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, experimental 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 converge for extreme Non-IID clients but oscillate 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.
论文检索
输入标题、作者或关键词，从 12,113 篇学术成果中精准定位
Lipschitz-based certification offers efficient, deterministic robustness guarantees but has struggled to scale in model size, training efficiency, and ImageNet performance. We introduce \emph{LipNeXt}, the first \emph{constraint-free} and \emph{convolution-free} 1-Lipschitz architecture for certified robustness. LipNeXt is built using two techniques: (1) a manifold optimization procedure that updates parameters directly on the orthogonal manifold and (2) a \emph{Spatial Shift Module} to model spatial pattern without convolutions. The full network uses orthogonal projections, spatial shifts, a simple 1-Lipschitz $\beta$-Abs nonlinearity, and $L_2$ spatial pooling to maintain tight Lipschitz control while enabling expressive feature mixing. Across CIFAR-10/100 and Tiny-ImageNet, LipNeXt achieves state-of-the-art clean and certified robust accuracy (CRA), and on ImageNet it scales to 1–2B large models, improving CRA over prior Lipschitz models (e.g., up to $+8\%$ at $\varepsilon{=}1$) while retaining efficient, stable low-precision training. These results demonstrate that Lipschitz-based certification can benefit from modern scaling trends without sacrificing determinism or efficiency.
Fine-tuning large language models (LLMs) has achieved significant success in downstream tasks. However, as the model size continues to grow, traditional fine-tuning methods have become increasingly impractical due to their high computational and memory costs. This has motivated researchers to explore parameter-efficient and memory-friendly fine-tuning strategies to enable scalable approaches, with Low-Rank Adaptation (LoRA) standing out as a representative work. However, the LoRA update is restricted to a low-rank subspace, which results in suboptimal performance compared to the full-parameter update. Recent research has also explored memory-efficient fine-tuning LLMs using just forward passes while suffer from high variance in gradient estimation and low convergence speed. To address the issues above, we propose a new alternating optimization framework called LMAO (**L**ow-rank and **M**emory-efficient Zeroth-Order **A**lternating **O**ptimization), which combines the advantages of LoRA and MeZO. This method alternately updates the low-rank components and zeroth-order directions during training. By performing three forward propagations and one backward propagation, each update is full-rank, thereby reducing feature loss and enabling efficient fine-tuning under strict memory constraints. We provide theoretical guarantees on the convergence and convergence rate of this method. Empirical results demonstrate that, in experiments on multiple models (e.g., OPT, RoBERTa-large), LMAO achieves performance comparable to first-order methods. This presents a practical and scalable solution for fine-tuning large-scale models. Our source code is available at [https://github.com/workelaina/LMAO](https://github.com/workelaina/LMAO).
We introduce Synthetic Bootstrapped Pretraining (SBP), a language model (LM) pretraining procedure that first learns a model of relations between documents from the pretraining dataset and then leverages it to synthesize a vast new corpus for joint training. While the standard pretraining teaches LMs to learn causal correlations among tokens within a single document, it is not designed to efficiently model the rich, learnable inter-document correlations that can potentially lead to better performance. We validate SBP by designing a compute-matched pretraining setup and pretrain a 3B-parameter and a 6B-parameter model on up to 1T tokens from scratch. We find SBP consistently improves upon a strong repetition baseline and delivers up to 60% of performance improvement attainable by an oracle upper bound with access to 20x more unique data. Qualitative analysis reveals that the synthesized documents go beyond mere paraphrases -- SBP first abstracts a core concept from the seed material and then crafts a new narration on top of it. Besides strong empirical performance, SBP admits a natural Bayesian interpretation: the synthesizer implicitly learns to abstract the latent concepts shared between related documents.
