From fe9cf2e2ec9df221b6f1242491583b458d579af0 Mon Sep 17 00:00:00 2001 From: Ronald Tse Date: Tue, 25 Aug 2026 11:17:23 +0800 Subject: [PATCH] release: ara-diac-1.0 (fp32, parity cer_delta 0.0pp on 500 samples) --- models.yaml | 19 +++++++++++++ models/ara-diac/ara-diac-1.0.metadata.yaml | 33 ++++++++++++---------- 2 files changed, 37 insertions(+), 15 deletions(-) diff --git a/models.yaml b/models.yaml index 2902e18..1193bbe 100644 --- a/models.yaml +++ b/models.yaml @@ -183,3 +183,22 @@ models: - {name: der_beam4, value: 16.43, source: rababa/docs/RESULTS.md#hebrew-diacritization} parity: {samples: 500, cer_delta: 0.0} license: BSD-3-Clause + ara-diac-1.0: + task: diacritization + scripts: [Arab, Arab] + precision: fp32 + filename: ara-diac-1.0-fp32.zip + parts: + - url: https://github.com/interscript/interscript-ml/releases/download/ara-diac-1.0/ara-diac-1.0-fp32.zip.part-00 + sha256: 6282a9291fb45ac7fcc3abe79d33b5913e91f87dcff8b305bf1d57ed98a3ed0b + size: 1500000000 + - url: https://github.com/interscript/interscript-ml/releases/download/ara-diac-1.0/ara-diac-1.0-fp32.zip.part-01 + sha256: 7e7254a89ec75493871f6f61a993315097dad1bb0ba63c4584453c309e8cb906 + size: 1280311342 + sha256: 9e6d9777fe2474b150450774aaa1c9e8a9051b502b184fde298b91636dcfdac6 + size: 2780311342 + metrics: + - {name: der_total_greedy, value: 2.5793, source: rababa/docs/RESULTS.md#r6-verdict-table-sadeeddiac-25-2026-08-21} + - {name: der_morph_greedy, value: 1.5317, source: rababa/docs/RESULTS.md#r6-verdict-table-sadeeddiac-25-2026-08-21} + parity: {samples: 500, cer_delta: 0.0} + license: BSD-3-Clause diff --git a/models/ara-diac/ara-diac-1.0.metadata.yaml b/models/ara-diac/ara-diac-1.0.metadata.yaml index 5a3fb64..fe78bd0 100644 --- a/models/ara-diac/ara-diac-1.0.metadata.yaml +++ b/models/ara-diac/ara-diac-1.0.metadata.yaml @@ -8,19 +8,22 @@ opset: 14 decoder: kv precision: fp32 license: BSD-3-Clause -trained_from: >- - rababa train_arabic_r6.py run-006-morph (morphological aux-task: - plain + "TAG: "-prefixed iʿrāb stream from qalsadi labels, init from - r5); checkpoint rababa-checkpoints:/rababa_arabic_byt5/run-006-morph/best +trained_from: 'rababa train_arabic_r6.py run-006-morph (morphological aux-task: plain + + "TAG: "-prefixed iʿrāb stream from qalsadi labels, init from r5); checkpoint rababa-checkpoints:/rababa_arabic_byt5/run-006-morph/best' metrics: - - name: der_total_greedy - value: 2.5793 - protocol: >- - greedy decode (the v1 runtime path); SadeedDiac-25, windowed - zero-skip at 1400 bytes, 1,200 paragraphs; ByT5-base r6 - source: rababa/docs/RESULTS.md#r6-verdict-table-sadeeddiac-25-2026-08-21 - - name: der_morph_greedy - value: 1.5317 - protocol: >- - same harness; Morphological DER (word-final case endings only) - source: rababa/docs/RESULTS.md#r6-verdict-table-sadeeddiac-25-2026-08-21 +- name: der_total_greedy + value: 2.5793 + protocol: greedy decode (the v1 runtime path); SadeedDiac-25, windowed zero-skip + at 1400 bytes, 1,200 paragraphs; ByT5-base r6 + source: rababa/docs/RESULTS.md#r6-verdict-table-sadeeddiac-25-2026-08-21 +- name: der_morph_greedy + value: 1.5317 + protocol: same harness; Morphological DER (word-final case endings only) + source: rababa/docs/RESULTS.md#r6-verdict-table-sadeeddiac-25-2026-08-21 +parity: + samples: 500 + cer_delta: 0.0 +sha256: + decoder-kv.onnx: aa9cc36870d2841b4c1e8863f9c4b1953cd8332f172644a8dcf87ea39fe28ff1 + decoder.onnx: 8d2302f5c9a58a6e204f6abe6ae7699a0662142da24c0c59f14b47476603b12c + encoder.onnx: 0b4df16da6626fa70f345b125a26fd0b5b41dc77655be175e54a1cd9bf93519d