PyTorch / OpenNMT / Modal
LINGO
A focused machine translation project for Kangri-Hindi that moves from dataset preparation through training and installable deployment.
return to projects
learned
- How data cleaning matters more than architecture changes for small parallel corpora.
- Why deployment format affects whether an MT model is actually usable.
- Where evaluation becomes difficult for regional and low-resource languages.
writeup queue
- Write a training-data and evaluation breakdown.
- Add examples comparing model output before and after cleanup.