Datasets:
Call for results: add your decision model to the leaderboard
LocalLLaMA/typed-decisions is now an official Hugging Face benchmark with four leaderboards (accuracy, kl_from_gold, brier, ece) on the test split: 400 cases, 2,000 decisions. Each case has five typed questions (yes/no, choice or score) and every answer is a probability distribution.
How to score. Send each whole case, the state plus all five questions, in one request, against the test split. The dataset README documents the request shape and the metric definitions.
How to add your result. Open a PR to your own model repo adding .eval_results/typed-decisions.yaml with one entry per task_id (accuracy, kl_from_gold, brier, ece) and dataset id LocalLLaMA/typed-decisions.
Say how the model was trained. Use the entry's notes field to say whether the model was trained on the train split. The README lists fine-tuned models in a separate table.