arXiv subjects

Zhaozhuo Xu

Publications and source records attributed to Zhaozhuo Xu.

At least 19 recordsLinked to original sources

arXiv2026

Sketching the Readout of Large Language Models for Scalable Data Attribution and Valuation

Data attribution and valuation are critical for understanding data-model synergy for Large Language Models (LLMs), yet existing gradient-based methods suffer from scalability challenges on LLMs. Inspired by human cognition, where decision making relies on a focused readout of relevant memories rather than replaying all pathways, we introduce RISE (Readout Influence Sketching Estimator). Instead of computing and indexing gradients across the entire LLM, RISE focuses on influence hotspots at the output layer, where influence signals concentrate, and the gradient admits a decomposed outer-product form. This enables a dual-channel representation combining a lexical residual channel (RH) and a semantic projected-error channel (GH). Applying CountSketch projections to these channels achieves strong compression while maintaining accurate attribution. Across the OLMo (1B-32B) and Pythia (14M-6.9B) families, RISE reduces index storage by up to 112$\times$ compared to RapidIn and scales to 32B parameters LLM, where gradient-based baselines such as RapidIn and ZO-Inf become memory-infeasible. We evaluate RISE on two paradigms: (1) retrospective attribution, retrieving influential training examples for specific predictions, and (2) prospective valuation, scoring candidate data utility zero-shot. We validate RISE on three tasks: Howdy backdoor data detection, Finance-Medical domain separation, and Brain Rot high-quality data selection. In a closed-loop Brain Rot study, continued pretraining on RISE-selected data yields consistent downstream improvements. Overall, RISE provides a practical and scalable primitive for influence analysis and training-data selection in modern large language models.

cs.LG↗

arXiv2026

A Replicate-and-Quantize Strategy for Plug-and-Play Load Balancing of Sparse Mixture-of-Experts LLMs

Sparse Mixture-of-Experts (SMoE) architectures are increasingly used to scale large language models efficiently, delivering strong accuracy under fixed compute budgets. However, SMoE models often suffer from severe load imbalance across experts, where a small subset of experts receives most tokens while others are underutilized. Prior work has focused mainly on training-time solutions such as routing regularization or auxiliary losses, leaving inference-time behavior, which is critical for deployment, less explored. We present a systematic analysis of expert routing during inference and identify three findings: (i) load imbalance persists and worsens with larger batch sizes, (ii) selection frequency does not reliably reflect expert importance, and (iii) overall expert workload and importance can be estimated using a small calibration set. These insights motivate inference-time mechanisms that rebalance workloads without retraining or router modification. We propose Replicate-and-Quantize (R&Q), a training-free and near-lossless framework for dynamic workload rebalancing. In each layer, heavy-hitter experts are replicated to increase parallel capacity, while less critical experts and replicas are quantized to remain within the original memory budget. We also introduce a Load-Imbalance Score (LIS) to measure routing skew by comparing heavy-hitter load to an equal allocation baseline. Experiments across representative SMoE models and benchmarks show up to 1.4x reduction in imbalance with accuracy maintained within +/-0.6%, enabling more predictable and efficient inference.

cs.LG↗

arXiv2026

When Implausible Tokens Get Reinforced: Tail-Aware Credit Calibration for LLM Reinforcement Learning

Reinforcement learning (RL) has achieved remarkable success in enhancing the reasoning capabilities of large language models (LLMs). However, widely used critic-free RL methods rely on uniform credit assignment, broadcasting the same advantage to all tokens regardless of their differences. We identify a critical failure mode of this design, which we refer to as Positive-Credit Contamination: low-probability tail tokens that are contextually erroneous receive identical positive credit to plausible ones within the same trajectory, resulting in the indiscriminate reinforcement of flawed reasoning behavior. To mitigate this issue, we propose Tail-Aware Credit calibratiOn (TACO), a method that calibrates uniform credit assignment to suppress undesirable positive updates. TACO first computes a tail-risk score that incorporates the local generation context to assess each token's risk of falling into the unreliable tail, distinguishing unexpected rarity from uncertainty-driven exploration. TACO then uses this score to tune positive credit for risky tokens without removing their gradients entirely, so that recurring useful rare patterns can accumulate reinforcement while incidental noise is progressively dampened. Experimental results across three LLMs and eight benchmarks show that TACO consistently outperforms GRPO-style baselines. Notably, TACO improves training stability, supporting sustained performance gains in long-horizon RL. The source code is available at: https://github.com/xiuyilou/TACO.

