CULTURE-CMT is a Chinese‑English benchmark that evaluates contextualized translation of social‑media comments by providing the original post and conversational context. It contains 572 manually verified examples grouped into Slang, Context, and Both, allowing diagnostic analysis of translation difficulty. The benchmark also introduces a retrieval‑augmented cultural‑effectiveness Judger that combines a calibrated scoring model with dynamically retrieved cultural knowledge.
By Linjuan Wu, Dingming Li, Yiwen Qiu, Xinze Lyu, Ye Guo, Daoxin Zhang, Zhe Xu, Weiming Lu
The paper introduces GRIPS, a three‑stage method that extracts structured pre‑SQL grounding information—such as table, column, value, and relation links—to guide text‑to‑SQL generation. Experiments on TempTabQA‑C and BIRD show that GRIPS‑Adaptive outperforms baseline SQL Static in 18 of 20 settings, boosting average accuracy by 11.8 percentage points and cutting model‑call token usage by 47.8%. The study demonstrates that explicit grounding improves accuracy and efficiency, while adding more pre‑SQL data does not always help.
By Ekata Mitra, Yufei Tao, Kamalamalini Nagasundaram, Ameeta Agrawal
MulTaBench is a new benchmark comprising 40 datasets split evenly between image‑tabular and text‑tabular tasks, designed to evaluate multimodal tabular learning where modalities provide complementary predictive signals. The study shows that tuning embeddings to the specific task—target‑aware representation learning—improves performance across both modalities, various tabular learners, encoder sizes, and embedding dimensions. MulTaBench is the largest image‑tabular benchmark to date, covering high‑impact domains such as healthcare and e‑commerce, and aims to spur research into joint modeling and multimodal tabular foundation models.
By Alan Arazi, Eilam Shapira, Shoham Grunblat, Mor Ventura, Elad Hoffer, Gioia Blayer, David Holzm\"uller, Lennart Purucker, Ga\"el Varoquaux, Frank Hutter, Roi Reichart
AtomMem is a long‑term memory system for large language models that stores high‑value atomic facts extracted from long‑form interactions. It organizes these facts into hierarchical event structures and temporal profiles, enabling coherent episodic context and dynamic tracking of user attributes. During retrieval, an associative memory graph connects fragmented memories, and experiments on the LoCoMo benchmark show state‑of‑the‑art performance across reasoning tasks.
By Yanyu Yao, Shangze Li, Zhi Zheng, Hui Zheng, Qi Liu, Tong Xu, Enhong Chen
FinForesight is a diagnostic backtesting benchmark designed to evaluate deep research systems for financial forecasting. It tests four forecasting tasks—valuation, return ranking, event direction, and future financial fundamentals—across Chinese, U.S., and Hong Kong markets, while also including two diagnostic tasks for counterfactual reasoning and financial calculation. The framework limits evidence to what was available at the forecast date and compares pretraining-era, post-cutoff, and live evaluations to assess how well controlled backtests predict future performance.
By Fengyuan Liu, Mukai Li, Jieshun You, Yilin Zheng, Chenjia Min, Yizhou Zhang, Harper Chen, Qi Liu
The paper introduces OLGA, an online label‑free gradient matching method for continual language model training. OLGA synthesizes examples whose gradients match the model’s own next‑token predictions, enabling efficient updates on small unlabeled data sets. It achieves significant accuracy gains on five text classification benchmarks, outperforms existing selection‑based baselines in both offline and online settings, and scales to large data by combining selection and distillation in a hybrid pipeline.
By Sajjad Pakdamansavoji, Mohammad Reza Rezaei, Ilija Subasic, Pedro Garzon, Radu Paradovschi, Zhao Chen
MoBiQuant introduces a Mixture-of-Bits quantization framework that allows large language models to be inferred at varying precisions during runtime. It addresses the poor generalization of existing post‑training quantization methods by mitigating precision‑dependent outlier migration through a recursive residual quantization scheme and a token‑aware router that selects the optimal precision per token. Experiments demonstrate that MoBiQuant matches or exceeds single‑precision PTQ performance while providing significant memory savings and up to 1.34× throughput gains over current any‑precision approaches.
