Long-term memory systems increasingly organize interaction histories into hierarchical contexts, where coarse summaries, mid-level events, and fine-grained evidence provide complementary signals for future queries. Yet after retrieval, these signals are often collapsed into a single generation context and used implicitly. We ask how retrieved hierarchical memory should shape generation after access, framing it as a post-retrieval calibration problem. We propose HiR, a training-free inference-time framework that contrasts full-hierarchy decoding with a coarse-only reference view, preserving the complete hierarchy while tempering over-reliance on broad abstractions. Extensive experiments and analyses show that HiR enables more effective and robust use of hierarchical memory without changing memory construction or retrieval; more broadly, our results suggest that memory research should consider not only what to store or retrieve, but also how retrieved information shapes generation.

Large Language Models (LLMs) are transforming chemical research by enabling specialized reasoning models and tool-augmented agents. Yet existing benchmarks typically evaluate reasoning or tool invocation in isolation, overlooking the dynamic interaction between internal deliberation and external execution. We present ChemEval2, a multimodal benchmark for tool-augmented chemical reasoning, containing 2,410 samples spanning fundamental principles, spectroscopic analysis, and complex reaction understanding. Built from professional textbooks and experimental data through automated generation and rigorous human annotation, ChemEval2 introduces a tool-augmented Chain-of-Thought evaluation framework that embeds tool-use checkpoints at stages where unaided language reasoning becomes unreliable. This design evaluates whether models can identify information gaps, invoke appropriate tools, and integrate tool outputs into reasoning trajectories. Across diverse text and multimodal models, we observe a process–outcome inconsistency: models can satisfy intermediate reasoning steps or execute useful tool calls yet still fail at final synthesis. These results provide evidence that coordinating computational tool use with downstream chemical reasoning remains a substantial challenge. We open-source ChemEval2 and an MCP-based suite of specialized chemical tools for reproducible.

Personalization has become crucial for adapting models to the diverse and evolving needs of users across cultural, temporal, and contextual dimensions. While existing methods often rely on centralized fine-tuning or static preference alignment within a single model, they struggle to achieve both real-time and high-quality personalization under the resource and privacy constraints of personal devices. To address this challenge, we propose CoSteer, a collaborative framework that enables tuning-free, real-time personalization via decoding-time adaptation. By leveraging logit differences between context-aware and -agnostic local small models, CoSteer steers cloud-based large models, ensuring effective personalization while preserving the large model’s capabilities. Personalization is handled locally, with only final tokens sent to the cloud, maintaining both user context and system efficiency. Through extensive experiments across a wide range of tasks, we demonstrate that CoSteer generates high-quality personalized content, ensuring both effectiveness and computational efficiency. Our results highlight its robustness across models and environments, confirming its practical applicability in real-world scenarios. Our code and data are available at https://anonymous.4open.science/r/Costeer-4977.

With the growing demand for processing multiple image sequences in real-world applications, various visual token pruning methods have emerged to mitigate the computational and context length constraints faced by Large Vision Language Models (LVLMs). However, most existing pruning approaches rely on static strategies that struggle to adapt across different architectural LVLMs and multi-image scenarios, and are additionally constrained by their dependence on attention computations that are incompatible with efficient techniques like FlashAttention.To address these limitations, we propose a training-free, Adaptive Visual Token Pruning (AVTP) framework, applicable to diverse LVLM architectures. We strategically determine pruning layers based on empirical analysis of visual attention distributions across various LVLMs, and implement adaptive pruning ratios in multi-image contexts where images of higher importance retain proportionally more tokens. We conduct extensive experiments across multiple LVLMs with different architectures to demonstrate the effectiveness and robustness of AVTP. Specifically, Qwen3VL-8B achieves 2× inference speedup while maintaining 96.1% of its original accuracy on multiple multi-image benchmarks, InternVL3.5-8B retains 94.1% accuracy, and LLaVA-OneVision-7B even exceeds its original baseline performance. Our code is available at https://github.com/zry13/AVTP.

Realizing personalized intelligence faces a core dilemma: sending user history to centralized large language models raises privacy concerns, while on-device small language models lack the reasoning capacity required for high-quality generation. Our pilot study shows that purely local enhancements remain insufficient to reliably bridge this gap. We therefore propose SpecSteer, an asymmetric collaborative inference framework that synergizes private on-device context with cloud-scale reasoning. SpecSteer casts collaboration as Bayesian knowledge fusion and repurposes speculative decoding as a distributed alignment protocol, yielding a Draft–Verify–Recover pipeline: the on-device model drafts personalized sequences; the cloud validates via a ratio-based mechanism that decouples reasoning verification from private context, filtering logical flaws without accessing raw user context; upon rejection, a steering recovery injects local intent during correction. Experiments demonstrate that SpecSteer successfully closes the reasoning gap and achieves superior personalized generation performance, while delivering a 2.36 X speedup over standard baselines.