cs.CL↗

arXiv2026

Forget Without Compromise: Nexus Sampling for Streaming KV-Cache Eviction Under Fixed Budgets

Long-context and agentic LLM workloads push the KV cache past any fixed memory budget, forcing the inference stack to permanently evict tokens at every step of a continuous-inference stream. Existing methods all share the same template, a per-step direct-attention score followed by deterministic top-$K$ selection, which converts a single below-cutoff step into an irreversible verdict and permanently erases any subtly important token that direct attention cannot single out from noise. To address this challenge, we propose Nexus Sampling, a training-free eviction method that pairs Nexus scoring, an iterative walk over direct attention that surfaces bridge tokens, with weighted reservoir sampling, which retains tokens with inclusion probability in place of deterministic top-$K$. Theoretically, we show that Nexus Sampling dominates deterministic top-$K$ in long-run survival of subtly important tokens. Empirically, at 80% KV cache eviction, Nexus Sampling matches dense attention within 1% on LongBench while outperforming top-$K$ baselines on retrieval-heavy tasks, with up to 10x smaller per-sequence cache memory.

cs.LG↗

arXiv2026

Learning at the Right Pace: Adaptive Data Scheduling Improves LLM Reinforcement Learning

Large Language Models (LLMs) achieve remarkable reasoning capabilities through reinforcement learning (RL) post-training. However, existing RL post-training commonly relies on uniform data sampling, which ignores the semantic structure of the training data and the changing capability of the training policy. To address these limitations, we propose Adaptive Data Scheduling (ADS), a dual-level data scheduling framework for pacing RL post-training that replaces uniform sampling with an adaptive distribution over semantic clusters and policy-boundary sample selection. At the cluster level, ADS organizes samples according to semantic patterns and maintains an adaptive inter-cluster distribution to solidify current training progress. At the sample level, ADS performs intra-cluster scheduling to continuously sample policy-boundary samples, which provides informative relative advantages. Experimental results across three LLMs and seven reasoning benchmarks demonstrate that ADS improves average accuracy by 5.2% over Group Relative Policy Optimization (GRPO). Notably, ADS consistently improves RL methods with different objective designs, highlighting its potential as a general data scheduling strategy for LLM RL post-training. The source code is available at: https://github.com/Richard-zrx/ADS.

cs.CL↗

arXiv2026

Randomized Antipodal Search Done Right for Data Pareto Improvement of LLM Unlearning

Large language models (LLMs) sometimes memorize undesirable knowledge, which must be removed after deployment. Prior work on machine unlearning has focused largely on optimization methods that adjust parameters to enforce forgetting while preserving retention. However, these approaches assume that the forget and retain sets are readily available, which rarely holds in practice. Unlearning is typically triggered by an undesired generation at inference time, making the retrieval of relevant data the central challenge. We introduce the notion of data Pareto improvement for LLM unlearning, which formalizes how retrieval can expand the achievable trade-off frontier between forgetting and retention. To realize this principle, we propose Randomized Antipodal Search on Linearized Influence Kernel (RASLIK), a retrieval algorithm that combines permutation-projection hashing with randomized antipodal search. RASLIK reduces selection variance, achieves sublinear complexity, and yields a double gain in both quality and efficiency. Across multiple models, datasets, and unlearning algorithms, RASLIK consistently outperforms deterministic baselines and even oracle sampling, establishing randomized search as a principled and scalable solution for data-centric unlearning.

cs.LG↗

arXiv2026

The Diminishing Returns of Early-Exit Decoding in Modern LLMs

In Large Language Model (LLM) inference, early-exit refers to stopping computation at an intermediate layer once the prediction is sufficiently confident, thereby reducing latency and cost. However, recent LLMs adopt improved pretraining recipes and architectures that reduce layer redundancy, potentially limiting early-exit opportunities. We re-evaluate layer-wise early-exit in modern LLMs and analyze how intermediate representations evolve during training. We introduce a metric to quantify a model's intrinsic suitability for early-exit and propose a benchmark for researchers to explore the potential early-exit benefits on different models and workloads. Our results show a diminishing trend in early-exit effectiveness across newer model generations. We further find that dense transformers generally offer greater early-exit potential than Mixture-of-Experts and State Space Models. In addition, larger models, particularly those with more than 20 billion parameters, and base pretrained models without specialized tuning tend to exhibit higher early-exit potential.

