Xiaoxuan Liu

Xiaoxuan Liu, Lianmin Zheng, Dequan Wang et al. · berkeley, tsinghua

Training large neural network (NN) models requires extensive memory resources, and Activation Compressed Training (ACT) is a promising approach to reduce training memory footprint. This paper presents GACT, an ACT framework to support a broad range of machine learning tasks for generic NN architectures with limited domain knowledge. By analyzing a linearized version of ACT's approximate gradient, we prove the convergence of GACT without prior knowledge on operator type or model architecture. To make training stable, we propose an algorithm that decides the compression ratio for each tensor by estimating its impact on the gradient at run time. We implement GACT as a PyTorch library that readily applies to any NN architecture. GACT reduces the activation memory for convolutional NNs, transformers, and graph NNs by up to 8.1x, enabling training with a 4.2x to 24.7x larger batch size, with negligible accuracy loss. We implement GACT as a PyTorch library at https://github.com/LiuXiaoxuanPKU/GACT-ICML.

Xiaoxuan Liu, Lanxiang Hu, Peter Bailis et al.

Speculative decoding is a pivotal technique to accelerate the inference of large language models (LLMs) by employing a smaller draft model to predict the target model's outputs. However, its efficacy can be limited due to the low predictive accuracy of the draft model, particularly when faced with diverse text inputs and a significant capability gap between the draft and target models. We introduce online speculative decoding to address this challenge. The main idea is to continuously update the (multiple) draft model(s) on observed user query data. Adapting to query distribution mitigates the shifts between the training distribution of the draft model and the query distribution, enabling the draft model to more accurately predict the target model's outputs. We develop a prototype of online speculative decoding based on knowledge distillation and evaluate it using both synthetic and real query data. The results show a substantial increase in the token acceptance rate by 0.1 to 0.65, bringing 1.42x to 2.17x latency reduction. Our code is available at https://github.com/LiuXiaoxuanPKU/OSD.

Zhikai Li, Xiaoxuan Liu, Banghua Zhu et al. · berkeley

Large Language Models (LLMs) have showcased remarkable impacts across a wide spectrum of natural language processing tasks. Fine-tuning these pretrained models on downstream datasets provides further significant performance gains; however, this process typically requires a large number of expensive, high-end GPUs. Although there have been efforts focused on parameter-efficient fine-tuning, they cannot fully unlock the powerful potential of full-parameter fine-tuning. In this paper, we propose QFT, a Quantized Full-parameter Tuning framework for LLMs that quantizes and stores all training states, including weights, gradients, and optimizer states, in INT8 format to reduce training memory, thereby enabling full-parameter fine-tuning on existing GPUs at an affordable cost. To ensure training performance, we make two key efforts: i) for quantized gradients and optimizer states, we theoretically prove that the Lion optimizer, with its property of consistent update magnitudes, is highly robust to quantization; ii) and for quantized weights, we employ the hybrid feature quantizer, which identifies and protects a small subset of sparse critical features while quantizing the remaining dense features, thus ensuring accurate weight updates without FP32 backups. Moreover, to support backpropagation in the integer context, we develop a stack-based gradient flow scheme with O(1) complexity, forming a unified integer training pipeline. As a result, QFT reduces the model state memory to 21% of the standard solution while achieving comparable performance, e.g., tuning a LLaMA-7B model requires only <30GB of memory, making it feasible on a single A6000 GPU.

Longfei Yun, Yihan Wu, Haoran Liu et al.

Modern large-scale ranking systems operate within a sophisticated landscape of competing objectives, operational constraints, and evolving product requirements. Progress in this domain is increasingly bottlenecked by the engineering context constraint: the arduous process of translating ambiguous product intent into reasonable, executable, verifiable hypotheses, rather than by modeling techniques alone. We present GEARS (Generative Engine for Agentic Ranking Systems), a framework that reframes ranking optimization as an autonomous discovery process within a programmable experimentation environment. Rather than treating optimization as static model selection, GEARS leverages Specialized Agent Skills to encapsulate ranking expert knowledge into reusable reasoning capabilities, enabling operators to steer systems via high-level intent vibe personalization. Furthermore, to ensure production reliability, the framework incorporates validation hooks to enforce statistical robustness and filter out brittle policies that overfit short-term signals. Experimental validation across diverse product surfaces demonstrates that GEARS consistently identifies superior, near-Pareto-efficient policies by synergizing algorithmic signals with deep ranking context while maintaining rigorous deployment stability.

