ningpy/brunei-malayu-translator
The ningpy/brunei-malayu-translator is a 0.8 billion parameter language model fine-tuned from Qwen3-0.6B by ningpy, specifically designed for translating Brunei Malay to Standard Malay. This version (v3) includes an enhanced vocabulary of approximately 1,660 medical and clinical dialect terms. It achieves a BLEU score of 60.57 and a chrF++ of 79.53 on its overall test set, making it highly effective for specialized translation tasks within the Malay language variants.
Loading preview...
Brunei Malay to Standard Malay Translator (Medical-Enhanced)
This model, developed by ningpy, is a specialized translation tool fine-tuned from the Qwen3-0.6B architecture. Its primary function is to accurately translate text from Brunei Malay to Standard Malay. A key differentiator for this v3 release is the significant enhancement of its medical and clinical vocabulary, incorporating approximately 1,660 dialect terms curated from a Brunei health-domain wordlist.
Key Capabilities & Performance
The model demonstrates strong performance on its held-out test set of 1,623 examples, achieving an overall BLEU score of 60.57 and a chrF++ score of 79.53. Notably, it shows robust performance on medical-specific sentences with a chrF++ of 89.14. This represents an improvement over its predecessor (v2), with a +2.7 increase in overall BLEU and +6.5 in adversarial BLEU.
Training involved LoRA fine-tuning on 29,789 sentence pairs, with medical sentences upsampled to improve domain-specific translation quality. Data sources included a general Brunei dictionary and a curated list of medical terms, with sentence-level expansion performed using OpenAI's gpt-4o-mini.
Good For
- Specialized Translation: Ideal for translating Brunei Malay text into Standard Malay, particularly in contexts requiring accurate handling of medical and clinical terminology.
- Domain-Specific Applications: Useful for applications in healthcare, public information, or any field where precise translation between these Malay dialects, especially concerning medical terms, is critical.
Limitations
Users should be aware that the test set was drawn from the same synthesis distribution as the training data, which might lead to lower real-world BLEU scores. Rare Brunei words appearing fewer than three times in the training data may also be mistranslated. The model currently does not support reverse translation from Standard Malay to Brunei Malay.