StrictPRMBench is a replayable evaluation benchmark designed to audit fixed‑score PRM aggregation rules. It standardizes mathematical reasoning pools, scorer evidence, and deletion interventions so that different reducers confront identical selection problems. Across seven checkpoints and two benchmarks, the benchmark reveals that product aggregation yields higher deletion‑induced selection instability than geometric mean, with statistically significant differences in most cells.
By Pengfei Zheng
arXiv:2606. 19808v1 Announce Type: new Abstract: Test-time reasoning is increasingly used as a serving-time control knob, but extra reasoning is not uniformly valuable: it can repair failed attempts, waste compute on already-correct answers, or introduce harmful answer changes.
By Sajib Acharjee Dip, Dawei Zhou, Liqing Zhang
The paper introduces Concise-9B, a version of the Qwen3.5-9B reasoning model that reduces unnecessary thinking tokens through concise self‑training. By applying two rounds of on‑policy rejection‑sampling fine‑tuning and a small share of answer‑synthesis targets, Concise-9B cuts token usage to roughly two‑thirds of the original model while maintaining comparable accuracy on a wide range of hard mathematics and reasoning benchmarks. The study also shows that most of the token savings occur after the answer has already formed, and that the approach’s effectiveness depends on the verbosity of the base model.
By Alexander Boesgaard Lorup
arXiv:2608. 05643v1 Announce Type: new Abstract: Test-time scaling improves LLM reasoning by using additional inference compute, but wider sampling alone can suffer from diminishing returns: new rollouts often repeat existing answer patterns instead of adding useful reasoning diversity.
By Ahsan Bilal, Muhammad Ahmed Mohsin, Muhammad Umer, Lena Trigg, Ali Subhan, Muhammad Ali, Dean F. Hougen
arXiv:2608. 15445v1 Announce Type: new Abstract: When a reward is correct on every training example yet consistent with more than one goal, a model can acquire an unintended one, a failure known as goal misgeneralization.
By Suyash Maniyar, Armaan Sandhu, Abhishek Mishra
The paper investigates whether large language models’ reasoning traces truly contain early, informative signals or merely reflect budget and difficulty confounds. Using a restart‑controlled truncation probe, the authors compare continuation success rates against from‑scratch restart curves across 178 problem‑model pairs, finding that only one case shows prefix‑limited success and that continuing a model’s own prefix generally outperforms restarting. A difficulty‑controlled test and two generation‑free analyses reveal that early internal signals do not carry outcome information beyond a problem‑difficulty baseline, underscoring the need for proper counterfactual controls.
By Yigit Utku Bulut
arXiv:2606. 16999v1 Announce Type: cross Abstract: Frozen small code models ( =45.
By Mehmet Iscan
arXiv:2609.05637v2 Announce Type: replace
Abstract: A popular way to improve Retrieval-Augmented Generation (RAG) is to rewrite the user's question into several variants and search with all of them....
By Sara Shanian, Xiaoqin Yi, Pavlo Ruban, Kurt MacDonald
The paper studies how voting over multiple large language model (LLM) responses can be optimized under a fixed call budget. It introduces a recoverability threshold that quantifies the gap between discovering a correct answer and ensuring it wins the plurality vote, showing that the candidate set can only grow while the set of reachable winners can only shrink. The authors also present a gold‑free locking certificate that identifies the earliest prefix where all remaining continuations produce the same fixed‑budget output, and demonstrate empirical gains in accuracy and call efficiency through input permutation and exact locking.
By Shaoang Li, Jian Li
arXiv:2609.38699v1 Announce Type: new
Abstract: A cumulative token cap can fall inside a mathematical derivation, forcing a test-time controller to choose between stopping at the cap (strict) and all...
By Guilin Zhang, Ziqi Tan, Wulan Guo, Kai Zhao, Hongyun Yang, Mei Luo, Qi Ning, Feng Yang
The paper critiques the use of pass@$k$ as a sole metric for evaluating reinforcement‑learning post‑training on language models, showing that it ignores diversity in model outputs. Experiments on Qwen2.5‑1.5B‑Instruct trained with Group Relative Policy Optimization (GRPO) and rejection‑sampling fine‑tuning (RFT) reveal that these methods shift token‑level, answer‑level, and unique‑answer diversity in opposite directions, yet pass@$k$ fails to capture these differences. Even when restricting to verifier‑correct completions, diversity gaps persist, and pass@$k$ does not consistently identify the better performing arm on hard problem subsets, indicating that the metric can mask true capability changes.
By Subham Rath, Raj Dandekar, Rajat Dandekar, Sreedath Panat
The paper investigates PRM‑Pruned Fragment Grafting (PPFG), an inference‑time technique that extracts high‑reward prefixes from a pruned chain‑of‑thought and grafts them into a sibling decoding process. Experiments on Qwen2.5‑7B‑Instruct with Math‑Shepherd across 500 MATH problems and multiple seeds show that PPFG performs statistically indistinguishable from a parallel‑CoT baseline, with only 14% of grafts targeting genuinely struggling chains. The study extends across three language models, six benchmarks, and multiple PRM configurations, concluding that PPFG’s inertness is not due to heuristic specifics and providing an equivalence‑testing framework for mechanism nulls.
By Khawaja Murad ul Hassan, Mehran Ebrahimi