Enze Zhu, Zhan Chen, Dingkai Wang et al.

Semantic segmentation of high-resolution remote sensing images is vital in downstream applications such as land-cover mapping, urban planning and disaster assessment.Existing Transformer-based methods suffer from the constraint between accuracy and efficiency, while the recently proposed Mamba is renowned for being efficient. Therefore, to overcome the dilemma, we propose UNetMamba, a UNet-like semantic segmentation model based on Mamba. It incorporates a mamba segmentation decoder (MSD) that can efficiently decode the complex information within high-resolution images, and a local supervision module (LSM), which is train-only but can significantly enhance the perception of local contents. Extensive experiments demonstrate that UNetMamba outperforms the state-of-the-art methods with mIoU increased by 0.87% on LoveDA and 0.39% on ISPRS Vaihingen, while achieving high efficiency through the lightweight design, less memory footprint and reduced computational cost. The source code is available at https://github.com/EnzeZhu2001/UNetMamba.

Beatriz Costa-Gomes, Pavel Tolmachev, Eloise Taysom et al.

We analyze over 500,000 de-identified health-related conversations with Microsoft Copilot from January 2026 to characterize what people ask conversational AI about health. We develop a hierarchical intent taxonomy of 12 primary categories using privacy-preserving LLM-based classification validated against expert human annotation, and apply LLM-driven topic-clustering for prevalent themes within each intent. Using this taxonomy, we characterize the intents and topics behind health queries, identify who these queries are about, and analyze how usage varies by device and time of day. Five findings stand out. First, nearly one in five conversations involve personal symptom assessment or condition discussion, and even the dominant general information category (40%) is concentrated on specific treatments and conditions, suggesting that this is a lower bound on personal health intent. Second, one in seven of these personal health queries concern someone other than the user, such as a child, a parent, a partner, suggesting that conversational AI can be a caregiving tool, not just a personal one. Third, personal queries about symptoms and emotional health queries increase markedly in the evening and nighttime hours, when traditional healthcare is most limited. Fourth, usage diverges sharply by device: mobile concentrates on personal health concerns, while desktop is dominated by professional and academic work. Fifth, a substantial share of queries focuses on navigating healthcare systems such as finding providers, and understanding insurance, highlighting friction in the delivery of existing healthcare. These patterns have direct implications for platform-specific design, safety considerations, and the responsible development of health AI.

Qiuyang Mang, Runyuan He, Suyang Zhong et al.

Since 2020, automated testing for Database Management Systems (DBMSs) has flourished, uncovering hundreds of bugs in widely-used systems. A cornerstone of these techniques is test oracle, which typically implements a mechanism to generate equivalent query pairs, thereby identifying bugs by checking the consistency between their results. However, while applying these oracles can be automated, their design remains a fundamentally manual endeavor. This paper explores the use of large language models (LLMs) to automate the discovery and instantiation of test oracles, addressing a long-standing bottleneck towards fully automated DBMS testing. Although LLMs demonstrate impressive creativity, they are prone to hallucinations that can produce numerous false positive bug reports. Furthermore, their significant monetary cost and latency mean that LLM invocations should be limited to ensure that bug detection is efficient and economical. To this end, we introduce Argus, a novel framework built upon the core concept of the Constrained Abstract Query - a SQL skeleton containing placeholders and their associated instantiation conditions (e.g., requiring a placeholder to be filled by a boolean column). Argus uses LLMs to generate pairs of these skeletons that are asserted to be semantically equivalent. This equivalence is then formally proven using a SQL equivalence solver to ensure soundness. Finally, the placeholders within the verified skeletons are instantiated with concrete, reusable SQL snippets that are also synthesized by LLMs to efficiently produce complex test cases. We implemented Argus and evaluated it on five extensively tested DBMSs, discovering 40 previously unknown bugs, 35 of which are logic bugs, with 36 confirmed and 26 already fixed by the developers.