By Dongwei Wang, Jinhee Kim, Seokho Han, Denis Gudovskiy, Yohei Nakata, Tomoyuki Okuno, KhayTze Peong, Kang Eun Jeon, Jong Hwan Ko, Yiran Chen, Huanrui Yang
The paper introduces GatedV, an attention variant that applies the SiLU activation only to the value matrix before aggregation, adding no extra trainable parameters. Experiments on a decoder‑only Transformer show that, with matched parameter counts and training budgets, GatedV consistently lowers held‑out perplexity—by 2.0% at 2.33M parameters and 2.4% at 26.16M parameters—across multiple seeds. Ablations indicate that the performance gains are not fully captured by simple affine rescaling of values, and a benchmark on NVIDIA DGX Spark reports a modest 0.7% increase in full‑context forward latency for the larger model.
By Li-Ming Yu, Hongyu Zhou
EffComp is a new framework for prompt compression in retrieval‑augmented question answering. It uses BERT‑style document rerankers and sentence selectors to perform fast, extractive compression at the sentence level, thereby avoiding hallucinations. The hybrid reinforcement‑plus‑supervised training optimizes compression ratio while maintaining LLM accuracy, and experiments show EffComp outperforms existing methods with low latency.
By Romtham Tungsununtham, Piyawat Lertvittayakumjorn, Atiwong Suchato, Peerapon Vateekul
AutoSpecNER is a new expert‑annotated dataset for fine‑grained named entity recognition in vehicle listings, comprising 659 advertisements and over 10,000 entities across 15 categories such as MODEL, ENGINE_SPEC, and BATTERY_CAPACITY. The annotation quality was validated with an inter‑annotator agreement score of 91.5%. Benchmarking shows that DeBERTa achieves a 90% micro‑F1, far surpassing a rule‑based baseline (43%) and the best large language model (77.8%).
By Jordan Lee, Filippos Ventirozos, Abdirahman Abdullahm, Ioanna Nteka, Peter Appleby, Matthew Shardlow
The paper introduces a continual learning approach for large language models called Multi‑Teacher On‑Policy Distillation (MOPD) that avoids storing intermediate expert snapshots by replaying prompts from earlier domains using the previous student as a teacher. Two replay strategies are examined: a shared replay budget across domains and a fixed budget per domain, with replay steps that either distill token distributions alone or also align hidden states, attention scores, and value relations. Experiments on a five‑domain benchmark (logic, math, code, safety, instruction following) demonstrate that replay alone closes most of the performance gap between sequential distillation and joint MOPD, achieving comparable results to recovery from all stored snapshots when using a fixed replay share per domain.
By Yiqiao Qiu
LF-VideoQA is a new dataset for long‑form, open‑ended question answering over narrative videos, built around theory‑of‑mind questions that require reasoning across scenes averaging 23 minutes. It contains over 6,300 question‑answer‑rubric triples generated automatically from annotated screenplays and validated by humans. The dataset challenges current vision‑language models, showing that even oracle models fall short and that visual evidence is essential for answering these questions.
By Shaden Shaar, Bradon Thymes, Majd Aldaye, Claire Cardie, Bharath Hariharan
arXiv:2611.02581v1 Announce Type: cross
Abstract: Automated liver ultrasound report generation improves the diagnosis of liver diseases by producing clinically structured descriptions of anatomical a...
By Haris Ghafoor, Minjee Seo, Seonaeng cho, Keungmo Yang, Jaejun Lee, Kyungho Yoon
arXiv:2605.00257v2 Announce Type: replace
Abstract: The inception of Large Language Models (LLMs) has catalyzed AI adoption in the finance sector, yet their reliability in complex, jurisdiction-speci...
