From 41440cd13bf4b20c1abd74d6d8f99779c57cea01 Mon Sep 17 00:00:00 2001 From: Ronald Tse Date: Tue, 18 Aug 2026 11:16:33 +0800 Subject: [PATCH 1/3] release: heb-diac-1.0 (gated fp32, token-exact on 600 samples) --- models.yaml | 13 +++++++++++++ 1 file changed, 13 insertions(+) diff --git a/models.yaml b/models.yaml index 97c6fc2..6f1b550 100644 --- a/models.yaml +++ b/models.yaml @@ -55,3 +55,16 @@ models: - {name: cer, value: 3.74, source: rababa-urdu/docs/RESULTS.md#diacritization-urdu-text--text--haraqat} parity: {samples: 11940, cer_delta: 0.0} license: BSD-3-Clause + heb-diac-1.0: + task: diacritization + scripts: [Hebr, Hebr] + precision: fp32 + filename: heb-diac-1.0-fp32.zip + url: https://github.com/interscript/ml-models/releases/download/heb-diac-1.0/heb-diac-1.0-fp32.zip + sha256: b32fa239e0c56708b6004942de4b9e6e9f265e7cc3828f41d560e96386e20dd8 + size: 2781139753 + metrics: + - {name: der_greedy, value: 29.0, source: rababa/docs/RESULTS.md#hebrew-diacritization} + - {name: der_beam4, value: 17.46, source: rababa/docs/RESULTS.md#hebrew-diacritization} + parity: {samples: 600, cer_delta: 0.0} + license: BSD-3-Clause From f275964eb0bb7ee81b688119411b5bf28ded93dd Mon Sep 17 00:00:00 2001 From: Ronald Tse Date: Tue, 18 Aug 2026 16:54:43 +0800 Subject: [PATCH 2/3] release: heb-diac-1.0 (fp32, parity cer_delta 0.0pp on 600 samples) --- models.yaml | 8 ++++- .../heb-diac-1.0/heb-diac-1.0.metadata.yaml | 29 +++++++++++++++++++ 2 files changed, 36 insertions(+), 1 deletion(-) create mode 100644 models/heb-diac-1.0/heb-diac-1.0.metadata.yaml diff --git a/models.yaml b/models.yaml index 6f1b550..04d3c7d 100644 --- a/models.yaml +++ b/models.yaml @@ -60,7 +60,13 @@ models: scripts: [Hebr, Hebr] precision: fp32 filename: heb-diac-1.0-fp32.zip - url: https://github.com/interscript/ml-models/releases/download/heb-diac-1.0/heb-diac-1.0-fp32.zip + parts: + - url: https://github.com/interscript/ml-models/releases/download/heb-diac-1.0/heb-diac-1.0-fp32.zip.part-00 + sha256: 5497f70cc3c96829e65f1eb0cc401c2d1ffeb4fa909acd4def824c2e019cbc37 + size: 1500000000 + - url: https://github.com/interscript/ml-models/releases/download/heb-diac-1.0/heb-diac-1.0-fp32.zip.part-01 + sha256: a7d2011ab4f9326a86e234cc49cb50daf82949e51efbf03a21d5796c46465e9d + size: 1281139753 sha256: b32fa239e0c56708b6004942de4b9e6e9f265e7cc3828f41d560e96386e20dd8 size: 2781139753 metrics: diff --git a/models/heb-diac-1.0/heb-diac-1.0.metadata.yaml b/models/heb-diac-1.0/heb-diac-1.0.metadata.yaml new file mode 100644 index 0000000..9a20d58 --- /dev/null +++ b/models/heb-diac-1.0/heb-diac-1.0.metadata.yaml @@ -0,0 +1,29 @@ +format: imf-v1 +id: heb-diac-1.0 +task: diacritization +source_script: Hebr +target: Hebr +tokenizer: bytes +opset: 14 +decoder: kv +precision: fp32 +license: BSD-3-Clause +trained_from: rababa train_hebrew_seeds.py s43 run-001; checkpoint rababa-checkpoints:/rababa_hebrew_byt5_s43/run-001/best +metrics: +- name: der_greedy + value: 29.0 + protocol: beam=1 greedy decode (the v1 runtime path); Nakdimon test split, 5,095 + examples; ByT5-base s43 + source: rababa/docs/RESULTS.md#hebrew-diacritization +- name: der_beam4 + value: 17.46 + protocol: beam=4 standard decode (reference quality; beam search is not in v1 runtimes); + Nakdimon test split, 5,095 examples; ByT5-base s43 + source: rababa/docs/RESULTS.md#hebrew-diacritization +parity: + samples: 600 + cer_delta: 0.0 +sha256: + decoder-kv.onnx: 2ff73f0f2ef81fb82231d8c6cb28b1383d9e8813360ffd6048eef47d051d624a + decoder.onnx: fdee8fd7e44d683ca885c9c3b1088e44f2c7d594e19212e438f9bf2b240ee168 + encoder.onnx: e41d2f50219de835a6c18883a685532736ab710e36bbe1e0eab2d1a80da4ced6 From b9bb24c44ccaaf9056a2a60c34f18a835f18f7c3 Mon Sep 17 00:00:00 2001 From: Ronald Tse Date: Tue, 18 Aug 2026 17:08:30 +0800 Subject: [PATCH 3/3] fix: metadata under family dir models/heb-diac/ --- .../heb-diac-1.0/heb-diac-1.0.metadata.yaml | 29 ------------------- 1 file changed, 29 deletions(-) delete mode 100644 models/heb-diac-1.0/heb-diac-1.0.metadata.yaml diff --git a/models/heb-diac-1.0/heb-diac-1.0.metadata.yaml b/models/heb-diac-1.0/heb-diac-1.0.metadata.yaml deleted file mode 100644 index 9a20d58..0000000 --- a/models/heb-diac-1.0/heb-diac-1.0.metadata.yaml +++ /dev/null @@ -1,29 +0,0 @@ -format: imf-v1 -id: heb-diac-1.0 -task: diacritization -source_script: Hebr -target: Hebr -tokenizer: bytes -opset: 14 -decoder: kv -precision: fp32 -license: BSD-3-Clause -trained_from: rababa train_hebrew_seeds.py s43 run-001; checkpoint rababa-checkpoints:/rababa_hebrew_byt5_s43/run-001/best -metrics: -- name: der_greedy - value: 29.0 - protocol: beam=1 greedy decode (the v1 runtime path); Nakdimon test split, 5,095 - examples; ByT5-base s43 - source: rababa/docs/RESULTS.md#hebrew-diacritization -- name: der_beam4 - value: 17.46 - protocol: beam=4 standard decode (reference quality; beam search is not in v1 runtimes); - Nakdimon test split, 5,095 examples; ByT5-base s43 - source: rababa/docs/RESULTS.md#hebrew-diacritization -parity: - samples: 600 - cer_delta: 0.0 -sha256: - decoder-kv.onnx: 2ff73f0f2ef81fb82231d8c6cb28b1383d9e8813360ffd6048eef47d051d624a - decoder.onnx: fdee8fd7e44d683ca885c9c3b1088e44f2c7d594e19212e438f9bf2b240ee168 - encoder.onnx: e41d2f50219de835a6c18883a685532736ab710e36bbe1e0eab2d1a80da4ced6