cs.CL↗

arXiv2026

Copyright Detective: A Forensic System to Evidence LLMs Flickering Copyright Leakage Risks

We present Copyright Detective, the first interactive forensic system for detecting, analyzing, and visualizing potential copyright risks in LLM outputs. The system treats copyright infringement versus compliance as an evidence discovery process rather than a static classification task due to the complex nature of copyright law. It integrates multiple detection paradigms, including content recall testing, paraphrase-level similarity analysis, persuasive jailbreak probing, and unlearning verification, within a unified and extensible framework. Through interactive prompting, response collection, and iterative workflows, our system enables systematic auditing of verbatim memorization and paraphrase-level leakage, supporting responsible deployment and transparent evaluation of LLM copyright risks even with black-box access.

cs.CL↗

arXiv2026

Scout Before You Attend: Sketch-and-Walk Sparse Attention for Efficient LLM Inference

Self-attention dominates the computational and memory cost of long-context LLM inference across both prefill and decode phases. To address this challenge, we introduce Sketch&Walk Attention, a training-free sparse attention method that determines sparsity with lightweight sketches and deterministic walk. Sketch&Walk applies Hadamard sketching to get inexpensive approximations of attention scores, then aggregates these estimates across layers via a walk mechanism that captures attention influence beyond direct interactions between tokens. The accumulated walk scores are used to select top-k attention blocks, enabling dynamic sparsity with a single training-free algorithm that applies uniformly to both the prefill and decode phases, together with custom sparse attention kernels. Across a wide range of models and tasks, Sketch&Walk maintains near-lossless accuracy at 20% attention density and can slightly outperform dense attention in some settings, while achieving up to 6x inference speedup.

cs.LG↗

arXiv2026

LLM Multi-Agent Systems: Challenges and Open Problems

This paper explores multi-agent systems and identify challenges that remain inadequately addressed. By leveraging the diverse capabilities and roles of individual agents, multi-agent systems can tackle complex tasks through agent collaboration. We discuss optimizing task allocation, fostering robust reasoning through iterative debates, managing complex and layered context information, and enhancing memory management to support the intricate interactions within multi-agent systems. We also explore potential applications of multi-agent systems in blockchain systems to shed light on their future development and application in real-world distributed systems.

cs.MA↗

arXiv2026

Sketch to Adapt: Fine-Tunable Sketches for Efficient LLM Adaptation

Adapting pre-trained large language models (LLMs) is crucial but challenging due to their enormous size. Parameter-efficient fine-tuning (PEFT) techniques typically employ additive adapters applied to frozen model weights. To further reduce memory usage, model weights are often compressed through quantization. However, existing PEFT methods often yield suboptimal model quality because they rely on restrictive assumptions, such as low-rank constraints on adapters to limit the number of trainable parameters. We find that sketching, a popular data compression technique, can serve as an efficient LLM adaptation strategy while avoiding the low-rank assumption. We introduce SketchTune, a compressive adaptation strategy that compresses LLM weights into compact fine-tunable sketches, integrating compression and adaptation into a unified framework. This integration eliminates the need for complex two-path computation in existing PEFT techniques, enabling faster and more memory-efficient training and inference. SketchTune is supported by mathematical insights into matrix classes that are better approximated using sketching rather than low-rank methods. Our extensive evaluations with Llama and Mistral models demonstrate that SketchTune outperforms leading PEFT methods across diverse tasks while using substantially smaller base models and comparable trainable parameters. As a highlight, SketchTune outperforms LoRA, DoRA, and S2FT on commonsense and math benchmarks using 2.6-3.5$\times$ smaller base models and exceeds LoftQ in accuracy by 14.48% on GSM8K with 7.3$\times$ fewer trainable parameters. Our code is available at https://github.com/LeanModels/SketchTune.

cs.LG↗

arXiv2025

Word Salad Chopper: Reasoning Models Waste A Ton Of Decoding Budget On Useless Repetitions, Self-Knowingly

