Sinhala MMLU Shared Task — ICTer 2026

Official submission organisation for the Sinhala MMLU Shared Task at ICTer 2026, Colombo, Sri Lanka.

For participating teams

Registered teams upload their final system here after the Codabench leaderboard closes on 22 October 2026. Do not upload before that date.

One repository per team, named team-<yourteamname>, containing:

README.md - model card and reproduction manifest (copy team-template)
run.sh - turns the released test file into predictions.jsonl
requirements.txt - pinned versions, or a Dockerfile
src/ - your code
trained weights, in this repo or a linked one

Trained weights are uploaded to this organisation as model repositories. The adapter if you trained one, full weights if you fine-tuned the whole model, and nothing if you used an open model unchanged. Do not re-upload unmodified base models.

Start by duplicating team-template, which has the required structure and the model card to fill in.

What to upload

Upload what you trained, not what you downloaded:

Your situation	Upload
No training. Open model used as-is	Nothing. Name the model id and commit in your card.
LoRA or other adapter	The adapter only.
Full fine-tune or merged model	The full weights.
Ensemble	One repository per trained component.

Joining

Questions: sinhalammlu.icter2026@gmail.com