Chunky Post-Training: Data Driven Failures of Generalization

LLM post-training involves many diverse datasets, each targeting a specific behavior. But these datasets encode incidental patterns alongside intended ones: correlations between formatting and content, narrow phrasings across diverse problems, and implicit associations arising from the discrete data curation process. These patterns are often invisible to developers yet salient to models, producing behaviors that surprise their creators, such as rejecting true facts presented in a particular question format. We call this chunky post-training: the model learns spurious correlations as a result of distinct chunks of post-training data. We introduce SURF, a black-box pipeline which surfaces these unintended behaviors at run time, and TURF, a tool that traces these failures back to specific post-training data. Applying these tools to frontier models (Claude 4.5, GPT-5.1, Grok 4.1, Gemini 3) and open models (Tülu 3), we show that chunky post-training produces miscalibrated behaviors, which often result from imbalanced or underspecified chunks of post-training data.

OAT: Ordered Action Tokenization

Autoregressive policies offer a compelling foundation for scalable robot learning by enabling discrete abstraction, token-level reasoning, and flexible inference. However, applying autoregressive modeling to continuous robot actions requires an effective action tokenization scheme. Existing approaches either rely on analytical discretization methods that produce prohibitively long token sequences, or learned latent tokenizers that lack structure, limiting their compatibility with next-token prediction. In this work, we identify three desiderata for action tokenization - high compression, total decodability, and a left-to-right causally ordered token space - and introduce Ordered Action Tokenization (OAT), a learned action tokenizer that satisfies all three. OAT discretizes action chunks into an ordered sequence of tokens using transformer with registers, finite scalar quantization, and ordering-inducing training mechanisms. The resulting token space aligns naturally with autoregressive generation and enables prefix-based detokenization, yielding an anytime trade-off between inference cost and action fidelity. Across more than 20 tasks spanning four simulation benchmarks and real-world settings, autoregressive policies equipped with OAT consistently outperform prior tokenization schemes and diffusion-based baselines, while offering significantly greater flexibility at inference time.

DeepRead: Document Structure-Aware Reasoning to Enhance Agentic Search

With the rapid progress of tool-using and agentic large language models (LLMs), Retrieval-Augmented Generation (RAG) is evolving from one-shot, passive retrieval into multi-turn, decision-driven evidence acquisition. Despite strong results in open-domain settings, existing agentic search frameworks commonly treat long documents as flat collections of chunks, underutilizing document-native priors such as hierarchical organization and sequential discourse structure. We introduce DeepRead, a structure-aware, multi-turn document reasoning agent that explicitly operationalizes these priors for long-document question answering. DeepRead leverages LLM-based OCR model to convert PDFs into structured Markdown that preserves headings and paragraph boundaries. It then indexes documents at the paragraph level and assigns each paragraph a coordinate-style metadata key encoding its section identity and in-section order. Building on this representation, DeepRead equips the LLM with two complementary tools: a Retrieve tool that localizes relevant paragraphs while exposing their structural coordinates (with lightweight scanning context), and a ReadSection tool that enables contiguous, order-preserving reading within a specified section and paragraph range. Our experiments demonstrate that DeepRead achieves significant improvements over Search-o1-style agentic search in document question answering. The synergistic effect between retrieval and reading tools is also validated. Our fine-grained behavioral analysis reveals a reading and reasoning paradigm resembling human-like ``locate then read'' behavior.

Viewpoint Matters: Dynamically Optimizing Viewpoints with Masked Autoencoder for Visual Manipulation

Robotic manipulation continues to be a challenge, and imitation learning (IL) enables robots to learn tasks from expert demonstrations. Current IL methods typically rely on fixed camera setups, where cameras are manually positioned in static locations, imposing significant limitations on adaptability and coverage. Inspired by human active perception, where humans dynamically adjust their viewpoint to capture the most relevant and least noisy information, we propose MAE-Select, a novel framework for active viewpoint selection in single-camera robotic systems. MAE-Select fully leverages pre-trained multi-view masked autoencoder representations and dynamically selects the next most informative viewpoint at each time chunk without requiring labeled viewpoints. Extensive experiments demonstrate that MAE-Select improves the capabilities of single-camera systems and, in some cases, even surpasses multi-camera setups. The project will be available at https://mae-select.github.io.

