small-models-for-glam
/

kraken-ppocrv6-tiny

+---
+id: https://doi.org/10.5281/zenodo.21788403
+authors:
+- name: Benjamin Kiessling
+  affiliation: ALMAnaCH, Inria Paris
+  orcid: https://orcid.org/0000-0001-9543-7827
+summary: PP-OCRv6 (tiny) multilingual handwritten and printed text recognition model
+license: apache-2.0
+software_name: kraken
+software_hints:
+- kraken_version>=7.1.0
+language:
+- ara
+- cat
+- ces
+- chu
+- cos
+- dan
+- deu
+- dum
+- eng
+- fas
+- fin
+- fra
+- frm
+- gez
+- gle
+- grc
+- heb
+- hun
+- ita
+- kat
+- lat
+- lav
+- lit
+- mal
+- mul
+- nld
+- nor
+- oci
+- ota
+- pcd
+- pol
+- por
+- ron
+- rus
+- slk
+- slv
+- spa
+- srp
+- swe
+- syr
+- ukr
+- urd
+- xcl
+- yid
+script:
+- Arab
+- Armn
+- Cyrl
+- Ethi
+- Geor
+- Grek
+- Hebr
+- Latn
+- Mlym
+- Syrc
+keywords:
+- handwritten-text-recognition
+- optical-character-recognition
+- htr
+- ocr
+- kraken
+- ppocr
+- ppocrv6
+- multilingual
+variant_of:
+- https://doi.org/10.5281/zenodo.21788405
+- https://doi.org/10.5281/zenodo.21788410
+model_type:
+- recognition
+metrics:
+  cer: 8.71
+  wer: 30.0
+  cer_macro: 10.99
+  wer_macro: 36.16
+library_name: kraken
+pipeline_tag: image-to-text
+tags:
+- handwritten-text-recognition
+- optical-character-recognition
+- htr
+- ocr
+- kraken
+- ppocr
+- ppocrv6
+- multilingual
+---
+# PP-OCRv6 (tiny) multilingual text recognition base model
+> **Hugging Face Hub mirror**
+>
+> This repository mirrors the [canonical Zenodo release](https://doi.org/10.5281/zenodo.21788403) by
+> [Benjamin Kiessling](https://orcid.org/0000-0001-9543-7827), ALMAnaCH, Inria Paris.
+> The weights are byte-identical to the Zenodo release and have not been modified by
+> [Small Models for GLAM](https://huggingface.co/small-models-for-glam).
+## Use from the Hugging Face Hub
+Install Kraken and the Hugging Face Hub CLI:
+```shell
+pip install "kraken>=7.1.0" huggingface_hub
+```
+Download the model and run recognition:
+```shell
+hf download small-models-for-glam/kraken-ppocrv6-tiny \
+    tiny.safetensors \
+    --local-dir ./kraken-ppocrv6-tiny
+kraken -i image.png output.txt \
+    segment -bl \
+    ocr -m ./kraken-ppocrv6-tiny/tiny.safetensors
+```
+## Description
+This is the **tiny** variant (~0.69M parameters) of a family of PP-OCRv6
+text-line recognition models (`tiny`, `small`, `medium`) for
+[kraken](https://kraken.re). The models are trained from scratch with baseline
++ bounding polygon data with a very diverse corpus containing historical,
+contemporary and born-digital document line images, handwritten and
+machine-printed, covering 44 languages across 10 scripts (Arabic, Armenian,
+Cyrillic, Ethiopic, Georgian, Greek, Hebrew, Latin, Malayalam, and Syriac).
+## Architecture
+PP-OCRv6 is a conventional CTC-based line recognizer consisting of a lightweight
+convolutional backbone and a non-recurrent sequence-modelling neck.
