Frametail

Running benchmarks

Launching benchmark jobs, choosing concurrency, and monitoring progress.

Preconditions

Ensure scorers are attached and dataset rows validate in the preview step. Fix schema issues before launching large runs to avoid partial failures.

Execution

Start runs from the benchmark detail page. Long jobs may queue — watch status chips and notifications rather than refreshing constantly.

Cost awareness

Video benchmarks consume compute and third-party inference depending on your configuration. Pilot with a subset of rows first.