TAB learns turn-adaptive token budgets via GRPO in a multi-objective MDP formulation, delivering up to 35% token savings on mathematical reasoning tasks compared to static baselines.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Not All Turns Are Equally Hard: Adaptive Thinking Budgets For Efficient Multi-Turn Reasoning
TAB learns turn-adaptive token budgets via GRPO in a multi-objective MDP formulation, delivering up to 35% token savings on mathematical reasoning tasks compared to static baselines.