Zile Guo, Zhan Chen, Enze Zhu et al.

Recent advances in world models have demonstrated strong capabilities in simulating physical reality, making them an increasingly important foundation for embodied intelligence. For UAV agents in particular, accurate prediction of complex 3D dynamics is essential for autonomous navigation and robust decision-making in unconstrained environments. However, under the highly dynamic camera trajectories typical of UAV views, existing world models often struggle to maintain spatiotemporal physical consistency. A key reason lies in the distribution bias of current training data: most existing datasets exhibit restricted 2.5D motion patterns, such as ground-constrained autonomous driving scenes or relatively smooth human-centric egocentric videos, and therefore lack realistic high-dynamic 6-DoF UAV motion priors. To address this gap, we present MotionScape, a large-scale real-world UAV-view video dataset with highly dynamic motion for world modeling. MotionScape contains over 30 hours of 4K UAV-view videos, totaling more than 4.5M frames. This novel dataset features semantically and geometrically aligned training samples, where diverse real-world UAV videos are tightly coupled with accurate 6-DoF camera trajectories and fine-grained natural language descriptions. To build the dataset, we develop an automated multi-stage processing pipeline that integrates CLIP-based relevance filtering, temporal segmentation, robust visual SLAM for trajectory recovery, and large-language-model-driven semantic annotation. Extensive experiments show that incorporating such semantically and geometrically aligned annotations effectively improves the ability of existing world models to simulate complex 3D dynamics and handle large viewpoint shifts, thereby benefiting decision-making and planning for UAV agents in complex environments. The dataset is publicly available at https://github.com/Thelegendzz/MotionScape

Xiaoxuan Liu, Jiaxiang Yu, Jongseok Park et al.

Speculative decoding (SD) has become a popular technique to accelerate Large Language Model (LLM) inference, yet its real-world effectiveness remains unclear as prior evaluations rely on research prototypes and unrealistically small batch sizes. We present, to our knowledge, the first systematic study of SD on a production-grade and widely deployed inference engine (vLLM), covering multiple SD variants ($n$-gram, EAGLE/EAGLE-3, Draft-Model, Multi-Token Prediction) across diverse workloads, model scales, and batch sizes. We analyze key factors governing SD performance, and quantify a theoretical upper bound on SD speedup. Our results show that verification by the target model dominates the execution, while acceptance length varies markedly across output token positions, requests, and datasets. Comparing measured performance with theoretical bounds reveals substantial gaps between observed and theoretical upper bounds, and we leverage this observation to highlight new research opportunities that our study opens up in improving SD.

Xiaoxuan Liu, Siddharth Jha, Alvin Cheung

As models continue to grow in size, the development of memory optimization methods (MOMs) has emerged as a solution to address the memory bottleneck encountered when training large models. To comprehensively examine the practical value of various MOMs, we have conducted a thorough analysis of existing literature from a systems perspective. Our analysis has revealed a notable challenge within the research community: the absence of standardized metrics for effectively evaluating the efficacy of MOMs. The scarcity of informative evaluation metrics hinders the ability of researchers and practitioners to compare and benchmark different approaches reliably. Consequently, drawing definitive conclusions and making informed decisions regarding the selection and application of MOMs becomes a challenging endeavor. To address the challenge, this paper summarizes the scenarios in which MOMs prove advantageous for model training. We propose the use of distinct evaluation metrics under different scenarios. By employing these metrics, we evaluate the prevailing MOMs and find that their benefits are not universal. We present insights derived from experiments and discuss the circumstances in which they can be advantageous.

Ran Wang, Xiaoxuan Liu, Hao Ren et al. · tsinghua

Structured decoding enables large language models (LLMs) to generate outputs in formats required by downstream systems, such as HTML or JSON. However, existing methods suffer from efficiency bottlenecks due to grammar compilation, state tracking, and mask creation. We observe that many real-world tasks embed strong prior knowledge about output structure. Leveraging this, we propose a decomposition of constraints into static and dynamic components -- precompiling static structures offline and instantiating dynamic arguments at runtime using grammar snippets. Instead of relying on pushdown automata, we employ a compositional set of operators to model regular formats, achieving lower transition latency. We introduce wgrammar, a lightweight decoding engine that integrates domain-aware simplification, constraint decomposition, and mask caching, achieving up to 250x speedup over existing systems. wgrammar's source code is publicly available at https://github.com/wrran/wgrammar.