Light Forcing: Accelerating Autoregressive Video Diffusion via Sparse Attention

Advanced autoregressive (AR) video generation models have improved visual fidelity and interactivity, but the quadratic complexity of attention remains a primary bottleneck for efficient deployment. While existing sparse attention solutions have shown promise on bidirectional models, we identify that applying these solutions to AR models leads to considerable performance degradation for two reasons: isolated consideration of chunk generation and insufficient utilization of past informative context. Motivated by these observations, we propose \textsc{Light Forcing}, the \textit{first} sparse attention solution tailored for AR video generation models. It incorporates a \textit{Chunk-Aware Growth} mechanism to quantitatively estimate the contribution of each chunk, which determines their sparsity allocation. This progressive sparsity increase strategy enables the current chunk to inherit prior knowledge in earlier chunks during generation. Additionally, we introduce a \textit{Hierarchical Sparse Attention} to capture informative historical and local context in a coarse-to-fine manner. Such two-level mask selection strategy (\ie, frame and block level) can adaptively handle diverse attention patterns. Extensive experiments demonstrate that our method outperforms existing sparse attention in quality (\eg, 84.5 on VBench) and efficiency (\eg, $1.2{\sim}1.3\times$ end-to-end speedup). Combined with FP8 quantization and LightVAE, \textsc{Light Forcing} further achieves a $2.3\times$ speedup and 19.7\,FPS on an RTX~5090 GPU. Code will be released at \href{https://github.com/chengtao-lv/LightForcing}{https://github.com/chengtao-lv/LightForcing}.

RAGTurk: Best Practices for Retrieval Augmented Generation in Turkish

Retrieval-Augmented Generation (RAG) enhances LLM factuality, yet design guidance remains English-centric, limiting insights for morphologically rich languages like Turkish. We address this by constructing a comprehensive Turkish RAG dataset derived from Turkish Wikipedia and CulturaX, comprising question-answer pairs and relevant passage chunks. We benchmark seven stages of the RAG pipeline, from query transformation and reranking to answer refinement, without task-specific fine-tuning. Our results show that complex methods like HyDE maximize accuracy (85%) that is considerably higher than the baseline (78.70%). Also a Pareto-optimal configuration using Cross-encoder Reranking and Context Augmentation achieves comparable performance (84.60%) with much lower cost. We further demonstrate that over-stacking generative modules can degrade performance by distorting morphological cues, whereas simple query clarification with robust reranking offers an effective solution.

FASA: Frequency-aware Sparse Attention

The deployment of Large Language Models (LLMs) faces a critical bottleneck when handling lengthy inputs: the prohibitive memory footprint of the Key Value (KV) cache. To address this bottleneck, the token pruning paradigm leverages attention sparsity to selectively retain a small, critical subset of tokens. However, existing approaches fall short, with static methods risking irreversible information loss and dynamic strategies employing heuristics that insufficiently capture the query-dependent nature of token importance. We propose FASA, a novel framework that achieves query-aware token eviction by dynamically predicting token importance. FASA stems from a novel insight into RoPE: the discovery of functional sparsity at the frequency-chunk (FC) level. Our key finding is that a small, identifiable subset of "dominant" FCs consistently exhibits high contextual agreement with the full attention head. This provides a robust and computationally free proxy for identifying salient tokens. %making them a powerful and efficient proxy for token importance. Building on this insight, FASA first identifies a critical set of tokens using dominant FCs, and then performs focused attention computation solely on this pruned subset. % Since accessing only a small fraction of the KV cache, FASA drastically lowers memory bandwidth requirements and computational cost. Across a spectrum of long-context tasks, from sequence modeling to complex CoT reasoning, FASA consistently outperforms all token-eviction baselines and achieves near-oracle accuracy, demonstrating remarkable robustness even under constraint budgets. Notably, on LongBench-V1, FASA reaches nearly 100\% of full-KV performance when only keeping 256 tokens, and achieves 2.56$\times$ speedup using just 18.9\% of the cache on AIME24.

