Lab
Passphrase for the runner portal. Not the admin key.
Set BENCHMARKER_RUN_HASH from benchmarker hash-passphrase, then restart.
Run evaluation
By default, models are evaluated on the official stratified hard set, and eligible results can enter the public ranker. You can also use a private bank for a single run by pasting data, uploading a file, providing an HTTPS URL, or loading a Hugging Face dataset. Results from custom banks do not enter the public ranker.