-
Add tb7s23-20261007-v2: per-seed relearn splits for seeds 2 and 3, bundle v2 (supersedes v1) (part 2)
-
Add tb7s23-20261007-v2: per-seed relearn splits for seeds 2 and 3, bundle v2 (supersedes v1)
-
Add tb7s23-20261007-v1: 7B internalization testbed, seeds 2 and 3 (code bundle and data) (part 2)
-
Add tb7s23-20261007-v1: 7B internalization testbed, seeds 2 and 3 (code bundle and data)
-
Publish single friend-facing SkillForget H1008 start page and verified release links
-
Release SkillForget H1008 v1: CPU accepted training capsule, data, frozen qualification, environment lock and runbook
-
Final controller admission fix: exclude only controller own CUDA metadata context from busy-device detection
-
Freeze final portable candidate03 with public model mirror, notices and full frozen parser tests
-
Mirror exact NVIDIA OpenMath-Nemotron-1.5B snapshot for SkillForget handoff with upstream model card
-
Portable H100 candidate02: source binding fix, locked CUDA extras and handoff instructions
-
Stage SkillForget H1008 portable CPU acceptance candidate (not yet qualified)
-
initial commit