Today
-
Preserve six final near-greedy RL policies
-
Preserve twelve bounded iteration policies
-
Preserve twelve matched bounded policy students
-
Preserve six public continuation value and action students
-
Preserve six public branch-feature students
-
Preserve six no-Qstar credit-horizon students
-
Preserve both fixed-visitation policy iteration rounds
-
Preserve twelve matched cost-sensitive policy students
-
Preserve both rounds of fixed and refreshed teacher students
-
Preserve six matched fitted policy-improvement students
-
Preserve predeclared exact policy-gradient checkpoints
-
Preserve six paired RL students from a preselected distilled initialization
-
Preserve two public-planner distillation replication students
-
Preserve verified public-model planner distillation student
-
Preserve analytic initialization and six verified residual policies
-
Preserve verified shared public-world myopic initialization
-
Preserve six verified direct-question on-policy actors
-
Preserve six verified direct-question on-policy actors
-
Preserve direct-question structured supervised actor
-
Preserve six verified direct-question on-policy actors
-
Preserve direct-question structured supervised actor
-
Preserve six verified direct-question on-policy actors
-
Preserve direct-question structured supervised actor
-
Preserve direct-question structured supervised actor
-
Preserve supervised executed-question policy checkpoint
-
Preserve paired canonical CE and top-one history policies
Yesterday
-
Preserve shared continuation policy trained on nonempty histories
-
Preserve shared 29-dimensional complete-slate scorer
-
Preserve public canonical-coordinate question-slate policy
-
Preserve train-only public-input whole-slate selector
3 days ago
-
Preserve decision and structured-CE grounding adapters
-
Preserve matched original and varied grounding adapters
-
Preserve deployed quadratic linear and misbound policy heads
-
Preserve deployment residual and calibration parameters
-
Preserve global grounding calibration parameter
-
Preserve grounding LoRA adapter
-
Preserve structured multimodal belief parameters
-
Preserve fixed root credit ridge policy parameters
-
Preserve fixed root credit ridge policy parameters
-
Preserve question-conditioned shared ridge parameters
-
Preserve full-context action LoRA
-
Preserve full-context action LoRA
-
Preserve six matched candidate regroup numeric heads
4 days ago
-
Preserve twelve frozen public set heads before evaluation
-
Preserve twelve frozen public set heads before evaluation
-
Preserve twelve frozen public set heads before evaluation
-
Preserve twelve frozen public set heads before evaluation
-
Preserve twelve frozen public set heads before evaluation
-
Preserve twelve frozen public set heads before evaluation
-
Preserve training-only diagnostic set head without deployment claims
