English to Hiligaynon, wherever Ilonggos need it
We are College of Computer Studies students from Central Philippine University. Our goal is to bring Ilonggos their first accurate, mobile-capable translator model.
The current model translates English into Hiligaynon. Hiligaynon-to-English translation is planned next.
The team
- Joshua Silubrico, Leader
- Jerome Cabugwason, Lead and AI Software Engineer
- Nicole Angela Gorriceta, Mobile Software Engineer
- Jonas Dupaya, Researcher
Recorded model results
We tested the model on a local CPU.
- Model format
- ONNX opset 18
- Tokenizer
- SentencePiece BPE with 8,000 tokens
- Translation direction
- English to Hiligaynon
- Initial load time
- 0.11 seconds
- Inference time after load
- 26 to 67 milliseconds for one sample
- Model memory use
- Approximately 100 MB of RAM
MaralNMT keeps the model loaded in a local worker. The public API limits request rates to protect the service.
These measurements describe runtime performance on the test computer. Results can change on other hardware. They do not measure translation quality.