aabbdev/StateBench-v1
StateBench v1 110 million verified state episodes for pretraining, SFT, and on-policy SDFT Train models to retain, edit, address, and transform state across long execution traces. Choose a configuration Configuration Scale Use when default 10,000,000 programs · 7.80 GiB You need 36 explicit families and varied surfaces dense-100m 100M episodes · 98.61 GiB You need high-throughput state training with grounded, structurally varied… See the full description on the dataset page: https://huggingface.co/datasets/aabbdev/StateBench-v1.
Consolidate StateBench to corrected dense-100m
Restore audited dense-100m v1.4 (part 2)
Restore audited dense-100m v1.4
Mark dense-100m restoration in progress
Consolidate StateBench to corrected dense-100m
Publish dense-100m v1.4 manifest
Replace dense-100m with audited v1.4 (part 2)
Replace dense-100m with audited v1.4
Warn about dense-100m v1.3 semantic issue
Document consolidated corrected dense-100m
Remove superseded dense configurations
Publish corrected dense-100m manifest
Replace dense-100m with corrected v1.3 (part 2)
Replace dense-100m with corrected v1.3
Document grounded-max-100m release
Publish grounded-max-100m manifest
Add StateBench grounded-max-100m (part 2)
Add StateBench grounded-max-100m
Document dense-max-100m release
Fix dataset card tables and capability coverage
Document dense-max-100m release
Document dense-max-100m release
Publish dense-max-100m manifest
Add StateBench dense-max-100m (part 2)
Add StateBench dense-max-100m
Rewrite dataset card for default and dense-100m
Document dense-100m configuration
Publish dense-100m manifest
Add StateBench dense-100m C++ curriculum (part 2)
Add StateBench dense-100m C++ curriculum
Publish audited StateBench v1.1 data revision
Redesign dataset card for pretraining, SFT, and SDFT
Publish StateBench v1 dataset shards (part 4)
Publish StateBench v1 dataset shards (part 3)
Publish StateBench v1 dataset shards (part 2)
Publish StateBench v1 dataset shards
initial commit