Zhan Chen, Zile Guo, Enze Zhu et al.

Video prediction is plagued by a fundamental trilemma: achieving high-resolution and perceptual quality typically comes at the cost of real-time speed, hindering its use in latency-critical applications. This challenge is most acute for autonomous UAVs in dense urban environments, where foreseeing events from high-resolution imagery is non-negotiable for safety. Existing methods, reliant on iterative generation (diffusion, autoregressive models) or quadratic-complexity attention, fail to meet these stringent demands on edge hardware. To break this long-standing trade-off, we introduce RAPTOR, a video prediction architecture that achieves real-time, high-resolution performance. RAPTOR's single-pass design avoids the error accumulation and latency of iterative approaches. Its core innovation is Efficient Video Attention (EVA), a novel translator module that factorizes spatiotemporal modeling. Instead of processing flattened spacetime tokens with $O((ST)^2)$ or $O(ST)$ complexity, EVA alternates operations along the spatial (S) and temporal (T) axes. This factorization reduces the time complexity to $O(S + T)$ and memory complexity to $O(max(S, T))$, enabling global context modeling at $512^2$ resolution and beyond, operating directly on dense feature maps with a patch-free design. Complementing this architecture is a 3-stage training curriculum that progressively refines predictions from coarse structure to sharp, temporally coherent details. Experiments show RAPTOR is the first predictor to exceed 30 FPS on a Jetson AGX Orin for $512^2$ video, setting a new state-of-the-art on UAVid, KTH, and a custom high-resolution dataset in PSNR, SSIM, and LPIPS. Critically, RAPTOR boosts the mission success rate in a real-world UAV navigation task by 18%, paving the way for safer and more anticipatory embodied agents.

Shiyi Cao, Shu Liu, Tyler Griggs et al.

Efficient deployment of large language models, particularly Mixture of Experts (MoE), on resource-constrained platforms presents significant challenges, especially in terms of computational efficiency and memory utilization. The MoE architecture, renowned for its ability to increase model capacity without a proportional increase in inference cost, greatly reduces the token generation latency compared with dense models. However, the large model size makes MoE models inaccessible to individuals without high-end GPUs. In this paper, we propose a high-throughput MoE batch inference system, that significantly outperforms past work. MoE-Lightning introduces a novel CPU-GPU-I/O pipelining schedule, CGOPipe, with paged weights to achieve high resource utilization, and a performance model, HRM, based on a Hierarchical Roofline Model we introduce to help find policies with higher throughput than existing systems. MoE-Lightning can achieve up to 10.3x higher throughput than state-of-the-art offloading-enabled LLM inference systems for Mixtral 8x7B on a single T4 GPU (16GB). When the theoretical system throughput is bounded by the GPU memory, MoE-Lightning can reach the throughput upper bound with 2-3x less CPU memory, significantly increasing resource utilization. MoE-Lightning also supports efficient batch inference for much larger MoEs (e.g., Mixtral 8x22B and DBRX) on multiple low-cost GPUs (e.g., 2-4 T4).

Harsha Nori, Mayank Daswani, Christopher Kelly et al.