Long-horizon sparse-reward tasks, such as goal-conditioned or robot manipulation tasks, remain challenging in offline reinforcement learning due to the credit assignment problem. Hierarchical methods have been proposed to tackle this problem by introducing sub-goal planning guided by value functions, which in principle can shorten the effective planning horizon for both high-level and low-level planners, and thereby avoiding the credit assignment problem. However, we demonstrate that the sub-goal selection mechanism is unreliable, as it relies on value functions suffering from generalization noise, which misguides value estimation and thus leads to sub-optimal sub-goals. In this work, to provide more reliable sub-goals, we novelly introduce a reliability-driven decision mechanism, and propose Reliability-Driven HRL (RD-HRL) as the solution. The reliability-driven decision mechanism provide decision-level targets for high-level policy, thereby providing noise-immune decision spaces for them, ensuring the reliability of sub-goals (which are termed as action-level targets in this paper). Comprehensive experimental results demonstrate that our approach RD-HRL outperforms baseline methods across multiple benchmarks, highlighting the competitive advantages of RD-HRL. Our code is anonymously available at \url{https://github.com/Looomo/RD-HRL-public}.
The deployment of large language models (LLMs) for specialized tasks on resource-constrained edge devices like smartphones and sensors presents a significant scalability problem. To run on such hardware, these massive models must be compressed using techniques like \emph{quantization or pruning} to reduce their memory and computational footprint. Concurrently, foundational LLMs are periodically updated by their developers with new data, making their $\textit{internal parameters shift over time}$. While parameter-efficient methods like Low-Rank Adaptation (LoRA) streamline personalization by fine-tuning only a small fraction of parameters, the resulting adapters are $\textbf{brittle}$; a LoRA trained for one specific compression scheme is incompatible with another, and an adapter trained on an older base model performs poorly on an updated one. This forces a costly cycle of retraining for each unique device and every new model release. To address this, we introduce a novel framework that creates a single, universally portable adapter that is both $\textbf{\textit{(i)} compression-aware and \textit{(ii)} temporally robust}$. We achieve this by augmenting the training process with a variety of simulated compression techniques during a single run, utilizing a quantized forward pass to build resilience while maintaining a full-precision backward pass for stable gradient optimization. $\textit{This method yields a unified adapter robust to diverse compression artifacts and the subtle parameter shifts from model evolution}$. Extensive experiments on models such as $\texttt{Llama-2, Llama-3.1, Gemma-2}$, and $\texttt{Mistral}$ across reasoning benchmarks like $\textit{SQA, MATH, and GSM8K}$ demonstrate that our single adapter achieves performance comparable to specialized adapters ($\textit{e.g.}$, QLoRA) that are individually retrained for each compression scheme. Furthermore, we show this single adapter maintains its high performance when applied to future, evolved versions of the base model, eliminating the need for periodic retraining. Our work pioneers an efficient paradigm for edge AI, creating portable model patches that bridge the gap between cloud-based personalization, the diverse hardware ecosystem, and the lifecycle of evolving LLMs.
This paper presents a method to analyze the inference patterns used by Large Language Models (LLMs) for judgment in a case study on legal LLMs, so as to identify potential incorrect representations of the LLM, according to human domain knowledge. Unlike traditional evaluations on language generation results, we propose to evaluate the correctness of the detailed inference patterns of an LLM behind its seemingly correct outputs. To this end, we quantify the interactions between input phrases used by the LLM as primitive inference patterns, because recent theoretical achievements have proven several mathematical guarantees of the faithfulness of the interaction-based explanation. We design a set of metrics to evaluate the detailed inference patterns of LLMs. Experiments show that even when the language generation results appear correct, a significant portion of the inference patterns used by the LLM for the legal judgment may represent misleading or irrelevant logic.