By Jatin Gupta, Akhil Sharma, Saransh Singhania, Ali Imam Abidi
EntropyCache is a training‑free KV caching technique for diffusion‑based large language models that uses the maximum entropy of newly decoded token distributions to decide when to recompute cached states. By leveraging two empirical observations—entropy correlating with cache drift and feature volatility persisting across multiple steps—EntropyCache makes a skip‑or‑recompute decision with only O(V) computation per step, independent of context length or model size. Experiments on LLaDA‑8B‑Instruct and Dream‑7B‑Instruct demonstrate speedups of 15.2×–26.4× on standard benchmarks and 22.4×–24.1× on chain‑of‑thought benchmarks while maintaining competitive accuracy and keeping decision overhead below 0.5% of inference time.
By Minsoo Cheong, Donghyun Son, Woosang Lim, Sungjoo Yoo
The paper introduces Adaptive-Policy Distillation (APD), a method that improves knowledge injection from a context-conditioned teacher model into a student model by dynamically adjusting the distillation policy and loss function at each token. APD addresses two structural failures of On‑Policy Distillation (OPD): the reverse‑KL supervision’s weak influence on unlikely document tokens and the degradation of coverage for longer target sequences. Experiments on QASPER and BioASQ show that APD outperforms OPD by up to 54.6%, achieving state‑of‑the‑art F1 and MRR scores.
By Hojun Cho, Kang Eun Jeon, Jaegul Choo
TrackMem is a per-speaker graph memory designed to handle evolving beliefs in long-horizon financial dialogue, tracking figures, perspectives, and speaker actions over time. It addresses limitations of existing memory systems that rely on similarity or recency, which often return outdated or irrelevant information in financial contexts. Evaluated on the FinLoCoMo benchmark of 15 investment conversations, TrackMem outperforms baselines, achieving 73.9% macro accuracy on the failure-mode split and significant gains on questions about revised figures.
By Austin M. Y. Cheung, Harsh Vardhan Gupta, Go Nam Lui, Lok Kan Chan, Ginny Y. Wong, Simon See
The paper introduces TTT‑Consolidate, a framework that leverages multi‑agent reasoning trajectories as adaptation data to improve subsequent reasoning. It consolidates these trajectories into parametric representations via test‑time training and then generates a new solution conditioned on the consolidated information, with meta‑learning used to optimize the consolidation process. Using the newly created OpenParallelThinking corpus of 12,810 math problems and trajectories from six LLMs, TTT‑Consolidate outperforms existing aggregation baselines by up to 20 percentage points on challenging math and STEM benchmarks, especially as model size grows.
By Haotian Wu, Zeming Chen, Hao Zhao, Antoine Bosselut
The study introduces a panel of three large language models that score each sentence of political manifestos on environmental protection and climate change, producing averaged manifesto-level ratings. Applied to 41 2024 European Parliament election manifestos, the model’s rankings align closely with the Chapel Hill Expert Survey but tend to rate parties as greener than experts, suggesting a gap between promises and practice. The approach offers a low‑cost, multilingual, and reproducible alternative to traditional expert surveys and hand‑coding, with inter‑model disagreement serving as a built‑in reliability check.
By Franco Corti, Sajid Siraj
The paper introduces a memory‑efficient audio synthesis architecture used in Siri Expressive Voices, converting semantic audio tokens into high‑fidelity audio via a residual vector quantization (RVQ) pipeline that decouples temporal and depth processing. A single Diffusion Transformer‑style depth decoder generates all RVQ levels autoregressively, while causal sliding‑window attention with fixed‑window key‑value caching keeps memory usage constant regardless of sequence length. Deployed on Apple’s AMX, the system achieves real‑time performance (≈10 ms per step) with only 21 MB peak runtime memory and 329 MB of on‑device assets, enabling continuous streaming of 20–320 s of audio and improving Mean Opinion Scores by +0.28 overall and +0.42 on conversational speech compared to the previous on‑device TTS system.
By Dongseong Hwang, Prasanth Yadla, Kaan Elgin, Shifas Padinjaru Veettil, Sivanand Achanta, Dipjyoti Paul, Ramya Rasipuram, Tyler Johnson, Emad Soroush, Chung-Cheng Chiu, Zhifeng Chen