Mixture-of-Experts (MoE) architectures enhance the scalability and efficiency of large language models (LLMs) by activating only a subset of parameters. However, the interpretability of individual experts and corresponding strategies for post-training adaptation to domain-specific tasks remain underexplored. In this work, we develop an interpretability framework for expert-level monosemanticity in MoE models, drawing inspiration from sparse autoencoders and offering new insights into the specialization patterns of domain experts. We demonstrate that the inherent sparsity of MoE models encourages stronger monosemantic behavior at the expert level, which allows for the identification of experts responsible for particular downstream tasks and aids in preserving cross-domain performance. Building on this, we propose an expert-frozen fine-tuning method that selectively updates domain-specific experts while freezing the rest, which also alleviates catastrophic forgetting. Experiments on specialized domains, including medicine and finance, show that our approach performs on par with or better than fully fine-tuned models on in-domain tasks, while achieving relative improvements of 19.58% and 37.09% in retaining performance on out-of-domain benchmarks. We also demonstrate consistent advantages over PEFT baselines such as LoRA, achieving relative improvements of 11.49% and 8.60% in other domains.

Reinforcement Learning from Verifiable Rewards (RLVR) has emerged as the standard paradigm for improving reasoning capability of large language models, while Multi-Token Prediction (MTP) has been a widely adopted module in pretraining. Combining the two is a natural approach, yet current RL practices detach MTP gradients because joint training degrades the performance. We revisit this failure from an optimization perspective. We show that the per-step effect of MTP on the RL objective can be decomposed into two terms: a first-order correlation and a second-order perturbation penalty. This decomposition unifies three MTP training regimes: Detach, Cross-Entropy loss, and Policy loss, and explains why each succeeds or fails. Further analysis of policy loss reveals that, although it aligns with intuition, performance still degrades: the correlation term decays while the quadratic penalty persists. Guided by the analysis, we propose Optimal Coefficient Calibration (OCC), an adaptive scheme that tracks the optimal coefficient online via a log-probability proxy at negligible cost. Across six competition-level mathematical reasoning benchmarks, OCC consistently matches or exceeds the detach baseline, delivering improved joint MTP-RL training performance.

Tool-Integrated Reasoning (TIR) is a fundamental capability for LLM agents to solve complex tasks by interacting with external tools iteratively. Reinforcement Learning (RL) has become the dominant paradigm for enabling this capability. However, existing approaches typically assign uniform trajectory-level advantages and treat all correct tool calls equally, ignoring the varying difficulty and learning value across trajectories and reasoning steps. This can lead to imprecise learning signals that do not adequately distinguish between trivial and challenging tool-use patterns. To address this limitation, we propose HiDiffTIR, a Hierarchical Difficulty-aware policy optimization framework for multi-turn TIR. HiDiffTIR performs difficulty-aware credit assignment at both trajectory and turn levels, enabling the policy to focus on more informative trajectories and harder reasoning steps. Notably, this fine-grained optimization is achieved without additional supervision, relying solely on group-level statistics derived from standard RL rollouts. Extensive experiments on three tool-using benchmarks demonstrate that HiDiffTIR consistently improves multi-turn TIR performance and tool invocation accuracy over strong RL baselines, highlighting the necessity of difficulty-aware credit assignment for effective policy optimization in tool-integrated LLM agents.

Adaptive context compression is vital for scaling Large Language Models (LLMs) to complex, multi-turn agent tasks. However, rule-based compression methods may discard task-critical nuances, while Reinforcement Learning (RL) approaches usually struggle to balance information retention and token efficiency under the sparse rewards inherent to long-horizon workflows. To bridge this gap, we propose ZipRL, a novel adaptive compression framework tailored for Reinforcement Learning from Verifiable Rewards (RLVR). ZipRL features a multi-granularity compression mechanism for active, non-uniform information reduction, coupled with Hindsight Response Replay (HRR)—a technique designed to densify training signals during RLVR optimization. Theoretically, we prove ZipRL’s superior task-relevant utility over uniform methods. Concretely, ZipRL utilizes coarse-to-fine prompts for macro-compression and incorporates HRR into GRPO via generalized advantage reshaping. Multiple models of varying versions and parameter scales validate the effectiveness of our approach. Benchmarks on five agent tasks show ZipRL outperforms state-of-the-art approaches by 27.9% and 34.7% across Qwen3-4B and Qwen3-8B models, while maintaining exceptional token efficiency and robustness under extreme 256-turn extrapolation stress tests. Our code is available at https://github.com/huzhexin/ZipRL.

Extracting skills from past successes is critical for the efficient evolution of Large Language Model (LLM) agents. Prevailing agent self-evolution paradigms typically rely on a core assumption: equipping LLMs with skill memories derived from successful trajectories will monotonically improve their problem-solving capabilities. However, probe analyses reveal that extracting skills solely from successful trajectories traps the model in a Skill Imitation Trap. For tasks that resemble past successes but require different tools, retrieving more skills paradoxically increases the model’s confidence in wrong tool calls—procedure skills raise the wrong-tool margin by 47% over a memory-free baseline. To overcome this limitation, we propose Boundary-Aware Skill Memory (BASM), which augments each skill with explicit boundary fields—applicability conditions, risk cues, avoidance rules, and recovery notes. These fields transform each retrieved skill from an unconditional action template into state-conditioned guidance: the agent applies the skill when its conditions hold, suppresses inapplicable tool calls when they do not, and issues targeted repairs when execution fails. Across three agent benchmarks and four model scales, BASM consistently outperforms success-distilled skill-memory baselines: it improves task success rate by up to 23.8% on AppWorld, accuracy by up to 5.0% on BFCL, and reduces attack success rate by 4.6% on AgentDojo, while simultaneously reducing average AppWorld steps by up to 6.6% relative to the memory-free baseline.