A Low-Cost Vision-Based Tactile Gripper with Pretraining Learning for Contact-Rich Manipulation

Robotic manipulation in contact-rich environments remains challenging, particularly when relying on conventional tactile sensors that suffer from limited sensing range, reliability, and cost-effectiveness. In this work, we present LVTG, a low-cost visuo-tactile gripper designed for stable, robust, and efficient physical interaction. Unlike existing visuo-tactile sensors, LVTG enables more effective and stable grasping of larger and heavier everyday objects, thanks to its enhanced tactile sensing area and greater opening angle. Its surface skin is made of highly wear-resistant material, significantly improving durability and extending operational lifespan. The integration of vision and tactile feedback allows LVTG to provide rich, high-fidelity sensory data, facilitating reliable perception during complex manipulation tasks. Furthermore, LVTG features a modular design that supports rapid maintenance and replacement. To effectively fuse vision and touch, We adopt a CLIP-inspired contrastive learning objective to align tactile embeddings with their corresponding visual observations, enabling a shared cross-modal representation space for visuo-tactile perception. This alignment improves the performance of an Action Chunking Transformer (ACT) policy in contact-rich manipulation, leading to more efficient data collection and more effective policy learning. Compared to the original ACT method, the proposed LVTG with pretraining achieves significantly higher success rates in manipulation tasks.

Reparameterization Flow Policy Optimization

Reparameterization Policy Gradient (RPG) has emerged as a powerful paradigm for model-based reinforcement learning, enabling high sample efficiency by backpropagating gradients through differentiable dynamics. However, prior RPG approaches have been predominantly restricted to Gaussian policies, limiting their performance and failing to leverage recent advances in generative models. In this work, we identify that flow policies, which generate actions via differentiable ODE integration, naturally align with the RPG framework, a connection not established in prior work. However, naively exploiting this synergy proves ineffective, often suffering from training instability and a lack of exploration. We propose Reparameterization Flow Policy Optimization (RFO). RFO computes policy gradients by backpropagating jointly through the flow generation process and system dynamics, unlocking high sample efficiency without requiring intractable log-likelihood calculations. RFO includes two tailored regularization terms for stability and exploration. We also propose a variant of RFO with action chunking. Extensive experiments on diverse locomotion and manipulation tasks, involving both rigid and soft bodies with state or visual inputs, demonstrate the effectiveness of RFO. Notably, on a challenging locomotion task controlling a soft-body quadruped, RFO achieves almost $2\times$ the reward of the state-of-the-art baseline.

A-RAG: Scaling Agentic Retrieval-Augmented Generation via Hierarchical Retrieval Interfaces

Frontier language models have demonstrated strong reasoning and long-horizon tool-use capabilities. However, existing RAG systems fail to leverage these capabilities. They still rely on two paradigms: (1) designing an algorithm that retrieves passages in a single shot and concatenates them into the model's input, or (2) predefining a workflow and prompting the model to execute it step-by-step. Neither paradigm allows the model to participate in retrieval decisions, preventing efficient scaling with model improvements. In this paper, we introduce A-RAG, an Agentic RAG framework that exposes hierarchical retrieval interfaces directly to the model. A-RAG provides three retrieval tools: keyword search, semantic search, and chunk read, enabling the agent to adaptively search and retrieve information across multiple granularities. Experiments on multiple open-domain QA benchmarks show that A-RAG consistently outperforms existing approaches with comparable or lower retrieved tokens, demonstrating that A-RAG effectively leverages model capabilities and dynamically adapts to different RAG tasks. We further systematically study how A-RAG scales with model size and test-time compute. We will release our code and evaluation suite to facilitate future research. Code and evaluation suite are available at https://github.com/Ayanami0730/arag.