TSRBench: A Comprehensive Multi-task Multi-modal Time Series Reasoning Benchmark for Generalist Models

Time series data is ubiquitous in real-world scenarios and crucial for critical applications ranging from energy management to traffic control. Consequently, the ability to reason over time series is a fundamental skill for generalist models to solve practical problems. However, this dimension is notably absent from existing benchmarks of generalist models. To bridge this gap, we introduce TSRBench, a comprehensive multi-modal benchmark designed to stress-test the full spectrum of time series reasoning capabilities. TSRBench features: i) a diverse set of 4125 problems from 14 domains, and is categorized into 4 major dimensions: Perception, Reasoning, Prediction, and Decision-Making. ii) 15 tasks from the 4 dimensions evaluating essential reasoning capabilities (e.g., numerical reasoning). Through extensive experiments, we evaluated over 30 leading proprietary and open-source LLMs, VLMs, and TSLLMs within TSRBench. Our findings reveal that: i) scaling laws hold for perception and reasoning but break down for prediction; ii) strong reasoning does not guarantee accurate context-aware forecasting, indicating a decoupling between semantic understanding and numerical prediction; and iii) despite the complementary nature of textual and visual represenations of time series as inputs, current multimodal models fail to effectively fuse them for reciprocal performance gains. TSRBench provides a standardized evaluation platform that not only highlights existing challenges but also offers valuable insights to advance generalist models. Our code and dataset are available at https://tsrbench.github.io/.

Token-Level LLM Collaboration via FusionRoute

Large language models (LLMs) exhibit strengths across diverse domains. However, achieving strong performance across these domains with a single general-purpose model typically requires scaling to sizes that are prohibitively expensive to train and deploy. On the other hand, while smaller domain-specialized models are much more efficient, they struggle to generalize beyond their training distributions. To address this dilemma, we propose FusionRoute, a robust and effective token-level multi-LLM collaboration framework in which a lightweight router simultaneously (i) selects the most suitable expert at each decoding step and (ii) contributes a complementary logit that refines or corrects the selected expert's next-token distribution via logit addition. Unlike existing token-level collaboration methods that rely solely on fixed expert outputs, we provide a theoretical analysis showing that pure expert-only routing is fundamentally limited: unless strong global coverage assumptions hold, it cannot in general realize the optimal decoding policy. By augmenting expert selection with a trainable complementary generator, FusionRoute expands the effective policy class and enables recovery of optimal value functions under mild conditions. Empirically, across both Llama-3 and Gemma-2 families and diverse benchmarks spanning mathematical reasoning, code generation, and instruction following, FusionRoute outperforms both sequence- and token-level collaboration, model merging, and direct fine-tuning, while remaining competitive with domain experts on their respective tasks.

PersonaLedger: Generating Realistic Financial Transactions with Persona Conditioned LLMs and Rule Grounded Feedback

Strict privacy regulations limit access to real transaction data, slowing open research in financial AI. Synthetic data can bridge this gap, but existing generators do not jointly achieve behavioral diversity and logical groundedness. Rule-driven simulators rely on hand-crafted workflows and shallow stochasticity, which miss the richness of human behavior. Learning-based generators such as GANs capture correlations yet often violate hard financial constraints and still require training on private data. We introduce PersonaLedger, a generation engine that uses a large language model conditioned on rich user personas to produce diverse transaction streams, coupled with an expert configurable programmatic engine that maintains correctness. The LLM and engine interact in a closed loop: after each event, the engine updates the user state, enforces financial rules, and returns a context aware "nextprompt" that guides the LLM toward feasible next actions. With this engine, we create a public dataset of 30 million transactions from 23,000 users and a benchmark suite with two tasks, illiquidity classification and identity theft segmentation. PersonaLedger offers a realistic, privacy preserving resource that supports rigorous evaluation of forecasting and anomaly detection models. PersonaLedger offers the community a rich, realistic, and privacy preserving resource -- complete with code, rules, and generation logs -- to accelerate innovation in financial AI and enable rigorous, reproducible evaluation.

