TL;DR
It is asked whether a single success-probability signal can allocate compute more efficiently in both phases on Countdown with Qwen2.5-0.5B.
This work casts per-step rollout collection as a budget-constrained sequential allocation problem and introduces SARA (Sequential Adaptive Rollout Allocation), a two-threshold, SPRT-style rule that commits effective groups, abandons saturated ones after a short probe, and reallocates the freed budget to fresh prompts,...
Per-Problem Base Anchoring is a deliberately simple proof-of-concept that improves both high-budget coverage over matched GRPO and ECCV-relevant vision-language agents when repeated visual, spatial, or chart-reasoning attempts are checked by external tools or verifiers.
TrajVal, a lightweight probe-based estimator that approximates per-task learnability from a short probe run and two endpoint evaluations, is proposed and it is found that learnability is reproducible across independently sampled training contexts and predictive of downstream utility.
A reasoning model is built that adaptively chooses how much to reason for each problem, and the brief modes end up more accurate than \textsc{Long}, which shows that the router sorts problems by difficulty rather than at random.
VarIance Guided Online Rollout allocation (VIGOR) is proposed which instead of allocating a fixed rollout budget per example, begins with a small number of rollouts for all examples in a batch and iteratively allocates additional rollouts to those with the highest group reward variance until a fixed total rollout budge...
Results show that test-time compute can be more effective when used to refine sampled trajectories rather than only to sample more candidates or rely on verifier-guided selection.
We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.
