What you get
Each example is one Bangla prompt with a label from 0 to 3. All three written forms of a prompt are in the same split, so a model never sees one form in training and another in testing.
| Split | Phase | Released | Use it to |
|---|---|---|---|
| Train | Development | Oct 20, 2026 | Train your system |
| Dev | Development | Oct 20, 2026 | Tune your system and practice submitting on CodaBench |
| Test | Evaluation | Jan 15, 2027 | Make your final predictions. New prompts, no labels. |
Three written forms
Every prompt appears three times, once in each form, always with the same label:
- Bangla script: written in Bangla letters.
- Romanized: Bangla typed in Latin letters.
- Code-mixed: Bangla mixed with English, Hindi, or Urdu.
A good system gives the same label to all three. Click the tabs in the example to switch forms.
আজ ঢাকায় আবহাওয়া কেমন থাকবে?
"What will the weather be like in Dhaka today?"
Same request, same label, in every form.
Starter kit Released Oct 20, 2026
Download it with the data. It gets you from zero to a valid CodaBench submission.
Train and dev sets
The labeled data for building and tuning your system.
Starter notebook
A Jupyter notebook (.ipynb) with a simple baseline that solves the task end to end. Start here and improve it.
prepare_submission_file notebook
Turns your predictions into the exact file format CodaBench accepts.
Download links will appear here on October 20, 2026.