Loading page…
ConnectingConnecting to the coordinator at https://api.34-93-70-171.sslip.io…
Loading page…
Every checkpoint is compared with the last accepted one on a fixed evaluation set. It passes only if, with 95% confidence, it is no more than 2.0 accuracy points worse. The policy was fixed before training started.
No evaluations yet
The first result appears when the first checkpoint is committed and verifiers have scored it.