---
license: cc-by-sa-4.0
language:
  - vi
tags:
  - vietnamese
  - diacritic-restoration
  - training
size_categories:
  - 100K<n<1M
task_categories:
  - text-generation
configs:
  - config_name: wiki_500k
    data_files:
      - split: train
        path: wiki_500k.jsonl
      - split: validation
        path: wiki_val_5k.jsonl
  - config_name: news_150k
    data_files:
      - split: train
        path: news_150k.jsonl
---

# `nrl-ai/vn-diacritic-train` — Vietnamese diacritic-restoration training data

Two register-distinct training corpora used to fine-tune Vietnamese
diacritic-restoration models in the
[`nom-vn`](https://github.com/nrl-ai/nom-vn) project. Each row is a
JSONL record:

```json
{"input": "diacritic-stripped text", "target": "correctly diacriticized original"}
```

Inputs are produced by
[`nom.text.strip_diacritics`](https://github.com/nrl-ai/nom-vn/blob/main/src/nom/text/strip_diacritics.py)
on the target. Both fields are NFC-normalized.

## Configs

### `wiki_500k`

500K (input, target) pairs from
[`hirine/wikipedia-vietnamese-1M296K-dataset`](https://huggingface.co/datasets/hirine/wikipedia-vietnamese-1M296K-dataset)
(CC-BY-SA-4.0). Encyclopedic register, broad topical coverage.

Filtering pipeline (deterministic, stride=7):

1. Sentence-split each article (regex on terminator + capital VN letter).
2. Drop sentences shorter than 30 chars or longer than 300.
3. Drop sentences with ASCII ratio > 95 % (URLs, tables, code blocks).
4. Drop sentences without diacritics (no training signal).
5. Drop sentences in the held-out diacritic eval set
   (see [`nrl-ai/vn-diacritic-eval`](https://huggingface.co/datasets/nrl-ai/vn-diacritic-eval)).
6. Deduplicate exact target.
7. Stride-sample every 7th eligible sentence — diverse without RNG.

Includes a 5K held-out validation split (`wiki_val_5k.jsonl`). 0
contamination against the diacritic eval slices (audited 2026-04-30).

**License:** CC-BY-SA-4.0 (inherited from the source corpus).

### `news_150k`

150K (input, target) pairs from
[`tmnam20/Vietnamese-News-dedup`](https://huggingface.co/datasets/tmnam20/Vietnamese-News-dedup)
(CC-BY-4.0). Modern news / business register — complements `wiki_500k`'s
encyclopedic tilt.

Same filters as `wiki_500k` plus:

- **NFC normalization is critical here.** The upstream `tmnam20`
  dataset ships ~79 % of its sentences in NFD-decomposed form (e.g.
  "Cộng" stored as 'C' + 'o' + COMBINING DOT BELOW + COMBINING
  CIRCUMFLEX rather than the precomposed U+1ED9). Training a model
  on NFD targets when the eval is NFC produces silent quality
  regressions (we hit a -15.45 pp business-register regression
  before catching this). All targets in this config are NFC-normalized
  before write.

- Stride=3 (vs wiki's 7) since news articles are denser and shorter.

**License:** CC-BY-4.0 (inherited from the source corpus). More
permissive than `wiki_500k` — derivatives don't need to be share-alike.

## Loading

```python
from datasets import load_dataset

# Wikipedia 500K
wiki = load_dataset("nrl-ai/vn-diacritic-train", "wiki_500k", split="train")
print(wiki[0])
# {'input': 'Hop dong nay duoc lap...', 'target': 'Hợp đồng này được lập...'}

# Mix wiki + news for register balance (recipe used by the published
# `nrl-ai/vn-diacritic-vit5-base` training)
import random
wiki = load_dataset("nrl-ai/vn-diacritic-train", "wiki_500k", split="train").shuffle(seed=42).select(range(350_000))
news = load_dataset("nrl-ai/vn-diacritic-train", "news_150k", split="train")
mixed = (wiki.to_list() + news.to_list())
random.Random(42).shuffle(mixed)
```

## Eval-leak protection

Both configs are scrubbed against [`nrl-ai/vn-diacritic-eval`][eval]
(business / formal / conversational / literary slices). Audited 2026-04-30
on the released JSONL files: 0 hits across all 4 splits.

[eval]: https://huggingface.co/datasets/nrl-ai/vn-diacritic-eval

## License posture

Repo-level license is **CC-BY-SA-4.0** (the most restrictive of the
two configs, applied for safety). Per-config licenses above. If you
only need the more permissive subset, use `news_150k` (CC-BY-4.0).

## Citation

```bibtex
@misc{nom_vn_diacritic_train_2026,
  title={Vietnamese diacritic-restoration training data},
  author={Nguyen, Viet-Anh and {Neural Research Lab}},
  year={2026},
  howpublished={\url{https://huggingface.co/datasets/nrl-ai/vn-diacritic-train}}
}
```

Cite upstream:

- [`hirine/wikipedia-vietnamese-1M296K-dataset`](https://huggingface.co/datasets/hirine/wikipedia-vietnamese-1M296K-dataset)
- [`tmnam20/Vietnamese-News-dedup`](https://huggingface.co/datasets/tmnam20/Vietnamese-News-dedup)