Load BUDOVA text corpus
Five lines of datasets.load_dataset() — plus how to filter by domain and inspect the NER annotations.
Coming soonColab tutorials
Notebooks
Runnable Colab notebooks that demonstrate how to work with the BUDOVA corpus — load data, train baselines, evaluate.
Fine-tune XLM-R on BUDOVA NER
Baseline named-entity recognition on the eight BUDOVA entity types. Includes evaluation on the held-out test set.
Coming soonASR evaluation on construction speech
Evaluate an off-the-shelf Ukrainian ASR on BUDOVA speech samples and compare WER across dialects and domains.
Coming soonLexicon-aware tokeniser
A minimal BPE tokeniser primed on BUDOVA lexicon that reduces token count for construction-domain prompts.
Coming soonCorpus statistics and coverage
Reproduce every number shown on the landing page — coverage, per-domain counts, IAA.
Coming soonPush your own subset to Hugging Face
Package an annotated subset from the platform export and publish it to a Hugging Face dataset repository.
Coming soonNotebooks are authored and released alongside each dataset version. Links become active once v1.0 ships.