Large Reasoning Models (LRMs) are often bottlenecked by the high cost of output tokens. We show that a significant portion of these tokens are useless self-repetitions - what we call "word salad" - that exhaust the decoding budget without adding value. Interestingly, we observe that LRMs are self-aware when trapped in these loops: the hidden states of <\n\n> tokens trailing each reasoning chunk exhibit patterns that allow us to detect word salad behavior on-the-fly via a single-layer linear classifier. Once detected, a simple chop appended by a straightforward regeneration prompt yields substantial length savings with minimal quality loss. Our work offers WordSaladChopper (WSC) - a lightweight, turnkey component for LRM that is minimally invasive to its reasoning trajectory by only removing semantically redundant tokens. Given its low overhead, strong savings, and the lack of semantic value of word salad tokens, we believe it is not too far-fetched to argue that WSC - or a similar component - is a must-have for all LRM applications with user experience in mind. Our code is publicly available at https://github.com/wenyaxie023/WordSaladChopper.

cs.CL↗

arXiv2025

Towards Fast LLM Fine-tuning through Zeroth-Order Optimization with Projected Gradient-Aligned Perturbations

Fine-tuning large language models (LLMs) using zeroth-order (ZO) optimization has emerged as a promising alternative to traditional gradient-based methods due to its reduced memory footprint requirement. However, existing ZO methods suffer from high variance in gradient estimation, leading to slow convergence and suboptimal performance on large-scale models. In this work, we propose P-GAP, a fast LLM fine-tuning approach through zeroth-order optimization with Projected Gradient-Aligned Perturbations. Specifically, we first estimate a low-dimensional gradient space and then align perturbations in projected gradients' direction within the space. This approach enables reduced the number of perturbed parameters and decreased variance, therefore accelerated convergence for LLM fine-tuning. Experiments on LLMs show that P-GAP consistently surpasses the baselines, achieving up to 6% increase in accuracy on classification tasks and up to 12% higher accuracy on generation tasks, with up to about 81% less training iterations and 70% less GPU hours. These results demonstrate that P-GAP enables fast, scalable, and resource-efficient ZO LLM fine-tuning.

cs.LG↗

arXiv2025

To Compress or Not? Pushing the Frontier of Lossless GenAI Model Weights Compression with Exponent Concentration

The scaling of Generative AI (GenAI) models into the hundreds of billions of parameters makes low-precision computation indispensable for efficient deployment. We argue that the fundamental solution lies in developing low-precision floating-point formats, which inherently provide numerical stability, memory savings, and hardware efficiency without dequantization overhead. In this paper, we present a theoretical and empirical study of an exponent concentration phenomenon in GenAI weights: exponents consistently exhibit low entropy across architectures and modalities. We show that this arises naturally from $α$-stable distributions induced by stochastic gradient descent, and we prove tight bounds on the entropy of exponents. Our analysis establishes a theoretical compression limit near FP4.67, which motivates the design of a practical FP8 format. Building on these insights, we propose Exponent-Concentrated FP8 (ECF8), a lossless compression framework with entropy-aware encoding and GPU-optimized decoding. Experiments on LLMs and DiTs up to 671B parameters demonstrate up to 26.9% memory savings and 177.1% throughput acceleration, with perfectly lossless computations, i.e., no deviation in model outputs. Our results establish exponent concentration as a statistical law of trained models and open a principled path for lossless low-precision floating-point design in the FP8 era.

cs.LG↗

arXiv2025

DEL-ToM: Inference-Time Scaling for Theory-of-Mind Reasoning via Dynamic Epistemic Logic

Theory-of-Mind (ToM) tasks pose a unique challenge for large language models (LLMs), which often lack the capability for dynamic logical reasoning. In this work, we propose DEL-ToM, a framework that improves verifiable ToM reasoning through inference-time scaling rather than architectural changes. Our approach decomposes ToM tasks into a sequence of belief updates grounded in Dynamic Epistemic Logic (DEL), enabling structured and verifiable dynamic logical reasoning. We use data generated automatically via a DEL simulator to train a verifier, which we call the Process Belief Model (PBM), to score each belief update step. During inference, the PBM evaluates candidate belief traces from the LLM and selects the highest-scoring one. This allows LLMs to allocate extra inference-time compute to yield more transparent reasoning. Experiments across model scales and benchmarks show that DEL-ToM consistently improves performance, demonstrating that verifiable belief supervision significantly enhances LLMs' ToM capabilities without retraining. Code is available at https://github.com/joel-wu/DEL-ToM.

