Merge branch 'main' of hf.co:spaces/CZLC/BenCzechMark

content.py

@@ -10,6 +10,8 @@ Here, you can compare models on tasks in the Czech language or submit your own m
 - Visit the **Submission** page to learn about how to submit your model.
 - Check out the **About** page for a brief overview of our evaluation protocol, win score mechanism, citation details, and future plans for this benchmark. 
 - __How scoring works__:
+  - On each task, we score every model using one of our metrics (Accuracy for multiple choice tasks, Word Perplexity for language modeling, AUROC for classification).
+  - On each task for each model pair, we perform a _duel_: a statistical significance test (with a 5% alpha level) to determine if the model's improvement in the metric is significant.
   - For each task, the __Duel Win Score__ reflects the proportion of duels a model has won.
   - Category scores are calculated by averaging scores across all tasks within that category. When viewing a specific category (other than Overall), the "Average" column displays the Category Duel Win Scores.
   - The __Overall__ Duel Win Score is the average across all category scores. When selecting the Overall category, the "Average" column shows the Overall Duel Win Score.