Artificial intelligence holds great promise for expanding access to expert medical knowledge and reasoning. However, most evaluations of language models rely on static vignettes and multiple-choice questions that fail to reflect the complexity and nuance of evidence-based medicine in real-world settings. In clinical practice, physicians iteratively formulate and revise diagnostic hypotheses, adapting each subsequent question and test to what they've just learned, and weigh the evolving evidence before committing to a final diagnosis. To emulate this iterative process, we introduce the Sequential Diagnosis Benchmark, which transforms 304 diagnostically challenging New England Journal of Medicine clinicopathological conference (NEJM-CPC) cases into stepwise diagnostic encounters. A physician or AI begins with a short case abstract and must iteratively request additional details from a gatekeeper model that reveals findings only when explicitly queried. Performance is assessed not just by diagnostic accuracy but also by the cost of physician visits and tests performed. We also present the MAI Diagnostic Orchestrator (MAI-DxO), a model-agnostic orchestrator that simulates a panel of physicians, proposes likely differential diagnoses and strategically selects high-value, cost-effective tests. When paired with OpenAI's o3 model, MAI-DxO achieves 80% diagnostic accuracy--four times higher than the 20% average of generalist physicians. MAI-DxO also reduces diagnostic costs by 20% compared to physicians, and 70% compared to off-the-shelf o3. When configured for maximum accuracy, MAI-DxO achieves 85.5% accuracy. These performance gains with MAI-DxO generalize across models from the OpenAI, Gemini, Claude, Grok, DeepSeek, and Llama families. We highlight how AI systems, when guided to think iteratively and act judiciously, can advance diagnostic precision and cost-effectiveness in clinical care.

Tyler Griggs, Xiaoxuan Liu, Jiaxiang Yu et al.

Large language models (LLMs) are increasingly integrated into many online services, yet they remain cost-prohibitive to deploy due to the requirement of expensive GPU instances. Prior work has addressed the high cost of LLM serving by improving the inference engine, but less attention has been given to selecting the most cost-efficient GPU type(s) for a specific LLM service. There is a large and growing landscape of GPU types and, within these options, higher cost does not always lead to increased performance. Instead, through a comprehensive investigation, we find that three key LLM service characteristics (request size, request rate, SLO) strongly influence GPU cost efficiency, and differing GPU types are most cost efficient for differing LLM service settings. As a result, the most cost-efficient allocation for a given service is typically a mix of heterogeneous GPU types. Based on this analysis, we introduce Mélange, a GPU allocation framework that navigates these diverse LLM service characteristics and heterogeneous GPU option space to automatically and efficiently derive the minimal-cost GPU allocation for a given LLM service. We formulate the GPU allocation task as a cost-aware bin packing problem where GPUs are bins and items are slices of the service workload. Our formulation's constraints account for a service's unique characteristics, allowing Mélange to be flexible to support diverse service settings and heterogeneity-aware to adapt the GPU allocation to a specific service. Compared to using only a single GPU type, Mélange reduces deployment costs by up to 77% in conversational settings, 33% in document-based settings, and 51% in a mixed setting.

Yao Lu, Song Bian, Lequn Chen et al.

In this paper, we investigate the intersection of large generative AI models and cloud-native computing architectures. Recent large models such as ChatGPT, while revolutionary in their capabilities, face challenges like escalating costs and demand for high-end GPUs. Drawing analogies between large-model-as-a-service (LMaaS) and cloud database-as-a-service (DBaaS), we describe an AI-native computing paradigm that harnesses the power of both cloud-native technologies (e.g., multi-tenancy and serverless computing) and advanced machine learning runtime (e.g., batched LoRA inference). These joint efforts aim to optimize costs-of-goods-sold (COGS) and improve resource accessibility. The journey of merging these two domains is just at the beginning and we hope to stimulate future research and development in this area.

Jasmine Chiat Ling Ong, Yilin Ning, Mingxuan Liu et al.

The integration of generative AI (GenAI) and large language models (LLMs) in healthcare presents both unprecedented opportunities and challenges, necessitating innovative regulatory approaches. GenAI and LLMs offer broad applications, from automating clinical workflows to personalizing diagnostics. However, the non-deterministic outputs, broad functionalities and complex integration of GenAI and LLMs challenge existing medical device regulatory frameworks, including the total product life cycle (TPLC) approach. Here we discuss the constraints of the TPLC approach to GenAI and LLM-based medical device regulation, and advocate for global collaboration in regulatory science research. This serves as the foundation for developing innovative approaches including adaptive policies and regulatory sandboxes, to test and refine governance in real-world settings. International harmonization, as seen with the International Medical Device Regulators Forum, is essential to manage implications of LLM on global health, including risks of widening health inequities driven by inherent model biases. By engaging multidisciplinary expertise, prioritizing iterative, data-driven approaches, and focusing on the needs of diverse populations, global regulatory science research enables the responsible and equitable advancement of LLM innovations in healthcare.