While large and diverse datasets have driven recent advances in large models, identifying the optimal data mixture for pre-training and post-training remains a significant open problem. We address this challenge with FastMix, a novel framework that automates data mixture discovery while training only a single proxy model. Instead of relying on predefined heuristics or resource-intensive simulations, FastMix jointly optimizes mixture coefficients and model parameters, substantially improving efficiency and scalability over prior approaches. At the core of FastMix is a reformulation of mixture selection as a bilevel optimization problem. Under this reformulation, we show that optimizing mixture ratios is mathematically equivalent to assigning per-source loss weights under uniform source sampling. This embeds the mixture coefficients directly into the differentiable iterative optimization objective, enabling efficient, gradient-based optimization of both mixture and model. To solve the optimization problem, FastMix implements an approximate iterative optimization procedure, alternating between (i) updating model parameters on data sampled according to current mixture ratios (inner loop) and (ii) updating mixture ratios based on validation feedback (outer loop). Across pre- and post-training, FastMix outperforms baselines while drastically reducing search cost: in pre-training, it attains an average score of 48.2 with 1.3 GPU-hours ($\times 550$ vs. RegMix; $\times 55$ vs. CLIMB), and in post-training (SFT) it leads with 65.4 with a $+5.5$ gain over the next best, completing search in 2.2 GPU-hours compared to the 115 GPU-hours required by CLIMB/RegMix.
Verifying the authenticity of AI-generated text has become increasingly important with the rapid advancement of large language models, and unbiased watermarking has emerged as a promising approach due to its ability to preserve output distribution without degrading quality. However, recent work reveals that unbiased watermarks can accumulate distributional bias over multiple generations and that existing robustness evaluations are inconsistent across studies. To address these issues, we introduce UWBench, the first open-source benchmark dedicated to the principled evaluation of unbiased watermarking methods. Our framework combines theoretical and empirical contributions: we propose a statistical metric to quantify multi-batch distribution shift, prove an impossibility result showing that no unbiased watermark can perfectly preserve the distribution under infinite queries, and develop a formal analysis of robustness against token-level modification attacks. Complementing this theory, we establish a three-axis evaluation protocol—unbiasedness, detectability, and robustness—and show that token modification attacks provide more stable robustness assessments than paraphrasing-based methods. Together, UWBench offers the community a standardized and reproducible platform for advancing the design and evaluation of unbiased watermarking algorithms.
Auto-regressive (AR) models, initially successful in language generation, have recently shown promise in visual generation tasks due to their superior sampling efficiency. Unlike image generation, video generation requires a substantially larger number of tokens to produce coherent temporal frames, resulting in significant overhead during decoding. We first make specific key observations: (i) MLP modules in the decode phase dominate the inference latency, and (ii) there exists high temporal redundancy in MLP outputs of adjacent frames. With the insights, we propose **FastCar** to accelerate the decode phase for the AR video generation by exploring the temporal redundancy. The Temporal Attention Score (TAS) is proposed to determine whether to apply the replay strategy (i.e., reusing cached MLP outputs from the previous frame to reduce redundant computations) with detailed theoretical analysis and justification. Furthermore, we develop a hardware accelerator on FPGA with Dynamic Resource Scheduling based on TAS to enable better resource utilization and faster inference. Experimental results demonstrate the effectiveness of our method, which outperforms traditional sparse attention approaches with more than 2.1x decoding speedup and higher energy efficiency on the edge. Furthermore, by combining FastCar and sparse attention, FastCar can boost the performance of sparse attention with alleviated drifting, demonstrating our unique advantages for high-resolution and long-duration video generation. Code: https://github.com/shawnricecake/fast-car
There is a growing interest in Universal Multimodal Embeddings (UME), where models are required to generate task-specific representations. While recent studies show that Multimodal Large Language Models (MLLMs) perform well on such tasks, they treat MLLMs solely as encoders, overlooking their generative capacity. However, such an encoding paradigm becomes less effective as instructions become more complex and require compositional reasoning. Inspired by the proven effectiveness of chain-of-thought reasoning, we propose a general Think-Then-Embed (TTE) framework for UME, composed of a reasoner and an embedder. The reasoner MLLM first generates reasoning traces that explain complex queries, followed by an embedder that produces representations conditioned on both the original query and the intermediate reasoning. This explicit reasoning step enables more nuanced understanding of complex multimodal instructions. Our contributions are threefold. First, by leveraging a powerful MLLM reasoner, we achieve state-of-the-art performance on the MMEB-V2 benchmark, surpassing proprietary models trained on massive in-house datasets. Second, to reduce the dependency on large MLLM reasoners, we finetune a smaller MLLM reasoner using high-quality embedding-centric reasoning traces, achieving the best performance among open-source models with a 7% absolute gain over recently proposed models. Third, we investigate strategies for integrating the reasoner and embedder into a unified model for improved efficiency without sacrificing performance.
