Value-Aware Stochastic KV Cache Eviction for Reasoning Models

Reasoning models improve accuracy through extended chains of thought, but their long outputs create a memory and compute bottleneck. KV cache eviction methods reduce this cost by evicting unimportant key-value pairs from the cache, yet they often yield worse accuracy than selection-based sparse attention alternatives, which keep the full KV cache. We identify key factors crucial to KV cache eviction accuracy. First, a small fraction of value states have abnormally large magnitudes, and evicting them causes catastrophic failure where models enter repetitive reasoning loops. Second, introducing stochasticity during eviction improves accuracy by increasing cache diversity. Based on these findings, we propose Value-aware Stochastic KV Cache Eviction (VaSE), a training-free recipe that protects large-magnitude value states and promotes diverse eviction decisions. Across six reasoning tasks, Qwen3 models using VaSE with 4x KV cache compression yield higher average accuracies than SOTA selection method at the same sparsity, while outperforming the strongest eviction method by more than 4%. Overall, VaSE bridges the gap between efficiency and accuracy, supporting FlashAttention2 and enabling a static memory footprint for reasoning models.

Dr. DocBench: A Comprehensive Benchmark for Expert-Level and Difficult Document Parsing

Document parsing and recognition are fundamental capabilities for vision-language models (VLMs) and document processing systems. However, existing Optical Character Recognition (OCR) and document parsing benchmarks are increasingly limited in coverage and difficulty: many focus on common document genres or uniformly sampled pages where modern parsers already perform strongly, while offering limited annotation for expert-domain structures such as chemical formula, music notation, complex tables, and cross-page layouts. We introduce Dr. DocBench, a difficulty-aware benchmark for expert-level document parsing. Built from a large-scale multilingual book corpus, Dr. DocBench spans 52 BISAC subject domains and selects challenging documents through parser-failure-based sampling, targeting cases where multiple state-of-the-art systems struggle. It contains 4,514 annotated pages from long documents averaging around 100 pages, with 65k high-quality page- and block-level annotations for layout, reading order, hierarchical relations, and domain-specific visual contents. Evaluations of pipeline-based parsers and general-purpose VLMs show that strong performance on existing benchmarks does not transfer to our expert-level document parsing. Our analysis reveals substantial failures across subjects, content types, and structural attributes, highlighting Dr. DocBench as a comprehensive testbed for diagnosing and advancing document intelligence.

Colosseum V2: Benchmarking Generalization for Vision Language Action Models

Vision-Language-Action (VLA) models demonstrate promising generalization in robotic manipulation, driven by advances in large-scale vision and language pre-training. This progress can be misleading. Despite the zero-shot perception and language capabilities of VLAs, their overall task performance often degrades under distribution shifts, revealing gaps in how these systems translate high-level understanding into robust behavior. To systematically study this gap, we introduce Colosseum V2, a large-scale simulation benchmark for evaluating VLA generalization in robot learning across diverse conditions. The benchmark comprises 28 tasks spanning 13 task categories and two robot morphologies, covering a wide range of manipulation primitives and long-horizon behaviors. Built on the ManiSkill simulator, Colosseum V2 enables fast, GPU-parallelized evaluation and supports both in-domain and out-of-domain testing at scale. We evaluate state-of-the-art methods, including Action Chunking Transformers (ACT) and Pi0.5, and reveal limitations in both base performance and generalization. We demonstrate strong correlations between simulation and real-world metrics that support the ecological validity of the benchmark. By standardizing tasks, metrics, and evaluation protocols within a unified benchmark, Colosseum V2 enables reproducible and fair comparisons, reduced evaluation overhead, and accelerated progress toward general-purpose robot policies.

PDDL-Mind: Large Language Models are Capable on Belief Reasoning with Reliable State Tracking

Large language models (LLMs) perform substantially below human level on existing theory-of-mind (ToM) benchmarks, even when augmented with chain-of-thought prompting or probabilistic belief updates. We argue that these failures primarily arise from unreliable implicit state tracking rather than limitations in high-level reasoning. We introduce PDDL-Mind, a neuro-symbolic framework that decouples environment state evolution from belief inference. By translating narrative descriptions into explicit states and actions expressed in Planning Domain Definition Language (PDDL), and by verifying action-induced state transitions against a predefined domain, PDDL-Mind provides LLMs with a logically consistent and explicit representation of world states for ToM tasks. Experiments on MMToM-QA, MuMA and FanToM show that PDDL-Mind achieves over 5% absolute accuracy gain over the best existing state-of-the-art method on ToM benchmark questions.

Expert Personas Improve LLM Alignment but Damage Accuracy: Bootstrapping Intent-Based Persona Routing with PRISM

Persona prompting can steer LLM generation towards a domain-specific tone and pattern. This behavior enables use cases in multi-agent systems where diverse interactions are crucial and human-centered tasks require high-level human alignment. Prior works provide mixed opinions on their utility: some report performance gains when using expert personas for certain domains and their contribution to data diversity in synthetic data creation, while others find near-zero or negative impact on general utility. To fully leverage the benefits of the LLM persona and avoid its harmfulness, a more comprehensive investigation of the mechanism is crucial. In this work, we study how model optimization, task type, prompt length, and placement can impact expert persona effectiveness across instruction-tuned and reasoning LLMs, and provide insight into conditions under which expert personas fail and succeed. Based on our findings, we developed a pipeline to fully leverage the benefits of an expert persona, named PRISM (Persona Routing via Intent-based Self-Modeling), which self-distills an intent-conditioned expert persona into a gated LoRA adapter through a bootstrapping process that requires no external data, models, or knowledge. PRISM enhances human preference and safety alignment on generative tasks while maintaining accuracy on discriminative tasks across all models, with minimal memory and computing overhead.