+The original architecture published as part of PaddlePaddle has been adapted
+for historical ATR by:
+* increasing line height to 128px
+* uncapping line width/CTC label budget
+* replacing the optimizer with Adam+Muon
+## Uses
+This is a base model that is supposed to produce usable output across a wide
+range of scripts and materials out of the box while also allowing fine-tuning
+with ease. It should offer similar accuracy and generalization to VLM-based
+recognizers without hallucinations and with vastly higher throughput. This
+medium variant model achieves the highest scores on the test set, `tiny` and
+`small` trade accuracy for inference speed and a smaller memory footprint.
+## Transcription guidelines, Normalization, and Transformations
+No attempt has been made to normalize the source datasets to a single set of
+transcription guidelines; the corpus mixes conventions, so inconsistent output
+is to be expected, in particular for Latin-script European manuscripts which
+mix large datasets such as CATMuS and TRIDIS that have very different
+approaches to transcription. Text was normalized to Unicode **NFD** and
+whitespace was normalized during training and evaluation.
+## Bias, Risks, and Limitations
+The training corpus is heavily skewed towards a handful of high-resource
+languages (English, French, German, Latin, Dutch, Middle French, ...). Languages
+with little real training material show markedly higher error rates and will
+require fine-tuning for practical use. Because transcription conventions are
+inconsistent across sources, the model may resolve abbreviations or expand
+glyphs unpredictably.
+The synthetic data used for training was created with the pangoline tool which
+is limited to approximating modern, machine-printed text. For the
+languages/scripts present only as synthetic data (Classical Armenian, Geʽez)
+and to a lesser extent those sharing the Latin script (Irish, Latvian,
+Lithuanian, Romanian, Serbian, Slovenian), real-world accuracy is probably
+limited.
+## How to Get Started with the Model
+Install `kraken` (>= 7.1.0), download the model, and run recognition on an
+input image:
+```shell
+kraken -i image.png output.txt \
+    segment -bl \
+    ocr -m tiny.safetensors
+```
+For more information, refer to the [documentation](https://kraken.re).
+## Training Details
+### Training Data
+The model was trained on publicly available and restricted (private) datasets.
+Datasets marked as *private* are part of the training mixture but are not
+redistributable. A **†** marks languages that were additionally
+augmented with synthetic training data (see below).
+| Language | Script | Datasets |
+|----------|:------:|----------|
+| Ancient Greek † | Greek | [EPARCHOS](https://zenodo.org/records/4095301), [HPGTR](https://github.com/vivianpl/HPGTR), [HTR_CPgr23](https://gitlab.huma-num.fr/ecrinum/anthologia/htr_cpgr23), [Stavronikita Monastery Greek Handwritten Document Collection No. 114](https://zenodo.org/records/5578251), [Stavronikita Monastery Greek Handwritten Document Collection No. 53](https://zenodo.org/records/5595669), [Stavronikita Monastery Greek Handwritten Document Collection No. 79](https://zenodo.org/records/5578136), 11 private datasets |