Agentic retrieval-augmented generation (RAG) systems enable large language models (LLMs) to solve complex tasks through multi-step interaction with external retrieval tools. However, such multi-step interaction often involves redundant search steps, incurring substantial computational cost and latency. Prior work limits search depth (i.e., the number of search steps) to reduce cost, but this often leads to underexploration of complex questions. To address this, we first investigate how search depth affects accuracy and find a minimal sufficient search depth that defines an accuracy-efficiency trade-off, jointly determined by question complexity and the agent’s capability. Furthermore, we propose AutoSearch, a reinforcement learning framework that evaluates each search step via self-generated intermediate answers. By a self-answering mechanism, AutoSearch identifies the minimal sufficient search depth and promotes efficient search by rewarding its attainment while penalizing over-searching. In addition, reward mechanisms are introduced to stabilize search behavior and improve answer quality on complex questions. Extensive experiments on multiple benchmarks show that AutoSearch achieves a superior accuracy-efficiency trade-off, alleviating over-searching while preserving search quality.

Memory is essential for LLM agents to accumulate task experience and reuse task-specific execution strategies. However, real-world deployment over boundary-agnostic and evolving task streams exposes a fundamental stability-plasticity dilemma. External retrieval-based memory can rapidly absorb new evidence, but it often fails to internalize recurring execution patterns and incurs inference-time retrieval overhead. Parametric memory enables stable and efficient execution once learned, but typically relies on explicit task boundaries and fixed parameter budgets. Inspired by the human brain, which balances plasticity and stability through complementary episodic storage and gradual consolidation, we propose UniMem, a self-routing framework for autonomous memory management. UniMem uses learnable routing tokens as memory controllers, enabling adaptive coordination between complementary memory pathways: novel or sparse tasks are retained in an episodic buffer for retrieval-augmented execution, while recurring and reliable patterns are consolidated into expandable parametric memory. By decoupling task identification from task execution with routing tokens and parametric memory blocks, UniMem expands memory on demand without task labels during deployment or uncontrolled parameter growth. Experiments on long-horizon streaming task sequences show that UniMem typically achieves strong aggregate results while maintaining routing and execution stability, with an average gain of 4.0 EM over Replay LoRA.

Training LLMs for tool use requires more than final-answer supervision: models need trajectories that specify how to decompose a question, call tools across turns, use retrieved evidence, and recover from earlier errors. Existing data construction pipelines often rely on large-scale real-world tool API calls or simple templates, making them costly or limited in interaction coverage. We introduce ToolForge, a virtual-tool data synthesis pipeline for constructing tool-use datasets with three targeted abilities: multi-hop reasoning, multi-turn tool use, and self-reflection. Starting from QA seed data, ToolForge builds an extensible tool space from 19 base virtual tools and automatically generated variants, selects among four tool-use paradigms, simulates both successful and misleading retrieval outcomes, and generates validated tool-use dialogues without large-scale real-world tool API execution. A Multi-Layer Validation module filters invalid traces and is audited with hard negatives mined by Monte Carlo Tree Search. Fine-tuning Qwen3-8B on only 4,250 ToolForge samples produces a model named ToolForge-8B, which improves over the Tool-Integrated Reasoning baseline by 21.6% on five out-of-domain QA benchmarks and remains competitive on BFCL and API-Bank. We further quantify the pipeline construction cost, including tool initialization, dialogue generation, validation, and SFT, and compare ToolForge with existing synthetic tool-data sources. Our code and dataset are publicly available at https://github.com/Buycar-arb/ToolForge.

Recent advances in large language models (LLMs) have significantly improved automated code generation, enabling tools such as GitHub Copilot and CodeWhisperer to assist developers in a wide range of programming tasks. However, the translation of complex mobile UI designs into high-fidelity front-end code remains a challenging and underexplored area, especially as modern app interfaces become increasingly intricate. In this work, we propose UIOrchestra, a collaborative multi-agent system designed for the AppUI2Code task, which aims to reconstruct static single-page applications from design mockups. UIOrchestra integrates three specialized agents, layout description, code generation, and difference analysis agent that work collaboratively to address the limitations of single-model approaches. To facilitate robust evaluation, we introduce APPUI, the first benchmark dataset for AppUI2Code, constructed through a human-in-the-loop process to ensure data quality and coverage. Experimental results demonstrate that UIOrchestra outperforms existing methods in reconstructing complex app pages and highlight the necessity of multi-agent collaboration for this task. We hope our work will inspire further research on leveraging LLMs for front-end automation. The code and data will be released upon paper acceptance.

Ensembling large language models (LLMs) can effectively combine diverse strengths of different models, offering a promising approach to enhance performance across various tasks. However, existing methods typically rely on fixed weighting strategies that fail to adapt to the dynamic, context-dependent characteristics of LLM capabilities. In this work, we propose Reinforcement Learning-Assisted Ensemble for LLMs (RLAE), a novel framework that reformulates LLM ensemble through the lens of a Markov Decision Process (MDP). Our approach introduces a RL agent that dynamically adjusts ensemble weights by considering both input context and intermediate generation states, with the agent being trained using rewards that directly correspond to the quality of final outputs. We implement RLAE using both single-agent and multi-agent reinforcement learning algorithms (RLAE_PPO and RLAE_MAPPO ), demonstrating substantial improvements over conventional ensemble methods. Extensive evaluations on a diverse set of tasks show that RLAE outperforms existing approaches by up to 3.3\\% accuracy points, offering a more effective framework for LLM ensembling. Furthermore, our method exhibits superior generalization capabilities across different tasks without the need for retraining, while simultaneously achieving lower time latency. The source code is available at here.