Multimodal Large Language Models (MLLMs) that directly process RGB inputs for tasks like 3D localization and navigation have shown remarkable potential. However, we argue that these ``RGB-only'' approaches are fundamentally flawed in their ability to generalize across cameras. By ignoring camera parameters, they entangle an object's physical properties with the camera's perspective, creating an irresolvable ambiguity. We show this leads MLLMs to overfit to the training camera distribution, rather than learning true and generalizable 3D geometric principles. To address this, we propose Camera-Aware MLLM framework for spatial MLLMs. It learns generalizable spatial reasoning by: (i) injecting camera intrinsics via a dense embedding that conditions each visual token; (ii) introducing a camera-aware data augmentation strategy that synthetically varies camera parameters, forcing the model to disentangle camera properties from scene content; and (iii) distilling geometric priors from a 3D vision foundation model. Extensive experiments demonstrate that camera-aware MLLMs substantially outperform their naive counterparts, particularly in cross-camera generalization tests on spatially-grounded tasks, indicating that camera-awareness is not only beneficial but also a prerequisite for robust and generalizable spatial intelligence in MLLMs.
Large reasoning models (LRMs) have achieved remarkable progress on complex tasks by generating extended chains of thought (CoT). However, their uncontrolled output lengths pose significant challenges for real-world deployment, where inference-time budgets on tokens, latency, or compute are strictly constrained. We propose Elastic Reasoning, a novel framework for scalable chain of thoughts that explicitly separates reasoning into two phases—thinking and solution—with independently allocated budgets. At test time, Elastic Reasoning prioritizes the completeness of solution segments, significantly improving reliability under tight resource constraints. To train models that are robust to truncated thinking, we introduce a lightweight budget-constrained rollout strategy, integrated into GRPO, which teaches the model to reason adaptively when the thinking process is cut short and generalizes effectively to unseen budget constraints without additional training. Empirical results on mathematical (AIME, MATH500) and programming (LiveCodeBench, Codeforces) benchmarks demonstrate that Elastic Reasoning performs robustly under strict budget constraints, while incurring significantly lower training cost than baseline methods. Remarkably, our approach also produces more concise and efficient reasoning even in unconstrained settings. Elastic Reasoning offers a principled and practical solution to the pressing challenge of controllable reasoning at scale. Code is available in the supplementary material.
The rapid development and widespread adoption of Audio Large Language Models (ALLMs) require a rigorous assessment of their trustworthiness. However, existing evaluation frameworks, primarily designed for text, are not equipped to handle the unique vulnerabilities introduced by audio’s acoustic properties. We find that significant trustworthiness risks in ALLMs arise from non-semantic acoustic cues, such as timbre, accent, and background noise, which can be used to manipulate model behavior. To address this gap, we propose AudioTrust, the first framework for large-scale and systematic evaluation of ALLM trustworthiness concerning these audio-specific risks. AudioTrust spans six key dimensions: fairness, hallucination, safety, privacy, robustness, and authenticition. It is implemented through 26 distinct sub-tasks and a curated dataset of over 4,420 audio samples collected from real-world scenarios (e.g., daily conversations, emergency calls, and voice assistant interactions), purposefully constructed to probe the trustworthiness of ALLMs across multiple dimensions. Our comprehensive evaluation includes 18 distinct experimental configurations and employs human-validated automated pipelines to objectively and scalably quantify model outputs. Experimental results reveal the boundaries and limitations of 14 state-of-the-art (SOTA) open-source and closed-source ALLMs when confronted with diverse high-risk audio scenarios, thereby offering critical insights into the secure and trustworthy deployment of future audio models. Our platform and benchmark are publicly available at https://github.com/JusperLee/AudioTrust.