Auditing Multi-Agent LLM Reasoning Trees Outperforms Majority Vote and LLM-as-Judge

Multi-agent systems (MAS) can substantially extend the reasoning capacity of large language models (LLMs), yet most frameworks still aggregate agent outputs with majority voting. This heuristic discards the evidential structure of reasoning traces and is brittle under the confabulation consensus, where agents share correlated biases and converge on the same incorrect rationale. We introduce AgentAuditor, which replaces voting with a path search over a Reasoning Tree that explicitly represents agreements and divergences among agent traces. AgentAuditor resolves conflicts by comparing reasoning branches at critical divergence points, turning global adjudication into efficient, localized verification. We further propose Anti-Consensus Preference Optimization (ACPO), which trains the adjudicator on majority-failure cases and rewards evidence-based minority selections over popular errors. AgentAuditor is agnostic to MAS setting, and we find across 5 popular settings that it yields up to 5% absolute accuracy improvement over a majority vote, and up to 3% over using LLM-as-Judge.

Words that make SENSE: Sensorimotor Norms in Learned Lexical Token Representations

While word embeddings derive meaning from co-occurrence patterns, human language understanding is grounded in sensory and motor experience. We present $\text{SENSE}$ $(\textbf{S}\text{ensorimotor }$ $\textbf{E}\text{mbedding }$ $\textbf{N}\text{orm }$ $\textbf{S}\text{coring }$ $\textbf{E}\text{ngine})$, a learned projection model that predicts Lancaster sensorimotor norms from word lexical embeddings. We also conducted a behavioral study where 281 participants selected which among candidate nonce words evoked specific sensorimotor associations, finding statistically significant correlations between human selection rates and $\text{SENSE}$ ratings across 6 of the 11 modalities. Sublexical analysis of these nonce words selection rates revealed systematic phonosthemic patterns for the interoceptive norm, suggesting a path towards computationally proposing candidate phonosthemes from text data.

Maestro: Learning to Collaborate via Conditional Listwise Policy Optimization for Multi-Agent LLMs

Multi-agent systems (MAS) built on Large Language Models (LLMs) are being used to approach complex problems and can surpass single model inference. However, their success hinges on navigating a fundamental cognitive tension: the need to balance broad, divergent exploration of the solution space with a principled, convergent synthesis to the optimal solution. Existing paradigms often struggle to manage this duality, leading to premature consensus, error propagation, and a critical credit assignment problem that fails to distinguish between genuine reasoning and superficially plausible arguments. To resolve this core challenge, we propose the Multi-Agent Exploration-Synthesis framework Through Role Orchestration (Maestro), a principled paradigm for collaboration that structurally decouples these cognitive modes. Maestro uses a collective of parallel Execution Agents for diverse exploration and a specialized Central Agent for convergent, evaluative synthesis. To operationalize this critical synthesis phase, we introduce Conditional Listwise Policy Optimization (CLPO), a reinforcement learning objective that disentangles signals for strategic decisions and tactical rationales. By combining decision-focused policy gradients with a list-wise ranking loss over justifications, CLPO achieves clean credit assignment and stronger comparative supervision. Experiments on mathematical reasoning and general problem-solving benchmarks demonstrate that Maestro, coupled with CLPO, consistently outperforms existing state-of-the-art multi-agent approaches, delivering absolute accuracy gains of 6% on average and up to 10% at best.

VeriMoA: A Mixture-of-Agents Framework for Spec-to-HDL Generation

Automation of Register Transfer Level (RTL) design can help developers meet increasing computational demands. Large Language Models (LLMs) show promise for Hardware Description Language (HDL) generation, but face challenges due to limited parametric knowledge and domain-specific constraints. While prompt engineering and fine-tuning have limitations in knowledge coverage and training costs, multi-agent architectures offer a training-free paradigm to enhance reasoning through collaborative generation. However, current multi-agent approaches suffer from two critical deficiencies: susceptibility to noise propagation and constrained reasoning space exploration. We propose VeriMoA, a training-free mixture-of-agents (MoA) framework with two synergistic innovations. First, a quality-guided caching mechanism to maintain all intermediate HDL outputs and enables quality-based ranking and selection across the entire generation process, encouraging knowledge accumulation over layers of reasoning. Second, a multi-path generation strategy that leverages C++ and Python as intermediate representations, decomposing specification-to-HDL translation into two-stage processes that exploit LLM fluency in high-resource languages while promoting solution diversity. Comprehensive experiments on VerilogEval 2.0 and RTLLM 2.0 benchmarks demonstrate that VeriMoA achieves 15--30% improvements in Pass@1 across diverse LLM backbones, especially enabling smaller models to match larger models and fine-tuned alternatives without requiring costly training.

Phonological Representation Learning for Isolated Signs Improves Out-of-Vocabulary Generalization

Sign language datasets are often not representative in terms of vocabulary, underscoring the need for models that generalize to unseen signs. Vector quantization is a promising approach for learning discrete, token-like representations, but it has not been evaluated whether the learned units capture spurious correlations that hinder out-of-vocabulary performance. This work investigates two phonological inductive biases: Parameter Disentanglement, an architectural bias, and Phonological Semi-Supervision, a regularization technique, to improve isolated sign recognition of known signs and reconstruction quality of unseen signs with a vector-quantized autoencoder. The primary finding is that the learned representations from the proposed model are more effective for one-shot reconstruction of unseen signs and more discriminative for sign identification compared to a controlled baseline. This work provides a quantitative analysis of how explicit, linguistically-motivated biases can improve the generalization of learned representations of sign language.