tranguyenxuwu/javicorpus-chatml-translation

JaViCorpus ChatML Translation Dataset (Full)

Bidirectional Japanese↔Vietnamese translation dataset in ChatML format, built from high-quality parallel corpora for SFT fine-tuning of large language models.

Dataset Description

This dataset contains ~480K ChatML-formatted translation examples derived from the ngovinhtn/JaViCorpus parallel corpus collection.

Each example is a 3-turn conversation (system → user → assistant) where the user provides a sentence in the source language and the assistant provides the translation.

Source Corpora

Corpus	Sentence Pairs	Domain
TEDjavi_106K	106,000	TED Talk subtitles
wiki_20K	20,000	Wikipedia / ALT Treebank
Tatoeba_2K	2,000	Simple everyday sentences
Glosbe282K	~210,000 (filtered)	Mixed domains

Key Features

• Bidirectional: Each sentence pair generates both JA→VI and VI→JA examples (2× amplification)
• 8 system prompt variants per direction to prevent prompt overfitting
• Vietnamese Unicode normalization applied via NFC standard
• Quality filtered: Removes empty pairs, punctuation-only, and gross length misalignments (>10× ratio)

Data Format

Each example follows the ChatML conversation format:

{
  "messages": [
    {"role": "system", "content": "Translate the given Japanese text to Vietnamese. Ensure the translation is fluent and preserves the original meaning."},
    {"role": "user", "content": "今日はとても良い天気ですね。"},
    {"role": "assistant", "content": "hôm nay thời tiết rất đẹp nhỉ."}
  ]
}

Splits

Split	Examples	Description
train	~456,000	95% of main corpora, shuffled
validation	~24,000	5% of main corpora, shuffled
test	986	TED dev2010 + tst2010 (held-out)

Usage

from datasets import load_dataset

ds = load_dataset("tranguyenxuwu/javicorpus-chatml-translation")
print(ds["train"][0])

For SFT Training (e.g., with TRL/Unsloth)

from trl import SFTTrainer

trainer = SFTTrainer(
    model=model,
    train_dataset=ds["train"],
    eval_dataset=ds["validation"],
    max_seq_length=4096,
)
trainer.train()

Intended Use

• Fine-tuning LLMs for Japanese↔Vietnamese machine translation
• Benchmarking translation quality on the JA-VI language pair
• Research on low-resource Asian language pair translation

Limitations

• Single sentence pairs: Each example is one sentence, so models fine-tuned on this data will tend to produce single-sentence outputs
• Lowercase Vietnamese: TED subtitle data is lowercase, which propagates to model outputs
• Domain bias: Predominantly subtitles and encyclopedic text — may not generalize to legal, medical, or highly technical domains

Citation

If you use this dataset, please cite the original corpus:

@inproceedings{Ngo2018Combining,
  author    = {Thi{-}Vinh Ngo and Thanh{-}Le Ha and Phuong{-}Thai Nguyen and Le{-}Minh Nguyen},
  title     = {Combining Advanced Methods in Japanese-Vietnamese Neural Machine Translation},
  booktitle = {Proceedings of the 10th International Conference on Knowledge and Systems Engineering (KSE 2018)},
  year      = {2018},
  address   = {Hochiminh City, Vietnam},
  url       = {https://arxiv.org/pdf/1805.07133.pdf}
}

License

The code is licensed under Apache 2.0. The parallel corpora follow the licensing policies of their original sources (TED, Glosbe, OPUS, ALT) and are restricted to research purposes only — no commercial usage permitted.