Continuous-space video generation has advanced rapidly, while discrete approaches lag behind due to error accumulation and long-context inconsistency. In this work, we revisit discrete generative modeling and present Uniform discRete diffuSion with metric pAth (URSA), a simple yet powerful framework that bridges the gap with continuous approaches for the scalable video generation. At its core, URSA formulates the video generation task as an iterative global refinement of discrete spatiotemporal tokens. It integrates two key designs: a Linearized Metric Path and a Resolution-dependent Timestep Shifting mechanism. These designs enable URSA to scale efficiently to high-resolution image synthesis and long-duration video generation, while requiring significantly fewer inference steps. Additionally, we introduce an asynchronous temporal fine-tuning strategy that unifies versatile tasks within a single model, including interpolation and image-to-video generation. Extensive experiments on challenging video and image generation benchmarks demonstrate that URSA consistently outperforms existing discrete methods and achieves performance comparable to state-of-the-art continuous diffusion methods. Code and models are available at https://github.com/baaivision/URSA.
Recent advances in few-step generative models (typically $1$-$8$ steps), such as consistency models, have yielded impressive performance. However, their broader adoption is hindered by significant challenges, including substantial computational overhead, the reliance on complex multi-component loss functions, and intricate multi-stage training strategies that lack end-to-end simplicity. These limitations impede their scalability and stability, especially when applied to large-scale models. To address these issues, we introduce **$N$-th order Recursive Consistent velocity field estimation for Generative Modeling (RCGM)**, a novel framework that unifies many existing approaches. Within this framework, we reveal that conventional one-step methods, such as consistency and MeanFlow models, are special cases of 1st-order RCGM. This insight enables a natural extension to higher-order scenarios ($N \geq 2$), which exhibit markedly improved training stability and achieve state-of-the-art (SOTA) performance. For instance, on ImageNet $256\times256$, RCGM enables a $675\text{M}$ parameter diffusion transformer to achieve a $1.48$ FID score in just $2$ sampling steps. Crucially, RCGM facilitates the stable full-parameter training of a large-scale ($20\textrm{B}$) unified multi-modal model, attaining a $0.86$ GenEval score in $2$ steps. In contrast, conventional 1st-order approaches, such as consistency and MeanFlow models, typically suffer from training instability, model collapse, or memory constraints under comparable settings. Code is available at: https://github.com/LINs-lab/RCGM.
Although Video Large Language Models (VLLMs) have shown remarkable capabilities in video understanding, they are required to process high volumes of visual tokens, causing significant computational inefficiency. Existing VLLMs acceleration frameworks usually compress spatial and temporal redundancy independently, which overlooks the spatiotemporal relationships, thereby leading to suboptimal spatiotemporal compression. The highly correlated visual features are likely to change in spatial position, scale, orientation, and other attributes over time due to the dynamic nature of video. Building on this insight, we introduce FlashVID, a training-free inference acceleration framework for VLLMs. Specifically, FlashVID utilizes Attention and Diversity-based Token Selection (ADTS) to select the most representative tokens for basic video representation, then applies Tree-based Spatiotemporal Token Merging (TSTM) for fine-grained spatiotemporal redundancy elimination. Extensive experiments conducted on three representative VLLMs across five video understanding benchmarks demonstrate the effectiveness and generalization of our method. Notably, by retaining only $\textbf{10}$% of visual tokens, FlashVID preserves $\textbf{99.1}$% of the performance of LLaVA-OneVision. Consequently, FlashVID can serve as a training-free and plug-and-play module for extending long video frames, which enables a $\textbf{10$\times$}$ increase in video frame input to Qwen2.5-VL, resulting in a relative improvement of $\textbf{8.6}$% within the same computational budget. Code is available at https://github.com/Fanziyang-v/FlashVID.