+| Arabic † | Arabic | [Agapet](https://zenodo.org/records/15473122), [arabic_ms_data](https://github.com/OpenITI/arabic_ms_data), [iskandar](https://github.com/OpenITI/arabic_ms_data), [Muharaf: Manuscripts of Handwritten Arabic Dataset](https://github.com/MehreenMehreen/muharaf), [OpenITI Arabic Print Data](https://github.com/OpenITI/arabic_print_data), [RASAM dataset](https://github.com/calfa-co/rasam-dataset), [TariMa](https://github.com/calfa-co/tarima) |
+| Catalan | Latin | [FONDUE-CA-PRINT-20](https://github.com/FoNDUE-HTR/FONDUE-CA-PRINT-20), [htromance-spa](https://github.com/HTRomance-Project/middle-ages-in-spain), 1 private dataset |
+| Church Slavonic | Cyrillic | 3 private datasets |
+| Classical Armenian † | Armenian | *synthetic only* |
+| Corsican | Latin | [OCR Corse](https://github.com/PSL-Chartes-HTR-Students/HN2021-OCR-Poesie-Corse) |
+| Czech † | Latin | [2024--medieval-czech-main](https://github.com/HTR-School-Vienna/2024--medieval-czech), [2023--medieval-czech](https://github.com/HTR-School-Vienna/2023--medieval-czech), [HTR Winter School 2025 - Medieval Czech - Biblioteka Jagiellonska BJ Rkp 441 IV](https://github.com/HTR-School-Vienna/2025--medieval-czech), [Paderov Bible handwriting ground truth](https://zenodo.org/records/7467034), [ehri](https://github.com/FloChiff/ehri-dataset) |
+| Danish | Latin | [ehri](https://github.com/FloChiff/ehri-dataset) |
+| Dutch † | Latin | [6000 ground truth of VOC and notarial deeds / HTR of VOC, WIC and notarial deeds](https://zenodo.org/records/11209325), [ARletta](https://doi.org/10.5281/zenodo.11191457), [Dagboek Ernest Clarysse](https://zenodo.org/records/13769222), [FONDUE-NE-MSS-17-PR](https://github.com/FoNDUE-HTR/FONDUE-NE-MSS-17-PR) |
+| English | Latin | [FONDUE-EN-PRINT-20](https://github.com/FoNDUE-HTR/FONDUE-EN-PRINT-20), [IAM Handwriting Database](https://fki.tic.heia-fr.ch/databases/iam-handwriting-database), [jcrs_train](https://zenodo.org/records/4243023), [jcrs_val](https://zenodo.org/records/4243023), [JosephHookerHTR](https://doi.org/10.5281/zenodo.8038689), [OCR-D gt_structure_text](https://github.com/OCR-D/gt_structure_text), [sloanelab](https://github.com/sloanelab-org/HTR-Model), [The Revolutionary City / RevCity documentation](https://americanphilosophicalsociety.github.io/RevCityDocs/), [Memorials for Jane Lathrop Stanford](https://github.com/PSL-Chartes-HTR-Students/HN2021-Memorials_Jane_Lathrop_Stanford), [ehri](https://github.com/FloChiff/ehri-dataset), 2 private datasets |
+| Finnish † | Latin | [NewsEye/READ OCR Finnish Newspapers](https://zenodo.org/records/4599472) |
+| French | Latin | [Antoine Verard extracts](https://github.com/LaurieHoeben/Verard-corpus), [Copiste-d-un-jour](https://github.com/banq-dcn/Copiste-d-un-jour), [corpus-HTR-lignes-mixtes](https://github.com/Front-Justice/corpus-HTR-lignes-mixtes), [dataset-celestine-doniau-danest](https://github.com/alix-tz/dataset-celestine-doniau-danest), [FONDUE-FR-MSS-19](https://github.com/FoNDUE-HTR/FONDUE-FR-MSS-19), [FONDUE-FR-MSS-19-PR](https://github.com/FoNDUE-HTR/FONDUE-FR-MSS-19-PR), [FONDUE-FR-PRINT-20](https://github.com/FoNDUE-HTR/FONDUE-FR-PRINT-20), [FONDUE-MLT-ART](https://github.com/FoNDUE-HTR/FONDUE-MLT-ART), [genauto-td-htr](https://github.com/jpmjpmjpm/genauto-td-htr), [HTR Front Justice](https://github.com/Front-Justice/htr-front-justice), [La