Benchmarks

Task NameDataset NameSOTA ResultTrendResults

HOPE 8-domain, clean-regime 256M (3 seeds)

35.8Avg PPL

HOPE

61.36Delta-1 Accuracy

HOPE 8-domain (test)

9.1Forgetting

HOPE

6.9mIoU

HOPE 8-domain, high-overlap non-adaptive stream 256M

9.4Forgetting

HOPE extended

0.478Density

HOPE BOP v2 (test)

41.1AP

HOPE 16-domain stream 256M (long continual sequence)

10.2Forgetting

HOPE BOP v2 (test)

33.5AP

HOPE v2

0.095R Error @5