RAGPart & RAGMask: Retrieval-Stage Defenses Against Corpus Poisoning in Retrieval-Augmented Generation

Retrieval-Augmented Generation (RAG) has emerged as a promising paradigm to enhance large language models (LLMs) with external knowledge, reducing hallucinations and compensating for outdated information. However, recent studies have exposed a critical vulnerability in RAG pipelines corpus poisoning where adversaries inject malicious documents into the retrieval corpus to manipulate model outputs. In this work, we propose two complementary retrieval-stage defenses: RAGPart and RAGMask. Our defenses operate directly on the retriever, making them computationally lightweight and requiring no modification to the generation model. RAGPart leverages the inherent training dynamics of dense retrievers, exploiting document partitioning to mitigate the effect of poisoned points. In contrast, RAGMask identifies suspicious tokens based on significant similarity shifts under targeted token masking. Across two benchmarks, four poisoning strategies, and four state-of-the-art retrievers, our defenses consistently reduce attack success rates while preserving utility under benign conditions. We further introduce an interpretable attack to stress-test our defenses. Our findings highlight the potential and limitations of retrieval-stage defenses, providing practical insights for robust RAG deployments.

MomaGraph: State-Aware Unified Scene Graphs with Vision-Language Model for Embodied Task Planning

Mobile manipulators in households must both navigate and manipulate. This requires a compact, semantically rich scene representation that captures where objects are, how they function, and which parts are actionable. Scene graphs are a natural choice, yet prior work often separates spatial and functional relations, treats scenes as static snapshots without object states or temporal updates, and overlooks information most relevant for accomplishing the current task. To address these limitations, we introduce MomaGraph, a unified scene representation for embodied agents that integrates spatial-functional relationships and part-level interactive elements. However, advancing such a representation requires both suitable data and rigorous evaluation, which have been largely missing. We thus contribute MomaGraph-Scenes, the first large-scale dataset of richly annotated, task-driven scene graphs in household environments, along with MomaGraph-Bench, a systematic evaluation suite spanning six reasoning capabilities from high-level planning to fine-grained scene understanding. Built upon this foundation, we further develop MomaGraph-R1, a 7B vision-language model trained with reinforcement learning on MomaGraph-Scenes. MomaGraph-R1 predicts task-oriented scene graphs and serves as a zero-shot task planner under a Graph-then-Plan framework. Extensive experiments demonstrate that our model achieves state-of-the-art results among open-source models, reaching 71.6% accuracy on the benchmark (+11.4% over the best baseline), while generalizing across public benchmarks and transferring effectively to real-robot experiments.

Lemon: A Unified and Scalable 3D Multimodal Model for Universal Spatial Understanding

Scaling large multimodal models (LMMs) to 3D understanding poses unique challenges: point cloud data is sparse and irregular, existing models rely on fragmented architectures with modality-specific encoders, and training pipelines often suffer from instability and poor scalability. We introduce Lemon, a unified transformer architecture that addresses these challenges by jointly processing 3D point cloud patches and language tokens as a single sequence. Unlike prior work that relies on modality-specific encoders and cross-modal alignment modules, this design enables early spatial-linguistic fusion, eliminates redundant encoders, improves parameter efficiency, and supports more effective model scaling. To handle the complexity of 3D data, we develop a structured patchification and tokenization scheme that preserves spatial context, and a three-stage training curriculum that progressively builds capabilities from object-level recognition to scene-level spatial reasoning. Lemon establishes new state-of-the-art performance across comprehensive 3D understanding and reasoning tasks, from object recognition and captioning to spatial reasoning in 3D scenes, while demonstrating robust scaling properties as model size and training data increase. Our work provides a unified foundation for advancing 3D spatial intelligence in real-world applications.