Correspondance Doucet-Rene Jean](https://nakala.fr/10.34847/nkl.48ad8b8d), [Memoire sur St Domingue par H. M. Michel](https://zenodo.org/records/13784411), [Moonshines](https://github.com/alix-tz/moonshines), [NewsEye READ AS French Newspapers](https://zenodo.org/records/5654841), [NuBIS-OCR](https://github.com/ksefil/NuBIS-OCR), [Recensement Valaisan (Valais Time Machine)](https://github.com/PonteIneptique/valais-recensement), [Tapus Corpus](https://github.com/HTR-United/tapuscorpus), [TIMEUS Corpus](https://github.com/HTR-United/timeuscorpus), [TitresNobiliaires_17_18](https://github.com/HisMoDoc-HTR/TitresNobiliaires_17_18), [CREMMA Manuscrits du 20e](https://github.com/HTR-United/CREMMA-MSS-20), [CREMMA Wikipedia](https://github.com/HTR-United/cremma-wikipedia), [Maxime Kovalewsky - Coutume contemporaine et loi ancienne (1893)](https://github.com/PSL-Chartes-HTR-Students/HN2021-Kovalewsky-1893), [HTRomance, Modern Roman languages corpus](https://github.com/HTRomance-Project/modern-roman-languages), [Peraire Ground Truth](https://github.com/alix-tz/peraire-ground-truth), [PARES](https://zenodo.org/records/15719530), 1 private dataset |
+| Georgian † | Georgian | 15 private datasets |
+| German | Latin | [2024--medieval-german](https://github.com/JulianHelmchen/2024--medieval-german), [2025--Early-Modern-German](https://github.com/HTR-School-Vienna/2025--early-modern-german), [Bullinger Digital Gwalther handwriting ground truth](https://zenodo.org/records/5153263), [charlottenburger-amtsschrifttum](https://github.com/UB-Mannheim/charlottenburger-amtsschrifttum), [Chronicling Germany](https://github.com/Digital-History-Bonn/Chronicling-Germany-Code), [dach-gt](https://github.com/UB-Mannheim/dach-gt), [DigiTheo Ground Truth](https://github.com/tboenig/DTGT), [Dresdner Hofdiarium](https://zenodo.org/records/15303398), [Fibeln](https://github.com/UB-Mannheim/Fibeln), [FONDUE-DE-MSS-16-PR](https://github.com/FoNDUE-HTR/FONDUE-DE-MSS-16-PR), [FONDUE-DE-MSS-18](https://github.com/FoNDUE-HTR/FONDUE-DE-MSS-18), [FONDUE-DE-MSS-19-PR](https://github.com/FoNDUE-HTR/FONDUE-DE-MSS-19-PR), [FONDUE-DE-MSS-20-PR](https://github.com/FoNDUE-HTR/FONDUE-DE-MSS-20-PR), [FONDUE-MLT-ART](https://github.com/FoNDUE-HTR/FONDUE-MLT-ART), [FONDUE-MLT-PRINT-TEST](https://github.com/FoNDUE-HTR/FONDUE-MLT-PRINT-TEST), [FoNDUE_Kunsthistorisches-UZH_Archivdatenbank](https://github.com/FoNDUE-HTR/FoNDUE_Kunsthistorisches-UZH_Archivdatenbank), [Ground truth for Neue Zurcher Zeitung black letter](https://zenodo.org/records/3333627), [Hakenkreuzbanner](https://github.com/UB-Mannheim/Hakenkreuzbanner), [inzigkofen](https://zenodo.org/records/17978574), [Klosterneuburg, Stiftsbibl., Cod. 48](https://zenodo.org/records/7466928), [koenigsfelden](https://zenodo.org/records/5179361), [mkn-kurrent-gt](https://github.com/bertsky/mkn-kurrent-gt), [NewsEye / READ OCR Austrian Newspapers](https://zenodo.org/records/3387369), [nuremberg_letterbooks](https://zenodo.org/records/13881575), [OCR-D gt_structure_text](https://github.com/OCR-D/gt_structure_text), [reichsanzeiger-gt](https://github.com/UB-Mannheim/reichsanzeiger-gt), [Training Data Incunabula Reichenau](https://zenodo.org/records/11046062), [Weisthuemer](https://github.com/UB-Mannheim/Weisthuemer), [gt-fraktur](https://github.com/ubtue/gt-fraktur), [german_kurrent_handwritten_text_lines](https://zenodo.org/records/17252677), [Ground Truth (Tagebücher Edwin Hennig)](https://github.com/ubtue/Ground-Truth), [ehri](https://github.com/FloChiff/ehri-dataset), [Fanny loves Wilhelm](https://zenodo.org/records/15223355), [Frauen im Fokus](https://zenodo.org/records/11038498), [Graphemic Early New German](https://zenodo.org/records/18441031), 1 private dataset |
