diff --git a/models.yaml b/models.yaml index df18b2d..2ba914d 100644 --- a/models.yaml +++ b/models.yaml @@ -149,3 +149,22 @@ models: - {name: sb_homograph_ezafe_norm, value: 77.34, source: interscript/interscript-ml docs/RESULTS.md#fas-g2p-1.0} parity: {samples: 1000, cer_delta: 0.0} license: BSD-3-Clause + heb-diac-1.1: + task: diacritization + scripts: [Hebr, Hebr] + precision: fp32 + filename: heb.zip + parts: + - url: https://github.com/interscript/interscript-ml/releases/download/heb-diac-1.1/heb.zip.part-00 + sha256: 8dac0d62e9b0ac4b137412eaf3a542953f11575f12f66f76cb93d3f5d788fc1b + size: 1500000000 + - url: https://github.com/interscript/interscript-ml/releases/download/heb-diac-1.1/heb.zip.part-01 + sha256: 230c5ab731ea5f001f802d60b9e9a328c10b43bc311a19ef1974f114384313ed + size: 1280987905 + sha256: f0022dde8d06e46e8f08cf0b2f7bbb887a05ed686b90f1d5aa82ca3ef690a0b0 + size: 2780987905 + metrics: + - {name: der_greedy, value: 16.44, source: rababa/docs/RESULTS.md#hebrew-diacritization} + - {name: der_beam4, value: 16.43, source: rababa/docs/RESULTS.md#hebrew-diacritization} + parity: {samples: 500, cer_delta: 0.0} + license: BSD-3-Clause diff --git a/models/heb-diac/heb-diac-1.1.metadata.yaml b/models/heb-diac/heb-diac-1.1.metadata.yaml index 9d69dd9..71a94f4 100644 --- a/models/heb-diac/heb-diac-1.1.metadata.yaml +++ b/models/heb-diac/heb-diac-1.1.metadata.yaml @@ -8,21 +8,24 @@ opset: 14 decoder: kv precision: fp32 license: BSD-3-Clause -trained_from: >- - rababa train_hebrew_s46.py — phonikud curriculum (1.5M knesset - weak-pretrain + 73.8K hewiki garnish, then gold FT); checkpoint - rababa-checkpoints:/rababa_hebrew/run-s46-phonikud-plus/run-002-gold-ft/best +trained_from: rababa train_hebrew_s46.py — phonikud curriculum (1.5M knesset weak-pretrain + + 73.8K hewiki garnish, then gold FT); checkpoint rababa-checkpoints:/rababa_hebrew/run-s46-phonikud-plus/run-002-gold-ft/best metrics: - - name: der_greedy - value: 16.44 - protocol: >- - greedy decode (the v1 runtime path); Nakdimon test split, - 5,095 examples; ByT5-base s46 - source: rababa/docs/RESULTS.md#hebrew-diacritization - - name: der_beam4 - value: 16.43 - protocol: >- - beam=4 standard decode (reference quality); Nakdimon test - split, 5,095 examples; ByT5-base s46. Greedy ≈ beam-4 for this - model — the runtime path loses nothing. - source: rababa/docs/RESULTS.md#hebrew-diacritization +- name: der_greedy + value: 16.44 + protocol: greedy decode (the v1 runtime path); Nakdimon test split, 5,095 examples; + ByT5-base s46 + source: rababa/docs/RESULTS.md#hebrew-diacritization +- name: der_beam4 + value: 16.43 + protocol: beam=4 standard decode (reference quality); Nakdimon test split, 5,095 + examples; ByT5-base s46. Greedy ≈ beam-4 for this model — the runtime path loses + nothing. + source: rababa/docs/RESULTS.md#hebrew-diacritization +parity: + samples: 500 + cer_delta: 0.0 +sha256: + decoder-kv.onnx: f5c6d56fe4fcc7ad99a8fa9b0936f5554c9a8dd11c612e5a0eb9119b4c5d58bc + decoder.onnx: 6665ab15d65808d0ebd9ae569713457737cebb4c6ca39cc4953ecc7231f3b1c8 + encoder.onnx: 09c3d1533a06e98efdf36b3dd3c23c68f0fc80f296431ffcbec919438e409708