index-v5

Interscript ML model index (index-v5). Runtimes resolve DEFAULT_INDEX_URL against this release asset + sha256 sidecar.

index-v4

Interscript ML model index (index-v4). Runtimes resolve DEFAULT_INDEX_URL against this release asset + sha256 sidecar.

frontier-predictions-v1

Raw per-paragraph predictions behind every published Arabic client-frontier verdict (Paper B, RESULTS.md). Re-score any run:

pip install interscript-ml-tools[sadeed]
interscript-sadeed-eval score --preds <file> --data Misraj/SadeedDiac-25 --key student

run-006 carries the 2026-09-05 correction (RESULTS.md): its rows re-score to 5.0821, the corrected 2.0 number — the tooling that published this release is the tooling that caught it.

index-v3

Interscript ML model index (index-v3). Runtimes resolve DEFAULT_INDEX_URL against this release asset + sha256 sidecar.

ara-diac-layerdrop-1.0

ara-diac-layerdrop-1.0-int4

IMF v1 (int4, decoder kv, opset 14). Trained from the depth-cut rung: ByT5-small with the encoder halved 12->6 (surviving layers copied VERBATIM from pretraining - both width-cut approaches collapsed at 74.68/82.96 while this works), Muon optimizer, r7 teacher labels (the 2.0 recipe), 6 epochs. ~190M parameters (63% of ByT5-small). Full-set windowed DER-CE 5.784, delta CI [3.033, 3.491] vs in-run teacher 2.2921 — the depth premium over the full-depth G2a peer (4.5701, CI [1.911, 2.352]) is 1.21pp with non-overlapping intervals. The int4 variant (~95MB) is the browser-budget tier; margin-gated before ship. Checkpoint rababa-checkpoints:/rababa_arabic_distill_small/run-009-layerdrop-6ep/best..

field	value
task	diacritization (Arab → Arab)
artifact	ara-diac-layerdrop-1.0-int4.zip (0.14 GiB)
der_teacher_fullset	2.2921 — windowed DER-CE (1400-byte windows, word-boundary split, greedy, haraqat-projected, Misraj evaluator); full 1,200-paragraph SadeedDiac-25
der_student_fullset	5.784 — same harness; paired bootstrap delta 3.2455 [3.033, 3.491]
parity	cer_delta 0.0589pp on 2480 samples
sha256	352e02105645a6e283e2c3602505bea0e90085b741666fd863e478e4835e7bc9
license	BSD-3-Clause

Runtimes reassemble split parts transparently and verify every sha256:

from secryst import Model
model = Model.load("ara-diac-layerdrop-1.0-int4")

ara-diac-small-2.1

ara-diac-small-2.1-fp16

IMF v1 (fp16, decoder kv, opset 14). 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..

field	value
task	diacritization (Arab → Arab)
artifact	ara-diac-small-2.1-fp16.zip (0.66 GiB)
der_teacher_fullset	2.289 — 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)
der_student_fullset	4.5701 — 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
parity	cer_delta 0.0714pp on 2480 samples
sha256	395ee72137d1cdfa9cd53bf167ad0d2f5fe5cbf9a5b62f118e0c9e0fa63c972d
license	BSD-3-Clause

Runtimes reassemble split parts transparently and verify every sha256:

from secryst import Model
model = Model.load("ara-diac-small-2.1-fp16")

index-v2

Interscript ML model index (index-v2). Runtimes resolve DEFAULT_INDEX_URL against this release asset + sha256 sidecar.

ara-diac-small-2.0

ara-diac-small-2.0

IMF v1 (fp32, decoder kv, opset 14). 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-006-r7-muon/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..

field	value
task	diacritization (Arab → Arab)
artifact	clean-fp32.zip (1.32 GiB)
der_teacher_fullset	2.289 — 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)
der_student_fullset	4.8218 — 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
parity	cer_delta 0.1187pp on 600 samples
sha256	d9aa95d0e8fd0d7da80fd2f7dabedf0ac65f55ba984de565d252f5e14906d98c
license	BSD-3-Clause

Runtimes reassemble split parts transparently and verify every sha256:

from secryst import Model
model = Model.load("ara-diac-small-2.0")

ara-diac-2.0

ara-diac-2.0

IMF v1 (int8, decoder kv, opset 14). 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 at 0.85% mix + 400 gold WikiNews-2014 lines, init from run-006-morph); checkpoint rababa-checkpoints:/rababa_arabic_byt5/run-007-news/best.

field	value
task	diacritization (Arab → Arab)
artifact	ara-diac-2.0-int8.zip (0.48 GiB)
der_total_greedy	2.2864 — greedy decode (the v1 runtime path); SadeedDiac-25, windowed zero-skip at 1400 bytes, full 1,200 paragraphs; ByT5-base r7
der_morph_greedy	1.3343 — same harness; Morphological DER (word-final case endings only)
wer_wikinews_multiref	17.3794 — WikiNews-2024 multi-reference, QCRI EvalDiac protocol (full mode), 356 texts / 10,616 words
der_wikinews_multiref	11.8273 — same harness
parity	cer_delta 0.0947pp on 600 samples
sha256	7f1d55d43f97564ad62fc14a7e0cabe52c01d8298b747677a75900de748bb6c6
license	BSD-3-Clause

Runtimes reassemble split parts transparently and verify every sha256:

from secryst import Model
model = Model.load("ara-diac-2.0")

index-v1

Interscript ML model index (index-v1). Runtimes resolve DEFAULT_INDEX_URL against this release asset + sha256 sidecar.