+| German (shorthand) | Latin | 1 private dataset |
+| Geʽez † | Ethiopic | *synthetic only* |
+| Hebrew | Hebrew | [2025-hebrew](https://github.com/HTR-School-Vienna/2025-hebrew), 2 private datasets |
+| Hungarian | Latin | [ehri](https://github.com/FloChiff/ehri-dataset) |
+| Irish † | Latin | *synthetic only* |
+| Italian | Latin | [Diario del Soldato Bruno Celestino](https://zenodo.org/records/13760586), [EpiSearch HTR](https://github.com/vedph/episearch-htr), [FONDUE-IT-PRINT-20](https://github.com/FoNDUE-HTR/FONDUE-IT-PRINT-20), [FONDUE-IT-PRINT-20-PR](https://github.com/FoNDUE-HTR/FONDUE-IT-PRINT-20-PR), [HTRogène Medieval Italian Manuscripts](https://github.com/HTRogene/italian), [HTRomance, Medieval Italian corpus of ground-truth for Handwritten Text Recognition](https://github.com/HTRomance-Project/medieval-italian), [leopardi](https://github.com/aimagelab/HWD/releases/tag/leopardi), [LiDi1.0-project](https://github.com/Giorgiaagostini/LiDi1.0-project), [LAM](https://aimagelab-legacy.ing.unimore.it/imagelab/page.asp?IdPage=46), 1 private dataset |
+| Latin | Latin | [2025--late-medieval-latin-main](https://htr-school-vienna.github.io/2025--late-medieval-latin/), [burchards-dekret-digital](https://gitlab.rlp.net/adwmainz/projekte/burchards-dekret-digital/data.git), [Caroline Minuscule ground truth](https://github.com/rescribe/carolineminuscule-groundtruth), [Carolingian Latin Group HTR Wien Winter School 2025](https://github.com/HTR-School-Vienna/2025--Carolingian_Latin-), [CREMMA Medii Aevi](https://github.com/HTR-United/CREMMA-Medieval-LAT), [DISTINGUO Latin ground truth](https://zenodo.org/records/4780947), [Eutyches](https://github.com/malamatenia/Eutyches), [FONDUE-LA-MSS-16-PR](https://github.com/FoNDUE-HTR/FONDUE-LA-MSS-16-PR), [FONDUE-LA-MSS-17-PR](https://github.com/FoNDUE-HTR/FONDUE-LA-MSS-17-PR), [FONDUE-LA-MSS-MA](https://github.com/FoNDUE-HTR/FONDUE-LA-MSS-MA), [FONDUE-LA-PRINT-16](https://github.com/FoNDUE-HTR/FONDUE-LA-PRINT-16), [HTR Winter School 2023/2024 - Late Medieval Latin, ONB 3891](https://github.com/HTR-School-Vienna/2023--late-medieval-latin), [HTR Winter School 2024/2025 - Late Medieval Latin, ONB 4135; ONB 4680](https://github.com/HTR-School-Vienna/2024--late-medieval-latin), [HTRogène Medieval Latin Manuscripts](https://github.com/HTRogene/latin), [HTRomance, Medieval Latin corpus of ground-truth for Handwritten Text Recognition](https://github.com/HTRomance-Project/medieval-latin), [notarial_charter](https://dl.acm.org/doi/abs/10.1007/978-3-032-04624-6_20), [nubis](https://github.com/ksefil/NuBIS-OCR), [OCR-D gt_structure_text](https://github.com/OCR-D/gt_structure_text), [Paris Bible Project](https://github.com/parisbible/ground_truth), [Training Data Incunabula Reichenau](https://zenodo.org/records/11046062), [Wien ONB Cod 2160 ground truth](https://zenodo.org/records/7537204) |
