streaming on-policy distillationsupervision schedulingpublished Nov 18, 2026 · arXiv:2611.22298
Streaming On-Policy Distillation: Decoupling Three Clocks for Stage-Wise Supervision Consumption
model evaluationreadout architecturespublished Nov 14, 2026 · arXiv:2611.15592
Small Heads, Strong Decisions: A Controlled Study of Typed-Decision Models
reinforcement learningdata selectionpublished Nov 5, 2026 · arXiv:2611.06395
GIST: Gibbs-Targeted Hierarchical Data Selection for Offline RLVR
KV-cache reuseselective recomputationpublished Oct 9, 2026 · arXiv:2610.11358
RaReCache: Bridging the Gap in Cross-Model KV Cache Reuse via Rank disagreement-based Selective Recomputation
autoregressive inferenceprefill schedulingpublished Oct 1, 2026 · arXiv:2609.38386
Decode-Latency Feedback Prefill: A Model-Free Controller and Its Generalization Limits
AI evaluator auditingcounterfactual reasoningpublished Aug 24, 2026 · arXiv:2608.20938
No Judgment Without a Reason: Counterfactual Receipts for Versioned AI Evaluators