Chain-of-Thought (CoT) reasoning is widely used to improve LLM performance, and recent foundation recommender models adopt it by generating textual reasoning before predicting target items represented by Semantic IDs (SIDs). However, we observe that enabling thinking mode in models such as OpenOneRec can degrade recommendation quality by up to 25%. We investigate this failure and identify Linguistic Inertia: when a textual CoT segment is inserted before SID generation, the model relies more on natural-language context and less on historical SID evidence. Further analyses show that this effect is amplified by reduced access to historical information and longer CoT lengths. To mitigate it, we propose Linguistic-Inertia-Calibrated Decoding (LICD), a training-free framework that combines Reasoning-Chain Compression and Bias-Subtracted Contrastive Inference. Experiments on three large-scale benchmarks show that LICD consistently outperforms both no-thinking and original-thinking baselines. Our code is available at https://github.com/USTC-StarTeam/LICD.

Agentic LLM pipelines face escalating inference costs as context accumulates across retrieval, tool use, and multi-turn interactions. To control latency, deployments routinely compress inputs, but this degrades task accuracy. Speculative decoding (SD) accelerates generation losslessly, yet it assumes the drafter and verifier share an identical context, preventing SD from resolving the accuracy–overhead trade-off. We propose AsymSpec, an asymmetric speculative decoding framework that breaks this symmetry: a lightweight drafter reads the full input while the large verifier operates on the compressed view. The drafter steers the verifier via a contrastive 𝛿-fusion of logits, modulated by a divergence-aware acceptance gate that preserves verification stability and high draft acceptance rates. Evaluated across four agentic capabilities and two end-to-end agent benchmarks, AsymSpec reaches ≈ 90% of full-context accuracy on average, delivering throughput speedups of 1.3×–1.7× at 0.2×–0.3× the compute cost on isolated text capabilities. These results show that asymmetric context access yields substantial gains precisely when compression discards critical reasoning signals.

Faithful chart generation in real-world data-science workflows requires grounding visualizations in scattered evidence, computing chart-ready quantities, and rendering them accurately. Modern LLMs can produce visually plausible, instruction-compliant charts, yet data-level hallucinations remain difficult to detect in long, noisy, and multimodal contexts. To measure this gap, we introduce , an expert-annotated benchmark of 1,482 task-conditioned chart-generation instances drawn from real-world scientific papers, financial filings, and ecosystem reports. DEEPCHART formulates chart generation as an Extract–Reason–Visualize pipeline and evaluates source-data extraction, derived-data reasoning, and chart rendering stage by stage. Experiments with state-of-the-art models show that visually plausible charts often conceal data-level hallucinations, with extraction and reasoning errors common in realistic long and multimodal settings. These findings suggest that larger context windows alone are insufficient; faithful chart generation also requires reliable evidence extraction and quantitative reasoning before rendering.

Industrial recommendation training increasingly relies on decentralized multi-GPU infrastructure, yet sparse embedding tables remain a major bottleneck due to significant storage, lookup, and communication overhead. Most existing works typically optimize only one of these bottlenecks or are designed for parameter server architectures, making them less effective in industrial decentralized training stacks. To tackle these challenges, we propose MixEmb, a practical framework for efficient decentralized embedding training through importance-aware quantization and caching. We first design a novel importance metric, i.e., embedding score, which can distinguish the contribution of embedding vectors based on spatial heterogeneity and temporal dynamics. Guided by embedding scores, MixEmb employs mixed precision quantization and hierarchical caching at fine-grained vector level. Besides, our key finding is that quantization and caching decisions are tightly coupled. Instead of naive combination, we delicately integrate them into a unified training framework through coupled co-design. A theoretical convergence analysis is also provided for MixEmb. We implement and evaluate MixEmb in real-world training scenarios. Across three workloads, MixEmb attains up to 1.55× speedup and 3.54× compression ratio compared to state-of-the-art methods.