+| Latvian † | Latin | *synthetic only* |
+| Lithuanian † | Latin | *synthetic only* |
+| Malayalam | Malayalam | [Ground Truth data for printed Malayalam](https://doi.org/10.11588/data/L2KRZO) |
+| Middle Dutch | Latin | [data](https://zenodo.org/records/10005366) |
+| Middle French | Latin | [Cremma Medieval](https://github.com/HTR-United/cremma-medieval), [De la généalogie des dieux](https://github.com/PSL-Chartes-HTR-Students/HN2021-Boccace), [Données imprimés du 16e siècle](https://github.com/Gallicorpora/HTR-imprime-16e-siecle), [Données HTR incunables du 15e siècle](https://github.com/Gallicorpora/HTR-incunable-15e-siecle), [Données HTR manuscrits du 15e siècle](https://github.com/Gallicorpora/HTR-MSS-15e-Siecle), [Données imprimés du 18e siècle](https://github.com/Gallicorpora/HTR-imprime-18e-siecle), [Données imprimés gothiques du 16e siècle](https://github.com/Gallicorpora/HTR-imprime-16e-siecle), [Fabliaux](https://github.com/CIHAM-HTR/Fabliaux), [FONDUE-FR-AAEB-16](https://github.com/FoNDUE-HTR/FONDUE-FR-AAEB-16), [FONDUE-FR-AAEB-17](https://github.com/FoNDUE-HTR/FONDUE-FR-AAEB-17), [FONDUE-FR-MSS-18](https://github.com/FoNDUE-HTR/FONDUE-FR-MSS-18), [FONDUE-FR-PRINT-16](https://github.com/FoNDUE-HTR/FONDUE-FR-PRINT-16), [FONDUE-FR-PRINT-17](https://github.com/FoNDUE-HTR/FONDUE-FR-PRINT-17), [HTR-SETAF-Jean-Michel](https://github.com/SETAFDH/HTR-SETAF-Jean-Michel), [HTR-SETAF-LesFaictzJCH](https://github.com/SETAFDH/HTR-SETAF-LesFaictzJCH), [HTR-SETAF-Pierre-de-Vingle](https://github.com/SETAFDH/HTR-SETAF-Pierre-de-Vingle), [HTRogene French](https://github.com/HTRogene/french), [HTRomance, Medieval French corpus of ground-truth for Handwritten Text Recognition](https://github.com/HTRomance-Project/medieval-french), [Imprimés 17e siècle](https://github.com/Gallicorpora/HTR-imprime-17e-siecle), [Liber](https://github.com/CIHAM-HTR/Liber), [OCR17plus](https://github.com/e-ditiones/OCR17), [TNAH-2021-DecameronFR](https://github.com/PSL-Chartes-HTR-Students/TNAH-2021-DecameronFR), [transcription-chastel](https://github.com/LaBretelle/transcription-chastel) |
+| Multilingual (mixed) | Latin | [Training Data Incunabula Reichenau](https://zenodo.org/records/11046062), [TranscriboQuest25_MedVernacReligio](https://doi.org/10.5281/zenodo.15030337) |
+| Norwegian | Latin | [NorHand v3 / Dataset for Handwritten Text Recognition in Norwegian](https://zenodo.org/records/10255840) |
+| Occitan | Latin | [HTRogène Medieval Occitan Manuscripts](https://github.com/HTRogene/occitan), 1 private dataset |
+| Ottoman Turkish | Arabic | [mehmed_ibn_mehmed_uskubi_cukrikcizade_altiparmak_risale](https://github.com/OpenITI/arabic_ms_data), [OpenITI Arabic-script OCR Catalyst Project print/typeface data](https://github.com/OpenITI/arabic_print_data) |
+| Persian † | Arabic | [hafiz_divan](https://github.com/OpenITI/arabic_ms_data), [OpenITI Arabic-script OCR Catalyst Project print/typeface data](https://github.com/OpenITI/arabic_print_data), [sadi_gulistan](https://github.com/OpenITI/arabic_ms_data) |
+| Picard | Latin | 1 private dataset |
+| Polish † | Latin | [ehri](https://github.com/FloChiff/ehri-dataset) |
+| Portuguese | Latin | [iForal-Dataset](https://github.com/Arch-W/iForal-Dataset), [Portuguese Handwriting 16th-19th c.](https://zenodo.org/records/13986218) |