Large Reasoning Models (LRMs) have advanced rapidly, yet existing benchmarks on mathematics, code, and common-sense reasoning remain limited: they lack long-context evaluation, offer insufficient challenge, and provide answers that are difficult to verify programmatically. We introduce GrAlgoBench, a benchmark designed to evaluate LRMs through graph algorithm problems. Such problems are particularly well-suited for probing reasoning abilities: they demand long-context reasoning, allow fine-grained control of difficulty levels, and enable standardized programmatic evaluation. Across nine tasks, our systematic experiments reveal two major weaknesses of current LRMs. First, accuracy deteriorates sharply with longer contexts input—falling below 50% once graphs exceed 120 nodes—driven by frequent execution errors, weak memory, and redundant reasoning. Second, LRMs suffer from an "over-thinking" phenomenon, primarily driven by extensive yet largely ineffective self-verification, which inflates reasoning traces without improving correctness. By exposing these limitations, GrAlgoBench establishes graph algorithm problems as a rigorous, multidimensional, and practically relevant testbed for advancing the study of reasoning in LRMs. Code is available at https://anonymous.4open.science/r/GrAlgoBench-7D17.
Generative models typically sample outputs independently, and recent inference-time guidance and scaling algorithms focus on improving the quality of individual samples. However, in real-world applications, users are often presented with a set of multiple images (e.g., 4-8) for each prompt, where independent sampling tends to lead to redundant results, limiting user choices and hindering idea exploration. In this work, we introduce a scalable group inference method that improves both the diversity and quality of a group of samples. We formulate group inference as a quadratic integer assignment problem: candidate outputs are modeled as graph nodes, and a subset is selected to optimize sample quality (unary term) while maximizing group diversity (binary term). To substantially improve runtime efficiency, we use intermediate predictions of the final sample at each step to progressively prune the candidate set, allowing our method to scale up efficiently to large input candidate sets. Extensive experiments show that our method significantly improves group diversity and quality compared to independent sampling baselines and recent inference algorithms. Our framework generalizes across a wide range of tasks, including text-to-image, image-to-image, and image prompting, enabling generative models to treat multiple outputs as cohesive groups rather than independent samples.
Ultra-long generation by large language models (LLMs) is a widely demanded scenario, yet it remains a significant challenge due to their maximum generation length limit and overall quality degradation as sequence length increases. Previous approaches, exemplified by LongWriter, typically rely on ''teaching'', which involves supervised fine-tuning (SFT) on synthetic long-form outputs. However, this strategy heavily depends on synthetic SFT data, which is difficult and costly to construct, often lacks coherence and consistency, and tends to be overly artificial and structurally monotonous. In this work, we propose an incentivization-based approach that, starting entirely from scratch and without relying on any annotated or synthetic data, leverages reinforcement learning (RL) to foster the emergence of ultra-long, high-quality text generation capabilities in LLMs. We perform RL training starting from a base model, similar to R1-Zero, guiding it to engage in reasoning that facilitates planning and refinement during the writing process. To support this, we employ specialized reward models that steer the LLM towards improved length control, writing quality, and structural formatting. Experimental evaluations show that our LongWriter-Zero model, trained from Qwen2.5-32B, consistently outperforms traditional SFT methods on long-form writing tasks, achieving state-of-the-art results across all metrics on WritingBench and Arena-Write, and even surpassing 100B+ models such as DeepSeek R1 and Qwen3-235B.