Datasets:

Archatext
/

AraMS-28k-HTR

Exception:    SplitsNotFoundError
Message:      The split names could not be parsed from the dataset config.
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
                  for split_generator in builder._split_generators(
                                         ~~~~~~~~~~~~~~~~~~~~~~~~~^
                      StreamingDownloadManager(base_path=builder.base_path, download_config=download_config)
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/folder_based_builder/folder_based_builder.py", line 246, in _split_generators
                  raise ValueError(
                      "`file_name`, `*_file_name`, `file_names` or `*_file_names` must be present as dictionary key in metadata files"
                  )
              ValueError: `file_name`, `*_file_name`, `file_names` or `*_file_names` must be present as dictionary key in metadata files
              
              The above exception was the direct cause of the following exception:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/split_names.py", line 68, in compute_split_names_from_streaming_response
                  for split in get_dataset_split_names(
                               ~~~~~~~~~~~~~~~~~~~~~~~^
                      path=dataset,
                      ^^^^^^^^^^^^^
                      config_name=config,
                      ^^^^^^^^^^^^^^^^^^^
                      token=hf_token,
                      ^^^^^^^^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
                  info = get_dataset_config_info(
                      path,
                  ...<6 lines>...
                      **config_kwargs,
                  )
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
                  raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
              datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

AraMS-28k-HTR

Recognition-ready release of AraMS-28k: cropped line images paired with normalized ground-truth text, ready for direct use with standard HTR/CTC/seq2seq training pipelines (including Kraken).

This is a derived format. For the full annotation release (page images, geometry, raw + normalized transcriptions, insertion anchors, review metadata), see AraMS-28k.

For the script that generates this release from AraMS-28k (scripts/build_data.py), see **GitHub: **.

Samples from data

Naskh, Ruq'ah, and Maghrebi hands sampled from the corpus.

This repo distributes a single archive, AraMS-28k-HTR.zip. It is download and extract locally before use.

AraMS-28k-HTR/
├── images/{book_id}_{page_id}_{line_idx}.png
├── images/{book_id}_{page_id}_{line_idx}.gt.txt
├── dataset_stats.json
├── metadata.csv
├── split_used.json
├── train_manifest.txt
├── val_manifest.txt
└── test_manifest.txt

Each .png is a cropped, single-line image. Each .gt.txt holds the corresponding gt_normalized transcription — diacritic-stripped and matched to what's visually present in the manuscript hand, not the fully vocalized reference (see the AraMS-28k paper, Sec. 4, for why gt_normalized rather than gt_raw is the recommended training target). manifest.csv lists every record with book_id, page_id, line_idx, line_type, split, and path.

Both main-text and margin lines are included (margins are not filtered out).

Download & Checksum

File	SHA-256
`AraMS-28k-HTR.zip`	`8c4acccefae78f3940deb26af91a72d6aad80f67de175a287669dd9e297b2091`

# download the zip from the "Files" tab above, then verify:
sha256sum AraMS-28k-HTR.zip
# should match the hash above

unzip AraMS-28k-HTR.zip

Split

Splits are formed at the book level, so no manuscript appears in more than one split — matching the splits distributed with AraMS-28k (splits/{train,val,test}_books.txt in the full release).

Split	Books	Lines
train	9	19,739
val	2	1,486
test	3 (one per script)	6,744

Quickstart: finetuning Kraken

unzip AraMS-28k-HTR.zip


ketos train \
  --load YOUR_BASE_MODEL.mlmodel \
  -t train_manifest.txt \
  -e val_manifest.txt

# optional: compile to binary first for faster training
ketos compile -f binary -o train.arrow $(cat train_manifest.txt)
ketos compile -f binary -o val.arrow   $(cat val_manifest.txt)

Reproducing this release from AraMS-28k

python scripts/build_htr_data.py \
    --input_dir ./anntations \
    --images_root . \
    --output_dir ./AraMS-28k-HTR \
    --split_cfg splits/split.json

Script and full instructions: GitHub repo.

License

Released under CC BY-NC-SA 4.0, matching the parent AraMS-28k release. See the AraMS-28k dataset card for licensing details on underlying reference transcriptions.

Citation

@article{guechaoui2026arams28k,
  title={AraMS-28k: The Largest Publicly Released Line-Level Dataset of Historical Arabic Manuscripts with Margin and Insertion-Anchor Annotations},
  author={Guechaoui, Mohamed and Zellagui, Mohamed Diaa and Chaib, Souleyman and Dhelim, Sahraoui},
  journal={arXiv preprint arXiv:2608.26921},
  year={2026},
  doi={10.48550/arXiv.2608.26921}
}