Hold Onto That Thought: Assessing KV Cache Compression On Reasoning

Large language models (LLMs) have demonstrated remarkable performance on long-context tasks, but are often bottlenecked by memory constraints. Namely, the KV cache, which is used to significantly speed up attention computations, grows linearly with context length. A suite of compression algorithms has been introduced to alleviate cache growth by evicting unimportant tokens. However, several popular strategies are targeted towards the prefill phase, i.e., processing long prompt context, and their performance is rarely assessed on reasoning tasks requiring long decoding. In particular, short but complex prompts, such as those in benchmarks like GSM8K and MATH500, often benefit from multi-step reasoning and self-reflection, resulting in thinking sequences thousands of tokens long. In this work, we benchmark the performance of several popular compression strategies on long-reasoning tasks. For the non-reasoning Llama-3.1-8B-Instruct, we determine that no singular strategy fits all, and that performance is heavily influenced by dataset type. However, we discover that H2O and our decoding-enabled variant of SnapKV are dominant strategies for reasoning models, indicating the utility of heavy-hitter tracking for reasoning traces. We also find that eviction strategies at low budgets can produce longer reasoning traces, revealing a tradeoff between cache size and inference costs.

Towards Mitigating Hallucinations in Large Vision-Language Models by Refining Textual Embeddings

In this work, we identify an inherent bias in prevailing LVLM architectures toward the language modality, largely resulting from the common practice of simply appending visual embeddings to the input text sequence. To address this, we propose a simple yet effective method that refines textual embeddings by integrating average-pooled visual features. Our approach demonstrably improves visual grounding and significantly reduces hallucinations on established benchmarks. While average pooling offers a straightforward, robust, and efficient means of incorporating visual information, we believe that more sophisticated fusion methods could further enhance visual grounding and cross-modal alignment. Given that the primary focus of this work is to highlight the modality imbalance and its impact on hallucinations -- and to show that refining textual embeddings with visual information mitigates this issue -- we leave exploration of advanced fusion strategies for future work.

MIRA: Towards Mitigating Reward Hacking in Inference-Time Alignment of T2I Diffusion Models

Diffusion models excel at generating images conditioned on text prompts, but the resulting images often do not satisfy user-specific criteria measured by scalar rewards such as Aesthetic Scores. This alignment typically requires fine-tuning, which is computationally demanding. Recently, inference-time alignment via noise optimization has emerged as an efficient alternative, modifying initial input noise to steer the diffusion denoising process towards generating high-reward images. However, this approach suffers from reward hacking, where the model produces images that score highly, yet deviate significantly from the original prompt. We show that noise-space regularization is insufficient and that preventing reward hacking requires an explicit image-space constraint. To this end, we propose MIRA (MItigating Reward hAcking), a training-free, inference-time alignment method. MIRA introduces an image-space, score-based KL surrogate that regularizes the sampling trajectory with a frozen backbone, constraining the output distribution so reward can increase without off-distribution drift (reward hacking). We derive a tractable approximation to KL using diffusion scores. Across SDv1.5 and SDXL, multiple rewards (Aesthetic, HPSv2, PickScore), and public datasets (e.g., Animal-Animal, HPDv2), MIRA achieves >60\% win rate vs. strong baselines while preserving prompt adherence; mechanism plots show reward gains with near-zero drift, whereas DNO drifts as compute increases. We further introduce MIRA-DPO, mapping preference optimization to inference time with a frozen backbone, extending MIRA to non-differentiable rewards without fine-tuning.

Zebra-CoT: A Dataset for Interleaved Vision Language Reasoning

Humans often use visual aids, for example diagrams or sketches, when solving complex problems. Training multimodal models to do the same, known as Visual Chain of Thought (Visual CoT), is challenging due to: (1) poor off-the-shelf visual CoT performance, which hinders reinforcement learning, and (2) the lack of high-quality visual CoT training data. We introduce $\textbf{Zebra-CoT}$, a diverse large-scale dataset with 182,384 samples, containing logically coherent interleaved text-image reasoning traces. We focus on four categories of tasks where sketching or visual reasoning is especially natural, spanning scientific questions such as geometry, physics, and algorithms; 2D visual reasoning tasks like visual search and jigsaw puzzles; 3D reasoning tasks including 3D multi-hop inference, embodied and robot planning; visual logic problems and strategic games like chess. Fine-tuning the Anole-7B model on the Zebra-CoT training corpus results in an improvement of +12% in our test-set accuracy and yields up to +13% performance gain on standard VLM benchmark evaluations. Fine-tuning Bagel-7B yields a model that generates high-quality interleaved visual reasoning chains, underscoring Zebra-CoT's effectiveness for developing multimodal reasoning abilities. We open-source our dataset and models to support development and evaluation of visual CoT.