --- a/models/ara-diac/ara-diac-2.0.metadata.yaml +++ b/models/ara-diac/ara-diac-2.0.metadata.yaml @@ -6,7 +6,7 @@ target: Arab tokenizer: bytes opset: 14 decoder: kv -precision: fp32 +precision: int8 license: BSD-3-Clause trained_from: 'rababa train_arabic_r7.py run-007-news (news-domain adaptation of the r6 morphological-aux teacher: r5-units anchor + 13,986 teacher-labeled news units @@ -33,8 +33,8 @@ metrics: source: rababa/docs/RESULTS.md#r7-ood-verdict-table-wikinews-2024-multiref-2026-08-28 parity: samples: 600 - cer_delta: 0.0903 + cer_delta: 0.0947 sha256: - decoder-kv.onnx: 1c012908b92efa3f051c667d82142d7cee990443f74a9cf36b957d669a66f27b - decoder.onnx: 939dd3498d819a793e90aa95122ab33b6925a005237766bf9be66968f2645867 - encoder.onnx: abd659bd4818a361e4268c23aa8644aa867f0667f32742b688f42eb5fe7aa8cc + decoder-kv.onnx: 8d3c2409b500f44db5b92cb206dc9bd7789db9006f5d7e54b58aea42c94a2ed0 + decoder.onnx: d5edc73942cab51a7893fcc5da1e4f2479d187465385d1d5d31e9a96dbc6e6d2 + encoder.onnx: 4268e6245204d00ed32ec22c1b94b02e5020bdf426fb7d15b8964825387349c7 From 33ade48e64de00409797de594884768c97103958 Mon Sep 17 00:00:00 2001 From: Ronald Tse Date: Sun, 30 Aug 2026 18:49:02 +0200 Subject: [PATCH 2/2] fix(release): restore the fp32 parts entry; int8 as its own id --- models.yaml | 21 +++++++++++++++++++++ 1 file changed, 21 insertions(+) diff --git a/models.yaml b/models.yaml index 9cd9324..47decc9 100644 --- a/models.yaml +++ b/models.yaml @@ -257,6 +257,27 @@ models: parity: {samples: 600, cer_delta: 0.1393} license: BSD-3-Clause ara-diac-2.0: + task: diacritization + scripts: [Arab, Arab] + precision: fp32 + filename: ara-diac-2.0-fp32.zip + parts: + - url: https://github.com/interscript/interscript-ml/releases/download/ara-diac-2.0/ara-diac-2.0-fp32.zip.part-00 + sha256: 28c272b5ce58031c356e75823eb581f12351a1675a52d8e4000f851c80d5e14d + size: 1500000000 + - url: https://github.com/interscript/interscript-ml/releases/download/ara-diac-2.0/ara-diac-2.0-fp32.zip.part-01 + sha256: 86016da35e14a8fc05d4fbf3b1ad9ef24c29f9f7893f1716e40bef4b491effa2 + size: 1280299051 + sha256: 872c655ec6750bf44e5986f958de26767eafa13f8e0a7c64340bc35fd8a4550e + size: 2780299051 + metrics: + - {name: der_total_greedy, value: 2.2864, source: rababa/docs/RESULTS.md#r7-verdict-table-sadeeddiac-25-2026-08-28} + - {name: der_morph_greedy, value: 1.3343, source: rababa/docs/RESULTS.md#r7-verdict-table-sadeeddiac-25-2026-08-28} + - {name: wer_wikinews_multiref, value: 17.3794, source: rababa/docs/RESULTS.md#r7-ood-verdict-table-wikinews-2024-multiref-2026-08-28} + - {name: der_wikinews_multiref, value: 11.8273, source: rababa/docs/RESULTS.md#r7-ood-verdict-table-wikinews-2024-multiref-2026-08-28} + parity: {samples: 600, cer_delta: 0.0903} + license: BSD-3-Clause + ara-diac-2.0-int8: task: diacritization scripts: [Arab, Arab] precision: int8