cs.AI↗

arXiv2025

Context Engineering for Trustworthiness: Rescorla Wagner Steering Under Mixed and Inappropriate Contexts

Incorporating external context can significantly enhance the response quality of Large Language Models (LLMs). However, real-world contexts often mix relevant information with disproportionate inappropriate content, posing reliability risks. How do LLMs process and prioritize mixed context? To study this, we introduce the Poisoned Context Testbed, pairing queries with real-world contexts containing relevant and inappropriate content. Inspired by associative learning in animals, we adapt the Rescorla-Wagner (RW) model from neuroscience to quantify how competing contextual signals influence LLM outputs. Our adapted model reveals a consistent behavioral pattern: LLMs exhibit a strong tendency to incorporate information that is less prevalent in the context. This susceptibility is harmful in real-world settings, where small amounts of inappropriate content can substantially degrade response quality. Empirical evaluations on our testbed further confirm this vulnerability. To tackle this, we introduce RW-Steering, a two-stage finetuning-based approach that enables the model to internally identify and ignore inappropriate signals. Unlike prior methods that rely on extensive supervision across diverse context mixtures, RW-Steering generalizes robustly across varying proportions of inappropriate content. Experiments show that our best fine-tuned model improves response quality by 39.8% and reverses the undesirable behavior curve, establishing RW-Steering as a robust, generalizable context engineering solution for improving LLM safety in real-world use.

cs.CL↗

arXiv2025

OAT-Rephrase: Optimization-Aware Training Data Rephrasing for Zeroth-Order LLM Fine-Tuning

Fine-tuning large language models (LLMs) using zeroth-order optimization (ZO) offers a memory-efficient alternative to gradient-based methods but suffers from slower convergence and unstable optimization due to noisy gradient estimates. This paper introduces OAT-Rephrase, an Optimization-Aware Training data rephrasing strategy that leverages an LLM to rephrase training instances based on its understanding of the ZO dynamics, specifically MeZO, derived directly from its paper. The approach incorporates a dual-stage pipeline featuring a rewriter LLM and a semantic judge, ensuring all rephrasings retain task relevance and logical consistency. Evaluations across five classification tasks and three LLM architectures demonstrate that OAT-Rephrase consistently improves MeZO fine-tuning performance, often narrowing or eliminating the gap with first-order methods. Our findings suggest that optimization-aware rephrasing serves as a reusable and low-overhead enhancement for zeroth-order tuning regimes.

cs.LG↗

arXiv2025

Mitigating Non-IID Drift in Zeroth-Order Federated LLM Fine-Tuning with Transferable Sparsity

Federated Learning enables collaborative fine-tuning of Large Language Models (LLMs) across decentralized Non-Independent and Identically Distributed (Non-IID) clients, but such models' massive parameter sizes lead to significant memory and communication challenges. This work introduces Meerkat, a sparse zeroth-order optimization (ZO) method designed for federated LLM fine-tuning. By limiting fine-tuning to a transferable, static, extremely sparse subset of parameters, Meerkat achieves remarkable communication efficiency, enabling cost-effective high-frequency synchronization. With theoretical analysis and experiments, we show that this high-frequency communication effectively mitigates Non-IID data challenges and leads to superior performance compared to full-parameter ZO. Furthermore, experiment results show that Meerkat outperforms existing sparsity baselines with better performance at the same communication frequency. To further handle Non-IID drift, Meerkat leverages traceable local updates and forms a virtual path for each client. This virtual path mechanism reveals the GradIP phenomenon: the inner products between LLM pre-training gradients maintained by server and client gradients estimated via ZO converges for extreme Non-IID clients but oscillates for IID ones. This distinct behavior provides a signal for identifying clients with extreme data heterogeneity. Using this signal, Meerkat-vp is proposed to analyze GradIP trajectories to identify extreme Non-IID clients and applies early stopping to enhance aggregated model quality. Experiments confirm that Meerkat and Meerkat-vp significantly improve the efficiency and effectiveness of ZO federated LLM fine-tuning.

cs.LG↗