+| Romanian † | Latin | *synthetic only* |
+| Russian † | Cyrillic | 2 private datasets |
+| Serbian (Cyrillic) † | Cyrillic | *synthetic only* |
+| Slovak | Latin | [Slovensky Supermodel P&T1](https://zenodo.org/records/11218527), [ehri](https://github.com/FloChiff/ehri-dataset) |
+| Slovenian † | Latin | *synthetic only* |
+| Spanish | Latin | [FoNDUE Spanish chapbooks 19th c. Dataset](https://github.com/DesenrollandoElCordel/FoNDUE-Spanish-chapbooks-Dataset), [FONDUE-ES-MSS-19-PR](https://github.com/FoNDUE-HTR/FONDUE-ES-MSS-19-PR), [FONDUE-ES-PRINT-19](https://github.com/FoNDUE-HTR/FONDUE-ES-PRINT-19), [HTR - Araucania manuscript XIX](https://github.com/Proyecto-Ocupacion-Araucania-UChile/HTR_Araucania_XIX), [HTRogène Medieval Spanish Manuscripts](https://github.com/HTRogene/spanish), [HTRomance, Medieval Spain corpus of ground-truth for Handwritten Text Recognition](https://github.com/HTRomance-Project/middle-ages-in-spain), [ohg](https://zenodo.org/records/1322666), 3 private datasets |
+| Swedish | Latin | [Finnish Court Records-sub500](https://zenodo.org/records/3945088), [kat57 Swedish ground truth dataset](https://zenodo.org/records/14679534), [NewsEye / READ OCR training dataset from Swedish Newspapers](https://zenodo.org/records/4599624), [riskarchiv](https://huggingface.co/collections/Riksarkivet/training-data-for-swedish-lion-libre) |
+| Syriac † | Syriac | [zenodo.18157525](https://doi.org/10.5281/zenodo.18157525), [winter_school_vienna](https://zenodo.org/records/14714089), 2 private datasets |
+| Ukrainian | Cyrillic | 1 private dataset |
+| Urdu | Arabic | [OpenITI Arabic-script OCR Catalyst Project print/typeface data](https://github.com/OpenITI/arabic_print_data) |
+| Yiddish | Hebrew | 7 private datasets |
+### Synthetic Training Data
+Synthetic line images were generated as additional training material for
+18 languages: Ancient Greek, Arabic, Classical Armenian, Czech, Dutch, Finnish, Georgian, Geʽez, Irish, Latvian, Lithuanian, Persian, Polish, Romanian, Russian, Serbian, Slovenian, Syriac.
+Of these, eight are present **only** as synthetic data: Classical Armenian, Geʽez, Irish, Latvian, Lithuanian, Romanian, Serbian (Cyrillic), Slovenian.
+### Training Procedure and Hyperparameters
+The model was trained with kraken (feature/ppocrv6_rec branch).
+| | |
+|---|---|
+| Hardware | 3 × NVIDIA H100 |
+| Precision | bf16-mixed |
+| Optimizer | AdamW + Muon (momentum 0.95, weight decay 0.01) |
+| Learning rate | 1.7e-3 (cosine schedule, 1500 step warmup, min 1e-6) |
+| Batch size | 128 |
+| Gradient clipping | 1.0 |
+| Epochs | 16 |
+| Normalization | NFD + whitespace |
+| Augmentation | enabled |
+## Evaluation
+### Testing Data
+Metrics are computed on a held-out test set for each language. The test split
+was obtained by random 5% split with an upper limit of 100 document pages per
+language. No attempt has been made to split in a manner that separates documents
+between train and test. The scores below are therefore best read as in-domain
+generalization.
+Evaluations with purely synthetic data are marked with **‡**. CER and WER are
+the character- and word-level error rates, computed with `torchmetrics` using
+greedy CTC decoding and the NFD + whitespace normalization (equivalent to
+`ketos test -u NFD -n`).
