🇬🇧 → 🇲🇾 English to Malay Translator
A custom 6+2 Tied Transformer trained from scratch on 2M OpenSubtitles sentence pairs.
Model: AstralPotato/en-ms-transformer | chrF: 52.14 (case-normalized, cleaned, beam=5) | Params: ~27M
2 10
Examples
ℹ️ About this model
- Architecture: 6-layer encoder + 2-layer decoder, pre-norm Transformer
- Tokenizer: 16K shared BPE
- Training data: 2M filtered OpenSubtitles en-ms pairs
- Trained for: IT3103 Advanced Topics in AI — Assignment 2, 2025 Semester 2
The model outputs lowercase text (tokenizer normalizes to lowercase), with post-processing to capitalize the first letter. Beam search is slower but may produce higher-quality translations.