Another try with bokeh2iframe()

Try to make bokeh plots low-weight with bokeh2html()

Added cache self.leaderboard_scatter_plots

Added cache self.leaderboard_heatmaps

Heatmap: Added model_size_threshold constant

Heatmap: Fixed typo from singular to plural in axis labels

Heatmap: Rotate the graph from horizontal to vertical for easier reading and add the transpose parameter for easier switching in the future

Heatmap: First the larger models, then the smaller ones

Heatmap: Aesthetic removal of gaps in axis labels

Heatmap: Change from abbreviations to category and task names

Heatmap: Sort by category only when viewing all categories

Heatmap: Second heatmap added

Heatmap: Fix space typo

Heatmap: If the size of the heatmap is not specified, it will be calculated

Heatmap: Abbreviations for categories, created functions _prepare_tasks_categories_abbreviations() and get_leaderboard_heatmap(), and modification of create_heatmap()

Added the "precision" attribute to the leaderboard table

The "chat_template" attribute in the table is now a sign of its presence

Truncate value of "chat_template" for table on web

Interpretation of None changed from "null" to "none"

Changed order of attr. in LB table

Changed abbr. CAT → CHAT and explanation of abbr. ML

Added more model attributes to the leaderboard table

Added selector of p_value

Added info message

Uploading corrected tournament_results

Optimization of the significance correction

Added tqdm to _correct_significance_in_tournament_results

Correcting significance in `tournament_results` with FDR correction method

Revert "Bokeh figure - changed Y axis label"

Removed unnecessary calls of print()

Renamed `get_bokeh_figure` to `get_leaderboard_scatter_plot`

Bokeh figure - changed Y axis label

Bokeh figure - change category according to the dropdown

Try analyze winscore with bokeh

Rename the submission date column from Release to Submitted

Reduced processing in the critical section

Tuning the duration of some notifications

Upload only when it could be changed

Do not read big JSON file twice

More info notifications

Default sorting of leaderboard is by "average_score"

For testing purpose improved fake tournament

Locks only on confirmation of submission

Now only executes the duels that are missing

Upload to the dataset the map from "submission_id" to "model_title"

Get task names only from "tasks_metadata.json"

Use specified metric for every task

Prepare to pandas 3.0

Repeat when connection to check significance server is timed out

Better check significance server reachable test