Siddharth Jha, Coleman Hooper, Xiaoxuan Liu et al.

Many applications must provide low-latency LLM service to users or risk unacceptable user experience. However, over-provisioning resources to serve fluctuating request patterns is often prohibitively expensive. In this work, we present a best-effort serving system that employs deep reinforcement learning to adjust service quality based on the task distribution and system load. Our best-effort system can maintain availability with over 10x higher client request rates, serves above 96% of peak performance 4.1x more often, and serves above 98% of peak performance 2.3x more often than static serving on unpredictable workloads. Our learned router is robust to shifts in both the arrival and task distribution. Compared to static serving, learned best-effort serving allows for cost-efficient serving through increased hardware utility. Additionally, we argue that learned best-effort LLM serving is applicable in wide variety of settings and provides application developers great flexibility to meet their specific needs.

Mert Cemri, Nived Rajaraman, Rishabh Tiwari et al. · berkeley

Scaling test-time compute has driven the recent advances in the reasoning capabilities of large language models (LLMs), typically by allocating additional computation for more thorough exploration. However, increased compute often comes at the expense of higher user-facing latency, directly impacting user experience. Current test-time scaling methods primarily optimize for accuracy based on total compute resources (FLOPS), often overlooking latency constraints. To address this gap, we propose $\texttt{SPECS}$, a latency-aware test-time scaling method inspired by speculative decoding. $\texttt{SPECS}$~uses a smaller, faster model to generate candidate sequences efficiently, and evaluates these candidates using signals from both a larger target model and a dedicated reward model. We introduce new integration strategies, including reward-guided soft verification and a reward-based deferral mechanism. Empirical results on MATH500, AMC23 and OlympiadBench datasets show that $\texttt{SPECS}$~matches or surpasses beam search accuracy while reducing latency by up to $\sim$19.1\%. Our theoretical analysis shows that our algorithm converges to the solution of a KL-regularized reinforcement learning objective with increasing beam width.

Yue Zhang, Jiaxin Zhang, Qiuyu Ren et al.

We introduce \textbf{GAUSS} (\textbf{G}eneral \textbf{A}ssessment of \textbf{U}nderlying \textbf{S}tructured \textbf{S}kills in Mathematics), a benchmark that evaluates LLMs' mathematical abilities across twelve core skill dimensions, grouped into three domains: knowledge and understanding, problem solving and communication, and meta-skills and creativity. By categorizing problems according to cognitive skills and designing tasks that isolate specific abilities, GAUSS constructs comprehensive, fine-grained, and interpretable profiles of models' mathematical abilities. These profiles faithfully represent their underlying mathematical intelligence. To exemplify how to use the \textsc{GAUSS} benchmark, we have derived the skill profile of \textsc{GPT-5-thinking}, revealing its strengths and weaknesses as well as its differences relative to \textsc{o4-mini-high}, thereby underscoring the value of multidimensional, skill-based evaluation.

Xiaoxuan Liu, Jongseok Park, Langxiang Hu et al.

Large Language Model (LLM) serving systems batch concurrent user requests to achieve efficient serving. However, in real-world deployments, such inter-request parallelism from batching is often limited by external factors such as low request rates or memory constraints. Recent works focus on intra-request parallelism from speculative decoding as a solution to this problem. Unfortunately, benefits from intra-request parallelism are often fragile, as speculative decoding causes overhead, and speculated tokens may miss. We observe that speculative decoding may degrade LLM serving performance if added naively without tuning to the incoming requests and the speculation method. To alleviate the need for expert tuning and make speculative decoding more robust, we present TurboSpec, a speculation control system that automatically profiles the execution environment and utilizes a feedback-based algorithm to dynamically adjust the amount of intra-request parallelism in LLM serving. TurboSpec predicts "goodput" - the amount of successfully generated tokens - to evaluate and adjust intra-request parallelism amount to that with the highest goodput in runtime. We implement TurboSpec on a real-world LLM serving system vLLM and demonstrate its effectiveness across diverse workloads and hardware configurations, providing consistent performance improvements across all test scenarios.