arXiv:2604. 25907v3 Announce Type: replace Abstract: The standard recipe for training reasoning models---SFT on rationale annotations followed by RLVR---relies on rationale annotations being available and clean.
arXiv:2609.01274v1 Announce Type: new Abstract: Reinforcement learning with verifiable rewards (RLVR) improves language-model reasoning, but how these gains relate to inference-time decoding and sear...
The paper investigates why reinforcement learning with verifiable rewards (RLVR) reduces the diversity of solutions in reasoning tasks. By analyzing the Countdown task, the authors show that RLVR contracts the solution space mainly at the entrance—before the first arithmetic operation—causing a 67% drop in solution coverage. They demonstrate that providing an unselected entrance prefix or applying entrance‑targeted interventions can restore or even improve coverage without harming accuracy.
arXiv:2606. 21253v2 Announce Type: replace Abstract: Continual learning that is gradient-free, local, online, and append-only is attractive for edge and streaming deployment, but its value is usually argued informally.
arXiv:2509. 00924v2 Announce Type: replace Abstract: Transformers often transfer across related tasks with little fine-tuning; we study this question in noisy nonparametric regression for continuous maps $f,g:[0,1]^d\to\mathbb{R}^D$, representing the source and shifted tasks, using a single-query, single-head transformer $\mathcal{T}(\cdot)=V\operatorname{Softmax}(-\alpha(\cdot,K))$ with $\ell^{\infty}$-based alignment scores $\alpha$.
The paper challenges the common practice of selecting latent world models based on one‑step prediction loss, arguing that this metric poorly predicts planning performance. Using a frozen‑decoder protocol, the authors decompose one‑step error into encoder‑inherent and predictor‑added components and evaluate goal‑conditioned planning across three continuous‑control environments. Their experiments show that improvements in one‑step fidelity do not correlate with planning success, with correlation coefficients near zero and inconsistent exchange rates across tasks.
arXiv:2607. 24741v2 Announce Type: replace-cross Abstract: Positive two-marginal entropic optimal transport is solved by a nonlinear, positive, order-preserving, homogeneous Sinkhorn map.
arXiv:2606. 09059v1 Announce Type: cross Abstract: Two-stage post-training -- a Stage-1 warm-start (supervised fine-tuning, SFT, or on-policy distillation, OPD) followed by Stage-2 reinforcement learning (RL) -- is increasingly used for vision-language models (VLMs).
arXiv:2609.37066v1 Announce Type: cross Abstract: Post-training is central to mathematical reasoning in modern large language models (LLMs), but endpoint pass@1 alone underidentifies what has changed...
The paper introduces belief‑shift branching, a method for placing forks in tree‑structured reinforcement learning rollouts by identifying points where a model’s answer belief changes most. Unlike traditional structural or entropy‑based approaches, belief‑shift uses a probe, logit‑lens depth profile, or learned activation direction to locate pivots in the value curve, incurring minimal computational overhead. Experiments across multiple models and benchmarks show that belief‑shift forking consistently outperforms baseline methods, yielding significant gains in mathematics and code tasks.
arXiv:2606. 01682v1 Announce Type: cross Abstract: Selecting the best response from multiple small-model samples using a stronger scorer is a simple inference-time strategy, but fails when the small model has already committed to incorrect reasoning paths.
The paper introduces OnQ, a low‑bit recovery pipeline for quantized large language models that uses on‑policy distillation and candidate‑wise Dynamic KL to address trajectory divergence and distributional misalignment. OnQ starts with block‑wise calibration, then lets the quantized student generate responses and receive full‑precision teacher supervision on its own prefixes, dynamically weighting forward and reverse KL losses per candidate. Experiments on Qwen3 models show that OnQ outperforms baselines on nine benchmarks, especially long‑horizon reasoning, and produces shorter, less truncated responses.
arXiv:2606. 27771v1 Announce Type: new Abstract: Reinforcement learning (RL) post-training improves the reward alignment of flow-based generators, but often degrades perceptual quality in ways that are not captured by the reward proxy.