Paper Detail
Lijie Yang, Hongyin Luo, Jiawei Zhao, Tri Dao, Ravi Netravali
Test-time compute scaling is a primary driver of performance in large reasoning models (LRMs), but extreme inefficiency bounds current approaches, shifting the critical question from how much compute to spend, to where to allocate it. We formalize test-time reasoning as a constrained compute allocation problem over partial trajectories. Under a fixed hardware budget, existing paradigms fail to actively allocate the compute to the most promising partial progress: traditional parallel sampling treats traces independently and induces severe memory bottlenecks, while subtractive pruning starves hardware and fails to actively and sufficiently shift the output distribution. To overcome this dichotomy, we introduce Gambit, an inference algorithm that executes thought-level beam search. By periodically pruning unpromising trajectories and immediately branching from high-quality prefixes, Gambit dynamically concentrates compute onto the most promising reasoning traces via a light-weight scorer probing hidden states while maintaining continuous high hardware utilization. Extensive evaluations across multiple models and benchmarks demonstrate that Gambit strictly dominates existing baselines. Under identical hardware constraints, our method yields up to a +6.7\% absolute accuracy gain on HMMT-24 and +3.3\% on AIME-25 over pruning baselines, delivers >2times higher throughput on trace completion, and reduces total token consumption by up to 68.5\% relative to standard parallel sampling.
No structured notes yet. Add `summary_sections`, `why_relevant`, `claim_impact`, or `next_action` in `papers.jsonl` to enrich this view.
No ranking explanation is available yet.
No tags.
@misc{yang2026thought,
title = {Thought-Level Beam Search for Reasoning},
author = {Lijie Yang and Hongyin Luo and Jiawei Zhao and Tri Dao and Ravi Netravali},
year = {2026},
abstract = {Test-time compute scaling is a primary driver of performance in large reasoning models (LRMs), but extreme inefficiency bounds current approaches, shifting the critical question from how much compute to spend, to where to allocate it. We formalize test-time reasoning as a constrained compute allocation problem over partial trajectories. Under a fixed hardware budget, existing paradigms fail to actively allocate the compute to the most promising partial progress: traditional parallel sampling tre},
url = {https://huggingface.co/papers/2608.08020},
keywords = {test-time compute scaling, large reasoning models, constrained compute allocation, partial trajectories, parallel sampling, subtractive pruning, thought-level beam search, Gambit, hidden states, hardware utilization, code available, huggingface daily},
eprint = {2608.08020},
archiveprefix = {arXiv},
}
{}