streaming on-policy distillationsupervision schedulingpublished Nov 18, 2026 · arXiv:2611.22298
Streaming On-Policy Distillation: Decoupling Three Clocks for Stage-Wise Supervision Consumption
Monte Carlo tree searchreinforcement learningpublished Nov 18, 2026 · arXiv:2611.22301
MCTS Is Counting the Wrong Thing: Learning Exploration Counts for LLM Reasoning
speculative decodingLLM inferencepublished Nov 18, 2026 · arXiv:2611.22296
SpecSearch: Towards Hardware-Aware Efficient Speculative Decoding
model evaluationreadout architecturespublished Nov 14, 2026 · arXiv:2611.15592
Small Heads, Strong Decisions: A Controlled Study of Typed-Decision Models
forward-KL distillationreasoning distillationpublished Nov 14, 2026 · arXiv:2611.14850
Effective Teacher Support for Reasoning Distillation
block draftingtree verificationpublished Nov 11, 2026 · arXiv:2611.13017
Multi-Future Drafting: Mixture Block Drafters for Tree-Verified Speculative Decoding