Sign in
Benchmark
SampleGeneration

Reports how many samples the given configuration would generate, and for which participants.

Takes the same body as POST /benchmark/{benchmarkId}/sample-generation, so the same payload can be previewed and then submitted unchanged. Writes nothing and generates nothing. The numbers are a snapshot: in target mode they are measured against what exists and what is still being generated, both of which move, so a run submitted later may differ.

post/benchmark/{benchmarkId}/sample-generation/preview

Path parameters

benchmarkIdstring required

The benchmark the configuration would run against.

Request body

mode'Additive' | 'Target'
samplesPerPromptinteger required

How many samples to request per matching prompt (1 to 16).

participantIdsstring[] nullable

Restrict generation to specific participants. Defaults to all participants in the benchmark that have a configured faucet.

promptIdentifiersstring[] nullable

Restrict generation to specific prompt identifiers. Defaults to every prompt in the benchmark.

tagsstring[] nullable

Restrict generation to prompts having at least one of these tags.

Response

OK

totalCountinteger required

How many generation items the run would queue.

queuedSampleCountinteger required

How many samples those items would produce, and therefore bill for.

skippedParticipantIdsstring[] required

Participants that matched the selection but would be skipped because they have no faucet configured. Empty when explicit participant ids were supplied.

satisfiedParticipantIdsstring[] required

Participants the run would generate nothing for because they already hold the requested count. Distinct from SkippedParticipantIds, which is about participants that cannot generate at all. Always empty in additive mode.

Changes

Changed in 1 of the 14 revisions of this API.1