English-to-Dagbani Seq2Seq Translation Model — translation_v1
A custom Seq2Seq Encoder-Decoder Transformer translation model trained from scratch to translate English to Dagbani (ISO 639-3 dag).
Metrics (Gold Standard Test Set)
| Metric | Value |
|---|---|
| Corpus BLEU | 0.00 |
| Corpus chrF++ | 0.00 |
| Parameters | 81.2M |
| Bilingual Tokenizer | dag_en_unigram_24k (SentencePiece unigram, joint 24k) |
Training Details
- Dataset: Clean English-Dagbani parallel corpus (~78k sentence pairs).
- Architecture: 6 encoder layers, 6 decoder layers, SwiGLU activation, RoPE positional embedding.
- Source Repository: https://github.com/pious2847/Dagbani-NLP_V2
Usage
The checkpoint can be loaded and executed using the source code in src/translation/eval_translation.py.