+### Metrics
+| Language | Lines | CER (%) | WER (%) |
+|----------|------:|--------:|--------:|
+| Ancient Greek | 451 | 22.98 | 87.79 |
+| Arabic | 1,926 | 20.11 | 68.77 |
+| Catalan | 128 | 5.47 | 26.59 |
+| Church Slavonic | 6,599 | 17.81 | 65.44 |
+| Classical Armenian ‡ | 3,580 | 0.64 | 3.23 |
+| Corsican | 50 | 1.88 | 13.33 |
+| Czech | 922 | 15.47 | 61.28 |
+| Danish | 25 | 1.29 | 10.89 |
+| Dutch | 4,384 | 14.31 | 51.81 |
+| English | 2,005 | 14.43 | 46.77 |
+| Finnish | 7,907 | 1.16 | 6.83 |
+| French | 5,967 | 14.02 | 31.40 |
+| Georgian | 394 | 28.72 | 85.71 |
+| German | 3,007 | 4.61 | 17.96 |
+| German (shorthand) | 830 | 41.20 | 84.51 |
+| Geʽez ‡ | 2,992 | 1.27 | 5.58 |
+| Hebrew | 3,426 | 9.11 | 26.32 |
+| Hungarian | 38 | 3.20 | 23.05 |
+| Irish ‡ | 2,824 | 1.00 | 4.15 |
+| Italian | 2,611 | 7.22 | 28.24 |
+| Latin | 5,748 | 14.03 | 44.34 |
+| Latvian ‡ | 2,397 | 1.25 | 6.48 |
+| Lithuanian ‡ | 2,615 | 1.30 | 6.73 |
+| Malayalam | 59 | 48.79 | 99.21 |
+| Middle Dutch | 3,014 | 12.44 | 43.63 |
+| Middle French | 3,970 | 8.70 | 32.38 |
+| Multilingual (mixed) | 241 | 3.49 | 17.99 |
+| Norwegian | 2,335 | 14.96 | 48.04 |
+| Ottoman Turkish | 451 | 10.70 | 48.25 |
+| Persian | 990 | 9.34 | 36.76 |
+| Polish | 2,766 | 1.93 | 10.21 |
+| Portuguese | 2,763 | 23.53 | 71.50 |
+| Romanian ‡ | 2,412 | 1.50 | 6.87 |
+| Russian | 3,053 | 25.11 | 68.33 |
+| Serbian (Cyrillic) ‡ | 2,789 | 0.58 | 2.71 |
+| Slovak | 1,550 | 5.11 | 21.50 |
+| Slovenian ‡ | 2,741 | 2.42 | 6.87 |
+| Spanish | 5,465 | 6.19 | 21.49 |
+| Swedish | 3,555 | 11.89 | 45.04 |
+| Syriac | 1,801 | 10.74 | 44.40 |
+| Ukrainian | 1,253 | 13.71 | 47.56 |
+| Urdu | 1,656 | 10.20 | 41.89 |
+| Yiddish | 4,320 | 8.96 | 32.79 |
+| **Aggregate (micro-average)** | 108,010 | **8.71** | **30.00** |
+| **Aggregate (macro-average)** | | **10.99** | **36.16** |
+## License
+Released under the Apache-2.0 license.
+## Acknowledgements
+Training of this model was funded by the European Union under Grant Agreement
+No.~101132163 (ATRIUM) and No.~101071829 (MiDRASH). Views and opinions
+expressed are those of the authors only and do not necessarily reflect those of
+the European Union.
+This project also received funding from the BPI Scribe project.
+## Citation
+If you use this model, please cite kraken and if possible credit the dataset
+providers linked in the front matter and the table above.

tiny.safetensors ADDED Viewed

	@@ -0,0 +1,3 @@

+version https://git-lfs.github.com/spec/v1
+oid sha256:972700b1c72de14bdde028f9b06d77f13c88ec60da5e942bb9e57aca421fa361
+size 2837508