Creates a new leaderboard.
Request body
If the leaderboard should be added to a preexisting benchmark, the benchmark id.
When no BenchmarkId is provided, the name of the new benchmark to create.
Optional plain-text credit for the people or sources behind the benchmark created via BenchmarkName. Limited to 2000 characters; ignored when BenchmarkId is set.
The name of the leaderboard.
The instruction datapoints will be matched up against.
Indicates if the prompt is shown on the rapids.
Whether the prompt asset should be shown on the rapids.
Total amount of responses that get collected per run.
The minimum amount of responses that need to be collected per comparison.
Whether the results should be inversed, selecting the worse model.
Optional audience id; defaults to the global audience when unset.
Optional priority override applied to every run's job; unset uses the default.
Multiplier applied to this leaderboard's match results when they contribute to the parent benchmark's overall / combined scoreboards. Defaults to 1.0; unbounded.
Additive offset applied to displayed scores on this leaderboard's scoreboard; defaults to 0.
Multiplicative factor applied to displayed scores on this leaderboard's scoreboard, relative to the Bradley-Terry center. Defaults to 1.0; must be strictly positive.
Whether the leaderboard should be created hidden.
Whether this leaderboard is excluded from its benchmark's overall / combined scoreboard, making it a private experiment of the owning organization. Defaults to false.
Prompt tag values to restrict matchups to; only prompts carrying at least one are used. Empty means no restriction.
Prompt tag values the leaderboard never creates matchups for; takes precedence over includedTags, even when another tag is included.
Set to true to skip the initial standings run. New participants are still compared against the whole existing field and boost runs still work; there is just no run collecting the first responses. Defaults to false.
Response
OK
The id of the created leaderboard.
The id of the benchmark the leaderboard belongs to.
Whether the prompt is shown on the rapids.
Whether the prompt asset is shown on the rapids.
The response budget per run.
The minimum responses per comparison.
Whether the leaderboard results are inversed.