release: ara-diac-small-2.1-int8 (int8, parity cer_delta 0.0857pp on 2480 samples)

Author: Ronald Tse

models.yaml

@@ -292,3 +292,16 @@ models:
       - {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: 0.0947}
     license: BSD-3-Clause
+  ara-diac-small-2.1-int8:
+    task: diacritization
+    scripts: [Arab, Arab]
+    precision: int8
+    filename: ara-diac-small-2.1-int8.zip
+    url: https://github.com/interscript/interscript-ml/releases/download/ara-diac-small-2.1/ara-diac-small-2.1-int8.zip
+    sha256: 9b4e4aad2ab53a8cfbec432a78e27b409376627243aec3fd043f06574552a32b
+    size: 267965657
+    metrics:
+      - {name: der_teacher_fullset, value: 2.289, source: interscript/interscript-ml docs/RESULTS.md#run-006-r7-muon}
+      - {name: der_student_fullset, value: 4.5701, source: interscript/interscript-ml docs/RESULTS.md#run-006-r7-muon}
+    parity: {samples: 2480, cer_delta: 0.0857}
+    license: BSD-3-Clause

models/ara-diac-small-2.1/ara-diac-small-2.1-int8.metadata.yaml

@@ -0,0 +1,39 @@
+format: imf-v1
+id: ara-diac-small-2.1
+task: diacritization
+source_script: Arab
+target: Arab
+tokenizer: bytes
+opset: 14
+decoder: kv
+precision: int8
+license: BSD-3-Clause
+trained_from: 'sequence-level KD from the r7 canonical teacher (rababa_arabic_byt5/run-007-news/best,
+  2.2864 windowed DER-CE full protocol): fresh greedy r7 labels on the same r5-units
+  corpus/limits as ara-diac-small-1.0, Muon optimizer (E3-adopted), vanilla ByT5-small
+  (E4, pre-registered gate <= 6.26). Checkpoint rababa-checkpoints:/rababa_arabic_distill_small/run-007-r7-muon-6ep/best.
+  The two measured wins compound: 8.259 -> 4.8218 full-set windowed DER-CE (teacher
+  reproduces 2.289 in-run vs documented 2.2864) — a 42% error reduction on the 1.0
+  release at the same architecture and artifact size. Still misses the strict teacher+0.5pp
+  gate (+2.53pp; miss disclosed); the E2/E3 factorial attributes the residual to domain
+  coverage.'
+metrics:
+- name: der_teacher_fullset
+  value: 2.289
+  protocol: windowed DER-CE (1400-byte windows, word-boundary split, greedy, haraqat-projected,
+    Misraj evaluator); full 1,200-paragraph SadeedDiac-25; in-run reproduction of
+    the documented 2.2864 (r7 canonical teacher)
+  source: interscript/interscript-ml docs/RESULTS.md#run-006-r7-muon
+- name: der_student_fullset
+  value: 4.5701
+  protocol: same full-set harness; E4 (r7 teacher labels + Muon, vanilla ByT5-small)
+    vs the 8.259 AdamW/r6-labels 1.0 release — a 42% error reduction at identical
+    architecture and artifact size
+  source: interscript/interscript-ml docs/RESULTS.md#run-006-r7-muon
+parity:
+  samples: 2480
+  cer_delta: 0.0857
+sha256:
+  decoder-kv.onnx: 7fe2953bafb0955986b2ce2036da6077df5b7f1d9a9295265a1aa24b85961813
+  decoder.onnx: affcca046189862be499aa6ce0d724a71d40507bd418882241d51cf6a277707a
+  encoder.onnx: 61afa9ded2e61966a2966f906650715183f95cccc24bac699ed4c4b2d766eecd