Generative listwise reranking leverages global context for superior retrieval but is plagued by intrinsic position bias, where models exhibit structural sensitivity to input order independent of relevance. Existing mitigations present a dilemma: inference-time aggregation incurs prohibitive latency, while training-based methods often fail to eradicate ingrained priors, particularly in compact models. To resolve this dilemma, we propose CapCal (Content-Agnostic Probability Calibration), a training-free framework that mechanically decouples positional bias from ranking decisions. By estimating the bias distribution via content-free placeholders, CapCal rectifies output logits through an entropy-adaptive contrastive mechanism. Evaluations across 10 benchmarks confirm that CapCal achieves superior performance among training-free methods while preserving single-pass efficiency. Notably, it unlocks the latent potential of lightweight models (e.g., 0.6B), delivering absolute NDCG gains exceeding 10 points and outperforming computationally expensive data augmentation strategies.

The evolution of Large Language Models (LLMs) is shifting the focus from single, verifiable tasks toward complex, open-ended real-world scenarios, imposing significant challenges on the post-training phase. In these settings, the scale and complexity of reward systems have grown significantly, transitioning toward multi-objective formulations that encompass a comprehensive spectrum of model capabilities and application contexts. However, traditional methods typically rely on fixed reward weights, ignoring non-stationary learning dynamics and struggling with data heterogeneity across dimensions. To address these issues, we propose SPARD, a framework that establishes an automated, self-paced curriculum by perceiving learning progress to dynamically adjust multi-objective reward weights and data importance, thereby synchronizing learning intent with data utility for optimal performance. Extensive experiments across multiple benchmarks demonstrate that SPARD significantly enhances model capabilities across all domains. Our code is publicly available at https://github.com/USTC-StarTeam/SPARD.

Recent advancements in large language models (LLMs) have revolutionized natural language processing through their remarkable capabilities in understanding and executing diverse tasks. While supervised fine-tuning, particularly in Retrieval-Augmented Generation (RAG) scenarios, effectively enhances task-specific performance, it often leads to catastrophic forgetting, where models lose their previously acquired knowledge and general capabilities. Existing solutions either require access to general instruction data or face limitations in preserving the model’s original distribution. To overcome these limitations, we propose SelfAug, a self-distribution alignment method that aligns input sequence logits to preserve the model’s semantic distribution, thereby mitigating catastrophic forgetting and improving downstream performance. Extensive experiments demonstrate that SelfAug achieves a superior balance between downstream learning and general capability retention. Our comprehensive empirical analysis reveals a direct correlation between distribution shifts and the severity of catastrophic forgetting in RAG scenarios, highlighting how the absence of RAG capabilities in general instruction tuning leads to significant distribution shifts during fine-tuning. Our findings not only advance the understanding of catastrophic forgetting in RAG contexts but also provide a practical solution applicable across diverse fine-tuning scenarios.

Event extraction (EE) is a fundamental task in natural language processing (NLP) that involves identifying and extracting event information from unstructured text. Effective EE in real-world scenarios requires two key steps: selecting appropriate schemas from hundreds of candidates and executing the extraction process.Existing research exhibits two critical gaps: (1) the rigid schema fixation in existing pipeline systems, and (2) the absence of benchmarks for evaluating joint schema matching and extraction.Although large language models (LLMs) offer potential solutions, their schema hallucination tendencies and context window limitations pose challenges for practical deployment. In response, we propose Adaptive Schema-aware Event Extraction (ASEE), a novel paradigm combining schema paraphrasing with schema retrieval-augmented generation. ASEE adeptly retrieves paraphrased schemas and accurately generates targeted structures.To facilitate rigorous evaluation, we construct the Multi-Dimensional Schema-aware Event Extraction (MD-SEE) benchmark, which systematically consolidates 12 datasets across diverse domains, complexity levels, and language settings.Extensive evaluations on MD-SEE show that our proposed ASEE demonstrates strong adaptability across various scenarios, significantly improving the accuracy of event extraction. Our codes and datasets are available at https://github.com/USTC-StarTeam/ASEE.git

Generating knowledge-intensive and comprehensive long texts, such as encyclopedia articles, remains significant challenges for Large Language Models. It requires not only the precise integration of facts but also the maintenance of thematic coherence throughout the article. Existing methods, such as multi-agent discussion, often struggle with issues like hallucinations, topic incoherence, and significant latency. To address these challenges, we propose RAPID, an efficient Retrieval-Augmented long text generation framework with writing Planning and Information Discovery. RAPID consists of three main modules: (1) Retrieval-augmented preliminary outline generation to reduce hallucinations, (2) Attribute-constrained search for efficient information discovery, (3) Plan-guided article generation for enhanced coherence. Extensive experiments on our newly compiled benchmark dataset, FreshWiki-2024, demonstrate that RAPID significantly outperforms state-of-the-art methods across a wide range of evaluation metrics (long-text generation, outline quality, latency, etc). Our work provides a robust and efficient solution to the challenges of automated long-text generation.